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 #include <net/protocol.h>
  31
  32 #if IS_ENABLED(CONFIG_IPV6)
  33 #include <net/ip6_tunnel.h>
  34 #include <net/ip6_checksum.h>
  35 #endif
  36
  37 #define VXLAN_VERSION   "0.1"
  38
  39 #define PORT_HASH_BITS  8
  40 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  41 #define FDB_AGE_DEFAULT 300 /* 5 min */
  42 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  43
  44 /* UDP port for VXLAN traffic.
  45  * The IANA assigned port is 4789, but the Linux default is 8472
  46  * for compatibility with early adopters.
  47  */
  48 static unsigned short vxlan_port __read_mostly = 8472;
  49 module_param_named(udp_port, vxlan_port, ushort, 0444);
  50 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  51
  52 static bool log_ecn_error = true;
  53 module_param(log_ecn_error, bool, 0644);
  54 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  55
  56 static int vxlan_net_id;
  57 static struct rtnl_link_ops vxlan_link_ops;
  58
  59 static const u8 all_zeros_mac[ETH_ALEN + 2];
  60
  61 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  62
  63 /* per-network namespace private data for this module */
  64 struct vxlan_net {
  65         struct list_head  vxlan_list;
  66         struct hlist_head sock_list[PORT_HASH_SIZE];
  67         spinlock_t        sock_lock;
  68 };
  69
  70 /* Forwarding table entry */
  71 struct vxlan_fdb {
  72         struct hlist_node hlist;        /* linked list of entries */
  73         struct rcu_head   rcu;
  74         unsigned long     updated;      /* jiffies */
  75         unsigned long     used;
  76         struct list_head  remotes;
  77         u8                eth_addr[ETH_ALEN];
  78         u16               state;        /* see ndm_state */
  79         u8                flags;        /* see ndm_flags */
  80 };
  81
  82 /* salt for hash table */
  83 static u32 vxlan_salt __read_mostly;
  84
  85 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  86 {
  87         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  88                ip_tunnel_collect_metadata();
  89 }
  90
  91 #if IS_ENABLED(CONFIG_IPV6)
  92 static inline
  93 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
  94 {
  95         if (a->sa.sa_family != b->sa.sa_family)
  96                 return false;
  97         if (a->sa.sa_family == AF_INET6)
  98                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
  99         else
 100                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 101 }
 102
 103 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 104 {
 105         if (ipa->sa.sa_family == AF_INET6)
 106                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 107         else
 108                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 109 }
 110
 111 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 112 {
 113         if (ipa->sa.sa_family == AF_INET6)
 114                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 115         else
 116                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 117 }
 118
 119 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 120 {
 121         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 122                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 123                 ip->sa.sa_family = AF_INET6;
 124                 return 0;
 125         } else if (nla_len(nla) >= sizeof(__be32)) {
 126                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 127                 ip->sa.sa_family = AF_INET;
 128                 return 0;
 129         } else {
 130                 return -EAFNOSUPPORT;
 131         }
 132 }
 133
 134 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 135                               const union vxlan_addr *ip)
 136 {
 137         if (ip->sa.sa_family == AF_INET6)
 138                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 139         else
 140                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 141 }
 142
 143 #else /* !CONFIG_IPV6 */
 144
 145 static inline
 146 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 147 {
 148         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 149 }
 150
 151 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 152 {
 153         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 154 }
 155
 156 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 157 {
 158         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 159 }
 160
 161 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 162 {
 163         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 164                 return -EAFNOSUPPORT;
 165         } else if (nla_len(nla) >= sizeof(__be32)) {
 166                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 167                 ip->sa.sa_family = AF_INET;
 168                 return 0;
 169         } else {
 170                 return -EAFNOSUPPORT;
 171         }
 172 }
 173
 174 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 175                               const union vxlan_addr *ip)
 176 {
 177         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 178 }
 179 #endif
 180
 181 /* Virtual Network hash table head */
 182 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 183 {
 184         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 185 }
 186
 187 /* Socket hash table head */
 188 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 189 {
 190         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 191
 192         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 193 }
 194
 195 /* First remote destination for a forwarding entry.
 196  * Guaranteed to be non-NULL because remotes are never deleted.
 197  */
 198 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 199 {
 200         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 201 }
 202
 203 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 204 {
 205         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 206 }
 207
 208 /* Find VXLAN socket based on network namespace, address family and UDP port
 209  * and enabled unshareable flags.
 210  */
 211 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 212                                           __be16 port, u32 flags)
 213 {
 214         struct vxlan_sock *vs;
 215
 216         flags &= VXLAN_F_RCV_FLAGS;
 217
 218         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 219                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 220                     vxlan_get_sk_family(vs) == family &&
 221                     vs->flags == flags)
 222                         return vs;
 223         }
 224         return NULL;
 225 }
 226
 227 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
 228 {
 229         struct vxlan_dev *vxlan;
 230
 231         /* For flow based devices, map all packets to VNI 0 */
 232         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 233                 vni = 0;
 234
 235         hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
 236                 if (vxlan->default_dst.remote_vni == vni)
 237                         return vxlan;
 238         }
 239
 240         return NULL;
 241 }
 242
 243 /* Look up VNI in a per net namespace table */
 244 static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
 245                                         sa_family_t family, __be16 port,
 246                                         u32 flags)
 247 {
 248         struct vxlan_sock *vs;
 249
 250         vs = vxlan_find_sock(net, family, port, flags);
 251         if (!vs)
 252                 return NULL;
 253
 254         return vxlan_vs_find_vni(vs, vni);
 255 }
 256
 257 /* Fill in neighbour message in skbuff. */
 258 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 259                           const struct vxlan_fdb *fdb,
 260                           u32 portid, u32 seq, int type, unsigned int flags,
 261                           const struct vxlan_rdst *rdst)
 262 {
 263         unsigned long now = jiffies;
 264         struct nda_cacheinfo ci;
 265         struct nlmsghdr *nlh;
 266         struct ndmsg *ndm;
 267         bool send_ip, send_eth;
 268
 269         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 270         if (nlh == NULL)
 271                 return -EMSGSIZE;
 272
 273         ndm = nlmsg_data(nlh);
 274         memset(ndm, 0, sizeof(*ndm));
 275
 276         send_eth = send_ip = true;
 277
 278         if (type == RTM_GETNEIGH) {
 279                 ndm->ndm_family = AF_INET;
 280                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 281                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 282         } else
 283                 ndm->ndm_family = AF_BRIDGE;
 284         ndm->ndm_state = fdb->state;
 285         ndm->ndm_ifindex = vxlan->dev->ifindex;
 286         ndm->ndm_flags = fdb->flags;
 287         ndm->ndm_type = RTN_UNICAST;
 288
 289         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 290             nla_put_s32(skb, NDA_LINK_NETNSID,
 291                         peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
 292                 goto nla_put_failure;
 293
 294         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 295                 goto nla_put_failure;
 296
 297         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 298                 goto nla_put_failure;
 299
 300         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 301             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 302                 goto nla_put_failure;
 303         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 304             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 305                 goto nla_put_failure;
 306         if (rdst->remote_ifindex &&
 307             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 308                 goto nla_put_failure;
 309
 310         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 311         ci.ndm_confirmed = 0;
 312         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 313         ci.ndm_refcnt    = 0;
 314
 315         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 316                 goto nla_put_failure;
 317
 318         nlmsg_end(skb, nlh);
 319         return 0;
 320
 321 nla_put_failure:
 322         nlmsg_cancel(skb, nlh);
 323         return -EMSGSIZE;
 324 }
 325
 326 static inline size_t vxlan_nlmsg_size(void)
 327 {
 328         return NLMSG_ALIGN(sizeof(struct ndmsg))
 329                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 330                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 331                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 332                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 333                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 334                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 335                 + nla_total_size(sizeof(struct nda_cacheinfo));
 336 }
 337
 338 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 339                              struct vxlan_rdst *rd, int type)
 340 {
 341         struct net *net = dev_net(vxlan->dev);
 342         struct sk_buff *skb;
 343         int err = -ENOBUFS;
 344
 345         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 346         if (skb == NULL)
 347                 goto errout;
 348
 349         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 350         if (err < 0) {
 351                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 352                 WARN_ON(err == -EMSGSIZE);
 353                 kfree_skb(skb);
 354                 goto errout;
 355         }
 356
 357         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 358         return;
 359 errout:
 360         if (err < 0)
 361                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 362 }
 363
 364 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 365 {
 366         struct vxlan_dev *vxlan = netdev_priv(dev);
 367         struct vxlan_fdb f = {
 368                 .state = NUD_STALE,
 369         };
 370         struct vxlan_rdst remote = {
 371                 .remote_ip = *ipa, /* goes to NDA_DST */
 372                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 373         };
 374
 375         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 376 }
 377
 378 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 379 {
 380         struct vxlan_fdb f = {
 381                 .state = NUD_STALE,
 382         };
 383         struct vxlan_rdst remote = { };
 384
 385         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 386
 387         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 388 }
 389
 390 /* Hash Ethernet address */
 391 static u32 eth_hash(const unsigned char *addr)
 392 {
 393         u64 value = get_unaligned((u64 *)addr);
 394
 395         /* only want 6 bytes */
 396 #ifdef __BIG_ENDIAN
 397         value >>= 16;
 398 #else
 399         value <<= 16;
 400 #endif
 401         return hash_64(value, FDB_HASH_BITS);
 402 }
 403
 404 /* Hash chain to use given mac address */
 405 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 406                                                 const u8 *mac)
 407 {
 408         return &vxlan->fdb_head[eth_hash(mac)];
 409 }
 410
 411 /* Look up Ethernet address in forwarding table */
 412 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 413                                         const u8 *mac)
 414 {
 415         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 416         struct vxlan_fdb *f;
 417
 418         hlist_for_each_entry_rcu(f, head, hlist) {
 419                 if (ether_addr_equal(mac, f->eth_addr))
 420                         return f;
 421         }
 422
 423         return NULL;
 424 }
 425
 426 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 427                                         const u8 *mac)
 428 {
 429         struct vxlan_fdb *f;
 430
 431         f = __vxlan_find_mac(vxlan, mac);
 432         if (f)
 433                 f->used = jiffies;
 434
 435         return f;
 436 }
 437
 438 /* caller should hold vxlan->hash_lock */
 439 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 440                                               union vxlan_addr *ip, __be16 port,
 441                                               __be32 vni, __u32 ifindex)
 442 {
 443         struct vxlan_rdst *rd;
 444
 445         list_for_each_entry(rd, &f->remotes, list) {
 446                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 447                     rd->remote_port == port &&
 448                     rd->remote_vni == vni &&
 449                     rd->remote_ifindex == ifindex)
 450                         return rd;
 451         }
 452
 453         return NULL;
 454 }
 455
 456 /* Replace destination of unicast mac */
 457 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 458                              union vxlan_addr *ip, __be16 port, __be32 vni,
 459                              __u32 ifindex)
 460 {
 461         struct vxlan_rdst *rd;
 462
 463         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 464         if (rd)
 465                 return 0;
 466
 467         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 468         if (!rd)
 469                 return 0;
 470
 471         dst_cache_reset(&rd->dst_cache);
 472         rd->remote_ip = *ip;
 473         rd->remote_port = port;
 474         rd->remote_vni = vni;
 475         rd->remote_ifindex = ifindex;
 476         return 1;
 477 }
 478
 479 /* Add/update destinations for multicast */
 480 static int vxlan_fdb_append(struct vxlan_fdb *f,
 481                             union vxlan_addr *ip, __be16 port, __be32 vni,
 482                             __u32 ifindex, struct vxlan_rdst **rdp)
 483 {
 484         struct vxlan_rdst *rd;
 485
 486         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 487         if (rd)
 488                 return 0;
 489
 490         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 491         if (rd == NULL)
 492                 return -ENOBUFS;
 493
 494         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 495                 kfree(rd);
 496                 return -ENOBUFS;
 497         }
 498
 499         rd->remote_ip = *ip;
 500         rd->remote_port = port;
 501         rd->remote_vni = vni;
 502         rd->remote_ifindex = ifindex;
 503
 504         list_add_tail_rcu(&rd->list, &f->remotes);
 505
 506         *rdp = rd;
 507         return 1;
 508 }
 509
 510 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 511                                           unsigned int off,
 512                                           struct vxlanhdr *vh, size_t hdrlen,
 513                                           __be32 vni_field,
 514                                           struct gro_remcsum *grc,
 515                                           bool nopartial)
 516 {
 517         size_t start, offset;
 518
 519         if (skb->remcsum_offload)
 520                 return vh;
 521
 522         if (!NAPI_GRO_CB(skb)->csum_valid)
 523                 return NULL;
 524
 525         start = vxlan_rco_start(vni_field);
 526         offset = start + vxlan_rco_offset(vni_field);
 527
 528         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 529                                      start, offset, grc, nopartial);
 530
 531         skb->remcsum_offload = 1;
 532
 533         return vh;
 534 }
 535
 536 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
 537                                           struct sk_buff **head,
 538                                           struct sk_buff *skb)
 539 {
 540         struct sk_buff *p, **pp = NULL;
 541         struct vxlanhdr *vh, *vh2;
 542         unsigned int hlen, off_vx;
 543         int flush = 1;
 544         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 545         __be32 flags;
 546         struct gro_remcsum grc;
 547
 548         skb_gro_remcsum_init(&grc);
 549
 550         off_vx = skb_gro_offset(skb);
 551         hlen = off_vx + sizeof(*vh);
 552         vh   = skb_gro_header_fast(skb, off_vx);
 553         if (skb_gro_header_hard(skb, hlen)) {
 554                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 555                 if (unlikely(!vh))
 556                         goto out;
 557         }
 558
 559         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 560
 561         flags = vh->vx_flags;
 562
 563         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 564                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 565                                        vh->vx_vni, &grc,
 566                                        !!(vs->flags &
 567                                           VXLAN_F_REMCSUM_NOPARTIAL));
 568
 569                 if (!vh)
 570                         goto out;
 571         }
 572
 573         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 574
 575         for (p = *head; p; p = p->next) {
 576                 if (!NAPI_GRO_CB(p)->same_flow)
 577                         continue;
 578
 579                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 580                 if (vh->vx_flags != vh2->vx_flags ||
 581                     vh->vx_vni != vh2->vx_vni) {
 582                         NAPI_GRO_CB(p)->same_flow = 0;
 583                         continue;
 584                 }
 585         }
 586
 587         pp = eth_gro_receive(head, skb);
 588         flush = 0;
 589
 590 out:
 591         skb_gro_remcsum_cleanup(skb, &grc);
 592         NAPI_GRO_CB(skb)->flush |= flush;
 593
 594         return pp;
 595 }
 596
 597 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 598 {
 599         /* Sets 'skb->inner_mac_header' since we are always called with
 600          * 'skb->encapsulation' set.
