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 list_head *head,
 573                                          struct sk_buff *skb)
 574 {
 575         struct sk_buff *pp = NULL;
 576         struct sk_buff *p;
 577         struct vxlanhdr *vh, *vh2;
 578         unsigned int hlen, off_vx;
 579         int flush = 1;
 580         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 581         __be32 flags;
 582         struct gro_remcsum grc;
 583
 584         skb_gro_remcsum_init(&grc);
 585
 586         off_vx = skb_gro_offset(skb);
 587         hlen = off_vx + sizeof(*vh);
 588         vh   = skb_gro_header_fast(skb, off_vx);
 589         if (skb_gro_header_hard(skb, hlen)) {
 590                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 591                 if (unlikely(!vh))
 592                         goto out;
 593         }
 594
 595         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 596
 597         flags = vh->vx_flags;
 598
 599         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 600                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 601                                        vh->vx_vni, &grc,
 602                                        !!(vs->flags &
 603                                           VXLAN_F_REMCSUM_NOPARTIAL));
 604
 605                 if (!vh)
 606                         goto out;
 607         }
 608
 609         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 610
 611         list_for_each_entry(p, head, list) {
 612                 if (!NAPI_GRO_CB(p)->same_flow)
 613                         continue;
 614
 615                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 616                 if (vh->vx_flags != vh2->vx_flags ||
 617                     vh->vx_vni != vh2->vx_vni) {
 618                         NAPI_GRO_CB(p)->same_flow = 0;
 619                         continue;
 620                 }
 621         }
 622
 623         pp = call_gro_receive(eth_gro_receive, head, skb);
 624         flush = 0;
 625
 626 out:
 627         skb_gro_flush_final_remcsum(skb, pp, flush, &grc);
 628
 629         return pp;
 630 }
 631
 632 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 633 {
 634         /* Sets 'skb->inner_mac_header' since we are always called with
 635          * 'skb->encapsulation' set.
 636          */
 637         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 638 }
 639
 640 static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan,
 641                                          const u8 *mac, __u16 state,
 642                                          __be32 src_vni, __u8 ndm_flags)
 643 {
 644         struct vxlan_fdb *f;
 645
 646         f = kmalloc(sizeof(*f), GFP_ATOMIC);
 647         if (!f)
 648                 return NULL;
 649         f->state = state;
 650         f->flags = ndm_flags;
 651         f->updated = f->used = jiffies;
 652         f->vni = src_vni;
 653         INIT_LIST_HEAD(&f->remotes);
 654         memcpy(f->eth_addr, mac, ETH_ALEN);
 655
 656         return f;
 657 }
 658
 659 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 660                             const u8 *mac, union vxlan_addr *ip,
 661                             __u16 state, __be16 port, __be32 src_vni,
 662                             __be32 vni, __u32 ifindex, __u8 ndm_flags,
 663                             struct vxlan_fdb **fdb)
 664 {
 665         struct vxlan_rdst *rd = NULL;
 666         struct vxlan_fdb *f;
 667         int rc;
 668
 669         if (vxlan->cfg.addrmax &&
 670             vxlan->addrcnt >= vxlan->cfg.addrmax)
 671                 return -ENOSPC;
 672
 673         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 674         f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags);
 675         if (!f)
 676                 return -ENOMEM;
 677
 678         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 679         if (rc < 0) {
 680                 kfree(f);
 681                 return rc;
 682         }
 683
 684         ++vxlan->addrcnt;
 685         hlist_add_head_rcu(&f->hlist,
 686                            vxlan_fdb_head(vxlan, mac, src_vni));
 687
 688         *fdb = f;
 689
 690         return 0;
 691 }
 692
 693 /* Add new entry to forwarding table -- assumes lock held */
 694 static int vxlan_fdb_update(struct vxlan_dev *vxlan,
 695                             const u8 *mac, union vxlan_addr *ip,
 696                             __u16 state, __u16 flags,
 697                             __be16 port, __be32 src_vni, __be32 vni,
 698                             __u32 ifindex, __u8 ndm_flags)
 699 {
 700         struct vxlan_rdst *rd = NULL;
 701         struct vxlan_fdb *f;
 702         int notify = 0;
 703         int rc;
 704
 705         f = __vxlan_find_mac(vxlan, mac, src_vni);
 706         if (f) {
 707                 if (flags & NLM_F_EXCL) {
 708                         netdev_dbg(vxlan->dev,
 709                                    "lost race to create %pM\n", mac);
 710                         return -EEXIST;
 711                 }
 712                 if (f->state != state) {
 713                         f->state = state;
 714                         f->updated = jiffies;
 715                         notify = 1;
 716                 }
 717                 if (f->flags != ndm_flags) {
 718                         f->flags = ndm_flags;
 719                         f->updated = jiffies;
 720                         notify = 1;
 721                 }
 722                 if ((flags & NLM_F_REPLACE)) {
 723                         /* Only change unicasts */
 724                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 725                              is_zero_ether_addr(f->eth_addr))) {
 726                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 727                                                            ifindex);
 728                         } else
 729                                 return -EOPNOTSUPP;
 730                 }
 731                 if ((flags & NLM_F_APPEND) &&
 732                     (is_multicast_ether_addr(f->eth_addr) ||
 733                      is_zero_ether_addr(f->eth_addr))) {
 734                         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 735
 736                         if (rc < 0)
 737                                 return rc;
 738                         notify |= rc;
 739                 }
 740         } else {
 741                 if (!(flags & NLM_F_CREATE))
 742                         return -ENOENT;
 743
 744                 /* Disallow replace to add a multicast entry */
 745                 if ((flags & NLM_F_REPLACE) &&
 746                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 747                         return -EOPNOTSUPP;
 748
 749                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 750                 rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni,
 751                                       vni, ifindex, ndm_flags, &f);
 752                 if (rc < 0)
 753                         return rc;
 754                 notify = 1;
 755         }
 756
 757         if (notify) {
 758                 if (rd == NULL)
 759                         rd = first_remote_rtnl(f);
 760                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 761         }
 762
 763         return 0;
 764 }
 765
 766 static void vxlan_fdb_free(struct rcu_head *head)
 767 {
 768         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 769         struct vxlan_rdst *rd, *nd;
 770
 771         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 772                 dst_cache_destroy(&rd->dst_cache);
 773                 kfree(rd);
 774         }
 775         kfree(f);
 776 }
 777
 778 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 779                               bool do_notify)
 780 {
 781         netdev_dbg(vxlan->dev,
 782                     "delete %pM\n", f->eth_addr);
 783
 784         --vxlan->addrcnt;
 785         if (do_notify)
 786                 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 787
 788         hlist_del_rcu(&f->hlist);
 789         call_rcu(&f->rcu, vxlan_fdb_free);
 790 }
 791
 792 static void vxlan_dst_free(struct rcu_head *head)
 793 {
 794         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 795
 796         dst_cache_destroy(&rd->dst_cache);
 797         kfree(rd);
 798 }
 799
 800 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 801                                   struct vxlan_rdst *rd)
 802 {
 803         list_del_rcu(&rd->list);
 804         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 805         call_rcu(&rd->rcu, vxlan_dst_free);
 806 }
 807
 808 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 809                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
 810                            __be32 *vni, u32 *ifindex)
 811 {
 812         struct net *net = dev_net(vxlan->dev);
 813         int err;
 814
 815         if (tb[NDA_DST]) {
 816                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 817                 if (err)
 818                         return err;
 819         } else {
 820                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 821                 if (remote->sa.sa_family == AF_INET) {
 822                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 823                         ip->sa.sa_family = AF_INET;
 824 #if IS_ENABLED(CONFIG_IPV6)
 825                 } else {
 826                         ip->sin6.sin6_addr = in6addr_any;
 827                         ip->sa.sa_family = AF_INET6;
 828 #endif
 829                 }
 830         }
 831
 832         if (tb[NDA_PORT]) {
 833                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 834                         return -EINVAL;
 835                 *port = nla_get_be16(tb[NDA_PORT]);
 836         } else {
 837                 *port = vxlan->cfg.dst_port;
 838         }
 839
 840         if (tb[NDA_VNI]) {
 841                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 842                         return -EINVAL;
 843                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 844         } else {
 845                 *vni = vxlan->default_dst.remote_vni;
 846         }
 847
 848         if (tb[NDA_SRC_VNI]) {
 849                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
 850                         return -EINVAL;
 851                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
 852         } else {
 853                 *src_vni = vxlan->default_dst.remote_vni;
 854         }
 855
 856         if (tb[NDA_IFINDEX]) {
 857                 struct net_device *tdev;
 858
 859                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 860                         return -EINVAL;
 861                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 862                 tdev = __dev_get_by_index(net, *ifindex);
 863                 if (!tdev)
 864                         return -EADDRNOTAVAIL;
 865         } else {
 866                 *ifindex = 0;
 867         }
 868
 869         return 0;
 870 }
 871
 872 /* Add static entry (via netlink) */
 873 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 874                          struct net_device *dev,
 875                          const unsigned char *addr, u16 vid, u16 flags)
 876 {
 877         struct vxlan_dev *vxlan = netdev_priv(dev);
 878         /* struct net *net = dev_net(vxlan->dev); */
 879         union vxlan_addr ip;
 880         __be16 port;
 881         __be32 src_vni, vni;
 882         u32 ifindex;
 883         int err;
 884
 885         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 886                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 887                         ndm->ndm_state);
 888                 return -EINVAL;
 889         }
 890
 891         if (tb[NDA_DST] == NULL)
 892                 return -EINVAL;
 893
 894         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 895         if (err)
 896                 return err;
 897
 898         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 899                 return -EAFNOSUPPORT;
 900
 901         spin_lock_bh(&vxlan->hash_lock);
 902         err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags,
 903                                port, src_vni, vni, ifindex, ndm->ndm_flags);
 904         spin_unlock_bh(&vxlan->hash_lock);
 905
 906         return err;
 907 }
 908
 909 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
 910                               const unsigned char *addr, union vxlan_addr ip,
 911                               __be16 port, __be32 src_vni, __be32 vni,
 912                               u32 ifindex, u16 vid)
 913 {
 914         struct vxlan_fdb *f;
 915         struct vxlan_rdst *rd = NULL;
 916         int err = -ENOENT;
 917
 918         f = vxlan_find_mac(vxlan, addr, src_vni);
 919         if (!f)
 920                 return err;
 921
 922         if (!vxlan_addr_any(&ip)) {
 923                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 924                 if (!rd)
 925                         goto out;
 926         }
 927
 928         /* remove a destination if it's not the only one on the list,
 929          * otherwise destroy the fdb entry
 930          */
 931         if (rd && !list_is_singular(&f->remotes)) {
 932                 vxlan_fdb_dst_destroy(vxlan, f, rd);
 933                 goto out;
 934         }
 935
 936         vxlan_fdb_destroy(vxlan, f, true);
 937
 938 out:
 939         return 0;
 940 }
 941
 942 /* Delete entry (via netlink) */
 943 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 944                             struct net_device *dev,
 945                             const unsigned char *addr, u16 vid)
 946 {
 947         struct vxlan_dev *vxlan = netdev_priv(dev);
 948         union vxlan_addr ip;
 949         __be32 src_vni, vni;
 950         __be16 port;
 951         u32 ifindex;
 952         int err;
 953
 954         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 955         if (err)
 956                 return err;
 957
 958         spin_lock_bh(&vxlan->hash_lock);
 959         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
 960                                  vid);
 961         spin_unlock_bh(&vxlan->hash_lock);
 962
 963         return err;
 964 }
 965
 966 /* Dump forwarding table */
 967 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 968                           struct net_device *dev,
 969                           struct net_device *filter_dev, int *idx)
 970 {
 971         struct vxlan_dev *vxlan = netdev_priv(dev);
 972         unsigned int h;
 973         int err = 0;
 974
 975         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 976                 struct vxlan_fdb *f;
 977
 978                 rcu_read_lock();
 979                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 980                         struct vxlan_rdst *rd;
 981
 982                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 983                                 if (*idx < cb->args[2])
 984                                         goto skip;
 985
 986                                 err = vxlan_fdb_info(skb, vxlan, f,
 987                                                      NETLINK_CB(cb->skb).portid,
 988                                                      cb->nlh->nlmsg_seq,
 989                                                      RTM_NEWNEIGH,
 990                                                      NLM_F_MULTI, rd);
 991                                 if (err < 0) {
 992                                         rcu_read_unlock();
 993                                         goto out;
 994                                 }
 995 skip:
 996                                 *idx += 1;
 997                         }
 998                 }
 999                 rcu_read_unlock();
1000         }
1001 out:
1002         return err;
1003 }
1004
1005 /* Watch incoming packets to learn mapping between Ethernet address
1006  * and Tunnel endpoint.
