drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/errno.h>
  16 #include <linux/slab.h>
  17 #include <linux/udp.h>
  18 #include <linux/igmp.h>
  19 #include <linux/if_ether.h>
  20 #include <linux/ethtool.h>
  21 #include <net/arp.h>
  22 #include <net/ndisc.h>
  23 #include <net/ip.h>
  24 #include <net/icmp.h>
  25 #include <net/rtnetlink.h>
  26 #include <net/inet_ecn.h>
  27 #include <net/net_namespace.h>
  28 #include <net/netns/generic.h>
  29 #include <net/tun_proto.h>
  30 #include <net/vxlan.h>
  31
  32 #if IS_ENABLED(CONFIG_IPV6)
  33 #include <net/ip6_tunnel.h>
  34 #include <net/ip6_checksum.h>
  35 #endif
  36
  37 #define VXLAN_VERSION   "0.1"
  38
  39 #define PORT_HASH_BITS  8
  40 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  41 #define FDB_AGE_DEFAULT 300 /* 5 min */
  42 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  43
  44 /* UDP port for VXLAN traffic.
  45  * The IANA assigned port is 4789, but the Linux default is 8472
  46  * for compatibility with early adopters.
  47  */
  48 static unsigned short vxlan_port __read_mostly = 8472;
  49 module_param_named(udp_port, vxlan_port, ushort, 0444);
  50 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  51
  52 static bool log_ecn_error = true;
  53 module_param(log_ecn_error, bool, 0644);
  54 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  55
  56 static unsigned int vxlan_net_id;
  57 static struct rtnl_link_ops vxlan_link_ops;
  58
  59 static const u8 all_zeros_mac[ETH_ALEN + 2];
  60
  61 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  62
  63 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
  64
  65 /* per-network namespace private data for this module */
  66 struct vxlan_net {
  67         struct list_head  vxlan_list;
  68         struct hlist_head sock_list[PORT_HASH_SIZE];
  69         spinlock_t        sock_lock;
  70 };
  71
  72 /* Forwarding table entry */
  73 struct vxlan_fdb {
  74         struct hlist_node hlist;        /* linked list of entries */
  75         struct rcu_head   rcu;
  76         unsigned long     updated;      /* jiffies */
  77         unsigned long     used;
  78         struct list_head  remotes;
  79         u8                eth_addr[ETH_ALEN];
  80         u16               state;        /* see ndm_state */
  81         __be32            vni;
  82         u8                flags;        /* see ndm_flags */
  83 };
  84
  85 /* salt for hash table */
  86 static u32 vxlan_salt __read_mostly;
  87
  88 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  89 {
  90         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  91                ip_tunnel_collect_metadata();
  92 }
  93
  94 #if IS_ENABLED(CONFIG_IPV6)
  95 static inline
  96 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
  97 {
  98         if (a->sa.sa_family != b->sa.sa_family)
  99                 return false;
 100         if (a->sa.sa_family == AF_INET6)
 101                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 102         else
 103                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 104 }
 105
 106 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 107 {
 108         if (ipa->sa.sa_family == AF_INET6)
 109                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 110         else
 111                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 112 }
 113
 114 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 115 {
 116         if (ipa->sa.sa_family == AF_INET6)
 117                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 118         else
 119                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 120 }
 121
 122 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 123 {
 124         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 125                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 126                 ip->sa.sa_family = AF_INET6;
 127                 return 0;
 128         } else if (nla_len(nla) >= sizeof(__be32)) {
 129                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 130                 ip->sa.sa_family = AF_INET;
 131                 return 0;
 132         } else {
 133                 return -EAFNOSUPPORT;
 134         }
 135 }
 136
 137 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 138                               const union vxlan_addr *ip)
 139 {
 140         if (ip->sa.sa_family == AF_INET6)
 141                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 142         else
 143                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 144 }
 145
 146 #else /* !CONFIG_IPV6 */
 147
 148 static inline
 149 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 150 {
 151         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 152 }
 153
 154 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 155 {
 156         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 157 }
 158
 159 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 160 {
 161         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 162 }
 163
 164 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 165 {
 166         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 167                 return -EAFNOSUPPORT;
 168         } else if (nla_len(nla) >= sizeof(__be32)) {
 169                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 170                 ip->sa.sa_family = AF_INET;
 171                 return 0;
 172         } else {
 173                 return -EAFNOSUPPORT;
 174         }
 175 }
 176
 177 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 178                               const union vxlan_addr *ip)
 179 {
 180         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 181 }
 182 #endif
 183
 184 /* Virtual Network hash table head */
 185 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 186 {
 187         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 188 }
 189
 190 /* Socket hash table head */
 191 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 192 {
 193         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 194
 195         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 196 }
 197
 198 /* First remote destination for a forwarding entry.
 199  * Guaranteed to be non-NULL because remotes are never deleted.
 200  */
 201 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 202 {
 203         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 204 }
 205
 206 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 207 {
 208         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 209 }
 210
 211 /* Find VXLAN socket based on network namespace, address family and UDP port
 212  * and enabled unshareable flags.
 213  */
 214 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 215                                           __be16 port, u32 flags)
 216 {
 217         struct vxlan_sock *vs;
 218
 219         flags &= VXLAN_F_RCV_FLAGS;
 220
 221         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 222                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 223                     vxlan_get_sk_family(vs) == family &&
 224                     vs->flags == flags)
 225                         return vs;
 226         }
 227         return NULL;
 228 }
 229
 230 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, int ifindex,
 231                                            __be32 vni)
 232 {
 233         struct vxlan_dev_node *node;
 234
 235         /* For flow based devices, map all packets to VNI 0 */
 236         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 237                 vni = 0;
 238
 239         hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
 240                 if (node->vxlan->default_dst.remote_vni != vni)
 241                         continue;
 242
 243                 if (IS_ENABLED(CONFIG_IPV6)) {
 244                         const struct vxlan_config *cfg = &node->vxlan->cfg;
 245
 246                         if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) &&
 247                             cfg->remote_ifindex != ifindex)
 248                                 continue;
 249                 }
 250
 251                 return node->vxlan;
 252         }
 253
 254         return NULL;
 255 }
 256
 257 /* Look up VNI in a per net namespace table */
 258 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex,
 259                                         __be32 vni, sa_family_t family,
 260                                         __be16 port, u32 flags)
 261 {
 262         struct vxlan_sock *vs;
 263
 264         vs = vxlan_find_sock(net, family, port, flags);
 265         if (!vs)
 266                 return NULL;
 267
 268         return vxlan_vs_find_vni(vs, ifindex, vni);
 269 }
 270
 271 /* Fill in neighbour message in skbuff. */
 272 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 273                           const struct vxlan_fdb *fdb,
 274                           u32 portid, u32 seq, int type, unsigned int flags,
 275                           const struct vxlan_rdst *rdst)
 276 {
 277         unsigned long now = jiffies;
 278         struct nda_cacheinfo ci;
 279         struct nlmsghdr *nlh;
 280         struct ndmsg *ndm;
 281         bool send_ip, send_eth;
 282
 283         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 284         if (nlh == NULL)
 285                 return -EMSGSIZE;
 286
 287         ndm = nlmsg_data(nlh);
 288         memset(ndm, 0, sizeof(*ndm));
 289
 290         send_eth = send_ip = true;
 291
 292         if (type == RTM_GETNEIGH) {
 293                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 294                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 295                 ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
 296         } else
 297                 ndm->ndm_family = AF_BRIDGE;
 298         ndm->ndm_state = fdb->state;
 299         ndm->ndm_ifindex = vxlan->dev->ifindex;
 300         ndm->ndm_flags = fdb->flags;
 301         ndm->ndm_type = RTN_UNICAST;
 302
 303         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 304             nla_put_s32(skb, NDA_LINK_NETNSID,
 305                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 306                 goto nla_put_failure;
 307
 308         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 309                 goto nla_put_failure;
 310
 311         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 312                 goto nla_put_failure;
 313
 314         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 315             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 316                 goto nla_put_failure;
 317         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 318             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 319                 goto nla_put_failure;
 320         if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
 321             nla_put_u32(skb, NDA_SRC_VNI,
 322                         be32_to_cpu(fdb->vni)))
 323                 goto nla_put_failure;
 324         if (rdst->remote_ifindex &&
 325             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 326                 goto nla_put_failure;
 327
 328         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 329         ci.ndm_confirmed = 0;
 330         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 331         ci.ndm_refcnt    = 0;
 332
 333         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 334                 goto nla_put_failure;
 335
 336         nlmsg_end(skb, nlh);
 337         return 0;
 338
 339 nla_put_failure:
 340         nlmsg_cancel(skb, nlh);
 341         return -EMSGSIZE;
 342 }
 343
 344 static inline size_t vxlan_nlmsg_size(void)
 345 {
 346         return NLMSG_ALIGN(sizeof(struct ndmsg))
 347                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 348                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 349                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 350                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 351                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 352                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 353                 + nla_total_size(sizeof(struct nda_cacheinfo));
 354 }
 355
 356 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 357                              struct vxlan_rdst *rd, int type)
 358 {
 359         struct net *net = dev_net(vxlan->dev);
 360         struct sk_buff *skb;
 361         int err = -ENOBUFS;
 362
 363         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 364         if (skb == NULL)
 365                 goto errout;
 366
 367         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 368         if (err < 0) {
 369                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 370                 WARN_ON(err == -EMSGSIZE);
 371                 kfree_skb(skb);
 372                 goto errout;
 373         }
 374
 375         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 376         return;
 377 errout:
 378         if (err < 0)
 379                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 380 }
 381
 382 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 383 {
 384         struct vxlan_dev *vxlan = netdev_priv(dev);
 385         struct vxlan_fdb f = {
 386                 .state = NUD_STALE,
 387         };
 388         struct vxlan_rdst remote = {
 389                 .remote_ip = *ipa, /* goes to NDA_DST */
 390                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 391         };
 392
 393         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 394 }
 395
 396 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 397 {
 398         struct vxlan_fdb f = {
 399                 .state = NUD_STALE,
 400         };
 401         struct vxlan_rdst remote = { };
 402
 403         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 404
 405         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 406 }
 407
 408 /* Hash Ethernet address */
 409 static u32 eth_hash(const unsigned char *addr)
 410 {
 411         u64 value = get_unaligned((u64 *)addr);
 412
 413         /* only want 6 bytes */
 414 #ifdef __BIG_ENDIAN
 415         value >>= 16;
 416 #else
 417         value <<= 16;
 418 #endif
 419         return hash_64(value, FDB_HASH_BITS);
 420 }
 421
 422 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
 423 {
 424         /* use 1 byte of OUI and 3 bytes of NIC */
 425         u32 key = get_unaligned((u32 *)(addr + 2));
 426
 427         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
 428 }
 429
 430 /* Hash chain to use given mac address */
 431 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 432                                                 const u8 *mac, __be32 vni)
 433 {
 434         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
 435                 return &vxlan->fdb_head[eth_vni_hash(mac, vni)];
 436         else
 437                 return &vxlan->fdb_head[eth_hash(mac)];
 438 }
 439
 440 /* Look up Ethernet address in forwarding table */
 441 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 442                                           const u8 *mac, __be32 vni)
 443 {
 444         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 445         struct vxlan_fdb *f;
 446
 447         hlist_for_each_entry_rcu(f, head, hlist) {
 448                 if (ether_addr_equal(mac, f->eth_addr)) {
 449                         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
 450                                 if (vni == f->vni)
 451                                         return f;
 452                         } else {
 453                                 return f;
 454                         }
 455                 }
 456         }
 457
 458         return NULL;
 459 }
 460
 461 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 462                                         const u8 *mac, __be32 vni)
 463 {
 464         struct vxlan_fdb *f;
 465
 466         f = __vxlan_find_mac(vxlan, mac, vni);
 467         if (f)
 468                 f->used = jiffies;
 469
 470         return f;
 471 }
 472
 473 /* caller should hold vxlan->hash_lock */
 474 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 475                                               union vxlan_addr *ip, __be16 port,
 476                                               __be32 vni, __u32 ifindex)
 477 {
 478         struct vxlan_rdst *rd;
 479
 480         list_for_each_entry(rd, &f->remotes, list) {
 481                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 482                     rd->remote_port == port &&
 483                     rd->remote_vni == vni &&
 484                     rd->remote_ifindex == ifindex)
 485                         return rd;
 486         }
 487
 488         return NULL;
 489 }
 490
 491 /* Replace destination of unicast mac */
 492 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 493                              union vxlan_addr *ip, __be16 port, __be32 vni,
 494                              __u32 ifindex)
 495 {
 496         struct vxlan_rdst *rd;
 497
 498         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 499         if (rd)
 500                 return 0;
 501
 502         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 503         if (!rd)
 504                 return 0;
 505
 506         dst_cache_reset(&rd->dst_cache);
 507         rd->remote_ip = *ip;
 508         rd->remote_port = port;
 509         rd->remote_vni = vni;
 510         rd->remote_ifindex = ifindex;
 511         return 1;
 512 }
 513
 514 /* Add/update destinations for multicast */
 515 static int vxlan_fdb_append(struct vxlan_fdb *f,
 516                             union vxlan_addr *ip, __be16 port, __be32 vni,
 517                             __u32 ifindex, struct vxlan_rdst **rdp)
 518 {
 519         struct vxlan_rdst *rd;
 520
 521         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 522         if (rd)
 523                 return 0;
 524
 525         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 526         if (rd == NULL)
 527                 return -ENOBUFS;
 528
 529         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 530                 kfree(rd);
 531                 return -ENOBUFS;
 532         }
 533
 534         rd->remote_ip = *ip;
 535         rd->remote_port = port;
 536         rd->remote_vni = vni;
 537         rd->remote_ifindex = ifindex;
 538
 539         list_add_tail_rcu(&rd->list, &f->remotes);
 540
 541         *rdp = rd;
 542         return 1;
 543 }
 544
 545 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 546                                           unsigned int off,
 547                                           struct vxlanhdr *vh, size_t hdrlen,
 548                                           __be32 vni_field,
 549                                           struct gro_remcsum *grc,
 550                                           bool nopartial)
 551 {
 552         size_t start, offset;
 553
 554         if (skb->remcsum_offload)
 555                 return vh;
 556
 557         if (!NAPI_GRO_CB(skb)->csum_valid)
 558                 return NULL;
 559
 560         start = vxlan_rco_start(vni_field);
 561         offset = start + vxlan_rco_offset(vni_field);
 562
 563         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 564                                      start, offset, grc, nopartial);
 565
 566         skb->remcsum_offload = 1;
 567
 568         return vh;
 569 }
 570
 571 static struct sk_buff *vxlan_gro_receive(struct sock *sk,
 572                                          struct 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                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 979                         struct vxlan_rdst *rd;
 980
 981                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 982                                 if (*idx < cb->args[2])
 983                                         goto skip;
 984
 985                                 err = vxlan_fdb_info(skb, vxlan, f,
 986                                                      NETLINK_CB(cb->skb).portid,
 987                                                      cb->nlh->nlmsg_seq,
 988                                                      RTM_NEWNEIGH,
 989                                                      NLM_F_MULTI, rd);
 990                                 if (err < 0)
 991                                         goto out;
 992 skip:
 993                                 *idx += 1;
 994                         }
 995                 }
 996         }
 997 out:
 998         return err;
 999 }
1000
1001 /* Watch incoming packets to learn mapping between Ethernet address
1002  * and Tunnel endpoint.