 601          */
 602         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 603 }
 604
 605 /* Add new entry to forwarding table -- assumes lock held */
 606 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 607                             const u8 *mac, union vxlan_addr *ip,
 608                             __u16 state, __u16 flags,
 609                             __be16 port, __be32 vni, __u32 ifindex,
 610                             __u8 ndm_flags)
 611 {
 612         struct vxlan_rdst *rd = NULL;
 613         struct vxlan_fdb *f;
 614         int notify = 0;
 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                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 646                                                   &rd);
 647
 648                         if (rc < 0)
 649                                 return rc;
 650                         notify |= rc;
 651                 }
 652         } else {
 653                 if (!(flags & NLM_F_CREATE))
 654                         return -ENOENT;
 655
 656                 if (vxlan->cfg.addrmax &&
 657                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 658                         return -ENOSPC;
 659
 660                 /* Disallow replace to add a multicast entry */
 661                 if ((flags & NLM_F_REPLACE) &&
 662                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 663                         return -EOPNOTSUPP;
 664
 665                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 666                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 667                 if (!f)
 668                         return -ENOMEM;
 669
 670                 notify = 1;
 671                 f->state = state;
 672                 f->flags = ndm_flags;
 673                 f->updated = f->used = jiffies;
 674                 INIT_LIST_HEAD(&f->remotes);
 675                 memcpy(f->eth_addr, mac, ETH_ALEN);
 676
 677                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 678
 679                 ++vxlan->addrcnt;
 680                 hlist_add_head_rcu(&f->hlist,
 681                                    vxlan_fdb_head(vxlan, mac));
 682         }
 683
 684         if (notify) {
 685                 if (rd == NULL)
 686                         rd = first_remote_rtnl(f);
 687                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 688         }
 689
 690         return 0;
 691 }
 692
 693 static void vxlan_fdb_free(struct rcu_head *head)
 694 {
 695         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 696         struct vxlan_rdst *rd, *nd;
 697
 698         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 699                 dst_cache_destroy(&rd->dst_cache);
 700                 kfree(rd);
 701         }
 702         kfree(f);
 703 }
 704
 705 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 706 {
 707         netdev_dbg(vxlan->dev,
 708                     "delete %pM\n", f->eth_addr);
 709
 710         --vxlan->addrcnt;
 711         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 712
 713         hlist_del_rcu(&f->hlist);
 714         call_rcu(&f->rcu, vxlan_fdb_free);
 715 }
 716
 717 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 718                            union vxlan_addr *ip, __be16 *port, __be32 *vni,
 719                            u32 *ifindex)
 720 {
 721         struct net *net = dev_net(vxlan->dev);
 722         int err;
 723
 724         if (tb[NDA_DST]) {
 725                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 726                 if (err)
 727                         return err;
 728         } else {
 729                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 730                 if (remote->sa.sa_family == AF_INET) {
 731                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 732                         ip->sa.sa_family = AF_INET;
 733 #if IS_ENABLED(CONFIG_IPV6)
 734                 } else {
 735                         ip->sin6.sin6_addr = in6addr_any;
 736                         ip->sa.sa_family = AF_INET6;
 737 #endif
 738                 }
 739         }
 740
 741         if (tb[NDA_PORT]) {
 742                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 743                         return -EINVAL;
 744                 *port = nla_get_be16(tb[NDA_PORT]);
 745         } else {
 746                 *port = vxlan->cfg.dst_port;
 747         }
 748
 749         if (tb[NDA_VNI]) {
 750                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 751                         return -EINVAL;
 752                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 753         } else {
 754                 *vni = vxlan->default_dst.remote_vni;
 755         }
 756
 757         if (tb[NDA_IFINDEX]) {
 758                 struct net_device *tdev;
 759
 760                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 761                         return -EINVAL;
 762                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 763                 tdev = __dev_get_by_index(net, *ifindex);
 764                 if (!tdev)
 765                         return -EADDRNOTAVAIL;
 766         } else {
 767                 *ifindex = 0;
 768         }
 769
 770         return 0;
 771 }
 772
 773 /* Add static entry (via netlink) */
 774 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 775                          struct net_device *dev,
 776                          const unsigned char *addr, u16 vid, u16 flags)
 777 {
 778         struct vxlan_dev *vxlan = netdev_priv(dev);
 779         /* struct net *net = dev_net(vxlan->dev); */
 780         union vxlan_addr ip;
 781         __be16 port;
 782         __be32 vni;
 783         u32 ifindex;
 784         int err;
 785
 786         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 787                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 788                         ndm->ndm_state);
 789                 return -EINVAL;
 790         }
 791
 792         if (tb[NDA_DST] == NULL)
 793                 return -EINVAL;
 794
 795         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 796         if (err)
 797                 return err;
 798
 799         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 800                 return -EAFNOSUPPORT;
 801
 802         spin_lock_bh(&vxlan->hash_lock);
 803         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 804                                port, vni, ifindex, ndm->ndm_flags);
 805         spin_unlock_bh(&vxlan->hash_lock);
 806
 807         return err;
 808 }
 809
 810 /* Delete entry (via netlink) */
 811 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 812                             struct net_device *dev,
 813                             const unsigned char *addr, u16 vid)
 814 {
 815         struct vxlan_dev *vxlan = netdev_priv(dev);
 816         struct vxlan_fdb *f;
 817         struct vxlan_rdst *rd = NULL;
 818         union vxlan_addr ip;
 819         __be16 port;
 820         __be32 vni;
 821         u32 ifindex;
 822         int err;
 823
 824         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 825         if (err)
 826                 return err;
 827
 828         err = -ENOENT;
 829
 830         spin_lock_bh(&vxlan->hash_lock);
 831         f = vxlan_find_mac(vxlan, addr);
 832         if (!f)
 833                 goto out;
 834
 835         if (!vxlan_addr_any(&ip)) {
 836                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 837                 if (!rd)
 838                         goto out;
 839         }
 840
 841         err = 0;
 842
 843         /* remove a destination if it's not the only one on the list,
 844          * otherwise destroy the fdb entry
 845          */
 846         if (rd && !list_is_singular(&f->remotes)) {
 847                 list_del_rcu(&rd->list);
 848                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 849                 kfree_rcu(rd, rcu);
 850                 goto out;
 851         }
 852
 853         vxlan_fdb_destroy(vxlan, f);
 854
 855 out:
 856         spin_unlock_bh(&vxlan->hash_lock);
 857
 858         return err;
 859 }
 860
 861 /* Dump forwarding table */
 862 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 863                           struct net_device *dev,
 864                           struct net_device *filter_dev, int idx)
 865 {
 866         struct vxlan_dev *vxlan = netdev_priv(dev);
 867         unsigned int h;
 868
 869         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 870                 struct vxlan_fdb *f;
 871                 int err;
 872
 873                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 874                         struct vxlan_rdst *rd;
 875
 876                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 877                                 if (idx < cb->args[0])
 878                                         goto skip;
 879
 880                                 err = vxlan_fdb_info(skb, vxlan, f,
 881                                                      NETLINK_CB(cb->skb).portid,
 882                                                      cb->nlh->nlmsg_seq,
 883                                                      RTM_NEWNEIGH,
 884                                                      NLM_F_MULTI, rd);
 885                                 if (err < 0) {
 886                                         cb->args[1] = err;
 887                                         goto out;
 888                                 }
 889 skip:
 890                                 ++idx;
 891                         }
 892                 }
 893         }
 894 out:
 895         return idx;
 896 }
 897
 898 /* Watch incoming packets to learn mapping between Ethernet address
 899  * and Tunnel endpoint.
 900  * Return true if packet is bogus and should be dropped.
 901  */
 902 static bool vxlan_snoop(struct net_device *dev,
 903                         union vxlan_addr *src_ip, const u8 *src_mac)
 904 {
 905         struct vxlan_dev *vxlan = netdev_priv(dev);
 906         struct vxlan_fdb *f;
 907
 908         f = vxlan_find_mac(vxlan, src_mac);
 909         if (likely(f)) {
 910                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 911
 912                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 913                         return false;
 914
 915                 /* Don't migrate static entries, drop packets */
 916                 if (f->state & NUD_NOARP)
 917                         return true;
 918
 919                 if (net_ratelimit())
 920                         netdev_info(dev,
 921                                     "%pM migrated from %pIS to %pIS\n",
 922                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 923
 924                 rdst->remote_ip = *src_ip;
 925                 f->updated = jiffies;
 926                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 927         } else {
 928                 /* learned new entry */
 929                 spin_lock(&vxlan->hash_lock);
 930
 931                 /* close off race between vxlan_flush and incoming packets */
 932                 if (netif_running(dev))
 933                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 934                                          NUD_REACHABLE,
 935                                          NLM_F_EXCL|NLM_F_CREATE,
 936                                          vxlan->cfg.dst_port,
 937                                          vxlan->default_dst.remote_vni,
 938                                          0, NTF_SELF);
 939                 spin_unlock(&vxlan->hash_lock);
 940         }
 941
 942         return false;
 943 }
 944
 945 /* See if multicast group is already in use by other ID */
 946 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
 947 {
 948         struct vxlan_dev *vxlan;
 949         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
 950
 951         /* The vxlan_sock is only used by dev, leaving group has
 952          * no effect on other vxlan devices.