1007  * Return true if packet is bogus and should be dropped.
1008  */
1009 static bool vxlan_snoop(struct net_device *dev,
1010                         union vxlan_addr *src_ip, const u8 *src_mac,
1011                         u32 src_ifindex, __be32 vni)
1012 {
1013         struct vxlan_dev *vxlan = netdev_priv(dev);
1014         struct vxlan_fdb *f;
1015         u32 ifindex = 0;
1016
1017 #if IS_ENABLED(CONFIG_IPV6)
1018         if (src_ip->sa.sa_family == AF_INET6 &&
1019             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1020                 ifindex = src_ifindex;
1021 #endif
1022
1023         f = vxlan_find_mac(vxlan, src_mac, vni);
1024         if (likely(f)) {
1025                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1026
1027                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1028                            rdst->remote_ifindex == ifindex))
1029                         return false;
1030
1031                 /* Don't migrate static entries, drop packets */
1032                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1033                         return true;
1034
1035                 if (net_ratelimit())
1036                         netdev_info(dev,
1037                                     "%pM migrated from %pIS to %pIS\n",
1038                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1039
1040                 rdst->remote_ip = *src_ip;
1041                 f->updated = jiffies;
1042                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1043         } else {
1044                 /* learned new entry */
1045                 spin_lock(&vxlan->hash_lock);
1046
1047                 /* close off race between vxlan_flush and incoming packets */
1048                 if (netif_running(dev))
1049                         vxlan_fdb_update(vxlan, src_mac, src_ip,
1050                                          NUD_REACHABLE,
1051                                          NLM_F_EXCL|NLM_F_CREATE,
1052                                          vxlan->cfg.dst_port,
1053                                          vni,
1054                                          vxlan->default_dst.remote_vni,
1055                                          ifindex, NTF_SELF);
1056                 spin_unlock(&vxlan->hash_lock);
1057         }
1058
1059         return false;
1060 }
1061
1062 /* See if multicast group is already in use by other ID */
1063 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1064 {
1065         struct vxlan_dev *vxlan;
1066         struct vxlan_sock *sock4;
1067 #if IS_ENABLED(CONFIG_IPV6)
1068         struct vxlan_sock *sock6;
1069 #endif
1070         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1071
1072         sock4 = rtnl_dereference(dev->vn4_sock);
1073
1074         /* The vxlan_sock is only used by dev, leaving group has
1075          * no effect on other vxlan devices.
1076          */
1077         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1078                 return false;
1079 #if IS_ENABLED(CONFIG_IPV6)
1080         sock6 = rtnl_dereference(dev->vn6_sock);
1081         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1082                 return false;
1083 #endif
1084
1085         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1086                 if (!netif_running(vxlan->dev) || vxlan == dev)
1087                         continue;
1088
1089                 if (family == AF_INET &&
1090                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1091                         continue;
1092 #if IS_ENABLED(CONFIG_IPV6)
1093                 if (family == AF_INET6 &&
1094                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1095                         continue;
1096 #endif
1097
1098                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1099                                       &dev->default_dst.remote_ip))
1100                         continue;
1101
1102                 if (vxlan->default_dst.remote_ifindex !=
1103                     dev->default_dst.remote_ifindex)
1104                         continue;
1105
1106                 return true;
1107         }
1108
1109         return false;
1110 }
1111
1112 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1113 {
1114         struct vxlan_net *vn;
1115
1116         if (!vs)
1117                 return false;
1118         if (!refcount_dec_and_test(&vs->refcnt))
1119                 return false;
1120
1121         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1122         spin_lock(&vn->sock_lock);
1123         hlist_del_rcu(&vs->hlist);
1124         udp_tunnel_notify_del_rx_port(vs->sock,
1125                                       (vs->flags & VXLAN_F_GPE) ?
1126                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1127                                       UDP_TUNNEL_TYPE_VXLAN);
1128         spin_unlock(&vn->sock_lock);
1129
1130         return true;
1131 }
1132
1133 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1134 {
1135         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1136 #if IS_ENABLED(CONFIG_IPV6)
1137         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1138
1139         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1140 #endif
1141
1142         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1143         synchronize_net();
1144
1145         vxlan_vs_del_dev(vxlan);
1146
1147         if (__vxlan_sock_release_prep(sock4)) {
1148                 udp_tunnel_sock_release(sock4->sock);
1149                 kfree(sock4);
1150         }
1151
1152 #if IS_ENABLED(CONFIG_IPV6)
1153         if (__vxlan_sock_release_prep(sock6)) {
1154                 udp_tunnel_sock_release(sock6->sock);
1155                 kfree(sock6);
1156         }
1157 #endif
1158 }
1159
1160 /* Update multicast group membership when first VNI on
1161  * multicast address is brought up
1162  */
1163 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1164 {
1165         struct sock *sk;
1166         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1167         int ifindex = vxlan->default_dst.remote_ifindex;
1168         int ret = -EINVAL;
1169
1170         if (ip->sa.sa_family == AF_INET) {
1171                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1172                 struct ip_mreqn mreq = {
1173                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1174                         .imr_ifindex            = ifindex,
1175                 };
1176
1177                 sk = sock4->sock->sk;
1178                 lock_sock(sk);
1179                 ret = ip_mc_join_group(sk, &mreq);
1180                 release_sock(sk);
1181 #if IS_ENABLED(CONFIG_IPV6)
1182         } else {
1183                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1184
1185                 sk = sock6->sock->sk;
1186                 lock_sock(sk);
1187                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1188                                                    &ip->sin6.sin6_addr);
1189                 release_sock(sk);
1190 #endif
1191         }
1192
1193         return ret;
1194 }
1195
1196 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1197 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1198 {
1199         struct sock *sk;
1200         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1201         int ifindex = vxlan->default_dst.remote_ifindex;
1202         int ret = -EINVAL;
1203
1204         if (ip->sa.sa_family == AF_INET) {
1205                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1206                 struct ip_mreqn mreq = {
1207                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1208                         .imr_ifindex            = ifindex,
1209                 };
1210
1211                 sk = sock4->sock->sk;
1212                 lock_sock(sk);
1213                 ret = ip_mc_leave_group(sk, &mreq);
1214                 release_sock(sk);
1215 #if IS_ENABLED(CONFIG_IPV6)
1216         } else {
1217                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1218
1219                 sk = sock6->sock->sk;
1220                 lock_sock(sk);
1221                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1222                                                    &ip->sin6.sin6_addr);
1223                 release_sock(sk);
1224 #endif
1225         }
1226
1227         return ret;
1228 }
1229
1230 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1231                           struct sk_buff *skb, u32 vxflags)
1232 {
1233         size_t start, offset;
1234
1235         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1236                 goto out;
1237
1238         start = vxlan_rco_start(unparsed->vx_vni);
1239         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1240
1241         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1242                 return false;
1243
1244         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1245                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1246 out:
1247         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1248         unparsed->vx_vni &= VXLAN_VNI_MASK;
1249         return true;
1250 }
1251
1252 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1253                                 struct sk_buff *skb, u32 vxflags,
1254                                 struct vxlan_metadata *md)
1255 {
1256         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1257         struct metadata_dst *tun_dst;
1258
1259         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1260                 goto out;
1261
1262         md->gbp = ntohs(gbp->policy_id);
1263
1264         tun_dst = (struct metadata_dst *)skb_dst(skb);
1265         if (tun_dst) {
1266                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1267                 tun_dst->u.tun_info.options_len = sizeof(*md);
1268         }
1269         if (gbp->dont_learn)
1270                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1271
1272         if (gbp->policy_applied)
1273                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1274
1275         /* In flow-based mode, GBP is carried in dst_metadata */
1276         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1277                 skb->mark = md->gbp;
1278 out:
1279         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1280 }
1281
1282 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1283                                 __be16 *protocol,
1284                                 struct sk_buff *skb, u32 vxflags)
1285 {
1286         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1287
1288         /* Need to have Next Protocol set for interfaces in GPE mode. */
1289         if (!gpe->np_applied)
1290                 return false;
1291         /* "The initial version is 0. If a receiver does not support the
1292          * version indicated it MUST drop the packet.
1293          */
1294         if (gpe->version != 0)
1295                 return false;
1296         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1297          * processing MUST occur." However, we don't implement OAM
1298          * processing, thus drop the packet.
1299          */
1300         if (gpe->oam_flag)
1301                 return false;
1302
1303         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1304         if (!*protocol)
1305                 return false;
1306
1307         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1308         return true;
1309 }
1310
1311 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1312                           struct vxlan_sock *vs,
1313                           struct sk_buff *skb, __be32 vni)
1314 {
1315         union vxlan_addr saddr;
1316         u32 ifindex = skb->dev->ifindex;
1317
1318         skb_reset_mac_header(skb);
1319         skb->protocol = eth_type_trans(skb, vxlan->dev);
1320         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1321
1322         /* Ignore packet loops (and multicast echo) */
1323         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1324                 return false;
1325
1326         /* Get address from the outer IP header */
1327         if (vxlan_get_sk_family(vs) == AF_INET) {
1328                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1329                 saddr.sa.sa_family = AF_INET;
1330 #if IS_ENABLED(CONFIG_IPV6)
1331         } else {
1332                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1333                 saddr.sa.sa_family = AF_INET6;
1334 #endif
1335         }
1336
1337         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1338             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1339                 return false;
1340
1341         return true;
1342 }
1343
1344 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1345                                   struct sk_buff *skb)
1346 {
1347         int err = 0;
1348
1349         if (vxlan_get_sk_family(vs) == AF_INET)
1350                 err = IP_ECN_decapsulate(oiph, skb);
1351 #if IS_ENABLED(CONFIG_IPV6)
1352         else
1353                 err = IP6_ECN_decapsulate(oiph, skb);
1354 #endif
1355
1356         if (unlikely(err) && log_ecn_error) {
1357                 if (vxlan_get_sk_family(vs) == AF_INET)
1358                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1359                                              &((struct iphdr *)oiph)->saddr,
1360                                              ((struct iphdr *)oiph)->tos);
1361                 else
1362                         net_info_ratelimited("non-ECT from %pI6\n",
1363                                              &((struct ipv6hdr *)oiph)->saddr);
1364         }
1365         return err <= 1;
1366 }
1367
1368 /* Callback from net/ipv4/udp.c to receive packets */
1369 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1370 {
1371         struct pcpu_sw_netstats *stats;
1372         struct vxlan_dev *vxlan;
1373         struct vxlan_sock *vs;
1374         struct vxlanhdr unparsed;
1375         struct vxlan_metadata _md;
1376         struct vxlan_metadata *md = &_md;
1377         __be16 protocol = htons(ETH_P_TEB);
1378         bool raw_proto = false;
1379         void *oiph;
1380         __be32 vni = 0;
1381
1382         /* Need UDP and VXLAN header to be present */
1383         if (!pskb_may_pull(skb, VXLAN_HLEN))
1384                 goto drop;
1385
1386         unparsed = *vxlan_hdr(skb);
1387         /* VNI flag always required to be set */
1388         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1389                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1390                            ntohl(vxlan_hdr(skb)->vx_flags),
1391                            ntohl(vxlan_hdr(skb)->vx_vni));
1392                 /* Return non vxlan pkt */
1393                 goto drop;
1394         }
1395         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1396         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1397
1398         vs = rcu_dereference_sk_user_data(sk);
1399         if (!vs)
1400                 goto drop;
1401
1402         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1403
1404         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1405         if (!vxlan)
1406                 goto drop;
1407
1408         /* For backwards compatibility, only allow reserved fields to be
1409          * used by VXLAN extensions if explicitly requested.