1003  * Return true if packet is bogus and should be dropped.
1004  */
1005 static bool vxlan_snoop(struct net_device *dev,
1006                         union vxlan_addr *src_ip, const u8 *src_mac,
1007                         u32 src_ifindex, __be32 vni)
1008 {
1009         struct vxlan_dev *vxlan = netdev_priv(dev);
1010         struct vxlan_fdb *f;
1011         u32 ifindex = 0;
1012
1013 #if IS_ENABLED(CONFIG_IPV6)
1014         if (src_ip->sa.sa_family == AF_INET6 &&
1015             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1016                 ifindex = src_ifindex;
1017 #endif
1018
1019         f = vxlan_find_mac(vxlan, src_mac, vni);
1020         if (likely(f)) {
1021                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1022
1023                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1024                            rdst->remote_ifindex == ifindex))
1025                         return false;
1026
1027                 /* Don't migrate static entries, drop packets */
1028                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1029                         return true;
1030
1031                 if (net_ratelimit())
1032                         netdev_info(dev,
1033                                     "%pM migrated from %pIS to %pIS\n",
1034                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1035
1036                 rdst->remote_ip = *src_ip;
1037                 f->updated = jiffies;
1038                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1039         } else {
1040                 /* learned new entry */
1041                 spin_lock(&vxlan->hash_lock);
1042
1043                 /* close off race between vxlan_flush and incoming packets */
1044                 if (netif_running(dev))
1045                         vxlan_fdb_update(vxlan, src_mac, src_ip,
1046                                          NUD_REACHABLE,
1047                                          NLM_F_EXCL|NLM_F_CREATE,
1048                                          vxlan->cfg.dst_port,
1049                                          vni,
1050                                          vxlan->default_dst.remote_vni,
1051                                          ifindex, NTF_SELF);
1052                 spin_unlock(&vxlan->hash_lock);
1053         }
1054
1055         return false;
1056 }
1057
1058 /* See if multicast group is already in use by other ID */
1059 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1060 {
1061         struct vxlan_dev *vxlan;
1062         struct vxlan_sock *sock4;
1063 #if IS_ENABLED(CONFIG_IPV6)
1064         struct vxlan_sock *sock6;
1065 #endif
1066         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1067
1068         sock4 = rtnl_dereference(dev->vn4_sock);
1069
1070         /* The vxlan_sock is only used by dev, leaving group has
1071          * no effect on other vxlan devices.
1072          */
1073         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1074                 return false;
1075 #if IS_ENABLED(CONFIG_IPV6)
1076         sock6 = rtnl_dereference(dev->vn6_sock);
1077         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1078                 return false;
1079 #endif
1080
1081         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1082                 if (!netif_running(vxlan->dev) || vxlan == dev)
1083                         continue;
1084
1085                 if (family == AF_INET &&
1086                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1087                         continue;
1088 #if IS_ENABLED(CONFIG_IPV6)
1089                 if (family == AF_INET6 &&
1090                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1091                         continue;
1092 #endif
1093
1094                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1095                                       &dev->default_dst.remote_ip))
1096                         continue;
1097
1098                 if (vxlan->default_dst.remote_ifindex !=
1099                     dev->default_dst.remote_ifindex)
1100                         continue;
1101
1102                 return true;
1103         }
1104
1105         return false;
1106 }
1107
1108 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1109 {
1110         struct vxlan_net *vn;
1111
1112         if (!vs)
1113                 return false;
1114         if (!refcount_dec_and_test(&vs->refcnt))
1115                 return false;
1116
1117         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1118         spin_lock(&vn->sock_lock);
1119         hlist_del_rcu(&vs->hlist);
1120         udp_tunnel_notify_del_rx_port(vs->sock,
1121                                       (vs->flags & VXLAN_F_GPE) ?
1122                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1123                                       UDP_TUNNEL_TYPE_VXLAN);
1124         spin_unlock(&vn->sock_lock);
1125
1126         return true;
1127 }
1128
1129 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1130 {
1131         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1132 #if IS_ENABLED(CONFIG_IPV6)
1133         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1134
1135         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1136 #endif
1137
1138         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1139         synchronize_net();
1140
1141         vxlan_vs_del_dev(vxlan);
1142
1143         if (__vxlan_sock_release_prep(sock4)) {
1144                 udp_tunnel_sock_release(sock4->sock);
1145                 kfree(sock4);
1146         }
1147
1148 #if IS_ENABLED(CONFIG_IPV6)
1149         if (__vxlan_sock_release_prep(sock6)) {
1150                 udp_tunnel_sock_release(sock6->sock);
1151                 kfree(sock6);
1152         }
1153 #endif
1154 }
1155
1156 /* Update multicast group membership when first VNI on
1157  * multicast address is brought up
1158  */
1159 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1160 {
1161         struct sock *sk;
1162         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1163         int ifindex = vxlan->default_dst.remote_ifindex;
1164         int ret = -EINVAL;
1165
1166         if (ip->sa.sa_family == AF_INET) {
1167                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1168                 struct ip_mreqn mreq = {
1169                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1170                         .imr_ifindex            = ifindex,
1171                 };
1172
1173                 sk = sock4->sock->sk;
1174                 lock_sock(sk);
1175                 ret = ip_mc_join_group(sk, &mreq);
1176                 release_sock(sk);
1177 #if IS_ENABLED(CONFIG_IPV6)
1178         } else {
1179                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1180
1181                 sk = sock6->sock->sk;
1182                 lock_sock(sk);
1183                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1184                                                    &ip->sin6.sin6_addr);
1185                 release_sock(sk);
1186 #endif
1187         }
1188
1189         return ret;
1190 }
1191
1192 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1193 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1194 {
1195         struct sock *sk;
1196         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1197         int ifindex = vxlan->default_dst.remote_ifindex;
1198         int ret = -EINVAL;
1199
1200         if (ip->sa.sa_family == AF_INET) {
1201                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1202                 struct ip_mreqn mreq = {
1203                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1204                         .imr_ifindex            = ifindex,
1205                 };
1206
1207                 sk = sock4->sock->sk;
1208                 lock_sock(sk);
1209                 ret = ip_mc_leave_group(sk, &mreq);
1210                 release_sock(sk);
1211 #if IS_ENABLED(CONFIG_IPV6)
1212         } else {
1213                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1214
1215                 sk = sock6->sock->sk;
1216                 lock_sock(sk);
1217                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1218                                                    &ip->sin6.sin6_addr);
1219                 release_sock(sk);
1220 #endif
1221         }
1222
1223         return ret;
1224 }
1225
1226 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1227                           struct sk_buff *skb, u32 vxflags)
1228 {
1229         size_t start, offset;
1230
1231         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1232                 goto out;
1233
1234         start = vxlan_rco_start(unparsed->vx_vni);
1235         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1236
1237         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1238                 return false;
1239
1240         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1241                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1242 out:
1243         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1244         unparsed->vx_vni &= VXLAN_VNI_MASK;
1245         return true;
1246 }
1247
1248 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1249                                 struct sk_buff *skb, u32 vxflags,
1250                                 struct vxlan_metadata *md)
1251 {
1252         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1253         struct metadata_dst *tun_dst;
1254
1255         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1256                 goto out;
1257
1258         md->gbp = ntohs(gbp->policy_id);
1259
1260         tun_dst = (struct metadata_dst *)skb_dst(skb);
1261         if (tun_dst) {
1262                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1263                 tun_dst->u.tun_info.options_len = sizeof(*md);
1264         }
1265         if (gbp->dont_learn)
1266                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1267
1268         if (gbp->policy_applied)
1269                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1270
1271         /* In flow-based mode, GBP is carried in dst_metadata */
1272         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1273                 skb->mark = md->gbp;
1274 out:
1275         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1276 }
1277
1278 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1279                                 __be16 *protocol,
1280                                 struct sk_buff *skb, u32 vxflags)
1281 {
1282         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1283
1284         /* Need to have Next Protocol set for interfaces in GPE mode. */
1285         if (!gpe->np_applied)
1286                 return false;
1287         /* "The initial version is 0. If a receiver does not support the
1288          * version indicated it MUST drop the packet.
1289          */
1290         if (gpe->version != 0)
1291                 return false;
1292         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1293          * processing MUST occur." However, we don't implement OAM
1294          * processing, thus drop the packet.