 953          */
 954         if (family == AF_INET && dev->vn4_sock &&
 955             atomic_read(&dev->vn4_sock->refcnt) == 1)
 956                 return false;
 957 #if IS_ENABLED(CONFIG_IPV6)
 958         if (family == AF_INET6 && dev->vn6_sock &&
 959             atomic_read(&dev->vn6_sock->refcnt) == 1)
 960                 return false;
 961 #endif
 962
 963         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
 964                 if (!netif_running(vxlan->dev) || vxlan == dev)
 965                         continue;
 966
 967                 if (family == AF_INET && vxlan->vn4_sock != dev->vn4_sock)
 968                         continue;
 969 #if IS_ENABLED(CONFIG_IPV6)
 970                 if (family == AF_INET6 && vxlan->vn6_sock != dev->vn6_sock)
 971                         continue;
 972 #endif
 973
 974                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
 975                                       &dev->default_dst.remote_ip))
 976                         continue;
 977
 978                 if (vxlan->default_dst.remote_ifindex !=
 979                     dev->default_dst.remote_ifindex)
 980                         continue;
 981
 982                 return true;
 983         }
 984
 985         return false;
 986 }
 987
 988 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
 989 {
 990         struct vxlan_net *vn;
 991
 992         if (!vs)
 993                 return false;
 994         if (!atomic_dec_and_test(&vs->refcnt))
 995                 return false;
 996
 997         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
 998         spin_lock(&vn->sock_lock);
 999         hlist_del_rcu(&vs->hlist);
1000         udp_tunnel_notify_del_rx_port(vs->sock,
1001                                       (vs->flags & VXLAN_F_GPE) ?
1002                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1003                                       UDP_TUNNEL_TYPE_VXLAN);
1004         spin_unlock(&vn->sock_lock);
1005
1006         return true;
1007 }
1008
1009 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1010 {
1011         bool ipv4 = __vxlan_sock_release_prep(vxlan->vn4_sock);
1012 #if IS_ENABLED(CONFIG_IPV6)
1013         bool ipv6 = __vxlan_sock_release_prep(vxlan->vn6_sock);
1014 #endif
1015
1016         synchronize_net();
1017
1018         if (ipv4) {
1019                 udp_tunnel_sock_release(vxlan->vn4_sock->sock);
1020                 kfree(vxlan->vn4_sock);
1021         }
1022
1023 #if IS_ENABLED(CONFIG_IPV6)
1024         if (ipv6) {
1025                 udp_tunnel_sock_release(vxlan->vn6_sock->sock);
1026                 kfree(vxlan->vn6_sock);
1027         }
1028 #endif
1029 }
1030
1031 /* Update multicast group membership when first VNI on
1032  * multicast address is brought up
1033  */
1034 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1035 {
1036         struct sock *sk;
1037         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1038         int ifindex = vxlan->default_dst.remote_ifindex;
1039         int ret = -EINVAL;
1040
1041         if (ip->sa.sa_family == AF_INET) {
1042                 struct ip_mreqn mreq = {
1043                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1044                         .imr_ifindex            = ifindex,
1045                 };
1046
1047                 sk = vxlan->vn4_sock->sock->sk;
1048                 lock_sock(sk);
1049                 ret = ip_mc_join_group(sk, &mreq);
1050                 release_sock(sk);
1051 #if IS_ENABLED(CONFIG_IPV6)
1052         } else {
1053                 sk = vxlan->vn6_sock->sock->sk;
1054                 lock_sock(sk);
1055                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1056                                                    &ip->sin6.sin6_addr);
1057                 release_sock(sk);
1058 #endif
1059         }
1060
1061         return ret;
1062 }
1063
1064 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1065 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1066 {
1067         struct sock *sk;
1068         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1069         int ifindex = vxlan->default_dst.remote_ifindex;
1070         int ret = -EINVAL;
1071
1072         if (ip->sa.sa_family == AF_INET) {
1073                 struct ip_mreqn mreq = {
1074                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1075                         .imr_ifindex            = ifindex,
1076                 };
1077
1078                 sk = vxlan->vn4_sock->sock->sk;
1079                 lock_sock(sk);
1080                 ret = ip_mc_leave_group(sk, &mreq);
1081                 release_sock(sk);
1082 #if IS_ENABLED(CONFIG_IPV6)
1083         } else {
1084                 sk = vxlan->vn6_sock->sock->sk;
1085                 lock_sock(sk);
1086                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1087                                                    &ip->sin6.sin6_addr);
1088                 release_sock(sk);
1089 #endif
1090         }
1091
1092         return ret;
1093 }
1094
1095 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1096                           struct sk_buff *skb, u32 vxflags)
1097 {
1098         size_t start, offset;
1099
1100         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1101                 goto out;
1102
1103         start = vxlan_rco_start(unparsed->vx_vni);
1104         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1105
1106         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1107                 return false;
1108
1109         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1110                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1111 out:
1112         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1113         unparsed->vx_vni &= VXLAN_VNI_MASK;
1114         return true;
1115 }
1116
1117 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1118                                 struct sk_buff *skb, u32 vxflags,
1119                                 struct vxlan_metadata *md)
1120 {
1121         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1122         struct metadata_dst *tun_dst;
1123
1124         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1125                 goto out;
1126
1127         md->gbp = ntohs(gbp->policy_id);
1128
1129         tun_dst = (struct metadata_dst *)skb_dst(skb);
1130         if (tun_dst) {
1131                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1132                 tun_dst->u.tun_info.options_len = sizeof(*md);
1133         }
1134         if (gbp->dont_learn)
1135                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1136
1137         if (gbp->policy_applied)
1138                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1139
1140         /* In flow-based mode, GBP is carried in dst_metadata */
1141         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1142                 skb->mark = md->gbp;
1143 out:
1144         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1145 }
1146
1147 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1148                                 __be16 *protocol,
1149                                 struct sk_buff *skb, u32 vxflags)
1150 {
1151         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1152
1153         /* Need to have Next Protocol set for interfaces in GPE mode. */
1154         if (!gpe->np_applied)
1155                 return false;
1156         /* "The initial version is 0. If a receiver does not support the
1157          * version indicated it MUST drop the packet.
1158          */
1159         if (gpe->version != 0)
1160                 return false;
1161         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1162          * processing MUST occur." However, we don't implement OAM
1163          * processing, thus drop the packet.
1164          */
1165         if (gpe->oam_flag)
1166                 return false;
1167
1168         switch (gpe->next_protocol) {
1169         case VXLAN_GPE_NP_IPV4:
1170                 *protocol = htons(ETH_P_IP);
1171                 break;
1172         case VXLAN_GPE_NP_IPV6:
1173                 *protocol = htons(ETH_P_IPV6);
1174                 break;
1175         case VXLAN_GPE_NP_ETHERNET:
1176                 *protocol = htons(ETH_P_TEB);
1177                 break;
1178         default:
1179                 return false;
1180         }
1181
1182         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1183         return true;
1184 }
1185
1186 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1187                           struct vxlan_sock *vs,
1188                           struct sk_buff *skb)
1189 {
1190         union vxlan_addr saddr;
1191
1192         skb_reset_mac_header(skb);
1193         skb->protocol = eth_type_trans(skb, vxlan->dev);
1194         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1195
1196         /* Ignore packet loops (and multicast echo) */
1197         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1198                 return false;
1199
1200         /* Get address from the outer IP header */
1201         if (vxlan_get_sk_family(vs) == AF_INET) {
1202                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1203                 saddr.sa.sa_family = AF_INET;
1204 #if IS_ENABLED(CONFIG_IPV6)
1205         } else {
1206                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1207                 saddr.sa.sa_family = AF_INET6;
1208 #endif
1209         }
1210
1211         if ((vxlan->flags & VXLAN_F_LEARN) &&
1212             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1213                 return false;
1214
1215         return true;
1216 }
1217
1218 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1219                                   struct sk_buff *skb)
1220 {
1221         int err = 0;
1222
1223         if (vxlan_get_sk_family(vs) == AF_INET)
1224                 err = IP_ECN_decapsulate(oiph, skb);
1225 #if IS_ENABLED(CONFIG_IPV6)
1226         else
1227                 err = IP6_ECN_decapsulate(oiph, skb);
1228 #endif
1229
1230         if (unlikely(err) && log_ecn_error) {
1231                 if (vxlan_get_sk_family(vs) == AF_INET)
1232                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1233                                              &((struct iphdr *)oiph)->saddr,
1234                                              ((struct iphdr *)oiph)->tos);
1235                 else
1236                         net_info_ratelimited("non-ECT from %pI6\n",
1237                                              &((struct ipv6hdr *)oiph)->saddr);
1238         }
1239         return err <= 1;
1240 }
1241
1242 /* Callback from net/ipv4/udp.c to receive packets */
1243 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1244 {
1245         struct pcpu_sw_netstats *stats;
1246         struct vxlan_dev *vxlan;
1247         struct vxlan_sock *vs;
1248         struct vxlanhdr unparsed;
1249         struct vxlan_metadata _md;
1250         struct vxlan_metadata *md = &_md;
1251         __be16 protocol = htons(ETH_P_TEB);
1252         bool raw_proto = false;
1253         void *oiph;
1254
1255         /* Need UDP and VXLAN header to be present */
1256         if (!pskb_may_pull(skb, VXLAN_HLEN))
1257                 goto drop;
1258
1259         unparsed = *vxlan_hdr(skb);
1260         /* VNI flag always required to be set */
1261         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1262                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1263                            ntohl(vxlan_hdr(skb)->vx_flags),
1264                            ntohl(vxlan_hdr(skb)->vx_vni));
1265                 /* Return non vxlan pkt */
1266                 goto drop;
1267         }
1268         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1269         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1270
1271         vs = rcu_dereference_sk_user_data(sk);
1272         if (!vs)
1273                 goto drop;
1274
1275         vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1276         if (!vxlan)
1277                 goto drop;
1278
1279         /* For backwards compatibility, only allow reserved fields to be
1280          * used by VXLAN extensions if explicitly requested.
1281          */
1282         if (vs->flags & VXLAN_F_GPE) {
1283                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1284                         goto drop;
1285                 raw_proto = true;
1286         }
1287
1288         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1289                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1290                         goto drop;
1291
1292         if (vxlan_collect_metadata(vs)) {
1293                 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1294                 struct metadata_dst *tun_dst;
1295
1296                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1297                                          vxlan_vni_to_tun_id(vni), sizeof(*md));
1298
1299                 if (!tun_dst)
1300                         goto drop;
1301
1302                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1303
1304                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1305         } else {
1306                 memset(md, 0, sizeof(*md));
1307         }
1308
1309         if (vs->flags & VXLAN_F_REMCSUM_RX)
1310                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1311                         goto drop;
1312         if (vs->flags & VXLAN_F_GBP)
1313                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1314         /* Note that GBP and GPE can never be active together. This is
1315          * ensured in vxlan_dev_configure.
1316          */
1317
1318         if (unparsed.vx_flags || unparsed.vx_vni) {
1319                 /* If there are any unprocessed flags remaining treat
1320                  * this as a malformed packet. This behavior diverges from
1321                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1322                  * in reserved fields are to be ignored. The approach here
1323                  * maintains compatibility with previous stack code, and also
1324                  * is more robust and provides a little more security in
1325                  * adding extensions to VXLAN.
