drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/errno.h>
  16 #include <linux/slab.h>
  17 #include <linux/udp.h>
  18 #include <linux/igmp.h>
  19 #include <linux/if_ether.h>
  20 #include <linux/ethtool.h>
  21 #include <net/arp.h>
  22 #include <net/ndisc.h>
  23 #include <net/ip.h>
  24 #include <net/icmp.h>
  25 #include <net/rtnetlink.h>
  26 #include <net/inet_ecn.h>
  27 #include <net/net_namespace.h>
  28 #include <net/netns/generic.h>
  29 #include <net/tun_proto.h>
  30 #include <net/vxlan.h>
  31
  32 #if IS_ENABLED(CONFIG_IPV6)
  33 #include <net/ip6_tunnel.h>
  34 #include <net/ip6_checksum.h>
  35 #endif
  36
  37 #define VXLAN_VERSION   "0.1"
  38
  39 #define PORT_HASH_BITS  8
  40 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  41 #define FDB_AGE_DEFAULT 300 /* 5 min */
  42 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  43
  44 /* UDP port for VXLAN traffic.
  45  * The IANA assigned port is 4789, but the Linux default is 8472
  46  * for compatibility with early adopters.
  47  */
  48 static unsigned short vxlan_port __read_mostly = 8472;
  49 module_param_named(udp_port, vxlan_port, ushort, 0444);
  50 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  51
  52 static bool log_ecn_error = true;
  53 module_param(log_ecn_error, bool, 0644);
  54 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  55
  56 static unsigned int vxlan_net_id;
  57 static struct rtnl_link_ops vxlan_link_ops;
  58
  59 static const u8 all_zeros_mac[ETH_ALEN + 2];
  60
  61 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  62
  63 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
  64
  65 /* per-network namespace private data for this module */
  66 struct vxlan_net {
  67         struct list_head  vxlan_list;
  68         struct hlist_head sock_list[PORT_HASH_SIZE];
  69         spinlock_t        sock_lock;
  70 };
  71
  72 /* Forwarding table entry */
  73 struct vxlan_fdb {
  74         struct hlist_node hlist;        /* linked list of entries */
  75         struct rcu_head   rcu;
  76         unsigned long     updated;      /* jiffies */
  77         unsigned long     used;
  78         struct list_head  remotes;
  79         u8                eth_addr[ETH_ALEN];
  80         u16               state;        /* see ndm_state */
  81         __be32            vni;
  82         u8                flags;        /* see ndm_flags */
  83 };
  84
  85 /* salt for hash table */
  86 static u32 vxlan_salt __read_mostly;
  87
  88 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  89 {
  90         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  91                ip_tunnel_collect_metadata();
  92 }
  93
  94 #if IS_ENABLED(CONFIG_IPV6)
  95 static inline
  96 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
  97 {
  98         if (a->sa.sa_family != b->sa.sa_family)
  99                 return false;
 100         if (a->sa.sa_family == AF_INET6)
 101                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 102         else
 103                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 104 }
 105
 106 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 107 {
 108         if (ipa->sa.sa_family == AF_INET6)
 109                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 110         else
 111                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 112 }
 113
 114 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 115 {
 116         if (ipa->sa.sa_family == AF_INET6)
 117                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 118         else
 119                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 120 }
 121
 122 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 123 {
 124         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 125                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 126                 ip->sa.sa_family = AF_INET6;
 127                 return 0;
 128         } else if (nla_len(nla) >= sizeof(__be32)) {
 129                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 130                 ip->sa.sa_family = AF_INET;
 131                 return 0;
 132         } else {
 133                 return -EAFNOSUPPORT;
 134         }
 135 }
 136
 137 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 138                               const union vxlan_addr *ip)
 139 {
 140         if (ip->sa.sa_family == AF_INET6)
 141                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 142         else
 143                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 144 }
 145
 146 #else /* !CONFIG_IPV6 */
 147
 148 static inline
 149 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 150 {
 151         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 152 }
 153
 154 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 155 {
 156         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 157 }
 158
 159 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 160 {
 161         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 162 }
 163
 164 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 165 {
 166         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 167                 return -EAFNOSUPPORT;
 168         } else if (nla_len(nla) >= sizeof(__be32)) {
 169                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 170                 ip->sa.sa_family = AF_INET;
 171                 return 0;
 172         } else {
 173                 return -EAFNOSUPPORT;
 174         }
 175 }
 176
 177 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 178                               const union vxlan_addr *ip)
 179 {
 180         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 181 }
 182 #endif
 183
 184 /* Virtual Network hash table head */
 185 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 186 {
 187         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 188 }
 189
 190 /* Socket hash table head */
 191 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 192 {
 193         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 194
 195         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 196 }
 197
 198 /* First remote destination for a forwarding entry.
 199  * Guaranteed to be non-NULL because remotes are never deleted.
 200  */
 201 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 202 {
 203         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 204 }
 205
 206 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 207 {
 208         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 209 }
 210
 211 /* Find VXLAN socket based on network namespace, address family and UDP port
 212  * and enabled unshareable flags.
 213  */
 214 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 215                                           __be16 port, u32 flags)
 216 {
 217         struct vxlan_sock *vs;
 218
 219         flags &= VXLAN_F_RCV_FLAGS;
 220
 221         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 222                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 223                     vxlan_get_sk_family(vs) == family &&
 224                     vs->flags == flags)
 225                         return vs;
 226         }
 227         return NULL;
 228 }
 229
 230 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, int ifindex,
 231                                            __be32 vni)
 232 {
 233         struct vxlan_dev_node *node;
 234
 235         /* For flow based devices, map all packets to VNI 0 */
 236         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 237                 vni = 0;
 238
 239         hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
 240                 if (node->vxlan->default_dst.remote_vni != vni)
 241                         continue;
 242
 243                 if (IS_ENABLED(CONFIG_IPV6)) {
 244                         const struct vxlan_config *cfg = &node->vxlan->cfg;
 245
 246                         if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) &&
 247                             cfg->remote_ifindex != ifindex)
 248                                 continue;
 249                 }
 250
 251                 return node->vxlan;
 252         }
 253
 254         return NULL;
 255 }
 256
 257 /* Look up VNI in a per net namespace table */
 258 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex,
 259                                         __be32 vni, sa_family_t family,
 260                                         __be16 port, u32 flags)
 261 {
 262         struct vxlan_sock *vs;
 263
 264         vs = vxlan_find_sock(net, family, port, flags);
 265         if (!vs)
 266                 return NULL;
 267
 268         return vxlan_vs_find_vni(vs, ifindex, vni);
 269 }
 270
 271 /* Fill in neighbour message in skbuff. */
 272 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 273                           const struct vxlan_fdb *fdb,
 274                           u32 portid, u32 seq, int type, unsigned int flags,
 275                           const struct vxlan_rdst *rdst)
 276 {
 277         unsigned long now = jiffies;
 278         struct nda_cacheinfo ci;
 279         struct nlmsghdr *nlh;
 280         struct ndmsg *ndm;
 281         bool send_ip, send_eth;
 282
 283         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 284         if (nlh == NULL)
 285                 return -EMSGSIZE;
 286
 287         ndm = nlmsg_data(nlh);
 288         memset(ndm, 0, sizeof(*ndm));
 289
 290         send_eth = send_ip = true;
 291
 292         if (type == RTM_GETNEIGH) {
 293                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 294                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 295                 ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
 296         } else
 297                 ndm->ndm_family = AF_BRIDGE;
 298         ndm->ndm_state = fdb->state;
 299         ndm->ndm_ifindex = vxlan->dev->ifindex;
 300         ndm->ndm_flags = fdb->flags;
 301         ndm->ndm_type = RTN_UNICAST;
 302
 303         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 304             nla_put_s32(skb, NDA_LINK_NETNSID,
 305                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 306                 goto nla_put_failure;
 307
 308         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 309                 goto nla_put_failure;
 310
 311         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 312                 goto nla_put_failure;
 313
 314         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 315             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 316                 goto nla_put_failure;
 317         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 318             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 319                 goto nla_put_failure;
 320         if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
 321             nla_put_u32(skb, NDA_SRC_VNI,
 322                         be32_to_cpu(fdb->vni)))
 323                 goto nla_put_failure;
 324         if (rdst->remote_ifindex &&
 325             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 326                 goto nla_put_failure;
 327
 328         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 329         ci.ndm_confirmed = 0;
 330         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 331         ci.ndm_refcnt    = 0;
 332
 333         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 334                 goto nla_put_failure;
 335
 336         nlmsg_end(skb, nlh);
 337         return 0;
 338
 339 nla_put_failure:
 340         nlmsg_cancel(skb, nlh);
 341         return -EMSGSIZE;
 342 }
 343
 344 static inline size_t vxlan_nlmsg_size(void)
 345 {
 346         return NLMSG_ALIGN(sizeof(struct ndmsg))
 347                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 348                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 349                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 350                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 351                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 352                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 353                 + nla_total_size(sizeof(struct nda_cacheinfo));
 354 }
 355
 356 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 357                              struct vxlan_rdst *rd, int type)
 358 {
 359         struct net *net = dev_net(vxlan->dev);
 360         struct sk_buff *skb;
 361         int err = -ENOBUFS;
 362
 363         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 364         if (skb == NULL)
 365                 goto errout;
 366
 367         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 368         if (err < 0) {
 369                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 370                 WARN_ON(err == -EMSGSIZE);
 371                 kfree_skb(skb);
 372                 goto errout;
 373         }
 374
 375         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 376         return;
 377 errout:
 378         if (err < 0)
 379                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 380 }
 381
 382 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 383 {
 384         struct vxlan_dev *vxlan = netdev_priv(dev);
 385         struct vxlan_fdb f = {
 386                 .state = NUD_STALE,
 387         };
 388         struct vxlan_rdst remote = {
 389                 .remote_ip = *ipa, /* goes to NDA_DST */
 390                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 391         };
 392
 393         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 394 }
 395
 396 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 397 {
 398         struct vxlan_fdb f = {
 399                 .state = NUD_STALE,
 400         };
 401         struct vxlan_rdst remote = { };
 402
 403         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 404
 405         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 406 }
 407
 408 /* Hash Ethernet address */
 409 static u32 eth_hash(const unsigned char *addr)
 410 {
 411         u64 value = get_unaligned((u64 *)addr);
 412
 413         /* only want 6 bytes */
 414 #ifdef __BIG_ENDIAN
 415         value >>= 16;
 416 #else
 417         value <<= 16;
 418 #endif
 419         return hash_64(value, FDB_HASH_BITS);
 420 }
 421
 422 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
 423 {
 424         /* use 1 byte of OUI and 3 bytes of NIC */
 425         u32 key = get_unaligned((u32 *)(addr + 2));
 426
 427         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
 428 }
 429
 430 /* Hash chain to use given mac address */
 431 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 432                                                 const u8 *mac, __be32 vni)
 433 {
 434         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
 435                 return &vxlan->fdb_head[eth_vni_hash(mac, vni)];
 436         else
 437                 return &vxlan->fdb_head[eth_hash(mac)];
 438 }
 439
 440 /* Look up Ethernet address in forwarding table */
 441 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 442                                           const u8 *mac, __be32 vni)
 443 {
 444         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 445         struct vxlan_fdb *f;
 446
 447         hlist_for_each_entry_rcu(f, head, hlist) {
 448                 if (ether_addr_equal(mac, f->eth_addr)) {
 449                         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
 450                                 if (vni == f->vni)
 451                                         return f;
 452                         } else {
 453                                 return f;
 454                         }
 455                 }
 456         }
 457
 458         return NULL;
 459 }
 460
 461 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 462                                         const u8 *mac, __be32 vni)
 463 {
 464         struct vxlan_fdb *f;
 465
 466         f = __vxlan_find_mac(vxlan, mac, vni);
 467         if (f)
 468                 f->used = jiffies;
 469
 470         return f;
 471 }
 472
 473 /* caller should hold vxlan->hash_lock */
 474 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 475                                               union vxlan_addr *ip, __be16 port,
 476                                               __be32 vni, __u32 ifindex)
 477 {
 478         struct vxlan_rdst *rd;
 479
 480         list_for_each_entry(rd, &f->remotes, list) {
 481                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 482                     rd->remote_port == port &&
 483                     rd->remote_vni == vni &&
 484                     rd->remote_ifindex == ifindex)
 485                         return rd;
 486         }
 487
 488         return NULL;
 489 }
 490
 491 /* Replace destination of unicast mac */
 492 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 493                              union vxlan_addr *ip, __be16 port, __be32 vni,
 494                              __u32 ifindex)
 495 {
 496         struct vxlan_rdst *rd;