1295          */
1296         if (gpe->oam_flag)
1297                 return false;
1298
1299         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1300         if (!*protocol)
1301                 return false;
1302
1303         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1304         return true;
1305 }
1306
1307 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1308                           struct vxlan_sock *vs,
1309                           struct sk_buff *skb, __be32 vni)
1310 {
1311         union vxlan_addr saddr;
1312         u32 ifindex = skb->dev->ifindex;
1313
1314         skb_reset_mac_header(skb);
1315         skb->protocol = eth_type_trans(skb, vxlan->dev);
1316         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1317
1318         /* Ignore packet loops (and multicast echo) */
1319         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1320                 return false;
1321
1322         /* Get address from the outer IP header */
1323         if (vxlan_get_sk_family(vs) == AF_INET) {
1324                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1325                 saddr.sa.sa_family = AF_INET;
1326 #if IS_ENABLED(CONFIG_IPV6)
1327         } else {
1328                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1329                 saddr.sa.sa_family = AF_INET6;
1330 #endif
1331         }
1332
1333         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1334             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1335                 return false;
1336
1337         return true;
1338 }
1339
1340 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1341                                   struct sk_buff *skb)
1342 {
1343         int err = 0;
1344
1345         if (vxlan_get_sk_family(vs) == AF_INET)
1346                 err = IP_ECN_decapsulate(oiph, skb);
1347 #if IS_ENABLED(CONFIG_IPV6)
1348         else
1349                 err = IP6_ECN_decapsulate(oiph, skb);
1350 #endif
1351
1352         if (unlikely(err) && log_ecn_error) {
1353                 if (vxlan_get_sk_family(vs) == AF_INET)
1354                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1355                                              &((struct iphdr *)oiph)->saddr,
1356                                              ((struct iphdr *)oiph)->tos);
1357                 else
1358                         net_info_ratelimited("non-ECT from %pI6\n",
1359                                              &((struct ipv6hdr *)oiph)->saddr);
1360         }
1361         return err <= 1;
1362 }
1363
1364 /* Callback from net/ipv4/udp.c to receive packets */
1365 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1366 {
1367         struct pcpu_sw_netstats *stats;
1368         struct vxlan_dev *vxlan;
1369         struct vxlan_sock *vs;
1370         struct vxlanhdr unparsed;
1371         struct vxlan_metadata _md;
1372         struct vxlan_metadata *md = &_md;
1373         __be16 protocol = htons(ETH_P_TEB);
1374         bool raw_proto = false;
1375         void *oiph;
1376         __be32 vni = 0;
1377
1378         /* Need UDP and VXLAN header to be present */
1379         if (!pskb_may_pull(skb, VXLAN_HLEN))
1380                 goto drop;
1381
1382         unparsed = *vxlan_hdr(skb);
1383         /* VNI flag always required to be set */
1384         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1385                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1386                            ntohl(vxlan_hdr(skb)->vx_flags),
1387                            ntohl(vxlan_hdr(skb)->vx_vni));
1388                 /* Return non vxlan pkt */
1389                 goto drop;
1390         }
1391         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1392         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1393
1394         vs = rcu_dereference_sk_user_data(sk);
1395         if (!vs)
1396                 goto drop;
1397
1398         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1399
1400         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1401         if (!vxlan)
1402                 goto drop;
1403
1404         /* For backwards compatibility, only allow reserved fields to be
1405          * used by VXLAN extensions if explicitly requested.
1406          */
1407         if (vs->flags & VXLAN_F_GPE) {
1408                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1409                         goto drop;
1410                 raw_proto = true;
1411         }
1412
1413         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1414                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1415                         goto drop;
1416
1417         if (vxlan_collect_metadata(vs)) {
1418                 struct metadata_dst *tun_dst;
1419
1420                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1421                                          key32_to_tunnel_id(vni), sizeof(*md));
1422
1423                 if (!tun_dst)
1424                         goto drop;
1425
1426                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1427
1428                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1429         } else {
1430                 memset(md, 0, sizeof(*md));
1431         }
1432
1433         if (vs->flags & VXLAN_F_REMCSUM_RX)
1434                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1435                         goto drop;
1436         if (vs->flags & VXLAN_F_GBP)
1437                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1438         /* Note that GBP and GPE can never be active together. This is
1439          * ensured in vxlan_dev_configure.
1440          */
1441
1442         if (unparsed.vx_flags || unparsed.vx_vni) {
1443                 /* If there are any unprocessed flags remaining treat
1444                  * this as a malformed packet. This behavior diverges from
1445                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1446                  * in reserved fields are to be ignored. The approach here
1447                  * maintains compatibility with previous stack code, and also
1448                  * is more robust and provides a little more security in
1449                  * adding extensions to VXLAN.
1450                  */
1451                 goto drop;
1452         }
1453
1454         if (!raw_proto) {
1455                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1456                         goto drop;
1457         } else {
1458                 skb_reset_mac_header(skb);
1459                 skb->dev = vxlan->dev;
1460                 skb->pkt_type = PACKET_HOST;
1461         }
1462
1463         oiph = skb_network_header(skb);
1464         skb_reset_network_header(skb);
1465
1466         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1467                 ++vxlan->dev->stats.rx_frame_errors;
1468                 ++vxlan->dev->stats.rx_errors;
1469                 goto drop;
1470         }
1471
1472         stats = this_cpu_ptr(vxlan->dev->tstats);
1473         u64_stats_update_begin(&stats->syncp);
1474         stats->rx_packets++;
1475         stats->rx_bytes += skb->len;
1476         u64_stats_update_end(&stats->syncp);
1477
1478         gro_cells_receive(&vxlan->gro_cells, skb);
1479         return 0;
1480
1481 drop:
1482         /* Consume bad packet */
1483         kfree_skb(skb);
1484         return 0;
1485 }
1486
1487 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1488 {
1489         struct vxlan_dev *vxlan = netdev_priv(dev);
1490         struct arphdr *parp;
1491         u8 *arpptr, *sha;
1492         __be32 sip, tip;
1493         struct neighbour *n;
1494
1495         if (dev->flags & IFF_NOARP)
1496                 goto out;
1497
1498         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1499                 dev->stats.tx_dropped++;
1500                 goto out;
1501         }
1502         parp = arp_hdr(skb);
1503
1504         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1505              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1506             parp->ar_pro != htons(ETH_P_IP) ||
1507             parp->ar_op != htons(ARPOP_REQUEST) ||
1508             parp->ar_hln != dev->addr_len ||
1509             parp->ar_pln != 4)
1510                 goto out;
1511         arpptr = (u8 *)parp + sizeof(struct arphdr);
1512         sha = arpptr;
1513         arpptr += dev->addr_len;        /* sha */
1514         memcpy(&sip, arpptr, sizeof(sip));
1515         arpptr += sizeof(sip);
1516         arpptr += dev->addr_len;        /* tha */
1517         memcpy(&tip, arpptr, sizeof(tip));
1518
1519         if (ipv4_is_loopback(tip) ||
1520             ipv4_is_multicast(tip))
1521                 goto out;
1522
1523         n = neigh_lookup(&arp_tbl, &tip, dev);
1524
1525         if (n) {
1526                 struct vxlan_fdb *f;
1527                 struct sk_buff  *reply;
1528
1529                 if (!(n->nud_state & NUD_CONNECTED)) {
1530                         neigh_release(n);
1531                         goto out;
1532                 }
1533
1534                 f = vxlan_find_mac(vxlan, n->ha, vni);
1535                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1536                         /* bridge-local neighbor */
1537                         neigh_release(n);
1538                         goto out;
1539                 }
1540
1541                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1542                                 n->ha, sha);
1543
1544                 neigh_release(n);
1545
1546                 if (reply == NULL)
1547                         goto out;
1548
1549                 skb_reset_mac_header(reply);
1550                 __skb_pull(reply, skb_network_offset(reply));
1551                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1552                 reply->pkt_type = PACKET_HOST;
1553
1554                 if (netif_rx_ni(reply) == NET_RX_DROP)
1555                         dev->stats.rx_dropped++;
1556         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1557                 union vxlan_addr ipa = {
1558                         .sin.sin_addr.s_addr = tip,
1559                         .sin.sin_family = AF_INET,
1560                 };
1561
1562                 vxlan_ip_miss(dev, &ipa);
1563         }
1564 out:
1565         consume_skb(skb);
1566         return NETDEV_TX_OK;
1567 }
1568
1569 #if IS_ENABLED(CONFIG_IPV6)
1570 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1571         struct neighbour *n, bool isrouter)
1572 {
1573         struct net_device *dev = request->dev;
1574         struct sk_buff *reply;
1575         struct nd_msg *ns, *na;
1576         struct ipv6hdr *pip6;
1577         u8 *daddr;
1578         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1579         int ns_olen;
1580         int i, len;
1581
1582         if (dev == NULL || !pskb_may_pull(request, request->len))
1583                 return NULL;
1584
1585         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1586                 sizeof(*na) + na_olen + dev->needed_tailroom;
1587         reply = alloc_skb(len, GFP_ATOMIC);
1588         if (reply == NULL)
1589                 return NULL;
1590
1591         reply->protocol = htons(ETH_P_IPV6);
1592         reply->dev = dev;
1593         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1594         skb_push(reply, sizeof(struct ethhdr));
1595         skb_reset_mac_header(reply);
1596
1597         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1598
1599         daddr = eth_hdr(request)->h_source;
1600         ns_olen = request->len - skb_network_offset(request) -
1601                 sizeof(struct ipv6hdr) - sizeof(*ns);
1602         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1603                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1604                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1605                         break;
1606                 }
1607         }
1608
1609         /* Ethernet header */
1610         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1611         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1612         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1613         reply->protocol = htons(ETH_P_IPV6);
1614
1615         skb_pull(reply, sizeof(struct ethhdr));
1616         skb_reset_network_header(reply);
1617         skb_put(reply, sizeof(struct ipv6hdr));
1618
1619         /* IPv6 header */
1620
1621         pip6 = ipv6_hdr(reply);
1622         memset(pip6, 0, sizeof(struct ipv6hdr));
1623         pip6->version = 6;
1624         pip6->priority = ipv6_hdr(request)->priority;
1625         pip6->nexthdr = IPPROTO_ICMPV6;
1626         pip6->hop_limit = 255;
1627         pip6->daddr = ipv6_hdr(request)->saddr;
1628         pip6->saddr = *(struct in6_addr *)n->primary_key;
1629
1630         skb_pull(reply, sizeof(struct ipv6hdr));
1631         skb_reset_transport_header(reply);
1632
1633         /* Neighbor Advertisement */
1634         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1635         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1636         na->icmph.icmp6_router = isrouter;
1637         na->icmph.icmp6_override = 1;
1638         na->icmph.icmp6_solicited = 1;
1639         na->target = ns->target;
1640         ether_addr_copy(&na->opt[2], n->ha);
1641         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1642         na->opt[1] = na_olen >> 3;
1643
1644         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1645                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1646                 csum_partial(na, sizeof(*na)+na_olen, 0));
1647
1648         pip6->payload_len = htons(sizeof(*na)+na_olen);
1649
1650         skb_push(reply, sizeof(struct ipv6hdr));
1651
1652         reply->ip_summed = CHECKSUM_UNNECESSARY;
1653
1654         return reply;
1655 }
1656
1657 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1658 {
1659         struct vxlan_dev *vxlan = netdev_priv(dev);
1660         const struct in6_addr *daddr;
1661         const struct ipv6hdr *iphdr;
1662         struct inet6_dev *in6_dev;
1663         struct neighbour *n;
1664         struct nd_msg *msg;
1665
1666         in6_dev = __in6_dev_get(dev);
1667         if (!in6_dev)
1668                 goto out;
1669
1670         iphdr = ipv6_hdr(skb);
1671         daddr = &iphdr->daddr;
1672         msg = (struct nd_msg *)(iphdr + 1);
1673
1674         if (ipv6_addr_loopback(daddr) ||
1675             ipv6_addr_is_multicast(&msg->target))
1676                 goto out;
1677
1678         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1679
1680         if (n) {
1681                 struct vxlan_fdb *f;
1682                 struct sk_buff *reply;
1683
1684                 if (!(n->nud_state & NUD_CONNECTED)) {
1685                         neigh_release(n);
1686                         goto out;
1687                 }
1688
1689                 f = vxlan_find_mac(vxlan, n->ha, vni);
1690                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1691                         /* bridge-local neighbor */
1692                         neigh_release(n);
1693                         goto out;
1694                 }
1695
1696                 reply = vxlan_na_create(skb, n,
1697                                         !!(f ? f->flags & NTF_ROUTER : 0));
1698
1699                 neigh_release(n);
1700
1701                 if (reply == NULL)
1702                         goto out;
1703
1704                 if (netif_rx_ni(reply) == NET_RX_DROP)
1705                         dev->stats.rx_dropped++;
1706
1707         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1708                 union vxlan_addr ipa = {
1709                         .sin6.sin6_addr = msg->target,
1710                         .sin6.sin6_family = AF_INET6,
1711                 };
1712
1713                 vxlan_ip_miss(dev, &ipa);
1714         }
1715
1716 out:
1717         consume_skb(skb);
1718         return NETDEV_TX_OK;
1719 }
1720 #endif
1721
1722 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1723 {
1724         struct vxlan_dev *vxlan = netdev_priv(dev);
1725         struct neighbour *n;
1726
1727         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1728                 return false;
1729
1730         n = NULL;
1731         switch (ntohs(eth_hdr(skb)->h_proto)) {
1732         case ETH_P_IP:
1733         {
1734                 struct iphdr *pip;
1735
1736                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1737                         return false;
1738                 pip = ip_hdr(skb);
1739                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1740                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1741                         union vxlan_addr ipa = {
1742                                 .sin.sin_addr.s_addr = pip->daddr,
1743                                 .sin.sin_family = AF_INET,
1744                         };
1745
1746                         vxlan_ip_miss(dev, &ipa);
1747                         return false;
1748                 }
1749
1750                 break;
1751         }
1752 #if IS_ENABLED(CONFIG_IPV6)
1753         case ETH_P_IPV6:
1754         {
1755                 struct ipv6hdr *pip6;
1756
1757                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1758                         return false;
1759                 pip6 = ipv6_hdr(skb);
1760                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1761                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1762                         union vxlan_addr ipa = {
1763                                 .sin6.sin6_addr = pip6->daddr,
1764                                 .sin6.sin6_family = AF_INET6,
1765                         };
1766
1767                         vxlan_ip_miss(dev, &ipa);
1768                         return false;
1769                 }
1770
1771                 break;
1772         }
1773 #endif
1774         default:
1775                 return false;
1776         }
1777
1778         if (n) {
1779                 bool diff;
1780
1781                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1782                 if (diff) {
1783                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1784                                 dev->addr_len);
1785                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1786                 }
1787                 neigh_release(n);
1788                 return diff;
1789         }
1790
1791         return false;
1792 }
1793
1794 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1795                                 struct vxlan_metadata *md)
1796 {
1797         struct vxlanhdr_gbp *gbp;
1798
1799         if (!md->gbp)
1800                 return;
1801
1802         gbp = (struct vxlanhdr_gbp *)vxh;
1803         vxh->vx_flags |= VXLAN_HF_GBP;
1804
1805         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1806                 gbp->dont_learn = 1;
1807
1808         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1809                 gbp->policy_applied = 1;
1810
1811         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1812 }
1813
1814 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1815                                __be16 protocol)