1326                  */
1327                 goto drop;
1328         }
1329
1330         if (!raw_proto) {
1331                 if (!vxlan_set_mac(vxlan, vs, skb))
1332                         goto drop;
1333         } else {
1334                 skb_reset_mac_header(skb);
1335                 skb->dev = vxlan->dev;
1336                 skb->pkt_type = PACKET_HOST;
1337         }
1338
1339         oiph = skb_network_header(skb);
1340         skb_reset_network_header(skb);
1341
1342         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1343                 ++vxlan->dev->stats.rx_frame_errors;
1344                 ++vxlan->dev->stats.rx_errors;
1345                 goto drop;
1346         }
1347
1348         stats = this_cpu_ptr(vxlan->dev->tstats);
1349         u64_stats_update_begin(&stats->syncp);
1350         stats->rx_packets++;
1351         stats->rx_bytes += skb->len;
1352         u64_stats_update_end(&stats->syncp);
1353
1354         gro_cells_receive(&vxlan->gro_cells, skb);
1355         return 0;
1356
1357 drop:
1358         /* Consume bad packet */
1359         kfree_skb(skb);
1360         return 0;
1361 }
1362
1363 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1364 {
1365         struct vxlan_dev *vxlan = netdev_priv(dev);
1366         struct arphdr *parp;
1367         u8 *arpptr, *sha;
1368         __be32 sip, tip;
1369         struct neighbour *n;
1370
1371         if (dev->flags & IFF_NOARP)
1372                 goto out;
1373
1374         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1375                 dev->stats.tx_dropped++;
1376                 goto out;
1377         }
1378         parp = arp_hdr(skb);
1379
1380         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1381              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1382             parp->ar_pro != htons(ETH_P_IP) ||
1383             parp->ar_op != htons(ARPOP_REQUEST) ||
1384             parp->ar_hln != dev->addr_len ||
1385             parp->ar_pln != 4)
1386                 goto out;
1387         arpptr = (u8 *)parp + sizeof(struct arphdr);
1388         sha = arpptr;
1389         arpptr += dev->addr_len;        /* sha */
1390         memcpy(&sip, arpptr, sizeof(sip));
1391         arpptr += sizeof(sip);
1392         arpptr += dev->addr_len;        /* tha */
1393         memcpy(&tip, arpptr, sizeof(tip));
1394
1395         if (ipv4_is_loopback(tip) ||
1396             ipv4_is_multicast(tip))
1397                 goto out;
1398
1399         n = neigh_lookup(&arp_tbl, &tip, dev);
1400
1401         if (n) {
1402                 struct vxlan_fdb *f;
1403                 struct sk_buff  *reply;
1404
1405                 if (!(n->nud_state & NUD_CONNECTED)) {
1406                         neigh_release(n);
1407                         goto out;
1408                 }
1409
1410                 f = vxlan_find_mac(vxlan, n->ha);
1411                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1412                         /* bridge-local neighbor */
1413                         neigh_release(n);
1414                         goto out;
1415                 }
1416
1417                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1418                                 n->ha, sha);
1419
1420                 neigh_release(n);
1421
1422                 if (reply == NULL)
1423                         goto out;
1424
1425                 skb_reset_mac_header(reply);
1426                 __skb_pull(reply, skb_network_offset(reply));
1427                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1428                 reply->pkt_type = PACKET_HOST;
1429
1430                 if (netif_rx_ni(reply) == NET_RX_DROP)
1431                         dev->stats.rx_dropped++;
1432         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1433                 union vxlan_addr ipa = {
1434                         .sin.sin_addr.s_addr = tip,
1435                         .sin.sin_family = AF_INET,
1436                 };
1437
1438                 vxlan_ip_miss(dev, &ipa);
1439         }
1440 out:
1441         consume_skb(skb);
1442         return NETDEV_TX_OK;
1443 }
1444
1445 #if IS_ENABLED(CONFIG_IPV6)
1446 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1447         struct neighbour *n, bool isrouter)
1448 {
1449         struct net_device *dev = request->dev;
1450         struct sk_buff *reply;
1451         struct nd_msg *ns, *na;
1452         struct ipv6hdr *pip6;
1453         u8 *daddr;
1454         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1455         int ns_olen;
1456         int i, len;
1457
1458         if (dev == NULL)
1459                 return NULL;
1460
1461         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1462                 sizeof(*na) + na_olen + dev->needed_tailroom;
1463         reply = alloc_skb(len, GFP_ATOMIC);
1464         if (reply == NULL)
1465                 return NULL;
1466
1467         reply->protocol = htons(ETH_P_IPV6);
1468         reply->dev = dev;
1469         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1470         skb_push(reply, sizeof(struct ethhdr));
1471         skb_reset_mac_header(reply);
1472
1473         ns = (struct nd_msg *)skb_transport_header(request);
1474
1475         daddr = eth_hdr(request)->h_source;
1476         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1477         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1478                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1479                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1480                         break;
1481                 }
1482         }
1483
1484         /* Ethernet header */
1485         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1486         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1487         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1488         reply->protocol = htons(ETH_P_IPV6);
1489
1490         skb_pull(reply, sizeof(struct ethhdr));
1491         skb_reset_network_header(reply);
1492         skb_put(reply, sizeof(struct ipv6hdr));
1493
1494         /* IPv6 header */
1495
1496         pip6 = ipv6_hdr(reply);
1497         memset(pip6, 0, sizeof(struct ipv6hdr));
1498         pip6->version = 6;
1499         pip6->priority = ipv6_hdr(request)->priority;
1500         pip6->nexthdr = IPPROTO_ICMPV6;
1501         pip6->hop_limit = 255;
1502         pip6->daddr = ipv6_hdr(request)->saddr;
1503         pip6->saddr = *(struct in6_addr *)n->primary_key;
1504
1505         skb_pull(reply, sizeof(struct ipv6hdr));
1506         skb_reset_transport_header(reply);
1507
1508         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1509
1510         /* Neighbor Advertisement */
1511         memset(na, 0, sizeof(*na)+na_olen);
1512         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1513         na->icmph.icmp6_router = isrouter;
1514         na->icmph.icmp6_override = 1;
1515         na->icmph.icmp6_solicited = 1;
1516         na->target = ns->target;
1517         ether_addr_copy(&na->opt[2], n->ha);
1518         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1519         na->opt[1] = na_olen >> 3;
1520
1521         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1522                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1523                 csum_partial(na, sizeof(*na)+na_olen, 0));
1524
1525         pip6->payload_len = htons(sizeof(*na)+na_olen);
1526
1527         skb_push(reply, sizeof(struct ipv6hdr));
1528
1529         reply->ip_summed = CHECKSUM_UNNECESSARY;
1530
1531         return reply;
1532 }
1533
1534 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1535 {
1536         struct vxlan_dev *vxlan = netdev_priv(dev);
1537         struct nd_msg *msg;
1538         const struct ipv6hdr *iphdr;
1539         const struct in6_addr *saddr, *daddr;
1540         struct neighbour *n;
1541         struct inet6_dev *in6_dev;
1542
1543         in6_dev = __in6_dev_get(dev);
1544         if (!in6_dev)
1545                 goto out;
1546
1547         iphdr = ipv6_hdr(skb);
1548         saddr = &iphdr->saddr;
1549         daddr = &iphdr->daddr;
1550
1551         msg = (struct nd_msg *)skb_transport_header(skb);
1552         if (msg->icmph.icmp6_code != 0 ||
1553             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1554                 goto out;
1555
1556         if (ipv6_addr_loopback(daddr) ||
1557             ipv6_addr_is_multicast(&msg->target))
1558                 goto out;
1559
1560         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1561
1562         if (n) {
1563                 struct vxlan_fdb *f;
1564                 struct sk_buff *reply;
1565
1566                 if (!(n->nud_state & NUD_CONNECTED)) {
1567                         neigh_release(n);
1568                         goto out;
1569                 }
1570
1571                 f = vxlan_find_mac(vxlan, n->ha);
1572                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1573                         /* bridge-local neighbor */
1574                         neigh_release(n);
1575                         goto out;
1576                 }
1577
1578                 reply = vxlan_na_create(skb, n,
1579                                         !!(f ? f->flags & NTF_ROUTER : 0));
1580
1581                 neigh_release(n);
1582
1583                 if (reply == NULL)
1584                         goto out;
1585
1586                 if (netif_rx_ni(reply) == NET_RX_DROP)
1587                         dev->stats.rx_dropped++;
1588
1589         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1590                 union vxlan_addr ipa = {
1591                         .sin6.sin6_addr = msg->target,
1592                         .sin6.sin6_family = AF_INET6,
1593                 };
1594
1595                 vxlan_ip_miss(dev, &ipa);
1596         }
1597
1598 out:
1599         consume_skb(skb);
1600         return NETDEV_TX_OK;
1601 }
1602 #endif
1603
1604 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1605 {
1606         struct vxlan_dev *vxlan = netdev_priv(dev);
1607         struct neighbour *n;
1608
1609         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1610                 return false;
1611
1612         n = NULL;
1613         switch (ntohs(eth_hdr(skb)->h_proto)) {
1614         case ETH_P_IP:
1615         {
1616                 struct iphdr *pip;
1617
1618                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1619                         return false;
1620                 pip = ip_hdr(skb);
1621                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1622                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1623                         union vxlan_addr ipa = {
1624                                 .sin.sin_addr.s_addr = pip->daddr,
1625                                 .sin.sin_family = AF_INET,
1626                         };
1627
1628                         vxlan_ip_miss(dev, &ipa);
1629                         return false;
1630                 }
1631
1632                 break;
1633         }
1634 #if IS_ENABLED(CONFIG_IPV6)
1635         case ETH_P_IPV6:
1636         {
1637                 struct ipv6hdr *pip6;
1638
1639                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1640                         return false;
1641                 pip6 = ipv6_hdr(skb);
1642                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1643                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1644                         union vxlan_addr ipa = {
1645                                 .sin6.sin6_addr = pip6->daddr,
1646                                 .sin6.sin6_family = AF_INET6,
1647                         };
1648
1649                         vxlan_ip_miss(dev, &ipa);
1650                         return false;
1651                 }
1652
1653                 break;
1654         }
1655 #endif
1656         default:
1657                 return false;
1658         }
1659
1660         if (n) {
1661                 bool diff;
1662
1663                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1664                 if (diff) {
1665                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1666                                 dev->addr_len);
1667                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1668                 }
1669                 neigh_release(n);
1670                 return diff;
1671         }
1672
1673         return false;
1674 }
1675
1676 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1677                                 struct vxlan_metadata *md)
1678 {
1679         struct vxlanhdr_gbp *gbp;
1680
1681         if (!md->gbp)
1682                 return;
1683
1684         gbp = (struct vxlanhdr_gbp *)vxh;
1685         vxh->vx_flags |= VXLAN_HF_GBP;
1686
1687         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1688                 gbp->dont_learn = 1;
1689
1690         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1691                 gbp->policy_applied = 1;
1692
1693         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1694 }
1695
1696 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1697                                __be16 protocol)
1698 {
1699         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1700
1701         gpe->np_applied = 1;
1702
1703         switch (protocol) {
1704         case htons(ETH_P_IP):
1705                 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1706                 return 0;
1707         case htons(ETH_P_IPV6):
1708                 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1709                 return 0;
1710         case htons(ETH_P_TEB):
1711                 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1712                 return 0;
1713         }
1714         return -EPFNOSUPPORT;
1715 }
1716
1717 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1718                            int iphdr_len, __be32 vni,
1719                            struct vxlan_metadata *md, u32 vxflags,
1720                            bool udp_sum)
1721 {
1722         struct vxlanhdr *vxh;
1723         int min_headroom;
1724         int err;
1725         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1726         __be16 inner_protocol = htons(ETH_P_TEB);
1727
1728         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1729             skb->ip_summed == CHECKSUM_PARTIAL) {
1730                 int csum_start = skb_checksum_start_offset(skb);
1731
1732                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1733                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1734                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1735                      skb->csum_offset == offsetof(struct tcphdr, check)))
1736                         type |= SKB_GSO_TUNNEL_REMCSUM;
1737         }
1738
1739         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1740                         + VXLAN_HLEN + iphdr_len
1741                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1742
1743         /* Need space for new headers (invalidates iph ptr) */
1744         err = skb_cow_head(skb, min_headroom);
1745         if (unlikely(err))
1746                 goto out_free;
1747
1748         skb = vlan_hwaccel_push_inside(skb);
1749         if (WARN_ON(!skb))
1750                 return -ENOMEM;
1751
1752         err = iptunnel_handle_offloads(skb, type);
1753         if (err)
1754                 goto out_free;
1755
1756         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1757         vxh->vx_flags = VXLAN_HF_VNI;
1758         vxh->vx_vni = vxlan_vni_field(vni);
1759