 497
 498         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 499         if (rd)
 500                 return 0;
 501
 502         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 503         if (!rd)
 504                 return 0;
 505
 506         dst_cache_reset(&rd->dst_cache);
 507         rd->remote_ip = *ip;
 508         rd->remote_port = port;
 509         rd->remote_vni = vni;
 510         rd->remote_ifindex = ifindex;
 511         return 1;
 512 }
 513
 514 /* Add/update destinations for multicast */
 515 static int vxlan_fdb_append(struct vxlan_fdb *f,
 516                             union vxlan_addr *ip, __be16 port, __be32 vni,
 517                             __u32 ifindex, struct vxlan_rdst **rdp)
 518 {
 519         struct vxlan_rdst *rd;
 520
 521         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 522         if (rd)
 523                 return 0;
 524
 525         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 526         if (rd == NULL)
 527                 return -ENOBUFS;
 528
 529         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 530                 kfree(rd);
 531                 return -ENOBUFS;
 532         }
 533
 534         rd->remote_ip = *ip;
 535         rd->remote_port = port;
 536         rd->remote_vni = vni;
 537         rd->remote_ifindex = ifindex;
 538
 539         list_add_tail_rcu(&rd->list, &f->remotes);
 540
 541         *rdp = rd;
 542         return 1;
 543 }
 544
 545 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 546                                           unsigned int off,
 547                                           struct vxlanhdr *vh, size_t hdrlen,
 548                                           __be32 vni_field,
 549                                           struct gro_remcsum *grc,
 550                                           bool nopartial)
 551 {
 552         size_t start, offset;
 553
 554         if (skb->remcsum_offload)
 555                 return vh;
 556
 557         if (!NAPI_GRO_CB(skb)->csum_valid)
 558                 return NULL;
 559
 560         start = vxlan_rco_start(vni_field);
 561         offset = start + vxlan_rco_offset(vni_field);
 562
 563         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 564                                      start, offset, grc, nopartial);
 565
 566         skb->remcsum_offload = 1;
 567
 568         return vh;
 569 }
 570
 571 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
 572                                           struct sk_buff **head,
 573                                           struct sk_buff *skb)
 574 {
 575         struct sk_buff *p, **pp = NULL;
 576         struct vxlanhdr *vh, *vh2;
 577         unsigned int hlen, off_vx;
 578         int flush = 1;
 579         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 580         __be32 flags;
 581         struct gro_remcsum grc;
 582
 583         skb_gro_remcsum_init(&grc);
 584
 585         off_vx = skb_gro_offset(skb);
 586         hlen = off_vx + sizeof(*vh);
 587         vh   = skb_gro_header_fast(skb, off_vx);
 588         if (skb_gro_header_hard(skb, hlen)) {
 589                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 590                 if (unlikely(!vh))
 591                         goto out;
 592         }
 593
 594         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 595
 596         flags = vh->vx_flags;
 597
 598         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 599                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 600                                        vh->vx_vni, &grc,
 601                                        !!(vs->flags &
 602                                           VXLAN_F_REMCSUM_NOPARTIAL));
 603
 604                 if (!vh)
 605                         goto out;
 606         }
 607
 608         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 609
 610         for (p = *head; p; p = p->next) {
 611                 if (!NAPI_GRO_CB(p)->same_flow)
 612                         continue;
 613
 614                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 615                 if (vh->vx_flags != vh2->vx_flags ||
 616                     vh->vx_vni != vh2->vx_vni) {
 617                         NAPI_GRO_CB(p)->same_flow = 0;
 618                         continue;
 619                 }
 620         }
 621
 622         pp = call_gro_receive(eth_gro_receive, head, skb);
 623         flush = 0;
 624
 625 out:
 626         skb_gro_flush_final_remcsum(skb, pp, flush, &grc);
 627
 628         return pp;
 629 }
 630
 631 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 632 {
 633         /* Sets 'skb->inner_mac_header' since we are always called with
 634          * 'skb->encapsulation' set.
 635          */
 636         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 637 }
 638
 639 /* Add new entry to forwarding table -- assumes lock held */
 640 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 641                             const u8 *mac, union vxlan_addr *ip,
 642                             __u16 state, __u16 flags,
 643                             __be16 port, __be32 src_vni, __be32 vni,
 644                             __u32 ifindex, __u8 ndm_flags)
 645 {
 646         struct vxlan_rdst *rd = NULL;
 647         struct vxlan_fdb *f;
 648         int notify = 0;
 649         int rc;
 650
 651         f = __vxlan_find_mac(vxlan, mac, src_vni);
 652         if (f) {
 653                 if (flags & NLM_F_EXCL) {
 654                         netdev_dbg(vxlan->dev,
 655                                    "lost race to create %pM\n", mac);
 656                         return -EEXIST;
 657                 }
 658                 if (f->state != state) {
 659                         f->state = state;
 660                         f->updated = jiffies;
 661                         notify = 1;
 662                 }
 663                 if (f->flags != ndm_flags) {
 664                         f->flags = ndm_flags;
 665                         f->updated = jiffies;
 666                         notify = 1;
 667                 }
 668                 if ((flags & NLM_F_REPLACE)) {
 669                         /* Only change unicasts */
 670                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 671                              is_zero_ether_addr(f->eth_addr))) {
 672                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 673                                                            ifindex);
 674                         } else
 675                                 return -EOPNOTSUPP;
 676                 }
 677                 if ((flags & NLM_F_APPEND) &&
 678                     (is_multicast_ether_addr(f->eth_addr) ||
 679                      is_zero_ether_addr(f->eth_addr))) {
 680                         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 681
 682                         if (rc < 0)
 683                                 return rc;
 684                         notify |= rc;
 685                 }
 686         } else {
 687                 if (!(flags & NLM_F_CREATE))
 688                         return -ENOENT;
 689
 690                 if (vxlan->cfg.addrmax &&
 691                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 692                         return -ENOSPC;
 693
 694                 /* Disallow replace to add a multicast entry */
 695                 if ((flags & NLM_F_REPLACE) &&
 696                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 697                         return -EOPNOTSUPP;
 698
 699                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 700                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 701                 if (!f)
 702                         return -ENOMEM;
 703
 704                 notify = 1;
 705                 f->state = state;
 706                 f->flags = ndm_flags;
 707                 f->updated = f->used = jiffies;
 708                 f->vni = src_vni;
 709                 INIT_LIST_HEAD(&f->remotes);
 710                 memcpy(f->eth_addr, mac, ETH_ALEN);
 711
 712                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 713                 if (rc < 0) {
 714                         kfree(f);
 715                         return rc;
 716                 }
 717
 718                 ++vxlan->addrcnt;
 719                 hlist_add_head_rcu(&f->hlist,
 720                                    vxlan_fdb_head(vxlan, mac, src_vni));
 721         }
 722
 723         if (notify) {
 724                 if (rd == NULL)
 725                         rd = first_remote_rtnl(f);
 726                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 727         }
 728
 729         return 0;
 730 }
 731
 732 static void vxlan_fdb_free(struct rcu_head *head)
 733 {
 734         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 735         struct vxlan_rdst *rd, *nd;
 736
 737         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 738                 dst_cache_destroy(&rd->dst_cache);
 739                 kfree(rd);
 740         }
 741         kfree(f);
 742 }
 743
 744 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 745 {
 746         netdev_dbg(vxlan->dev,
 747                     "delete %pM\n", f->eth_addr);
 748
 749         --vxlan->addrcnt;
 750         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 751
 752         hlist_del_rcu(&f->hlist);
 753         call_rcu(&f->rcu, vxlan_fdb_free);
 754 }
 755
 756 static void vxlan_dst_free(struct rcu_head *head)
 757 {
 758         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 759
 760         dst_cache_destroy(&rd->dst_cache);
 761         kfree(rd);
 762 }
 763
 764 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 765                                   struct vxlan_rdst *rd)
 766 {
 767         list_del_rcu(&rd->list);
 768         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 769         call_rcu(&rd->rcu, vxlan_dst_free);
 770 }
 771
 772 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 773                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
 774                            __be32 *vni, u32 *ifindex)
 775 {
 776         struct net *net = dev_net(vxlan->dev);
 777         int err;
 778
 779         if (tb[NDA_DST]) {
 780                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 781                 if (err)
 782                         return err;
 783         } else {
 784                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 785                 if (remote->sa.sa_family == AF_INET) {
 786                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 787                         ip->sa.sa_family = AF_INET;
 788 #if IS_ENABLED(CONFIG_IPV6)
 789                 } else {
 790                         ip->sin6.sin6_addr = in6addr_any;
 791                         ip->sa.sa_family = AF_INET6;
 792 #endif
 793                 }
 794         }
 795
 796         if (tb[NDA_PORT]) {
 797                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 798                         return -EINVAL;
 799                 *port = nla_get_be16(tb[NDA_PORT]);
 800         } else {
 801                 *port = vxlan->cfg.dst_port;
 802         }
 803
 804         if (tb[NDA_VNI]) {
 805                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 806                         return -EINVAL;
 807                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 808         } else {
 809                 *vni = vxlan->default_dst.remote_vni;
 810         }
 811
 812         if (tb[NDA_SRC_VNI]) {
 813                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
 814                         return -EINVAL;
 815                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
 816         } else {
 817                 *src_vni = vxlan->default_dst.remote_vni;
 818         }
 819
 820         if (tb[NDA_IFINDEX]) {
 821                 struct net_device *tdev;
 822
 823                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 824                         return -EINVAL;
 825                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 826                 tdev = __dev_get_by_index(net, *ifindex);
 827                 if (!tdev)
 828                         return -EADDRNOTAVAIL;
 829         } else {
 830                 *ifindex = 0;
 831         }
 832
 833         return 0;
 834 }
 835
 836 /* Add static entry (via netlink) */
 837 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 838                          struct net_device *dev,
 839                          const unsigned char *addr, u16 vid, u16 flags)
 840 {
 841         struct vxlan_dev *vxlan = netdev_priv(dev);
 842         /* struct net *net = dev_net(vxlan->dev); */
 843         union vxlan_addr ip;
 844         __be16 port;
 845         __be32 src_vni, vni;
 846         u32 ifindex;
 847         int err;
 848
 849         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 850                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 851                         ndm->ndm_state);
 852                 return -EINVAL;
 853         }
 854
 855         if (tb[NDA_DST] == NULL)
 856                 return -EINVAL;
 857
 858         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 859         if (err)
 860                 return err;
 861
 862         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 863                 return -EAFNOSUPPORT;
 864
 865         spin_lock_bh(&vxlan->hash_lock);
 866         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 867                                port, src_vni, vni, ifindex, ndm->ndm_flags);
 868         spin_unlock_bh(&vxlan->hash_lock);
 869
 870         return err;
 871 }
 872
 873 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
 874                               const unsigned char *addr, union vxlan_addr ip,
 875                               __be16 port, __be32 src_vni, __be32 vni,
 876                               u32 ifindex, u16 vid)
 877 {
 878         struct vxlan_fdb *f;
 879         struct vxlan_rdst *rd = NULL;
 880         int err = -ENOENT;
 881
 882         f = vxlan_find_mac(vxlan, addr, src_vni);
 883         if (!f)
 884                 return err;
 885
 886         if (!vxlan_addr_any(&ip)) {
 887                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 888                 if (!rd)
 889                         goto out;
 890         }
 891
 892         /* remove a destination if it's not the only one on the list,
 893          * otherwise destroy the fdb entry
 894          */
 895         if (rd && !list_is_singular(&f->remotes)) {
 896                 vxlan_fdb_dst_destroy(vxlan, f, rd);
 897                 goto out;
 898         }
 899
 900         vxlan_fdb_destroy(vxlan, f);
 901
 902 out:
 903         return 0;
 904 }
 905
 906 /* Delete entry (via netlink) */
 907 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 908                             struct net_device *dev,
 909                             const unsigned char *addr, u16 vid)
 910 {
 911         struct vxlan_dev *vxlan = netdev_priv(dev);
 912         union vxlan_addr ip;
 913         __be32 src_vni, vni;
 914         __be16 port;
 915         u32 ifindex;
 916         int err;
 917
 918         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 919         if (err)
 920                 return err;
 921
 922         spin_lock_bh(&vxlan->hash_lock);
 923         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
 924                                  vid);
 925         spin_unlock_bh(&vxlan->hash_lock);
 926
 927         return err;
 928 }
 929
 930 /* Dump forwarding table */
 931 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 932                           struct net_device *dev,
 933                           struct net_device *filter_dev, int *idx)
 934 {
 935         struct vxlan_dev *vxlan = netdev_priv(dev);
 936         unsigned int h;
 937         int err = 0;
 938
 939         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 940                 struct vxlan_fdb *f;
 941
 942                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 943                         struct vxlan_rdst *rd;
 944
 945                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 946                                 if (*idx < cb->args[2])
 947                                         goto skip;
 948
 949                                 err = vxlan_fdb_info(skb, vxlan, f,
 950                                                      NETLINK_CB(cb->skb).portid,
 951                                                      cb->nlh->nlmsg_seq,
 952                                                      RTM_NEWNEIGH,
 953                                                      NLM_F_MULTI, rd);
 954                                 if (err < 0)
 955                                         goto out;
 956 skip:
 957                                 *idx += 1;
 958                         }
 959                 }
 960         }
 961 out:
 962         return err;
 963 }
 964
 965 /* Watch incoming packets to learn mapping between Ethernet address
 966  * and Tunnel endpoint.