1410          */
1411         if (vs->flags & VXLAN_F_GPE) {
1412                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1413                         goto drop;
1414                 raw_proto = true;
1415         }
1416
1417         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1418                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1419                         goto drop;
1420
1421         if (vxlan_collect_metadata(vs)) {
1422                 struct metadata_dst *tun_dst;
1423
1424                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1425                                          key32_to_tunnel_id(vni), sizeof(*md));
1426
1427                 if (!tun_dst)
1428                         goto drop;
1429
1430                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1431
1432                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1433         } else {
1434                 memset(md, 0, sizeof(*md));
1435         }
1436
1437         if (vs->flags & VXLAN_F_REMCSUM_RX)
1438                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1439                         goto drop;
1440         if (vs->flags & VXLAN_F_GBP)
1441                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1442         /* Note that GBP and GPE can never be active together. This is
1443          * ensured in vxlan_dev_configure.
1444          */
1445
1446         if (unparsed.vx_flags || unparsed.vx_vni) {
1447                 /* If there are any unprocessed flags remaining treat
1448                  * this as a malformed packet. This behavior diverges from
1449                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1450                  * in reserved fields are to be ignored. The approach here
1451                  * maintains compatibility with previous stack code, and also
1452                  * is more robust and provides a little more security in
1453                  * adding extensions to VXLAN.
1454                  */
1455                 goto drop;
1456         }
1457
1458         if (!raw_proto) {
1459                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1460                         goto drop;
1461         } else {
1462                 skb_reset_mac_header(skb);
1463                 skb->dev = vxlan->dev;
1464                 skb->pkt_type = PACKET_HOST;
1465         }
1466
1467         oiph = skb_network_header(skb);
1468         skb_reset_network_header(skb);
1469
1470         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1471                 ++vxlan->dev->stats.rx_frame_errors;
1472                 ++vxlan->dev->stats.rx_errors;
1473                 goto drop;
1474         }
1475
1476         rcu_read_lock();
1477
1478         if (unlikely(!(vxlan->dev->flags & IFF_UP))) {
1479                 rcu_read_unlock();
1480                 atomic_long_inc(&vxlan->dev->rx_dropped);
1481                 goto drop;
1482         }
1483
1484         stats = this_cpu_ptr(vxlan->dev->tstats);
1485         u64_stats_update_begin(&stats->syncp);
1486         stats->rx_packets++;
1487         stats->rx_bytes += skb->len;
1488         u64_stats_update_end(&stats->syncp);
1489
1490         gro_cells_receive(&vxlan->gro_cells, skb);
1491
1492         rcu_read_unlock();
1493
1494         return 0;
1495
1496 drop:
1497         /* Consume bad packet */
1498         kfree_skb(skb);
1499         return 0;
1500 }
1501
1502 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1503 {
1504         struct vxlan_dev *vxlan = netdev_priv(dev);
1505         struct arphdr *parp;
1506         u8 *arpptr, *sha;
1507         __be32 sip, tip;
1508         struct neighbour *n;
1509
1510         if (dev->flags & IFF_NOARP)
1511                 goto out;
1512
1513         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1514                 dev->stats.tx_dropped++;
1515                 goto out;
1516         }
1517         parp = arp_hdr(skb);
1518
1519         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1520              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1521             parp->ar_pro != htons(ETH_P_IP) ||
1522             parp->ar_op != htons(ARPOP_REQUEST) ||
1523             parp->ar_hln != dev->addr_len ||
1524             parp->ar_pln != 4)
1525                 goto out;
1526         arpptr = (u8 *)parp + sizeof(struct arphdr);
1527         sha = arpptr;
1528         arpptr += dev->addr_len;        /* sha */
1529         memcpy(&sip, arpptr, sizeof(sip));
1530         arpptr += sizeof(sip);
1531         arpptr += dev->addr_len;        /* tha */
1532         memcpy(&tip, arpptr, sizeof(tip));
1533
1534         if (ipv4_is_loopback(tip) ||
1535             ipv4_is_multicast(tip))
1536                 goto out;
1537
1538         n = neigh_lookup(&arp_tbl, &tip, dev);
1539
1540         if (n) {
1541                 struct vxlan_fdb *f;
1542                 struct sk_buff  *reply;
1543
1544                 if (!(n->nud_state & NUD_CONNECTED)) {
1545                         neigh_release(n);
1546                         goto out;
1547                 }
1548
1549                 f = vxlan_find_mac(vxlan, n->ha, vni);
1550                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1551                         /* bridge-local neighbor */
1552                         neigh_release(n);
1553                         goto out;
1554                 }
1555
1556                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1557                                 n->ha, sha);
1558
1559                 neigh_release(n);
1560
1561                 if (reply == NULL)
1562                         goto out;
1563
1564                 skb_reset_mac_header(reply);
1565                 __skb_pull(reply, skb_network_offset(reply));
1566                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1567                 reply->pkt_type = PACKET_HOST;
1568
1569                 if (netif_rx_ni(reply) == NET_RX_DROP)
1570                         dev->stats.rx_dropped++;
1571         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1572                 union vxlan_addr ipa = {
1573                         .sin.sin_addr.s_addr = tip,
1574                         .sin.sin_family = AF_INET,
1575                 };
1576
1577                 vxlan_ip_miss(dev, &ipa);
1578         }
1579 out:
1580         consume_skb(skb);
1581         return NETDEV_TX_OK;
1582 }
1583
1584 #if IS_ENABLED(CONFIG_IPV6)
1585 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1586         struct neighbour *n, bool isrouter)
1587 {
1588         struct net_device *dev = request->dev;
1589         struct sk_buff *reply;
1590         struct nd_msg *ns, *na;
1591         struct ipv6hdr *pip6;
1592         u8 *daddr;
1593         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1594         int ns_olen;
1595         int i, len;
1596
1597         if (dev == NULL || !pskb_may_pull(request, request->len))
1598                 return NULL;
1599
1600         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1601                 sizeof(*na) + na_olen + dev->needed_tailroom;
1602         reply = alloc_skb(len, GFP_ATOMIC);
1603         if (reply == NULL)
1604                 return NULL;
1605
1606         reply->protocol = htons(ETH_P_IPV6);
1607         reply->dev = dev;
1608         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1609         skb_push(reply, sizeof(struct ethhdr));
1610         skb_reset_mac_header(reply);
1611
1612         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1613
1614         daddr = eth_hdr(request)->h_source;
1615         ns_olen = request->len - skb_network_offset(request) -
1616                 sizeof(struct ipv6hdr) - sizeof(*ns);
1617         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1618                 if (!ns->opt[i + 1]) {
1619                         kfree_skb(reply);
1620                         return NULL;
1621                 }
1622                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1623                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1624                         break;
1625                 }
1626         }
1627
1628         /* Ethernet header */
1629         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1630         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1631         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1632         reply->protocol = htons(ETH_P_IPV6);
1633
1634         skb_pull(reply, sizeof(struct ethhdr));
1635         skb_reset_network_header(reply);
1636         skb_put(reply, sizeof(struct ipv6hdr));
1637
1638         /* IPv6 header */
1639
1640         pip6 = ipv6_hdr(reply);
1641         memset(pip6, 0, sizeof(struct ipv6hdr));
1642         pip6->version = 6;
1643         pip6->priority = ipv6_hdr(request)->priority;
1644         pip6->nexthdr = IPPROTO_ICMPV6;
1645         pip6->hop_limit = 255;
1646         pip6->daddr = ipv6_hdr(request)->saddr;
1647         pip6->saddr = *(struct in6_addr *)n->primary_key;
1648
1649         skb_pull(reply, sizeof(struct ipv6hdr));
1650         skb_reset_transport_header(reply);
1651
1652         /* Neighbor Advertisement */
1653         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1654         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1655         na->icmph.icmp6_router = isrouter;
1656         na->icmph.icmp6_override = 1;
1657         na->icmph.icmp6_solicited = 1;
1658         na->target = ns->target;
1659         ether_addr_copy(&na->opt[2], n->ha);
1660         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1661         na->opt[1] = na_olen >> 3;
1662
1663         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1664                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1665                 csum_partial(na, sizeof(*na)+na_olen, 0));
1666
1667         pip6->payload_len = htons(sizeof(*na)+na_olen);
1668
1669         skb_push(reply, sizeof(struct ipv6hdr));
1670
1671         reply->ip_summed = CHECKSUM_UNNECESSARY;
1672
1673         return reply;
1674 }
1675
1676 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1677 {
1678         struct vxlan_dev *vxlan = netdev_priv(dev);
1679         const struct in6_addr *daddr;
1680         const struct ipv6hdr *iphdr;
1681         struct inet6_dev *in6_dev;
1682         struct neighbour *n;
1683         struct nd_msg *msg;
1684
1685         rcu_read_lock();
1686         in6_dev = __in6_dev_get(dev);
1687         if (!in6_dev)
1688                 goto out;
1689
1690         iphdr = ipv6_hdr(skb);
1691         daddr = &iphdr->daddr;
1692         msg = (struct nd_msg *)(iphdr + 1);
1693
1694         if (ipv6_addr_loopback(daddr) ||
1695             ipv6_addr_is_multicast(&msg->target))
1696                 goto out;
1697
1698         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1699
1700         if (n) {
1701                 struct vxlan_fdb *f;
1702                 struct sk_buff *reply;
1703
1704                 if (!(n->nud_state & NUD_CONNECTED)) {
1705                         neigh_release(n);
1706                         goto out;
1707                 }
1708
1709                 f = vxlan_find_mac(vxlan, n->ha, vni);
1710                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1711                         /* bridge-local neighbor */
1712                         neigh_release(n);
1713                         goto out;
1714                 }
1715
1716                 reply = vxlan_na_create(skb, n,
1717                                         !!(f ? f->flags & NTF_ROUTER : 0));
1718
1719                 neigh_release(n);
1720
1721                 if (reply == NULL)
1722                         goto out;
1723
1724                 if (netif_rx_ni(reply) == NET_RX_DROP)
1725                         dev->stats.rx_dropped++;
1726
1727         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1728                 union vxlan_addr ipa = {
1729                         .sin6.sin6_addr = msg->target,
1730                         .sin6.sin6_family = AF_INET6,
1731                 };
1732
1733                 vxlan_ip_miss(dev, &ipa);
1734         }
1735
1736 out:
1737         rcu_read_unlock();
1738         consume_skb(skb);
1739         return NETDEV_TX_OK;
1740 }
1741 #endif
1742
1743 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1744 {
1745         struct vxlan_dev *vxlan = netdev_priv(dev);
1746         struct neighbour *n;
1747
1748         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1749                 return false;
1750
1751         n = NULL;
1752         switch (ntohs(eth_hdr(skb)->h_proto)) {
1753         case ETH_P_IP:
1754         {
1755                 struct iphdr *pip;
1756
1757                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1758                         return false;
1759                 pip = ip_hdr(skb);
1760                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1761                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1762                         union vxlan_addr ipa = {
1763                                 .sin.sin_addr.s_addr = pip->daddr,
1764                                 .sin.sin_family = AF_INET,
1765                         };
1766
1767                         vxlan_ip_miss(dev, &ipa);
1768                         return false;
1769                 }
1770
1771                 break;
1772         }
1773 #if IS_ENABLED(CONFIG_IPV6)
1774         case ETH_P_IPV6:
1775         {
1776                 struct ipv6hdr *pip6;
1777
1778                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1779                         return false;
1780                 pip6 = ipv6_hdr(skb);
1781                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1782                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1783                         union vxlan_addr ipa = {
1784                                 .sin6.sin6_addr = pip6->daddr,
1785                                 .sin6.sin6_family = AF_INET6,
1786                         };
1787
1788                         vxlan_ip_miss(dev, &ipa);
1789                         return false;
1790                 }
1791
1792                 break;
1793         }
1794 #endif
1795         default:
1796                 return false;
1797         }
1798
1799         if (n) {
1800                 bool diff;
1801
1802                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1803                 if (diff) {
1804                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1805                                 dev->addr_len);
1806                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1807                 }
1808                 neigh_release(n);
1809                 return diff;
1810         }
1811
1812         return false;
1813 }
1814
1815 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1816                                 struct vxlan_metadata *md)
1817 {
1818         struct vxlanhdr_gbp *gbp;
1819
1820         if (!md->gbp)
1821                 return;
1822
1823         gbp = (struct vxlanhdr_gbp *)vxh;
1824         vxh->vx_flags |= VXLAN_HF_GBP;
1825
1826         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1827                 gbp->dont_learn = 1;
1828
1829         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1830                 gbp->policy_applied = 1;
1831
1832         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1833 }
1834
1835 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1836                                __be16 protocol)
1837 {
1838         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1839
1840         gpe->np_applied = 1;
1841         gpe->next_protocol = tun_p_from_eth_p(protocol);
1842         if (!gpe->next_protocol)
1843                 return -EPFNOSUPPORT;
1844         return 0;
1845 }
1846
1847 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1848                            int iphdr_len, __be32 vni,
1849                            struct vxlan_metadata *md, u32 vxflags,
1850                            bool udp_sum)
1851 {
1852         struct vxlanhdr *vxh;
1853         int min_headroom;
1854         int err;
1855         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1856         __be16 inner_protocol = htons(ETH_P_TEB);
1857
1858         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1859             skb->ip_summed == CHECKSUM_PARTIAL) {
1860                 int csum_start = skb_checksum_start_offset(skb);
1861
1862                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1863                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1864                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1865                      skb->csum_offset == offsetof(struct tcphdr, check)))
1866                         type |= SKB_GSO_TUNNEL_REMCSUM;
1867         }
1868
1869         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1870                         + VXLAN_HLEN + iphdr_len;
1871
1872         /* Need space for new headers (invalidates iph ptr) */
1873         err = skb_cow_head(skb, min_headroom);
1874         if (unlikely(err))
1875                 return err;
1876
1877         err = iptunnel_handle_offloads(skb, type);
1878         if (err)
1879                 return err;
1880
1881         vxh = __skb_push(skb, sizeof(*vxh));
1882         vxh->vx_flags = VXLAN_HF_VNI;
1883         vxh->vx_vni = vxlan_vni_field(vni);
1884
1885         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1886                 unsigned int start;
