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 -ENOMEM;
 528
 529         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 530                 kfree(rd);
 531                 return -ENOMEM;
 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_flush_final_remcsum(skb, pp, flush, &grc);
 627
 628         return pp;
 629 }
 630
 631 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 632 {
 633         /* Sets 'skb->inner_mac_header' since we are always called with
 634          * 'skb->encapsulation' set.
 635          */
 636         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 637 }
 638
 639 static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan,
 640                                          const u8 *mac, __u16 state,
 641                                          __be32 src_vni, __u8 ndm_flags)
 642 {
 643         struct vxlan_fdb *f;
 644
 645         f = kmalloc(sizeof(*f), GFP_ATOMIC);
 646         if (!f)
 647                 return NULL;
 648         f->state = state;
 649         f->flags = ndm_flags;
 650         f->updated = f->used = jiffies;
 651         f->vni = src_vni;
 652         INIT_LIST_HEAD(&f->remotes);
 653         memcpy(f->eth_addr, mac, ETH_ALEN);
 654
 655         return f;
 656 }
 657
 658 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 659                             const u8 *mac, union vxlan_addr *ip,
 660                             __u16 state, __be16 port, __be32 src_vni,
 661                             __be32 vni, __u32 ifindex, __u8 ndm_flags,
 662                             struct vxlan_fdb **fdb)
 663 {
 664         struct vxlan_rdst *rd = NULL;
 665         struct vxlan_fdb *f;
 666         int rc;
 667
 668         if (vxlan->cfg.addrmax &&
 669             vxlan->addrcnt >= vxlan->cfg.addrmax)
 670                 return -ENOSPC;
 671
 672         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 673         f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags);
 674         if (!f)
 675                 return -ENOMEM;
 676
 677         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 678         if (rc < 0) {
 679                 kfree(f);
 680                 return rc;
 681         }
 682
 683         ++vxlan->addrcnt;
 684         hlist_add_head_rcu(&f->hlist,
 685                            vxlan_fdb_head(vxlan, mac, src_vni));
 686
 687         *fdb = f;
 688
 689         return 0;
 690 }
 691
 692 /* Add new entry to forwarding table -- assumes lock held */
 693 static int vxlan_fdb_update(struct vxlan_dev *vxlan,
 694                             const u8 *mac, union vxlan_addr *ip,
 695                             __u16 state, __u16 flags,
 696                             __be16 port, __be32 src_vni, __be32 vni,
 697                             __u32 ifindex, __u8 ndm_flags)
 698 {
 699         struct vxlan_rdst *rd = NULL;
 700         struct vxlan_fdb *f;
 701         int notify = 0;
 702         int rc;
 703
 704         f = __vxlan_find_mac(vxlan, mac, src_vni);
 705         if (f) {
 706                 if (flags & NLM_F_EXCL) {
 707                         netdev_dbg(vxlan->dev,
 708                                    "lost race to create %pM\n", mac);
 709                         return -EEXIST;
 710                 }
 711                 if (f->state != state) {
 712                         f->state = state;
 713                         f->updated = jiffies;
 714                         notify = 1;
 715                 }
 716                 if (f->flags != ndm_flags) {
 717                         f->flags = ndm_flags;
 718                         f->updated = jiffies;
 719                         notify = 1;
 720                 }
 721                 if ((flags & NLM_F_REPLACE)) {
 722                         /* Only change unicasts */
 723                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 724                              is_zero_ether_addr(f->eth_addr))) {
 725                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 726                                                            ifindex);
 727                         } else
 728                                 return -EOPNOTSUPP;
 729                 }
 730                 if ((flags & NLM_F_APPEND) &&
 731                     (is_multicast_ether_addr(f->eth_addr) ||
 732                      is_zero_ether_addr(f->eth_addr))) {
 733                         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 734
 735                         if (rc < 0)
 736                                 return rc;
 737                         notify |= rc;
 738                 }
 739         } else {
 740                 if (!(flags & NLM_F_CREATE))
 741                         return -ENOENT;
 742
 743                 /* Disallow replace to add a multicast entry */
 744                 if ((flags & NLM_F_REPLACE) &&
 745                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 746                         return -EOPNOTSUPP;
 747
 748                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 749                 rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni,
 750                                       vni, ifindex, ndm_flags, &f);
 751                 if (rc < 0)
 752                         return rc;
 753                 notify = 1;
 754         }
 755
 756         if (notify) {
 757                 if (rd == NULL)
 758                         rd = first_remote_rtnl(f);
 759                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 760         }
 761
 762         return 0;
 763 }
 764
 765 static void vxlan_fdb_free(struct rcu_head *head)
 766 {
 767         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 768         struct vxlan_rdst *rd, *nd;
 769
 770         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 771                 dst_cache_destroy(&rd->dst_cache);
 772                 kfree(rd);
 773         }
 774         kfree(f);
 775 }
 776
 777 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 778                               bool do_notify)
 779 {
 780         netdev_dbg(vxlan->dev,
 781                     "delete %pM\n", f->eth_addr);
 782
 783         --vxlan->addrcnt;
 784         if (do_notify)
 785                 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 786
 787         hlist_del_rcu(&f->hlist);
 788         call_rcu(&f->rcu, vxlan_fdb_free);
 789 }
 790
 791 static void vxlan_dst_free(struct rcu_head *head)
 792 {
 793         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 794
 795         dst_cache_destroy(&rd->dst_cache);
 796         kfree(rd);
 797 }
 798
 799 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 800                                   struct vxlan_rdst *rd)
 801 {
 802         list_del_rcu(&rd->list);
 803         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 804         call_rcu(&rd->rcu, vxlan_dst_free);
 805 }
 806
 807 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 808                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
 809                            __be32 *vni, u32 *ifindex)
 810 {
 811         struct net *net = dev_net(vxlan->dev);
 812         int err;
 813
 814         if (tb[NDA_DST]) {
 815                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 816                 if (err)
 817                         return err;
 818         } else {
 819                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 820                 if (remote->sa.sa_family == AF_INET) {
 821                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 822                         ip->sa.sa_family = AF_INET;
 823 #if IS_ENABLED(CONFIG_IPV6)
 824                 } else {
 825                         ip->sin6.sin6_addr = in6addr_any;
 826                         ip->sa.sa_family = AF_INET6;
 827 #endif
 828                 }
 829         }
 830
 831         if (tb[NDA_PORT]) {
 832                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 833                         return -EINVAL;
 834                 *port = nla_get_be16(tb[NDA_PORT]);
 835         } else {
 836                 *port = vxlan->cfg.dst_port;
 837         }
 838
 839         if (tb[NDA_VNI]) {
 840                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 841                         return -EINVAL;
 842                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 843         } else {
 844                 *vni = vxlan->default_dst.remote_vni;
 845         }
 846
 847         if (tb[NDA_SRC_VNI]) {
 848                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
 849                         return -EINVAL;
 850                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
 851         } else {
 852                 *src_vni = vxlan->default_dst.remote_vni;
 853         }
 854
 855         if (tb[NDA_IFINDEX]) {
 856                 struct net_device *tdev;
 857
 858                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 859                         return -EINVAL;
 860                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 861                 tdev = __dev_get_by_index(net, *ifindex);
 862                 if (!tdev)
 863                         return -EADDRNOTAVAIL;
 864         } else {
 865                 *ifindex = 0;
 866         }
 867
 868         return 0;
 869 }
 870
 871 /* Add static entry (via netlink) */
 872 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 873                          struct net_device *dev,
 874                          const unsigned char *addr, u16 vid, u16 flags)
 875 {
 876         struct vxlan_dev *vxlan = netdev_priv(dev);
 877         /* struct net *net = dev_net(vxlan->dev); */
 878         union vxlan_addr ip;
 879         __be16 port;
 880         __be32 src_vni, vni;
 881         u32 ifindex;
 882         int err;
 883
 884         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 885                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 886                         ndm->ndm_state);
 887                 return -EINVAL;
 888         }
 889
 890         if (tb[NDA_DST] == NULL)
 891                 return -EINVAL;
 892
 893         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 894         if (err)
 895                 return err;
 896
 897         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 898                 return -EAFNOSUPPORT;
 899
 900         spin_lock_bh(&vxlan->hash_lock);
 901         err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags,
 902                                port, src_vni, vni, ifindex, ndm->ndm_flags);
 903         spin_unlock_bh(&vxlan->hash_lock);
 904
 905         return err;
 906 }
 907
 908 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
 909                               const unsigned char *addr, union vxlan_addr ip,
 910                               __be16 port, __be32 src_vni, u32 vni, u32 ifindex,
 911                               u16 vid)
 912 {
 913         struct vxlan_fdb *f;
 914         struct vxlan_rdst *rd = NULL;
 915         int err = -ENOENT;
 916
 917         f = vxlan_find_mac(vxlan, addr, src_vni);
 918         if (!f)
 919                 return err;
 920
 921         if (!vxlan_addr_any(&ip)) {
 922                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 923                 if (!rd)
 924                         goto out;
 925         }
 926
 927         /* remove a destination if it's not the only one on the list,
 928          * otherwise destroy the fdb entry
 929          */
 930         if (rd && !list_is_singular(&f->remotes)) {
 931                 vxlan_fdb_dst_destroy(vxlan, f, rd);
 932                 goto out;
 933         }
 934
 935         vxlan_fdb_destroy(vxlan, f, true);
 936
 937 out:
 938         return 0;
 939 }
 940
 941 /* Delete entry (via netlink) */
 942 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 943                             struct net_device *dev,
 944                             const unsigned char *addr, u16 vid)
 945 {
 946         struct vxlan_dev *vxlan = netdev_priv(dev);
 947         union vxlan_addr ip;
 948         __be32 src_vni, vni;
 949         __be16 port;
 950         u32 ifindex;
 951         int err;
 952
 953         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 954         if (err)
 955                 return err;
 956
 957         spin_lock_bh(&vxlan->hash_lock);
 958         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
 959                                  vid);
 960         spin_unlock_bh(&vxlan->hash_lock);
 961
 962         return err;
 963 }
 964
 965 /* Dump forwarding table */
 966 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 967                           struct net_device *dev,
 968                           struct net_device *filter_dev, int *idx)
 969 {
 970         struct vxlan_dev *vxlan = netdev_priv(dev);
 971         unsigned int h;
 972         int err = 0;
 973
 974         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 975                 struct vxlan_fdb *f;
 976
 977                 rcu_read_lock();
 978                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 979                         struct vxlan_rdst *rd;
 980
 981                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 982                                 if (*idx < cb->args[2])
 983                                         goto skip;
 984
 985                                 err = vxlan_fdb_info(skb, vxlan, f,
 986                                                      NETLINK_CB(cb->skb).portid,
 987                                                      cb->nlh->nlmsg_seq,
 988                                                      RTM_NEWNEIGH,
 989                                                      NLM_F_MULTI, rd);
 990                                 if (err < 0) {
 991                                         rcu_read_unlock();
 992                                         goto out;
 993                                 }
 994 skip:
 995                                 *idx += 1;
 996                         }
 997                 }
 998                 rcu_read_unlock();
 999         }
1000 out:
1001         return err;
1002 }
1003
1004 /* Watch incoming packets to learn mapping between Ethernet address
1005  * and Tunnel endpoint.
1006  * Return true if packet is bogus and should be dropped.