 967  * Return true if packet is bogus and should be dropped.
 968  */
 969 static bool vxlan_snoop(struct net_device *dev,
 970                         union vxlan_addr *src_ip, const u8 *src_mac,
 971                         u32 src_ifindex, __be32 vni)
 972 {
 973         struct vxlan_dev *vxlan = netdev_priv(dev);
 974         struct vxlan_fdb *f;
 975         u32 ifindex = 0;
 976
 977 #if IS_ENABLED(CONFIG_IPV6)
 978         if (src_ip->sa.sa_family == AF_INET6 &&
 979             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
 980                 ifindex = src_ifindex;
 981 #endif
 982
 983         f = vxlan_find_mac(vxlan, src_mac, vni);
 984         if (likely(f)) {
 985                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 986
 987                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
 988                            rdst->remote_ifindex == ifindex))
 989                         return false;
 990
 991                 /* Don't migrate static entries, drop packets */
 992                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
 993                         return true;
 994
 995                 if (net_ratelimit())
 996                         netdev_info(dev,
 997                                     "%pM migrated from %pIS to %pIS\n",
 998                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 999
1000                 rdst->remote_ip = *src_ip;
1001                 f->updated = jiffies;
1002                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1003         } else {
1004                 /* learned new entry */
1005                 spin_lock(&vxlan->hash_lock);
1006
1007                 /* close off race between vxlan_flush and incoming packets */
1008                 if (netif_running(dev))
1009                         vxlan_fdb_create(vxlan, src_mac, src_ip,
1010                                          NUD_REACHABLE,
1011                                          NLM_F_EXCL|NLM_F_CREATE,
1012                                          vxlan->cfg.dst_port,
1013                                          vni,
1014                                          vxlan->default_dst.remote_vni,
1015                                          ifindex, NTF_SELF);
1016                 spin_unlock(&vxlan->hash_lock);
1017         }
1018
1019         return false;
1020 }
1021
1022 /* See if multicast group is already in use by other ID */
1023 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1024 {
1025         struct vxlan_dev *vxlan;
1026         struct vxlan_sock *sock4;
1027 #if IS_ENABLED(CONFIG_IPV6)
1028         struct vxlan_sock *sock6;
1029 #endif
1030         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1031
1032         sock4 = rtnl_dereference(dev->vn4_sock);
1033
1034         /* The vxlan_sock is only used by dev, leaving group has
1035          * no effect on other vxlan devices.
1036          */
1037         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1038                 return false;
1039 #if IS_ENABLED(CONFIG_IPV6)
1040         sock6 = rtnl_dereference(dev->vn6_sock);
1041         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1042                 return false;
1043 #endif
1044
1045         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1046                 if (!netif_running(vxlan->dev) || vxlan == dev)
1047                         continue;
1048
1049                 if (family == AF_INET &&
1050                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1051                         continue;
1052 #if IS_ENABLED(CONFIG_IPV6)
1053                 if (family == AF_INET6 &&
1054                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1055                         continue;
1056 #endif
1057
1058                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1059                                       &dev->default_dst.remote_ip))
1060                         continue;
1061
1062                 if (vxlan->default_dst.remote_ifindex !=
1063                     dev->default_dst.remote_ifindex)
1064                         continue;
1065
1066                 return true;
1067         }
1068
1069         return false;
1070 }
1071
1072 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1073 {
1074         struct vxlan_net *vn;
1075
1076         if (!vs)
1077                 return false;
1078         if (!refcount_dec_and_test(&vs->refcnt))
1079                 return false;
1080
1081         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1082         spin_lock(&vn->sock_lock);
1083         hlist_del_rcu(&vs->hlist);
1084         udp_tunnel_notify_del_rx_port(vs->sock,
1085                                       (vs->flags & VXLAN_F_GPE) ?
1086                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1087                                       UDP_TUNNEL_TYPE_VXLAN);
1088         spin_unlock(&vn->sock_lock);
1089
1090         return true;
1091 }
1092
1093 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1094 {
1095         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1096 #if IS_ENABLED(CONFIG_IPV6)
1097         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1098
1099         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1100 #endif
1101
1102         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1103         synchronize_net();
1104
1105         vxlan_vs_del_dev(vxlan);
1106
1107         if (__vxlan_sock_release_prep(sock4)) {
1108                 udp_tunnel_sock_release(sock4->sock);
1109                 kfree(sock4);
1110         }
1111
1112 #if IS_ENABLED(CONFIG_IPV6)
1113         if (__vxlan_sock_release_prep(sock6)) {
1114                 udp_tunnel_sock_release(sock6->sock);
1115                 kfree(sock6);
1116         }
1117 #endif
1118 }
1119
1120 /* Update multicast group membership when first VNI on
1121  * multicast address is brought up
1122  */
1123 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1124 {
1125         struct sock *sk;
1126         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1127         int ifindex = vxlan->default_dst.remote_ifindex;
1128         int ret = -EINVAL;
1129
1130         if (ip->sa.sa_family == AF_INET) {
1131                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1132                 struct ip_mreqn mreq = {
1133                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1134                         .imr_ifindex            = ifindex,
1135                 };
1136
1137                 sk = sock4->sock->sk;
1138                 lock_sock(sk);
1139                 ret = ip_mc_join_group(sk, &mreq);
1140                 release_sock(sk);
1141 #if IS_ENABLED(CONFIG_IPV6)
1142         } else {
1143                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1144
1145                 sk = sock6->sock->sk;
1146                 lock_sock(sk);
1147                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1148                                                    &ip->sin6.sin6_addr);
1149                 release_sock(sk);
1150 #endif
1151         }
1152
1153         return ret;
1154 }
1155
1156 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1157 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1158 {
1159         struct sock *sk;
1160         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1161         int ifindex = vxlan->default_dst.remote_ifindex;
1162         int ret = -EINVAL;
1163
1164         if (ip->sa.sa_family == AF_INET) {
1165                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1166                 struct ip_mreqn mreq = {
1167                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1168                         .imr_ifindex            = ifindex,
1169                 };
1170
1171                 sk = sock4->sock->sk;
1172                 lock_sock(sk);
1173                 ret = ip_mc_leave_group(sk, &mreq);
1174                 release_sock(sk);
1175 #if IS_ENABLED(CONFIG_IPV6)
1176         } else {
1177                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1178
1179                 sk = sock6->sock->sk;
1180                 lock_sock(sk);
1181                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1182                                                    &ip->sin6.sin6_addr);
1183                 release_sock(sk);
1184 #endif
1185         }
1186
1187         return ret;
1188 }
1189
1190 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1191                           struct sk_buff *skb, u32 vxflags)
1192 {
1193         size_t start, offset;
1194
1195         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1196                 goto out;
1197
1198         start = vxlan_rco_start(unparsed->vx_vni);
1199         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1200
1201         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1202                 return false;
1203
1204         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1205                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1206 out:
1207         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1208         unparsed->vx_vni &= VXLAN_VNI_MASK;
1209         return true;
1210 }
1211
1212 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1213                                 struct sk_buff *skb, u32 vxflags,
1214                                 struct vxlan_metadata *md)
1215 {
1216         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1217         struct metadata_dst *tun_dst;
1218
1219         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1220                 goto out;
1221
1222         md->gbp = ntohs(gbp->policy_id);
1223
1224         tun_dst = (struct metadata_dst *)skb_dst(skb);
1225         if (tun_dst) {
1226                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1227                 tun_dst->u.tun_info.options_len = sizeof(*md);
1228         }
1229         if (gbp->dont_learn)
1230                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1231
1232         if (gbp->policy_applied)
1233                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1234
1235         /* In flow-based mode, GBP is carried in dst_metadata */
1236         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1237                 skb->mark = md->gbp;
1238 out:
1239         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1240 }
1241
1242 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1243                                 __be16 *protocol,
1244                                 struct sk_buff *skb, u32 vxflags)
1245 {
1246         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1247
1248         /* Need to have Next Protocol set for interfaces in GPE mode. */
1249         if (!gpe->np_applied)
1250                 return false;
1251         /* "The initial version is 0. If a receiver does not support the
1252          * version indicated it MUST drop the packet.
1253          */
1254         if (gpe->version != 0)
1255                 return false;
1256         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1257          * processing MUST occur." However, we don't implement OAM
1258          * processing, thus drop the packet.
1259          */
1260         if (gpe->oam_flag)
1261                 return false;
1262
1263         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1264         if (!*protocol)
1265                 return false;
1266
1267         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1268         return true;
1269 }
1270
1271 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1272                           struct vxlan_sock *vs,
1273                           struct sk_buff *skb, __be32 vni)
1274 {
1275         union vxlan_addr saddr;
1276         u32 ifindex = skb->dev->ifindex;
1277
1278         skb_reset_mac_header(skb);
1279         skb->protocol = eth_type_trans(skb, vxlan->dev);
1280         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1281
1282         /* Ignore packet loops (and multicast echo) */
1283         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1284                 return false;
1285
1286         /* Get address from the outer IP header */
1287         if (vxlan_get_sk_family(vs) == AF_INET) {
1288                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1289                 saddr.sa.sa_family = AF_INET;
1290 #if IS_ENABLED(CONFIG_IPV6)
1291         } else {
1292                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1293                 saddr.sa.sa_family = AF_INET6;
1294 #endif
1295         }
1296
1297         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1298             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1299                 return false;
1300
1301         return true;
1302 }
1303
1304 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1305                                   struct sk_buff *skb)
1306 {
1307         int err = 0;
1308
1309         if (vxlan_get_sk_family(vs) == AF_INET)
1310                 err = IP_ECN_decapsulate(oiph, skb);
1311 #if IS_ENABLED(CONFIG_IPV6)
1312         else
1313                 err = IP6_ECN_decapsulate(oiph, skb);
1314 #endif
1315
1316         if (unlikely(err) && log_ecn_error) {
1317                 if (vxlan_get_sk_family(vs) == AF_INET)
1318                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1319                                              &((struct iphdr *)oiph)->saddr,
1320                                              ((struct iphdr *)oiph)->tos);
1321                 else
1322                         net_info_ratelimited("non-ECT from %pI6\n",
1323                                              &((struct ipv6hdr *)oiph)->saddr);
1324         }
1325         return err <= 1;
1326 }
1327
1328 /* Callback from net/ipv4/udp.c to receive packets */
1329 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1330 {
1331         struct pcpu_sw_netstats *stats;
1332         struct vxlan_dev *vxlan;
1333         struct vxlan_sock *vs;
1334         struct vxlanhdr unparsed;
1335         struct vxlan_metadata _md;
1336         struct vxlan_metadata *md = &_md;
1337         __be16 protocol = htons(ETH_P_TEB);
1338         bool raw_proto = false;
1339         void *oiph;
1340         __be32 vni = 0;
1341
1342         /* Need UDP and VXLAN header to be present */
1343         if (!pskb_may_pull(skb, VXLAN_HLEN))
1344                 goto drop;
1345
1346         unparsed = *vxlan_hdr(skb);
1347         /* VNI flag always required to be set */
1348         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1349                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1350                            ntohl(vxlan_hdr(skb)->vx_flags),
1351                            ntohl(vxlan_hdr(skb)->vx_vni));
1352                 /* Return non vxlan pkt */
1353                 goto drop;
1354         }
1355         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1356         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1357
1358         vs = rcu_dereference_sk_user_data(sk);
1359         if (!vs)
1360                 goto drop;
1361
1362         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1363
1364         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1365         if (!vxlan)
1366                 goto drop;
1367
1368         /* For backwards compatibility, only allow reserved fields to be
1369          * used by VXLAN extensions if explicitly requested.
