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/tun_proto.h>
  30 #include <net/vxlan.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 unsigned 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 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
  64
  65 /* per-network namespace private data for this module */
  66 struct vxlan_net {
  67         struct list_head  vxlan_list;
  68         struct hlist_head sock_list[PORT_HASH_SIZE];
  69         spinlock_t        sock_lock;
  70 };
  71
  72 /* Forwarding table entry */
  73 struct vxlan_fdb {
  74         struct hlist_node hlist;        /* linked list of entries */
  75         struct rcu_head   rcu;
  76         unsigned long     updated;      /* jiffies */
  77         unsigned long     used;
  78         struct list_head  remotes;
  79         u8                eth_addr[ETH_ALEN];
  80         u16               state;        /* see ndm_state */
  81         __be32            vni;
  82         u8                flags;        /* see ndm_flags */
  83 };
  84
  85 /* salt for hash table */
  86 static u32 vxlan_salt __read_mostly;
  87
  88 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  89 {
  90         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  91                ip_tunnel_collect_metadata();
  92 }
  93
  94 #if IS_ENABLED(CONFIG_IPV6)
  95 static inline
  96 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
  97 {
  98         if (a->sa.sa_family != b->sa.sa_family)
  99                 return false;
 100         if (a->sa.sa_family == AF_INET6)
 101                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 102         else
 103                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 104 }
 105
 106 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 107 {
 108         if (ipa->sa.sa_family == AF_INET6)
 109                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 110         else
 111                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 112 }
 113
 114 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 115 {
 116         if (ipa->sa.sa_family == AF_INET6)
 117                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 118         else
 119                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 120 }
 121
 122 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 123 {
 124         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 125                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 126                 ip->sa.sa_family = AF_INET6;
 127                 return 0;
 128         } else if (nla_len(nla) >= sizeof(__be32)) {
 129                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 130                 ip->sa.sa_family = AF_INET;
 131                 return 0;
 132         } else {
 133                 return -EAFNOSUPPORT;
 134         }
 135 }
 136
 137 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 138                               const union vxlan_addr *ip)
 139 {
 140         if (ip->sa.sa_family == AF_INET6)
 141                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 142         else
 143                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 144 }
 145
 146 #else /* !CONFIG_IPV6 */
 147
 148 static inline
 149 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 150 {
 151         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 152 }
 153
 154 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 155 {
 156         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 157 }
 158
 159 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 160 {
 161         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 162 }
 163
 164 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 165 {
 166         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 167                 return -EAFNOSUPPORT;
 168         } else if (nla_len(nla) >= sizeof(__be32)) {
 169                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 170                 ip->sa.sa_family = AF_INET;
 171                 return 0;
 172         } else {
 173                 return -EAFNOSUPPORT;
 174         }
 175 }
 176
 177 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 178                               const union vxlan_addr *ip)
 179 {
 180         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 181 }
 182 #endif
 183
 184 /* Virtual Network hash table head */
 185 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 186 {
 187         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 188 }
 189
 190 /* Socket hash table head */
 191 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 192 {
 193         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 194
 195         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 196 }
 197
 198 /* First remote destination for a forwarding entry.
 199  * Guaranteed to be non-NULL because remotes are never deleted.
 200  */
 201 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 202 {
 203         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 204 }
 205
 206 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 207 {
 208         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 209 }
 210
 211 /* Find VXLAN socket based on network namespace, address family and UDP port
 212  * and enabled unshareable flags.
 213  */
 214 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 215                                           __be16 port, u32 flags)
 216 {
 217         struct vxlan_sock *vs;
 218
 219         flags &= VXLAN_F_RCV_FLAGS;
 220
 221         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 222                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 223                     vxlan_get_sk_family(vs) == family &&
 224                     vs->flags == flags)
 225                         return vs;
 226         }
 227         return NULL;
 228 }
 229
 230 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, int ifindex,
 231                                            __be32 vni)
 232 {
 233         struct vxlan_dev_node *node;
 234
 235         /* For flow based devices, map all packets to VNI 0 */
 236         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 237                 vni = 0;
 238
 239         hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
 240                 if (node->vxlan->default_dst.remote_vni != vni)
 241                         continue;
 242
 243                 if (IS_ENABLED(CONFIG_IPV6)) {
 244                         const struct vxlan_config *cfg = &node->vxlan->cfg;
 245
 246                         if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) &&
 247                             cfg->remote_ifindex != ifindex)
 248                                 continue;
 249                 }
 250
 251                 return node->vxlan;
 252         }
 253
 254         return NULL;
 255 }
 256
 257 /* Look up VNI in a per net namespace table */
 258 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex,
 259                                         __be32 vni, sa_family_t family,
 260                                         __be16 port, u32 flags)
 261 {
 262         struct vxlan_sock *vs;
 263
 264         vs = vxlan_find_sock(net, family, port, flags);
 265         if (!vs)
 266                 return NULL;
 267
 268         return vxlan_vs_find_vni(vs, ifindex, vni);
 269 }
 270
 271 /* Fill in neighbour message in skbuff. */
 272 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 273                           const struct vxlan_fdb *fdb,
 274                           u32 portid, u32 seq, int type, unsigned int flags,
 275                           const struct vxlan_rdst *rdst)
 276 {
 277         unsigned long now = jiffies;
 278         struct nda_cacheinfo ci;
 279         struct nlmsghdr *nlh;
 280         struct ndmsg *ndm;
 281         bool send_ip, send_eth;
 282
 283         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 284         if (nlh == NULL)
 285                 return -EMSGSIZE;
 286
 287         ndm = nlmsg_data(nlh);
 288         memset(ndm, 0, sizeof(*ndm));
 289
 290         send_eth = send_ip = true;
 291
 292         if (type == RTM_GETNEIGH) {
 293                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 294                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 295                 ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
 296         } else
 297                 ndm->ndm_family = AF_BRIDGE;
 298         ndm->ndm_state = fdb->state;
 299         ndm->ndm_ifindex = vxlan->dev->ifindex;
 300         ndm->ndm_flags = fdb->flags;
 301         ndm->ndm_type = RTN_UNICAST;
 302
 303         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 304             nla_put_s32(skb, NDA_LINK_NETNSID,
 305                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 306                 goto nla_put_failure;
 307
 308         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 309                 goto nla_put_failure;
 310
 311         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 312                 goto nla_put_failure;
 313
 314         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 315             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 316                 goto nla_put_failure;
 317         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 318             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 319                 goto nla_put_failure;
 320         if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
 321             nla_put_u32(skb, NDA_SRC_VNI,
 322                         be32_to_cpu(fdb->vni)))
 323                 goto nla_put_failure;
 324         if (rdst->remote_ifindex &&
 325             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 326                 goto nla_put_failure;
 327
 328         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 329         ci.ndm_confirmed = 0;
 330         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 331         ci.ndm_refcnt    = 0;
 332
 333         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 334                 goto nla_put_failure;
 335
 336         nlmsg_end(skb, nlh);
 337         return 0;
 338
 339 nla_put_failure:
 340         nlmsg_cancel(skb, nlh);
 341         return -EMSGSIZE;
 342 }
 343
 344 static inline size_t vxlan_nlmsg_size(void)
 345 {
 346         return NLMSG_ALIGN(sizeof(struct ndmsg))
 347                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 348                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 349                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 350                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 351                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 352                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 353                 + nla_total_size(sizeof(struct nda_cacheinfo));
 354 }
 355
 356 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 357                              struct vxlan_rdst *rd, int type)
 358 {
 359         struct net *net = dev_net(vxlan->dev);
 360         struct sk_buff *skb;
 361         int err = -ENOBUFS;
 362
 363         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 364         if (skb == NULL)
 365                 goto errout;
 366
 367         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 368         if (err < 0) {
 369                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 370                 WARN_ON(err == -EMSGSIZE);
 371                 kfree_skb(skb);
 372                 goto errout;
 373         }
 374
 375         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 376         return;
 377 errout:
 378         if (err < 0)
 379                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 380 }
 381
 382 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 383 {
 384         struct vxlan_dev *vxlan = netdev_priv(dev);
 385         struct vxlan_fdb f = {
 386                 .state = NUD_STALE,
 387         };
 388         struct vxlan_rdst remote = {
 389                 .remote_ip = *ipa, /* goes to NDA_DST */
 390                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 391         };
 392
 393         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 394 }
 395
 396 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 397 {
 398         struct vxlan_fdb f = {
 399                 .state = NUD_STALE,
 400         };
 401         struct vxlan_rdst remote = { };
 402
 403         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 404
 405         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 406 }
 407
 408 /* Hash Ethernet address */
 409 static u32 eth_hash(const unsigned char *addr)
 410 {
 411         u64 value = get_unaligned((u64 *)addr);
 412
 413         /* only want 6 bytes */
 414 #ifdef __BIG_ENDIAN
 415         value >>= 16;
 416 #else
 417         value <<= 16;
 418 #endif
 419         return hash_64(value, FDB_HASH_BITS);
 420 }
 421
 422 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
 423 {
 424         /* use 1 byte of OUI and 3 bytes of NIC */
 425         u32 key = get_unaligned((u32 *)(addr + 2));
 426
 427         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
 428 }
 429
 430 /* Hash chain to use given mac address */
 431 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 432                                                 const u8 *mac, __be32 vni)
 433 {
 434         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
 435                 return &vxlan->fdb_head[eth_vni_hash(mac, vni)];
 436         else
 437                 return &vxlan->fdb_head[eth_hash(mac)];
 438 }
 439
 440 /* Look up Ethernet address in forwarding table */
 441 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 442                                           const u8 *mac, __be32 vni)
 443 {
 444         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 445         struct vxlan_fdb *f;
 446
 447         hlist_for_each_entry_rcu(f, head, hlist) {
 448                 if (ether_addr_equal(mac, f->eth_addr)) {
 449                         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
 450                                 if (vni == f->vni)
 451                                         return f;
 452                         } else {
 453                                 return f;
 454                         }
 455                 }
 456         }
 457
 458         return NULL;
 459 }
 460
 461 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 462                                         const u8 *mac, __be32 vni)
 463 {
 464         struct vxlan_fdb *f;
 465
 466         f = __vxlan_find_mac(vxlan, mac, vni);
 467         if (f)
 468                 f->used = jiffies;
 469
 470         return f;
 471 }
 472
 473 /* caller should hold vxlan->hash_lock */
 474 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 475                                               union vxlan_addr *ip, __be16 port,
 476                                               __be32 vni, __u32 ifindex)
 477 {
 478         struct vxlan_rdst *rd;
 479
 480         list_for_each_entry(rd, &f->remotes, list) {
 481                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 482                     rd->remote_port == port &&
 483                     rd->remote_vni == vni &&
 484                     rd->remote_ifindex == ifindex)
 485                         return rd;
 486         }
 487
 488         return NULL;
 489 }
 490
 491 /* Replace destination of unicast mac */
 492 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 493                              union vxlan_addr *ip, __be16 port, __be32 vni,
 494                              __u32 ifindex)
 495 {
 496         struct vxlan_rdst *rd;
 497
 498         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 499         if (rd)
 500                 return 0;
 501
 502         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 503         if (!rd)
 504                 return 0;
 505
 506         dst_cache_reset(&rd->dst_cache);
 507         rd->remote_ip = *ip;
 508         rd->remote_port = port;
 509         rd->remote_vni = vni;
 510         rd->remote_ifindex = ifindex;
 511         return 1;
 512 }
 513
 514 /* Add/update destinations for multicast */
 515 static int vxlan_fdb_append(struct vxlan_fdb *f,
 516                             union vxlan_addr *ip, __be16 port, __be32 vni,
 517                             __u32 ifindex, struct vxlan_rdst **rdp)
 518 {
 519         struct vxlan_rdst *rd;
 520
 521         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 522         if (rd)
 523                 return 0;
 524
 525         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 526         if (rd == NULL)
 527                 return -ENOBUFS;
 528
 529         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 530                 kfree(rd);
 531                 return -ENOBUFS;
 532         }
 533
 534         rd->remote_ip = *ip;
 535         rd->remote_port = port;
 536         rd->remote_vni = vni;
 537         rd->remote_ifindex = ifindex;
 538
 539         list_add_tail_rcu(&rd->list, &f->remotes);
 540
 541         *rdp = rd;
 542         return 1;
 543 }
 544
 545 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 546                                           unsigned int off,
 547                                           struct vxlanhdr *vh, size_t hdrlen,
 548                                           __be32 vni_field,
 549                                           struct gro_remcsum *grc,
 550                                           bool nopartial)
 551 {
 552         size_t start, offset;
 553
 554         if (skb->remcsum_offload)
 555                 return vh;
 556
 557         if (!NAPI_GRO_CB(skb)->csum_valid)
 558                 return NULL;
 559
 560         start = vxlan_rco_start(vni_field);
 561         offset = start + vxlan_rco_offset(vni_field);
 562
 563         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 564                                      start, offset, grc, nopartial);
 565
 566         skb->remcsum_offload = 1;
 567
 568         return vh;
 569 }
 570
 571 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
 572                                           struct sk_buff **head,
 573                                           struct sk_buff *skb)
 574 {
 575         struct sk_buff *p, **pp = NULL;
 576         struct vxlanhdr *vh, *vh2;
 577         unsigned int hlen, off_vx;
 578         int flush = 1;
 579         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 580         __be32 flags;
 581         struct gro_remcsum grc;
 582
 583         skb_gro_remcsum_init(&grc);
 584
 585         off_vx = skb_gro_offset(skb);
 586         hlen = off_vx + sizeof(*vh);
 587         vh   = skb_gro_header_fast(skb, off_vx);
 588         if (skb_gro_header_hard(skb, hlen)) {
 589                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 590                 if (unlikely(!vh))
 591                         goto out;
 592         }
 593
 594         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 595
 596         flags = vh->vx_flags;
 597
 598         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 599                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 600                                        vh->vx_vni, &grc,
 601                                        !!(vs->flags &
 602                                           VXLAN_F_REMCSUM_NOPARTIAL));
 603
 604                 if (!vh)
 605                         goto out;
 606         }
 607
 608         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 609
 610         for (p = *head; p; p = p->next) {
 611                 if (!NAPI_GRO_CB(p)->same_flow)
 612                         continue;
 613
 614                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 615                 if (vh->vx_flags != vh2->vx_flags ||
 616                     vh->vx_vni != vh2->vx_vni) {
 617                         NAPI_GRO_CB(p)->same_flow = 0;
 618                         continue;
 619                 }
 620         }
 621
 622         pp = call_gro_receive(eth_gro_receive, head, skb);
 623         flush = 0;
 624
 625 out:
 626         skb_gro_remcsum_cleanup(skb, &grc);
 627         skb->remcsum_offload = 0;
 628         NAPI_GRO_CB(skb)->flush |= flush;
 629
 630         return pp;
 631 }
 632
 633 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 634 {
 635         /* Sets 'skb->inner_mac_header' since we are always called with
 636          * 'skb->encapsulation' set.
