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         int rc;
 615
 616         f = __vxlan_find_mac(vxlan, mac);
 617         if (f) {
 618                 if (flags & NLM_F_EXCL) {
 619                         netdev_dbg(vxlan->dev,
 620                                    "lost race to create %pM\n", mac);
 621                         return -EEXIST;
 622                 }
 623                 if (f->state != state) {
 624                         f->state = state;
 625                         f->updated = jiffies;
 626                         notify = 1;
 627                 }
 628                 if (f->flags != ndm_flags) {
 629                         f->flags = ndm_flags;
 630                         f->updated = jiffies;
 631                         notify = 1;
 632                 }
 633                 if ((flags & NLM_F_REPLACE)) {
 634                         /* Only change unicasts */
 635                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 636                              is_zero_ether_addr(f->eth_addr))) {
 637                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 638                                                            ifindex);
 639                         } else
 640                                 return -EOPNOTSUPP;
 641                 }
 642                 if ((flags & NLM_F_APPEND) &&
 643                     (is_multicast_ether_addr(f->eth_addr) ||
 644                      is_zero_ether_addr(f->eth_addr))) {
 645                         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 646
 647                         if (rc < 0)
 648                                 return rc;
 649                         notify |= rc;
 650                 }
 651         } else {
 652                 if (!(flags & NLM_F_CREATE))
 653                         return -ENOENT;
 654
 655                 if (vxlan->cfg.addrmax &&
 656                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 657                         return -ENOSPC;
 658
 659                 /* Disallow replace to add a multicast entry */
 660                 if ((flags & NLM_F_REPLACE) &&
 661                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 662                         return -EOPNOTSUPP;
 663
 664                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 665                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 666                 if (!f)
 667                         return -ENOMEM;
 668
 669                 notify = 1;
 670                 f->state = state;
 671                 f->flags = ndm_flags;
 672                 f->updated = f->used = jiffies;
 673                 INIT_LIST_HEAD(&f->remotes);
 674                 memcpy(f->eth_addr, mac, ETH_ALEN);
 675
 676                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 677                 if (rc < 0) {
 678                         kfree(f);
 679                         return rc;
 680                 }
 681
 682                 ++vxlan->addrcnt;
 683                 hlist_add_head_rcu(&f->hlist,
 684                                    vxlan_fdb_head(vxlan, mac));
 685         }
 686
 687         if (notify) {
 688                 if (rd == NULL)
 689                         rd = first_remote_rtnl(f);
 690                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 691         }
 692
 693         return 0;
 694 }
 695
 696 static void vxlan_fdb_free(struct rcu_head *head)
 697 {
 698         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 699         struct vxlan_rdst *rd, *nd;
 700
 701         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 702                 dst_cache_destroy(&rd->dst_cache);
 703                 kfree(rd);
 704         }
 705         kfree(f);
 706 }
 707
 708 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 709 {
 710         netdev_dbg(vxlan->dev,
 711                     "delete %pM\n", f->eth_addr);
 712
 713         --vxlan->addrcnt;
 714         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 715
 716         hlist_del_rcu(&f->hlist);
 717         call_rcu(&f->rcu, vxlan_fdb_free);
 718 }
 719
 720 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 721                            union vxlan_addr *ip, __be16 *port, __be32 *vni,
 722                            u32 *ifindex)
 723 {
 724         struct net *net = dev_net(vxlan->dev);
 725         int err;
 726
 727         if (tb[NDA_DST]) {
 728                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 729                 if (err)
 730                         return err;
 731         } else {
 732                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 733                 if (remote->sa.sa_family == AF_INET) {
 734                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 735                         ip->sa.sa_family = AF_INET;
 736 #if IS_ENABLED(CONFIG_IPV6)
 737                 } else {
 738                         ip->sin6.sin6_addr = in6addr_any;
 739                         ip->sa.sa_family = AF_INET6;
 740 #endif
 741                 }
 742         }
 743
 744         if (tb[NDA_PORT]) {
 745                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 746                         return -EINVAL;
 747                 *port = nla_get_be16(tb[NDA_PORT]);
 748         } else {
 749                 *port = vxlan->cfg.dst_port;
 750         }
 751
 752         if (tb[NDA_VNI]) {
 753                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 754                         return -EINVAL;
 755                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 756         } else {
 757                 *vni = vxlan->default_dst.remote_vni;
 758         }
 759
 760         if (tb[NDA_IFINDEX]) {
 761                 struct net_device *tdev;
 762
 763                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 764                         return -EINVAL;
 765                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 766                 tdev = __dev_get_by_index(net, *ifindex);
 767                 if (!tdev)
 768                         return -EADDRNOTAVAIL;
 769         } else {
 770                 *ifindex = 0;
 771         }
 772
 773         return 0;
 774 }
 775
 776 /* Add static entry (via netlink) */
 777 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 778                          struct net_device *dev,
 779                          const unsigned char *addr, u16 vid, u16 flags)
 780 {
 781         struct vxlan_dev *vxlan = netdev_priv(dev);
 782         /* struct net *net = dev_net(vxlan->dev); */
 783         union vxlan_addr ip;
 784         __be16 port;
 785         __be32 vni;
 786         u32 ifindex;
 787         int err;
 788
 789         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 790                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 791                         ndm->ndm_state);
 792                 return -EINVAL;
 793         }
 794
 795         if (tb[NDA_DST] == NULL)
 796                 return -EINVAL;
 797
 798         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 799         if (err)
 800                 return err;
 801
 802         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 803                 return -EAFNOSUPPORT;
 804
 805         spin_lock_bh(&vxlan->hash_lock);
 806         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 807                                port, vni, ifindex, ndm->ndm_flags);
 808         spin_unlock_bh(&vxlan->hash_lock);
 809
 810         return err;
 811 }
 812
 813 /* Delete entry (via netlink) */
 814 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 815                             struct net_device *dev,
 816                             const unsigned char *addr, u16 vid)
 817 {
 818         struct vxlan_dev *vxlan = netdev_priv(dev);
 819         struct vxlan_fdb *f;
 820         struct vxlan_rdst *rd = NULL;
 821         union vxlan_addr ip;
 822         __be16 port;
 823         __be32 vni;
 824         u32 ifindex;
 825         int err;
 826
 827         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 828         if (err)
 829                 return err;
 830
 831         err = -ENOENT;
 832
 833         spin_lock_bh(&vxlan->hash_lock);
 834         f = vxlan_find_mac(vxlan, addr);
 835         if (!f)
 836                 goto out;
 837
 838         if (!vxlan_addr_any(&ip)) {
 839                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 840                 if (!rd)
 841                         goto out;
 842         }
 843
 844         err = 0;
 845
 846         /* remove a destination if it's not the only one on the list,
 847          * otherwise destroy the fdb entry
 848          */
 849         if (rd && !list_is_singular(&f->remotes)) {
 850                 list_del_rcu(&rd->list);
 851                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 852                 kfree_rcu(rd, rcu);
 853                 goto out;
 854         }
 855
 856         vxlan_fdb_destroy(vxlan, f);
 857
 858 out:
 859         spin_unlock_bh(&vxlan->hash_lock);
 860
 861         return err;
 862 }
 863
 864 /* Dump forwarding table */
 865 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 866                           struct net_device *dev,
 867                           struct net_device *filter_dev, int *idx)
 868 {
 869         struct vxlan_dev *vxlan = netdev_priv(dev);
 870         unsigned int h;
 871         int err = 0;
 872
 873         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 874                 struct vxlan_fdb *f;
 875
 876                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 877                         struct vxlan_rdst *rd;
 878
 879                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 880                                 if (*idx < cb->args[2])
 881                                         goto skip;
 882
 883                                 err = vxlan_fdb_info(skb, vxlan, f,
 884                                                      NETLINK_CB(cb->skb).portid,
 885                                                      cb->nlh->nlmsg_seq,
 886                                                      RTM_NEWNEIGH,
 887                                                      NLM_F_MULTI, rd);
 888                                 if (err < 0)
 889                                         goto out;
 890 skip:
 891                                 *idx += 1;
 892                         }
 893                 }
 894         }
 895 out:
 896         return err;
 897 }
 898
 899 /* Watch incoming packets to learn mapping between Ethernet address
 900  * and Tunnel endpoint.
 901  * Return true if packet is bogus and should be dropped.
 902  */
 903 static bool vxlan_snoop(struct net_device *dev,
 904                         union vxlan_addr *src_ip, const u8 *src_mac)
 905 {
 906         struct vxlan_dev *vxlan = netdev_priv(dev);
 907         struct vxlan_fdb *f;
 908
 909         f = vxlan_find_mac(vxlan, src_mac);
 910         if (likely(f)) {
 911                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 912
 913                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 914                         return false;
 915
 916                 /* Don't migrate static entries, drop packets */
 917                 if (f->state & NUD_NOARP)
 918                         return true;
 919
 920                 if (net_ratelimit())
 921                         netdev_info(dev,
 922                                     "%pM migrated from %pIS to %pIS\n",
 923                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 924
 925                 rdst->remote_ip = *src_ip;
 926                 f->updated = jiffies;
 927                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 928         } else {
 929                 /* learned new entry */
 930                 spin_lock(&vxlan->hash_lock);
 931
 932                 /* close off race between vxlan_flush and incoming packets */
 933                 if (netif_running(dev))
 934                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 935                                          NUD_REACHABLE,
 936                                          NLM_F_EXCL|NLM_F_CREATE,
 937                                          vxlan->cfg.dst_port,
 938                                          vxlan->default_dst.remote_vni,
 939                                          0, NTF_SELF);
 940                 spin_unlock(&vxlan->hash_lock);
 941         }
 942
 943         return false;
 944 }
 945
 946 /* See if multicast group is already in use by other ID */
 947 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
 948 {
 949         struct vxlan_dev *vxlan;
 950         struct vxlan_sock *sock4;
 951 #if IS_ENABLED(CONFIG_IPV6)
 952         struct vxlan_sock *sock6;
 953 #endif
 954         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
 955
 956         sock4 = rtnl_dereference(dev->vn4_sock);
 957
 958         /* The vxlan_sock is only used by dev, leaving group has
 959          * no effect on other vxlan devices.