1370          */
1371         if (vs->flags & VXLAN_F_GPE) {
1372                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1373                         goto drop;
1374                 raw_proto = true;
1375         }
1376
1377         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1378                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1379                         goto drop;
1380
1381         if (vxlan_collect_metadata(vs)) {
1382                 struct metadata_dst *tun_dst;
1383
1384                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1385                                          key32_to_tunnel_id(vni), sizeof(*md));
1386
1387                 if (!tun_dst)
1388                         goto drop;
1389
1390                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1391
1392                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1393         } else {
1394                 memset(md, 0, sizeof(*md));
1395         }
1396
1397         if (vs->flags & VXLAN_F_REMCSUM_RX)
1398                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1399                         goto drop;
1400         if (vs->flags & VXLAN_F_GBP)
1401                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1402         /* Note that GBP and GPE can never be active together. This is
1403          * ensured in vxlan_dev_configure.
1404          */
1405
1406         if (unparsed.vx_flags || unparsed.vx_vni) {
1407                 /* If there are any unprocessed flags remaining treat
1408                  * this as a malformed packet. This behavior diverges from
1409                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1410                  * in reserved fields are to be ignored. The approach here
1411                  * maintains compatibility with previous stack code, and also
1412                  * is more robust and provides a little more security in
1413                  * adding extensions to VXLAN.
1414                  */
1415                 goto drop;
1416         }
1417
1418         if (!raw_proto) {
1419                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1420                         goto drop;
1421         } else {
1422                 skb_reset_mac_header(skb);
1423                 skb->dev = vxlan->dev;
1424                 skb->pkt_type = PACKET_HOST;
1425         }
1426
1427         oiph = skb_network_header(skb);
1428         skb_reset_network_header(skb);
1429
1430         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1431                 ++vxlan->dev->stats.rx_frame_errors;
1432                 ++vxlan->dev->stats.rx_errors;
1433                 goto drop;
1434         }
1435
1436         stats = this_cpu_ptr(vxlan->dev->tstats);
1437         u64_stats_update_begin(&stats->syncp);
1438         stats->rx_packets++;
1439         stats->rx_bytes += skb->len;
1440         u64_stats_update_end(&stats->syncp);
1441
1442         gro_cells_receive(&vxlan->gro_cells, skb);
1443         return 0;
1444
1445 drop:
1446         /* Consume bad packet */
1447         kfree_skb(skb);
1448         return 0;
1449 }
1450
1451 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1452 {
1453         struct vxlan_dev *vxlan = netdev_priv(dev);
1454         struct arphdr *parp;
1455         u8 *arpptr, *sha;
1456         __be32 sip, tip;
1457         struct neighbour *n;
1458
1459         if (dev->flags & IFF_NOARP)
1460                 goto out;
1461
1462         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1463                 dev->stats.tx_dropped++;
1464                 goto out;
1465         }
1466         parp = arp_hdr(skb);
1467
1468         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1469              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1470             parp->ar_pro != htons(ETH_P_IP) ||
1471             parp->ar_op != htons(ARPOP_REQUEST) ||
1472             parp->ar_hln != dev->addr_len ||
1473             parp->ar_pln != 4)
1474                 goto out;
1475         arpptr = (u8 *)parp + sizeof(struct arphdr);
1476         sha = arpptr;
1477         arpptr += dev->addr_len;        /* sha */
1478         memcpy(&sip, arpptr, sizeof(sip));
1479         arpptr += sizeof(sip);
1480         arpptr += dev->addr_len;        /* tha */
1481         memcpy(&tip, arpptr, sizeof(tip));
1482
1483         if (ipv4_is_loopback(tip) ||
1484             ipv4_is_multicast(tip))
1485                 goto out;
1486
1487         n = neigh_lookup(&arp_tbl, &tip, dev);
1488
1489         if (n) {
1490                 struct vxlan_fdb *f;
1491                 struct sk_buff  *reply;
1492
1493                 if (!(n->nud_state & NUD_CONNECTED)) {
1494                         neigh_release(n);
1495                         goto out;
1496                 }
1497
1498                 f = vxlan_find_mac(vxlan, n->ha, vni);
1499                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1500                         /* bridge-local neighbor */
1501                         neigh_release(n);
1502                         goto out;
1503                 }
1504
1505                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1506                                 n->ha, sha);
1507
1508                 neigh_release(n);
1509
1510                 if (reply == NULL)
1511                         goto out;
1512
1513                 skb_reset_mac_header(reply);
1514                 __skb_pull(reply, skb_network_offset(reply));
1515                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1516                 reply->pkt_type = PACKET_HOST;
1517
1518                 if (netif_rx_ni(reply) == NET_RX_DROP)
1519                         dev->stats.rx_dropped++;
1520         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1521                 union vxlan_addr ipa = {
1522                         .sin.sin_addr.s_addr = tip,
1523                         .sin.sin_family = AF_INET,
1524                 };
1525
1526                 vxlan_ip_miss(dev, &ipa);
1527         }
1528 out:
1529         consume_skb(skb);
1530         return NETDEV_TX_OK;
1531 }
1532
1533 #if IS_ENABLED(CONFIG_IPV6)
1534 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1535         struct neighbour *n, bool isrouter)
1536 {
1537         struct net_device *dev = request->dev;
1538         struct sk_buff *reply;
1539         struct nd_msg *ns, *na;
1540         struct ipv6hdr *pip6;
1541         u8 *daddr;
1542         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1543         int ns_olen;
1544         int i, len;
1545
1546         if (dev == NULL || !pskb_may_pull(request, request->len))
1547                 return NULL;
1548
1549         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1550                 sizeof(*na) + na_olen + dev->needed_tailroom;
1551         reply = alloc_skb(len, GFP_ATOMIC);
1552         if (reply == NULL)
1553                 return NULL;
1554
1555         reply->protocol = htons(ETH_P_IPV6);
1556         reply->dev = dev;
1557         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1558         skb_push(reply, sizeof(struct ethhdr));
1559         skb_reset_mac_header(reply);
1560
1561         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1562
1563         daddr = eth_hdr(request)->h_source;
1564         ns_olen = request->len - skb_network_offset(request) -
1565                 sizeof(struct ipv6hdr) - sizeof(*ns);
1566         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1567                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1568                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1569                         break;
1570                 }
1571         }
1572
1573         /* Ethernet header */
1574         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1575         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1576         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1577         reply->protocol = htons(ETH_P_IPV6);
1578
1579         skb_pull(reply, sizeof(struct ethhdr));
1580         skb_reset_network_header(reply);
1581         skb_put(reply, sizeof(struct ipv6hdr));
1582
1583         /* IPv6 header */
1584
1585         pip6 = ipv6_hdr(reply);
1586         memset(pip6, 0, sizeof(struct ipv6hdr));
1587         pip6->version = 6;
1588         pip6->priority = ipv6_hdr(request)->priority;
1589         pip6->nexthdr = IPPROTO_ICMPV6;
1590         pip6->hop_limit = 255;
1591         pip6->daddr = ipv6_hdr(request)->saddr;
1592         pip6->saddr = *(struct in6_addr *)n->primary_key;
1593
1594         skb_pull(reply, sizeof(struct ipv6hdr));
1595         skb_reset_transport_header(reply);
1596
1597         /* Neighbor Advertisement */
1598         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1599         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1600         na->icmph.icmp6_router = isrouter;
1601         na->icmph.icmp6_override = 1;
1602         na->icmph.icmp6_solicited = 1;
1603         na->target = ns->target;
1604         ether_addr_copy(&na->opt[2], n->ha);
1605         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1606         na->opt[1] = na_olen >> 3;
1607
1608         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1609                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1610                 csum_partial(na, sizeof(*na)+na_olen, 0));
1611
1612         pip6->payload_len = htons(sizeof(*na)+na_olen);
1613
1614         skb_push(reply, sizeof(struct ipv6hdr));
1615
1616         reply->ip_summed = CHECKSUM_UNNECESSARY;
1617
1618         return reply;
1619 }
1620
1621 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1622 {
1623         struct vxlan_dev *vxlan = netdev_priv(dev);
1624         const struct in6_addr *daddr;
1625         const struct ipv6hdr *iphdr;
1626         struct inet6_dev *in6_dev;
1627         struct neighbour *n;
1628         struct nd_msg *msg;
1629
1630         in6_dev = __in6_dev_get(dev);
1631         if (!in6_dev)
1632                 goto out;
1633
1634         iphdr = ipv6_hdr(skb);
1635         daddr = &iphdr->daddr;
1636         msg = (struct nd_msg *)(iphdr + 1);
1637
1638         if (ipv6_addr_loopback(daddr) ||
1639             ipv6_addr_is_multicast(&msg->target))
1640                 goto out;
1641
1642         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1643
1644         if (n) {
1645                 struct vxlan_fdb *f;
1646                 struct sk_buff *reply;
1647
1648                 if (!(n->nud_state & NUD_CONNECTED)) {
1649                         neigh_release(n);
1650                         goto out;
1651                 }
1652
1653                 f = vxlan_find_mac(vxlan, n->ha, vni);
1654                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1655                         /* bridge-local neighbor */
1656                         neigh_release(n);
1657                         goto out;
1658                 }
1659
1660                 reply = vxlan_na_create(skb, n,
1661                                         !!(f ? f->flags & NTF_ROUTER : 0));
1662
1663                 neigh_release(n);
1664
1665                 if (reply == NULL)
1666                         goto out;
1667
1668                 if (netif_rx_ni(reply) == NET_RX_DROP)
1669                         dev->stats.rx_dropped++;
1670
1671         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1672                 union vxlan_addr ipa = {
1673                         .sin6.sin6_addr = msg->target,
1674                         .sin6.sin6_family = AF_INET6,
1675                 };
1676
1677                 vxlan_ip_miss(dev, &ipa);
1678         }
1679
1680 out:
1681         consume_skb(skb);
1682         return NETDEV_TX_OK;
1683 }
1684 #endif
1685
1686 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1687 {
1688         struct vxlan_dev *vxlan = netdev_priv(dev);
1689         struct neighbour *n;
1690
1691         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1692                 return false;
1693
1694         n = NULL;
1695         switch (ntohs(eth_hdr(skb)->h_proto)) {
1696         case ETH_P_IP:
1697         {
1698                 struct iphdr *pip;
1699
1700                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1701                         return false;
1702                 pip = ip_hdr(skb);
1703                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1704                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1705                         union vxlan_addr ipa = {
1706                                 .sin.sin_addr.s_addr = pip->daddr,
1707                                 .sin.sin_family = AF_INET,
1708                         };
1709
1710                         vxlan_ip_miss(dev, &ipa);
1711                         return false;
1712                 }
1713
1714                 break;
1715         }
1716 #if IS_ENABLED(CONFIG_IPV6)
1717         case ETH_P_IPV6:
1718         {
1719                 struct ipv6hdr *pip6;
1720
1721                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1722                         return false;
1723                 pip6 = ipv6_hdr(skb);
1724                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1725                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1726                         union vxlan_addr ipa = {
1727                                 .sin6.sin6_addr = pip6->daddr,
1728                                 .sin6.sin6_family = AF_INET6,
1729                         };
1730
1731                         vxlan_ip_miss(dev, &ipa);
1732                         return false;
1733                 }
1734
1735                 break;
1736         }
1737 #endif
1738         default:
1739                 return false;
1740         }
1741
1742         if (n) {
1743                 bool diff;
1744
1745                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1746                 if (diff) {
1747                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1748                                 dev->addr_len);
1749                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1750                 }
1751                 neigh_release(n);
1752                 return diff;
1753         }
1754
1755         return false;
1756 }
1757
1758 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1759                                 struct vxlan_metadata *md)
1760 {
1761         struct vxlanhdr_gbp *gbp;
1762
1763         if (!md->gbp)
1764                 return;
1765
1766         gbp = (struct vxlanhdr_gbp *)vxh;
1767         vxh->vx_flags |= VXLAN_HF_GBP;
1768
1769         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1770                 gbp->dont_learn = 1;
1771
1772         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1773                 gbp->policy_applied = 1;
1774
1775         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1776 }
1777
1778 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1779                                __be16 protocol)
1780 {
1781         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1782
1783         gpe->np_applied = 1;
1784         gpe->next_protocol = tun_p_from_eth_p(protocol);
1785         if (!gpe->next_protocol)
1786                 return -EPFNOSUPPORT;
1787         return 0;
1788 }
1789
1790 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1791                            int iphdr_len, __be32 vni,
1792                            struct vxlan_metadata *md, u32 vxflags,
1793                            bool udp_sum)
1794 {
1795         struct vxlanhdr *vxh;
1796         int min_headroom;
1797         int err;
1798         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1799         __be16 inner_protocol = htons(ETH_P_TEB);
1800
1801         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1802             skb->ip_summed == CHECKSUM_PARTIAL) {
1803                 int csum_start = skb_checksum_start_offset(skb);
1804
1805                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1806                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1807                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1808                      skb->csum_offset == offsetof(struct tcphdr, check)))