1887
1888                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1889                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1890                 vxh->vx_flags |= VXLAN_HF_RCO;
1891
1892                 if (!skb_is_gso(skb)) {
1893                         skb->ip_summed = CHECKSUM_NONE;
1894                         skb->encapsulation = 0;
1895                 }
1896         }
1897
1898         if (vxflags & VXLAN_F_GBP)
1899                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1900         if (vxflags & VXLAN_F_GPE) {
1901                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1902                 if (err < 0)
1903                         return err;
1904                 inner_protocol = skb->protocol;
1905         }
1906
1907         skb_set_inner_protocol(skb, inner_protocol);
1908         return 0;
1909 }
1910
1911 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1912                                       struct vxlan_sock *sock4,
1913                                       struct sk_buff *skb, int oif, u8 tos,
1914                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
1915                                       struct dst_cache *dst_cache,
1916                                       const struct ip_tunnel_info *info)
1917 {
1918         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1919         struct rtable *rt = NULL;
1920         struct flowi4 fl4;
1921
1922         if (!sock4)
1923                 return ERR_PTR(-EIO);
1924
1925         if (tos && !info)
1926                 use_cache = false;
1927         if (use_cache) {
1928                 rt = dst_cache_get_ip4(dst_cache, saddr);
1929                 if (rt)
1930                         return rt;
1931         }
1932
1933         memset(&fl4, 0, sizeof(fl4));
1934         fl4.flowi4_oif = oif;
1935         fl4.flowi4_tos = RT_TOS(tos);
1936         fl4.flowi4_mark = skb->mark;
1937         fl4.flowi4_proto = IPPROTO_UDP;
1938         fl4.daddr = daddr;
1939         fl4.saddr = *saddr;
1940         fl4.fl4_dport = dport;
1941         fl4.fl4_sport = sport;
1942
1943         rt = ip_route_output_key(vxlan->net, &fl4);
1944         if (likely(!IS_ERR(rt))) {
1945                 if (rt->dst.dev == dev) {
1946                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1947                         ip_rt_put(rt);
1948                         return ERR_PTR(-ELOOP);
1949                 }
1950
1951                 *saddr = fl4.saddr;
1952                 if (use_cache)
1953                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1954         } else {
1955                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1956                 return ERR_PTR(-ENETUNREACH);
1957         }
1958         return rt;
1959 }
1960
1961 #if IS_ENABLED(CONFIG_IPV6)
1962 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1963                                           struct net_device *dev,
1964                                           struct vxlan_sock *sock6,
1965                                           struct sk_buff *skb, int oif, u8 tos,
1966                                           __be32 label,
1967                                           const struct in6_addr *daddr,
1968                                           struct in6_addr *saddr,
1969                                           __be16 dport, __be16 sport,
1970                                           struct dst_cache *dst_cache,
1971                                           const struct ip_tunnel_info *info)
1972 {
1973         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1974         struct dst_entry *ndst;
1975         struct flowi6 fl6;
1976
1977         if (!sock6)
1978                 return ERR_PTR(-EIO);
1979
1980         if (tos && !info)
1981                 use_cache = false;
1982         if (use_cache) {
1983                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1984                 if (ndst)
1985                         return ndst;
1986         }
1987
1988         memset(&fl6, 0, sizeof(fl6));
1989         fl6.flowi6_oif = oif;
1990         fl6.daddr = *daddr;
1991         fl6.saddr = *saddr;
1992         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1993         fl6.flowi6_mark = skb->mark;
1994         fl6.flowi6_proto = IPPROTO_UDP;
1995         fl6.fl6_dport = dport;
1996         fl6.fl6_sport = sport;
1997
1998         ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk,
1999                                                &fl6, NULL);
2000         if (unlikely(IS_ERR(ndst))) {
2001                 netdev_dbg(dev, "no route to %pI6\n", daddr);
2002                 return ERR_PTR(-ENETUNREACH);
2003         }
2004
2005         if (unlikely(ndst->dev == dev)) {
2006                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
2007                 dst_release(ndst);
2008                 return ERR_PTR(-ELOOP);
2009         }
2010
2011         *saddr = fl6.saddr;
2012         if (use_cache)
2013                 dst_cache_set_ip6(dst_cache, ndst, saddr);
2014         return ndst;
2015 }
2016 #endif
2017
2018 /* Bypass encapsulation if the destination is local */
2019 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2020                                struct vxlan_dev *dst_vxlan, __be32 vni)
2021 {
2022         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2023         union vxlan_addr loopback;
2024         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2025         struct net_device *dev;
2026         int len = skb->len;
2027
2028         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2029         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2030         skb->pkt_type = PACKET_HOST;
2031         skb->encapsulation = 0;
2032         skb->dev = dst_vxlan->dev;
2033         __skb_pull(skb, skb_network_offset(skb));
2034
2035         if (remote_ip->sa.sa_family == AF_INET) {
2036                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2037                 loopback.sa.sa_family =  AF_INET;
2038 #if IS_ENABLED(CONFIG_IPV6)
2039         } else {
2040                 loopback.sin6.sin6_addr = in6addr_loopback;
2041                 loopback.sa.sa_family =  AF_INET6;
2042 #endif
2043         }
2044
2045         rcu_read_lock();
2046         dev = skb->dev;
2047         if (unlikely(!(dev->flags & IFF_UP))) {
2048                 kfree_skb(skb);
2049                 goto drop;
2050         }
2051
2052         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
2053                 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni);
2054
2055         u64_stats_update_begin(&tx_stats->syncp);
2056         tx_stats->tx_packets++;
2057         tx_stats->tx_bytes += len;
2058         u64_stats_update_end(&tx_stats->syncp);
2059
2060         if (netif_rx(skb) == NET_RX_SUCCESS) {
2061                 u64_stats_update_begin(&rx_stats->syncp);
2062                 rx_stats->rx_packets++;
2063                 rx_stats->rx_bytes += len;
2064                 u64_stats_update_end(&rx_stats->syncp);
2065         } else {
2066 drop:
2067                 dev->stats.rx_dropped++;
2068         }
2069         rcu_read_unlock();
2070 }
2071
2072 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2073                                  struct vxlan_dev *vxlan,
2074                                  union vxlan_addr *daddr,
2075                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2076                                  struct dst_entry *dst,
2077                                  u32 rt_flags)
2078 {
2079 #if IS_ENABLED(CONFIG_IPV6)
2080         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2081          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2082          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2083          */
2084         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2085 #endif
2086         /* Bypass encapsulation if the destination is local */
2087         if (rt_flags & RTCF_LOCAL &&
2088             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2089                 struct vxlan_dev *dst_vxlan;
2090
2091                 dst_release(dst);
2092                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2093                                            daddr->sa.sa_family, dst_port,
2094                                            vxlan->cfg.flags);
2095                 if (!dst_vxlan) {
2096                         dev->stats.tx_errors++;
2097                         kfree_skb(skb);
2098
2099                         return -ENOENT;
2100                 }
2101                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2102                 return 1;
2103         }
2104
2105         return 0;
2106 }
2107
2108 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2109                            __be32 default_vni, struct vxlan_rdst *rdst,
2110                            bool did_rsc)
2111 {
2112         struct dst_cache *dst_cache;
2113         struct ip_tunnel_info *info;
2114         struct vxlan_dev *vxlan = netdev_priv(dev);
2115         const struct iphdr *old_iph = ip_hdr(skb);
2116         union vxlan_addr *dst;
2117         union vxlan_addr remote_ip, local_ip;
2118         struct vxlan_metadata _md;
2119         struct vxlan_metadata *md = &_md;
2120         __be16 src_port = 0, dst_port;
2121         struct dst_entry *ndst = NULL;
2122         __be32 vni, label;
2123         __u8 tos, ttl;
2124         int ifindex;
2125         int err;
2126         u32 flags = vxlan->cfg.flags;
2127         bool udp_sum = false;
2128         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2129
2130         info = skb_tunnel_info(skb);
2131
2132         if (rdst) {
2133                 dst = &rdst->remote_ip;
2134                 if (vxlan_addr_any(dst)) {
2135                         if (did_rsc) {
2136                                 /* short-circuited back to local bridge */
2137                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2138                                 return;
2139                         }
2140                         goto drop;
2141                 }
2142
2143                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2144                 vni = (rdst->remote_vni) ? : default_vni;
2145                 ifindex = rdst->remote_ifindex;
2146                 local_ip = vxlan->cfg.saddr;
2147                 dst_cache = &rdst->dst_cache;
2148                 md->gbp = skb->mark;
2149                 if (flags & VXLAN_F_TTL_INHERIT) {
2150                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2151                 } else {
2152                         ttl = vxlan->cfg.ttl;
2153                         if (!ttl && vxlan_addr_multicast(dst))
2154                                 ttl = 1;
2155                 }
2156
2157                 tos = vxlan->cfg.tos;
2158                 if (tos == 1)
2159                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2160
2161                 if (dst->sa.sa_family == AF_INET)
2162                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2163                 else
2164                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2165                 label = vxlan->cfg.label;
2166         } else {
2167                 if (!info) {
2168                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2169                                   dev->name);
2170                         goto drop;
2171                 }
2172                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2173                 if (remote_ip.sa.sa_family == AF_INET) {
2174                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2175                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2176                 } else {
2177                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2178                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2179                 }
2180                 dst = &remote_ip;
2181                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2182                 vni = tunnel_id_to_key32(info->key.tun_id);
2183                 ifindex = 0;
2184                 dst_cache = &info->dst_cache;
2185                 if (info->key.tun_flags & TUNNEL_VXLAN_OPT) {
2186                         if (info->options_len < sizeof(*md))
2187                                 goto drop;
2188                         md = ip_tunnel_info_opts(info);
2189                 }
2190                 ttl = info->key.ttl;
2191                 tos = info->key.tos;
2192                 label = info->key.label;
2193                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2194         }
2195         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2196                                      vxlan->cfg.port_max, true);
2197
2198         rcu_read_lock();
2199         if (dst->sa.sa_family == AF_INET) {
2200                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2201                 struct rtable *rt;
2202                 __be16 df = 0;
2203
2204                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2205                                      dst->sin.sin_addr.s_addr,
2206                                      &local_ip.sin.sin_addr.s_addr,
2207                                      dst_port, src_port,
2208                                      dst_cache, info);
2209                 if (IS_ERR(rt)) {
2210                         err = PTR_ERR(rt);
2211                         goto tx_error;
2212                 }
2213
2214                 /* Bypass encapsulation if the destination is local */
2215                 if (!info) {
2216                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2217                                                     dst_port, ifindex, vni,
2218                                                     &rt->dst, rt->rt_flags);
2219                         if (err)
2220                                 goto out_unlock;
2221                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2222                         df = htons(IP_DF);
2223                 }
2224
2225                 ndst = &rt->dst;
2226                 skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM);
2227
2228                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2229                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2230                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2231                                       vni, md, flags, udp_sum);
2232                 if (err < 0)
2233                         goto tx_error;
2234
2235                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2236                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2237                                     src_port, dst_port, xnet, !udp_sum);
2238 #if IS_ENABLED(CONFIG_IPV6)
2239         } else {
2240                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2241
2242                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2243                                         label, &dst->sin6.sin6_addr,
2244                                         &local_ip.sin6.sin6_addr,
2245                                         dst_port, src_port,
2246                                         dst_cache, info);
2247                 if (IS_ERR(ndst)) {
2248                         err = PTR_ERR(ndst);
2249                         ndst = NULL;
2250                         goto tx_error;
2251                 }
2252
2253                 if (!info) {
2254                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2255
2256                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2257                                                     dst_port, ifindex, vni,
2258                                                     ndst, rt6i_flags);
2259                         if (err)
2260                                 goto out_unlock;
2261                 }
2262
2263                 skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM);
2264
2265                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2266                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2267                 skb_scrub_packet(skb, xnet);
2268                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2269                                       vni, md, flags, udp_sum);
2270                 if (err < 0)
2271                         goto tx_error;
2272
2273                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2274                                      &local_ip.sin6.sin6_addr,
2275                                      &dst->sin6.sin6_addr, tos, ttl,
2276                                      label, src_port, dst_port, !udp_sum);
2277 #endif
2278         }
2279 out_unlock:
2280         rcu_read_unlock();
2281         return;
2282
2283 drop:
2284         dev->stats.tx_dropped++;
2285         dev_kfree_skb(skb);
2286         return;
2287
2288 tx_error:
2289         rcu_read_unlock();
2290         if (err == -ELOOP)
2291                 dev->stats.collisions++;
2292         else if (err == -ENETUNREACH)
2293                 dev->stats.tx_carrier_errors++;
2294         dst_release(ndst);
2295         dev->stats.tx_errors++;
2296         kfree_skb(skb);
2297 }
2298
2299 /* Transmit local packets over Vxlan
2300  *
2301  * Outer IP header inherits ECN and DF from inner header.