 960          */
 961         if (family == AF_INET && sock4 && atomic_read(&sock4->refcnt) == 1)
 962                 return false;
 963 #if IS_ENABLED(CONFIG_IPV6)
 964         sock6 = rtnl_dereference(dev->vn6_sock);
 965         if (family == AF_INET6 && sock6 && atomic_read(&sock6->refcnt) == 1)
 966                 return false;
 967 #endif
 968
 969         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
 970                 if (!netif_running(vxlan->dev) || vxlan == dev)
 971                         continue;
 972
 973                 if (family == AF_INET &&
 974                     rtnl_dereference(vxlan->vn4_sock) != sock4)
 975                         continue;
 976 #if IS_ENABLED(CONFIG_IPV6)
 977                 if (family == AF_INET6 &&
 978                     rtnl_dereference(vxlan->vn6_sock) != sock6)
 979                         continue;
 980 #endif
 981
 982                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
 983                                       &dev->default_dst.remote_ip))
 984                         continue;
 985
 986                 if (vxlan->default_dst.remote_ifindex !=
 987                     dev->default_dst.remote_ifindex)
 988                         continue;
 989
 990                 return true;
 991         }
 992
 993         return false;
 994 }
 995
 996 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
 997 {
 998         struct vxlan_net *vn;
 999
1000         if (!vs)
1001                 return false;
1002         if (!atomic_dec_and_test(&vs->refcnt))
1003                 return false;
1004
1005         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1006         spin_lock(&vn->sock_lock);
1007         hlist_del_rcu(&vs->hlist);
1008         udp_tunnel_notify_del_rx_port(vs->sock,
1009                                       (vs->flags & VXLAN_F_GPE) ?
1010                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1011                                       UDP_TUNNEL_TYPE_VXLAN);
1012         spin_unlock(&vn->sock_lock);
1013
1014         return true;
1015 }
1016
1017 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1018 {
1019         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1020 #if IS_ENABLED(CONFIG_IPV6)
1021         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1022
1023         rcu_assign_pointer(vxlan->vn6_sock, NULL);
1024 #endif
1025
1026         rcu_assign_pointer(vxlan->vn4_sock, NULL);
1027         synchronize_net();
1028
1029         if (__vxlan_sock_release_prep(sock4)) {
1030                 udp_tunnel_sock_release(sock4->sock);
1031                 kfree(sock4);
1032         }
1033
1034 #if IS_ENABLED(CONFIG_IPV6)
1035         if (__vxlan_sock_release_prep(sock6)) {
1036                 udp_tunnel_sock_release(sock6->sock);
1037                 kfree(sock6);
1038         }
1039 #endif
1040 }
1041
1042 /* Update multicast group membership when first VNI on
1043  * multicast address is brought up
1044  */
1045 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1046 {
1047         struct sock *sk;
1048         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1049         int ifindex = vxlan->default_dst.remote_ifindex;
1050         int ret = -EINVAL;
1051
1052         if (ip->sa.sa_family == AF_INET) {
1053                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1054                 struct ip_mreqn mreq = {
1055                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1056                         .imr_ifindex            = ifindex,
1057                 };
1058
1059                 sk = sock4->sock->sk;
1060                 lock_sock(sk);
1061                 ret = ip_mc_join_group(sk, &mreq);
1062                 release_sock(sk);
1063 #if IS_ENABLED(CONFIG_IPV6)
1064         } else {
1065                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1066
1067                 sk = sock6->sock->sk;
1068                 lock_sock(sk);
1069                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1070                                                    &ip->sin6.sin6_addr);
1071                 release_sock(sk);
1072 #endif
1073         }
1074
1075         return ret;
1076 }
1077
1078 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1079 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1080 {
1081         struct sock *sk;
1082         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1083         int ifindex = vxlan->default_dst.remote_ifindex;
1084         int ret = -EINVAL;
1085
1086         if (ip->sa.sa_family == AF_INET) {
1087                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1088                 struct ip_mreqn mreq = {
1089                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1090                         .imr_ifindex            = ifindex,
1091                 };
1092
1093                 sk = sock4->sock->sk;
1094                 lock_sock(sk);
1095                 ret = ip_mc_leave_group(sk, &mreq);
1096                 release_sock(sk);
1097 #if IS_ENABLED(CONFIG_IPV6)
1098         } else {
1099                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1100
1101                 sk = sock6->sock->sk;
1102                 lock_sock(sk);
1103                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1104                                                    &ip->sin6.sin6_addr);
1105                 release_sock(sk);
1106 #endif
1107         }
1108
1109         return ret;
1110 }
1111
1112 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1113                           struct sk_buff *skb, u32 vxflags)
1114 {
1115         size_t start, offset;
1116
1117         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1118                 goto out;
1119
1120         start = vxlan_rco_start(unparsed->vx_vni);
1121         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1122
1123         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1124                 return false;
1125
1126         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1127                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1128 out:
1129         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1130         unparsed->vx_vni &= VXLAN_VNI_MASK;
1131         return true;
1132 }
1133
1134 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1135                                 struct sk_buff *skb, u32 vxflags,
1136                                 struct vxlan_metadata *md)
1137 {
1138         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1139         struct metadata_dst *tun_dst;
1140
1141         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1142                 goto out;
1143
1144         md->gbp = ntohs(gbp->policy_id);
1145
1146         tun_dst = (struct metadata_dst *)skb_dst(skb);
1147         if (tun_dst) {
1148                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1149                 tun_dst->u.tun_info.options_len = sizeof(*md);
1150         }
1151         if (gbp->dont_learn)
1152                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1153
1154         if (gbp->policy_applied)
1155                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1156
1157         /* In flow-based mode, GBP is carried in dst_metadata */
1158         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1159                 skb->mark = md->gbp;
1160 out:
1161         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1162 }
1163
1164 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1165                                 __be16 *protocol,
1166                                 struct sk_buff *skb, u32 vxflags)
1167 {
1168         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1169
1170         /* Need to have Next Protocol set for interfaces in GPE mode. */
1171         if (!gpe->np_applied)
1172                 return false;
1173         /* "The initial version is 0. If a receiver does not support the
1174          * version indicated it MUST drop the packet.
1175          */
1176         if (gpe->version != 0)
1177                 return false;
1178         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1179          * processing MUST occur." However, we don't implement OAM
1180          * processing, thus drop the packet.
1181          */
1182         if (gpe->oam_flag)
1183                 return false;
1184
1185         switch (gpe->next_protocol) {
1186         case VXLAN_GPE_NP_IPV4:
1187                 *protocol = htons(ETH_P_IP);
1188                 break;
1189         case VXLAN_GPE_NP_IPV6:
1190                 *protocol = htons(ETH_P_IPV6);
1191                 break;
1192         case VXLAN_GPE_NP_ETHERNET:
1193                 *protocol = htons(ETH_P_TEB);
1194                 break;
1195         default:
1196                 return false;
1197         }
1198
1199         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1200         return true;
1201 }
1202
1203 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1204                           struct vxlan_sock *vs,
1205                           struct sk_buff *skb)
1206 {
1207         union vxlan_addr saddr;
1208
1209         skb_reset_mac_header(skb);
1210         skb->protocol = eth_type_trans(skb, vxlan->dev);
1211         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1212
1213         /* Ignore packet loops (and multicast echo) */
1214         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1215                 return false;
1216
1217         /* Get address from the outer IP header */
1218         if (vxlan_get_sk_family(vs) == AF_INET) {
1219                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1220                 saddr.sa.sa_family = AF_INET;
1221 #if IS_ENABLED(CONFIG_IPV6)
1222         } else {
1223                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1224                 saddr.sa.sa_family = AF_INET6;
1225 #endif
1226         }
1227
1228         if ((vxlan->flags & VXLAN_F_LEARN) &&
1229             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1230                 return false;
1231
1232         return true;
1233 }
1234
1235 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1236                                   struct sk_buff *skb)
1237 {
1238         int err = 0;
1239
1240         if (vxlan_get_sk_family(vs) == AF_INET)
1241                 err = IP_ECN_decapsulate(oiph, skb);
1242 #if IS_ENABLED(CONFIG_IPV6)
1243         else
1244                 err = IP6_ECN_decapsulate(oiph, skb);
1245 #endif
1246
1247         if (unlikely(err) && log_ecn_error) {
1248                 if (vxlan_get_sk_family(vs) == AF_INET)
1249                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1250                                              &((struct iphdr *)oiph)->saddr,
1251                                              ((struct iphdr *)oiph)->tos);
1252                 else
1253                         net_info_ratelimited("non-ECT from %pI6\n",
1254                                              &((struct ipv6hdr *)oiph)->saddr);
1255         }
1256         return err <= 1;
1257 }
1258
1259 /* Callback from net/ipv4/udp.c to receive packets */
1260 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1261 {
1262         struct pcpu_sw_netstats *stats;
1263         struct vxlan_dev *vxlan;
1264         struct vxlan_sock *vs;
1265         struct vxlanhdr unparsed;
1266         struct vxlan_metadata _md;
1267         struct vxlan_metadata *md = &_md;
1268         __be16 protocol = htons(ETH_P_TEB);
1269         bool raw_proto = false;
1270         void *oiph;
1271
1272         /* Need UDP and VXLAN header to be present */
1273         if (!pskb_may_pull(skb, VXLAN_HLEN))
1274                 goto drop;
1275
1276         unparsed = *vxlan_hdr(skb);
1277         /* VNI flag always required to be set */
1278         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1279                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1280                            ntohl(vxlan_hdr(skb)->vx_flags),
1281                            ntohl(vxlan_hdr(skb)->vx_vni));
1282                 /* Return non vxlan pkt */
1283                 goto drop;
1284         }
1285         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1286         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1287
1288         vs = rcu_dereference_sk_user_data(sk);
1289         if (!vs)
1290                 goto drop;
1291
1292         vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1293         if (!vxlan)
1294                 goto drop;
1295
1296         /* For backwards compatibility, only allow reserved fields to be
1297          * used by VXLAN extensions if explicitly requested.
1298          */
1299         if (vs->flags & VXLAN_F_GPE) {
1300                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1301                         goto drop;
1302                 raw_proto = true;
1303         }
1304
1305         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1306                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1307                         goto drop;
1308
1309         if (vxlan_collect_metadata(vs)) {
1310                 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1311                 struct metadata_dst *tun_dst;
1312
1313                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1314                                          key32_to_tunnel_id(vni), sizeof(*md));
1315
1316                 if (!tun_dst)
1317                         goto drop;
1318
1319                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1320
1321                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1322         } else {
1323                 memset(md, 0, sizeof(*md));
1324         }
1325
1326         if (vs->flags & VXLAN_F_REMCSUM_RX)
1327                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1328                         goto drop;
1329         if (vs->flags & VXLAN_F_GBP)
1330                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1331         /* Note that GBP and GPE can never be active together. This is
1332          * ensured in vxlan_dev_configure.