1809                         type |= SKB_GSO_TUNNEL_REMCSUM;
1810         }
1811
1812         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1813                         + VXLAN_HLEN + iphdr_len;
1814
1815         /* Need space for new headers (invalidates iph ptr) */
1816         err = skb_cow_head(skb, min_headroom);
1817         if (unlikely(err))
1818                 return err;
1819
1820         err = iptunnel_handle_offloads(skb, type);
1821         if (err)
1822                 return err;
1823
1824         vxh = __skb_push(skb, sizeof(*vxh));
1825         vxh->vx_flags = VXLAN_HF_VNI;
1826         vxh->vx_vni = vxlan_vni_field(vni);
1827
1828         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1829                 unsigned int start;
1830
1831                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1832                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1833                 vxh->vx_flags |= VXLAN_HF_RCO;
1834
1835                 if (!skb_is_gso(skb)) {
1836                         skb->ip_summed = CHECKSUM_NONE;
1837                         skb->encapsulation = 0;
1838                 }
1839         }
1840
1841         if (vxflags & VXLAN_F_GBP)
1842                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1843         if (vxflags & VXLAN_F_GPE) {
1844                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1845                 if (err < 0)
1846                         return err;
1847                 inner_protocol = skb->protocol;
1848         }
1849
1850         skb_set_inner_protocol(skb, inner_protocol);
1851         return 0;
1852 }
1853
1854 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1855                                       struct vxlan_sock *sock4,
1856                                       struct sk_buff *skb, int oif, u8 tos,
1857                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
1858                                       struct dst_cache *dst_cache,
1859                                       const struct ip_tunnel_info *info)
1860 {
1861         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1862         struct rtable *rt = NULL;
1863         struct flowi4 fl4;
1864
1865         if (!sock4)
1866                 return ERR_PTR(-EIO);
1867
1868         if (tos && !info)
1869                 use_cache = false;
1870         if (use_cache) {
1871                 rt = dst_cache_get_ip4(dst_cache, saddr);
1872                 if (rt)
1873                         return rt;
1874         }
1875
1876         memset(&fl4, 0, sizeof(fl4));
1877         fl4.flowi4_oif = oif;
1878         fl4.flowi4_tos = RT_TOS(tos);
1879         fl4.flowi4_mark = skb->mark;
1880         fl4.flowi4_proto = IPPROTO_UDP;
1881         fl4.daddr = daddr;
1882         fl4.saddr = *saddr;
1883         fl4.fl4_dport = dport;
1884         fl4.fl4_sport = sport;
1885
1886         rt = ip_route_output_key(vxlan->net, &fl4);
1887         if (likely(!IS_ERR(rt))) {
1888                 if (rt->dst.dev == dev) {
1889                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1890                         ip_rt_put(rt);
1891                         return ERR_PTR(-ELOOP);
1892                 }
1893
1894                 *saddr = fl4.saddr;
1895                 if (use_cache)
1896                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1897         } else {
1898                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1899                 return ERR_PTR(-ENETUNREACH);
1900         }
1901         return rt;
1902 }
1903
1904 #if IS_ENABLED(CONFIG_IPV6)
1905 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1906                                           struct net_device *dev,
1907                                           struct vxlan_sock *sock6,
1908                                           struct sk_buff *skb, int oif, u8 tos,
1909                                           __be32 label,
1910                                           const struct in6_addr *daddr,
1911                                           struct in6_addr *saddr,
1912                                           __be16 dport, __be16 sport,
1913                                           struct dst_cache *dst_cache,
1914                                           const struct ip_tunnel_info *info)
1915 {
1916         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1917         struct dst_entry *ndst;
1918         struct flowi6 fl6;
1919         int err;
1920
1921         if (!sock6)
1922                 return ERR_PTR(-EIO);
1923
1924         if (tos && !info)
1925                 use_cache = false;
1926         if (use_cache) {
1927                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1928                 if (ndst)
1929                         return ndst;
1930         }
1931
1932         memset(&fl6, 0, sizeof(fl6));
1933         fl6.flowi6_oif = oif;
1934         fl6.daddr = *daddr;
1935         fl6.saddr = *saddr;
1936         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1937         fl6.flowi6_mark = skb->mark;
1938         fl6.flowi6_proto = IPPROTO_UDP;
1939         fl6.fl6_dport = dport;
1940         fl6.fl6_sport = sport;
1941
1942         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1943                                          sock6->sock->sk,
1944                                          &ndst, &fl6);
1945         if (unlikely(err < 0)) {
1946                 netdev_dbg(dev, "no route to %pI6\n", daddr);
1947                 return ERR_PTR(-ENETUNREACH);
1948         }
1949
1950         if (unlikely(ndst->dev == dev)) {
1951                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
1952                 dst_release(ndst);
1953                 return ERR_PTR(-ELOOP);
1954         }
1955
1956         *saddr = fl6.saddr;
1957         if (use_cache)
1958                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1959         return ndst;
1960 }
1961 #endif
1962
1963 /* Bypass encapsulation if the destination is local */
1964 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1965                                struct vxlan_dev *dst_vxlan, __be32 vni)
1966 {
1967         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1968         union vxlan_addr loopback;
1969         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1970         struct net_device *dev = skb->dev;
1971         int len = skb->len;
1972
1973         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1974         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1975         skb->pkt_type = PACKET_HOST;
1976         skb->encapsulation = 0;
1977         skb->dev = dst_vxlan->dev;
1978         __skb_pull(skb, skb_network_offset(skb));
1979
1980         if (remote_ip->sa.sa_family == AF_INET) {
1981                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1982                 loopback.sa.sa_family =  AF_INET;
1983 #if IS_ENABLED(CONFIG_IPV6)
1984         } else {
1985                 loopback.sin6.sin6_addr = in6addr_loopback;
1986                 loopback.sa.sa_family =  AF_INET6;
1987 #endif
1988         }
1989
1990         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
1991                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source, 0,
1992                             vni);
1993
1994         u64_stats_update_begin(&tx_stats->syncp);
1995         tx_stats->tx_packets++;
1996         tx_stats->tx_bytes += len;
1997         u64_stats_update_end(&tx_stats->syncp);
1998
1999         if (netif_rx(skb) == NET_RX_SUCCESS) {
2000                 u64_stats_update_begin(&rx_stats->syncp);
2001                 rx_stats->rx_packets++;
2002                 rx_stats->rx_bytes += len;
2003                 u64_stats_update_end(&rx_stats->syncp);
2004         } else {
2005                 dev->stats.rx_dropped++;
2006         }
2007 }
2008
2009 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2010                                  struct vxlan_dev *vxlan,
2011                                  union vxlan_addr *daddr,
2012                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2013                                  struct dst_entry *dst,
2014                                  u32 rt_flags)
2015 {
2016 #if IS_ENABLED(CONFIG_IPV6)
2017         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2018          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2019          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2020          */
2021         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2022 #endif
2023         /* Bypass encapsulation if the destination is local */
2024         if (rt_flags & RTCF_LOCAL &&
2025             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2026                 struct vxlan_dev *dst_vxlan;
2027
2028                 dst_release(dst);
2029                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2030                                            daddr->sa.sa_family, dst_port,
2031                                            vxlan->cfg.flags);
2032                 if (!dst_vxlan) {
2033                         dev->stats.tx_errors++;
2034                         kfree_skb(skb);
2035
2036                         return -ENOENT;
2037                 }
2038                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2039                 return 1;
2040         }
2041
2042         return 0;
2043 }
2044
2045 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2046                            __be32 default_vni, struct vxlan_rdst *rdst,
2047                            bool did_rsc)
2048 {
2049         struct dst_cache *dst_cache;
2050         struct ip_tunnel_info *info;
2051         struct vxlan_dev *vxlan = netdev_priv(dev);
2052         const struct iphdr *old_iph = ip_hdr(skb);
2053         union vxlan_addr *dst;
2054         union vxlan_addr remote_ip, local_ip;
2055         struct vxlan_metadata _md;
2056         struct vxlan_metadata *md = &_md;
2057         __be16 src_port = 0, dst_port;
2058         struct dst_entry *ndst = NULL;
2059         __be32 vni, label;
2060         __u8 tos, ttl;
2061         int ifindex;
2062         int err;
2063         u32 flags = vxlan->cfg.flags;
2064         bool udp_sum = false;
2065         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2066
2067         info = skb_tunnel_info(skb);
2068
2069         if (rdst) {
2070                 dst = &rdst->remote_ip;
2071                 if (vxlan_addr_any(dst)) {
2072                         if (did_rsc) {
2073                                 /* short-circuited back to local bridge */
2074                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2075                                 return;
2076                         }
2077                         goto drop;
2078                 }
2079
2080                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2081                 vni = (rdst->remote_vni) ? : default_vni;
2082                 ifindex = rdst->remote_ifindex;
2083                 local_ip = vxlan->cfg.saddr;
2084                 dst_cache = &rdst->dst_cache;
2085                 md->gbp = skb->mark;
2086                 if (flags & VXLAN_F_TTL_INHERIT) {
2087                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2088                 } else {
2089                         ttl = vxlan->cfg.ttl;
2090                         if (!ttl && vxlan_addr_multicast(dst))
2091                                 ttl = 1;
2092                 }
2093
2094                 tos = vxlan->cfg.tos;
2095                 if (tos == 1)
2096                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2097
2098                 if (dst->sa.sa_family == AF_INET)
2099                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2100                 else
2101                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2102                 label = vxlan->cfg.label;
2103         } else {
2104                 if (!info) {
2105                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2106                                   dev->name);
2107                         goto drop;
2108                 }
2109                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2110                 if (remote_ip.sa.sa_family == AF_INET) {
2111                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2112                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2113                 } else {
2114                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2115                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2116                 }
2117                 dst = &remote_ip;
2118                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2119                 vni = tunnel_id_to_key32(info->key.tun_id);
2120                 ifindex = 0;
2121                 dst_cache = &info->dst_cache;
2122                 if (info->options_len)
2123                         md = ip_tunnel_info_opts(info);
2124                 ttl = info->key.ttl;
2125                 tos = info->key.tos;
2126                 label = info->key.label;
2127                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2128         }
2129         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2130                                      vxlan->cfg.port_max, true);
2131
2132         rcu_read_lock();
2133         if (dst->sa.sa_family == AF_INET) {
2134                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2135                 struct rtable *rt;
2136                 __be16 df = 0;
2137
2138                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2139                                      dst->sin.sin_addr.s_addr,
2140                                      &local_ip.sin.sin_addr.s_addr,
2141                                      dst_port, src_port,
2142                                      dst_cache, info);
2143                 if (IS_ERR(rt)) {
2144                         err = PTR_ERR(rt);
2145                         goto tx_error;
2146                 }
2147
2148                 /* Bypass encapsulation if the destination is local */
2149                 if (!info) {
2150                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2151                                                     dst_port, ifindex, vni,
2152                                                     &rt->dst, rt->rt_flags);
2153                         if (err)
2154                                 goto out_unlock;
2155                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2156                         df = htons(IP_DF);
2157                 }
2158
2159                 ndst = &rt->dst;
2160                 if (skb_dst(skb)) {
2161                         int mtu = dst_mtu(ndst) - VXLAN_HEADROOM;
2162
2163                         skb_dst_update_pmtu(skb, mtu);
2164                 }
2165
2166                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2167                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2168                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2169                                       vni, md, flags, udp_sum);
2170                 if (err < 0)
2171                         goto tx_error;
2172
2173                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2174                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2175                                     src_port, dst_port, xnet, !udp_sum);
2176 #if IS_ENABLED(CONFIG_IPV6)
2177         } else {
2178                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2179
2180                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2181                                         label, &dst->sin6.sin6_addr,
2182                                         &local_ip.sin6.sin6_addr,
2183                                         dst_port, src_port,
2184                                         dst_cache, info);
2185                 if (IS_ERR(ndst)) {
2186                         err = PTR_ERR(ndst);
2187                         ndst = NULL;
2188                         goto tx_error;
2189                 }