1007  */
1008 static bool vxlan_snoop(struct net_device *dev,
1009                         union vxlan_addr *src_ip, const u8 *src_mac,
1010                         u32 src_ifindex, __be32 vni)
1011 {
1012         struct vxlan_dev *vxlan = netdev_priv(dev);
1013         struct vxlan_fdb *f;
1014         u32 ifindex = 0;
1015
1016 #if IS_ENABLED(CONFIG_IPV6)
1017         if (src_ip->sa.sa_family == AF_INET6 &&
1018             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1019                 ifindex = src_ifindex;
1020 #endif
1021
1022         f = vxlan_find_mac(vxlan, src_mac, vni);
1023         if (likely(f)) {
1024                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1025
1026                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1027                            rdst->remote_ifindex == ifindex))
1028                         return false;
1029
1030                 /* Don't migrate static entries, drop packets */
1031                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1032                         return true;
1033
1034                 if (net_ratelimit())
1035                         netdev_info(dev,
1036                                     "%pM migrated from %pIS to %pIS\n",
1037                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1038
1039                 rdst->remote_ip = *src_ip;
1040                 f->updated = jiffies;
1041                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1042         } else {
1043                 /* learned new entry */
1044                 spin_lock(&vxlan->hash_lock);
1045
1046                 /* close off race between vxlan_flush and incoming packets */
1047                 if (netif_running(dev))
1048                         vxlan_fdb_update(vxlan, src_mac, src_ip,
1049                                          NUD_REACHABLE,
1050                                          NLM_F_EXCL|NLM_F_CREATE,
1051                                          vxlan->cfg.dst_port,
1052                                          vni,
1053                                          vxlan->default_dst.remote_vni,
1054                                          ifindex, NTF_SELF);
1055                 spin_unlock(&vxlan->hash_lock);
1056         }
1057
1058         return false;
1059 }
1060
1061 /* See if multicast group is already in use by other ID */
1062 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1063 {
1064         struct vxlan_dev *vxlan;
1065         struct vxlan_sock *sock4;
1066 #if IS_ENABLED(CONFIG_IPV6)
1067         struct vxlan_sock *sock6;
1068 #endif
1069         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1070
1071         sock4 = rtnl_dereference(dev->vn4_sock);
1072
1073         /* The vxlan_sock is only used by dev, leaving group has
1074          * no effect on other vxlan devices.
1075          */
1076         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1077                 return false;
1078 #if IS_ENABLED(CONFIG_IPV6)
1079         sock6 = rtnl_dereference(dev->vn6_sock);
1080         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1081                 return false;
1082 #endif
1083
1084         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1085                 if (!netif_running(vxlan->dev) || vxlan == dev)
1086                         continue;
1087
1088                 if (family == AF_INET &&
1089                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1090                         continue;
1091 #if IS_ENABLED(CONFIG_IPV6)
1092                 if (family == AF_INET6 &&
1093                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1094                         continue;
1095 #endif
1096
1097                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1098                                       &dev->default_dst.remote_ip))
1099                         continue;
1100
1101                 if (vxlan->default_dst.remote_ifindex !=
1102                     dev->default_dst.remote_ifindex)
1103                         continue;
1104
1105                 return true;
1106         }
1107
1108         return false;
1109 }
1110
1111 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1112 {
1113         struct vxlan_net *vn;
1114
1115         if (!vs)
1116                 return false;
1117         if (!refcount_dec_and_test(&vs->refcnt))
1118                 return false;
1119
1120         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1121         spin_lock(&vn->sock_lock);
1122         hlist_del_rcu(&vs->hlist);
1123         udp_tunnel_notify_del_rx_port(vs->sock,
1124                                       (vs->flags & VXLAN_F_GPE) ?
1125                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1126                                       UDP_TUNNEL_TYPE_VXLAN);
1127         spin_unlock(&vn->sock_lock);
1128
1129         return true;
1130 }
1131
1132 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1133 {
1134         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1135 #if IS_ENABLED(CONFIG_IPV6)
1136         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1137
1138         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1139 #endif
1140
1141         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1142         synchronize_net();
1143
1144         vxlan_vs_del_dev(vxlan);
1145
1146         if (__vxlan_sock_release_prep(sock4)) {
1147                 udp_tunnel_sock_release(sock4->sock);
1148                 kfree(sock4);
1149         }
1150
1151 #if IS_ENABLED(CONFIG_IPV6)
1152         if (__vxlan_sock_release_prep(sock6)) {
1153                 udp_tunnel_sock_release(sock6->sock);
1154                 kfree(sock6);
1155         }
1156 #endif
1157 }
1158
1159 /* Update multicast group membership when first VNI on
1160  * multicast address is brought up
1161  */
1162 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1163 {
1164         struct sock *sk;
1165         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1166         int ifindex = vxlan->default_dst.remote_ifindex;
1167         int ret = -EINVAL;
1168
1169         if (ip->sa.sa_family == AF_INET) {
1170                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1171                 struct ip_mreqn mreq = {
1172                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1173                         .imr_ifindex            = ifindex,
1174                 };
1175
1176                 sk = sock4->sock->sk;
1177                 lock_sock(sk);
1178                 ret = ip_mc_join_group(sk, &mreq);
1179                 release_sock(sk);
1180 #if IS_ENABLED(CONFIG_IPV6)
1181         } else {
1182                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1183
1184                 sk = sock6->sock->sk;
1185                 lock_sock(sk);
1186                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1187                                                    &ip->sin6.sin6_addr);
1188                 release_sock(sk);
1189 #endif
1190         }
1191
1192         return ret;
1193 }
1194
1195 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1196 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1197 {
1198         struct sock *sk;
1199         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1200         int ifindex = vxlan->default_dst.remote_ifindex;
1201         int ret = -EINVAL;
1202
1203         if (ip->sa.sa_family == AF_INET) {
1204                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1205                 struct ip_mreqn mreq = {
1206                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1207                         .imr_ifindex            = ifindex,
1208                 };
1209
1210                 sk = sock4->sock->sk;
1211                 lock_sock(sk);
1212                 ret = ip_mc_leave_group(sk, &mreq);
1213                 release_sock(sk);
1214 #if IS_ENABLED(CONFIG_IPV6)
1215         } else {
1216                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1217
1218                 sk = sock6->sock->sk;
1219                 lock_sock(sk);
1220                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1221                                                    &ip->sin6.sin6_addr);
1222                 release_sock(sk);
1223 #endif
1224         }
1225
1226         return ret;
1227 }
1228
1229 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1230                           struct sk_buff *skb, u32 vxflags)
1231 {
1232         size_t start, offset;
1233
1234         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1235                 goto out;
1236
1237         start = vxlan_rco_start(unparsed->vx_vni);
1238         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1239
1240         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1241                 return false;
1242
1243         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1244                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1245 out:
1246         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1247         unparsed->vx_vni &= VXLAN_VNI_MASK;
1248         return true;
1249 }
1250
1251 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1252                                 struct sk_buff *skb, u32 vxflags,
1253                                 struct vxlan_metadata *md)
1254 {
1255         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1256         struct metadata_dst *tun_dst;
1257
1258         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1259                 goto out;
1260
1261         md->gbp = ntohs(gbp->policy_id);
1262
1263         tun_dst = (struct metadata_dst *)skb_dst(skb);
1264         if (tun_dst) {
1265                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1266                 tun_dst->u.tun_info.options_len = sizeof(*md);
1267         }
1268         if (gbp->dont_learn)
1269                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1270
1271         if (gbp->policy_applied)
1272                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1273
1274         /* In flow-based mode, GBP is carried in dst_metadata */
1275         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1276                 skb->mark = md->gbp;
1277 out:
1278         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1279 }
1280
1281 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1282                                 __be16 *protocol,
1283                                 struct sk_buff *skb, u32 vxflags)
1284 {
1285         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1286
1287         /* Need to have Next Protocol set for interfaces in GPE mode. */
1288         if (!gpe->np_applied)
1289                 return false;
1290         /* "The initial version is 0. If a receiver does not support the
1291          * version indicated it MUST drop the packet.
1292          */
1293         if (gpe->version != 0)
1294                 return false;
1295         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1296          * processing MUST occur." However, we don't implement OAM
1297          * processing, thus drop the packet.
1298          */
1299         if (gpe->oam_flag)
1300                 return false;
1301
1302         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1303         if (!*protocol)
1304                 return false;
1305
1306         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1307         return true;
1308 }
1309
1310 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1311                           struct vxlan_sock *vs,
1312                           struct sk_buff *skb, __be32 vni)
1313 {
1314         union vxlan_addr saddr;
1315         u32 ifindex = skb->dev->ifindex;
1316
1317         skb_reset_mac_header(skb);
1318         skb->protocol = eth_type_trans(skb, vxlan->dev);
1319         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1320
1321         /* Ignore packet loops (and multicast echo) */
1322         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1323                 return false;
1324
1325         /* Get address from the outer IP header */
1326         if (vxlan_get_sk_family(vs) == AF_INET) {
1327                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1328                 saddr.sa.sa_family = AF_INET;
1329 #if IS_ENABLED(CONFIG_IPV6)
1330         } else {
1331                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1332                 saddr.sa.sa_family = AF_INET6;
1333 #endif
1334         }
1335
1336         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1337             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1338                 return false;
1339
1340         return true;
1341 }
1342
1343 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1344                                   struct sk_buff *skb)
1345 {
1346         int err = 0;
1347
1348         if (vxlan_get_sk_family(vs) == AF_INET)
1349                 err = IP_ECN_decapsulate(oiph, skb);
1350 #if IS_ENABLED(CONFIG_IPV6)
1351         else
1352                 err = IP6_ECN_decapsulate(oiph, skb);
1353 #endif
1354
1355         if (unlikely(err) && log_ecn_error) {
1356                 if (vxlan_get_sk_family(vs) == AF_INET)
1357                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1358                                              &((struct iphdr *)oiph)->saddr,
1359                                              ((struct iphdr *)oiph)->tos);
1360                 else
1361                         net_info_ratelimited("non-ECT from %pI6\n",
1362                                              &((struct ipv6hdr *)oiph)->saddr);
1363         }
1364         return err <= 1;
1365 }
1366
1367 /* Callback from net/ipv4/udp.c to receive packets */
1368 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1369 {
1370         struct pcpu_sw_netstats *stats;
1371         struct vxlan_dev *vxlan;
1372         struct vxlan_sock *vs;
1373         struct vxlanhdr unparsed;
1374         struct vxlan_metadata _md;
1375         struct vxlan_metadata *md = &_md;
1376         __be16 protocol = htons(ETH_P_TEB);
1377         bool raw_proto = false;
1378         void *oiph;
1379         __be32 vni = 0;
1380
1381         /* Need UDP and VXLAN header to be present */
1382         if (!pskb_may_pull(skb, VXLAN_HLEN))
1383                 goto drop;
1384
1385         unparsed = *vxlan_hdr(skb);
1386         /* VNI flag always required to be set */
1387         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1388                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1389                            ntohl(vxlan_hdr(skb)->vx_flags),
1390                            ntohl(vxlan_hdr(skb)->vx_vni));
1391                 /* Return non vxlan pkt */
1392                 goto drop;
1393         }
1394         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1395         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1396
1397         vs = rcu_dereference_sk_user_data(sk);
1398         if (!vs)
1399                 goto drop;
1400
1401         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1402
1403         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1404         if (!vxlan)
1405                 goto drop;
1406
1407         /* For backwards compatibility, only allow reserved fields to be
1408          * used by VXLAN extensions if explicitly requested.
1409          */
1410         if (vs->flags & VXLAN_F_GPE) {
1411                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1412                         goto drop;
1413                 raw_proto = true;
1414         }
1415
1416         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1417                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1418                         goto drop;
1419
1420         if (vxlan_collect_metadata(vs)) {
1421                 struct metadata_dst *tun_dst;
1422
1423                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1424                                          key32_to_tunnel_id(vni), sizeof(*md));
1425
1426                 if (!tun_dst)
1427                         goto drop;
1428
1429                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1430
1431                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1432         } else {
1433                 memset(md, 0, sizeof(*md));
1434         }
1435
1436         if (vs->flags & VXLAN_F_REMCSUM_RX)
1437                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1438                         goto drop;
1439         if (vs->flags & VXLAN_F_GBP)
1440                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1441         /* Note that GBP and GPE can never be active together. This is
1442          * ensured in vxlan_dev_configure.
1443          */
1444
1445         if (unparsed.vx_flags || unparsed.vx_vni) {
1446                 /* If there are any unprocessed flags remaining treat
1447                  * this as a malformed packet. This behavior diverges from
1448                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1449                  * in reserved fields are to be ignored. The approach here
1450                  * maintains compatibility with previous stack code, and also
1451                  * is more robust and provides a little more security in
1452                  * adding extensions to VXLAN.