1760         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1761                 unsigned int start;
1762
1763                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1764                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1765                 vxh->vx_flags |= VXLAN_HF_RCO;
1766
1767                 if (!skb_is_gso(skb)) {
1768                         skb->ip_summed = CHECKSUM_NONE;
1769                         skb->encapsulation = 0;
1770                 }
1771         }
1772
1773         if (vxflags & VXLAN_F_GBP)
1774                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1775         if (vxflags & VXLAN_F_GPE) {
1776                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1777                 if (err < 0)
1778                         goto out_free;
1779                 inner_protocol = skb->protocol;
1780         }
1781
1782         skb_set_inner_protocol(skb, inner_protocol);
1783         return 0;
1784
1785 out_free:
1786         kfree_skb(skb);
1787         return err;
1788 }
1789
1790 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1791                                       struct sk_buff *skb, int oif, u8 tos,
1792                                       __be32 daddr, __be32 *saddr,
1793                                       struct dst_cache *dst_cache,
1794                                       const struct ip_tunnel_info *info)
1795 {
1796         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1797         struct rtable *rt = NULL;
1798         struct flowi4 fl4;
1799
1800         if (tos && !info)
1801                 use_cache = false;
1802         if (use_cache) {
1803                 rt = dst_cache_get_ip4(dst_cache, saddr);
1804                 if (rt)
1805                         return rt;
1806         }
1807
1808         memset(&fl4, 0, sizeof(fl4));
1809         fl4.flowi4_oif = oif;
1810         fl4.flowi4_tos = RT_TOS(tos);
1811         fl4.flowi4_mark = skb->mark;
1812         fl4.flowi4_proto = IPPROTO_UDP;
1813         fl4.daddr = daddr;
1814         fl4.saddr = *saddr;
1815
1816         rt = ip_route_output_key(vxlan->net, &fl4);
1817         if (!IS_ERR(rt)) {
1818                 *saddr = fl4.saddr;
1819                 if (use_cache)
1820                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1821         }
1822         return rt;
1823 }
1824
1825 #if IS_ENABLED(CONFIG_IPV6)
1826 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1827                                           struct sk_buff *skb, int oif, u8 tos,
1828                                           __be32 label,
1829                                           const struct in6_addr *daddr,
1830                                           struct in6_addr *saddr,
1831                                           struct dst_cache *dst_cache,
1832                                           const struct ip_tunnel_info *info)
1833 {
1834         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1835         struct dst_entry *ndst;
1836         struct flowi6 fl6;
1837         int err;
1838
1839         if (tos && !info)
1840                 use_cache = false;
1841         if (use_cache) {
1842                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1843                 if (ndst)
1844                         return ndst;
1845         }
1846
1847         memset(&fl6, 0, sizeof(fl6));
1848         fl6.flowi6_oif = oif;
1849         fl6.daddr = *daddr;
1850         fl6.saddr = *saddr;
1851         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1852         fl6.flowi6_mark = skb->mark;
1853         fl6.flowi6_proto = IPPROTO_UDP;
1854
1855         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1856                                          vxlan->vn6_sock->sock->sk,
1857                                          &ndst, &fl6);
1858         if (err < 0)
1859                 return ERR_PTR(err);
1860
1861         *saddr = fl6.saddr;
1862         if (use_cache)
1863                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1864         return ndst;
1865 }
1866 #endif
1867
1868 /* Bypass encapsulation if the destination is local */
1869 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1870                                struct vxlan_dev *dst_vxlan)
1871 {
1872         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1873         union vxlan_addr loopback;
1874         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1875         struct net_device *dev = skb->dev;
1876         int len = skb->len;
1877
1878         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1879         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1880         skb->pkt_type = PACKET_HOST;
1881         skb->encapsulation = 0;
1882         skb->dev = dst_vxlan->dev;
1883         __skb_pull(skb, skb_network_offset(skb));
1884
1885         if (remote_ip->sa.sa_family == AF_INET) {
1886                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1887                 loopback.sa.sa_family =  AF_INET;
1888 #if IS_ENABLED(CONFIG_IPV6)
1889         } else {
1890                 loopback.sin6.sin6_addr = in6addr_loopback;
1891                 loopback.sa.sa_family =  AF_INET6;
1892 #endif
1893         }
1894
1895         if (dst_vxlan->flags & VXLAN_F_LEARN)
1896                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1897
1898         u64_stats_update_begin(&tx_stats->syncp);
1899         tx_stats->tx_packets++;
1900         tx_stats->tx_bytes += len;
1901         u64_stats_update_end(&tx_stats->syncp);
1902
1903         if (netif_rx(skb) == NET_RX_SUCCESS) {
1904                 u64_stats_update_begin(&rx_stats->syncp);
1905                 rx_stats->rx_packets++;
1906                 rx_stats->rx_bytes += len;
1907                 u64_stats_update_end(&rx_stats->syncp);
1908         } else {
1909                 dev->stats.rx_dropped++;
1910         }
1911 }
1912
1913 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1914                            struct vxlan_rdst *rdst, bool did_rsc)
1915 {
1916         struct dst_cache *dst_cache;
1917         struct ip_tunnel_info *info;
1918         struct vxlan_dev *vxlan = netdev_priv(dev);
1919         struct sock *sk;
1920         struct rtable *rt = NULL;
1921         const struct iphdr *old_iph;
1922         union vxlan_addr *dst;
1923         union vxlan_addr remote_ip, local_ip;
1924         union vxlan_addr *src;
1925         struct vxlan_metadata _md;
1926         struct vxlan_metadata *md = &_md;
1927         __be16 src_port = 0, dst_port;
1928         __be32 vni, label;
1929         __be16 df = 0;
1930         __u8 tos, ttl;
1931         int err;
1932         u32 flags = vxlan->flags;
1933         bool udp_sum = false;
1934         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1935
1936         info = skb_tunnel_info(skb);
1937
1938         if (rdst) {
1939                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1940                 vni = rdst->remote_vni;
1941                 dst = &rdst->remote_ip;
1942                 src = &vxlan->cfg.saddr;
1943                 dst_cache = &rdst->dst_cache;
1944         } else {
1945                 if (!info) {
1946                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1947                                   dev->name);
1948                         goto drop;
1949                 }
1950                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1951                 vni = vxlan_tun_id_to_vni(info->key.tun_id);
1952                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1953                 if (remote_ip.sa.sa_family == AF_INET) {
1954                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1955                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
1956                 } else {
1957                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1958                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
1959                 }
1960                 dst = &remote_ip;
1961                 src = &local_ip;
1962                 dst_cache = &info->dst_cache;
1963         }
1964
1965         if (vxlan_addr_any(dst)) {
1966                 if (did_rsc) {
1967                         /* short-circuited back to local bridge */
1968                         vxlan_encap_bypass(skb, vxlan, vxlan);
1969                         return;
1970                 }
1971                 goto drop;
1972         }
1973
1974         old_iph = ip_hdr(skb);
1975
1976         ttl = vxlan->cfg.ttl;
1977         if (!ttl && vxlan_addr_multicast(dst))
1978                 ttl = 1;
1979
1980         tos = vxlan->cfg.tos;
1981         if (tos == 1)
1982                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1983
1984         label = vxlan->cfg.label;
1985         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
1986                                      vxlan->cfg.port_max, true);
1987
1988         if (info) {
1989                 ttl = info->key.ttl;
1990                 tos = info->key.tos;
1991                 label = info->key.label;
1992                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
1993
1994                 if (info->options_len)
1995                         md = ip_tunnel_info_opts(info);
1996         } else {
1997                 md->gbp = skb->mark;
1998         }
1999
2000         if (dst->sa.sa_family == AF_INET) {
2001                 if (!vxlan->vn4_sock)
2002                         goto drop;
2003                 sk = vxlan->vn4_sock->sock->sk;
2004
2005                 rt = vxlan_get_route(vxlan, skb,
2006                                      rdst ? rdst->remote_ifindex : 0, tos,
2007                                      dst->sin.sin_addr.s_addr,
2008                                      &src->sin.sin_addr.s_addr,
2009                                      dst_cache, info);
2010                 if (IS_ERR(rt)) {
2011                         netdev_dbg(dev, "no route to %pI4\n",
2012                                    &dst->sin.sin_addr.s_addr);
2013                         dev->stats.tx_carrier_errors++;
2014                         goto tx_error;
2015                 }
2016
2017                 if (rt->dst.dev == dev) {
2018                         netdev_dbg(dev, "circular route to %pI4\n",
2019                                    &dst->sin.sin_addr.s_addr);
2020                         dev->stats.collisions++;
2021                         goto rt_tx_error;
2022                 }
2023
2024                 /* Bypass encapsulation if the destination is local */
2025                 if (!info && rt->rt_flags & RTCF_LOCAL &&
2026                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2027                         struct vxlan_dev *dst_vxlan;
2028
2029                         ip_rt_put(rt);
2030                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2031                                                    dst->sa.sa_family, dst_port,
2032                                                    vxlan->flags);
2033                         if (!dst_vxlan)
2034                                 goto tx_error;
2035                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2036                         return;
2037                 }
2038
2039                 if (!info)
2040                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2041                 else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2042                         df = htons(IP_DF);
2043
2044                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2045                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2046                 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2047                                       vni, md, flags, udp_sum);
2048                 if (err < 0)
2049                         goto xmit_tx_error;
2050
2051                 udp_tunnel_xmit_skb(rt, sk, skb, src->sin.sin_addr.s_addr,
2052                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2053                                     src_port, dst_port, xnet, !udp_sum);
2054 #if IS_ENABLED(CONFIG_IPV6)
2055         } else {
2056                 struct dst_entry *ndst;
2057                 u32 rt6i_flags;
2058
2059                 if (!vxlan->vn6_sock)
2060                         goto drop;
2061                 sk = vxlan->vn6_sock->sock->sk;
2062
2063                 ndst = vxlan6_get_route(vxlan, skb,
2064                                         rdst ? rdst->remote_ifindex : 0, tos,
2065                                         label, &dst->sin6.sin6_addr,
2066                                         &src->sin6.sin6_addr,
2067                                         dst_cache, info);
2068                 if (IS_ERR(ndst)) {
2069                         netdev_dbg(dev, "no route to %pI6\n",
2070                                    &dst->sin6.sin6_addr);
2071                         dev->stats.tx_carrier_errors++;
2072                         goto tx_error;
2073                 }
2074
2075                 if (ndst->dev == dev) {
2076                         netdev_dbg(dev, "circular route to %pI6\n",
2077                                    &dst->sin6.sin6_addr);
2078                         dst_release(ndst);
2079                         dev->stats.collisions++;
2080                         goto tx_error;
2081                 }
2082
2083                 /* Bypass encapsulation if the destination is local */
2084                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2085                 if (!info && rt6i_flags & RTF_LOCAL &&
2086                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2087                         struct vxlan_dev *dst_vxlan;
2088
2089                         dst_release(ndst);
2090                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2091                                                    dst->sa.sa_family, dst_port,
2092                                                    vxlan->flags);
2093                         if (!dst_vxlan)
2094                                 goto tx_error;
2095                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2096                         return;
2097                 }
2098
2099                 if (!info)
2100                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2101
2102                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2103                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2104                 skb_scrub_packet(skb, xnet);
2105                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2106                                       vni, md, flags, udp_sum);
2107                 if (err < 0) {
2108                         dst_release(ndst);
2109                         return;
2110                 }
2111                 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2112                                      &src->sin6.sin6_addr,
2113                                      &dst->sin6.sin6_addr, tos, ttl,
2114                                      label, src_port, dst_port, !udp_sum);
2115 #endif
2116         }
2117
2118         return;
2119
2120 drop:
2121         dev->stats.tx_dropped++;
2122         goto tx_free;
2123
2124 xmit_tx_error:
2125         /* skb is already freed. */
2126         skb = NULL;
2127 rt_tx_error:
2128         ip_rt_put(rt);
2129 tx_error:
2130         dev->stats.tx_errors++;
2131 tx_free:
2132         dev_kfree_skb(skb);
2133 }
2134
2135 /* Transmit local packets over Vxlan
2136  *
2137  * Outer IP header inherits ECN and DF from inner header.
2138  * Outer UDP destination is the VXLAN assigned port.