2302  * Outer UDP destination is the VXLAN assigned port.
2303  *           source port is based on hash of flow
2304  */
2305 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2306 {
2307         struct vxlan_dev *vxlan = netdev_priv(dev);
2308         struct vxlan_rdst *rdst, *fdst = NULL;
2309         const struct ip_tunnel_info *info;
2310         bool did_rsc = false;
2311         struct vxlan_fdb *f;
2312         struct ethhdr *eth;
2313         __be32 vni = 0;
2314
2315         info = skb_tunnel_info(skb);
2316
2317         skb_reset_mac_header(skb);
2318
2319         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2320                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2321                     info->mode & IP_TUNNEL_INFO_TX) {
2322                         vni = tunnel_id_to_key32(info->key.tun_id);
2323                 } else {
2324                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2325                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2326                         else
2327                                 kfree_skb(skb);
2328                         return NETDEV_TX_OK;
2329                 }
2330         }
2331
2332         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2333                 eth = eth_hdr(skb);
2334                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2335                         return arp_reduce(dev, skb, vni);
2336 #if IS_ENABLED(CONFIG_IPV6)
2337                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2338                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2339                                             sizeof(struct nd_msg)) &&
2340                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2341                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2342
2343                         if (m->icmph.icmp6_code == 0 &&
2344                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2345                                 return neigh_reduce(dev, skb, vni);
2346                 }
2347 #endif
2348         }
2349
2350         eth = eth_hdr(skb);
2351         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2352         did_rsc = false;
2353
2354         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2355             (ntohs(eth->h_proto) == ETH_P_IP ||
2356              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2357                 did_rsc = route_shortcircuit(dev, skb);
2358                 if (did_rsc)
2359                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2360         }
2361
2362         if (f == NULL) {
2363                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2364                 if (f == NULL) {
2365                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2366                             !is_multicast_ether_addr(eth->h_dest))
2367                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2368
2369                         dev->stats.tx_dropped++;
2370                         kfree_skb(skb);
2371                         return NETDEV_TX_OK;
2372                 }
2373         }
2374
2375         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2376                 struct sk_buff *skb1;
2377
2378                 if (!fdst) {
2379                         fdst = rdst;
2380                         continue;
2381                 }
2382                 skb1 = skb_clone(skb, GFP_ATOMIC);
2383                 if (skb1)
2384                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2385         }
2386
2387         if (fdst)
2388                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2389         else
2390                 kfree_skb(skb);
2391         return NETDEV_TX_OK;
2392 }
2393
2394 /* Walk the forwarding table and purge stale entries */
2395 static void vxlan_cleanup(struct timer_list *t)
2396 {
2397         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2398         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2399         unsigned int h;
2400
2401         if (!netif_running(vxlan->dev))
2402                 return;
2403
2404         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2405                 struct hlist_node *p, *n;
2406
2407                 spin_lock_bh(&vxlan->hash_lock);
2408                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2409                         struct vxlan_fdb *f
2410                                 = container_of(p, struct vxlan_fdb, hlist);
2411                         unsigned long timeout;
2412
2413                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2414                                 continue;
2415
2416                         if (f->flags & NTF_EXT_LEARNED)
2417                                 continue;
2418
2419                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2420                         if (time_before_eq(timeout, jiffies)) {
2421                                 netdev_dbg(vxlan->dev,
2422                                            "garbage collect %pM\n",
2423                                            f->eth_addr);
2424                                 f->state = NUD_STALE;
2425                                 vxlan_fdb_destroy(vxlan, f, true);
2426                         } else if (time_before(timeout, next_timer))
2427                                 next_timer = timeout;
2428                 }
2429                 spin_unlock_bh(&vxlan->hash_lock);
2430         }
2431
2432         mod_timer(&vxlan->age_timer, next_timer);
2433 }
2434
2435 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2436 {
2437         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2438
2439         spin_lock(&vn->sock_lock);
2440         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2441 #if IS_ENABLED(CONFIG_IPV6)
2442         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2443 #endif
2444         spin_unlock(&vn->sock_lock);
2445 }
2446
2447 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2448                              struct vxlan_dev_node *node)
2449 {
2450         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2451         __be32 vni = vxlan->default_dst.remote_vni;
2452
2453         node->vxlan = vxlan;
2454         spin_lock(&vn->sock_lock);
2455         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2456         spin_unlock(&vn->sock_lock);
2457 }
2458
2459 /* Setup stats when device is created */
2460 static int vxlan_init(struct net_device *dev)
2461 {
2462         struct vxlan_dev *vxlan = netdev_priv(dev);
2463         int err;
2464
2465         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2466         if (!dev->tstats)
2467                 return -ENOMEM;
2468
2469         err = gro_cells_init(&vxlan->gro_cells, dev);
2470         if (err) {
2471                 free_percpu(dev->tstats);
2472                 return err;
2473         }
2474
2475         return 0;
2476 }
2477
2478 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2479 {
2480         struct vxlan_fdb *f;
2481
2482         spin_lock_bh(&vxlan->hash_lock);
2483         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2484         if (f)
2485                 vxlan_fdb_destroy(vxlan, f, true);
2486         spin_unlock_bh(&vxlan->hash_lock);
2487 }
2488
2489 static void vxlan_uninit(struct net_device *dev)
2490 {
2491         struct vxlan_dev *vxlan = netdev_priv(dev);
2492
2493         gro_cells_destroy(&vxlan->gro_cells);
2494
2495         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2496
2497         free_percpu(dev->tstats);
2498 }
2499
2500 /* Start ageing timer and join group when device is brought up */
2501 static int vxlan_open(struct net_device *dev)
2502 {
2503         struct vxlan_dev *vxlan = netdev_priv(dev);
2504         int ret;
2505
2506         ret = vxlan_sock_add(vxlan);
2507         if (ret < 0)
2508                 return ret;
2509
2510         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2511                 ret = vxlan_igmp_join(vxlan);
2512                 if (ret == -EADDRINUSE)
2513                         ret = 0;
2514                 if (ret) {
2515                         vxlan_sock_release(vxlan);
2516                         return ret;
2517                 }
2518         }
2519
2520         if (vxlan->cfg.age_interval)
2521                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2522
2523         return ret;
2524 }
2525
2526 /* Purge the forwarding table */
2527 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2528 {
2529         unsigned int h;
2530
2531         spin_lock_bh(&vxlan->hash_lock);
2532         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2533                 struct hlist_node *p, *n;
2534                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2535                         struct vxlan_fdb *f
2536                                 = container_of(p, struct vxlan_fdb, hlist);
2537                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2538                                 continue;
2539                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2540                         if (!is_zero_ether_addr(f->eth_addr))
2541                                 vxlan_fdb_destroy(vxlan, f, true);
2542                 }
2543         }
2544         spin_unlock_bh(&vxlan->hash_lock);
2545 }
2546
2547 /* Cleanup timer and forwarding table on shutdown */
2548 static int vxlan_stop(struct net_device *dev)
2549 {
2550         struct vxlan_dev *vxlan = netdev_priv(dev);
2551         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2552         int ret = 0;
2553
2554         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2555             !vxlan_group_used(vn, vxlan))
2556                 ret = vxlan_igmp_leave(vxlan);
2557
2558         del_timer_sync(&vxlan->age_timer);
2559
2560         vxlan_flush(vxlan, false);
2561         vxlan_sock_release(vxlan);
2562
2563         return ret;
2564 }
2565
2566 /* Stub, nothing needs to be done. */
2567 static void vxlan_set_multicast_list(struct net_device *dev)
2568 {
2569 }
2570
2571 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2572 {
2573         struct vxlan_dev *vxlan = netdev_priv(dev);
2574         struct vxlan_rdst *dst = &vxlan->default_dst;
2575         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2576                                                          dst->remote_ifindex);
2577         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
2578
2579         /* This check is different than dev->max_mtu, because it looks at
2580          * the lowerdev->mtu, rather than the static dev->max_mtu
2581          */
2582         if (lowerdev) {
2583                 int max_mtu = lowerdev->mtu -
2584                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2585                 if (new_mtu > max_mtu)
2586                         return -EINVAL;
2587         }
2588
2589         dev->mtu = new_mtu;
2590         return 0;
2591 }
2592
2593 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2594 {
2595         struct vxlan_dev *vxlan = netdev_priv(dev);
2596         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2597         __be16 sport, dport;
2598
2599         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2600                                   vxlan->cfg.port_max, true);
2601         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2602
2603         if (ip_tunnel_info_af(info) == AF_INET) {
2604                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2605                 struct rtable *rt;
2606
2607                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2608                                      info->key.u.ipv4.dst,
2609                                      &info->key.u.ipv4.src, dport, sport,
2610                                      &info->dst_cache, info);
2611                 if (IS_ERR(rt))
2612                         return PTR_ERR(rt);
2613                 ip_rt_put(rt);
2614         } else {
2615 #if IS_ENABLED(CONFIG_IPV6)
2616                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2617                 struct dst_entry *ndst;
2618
2619                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2620                                         info->key.label, &info->key.u.ipv6.dst,
2621                                         &info->key.u.ipv6.src, dport, sport,
2622                                         &info->dst_cache, info);
2623                 if (IS_ERR(ndst))
2624                         return PTR_ERR(ndst);
2625                 dst_release(ndst);
2626 #else /* !CONFIG_IPV6 */
2627                 return -EPFNOSUPPORT;
2628 #endif
2629         }
2630         info->key.tp_src = sport;
2631         info->key.tp_dst = dport;
2632         return 0;
2633 }
2634
2635 static const struct net_device_ops vxlan_netdev_ether_ops = {
2636         .ndo_init               = vxlan_init,
2637         .ndo_uninit             = vxlan_uninit,
2638         .ndo_open               = vxlan_open,
2639         .ndo_stop               = vxlan_stop,
2640         .ndo_start_xmit         = vxlan_xmit,
2641         .ndo_get_stats64        = ip_tunnel_get_stats64,
2642         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2643         .ndo_change_mtu         = vxlan_change_mtu,
2644         .ndo_validate_addr      = eth_validate_addr,
2645         .ndo_set_mac_address    = eth_mac_addr,
2646         .ndo_fdb_add            = vxlan_fdb_add,
2647         .ndo_fdb_del            = vxlan_fdb_delete,
2648         .ndo_fdb_dump           = vxlan_fdb_dump,
2649         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2650 };
2651
2652 static const struct net_device_ops vxlan_netdev_raw_ops = {
2653         .ndo_init               = vxlan_init,
2654         .ndo_uninit             = vxlan_uninit,
2655         .ndo_open               = vxlan_open,
2656         .ndo_stop               = vxlan_stop,
2657         .ndo_start_xmit         = vxlan_xmit,
2658         .ndo_get_stats64        = ip_tunnel_get_stats64,
2659         .ndo_change_mtu         = vxlan_change_mtu,
2660         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2661 };
2662
2663 /* Info for udev, that this is a virtual tunnel endpoint */
2664 static struct device_type vxlan_type = {
2665         .name = "vxlan",
2666 };
2667
2668 /* Calls the ndo_udp_tunnel_add of the caller in order to
2669  * supply the listening VXLAN udp ports. Callers are expected
2670  * to implement the ndo_udp_tunnel_add.