1453                  */
1454                 goto drop;
1455         }
1456
1457         if (!raw_proto) {
1458                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1459                         goto drop;
1460         } else {
1461                 skb_reset_mac_header(skb);
1462                 skb->dev = vxlan->dev;
1463                 skb->pkt_type = PACKET_HOST;
1464         }
1465
1466         oiph = skb_network_header(skb);
1467         skb_reset_network_header(skb);
1468
1469         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1470                 ++vxlan->dev->stats.rx_frame_errors;
1471                 ++vxlan->dev->stats.rx_errors;
1472                 goto drop;
1473         }
1474
1475         rcu_read_lock();
1476
1477         if (unlikely(!(vxlan->dev->flags & IFF_UP))) {
1478                 rcu_read_unlock();
1479                 atomic_long_inc(&vxlan->dev->rx_dropped);
1480                 goto drop;
1481         }
1482
1483         stats = this_cpu_ptr(vxlan->dev->tstats);
1484         u64_stats_update_begin(&stats->syncp);
1485         stats->rx_packets++;
1486         stats->rx_bytes += skb->len;
1487         u64_stats_update_end(&stats->syncp);
1488
1489         gro_cells_receive(&vxlan->gro_cells, skb);
1490
1491         rcu_read_unlock();
1492
1493         return 0;
1494
1495 drop:
1496         /* Consume bad packet */
1497         kfree_skb(skb);
1498         return 0;
1499 }
1500
1501 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1502 {
1503         struct vxlan_dev *vxlan = netdev_priv(dev);
1504         struct arphdr *parp;
1505         u8 *arpptr, *sha;
1506         __be32 sip, tip;
1507         struct neighbour *n;
1508
1509         if (dev->flags & IFF_NOARP)
1510                 goto out;
1511
1512         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1513                 dev->stats.tx_dropped++;
1514                 goto out;
1515         }
1516         parp = arp_hdr(skb);
1517
1518         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1519              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1520             parp->ar_pro != htons(ETH_P_IP) ||
1521             parp->ar_op != htons(ARPOP_REQUEST) ||
1522             parp->ar_hln != dev->addr_len ||
1523             parp->ar_pln != 4)
1524                 goto out;
1525         arpptr = (u8 *)parp + sizeof(struct arphdr);
1526         sha = arpptr;
1527         arpptr += dev->addr_len;        /* sha */
1528         memcpy(&sip, arpptr, sizeof(sip));
1529         arpptr += sizeof(sip);
1530         arpptr += dev->addr_len;        /* tha */
1531         memcpy(&tip, arpptr, sizeof(tip));
1532
1533         if (ipv4_is_loopback(tip) ||
1534             ipv4_is_multicast(tip))
1535                 goto out;
1536
1537         n = neigh_lookup(&arp_tbl, &tip, dev);
1538
1539         if (n) {
1540                 struct vxlan_fdb *f;
1541                 struct sk_buff  *reply;
1542
1543                 if (!(n->nud_state & NUD_CONNECTED)) {
1544                         neigh_release(n);
1545                         goto out;
1546                 }
1547
1548                 f = vxlan_find_mac(vxlan, n->ha, vni);
1549                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1550                         /* bridge-local neighbor */
1551                         neigh_release(n);
1552                         goto out;
1553                 }
1554
1555                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1556                                 n->ha, sha);
1557
1558                 neigh_release(n);
1559
1560                 if (reply == NULL)
1561                         goto out;
1562
1563                 skb_reset_mac_header(reply);
1564                 __skb_pull(reply, skb_network_offset(reply));
1565                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1566                 reply->pkt_type = PACKET_HOST;
1567
1568                 if (netif_rx_ni(reply) == NET_RX_DROP)
1569                         dev->stats.rx_dropped++;
1570         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1571                 union vxlan_addr ipa = {
1572                         .sin.sin_addr.s_addr = tip,
1573                         .sin.sin_family = AF_INET,
1574                 };
1575
1576                 vxlan_ip_miss(dev, &ipa);
1577         }
1578 out:
1579         consume_skb(skb);
1580         return NETDEV_TX_OK;
1581 }
1582
1583 #if IS_ENABLED(CONFIG_IPV6)
1584 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1585         struct neighbour *n, bool isrouter)
1586 {
1587         struct net_device *dev = request->dev;
1588         struct sk_buff *reply;
1589         struct nd_msg *ns, *na;
1590         struct ipv6hdr *pip6;
1591         u8 *daddr;
1592         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1593         int ns_olen;
1594         int i, len;
1595
1596         if (dev == NULL || !pskb_may_pull(request, request->len))
1597                 return NULL;
1598
1599         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1600                 sizeof(*na) + na_olen + dev->needed_tailroom;
1601         reply = alloc_skb(len, GFP_ATOMIC);
1602         if (reply == NULL)
1603                 return NULL;
1604
1605         reply->protocol = htons(ETH_P_IPV6);
1606         reply->dev = dev;
1607         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1608         skb_push(reply, sizeof(struct ethhdr));
1609         skb_reset_mac_header(reply);
1610
1611         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1612
1613         daddr = eth_hdr(request)->h_source;
1614         ns_olen = request->len - skb_network_offset(request) -
1615                 sizeof(struct ipv6hdr) - sizeof(*ns);
1616         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1617                 if (!ns->opt[i + 1]) {
1618                         kfree_skb(reply);
1619                         return NULL;
1620                 }
1621                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1622                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1623                         break;
1624                 }
1625         }
1626
1627         /* Ethernet header */
1628         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1629         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1630         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1631         reply->protocol = htons(ETH_P_IPV6);
1632
1633         skb_pull(reply, sizeof(struct ethhdr));
1634         skb_reset_network_header(reply);
1635         skb_put(reply, sizeof(struct ipv6hdr));
1636
1637         /* IPv6 header */
1638
1639         pip6 = ipv6_hdr(reply);
1640         memset(pip6, 0, sizeof(struct ipv6hdr));
1641         pip6->version = 6;
1642         pip6->priority = ipv6_hdr(request)->priority;
1643         pip6->nexthdr = IPPROTO_ICMPV6;
1644         pip6->hop_limit = 255;
1645         pip6->daddr = ipv6_hdr(request)->saddr;
1646         pip6->saddr = *(struct in6_addr *)n->primary_key;
1647
1648         skb_pull(reply, sizeof(struct ipv6hdr));
1649         skb_reset_transport_header(reply);
1650
1651         /* Neighbor Advertisement */
1652         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1653         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1654         na->icmph.icmp6_router = isrouter;
1655         na->icmph.icmp6_override = 1;
1656         na->icmph.icmp6_solicited = 1;
1657         na->target = ns->target;
1658         ether_addr_copy(&na->opt[2], n->ha);
1659         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1660         na->opt[1] = na_olen >> 3;
1661
1662         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1663                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1664                 csum_partial(na, sizeof(*na)+na_olen, 0));
1665
1666         pip6->payload_len = htons(sizeof(*na)+na_olen);
1667
1668         skb_push(reply, sizeof(struct ipv6hdr));
1669
1670         reply->ip_summed = CHECKSUM_UNNECESSARY;
1671
1672         return reply;
1673 }
1674
1675 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1676 {
1677         struct vxlan_dev *vxlan = netdev_priv(dev);
1678         const struct in6_addr *daddr;
1679         const struct ipv6hdr *iphdr;
1680         struct inet6_dev *in6_dev;
1681         struct neighbour *n;
1682         struct nd_msg *msg;
1683
1684         rcu_read_lock();
1685         in6_dev = __in6_dev_get(dev);
1686         if (!in6_dev)
1687                 goto out;
1688
1689         iphdr = ipv6_hdr(skb);
1690         daddr = &iphdr->daddr;
1691         msg = (struct nd_msg *)(iphdr + 1);
1692
1693         if (ipv6_addr_loopback(daddr) ||
1694             ipv6_addr_is_multicast(&msg->target))
1695                 goto out;
1696
1697         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1698
1699         if (n) {
1700                 struct vxlan_fdb *f;
1701                 struct sk_buff *reply;
1702
1703                 if (!(n->nud_state & NUD_CONNECTED)) {
1704                         neigh_release(n);
1705                         goto out;
1706                 }
1707
1708                 f = vxlan_find_mac(vxlan, n->ha, vni);
1709                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1710                         /* bridge-local neighbor */
1711                         neigh_release(n);
1712                         goto out;
1713                 }
1714
1715                 reply = vxlan_na_create(skb, n,
1716                                         !!(f ? f->flags & NTF_ROUTER : 0));
1717
1718                 neigh_release(n);
1719
1720                 if (reply == NULL)
1721                         goto out;
1722
1723                 if (netif_rx_ni(reply) == NET_RX_DROP)
1724                         dev->stats.rx_dropped++;
1725
1726         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1727                 union vxlan_addr ipa = {
1728                         .sin6.sin6_addr = msg->target,
1729                         .sin6.sin6_family = AF_INET6,
1730                 };
1731
1732                 vxlan_ip_miss(dev, &ipa);
1733         }
1734
1735 out:
1736         rcu_read_unlock();
1737         consume_skb(skb);
1738         return NETDEV_TX_OK;
1739 }
1740 #endif
1741
1742 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1743 {
1744         struct vxlan_dev *vxlan = netdev_priv(dev);
1745         struct neighbour *n;
1746
1747         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1748                 return false;
1749
1750         n = NULL;
1751         switch (ntohs(eth_hdr(skb)->h_proto)) {
1752         case ETH_P_IP:
1753         {
1754                 struct iphdr *pip;
1755
1756                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1757                         return false;
1758                 pip = ip_hdr(skb);
1759                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1760                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1761                         union vxlan_addr ipa = {
1762                                 .sin.sin_addr.s_addr = pip->daddr,
1763                                 .sin.sin_family = AF_INET,
1764                         };
1765
1766                         vxlan_ip_miss(dev, &ipa);
1767                         return false;
1768                 }
1769
1770                 break;
1771         }
1772 #if IS_ENABLED(CONFIG_IPV6)
1773         case ETH_P_IPV6:
1774         {
1775                 struct ipv6hdr *pip6;
1776
1777                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1778                         return false;
1779                 pip6 = ipv6_hdr(skb);
1780                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1781                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1782                         union vxlan_addr ipa = {
1783                                 .sin6.sin6_addr = pip6->daddr,
1784                                 .sin6.sin6_family = AF_INET6,
1785                         };
1786
1787                         vxlan_ip_miss(dev, &ipa);
1788                         return false;
1789                 }
1790
1791                 break;
1792         }
1793 #endif
1794         default:
1795                 return false;
1796         }
1797
1798         if (n) {
1799                 bool diff;
1800
1801                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1802                 if (diff) {
1803                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1804                                 dev->addr_len);
1805                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1806                 }
1807                 neigh_release(n);
1808                 return diff;
1809         }
1810
1811         return false;
1812 }
1813
1814 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1815                                 struct vxlan_metadata *md)
1816 {
1817         struct vxlanhdr_gbp *gbp;
1818
1819         if (!md->gbp)
1820                 return;
1821
1822         gbp = (struct vxlanhdr_gbp *)vxh;
1823         vxh->vx_flags |= VXLAN_HF_GBP;
1824
1825         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1826                 gbp->dont_learn = 1;
1827
1828         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1829                 gbp->policy_applied = 1;
1830
1831         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1832 }
1833
1834 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1835                                __be16 protocol)
1836 {
1837         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1838
1839         gpe->np_applied = 1;
1840         gpe->next_protocol = tun_p_from_eth_p(protocol);
1841         if (!gpe->next_protocol)
1842                 return -EPFNOSUPPORT;
1843         return 0;
1844 }
1845
1846 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1847                            int iphdr_len, __be32 vni,
1848                            struct vxlan_metadata *md, u32 vxflags,
1849                            bool udp_sum)
1850 {
1851         struct vxlanhdr *vxh;
1852         int min_headroom;
1853         int err;
1854         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1855         __be16 inner_protocol = htons(ETH_P_TEB);
1856
1857         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1858             skb->ip_summed == CHECKSUM_PARTIAL) {
1859                 int csum_start = skb_checksum_start_offset(skb);
1860
1861                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1862                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1863                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1864                      skb->csum_offset == offsetof(struct tcphdr, check)))
1865                         type |= SKB_GSO_TUNNEL_REMCSUM;
1866         }
1867
1868         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1869                         + VXLAN_HLEN + iphdr_len;
1870
1871         /* Need space for new headers (invalidates iph ptr) */
1872         err = skb_cow_head(skb, min_headroom);
1873         if (unlikely(err))
1874                 return err;
1875
1876         err = iptunnel_handle_offloads(skb, type);
1877         if (err)
1878                 return err;
1879
1880         vxh = __skb_push(skb, sizeof(*vxh));
1881         vxh->vx_flags = VXLAN_HF_VNI;
1882         vxh->vx_vni = vxlan_vni_field(vni);
1883
1884         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1885                 unsigned int start;
1886
1887                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1888                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1889                 vxh->vx_flags |= VXLAN_HF_RCO;
1890
1891                 if (!skb_is_gso(skb)) {
1892                         skb->ip_summed = CHECKSUM_NONE;
1893                         skb->encapsulation = 0;
1894                 }
1895         }
1896
1897         if (vxflags & VXLAN_F_GBP)
1898                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1899         if (vxflags & VXLAN_F_GPE) {
1900                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1901                 if (err < 0)
1902                         return err;
1903                 inner_protocol = skb->protocol;
1904         }
1905
1906         skb_set_inner_protocol(skb, inner_protocol);
1907         return 0;
1908 }
1909
1910 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1911                                       struct vxlan_sock *sock4,