2139  *           source port is based on hash of flow
2140  */
2141 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2142 {
2143         struct vxlan_dev *vxlan = netdev_priv(dev);
2144         const struct ip_tunnel_info *info;
2145         struct ethhdr *eth;
2146         bool did_rsc = false;
2147         struct vxlan_rdst *rdst, *fdst = NULL;
2148         struct vxlan_fdb *f;
2149
2150         info = skb_tunnel_info(skb);
2151
2152         skb_reset_mac_header(skb);
2153
2154         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2155                 if (info && info->mode & IP_TUNNEL_INFO_TX)
2156                         vxlan_xmit_one(skb, dev, NULL, false);
2157                 else
2158                         kfree_skb(skb);
2159                 return NETDEV_TX_OK;
2160         }
2161
2162         if (vxlan->flags & VXLAN_F_PROXY) {
2163                 eth = eth_hdr(skb);
2164                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2165                         return arp_reduce(dev, skb);
2166 #if IS_ENABLED(CONFIG_IPV6)
2167                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2168                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2169                                        + sizeof(struct nd_msg)) &&
2170                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2171                                 struct nd_msg *msg;
2172
2173                                 msg = (struct nd_msg *)skb_transport_header(skb);
2174                                 if (msg->icmph.icmp6_code == 0 &&
2175                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2176                                         return neigh_reduce(dev, skb);
2177                 }
2178 #endif
2179         }
2180
2181         eth = eth_hdr(skb);
2182         f = vxlan_find_mac(vxlan, eth->h_dest);
2183         did_rsc = false;
2184
2185         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2186             (ntohs(eth->h_proto) == ETH_P_IP ||
2187              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2188                 did_rsc = route_shortcircuit(dev, skb);
2189                 if (did_rsc)
2190                         f = vxlan_find_mac(vxlan, eth->h_dest);
2191         }
2192
2193         if (f == NULL) {
2194                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2195                 if (f == NULL) {
2196                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2197                             !is_multicast_ether_addr(eth->h_dest))
2198                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2199
2200                         dev->stats.tx_dropped++;
2201                         kfree_skb(skb);
2202                         return NETDEV_TX_OK;
2203                 }
2204         }
2205
2206         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2207                 struct sk_buff *skb1;
2208
2209                 if (!fdst) {
2210                         fdst = rdst;
2211                         continue;
2212                 }
2213                 skb1 = skb_clone(skb, GFP_ATOMIC);
2214                 if (skb1)
2215                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2216         }
2217
2218         if (fdst)
2219                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2220         else
2221                 kfree_skb(skb);
2222         return NETDEV_TX_OK;
2223 }
2224
2225 /* Walk the forwarding table and purge stale entries */
2226 static void vxlan_cleanup(unsigned long arg)
2227 {
2228         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2229         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2230         unsigned int h;
2231
2232         if (!netif_running(vxlan->dev))
2233                 return;
2234
2235         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2236                 struct hlist_node *p, *n;
2237
2238                 spin_lock_bh(&vxlan->hash_lock);
2239                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2240                         struct vxlan_fdb *f
2241                                 = container_of(p, struct vxlan_fdb, hlist);
2242                         unsigned long timeout;
2243
2244                         if (f->state & NUD_PERMANENT)
2245                                 continue;
2246
2247                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2248                         if (time_before_eq(timeout, jiffies)) {
2249                                 netdev_dbg(vxlan->dev,
2250                                            "garbage collect %pM\n",
2251                                            f->eth_addr);
2252                                 f->state = NUD_STALE;
2253                                 vxlan_fdb_destroy(vxlan, f);
2254                         } else if (time_before(timeout, next_timer))
2255                                 next_timer = timeout;
2256                 }
2257                 spin_unlock_bh(&vxlan->hash_lock);
2258         }
2259
2260         mod_timer(&vxlan->age_timer, next_timer);
2261 }
2262
2263 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2264 {
2265         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2266         __be32 vni = vxlan->default_dst.remote_vni;
2267
2268         spin_lock(&vn->sock_lock);
2269         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2270         spin_unlock(&vn->sock_lock);
2271 }
2272
2273 /* Setup stats when device is created */
2274 static int vxlan_init(struct net_device *dev)
2275 {
2276         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2277         if (!dev->tstats)
2278                 return -ENOMEM;
2279
2280         return 0;
2281 }
2282
2283 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2284 {
2285         struct vxlan_fdb *f;
2286
2287         spin_lock_bh(&vxlan->hash_lock);
2288         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2289         if (f)
2290                 vxlan_fdb_destroy(vxlan, f);
2291         spin_unlock_bh(&vxlan->hash_lock);
2292 }
2293
2294 static void vxlan_uninit(struct net_device *dev)
2295 {
2296         struct vxlan_dev *vxlan = netdev_priv(dev);
2297
2298         vxlan_fdb_delete_default(vxlan);
2299
2300         free_percpu(dev->tstats);
2301 }
2302
2303 /* Start ageing timer and join group when device is brought up */
2304 static int vxlan_open(struct net_device *dev)
2305 {
2306         struct vxlan_dev *vxlan = netdev_priv(dev);
2307         int ret;
2308
2309         ret = vxlan_sock_add(vxlan);
2310         if (ret < 0)
2311                 return ret;
2312
2313         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2314                 ret = vxlan_igmp_join(vxlan);
2315                 if (ret == -EADDRINUSE)
2316                         ret = 0;
2317                 if (ret) {
2318                         vxlan_sock_release(vxlan);
2319                         return ret;
2320                 }
2321         }
2322
2323         if (vxlan->cfg.age_interval)
2324                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2325
2326         return ret;
2327 }
2328
2329 /* Purge the forwarding table */
2330 static void vxlan_flush(struct vxlan_dev *vxlan)
2331 {
2332         unsigned int h;
2333
2334         spin_lock_bh(&vxlan->hash_lock);
2335         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2336                 struct hlist_node *p, *n;
2337                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2338                         struct vxlan_fdb *f
2339                                 = container_of(p, struct vxlan_fdb, hlist);
2340                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2341                         if (!is_zero_ether_addr(f->eth_addr))
2342                                 vxlan_fdb_destroy(vxlan, f);
2343                 }
2344         }
2345         spin_unlock_bh(&vxlan->hash_lock);
2346 }
2347
2348 /* Cleanup timer and forwarding table on shutdown */
2349 static int vxlan_stop(struct net_device *dev)
2350 {
2351         struct vxlan_dev *vxlan = netdev_priv(dev);
2352         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2353         int ret = 0;
2354
2355         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2356             !vxlan_group_used(vn, vxlan))
2357                 ret = vxlan_igmp_leave(vxlan);
2358
2359         del_timer_sync(&vxlan->age_timer);
2360
2361         vxlan_flush(vxlan);
2362         vxlan_sock_release(vxlan);
2363
2364         return ret;
2365 }
2366
2367 /* Stub, nothing needs to be done. */
2368 static void vxlan_set_multicast_list(struct net_device *dev)
2369 {
2370 }
2371
2372 static int __vxlan_change_mtu(struct net_device *dev,
2373                               struct net_device *lowerdev,
2374                               struct vxlan_rdst *dst, int new_mtu, bool strict)
2375 {
2376         int max_mtu = IP_MAX_MTU;
2377
2378         if (lowerdev)
2379                 max_mtu = lowerdev->mtu;
2380
2381         if (dst->remote_ip.sa.sa_family == AF_INET6)
2382                 max_mtu -= VXLAN6_HEADROOM;
2383         else
2384                 max_mtu -= VXLAN_HEADROOM;
2385
2386         if (new_mtu < 68)
2387                 return -EINVAL;
2388
2389         if (new_mtu > max_mtu) {
2390                 if (strict)
2391                         return -EINVAL;
2392
2393                 new_mtu = max_mtu;
2394         }
2395
2396         dev->mtu = new_mtu;
2397         return 0;
2398 }
2399
2400 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2401 {
2402         struct vxlan_dev *vxlan = netdev_priv(dev);
2403         struct vxlan_rdst *dst = &vxlan->default_dst;
2404         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2405                                                          dst->remote_ifindex);
2406         return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2407 }
2408
2409 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2410 {
2411         struct vxlan_dev *vxlan = netdev_priv(dev);
2412         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2413         __be16 sport, dport;
2414
2415         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2416                                   vxlan->cfg.port_max, true);
2417         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2418
2419         if (ip_tunnel_info_af(info) == AF_INET) {
2420                 struct rtable *rt;
2421
2422                 if (!vxlan->vn4_sock)
2423                         return -EINVAL;
2424                 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2425                                      info->key.u.ipv4.dst,
2426                                      &info->key.u.ipv4.src, NULL, info);
2427                 if (IS_ERR(rt))
2428                         return PTR_ERR(rt);
2429                 ip_rt_put(rt);
2430         } else {
2431 #if IS_ENABLED(CONFIG_IPV6)
2432                 struct dst_entry *ndst;
2433
2434                 if (!vxlan->vn6_sock)
2435                         return -EINVAL;
2436                 ndst = vxlan6_get_route(vxlan, skb, 0, info->key.tos,
2437                                         info->key.label, &info->key.u.ipv6.dst,
2438                                         &info->key.u.ipv6.src, NULL, info);
2439                 if (IS_ERR(ndst))
2440                         return PTR_ERR(ndst);
2441                 dst_release(ndst);
2442 #else /* !CONFIG_IPV6 */
2443                 return -EPFNOSUPPORT;
2444 #endif
2445         }
2446         info->key.tp_src = sport;
2447         info->key.tp_dst = dport;
2448         return 0;
2449 }
2450
2451 static const struct net_device_ops vxlan_netdev_ether_ops = {
2452         .ndo_init               = vxlan_init,
2453         .ndo_uninit             = vxlan_uninit,
2454         .ndo_open               = vxlan_open,
2455         .ndo_stop               = vxlan_stop,
2456         .ndo_start_xmit         = vxlan_xmit,
2457         .ndo_get_stats64        = ip_tunnel_get_stats64,
2458         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2459         .ndo_change_mtu         = vxlan_change_mtu,
2460         .ndo_validate_addr      = eth_validate_addr,
2461         .ndo_set_mac_address    = eth_mac_addr,
2462         .ndo_fdb_add            = vxlan_fdb_add,
2463         .ndo_fdb_del            = vxlan_fdb_delete,
2464         .ndo_fdb_dump           = vxlan_fdb_dump,
2465         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2466 };
2467
2468 static const struct net_device_ops vxlan_netdev_raw_ops = {
2469         .ndo_init               = vxlan_init,
2470         .ndo_uninit             = vxlan_uninit,
2471         .ndo_open               = vxlan_open,
2472         .ndo_stop               = vxlan_stop,
2473         .ndo_start_xmit         = vxlan_xmit,
2474         .ndo_get_stats64        = ip_tunnel_get_stats64,
2475         .ndo_change_mtu         = vxlan_change_mtu,
2476         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2477 };
2478
2479 /* Info for udev, that this is a virtual tunnel endpoint */
2480 static struct device_type vxlan_type = {
2481         .name = "vxlan",
2482 };
2483
2484 /* Calls the ndo_udp_tunnel_add of the caller in order to
2485  * supply the listening VXLAN udp ports. Callers are expected
2486  * to implement the ndo_udp_tunnel_add.
2487  */
2488 static void vxlan_push_rx_ports(struct net_device *dev)
2489 {
2490         struct vxlan_sock *vs;
2491         struct net *net = dev_net(dev);
2492         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2493         unsigned int i;
2494
2495         spin_lock(&vn->sock_lock);
2496         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2497                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2498                         udp_tunnel_push_rx_port(dev, vs->sock,
2499                                                 (vs->flags & VXLAN_F_GPE) ?