 637          */
 638         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 639 }
 640
 641 /* Add new entry to forwarding table -- assumes lock held */
 642 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 643                             const u8 *mac, union vxlan_addr *ip,
 644                             __u16 state, __u16 flags,
 645                             __be16 port, __be32 src_vni, __be32 vni,
 646                             __u32 ifindex, __u8 ndm_flags)
 647 {
 648         struct vxlan_rdst *rd = NULL;
 649         struct vxlan_fdb *f;
 650         int notify = 0;
 651         int rc;
 652
 653         f = __vxlan_find_mac(vxlan, mac, src_vni);
 654         if (f) {
 655                 if (flags & NLM_F_EXCL) {
 656                         netdev_dbg(vxlan->dev,
 657                                    "lost race to create %pM\n", mac);
 658                         return -EEXIST;
 659                 }
 660                 if (f->state != state) {
 661                         f->state = state;
 662                         f->updated = jiffies;
 663                         notify = 1;
 664                 }
 665                 if (f->flags != ndm_flags) {
 666                         f->flags = ndm_flags;
 667                         f->updated = jiffies;
 668                         notify = 1;
 669                 }
 670                 if ((flags & NLM_F_REPLACE)) {
 671                         /* Only change unicasts */
 672                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 673                              is_zero_ether_addr(f->eth_addr))) {
 674                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 675                                                            ifindex);
 676                         } else
 677                                 return -EOPNOTSUPP;
 678                 }
 679                 if ((flags & NLM_F_APPEND) &&
 680                     (is_multicast_ether_addr(f->eth_addr) ||
 681                      is_zero_ether_addr(f->eth_addr))) {
 682                         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 683
 684                         if (rc < 0)
 685                                 return rc;
 686                         notify |= rc;
 687                 }
 688         } else {
 689                 if (!(flags & NLM_F_CREATE))
 690                         return -ENOENT;
 691
 692                 if (vxlan->cfg.addrmax &&
 693                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 694                         return -ENOSPC;
 695
 696                 /* Disallow replace to add a multicast entry */
 697                 if ((flags & NLM_F_REPLACE) &&
 698                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 699                         return -EOPNOTSUPP;
 700
 701                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 702                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 703                 if (!f)
 704                         return -ENOMEM;
 705
 706                 notify = 1;
 707                 f->state = state;
 708                 f->flags = ndm_flags;
 709                 f->updated = f->used = jiffies;
 710                 f->vni = src_vni;
 711                 INIT_LIST_HEAD(&f->remotes);
 712                 memcpy(f->eth_addr, mac, ETH_ALEN);
 713
 714                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 715                 if (rc < 0) {
 716                         kfree(f);
 717                         return rc;
 718                 }
 719
 720                 ++vxlan->addrcnt;
 721                 hlist_add_head_rcu(&f->hlist,
 722                                    vxlan_fdb_head(vxlan, mac, src_vni));
 723         }
 724
 725         if (notify) {
 726                 if (rd == NULL)
 727                         rd = first_remote_rtnl(f);
 728                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 729         }
 730
 731         return 0;
 732 }
 733
 734 static void vxlan_fdb_free(struct rcu_head *head)
 735 {
 736         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 737         struct vxlan_rdst *rd, *nd;
 738
 739         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 740                 dst_cache_destroy(&rd->dst_cache);
 741                 kfree(rd);
 742         }
 743         kfree(f);
 744 }
 745
 746 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 747 {
 748         netdev_dbg(vxlan->dev,
 749                     "delete %pM\n", f->eth_addr);
 750
 751         --vxlan->addrcnt;
 752         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 753
 754         hlist_del_rcu(&f->hlist);
 755         call_rcu(&f->rcu, vxlan_fdb_free);
 756 }
 757
 758 static void vxlan_dst_free(struct rcu_head *head)
 759 {
 760         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 761
 762         dst_cache_destroy(&rd->dst_cache);
 763         kfree(rd);
 764 }
 765
 766 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 767                                   struct vxlan_rdst *rd)
 768 {
 769         list_del_rcu(&rd->list);
 770         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 771         call_rcu(&rd->rcu, vxlan_dst_free);
 772 }
 773
 774 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 775                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
 776                            __be32 *vni, u32 *ifindex)
 777 {
 778         struct net *net = dev_net(vxlan->dev);
 779         int err;
 780
 781         if (tb[NDA_DST]) {
 782                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 783                 if (err)
 784                         return err;
 785         } else {
 786                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 787                 if (remote->sa.sa_family == AF_INET) {
 788                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 789                         ip->sa.sa_family = AF_INET;
 790 #if IS_ENABLED(CONFIG_IPV6)
 791                 } else {
 792                         ip->sin6.sin6_addr = in6addr_any;
 793                         ip->sa.sa_family = AF_INET6;
 794 #endif
 795                 }
 796         }
 797
 798         if (tb[NDA_PORT]) {
 799                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 800                         return -EINVAL;
 801                 *port = nla_get_be16(tb[NDA_PORT]);
 802         } else {
 803                 *port = vxlan->cfg.dst_port;
 804         }
 805
 806         if (tb[NDA_VNI]) {
 807                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 808                         return -EINVAL;
 809                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 810         } else {
 811                 *vni = vxlan->default_dst.remote_vni;
 812         }
 813
 814         if (tb[NDA_SRC_VNI]) {
 815                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
 816                         return -EINVAL;
 817                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
 818         } else {
 819                 *src_vni = vxlan->default_dst.remote_vni;
 820         }
 821
 822         if (tb[NDA_IFINDEX]) {
 823                 struct net_device *tdev;
 824
 825                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 826                         return -EINVAL;
 827                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 828                 tdev = __dev_get_by_index(net, *ifindex);
 829                 if (!tdev)
 830                         return -EADDRNOTAVAIL;
 831         } else {
 832                 *ifindex = 0;
 833         }
 834
 835         return 0;
 836 }
 837
 838 /* Add static entry (via netlink) */
 839 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 840                          struct net_device *dev,
 841                          const unsigned char *addr, u16 vid, u16 flags)
 842 {
 843         struct vxlan_dev *vxlan = netdev_priv(dev);
 844         /* struct net *net = dev_net(vxlan->dev); */
 845         union vxlan_addr ip;
 846         __be16 port;
 847         __be32 src_vni, vni;
 848         u32 ifindex;
 849         int err;
 850
 851         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 852                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 853                         ndm->ndm_state);
 854                 return -EINVAL;
 855         }
 856
 857         if (tb[NDA_DST] == NULL)
 858                 return -EINVAL;
 859
 860         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 861         if (err)
 862                 return err;
 863
 864         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 865                 return -EAFNOSUPPORT;
 866
 867         spin_lock_bh(&vxlan->hash_lock);
 868         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 869                                port, src_vni, vni, ifindex, ndm->ndm_flags);
 870         spin_unlock_bh(&vxlan->hash_lock);
 871
 872         return err;
 873 }
 874
 875 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
 876                               const unsigned char *addr, union vxlan_addr ip,
 877                               __be16 port, __be32 src_vni, __be32 vni,
 878                               u32 ifindex, u16 vid)
 879 {
 880         struct vxlan_fdb *f;
 881         struct vxlan_rdst *rd = NULL;
 882         int err = -ENOENT;
 883
 884         f = vxlan_find_mac(vxlan, addr, src_vni);
 885         if (!f)
 886                 return err;
 887
 888         if (!vxlan_addr_any(&ip)) {
 889                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 890                 if (!rd)
 891                         goto out;
 892         }
 893
 894         /* remove a destination if it's not the only one on the list,
 895          * otherwise destroy the fdb entry
 896          */
 897         if (rd && !list_is_singular(&f->remotes)) {
 898                 vxlan_fdb_dst_destroy(vxlan, f, rd);
 899                 goto out;
 900         }
 901
 902         vxlan_fdb_destroy(vxlan, f);
 903
 904 out:
 905         return 0;
 906 }
 907
 908 /* Delete entry (via netlink) */
 909 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 910                             struct net_device *dev,
 911                             const unsigned char *addr, u16 vid)
 912 {
 913         struct vxlan_dev *vxlan = netdev_priv(dev);
 914         union vxlan_addr ip;
 915         __be32 src_vni, vni;
 916         __be16 port;
 917         u32 ifindex;
 918         int err;
 919
 920         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 921         if (err)
 922                 return err;
 923
 924         spin_lock_bh(&vxlan->hash_lock);
 925         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
 926                                  vid);
 927         spin_unlock_bh(&vxlan->hash_lock);
 928
 929         return err;
 930 }
 931
 932 /* Dump forwarding table */
 933 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 934                           struct net_device *dev,
 935                           struct net_device *filter_dev, int *idx)
 936 {
 937         struct vxlan_dev *vxlan = netdev_priv(dev);
 938         unsigned int h;
 939         int err = 0;
 940
 941         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 942                 struct vxlan_fdb *f;
 943
 944                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 945                         struct vxlan_rdst *rd;
 946
 947                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 948                                 if (*idx < cb->args[2])
 949                                         goto skip;
 950
 951                                 err = vxlan_fdb_info(skb, vxlan, f,
 952                                                      NETLINK_CB(cb->skb).portid,
 953                                                      cb->nlh->nlmsg_seq,
 954                                                      RTM_NEWNEIGH,
 955                                                      NLM_F_MULTI, rd);
 956                                 if (err < 0)
 957                                         goto out;
 958 skip:
 959                                 *idx += 1;
 960                         }
 961                 }
 962         }
 963 out:
 964         return err;
 965 }
 966
 967 /* Watch incoming packets to learn mapping between Ethernet address
 968  * and Tunnel endpoint.
 969  * Return true if packet is bogus and should be dropped.
 970  */
 971 static bool vxlan_snoop(struct net_device *dev,
 972                         union vxlan_addr *src_ip, const u8 *src_mac,
 973                         u32 src_ifindex, __be32 vni)
 974 {
 975         struct vxlan_dev *vxlan = netdev_priv(dev);
 976         struct vxlan_fdb *f;
 977         u32 ifindex = 0;
 978
 979 #if IS_ENABLED(CONFIG_IPV6)
 980         if (src_ip->sa.sa_family == AF_INET6 &&
 981             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
 982                 ifindex = src_ifindex;
 983 #endif
 984
 985         f = vxlan_find_mac(vxlan, src_mac, vni);
 986         if (likely(f)) {
 987                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 988
 989                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
 990                            rdst->remote_ifindex == ifindex))
 991                         return false;
 992
 993                 /* Don't migrate static entries, drop packets */
 994                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
 995                         return true;
 996
 997                 if (net_ratelimit())
 998                         netdev_info(dev,
 999                                     "%pM migrated from %pIS to %pIS\n",
1000                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1001
1002                 rdst->remote_ip = *src_ip;
1003                 f->updated = jiffies;
1004                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1005         } else {
1006                 /* learned new entry */
1007                 spin_lock(&vxlan->hash_lock);
1008
1009                 /* close off race between vxlan_flush and incoming packets */
1010                 if (netif_running(dev))
1011                         vxlan_fdb_create(vxlan, src_mac, src_ip,
1012                                          NUD_REACHABLE,
1013                                          NLM_F_EXCL|NLM_F_CREATE,
1014                                          vxlan->cfg.dst_port,
1015                                          vni,
1016                                          vxlan->default_dst.remote_vni,
1017                                          ifindex, NTF_SELF);
1018                 spin_unlock(&vxlan->hash_lock);
1019         }
1020
1021         return false;
1022 }
1023
1024 /* See if multicast group is already in use by other ID */
1025 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1026 {
1027         struct vxlan_dev *vxlan;
1028         struct vxlan_sock *sock4;
1029 #if IS_ENABLED(CONFIG_IPV6)
1030         struct vxlan_sock *sock6;
1031 #endif
1032         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1033
1034         sock4 = rtnl_dereference(dev->vn4_sock);
1035
1036         /* The vxlan_sock is only used by dev, leaving group has
1037          * no effect on other vxlan devices.