1333          */
1334
1335         if (unparsed.vx_flags || unparsed.vx_vni) {
1336                 /* If there are any unprocessed flags remaining treat
1337                  * this as a malformed packet. This behavior diverges from
1338                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1339                  * in reserved fields are to be ignored. The approach here
1340                  * maintains compatibility with previous stack code, and also
1341                  * is more robust and provides a little more security in
1342                  * adding extensions to VXLAN.
1343                  */
1344                 goto drop;
1345         }
1346
1347         if (!raw_proto) {
1348                 if (!vxlan_set_mac(vxlan, vs, skb))
1349                         goto drop;
1350         } else {
1351                 skb_reset_mac_header(skb);
1352                 skb->dev = vxlan->dev;
1353                 skb->pkt_type = PACKET_HOST;
1354         }
1355
1356         oiph = skb_network_header(skb);
1357         skb_reset_network_header(skb);
1358
1359         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1360                 ++vxlan->dev->stats.rx_frame_errors;
1361                 ++vxlan->dev->stats.rx_errors;
1362                 goto drop;
1363         }
1364
1365         stats = this_cpu_ptr(vxlan->dev->tstats);
1366         u64_stats_update_begin(&stats->syncp);
1367         stats->rx_packets++;
1368         stats->rx_bytes += skb->len;
1369         u64_stats_update_end(&stats->syncp);
1370
1371         gro_cells_receive(&vxlan->gro_cells, skb);
1372         return 0;
1373
1374 drop:
1375         /* Consume bad packet */
1376         kfree_skb(skb);
1377         return 0;
1378 }
1379
1380 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1381 {
1382         struct vxlan_dev *vxlan = netdev_priv(dev);
1383         struct arphdr *parp;
1384         u8 *arpptr, *sha;
1385         __be32 sip, tip;
1386         struct neighbour *n;
1387
1388         if (dev->flags & IFF_NOARP)
1389                 goto out;
1390
1391         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1392                 dev->stats.tx_dropped++;
1393                 goto out;
1394         }
1395         parp = arp_hdr(skb);
1396
1397         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1398              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1399             parp->ar_pro != htons(ETH_P_IP) ||
1400             parp->ar_op != htons(ARPOP_REQUEST) ||
1401             parp->ar_hln != dev->addr_len ||
1402             parp->ar_pln != 4)
1403                 goto out;
1404         arpptr = (u8 *)parp + sizeof(struct arphdr);
1405         sha = arpptr;
1406         arpptr += dev->addr_len;        /* sha */
1407         memcpy(&sip, arpptr, sizeof(sip));
1408         arpptr += sizeof(sip);
1409         arpptr += dev->addr_len;        /* tha */
1410         memcpy(&tip, arpptr, sizeof(tip));
1411
1412         if (ipv4_is_loopback(tip) ||
1413             ipv4_is_multicast(tip))
1414                 goto out;
1415
1416         n = neigh_lookup(&arp_tbl, &tip, dev);
1417
1418         if (n) {
1419                 struct vxlan_fdb *f;
1420                 struct sk_buff  *reply;
1421
1422                 if (!(n->nud_state & NUD_CONNECTED)) {
1423                         neigh_release(n);
1424                         goto out;
1425                 }
1426
1427                 f = vxlan_find_mac(vxlan, n->ha);
1428                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1429                         /* bridge-local neighbor */
1430                         neigh_release(n);
1431                         goto out;
1432                 }
1433
1434                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1435                                 n->ha, sha);
1436
1437                 neigh_release(n);
1438
1439                 if (reply == NULL)
1440                         goto out;
1441
1442                 skb_reset_mac_header(reply);
1443                 __skb_pull(reply, skb_network_offset(reply));
1444                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1445                 reply->pkt_type = PACKET_HOST;
1446
1447                 if (netif_rx_ni(reply) == NET_RX_DROP)
1448                         dev->stats.rx_dropped++;
1449         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1450                 union vxlan_addr ipa = {
1451                         .sin.sin_addr.s_addr = tip,
1452                         .sin.sin_family = AF_INET,
1453                 };
1454
1455                 vxlan_ip_miss(dev, &ipa);
1456         }
1457 out:
1458         consume_skb(skb);
1459         return NETDEV_TX_OK;
1460 }
1461
1462 #if IS_ENABLED(CONFIG_IPV6)
1463 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1464         struct neighbour *n, bool isrouter)
1465 {
1466         struct net_device *dev = request->dev;
1467         struct sk_buff *reply;
1468         struct nd_msg *ns, *na;
1469         struct ipv6hdr *pip6;
1470         u8 *daddr;
1471         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1472         int ns_olen;
1473         int i, len;
1474
1475         if (dev == NULL)
1476                 return NULL;
1477
1478         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1479                 sizeof(*na) + na_olen + dev->needed_tailroom;
1480         reply = alloc_skb(len, GFP_ATOMIC);
1481         if (reply == NULL)
1482                 return NULL;
1483
1484         reply->protocol = htons(ETH_P_IPV6);
1485         reply->dev = dev;
1486         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1487         skb_push(reply, sizeof(struct ethhdr));
1488         skb_reset_mac_header(reply);
1489
1490         ns = (struct nd_msg *)skb_transport_header(request);
1491
1492         daddr = eth_hdr(request)->h_source;
1493         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1494         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1495                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1496                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1497                         break;
1498                 }
1499         }
1500
1501         /* Ethernet header */
1502         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1503         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1504         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1505         reply->protocol = htons(ETH_P_IPV6);
1506
1507         skb_pull(reply, sizeof(struct ethhdr));
1508         skb_reset_network_header(reply);
1509         skb_put(reply, sizeof(struct ipv6hdr));
1510
1511         /* IPv6 header */
1512
1513         pip6 = ipv6_hdr(reply);
1514         memset(pip6, 0, sizeof(struct ipv6hdr));
1515         pip6->version = 6;
1516         pip6->priority = ipv6_hdr(request)->priority;
1517         pip6->nexthdr = IPPROTO_ICMPV6;
1518         pip6->hop_limit = 255;
1519         pip6->daddr = ipv6_hdr(request)->saddr;
1520         pip6->saddr = *(struct in6_addr *)n->primary_key;
1521
1522         skb_pull(reply, sizeof(struct ipv6hdr));
1523         skb_reset_transport_header(reply);
1524
1525         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1526
1527         /* Neighbor Advertisement */
1528         memset(na, 0, sizeof(*na)+na_olen);
1529         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1530         na->icmph.icmp6_router = isrouter;
1531         na->icmph.icmp6_override = 1;
1532         na->icmph.icmp6_solicited = 1;
1533         na->target = ns->target;
1534         ether_addr_copy(&na->opt[2], n->ha);
1535         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1536         na->opt[1] = na_olen >> 3;
1537
1538         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1539                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1540                 csum_partial(na, sizeof(*na)+na_olen, 0));
1541
1542         pip6->payload_len = htons(sizeof(*na)+na_olen);
1543
1544         skb_push(reply, sizeof(struct ipv6hdr));
1545
1546         reply->ip_summed = CHECKSUM_UNNECESSARY;
1547
1548         return reply;
1549 }
1550
1551 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1552 {
1553         struct vxlan_dev *vxlan = netdev_priv(dev);
1554         struct nd_msg *msg;
1555         const struct ipv6hdr *iphdr;
1556         const struct in6_addr *saddr, *daddr;
1557         struct neighbour *n;
1558         struct inet6_dev *in6_dev;
1559
1560         in6_dev = __in6_dev_get(dev);
1561         if (!in6_dev)
1562                 goto out;
1563
1564         iphdr = ipv6_hdr(skb);
1565         saddr = &iphdr->saddr;
1566         daddr = &iphdr->daddr;
1567
1568         msg = (struct nd_msg *)skb_transport_header(skb);
1569         if (msg->icmph.icmp6_code != 0 ||
1570             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1571                 goto out;
1572
1573         if (ipv6_addr_loopback(daddr) ||
1574             ipv6_addr_is_multicast(&msg->target))
1575                 goto out;
1576
1577         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1578
1579         if (n) {
1580                 struct vxlan_fdb *f;
1581                 struct sk_buff *reply;
1582
1583                 if (!(n->nud_state & NUD_CONNECTED)) {
1584                         neigh_release(n);
1585                         goto out;
1586                 }
1587
1588                 f = vxlan_find_mac(vxlan, n->ha);
1589                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1590                         /* bridge-local neighbor */
1591                         neigh_release(n);
1592                         goto out;
1593                 }
1594
1595                 reply = vxlan_na_create(skb, n,
1596                                         !!(f ? f->flags & NTF_ROUTER : 0));
1597
1598                 neigh_release(n);
1599
1600                 if (reply == NULL)
1601                         goto out;
1602
1603                 if (netif_rx_ni(reply) == NET_RX_DROP)
1604                         dev->stats.rx_dropped++;
1605
1606         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1607                 union vxlan_addr ipa = {
1608                         .sin6.sin6_addr = msg->target,
1609                         .sin6.sin6_family = AF_INET6,
1610                 };
1611
1612                 vxlan_ip_miss(dev, &ipa);
1613         }
1614
1615 out:
1616         consume_skb(skb);
1617         return NETDEV_TX_OK;
1618 }
1619 #endif
1620
1621 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1622 {
1623         struct vxlan_dev *vxlan = netdev_priv(dev);
1624         struct neighbour *n;
1625
1626         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1627                 return false;
1628
1629         n = NULL;
1630         switch (ntohs(eth_hdr(skb)->h_proto)) {
1631         case ETH_P_IP:
1632         {
1633                 struct iphdr *pip;
1634
1635                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1636                         return false;
1637                 pip = ip_hdr(skb);
1638                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1639                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1640                         union vxlan_addr ipa = {
1641                                 .sin.sin_addr.s_addr = pip->daddr,
1642                                 .sin.sin_family = AF_INET,
1643                         };
1644
1645                         vxlan_ip_miss(dev, &ipa);
1646                         return false;
1647                 }
1648
1649                 break;
1650         }
1651 #if IS_ENABLED(CONFIG_IPV6)
1652         case ETH_P_IPV6:
1653         {
1654                 struct ipv6hdr *pip6;
1655
1656                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1657                         return false;
1658                 pip6 = ipv6_hdr(skb);
1659                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1660                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1661                         union vxlan_addr ipa = {
1662                                 .sin6.sin6_addr = pip6->daddr,
1663                                 .sin6.sin6_family = AF_INET6,
1664                         };
1665
1666                         vxlan_ip_miss(dev, &ipa);
1667                         return false;
1668                 }
1669
1670                 break;
1671         }
1672 #endif
1673         default:
1674                 return false;
1675         }
1676
1677         if (n) {
1678                 bool diff;
1679
1680                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1681                 if (diff) {
1682                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1683                                 dev->addr_len);
1684                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1685                 }
1686                 neigh_release(n);
1687                 return diff;
1688         }
1689
1690         return false;
1691 }
1692
1693 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1694                                 struct vxlan_metadata *md)
1695 {
1696         struct vxlanhdr_gbp *gbp;
1697
1698         if (!md->gbp)
1699                 return;
1700
1701         gbp = (struct vxlanhdr_gbp *)vxh;
1702         vxh->vx_flags |= VXLAN_HF_GBP;
1703
1704         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1705                 gbp->dont_learn = 1;
1706
1707         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1708                 gbp->policy_applied = 1;
1709