2500                                                 UDP_TUNNEL_TYPE_VXLAN_GPE :
2501                                                 UDP_TUNNEL_TYPE_VXLAN);
2502         }
2503         spin_unlock(&vn->sock_lock);
2504 }
2505
2506 /* Initialize the device structure. */
2507 static void vxlan_setup(struct net_device *dev)
2508 {
2509         struct vxlan_dev *vxlan = netdev_priv(dev);
2510         unsigned int h;
2511
2512         eth_hw_addr_random(dev);
2513         ether_setup(dev);
2514
2515         dev->destructor = free_netdev;
2516         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2517
2518         dev->features   |= NETIF_F_LLTX;
2519         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2520         dev->features   |= NETIF_F_RXCSUM;
2521         dev->features   |= NETIF_F_GSO_SOFTWARE;
2522
2523         dev->vlan_features = dev->features;
2524         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2525         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2526         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2527         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2528         netif_keep_dst(dev);
2529         dev->priv_flags |= IFF_NO_QUEUE;
2530
2531         INIT_LIST_HEAD(&vxlan->next);
2532         spin_lock_init(&vxlan->hash_lock);
2533
2534         init_timer_deferrable(&vxlan->age_timer);
2535         vxlan->age_timer.function = vxlan_cleanup;
2536         vxlan->age_timer.data = (unsigned long) vxlan;
2537
2538         vxlan->cfg.dst_port = htons(vxlan_port);
2539
2540         vxlan->dev = dev;
2541
2542         gro_cells_init(&vxlan->gro_cells, dev);
2543
2544         for (h = 0; h < FDB_HASH_SIZE; ++h)
2545                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2546 }
2547
2548 static void vxlan_ether_setup(struct net_device *dev)
2549 {
2550         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2551         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2552         dev->netdev_ops = &vxlan_netdev_ether_ops;
2553 }
2554
2555 static void vxlan_raw_setup(struct net_device *dev)
2556 {
2557         dev->header_ops = NULL;
2558         dev->type = ARPHRD_NONE;
2559         dev->hard_header_len = 0;
2560         dev->addr_len = 0;
2561         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2562         dev->netdev_ops = &vxlan_netdev_raw_ops;
2563 }
2564
2565 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2566         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2567         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2568         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2569         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2570         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2571         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2572         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2573         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2574         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2575         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2576         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2577         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2578         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2579         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2580         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2581         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2582         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2583         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2584         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2585         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2586         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2587         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2588         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2589         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2590         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2591         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2592         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2593 };
2594
2595 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2596 {
2597         if (tb[IFLA_ADDRESS]) {
2598                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2599                         pr_debug("invalid link address (not ethernet)\n");
2600                         return -EINVAL;
2601                 }
2602
2603                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2604                         pr_debug("invalid all zero ethernet address\n");
2605                         return -EADDRNOTAVAIL;
2606                 }
2607         }
2608
2609         if (!data)
2610                 return -EINVAL;
2611
2612         if (data[IFLA_VXLAN_ID]) {
2613                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2614                 if (id >= VXLAN_VID_MASK)
2615                         return -ERANGE;
2616         }
2617
2618         if (data[IFLA_VXLAN_PORT_RANGE]) {
2619                 const struct ifla_vxlan_port_range *p
2620                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2621
2622                 if (ntohs(p->high) < ntohs(p->low)) {
2623                         pr_debug("port range %u .. %u not valid\n",
2624                                  ntohs(p->low), ntohs(p->high));
2625                         return -EINVAL;
2626                 }
2627         }
2628
2629         return 0;
2630 }
2631
2632 static void vxlan_get_drvinfo(struct net_device *netdev,
2633                               struct ethtool_drvinfo *drvinfo)
2634 {
2635         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2636         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2637 }
2638
2639 static const struct ethtool_ops vxlan_ethtool_ops = {
2640         .get_drvinfo    = vxlan_get_drvinfo,
2641         .get_link       = ethtool_op_get_link,
2642 };
2643
2644 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2645                                         __be16 port, u32 flags)
2646 {
2647         struct socket *sock;
2648         struct udp_port_cfg udp_conf;
2649         int err;
2650
2651         memset(&udp_conf, 0, sizeof(udp_conf));
2652
2653         if (ipv6) {
2654                 udp_conf.family = AF_INET6;
2655                 udp_conf.use_udp6_rx_checksums =
2656                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2657                 udp_conf.ipv6_v6only = 1;
2658         } else {
2659                 udp_conf.family = AF_INET;
2660         }
2661
2662         udp_conf.local_udp_port = port;
2663
2664         /* Open UDP socket */
2665         err = udp_sock_create(net, &udp_conf, &sock);
2666         if (err < 0)
2667                 return ERR_PTR(err);
2668
2669         return sock;
2670 }
2671
2672 /* Create new listen socket if needed */
2673 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2674                                               __be16 port, u32 flags)
2675 {
2676         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2677         struct vxlan_sock *vs;
2678         struct socket *sock;
2679         unsigned int h;
2680         struct udp_tunnel_sock_cfg tunnel_cfg;
2681
2682         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2683         if (!vs)
2684                 return ERR_PTR(-ENOMEM);
2685
2686         for (h = 0; h < VNI_HASH_SIZE; ++h)
2687                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2688
2689         sock = vxlan_create_sock(net, ipv6, port, flags);
2690         if (IS_ERR(sock)) {
2691                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2692                         PTR_ERR(sock));
2693                 kfree(vs);
2694                 return ERR_CAST(sock);
2695         }
2696
2697         vs->sock = sock;
2698         atomic_set(&vs->refcnt, 1);
2699         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2700
2701         spin_lock(&vn->sock_lock);
2702         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2703         udp_tunnel_notify_add_rx_port(sock,
2704                                       (vs->flags & VXLAN_F_GPE) ?
2705                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2706                                       UDP_TUNNEL_TYPE_VXLAN);
2707         spin_unlock(&vn->sock_lock);
2708
2709         /* Mark socket as an encapsulation socket. */
2710         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2711         tunnel_cfg.sk_user_data = vs;
2712         tunnel_cfg.encap_type = 1;
2713         tunnel_cfg.encap_rcv = vxlan_rcv;
2714         tunnel_cfg.encap_destroy = NULL;
2715         tunnel_cfg.gro_receive = vxlan_gro_receive;
2716         tunnel_cfg.gro_complete = vxlan_gro_complete;
2717
2718         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2719
2720         return vs;
2721 }
2722
2723 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2724 {
2725         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2726         struct vxlan_sock *vs = NULL;
2727
2728         if (!vxlan->cfg.no_share) {
2729                 spin_lock(&vn->sock_lock);
2730                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2731                                      vxlan->cfg.dst_port, vxlan->flags);
2732                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2733                         spin_unlock(&vn->sock_lock);
2734                         return -EBUSY;
2735                 }
2736                 spin_unlock(&vn->sock_lock);
2737         }
2738         if (!vs)
2739                 vs = vxlan_socket_create(vxlan->net, ipv6,
2740                                          vxlan->cfg.dst_port, vxlan->flags);
2741         if (IS_ERR(vs))
2742                 return PTR_ERR(vs);
2743 #if IS_ENABLED(CONFIG_IPV6)
2744         if (ipv6)
2745                 vxlan->vn6_sock = vs;
2746         else
2747 #endif
2748                 vxlan->vn4_sock = vs;
2749         vxlan_vs_add_dev(vs, vxlan);
2750         return 0;
2751 }
2752
2753 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2754 {
2755         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2756         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2757         int ret = 0;
2758
2759         vxlan->vn4_sock = NULL;
2760 #if IS_ENABLED(CONFIG_IPV6)
2761         vxlan->vn6_sock = NULL;
2762         if (ipv6 || metadata)
2763                 ret = __vxlan_sock_add(vxlan, true);
2764 #endif
2765         if (!ret && (!ipv6 || metadata))
2766                 ret = __vxlan_sock_add(vxlan, false);
2767         if (ret < 0)
2768                 vxlan_sock_release(vxlan);
2769         return ret;
2770 }
2771
2772 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2773                                struct vxlan_config *conf)
2774 {
2775         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2776         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2777         struct vxlan_rdst *dst = &vxlan->default_dst;
2778         unsigned short needed_headroom = ETH_HLEN;
2779         int err;
2780         bool use_ipv6 = false;
2781         __be16 default_port = vxlan->cfg.dst_port;
2782         struct net_device *lowerdev = NULL;
2783
2784         if (conf->flags & VXLAN_F_GPE) {
2785                 /* For now, allow GPE only together with COLLECT_METADATA.
2786                  * This can be relaxed later; in such case, the other side
2787                  * of the PtP link will have to be provided.
2788                  */
2789                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2790                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2791                         pr_info("unsupported combination of extensions\n");
2792                         return -EINVAL;
2793                 }
2794
2795                 vxlan_raw_setup(dev);
2796         } else {
2797                 vxlan_ether_setup(dev);
2798         }
2799
2800         vxlan->net = src_net;
2801
2802         dst->remote_vni = conf->vni;
2803
2804         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2805
2806         /* Unless IPv6 is explicitly requested, assume IPv4 */
2807         if (!dst->remote_ip.sa.sa_family)
2808                 dst->remote_ip.sa.sa_family = AF_INET;
2809
2810         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2811             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2812                 if (!IS_ENABLED(CONFIG_IPV6))
2813                         return -EPFNOSUPPORT;
2814                 use_ipv6 = true;
2815                 vxlan->flags |= VXLAN_F_IPV6;
2816         }
2817
2818         if (conf->label && !use_ipv6) {
2819                 pr_info("label only supported in use with IPv6\n");
2820                 return -EINVAL;
2821         }
2822
2823         if (conf->remote_ifindex) {
2824                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2825                 dst->remote_ifindex = conf->remote_ifindex;
2826
2827                 if (!lowerdev) {
2828                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2829                         return -ENODEV;
2830                 }
2831
2832 #if IS_ENABLED(CONFIG_IPV6)
2833                 if (use_ipv6) {
2834                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2835                         if (idev && idev->cnf.disable_ipv6) {
2836                                 pr_info("IPv6 is disabled via sysctl\n");
2837                                 return -EPERM;
2838                         }
2839                 }
2840 #endif
2841
2842                 if (!conf->mtu)
2843                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2844
2845                 needed_headroom = lowerdev->hard_header_len;
2846         } else if (vxlan_addr_multicast(&dst->remote_ip)) {
2847                 pr_info("multicast destination requires interface to be specified\n");
2848                 return -EINVAL;
2849         }
2850
2851         if (conf->mtu) {
2852                 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2853                 if (err)
2854                         return err;
2855         }
2856
2857         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2858                 needed_headroom += VXLAN6_HEADROOM;
2859         else
2860                 needed_headroom += VXLAN_HEADROOM;
2861         dev->needed_headroom = needed_headroom;
2862
2863         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2864         if (!vxlan->cfg.dst_port) {
2865                 if (conf->flags & VXLAN_F_GPE)
2866                         vxlan->cfg.dst_port = 4790; /* IANA assigned VXLAN-GPE port */
2867                 else
2868                         vxlan->cfg.dst_port = default_port;
2869         }
2870         vxlan->flags |= conf->flags;
2871
2872         if (!vxlan->cfg.age_interval)
2873                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2874
2875         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2876                 if (tmp->cfg.vni == conf->vni &&
2877                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2878                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2879                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2880                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2881                     (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
2882                         pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
2883                         return -EEXIST;
2884                 }
2885         }
2886
2887         dev->ethtool_ops = &vxlan_ethtool_ops;
2888
2889         /* create an fdb entry for a valid default destination */
2890         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2891                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2892                                        &vxlan->default_dst.remote_ip,
2893                                        NUD_REACHABLE|NUD_PERMANENT,
2894                                        NLM_F_EXCL|NLM_F_CREATE,
2895                                        vxlan->cfg.dst_port,
2896                                        vxlan->default_dst.remote_vni,
2897                                        vxlan->default_dst.remote_ifindex,
2898                                        NTF_SELF);
2899                 if (err)
2900                         return err;
2901         }
2902
2903         err = register_netdevice(dev);
2904         if (err) {
2905                 vxlan_fdb_delete_default(vxlan);
2906                 return err;
2907         }
2908
2909         list_add(&vxlan->next, &vn->vxlan_list);
2910
2911         return 0;
2912 }
2913
2914 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2915                          struct nlattr *tb[], struct nlattr *data[])
2916 {
2917         struct vxlan_config conf;
2918
2919         memset(&conf, 0, sizeof(conf));
2920
2921         if (data[IFLA_VXLAN_ID])
2922                 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2923
2924         if (data[IFLA_VXLAN_GROUP]) {
2925                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2926         } else if (data[IFLA_VXLAN_GROUP6]) {
2927                 if (!IS_ENABLED(CONFIG_IPV6))
2928                         return -EPFNOSUPPORT;
2929