2190
2191                 if (!info) {
2192                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2193
2194                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2195                                                     dst_port, ifindex, vni,
2196                                                     ndst, rt6i_flags);
2197                         if (err)
2198                                 goto out_unlock;
2199                 }
2200
2201                 if (skb_dst(skb)) {
2202                         int mtu = dst_mtu(ndst) - VXLAN6_HEADROOM;
2203
2204                         skb_dst_update_pmtu(skb, mtu);
2205                 }
2206
2207                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2208                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2209                 skb_scrub_packet(skb, xnet);
2210                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2211                                       vni, md, flags, udp_sum);
2212                 if (err < 0)
2213                         goto tx_error;
2214
2215                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2216                                      &local_ip.sin6.sin6_addr,
2217                                      &dst->sin6.sin6_addr, tos, ttl,
2218                                      label, src_port, dst_port, !udp_sum);
2219 #endif
2220         }
2221 out_unlock:
2222         rcu_read_unlock();
2223         return;
2224
2225 drop:
2226         dev->stats.tx_dropped++;
2227         dev_kfree_skb(skb);
2228         return;
2229
2230 tx_error:
2231         rcu_read_unlock();
2232         if (err == -ELOOP)
2233                 dev->stats.collisions++;
2234         else if (err == -ENETUNREACH)
2235                 dev->stats.tx_carrier_errors++;
2236         dst_release(ndst);
2237         dev->stats.tx_errors++;
2238         kfree_skb(skb);
2239 }
2240
2241 /* Transmit local packets over Vxlan
2242  *
2243  * Outer IP header inherits ECN and DF from inner header.
2244  * Outer UDP destination is the VXLAN assigned port.
2245  *           source port is based on hash of flow
2246  */
2247 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2248 {
2249         struct vxlan_dev *vxlan = netdev_priv(dev);
2250         struct vxlan_rdst *rdst, *fdst = NULL;
2251         const struct ip_tunnel_info *info;
2252         bool did_rsc = false;
2253         struct vxlan_fdb *f;
2254         struct ethhdr *eth;
2255         __be32 vni = 0;
2256
2257         info = skb_tunnel_info(skb);
2258
2259         skb_reset_mac_header(skb);
2260
2261         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2262                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2263                     info->mode & IP_TUNNEL_INFO_TX) {
2264                         vni = tunnel_id_to_key32(info->key.tun_id);
2265                 } else {
2266                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2267                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2268                         else
2269                                 kfree_skb(skb);
2270                         return NETDEV_TX_OK;
2271                 }
2272         }
2273
2274         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2275                 eth = eth_hdr(skb);
2276                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2277                         return arp_reduce(dev, skb, vni);
2278 #if IS_ENABLED(CONFIG_IPV6)
2279                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2280                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2281                                             sizeof(struct nd_msg)) &&
2282                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2283                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2284
2285                         if (m->icmph.icmp6_code == 0 &&
2286                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2287                                 return neigh_reduce(dev, skb, vni);
2288                 }
2289 #endif
2290         }
2291
2292         eth = eth_hdr(skb);
2293         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2294         did_rsc = false;
2295
2296         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2297             (ntohs(eth->h_proto) == ETH_P_IP ||
2298              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2299                 did_rsc = route_shortcircuit(dev, skb);
2300                 if (did_rsc)
2301                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2302         }
2303
2304         if (f == NULL) {
2305                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2306                 if (f == NULL) {
2307                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2308                             !is_multicast_ether_addr(eth->h_dest))
2309                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2310
2311                         dev->stats.tx_dropped++;
2312                         kfree_skb(skb);
2313                         return NETDEV_TX_OK;
2314                 }
2315         }
2316
2317         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2318                 struct sk_buff *skb1;
2319
2320                 if (!fdst) {
2321                         fdst = rdst;
2322                         continue;
2323                 }
2324                 skb1 = skb_clone(skb, GFP_ATOMIC);
2325                 if (skb1)
2326                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2327         }
2328
2329         if (fdst)
2330                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2331         else
2332                 kfree_skb(skb);
2333         return NETDEV_TX_OK;
2334 }
2335
2336 /* Walk the forwarding table and purge stale entries */
2337 static void vxlan_cleanup(struct timer_list *t)
2338 {
2339         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2340         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2341         unsigned int h;
2342
2343         if (!netif_running(vxlan->dev))
2344                 return;
2345
2346         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2347                 struct hlist_node *p, *n;
2348
2349                 spin_lock_bh(&vxlan->hash_lock);
2350                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2351                         struct vxlan_fdb *f
2352                                 = container_of(p, struct vxlan_fdb, hlist);
2353                         unsigned long timeout;
2354
2355                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2356                                 continue;
2357
2358                         if (f->flags & NTF_EXT_LEARNED)
2359                                 continue;
2360
2361                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2362                         if (time_before_eq(timeout, jiffies)) {
2363                                 netdev_dbg(vxlan->dev,
2364                                            "garbage collect %pM\n",
2365                                            f->eth_addr);
2366                                 f->state = NUD_STALE;
2367                                 vxlan_fdb_destroy(vxlan, f);
2368                         } else if (time_before(timeout, next_timer))
2369                                 next_timer = timeout;
2370                 }
2371                 spin_unlock_bh(&vxlan->hash_lock);
2372         }
2373
2374         mod_timer(&vxlan->age_timer, next_timer);
2375 }
2376
2377 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2378 {
2379         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2380
2381         spin_lock(&vn->sock_lock);
2382         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2383 #if IS_ENABLED(CONFIG_IPV6)
2384         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2385 #endif
2386         spin_unlock(&vn->sock_lock);
2387 }
2388
2389 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2390                              struct vxlan_dev_node *node)
2391 {
2392         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2393         __be32 vni = vxlan->default_dst.remote_vni;
2394
2395         node->vxlan = vxlan;
2396         spin_lock(&vn->sock_lock);
2397         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2398         spin_unlock(&vn->sock_lock);
2399 }
2400
2401 /* Setup stats when device is created */
2402 static int vxlan_init(struct net_device *dev)
2403 {
2404         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2405         if (!dev->tstats)
2406                 return -ENOMEM;
2407
2408         return 0;
2409 }
2410
2411 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2412 {
2413         struct vxlan_fdb *f;
2414
2415         spin_lock_bh(&vxlan->hash_lock);
2416         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2417         if (f)
2418                 vxlan_fdb_destroy(vxlan, f);
2419         spin_unlock_bh(&vxlan->hash_lock);
2420 }
2421
2422 static void vxlan_uninit(struct net_device *dev)
2423 {
2424         struct vxlan_dev *vxlan = netdev_priv(dev);
2425
2426         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2427
2428         free_percpu(dev->tstats);
2429 }
2430
2431 /* Start ageing timer and join group when device is brought up */
2432 static int vxlan_open(struct net_device *dev)
2433 {
2434         struct vxlan_dev *vxlan = netdev_priv(dev);
2435         int ret;
2436
2437         ret = vxlan_sock_add(vxlan);
2438         if (ret < 0)
2439                 return ret;
2440
2441         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2442                 ret = vxlan_igmp_join(vxlan);
2443                 if (ret == -EADDRINUSE)
2444                         ret = 0;
2445                 if (ret) {
2446                         vxlan_sock_release(vxlan);
2447                         return ret;
2448                 }
2449         }
2450
2451         if (vxlan->cfg.age_interval)
2452                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2453
2454         return ret;
2455 }
2456
2457 /* Purge the forwarding table */
2458 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2459 {
2460         unsigned int h;
2461
2462         spin_lock_bh(&vxlan->hash_lock);
2463         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2464                 struct hlist_node *p, *n;
2465                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2466                         struct vxlan_fdb *f
2467                                 = container_of(p, struct vxlan_fdb, hlist);
2468                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2469                                 continue;
2470                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2471                         if (!is_zero_ether_addr(f->eth_addr))
2472                                 vxlan_fdb_destroy(vxlan, f);
2473                 }
2474         }
2475         spin_unlock_bh(&vxlan->hash_lock);
2476 }
2477
2478 /* Cleanup timer and forwarding table on shutdown */
2479 static int vxlan_stop(struct net_device *dev)
2480 {
2481         struct vxlan_dev *vxlan = netdev_priv(dev);
2482         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2483         int ret = 0;
2484
2485         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2486             !vxlan_group_used(vn, vxlan))
2487                 ret = vxlan_igmp_leave(vxlan);
2488
2489         del_timer_sync(&vxlan->age_timer);
2490
2491         vxlan_flush(vxlan, false);
2492         vxlan_sock_release(vxlan);
2493
2494         return ret;
2495 }
2496
2497 /* Stub, nothing needs to be done. */
2498 static void vxlan_set_multicast_list(struct net_device *dev)
2499 {
2500 }
2501
2502 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2503 {
2504         struct vxlan_dev *vxlan = netdev_priv(dev);
2505         struct vxlan_rdst *dst = &vxlan->default_dst;
2506         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2507                                                          dst->remote_ifindex);
2508         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
2509
2510         /* This check is different than dev->max_mtu, because it looks at
2511          * the lowerdev->mtu, rather than the static dev->max_mtu
2512          */
2513         if (lowerdev) {
2514                 int max_mtu = lowerdev->mtu -
2515                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2516                 if (new_mtu > max_mtu)
2517                         return -EINVAL;
2518         }
2519
2520         dev->mtu = new_mtu;
2521         return 0;
2522 }
2523
2524 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2525 {
2526         struct vxlan_dev *vxlan = netdev_priv(dev);
2527         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2528         __be16 sport, dport;
2529
2530         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2531                                   vxlan->cfg.port_max, true);
2532         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2533
2534         if (ip_tunnel_info_af(info) == AF_INET) {
2535                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2536                 struct rtable *rt;
2537
2538                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2539                                      info->key.u.ipv4.dst,
2540                                      &info->key.u.ipv4.src, dport, sport,
2541                                      &info->dst_cache, info);
2542                 if (IS_ERR(rt))
2543                         return PTR_ERR(rt);
2544                 ip_rt_put(rt);
2545         } else {
2546 #if IS_ENABLED(CONFIG_IPV6)
2547                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2548                 struct dst_entry *ndst;
2549
2550                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2551                                         info->key.label, &info->key.u.ipv6.dst,
2552                                         &info->key.u.ipv6.src, dport, sport,
2553                                         &info->dst_cache, info);
2554                 if (IS_ERR(ndst))
2555                         return PTR_ERR(ndst);
2556                 dst_release(ndst);
2557 #else /* !CONFIG_IPV6 */
2558                 return -EPFNOSUPPORT;
2559 #endif
2560         }
2561         info->key.tp_src = sport;
2562         info->key.tp_dst = dport;
2563         return 0;
2564 }
2565
2566 static const struct net_device_ops vxlan_netdev_ether_ops = {
2567         .ndo_init               = vxlan_init,
2568         .ndo_uninit             = vxlan_uninit,
2569         .ndo_open               = vxlan_open,
2570         .ndo_stop               = vxlan_stop,
2571         .ndo_start_xmit         = vxlan_xmit,
2572         .ndo_get_stats64        = ip_tunnel_get_stats64,
2573         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2574         .ndo_change_mtu         = vxlan_change_mtu,
2575         .ndo_validate_addr      = eth_validate_addr,
2576         .ndo_set_mac_address    = eth_mac_addr,
2577         .ndo_fdb_add            = vxlan_fdb_add,
2578         .ndo_fdb_del            = vxlan_fdb_delete,
2579         .ndo_fdb_dump           = vxlan_fdb_dump,
2580         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2581 };
2582
2583 static const struct net_device_ops vxlan_netdev_raw_ops = {
2584         .ndo_init               = vxlan_init,
2585         .ndo_uninit             = vxlan_uninit,
2586         .ndo_open               = vxlan_open,
2587         .ndo_stop               = vxlan_stop,
2588         .ndo_start_xmit         = vxlan_xmit,
2589         .ndo_get_stats64        = ip_tunnel_get_stats64,
2590         .ndo_change_mtu         = vxlan_change_mtu,
2591         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2592 };
2593
2594 /* Info for udev, that this is a virtual tunnel endpoint */
2595 static struct device_type vxlan_type = {
2596         .name = "vxlan",
2597 };
2598
2599 /* Calls the ndo_udp_tunnel_add of the caller in order to
2600  * supply the listening VXLAN udp ports. Callers are expected
2601  * to implement the ndo_udp_tunnel_add.