1710         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1711 }
1712
1713 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1714                                __be16 protocol)
1715 {
1716         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1717
1718         gpe->np_applied = 1;
1719
1720         switch (protocol) {
1721         case htons(ETH_P_IP):
1722                 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1723                 return 0;
1724         case htons(ETH_P_IPV6):
1725                 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1726                 return 0;
1727         case htons(ETH_P_TEB):
1728                 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1729                 return 0;
1730         }
1731         return -EPFNOSUPPORT;
1732 }
1733
1734 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1735                            int iphdr_len, __be32 vni,
1736                            struct vxlan_metadata *md, u32 vxflags,
1737                            bool udp_sum)
1738 {
1739         struct vxlanhdr *vxh;
1740         int min_headroom;
1741         int err;
1742         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1743         __be16 inner_protocol = htons(ETH_P_TEB);
1744
1745         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1746             skb->ip_summed == CHECKSUM_PARTIAL) {
1747                 int csum_start = skb_checksum_start_offset(skb);
1748
1749                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1750                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1751                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1752                      skb->csum_offset == offsetof(struct tcphdr, check)))
1753                         type |= SKB_GSO_TUNNEL_REMCSUM;
1754         }
1755
1756         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1757                         + VXLAN_HLEN + iphdr_len
1758                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1759
1760         /* Need space for new headers (invalidates iph ptr) */
1761         err = skb_cow_head(skb, min_headroom);
1762         if (unlikely(err))
1763                 goto out_free;
1764
1765         skb = vlan_hwaccel_push_inside(skb);
1766         if (WARN_ON(!skb))
1767                 return -ENOMEM;
1768
1769         err = iptunnel_handle_offloads(skb, type);
1770         if (err)
1771                 goto out_free;
1772
1773         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1774         vxh->vx_flags = VXLAN_HF_VNI;
1775         vxh->vx_vni = vxlan_vni_field(vni);
1776
1777         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1778                 unsigned int start;
1779
1780                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1781                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1782                 vxh->vx_flags |= VXLAN_HF_RCO;
1783
1784                 if (!skb_is_gso(skb)) {
1785                         skb->ip_summed = CHECKSUM_NONE;
1786                         skb->encapsulation = 0;
1787                 }
1788         }
1789
1790         if (vxflags & VXLAN_F_GBP)
1791                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1792         if (vxflags & VXLAN_F_GPE) {
1793                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1794                 if (err < 0)
1795                         goto out_free;
1796                 inner_protocol = skb->protocol;
1797         }
1798
1799         skb_set_inner_protocol(skb, inner_protocol);
1800         return 0;
1801
1802 out_free:
1803         kfree_skb(skb);
1804         return err;
1805 }
1806
1807 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1808                                       struct sk_buff *skb, int oif, u8 tos,
1809                                       __be32 daddr, __be32 *saddr,
1810                                       struct dst_cache *dst_cache,
1811                                       const struct ip_tunnel_info *info)
1812 {
1813         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1814         struct rtable *rt = NULL;
1815         struct flowi4 fl4;
1816
1817         if (tos && !info)
1818                 use_cache = false;
1819         if (use_cache) {
1820                 rt = dst_cache_get_ip4(dst_cache, saddr);
1821                 if (rt)
1822                         return rt;
1823         }
1824
1825         memset(&fl4, 0, sizeof(fl4));
1826         fl4.flowi4_oif = oif;
1827         fl4.flowi4_tos = RT_TOS(tos);
1828         fl4.flowi4_mark = skb->mark;
1829         fl4.flowi4_proto = IPPROTO_UDP;
1830         fl4.daddr = daddr;
1831         fl4.saddr = *saddr;
1832
1833         rt = ip_route_output_key(vxlan->net, &fl4);
1834         if (!IS_ERR(rt)) {
1835                 *saddr = fl4.saddr;
1836                 if (use_cache)
1837                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1838         }
1839         return rt;
1840 }
1841
1842 #if IS_ENABLED(CONFIG_IPV6)
1843 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1844                                           struct sk_buff *skb, int oif, u8 tos,
1845                                           __be32 label,
1846                                           const struct in6_addr *daddr,
1847                                           struct in6_addr *saddr,
1848                                           struct dst_cache *dst_cache,
1849                                           const struct ip_tunnel_info *info)
1850 {
1851         struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
1852         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1853         struct dst_entry *ndst;
1854         struct flowi6 fl6;
1855         int err;
1856
1857         if (!sock6)
1858                 return ERR_PTR(-EIO);
1859
1860         if (tos && !info)
1861                 use_cache = false;
1862         if (use_cache) {
1863                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1864                 if (ndst)
1865                         return ndst;
1866         }
1867
1868         memset(&fl6, 0, sizeof(fl6));
1869         fl6.flowi6_oif = oif;
1870         fl6.daddr = *daddr;
1871         fl6.saddr = *saddr;
1872         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1873         fl6.flowi6_mark = skb->mark;
1874         fl6.flowi6_proto = IPPROTO_UDP;
1875
1876         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1877                                          sock6->sock->sk,
1878                                          &ndst, &fl6);
1879         if (err < 0)
1880                 return ERR_PTR(err);
1881
1882         *saddr = fl6.saddr;
1883         if (use_cache)
1884                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1885         return ndst;
1886 }
1887 #endif
1888
1889 /* Bypass encapsulation if the destination is local */
1890 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1891                                struct vxlan_dev *dst_vxlan)
1892 {
1893         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1894         union vxlan_addr loopback;
1895         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1896         struct net_device *dev = skb->dev;
1897         int len = skb->len;
1898
1899         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1900         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1901         skb->pkt_type = PACKET_HOST;
1902         skb->encapsulation = 0;
1903         skb->dev = dst_vxlan->dev;
1904         __skb_pull(skb, skb_network_offset(skb));
1905
1906         if (remote_ip->sa.sa_family == AF_INET) {
1907                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1908                 loopback.sa.sa_family =  AF_INET;
1909 #if IS_ENABLED(CONFIG_IPV6)
1910         } else {
1911                 loopback.sin6.sin6_addr = in6addr_loopback;
1912                 loopback.sa.sa_family =  AF_INET6;
1913 #endif
1914         }
1915
1916         if (dst_vxlan->flags & VXLAN_F_LEARN)
1917                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1918
1919         u64_stats_update_begin(&tx_stats->syncp);
1920         tx_stats->tx_packets++;
1921         tx_stats->tx_bytes += len;
1922         u64_stats_update_end(&tx_stats->syncp);
1923
1924         if (netif_rx(skb) == NET_RX_SUCCESS) {
1925                 u64_stats_update_begin(&rx_stats->syncp);
1926                 rx_stats->rx_packets++;
1927                 rx_stats->rx_bytes += len;
1928                 u64_stats_update_end(&rx_stats->syncp);
1929         } else {
1930                 dev->stats.rx_dropped++;
1931         }
1932 }
1933
1934 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1935                            struct vxlan_rdst *rdst, bool did_rsc)
1936 {
1937         struct dst_cache *dst_cache;
1938         struct ip_tunnel_info *info;
1939         struct vxlan_dev *vxlan = netdev_priv(dev);
1940         struct sock *sk;
1941         struct rtable *rt = NULL;
1942         const struct iphdr *old_iph;
1943         union vxlan_addr *dst;
1944         union vxlan_addr remote_ip, local_ip;
1945         struct vxlan_metadata _md;
1946         struct vxlan_metadata *md = &_md;
1947         __be16 src_port = 0, dst_port;
1948         __be32 vni, label;
1949         __be16 df = 0;
1950         __u8 tos, ttl;
1951         int err;
1952         u32 flags = vxlan->flags;
1953         bool udp_sum = false;
1954         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1955
1956         info = skb_tunnel_info(skb);
1957
1958         rcu_read_lock();
1959         if (rdst) {
1960                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1961                 vni = rdst->remote_vni;
1962                 dst = &rdst->remote_ip;
1963                 local_ip = vxlan->cfg.saddr;
1964                 dst_cache = &rdst->dst_cache;
1965         } else {
1966                 if (!info) {
1967                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1968                                   dev->name);
1969                         goto drop;
1970                 }
1971                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1972                 vni = tunnel_id_to_key32(info->key.tun_id);
1973                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1974                 if (remote_ip.sa.sa_family == AF_INET) {
1975                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1976                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
1977                 } else {
1978                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1979                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
1980                 }
1981                 dst = &remote_ip;
1982                 dst_cache = &info->dst_cache;
1983         }
1984
1985         if (vxlan_addr_any(dst)) {
1986                 if (did_rsc) {
1987                         /* short-circuited back to local bridge */
1988                         vxlan_encap_bypass(skb, vxlan, vxlan);
1989                         goto out_unlock;
1990                 }
1991                 goto drop;
1992         }
1993
1994         old_iph = ip_hdr(skb);
1995
1996         ttl = vxlan->cfg.ttl;
1997         if (!ttl && vxlan_addr_multicast(dst))
1998                 ttl = 1;
1999
2000         tos = vxlan->cfg.tos;
2001         if (tos == 1)
2002                 tos = ip_tunnel_get_dsfield(old_iph, skb);
2003
2004         label = vxlan->cfg.label;
2005         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2006                                      vxlan->cfg.port_max, true);
2007
2008         if (info) {
2009                 ttl = info->key.ttl;
2010                 tos = info->key.tos;
2011                 label = info->key.label;
2012                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2013
2014                 if (info->options_len)
2015                         md = ip_tunnel_info_opts(info);
2016         } else {
2017                 md->gbp = skb->mark;
2018         }
2019
2020         if (dst->sa.sa_family == AF_INET) {
2021                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2022
2023                 if (!sock4)
2024                         goto drop;
2025                 sk = sock4->sock->sk;
2026
2027                 rt = vxlan_get_route(vxlan, skb,
2028                                      rdst ? rdst->remote_ifindex : 0, tos,
2029                                      dst->sin.sin_addr.s_addr,
2030                                      &local_ip.sin.sin_addr.s_addr,
2031                                      dst_cache, info);
2032                 if (IS_ERR(rt)) {
2033                         netdev_dbg(dev, "no route to %pI4\n",
2034                                    &dst->sin.sin_addr.s_addr);
2035                         dev->stats.tx_carrier_errors++;
2036                         goto tx_error;
2037                 }
2038
2039                 if (rt->dst.dev == dev) {
2040                         netdev_dbg(dev, "circular route to %pI4\n",
2041                                    &dst->sin.sin_addr.s_addr);
2042                         dev->stats.collisions++;
2043                         goto rt_tx_error;
2044                 }
2045
2046                 /* Bypass encapsulation if the destination is local */
2047                 if (!info && rt->rt_flags & RTCF_LOCAL &&
2048                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2049                         struct vxlan_dev *dst_vxlan;
2050
2051                         ip_rt_put(rt);