1038          */
1039         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1040                 return false;
1041 #if IS_ENABLED(CONFIG_IPV6)
1042         sock6 = rtnl_dereference(dev->vn6_sock);
1043         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1044                 return false;
1045 #endif
1046
1047         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1048                 if (!netif_running(vxlan->dev) || vxlan == dev)
1049                         continue;
1050
1051                 if (family == AF_INET &&
1052                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1053                         continue;
1054 #if IS_ENABLED(CONFIG_IPV6)
1055                 if (family == AF_INET6 &&
1056                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1057                         continue;
1058 #endif
1059
1060                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1061                                       &dev->default_dst.remote_ip))
1062                         continue;
1063
1064                 if (vxlan->default_dst.remote_ifindex !=
1065                     dev->default_dst.remote_ifindex)
1066                         continue;
1067
1068                 return true;
1069         }
1070
1071         return false;
1072 }
1073
1074 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1075 {
1076         struct vxlan_net *vn;
1077
1078         if (!vs)
1079                 return false;
1080         if (!refcount_dec_and_test(&vs->refcnt))
1081                 return false;
1082
1083         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1084         spin_lock(&vn->sock_lock);
1085         hlist_del_rcu(&vs->hlist);
1086         udp_tunnel_notify_del_rx_port(vs->sock,
1087                                       (vs->flags & VXLAN_F_GPE) ?
1088                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1089                                       UDP_TUNNEL_TYPE_VXLAN);
1090         spin_unlock(&vn->sock_lock);
1091
1092         return true;
1093 }
1094
1095 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1096 {
1097         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1098 #if IS_ENABLED(CONFIG_IPV6)
1099         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1100
1101         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1102 #endif
1103
1104         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1105         synchronize_net();
1106
1107         vxlan_vs_del_dev(vxlan);
1108
1109         if (__vxlan_sock_release_prep(sock4)) {
1110                 udp_tunnel_sock_release(sock4->sock);
1111                 kfree(sock4);
1112         }
1113
1114 #if IS_ENABLED(CONFIG_IPV6)
1115         if (__vxlan_sock_release_prep(sock6)) {
1116                 udp_tunnel_sock_release(sock6->sock);
1117                 kfree(sock6);
1118         }
1119 #endif
1120 }
1121
1122 /* Update multicast group membership when first VNI on
1123  * multicast address is brought up
1124  */
1125 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1126 {
1127         struct sock *sk;
1128         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1129         int ifindex = vxlan->default_dst.remote_ifindex;
1130         int ret = -EINVAL;
1131
1132         if (ip->sa.sa_family == AF_INET) {
1133                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1134                 struct ip_mreqn mreq = {
1135                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1136                         .imr_ifindex            = ifindex,
1137                 };
1138
1139                 sk = sock4->sock->sk;
1140                 lock_sock(sk);
1141                 ret = ip_mc_join_group(sk, &mreq);
1142                 release_sock(sk);
1143 #if IS_ENABLED(CONFIG_IPV6)
1144         } else {
1145                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1146
1147                 sk = sock6->sock->sk;
1148                 lock_sock(sk);
1149                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1150                                                    &ip->sin6.sin6_addr);
1151                 release_sock(sk);
1152 #endif
1153         }
1154
1155         return ret;
1156 }
1157
1158 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1159 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1160 {
1161         struct sock *sk;
1162         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1163         int ifindex = vxlan->default_dst.remote_ifindex;
1164         int ret = -EINVAL;
1165
1166         if (ip->sa.sa_family == AF_INET) {
1167                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1168                 struct ip_mreqn mreq = {
1169                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1170                         .imr_ifindex            = ifindex,
1171                 };
1172
1173                 sk = sock4->sock->sk;
1174                 lock_sock(sk);
1175                 ret = ip_mc_leave_group(sk, &mreq);
1176                 release_sock(sk);
1177 #if IS_ENABLED(CONFIG_IPV6)
1178         } else {
1179                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1180
1181                 sk = sock6->sock->sk;
1182                 lock_sock(sk);
1183                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1184                                                    &ip->sin6.sin6_addr);
1185                 release_sock(sk);
1186 #endif
1187         }
1188
1189         return ret;
1190 }
1191
1192 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1193                           struct sk_buff *skb, u32 vxflags)
1194 {
1195         size_t start, offset;
1196
1197         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1198                 goto out;
1199
1200         start = vxlan_rco_start(unparsed->vx_vni);
1201         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1202
1203         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1204                 return false;
1205
1206         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1207                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1208 out:
1209         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1210         unparsed->vx_vni &= VXLAN_VNI_MASK;
1211         return true;
1212 }
1213
1214 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1215                                 struct sk_buff *skb, u32 vxflags,
1216                                 struct vxlan_metadata *md)
1217 {
1218         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1219         struct metadata_dst *tun_dst;
1220
1221         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1222                 goto out;
1223
1224         md->gbp = ntohs(gbp->policy_id);
1225
1226         tun_dst = (struct metadata_dst *)skb_dst(skb);
1227         if (tun_dst) {
1228                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1229                 tun_dst->u.tun_info.options_len = sizeof(*md);
1230         }
1231         if (gbp->dont_learn)
1232                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1233
1234         if (gbp->policy_applied)
1235                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1236
1237         /* In flow-based mode, GBP is carried in dst_metadata */
1238         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1239                 skb->mark = md->gbp;
1240 out:
1241         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1242 }
1243
1244 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1245                                 __be16 *protocol,
1246                                 struct sk_buff *skb, u32 vxflags)
1247 {
1248         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1249
1250         /* Need to have Next Protocol set for interfaces in GPE mode. */
1251         if (!gpe->np_applied)
1252                 return false;
1253         /* "The initial version is 0. If a receiver does not support the
1254          * version indicated it MUST drop the packet.
1255          */
1256         if (gpe->version != 0)
1257                 return false;
1258         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1259          * processing MUST occur." However, we don't implement OAM
1260          * processing, thus drop the packet.
1261          */
1262         if (gpe->oam_flag)
1263                 return false;
1264
1265         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1266         if (!*protocol)
1267                 return false;
1268
1269         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1270         return true;
1271 }
1272
1273 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1274                           struct vxlan_sock *vs,
1275                           struct sk_buff *skb, __be32 vni)
1276 {
1277         union vxlan_addr saddr;
1278         u32 ifindex = skb->dev->ifindex;
1279
1280         skb_reset_mac_header(skb);
1281         skb->protocol = eth_type_trans(skb, vxlan->dev);
1282         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1283
1284         /* Ignore packet loops (and multicast echo) */
1285         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1286                 return false;
1287
1288         /* Get address from the outer IP header */
1289         if (vxlan_get_sk_family(vs) == AF_INET) {
1290                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1291                 saddr.sa.sa_family = AF_INET;
1292 #if IS_ENABLED(CONFIG_IPV6)
1293         } else {
1294                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1295                 saddr.sa.sa_family = AF_INET6;
1296 #endif
1297         }
1298
1299         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1300             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1301                 return false;
1302
1303         return true;
1304 }
1305
1306 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1307                                   struct sk_buff *skb)
1308 {
1309         int err = 0;
1310
1311         if (vxlan_get_sk_family(vs) == AF_INET)
1312                 err = IP_ECN_decapsulate(oiph, skb);
1313 #if IS_ENABLED(CONFIG_IPV6)
1314         else
1315                 err = IP6_ECN_decapsulate(oiph, skb);
1316 #endif
1317
1318         if (unlikely(err) && log_ecn_error) {
1319                 if (vxlan_get_sk_family(vs) == AF_INET)
1320                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1321                                              &((struct iphdr *)oiph)->saddr,
1322                                              ((struct iphdr *)oiph)->tos);
1323                 else
1324                         net_info_ratelimited("non-ECT from %pI6\n",
1325                                              &((struct ipv6hdr *)oiph)->saddr);
1326         }
1327         return err <= 1;
1328 }
1329
1330 /* Callback from net/ipv4/udp.c to receive packets */
1331 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1332 {
1333         struct pcpu_sw_netstats *stats;
1334         struct vxlan_dev *vxlan;
1335         struct vxlan_sock *vs;
1336         struct vxlanhdr unparsed;
1337         struct vxlan_metadata _md;
1338         struct vxlan_metadata *md = &_md;
1339         __be16 protocol = htons(ETH_P_TEB);
1340         bool raw_proto = false;
1341         void *oiph;
1342         __be32 vni = 0;
1343
1344         /* Need UDP and VXLAN header to be present */
1345         if (!pskb_may_pull(skb, VXLAN_HLEN))
1346                 goto drop;
1347
1348         unparsed = *vxlan_hdr(skb);
1349         /* VNI flag always required to be set */
1350         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1351                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1352                            ntohl(vxlan_hdr(skb)->vx_flags),
1353                            ntohl(vxlan_hdr(skb)->vx_vni));
1354                 /* Return non vxlan pkt */
1355                 goto drop;
1356         }
1357         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1358         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1359
1360         vs = rcu_dereference_sk_user_data(sk);
1361         if (!vs)
1362                 goto drop;
1363
1364         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1365
1366         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1367         if (!vxlan)
1368                 goto drop;
1369
1370         /* For backwards compatibility, only allow reserved fields to be
1371          * used by VXLAN extensions if explicitly requested.
1372          */
1373         if (vs->flags & VXLAN_F_GPE) {
1374                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1375                         goto drop;
1376                 raw_proto = true;
1377         }
1378
1379         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1380                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1381                         goto drop;
1382
1383         if (vxlan_collect_metadata(vs)) {
1384                 struct metadata_dst *tun_dst;
1385
1386                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1387                                          key32_to_tunnel_id(vni), sizeof(*md));
1388
1389                 if (!tun_dst)
1390                         goto drop;
1391
1392                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1393
1394                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1395         } else {
1396                 memset(md, 0, sizeof(*md));
1397         }
1398
1399         if (vs->flags & VXLAN_F_REMCSUM_RX)
1400                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1401                         goto drop;
1402         if (vs->flags & VXLAN_F_GBP)
1403                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1404         /* Note that GBP and GPE can never be active together. This is
1405          * ensured in vxlan_dev_configure.
1406          */
1407
1408         if (unparsed.vx_flags || unparsed.vx_vni) {
1409                 /* If there are any unprocessed flags remaining treat
1410                  * this as a malformed packet. This behavior diverges from
1411                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1412                  * in reserved fields are to be ignored. The approach here
1413                  * maintains compatibility with previous stack code, and also
1414                  * is more robust and provides a little more security in
1415                  * adding extensions to VXLAN.