1912                                       struct sk_buff *skb, int oif, u8 tos,
1913                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
1914                                       struct dst_cache *dst_cache,
1915                                       const struct ip_tunnel_info *info)
1916 {
1917         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1918         struct rtable *rt = NULL;
1919         struct flowi4 fl4;
1920
1921         if (!sock4)
1922                 return ERR_PTR(-EIO);
1923
1924         if (tos && !info)
1925                 use_cache = false;
1926         if (use_cache) {
1927                 rt = dst_cache_get_ip4(dst_cache, saddr);
1928                 if (rt)
1929                         return rt;
1930         }
1931
1932         memset(&fl4, 0, sizeof(fl4));
1933         fl4.flowi4_oif = oif;
1934         fl4.flowi4_tos = RT_TOS(tos);
1935         fl4.flowi4_mark = skb->mark;
1936         fl4.flowi4_proto = IPPROTO_UDP;
1937         fl4.daddr = daddr;
1938         fl4.saddr = *saddr;
1939         fl4.fl4_dport = dport;
1940         fl4.fl4_sport = sport;
1941
1942         rt = ip_route_output_key(vxlan->net, &fl4);
1943         if (likely(!IS_ERR(rt))) {
1944                 if (rt->dst.dev == dev) {
1945                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1946                         ip_rt_put(rt);
1947                         return ERR_PTR(-ELOOP);
1948                 }
1949
1950                 *saddr = fl4.saddr;
1951                 if (use_cache)
1952                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1953         } else {
1954                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1955                 return ERR_PTR(-ENETUNREACH);
1956         }
1957         return rt;
1958 }
1959
1960 #if IS_ENABLED(CONFIG_IPV6)
1961 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1962                                           struct net_device *dev,
1963                                           struct vxlan_sock *sock6,
1964                                           struct sk_buff *skb, int oif, u8 tos,
1965                                           __be32 label,
1966                                           const struct in6_addr *daddr,
1967                                           struct in6_addr *saddr,
1968                                           __be16 dport, __be16 sport,
1969                                           struct dst_cache *dst_cache,
1970                                           const struct ip_tunnel_info *info)
1971 {
1972         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1973         struct dst_entry *ndst;
1974         struct flowi6 fl6;
1975
1976         if (!sock6)
1977                 return ERR_PTR(-EIO);
1978
1979         if (tos && !info)
1980                 use_cache = false;
1981         if (use_cache) {
1982                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1983                 if (ndst)
1984                         return ndst;
1985         }
1986
1987         memset(&fl6, 0, sizeof(fl6));
1988         fl6.flowi6_oif = oif;
1989         fl6.daddr = *daddr;
1990         fl6.saddr = *saddr;
1991         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1992         fl6.flowi6_mark = skb->mark;
1993         fl6.flowi6_proto = IPPROTO_UDP;
1994         fl6.fl6_dport = dport;
1995         fl6.fl6_sport = sport;
1996
1997         ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk,
1998                                                &fl6, NULL);
1999         if (unlikely(IS_ERR(ndst))) {
2000                 netdev_dbg(dev, "no route to %pI6\n", daddr);
2001                 return ERR_PTR(-ENETUNREACH);
2002         }
2003
2004         if (unlikely(ndst->dev == dev)) {
2005                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
2006                 dst_release(ndst);
2007                 return ERR_PTR(-ELOOP);
2008         }
2009
2010         *saddr = fl6.saddr;
2011         if (use_cache)
2012                 dst_cache_set_ip6(dst_cache, ndst, saddr);
2013         return ndst;
2014 }
2015 #endif
2016
2017 /* Bypass encapsulation if the destination is local */
2018 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2019                                struct vxlan_dev *dst_vxlan, __be32 vni)
2020 {
2021         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2022         union vxlan_addr loopback;
2023         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2024         struct net_device *dev;
2025         int len = skb->len;
2026
2027         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2028         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2029         skb->pkt_type = PACKET_HOST;
2030         skb->encapsulation = 0;
2031         skb->dev = dst_vxlan->dev;
2032         __skb_pull(skb, skb_network_offset(skb));
2033
2034         if (remote_ip->sa.sa_family == AF_INET) {
2035                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2036                 loopback.sa.sa_family =  AF_INET;
2037 #if IS_ENABLED(CONFIG_IPV6)
2038         } else {
2039                 loopback.sin6.sin6_addr = in6addr_loopback;
2040                 loopback.sa.sa_family =  AF_INET6;
2041 #endif
2042         }
2043
2044         rcu_read_lock();
2045         dev = skb->dev;
2046         if (unlikely(!(dev->flags & IFF_UP))) {
2047                 kfree_skb(skb);
2048                 goto drop;
2049         }
2050
2051         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
2052                 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni);
2053
2054         u64_stats_update_begin(&tx_stats->syncp);
2055         tx_stats->tx_packets++;
2056         tx_stats->tx_bytes += len;
2057         u64_stats_update_end(&tx_stats->syncp);
2058
2059         if (netif_rx(skb) == NET_RX_SUCCESS) {
2060                 u64_stats_update_begin(&rx_stats->syncp);
2061                 rx_stats->rx_packets++;
2062                 rx_stats->rx_bytes += len;
2063                 u64_stats_update_end(&rx_stats->syncp);
2064         } else {
2065 drop:
2066                 dev->stats.rx_dropped++;
2067         }
2068         rcu_read_unlock();
2069 }
2070
2071 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2072                                  struct vxlan_dev *vxlan,
2073                                  union vxlan_addr *daddr,
2074                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2075                                  struct dst_entry *dst,
2076                                  u32 rt_flags)
2077 {
2078 #if IS_ENABLED(CONFIG_IPV6)
2079         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2080          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2081          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2082          */
2083         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2084 #endif
2085         /* Bypass encapsulation if the destination is local */
2086         if (rt_flags & RTCF_LOCAL &&
2087             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2088                 struct vxlan_dev *dst_vxlan;
2089
2090                 dst_release(dst);
2091                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2092                                            daddr->sa.sa_family, dst_port,
2093                                            vxlan->cfg.flags);
2094                 if (!dst_vxlan) {
2095                         dev->stats.tx_errors++;
2096                         kfree_skb(skb);
2097
2098                         return -ENOENT;
2099                 }
2100                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2101                 return 1;
2102         }
2103
2104         return 0;
2105 }
2106
2107 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2108                            __be32 default_vni, struct vxlan_rdst *rdst,
2109                            bool did_rsc)
2110 {
2111         struct dst_cache *dst_cache;
2112         struct ip_tunnel_info *info;
2113         struct vxlan_dev *vxlan = netdev_priv(dev);
2114         const struct iphdr *old_iph = ip_hdr(skb);
2115         union vxlan_addr *dst;
2116         union vxlan_addr remote_ip, local_ip;
2117         struct vxlan_metadata _md;
2118         struct vxlan_metadata *md = &_md;
2119         __be16 src_port = 0, dst_port;
2120         struct dst_entry *ndst = NULL;
2121         __be32 vni, label;
2122         __u8 tos, ttl;
2123         int ifindex;
2124         int err;
2125         u32 flags = vxlan->cfg.flags;
2126         bool udp_sum = false;
2127         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2128
2129         info = skb_tunnel_info(skb);
2130
2131         if (rdst) {
2132                 dst = &rdst->remote_ip;
2133                 if (vxlan_addr_any(dst)) {
2134                         if (did_rsc) {
2135                                 /* short-circuited back to local bridge */
2136                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2137                                 return;
2138                         }
2139                         goto drop;
2140                 }
2141
2142                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2143                 vni = (rdst->remote_vni) ? : default_vni;
2144                 ifindex = rdst->remote_ifindex;
2145                 local_ip = vxlan->cfg.saddr;
2146                 dst_cache = &rdst->dst_cache;
2147                 md->gbp = skb->mark;
2148                 ttl = vxlan->cfg.ttl;
2149                 if (!ttl && vxlan_addr_multicast(dst))
2150                         ttl = 1;
2151
2152                 tos = vxlan->cfg.tos;
2153                 if (tos == 1)
2154                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2155
2156                 if (dst->sa.sa_family == AF_INET)
2157                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2158                 else
2159                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2160                 label = vxlan->cfg.label;
2161         } else {
2162                 if (!info) {
2163                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2164                                   dev->name);
2165                         goto drop;
2166                 }
2167                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2168                 if (remote_ip.sa.sa_family == AF_INET) {
2169                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2170                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2171                 } else {
2172                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2173                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2174                 }
2175                 dst = &remote_ip;
2176                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2177                 vni = tunnel_id_to_key32(info->key.tun_id);
2178                 ifindex = 0;
2179                 dst_cache = &info->dst_cache;
2180                 if (info->options_len) {
2181                         if (info->options_len < sizeof(*md))
2182                                 goto drop;
2183                         md = ip_tunnel_info_opts(info);
2184                 }
2185                 ttl = info->key.ttl;
2186                 tos = info->key.tos;
2187                 label = info->key.label;
2188                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2189         }
2190         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2191                                      vxlan->cfg.port_max, true);
2192
2193         rcu_read_lock();
2194         if (dst->sa.sa_family == AF_INET) {
2195                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2196                 struct rtable *rt;
2197                 __be16 df = 0;
2198
2199                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2200                                      dst->sin.sin_addr.s_addr,
2201                                      &local_ip.sin.sin_addr.s_addr,
2202                                      dst_port, src_port,
2203                                      dst_cache, info);
2204                 if (IS_ERR(rt)) {
2205                         err = PTR_ERR(rt);
2206                         goto tx_error;
2207                 }
2208
2209                 /* Bypass encapsulation if the destination is local */
2210                 if (!info) {
2211                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2212                                                     dst_port, ifindex, vni,
2213                                                     &rt->dst, rt->rt_flags);
2214                         if (err)
2215                                 goto out_unlock;
2216                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2217                         df = htons(IP_DF);
2218                 }
2219
2220                 ndst = &rt->dst;
2221                 if (skb_dst(skb)) {
2222                         int mtu = dst_mtu(ndst) - VXLAN_HEADROOM;
2223
2224                         skb_dst_update_pmtu(skb, mtu);
2225                 }
2226
2227                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2228                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2229                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2230                                       vni, md, flags, udp_sum);
2231                 if (err < 0)
2232                         goto tx_error;
2233
2234                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2235                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2236                                     src_port, dst_port, xnet, !udp_sum);
2237 #if IS_ENABLED(CONFIG_IPV6)
2238         } else {
2239                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2240
2241                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2242                                         label, &dst->sin6.sin6_addr,
2243                                         &local_ip.sin6.sin6_addr,
2244                                         dst_port, src_port,
2245                                         dst_cache, info);
2246                 if (IS_ERR(ndst)) {
2247                         err = PTR_ERR(ndst);
2248                         ndst = NULL;
2249                         goto tx_error;
2250                 }
2251
2252                 if (!info) {
2253                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2254
2255                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2256                                                     dst_port, ifindex, vni,
2257                                                     ndst, rt6i_flags);
2258                         if (err)
2259                                 goto out_unlock;
2260                 }
2261
2262                 if (skb_dst(skb)) {
2263                         int mtu = dst_mtu(ndst) - VXLAN6_HEADROOM;
2264
2265                         skb_dst_update_pmtu(skb, mtu);
2266                 }
2267
2268                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2269                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2270                 skb_scrub_packet(skb, xnet);
2271                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2272                                       vni, md, flags, udp_sum);
2273                 if (err < 0)
2274                         goto tx_error;
2275
2276                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2277                                      &local_ip.sin6.sin6_addr,
2278                                      &dst->sin6.sin6_addr, tos, ttl,
2279                                      label, src_port, dst_port, !udp_sum);
2280 #endif
2281         }
2282 out_unlock:
2283         rcu_read_unlock();
2284         return;
2285
2286 drop:
2287         dev->stats.tx_dropped++;
2288         dev_kfree_skb(skb);
2289         return;
2290
2291 tx_error:
2292         rcu_read_unlock();
2293         if (err == -ELOOP)
2294                 dev->stats.collisions++;
2295         else if (err == -ENETUNREACH)
2296                 dev->stats.tx_carrier_errors++;
2297         dst_release(ndst);
2298         dev->stats.tx_errors++;
2299         kfree_skb(skb);
2300 }
2301
2302 /* Transmit local packets over Vxlan
2303  *
2304  * Outer IP header inherits ECN and DF from inner header.