2671  */
2672 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
2673 {
2674         struct vxlan_sock *vs;
2675         struct net *net = dev_net(dev);
2676         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2677         unsigned int i;
2678
2679         spin_lock(&vn->sock_lock);
2680         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2681                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2682                         unsigned short type;
2683
2684                         if (vs->flags & VXLAN_F_GPE)
2685                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
2686                         else
2687                                 type = UDP_TUNNEL_TYPE_VXLAN;
2688
2689                         if (push)
2690                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
2691                         else
2692                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
2693                 }
2694         }
2695         spin_unlock(&vn->sock_lock);
2696 }
2697
2698 /* Initialize the device structure. */
2699 static void vxlan_setup(struct net_device *dev)
2700 {
2701         struct vxlan_dev *vxlan = netdev_priv(dev);
2702         unsigned int h;
2703
2704         eth_hw_addr_random(dev);
2705         ether_setup(dev);
2706
2707         dev->needs_free_netdev = true;
2708         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2709
2710         dev->features   |= NETIF_F_LLTX;
2711         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2712         dev->features   |= NETIF_F_RXCSUM;
2713         dev->features   |= NETIF_F_GSO_SOFTWARE;
2714
2715         dev->vlan_features = dev->features;
2716         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2717         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2718         netif_keep_dst(dev);
2719         dev->priv_flags |= IFF_NO_QUEUE;
2720
2721         /* MTU range: 68 - 65535 */
2722         dev->min_mtu = ETH_MIN_MTU;
2723         dev->max_mtu = ETH_MAX_MTU;
2724
2725         INIT_LIST_HEAD(&vxlan->next);
2726         spin_lock_init(&vxlan->hash_lock);
2727
2728         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
2729
2730         vxlan->dev = dev;
2731
2732         for (h = 0; h < FDB_HASH_SIZE; ++h)
2733                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2734 }
2735
2736 static void vxlan_ether_setup(struct net_device *dev)
2737 {
2738         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2739         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2740         dev->netdev_ops = &vxlan_netdev_ether_ops;
2741 }
2742
2743 static void vxlan_raw_setup(struct net_device *dev)
2744 {
2745         dev->header_ops = NULL;
2746         dev->type = ARPHRD_NONE;
2747         dev->hard_header_len = 0;
2748         dev->addr_len = 0;
2749         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2750         dev->netdev_ops = &vxlan_netdev_raw_ops;
2751 }
2752
2753 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2754         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2755         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2756         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2757         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2758         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2759         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2760         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2761         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2762         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2763         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2764         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2765         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2766         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2767         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2768         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2769         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2770         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2771         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2772         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2773         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2774         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2775         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2776         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2777         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2778         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2779         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2780         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2781         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
2782 };
2783
2784 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
2785                           struct netlink_ext_ack *extack)
2786 {
2787         if (tb[IFLA_ADDRESS]) {
2788                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2789                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2790                                             "Provided link layer address is not Ethernet");
2791                         return -EINVAL;
2792                 }
2793
2794                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2795                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2796                                             "Provided Ethernet address is not unicast");
2797                         return -EADDRNOTAVAIL;
2798                 }
2799         }
2800
2801         if (tb[IFLA_MTU]) {
2802                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
2803
2804                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
2805                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
2806                                             "MTU must be between 68 and 65535");
2807                         return -EINVAL;
2808                 }
2809         }
2810
2811         if (!data) {
2812                 NL_SET_ERR_MSG(extack,
2813                                "Required attributes not provided to perform the operation");
2814                 return -EINVAL;
2815         }
2816
2817         if (data[IFLA_VXLAN_ID]) {
2818                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2819
2820                 if (id >= VXLAN_N_VID) {
2821                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID],
2822                                             "VXLAN ID must be lower than 16777216");
2823                         return -ERANGE;
2824                 }
2825         }
2826
2827         if (data[IFLA_VXLAN_PORT_RANGE]) {
2828                 const struct ifla_vxlan_port_range *p
2829                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2830
2831                 if (ntohs(p->high) < ntohs(p->low)) {
2832                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
2833                                             "Invalid source port range");
2834                         return -EINVAL;
2835                 }
2836         }
2837
2838         return 0;
2839 }
2840
2841 static void vxlan_get_drvinfo(struct net_device *netdev,
2842                               struct ethtool_drvinfo *drvinfo)
2843 {
2844         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2845         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2846 }
2847
2848 static const struct ethtool_ops vxlan_ethtool_ops = {
2849         .get_drvinfo    = vxlan_get_drvinfo,
2850         .get_link       = ethtool_op_get_link,
2851 };
2852
2853 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2854                                         __be16 port, u32 flags)
2855 {
2856         struct socket *sock;
2857         struct udp_port_cfg udp_conf;
2858         int err;
2859
2860         memset(&udp_conf, 0, sizeof(udp_conf));
2861
2862         if (ipv6) {
2863                 udp_conf.family = AF_INET6;
2864                 udp_conf.use_udp6_rx_checksums =
2865                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2866                 udp_conf.ipv6_v6only = 1;
2867         } else {
2868                 udp_conf.family = AF_INET;
2869         }
2870
2871         udp_conf.local_udp_port = port;
2872
2873         /* Open UDP socket */
2874         err = udp_sock_create(net, &udp_conf, &sock);
2875         if (err < 0)
2876                 return ERR_PTR(err);
2877
2878         return sock;
2879 }
2880
2881 /* Create new listen socket if needed */
2882 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2883                                               __be16 port, u32 flags)
2884 {
2885         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2886         struct vxlan_sock *vs;
2887         struct socket *sock;
2888         unsigned int h;
2889         struct udp_tunnel_sock_cfg tunnel_cfg;
2890
2891         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2892         if (!vs)
2893                 return ERR_PTR(-ENOMEM);
2894
2895         for (h = 0; h < VNI_HASH_SIZE; ++h)
2896                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2897
2898         sock = vxlan_create_sock(net, ipv6, port, flags);
2899         if (IS_ERR(sock)) {
2900                 kfree(vs);
2901                 return ERR_CAST(sock);
2902         }
2903
2904         vs->sock = sock;
2905         refcount_set(&vs->refcnt, 1);
2906         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2907
2908         spin_lock(&vn->sock_lock);
2909         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2910         udp_tunnel_notify_add_rx_port(sock,
2911                                       (vs->flags & VXLAN_F_GPE) ?
2912                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2913                                       UDP_TUNNEL_TYPE_VXLAN);
2914         spin_unlock(&vn->sock_lock);
2915
2916         /* Mark socket as an encapsulation socket. */
2917         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2918         tunnel_cfg.sk_user_data = vs;
2919         tunnel_cfg.encap_type = 1;
2920         tunnel_cfg.encap_rcv = vxlan_rcv;
2921         tunnel_cfg.encap_destroy = NULL;
2922         tunnel_cfg.gro_receive = vxlan_gro_receive;
2923         tunnel_cfg.gro_complete = vxlan_gro_complete;
2924
2925         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2926
2927         return vs;
2928 }
2929
2930 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2931 {
2932         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2933         struct vxlan_sock *vs = NULL;
2934         struct vxlan_dev_node *node;
2935
2936         if (!vxlan->cfg.no_share) {
2937                 spin_lock(&vn->sock_lock);
2938                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2939                                      vxlan->cfg.dst_port, vxlan->cfg.flags);
2940                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
2941                         spin_unlock(&vn->sock_lock);
2942                         return -EBUSY;
2943                 }
2944                 spin_unlock(&vn->sock_lock);
2945         }
2946         if (!vs)
2947                 vs = vxlan_socket_create(vxlan->net, ipv6,
2948                                          vxlan->cfg.dst_port, vxlan->cfg.flags);
2949         if (IS_ERR(vs))
2950                 return PTR_ERR(vs);
2951 #if IS_ENABLED(CONFIG_IPV6)
2952         if (ipv6) {
2953                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2954                 node = &vxlan->hlist6;
2955         } else
2956 #endif
2957         {
2958                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2959                 node = &vxlan->hlist4;
2960         }
2961         vxlan_vs_add_dev(vs, vxlan, node);
2962         return 0;
2963 }
2964
2965 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2966 {
2967         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
2968         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
2969         bool ipv4 = !ipv6 || metadata;
2970         int ret = 0;
2971
2972         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2973 #if IS_ENABLED(CONFIG_IPV6)
2974         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2975         if (ipv6) {
2976                 ret = __vxlan_sock_add(vxlan, true);
2977                 if (ret < 0 && ret != -EAFNOSUPPORT)
2978                         ipv4 = false;
2979         }
2980 #endif
2981         if (ipv4)
2982                 ret = __vxlan_sock_add(vxlan, false);
2983         if (ret < 0)
2984                 vxlan_sock_release(vxlan);
2985         return ret;
2986 }
2987
2988 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
2989                                  struct net_device **lower,
2990                                  struct vxlan_dev *old,
2991                                  struct netlink_ext_ack *extack)
2992 {
2993         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2994         struct vxlan_dev *tmp;
2995         bool use_ipv6 = false;
2996
2997         if (conf->flags & VXLAN_F_GPE) {
2998                 /* For now, allow GPE only together with
2999                  * COLLECT_METADATA. This can be relaxed later; in such
3000                  * case, the other side of the PtP link will have to be
3001                  * provided.