2930                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2931                 conf.remote_ip.sa.sa_family = AF_INET6;
2932         }
2933
2934         if (data[IFLA_VXLAN_LOCAL]) {
2935                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2936                 conf.saddr.sa.sa_family = AF_INET;
2937         } else if (data[IFLA_VXLAN_LOCAL6]) {
2938                 if (!IS_ENABLED(CONFIG_IPV6))
2939                         return -EPFNOSUPPORT;
2940
2941                 /* TODO: respect scope id */
2942                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2943                 conf.saddr.sa.sa_family = AF_INET6;
2944         }
2945
2946         if (data[IFLA_VXLAN_LINK])
2947                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2948
2949         if (data[IFLA_VXLAN_TOS])
2950                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2951
2952         if (data[IFLA_VXLAN_TTL])
2953                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2954
2955         if (data[IFLA_VXLAN_LABEL])
2956                 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
2957                              IPV6_FLOWLABEL_MASK;
2958
2959         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2960                 conf.flags |= VXLAN_F_LEARN;
2961
2962         if (data[IFLA_VXLAN_AGEING])
2963                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2964
2965         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2966                 conf.flags |= VXLAN_F_PROXY;
2967
2968         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2969                 conf.flags |= VXLAN_F_RSC;
2970
2971         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2972                 conf.flags |= VXLAN_F_L2MISS;
2973
2974         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2975                 conf.flags |= VXLAN_F_L3MISS;
2976
2977         if (data[IFLA_VXLAN_LIMIT])
2978                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2979
2980         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2981             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2982                 conf.flags |= VXLAN_F_COLLECT_METADATA;
2983
2984         if (data[IFLA_VXLAN_PORT_RANGE]) {
2985                 const struct ifla_vxlan_port_range *p
2986                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2987                 conf.port_min = ntohs(p->low);
2988                 conf.port_max = ntohs(p->high);
2989         }
2990
2991         if (data[IFLA_VXLAN_PORT])
2992                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2993
2994         if (data[IFLA_VXLAN_UDP_CSUM] &&
2995             !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2996                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
2997
2998         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2999             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3000                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3001
3002         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3003             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3004                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3005
3006         if (data[IFLA_VXLAN_REMCSUM_TX] &&
3007             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3008                 conf.flags |= VXLAN_F_REMCSUM_TX;
3009
3010         if (data[IFLA_VXLAN_REMCSUM_RX] &&
3011             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3012                 conf.flags |= VXLAN_F_REMCSUM_RX;
3013
3014         if (data[IFLA_VXLAN_GBP])
3015                 conf.flags |= VXLAN_F_GBP;
3016
3017         if (data[IFLA_VXLAN_GPE])
3018                 conf.flags |= VXLAN_F_GPE;
3019
3020         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3021                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3022
3023         if (tb[IFLA_MTU])
3024                 conf.mtu = nla_get_u32(tb[IFLA_MTU]);
3025
3026         return vxlan_dev_configure(src_net, dev, &conf);
3027 }
3028
3029 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3030 {
3031         struct vxlan_dev *vxlan = netdev_priv(dev);
3032         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3033
3034         spin_lock(&vn->sock_lock);
3035         if (!hlist_unhashed(&vxlan->hlist))
3036                 hlist_del_rcu(&vxlan->hlist);
3037         spin_unlock(&vn->sock_lock);
3038
3039         gro_cells_destroy(&vxlan->gro_cells);
3040         list_del(&vxlan->next);
3041         unregister_netdevice_queue(dev, head);
3042 }
3043
3044 static size_t vxlan_get_size(const struct net_device *dev)
3045 {
3046
3047         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3048                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3049                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3050                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3051                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3052                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3053                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3054                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3055                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3056                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3057                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3058                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3059                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3060                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3061                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3062                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3063                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3064                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3065                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3066                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3067                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3068                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3069                 0;
3070 }
3071
3072 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3073 {
3074         const struct vxlan_dev *vxlan = netdev_priv(dev);
3075         const struct vxlan_rdst *dst = &vxlan->default_dst;
3076         struct ifla_vxlan_port_range ports = {
3077                 .low =  htons(vxlan->cfg.port_min),
3078                 .high = htons(vxlan->cfg.port_max),
3079         };
3080
3081         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3082                 goto nla_put_failure;
3083
3084         if (!vxlan_addr_any(&dst->remote_ip)) {
3085                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3086                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3087                                             dst->remote_ip.sin.sin_addr.s_addr))
3088                                 goto nla_put_failure;
3089 #if IS_ENABLED(CONFIG_IPV6)
3090                 } else {
3091                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3092                                              &dst->remote_ip.sin6.sin6_addr))
3093                                 goto nla_put_failure;
3094 #endif
3095                 }
3096         }
3097
3098         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3099                 goto nla_put_failure;
3100
3101         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3102                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3103                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3104                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3105                                 goto nla_put_failure;
3106 #if IS_ENABLED(CONFIG_IPV6)
3107                 } else {
3108                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3109                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3110                                 goto nla_put_failure;
3111 #endif
3112                 }
3113         }
3114
3115         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3116             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3117             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3118             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3119                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3120             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3121                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3122             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3123             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3124                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3125             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3126                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3127             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3128                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3129             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3130             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3131             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3132             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3133                         !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3134             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3135                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3136             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3137                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3138             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3139                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3140             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3141                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3142                 goto nla_put_failure;
3143
3144         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3145                 goto nla_put_failure;
3146
3147         if (vxlan->flags & VXLAN_F_GBP &&
3148             nla_put_flag(skb, IFLA_VXLAN_GBP))
3149                 goto nla_put_failure;
3150
3151         if (vxlan->flags & VXLAN_F_GPE &&
3152             nla_put_flag(skb, IFLA_VXLAN_GPE))
3153                 goto nla_put_failure;
3154
3155         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3156             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3157                 goto nla_put_failure;
3158
3159         return 0;
3160
3161 nla_put_failure:
3162         return -EMSGSIZE;
3163 }
3164
3165 static struct net *vxlan_get_link_net(const struct net_device *dev)
3166 {
3167         struct vxlan_dev *vxlan = netdev_priv(dev);
3168
3169         return vxlan->net;
3170 }
3171
3172 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3173         .kind           = "vxlan",
3174         .maxtype        = IFLA_VXLAN_MAX,
3175         .policy         = vxlan_policy,
3176         .priv_size      = sizeof(struct vxlan_dev),
3177         .setup          = vxlan_setup,
3178         .validate       = vxlan_validate,
3179         .newlink        = vxlan_newlink,
3180         .dellink        = vxlan_dellink,
3181         .get_size       = vxlan_get_size,
3182         .fill_info      = vxlan_fill_info,
3183         .get_link_net   = vxlan_get_link_net,
3184 };
3185
3186 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3187                                     u8 name_assign_type,
3188                                     struct vxlan_config *conf)
3189 {
3190         struct nlattr *tb[IFLA_MAX + 1];
3191         struct net_device *dev;
3192         int err;
3193
3194         memset(&tb, 0, sizeof(tb));
3195
3196         dev = rtnl_create_link(net, name, name_assign_type,
3197                                &vxlan_link_ops, tb);
3198         if (IS_ERR(dev))
3199                 return dev;
3200
3201         err = vxlan_dev_configure(net, dev, conf);
3202         if (err < 0) {
3203                 free_netdev(dev);
3204                 return ERR_PTR(err);
3205         }
3206
3207         err = rtnl_configure_link(dev, NULL);
3208         if (err < 0) {
3209                 LIST_HEAD(list_kill);
3210
3211                 vxlan_dellink(dev, &list_kill);
3212                 unregister_netdevice_many(&list_kill);
3213                 return ERR_PTR(err);
3214         }
3215
3216         return dev;
3217 }
3218 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3219
3220 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3221                                              struct net_device *dev)
3222 {
3223         struct vxlan_dev *vxlan, *next;
3224         LIST_HEAD(list_kill);
3225
3226         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3227                 struct vxlan_rdst *dst = &vxlan->default_dst;
3228
3229                 /* In case we created vxlan device with carrier
3230                  * and we loose the carrier due to module unload
3231                  * we also need to remove vxlan device. In other
3232                  * cases, it's not necessary and remote_ifindex
3233                  * is 0 here, so no matches.
3234                  */
3235                 if (dst->remote_ifindex == dev->ifindex)
3236                         vxlan_dellink(vxlan->dev, &list_kill);
3237         }
3238
3239         unregister_netdevice_many(&list_kill);
3240 }
3241
3242 static int vxlan_netdevice_event(struct notifier_block *unused,
3243                                  unsigned long event, void *ptr)
3244 {
3245         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3246         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3247
3248         if (event == NETDEV_UNREGISTER)
3249                 vxlan_handle_lowerdev_unregister(vn, dev);
3250         else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3251                 vxlan_push_rx_ports(dev);
3252
3253         return NOTIFY_DONE;
3254 }
3255
3256 static struct notifier_block vxlan_notifier_block __read_mostly = {
3257         .notifier_call = vxlan_netdevice_event,
3258 };
3259
3260 static __net_init int vxlan_init_net(struct net *net)
3261 {
3262         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3263         unsigned int h;
3264
3265         INIT_LIST_HEAD(&vn->vxlan_list);
3266         spin_lock_init(&vn->sock_lock);
3267
3268         for (h = 0; h < PORT_HASH_SIZE; ++h)
3269                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3270
3271         return 0;
3272 }
3273
3274 static void __net_exit vxlan_exit_net(struct net *net)
3275 {
3276         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3277         struct vxlan_dev *vxlan, *next;
3278         struct net_device *dev, *aux;
3279         LIST_HEAD(list);
3280
3281         rtnl_lock();
3282         for_each_netdev_safe(net, dev, aux)
3283                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3284                         unregister_netdevice_queue(dev, &list);
3285
3286         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3287                 /* If vxlan->dev is in the same netns, it has already been added
3288                  * to the list by the previous loop.
3289                  */
3290                 if (!net_eq(dev_net(vxlan->dev), net)) {
3291                         gro_cells_destroy(&vxlan->gro_cells);
3292                         unregister_netdevice_queue(vxlan->dev, &list);
3293                 }
3294         }
3295
3296         unregister_netdevice_many(&list);
3297         rtnl_unlock();
3298 }
3299
3300 static struct pernet_operations vxlan_net_ops = {
3301         .init = vxlan_init_net,
3302         .exit = vxlan_exit_net,
3303         .id   = &vxlan_net_id,
3304         .size = sizeof(struct vxlan_net),
3305 };
3306
3307 static int __init vxlan_init_module(void)
3308 {
3309         int rc;
3310
3311         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3312
3313         rc = register_pernet_subsys(&vxlan_net_ops);
3314         if (rc)
3315                 goto out1;
3316
3317         rc = register_netdevice_notifier(&vxlan_notifier_block);
3318         if (rc)
3319                 goto out2;
3320
3321         rc = rtnl_link_register(&vxlan_link_ops);
3322         if (rc)
3323                 goto out3;
3324
3325         return 0;
3326 out3:
3327         unregister_netdevice_notifier(&vxlan_notifier_block);
3328 out2:
3329         unregister_pernet_subsys(&vxlan_net_ops);
3330 out1:
3331         return rc;
3332 }
3333 late_initcall(vxlan_init_module);
3334
3335 static void __exit vxlan_cleanup_module(void)
3336 {
3337         rtnl_link_unregister(&vxlan_link_ops);
3338         unregister_netdevice_notifier(&vxlan_notifier_block);
3339         unregister_pernet_subsys(&vxlan_net_ops);
3340         /* rcu_barrier() is called by netns */
3341 }
3342 module_exit(vxlan_cleanup_module);
3343
3344 MODULE_LICENSE("GPL");
3345 MODULE_VERSION(VXLAN_VERSION);
3346 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3347 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3348 MODULE_ALIAS_RTNL_LINK("vxlan");