2052                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2053                                                    dst->sa.sa_family, dst_port,
2054                                                    vxlan->flags);
2055                         if (!dst_vxlan)
2056                                 goto tx_error;
2057                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2058                         goto out_unlock;
2059                 }
2060
2061                 if (!info)
2062                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2063                 else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2064                         df = htons(IP_DF);
2065
2066                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2067                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2068                 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2069                                       vni, md, flags, udp_sum);
2070                 if (err < 0)
2071                         goto xmit_tx_error;
2072
2073                 udp_tunnel_xmit_skb(rt, sk, skb, local_ip.sin.sin_addr.s_addr,
2074                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2075                                     src_port, dst_port, xnet, !udp_sum);
2076 #if IS_ENABLED(CONFIG_IPV6)
2077         } else {
2078                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2079                 struct dst_entry *ndst;
2080                 u32 rt6i_flags;
2081
2082                 if (!sock6)
2083                         goto drop;
2084                 sk = sock6->sock->sk;
2085
2086                 ndst = vxlan6_get_route(vxlan, skb,
2087                                         rdst ? rdst->remote_ifindex : 0, tos,
2088                                         label, &dst->sin6.sin6_addr,
2089                                         &local_ip.sin6.sin6_addr,
2090                                         dst_cache, info);
2091                 if (IS_ERR(ndst)) {
2092                         netdev_dbg(dev, "no route to %pI6\n",
2093                                    &dst->sin6.sin6_addr);
2094                         dev->stats.tx_carrier_errors++;
2095                         goto tx_error;
2096                 }
2097
2098                 if (ndst->dev == dev) {
2099                         netdev_dbg(dev, "circular route to %pI6\n",
2100                                    &dst->sin6.sin6_addr);
2101                         dst_release(ndst);
2102                         dev->stats.collisions++;
2103                         goto tx_error;
2104                 }
2105
2106                 /* Bypass encapsulation if the destination is local */
2107                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2108                 if (!info && rt6i_flags & RTF_LOCAL &&
2109                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2110                         struct vxlan_dev *dst_vxlan;
2111
2112                         dst_release(ndst);
2113                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2114                                                    dst->sa.sa_family, dst_port,
2115                                                    vxlan->flags);
2116                         if (!dst_vxlan)
2117                                 goto tx_error;
2118                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2119                         goto out_unlock;
2120                 }
2121
2122                 if (!info)
2123                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2124
2125                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2126                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2127                 skb_scrub_packet(skb, xnet);
2128                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2129                                       vni, md, flags, udp_sum);
2130                 if (err < 0) {
2131                         dst_release(ndst);
2132                         dev->stats.tx_errors++;
2133                         goto out_unlock;
2134                 }
2135                 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2136                                      &local_ip.sin6.sin6_addr,
2137                                      &dst->sin6.sin6_addr, tos, ttl,
2138                                      label, src_port, dst_port, !udp_sum);
2139 #endif
2140         }
2141 out_unlock:
2142         rcu_read_unlock();
2143         return;
2144
2145 drop:
2146         dev->stats.tx_dropped++;
2147         goto tx_free;
2148
2149 xmit_tx_error:
2150         /* skb is already freed. */
2151         skb = NULL;
2152 rt_tx_error:
2153         ip_rt_put(rt);
2154 tx_error:
2155         dev->stats.tx_errors++;
2156 tx_free:
2157         dev_kfree_skb(skb);
2158         rcu_read_unlock();
2159 }
2160
2161 /* Transmit local packets over Vxlan
2162  *
2163  * Outer IP header inherits ECN and DF from inner header.
2164  * Outer UDP destination is the VXLAN assigned port.
2165  *           source port is based on hash of flow
2166  */
2167 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2168 {
2169         struct vxlan_dev *vxlan = netdev_priv(dev);
2170         const struct ip_tunnel_info *info;
2171         struct ethhdr *eth;
2172         bool did_rsc = false;
2173         struct vxlan_rdst *rdst, *fdst = NULL;
2174         struct vxlan_fdb *f;
2175
2176         info = skb_tunnel_info(skb);
2177
2178         skb_reset_mac_header(skb);
2179
2180         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2181                 if (info && info->mode & IP_TUNNEL_INFO_TX)
2182                         vxlan_xmit_one(skb, dev, NULL, false);
2183                 else
2184                         kfree_skb(skb);
2185                 return NETDEV_TX_OK;
2186         }
2187
2188         if (vxlan->flags & VXLAN_F_PROXY) {
2189                 eth = eth_hdr(skb);
2190                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2191                         return arp_reduce(dev, skb);
2192 #if IS_ENABLED(CONFIG_IPV6)
2193                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2194                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2195                                        + sizeof(struct nd_msg)) &&
2196                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2197                                 struct nd_msg *msg;
2198
2199                                 msg = (struct nd_msg *)skb_transport_header(skb);
2200                                 if (msg->icmph.icmp6_code == 0 &&
2201                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2202                                         return neigh_reduce(dev, skb);
2203                 }
2204 #endif
2205         }
2206
2207         eth = eth_hdr(skb);
2208         f = vxlan_find_mac(vxlan, eth->h_dest);
2209         did_rsc = false;
2210
2211         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2212             (ntohs(eth->h_proto) == ETH_P_IP ||
2213              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2214                 did_rsc = route_shortcircuit(dev, skb);
2215                 if (did_rsc)
2216                         f = vxlan_find_mac(vxlan, eth->h_dest);
2217         }
2218
2219         if (f == NULL) {
2220                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2221                 if (f == NULL) {
2222                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2223                             !is_multicast_ether_addr(eth->h_dest))
2224                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2225
2226                         dev->stats.tx_dropped++;
2227                         kfree_skb(skb);
2228                         return NETDEV_TX_OK;
2229                 }
2230         }
2231
2232         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2233                 struct sk_buff *skb1;
2234
2235                 if (!fdst) {
2236                         fdst = rdst;
2237                         continue;
2238                 }
2239                 skb1 = skb_clone(skb, GFP_ATOMIC);
2240                 if (skb1)
2241                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2242         }
2243
2244         if (fdst)
2245                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2246         else
2247                 kfree_skb(skb);
2248         return NETDEV_TX_OK;
2249 }
2250
2251 /* Walk the forwarding table and purge stale entries */
2252 static void vxlan_cleanup(unsigned long arg)
2253 {
2254         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2255         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2256         unsigned int h;
2257
2258         if (!netif_running(vxlan->dev))
2259                 return;
2260
2261         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2262                 struct hlist_node *p, *n;
2263
2264                 spin_lock_bh(&vxlan->hash_lock);
2265                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2266                         struct vxlan_fdb *f
2267                                 = container_of(p, struct vxlan_fdb, hlist);
2268                         unsigned long timeout;
2269
2270                         if (f->state & NUD_PERMANENT)
2271                                 continue;
2272
2273                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2274                         if (time_before_eq(timeout, jiffies)) {
2275                                 netdev_dbg(vxlan->dev,
2276                                            "garbage collect %pM\n",
2277                                            f->eth_addr);
2278                                 f->state = NUD_STALE;
2279                                 vxlan_fdb_destroy(vxlan, f);
2280                         } else if (time_before(timeout, next_timer))
2281                                 next_timer = timeout;
2282                 }
2283                 spin_unlock_bh(&vxlan->hash_lock);
2284         }
2285
2286         mod_timer(&vxlan->age_timer, next_timer);
2287 }
2288
2289 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2290 {
2291         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2292         __be32 vni = vxlan->default_dst.remote_vni;
2293
2294         spin_lock(&vn->sock_lock);
2295         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2296         spin_unlock(&vn->sock_lock);
2297 }
2298
2299 /* Setup stats when device is created */
2300 static int vxlan_init(struct net_device *dev)
2301 {
2302         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2303         if (!dev->tstats)
2304                 return -ENOMEM;
2305
2306         return 0;
2307 }
2308
2309 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2310 {
2311         struct vxlan_fdb *f;
2312
2313         spin_lock_bh(&vxlan->hash_lock);
2314         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2315         if (f)
2316                 vxlan_fdb_destroy(vxlan, f);
2317         spin_unlock_bh(&vxlan->hash_lock);
2318 }
2319
2320 static void vxlan_uninit(struct net_device *dev)
2321 {
2322         struct vxlan_dev *vxlan = netdev_priv(dev);
2323
2324         vxlan_fdb_delete_default(vxlan);
2325
2326         free_percpu(dev->tstats);
2327 }
2328
2329 /* Start ageing timer and join group when device is brought up */
2330 static int vxlan_open(struct net_device *dev)
2331 {
2332         struct vxlan_dev *vxlan = netdev_priv(dev);
2333         int ret;
2334
2335         ret = vxlan_sock_add(vxlan);
2336         if (ret < 0)
2337                 return ret;
2338
2339         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2340                 ret = vxlan_igmp_join(vxlan);
2341                 if (ret == -EADDRINUSE)
2342                         ret = 0;
2343                 if (ret) {
2344                         vxlan_sock_release(vxlan);
2345                         return ret;
2346                 }
2347         }
2348
2349         if (vxlan->cfg.age_interval)
2350                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2351
2352         return ret;
2353 }
2354
2355 /* Purge the forwarding table */
2356 static void vxlan_flush(struct vxlan_dev *vxlan)
2357 {
2358         unsigned int h;
2359
2360         spin_lock_bh(&vxlan->hash_lock);
2361         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2362                 struct hlist_node *p, *n;
2363                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2364                         struct vxlan_fdb *f
2365                                 = container_of(p, struct vxlan_fdb, hlist);
2366                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2367                         if (!is_zero_ether_addr(f->eth_addr))
2368                                 vxlan_fdb_destroy(vxlan, f);
2369                 }
2370         }
2371         spin_unlock_bh(&vxlan->hash_lock);
2372 }
2373
2374 /* Cleanup timer and forwarding table on shutdown */
2375 static int vxlan_stop(struct net_device *dev)
2376 {
2377         struct vxlan_dev *vxlan = netdev_priv(dev);
2378         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2379         int ret = 0;
2380
2381         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2382             !vxlan_group_used(vn, vxlan))
2383                 ret = vxlan_igmp_leave(vxlan);
2384
2385         del_timer_sync(&vxlan->age_timer);
2386
2387         vxlan_flush(vxlan);
2388         vxlan_sock_release(vxlan);
2389
2390         return ret;
2391 }
2392
2393 /* Stub, nothing needs to be done. */
2394 static void vxlan_set_multicast_list(struct net_device *dev)