2305  * Outer UDP destination is the VXLAN assigned port.
2306  *           source port is based on hash of flow
2307  */
2308 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2309 {
2310         struct vxlan_dev *vxlan = netdev_priv(dev);
2311         struct vxlan_rdst *rdst, *fdst = NULL;
2312         const struct ip_tunnel_info *info;
2313         bool did_rsc = false;
2314         struct vxlan_fdb *f;
2315         struct ethhdr *eth;
2316         __be32 vni = 0;
2317
2318         info = skb_tunnel_info(skb);
2319
2320         skb_reset_mac_header(skb);
2321
2322         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2323                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2324                     info->mode & IP_TUNNEL_INFO_TX) {
2325                         vni = tunnel_id_to_key32(info->key.tun_id);
2326                 } else {
2327                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2328                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2329                         else
2330                                 kfree_skb(skb);
2331                         return NETDEV_TX_OK;
2332                 }
2333         }
2334
2335         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2336                 eth = eth_hdr(skb);
2337                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2338                         return arp_reduce(dev, skb, vni);
2339 #if IS_ENABLED(CONFIG_IPV6)
2340                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2341                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2342                                             sizeof(struct nd_msg)) &&
2343                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2344                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2345
2346                         if (m->icmph.icmp6_code == 0 &&
2347                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2348                                 return neigh_reduce(dev, skb, vni);
2349                 }
2350 #endif
2351         }
2352
2353         eth = eth_hdr(skb);
2354         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2355         did_rsc = false;
2356
2357         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2358             (ntohs(eth->h_proto) == ETH_P_IP ||
2359              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2360                 did_rsc = route_shortcircuit(dev, skb);
2361                 if (did_rsc)
2362                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2363         }
2364
2365         if (f == NULL) {
2366                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2367                 if (f == NULL) {
2368                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2369                             !is_multicast_ether_addr(eth->h_dest))
2370                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2371
2372                         dev->stats.tx_dropped++;
2373                         kfree_skb(skb);
2374                         return NETDEV_TX_OK;
2375                 }
2376         }
2377
2378         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2379                 struct sk_buff *skb1;
2380
2381                 if (!fdst) {
2382                         fdst = rdst;
2383                         continue;
2384                 }
2385                 skb1 = skb_clone(skb, GFP_ATOMIC);
2386                 if (skb1)
2387                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2388         }
2389
2390         if (fdst)
2391                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2392         else
2393                 kfree_skb(skb);
2394         return NETDEV_TX_OK;
2395 }
2396
2397 /* Walk the forwarding table and purge stale entries */
2398 static void vxlan_cleanup(unsigned long arg)
2399 {
2400         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2401         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2402         unsigned int h;
2403
2404         if (!netif_running(vxlan->dev))
2405                 return;
2406
2407         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2408                 struct hlist_node *p, *n;
2409
2410                 spin_lock_bh(&vxlan->hash_lock);
2411                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2412                         struct vxlan_fdb *f
2413                                 = container_of(p, struct vxlan_fdb, hlist);
2414                         unsigned long timeout;
2415
2416                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2417                                 continue;
2418
2419                         if (f->flags & NTF_EXT_LEARNED)
2420                                 continue;
2421
2422                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2423                         if (time_before_eq(timeout, jiffies)) {
2424                                 netdev_dbg(vxlan->dev,
2425                                            "garbage collect %pM\n",
2426                                            f->eth_addr);
2427                                 f->state = NUD_STALE;
2428                                 vxlan_fdb_destroy(vxlan, f, true);
2429                         } else if (time_before(timeout, next_timer))
2430                                 next_timer = timeout;
2431                 }
2432                 spin_unlock_bh(&vxlan->hash_lock);
2433         }
2434
2435         mod_timer(&vxlan->age_timer, next_timer);
2436 }
2437
2438 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2439 {
2440         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2441
2442         spin_lock(&vn->sock_lock);
2443         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2444 #if IS_ENABLED(CONFIG_IPV6)
2445         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2446 #endif
2447         spin_unlock(&vn->sock_lock);
2448 }
2449
2450 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2451                              struct vxlan_dev_node *node)
2452 {
2453         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2454         __be32 vni = vxlan->default_dst.remote_vni;
2455
2456         node->vxlan = vxlan;
2457         spin_lock(&vn->sock_lock);
2458         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2459         spin_unlock(&vn->sock_lock);
2460 }
2461
2462 /* Setup stats when device is created */
2463 static int vxlan_init(struct net_device *dev)
2464 {
2465         struct vxlan_dev *vxlan = netdev_priv(dev);
2466         int err;
2467
2468         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2469         if (!dev->tstats)
2470                 return -ENOMEM;
2471
2472         err = gro_cells_init(&vxlan->gro_cells, dev);
2473         if (err) {
2474                 free_percpu(dev->tstats);
2475                 return err;
2476         }
2477
2478         return 0;
2479 }
2480
2481 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2482 {
2483         struct vxlan_fdb *f;
2484
2485         spin_lock_bh(&vxlan->hash_lock);
2486         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2487         if (f)
2488                 vxlan_fdb_destroy(vxlan, f, true);
2489         spin_unlock_bh(&vxlan->hash_lock);
2490 }
2491
2492 static void vxlan_uninit(struct net_device *dev)
2493 {
2494         struct vxlan_dev *vxlan = netdev_priv(dev);
2495
2496         gro_cells_destroy(&vxlan->gro_cells);
2497
2498         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2499
2500         free_percpu(dev->tstats);
2501 }
2502
2503 /* Start ageing timer and join group when device is brought up */
2504 static int vxlan_open(struct net_device *dev)
2505 {
2506         struct vxlan_dev *vxlan = netdev_priv(dev);
2507         int ret;
2508
2509         ret = vxlan_sock_add(vxlan);
2510         if (ret < 0)
2511                 return ret;
2512
2513         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2514                 ret = vxlan_igmp_join(vxlan);
2515                 if (ret == -EADDRINUSE)
2516                         ret = 0;
2517                 if (ret) {
2518                         vxlan_sock_release(vxlan);
2519                         return ret;
2520                 }
2521         }
2522
2523         if (vxlan->cfg.age_interval)
2524                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2525
2526         return ret;
2527 }
2528
2529 /* Purge the forwarding table */
2530 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2531 {
2532         unsigned int h;
2533
2534         spin_lock_bh(&vxlan->hash_lock);
2535         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2536                 struct hlist_node *p, *n;
2537                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2538                         struct vxlan_fdb *f
2539                                 = container_of(p, struct vxlan_fdb, hlist);
2540                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2541                                 continue;
2542                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2543                         if (!is_zero_ether_addr(f->eth_addr))
2544                                 vxlan_fdb_destroy(vxlan, f, true);
2545                 }
2546         }
2547         spin_unlock_bh(&vxlan->hash_lock);
2548 }
2549
2550 /* Cleanup timer and forwarding table on shutdown */
2551 static int vxlan_stop(struct net_device *dev)
2552 {
2553         struct vxlan_dev *vxlan = netdev_priv(dev);
2554         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2555         int ret = 0;
2556
2557         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2558             !vxlan_group_used(vn, vxlan))
2559                 ret = vxlan_igmp_leave(vxlan);
2560
2561         del_timer_sync(&vxlan->age_timer);
2562
2563         vxlan_flush(vxlan, false);
2564         vxlan_sock_release(vxlan);
2565
2566         return ret;
2567 }
2568
2569 /* Stub, nothing needs to be done. */
2570 static void vxlan_set_multicast_list(struct net_device *dev)
2571 {
2572 }
2573
2574 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2575 {
2576         struct vxlan_dev *vxlan = netdev_priv(dev);
2577         struct vxlan_rdst *dst = &vxlan->default_dst;
2578         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2579                                                          dst->remote_ifindex);
2580         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
2581
2582         /* This check is different than dev->max_mtu, because it looks at
2583          * the lowerdev->mtu, rather than the static dev->max_mtu
2584          */
2585         if (lowerdev) {
2586                 int max_mtu = lowerdev->mtu -
2587                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2588                 if (new_mtu > max_mtu)
2589                         return -EINVAL;
2590         }
2591
2592         dev->mtu = new_mtu;
2593         return 0;
2594 }
2595
2596 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2597 {
2598         struct vxlan_dev *vxlan = netdev_priv(dev);
2599         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2600         __be16 sport, dport;
2601
2602         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2603                                   vxlan->cfg.port_max, true);
2604         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2605
2606         if (ip_tunnel_info_af(info) == AF_INET) {
2607                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2608                 struct rtable *rt;
2609
2610                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2611                                      info->key.u.ipv4.dst,
2612                                      &info->key.u.ipv4.src, dport, sport,
2613                                      &info->dst_cache, info);
2614                 if (IS_ERR(rt))
2615                         return PTR_ERR(rt);
2616                 ip_rt_put(rt);
2617         } else {
2618 #if IS_ENABLED(CONFIG_IPV6)
2619                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2620                 struct dst_entry *ndst;
2621
2622                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2623                                         info->key.label, &info->key.u.ipv6.dst,
2624                                         &info->key.u.ipv6.src, dport, sport,
2625                                         &info->dst_cache, info);
2626                 if (IS_ERR(ndst))
2627                         return PTR_ERR(ndst);
2628                 dst_release(ndst);
2629 #else /* !CONFIG_IPV6 */
2630                 return -EPFNOSUPPORT;
2631 #endif
2632         }
2633         info->key.tp_src = sport;
2634         info->key.tp_dst = dport;
2635         return 0;
2636 }
2637
2638 static const struct net_device_ops vxlan_netdev_ether_ops = {
2639         .ndo_init               = vxlan_init,
2640         .ndo_uninit             = vxlan_uninit,
2641         .ndo_open               = vxlan_open,
2642         .ndo_stop               = vxlan_stop,
2643         .ndo_start_xmit         = vxlan_xmit,
2644         .ndo_get_stats64        = ip_tunnel_get_stats64,
2645         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2646         .ndo_change_mtu         = vxlan_change_mtu,
2647         .ndo_validate_addr      = eth_validate_addr,
2648         .ndo_set_mac_address    = eth_mac_addr,
2649         .ndo_fdb_add            = vxlan_fdb_add,
2650         .ndo_fdb_del            = vxlan_fdb_delete,
2651         .ndo_fdb_dump           = vxlan_fdb_dump,
2652         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2653 };
2654
2655 static const struct net_device_ops vxlan_netdev_raw_ops = {
2656         .ndo_init               = vxlan_init,
2657         .ndo_uninit             = vxlan_uninit,
2658         .ndo_open               = vxlan_open,
2659         .ndo_stop               = vxlan_stop,
2660         .ndo_start_xmit         = vxlan_xmit,
2661         .ndo_get_stats64        = ip_tunnel_get_stats64,
2662         .ndo_change_mtu         = vxlan_change_mtu,
2663         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2664 };
2665
2666 /* Info for udev, that this is a virtual tunnel endpoint */
2667 static struct device_type vxlan_type = {
2668         .name = "vxlan",
2669 };
2670
2671 /* Calls the ndo_udp_tunnel_add of the caller in order to
2672  * supply the listening VXLAN udp ports. Callers are expected
2673  * to implement the ndo_udp_tunnel_add.