1816 {
1817         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1818
1819         gpe->np_applied = 1;
1820         gpe->next_protocol = tun_p_from_eth_p(protocol);
1821         if (!gpe->next_protocol)
1822                 return -EPFNOSUPPORT;
1823         return 0;
1824 }
1825
1826 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1827                            int iphdr_len, __be32 vni,
1828                            struct vxlan_metadata *md, u32 vxflags,
1829                            bool udp_sum)
1830 {
1831         struct vxlanhdr *vxh;
1832         int min_headroom;
1833         int err;
1834         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1835         __be16 inner_protocol = htons(ETH_P_TEB);
1836
1837         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1838             skb->ip_summed == CHECKSUM_PARTIAL) {
1839                 int csum_start = skb_checksum_start_offset(skb);
1840
1841                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1842                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1843                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1844                      skb->csum_offset == offsetof(struct tcphdr, check)))
1845                         type |= SKB_GSO_TUNNEL_REMCSUM;
1846         }
1847
1848         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1849                         + VXLAN_HLEN + iphdr_len;
1850
1851         /* Need space for new headers (invalidates iph ptr) */
1852         err = skb_cow_head(skb, min_headroom);
1853         if (unlikely(err))
1854                 return err;
1855
1856         err = iptunnel_handle_offloads(skb, type);
1857         if (err)
1858                 return err;
1859
1860         vxh = __skb_push(skb, sizeof(*vxh));
1861         vxh->vx_flags = VXLAN_HF_VNI;
1862         vxh->vx_vni = vxlan_vni_field(vni);
1863
1864         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1865                 unsigned int start;
1866
1867                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1868                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1869                 vxh->vx_flags |= VXLAN_HF_RCO;
1870
1871                 if (!skb_is_gso(skb)) {
1872                         skb->ip_summed = CHECKSUM_NONE;
1873                         skb->encapsulation = 0;
1874                 }
1875         }
1876
1877         if (vxflags & VXLAN_F_GBP)
1878                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1879         if (vxflags & VXLAN_F_GPE) {
1880                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1881                 if (err < 0)
1882                         return err;
1883                 inner_protocol = skb->protocol;
1884         }
1885
1886         skb_set_inner_protocol(skb, inner_protocol);
1887         return 0;
1888 }
1889
1890 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1891                                       struct vxlan_sock *sock4,
1892                                       struct sk_buff *skb, int oif, u8 tos,
1893                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
1894                                       struct dst_cache *dst_cache,
1895                                       const struct ip_tunnel_info *info)
1896 {
1897         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1898         struct rtable *rt = NULL;
1899         struct flowi4 fl4;
1900
1901         if (!sock4)
1902                 return ERR_PTR(-EIO);
1903
1904         if (tos && !info)
1905                 use_cache = false;
1906         if (use_cache) {
1907                 rt = dst_cache_get_ip4(dst_cache, saddr);
1908                 if (rt)
1909                         return rt;
1910         }
1911
1912         memset(&fl4, 0, sizeof(fl4));
1913         fl4.flowi4_oif = oif;
1914         fl4.flowi4_tos = RT_TOS(tos);
1915         fl4.flowi4_mark = skb->mark;
1916         fl4.flowi4_proto = IPPROTO_UDP;
1917         fl4.daddr = daddr;
1918         fl4.saddr = *saddr;
1919         fl4.fl4_dport = dport;
1920         fl4.fl4_sport = sport;
1921
1922         rt = ip_route_output_key(vxlan->net, &fl4);
1923         if (likely(!IS_ERR(rt))) {
1924                 if (rt->dst.dev == dev) {
1925                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1926                         ip_rt_put(rt);
1927                         return ERR_PTR(-ELOOP);
1928                 }
1929
1930                 *saddr = fl4.saddr;
1931                 if (use_cache)
1932                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1933         } else {
1934                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1935                 return ERR_PTR(-ENETUNREACH);
1936         }
1937         return rt;
1938 }
1939
1940 #if IS_ENABLED(CONFIG_IPV6)
1941 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1942                                           struct net_device *dev,
1943                                           struct vxlan_sock *sock6,
1944                                           struct sk_buff *skb, int oif, u8 tos,
1945                                           __be32 label,
1946                                           const struct in6_addr *daddr,
1947                                           struct in6_addr *saddr,
1948                                           __be16 dport, __be16 sport,
1949                                           struct dst_cache *dst_cache,
1950                                           const struct ip_tunnel_info *info)
1951 {
1952         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1953         struct dst_entry *ndst;
1954         struct flowi6 fl6;
1955         int err;
1956
1957         if (!sock6)
1958                 return ERR_PTR(-EIO);
1959
1960         if (tos && !info)
1961                 use_cache = false;
1962         if (use_cache) {
1963                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1964                 if (ndst)
1965                         return ndst;
1966         }
1967
1968         memset(&fl6, 0, sizeof(fl6));
1969         fl6.flowi6_oif = oif;
1970         fl6.daddr = *daddr;
1971         fl6.saddr = *saddr;
1972         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1973         fl6.flowi6_mark = skb->mark;
1974         fl6.flowi6_proto = IPPROTO_UDP;
1975         fl6.fl6_dport = dport;
1976         fl6.fl6_sport = sport;
1977
1978         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1979                                          sock6->sock->sk,
1980                                          &ndst, &fl6);
1981         if (unlikely(err < 0)) {
1982                 netdev_dbg(dev, "no route to %pI6\n", daddr);
1983                 return ERR_PTR(-ENETUNREACH);
1984         }
1985
1986         if (unlikely(ndst->dev == dev)) {
1987                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
1988                 dst_release(ndst);
1989                 return ERR_PTR(-ELOOP);
1990         }
1991
1992         *saddr = fl6.saddr;
1993         if (use_cache)
1994                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1995         return ndst;
1996 }
1997 #endif
1998
1999 /* Bypass encapsulation if the destination is local */
2000 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2001                                struct vxlan_dev *dst_vxlan, __be32 vni)
2002 {
2003         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2004         union vxlan_addr loopback;
2005         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2006         struct net_device *dev = skb->dev;
2007         int len = skb->len;
2008
2009         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2010         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2011         skb->pkt_type = PACKET_HOST;
2012         skb->encapsulation = 0;
2013         skb->dev = dst_vxlan->dev;
2014         __skb_pull(skb, skb_network_offset(skb));
2015
2016         if (remote_ip->sa.sa_family == AF_INET) {
2017                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2018                 loopback.sa.sa_family =  AF_INET;
2019 #if IS_ENABLED(CONFIG_IPV6)
2020         } else {
2021                 loopback.sin6.sin6_addr = in6addr_loopback;
2022                 loopback.sa.sa_family =  AF_INET6;
2023 #endif
2024         }
2025
2026         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
2027                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source, 0,
2028                             vni);
2029
2030         u64_stats_update_begin(&tx_stats->syncp);
2031         tx_stats->tx_packets++;
2032         tx_stats->tx_bytes += len;
2033         u64_stats_update_end(&tx_stats->syncp);
2034
2035         if (netif_rx(skb) == NET_RX_SUCCESS) {
2036                 u64_stats_update_begin(&rx_stats->syncp);
2037                 rx_stats->rx_packets++;
2038                 rx_stats->rx_bytes += len;
2039                 u64_stats_update_end(&rx_stats->syncp);
2040         } else {
2041                 dev->stats.rx_dropped++;
2042         }
2043 }
2044
2045 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2046                                  struct vxlan_dev *vxlan,
2047                                  union vxlan_addr *daddr,
2048                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2049                                  struct dst_entry *dst,
2050                                  u32 rt_flags)
2051 {
2052 #if IS_ENABLED(CONFIG_IPV6)
2053         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2054          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2055          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2056          */
2057         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2058 #endif
2059         /* Bypass encapsulation if the destination is local */
2060         if (rt_flags & RTCF_LOCAL &&
2061             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2062                 struct vxlan_dev *dst_vxlan;
2063
2064                 dst_release(dst);
2065                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2066                                            daddr->sa.sa_family, dst_port,
2067                                            vxlan->cfg.flags);
2068                 if (!dst_vxlan) {
2069                         dev->stats.tx_errors++;
2070                         kfree_skb(skb);
2071
2072                         return -ENOENT;
2073                 }
2074                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2075                 return 1;
2076         }
2077
2078         return 0;
2079 }
2080
2081 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2082                            __be32 default_vni, struct vxlan_rdst *rdst,
2083                            bool did_rsc)
2084 {
2085         struct dst_cache *dst_cache;
2086         struct ip_tunnel_info *info;
2087         struct vxlan_dev *vxlan = netdev_priv(dev);
2088         const struct iphdr *old_iph = ip_hdr(skb);
2089         union vxlan_addr *dst;
2090         union vxlan_addr remote_ip, local_ip;
2091         struct vxlan_metadata _md;
2092         struct vxlan_metadata *md = &_md;
2093         __be16 src_port = 0, dst_port;
2094         struct dst_entry *ndst = NULL;
2095         __be32 vni, label;
2096         __u8 tos, ttl;
2097         int ifindex;
2098         int err;
2099         u32 flags = vxlan->cfg.flags;
2100         bool udp_sum = false;
2101         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2102
2103         info = skb_tunnel_info(skb);
2104
2105         if (rdst) {
2106                 dst = &rdst->remote_ip;
2107                 if (vxlan_addr_any(dst)) {
2108                         if (did_rsc) {
2109                                 /* short-circuited back to local bridge */
2110                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2111                                 return;
2112                         }
2113                         goto drop;
2114                 }
2115
2116                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2117                 vni = (rdst->remote_vni) ? : default_vni;
2118                 ifindex = rdst->remote_ifindex;
2119                 local_ip = vxlan->cfg.saddr;
2120                 dst_cache = &rdst->dst_cache;
2121                 md->gbp = skb->mark;
2122                 if (flags & VXLAN_F_TTL_INHERIT) {
2123                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2124                 } else {
2125                         ttl = vxlan->cfg.ttl;
2126                         if (!ttl && vxlan_addr_multicast(dst))
2127                                 ttl = 1;
2128                 }
2129
2130                 tos = vxlan->cfg.tos;
2131                 if (tos == 1)
2132                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2133
2134                 if (dst->sa.sa_family == AF_INET)
2135                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2136                 else
2137                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2138                 label = vxlan->cfg.label;
2139         } else {
2140                 if (!info) {
2141                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2142                                   dev->name);
2143                         goto drop;
2144                 }
2145                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2146                 if (remote_ip.sa.sa_family == AF_INET) {
2147                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2148                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2149                 } else {
2150                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2151                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2152                 }
2153                 dst = &remote_ip;
2154                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2155                 vni = tunnel_id_to_key32(info->key.tun_id);
2156                 ifindex = 0;
2157                 dst_cache = &info->dst_cache;
2158                 if (info->options_len &&
2159                     info->key.tun_flags & TUNNEL_VXLAN_OPT)
2160                         md = ip_tunnel_info_opts(info);
2161                 ttl = info->key.ttl;
2162                 tos = info->key.tos;
2163                 label = info->key.label;
2164                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2165         }
2166         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2167                                      vxlan->cfg.port_max, true);
2168
2169         rcu_read_lock();
2170         if (dst->sa.sa_family == AF_INET) {
2171                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2172                 struct rtable *rt;
2173                 __be16 df = 0;
2174
2175                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2176                                      dst->sin.sin_addr.s_addr,
2177                                      &local_ip.sin.sin_addr.s_addr,
2178                                      dst_port, src_port,
2179                                      dst_cache, info);
2180                 if (IS_ERR(rt)) {
2181                         err = PTR_ERR(rt);
2182                         goto tx_error;
2183                 }
2184
2185                 /* Bypass encapsulation if the destination is local */
2186                 if (!info) {
2187                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2188                                                     dst_port, ifindex, vni,
2189                                                     &rt->dst, rt->rt_flags);
2190                         if (err)
2191                                 goto out_unlock;
2192                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2193                         df = htons(IP_DF);
2194                 }
2195
2196                 ndst = &rt->dst;
2197                 skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM);
2198
2199                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2200                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2201                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2202                                       vni, md, flags, udp_sum);
2203                 if (err < 0)
2204                         goto tx_error;
2205
2206                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2207                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2208                                     src_port, dst_port, xnet, !udp_sum);
2209 #if IS_ENABLED(CONFIG_IPV6)
2210         } else {
2211                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2212
2213                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2214                                         label, &dst->sin6.sin6_addr,
2215                                         &local_ip.sin6.sin6_addr,
2216                                         dst_port, src_port,
2217                                         dst_cache, info);
2218                 if (IS_ERR(ndst)) {
2219                         err = PTR_ERR(ndst);
2220                         ndst = NULL;
2221                         goto tx_error;
2222                 }
2223
2224                 if (!info) {
2225                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2226
2227                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2228                                                     dst_port, ifindex, vni,
2229                                                     ndst, rt6i_flags);
2230                         if (err)
2231                                 goto out_unlock;
2232                 }
2233
2234                 skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM);
2235
2236                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2237                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2238                 skb_scrub_packet(skb, xnet);
2239                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2240                                       vni, md, flags, udp_sum);
2241                 if (err < 0)
2242                         goto tx_error;
2243
2244                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2245                                      &local_ip.sin6.sin6_addr,
2246                                      &dst->sin6.sin6_addr, tos, ttl,
2247                                      label, src_port, dst_port, !udp_sum);
2248 #endif
2249         }
2250 out_unlock:
2251         rcu_read_unlock();
2252         return;
2253
2254 drop:
2255         dev->stats.tx_dropped++;
2256         dev_kfree_skb(skb);
2257         return;
2258
2259 tx_error:
2260         rcu_read_unlock();
2261         if (err == -ELOOP)
2262                 dev->stats.collisions++;
2263         else if (err == -ENETUNREACH)
2264                 dev->stats.tx_carrier_errors++;
2265         dst_release(ndst);
2266         dev->stats.tx_errors++;
2267         kfree_skb(skb);
2268 }
2269
2270 /* Transmit local packets over Vxlan
2271  *
2272  * Outer IP header inherits ECN and DF from inner header.