3002                  */
3003                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
3004                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3005                         NL_SET_ERR_MSG(extack,
3006                                        "VXLAN GPE does not support this combination of attributes");
3007                         return -EINVAL;
3008                 }
3009         }
3010
3011         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
3012                 /* Unless IPv6 is explicitly requested, assume IPv4 */
3013                 conf->remote_ip.sa.sa_family = AF_INET;
3014                 conf->saddr.sa.sa_family = AF_INET;
3015         } else if (!conf->remote_ip.sa.sa_family) {
3016                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
3017         } else if (!conf->saddr.sa.sa_family) {
3018                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
3019         }
3020
3021         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
3022                 NL_SET_ERR_MSG(extack,
3023                                "Local and remote address must be from the same family");
3024                 return -EINVAL;
3025         }
3026
3027         if (vxlan_addr_multicast(&conf->saddr)) {
3028                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
3029                 return -EINVAL;
3030         }
3031
3032         if (conf->saddr.sa.sa_family == AF_INET6) {
3033                 if (!IS_ENABLED(CONFIG_IPV6)) {
3034                         NL_SET_ERR_MSG(extack,
3035                                        "IPv6 support not enabled in the kernel");
3036                         return -EPFNOSUPPORT;
3037                 }
3038                 use_ipv6 = true;
3039                 conf->flags |= VXLAN_F_IPV6;
3040
3041                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3042                         int local_type =
3043                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3044                         int remote_type =
3045                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3046
3047                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3048                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3049                                     (remote_type != IPV6_ADDR_ANY)) {
3050                                         NL_SET_ERR_MSG(extack,
3051                                                        "Invalid combination of local and remote address scopes");
3052                                         return -EINVAL;
3053                                 }
3054
3055                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3056                         } else {
3057                                 if (remote_type ==
3058                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3059                                         NL_SET_ERR_MSG(extack,
3060                                                        "Invalid combination of local and remote address scopes");
3061                                         return -EINVAL;
3062                                 }
3063
3064                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3065                         }
3066                 }
3067         }
3068
3069         if (conf->label && !use_ipv6) {
3070                 NL_SET_ERR_MSG(extack,
3071                                "Label attribute only applies to IPv6 VXLAN devices");
3072                 return -EINVAL;
3073         }
3074
3075         if (conf->remote_ifindex) {
3076                 struct net_device *lowerdev;
3077
3078                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3079                 if (!lowerdev) {
3080                         NL_SET_ERR_MSG(extack,
3081                                        "Invalid local interface, device not found");
3082                         return -ENODEV;
3083                 }
3084
3085 #if IS_ENABLED(CONFIG_IPV6)
3086                 if (use_ipv6) {
3087                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3088                         if (idev && idev->cnf.disable_ipv6) {
3089                                 NL_SET_ERR_MSG(extack,
3090                                                "IPv6 support disabled by administrator");
3091                                 return -EPERM;
3092                         }
3093                 }
3094 #endif
3095
3096                 *lower = lowerdev;
3097         } else {
3098                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3099                         NL_SET_ERR_MSG(extack,
3100                                        "Local interface required for multicast remote destination");
3101
3102                         return -EINVAL;
3103                 }
3104
3105 #if IS_ENABLED(CONFIG_IPV6)
3106                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3107                         NL_SET_ERR_MSG(extack,
3108                                        "Local interface required for link-local local/remote addresses");
3109                         return -EINVAL;
3110                 }
3111 #endif
3112
3113                 *lower = NULL;
3114         }
3115
3116         if (!conf->dst_port) {
3117                 if (conf->flags & VXLAN_F_GPE)
3118                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3119                 else
3120                         conf->dst_port = htons(vxlan_port);
3121         }
3122
3123         if (!conf->age_interval)
3124                 conf->age_interval = FDB_AGE_DEFAULT;
3125
3126         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3127                 if (tmp == old)
3128                         continue;
3129
3130                 if (tmp->cfg.vni != conf->vni)
3131                         continue;
3132                 if (tmp->cfg.dst_port != conf->dst_port)
3133                         continue;
3134                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3135                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3136                         continue;
3137
3138                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3139                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3140                         continue;
3141
3142                 NL_SET_ERR_MSG(extack,
3143                                "A VXLAN device with the specified VNI already exists");
3144                 return -EEXIST;
3145         }
3146
3147         return 0;
3148 }
3149
3150 static void vxlan_config_apply(struct net_device *dev,
3151                                struct vxlan_config *conf,
3152                                struct net_device *lowerdev,
3153                                struct net *src_net,
3154                                bool changelink)
3155 {
3156         struct vxlan_dev *vxlan = netdev_priv(dev);
3157         struct vxlan_rdst *dst = &vxlan->default_dst;
3158         unsigned short needed_headroom = ETH_HLEN;
3159         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3160         int max_mtu = ETH_MAX_MTU;
3161
3162         if (!changelink) {
3163                 if (conf->flags & VXLAN_F_GPE)
3164                         vxlan_raw_setup(dev);
3165                 else
3166                         vxlan_ether_setup(dev);
3167
3168                 if (conf->mtu)
3169                         dev->mtu = conf->mtu;
3170
3171                 vxlan->net = src_net;
3172         }
3173
3174         dst->remote_vni = conf->vni;
3175
3176         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3177
3178         if (lowerdev) {
3179                 dst->remote_ifindex = conf->remote_ifindex;
3180
3181                 dev->gso_max_size = lowerdev->gso_max_size;
3182                 dev->gso_max_segs = lowerdev->gso_max_segs;
3183
3184                 needed_headroom = lowerdev->hard_header_len;
3185                 needed_headroom += lowerdev->needed_headroom;
3186
3187                 dev->needed_tailroom = lowerdev->needed_tailroom;
3188
3189                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3190                                            VXLAN_HEADROOM);
3191                 if (max_mtu < ETH_MIN_MTU)
3192                         max_mtu = ETH_MIN_MTU;
3193
3194                 if (!changelink && !conf->mtu)
3195                         dev->mtu = max_mtu;
3196         }
3197
3198         if (dev->mtu > max_mtu)
3199                 dev->mtu = max_mtu;
3200
3201         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3202                 needed_headroom += VXLAN6_HEADROOM;
3203         else
3204                 needed_headroom += VXLAN_HEADROOM;
3205         dev->needed_headroom = needed_headroom;
3206
3207         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3208 }
3209
3210 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3211                                struct vxlan_config *conf, bool changelink,
3212                                struct netlink_ext_ack *extack)
3213 {
3214         struct vxlan_dev *vxlan = netdev_priv(dev);
3215         struct net_device *lowerdev;
3216         int ret;
3217
3218         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3219         if (ret)
3220                 return ret;
3221
3222         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3223
3224         return 0;
3225 }
3226
3227 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3228                               struct vxlan_config *conf,
3229                               struct netlink_ext_ack *extack)
3230 {
3231         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3232         struct vxlan_dev *vxlan = netdev_priv(dev);
3233         struct vxlan_fdb *f = NULL;
3234         bool unregister = false;
3235         int err;
3236
3237         err = vxlan_dev_configure(net, dev, conf, false, extack);
3238         if (err)
3239                 return err;
3240
3241         dev->ethtool_ops = &vxlan_ethtool_ops;
3242
3243         /* create an fdb entry for a valid default destination */
3244         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
3245                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3246                                        &vxlan->default_dst.remote_ip,
3247                                        NUD_REACHABLE | NUD_PERMANENT,
3248                                        vxlan->cfg.dst_port,
3249                                        vxlan->default_dst.remote_vni,
3250                                        vxlan->default_dst.remote_vni,
3251                                        vxlan->default_dst.remote_ifindex,
3252                                        NTF_SELF, &f);
3253                 if (err)
3254                         return err;
3255         }
3256
3257         err = register_netdevice(dev);
3258         if (err)
3259                 goto errout;
3260         unregister = true;
3261
3262         err = rtnl_configure_link(dev, NULL);
3263         if (err)
3264                 goto errout;
3265
3266         /* notify default fdb entry */
3267         if (f)
3268                 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH);
3269
3270         list_add(&vxlan->next, &vn->vxlan_list);
3271         return 0;
3272
3273 errout:
3274         /* unregister_netdevice() destroys the default FDB entry with deletion
3275          * notification. But the addition notification was not sent yet, so
3276          * destroy the entry by hand here.
3277          */
3278         if (f)
3279                 vxlan_fdb_destroy(vxlan, f, false);
3280         if (unregister)
3281                 unregister_netdevice(dev);
3282         return err;
3283 }
3284
3285 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3286                          struct net_device *dev, struct vxlan_config *conf,
3287                          bool changelink)
3288 {
3289         struct vxlan_dev *vxlan = netdev_priv(dev);
3290
3291         memset(conf, 0, sizeof(*conf));
3292
3293         /* if changelink operation, start with old existing cfg */
3294         if (changelink)
3295                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3296
3297         if (data[IFLA_VXLAN_ID]) {
3298                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3299
3300                 if (changelink && (vni != conf->vni))
3301                         return -EOPNOTSUPP;
3302                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3303         }
3304
3305         if (data[IFLA_VXLAN_GROUP]) {
3306                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET))
3307                         return -EOPNOTSUPP;
3308
3309                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3310                 conf->remote_ip.sa.sa_family = AF_INET;
3311         } else if (data[IFLA_VXLAN_GROUP6]) {
3312                 if (!IS_ENABLED(CONFIG_IPV6))
3313                         return -EPFNOSUPPORT;
3314
3315                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6))
3316                         return -EOPNOTSUPP;
3317
3318                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3319                 conf->remote_ip.sa.sa_family = AF_INET6;
3320         }
3321
3322         if (data[IFLA_VXLAN_LOCAL]) {
3323                 if (changelink && (conf->saddr.sa.sa_family != AF_INET))
3324                         return -EOPNOTSUPP;
3325
3326                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3327                 conf->saddr.sa.sa_family = AF_INET;
3328         } else if (data[IFLA_VXLAN_LOCAL6]) {
3329                 if (!IS_ENABLED(CONFIG_IPV6))
3330                         return -EPFNOSUPPORT;
3331
3332                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6))
3333                         return -EOPNOTSUPP;
3334
3335                 /* TODO: respect scope id */
3336                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3337                 conf->saddr.sa.sa_family = AF_INET6;
3338         }
3339
3340         if (data[IFLA_VXLAN_LINK])
3341                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3342
3343         if (data[IFLA_VXLAN_TOS])
3344                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3345
3346         if (data[IFLA_VXLAN_TTL])
3347                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3348
3349         if (data[IFLA_VXLAN_TTL_INHERIT]) {
3350                 if (changelink)
3351                         return -EOPNOTSUPP;
3352                 conf->flags |= VXLAN_F_TTL_INHERIT;
3353         }
3354
3355         if (data[IFLA_VXLAN_LABEL])
3356                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3357                              IPV6_FLOWLABEL_MASK;
3358
3359         if (data[IFLA_VXLAN_LEARNING]) {
3360                 if (nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3361                         conf->flags |= VXLAN_F_LEARN;
3362                 else
3363                         conf->flags &= ~VXLAN_F_LEARN;
3364         } else if (!changelink) {
3365                 /* default to learn on a new device */
3366                 conf->flags |= VXLAN_F_LEARN;
3367         }
3368
3369         if (data[IFLA_VXLAN_AGEING]) {
3370                 if (changelink)
3371                         return -EOPNOTSUPP;
3372                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3373         }
3374
3375         if (data[IFLA_VXLAN_PROXY]) {
3376                 if (changelink)
3377                         return -EOPNOTSUPP;
3378                 if (nla_get_u8(data[IFLA_VXLAN_PROXY]))
3379                         conf->flags |= VXLAN_F_PROXY;
3380         }
3381
3382         if (data[IFLA_VXLAN_RSC]) {
3383                 if (changelink)
3384                         return -EOPNOTSUPP;
3385                 if (nla_get_u8(data[IFLA_VXLAN_RSC]))
3386                         conf->flags |= VXLAN_F_RSC;
3387         }
3388
3389         if (data[IFLA_VXLAN_L2MISS]) {
3390                 if (changelink)
3391                         return -EOPNOTSUPP;
3392                 if (nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3393                         conf->flags |= VXLAN_F_L2MISS;
3394         }
3395
3396         if (data[IFLA_VXLAN_L3MISS]) {
3397                 if (changelink)
3398                         return -EOPNOTSUPP;
3399                 if (nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3400                         conf->flags |= VXLAN_F_L3MISS;
3401         }
3402
3403         if (data[IFLA_VXLAN_LIMIT]) {
3404                 if (changelink)
3405                         return -EOPNOTSUPP;
3406                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3407         }
3408
3409         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3410                 if (changelink)
3411                         return -EOPNOTSUPP;
3412                 if (nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3413                         conf->flags |= VXLAN_F_COLLECT_METADATA;
3414         }
3415
3416         if (data[IFLA_VXLAN_PORT_RANGE]) {
3417                 if (!changelink) {
3418                         const struct ifla_vxlan_port_range *p
3419                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3420                         conf->port_min = ntohs(p->low);
3421                         conf->port_max = ntohs(p->high);
3422                 } else {
3423                         return -EOPNOTSUPP;
3424                 }
3425         }
3426
3427         if (data[IFLA_VXLAN_PORT]) {
3428                 if (changelink)
3429                         return -EOPNOTSUPP;
3430                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3431         }
3432
3433         if (data[IFLA_VXLAN_UDP_CSUM]) {
3434                 if (changelink)
3435                         return -EOPNOTSUPP;
3436                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3437                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3438         }
3439
3440         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3441                 if (changelink)
3442                         return -EOPNOTSUPP;
3443                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3444                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3445         }
3446
3447         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3448                 if (changelink)
3449                         return -EOPNOTSUPP;
3450                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3451                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3452         }