1416                  */
1417                 goto drop;
1418         }
1419
1420         if (!raw_proto) {
1421                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1422                         goto drop;
1423         } else {
1424                 skb_reset_mac_header(skb);
1425                 skb->dev = vxlan->dev;
1426                 skb->pkt_type = PACKET_HOST;
1427         }
1428
1429         oiph = skb_network_header(skb);
1430         skb_reset_network_header(skb);
1431
1432         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1433                 ++vxlan->dev->stats.rx_frame_errors;
1434                 ++vxlan->dev->stats.rx_errors;
1435                 goto drop;
1436         }
1437
1438         stats = this_cpu_ptr(vxlan->dev->tstats);
1439         u64_stats_update_begin(&stats->syncp);
1440         stats->rx_packets++;
1441         stats->rx_bytes += skb->len;
1442         u64_stats_update_end(&stats->syncp);
1443
1444         gro_cells_receive(&vxlan->gro_cells, skb);
1445         return 0;
1446
1447 drop:
1448         /* Consume bad packet */
1449         kfree_skb(skb);
1450         return 0;
1451 }
1452
1453 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1454 {
1455         struct vxlan_dev *vxlan = netdev_priv(dev);
1456         struct arphdr *parp;
1457         u8 *arpptr, *sha;
1458         __be32 sip, tip;
1459         struct neighbour *n;
1460
1461         if (dev->flags & IFF_NOARP)
1462                 goto out;
1463
1464         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1465                 dev->stats.tx_dropped++;
1466                 goto out;
1467         }
1468         parp = arp_hdr(skb);
1469
1470         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1471              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1472             parp->ar_pro != htons(ETH_P_IP) ||
1473             parp->ar_op != htons(ARPOP_REQUEST) ||
1474             parp->ar_hln != dev->addr_len ||
1475             parp->ar_pln != 4)
1476                 goto out;
1477         arpptr = (u8 *)parp + sizeof(struct arphdr);
1478         sha = arpptr;
1479         arpptr += dev->addr_len;        /* sha */
1480         memcpy(&sip, arpptr, sizeof(sip));
1481         arpptr += sizeof(sip);
1482         arpptr += dev->addr_len;        /* tha */
1483         memcpy(&tip, arpptr, sizeof(tip));
1484
1485         if (ipv4_is_loopback(tip) ||
1486             ipv4_is_multicast(tip))
1487                 goto out;
1488
1489         n = neigh_lookup(&arp_tbl, &tip, dev);
1490
1491         if (n) {
1492                 struct vxlan_fdb *f;
1493                 struct sk_buff  *reply;
1494
1495                 if (!(n->nud_state & NUD_CONNECTED)) {
1496                         neigh_release(n);
1497                         goto out;
1498                 }
1499
1500                 f = vxlan_find_mac(vxlan, n->ha, vni);
1501                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1502                         /* bridge-local neighbor */
1503                         neigh_release(n);
1504                         goto out;
1505                 }
1506
1507                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1508                                 n->ha, sha);
1509
1510                 neigh_release(n);
1511
1512                 if (reply == NULL)
1513                         goto out;
1514
1515                 skb_reset_mac_header(reply);
1516                 __skb_pull(reply, skb_network_offset(reply));
1517                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1518                 reply->pkt_type = PACKET_HOST;
1519
1520                 if (netif_rx_ni(reply) == NET_RX_DROP)
1521                         dev->stats.rx_dropped++;
1522         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1523                 union vxlan_addr ipa = {
1524                         .sin.sin_addr.s_addr = tip,
1525                         .sin.sin_family = AF_INET,
1526                 };
1527
1528                 vxlan_ip_miss(dev, &ipa);
1529         }
1530 out:
1531         consume_skb(skb);
1532         return NETDEV_TX_OK;
1533 }
1534
1535 #if IS_ENABLED(CONFIG_IPV6)
1536 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1537         struct neighbour *n, bool isrouter)
1538 {
1539         struct net_device *dev = request->dev;
1540         struct sk_buff *reply;
1541         struct nd_msg *ns, *na;
1542         struct ipv6hdr *pip6;
1543         u8 *daddr;
1544         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1545         int ns_olen;
1546         int i, len;
1547
1548         if (dev == NULL || !pskb_may_pull(request, request->len))
1549                 return NULL;
1550
1551         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1552                 sizeof(*na) + na_olen + dev->needed_tailroom;
1553         reply = alloc_skb(len, GFP_ATOMIC);
1554         if (reply == NULL)
1555                 return NULL;
1556
1557         reply->protocol = htons(ETH_P_IPV6);
1558         reply->dev = dev;
1559         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1560         skb_push(reply, sizeof(struct ethhdr));
1561         skb_reset_mac_header(reply);
1562
1563         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1564
1565         daddr = eth_hdr(request)->h_source;
1566         ns_olen = request->len - skb_network_offset(request) -
1567                 sizeof(struct ipv6hdr) - sizeof(*ns);
1568         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1569                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1570                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1571                         break;
1572                 }
1573         }
1574
1575         /* Ethernet header */
1576         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1577         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1578         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1579         reply->protocol = htons(ETH_P_IPV6);
1580
1581         skb_pull(reply, sizeof(struct ethhdr));
1582         skb_reset_network_header(reply);
1583         skb_put(reply, sizeof(struct ipv6hdr));
1584
1585         /* IPv6 header */
1586
1587         pip6 = ipv6_hdr(reply);
1588         memset(pip6, 0, sizeof(struct ipv6hdr));
1589         pip6->version = 6;
1590         pip6->priority = ipv6_hdr(request)->priority;
1591         pip6->nexthdr = IPPROTO_ICMPV6;
1592         pip6->hop_limit = 255;
1593         pip6->daddr = ipv6_hdr(request)->saddr;
1594         pip6->saddr = *(struct in6_addr *)n->primary_key;
1595
1596         skb_pull(reply, sizeof(struct ipv6hdr));
1597         skb_reset_transport_header(reply);
1598
1599         /* Neighbor Advertisement */
1600         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1601         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1602         na->icmph.icmp6_router = isrouter;
1603         na->icmph.icmp6_override = 1;
1604         na->icmph.icmp6_solicited = 1;
1605         na->target = ns->target;
1606         ether_addr_copy(&na->opt[2], n->ha);
1607         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1608         na->opt[1] = na_olen >> 3;
1609
1610         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1611                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1612                 csum_partial(na, sizeof(*na)+na_olen, 0));
1613
1614         pip6->payload_len = htons(sizeof(*na)+na_olen);
1615
1616         skb_push(reply, sizeof(struct ipv6hdr));
1617
1618         reply->ip_summed = CHECKSUM_UNNECESSARY;
1619
1620         return reply;
1621 }
1622
1623 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1624 {
1625         struct vxlan_dev *vxlan = netdev_priv(dev);
1626         const struct in6_addr *daddr;
1627         const struct ipv6hdr *iphdr;
1628         struct inet6_dev *in6_dev;
1629         struct neighbour *n;
1630         struct nd_msg *msg;
1631
1632         in6_dev = __in6_dev_get(dev);
1633         if (!in6_dev)
1634                 goto out;
1635
1636         iphdr = ipv6_hdr(skb);
1637         daddr = &iphdr->daddr;
1638         msg = (struct nd_msg *)(iphdr + 1);
1639
1640         if (ipv6_addr_loopback(daddr) ||
1641             ipv6_addr_is_multicast(&msg->target))
1642                 goto out;
1643
1644         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1645
1646         if (n) {
1647                 struct vxlan_fdb *f;
1648                 struct sk_buff *reply;
1649
1650                 if (!(n->nud_state & NUD_CONNECTED)) {
1651                         neigh_release(n);
1652                         goto out;
1653                 }
1654
1655                 f = vxlan_find_mac(vxlan, n->ha, vni);
1656                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1657                         /* bridge-local neighbor */
1658                         neigh_release(n);
1659                         goto out;
1660                 }
1661
1662                 reply = vxlan_na_create(skb, n,
1663                                         !!(f ? f->flags & NTF_ROUTER : 0));
1664
1665                 neigh_release(n);
1666
1667                 if (reply == NULL)
1668                         goto out;
1669
1670                 if (netif_rx_ni(reply) == NET_RX_DROP)
1671                         dev->stats.rx_dropped++;
1672
1673         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1674                 union vxlan_addr ipa = {
1675                         .sin6.sin6_addr = msg->target,
1676                         .sin6.sin6_family = AF_INET6,
1677                 };
1678
1679                 vxlan_ip_miss(dev, &ipa);
1680         }
1681
1682 out:
1683         consume_skb(skb);
1684         return NETDEV_TX_OK;
1685 }
1686 #endif
1687
1688 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1689 {
1690         struct vxlan_dev *vxlan = netdev_priv(dev);
1691         struct neighbour *n;
1692
1693         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1694                 return false;
1695
1696         n = NULL;
1697         switch (ntohs(eth_hdr(skb)->h_proto)) {
1698         case ETH_P_IP:
1699         {
1700                 struct iphdr *pip;
1701
1702                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1703                         return false;
1704                 pip = ip_hdr(skb);
1705                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1706                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1707                         union vxlan_addr ipa = {
1708                                 .sin.sin_addr.s_addr = pip->daddr,
1709                                 .sin.sin_family = AF_INET,
1710                         };
1711
1712                         vxlan_ip_miss(dev, &ipa);
1713                         return false;
1714                 }
1715
1716                 break;
1717         }
1718 #if IS_ENABLED(CONFIG_IPV6)
1719         case ETH_P_IPV6:
1720         {
1721                 struct ipv6hdr *pip6;
1722
1723                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1724                         return false;
1725                 pip6 = ipv6_hdr(skb);
1726                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1727                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1728                         union vxlan_addr ipa = {
1729                                 .sin6.sin6_addr = pip6->daddr,
1730                                 .sin6.sin6_family = AF_INET6,
1731                         };
1732
1733                         vxlan_ip_miss(dev, &ipa);
1734                         return false;
1735                 }
1736
1737                 break;
1738         }
1739 #endif
1740         default:
1741                 return false;
1742         }
1743
1744         if (n) {
1745                 bool diff;
1746
1747                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1748                 if (diff) {
1749                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1750                                 dev->addr_len);
1751                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1752                 }
1753                 neigh_release(n);
1754                 return diff;
1755         }
1756
1757         return false;
1758 }
1759
1760 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1761                                 struct vxlan_metadata *md)
1762 {
1763         struct vxlanhdr_gbp *gbp;
1764
1765         if (!md->gbp)
1766                 return;
1767
1768         gbp = (struct vxlanhdr_gbp *)vxh;
1769         vxh->vx_flags |= VXLAN_HF_GBP;
1770
1771         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1772                 gbp->dont_learn = 1;
1773
1774         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1775                 gbp->policy_applied = 1;
1776
1777         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1778 }
1779
1780 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1781                                __be16 protocol)
1782 {
1783         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1784
1785         gpe->np_applied = 1;
1786         gpe->next_protocol = tun_p_from_eth_p(protocol);
1787         if (!gpe->next_protocol)
1788                 return -EPFNOSUPPORT;
1789         return 0;
1790 }
1791
1792 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1793                            int iphdr_len, __be32 vni,
1794                            struct vxlan_metadata *md, u32 vxflags,
1795                            bool udp_sum)
1796 {
1797         struct vxlanhdr *vxh;
1798         int min_headroom;
1799         int err;
1800         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1801         __be16 inner_protocol = htons(ETH_P_TEB);
1802
1803         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1804             skb->ip_summed == CHECKSUM_PARTIAL) {
1805                 int csum_start = skb_checksum_start_offset(skb);
1806
1807                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1808                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1809                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1810                      skb->csum_offset == offsetof(struct tcphdr, check)))
1811                         type |= SKB_GSO_TUNNEL_REMCSUM;
1812         }
1813
1814         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1815                         + VXLAN_HLEN + iphdr_len;
1816
1817         /* Need space for new headers (invalidates iph ptr) */
1818         err = skb_cow_head(skb, min_headroom);
1819         if (unlikely(err))
1820                 return err;
1821
1822         err = iptunnel_handle_offloads(skb, type);
1823         if (err)
1824                 return err;
1825
1826         vxh = __skb_push(skb, sizeof(*vxh));
1827         vxh->vx_flags = VXLAN_HF_VNI;
1828         vxh->vx_vni = vxlan_vni_field(vni);
1829
1830         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1831                 unsigned int start;
1832
1833                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1834                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1835                 vxh->vx_flags |= VXLAN_HF_RCO;
1836
1837                 if (!skb_is_gso(skb)) {
1838                         skb->ip_summed = CHECKSUM_NONE;
1839                         skb->encapsulation = 0;
1840                 }
1841         }
1842
1843         if (vxflags & VXLAN_F_GBP)
1844                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1845         if (vxflags & VXLAN_F_GPE) {
1846                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1847                 if (err < 0)
1848                         return err;
1849                 inner_protocol = skb->protocol;
1850         }
1851
1852         skb_set_inner_protocol(skb, inner_protocol);
1853         return 0;
1854 }
1855
1856 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1857                                       struct vxlan_sock *sock4,
1858                                       struct sk_buff *skb, int oif, u8 tos,
1859                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
1860                                       struct dst_cache *dst_cache,
1861                                       const struct ip_tunnel_info *info)
1862 {
1863         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1864         struct rtable *rt = NULL;
1865         struct flowi4 fl4;
1866
1867         if (!sock4)
1868                 return ERR_PTR(-EIO);
1869
1870         if (tos && !info)
1871                 use_cache = false;
1872         if (use_cache) {
1873                 rt = dst_cache_get_ip4(dst_cache, saddr);
1874                 if (rt)
1875                         return rt;
1876         }
1877
1878         memset(&fl4, 0, sizeof(fl4));
1879         fl4.flowi4_oif = oif;
1880         fl4.flowi4_tos = RT_TOS(tos);
1881         fl4.flowi4_mark = skb->mark;
1882         fl4.flowi4_proto = IPPROTO_UDP;
1883         fl4.daddr = daddr;
1884         fl4.saddr = *saddr;
1885         fl4.fl4_dport = dport;
1886         fl4.fl4_sport = sport;
1887
1888         rt = ip_route_output_key(vxlan->net, &fl4);
1889         if (likely(!IS_ERR(rt))) {
1890                 if (rt->dst.dev == dev) {
1891                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1892                         ip_rt_put(rt);
1893                         return ERR_PTR(-ELOOP);
1894                 }
1895
1896                 *saddr = fl4.saddr;
1897                 if (use_cache)
1898                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1899         } else {
1900                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1901                 return ERR_PTR(-ENETUNREACH);
1902         }
1903         return rt;
1904 }
1905
1906 #if IS_ENABLED(CONFIG_IPV6)
1907 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1908                                           struct net_device *dev,
1909                                           struct vxlan_sock *sock6,
1910                                           struct sk_buff *skb, int oif, u8 tos,
1911                                           __be32 label,
1912                                           const struct in6_addr *daddr,
1913                                           struct in6_addr *saddr,
1914                                           __be16 dport, __be16 sport,
1915                                           struct dst_cache *dst_cache,
1916                                           const struct ip_tunnel_info *info)
1917 {
1918         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1919         struct dst_entry *ndst;
1920         struct flowi6 fl6;
1921         int err;
1922
1923         if (!sock6)
1924                 return ERR_PTR(-EIO);
1925
1926         if (tos && !info)
1927                 use_cache = false;
1928         if (use_cache) {
1929                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1930                 if (ndst)
1931                         return ndst;
1932         }
1933
1934         memset(&fl6, 0, sizeof(fl6));
1935         fl6.flowi6_oif = oif;
1936         fl6.daddr = *daddr;
1937         fl6.saddr = *saddr;
1938         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1939         fl6.flowi6_mark = skb->mark;
1940         fl6.flowi6_proto = IPPROTO_UDP;
1941         fl6.fl6_dport = dport;
1942         fl6.fl6_sport = sport;
1943
1944         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1945                                          sock6->sock->sk,
1946                                          &ndst, &fl6);
1947         if (unlikely(err < 0)) {
1948                 netdev_dbg(dev, "no route to %pI6\n", daddr);
1949                 return ERR_PTR(-ENETUNREACH);
1950         }
1951
1952         if (unlikely(ndst->dev == dev)) {
1953                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
1954                 dst_release(ndst);
1955                 return ERR_PTR(-ELOOP);
1956         }