2273  * Outer UDP destination is the VXLAN assigned port.
2274  *           source port is based on hash of flow
2275  */
2276 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2277 {
2278         struct vxlan_dev *vxlan = netdev_priv(dev);
2279         struct vxlan_rdst *rdst, *fdst = NULL;
2280         const struct ip_tunnel_info *info;
2281         bool did_rsc = false;
2282         struct vxlan_fdb *f;
2283         struct ethhdr *eth;
2284         __be32 vni = 0;
2285
2286         info = skb_tunnel_info(skb);
2287
2288         skb_reset_mac_header(skb);
2289
2290         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2291                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2292                     info->mode & IP_TUNNEL_INFO_TX) {
2293                         vni = tunnel_id_to_key32(info->key.tun_id);
2294                 } else {
2295                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2296                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2297                         else
2298                                 kfree_skb(skb);
2299                         return NETDEV_TX_OK;
2300                 }
2301         }
2302
2303         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2304                 eth = eth_hdr(skb);
2305                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2306                         return arp_reduce(dev, skb, vni);
2307 #if IS_ENABLED(CONFIG_IPV6)
2308                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2309                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2310                                             sizeof(struct nd_msg)) &&
2311                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2312                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2313
2314                         if (m->icmph.icmp6_code == 0 &&
2315                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2316                                 return neigh_reduce(dev, skb, vni);
2317                 }
2318 #endif
2319         }
2320
2321         eth = eth_hdr(skb);
2322         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2323         did_rsc = false;
2324
2325         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2326             (ntohs(eth->h_proto) == ETH_P_IP ||
2327              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2328                 did_rsc = route_shortcircuit(dev, skb);
2329                 if (did_rsc)
2330                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2331         }
2332
2333         if (f == NULL) {
2334                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2335                 if (f == NULL) {
2336                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2337                             !is_multicast_ether_addr(eth->h_dest))
2338                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2339
2340                         dev->stats.tx_dropped++;
2341                         kfree_skb(skb);
2342                         return NETDEV_TX_OK;
2343                 }
2344         }
2345
2346         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2347                 struct sk_buff *skb1;
2348
2349                 if (!fdst) {
2350                         fdst = rdst;
2351                         continue;
2352                 }
2353                 skb1 = skb_clone(skb, GFP_ATOMIC);
2354                 if (skb1)
2355                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2356         }
2357
2358         if (fdst)
2359                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2360         else
2361                 kfree_skb(skb);
2362         return NETDEV_TX_OK;
2363 }
2364
2365 /* Walk the forwarding table and purge stale entries */
2366 static void vxlan_cleanup(struct timer_list *t)
2367 {
2368         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2369         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2370         unsigned int h;
2371
2372         if (!netif_running(vxlan->dev))
2373                 return;
2374
2375         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2376                 struct hlist_node *p, *n;
2377
2378                 spin_lock_bh(&vxlan->hash_lock);
2379                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2380                         struct vxlan_fdb *f
2381                                 = container_of(p, struct vxlan_fdb, hlist);
2382                         unsigned long timeout;
2383
2384                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2385                                 continue;
2386
2387                         if (f->flags & NTF_EXT_LEARNED)
2388                                 continue;
2389
2390                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2391                         if (time_before_eq(timeout, jiffies)) {
2392                                 netdev_dbg(vxlan->dev,
2393                                            "garbage collect %pM\n",
2394                                            f->eth_addr);
2395                                 f->state = NUD_STALE;
2396                                 vxlan_fdb_destroy(vxlan, f, true);
2397                         } else if (time_before(timeout, next_timer))
2398                                 next_timer = timeout;
2399                 }
2400                 spin_unlock_bh(&vxlan->hash_lock);
2401         }
2402
2403         mod_timer(&vxlan->age_timer, next_timer);
2404 }
2405
2406 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2407 {
2408         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2409
2410         spin_lock(&vn->sock_lock);
2411         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2412 #if IS_ENABLED(CONFIG_IPV6)
2413         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2414 #endif
2415         spin_unlock(&vn->sock_lock);
2416 }
2417
2418 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2419                              struct vxlan_dev_node *node)
2420 {
2421         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2422         __be32 vni = vxlan->default_dst.remote_vni;
2423
2424         node->vxlan = vxlan;
2425         spin_lock(&vn->sock_lock);
2426         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2427         spin_unlock(&vn->sock_lock);
2428 }
2429
2430 /* Setup stats when device is created */
2431 static int vxlan_init(struct net_device *dev)
2432 {
2433         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2434         if (!dev->tstats)
2435                 return -ENOMEM;
2436
2437         return 0;
2438 }
2439
2440 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2441 {
2442         struct vxlan_fdb *f;
2443
2444         spin_lock_bh(&vxlan->hash_lock);
2445         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2446         if (f)
2447                 vxlan_fdb_destroy(vxlan, f, true);
2448         spin_unlock_bh(&vxlan->hash_lock);
2449 }
2450
2451 static void vxlan_uninit(struct net_device *dev)
2452 {
2453         struct vxlan_dev *vxlan = netdev_priv(dev);
2454
2455         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2456
2457         free_percpu(dev->tstats);
2458 }
2459
2460 /* Start ageing timer and join group when device is brought up */
2461 static int vxlan_open(struct net_device *dev)
2462 {
2463         struct vxlan_dev *vxlan = netdev_priv(dev);
2464         int ret;
2465
2466         ret = vxlan_sock_add(vxlan);
2467         if (ret < 0)
2468                 return ret;
2469
2470         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2471                 ret = vxlan_igmp_join(vxlan);
2472                 if (ret == -EADDRINUSE)
2473                         ret = 0;
2474                 if (ret) {
2475                         vxlan_sock_release(vxlan);
2476                         return ret;
2477                 }
2478         }
2479
2480         if (vxlan->cfg.age_interval)
2481                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2482
2483         return ret;
2484 }
2485
2486 /* Purge the forwarding table */
2487 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2488 {
2489         unsigned int h;
2490
2491         spin_lock_bh(&vxlan->hash_lock);
2492         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2493                 struct hlist_node *p, *n;
2494                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2495                         struct vxlan_fdb *f
2496                                 = container_of(p, struct vxlan_fdb, hlist);
2497                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2498                                 continue;
2499                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2500                         if (!is_zero_ether_addr(f->eth_addr))
2501                                 vxlan_fdb_destroy(vxlan, f, true);
2502                 }
2503         }
2504         spin_unlock_bh(&vxlan->hash_lock);
2505 }
2506
2507 /* Cleanup timer and forwarding table on shutdown */
2508 static int vxlan_stop(struct net_device *dev)
2509 {
2510         struct vxlan_dev *vxlan = netdev_priv(dev);
2511         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2512         int ret = 0;
2513
2514         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2515             !vxlan_group_used(vn, vxlan))
2516                 ret = vxlan_igmp_leave(vxlan);
2517
2518         del_timer_sync(&vxlan->age_timer);
2519
2520         vxlan_flush(vxlan, false);
2521         vxlan_sock_release(vxlan);
2522
2523         return ret;
2524 }
2525
2526 /* Stub, nothing needs to be done. */
2527 static void vxlan_set_multicast_list(struct net_device *dev)
2528 {
2529 }
2530
2531 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2532 {
2533         struct vxlan_dev *vxlan = netdev_priv(dev);
2534         struct vxlan_rdst *dst = &vxlan->default_dst;
2535         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2536                                                          dst->remote_ifindex);
2537         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
2538
2539         /* This check is different than dev->max_mtu, because it looks at
2540          * the lowerdev->mtu, rather than the static dev->max_mtu
2541          */
2542         if (lowerdev) {
2543                 int max_mtu = lowerdev->mtu -
2544                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2545                 if (new_mtu > max_mtu)
2546                         return -EINVAL;
2547         }
2548
2549         dev->mtu = new_mtu;
2550         return 0;
2551 }
2552
2553 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2554 {
2555         struct vxlan_dev *vxlan = netdev_priv(dev);
2556         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2557         __be16 sport, dport;
2558
2559         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2560                                   vxlan->cfg.port_max, true);
2561         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2562
2563         if (ip_tunnel_info_af(info) == AF_INET) {
2564                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2565                 struct rtable *rt;
2566
2567                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2568                                      info->key.u.ipv4.dst,
2569                                      &info->key.u.ipv4.src, dport, sport,
2570                                      &info->dst_cache, info);
2571                 if (IS_ERR(rt))
2572                         return PTR_ERR(rt);
2573                 ip_rt_put(rt);
2574         } else {
2575 #if IS_ENABLED(CONFIG_IPV6)
2576                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2577                 struct dst_entry *ndst;
2578
2579                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2580                                         info->key.label, &info->key.u.ipv6.dst,
2581                                         &info->key.u.ipv6.src, dport, sport,
2582                                         &info->dst_cache, info);
2583                 if (IS_ERR(ndst))
2584                         return PTR_ERR(ndst);
2585                 dst_release(ndst);
2586 #else /* !CONFIG_IPV6 */
2587                 return -EPFNOSUPPORT;
2588 #endif
2589         }
2590         info->key.tp_src = sport;
2591         info->key.tp_dst = dport;
2592         return 0;
2593 }
2594
2595 static const struct net_device_ops vxlan_netdev_ether_ops = {
2596         .ndo_init               = vxlan_init,
2597         .ndo_uninit             = vxlan_uninit,
2598         .ndo_open               = vxlan_open,
2599         .ndo_stop               = vxlan_stop,
2600         .ndo_start_xmit         = vxlan_xmit,
2601         .ndo_get_stats64        = ip_tunnel_get_stats64,
2602         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2603         .ndo_change_mtu         = vxlan_change_mtu,
2604         .ndo_validate_addr      = eth_validate_addr,
2605         .ndo_set_mac_address    = eth_mac_addr,
2606         .ndo_fdb_add            = vxlan_fdb_add,
2607         .ndo_fdb_del            = vxlan_fdb_delete,
2608         .ndo_fdb_dump           = vxlan_fdb_dump,
2609         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2610 };
2611
2612 static const struct net_device_ops vxlan_netdev_raw_ops = {
2613         .ndo_init               = vxlan_init,
2614         .ndo_uninit             = vxlan_uninit,
2615         .ndo_open               = vxlan_open,
2616         .ndo_stop               = vxlan_stop,
2617         .ndo_start_xmit         = vxlan_xmit,
2618         .ndo_get_stats64        = ip_tunnel_get_stats64,
2619         .ndo_change_mtu         = vxlan_change_mtu,
2620         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2621 };
2622
2623 /* Info for udev, that this is a virtual tunnel endpoint */
2624 static struct device_type vxlan_type = {
2625         .name = "vxlan",
2626 };
2627
2628 /* Calls the ndo_udp_tunnel_add of the caller in order to
2629  * supply the listening VXLAN udp ports. Callers are expected
2630  * to implement the ndo_udp_tunnel_add.