3453
3454         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3455                 if (changelink)
3456                         return -EOPNOTSUPP;
3457                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3458                         conf->flags |= VXLAN_F_REMCSUM_TX;
3459         }
3460
3461         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3462                 if (changelink)
3463                         return -EOPNOTSUPP;
3464                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3465                         conf->flags |= VXLAN_F_REMCSUM_RX;
3466         }
3467
3468         if (data[IFLA_VXLAN_GBP]) {
3469                 if (changelink)
3470                         return -EOPNOTSUPP;
3471                 conf->flags |= VXLAN_F_GBP;
3472         }
3473
3474         if (data[IFLA_VXLAN_GPE]) {
3475                 if (changelink)
3476                         return -EOPNOTSUPP;
3477                 conf->flags |= VXLAN_F_GPE;
3478         }
3479
3480         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3481                 if (changelink)
3482                         return -EOPNOTSUPP;
3483                 conf->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3484         }
3485
3486         if (tb[IFLA_MTU]) {
3487                 if (changelink)
3488                         return -EOPNOTSUPP;
3489                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3490         }
3491
3492         return 0;
3493 }
3494
3495 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3496                          struct nlattr *tb[], struct nlattr *data[],
3497                          struct netlink_ext_ack *extack)
3498 {
3499         struct vxlan_config conf;
3500         int err;
3501
3502         err = vxlan_nl2conf(tb, data, dev, &conf, false);
3503         if (err)
3504                 return err;
3505
3506         return __vxlan_dev_create(src_net, dev, &conf, extack);
3507 }
3508
3509 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3510                             struct nlattr *data[],
3511                             struct netlink_ext_ack *extack)
3512 {
3513         struct vxlan_dev *vxlan = netdev_priv(dev);
3514         struct vxlan_rdst *dst = &vxlan->default_dst;
3515         struct vxlan_rdst old_dst;
3516         struct vxlan_config conf;
3517         int err;
3518
3519         err = vxlan_nl2conf(tb, data,
3520                             dev, &conf, true);
3521         if (err)
3522                 return err;
3523
3524         memcpy(&old_dst, dst, sizeof(struct vxlan_rdst));
3525
3526         err = vxlan_dev_configure(vxlan->net, dev, &conf, true, extack);
3527         if (err)
3528                 return err;
3529
3530         /* handle default dst entry */
3531         if (!vxlan_addr_equal(&dst->remote_ip, &old_dst.remote_ip)) {
3532                 spin_lock_bh(&vxlan->hash_lock);
3533                 if (!vxlan_addr_any(&old_dst.remote_ip))
3534                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
3535                                            old_dst.remote_ip,
3536                                            vxlan->cfg.dst_port,
3537                                            old_dst.remote_vni,
3538                                            old_dst.remote_vni,
3539                                            old_dst.remote_ifindex, 0);
3540
3541                 if (!vxlan_addr_any(&dst->remote_ip)) {
3542                         err = vxlan_fdb_update(vxlan, all_zeros_mac,
3543                                                &dst->remote_ip,
3544                                                NUD_REACHABLE | NUD_PERMANENT,
3545                                                NLM_F_APPEND | NLM_F_CREATE,
3546                                                vxlan->cfg.dst_port,
3547                                                dst->remote_vni,
3548                                                dst->remote_vni,
3549                                                dst->remote_ifindex,
3550                                                NTF_SELF);
3551                         if (err) {
3552                                 spin_unlock_bh(&vxlan->hash_lock);
3553                                 return err;
3554                         }
3555                 }
3556                 spin_unlock_bh(&vxlan->hash_lock);
3557         }
3558
3559         return 0;
3560 }
3561
3562 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3563 {
3564         struct vxlan_dev *vxlan = netdev_priv(dev);
3565
3566         vxlan_flush(vxlan, true);
3567
3568         list_del(&vxlan->next);
3569         unregister_netdevice_queue(dev, head);
3570 }
3571
3572 static size_t vxlan_get_size(const struct net_device *dev)
3573 {
3574
3575         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3576                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3577                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3578                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3579                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3580                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL_INHERIT */
3581                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3582                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3583                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3584                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3585                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3586                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3587                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3588                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3589                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3590                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3591                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3592                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3593                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3594                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3595                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3596                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3597                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3598                 0;
3599 }
3600
3601 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3602 {
3603         const struct vxlan_dev *vxlan = netdev_priv(dev);
3604         const struct vxlan_rdst *dst = &vxlan->default_dst;
3605         struct ifla_vxlan_port_range ports = {
3606                 .low =  htons(vxlan->cfg.port_min),
3607                 .high = htons(vxlan->cfg.port_max),
3608         };
3609
3610         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3611                 goto nla_put_failure;
3612
3613         if (!vxlan_addr_any(&dst->remote_ip)) {
3614                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3615                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3616                                             dst->remote_ip.sin.sin_addr.s_addr))
3617                                 goto nla_put_failure;
3618 #if IS_ENABLED(CONFIG_IPV6)
3619                 } else {
3620                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3621                                              &dst->remote_ip.sin6.sin6_addr))
3622                                 goto nla_put_failure;
3623 #endif
3624                 }
3625         }
3626
3627         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3628                 goto nla_put_failure;
3629
3630         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3631                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3632                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3633                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3634                                 goto nla_put_failure;
3635 #if IS_ENABLED(CONFIG_IPV6)
3636                 } else {
3637                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3638                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3639                                 goto nla_put_failure;
3640 #endif
3641                 }
3642         }
3643
3644         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3645             nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT,
3646                        !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) ||
3647             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3648             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3649             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3650                         !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
3651             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3652                         !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
3653             nla_put_u8(skb, IFLA_VXLAN_RSC,
3654                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
3655             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3656                         !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
3657             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3658                         !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
3659             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3660                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
3661             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3662             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3663             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3664             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3665                         !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3666             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3667                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3668             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3669                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3670             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3671                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
3672             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3673                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
3674                 goto nla_put_failure;
3675
3676         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3677                 goto nla_put_failure;
3678
3679         if (vxlan->cfg.flags & VXLAN_F_GBP &&
3680             nla_put_flag(skb, IFLA_VXLAN_GBP))
3681                 goto nla_put_failure;
3682
3683         if (vxlan->cfg.flags & VXLAN_F_GPE &&
3684             nla_put_flag(skb, IFLA_VXLAN_GPE))
3685                 goto nla_put_failure;
3686
3687         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3688             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3689                 goto nla_put_failure;
3690
3691         return 0;
3692
3693 nla_put_failure:
3694         return -EMSGSIZE;
3695 }
3696
3697 static struct net *vxlan_get_link_net(const struct net_device *dev)
3698 {
3699         struct vxlan_dev *vxlan = netdev_priv(dev);
3700
3701         return vxlan->net;
3702 }
3703
3704 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3705         .kind           = "vxlan",
3706         .maxtype        = IFLA_VXLAN_MAX,
3707         .policy         = vxlan_policy,
3708         .priv_size      = sizeof(struct vxlan_dev),
3709         .setup          = vxlan_setup,
3710         .validate       = vxlan_validate,
3711         .newlink        = vxlan_newlink,
3712         .changelink     = vxlan_changelink,
3713         .dellink        = vxlan_dellink,
3714         .get_size       = vxlan_get_size,
3715         .fill_info      = vxlan_fill_info,
3716         .get_link_net   = vxlan_get_link_net,
3717 };
3718
3719 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3720                                     u8 name_assign_type,
3721                                     struct vxlan_config *conf)
3722 {
3723         struct nlattr *tb[IFLA_MAX + 1];
3724         struct net_device *dev;
3725         int err;
3726
3727         memset(&tb, 0, sizeof(tb));
3728
3729         dev = rtnl_create_link(net, name, name_assign_type,
3730                                &vxlan_link_ops, tb);
3731         if (IS_ERR(dev))
3732                 return dev;
3733
3734         err = __vxlan_dev_create(net, dev, conf, NULL);
3735         if (err < 0) {
3736                 free_netdev(dev);
3737                 return ERR_PTR(err);
3738         }
3739
3740         err = rtnl_configure_link(dev, NULL);
3741         if (err < 0) {
3742                 LIST_HEAD(list_kill);
3743
3744                 vxlan_dellink(dev, &list_kill);
3745                 unregister_netdevice_many(&list_kill);
3746                 return ERR_PTR(err);
3747         }
3748
3749         return dev;
3750 }
3751 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3752
3753 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3754                                              struct net_device *dev)
3755 {
3756         struct vxlan_dev *vxlan, *next;
3757         LIST_HEAD(list_kill);
3758
3759         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3760                 struct vxlan_rdst *dst = &vxlan->default_dst;
3761
3762                 /* In case we created vxlan device with carrier
3763                  * and we loose the carrier due to module unload
3764                  * we also need to remove vxlan device. In other
3765                  * cases, it's not necessary and remote_ifindex
3766                  * is 0 here, so no matches.
3767                  */
3768                 if (dst->remote_ifindex == dev->ifindex)
3769                         vxlan_dellink(vxlan->dev, &list_kill);
3770         }
3771
3772         unregister_netdevice_many(&list_kill);
3773 }
3774
3775 static int vxlan_netdevice_event(struct notifier_block *unused,
3776                                  unsigned long event, void *ptr)
3777 {
3778         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3779         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3780
3781         if (event == NETDEV_UNREGISTER) {
3782                 vxlan_offload_rx_ports(dev, false);
3783                 vxlan_handle_lowerdev_unregister(vn, dev);
3784         } else if (event == NETDEV_REGISTER) {
3785                 vxlan_offload_rx_ports(dev, true);
3786         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
3787                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
3788                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
3789         }
3790
3791         return NOTIFY_DONE;
3792 }
3793
3794 static struct notifier_block vxlan_notifier_block __read_mostly = {
3795         .notifier_call = vxlan_netdevice_event,
3796 };
3797
3798 static __net_init int vxlan_init_net(struct net *net)
3799 {
3800         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3801         unsigned int h;
3802
3803         INIT_LIST_HEAD(&vn->vxlan_list);
3804         spin_lock_init(&vn->sock_lock);
3805
3806         for (h = 0; h < PORT_HASH_SIZE; ++h)
3807                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3808
3809         return 0;
3810 }
3811
3812 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
3813 {
3814         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3815         struct vxlan_dev *vxlan, *next;
3816         struct net_device *dev, *aux;
3817
3818         for_each_netdev_safe(net, dev, aux)
3819                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3820                         unregister_netdevice_queue(dev, head);
3821
3822         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3823                 /* If vxlan->dev is in the same netns, it has already been added
3824                  * to the list by the previous loop.
3825                  */
3826                 if (!net_eq(dev_net(vxlan->dev), net))
3827                         unregister_netdevice_queue(vxlan->dev, head);
3828         }
3829
3830 }
3831
3832 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
3833 {
3834         struct net *net;
3835         LIST_HEAD(list);
3836         unsigned int h;
3837
3838         rtnl_lock();
3839         list_for_each_entry(net, net_list, exit_list)
3840                 vxlan_destroy_tunnels(net, &list);
3841
3842         unregister_netdevice_many(&list);
3843         rtnl_unlock();
3844
3845         list_for_each_entry(net, net_list, exit_list) {
3846                 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3847
3848                 for (h = 0; h < PORT_HASH_SIZE; ++h)
3849                         WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
3850         }
3851 }
3852
3853 static struct pernet_operations vxlan_net_ops = {
3854         .init = vxlan_init_net,
3855         .exit_batch = vxlan_exit_batch_net,
3856         .id   = &vxlan_net_id,
3857         .size = sizeof(struct vxlan_net),
3858 };
3859
3860 static int __init vxlan_init_module(void)
3861 {
3862         int rc;
3863
3864         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3865
3866         rc = register_pernet_subsys(&vxlan_net_ops);
3867         if (rc)
3868                 goto out1;
3869
3870         rc = register_netdevice_notifier(&vxlan_notifier_block);
3871         if (rc)
3872                 goto out2;
3873
3874         rc = rtnl_link_register(&vxlan_link_ops);
3875         if (rc)
3876                 goto out3;
3877
3878         return 0;
3879 out3:
3880         unregister_netdevice_notifier(&vxlan_notifier_block);
3881 out2:
3882         unregister_pernet_subsys(&vxlan_net_ops);
3883 out1:
3884         return rc;
3885 }
3886 late_initcall(vxlan_init_module);
3887
3888 static void __exit vxlan_cleanup_module(void)
3889 {
3890         rtnl_link_unregister(&vxlan_link_ops);
3891         unregister_netdevice_notifier(&vxlan_notifier_block);
3892         unregister_pernet_subsys(&vxlan_net_ops);
3893         /* rcu_barrier() is called by netns */
3894 }
3895 module_exit(vxlan_cleanup_module);
3896
3897 MODULE_LICENSE("GPL");
3898 MODULE_VERSION(VXLAN_VERSION);
3899 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3900 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3901 MODULE_ALIAS_RTNL_LINK("vxlan");