2395 {
2396 }
2397
2398 static int __vxlan_change_mtu(struct net_device *dev,
2399                               struct net_device *lowerdev,
2400                               struct vxlan_rdst *dst, int new_mtu, bool strict)
2401 {
2402         int max_mtu = IP_MAX_MTU;
2403
2404         if (lowerdev)
2405                 max_mtu = lowerdev->mtu;
2406
2407         if (dst->remote_ip.sa.sa_family == AF_INET6)
2408                 max_mtu -= VXLAN6_HEADROOM;
2409         else
2410                 max_mtu -= VXLAN_HEADROOM;
2411
2412         if (new_mtu < 68)
2413                 return -EINVAL;
2414
2415         if (new_mtu > max_mtu) {
2416                 if (strict)
2417                         return -EINVAL;
2418
2419                 new_mtu = max_mtu;
2420         }
2421
2422         dev->mtu = new_mtu;
2423         return 0;
2424 }
2425
2426 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2427 {
2428         struct vxlan_dev *vxlan = netdev_priv(dev);
2429         struct vxlan_rdst *dst = &vxlan->default_dst;
2430         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2431                                                          dst->remote_ifindex);
2432         return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2433 }
2434
2435 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2436 {
2437         struct vxlan_dev *vxlan = netdev_priv(dev);
2438         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2439         __be16 sport, dport;
2440
2441         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2442                                   vxlan->cfg.port_max, true);
2443         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2444
2445         if (ip_tunnel_info_af(info) == AF_INET) {
2446                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2447                 struct rtable *rt;
2448
2449                 if (!sock4)
2450                         return -EINVAL;
2451                 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2452                                      info->key.u.ipv4.dst,
2453                                      &info->key.u.ipv4.src,
2454                                      &info->dst_cache, info);
2455                 if (IS_ERR(rt))
2456                         return PTR_ERR(rt);
2457                 ip_rt_put(rt);
2458         } else {
2459 #if IS_ENABLED(CONFIG_IPV6)
2460                 struct dst_entry *ndst;
2461
2462                 ndst = vxlan6_get_route(vxlan, skb, 0, info->key.tos,
2463                                         info->key.label, &info->key.u.ipv6.dst,
2464                                         &info->key.u.ipv6.src,
2465                                         &info->dst_cache, info);
2466                 if (IS_ERR(ndst))
2467                         return PTR_ERR(ndst);
2468                 dst_release(ndst);
2469 #else /* !CONFIG_IPV6 */
2470                 return -EPFNOSUPPORT;
2471 #endif
2472         }
2473         info->key.tp_src = sport;
2474         info->key.tp_dst = dport;
2475         return 0;
2476 }
2477
2478 static const struct net_device_ops vxlan_netdev_ether_ops = {
2479         .ndo_init               = vxlan_init,
2480         .ndo_uninit             = vxlan_uninit,
2481         .ndo_open               = vxlan_open,
2482         .ndo_stop               = vxlan_stop,
2483         .ndo_start_xmit         = vxlan_xmit,
2484         .ndo_get_stats64        = ip_tunnel_get_stats64,
2485         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2486         .ndo_change_mtu         = vxlan_change_mtu,
2487         .ndo_validate_addr      = eth_validate_addr,
2488         .ndo_set_mac_address    = eth_mac_addr,
2489         .ndo_fdb_add            = vxlan_fdb_add,
2490         .ndo_fdb_del            = vxlan_fdb_delete,
2491         .ndo_fdb_dump           = vxlan_fdb_dump,
2492         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2493 };
2494
2495 static const struct net_device_ops vxlan_netdev_raw_ops = {
2496         .ndo_init               = vxlan_init,
2497         .ndo_uninit             = vxlan_uninit,
2498         .ndo_open               = vxlan_open,
2499         .ndo_stop               = vxlan_stop,
2500         .ndo_start_xmit         = vxlan_xmit,
2501         .ndo_get_stats64        = ip_tunnel_get_stats64,
2502         .ndo_change_mtu         = vxlan_change_mtu,
2503         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2504 };
2505
2506 /* Info for udev, that this is a virtual tunnel endpoint */
2507 static struct device_type vxlan_type = {
2508         .name = "vxlan",
2509 };
2510
2511 /* Calls the ndo_udp_tunnel_add of the caller in order to
2512  * supply the listening VXLAN udp ports. Callers are expected
2513  * to implement the ndo_udp_tunnel_add.
2514  */
2515 static void vxlan_push_rx_ports(struct net_device *dev)
2516 {
2517         struct vxlan_sock *vs;
2518         struct net *net = dev_net(dev);
2519         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2520         unsigned int i;
2521
2522         spin_lock(&vn->sock_lock);
2523         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2524                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2525                         udp_tunnel_push_rx_port(dev, vs->sock,
2526                                                 (vs->flags & VXLAN_F_GPE) ?
2527                                                 UDP_TUNNEL_TYPE_VXLAN_GPE :
2528                                                 UDP_TUNNEL_TYPE_VXLAN);
2529         }
2530         spin_unlock(&vn->sock_lock);
2531 }
2532
2533 /* Initialize the device structure. */
2534 static void vxlan_setup(struct net_device *dev)
2535 {
2536         struct vxlan_dev *vxlan = netdev_priv(dev);
2537         unsigned int h;
2538
2539         eth_hw_addr_random(dev);
2540         ether_setup(dev);
2541
2542         dev->destructor = free_netdev;
2543         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2544
2545         dev->features   |= NETIF_F_LLTX;
2546         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2547         dev->features   |= NETIF_F_RXCSUM;
2548         dev->features   |= NETIF_F_GSO_SOFTWARE;
2549
2550         dev->vlan_features = dev->features;
2551         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2552         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2553         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2554         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2555         netif_keep_dst(dev);
2556         dev->priv_flags |= IFF_NO_QUEUE;
2557
2558         INIT_LIST_HEAD(&vxlan->next);
2559         spin_lock_init(&vxlan->hash_lock);
2560
2561         init_timer_deferrable(&vxlan->age_timer);
2562         vxlan->age_timer.function = vxlan_cleanup;
2563         vxlan->age_timer.data = (unsigned long) vxlan;
2564
2565         vxlan->cfg.dst_port = htons(vxlan_port);
2566
2567         vxlan->dev = dev;
2568
2569         gro_cells_init(&vxlan->gro_cells, dev);
2570
2571         for (h = 0; h < FDB_HASH_SIZE; ++h)
2572                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2573 }
2574
2575 static void vxlan_ether_setup(struct net_device *dev)
2576 {
2577         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2578         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2579         dev->netdev_ops = &vxlan_netdev_ether_ops;
2580 }
2581
2582 static void vxlan_raw_setup(struct net_device *dev)
2583 {
2584         dev->header_ops = NULL;
2585         dev->type = ARPHRD_NONE;
2586         dev->hard_header_len = 0;
2587         dev->addr_len = 0;
2588         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2589         dev->netdev_ops = &vxlan_netdev_raw_ops;
2590 }
2591
2592 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2593         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2594         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2595         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2596         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2597         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2598         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2599         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2600         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2601         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2602         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2603         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2604         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2605         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2606         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2607         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2608         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2609         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2610         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2611         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2612         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2613         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2614         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2615         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2616         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2617         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2618         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2619         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2620 };
2621
2622 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2623 {
2624         if (tb[IFLA_ADDRESS]) {
2625                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2626                         pr_debug("invalid link address (not ethernet)\n");
2627                         return -EINVAL;
2628                 }
2629
2630                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2631                         pr_debug("invalid all zero ethernet address\n");
2632                         return -EADDRNOTAVAIL;
2633                 }
2634         }
2635
2636         if (!data)
2637                 return -EINVAL;
2638
2639         if (data[IFLA_VXLAN_ID]) {
2640                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2641                 if (id >= VXLAN_N_VID)
2642                         return -ERANGE;
2643         }
2644
2645         if (data[IFLA_VXLAN_PORT_RANGE]) {
2646                 const struct ifla_vxlan_port_range *p
2647                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2648
2649                 if (ntohs(p->high) < ntohs(p->low)) {
2650                         pr_debug("port range %u .. %u not valid\n",
2651                                  ntohs(p->low), ntohs(p->high));
2652                         return -EINVAL;
2653                 }
2654         }
2655
2656         return 0;
2657 }
2658
2659 static void vxlan_get_drvinfo(struct net_device *netdev,
2660                               struct ethtool_drvinfo *drvinfo)
2661 {
2662         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2663         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2664 }
2665
2666 static const struct ethtool_ops vxlan_ethtool_ops = {
2667         .get_drvinfo    = vxlan_get_drvinfo,
2668         .get_link       = ethtool_op_get_link,
2669 };
2670
2671 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2672                                         __be16 port, u32 flags)
2673 {
2674         struct socket *sock;
2675         struct udp_port_cfg udp_conf;
2676         int err;
2677
2678         memset(&udp_conf, 0, sizeof(udp_conf));
2679
2680         if (ipv6) {
2681                 udp_conf.family = AF_INET6;
2682                 udp_conf.use_udp6_rx_checksums =
2683                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2684                 udp_conf.ipv6_v6only = 1;
2685         } else {
2686                 udp_conf.family = AF_INET;
2687         }
2688
2689         udp_conf.local_udp_port = port;
2690
2691         /* Open UDP socket */
2692         err = udp_sock_create(net, &udp_conf, &sock);
2693         if (err < 0)
2694                 return ERR_PTR(err);
2695
2696         return sock;
2697 }
2698
2699 /* Create new listen socket if needed */
2700 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2701                                               __be16 port, u32 flags)
2702 {
2703         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2704         struct vxlan_sock *vs;
2705         struct socket *sock;
2706         unsigned int h;
2707         struct udp_tunnel_sock_cfg tunnel_cfg;
2708
2709         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2710         if (!vs)
2711                 return ERR_PTR(-ENOMEM);
2712
2713         for (h = 0; h < VNI_HASH_SIZE; ++h)
2714                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2715
2716         sock = vxlan_create_sock(net, ipv6, port, flags);
2717         if (IS_ERR(sock)) {
2718                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2719                         PTR_ERR(sock));
2720                 kfree(vs);
2721                 return ERR_CAST(sock);
2722         }
2723
2724         vs->sock = sock;
2725         atomic_set(&vs->refcnt, 1);
2726         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2727
2728         spin_lock(&vn->sock_lock);
2729         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2730         udp_tunnel_notify_add_rx_port(sock,
2731                                       (vs->flags & VXLAN_F_GPE) ?