2674  */
2675 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
2676 {
2677         struct vxlan_sock *vs;
2678         struct net *net = dev_net(dev);
2679         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2680         unsigned int i;
2681
2682         spin_lock(&vn->sock_lock);
2683         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2684                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2685                         unsigned short type;
2686
2687                         if (vs->flags & VXLAN_F_GPE)
2688                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
2689                         else
2690                                 type = UDP_TUNNEL_TYPE_VXLAN;
2691
2692                         if (push)
2693                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
2694                         else
2695                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
2696                 }
2697         }
2698         spin_unlock(&vn->sock_lock);
2699 }
2700
2701 /* Initialize the device structure. */
2702 static void vxlan_setup(struct net_device *dev)
2703 {
2704         struct vxlan_dev *vxlan = netdev_priv(dev);
2705         unsigned int h;
2706
2707         eth_hw_addr_random(dev);
2708         ether_setup(dev);
2709
2710         dev->needs_free_netdev = true;
2711         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2712
2713         dev->features   |= NETIF_F_LLTX;
2714         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2715         dev->features   |= NETIF_F_RXCSUM;
2716         dev->features   |= NETIF_F_GSO_SOFTWARE;
2717
2718         dev->vlan_features = dev->features;
2719         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2720         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2721         netif_keep_dst(dev);
2722         dev->priv_flags |= IFF_NO_QUEUE;
2723
2724         /* MTU range: 68 - 65535 */
2725         dev->min_mtu = ETH_MIN_MTU;
2726         dev->max_mtu = ETH_MAX_MTU;
2727
2728         INIT_LIST_HEAD(&vxlan->next);
2729         spin_lock_init(&vxlan->hash_lock);
2730
2731         init_timer_deferrable(&vxlan->age_timer);
2732         vxlan->age_timer.function = vxlan_cleanup;
2733         vxlan->age_timer.data = (unsigned long) vxlan;
2734
2735         vxlan->dev = dev;
2736
2737         for (h = 0; h < FDB_HASH_SIZE; ++h)
2738                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2739 }
2740
2741 static void vxlan_ether_setup(struct net_device *dev)
2742 {
2743         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2744         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2745         dev->netdev_ops = &vxlan_netdev_ether_ops;
2746 }
2747
2748 static void vxlan_raw_setup(struct net_device *dev)
2749 {
2750         dev->header_ops = NULL;
2751         dev->type = ARPHRD_NONE;
2752         dev->hard_header_len = 0;
2753         dev->addr_len = 0;
2754         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2755         dev->netdev_ops = &vxlan_netdev_raw_ops;
2756 }
2757
2758 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2759         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2760         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2761         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2762         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2763         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2764         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2765         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2766         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2767         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2768         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2769         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2770         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2771         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2772         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2773         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2774         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2775         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2776         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2777         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2778         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2779         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2780         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2781         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2782         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2783         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2784         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2785         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2786 };
2787
2788 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
2789                           struct netlink_ext_ack *extack)
2790 {
2791         if (tb[IFLA_ADDRESS]) {
2792                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2793                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2794                                             "Provided link layer address is not Ethernet");
2795                         return -EINVAL;
2796                 }
2797
2798                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2799                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2800                                             "Provided Ethernet address is not unicast");
2801                         return -EADDRNOTAVAIL;
2802                 }
2803         }
2804
2805         if (tb[IFLA_MTU]) {
2806                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
2807
2808                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
2809                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
2810                                             "MTU must be between 68 and 65535");
2811                         return -EINVAL;
2812                 }
2813         }
2814
2815         if (!data) {
2816                 NL_SET_ERR_MSG(extack,
2817                                "Required attributes not provided to perform the operation");
2818                 return -EINVAL;
2819         }
2820
2821         if (data[IFLA_VXLAN_ID]) {
2822                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2823
2824                 if (id >= VXLAN_N_VID) {
2825                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID],
2826                                             "VXLAN ID must be lower than 16777216");
2827                         return -ERANGE;
2828                 }
2829         }
2830
2831         if (data[IFLA_VXLAN_PORT_RANGE]) {
2832                 const struct ifla_vxlan_port_range *p
2833                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2834
2835                 if (ntohs(p->high) < ntohs(p->low)) {
2836                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
2837                                             "Invalid source port range");
2838                         return -EINVAL;
2839                 }
2840         }
2841
2842         return 0;
2843 }
2844
2845 static void vxlan_get_drvinfo(struct net_device *netdev,
2846                               struct ethtool_drvinfo *drvinfo)
2847 {
2848         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2849         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2850 }
2851
2852 static const struct ethtool_ops vxlan_ethtool_ops = {
2853         .get_drvinfo    = vxlan_get_drvinfo,
2854         .get_link       = ethtool_op_get_link,
2855 };
2856
2857 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2858                                         __be16 port, u32 flags)
2859 {
2860         struct socket *sock;
2861         struct udp_port_cfg udp_conf;
2862         int err;
2863
2864         memset(&udp_conf, 0, sizeof(udp_conf));
2865
2866         if (ipv6) {
2867                 udp_conf.family = AF_INET6;
2868                 udp_conf.use_udp6_rx_checksums =
2869                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2870                 udp_conf.ipv6_v6only = 1;
2871         } else {
2872                 udp_conf.family = AF_INET;
2873         }
2874
2875         udp_conf.local_udp_port = port;
2876
2877         /* Open UDP socket */
2878         err = udp_sock_create(net, &udp_conf, &sock);
2879         if (err < 0)
2880                 return ERR_PTR(err);
2881
2882         return sock;
2883 }
2884
2885 /* Create new listen socket if needed */
2886 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2887                                               __be16 port, u32 flags)
2888 {
2889         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2890         struct vxlan_sock *vs;
2891         struct socket *sock;
2892         unsigned int h;
2893         struct udp_tunnel_sock_cfg tunnel_cfg;
2894
2895         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2896         if (!vs)
2897                 return ERR_PTR(-ENOMEM);
2898
2899         for (h = 0; h < VNI_HASH_SIZE; ++h)
2900                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2901
2902         sock = vxlan_create_sock(net, ipv6, port, flags);
2903         if (IS_ERR(sock)) {
2904                 kfree(vs);
2905                 return ERR_CAST(sock);
2906         }
2907
2908         vs->sock = sock;
2909         refcount_set(&vs->refcnt, 1);
2910         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2911
2912         spin_lock(&vn->sock_lock);
2913         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2914         udp_tunnel_notify_add_rx_port(sock,
2915                                       (vs->flags & VXLAN_F_GPE) ?
2916                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2917                                       UDP_TUNNEL_TYPE_VXLAN);
2918         spin_unlock(&vn->sock_lock);
2919
2920         /* Mark socket as an encapsulation socket. */
2921         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2922         tunnel_cfg.sk_user_data = vs;
2923         tunnel_cfg.encap_type = 1;
2924         tunnel_cfg.encap_rcv = vxlan_rcv;
2925         tunnel_cfg.encap_destroy = NULL;
2926         tunnel_cfg.gro_receive = vxlan_gro_receive;
2927         tunnel_cfg.gro_complete = vxlan_gro_complete;
2928
2929         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2930
2931         return vs;
2932 }
2933
2934 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2935 {
2936         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2937         struct vxlan_sock *vs = NULL;
2938         struct vxlan_dev_node *node;
2939
2940         if (!vxlan->cfg.no_share) {
2941                 spin_lock(&vn->sock_lock);
2942                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2943                                      vxlan->cfg.dst_port, vxlan->cfg.flags);
2944                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
2945                         spin_unlock(&vn->sock_lock);
2946                         return -EBUSY;
2947                 }
2948                 spin_unlock(&vn->sock_lock);
2949         }
2950         if (!vs)
2951                 vs = vxlan_socket_create(vxlan->net, ipv6,
2952                                          vxlan->cfg.dst_port, vxlan->cfg.flags);
2953         if (IS_ERR(vs))
2954                 return PTR_ERR(vs);
2955 #if IS_ENABLED(CONFIG_IPV6)
2956         if (ipv6) {
2957                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2958                 node = &vxlan->hlist6;
2959         } else
2960 #endif
2961         {
2962                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2963                 node = &vxlan->hlist4;
2964         }
2965         vxlan_vs_add_dev(vs, vxlan, node);
2966         return 0;
2967 }
2968
2969 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2970 {
2971         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
2972         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
2973         bool ipv4 = !ipv6 || metadata;
2974         int ret = 0;
2975
2976         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2977 #if IS_ENABLED(CONFIG_IPV6)
2978         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2979         if (ipv6) {
2980                 ret = __vxlan_sock_add(vxlan, true);
2981                 if (ret < 0 && ret != -EAFNOSUPPORT)
2982                         ipv4 = false;
2983         }
2984 #endif
2985         if (ipv4)
2986                 ret = __vxlan_sock_add(vxlan, false);
2987         if (ret < 0)
2988                 vxlan_sock_release(vxlan);
2989         return ret;
2990 }
2991
2992 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
2993                                  struct net_device **lower,
2994                                  struct vxlan_dev *old,
2995                                  struct netlink_ext_ack *extack)
2996 {
2997         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2998         struct vxlan_dev *tmp;
2999         bool use_ipv6 = false;
3000
3001         if (conf->flags & VXLAN_F_GPE) {
3002                 /* For now, allow GPE only together with
3003                  * COLLECT_METADATA. This can be relaxed later; in such
3004                  * case, the other side of the PtP link will have to be
3005                  * provided.