2602  */
2603 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
2604 {
2605         struct vxlan_sock *vs;
2606         struct net *net = dev_net(dev);
2607         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2608         unsigned int i;
2609
2610         spin_lock(&vn->sock_lock);
2611         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2612                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2613                         unsigned short type;
2614
2615                         if (vs->flags & VXLAN_F_GPE)
2616                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
2617                         else
2618                                 type = UDP_TUNNEL_TYPE_VXLAN;
2619
2620                         if (push)
2621                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
2622                         else
2623                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
2624                 }
2625         }
2626         spin_unlock(&vn->sock_lock);
2627 }
2628
2629 /* Initialize the device structure. */
2630 static void vxlan_setup(struct net_device *dev)
2631 {
2632         struct vxlan_dev *vxlan = netdev_priv(dev);
2633         unsigned int h;
2634
2635         eth_hw_addr_random(dev);
2636         ether_setup(dev);
2637
2638         dev->needs_free_netdev = true;
2639         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2640
2641         dev->features   |= NETIF_F_LLTX;
2642         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2643         dev->features   |= NETIF_F_RXCSUM;
2644         dev->features   |= NETIF_F_GSO_SOFTWARE;
2645
2646         dev->vlan_features = dev->features;
2647         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2648         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2649         netif_keep_dst(dev);
2650         dev->priv_flags |= IFF_NO_QUEUE;
2651
2652         /* MTU range: 68 - 65535 */
2653         dev->min_mtu = ETH_MIN_MTU;
2654         dev->max_mtu = ETH_MAX_MTU;
2655
2656         INIT_LIST_HEAD(&vxlan->next);
2657         spin_lock_init(&vxlan->hash_lock);
2658
2659         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
2660
2661         vxlan->dev = dev;
2662
2663         gro_cells_init(&vxlan->gro_cells, dev);
2664
2665         for (h = 0; h < FDB_HASH_SIZE; ++h)
2666                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2667 }
2668
2669 static void vxlan_ether_setup(struct net_device *dev)
2670 {
2671         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2672         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2673         dev->netdev_ops = &vxlan_netdev_ether_ops;
2674 }
2675
2676 static void vxlan_raw_setup(struct net_device *dev)
2677 {
2678         dev->header_ops = NULL;
2679         dev->type = ARPHRD_NONE;
2680         dev->hard_header_len = 0;
2681         dev->addr_len = 0;
2682         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2683         dev->netdev_ops = &vxlan_netdev_raw_ops;
2684 }
2685
2686 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2687         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2688         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2689         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2690         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2691         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2692         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2693         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2694         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2695         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2696         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2697         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2698         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2699         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2700         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2701         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2702         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2703         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2704         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2705         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2706         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2707         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2708         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2709         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2710         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2711         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2712         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2713         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2714         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
2715 };
2716
2717 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
2718                           struct netlink_ext_ack *extack)
2719 {
2720         if (tb[IFLA_ADDRESS]) {
2721                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2722                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2723                                             "Provided link layer address is not Ethernet");
2724                         return -EINVAL;
2725                 }
2726
2727                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2728                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2729                                             "Provided Ethernet address is not unicast");
2730                         return -EADDRNOTAVAIL;
2731                 }
2732         }
2733
2734         if (tb[IFLA_MTU]) {
2735                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
2736
2737                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
2738                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
2739                                             "MTU must be between 68 and 65535");
2740                         return -EINVAL;
2741                 }
2742         }
2743
2744         if (!data) {
2745                 NL_SET_ERR_MSG(extack,
2746                                "Required attributes not provided to perform the operation");
2747                 return -EINVAL;
2748         }
2749
2750         if (data[IFLA_VXLAN_ID]) {
2751                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2752
2753                 if (id >= VXLAN_N_VID) {
2754                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID],
2755                                             "VXLAN ID must be lower than 16777216");
2756                         return -ERANGE;
2757                 }
2758         }
2759
2760         if (data[IFLA_VXLAN_PORT_RANGE]) {
2761                 const struct ifla_vxlan_port_range *p
2762                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2763
2764                 if (ntohs(p->high) < ntohs(p->low)) {
2765                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
2766                                             "Invalid source port range");
2767                         return -EINVAL;
2768                 }
2769         }
2770
2771         return 0;
2772 }
2773
2774 static void vxlan_get_drvinfo(struct net_device *netdev,
2775                               struct ethtool_drvinfo *drvinfo)
2776 {
2777         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2778         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2779 }
2780
2781 static const struct ethtool_ops vxlan_ethtool_ops = {
2782         .get_drvinfo    = vxlan_get_drvinfo,
2783         .get_link       = ethtool_op_get_link,
2784 };
2785
2786 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2787                                         __be16 port, u32 flags)
2788 {
2789         struct socket *sock;
2790         struct udp_port_cfg udp_conf;
2791         int err;
2792
2793         memset(&udp_conf, 0, sizeof(udp_conf));
2794
2795         if (ipv6) {
2796                 udp_conf.family = AF_INET6;
2797                 udp_conf.use_udp6_rx_checksums =
2798                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2799                 udp_conf.ipv6_v6only = 1;
2800         } else {
2801                 udp_conf.family = AF_INET;
2802         }
2803
2804         udp_conf.local_udp_port = port;
2805
2806         /* Open UDP socket */
2807         err = udp_sock_create(net, &udp_conf, &sock);
2808         if (err < 0)
2809                 return ERR_PTR(err);
2810
2811         return sock;
2812 }
2813
2814 /* Create new listen socket if needed */
2815 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2816                                               __be16 port, u32 flags)
2817 {
2818         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2819         struct vxlan_sock *vs;
2820         struct socket *sock;
2821         unsigned int h;
2822         struct udp_tunnel_sock_cfg tunnel_cfg;
2823
2824         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2825         if (!vs)
2826                 return ERR_PTR(-ENOMEM);
2827
2828         for (h = 0; h < VNI_HASH_SIZE; ++h)
2829                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2830
2831         sock = vxlan_create_sock(net, ipv6, port, flags);
2832         if (IS_ERR(sock)) {
2833                 kfree(vs);
2834                 return ERR_CAST(sock);
2835         }
2836
2837         vs->sock = sock;
2838         refcount_set(&vs->refcnt, 1);
2839         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2840
2841         spin_lock(&vn->sock_lock);
2842         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2843         udp_tunnel_notify_add_rx_port(sock,
2844                                       (vs->flags & VXLAN_F_GPE) ?
2845                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2846                                       UDP_TUNNEL_TYPE_VXLAN);
2847         spin_unlock(&vn->sock_lock);
2848
2849         /* Mark socket as an encapsulation socket. */
2850         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2851         tunnel_cfg.sk_user_data = vs;
2852         tunnel_cfg.encap_type = 1;
2853         tunnel_cfg.encap_rcv = vxlan_rcv;
2854         tunnel_cfg.encap_destroy = NULL;
2855         tunnel_cfg.gro_receive = vxlan_gro_receive;
2856         tunnel_cfg.gro_complete = vxlan_gro_complete;
2857
2858         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2859
2860         return vs;
2861 }
2862
2863 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2864 {
2865         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2866         struct vxlan_sock *vs = NULL;
2867         struct vxlan_dev_node *node;
2868
2869         if (!vxlan->cfg.no_share) {
2870                 spin_lock(&vn->sock_lock);
2871                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2872                                      vxlan->cfg.dst_port, vxlan->cfg.flags);
2873                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
2874                         spin_unlock(&vn->sock_lock);
2875                         return -EBUSY;
2876                 }
2877                 spin_unlock(&vn->sock_lock);
2878         }
2879         if (!vs)
2880                 vs = vxlan_socket_create(vxlan->net, ipv6,
2881                                          vxlan->cfg.dst_port, vxlan->cfg.flags);
2882         if (IS_ERR(vs))
2883                 return PTR_ERR(vs);
2884 #if IS_ENABLED(CONFIG_IPV6)
2885         if (ipv6) {
2886                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2887                 node = &vxlan->hlist6;
2888         } else
2889 #endif
2890         {
2891                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2892                 node = &vxlan->hlist4;
2893         }
2894         vxlan_vs_add_dev(vs, vxlan, node);
2895         return 0;
2896 }
2897
2898 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2899 {
2900         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
2901         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
2902         bool ipv4 = !ipv6 || metadata;
2903         int ret = 0;
2904
2905         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2906 #if IS_ENABLED(CONFIG_IPV6)
2907         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2908         if (ipv6) {
2909                 ret = __vxlan_sock_add(vxlan, true);
2910                 if (ret < 0 && ret != -EAFNOSUPPORT)
2911                         ipv4 = false;
2912         }
2913 #endif
2914         if (ipv4)
2915                 ret = __vxlan_sock_add(vxlan, false);
2916         if (ret < 0)
2917                 vxlan_sock_release(vxlan);
2918         return ret;
2919 }
2920
2921 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
2922                                  struct net_device **lower,
2923                                  struct vxlan_dev *old,
2924                                  struct netlink_ext_ack *extack)
2925 {
2926         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2927         struct vxlan_dev *tmp;
2928         bool use_ipv6 = false;
2929
2930         if (conf->flags & VXLAN_F_GPE) {
2931                 /* For now, allow GPE only together with
2932                  * COLLECT_METADATA. This can be relaxed later; in such
2933                  * case, the other side of the PtP link will have to be
2934                  * provided.