2732                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2733                                       UDP_TUNNEL_TYPE_VXLAN);
2734         spin_unlock(&vn->sock_lock);
2735
2736         /* Mark socket as an encapsulation socket. */
2737         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2738         tunnel_cfg.sk_user_data = vs;
2739         tunnel_cfg.encap_type = 1;
2740         tunnel_cfg.encap_rcv = vxlan_rcv;
2741         tunnel_cfg.encap_destroy = NULL;
2742         tunnel_cfg.gro_receive = vxlan_gro_receive;
2743         tunnel_cfg.gro_complete = vxlan_gro_complete;
2744
2745         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2746
2747         return vs;
2748 }
2749
2750 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2751 {
2752         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2753         struct vxlan_sock *vs = NULL;
2754
2755         if (!vxlan->cfg.no_share) {
2756                 spin_lock(&vn->sock_lock);
2757                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2758                                      vxlan->cfg.dst_port, vxlan->flags);
2759                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2760                         spin_unlock(&vn->sock_lock);
2761                         return -EBUSY;
2762                 }
2763                 spin_unlock(&vn->sock_lock);
2764         }
2765         if (!vs)
2766                 vs = vxlan_socket_create(vxlan->net, ipv6,
2767                                          vxlan->cfg.dst_port, vxlan->flags);
2768         if (IS_ERR(vs))
2769                 return PTR_ERR(vs);
2770 #if IS_ENABLED(CONFIG_IPV6)
2771         if (ipv6)
2772                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2773         else
2774 #endif
2775                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2776         vxlan_vs_add_dev(vs, vxlan);
2777         return 0;
2778 }
2779
2780 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2781 {
2782         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2783         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2784         int ret = 0;
2785
2786         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2787 #if IS_ENABLED(CONFIG_IPV6)
2788         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2789         if (ipv6 || metadata)
2790                 ret = __vxlan_sock_add(vxlan, true);
2791 #endif
2792         if (!ret && (!ipv6 || metadata))
2793                 ret = __vxlan_sock_add(vxlan, false);
2794         if (ret < 0)
2795                 vxlan_sock_release(vxlan);
2796         return ret;
2797 }
2798
2799 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2800                                struct vxlan_config *conf)
2801 {
2802         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2803         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2804         struct vxlan_rdst *dst = &vxlan->default_dst;
2805         unsigned short needed_headroom = ETH_HLEN;
2806         int err;
2807         bool use_ipv6 = false;
2808         __be16 default_port = vxlan->cfg.dst_port;
2809         struct net_device *lowerdev = NULL;
2810
2811         if (conf->flags & VXLAN_F_GPE) {
2812                 /* For now, allow GPE only together with COLLECT_METADATA.
2813                  * This can be relaxed later; in such case, the other side
2814                  * of the PtP link will have to be provided.
2815                  */
2816                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2817                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2818                         pr_info("unsupported combination of extensions\n");
2819                         return -EINVAL;
2820                 }
2821
2822                 vxlan_raw_setup(dev);
2823         } else {
2824                 vxlan_ether_setup(dev);
2825         }
2826
2827         vxlan->net = src_net;
2828
2829         dst->remote_vni = conf->vni;
2830
2831         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2832
2833         /* Unless IPv6 is explicitly requested, assume IPv4 */
2834         if (!dst->remote_ip.sa.sa_family)
2835                 dst->remote_ip.sa.sa_family = AF_INET;
2836
2837         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2838             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2839                 if (!IS_ENABLED(CONFIG_IPV6))
2840                         return -EPFNOSUPPORT;
2841                 use_ipv6 = true;
2842                 vxlan->flags |= VXLAN_F_IPV6;
2843         }
2844
2845         if (conf->label && !use_ipv6) {
2846                 pr_info("label only supported in use with IPv6\n");
2847                 return -EINVAL;
2848         }
2849
2850         if (conf->remote_ifindex) {
2851                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2852                 dst->remote_ifindex = conf->remote_ifindex;
2853
2854                 if (!lowerdev) {
2855                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2856                         return -ENODEV;
2857                 }
2858
2859 #if IS_ENABLED(CONFIG_IPV6)
2860                 if (use_ipv6) {
2861                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2862                         if (idev && idev->cnf.disable_ipv6) {
2863                                 pr_info("IPv6 is disabled via sysctl\n");
2864                                 return -EPERM;
2865                         }
2866                 }
2867 #endif
2868
2869                 if (!conf->mtu)
2870                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2871
2872                 needed_headroom = lowerdev->hard_header_len;
2873         } else if (vxlan_addr_multicast(&dst->remote_ip)) {
2874                 pr_info("multicast destination requires interface to be specified\n");
2875                 return -EINVAL;
2876         }
2877
2878         if (conf->mtu) {
2879                 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2880                 if (err)
2881                         return err;
2882         }
2883
2884         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2885                 needed_headroom += VXLAN6_HEADROOM;
2886         else
2887                 needed_headroom += VXLAN_HEADROOM;
2888         dev->needed_headroom = needed_headroom;
2889
2890         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2891         if (!vxlan->cfg.dst_port) {
2892                 if (conf->flags & VXLAN_F_GPE)
2893                         vxlan->cfg.dst_port = htons(4790); /* IANA VXLAN-GPE port */
2894                 else
2895                         vxlan->cfg.dst_port = default_port;
2896         }
2897         vxlan->flags |= conf->flags;
2898
2899         if (!vxlan->cfg.age_interval)
2900                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2901
2902         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2903                 if (tmp->cfg.vni == conf->vni &&
2904                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2905                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2906                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2907                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2908                     (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
2909                         pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
2910                         return -EEXIST;
2911                 }
2912         }
2913
2914         dev->ethtool_ops = &vxlan_ethtool_ops;
2915
2916         /* create an fdb entry for a valid default destination */
2917         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2918                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2919                                        &vxlan->default_dst.remote_ip,
2920                                        NUD_REACHABLE|NUD_PERMANENT,
2921                                        NLM_F_EXCL|NLM_F_CREATE,
2922                                        vxlan->cfg.dst_port,
2923                                        vxlan->default_dst.remote_vni,
2924                                        vxlan->default_dst.remote_ifindex,
2925                                        NTF_SELF);
2926                 if (err)
2927                         return err;
2928         }
2929
2930         err = register_netdevice(dev);
2931         if (err) {
2932                 vxlan_fdb_delete_default(vxlan);
2933                 return err;
2934         }
2935
2936         list_add(&vxlan->next, &vn->vxlan_list);
2937
2938         return 0;
2939 }
2940
2941 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2942                          struct nlattr *tb[], struct nlattr *data[])
2943 {
2944         struct vxlan_config conf;
2945
2946         memset(&conf, 0, sizeof(conf));
2947
2948         if (data[IFLA_VXLAN_ID])
2949                 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2950
2951         if (data[IFLA_VXLAN_GROUP]) {
2952                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2953         } else if (data[IFLA_VXLAN_GROUP6]) {
2954                 if (!IS_ENABLED(CONFIG_IPV6))
2955                         return -EPFNOSUPPORT;
2956
2957                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2958                 conf.remote_ip.sa.sa_family = AF_INET6;
2959         }
2960
2961         if (data[IFLA_VXLAN_LOCAL]) {
2962                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2963                 conf.saddr.sa.sa_family = AF_INET;
2964         } else if (data[IFLA_VXLAN_LOCAL6]) {
2965                 if (!IS_ENABLED(CONFIG_IPV6))
2966                         return -EPFNOSUPPORT;
2967
2968                 /* TODO: respect scope id */
2969                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2970                 conf.saddr.sa.sa_family = AF_INET6;
2971         }
2972
2973         if (data[IFLA_VXLAN_LINK])
2974                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2975
2976         if (data[IFLA_VXLAN_TOS])
2977                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2978
2979         if (data[IFLA_VXLAN_TTL])
2980                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2981
2982         if (data[IFLA_VXLAN_LABEL])
2983                 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
2984                              IPV6_FLOWLABEL_MASK;
2985
2986         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2987                 conf.flags |= VXLAN_F_LEARN;
2988
2989         if (data[IFLA_VXLAN_AGEING])
2990                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2991
2992         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2993                 conf.flags |= VXLAN_F_PROXY;
2994
2995         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2996                 conf.flags |= VXLAN_F_RSC;
2997
2998         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2999                 conf.flags |= VXLAN_F_L2MISS;
3000
3001         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3002                 conf.flags |= VXLAN_F_L3MISS;
3003
3004         if (data[IFLA_VXLAN_LIMIT])
3005                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3006
3007         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
3008             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3009                 conf.flags |= VXLAN_F_COLLECT_METADATA;
3010
3011         if (data[IFLA_VXLAN_PORT_RANGE]) {
3012                 const struct ifla_vxlan_port_range *p
3013                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3014                 conf.port_min = ntohs(p->low);
3015                 conf.port_max = ntohs(p->high);
3016         }
3017
3018         if (data[IFLA_VXLAN_PORT])
3019                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3020
3021         if (data[IFLA_VXLAN_UDP_CSUM] &&
3022             !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3023                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3024
3025         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
3026             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3027                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3028
3029         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3030             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3031                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3032
3033         if (data[IFLA_VXLAN_REMCSUM_TX] &&
3034             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3035                 conf.flags |= VXLAN_F_REMCSUM_TX;
3036
3037         if (data[IFLA_VXLAN_REMCSUM_RX] &&
3038             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3039                 conf.flags |= VXLAN_F_REMCSUM_RX;
3040
3041         if (data[IFLA_VXLAN_GBP])
3042                 conf.flags |= VXLAN_F_GBP;
3043
3044         if (data[IFLA_VXLAN_GPE])
3045                 conf.flags |= VXLAN_F_GPE;
3046
3047         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3048                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3049
3050         if (tb[IFLA_MTU])
3051                 conf.mtu = nla_get_u32(tb[IFLA_MTU]);
3052