3006                  */
3007                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
3008                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3009                         NL_SET_ERR_MSG(extack,
3010                                        "VXLAN GPE does not support this combination of attributes");
3011                         return -EINVAL;
3012                 }
3013         }
3014
3015         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
3016                 /* Unless IPv6 is explicitly requested, assume IPv4 */
3017                 conf->remote_ip.sa.sa_family = AF_INET;
3018                 conf->saddr.sa.sa_family = AF_INET;
3019         } else if (!conf->remote_ip.sa.sa_family) {
3020                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
3021         } else if (!conf->saddr.sa.sa_family) {
3022                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
3023         }
3024
3025         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
3026                 NL_SET_ERR_MSG(extack,
3027                                "Local and remote address must be from the same family");
3028                 return -EINVAL;
3029         }
3030
3031         if (vxlan_addr_multicast(&conf->saddr)) {
3032                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
3033                 return -EINVAL;
3034         }
3035
3036         if (conf->saddr.sa.sa_family == AF_INET6) {
3037                 if (!IS_ENABLED(CONFIG_IPV6)) {
3038                         NL_SET_ERR_MSG(extack,
3039                                        "IPv6 support not enabled in the kernel");
3040                         return -EPFNOSUPPORT;
3041                 }
3042                 use_ipv6 = true;
3043                 conf->flags |= VXLAN_F_IPV6;
3044
3045                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3046                         int local_type =
3047                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3048                         int remote_type =
3049                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3050
3051                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3052                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3053                                     (remote_type != IPV6_ADDR_ANY)) {
3054                                         NL_SET_ERR_MSG(extack,
3055                                                        "Invalid combination of local and remote address scopes");
3056                                         return -EINVAL;
3057                                 }
3058
3059                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3060                         } else {
3061                                 if (remote_type ==
3062                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3063                                         NL_SET_ERR_MSG(extack,
3064                                                        "Invalid combination of local and remote address scopes");
3065                                         return -EINVAL;
3066                                 }
3067
3068                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3069                         }
3070                 }
3071         }
3072
3073         if (conf->label && !use_ipv6) {
3074                 NL_SET_ERR_MSG(extack,
3075                                "Label attribute only applies to IPv6 VXLAN devices");
3076                 return -EINVAL;
3077         }
3078
3079         if (conf->remote_ifindex) {
3080                 struct net_device *lowerdev;
3081
3082                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3083                 if (!lowerdev) {
3084                         NL_SET_ERR_MSG(extack,
3085                                        "Invalid local interface, device not found");
3086                         return -ENODEV;
3087                 }
3088
3089 #if IS_ENABLED(CONFIG_IPV6)
3090                 if (use_ipv6) {
3091                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3092                         if (idev && idev->cnf.disable_ipv6) {
3093                                 NL_SET_ERR_MSG(extack,
3094                                                "IPv6 support disabled by administrator");
3095                                 return -EPERM;
3096                         }
3097                 }
3098 #endif
3099
3100                 *lower = lowerdev;
3101         } else {
3102                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3103                         NL_SET_ERR_MSG(extack,
3104                                        "Local interface required for multicast remote destination");
3105
3106                         return -EINVAL;
3107                 }
3108
3109 #if IS_ENABLED(CONFIG_IPV6)
3110                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3111                         NL_SET_ERR_MSG(extack,
3112                                        "Local interface required for link-local local/remote addresses");
3113                         return -EINVAL;
3114                 }
3115 #endif
3116
3117                 *lower = NULL;
3118         }
3119
3120         if (!conf->dst_port) {
3121                 if (conf->flags & VXLAN_F_GPE)
3122                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3123                 else
3124                         conf->dst_port = htons(vxlan_port);
3125         }
3126
3127         if (!conf->age_interval)
3128                 conf->age_interval = FDB_AGE_DEFAULT;
3129
3130         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3131                 if (tmp == old)
3132                         continue;
3133
3134                 if (tmp->cfg.vni != conf->vni)
3135                         continue;
3136                 if (tmp->cfg.dst_port != conf->dst_port)
3137                         continue;
3138                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3139                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3140                         continue;
3141
3142                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3143                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3144                         continue;
3145
3146                 NL_SET_ERR_MSG(extack,
3147                                "A VXLAN device with the specified VNI already exists");
3148                 return -EEXIST;
3149         }
3150
3151         return 0;
3152 }
3153
3154 static void vxlan_config_apply(struct net_device *dev,
3155                                struct vxlan_config *conf,
3156                                struct net_device *lowerdev,
3157                                struct net *src_net,
3158                                bool changelink)
3159 {
3160         struct vxlan_dev *vxlan = netdev_priv(dev);
3161         struct vxlan_rdst *dst = &vxlan->default_dst;
3162         unsigned short needed_headroom = ETH_HLEN;
3163         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3164         int max_mtu = ETH_MAX_MTU;
3165
3166         if (!changelink) {
3167                 if (conf->flags & VXLAN_F_GPE)
3168                         vxlan_raw_setup(dev);
3169                 else
3170                         vxlan_ether_setup(dev);
3171
3172                 if (conf->mtu)
3173                         dev->mtu = conf->mtu;
3174
3175                 vxlan->net = src_net;
3176         }
3177
3178         dst->remote_vni = conf->vni;
3179
3180         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3181
3182         if (lowerdev) {
3183                 dst->remote_ifindex = conf->remote_ifindex;
3184
3185                 dev->gso_max_size = lowerdev->gso_max_size;
3186                 dev->gso_max_segs = lowerdev->gso_max_segs;
3187
3188                 needed_headroom = lowerdev->hard_header_len;
3189                 needed_headroom += lowerdev->needed_headroom;
3190
3191                 dev->needed_tailroom = lowerdev->needed_tailroom;
3192
3193                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3194                                            VXLAN_HEADROOM);
3195                 if (max_mtu < ETH_MIN_MTU)
3196                         max_mtu = ETH_MIN_MTU;
3197
3198                 if (!changelink && !conf->mtu)
3199                         dev->mtu = max_mtu;
3200         }
3201
3202         if (dev->mtu > max_mtu)
3203                 dev->mtu = max_mtu;
3204
3205         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3206                 needed_headroom += VXLAN6_HEADROOM;
3207         else
3208                 needed_headroom += VXLAN_HEADROOM;
3209         dev->needed_headroom = needed_headroom;
3210
3211         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3212 }
3213
3214 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3215                                struct vxlan_config *conf, bool changelink,
3216                                struct netlink_ext_ack *extack)
3217 {
3218         struct vxlan_dev *vxlan = netdev_priv(dev);
3219         struct net_device *lowerdev;
3220         int ret;
3221
3222         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3223         if (ret)
3224                 return ret;
3225
3226         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3227
3228         return 0;
3229 }
3230
3231 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3232                               struct vxlan_config *conf,
3233                               struct netlink_ext_ack *extack)
3234 {
3235         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3236         struct vxlan_dev *vxlan = netdev_priv(dev);
3237         struct vxlan_fdb *f = NULL;
3238         bool unregister = false;
3239         int err;
3240
3241         err = vxlan_dev_configure(net, dev, conf, false, extack);
3242         if (err)
3243                 return err;
3244
3245         dev->ethtool_ops = &vxlan_ethtool_ops;
3246
3247         /* create an fdb entry for a valid default destination */
3248         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
3249                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3250                                        &vxlan->default_dst.remote_ip,
3251                                        NUD_REACHABLE | NUD_PERMANENT,
3252                                        vxlan->cfg.dst_port,
3253                                        vxlan->default_dst.remote_vni,
3254                                        vxlan->default_dst.remote_vni,
3255                                        vxlan->default_dst.remote_ifindex,
3256                                        NTF_SELF, &f);
3257                 if (err)
3258                         return err;
3259         }
3260
3261         err = register_netdevice(dev);
3262         if (err)
3263                 goto errout;
3264         unregister = true;
3265
3266         err = rtnl_configure_link(dev, NULL);
3267         if (err)
3268                 goto errout;
3269
3270         /* notify default fdb entry */
3271         if (f)
3272                 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH);
3273
3274         list_add(&vxlan->next, &vn->vxlan_list);
3275         return 0;
3276
3277 errout:
3278         /* unregister_netdevice() destroys the default FDB entry with deletion
3279          * notification. But the addition notification was not sent yet, so
3280          * destroy the entry by hand here.
3281          */
3282         if (f)
3283                 vxlan_fdb_destroy(vxlan, f, false);
3284         if (unregister)
3285                 unregister_netdevice(dev);
3286         return err;
3287 }
3288
3289 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3290                          struct net_device *dev, struct vxlan_config *conf,
3291                          bool changelink)
3292 {
3293         struct vxlan_dev *vxlan = netdev_priv(dev);
3294
3295         memset(conf, 0, sizeof(*conf));
3296
3297         /* if changelink operation, start with old existing cfg */
3298         if (changelink)
3299                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3300
3301         if (data[IFLA_VXLAN_ID]) {
3302                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3303
3304                 if (changelink && (vni != conf->vni))
3305                         return -EOPNOTSUPP;
3306                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3307         }
3308
3309         if (data[IFLA_VXLAN_GROUP]) {
3310                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET))
3311                         return -EOPNOTSUPP;
3312
3313                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3314                 conf->remote_ip.sa.sa_family = AF_INET;
3315         } else if (data[IFLA_VXLAN_GROUP6]) {
3316                 if (!IS_ENABLED(CONFIG_IPV6))
3317                         return -EPFNOSUPPORT;
3318
3319                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6))
3320                         return -EOPNOTSUPP;
3321
3322                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3323                 conf->remote_ip.sa.sa_family = AF_INET6;
3324         }
3325
3326         if (data[IFLA_VXLAN_LOCAL]) {
3327                 if (changelink && (conf->saddr.sa.sa_family != AF_INET))
3328                         return -EOPNOTSUPP;
3329
3330                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3331                 conf->saddr.sa.sa_family = AF_INET;
3332         } else if (data[IFLA_VXLAN_LOCAL6]) {
3333                 if (!IS_ENABLED(CONFIG_IPV6))
3334                         return -EPFNOSUPPORT;
3335
3336                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6))
3337                         return -EOPNOTSUPP;
3338
3339                 /* TODO: respect scope id */
3340                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3341                 conf->saddr.sa.sa_family = AF_INET6;
3342         }
3343
3344         if (data[IFLA_VXLAN_LINK])
3345                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3346
3347         if (data[IFLA_VXLAN_TOS])
3348                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3349
3350         if (data[IFLA_VXLAN_TTL])
3351                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3352
3353         if (data[IFLA_VXLAN_LABEL])
3354                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3355                              IPV6_FLOWLABEL_MASK;
3356
3357         if (data[IFLA_VXLAN_LEARNING]) {
3358                 if (nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3359                         conf->flags |= VXLAN_F_LEARN;
3360                 else
3361                         conf->flags &= ~VXLAN_F_LEARN;
3362         } else if (!changelink) {
3363                 /* default to learn on a new device */
3364                 conf->flags |= VXLAN_F_LEARN;
3365         }
3366
3367         if (data[IFLA_VXLAN_AGEING]) {
3368                 if (changelink)
3369                         return -EOPNOTSUPP;
3370                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3371         }
3372
3373         if (data[IFLA_VXLAN_PROXY]) {
3374                 if (changelink)
3375                         return -EOPNOTSUPP;
3376                 if (nla_get_u8(data[IFLA_VXLAN_PROXY]))
3377                         conf->flags |= VXLAN_F_PROXY;
3378         }
3379
3380         if (data[IFLA_VXLAN_RSC]) {
3381                 if (changelink)
3382                         return -EOPNOTSUPP;
3383                 if (nla_get_u8(data[IFLA_VXLAN_RSC]))
3384                         conf->flags |= VXLAN_F_RSC;
3385         }
3386
3387         if (data[IFLA_VXLAN_L2MISS]) {
3388                 if (changelink)
3389                         return -EOPNOTSUPP;
3390                 if (nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3391                         conf->flags |= VXLAN_F_L2MISS;
3392         }
3393
3394         if (data[IFLA_VXLAN_L3MISS]) {
3395                 if (changelink)
3396                         return -EOPNOTSUPP;
3397                 if (nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3398                         conf->flags |= VXLAN_F_L3MISS;
3399         }
3400
3401         if (data[IFLA_VXLAN_LIMIT]) {
3402                 if (changelink)
3403                         return -EOPNOTSUPP;
3404                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3405         }
3406
3407         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3408                 if (changelink)
3409                         return -EOPNOTSUPP;
3410                 if (nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3411                         conf->flags |= VXLAN_F_COLLECT_METADATA;
3412         }
3413
3414         if (data[IFLA_VXLAN_PORT_RANGE]) {
3415                 if (!changelink) {
3416                         const struct ifla_vxlan_port_range *p
3417                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3418                         conf->port_min = ntohs(p->low);
3419                         conf->port_max = ntohs(p->high);
3420                 } else {
3421                         return -EOPNOTSUPP;
3422                 }
3423         }
3424
3425         if (data[IFLA_VXLAN_PORT]) {
3426                 if (changelink)
3427                         return -EOPNOTSUPP;
3428                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3429         }
3430
3431         if (data[IFLA_VXLAN_UDP_CSUM]) {
3432                 if (changelink)
3433                         return -EOPNOTSUPP;
3434                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3435                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3436         }
3437
3438         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3439                 if (changelink)
3440                         return -EOPNOTSUPP;
3441                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3442                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3443         }