1957
1958         *saddr = fl6.saddr;
1959         if (use_cache)
1960                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1961         return ndst;
1962 }
1963 #endif
1964
1965 /* Bypass encapsulation if the destination is local */
1966 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1967                                struct vxlan_dev *dst_vxlan, __be32 vni)
1968 {
1969         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1970         union vxlan_addr loopback;
1971         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1972         struct net_device *dev = skb->dev;
1973         int len = skb->len;
1974
1975         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1976         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1977         skb->pkt_type = PACKET_HOST;
1978         skb->encapsulation = 0;
1979         skb->dev = dst_vxlan->dev;
1980         __skb_pull(skb, skb_network_offset(skb));
1981
1982         if (remote_ip->sa.sa_family == AF_INET) {
1983                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1984                 loopback.sa.sa_family =  AF_INET;
1985 #if IS_ENABLED(CONFIG_IPV6)
1986         } else {
1987                 loopback.sin6.sin6_addr = in6addr_loopback;
1988                 loopback.sa.sa_family =  AF_INET6;
1989 #endif
1990         }
1991
1992         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
1993                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source, 0,
1994                             vni);
1995
1996         u64_stats_update_begin(&tx_stats->syncp);
1997         tx_stats->tx_packets++;
1998         tx_stats->tx_bytes += len;
1999         u64_stats_update_end(&tx_stats->syncp);
2000
2001         if (netif_rx(skb) == NET_RX_SUCCESS) {
2002                 u64_stats_update_begin(&rx_stats->syncp);
2003                 rx_stats->rx_packets++;
2004                 rx_stats->rx_bytes += len;
2005                 u64_stats_update_end(&rx_stats->syncp);
2006         } else {
2007                 dev->stats.rx_dropped++;
2008         }
2009 }
2010
2011 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2012                                  struct vxlan_dev *vxlan,
2013                                  union vxlan_addr *daddr,
2014                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2015                                  struct dst_entry *dst,
2016                                  u32 rt_flags)
2017 {
2018 #if IS_ENABLED(CONFIG_IPV6)
2019         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2020          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2021          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2022          */
2023         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2024 #endif
2025         /* Bypass encapsulation if the destination is local */
2026         if (rt_flags & RTCF_LOCAL &&
2027             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2028                 struct vxlan_dev *dst_vxlan;
2029
2030                 dst_release(dst);
2031                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2032                                            daddr->sa.sa_family, dst_port,
2033                                            vxlan->cfg.flags);
2034                 if (!dst_vxlan) {
2035                         dev->stats.tx_errors++;
2036                         kfree_skb(skb);
2037
2038                         return -ENOENT;
2039                 }
2040                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2041                 return 1;
2042         }
2043
2044         return 0;
2045 }
2046
2047 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2048                            __be32 default_vni, struct vxlan_rdst *rdst,
2049                            bool did_rsc)
2050 {
2051         struct dst_cache *dst_cache;
2052         struct ip_tunnel_info *info;
2053         struct vxlan_dev *vxlan = netdev_priv(dev);
2054         const struct iphdr *old_iph = ip_hdr(skb);
2055         union vxlan_addr *dst;
2056         union vxlan_addr remote_ip, local_ip;
2057         struct vxlan_metadata _md;
2058         struct vxlan_metadata *md = &_md;
2059         __be16 src_port = 0, dst_port;
2060         struct dst_entry *ndst = NULL;
2061         __be32 vni, label;
2062         __u8 tos, ttl;
2063         int ifindex;
2064         int err;
2065         u32 flags = vxlan->cfg.flags;
2066         bool udp_sum = false;
2067         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2068
2069         info = skb_tunnel_info(skb);
2070
2071         if (rdst) {
2072                 dst = &rdst->remote_ip;
2073                 if (vxlan_addr_any(dst)) {
2074                         if (did_rsc) {
2075                                 /* short-circuited back to local bridge */
2076                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2077                                 return;
2078                         }
2079                         goto drop;
2080                 }
2081
2082                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2083                 vni = (rdst->remote_vni) ? : default_vni;
2084                 ifindex = rdst->remote_ifindex;
2085                 local_ip = vxlan->cfg.saddr;
2086                 dst_cache = &rdst->dst_cache;
2087                 md->gbp = skb->mark;
2088                 if (flags & VXLAN_F_TTL_INHERIT) {
2089                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2090                 } else {
2091                         ttl = vxlan->cfg.ttl;
2092                         if (!ttl && vxlan_addr_multicast(dst))
2093                                 ttl = 1;
2094                 }
2095
2096                 tos = vxlan->cfg.tos;
2097                 if (tos == 1)
2098                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2099
2100                 if (dst->sa.sa_family == AF_INET)
2101                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2102                 else
2103                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2104                 label = vxlan->cfg.label;
2105         } else {
2106                 if (!info) {
2107                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2108                                   dev->name);
2109                         goto drop;
2110                 }
2111                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2112                 if (remote_ip.sa.sa_family == AF_INET) {
2113                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2114                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2115                 } else {
2116                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2117                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2118                 }
2119                 dst = &remote_ip;
2120                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2121                 vni = tunnel_id_to_key32(info->key.tun_id);
2122                 ifindex = 0;
2123                 dst_cache = &info->dst_cache;
2124                 if (info->options_len)
2125                         md = ip_tunnel_info_opts(info);
2126                 ttl = info->key.ttl;
2127                 tos = info->key.tos;
2128                 label = info->key.label;
2129                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2130         }
2131         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2132                                      vxlan->cfg.port_max, true);
2133
2134         rcu_read_lock();
2135         if (dst->sa.sa_family == AF_INET) {
2136                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2137                 struct rtable *rt;
2138                 __be16 df = 0;
2139
2140                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2141                                      dst->sin.sin_addr.s_addr,
2142                                      &local_ip.sin.sin_addr.s_addr,
2143                                      dst_port, src_port,
2144                                      dst_cache, info);
2145                 if (IS_ERR(rt)) {
2146                         err = PTR_ERR(rt);
2147                         goto tx_error;
2148                 }
2149
2150                 /* Bypass encapsulation if the destination is local */
2151                 if (!info) {
2152                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2153                                                     dst_port, ifindex, vni,
2154                                                     &rt->dst, rt->rt_flags);
2155                         if (err)
2156                                 goto out_unlock;
2157                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2158                         df = htons(IP_DF);
2159                 }
2160
2161                 ndst = &rt->dst;
2162                 if (skb_dst(skb)) {
2163                         int mtu = dst_mtu(ndst) - VXLAN_HEADROOM;
2164
2165                         skb_dst_update_pmtu(skb, mtu);
2166                 }
2167
2168                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2169                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2170                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2171                                       vni, md, flags, udp_sum);
2172                 if (err < 0)
2173                         goto tx_error;
2174
2175                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2176                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2177                                     src_port, dst_port, xnet, !udp_sum);
2178 #if IS_ENABLED(CONFIG_IPV6)
2179         } else {
2180                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2181
2182                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2183                                         label, &dst->sin6.sin6_addr,
2184                                         &local_ip.sin6.sin6_addr,
2185                                         dst_port, src_port,
2186                                         dst_cache, info);
2187                 if (IS_ERR(ndst)) {
2188                         err = PTR_ERR(ndst);
2189                         ndst = NULL;
2190                         goto tx_error;
2191                 }
2192
2193                 if (!info) {
2194                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2195
2196                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2197                                                     dst_port, ifindex, vni,
2198                                                     ndst, rt6i_flags);
2199                         if (err)
2200                                 goto out_unlock;
2201                 }
2202
2203                 if (skb_dst(skb)) {
2204                         int mtu = dst_mtu(ndst) - VXLAN6_HEADROOM;
2205
2206                         skb_dst_update_pmtu(skb, mtu);
2207                 }
2208
2209                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2210                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2211                 skb_scrub_packet(skb, xnet);
2212                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2213                                       vni, md, flags, udp_sum);
2214                 if (err < 0)
2215                         goto tx_error;
2216
2217                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2218                                      &local_ip.sin6.sin6_addr,
2219                                      &dst->sin6.sin6_addr, tos, ttl,
2220                                      label, src_port, dst_port, !udp_sum);
2221 #endif
2222         }
2223 out_unlock:
2224         rcu_read_unlock();
2225         return;
2226
2227 drop:
2228         dev->stats.tx_dropped++;
2229         dev_kfree_skb(skb);
2230         return;
2231
2232 tx_error:
2233         rcu_read_unlock();
2234         if (err == -ELOOP)
2235                 dev->stats.collisions++;
2236         else if (err == -ENETUNREACH)
2237                 dev->stats.tx_carrier_errors++;
2238         dst_release(ndst);
2239         dev->stats.tx_errors++;
2240         kfree_skb(skb);
2241 }
2242
2243 /* Transmit local packets over Vxlan
2244  *
2245  * Outer IP header inherits ECN and DF from inner header.
2246  * Outer UDP destination is the VXLAN assigned port.
2247  *           source port is based on hash of flow
2248  */
2249 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2250 {
2251         struct vxlan_dev *vxlan = netdev_priv(dev);
2252         struct vxlan_rdst *rdst, *fdst = NULL;
2253         const struct ip_tunnel_info *info;
2254         bool did_rsc = false;
2255         struct vxlan_fdb *f;
2256         struct ethhdr *eth;
2257         __be32 vni = 0;
2258
2259         info = skb_tunnel_info(skb);
2260
2261         skb_reset_mac_header(skb);
2262
2263         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2264                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2265                     info->mode & IP_TUNNEL_INFO_TX) {
2266                         vni = tunnel_id_to_key32(info->key.tun_id);
2267                 } else {
2268                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2269                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2270                         else
2271                                 kfree_skb(skb);
2272                         return NETDEV_TX_OK;
2273                 }
2274         }
2275
2276         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2277                 eth = eth_hdr(skb);
2278                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2279                         return arp_reduce(dev, skb, vni);
2280 #if IS_ENABLED(CONFIG_IPV6)
2281                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2282                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2283                                             sizeof(struct nd_msg)) &&
2284                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2285                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2286
2287                         if (m->icmph.icmp6_code == 0 &&
2288                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2289                                 return neigh_reduce(dev, skb, vni);
2290                 }
2291 #endif
2292         }
2293
2294         eth = eth_hdr(skb);
2295         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2296         did_rsc = false;
2297
2298         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2299             (ntohs(eth->h_proto) == ETH_P_IP ||
2300              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2301                 did_rsc = route_shortcircuit(dev, skb);
2302                 if (did_rsc)
2303                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2304         }
2305
2306         if (f == NULL) {
2307                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2308                 if (f == NULL) {
2309                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2310                             !is_multicast_ether_addr(eth->h_dest))
2311                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2312
2313                         dev->stats.tx_dropped++;
2314                         kfree_skb(skb);
2315                         return NETDEV_TX_OK;
2316                 }
2317         }
2318
2319         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2320                 struct sk_buff *skb1;
2321
2322                 if (!fdst) {
2323                         fdst = rdst;
2324                         continue;
2325                 }
2326                 skb1 = skb_clone(skb, GFP_ATOMIC);
2327                 if (skb1)
2328                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2329         }
2330
2331         if (fdst)
2332                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2333         else
2334                 kfree_skb(skb);
2335         return NETDEV_TX_OK;
2336 }
2337
2338 /* Walk the forwarding table and purge stale entries */
2339 static void vxlan_cleanup(struct timer_list *t)
2340 {
2341         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2342         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2343         unsigned int h;
2344
2345         if (!netif_running(vxlan->dev))
2346                 return;
2347
2348         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2349                 struct hlist_node *p, *n;
2350
2351                 spin_lock_bh(&vxlan->hash_lock);
2352                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2353                         struct vxlan_fdb *f
2354                                 = container_of(p, struct vxlan_fdb, hlist);
2355                         unsigned long timeout;
2356
2357                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2358                                 continue;
2359
2360                         if (f->flags & NTF_EXT_LEARNED)
2361                                 continue;
2362
2363                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2364                         if (time_before_eq(timeout, jiffies)) {
2365                                 netdev_dbg(vxlan->dev,
2366                                            "garbage collect %pM\n",
2367                                            f->eth_addr);
2368                                 f->state = NUD_STALE;
2369                                 vxlan_fdb_destroy(vxlan, f);
2370                         } else if (time_before(timeout, next_timer))
2371                                 next_timer = timeout;
2372                 }
2373                 spin_unlock_bh(&vxlan->hash_lock);
2374         }
2375
2376         mod_timer(&vxlan->age_timer, next_timer);
2377 }
2378
2379 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2380 {
2381         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2382
2383         spin_lock(&vn->sock_lock);
2384         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2385 #if IS_ENABLED(CONFIG_IPV6)
2386         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2387 #endif
2388         spin_unlock(&vn->sock_lock);
2389 }
2390
2391 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2392                              struct vxlan_dev_node *node)
2393 {
2394         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2395         __be32 vni = vxlan->default_dst.remote_vni;
2396
2397         node->vxlan = vxlan;
2398         spin_lock(&vn->sock_lock);
2399         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2400         spin_unlock(&vn->sock_lock);
2401 }
2402
2403 /* Setup stats when device is created */
2404 static int vxlan_init(struct net_device *dev)
2405 {
2406         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2407         if (!dev->tstats)
2408                 return -ENOMEM;
2409
2410         return 0;
2411 }
2412
2413 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2414 {
2415         struct vxlan_fdb *f;
2416
2417         spin_lock_bh(&vxlan->hash_lock);
2418         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2419         if (f)
2420                 vxlan_fdb_destroy(vxlan, f);
2421         spin_unlock_bh(&vxlan->hash_lock);
2422 }
2423
2424 static void vxlan_uninit(struct net_device *dev)
2425 {
2426         struct vxlan_dev *vxlan = netdev_priv(dev);
2427
2428         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2429
2430         free_percpu(dev->tstats);
2431 }
2432
2433 /* Start ageing timer and join group when device is brought up */
2434 static int vxlan_open(struct net_device *dev)
2435 {
2436         struct vxlan_dev *vxlan = netdev_priv(dev);
2437         int ret;
2438
2439         ret = vxlan_sock_add(vxlan);
2440         if (ret < 0)
2441                 return ret;
2442