3053         return vxlan_dev_configure(src_net, dev, &conf);
3054 }
3055
3056 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3057 {
3058         struct vxlan_dev *vxlan = netdev_priv(dev);
3059         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3060
3061         spin_lock(&vn->sock_lock);
3062         if (!hlist_unhashed(&vxlan->hlist))
3063                 hlist_del_rcu(&vxlan->hlist);
3064         spin_unlock(&vn->sock_lock);
3065
3066         gro_cells_destroy(&vxlan->gro_cells);
3067         list_del(&vxlan->next);
3068         unregister_netdevice_queue(dev, head);
3069 }
3070
3071 static size_t vxlan_get_size(const struct net_device *dev)
3072 {
3073
3074         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3075                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3076                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3077                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3078                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3079                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3080                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3081                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3082                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3083                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3084                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3085                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3086                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3087                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3088                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3089                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3090                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3091                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3092                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3093                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3094                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3095                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3096                 0;
3097 }
3098
3099 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3100 {
3101         const struct vxlan_dev *vxlan = netdev_priv(dev);
3102         const struct vxlan_rdst *dst = &vxlan->default_dst;
3103         struct ifla_vxlan_port_range ports = {
3104                 .low =  htons(vxlan->cfg.port_min),
3105                 .high = htons(vxlan->cfg.port_max),
3106         };
3107
3108         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3109                 goto nla_put_failure;
3110
3111         if (!vxlan_addr_any(&dst->remote_ip)) {
3112                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3113                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3114                                             dst->remote_ip.sin.sin_addr.s_addr))
3115                                 goto nla_put_failure;
3116 #if IS_ENABLED(CONFIG_IPV6)
3117                 } else {
3118                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3119                                              &dst->remote_ip.sin6.sin6_addr))
3120                                 goto nla_put_failure;
3121 #endif
3122                 }
3123         }
3124
3125         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3126                 goto nla_put_failure;
3127
3128         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3129                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3130                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3131                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3132                                 goto nla_put_failure;
3133 #if IS_ENABLED(CONFIG_IPV6)
3134                 } else {
3135                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3136                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3137                                 goto nla_put_failure;
3138 #endif
3139                 }
3140         }
3141
3142         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3143             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3144             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3145             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3146                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3147             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3148                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3149             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3150             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3151                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3152             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3153                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3154             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3155                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3156             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3157             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3158             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3159             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3160                         !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3161             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3162                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3163             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3164                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3165             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3166                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3167             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3168                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3169                 goto nla_put_failure;
3170
3171         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3172                 goto nla_put_failure;
3173
3174         if (vxlan->flags & VXLAN_F_GBP &&
3175             nla_put_flag(skb, IFLA_VXLAN_GBP))
3176                 goto nla_put_failure;
3177
3178         if (vxlan->flags & VXLAN_F_GPE &&
3179             nla_put_flag(skb, IFLA_VXLAN_GPE))
3180                 goto nla_put_failure;
3181
3182         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3183             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3184                 goto nla_put_failure;
3185
3186         return 0;
3187
3188 nla_put_failure:
3189         return -EMSGSIZE;
3190 }
3191
3192 static struct net *vxlan_get_link_net(const struct net_device *dev)
3193 {
3194         struct vxlan_dev *vxlan = netdev_priv(dev);
3195
3196         return vxlan->net;
3197 }
3198
3199 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3200         .kind           = "vxlan",
3201         .maxtype        = IFLA_VXLAN_MAX,
3202         .policy         = vxlan_policy,
3203         .priv_size      = sizeof(struct vxlan_dev),
3204         .setup          = vxlan_setup,
3205         .validate       = vxlan_validate,
3206         .newlink        = vxlan_newlink,
3207         .dellink        = vxlan_dellink,
3208         .get_size       = vxlan_get_size,
3209         .fill_info      = vxlan_fill_info,
3210         .get_link_net   = vxlan_get_link_net,
3211 };
3212
3213 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3214                                     u8 name_assign_type,
3215                                     struct vxlan_config *conf)
3216 {
3217         struct nlattr *tb[IFLA_MAX + 1];
3218         struct net_device *dev;
3219         int err;
3220
3221         memset(&tb, 0, sizeof(tb));
3222
3223         dev = rtnl_create_link(net, name, name_assign_type,
3224                                &vxlan_link_ops, tb);
3225         if (IS_ERR(dev))
3226                 return dev;
3227
3228         err = vxlan_dev_configure(net, dev, conf);
3229         if (err < 0) {
3230                 free_netdev(dev);
3231                 return ERR_PTR(err);
3232         }
3233
3234         err = rtnl_configure_link(dev, NULL);
3235         if (err < 0) {
3236                 LIST_HEAD(list_kill);
3237
3238                 vxlan_dellink(dev, &list_kill);
3239                 unregister_netdevice_many(&list_kill);
3240                 return ERR_PTR(err);
3241         }
3242
3243         return dev;
3244 }
3245 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3246
3247 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3248                                              struct net_device *dev)
3249 {
3250         struct vxlan_dev *vxlan, *next;
3251         LIST_HEAD(list_kill);
3252
3253         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3254                 struct vxlan_rdst *dst = &vxlan->default_dst;
3255
3256                 /* In case we created vxlan device with carrier
3257                  * and we loose the carrier due to module unload
3258                  * we also need to remove vxlan device. In other
3259                  * cases, it's not necessary and remote_ifindex
3260                  * is 0 here, so no matches.
3261                  */
3262                 if (dst->remote_ifindex == dev->ifindex)
3263                         vxlan_dellink(vxlan->dev, &list_kill);
3264         }
3265
3266         unregister_netdevice_many(&list_kill);
3267 }
3268
3269 static int vxlan_netdevice_event(struct notifier_block *unused,
3270                                  unsigned long event, void *ptr)
3271 {
3272         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3273         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3274
3275         if (event == NETDEV_UNREGISTER)
3276                 vxlan_handle_lowerdev_unregister(vn, dev);
3277         else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3278                 vxlan_push_rx_ports(dev);
3279
3280         return NOTIFY_DONE;
3281 }
3282
3283 static struct notifier_block vxlan_notifier_block __read_mostly = {
3284         .notifier_call = vxlan_netdevice_event,
3285 };
3286
3287 static __net_init int vxlan_init_net(struct net *net)
3288 {
3289         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3290         unsigned int h;
3291
3292         INIT_LIST_HEAD(&vn->vxlan_list);
3293         spin_lock_init(&vn->sock_lock);
3294
3295         for (h = 0; h < PORT_HASH_SIZE; ++h)
3296                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3297
3298         return 0;
3299 }
3300
3301 static void __net_exit vxlan_exit_net(struct net *net)
3302 {
3303         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3304         struct vxlan_dev *vxlan, *next;
3305         struct net_device *dev, *aux;
3306         LIST_HEAD(list);
3307
3308         rtnl_lock();
3309         for_each_netdev_safe(net, dev, aux)
3310                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3311                         unregister_netdevice_queue(dev, &list);
3312
3313         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3314                 /* If vxlan->dev is in the same netns, it has already been added
3315                  * to the list by the previous loop.
3316                  */
3317                 if (!net_eq(dev_net(vxlan->dev), net)) {
3318                         gro_cells_destroy(&vxlan->gro_cells);
3319                         unregister_netdevice_queue(vxlan->dev, &list);
3320                 }
3321         }
3322
3323         unregister_netdevice_many(&list);
3324         rtnl_unlock();
3325 }
3326
3327 static struct pernet_operations vxlan_net_ops = {
3328         .init = vxlan_init_net,
3329         .exit = vxlan_exit_net,
3330         .id   = &vxlan_net_id,
3331         .size = sizeof(struct vxlan_net),
3332 };
3333
3334 static int __init vxlan_init_module(void)
3335 {
3336         int rc;
3337
3338         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3339
3340         rc = register_pernet_subsys(&vxlan_net_ops);
3341         if (rc)
3342                 goto out1;
3343
3344         rc = register_netdevice_notifier(&vxlan_notifier_block);
3345         if (rc)
3346                 goto out2;
3347
3348         rc = rtnl_link_register(&vxlan_link_ops);
3349         if (rc)
3350                 goto out3;
3351
3352         return 0;
3353 out3:
3354         unregister_netdevice_notifier(&vxlan_notifier_block);
3355 out2:
3356         unregister_pernet_subsys(&vxlan_net_ops);
3357 out1:
3358         return rc;
3359 }
3360 late_initcall(vxlan_init_module);
3361
3362 static void __exit vxlan_cleanup_module(void)
3363 {
3364         rtnl_link_unregister(&vxlan_link_ops);
3365         unregister_netdevice_notifier(&vxlan_notifier_block);
3366         unregister_pernet_subsys(&vxlan_net_ops);
3367         /* rcu_barrier() is called by netns */
3368 }
3369 module_exit(vxlan_cleanup_module);
3370
3371 MODULE_LICENSE("GPL");
3372 MODULE_VERSION(VXLAN_VERSION);
3373 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3374 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3375 MODULE_ALIAS_RTNL_LINK("vxlan");