3444
3445         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3446                 if (changelink)
3447                         return -EOPNOTSUPP;
3448                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3449                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3450         }
3451
3452         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3453                 if (changelink)
3454                         return -EOPNOTSUPP;
3455                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3456                         conf->flags |= VXLAN_F_REMCSUM_TX;
3457         }
3458
3459         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3460                 if (changelink)
3461                         return -EOPNOTSUPP;
3462                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3463                         conf->flags |= VXLAN_F_REMCSUM_RX;
3464         }
3465
3466         if (data[IFLA_VXLAN_GBP]) {
3467                 if (changelink)
3468                         return -EOPNOTSUPP;
3469                 conf->flags |= VXLAN_F_GBP;
3470         }
3471
3472         if (data[IFLA_VXLAN_GPE]) {
3473                 if (changelink)
3474                         return -EOPNOTSUPP;
3475                 conf->flags |= VXLAN_F_GPE;
3476         }
3477
3478         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3479                 if (changelink)
3480                         return -EOPNOTSUPP;
3481                 conf->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3482         }
3483
3484         if (tb[IFLA_MTU]) {
3485                 if (changelink)
3486                         return -EOPNOTSUPP;
3487                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3488         }
3489
3490         return 0;
3491 }
3492
3493 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3494                          struct nlattr *tb[], struct nlattr *data[],
3495                          struct netlink_ext_ack *extack)
3496 {
3497         struct vxlan_config conf;
3498         int err;
3499
3500         err = vxlan_nl2conf(tb, data, dev, &conf, false);
3501         if (err)
3502                 return err;
3503
3504         return __vxlan_dev_create(src_net, dev, &conf, extack);
3505 }
3506
3507 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3508                             struct nlattr *data[],
3509                             struct netlink_ext_ack *extack)
3510 {
3511         struct vxlan_dev *vxlan = netdev_priv(dev);
3512         struct vxlan_rdst *dst = &vxlan->default_dst;
3513         struct vxlan_rdst old_dst;
3514         struct vxlan_config conf;
3515         int err;
3516
3517         err = vxlan_nl2conf(tb, data,
3518                             dev, &conf, true);
3519         if (err)
3520                 return err;
3521
3522         memcpy(&old_dst, dst, sizeof(struct vxlan_rdst));
3523
3524         err = vxlan_dev_configure(vxlan->net, dev, &conf, true, extack);
3525         if (err)
3526                 return err;
3527
3528         /* handle default dst entry */
3529         if (!vxlan_addr_equal(&dst->remote_ip, &old_dst.remote_ip)) {
3530                 spin_lock_bh(&vxlan->hash_lock);
3531                 if (!vxlan_addr_any(&old_dst.remote_ip))
3532                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
3533                                            old_dst.remote_ip,
3534                                            vxlan->cfg.dst_port,
3535                                            old_dst.remote_vni,
3536                                            old_dst.remote_vni,
3537                                            old_dst.remote_ifindex, 0);
3538
3539                 if (!vxlan_addr_any(&dst->remote_ip)) {
3540                         err = vxlan_fdb_update(vxlan, all_zeros_mac,
3541                                                &dst->remote_ip,
3542                                                NUD_REACHABLE | NUD_PERMANENT,
3543                                                NLM_F_APPEND | NLM_F_CREATE,
3544                                                vxlan->cfg.dst_port,
3545                                                dst->remote_vni,
3546                                                dst->remote_vni,
3547                                                dst->remote_ifindex,
3548                                                NTF_SELF);
3549                         if (err) {
3550                                 spin_unlock_bh(&vxlan->hash_lock);
3551                                 return err;
3552                         }
3553                 }
3554                 spin_unlock_bh(&vxlan->hash_lock);
3555         }
3556
3557         return 0;
3558 }
3559
3560 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3561 {
3562         struct vxlan_dev *vxlan = netdev_priv(dev);
3563
3564         vxlan_flush(vxlan, true);
3565
3566         list_del(&vxlan->next);
3567         unregister_netdevice_queue(dev, head);
3568 }
3569
3570 static size_t vxlan_get_size(const struct net_device *dev)
3571 {
3572
3573         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3574                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3575                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3576                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3577                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3578                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3579                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3580                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3581                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3582                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3583                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3584                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3585                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3586                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3587                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3588                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3589                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3590                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3591                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3592                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3593                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3594                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3595                 0;
3596 }
3597
3598 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3599 {
3600         const struct vxlan_dev *vxlan = netdev_priv(dev);
3601         const struct vxlan_rdst *dst = &vxlan->default_dst;
3602         struct ifla_vxlan_port_range ports = {
3603                 .low =  htons(vxlan->cfg.port_min),
3604                 .high = htons(vxlan->cfg.port_max),
3605         };
3606
3607         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3608                 goto nla_put_failure;
3609
3610         if (!vxlan_addr_any(&dst->remote_ip)) {
3611                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3612                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3613                                             dst->remote_ip.sin.sin_addr.s_addr))
3614                                 goto nla_put_failure;
3615 #if IS_ENABLED(CONFIG_IPV6)
3616                 } else {
3617                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3618                                              &dst->remote_ip.sin6.sin6_addr))
3619                                 goto nla_put_failure;
3620 #endif
3621                 }
3622         }
3623
3624         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3625                 goto nla_put_failure;
3626
3627         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3628                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3629                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3630                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3631                                 goto nla_put_failure;
3632 #if IS_ENABLED(CONFIG_IPV6)
3633                 } else {
3634                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3635                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3636                                 goto nla_put_failure;
3637 #endif
3638                 }
3639         }
3640
3641         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3642             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3643             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3644             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3645                         !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
3646             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3647                         !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
3648             nla_put_u8(skb, IFLA_VXLAN_RSC,
3649                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
3650             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3651                         !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
3652             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3653                         !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
3654             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3655                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
3656             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3657             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3658             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3659             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3660                         !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3661             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3662                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3663             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3664                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3665             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3666                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
3667             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3668                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
3669                 goto nla_put_failure;
3670
3671         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3672                 goto nla_put_failure;
3673
3674         if (vxlan->cfg.flags & VXLAN_F_GBP &&
3675             nla_put_flag(skb, IFLA_VXLAN_GBP))
3676                 goto nla_put_failure;
3677
3678         if (vxlan->cfg.flags & VXLAN_F_GPE &&
3679             nla_put_flag(skb, IFLA_VXLAN_GPE))
3680                 goto nla_put_failure;
3681
3682         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3683             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3684                 goto nla_put_failure;
3685
3686         return 0;
3687
3688 nla_put_failure:
3689         return -EMSGSIZE;
3690 }
3691
3692 static struct net *vxlan_get_link_net(const struct net_device *dev)
3693 {
3694         struct vxlan_dev *vxlan = netdev_priv(dev);
3695
3696         return vxlan->net;
3697 }
3698
3699 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3700         .kind           = "vxlan",
3701         .maxtype        = IFLA_VXLAN_MAX,
3702         .policy         = vxlan_policy,
3703         .priv_size      = sizeof(struct vxlan_dev),
3704         .setup          = vxlan_setup,
3705         .validate       = vxlan_validate,
3706         .newlink        = vxlan_newlink,
3707         .changelink     = vxlan_changelink,
3708         .dellink        = vxlan_dellink,
3709         .get_size       = vxlan_get_size,
3710         .fill_info      = vxlan_fill_info,
3711         .get_link_net   = vxlan_get_link_net,
3712 };
3713
3714 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3715                                     u8 name_assign_type,
3716                                     struct vxlan_config *conf)
3717 {
3718         struct nlattr *tb[IFLA_MAX + 1];
3719         struct net_device *dev;
3720         int err;
3721
3722         memset(&tb, 0, sizeof(tb));
3723
3724         dev = rtnl_create_link(net, name, name_assign_type,
3725                                &vxlan_link_ops, tb);
3726         if (IS_ERR(dev))
3727                 return dev;
3728
3729         err = __vxlan_dev_create(net, dev, conf, NULL);
3730         if (err < 0) {
3731                 free_netdev(dev);
3732                 return ERR_PTR(err);
3733         }
3734
3735         err = rtnl_configure_link(dev, NULL);
3736         if (err < 0) {
3737                 LIST_HEAD(list_kill);
3738
3739                 vxlan_dellink(dev, &list_kill);
3740                 unregister_netdevice_many(&list_kill);
3741                 return ERR_PTR(err);
3742         }
3743
3744         return dev;
3745 }
3746 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3747
3748 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3749                                              struct net_device *dev)
3750 {
3751         struct vxlan_dev *vxlan, *next;
3752         LIST_HEAD(list_kill);
3753
3754         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3755                 struct vxlan_rdst *dst = &vxlan->default_dst;
3756
3757                 /* In case we created vxlan device with carrier
3758                  * and we loose the carrier due to module unload
3759                  * we also need to remove vxlan device. In other
3760                  * cases, it's not necessary and remote_ifindex
3761                  * is 0 here, so no matches.
3762                  */
3763                 if (dst->remote_ifindex == dev->ifindex)
3764                         vxlan_dellink(vxlan->dev, &list_kill);
3765         }
3766
3767         unregister_netdevice_many(&list_kill);
3768 }
3769
3770 static int vxlan_netdevice_event(struct notifier_block *unused,
3771                                  unsigned long event, void *ptr)
3772 {
3773         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3774         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3775
3776         if (event == NETDEV_UNREGISTER) {
3777                 vxlan_offload_rx_ports(dev, false);
3778                 vxlan_handle_lowerdev_unregister(vn, dev);
3779         } else if (event == NETDEV_REGISTER) {
3780                 vxlan_offload_rx_ports(dev, true);
3781         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
3782                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
3783                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
3784         }
3785
3786         return NOTIFY_DONE;
3787 }
3788
3789 static struct notifier_block vxlan_notifier_block __read_mostly = {
3790         .notifier_call = vxlan_netdevice_event,
3791 };
3792
3793 static __net_init int vxlan_init_net(struct net *net)
3794 {
3795         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3796         unsigned int h;
3797
3798         INIT_LIST_HEAD(&vn->vxlan_list);
3799         spin_lock_init(&vn->sock_lock);
3800
3801         for (h = 0; h < PORT_HASH_SIZE; ++h)
3802                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3803
3804         return 0;
3805 }
3806
3807 static void __net_exit vxlan_exit_net(struct net *net)
3808 {
3809         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3810         struct vxlan_dev *vxlan, *next;
3811         struct net_device *dev, *aux;
3812         LIST_HEAD(list);
3813
3814         rtnl_lock();
3815         for_each_netdev_safe(net, dev, aux)
3816                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3817                         unregister_netdevice_queue(dev, &list);
3818
3819         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3820                 /* If vxlan->dev is in the same netns, it has already been added
3821                  * to the list by the previous loop.
3822                  */
3823                 if (!net_eq(dev_net(vxlan->dev), net))
3824                         unregister_netdevice_queue(vxlan->dev, &list);
3825         }
3826
3827         unregister_netdevice_many(&list);
3828         rtnl_unlock();
3829 }
3830
3831 static struct pernet_operations vxlan_net_ops = {
3832         .init = vxlan_init_net,
3833         .exit = vxlan_exit_net,
3834         .id   = &vxlan_net_id,
3835         .size = sizeof(struct vxlan_net),
3836 };
3837
3838 static int __init vxlan_init_module(void)
3839 {
3840         int rc;
3841
3842         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3843
3844         rc = register_pernet_subsys(&vxlan_net_ops);
3845         if (rc)
3846                 goto out1;
3847
3848         rc = register_netdevice_notifier(&vxlan_notifier_block);
3849         if (rc)
3850                 goto out2;
3851
3852         rc = rtnl_link_register(&vxlan_link_ops);
3853         if (rc)
3854                 goto out3;
3855
3856         return 0;
3857 out3:
3858         unregister_netdevice_notifier(&vxlan_notifier_block);
3859 out2:
3860         unregister_pernet_subsys(&vxlan_net_ops);
3861 out1:
3862         return rc;
3863 }
3864 late_initcall(vxlan_init_module);
3865
3866 static void __exit vxlan_cleanup_module(void)
3867 {
3868         rtnl_link_unregister(&vxlan_link_ops);
3869         unregister_netdevice_notifier(&vxlan_notifier_block);
3870         unregister_pernet_subsys(&vxlan_net_ops);
3871         /* rcu_barrier() is called by netns */
3872 }
3873 module_exit(vxlan_cleanup_module);
3874
3875 MODULE_LICENSE("GPL");
3876 MODULE_VERSION(VXLAN_VERSION);
3877 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3878 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3879 MODULE_ALIAS_RTNL_LINK("vxlan");