2631  */
2632 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
2633 {
2634         struct vxlan_sock *vs;
2635         struct net *net = dev_net(dev);
2636         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2637         unsigned int i;
2638
2639         spin_lock(&vn->sock_lock);
2640         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2641                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2642                         unsigned short type;
2643
2644                         if (vs->flags & VXLAN_F_GPE)
2645                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
2646                         else
2647                                 type = UDP_TUNNEL_TYPE_VXLAN;
2648
2649                         if (push)
2650                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
2651                         else
2652                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
2653                 }
2654         }
2655         spin_unlock(&vn->sock_lock);
2656 }
2657
2658 /* Initialize the device structure. */
2659 static void vxlan_setup(struct net_device *dev)
2660 {
2661         struct vxlan_dev *vxlan = netdev_priv(dev);
2662         unsigned int h;
2663
2664         eth_hw_addr_random(dev);
2665         ether_setup(dev);
2666
2667         dev->needs_free_netdev = true;
2668         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2669
2670         dev->features   |= NETIF_F_LLTX;
2671         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2672         dev->features   |= NETIF_F_RXCSUM;
2673         dev->features   |= NETIF_F_GSO_SOFTWARE;
2674
2675         dev->vlan_features = dev->features;
2676         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2677         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2678         netif_keep_dst(dev);
2679         dev->priv_flags |= IFF_NO_QUEUE;
2680
2681         /* MTU range: 68 - 65535 */
2682         dev->min_mtu = ETH_MIN_MTU;
2683         dev->max_mtu = ETH_MAX_MTU;
2684
2685         INIT_LIST_HEAD(&vxlan->next);
2686         spin_lock_init(&vxlan->hash_lock);
2687
2688         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
2689
2690         vxlan->dev = dev;
2691
2692         gro_cells_init(&vxlan->gro_cells, dev);
2693
2694         for (h = 0; h < FDB_HASH_SIZE; ++h)
2695                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2696 }
2697
2698 static void vxlan_ether_setup(struct net_device *dev)
2699 {
2700         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2701         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2702         dev->netdev_ops = &vxlan_netdev_ether_ops;
2703 }
2704
2705 static void vxlan_raw_setup(struct net_device *dev)
2706 {
2707         dev->header_ops = NULL;
2708         dev->type = ARPHRD_NONE;
2709         dev->hard_header_len = 0;
2710         dev->addr_len = 0;
2711         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2712         dev->netdev_ops = &vxlan_netdev_raw_ops;
2713 }
2714
2715 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2716         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2717         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2718         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2719         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2720         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2721         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2722         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2723         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2724         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2725         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2726         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2727         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2728         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2729         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2730         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2731         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2732         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2733         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2734         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2735         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2736         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2737         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2738         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2739         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2740         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2741         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2742         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2743         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
2744 };
2745
2746 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
2747                           struct netlink_ext_ack *extack)
2748 {
2749         if (tb[IFLA_ADDRESS]) {
2750                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2751                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2752                                             "Provided link layer address is not Ethernet");
2753                         return -EINVAL;
2754                 }
2755
2756                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2757                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2758                                             "Provided Ethernet address is not unicast");
2759                         return -EADDRNOTAVAIL;
2760                 }
2761         }
2762
2763         if (tb[IFLA_MTU]) {
2764                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
2765
2766                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
2767                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
2768                                             "MTU must be between 68 and 65535");
2769                         return -EINVAL;
2770                 }
2771         }
2772
2773         if (!data) {
2774                 NL_SET_ERR_MSG(extack,
2775                                "Required attributes not provided to perform the operation");
2776                 return -EINVAL;
2777         }
2778
2779         if (data[IFLA_VXLAN_ID]) {
2780                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2781
2782                 if (id >= VXLAN_N_VID) {
2783                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID],
2784                                             "VXLAN ID must be lower than 16777216");
2785                         return -ERANGE;
2786                 }
2787         }
2788
2789         if (data[IFLA_VXLAN_PORT_RANGE]) {
2790                 const struct ifla_vxlan_port_range *p
2791                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2792
2793                 if (ntohs(p->high) < ntohs(p->low)) {
2794                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
2795                                             "Invalid source port range");
2796                         return -EINVAL;
2797                 }
2798         }
2799
2800         return 0;
2801 }
2802
2803 static void vxlan_get_drvinfo(struct net_device *netdev,
2804                               struct ethtool_drvinfo *drvinfo)
2805 {
2806         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2807         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2808 }
2809
2810 static const struct ethtool_ops vxlan_ethtool_ops = {
2811         .get_drvinfo    = vxlan_get_drvinfo,
2812         .get_link       = ethtool_op_get_link,
2813 };
2814
2815 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2816                                         __be16 port, u32 flags)
2817 {
2818         struct socket *sock;
2819         struct udp_port_cfg udp_conf;
2820         int err;
2821
2822         memset(&udp_conf, 0, sizeof(udp_conf));
2823
2824         if (ipv6) {
2825                 udp_conf.family = AF_INET6;
2826                 udp_conf.use_udp6_rx_checksums =
2827                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2828                 udp_conf.ipv6_v6only = 1;
2829         } else {
2830                 udp_conf.family = AF_INET;
2831         }
2832
2833         udp_conf.local_udp_port = port;
2834
2835         /* Open UDP socket */
2836         err = udp_sock_create(net, &udp_conf, &sock);
2837         if (err < 0)
2838                 return ERR_PTR(err);
2839
2840         return sock;
2841 }
2842
2843 /* Create new listen socket if needed */
2844 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2845                                               __be16 port, u32 flags)
2846 {
2847         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2848         struct vxlan_sock *vs;
2849         struct socket *sock;
2850         unsigned int h;
2851         struct udp_tunnel_sock_cfg tunnel_cfg;
2852
2853         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2854         if (!vs)
2855                 return ERR_PTR(-ENOMEM);
2856
2857         for (h = 0; h < VNI_HASH_SIZE; ++h)
2858                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2859
2860         sock = vxlan_create_sock(net, ipv6, port, flags);
2861         if (IS_ERR(sock)) {
2862                 kfree(vs);
2863                 return ERR_CAST(sock);
2864         }
2865
2866         vs->sock = sock;
2867         refcount_set(&vs->refcnt, 1);
2868         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2869
2870         spin_lock(&vn->sock_lock);
2871         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2872         udp_tunnel_notify_add_rx_port(sock,
2873                                       (vs->flags & VXLAN_F_GPE) ?
2874                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2875                                       UDP_TUNNEL_TYPE_VXLAN);
2876         spin_unlock(&vn->sock_lock);
2877
2878         /* Mark socket as an encapsulation socket. */
2879         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2880         tunnel_cfg.sk_user_data = vs;
2881         tunnel_cfg.encap_type = 1;
2882         tunnel_cfg.encap_rcv = vxlan_rcv;
2883         tunnel_cfg.encap_destroy = NULL;
2884         tunnel_cfg.gro_receive = vxlan_gro_receive;
2885         tunnel_cfg.gro_complete = vxlan_gro_complete;
2886
2887         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2888
2889         return vs;
2890 }
2891
2892 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2893 {
2894         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2895         struct vxlan_sock *vs = NULL;
2896         struct vxlan_dev_node *node;
2897
2898         if (!vxlan->cfg.no_share) {
2899                 spin_lock(&vn->sock_lock);
2900                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2901                                      vxlan->cfg.dst_port, vxlan->cfg.flags);
2902                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
2903                         spin_unlock(&vn->sock_lock);
2904                         return -EBUSY;
2905                 }
2906                 spin_unlock(&vn->sock_lock);
2907         }
2908         if (!vs)
2909                 vs = vxlan_socket_create(vxlan->net, ipv6,
2910                                          vxlan->cfg.dst_port, vxlan->cfg.flags);
2911         if (IS_ERR(vs))
2912                 return PTR_ERR(vs);
2913 #if IS_ENABLED(CONFIG_IPV6)
2914         if (ipv6) {
2915                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2916                 node = &vxlan->hlist6;
2917         } else
2918 #endif
2919         {
2920                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2921                 node = &vxlan->hlist4;
2922         }
2923         vxlan_vs_add_dev(vs, vxlan, node);
2924         return 0;
2925 }
2926
2927 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2928 {
2929         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
2930         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
2931         bool ipv4 = !ipv6 || metadata;
2932         int ret = 0;
2933
2934         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2935 #if IS_ENABLED(CONFIG_IPV6)
2936         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2937         if (ipv6) {
2938                 ret = __vxlan_sock_add(vxlan, true);
2939                 if (ret < 0 && ret != -EAFNOSUPPORT)
2940                         ipv4 = false;
2941         }
2942 #endif
2943         if (ipv4)
2944                 ret = __vxlan_sock_add(vxlan, false);
2945         if (ret < 0)
2946                 vxlan_sock_release(vxlan);
2947         return ret;
2948 }
2949
2950 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
2951                                  struct net_device **lower,
2952                                  struct vxlan_dev *old,
2953                                  struct netlink_ext_ack *extack)
2954 {
2955         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2956         struct vxlan_dev *tmp;
2957         bool use_ipv6 = false;
2958
2959         if (conf->flags & VXLAN_F_GPE) {
2960                 /* For now, allow GPE only together with
2961                  * COLLECT_METADATA. This can be relaxed later; in such
2962                  * case, the other side of the PtP link will have to be
2963                  * provided.