2443         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2444                 ret = vxlan_igmp_join(vxlan);
2445                 if (ret == -EADDRINUSE)
2446                         ret = 0;
2447                 if (ret) {
2448                         vxlan_sock_release(vxlan);
2449                         return ret;
2450                 }
2451         }
2452
2453         if (vxlan->cfg.age_interval)
2454                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2455
2456         return ret;
2457 }
2458
2459 /* Purge the forwarding table */
2460 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2461 {
2462         unsigned int h;
2463
2464         spin_lock_bh(&vxlan->hash_lock);
2465         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2466                 struct hlist_node *p, *n;
2467                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2468                         struct vxlan_fdb *f
2469                                 = container_of(p, struct vxlan_fdb, hlist);
2470                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2471                                 continue;
2472                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2473                         if (!is_zero_ether_addr(f->eth_addr))
2474                                 vxlan_fdb_destroy(vxlan, f);
2475                 }
2476         }
2477         spin_unlock_bh(&vxlan->hash_lock);
2478 }
2479
2480 /* Cleanup timer and forwarding table on shutdown */
2481 static int vxlan_stop(struct net_device *dev)
2482 {
2483         struct vxlan_dev *vxlan = netdev_priv(dev);
2484         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2485         int ret = 0;
2486
2487         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2488             !vxlan_group_used(vn, vxlan))
2489                 ret = vxlan_igmp_leave(vxlan);
2490
2491         del_timer_sync(&vxlan->age_timer);
2492
2493         vxlan_flush(vxlan, false);
2494         vxlan_sock_release(vxlan);
2495
2496         return ret;
2497 }
2498
2499 /* Stub, nothing needs to be done. */
2500 static void vxlan_set_multicast_list(struct net_device *dev)
2501 {
2502 }
2503
2504 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2505 {
2506         struct vxlan_dev *vxlan = netdev_priv(dev);
2507         struct vxlan_rdst *dst = &vxlan->default_dst;
2508         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2509                                                          dst->remote_ifindex);
2510         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
2511
2512         /* This check is different than dev->max_mtu, because it looks at
2513          * the lowerdev->mtu, rather than the static dev->max_mtu
2514          */
2515         if (lowerdev) {
2516                 int max_mtu = lowerdev->mtu -
2517                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2518                 if (new_mtu > max_mtu)
2519                         return -EINVAL;
2520         }
2521
2522         dev->mtu = new_mtu;
2523         return 0;
2524 }
2525
2526 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2527 {
2528         struct vxlan_dev *vxlan = netdev_priv(dev);
2529         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2530         __be16 sport, dport;
2531
2532         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2533                                   vxlan->cfg.port_max, true);
2534         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2535
2536         if (ip_tunnel_info_af(info) == AF_INET) {
2537                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2538                 struct rtable *rt;
2539
2540                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2541                                      info->key.u.ipv4.dst,
2542                                      &info->key.u.ipv4.src, dport, sport,
2543                                      &info->dst_cache, info);
2544                 if (IS_ERR(rt))
2545                         return PTR_ERR(rt);
2546                 ip_rt_put(rt);
2547         } else {
2548 #if IS_ENABLED(CONFIG_IPV6)
2549                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2550                 struct dst_entry *ndst;
2551
2552                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2553                                         info->key.label, &info->key.u.ipv6.dst,
2554                                         &info->key.u.ipv6.src, dport, sport,
2555                                         &info->dst_cache, info);
2556                 if (IS_ERR(ndst))
2557                         return PTR_ERR(ndst);
2558                 dst_release(ndst);
2559 #else /* !CONFIG_IPV6 */
2560                 return -EPFNOSUPPORT;
2561 #endif
2562         }
2563         info->key.tp_src = sport;
2564         info->key.tp_dst = dport;
2565         return 0;
2566 }
2567
2568 static const struct net_device_ops vxlan_netdev_ether_ops = {
2569         .ndo_init               = vxlan_init,
2570         .ndo_uninit             = vxlan_uninit,
2571         .ndo_open               = vxlan_open,
2572         .ndo_stop               = vxlan_stop,
2573         .ndo_start_xmit         = vxlan_xmit,
2574         .ndo_get_stats64        = ip_tunnel_get_stats64,
2575         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2576         .ndo_change_mtu         = vxlan_change_mtu,
2577         .ndo_validate_addr      = eth_validate_addr,
2578         .ndo_set_mac_address    = eth_mac_addr,
2579         .ndo_fdb_add            = vxlan_fdb_add,
2580         .ndo_fdb_del            = vxlan_fdb_delete,
2581         .ndo_fdb_dump           = vxlan_fdb_dump,
2582         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2583 };
2584
2585 static const struct net_device_ops vxlan_netdev_raw_ops = {
2586         .ndo_init               = vxlan_init,
2587         .ndo_uninit             = vxlan_uninit,
2588         .ndo_open               = vxlan_open,
2589         .ndo_stop               = vxlan_stop,
2590         .ndo_start_xmit         = vxlan_xmit,
2591         .ndo_get_stats64        = ip_tunnel_get_stats64,
2592         .ndo_change_mtu         = vxlan_change_mtu,
2593         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2594 };
2595
2596 /* Info for udev, that this is a virtual tunnel endpoint */
2597 static struct device_type vxlan_type = {
2598         .name = "vxlan",
2599 };
2600
2601 /* Calls the ndo_udp_tunnel_add of the caller in order to
2602  * supply the listening VXLAN udp ports. Callers are expected
2603  * to implement the ndo_udp_tunnel_add.
2604  */
2605 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
2606 {
2607         struct vxlan_sock *vs;
2608         struct net *net = dev_net(dev);
2609         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2610         unsigned int i;
2611
2612         spin_lock(&vn->sock_lock);
2613         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2614                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2615                         unsigned short type;
2616
2617                         if (vs->flags & VXLAN_F_GPE)
2618                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
2619                         else
2620                                 type = UDP_TUNNEL_TYPE_VXLAN;
2621
2622                         if (push)
2623                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
2624                         else
2625                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
2626                 }
2627         }
2628         spin_unlock(&vn->sock_lock);
2629 }
2630
2631 /* Initialize the device structure. */
2632 static void vxlan_setup(struct net_device *dev)
2633 {
2634         struct vxlan_dev *vxlan = netdev_priv(dev);
2635         unsigned int h;
2636
2637         eth_hw_addr_random(dev);
2638         ether_setup(dev);
2639
2640         dev->needs_free_netdev = true;
2641         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2642
2643         dev->features   |= NETIF_F_LLTX;
2644         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2645         dev->features   |= NETIF_F_RXCSUM;
2646         dev->features   |= NETIF_F_GSO_SOFTWARE;
2647
2648         dev->vlan_features = dev->features;
2649         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2650         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2651         netif_keep_dst(dev);
2652         dev->priv_flags |= IFF_NO_QUEUE;
2653
2654         /* MTU range: 68 - 65535 */
2655         dev->min_mtu = ETH_MIN_MTU;
2656         dev->max_mtu = ETH_MAX_MTU;
2657
2658         INIT_LIST_HEAD(&vxlan->next);
2659         spin_lock_init(&vxlan->hash_lock);
2660
2661         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
2662
2663         vxlan->dev = dev;
2664
2665         gro_cells_init(&vxlan->gro_cells, dev);
2666
2667         for (h = 0; h < FDB_HASH_SIZE; ++h)
2668                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2669 }
2670
2671 static void vxlan_ether_setup(struct net_device *dev)
2672 {
2673         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2674         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2675         dev->netdev_ops = &vxlan_netdev_ether_ops;
2676 }
2677
2678 static void vxlan_raw_setup(struct net_device *dev)
2679 {
2680         dev->header_ops = NULL;
2681         dev->type = ARPHRD_NONE;
2682         dev->hard_header_len = 0;
2683         dev->addr_len = 0;
2684         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2685         dev->netdev_ops = &vxlan_netdev_raw_ops;
2686 }
2687
2688 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2689         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2690         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2691         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2692         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2693         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2694         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2695         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2696         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2697         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2698         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2699         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2700         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2701         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2702         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2703         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2704         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2705         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2706         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2707         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2708         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2709         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2710         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2711         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2712         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2713         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2714         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2715         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2716         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
2717 };
2718
2719 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
2720                           struct netlink_ext_ack *extack)
2721 {
2722         if (tb[IFLA_ADDRESS]) {
2723                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2724                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2725                                             "Provided link layer address is not Ethernet");
2726                         return -EINVAL;
2727                 }
2728
2729                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2730                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2731                                             "Provided Ethernet address is not unicast");
2732                         return -EADDRNOTAVAIL;
2733                 }
2734         }
2735
2736         if (tb[IFLA_MTU]) {
2737                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
2738
2739                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
2740                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
2741                                             "MTU must be between 68 and 65535");
2742                         return -EINVAL;
2743                 }
2744         }
2745
2746         if (!data) {
2747                 NL_SET_ERR_MSG(extack,
2748                                "Required attributes not provided to perform the operation");
2749                 return -EINVAL;
2750         }
2751
2752         if (data[IFLA_VXLAN_ID]) {
2753                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2754
2755                 if (id >= VXLAN_N_VID) {
2756                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID],
2757                                             "VXLAN ID must be lower than 16777216");
2758                         return -ERANGE;
2759                 }
2760         }
2761
2762         if (data[IFLA_VXLAN_PORT_RANGE]) {
2763                 const struct ifla_vxlan_port_range *p
2764                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2765
2766                 if (ntohs(p->high) < ntohs(p->low)) {
2767                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
2768                                             "Invalid source port range");
2769                         return -EINVAL;
2770                 }
2771         }
2772
2773         return 0;
2774 }
2775
2776 static void vxlan_get_drvinfo(struct net_device *netdev,
2777                               struct ethtool_drvinfo *drvinfo)
2778 {
2779         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2780         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2781 }
2782
2783 static const struct ethtool_ops vxlan_ethtool_ops = {
2784         .get_drvinfo    = vxlan_get_drvinfo,
2785         .get_link       = ethtool_op_get_link,
2786 };
2787
2788 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2789                                         __be16 port, u32 flags)
2790 {
2791         struct socket *sock;
2792         struct udp_port_cfg udp_conf;
2793         int err;
2794
2795         memset(&udp_conf, 0, sizeof(udp_conf));
2796
2797         if (ipv6) {
2798                 udp_conf.family = AF_INET6;
2799                 udp_conf.use_udp6_rx_checksums =
2800                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2801                 udp_conf.ipv6_v6only = 1;
2802         } else {
2803                 udp_conf.family = AF_INET;
2804         }
2805
2806         udp_conf.local_udp_port = port;
2807
2808         /* Open UDP socket */
2809         err = udp_sock_create(net, &udp_conf, &sock);
2810         if (err < 0)
2811                 return ERR_PTR(err);
2812
2813         return sock;
2814 }
2815
2816 /* Create new listen socket if needed */
2817 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2818                                               __be16 port, u32 flags)
2819 {
2820         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2821         struct vxlan_sock *vs;
2822         struct socket *sock;
2823         unsigned int h;
2824         struct udp_tunnel_sock_cfg tunnel_cfg;
2825
2826         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2827         if (!vs)
2828                 return ERR_PTR(-ENOMEM);
2829
2830         for (h = 0; h < VNI_HASH_SIZE; ++h)
2831                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2832
2833         sock = vxlan_create_sock(net, ipv6, port, flags);
2834         if (IS_ERR(sock)) {
2835                 kfree(vs);
2836                 return ERR_CAST(sock);
2837         }
2838
2839         vs->sock = sock;
2840         refcount_set(&vs->refcnt, 1);
2841         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2842
2843         spin_lock(&vn->sock_lock);
2844         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2845         udp_tunnel_notify_add_rx_port(sock,
2846                                       (vs->flags & VXLAN_F_GPE) ?
2847                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2848                                       UDP_TUNNEL_TYPE_VXLAN);
2849         spin_unlock(&vn->sock_lock);
2850
2851         /* Mark socket as an encapsulation socket. */
2852         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2853         tunnel_cfg.sk_user_data = vs;
2854         tunnel_cfg.encap_type = 1;
2855         tunnel_cfg.encap_rcv = vxlan_rcv;
2856         tunnel_cfg.encap_destroy = NULL;
2857         tunnel_cfg.gro_receive = vxlan_gro_receive;
2858         tunnel_cfg.gro_complete = vxlan_gro_complete;
2859
2860         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2861
2862         return vs;
2863 }
2864
2865 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2866 {
2867         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2868         struct vxlan_sock *vs = NULL;
2869         struct vxlan_dev_node *node;
2870
2871         if (!vxlan->cfg.no_share) {
2872                 spin_lock(&vn->sock_lock);
2873                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2874                                      vxlan->cfg.dst_port, vxlan->cfg.flags);
2875                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
2876                         spin_unlock(&vn->sock_lock);
2877                         return -EBUSY;
2878                 }
2879                 spin_unlock(&vn->sock_lock);
2880         }
2881         if (!vs)
2882                 vs = vxlan_socket_create(vxlan->net, ipv6,
2883                                          vxlan->cfg.dst_port, vxlan->cfg.flags);
2884         if (IS_ERR(vs))
2885                 return PTR_ERR(vs);
2886 #if IS_ENABLED(CONFIG_IPV6)
2887         if (ipv6) {
2888                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2889                 node = &vxlan->hlist6;
2890         } else
2891 #endif
2892         {
2893                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2894                 node = &vxlan->hlist4;
2895         }
2896         vxlan_vs_add_dev(vs, vxlan, node);
2897         return 0;
2898 }
2899
2900 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2901 {
2902         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
2903         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
2904         bool ipv4 = !ipv6 || metadata;
2905         int ret = 0;
2906
2907         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2908 #if IS_ENABLED(CONFIG_IPV6)
2909         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2910         if (ipv6) {
2911                 ret = __vxlan_sock_add(vxlan, true);
2912                 if (ret < 0 && ret != -EAFNOSUPPORT)
2913                         ipv4 = false;
2914         }
2915 #endif
2916         if (ipv4)
2917                 ret = __vxlan_sock_add(vxlan, false);
2918         if (ret < 0)
2919                 vxlan_sock_release(vxlan);
2920         return ret;
2921 }
2922
2923 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
2924                                  struct net_device **lower,
2925                                  struct vxlan_dev *old,
2926                                  struct netlink_ext_ack *extack)
2927 {
2928         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2929         struct vxlan_dev *tmp;
2930         bool use_ipv6 = false;
2931
2932         if (conf->flags & VXLAN_F_GPE) {
2933                 /* For now, allow GPE only together with
2934                  * COLLECT_METADATA. This can be relaxed later; in such
2935                  * case, the other side of the PtP link will have to be
2936                  * provided.