2964                  */
2965                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2966                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2967                         NL_SET_ERR_MSG(extack,
2968                                        "VXLAN GPE does not support this combination of attributes");
2969                         return -EINVAL;
2970                 }
2971         }
2972
2973         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
2974                 /* Unless IPv6 is explicitly requested, assume IPv4 */
2975                 conf->remote_ip.sa.sa_family = AF_INET;
2976                 conf->saddr.sa.sa_family = AF_INET;
2977         } else if (!conf->remote_ip.sa.sa_family) {
2978                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
2979         } else if (!conf->saddr.sa.sa_family) {
2980                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
2981         }
2982
2983         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
2984                 NL_SET_ERR_MSG(extack,
2985                                "Local and remote address must be from the same family");
2986                 return -EINVAL;
2987         }
2988
2989         if (vxlan_addr_multicast(&conf->saddr)) {
2990                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
2991                 return -EINVAL;
2992         }
2993
2994         if (conf->saddr.sa.sa_family == AF_INET6) {
2995                 if (!IS_ENABLED(CONFIG_IPV6)) {
2996                         NL_SET_ERR_MSG(extack,
2997                                        "IPv6 support not enabled in the kernel");
2998                         return -EPFNOSUPPORT;
2999                 }
3000                 use_ipv6 = true;
3001                 conf->flags |= VXLAN_F_IPV6;
3002
3003                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3004                         int local_type =
3005                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3006                         int remote_type =
3007                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3008
3009                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3010                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3011                                     (remote_type != IPV6_ADDR_ANY)) {
3012                                         NL_SET_ERR_MSG(extack,
3013                                                        "Invalid combination of local and remote address scopes");
3014                                         return -EINVAL;
3015                                 }
3016
3017                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3018                         } else {
3019                                 if (remote_type ==
3020                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3021                                         NL_SET_ERR_MSG(extack,
3022                                                        "Invalid combination of local and remote address scopes");
3023                                         return -EINVAL;
3024                                 }
3025
3026                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3027                         }
3028                 }
3029         }
3030
3031         if (conf->label && !use_ipv6) {
3032                 NL_SET_ERR_MSG(extack,
3033                                "Label attribute only applies to IPv6 VXLAN devices");
3034                 return -EINVAL;
3035         }
3036
3037         if (conf->remote_ifindex) {
3038                 struct net_device *lowerdev;
3039
3040                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3041                 if (!lowerdev) {
3042                         NL_SET_ERR_MSG(extack,
3043                                        "Invalid local interface, device not found");
3044                         return -ENODEV;
3045                 }
3046
3047 #if IS_ENABLED(CONFIG_IPV6)
3048                 if (use_ipv6) {
3049                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3050                         if (idev && idev->cnf.disable_ipv6) {
3051                                 NL_SET_ERR_MSG(extack,
3052                                                "IPv6 support disabled by administrator");
3053                                 return -EPERM;
3054                         }
3055                 }
3056 #endif
3057
3058                 *lower = lowerdev;
3059         } else {
3060                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3061                         NL_SET_ERR_MSG(extack,
3062                                        "Local interface required for multicast remote destination");
3063
3064                         return -EINVAL;
3065                 }
3066
3067 #if IS_ENABLED(CONFIG_IPV6)
3068                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3069                         NL_SET_ERR_MSG(extack,
3070                                        "Local interface required for link-local local/remote addresses");
3071                         return -EINVAL;
3072                 }
3073 #endif
3074
3075                 *lower = NULL;
3076         }
3077
3078         if (!conf->dst_port) {
3079                 if (conf->flags & VXLAN_F_GPE)
3080                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3081                 else
3082                         conf->dst_port = htons(vxlan_port);
3083         }
3084
3085         if (!conf->age_interval)
3086                 conf->age_interval = FDB_AGE_DEFAULT;
3087
3088         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3089                 if (tmp == old)
3090                         continue;
3091
3092                 if (tmp->cfg.vni != conf->vni)
3093                         continue;
3094                 if (tmp->cfg.dst_port != conf->dst_port)
3095                         continue;
3096                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3097                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3098                         continue;
3099
3100                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3101                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3102                         continue;
3103
3104                 NL_SET_ERR_MSG(extack,
3105                                "A VXLAN device with the specified VNI already exists");
3106                 return -EEXIST;
3107         }
3108
3109         return 0;
3110 }
3111
3112 static void vxlan_config_apply(struct net_device *dev,
3113                                struct vxlan_config *conf,
3114                                struct net_device *lowerdev,
3115                                struct net *src_net,
3116                                bool changelink)
3117 {
3118         struct vxlan_dev *vxlan = netdev_priv(dev);
3119         struct vxlan_rdst *dst = &vxlan->default_dst;
3120         unsigned short needed_headroom = ETH_HLEN;
3121         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3122         int max_mtu = ETH_MAX_MTU;
3123
3124         if (!changelink) {
3125                 if (conf->flags & VXLAN_F_GPE)
3126                         vxlan_raw_setup(dev);
3127                 else
3128                         vxlan_ether_setup(dev);
3129
3130                 if (conf->mtu)
3131                         dev->mtu = conf->mtu;
3132
3133                 vxlan->net = src_net;
3134         }
3135
3136         dst->remote_vni = conf->vni;
3137
3138         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3139
3140         if (lowerdev) {
3141                 dst->remote_ifindex = conf->remote_ifindex;
3142
3143                 dev->gso_max_size = lowerdev->gso_max_size;
3144                 dev->gso_max_segs = lowerdev->gso_max_segs;
3145
3146                 needed_headroom = lowerdev->hard_header_len;
3147
3148                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3149                                            VXLAN_HEADROOM);
3150                 if (max_mtu < ETH_MIN_MTU)
3151                         max_mtu = ETH_MIN_MTU;
3152
3153                 if (!changelink && !conf->mtu)
3154                         dev->mtu = max_mtu;
3155         }
3156
3157         if (dev->mtu > max_mtu)
3158                 dev->mtu = max_mtu;
3159
3160         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3161                 needed_headroom += VXLAN6_HEADROOM;
3162         else
3163                 needed_headroom += VXLAN_HEADROOM;
3164         dev->needed_headroom = needed_headroom;
3165
3166         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3167 }
3168
3169 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3170                                struct vxlan_config *conf, bool changelink,
3171                                struct netlink_ext_ack *extack)
3172 {
3173         struct vxlan_dev *vxlan = netdev_priv(dev);
3174         struct net_device *lowerdev;
3175         int ret;
3176
3177         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3178         if (ret)
3179                 return ret;
3180
3181         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3182
3183         return 0;
3184 }
3185
3186 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3187                               struct vxlan_config *conf,
3188                               struct netlink_ext_ack *extack)
3189 {
3190         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3191         struct vxlan_dev *vxlan = netdev_priv(dev);
3192         struct vxlan_fdb *f = NULL;
3193         int err;
3194
3195         err = vxlan_dev_configure(net, dev, conf, false, extack);
3196         if (err)
3197                 return err;
3198
3199         dev->ethtool_ops = &vxlan_ethtool_ops;
3200
3201         /* create an fdb entry for a valid default destination */
3202         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
3203                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3204                                        &vxlan->default_dst.remote_ip,
3205                                        NUD_REACHABLE | NUD_PERMANENT,
3206                                        vxlan->cfg.dst_port,
3207                                        vxlan->default_dst.remote_vni,
3208                                        vxlan->default_dst.remote_vni,
3209                                        vxlan->default_dst.remote_ifindex,
3210                                        NTF_SELF, &f);
3211                 if (err)
3212                         return err;
3213         }
3214
3215         err = register_netdevice(dev);
3216         if (err)
3217                 goto errout;
3218
3219         err = rtnl_configure_link(dev, NULL);
3220         if (err) {
3221                 unregister_netdevice(dev);
3222                 goto errout;
3223         }
3224
3225         /* notify default fdb entry */
3226         if (f)
3227                 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH);
3228
3229         list_add(&vxlan->next, &vn->vxlan_list);
3230         return 0;
3231 errout:
3232         if (f)
3233                 vxlan_fdb_destroy(vxlan, f, false);
3234         return err;
3235 }
3236
3237 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3238                          struct net_device *dev, struct vxlan_config *conf,
3239                          bool changelink)
3240 {
3241         struct vxlan_dev *vxlan = netdev_priv(dev);
3242
3243         memset(conf, 0, sizeof(*conf));
3244
3245         /* if changelink operation, start with old existing cfg */
3246         if (changelink)
3247                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3248
3249         if (data[IFLA_VXLAN_ID]) {
3250                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3251
3252                 if (changelink && (vni != conf->vni))
3253                         return -EOPNOTSUPP;
3254                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3255         }
3256
3257         if (data[IFLA_VXLAN_GROUP]) {
3258                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET))
3259                         return -EOPNOTSUPP;
3260
3261                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3262                 conf->remote_ip.sa.sa_family = AF_INET;
3263         } else if (data[IFLA_VXLAN_GROUP6]) {
3264                 if (!IS_ENABLED(CONFIG_IPV6))
3265                         return -EPFNOSUPPORT;
3266
3267                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6))
3268                         return -EOPNOTSUPP;
3269
3270                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3271                 conf->remote_ip.sa.sa_family = AF_INET6;
3272         }
3273
3274         if (data[IFLA_VXLAN_LOCAL]) {
3275                 if (changelink && (conf->saddr.sa.sa_family != AF_INET))
3276                         return -EOPNOTSUPP;
3277
3278                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3279                 conf->saddr.sa.sa_family = AF_INET;
3280         } else if (data[IFLA_VXLAN_LOCAL6]) {
3281                 if (!IS_ENABLED(CONFIG_IPV6))
3282                         return -EPFNOSUPPORT;
3283
3284                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6))
3285                         return -EOPNOTSUPP;
3286
3287                 /* TODO: respect scope id */
3288                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3289                 conf->saddr.sa.sa_family = AF_INET6;
3290         }
3291
3292         if (data[IFLA_VXLAN_LINK])
3293                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3294
3295         if (data[IFLA_VXLAN_TOS])
3296                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3297
3298         if (data[IFLA_VXLAN_TTL])
3299                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3300
3301         if (data[IFLA_VXLAN_TTL_INHERIT]) {
3302                 if (changelink)
3303                         return -EOPNOTSUPP;
3304                 conf->flags |= VXLAN_F_TTL_INHERIT;
3305         }
3306
3307         if (data[IFLA_VXLAN_LABEL])
3308                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3309                              IPV6_FLOWLABEL_MASK;
3310
3311         if (data[IFLA_VXLAN_LEARNING]) {
3312                 if (nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3313                         conf->flags |= VXLAN_F_LEARN;
3314                 else
3315                         conf->flags &= ~VXLAN_F_LEARN;
3316         } else if (!changelink) {
3317                 /* default to learn on a new device */
3318                 conf->flags |= VXLAN_F_LEARN;
3319         }
3320
3321         if (data[IFLA_VXLAN_AGEING]) {
3322                 if (changelink)
3323                         return -EOPNOTSUPP;
3324                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3325         }
3326
3327         if (data[IFLA_VXLAN_PROXY]) {
3328                 if (changelink)
3329                         return -EOPNOTSUPP;
3330                 if (nla_get_u8(data[IFLA_VXLAN_PROXY]))
3331                         conf->flags |= VXLAN_F_PROXY;
3332         }
3333
3334         if (data[IFLA_VXLAN_RSC]) {
3335                 if (changelink)
3336                         return -EOPNOTSUPP;
3337                 if (nla_get_u8(data[IFLA_VXLAN_RSC]))
3338                         conf->flags |= VXLAN_F_RSC;
3339         }
3340
3341         if (data[IFLA_VXLAN_L2MISS]) {
3342                 if (changelink)
3343                         return -EOPNOTSUPP;
3344                 if (nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3345                         conf->flags |= VXLAN_F_L2MISS;
3346         }
3347
3348         if (data[IFLA_VXLAN_L3MISS]) {
3349                 if (changelink)
3350                         return -EOPNOTSUPP;
3351                 if (nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3352                         conf->flags |= VXLAN_F_L3MISS;
3353         }
3354
3355         if (data[IFLA_VXLAN_LIMIT]) {
3356                 if (changelink)
3357                         return -EOPNOTSUPP;
3358                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3359         }
3360
3361         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3362                 if (changelink)
3363                         return -EOPNOTSUPP;
3364                 if (nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3365                         conf->flags |= VXLAN_F_COLLECT_METADATA;
3366         }
3367
3368         if (data[IFLA_VXLAN_PORT_RANGE]) {
3369                 if (!changelink) {
3370                         const struct ifla_vxlan_port_range *p
3371                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3372                         conf->port_min = ntohs(p->low);
3373                         conf->port_max = ntohs(p->high);
3374                 } else {
3375                         return -EOPNOTSUPP;
3376                 }
3377         }
3378
3379         if (data[IFLA_VXLAN_PORT]) {
3380                 if (changelink)
3381                         return -EOPNOTSUPP;
3382                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3383         }
3384
3385         if (data[IFLA_VXLAN_UDP_CSUM]) {
3386                 if (changelink)
3387                         return -EOPNOTSUPP;
3388                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3389                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3390         }
3391
3392         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3393                 if (changelink)
3394                         return -EOPNOTSUPP;
3395                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3396                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3397         }
3398
3399         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3400                 if (changelink)
3401                         return -EOPNOTSUPP;
3402                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3403                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3404         }
3405
3406         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3407                 if (changelink)
3408                         return -EOPNOTSUPP;