2937                  */
2938                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2939                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2940                         NL_SET_ERR_MSG(extack,
2941                                        "VXLAN GPE does not support this combination of attributes");
2942                         return -EINVAL;
2943                 }
2944         }
2945
2946         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
2947                 /* Unless IPv6 is explicitly requested, assume IPv4 */
2948                 conf->remote_ip.sa.sa_family = AF_INET;
2949                 conf->saddr.sa.sa_family = AF_INET;
2950         } else if (!conf->remote_ip.sa.sa_family) {
2951                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
2952         } else if (!conf->saddr.sa.sa_family) {
2953                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
2954         }
2955
2956         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
2957                 NL_SET_ERR_MSG(extack,
2958                                "Local and remote address must be from the same family");
2959                 return -EINVAL;
2960         }
2961
2962         if (vxlan_addr_multicast(&conf->saddr)) {
2963                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
2964                 return -EINVAL;
2965         }
2966
2967         if (conf->saddr.sa.sa_family == AF_INET6) {
2968                 if (!IS_ENABLED(CONFIG_IPV6)) {
2969                         NL_SET_ERR_MSG(extack,
2970                                        "IPv6 support not enabled in the kernel");
2971                         return -EPFNOSUPPORT;
2972                 }
2973                 use_ipv6 = true;
2974                 conf->flags |= VXLAN_F_IPV6;
2975
2976                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2977                         int local_type =
2978                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
2979                         int remote_type =
2980                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
2981
2982                         if (local_type & IPV6_ADDR_LINKLOCAL) {
2983                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
2984                                     (remote_type != IPV6_ADDR_ANY)) {
2985                                         NL_SET_ERR_MSG(extack,
2986                                                        "Invalid combination of local and remote address scopes");
2987                                         return -EINVAL;
2988                                 }
2989
2990                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
2991                         } else {
2992                                 if (remote_type ==
2993                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
2994                                         NL_SET_ERR_MSG(extack,
2995                                                        "Invalid combination of local and remote address scopes");
2996                                         return -EINVAL;
2997                                 }
2998
2999                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3000                         }
3001                 }
3002         }
3003
3004         if (conf->label && !use_ipv6) {
3005                 NL_SET_ERR_MSG(extack,
3006                                "Label attribute only applies to IPv6 VXLAN devices");
3007                 return -EINVAL;
3008         }
3009
3010         if (conf->remote_ifindex) {
3011                 struct net_device *lowerdev;
3012
3013                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3014                 if (!lowerdev) {
3015                         NL_SET_ERR_MSG(extack,
3016                                        "Invalid local interface, device not found");
3017                         return -ENODEV;
3018                 }
3019
3020 #if IS_ENABLED(CONFIG_IPV6)
3021                 if (use_ipv6) {
3022                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3023                         if (idev && idev->cnf.disable_ipv6) {
3024                                 NL_SET_ERR_MSG(extack,
3025                                                "IPv6 support disabled by administrator");
3026                                 return -EPERM;
3027                         }
3028                 }
3029 #endif
3030
3031                 *lower = lowerdev;
3032         } else {
3033                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3034                         NL_SET_ERR_MSG(extack,
3035                                        "Local interface required for multicast remote destination");
3036
3037                         return -EINVAL;
3038                 }
3039
3040 #if IS_ENABLED(CONFIG_IPV6)
3041                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3042                         NL_SET_ERR_MSG(extack,
3043                                        "Local interface required for link-local local/remote addresses");
3044                         return -EINVAL;
3045                 }
3046 #endif
3047
3048                 *lower = NULL;
3049         }
3050
3051         if (!conf->dst_port) {
3052                 if (conf->flags & VXLAN_F_GPE)
3053                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3054                 else
3055                         conf->dst_port = htons(vxlan_port);
3056         }
3057
3058         if (!conf->age_interval)
3059                 conf->age_interval = FDB_AGE_DEFAULT;
3060
3061         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3062                 if (tmp == old)
3063                         continue;
3064
3065                 if (tmp->cfg.vni != conf->vni)
3066                         continue;
3067                 if (tmp->cfg.dst_port != conf->dst_port)
3068                         continue;
3069                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3070                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3071                         continue;
3072
3073                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3074                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3075                         continue;
3076
3077                 NL_SET_ERR_MSG(extack,
3078                                "A VXLAN device with the specified VNI already exists");
3079                 return -EEXIST;
3080         }
3081
3082         return 0;
3083 }
3084
3085 static void vxlan_config_apply(struct net_device *dev,
3086                                struct vxlan_config *conf,
3087                                struct net_device *lowerdev,
3088                                struct net *src_net,
3089                                bool changelink)
3090 {
3091         struct vxlan_dev *vxlan = netdev_priv(dev);
3092         struct vxlan_rdst *dst = &vxlan->default_dst;
3093         unsigned short needed_headroom = ETH_HLEN;
3094         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3095         int max_mtu = ETH_MAX_MTU;
3096
3097         if (!changelink) {
3098                 if (conf->flags & VXLAN_F_GPE)
3099                         vxlan_raw_setup(dev);
3100                 else
3101                         vxlan_ether_setup(dev);
3102
3103                 if (conf->mtu)
3104                         dev->mtu = conf->mtu;
3105
3106                 vxlan->net = src_net;
3107         }
3108
3109         dst->remote_vni = conf->vni;
3110
3111         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3112
3113         if (lowerdev) {
3114                 dst->remote_ifindex = conf->remote_ifindex;
3115
3116                 dev->gso_max_size = lowerdev->gso_max_size;
3117                 dev->gso_max_segs = lowerdev->gso_max_segs;
3118
3119                 needed_headroom = lowerdev->hard_header_len;
3120
3121                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3122                                            VXLAN_HEADROOM);
3123                 if (max_mtu < ETH_MIN_MTU)
3124                         max_mtu = ETH_MIN_MTU;
3125
3126                 if (!changelink && !conf->mtu)
3127                         dev->mtu = max_mtu;
3128         }
3129
3130         if (dev->mtu > max_mtu)
3131                 dev->mtu = max_mtu;
3132
3133         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3134                 needed_headroom += VXLAN6_HEADROOM;
3135         else
3136                 needed_headroom += VXLAN_HEADROOM;
3137         dev->needed_headroom = needed_headroom;
3138
3139         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3140 }
3141
3142 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3143                                struct vxlan_config *conf, bool changelink,
3144                                struct netlink_ext_ack *extack)
3145 {
3146         struct vxlan_dev *vxlan = netdev_priv(dev);
3147         struct net_device *lowerdev;
3148         int ret;
3149
3150         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3151         if (ret)
3152                 return ret;
3153
3154         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3155
3156         return 0;
3157 }
3158
3159 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3160                               struct vxlan_config *conf,
3161                               struct netlink_ext_ack *extack)
3162 {
3163         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3164         struct vxlan_dev *vxlan = netdev_priv(dev);
3165         int err;
3166
3167         err = vxlan_dev_configure(net, dev, conf, false, extack);
3168         if (err)
3169                 return err;
3170
3171         dev->ethtool_ops = &vxlan_ethtool_ops;
3172
3173         /* create an fdb entry for a valid default destination */
3174         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
3175                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3176                                        &vxlan->default_dst.remote_ip,
3177                                        NUD_REACHABLE | NUD_PERMANENT,
3178                                        NLM_F_EXCL | NLM_F_CREATE,
3179                                        vxlan->cfg.dst_port,
3180                                        vxlan->default_dst.remote_vni,
3181                                        vxlan->default_dst.remote_vni,
3182                                        vxlan->default_dst.remote_ifindex,
3183                                        NTF_SELF);
3184                 if (err)
3185                         return err;
3186         }
3187
3188         err = register_netdevice(dev);
3189         if (err) {
3190                 vxlan_fdb_delete_default(vxlan, vxlan->default_dst.remote_vni);
3191                 return err;
3192         }
3193
3194         list_add(&vxlan->next, &vn->vxlan_list);
3195         return 0;
3196 }
3197
3198 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3199                          struct net_device *dev, struct vxlan_config *conf,
3200                          bool changelink)
3201 {
3202         struct vxlan_dev *vxlan = netdev_priv(dev);
3203
3204         memset(conf, 0, sizeof(*conf));
3205
3206         /* if changelink operation, start with old existing cfg */
3207         if (changelink)
3208                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3209
3210         if (data[IFLA_VXLAN_ID]) {
3211                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3212
3213                 if (changelink && (vni != conf->vni))
3214                         return -EOPNOTSUPP;
3215                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3216         }
3217
3218         if (data[IFLA_VXLAN_GROUP]) {
3219                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET))
3220                         return -EOPNOTSUPP;
3221
3222                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3223                 conf->remote_ip.sa.sa_family = AF_INET;
3224         } else if (data[IFLA_VXLAN_GROUP6]) {
3225                 if (!IS_ENABLED(CONFIG_IPV6))
3226                         return -EPFNOSUPPORT;
3227
3228                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6))
3229                         return -EOPNOTSUPP;
3230
3231                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3232                 conf->remote_ip.sa.sa_family = AF_INET6;
3233         }
3234
3235         if (data[IFLA_VXLAN_LOCAL]) {
3236                 if (changelink && (conf->saddr.sa.sa_family != AF_INET))
3237                         return -EOPNOTSUPP;
3238
3239                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3240                 conf->saddr.sa.sa_family = AF_INET;
3241         } else if (data[IFLA_VXLAN_LOCAL6]) {
3242                 if (!IS_ENABLED(CONFIG_IPV6))
3243                         return -EPFNOSUPPORT;
3244
3245                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6))
3246                         return -EOPNOTSUPP;
3247
3248                 /* TODO: respect scope id */
3249                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3250                 conf->saddr.sa.sa_family = AF_INET6;
3251         }
3252
3253         if (data[IFLA_VXLAN_LINK])
3254                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3255
3256         if (data[IFLA_VXLAN_TOS])
3257                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3258
3259         if (data[IFLA_VXLAN_TTL])
3260                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3261
3262         if (data[IFLA_VXLAN_TTL_INHERIT]) {
3263                 if (changelink)
3264                         return -EOPNOTSUPP;
3265                 conf->flags |= VXLAN_F_TTL_INHERIT;
3266         }
3267
3268         if (data[IFLA_VXLAN_LABEL])
3269                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3270                              IPV6_FLOWLABEL_MASK;
3271
3272         if (data[IFLA_VXLAN_LEARNING]) {
3273                 if (nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3274                         conf->flags |= VXLAN_F_LEARN;
3275                 else
3276                         conf->flags &= ~VXLAN_F_LEARN;
3277         } else if (!changelink) {
3278                 /* default to learn on a new device */
3279                 conf->flags |= VXLAN_F_LEARN;
3280         }
3281
3282         if (data[IFLA_VXLAN_AGEING]) {
3283                 if (changelink)
3284                         return -EOPNOTSUPP;
3285                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3286         }
3287
3288         if (data[IFLA_VXLAN_PROXY]) {
3289                 if (changelink)
3290                         return -EOPNOTSUPP;
3291                 if (nla_get_u8(data[IFLA_VXLAN_PROXY]))
3292                         conf->flags |= VXLAN_F_PROXY;
3293         }
3294
3295         if (data[IFLA_VXLAN_RSC]) {
3296                 if (changelink)
3297                         return -EOPNOTSUPP;
3298                 if (nla_get_u8(data[IFLA_VXLAN_RSC]))
3299                         conf->flags |= VXLAN_F_RSC;
3300         }
3301
3302         if (data[IFLA_VXLAN_L2MISS]) {
3303                 if (changelink)
3304                         return -EOPNOTSUPP;
3305                 if (nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3306                         conf->flags |= VXLAN_F_L2MISS;
3307         }
3308
3309         if (data[IFLA_VXLAN_L3MISS]) {
3310                 if (changelink)
3311                         return -EOPNOTSUPP;
3312                 if (nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3313                         conf->flags |= VXLAN_F_L3MISS;
3314         }
3315
3316         if (data[IFLA_VXLAN_LIMIT]) {
3317                 if (changelink)
3318                         return -EOPNOTSUPP;
3319                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3320         }
3321
3322         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3323                 if (changelink)
3324                         return -EOPNOTSUPP;
3325                 if (nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3326                         conf->flags |= VXLAN_F_COLLECT_METADATA;
3327         }
3328
3329         if (data[IFLA_VXLAN_PORT_RANGE]) {
3330                 if (!changelink) {
3331                         const struct ifla_vxlan_port_range *p
3332                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3333                         conf->port_min = ntohs(p->low);
3334                         conf->port_max = ntohs(p->high);
3335                 } else {
3336                         return -EOPNOTSUPP;
3337                 }
3338         }
3339
3340         if (data[IFLA_VXLAN_PORT]) {
3341                 if (changelink)
3342                         return -EOPNOTSUPP;
3343                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3344         }
3345
3346         if (data[IFLA_VXLAN_UDP_CSUM]) {
3347                 if (changelink)
3348                         return -EOPNOTSUPP;
3349                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3350                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3351         }
3352
3353         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3354                 if (changelink)
3355                         return -EOPNOTSUPP;
3356                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3357                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3358         }
3359
3360         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3361                 if (changelink)
3362                         return -EOPNOTSUPP;
3363                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3364                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3365         }
3366
3367         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3368                 if (changelink)
3369                         return -EOPNOTSUPP;