3409                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3410                         conf->flags |= VXLAN_F_REMCSUM_TX;
3411         }
3412
3413         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3414                 if (changelink)
3415                         return -EOPNOTSUPP;
3416                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3417                         conf->flags |= VXLAN_F_REMCSUM_RX;
3418         }
3419
3420         if (data[IFLA_VXLAN_GBP]) {
3421                 if (changelink)
3422                         return -EOPNOTSUPP;
3423                 conf->flags |= VXLAN_F_GBP;
3424         }
3425
3426         if (data[IFLA_VXLAN_GPE]) {
3427                 if (changelink)
3428                         return -EOPNOTSUPP;
3429                 conf->flags |= VXLAN_F_GPE;
3430         }
3431
3432         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3433                 if (changelink)
3434                         return -EOPNOTSUPP;
3435                 conf->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3436         }
3437
3438         if (tb[IFLA_MTU]) {
3439                 if (changelink)
3440                         return -EOPNOTSUPP;
3441                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3442         }
3443
3444         return 0;
3445 }
3446
3447 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3448                          struct nlattr *tb[], struct nlattr *data[],
3449                          struct netlink_ext_ack *extack)
3450 {
3451         struct vxlan_config conf;
3452         int err;
3453
3454         err = vxlan_nl2conf(tb, data, dev, &conf, false);
3455         if (err)
3456                 return err;
3457
3458         return __vxlan_dev_create(src_net, dev, &conf, extack);
3459 }
3460
3461 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3462                             struct nlattr *data[],
3463                             struct netlink_ext_ack *extack)
3464 {
3465         struct vxlan_dev *vxlan = netdev_priv(dev);
3466         struct vxlan_rdst *dst = &vxlan->default_dst;
3467         struct vxlan_rdst old_dst;
3468         struct vxlan_config conf;
3469         struct vxlan_fdb *f = NULL;
3470         int err;
3471
3472         err = vxlan_nl2conf(tb, data,
3473                             dev, &conf, true);
3474         if (err)
3475                 return err;
3476
3477         memcpy(&old_dst, dst, sizeof(struct vxlan_rdst));
3478
3479         err = vxlan_dev_configure(vxlan->net, dev, &conf, true, extack);
3480         if (err)
3481                 return err;
3482
3483         /* handle default dst entry */
3484         if (!vxlan_addr_equal(&dst->remote_ip, &old_dst.remote_ip)) {
3485                 spin_lock_bh(&vxlan->hash_lock);
3486                 if (!vxlan_addr_any(&old_dst.remote_ip))
3487                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
3488                                            old_dst.remote_ip,
3489                                            vxlan->cfg.dst_port,
3490                                            old_dst.remote_vni,
3491                                            old_dst.remote_vni,
3492                                            old_dst.remote_ifindex, 0);
3493
3494                 if (!vxlan_addr_any(&dst->remote_ip)) {
3495                         err = vxlan_fdb_create(vxlan, all_zeros_mac,
3496                                                &dst->remote_ip,
3497                                                NUD_REACHABLE | NUD_PERMANENT,
3498                                                vxlan->cfg.dst_port,
3499                                                dst->remote_vni,
3500                                                dst->remote_vni,
3501                                                dst->remote_ifindex,
3502                                                NTF_SELF, &f);
3503                         if (err) {
3504                                 spin_unlock_bh(&vxlan->hash_lock);
3505                                 return err;
3506                         }
3507                         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH);
3508                 }
3509                 spin_unlock_bh(&vxlan->hash_lock);
3510         }
3511
3512         return 0;
3513 }
3514
3515 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3516 {
3517         struct vxlan_dev *vxlan = netdev_priv(dev);
3518
3519         vxlan_flush(vxlan, true);
3520
3521         gro_cells_destroy(&vxlan->gro_cells);
3522         list_del(&vxlan->next);
3523         unregister_netdevice_queue(dev, head);
3524 }
3525
3526 static size_t vxlan_get_size(const struct net_device *dev)
3527 {
3528
3529         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3530                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3531                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3532                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3533                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3534                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL_INHERIT */
3535                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3536                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3537                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3538                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3539                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3540                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3541                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3542                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3543                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3544                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3545                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3546                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3547                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3548                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3549                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3550                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3551                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3552                 0;
3553 }
3554
3555 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3556 {
3557         const struct vxlan_dev *vxlan = netdev_priv(dev);
3558         const struct vxlan_rdst *dst = &vxlan->default_dst;
3559         struct ifla_vxlan_port_range ports = {
3560                 .low =  htons(vxlan->cfg.port_min),
3561                 .high = htons(vxlan->cfg.port_max),
3562         };
3563
3564         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3565                 goto nla_put_failure;
3566
3567         if (!vxlan_addr_any(&dst->remote_ip)) {
3568                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3569                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3570                                             dst->remote_ip.sin.sin_addr.s_addr))
3571                                 goto nla_put_failure;
3572 #if IS_ENABLED(CONFIG_IPV6)
3573                 } else {
3574                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3575                                              &dst->remote_ip.sin6.sin6_addr))
3576                                 goto nla_put_failure;
3577 #endif
3578                 }
3579         }
3580
3581         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3582                 goto nla_put_failure;
3583
3584         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3585                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3586                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3587                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3588                                 goto nla_put_failure;
3589 #if IS_ENABLED(CONFIG_IPV6)
3590                 } else {
3591                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3592                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3593                                 goto nla_put_failure;
3594 #endif
3595                 }
3596         }
3597
3598         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3599             nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT,
3600                        !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) ||
3601             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3602             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3603             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3604                         !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
3605             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3606                         !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
3607             nla_put_u8(skb, IFLA_VXLAN_RSC,
3608                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
3609             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3610                         !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
3611             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3612                         !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
3613             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3614                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
3615             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3616             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3617             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3618             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3619                         !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3620             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3621                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3622             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3623                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3624             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3625                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
3626             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3627                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
3628                 goto nla_put_failure;
3629
3630         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3631                 goto nla_put_failure;
3632
3633         if (vxlan->cfg.flags & VXLAN_F_GBP &&
3634             nla_put_flag(skb, IFLA_VXLAN_GBP))
3635                 goto nla_put_failure;
3636
3637         if (vxlan->cfg.flags & VXLAN_F_GPE &&
3638             nla_put_flag(skb, IFLA_VXLAN_GPE))
3639                 goto nla_put_failure;
3640
3641         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3642             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3643                 goto nla_put_failure;
3644
3645         return 0;
3646
3647 nla_put_failure:
3648         return -EMSGSIZE;
3649 }
3650
3651 static struct net *vxlan_get_link_net(const struct net_device *dev)
3652 {
3653         struct vxlan_dev *vxlan = netdev_priv(dev);
3654
3655         return vxlan->net;
3656 }
3657
3658 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3659         .kind           = "vxlan",
3660         .maxtype        = IFLA_VXLAN_MAX,
3661         .policy         = vxlan_policy,
3662         .priv_size      = sizeof(struct vxlan_dev),
3663         .setup          = vxlan_setup,
3664         .validate       = vxlan_validate,
3665         .newlink        = vxlan_newlink,
3666         .changelink     = vxlan_changelink,
3667         .dellink        = vxlan_dellink,
3668         .get_size       = vxlan_get_size,
3669         .fill_info      = vxlan_fill_info,
3670         .get_link_net   = vxlan_get_link_net,
3671 };
3672
3673 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3674                                     u8 name_assign_type,
3675                                     struct vxlan_config *conf)
3676 {
3677         struct nlattr *tb[IFLA_MAX + 1];
3678         struct net_device *dev;
3679         int err;
3680
3681         memset(&tb, 0, sizeof(tb));
3682
3683         dev = rtnl_create_link(net, name, name_assign_type,
3684                                &vxlan_link_ops, tb);
3685         if (IS_ERR(dev))
3686                 return dev;
3687
3688         err = __vxlan_dev_create(net, dev, conf, NULL);
3689         if (err < 0) {
3690                 free_netdev(dev);
3691                 return ERR_PTR(err);
3692         }
3693
3694         err = rtnl_configure_link(dev, NULL);
3695         if (err < 0) {
3696                 LIST_HEAD(list_kill);
3697
3698                 vxlan_dellink(dev, &list_kill);
3699                 unregister_netdevice_many(&list_kill);
3700                 return ERR_PTR(err);
3701         }
3702
3703         return dev;
3704 }
3705 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3706
3707 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3708                                              struct net_device *dev)
3709 {
3710         struct vxlan_dev *vxlan, *next;
3711         LIST_HEAD(list_kill);
3712
3713         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3714                 struct vxlan_rdst *dst = &vxlan->default_dst;
3715
3716                 /* In case we created vxlan device with carrier
3717                  * and we loose the carrier due to module unload
3718                  * we also need to remove vxlan device. In other
3719                  * cases, it's not necessary and remote_ifindex
3720                  * is 0 here, so no matches.
3721                  */
3722                 if (dst->remote_ifindex == dev->ifindex)
3723                         vxlan_dellink(vxlan->dev, &list_kill);
3724         }
3725
3726         unregister_netdevice_many(&list_kill);
3727 }
3728
3729 static int vxlan_netdevice_event(struct notifier_block *unused,
3730                                  unsigned long event, void *ptr)
3731 {
3732         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3733         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3734
3735         if (event == NETDEV_UNREGISTER) {
3736                 vxlan_offload_rx_ports(dev, false);
3737                 vxlan_handle_lowerdev_unregister(vn, dev);
3738         } else if (event == NETDEV_REGISTER) {
3739                 vxlan_offload_rx_ports(dev, true);
3740         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
3741                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
3742                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
3743         }
3744
3745         return NOTIFY_DONE;
3746 }
3747
3748 static struct notifier_block vxlan_notifier_block __read_mostly = {
3749         .notifier_call = vxlan_netdevice_event,
3750 };
3751
3752 static __net_init int vxlan_init_net(struct net *net)
3753 {
3754         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3755         unsigned int h;
3756
3757         INIT_LIST_HEAD(&vn->vxlan_list);
3758         spin_lock_init(&vn->sock_lock);
3759
3760         for (h = 0; h < PORT_HASH_SIZE; ++h)
3761                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3762
3763         return 0;
3764 }
3765
3766 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
3767 {
3768         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3769         struct vxlan_dev *vxlan, *next;
3770         struct net_device *dev, *aux;
3771         unsigned int h;
3772
3773         for_each_netdev_safe(net, dev, aux)
3774                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3775                         unregister_netdevice_queue(dev, head);
3776
3777         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3778                 /* If vxlan->dev is in the same netns, it has already been added
3779                  * to the list by the previous loop.
3780                  */
3781                 if (!net_eq(dev_net(vxlan->dev), net)) {
3782                         gro_cells_destroy(&vxlan->gro_cells);
3783                         unregister_netdevice_queue(vxlan->dev, head);
3784                 }
3785         }
3786
3787         for (h = 0; h < PORT_HASH_SIZE; ++h)
3788                 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
3789 }
3790
3791 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
3792 {
3793         struct net *net;
3794         LIST_HEAD(list);
3795
3796         rtnl_lock();
3797         list_for_each_entry(net, net_list, exit_list)
3798                 vxlan_destroy_tunnels(net, &list);
3799
3800         unregister_netdevice_many(&list);
3801         rtnl_unlock();
3802 }
3803
3804 static struct pernet_operations vxlan_net_ops = {
3805         .init = vxlan_init_net,
3806         .exit_batch = vxlan_exit_batch_net,
3807         .id   = &vxlan_net_id,
3808         .size = sizeof(struct vxlan_net),
3809 };
3810
3811 static int __init vxlan_init_module(void)
3812 {
3813         int rc;
3814
3815         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3816
3817         rc = register_pernet_subsys(&vxlan_net_ops);
3818         if (rc)
3819                 goto out1;
3820
3821         rc = register_netdevice_notifier(&vxlan_notifier_block);
3822         if (rc)
3823                 goto out2;
3824
3825         rc = rtnl_link_register(&vxlan_link_ops);
3826         if (rc)
3827                 goto out3;
3828
3829         return 0;
3830 out3:
3831         unregister_netdevice_notifier(&vxlan_notifier_block);
3832 out2:
3833         unregister_pernet_subsys(&vxlan_net_ops);
3834 out1:
3835         return rc;
3836 }
3837 late_initcall(vxlan_init_module);
3838
3839 static void __exit vxlan_cleanup_module(void)
3840 {
3841         rtnl_link_unregister(&vxlan_link_ops);
3842         unregister_netdevice_notifier(&vxlan_notifier_block);
3843         unregister_pernet_subsys(&vxlan_net_ops);
3844         /* rcu_barrier() is called by netns */
3845 }
3846 module_exit(vxlan_cleanup_module);
3847
3848 MODULE_LICENSE("GPL");
3849 MODULE_VERSION(VXLAN_VERSION);
3850 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3851 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3852 MODULE_ALIAS_RTNL_LINK("vxlan");