3370                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3371                         conf->flags |= VXLAN_F_REMCSUM_TX;
3372         }
3373
3374         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3375                 if (changelink)
3376                         return -EOPNOTSUPP;
3377                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3378                         conf->flags |= VXLAN_F_REMCSUM_RX;
3379         }
3380
3381         if (data[IFLA_VXLAN_GBP]) {
3382                 if (changelink)
3383                         return -EOPNOTSUPP;
3384                 conf->flags |= VXLAN_F_GBP;
3385         }
3386
3387         if (data[IFLA_VXLAN_GPE]) {
3388                 if (changelink)
3389                         return -EOPNOTSUPP;
3390                 conf->flags |= VXLAN_F_GPE;
3391         }
3392
3393         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3394                 if (changelink)
3395                         return -EOPNOTSUPP;
3396                 conf->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3397         }
3398
3399         if (tb[IFLA_MTU]) {
3400                 if (changelink)
3401                         return -EOPNOTSUPP;
3402                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3403         }
3404
3405         return 0;
3406 }
3407
3408 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3409                          struct nlattr *tb[], struct nlattr *data[],
3410                          struct netlink_ext_ack *extack)
3411 {
3412         struct vxlan_config conf;
3413         int err;
3414
3415         err = vxlan_nl2conf(tb, data, dev, &conf, false);
3416         if (err)
3417                 return err;
3418
3419         return __vxlan_dev_create(src_net, dev, &conf, extack);
3420 }
3421
3422 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3423                             struct nlattr *data[],
3424                             struct netlink_ext_ack *extack)
3425 {
3426         struct vxlan_dev *vxlan = netdev_priv(dev);
3427         struct vxlan_rdst *dst = &vxlan->default_dst;
3428         struct vxlan_rdst old_dst;
3429         struct vxlan_config conf;
3430         int err;
3431
3432         err = vxlan_nl2conf(tb, data,
3433                             dev, &conf, true);
3434         if (err)
3435                 return err;
3436
3437         memcpy(&old_dst, dst, sizeof(struct vxlan_rdst));
3438
3439         err = vxlan_dev_configure(vxlan->net, dev, &conf, true, extack);
3440         if (err)
3441                 return err;
3442
3443         /* handle default dst entry */
3444         if (!vxlan_addr_equal(&dst->remote_ip, &old_dst.remote_ip)) {
3445                 spin_lock_bh(&vxlan->hash_lock);
3446                 if (!vxlan_addr_any(&old_dst.remote_ip))
3447                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
3448                                            old_dst.remote_ip,
3449                                            vxlan->cfg.dst_port,
3450                                            old_dst.remote_vni,
3451                                            old_dst.remote_vni,
3452                                            old_dst.remote_ifindex, 0);
3453
3454                 if (!vxlan_addr_any(&dst->remote_ip)) {
3455                         err = vxlan_fdb_create(vxlan, all_zeros_mac,
3456                                                &dst->remote_ip,
3457                                                NUD_REACHABLE | NUD_PERMANENT,
3458                                                NLM_F_CREATE | NLM_F_APPEND,
3459                                                vxlan->cfg.dst_port,
3460                                                dst->remote_vni,
3461                                                dst->remote_vni,
3462                                                dst->remote_ifindex,
3463                                                NTF_SELF);
3464                         if (err) {
3465                                 spin_unlock_bh(&vxlan->hash_lock);
3466                                 return err;
3467                         }
3468                 }
3469                 spin_unlock_bh(&vxlan->hash_lock);
3470         }
3471
3472         return 0;
3473 }
3474
3475 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3476 {
3477         struct vxlan_dev *vxlan = netdev_priv(dev);
3478
3479         vxlan_flush(vxlan, true);
3480
3481         gro_cells_destroy(&vxlan->gro_cells);
3482         list_del(&vxlan->next);
3483         unregister_netdevice_queue(dev, head);
3484 }
3485
3486 static size_t vxlan_get_size(const struct net_device *dev)
3487 {
3488
3489         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3490                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3491                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3492                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3493                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3494                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3495                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3496                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3497                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3498                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3499                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3500                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3501                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3502                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3503                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3504                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3505                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3506                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3507                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3508                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3509                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3510                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3511                 0;
3512 }
3513
3514 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3515 {
3516         const struct vxlan_dev *vxlan = netdev_priv(dev);
3517         const struct vxlan_rdst *dst = &vxlan->default_dst;
3518         struct ifla_vxlan_port_range ports = {
3519                 .low =  htons(vxlan->cfg.port_min),
3520                 .high = htons(vxlan->cfg.port_max),
3521         };
3522
3523         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3524                 goto nla_put_failure;
3525
3526         if (!vxlan_addr_any(&dst->remote_ip)) {
3527                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3528                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3529                                             dst->remote_ip.sin.sin_addr.s_addr))
3530                                 goto nla_put_failure;
3531 #if IS_ENABLED(CONFIG_IPV6)
3532                 } else {
3533                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3534                                              &dst->remote_ip.sin6.sin6_addr))
3535                                 goto nla_put_failure;
3536 #endif
3537                 }
3538         }
3539
3540         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3541                 goto nla_put_failure;
3542
3543         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3544                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3545                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3546                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3547                                 goto nla_put_failure;
3548 #if IS_ENABLED(CONFIG_IPV6)
3549                 } else {
3550                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3551                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3552                                 goto nla_put_failure;
3553 #endif
3554                 }
3555         }
3556
3557         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3558             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3559             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3560             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3561                         !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
3562             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3563                         !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
3564             nla_put_u8(skb, IFLA_VXLAN_RSC,
3565                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
3566             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3567                         !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
3568             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3569                         !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
3570             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3571                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
3572             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3573             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3574             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3575             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3576                         !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3577             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3578                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3579             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3580                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3581             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3582                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
3583             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3584                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
3585                 goto nla_put_failure;
3586
3587         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3588                 goto nla_put_failure;
3589
3590         if (vxlan->cfg.flags & VXLAN_F_GBP &&
3591             nla_put_flag(skb, IFLA_VXLAN_GBP))
3592                 goto nla_put_failure;
3593
3594         if (vxlan->cfg.flags & VXLAN_F_GPE &&
3595             nla_put_flag(skb, IFLA_VXLAN_GPE))
3596                 goto nla_put_failure;
3597
3598         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3599             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3600                 goto nla_put_failure;
3601
3602         return 0;
3603
3604 nla_put_failure:
3605         return -EMSGSIZE;
3606 }
3607
3608 static struct net *vxlan_get_link_net(const struct net_device *dev)
3609 {
3610         struct vxlan_dev *vxlan = netdev_priv(dev);
3611
3612         return vxlan->net;
3613 }
3614
3615 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3616         .kind           = "vxlan",
3617         .maxtype        = IFLA_VXLAN_MAX,
3618         .policy         = vxlan_policy,
3619         .priv_size      = sizeof(struct vxlan_dev),
3620         .setup          = vxlan_setup,
3621         .validate       = vxlan_validate,
3622         .newlink        = vxlan_newlink,
3623         .changelink     = vxlan_changelink,
3624         .dellink        = vxlan_dellink,
3625         .get_size       = vxlan_get_size,
3626         .fill_info      = vxlan_fill_info,
3627         .get_link_net   = vxlan_get_link_net,
3628 };
3629
3630 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3631                                     u8 name_assign_type,
3632                                     struct vxlan_config *conf)
3633 {
3634         struct nlattr *tb[IFLA_MAX + 1];
3635         struct net_device *dev;
3636         int err;
3637
3638         memset(&tb, 0, sizeof(tb));
3639
3640         dev = rtnl_create_link(net, name, name_assign_type,
3641                                &vxlan_link_ops, tb);
3642         if (IS_ERR(dev))
3643                 return dev;
3644
3645         err = __vxlan_dev_create(net, dev, conf, NULL);
3646         if (err < 0) {
3647                 free_netdev(dev);
3648                 return ERR_PTR(err);
3649         }
3650
3651         err = rtnl_configure_link(dev, NULL);
3652         if (err < 0) {
3653                 LIST_HEAD(list_kill);
3654
3655                 vxlan_dellink(dev, &list_kill);
3656                 unregister_netdevice_many(&list_kill);
3657                 return ERR_PTR(err);
3658         }
3659
3660         return dev;
3661 }
3662 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3663
3664 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3665                                              struct net_device *dev)
3666 {
3667         struct vxlan_dev *vxlan, *next;
3668         LIST_HEAD(list_kill);
3669
3670         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3671                 struct vxlan_rdst *dst = &vxlan->default_dst;
3672
3673                 /* In case we created vxlan device with carrier
3674                  * and we loose the carrier due to module unload
3675                  * we also need to remove vxlan device. In other
3676                  * cases, it's not necessary and remote_ifindex
3677                  * is 0 here, so no matches.
3678                  */
3679                 if (dst->remote_ifindex == dev->ifindex)
3680                         vxlan_dellink(vxlan->dev, &list_kill);
3681         }
3682
3683         unregister_netdevice_many(&list_kill);
3684 }
3685
3686 static int vxlan_netdevice_event(struct notifier_block *unused,
3687                                  unsigned long event, void *ptr)
3688 {
3689         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3690         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3691
3692         if (event == NETDEV_UNREGISTER) {
3693                 vxlan_offload_rx_ports(dev, false);
3694                 vxlan_handle_lowerdev_unregister(vn, dev);
3695         } else if (event == NETDEV_REGISTER) {
3696                 vxlan_offload_rx_ports(dev, true);
3697         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
3698                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
3699                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
3700         }
3701
3702         return NOTIFY_DONE;
3703 }
3704
3705 static struct notifier_block vxlan_notifier_block __read_mostly = {
3706         .notifier_call = vxlan_netdevice_event,
3707 };
3708
3709 static __net_init int vxlan_init_net(struct net *net)
3710 {
3711         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3712         unsigned int h;
3713
3714         INIT_LIST_HEAD(&vn->vxlan_list);
3715         spin_lock_init(&vn->sock_lock);
3716
3717         for (h = 0; h < PORT_HASH_SIZE; ++h)
3718                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3719
3720         return 0;
3721 }
3722
3723 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
3724 {
3725         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3726         struct vxlan_dev *vxlan, *next;
3727         struct net_device *dev, *aux;
3728         unsigned int h;
3729
3730         for_each_netdev_safe(net, dev, aux)
3731                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3732                         unregister_netdevice_queue(dev, head);
3733
3734         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3735                 /* If vxlan->dev is in the same netns, it has already been added
3736                  * to the list by the previous loop.
3737                  */
3738                 if (!net_eq(dev_net(vxlan->dev), net)) {
3739                         gro_cells_destroy(&vxlan->gro_cells);
3740                         unregister_netdevice_queue(vxlan->dev, head);
3741                 }
3742         }
3743
3744         for (h = 0; h < PORT_HASH_SIZE; ++h)
3745                 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
3746 }
3747
3748 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
3749 {
3750         struct net *net;
3751         LIST_HEAD(list);
3752
3753         rtnl_lock();
3754         list_for_each_entry(net, net_list, exit_list)
3755                 vxlan_destroy_tunnels(net, &list);
3756
3757         unregister_netdevice_many(&list);
3758         rtnl_unlock();
3759 }
3760
3761 static struct pernet_operations vxlan_net_ops = {
3762         .init = vxlan_init_net,
3763         .exit_batch = vxlan_exit_batch_net,
3764         .id   = &vxlan_net_id,
3765         .size = sizeof(struct vxlan_net),
3766 };
3767
3768 static int __init vxlan_init_module(void)
3769 {
3770         int rc;
3771
3772         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3773
3774         rc = register_pernet_subsys(&vxlan_net_ops);
3775         if (rc)
3776                 goto out1;
3777
3778         rc = register_netdevice_notifier(&vxlan_notifier_block);
3779         if (rc)
3780                 goto out2;
3781
3782         rc = rtnl_link_register(&vxlan_link_ops);
3783         if (rc)
3784                 goto out3;
3785
3786         return 0;
3787 out3:
3788         unregister_netdevice_notifier(&vxlan_notifier_block);
3789 out2:
3790         unregister_pernet_subsys(&vxlan_net_ops);
3791 out1:
3792         return rc;
3793 }
3794 late_initcall(vxlan_init_module);
3795
3796 static void __exit vxlan_cleanup_module(void)
3797 {
3798         rtnl_link_unregister(&vxlan_link_ops);
3799         unregister_netdevice_notifier(&vxlan_notifier_block);
3800         unregister_pernet_subsys(&vxlan_net_ops);
3801         /* rcu_barrier() is called by netns */
3802 }
3803 module_exit(vxlan_cleanup_module);
3804
3805 MODULE_LICENSE("GPL");
3806 MODULE_VERSION(VXLAN_VERSION);
3807 MODULE_AUTHOR("Stephen Hemminger <[email protected]>");
3808 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3809 MODULE_ALIAS_RTNL_LINK("vxlan");