2935                  */
2936                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2937                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2938                         NL_SET_ERR_MSG(extack,
2939                                        "VXLAN GPE does not support this combination of attributes");
2940                         return -EINVAL;
2941                 }
2942         }
2943
2944         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
2945                 /* Unless IPv6 is explicitly requested, assume IPv4 */
2946                 conf->remote_ip.sa.sa_family = AF_INET;
2947                 conf->saddr.sa.sa_family = AF_INET;
2948         } else if (!conf->remote_ip.sa.sa_family) {
2949                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
2950         } else if (!conf->saddr.sa.sa_family) {
2951                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
2952         }
2953
2954         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
2955                 NL_SET_ERR_MSG(extack,
2956                                "Local and remote address must be from the same family");
2957                 return -EINVAL;
2958         }
2959
2960         if (vxlan_addr_multicast(&conf->saddr)) {
2961                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
2962                 return -EINVAL;
2963         }
2964
2965         if (conf->saddr.sa.sa_family == AF_INET6) {
2966                 if (!IS_ENABLED(CONFIG_IPV6)) {
2967                         NL_SET_ERR_MSG(extack,
2968                                        "IPv6 support not enabled in the kernel");
2969                         return -EPFNOSUPPORT;
2970                 }
2971                 use_ipv6 = true;
2972                 conf->flags |= VXLAN_F_IPV6;
2973
2974                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2975                         int local_type =
2976                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
2977                         int remote_type =
2978                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
2979
2980                         if (local_type & IPV6_ADDR_LINKLOCAL) {
2981                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
2982                                     (remote_type != IPV6_ADDR_ANY)) {
2983                                         NL_SET_ERR_MSG(extack,
2984                                                        "Invalid combination of local and remote address scopes");
2985                                         return -EINVAL;
2986                                 }
2987
2988                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
2989                         } else {
2990                                 if (remote_type ==
2991                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
2992                                         NL_SET_ERR_MSG(extack,
2993                                                        "Invalid combination of local and remote address scopes");
2994                                         return -EINVAL;
2995                                 }
2996
2997                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
2998                         }
2999                 }
3000         }
3001
3002         if (conf->label && !use_ipv6) {
3003                 NL_SET_ERR_MSG(extack,
3004                                "Label attribute only applies to IPv6 VXLAN devices");
3005                 return -EINVAL;
3006         }
3007
3008         if (conf->remote_ifindex) {
3009                 struct net_device *lowerdev;
3010
3011                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3012                 if (!lowerdev) {
3013                         NL_SET_ERR_MSG(extack,
3014                                        "Invalid local interface, device not found");
3015                         return -ENODEV;
3016                 }
3017
3018 #if IS_ENABLED(CONFIG_IPV6)
3019                 if (use_ipv6) {
3020                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3021                         if (idev && idev->cnf.disable_ipv6) {
3022                                 NL_SET_ERR_MSG(extack,
3023                                                "IPv6 support disabled by administrator");
3024                                 return -EPERM;
3025                         }
3026                 }
3027 #endif
3028
3029                 *lower = lowerdev;
3030         } else {
3031                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3032                         NL_SET_ERR_MSG(extack,
3033                                        "Local interface required for multicast remote destination");
3034
3035                         return -EINVAL;
3036                 }
3037
3038 #if IS_ENABLED(CONFIG_IPV6)
3039                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3040                         NL_SET_ERR_MSG(extack,
3041                                        "Local interface required for link-local local/remote addresses");
3042                         return -EINVAL;
3043                 }
3044 #endif
3045
3046                 *lower = NULL;
3047         }
3048
3049         if (!conf->dst_port) {
3050                 if (conf->flags & VXLAN_F_GPE)
3051                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3052                 else
3053                         conf->dst_port = htons(vxlan_port);
3054         }
3055
3056         if (!conf->age_interval)
3057                 conf->age_interval = FDB_AGE_DEFAULT;
3058
3059         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3060                 if (tmp == old)
3061                         continue;
3062
3063                 if (tmp->cfg.vni != conf->vni)
3064                         continue;
3065                 if (tmp->cfg.dst_port != conf->dst_port)
3066                         continue;
3067                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3068                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3069                         continue;
3070
3071                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3072                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3073                         continue;
3074
3075                 NL_SET_ERR_MSG(extack,
3076                                "A VXLAN device with the specified VNI already exists");
3077                 return -EEXIST;
3078         }
3079
3080         return 0;
3081 }
3082
3083 static void vxlan_config_apply(struct net_device *dev,
3084                                struct vxlan_config *conf,
3085                                struct net_device *lowerdev,
3086                                struct net *src_net,
3087                                bool changelink)
3088 {
3089         struct vxlan_dev *vxlan = netdev_priv(dev);
3090         struct vxlan_rdst *dst = &vxlan->default_dst;
3091         unsigned short needed_headroom = ETH_HLEN;
3092         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3093         int max_mtu = ETH_MAX_MTU;
3094
3095         if (!changelink) {
3096                 if (conf->flags & VXLAN_F_GPE)
3097                         vxlan_raw_setup(dev);
3098                 else
3099                         vxlan_ether_setup(dev);
3100
3101                 if (conf->mtu)
3102                         dev->mtu = conf->mtu;
3103
3104                 vxlan->net = src_net;
3105         }
3106
3107         dst->remote_vni = conf->vni;
3108
3109         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3110
3111         if (lowerdev) {
3112                 dst->remote_ifindex = conf->remote_ifindex;
3113
3114                 dev->gso_max_size = lowerdev->gso_max_size;
3115                 dev->gso_max_segs = lowerdev->gso_max_segs;
3116
3117                 needed_headroom = lowerdev->hard_header_len;
3118
3119                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3120                                            VXLAN_HEADROOM);
3121                 if (max_mtu < ETH_MIN_MTU)
3122                         max_mtu = ETH_MIN_MTU;
3123
3124                 if (!changelink && !conf->mtu)
3125                         dev->mtu = max_mtu;
3126         }
3127
3128         if (dev->mtu > max_mtu)
3129                 dev->mtu = max_mtu;
3130
3131         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3132                 needed_headroom += VXLAN6_HEADROOM;
3133         else
3134                 needed_headroom += VXLAN_HEADROOM;
3135         dev->needed_headroom = needed_headroom;
3136
3137         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3138 }
3139
3140 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3141                                struct vxlan_config *conf, bool changelink,
3142                                struct netlink_ext_ack *extack)
3143 {
3144         struct vxlan_dev *vxlan = netdev_priv(dev);
3145         struct net_device *lowerdev;
3146         int ret;
3147
3148         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3149         if (ret)
3150                 return ret;
3151
3152         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3153
3154         return 0;
3155 }
3156
3157 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3158                               struct vxlan_config *conf,
3159                               struct netlink_ext_ack *extack)
3160 {
3161         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3162         struct vxlan_dev *vxlan = netdev_priv(dev);
3163         int err;
3164
3165         err = vxlan_dev_configure(net, dev, conf, false, extack);
3166         if (err)
3167                 return err;
3168
3169         dev->ethtool_ops = &vxlan_ethtool_ops;
3170
3171         /* create an fdb entry for a valid default destination */
3172         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
3173                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3174                                        &vxlan->default_dst.remote_ip,
3175                                        NUD_REACHABLE | NUD_PERMANENT,
3176                                        NLM_F_EXCL | NLM_F_CREATE,
3177                                        vxlan->cfg.dst_port,
3178                                        vxlan->default_dst.remote_vni,
3179                                        vxlan->default_dst.remote_vni,
3180                                        vxlan->default_dst.remote_ifindex,
3181                                        NTF_SELF);
3182                 if (err)
3183                         return err;
3184         }
3185
3186         err = register_netdevice(dev);
3187         if (err) {
3188                 vxlan_fdb_delete_default(vxlan, vxlan->default_dst.remote_vni);
3189                 return err;
3190         }
3191
3192         list_add(&vxlan->next, &vn->vxlan_list);
3193         return 0;
3194 }
3195
3196 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3197                          struct net_device *dev, struct vxlan_config *conf,
3198                          bool changelink)
3199 {
3200         struct vxlan_dev *vxlan = netdev_priv(dev);
3201
3202         memset(conf, 0, sizeof(*conf));
3203
3204         /* if changelink operation, start with old existing cfg */
3205         if (changelink)
3206                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3207
3208         if (data[IFLA_VXLAN_ID]) {
3209                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3210
3211                 if (changelink && (vni != conf->vni))
3212                         return -EOPNOTSUPP;
3213                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3214         }
3215
3216         if (data[IFLA_VXLAN_GROUP]) {
3217                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET))
3218                         return -EOPNOTSUPP;
3219
3220                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3221                 conf->remote_ip.sa.sa_family = AF_INET;
3222         } else if (data[IFLA_VXLAN_GROUP6]) {
3223                 if (!IS_ENABLED(CONFIG_IPV6))
3224                         return -EPFNOSUPPORT;
3225
3226                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6))
3227                         return -EOPNOTSUPP;
3228
3229                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3230                 conf->remote_ip.sa.sa_family = AF_INET6;
3231         }
3232
3233         if (data[IFLA_VXLAN_LOCAL]) {
3234                 if (changelink && (conf->saddr.sa.sa_family != AF_INET))
3235                         return -EOPNOTSUPP;
3236
3237                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3238                 conf->saddr.sa.sa_family = AF_INET;
3239         } else if (data[IFLA_VXLAN_LOCAL6]) {
3240                 if (!IS_ENABLED(CONFIG_IPV6))
3241                         return -EPFNOSUPPORT;
3242
3243                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6))
3244                         return -EOPNOTSUPP;
3245
3246                 /* TODO: respect scope id */
3247                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3248                 conf->saddr.sa.sa_family = AF_INET6;
3249         }
3250
3251         if (data[IFLA_VXLAN_LINK])
3252                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3253
3254         if (data[IFLA_VXLAN_TOS])
3255                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3256
3257         if (data[IFLA_VXLAN_TTL])
3258                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3259
3260         if (data[IFLA_VXLAN_TTL_INHERIT]) {
3261                 if (changelink)
3262                         return -EOPNOTSUPP;
3263                 conf->flags |= VXLAN_F_TTL_INHERIT;
3264         }
3265
3266         if (data[IFLA_VXLAN_LABEL])
3267                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3268                              IPV6_FLOWLABEL_MASK;
3269
3270         if (data[IFLA_VXLAN_LEARNING]) {
3271                 if (nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3272                         conf->flags |= VXLAN_F_LEARN;
3273                 else
3274                         conf->flags &= ~VXLAN_F_LEARN;
3275         } else if (!changelink) {
3276                 /* default to learn on a new device */
3277                 conf->flags |= VXLAN_F_LEARN;
3278         }
3279
3280         if (data[IFLA_VXLAN_AGEING]) {
3281                 if (changelink)
3282                         return -EOPNOTSUPP;
3283                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3284         }
3285
3286         if (data[IFLA_VXLAN_PROXY]) {
3287                 if (changelink)
3288                         return -EOPNOTSUPP;
3289                 if (nla_get_u8(data[IFLA_VXLAN_PROXY]))
3290                         conf->flags |= VXLAN_F_PROXY;
3291         }
3292
3293         if (data[IFLA_VXLAN_RSC]) {
3294                 if (changelink)
3295                         return -EOPNOTSUPP;
3296                 if (nla_get_u8(data[IFLA_VXLAN_RSC]))
3297                         conf->flags |= VXLAN_F_RSC;
3298         }
3299
3300         if (data[IFLA_VXLAN_L2MISS]) {
3301                 if (changelink)
3302                         return -EOPNOTSUPP;
3303                 if (nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3304                         conf->flags |= VXLAN_F_L2MISS;
3305         }
3306
3307         if (data[IFLA_VXLAN_L3MISS]) {
3308                 if (changelink)
3309                         return -EOPNOTSUPP;
3310                 if (nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3311                         conf->flags |= VXLAN_F_L3MISS;
3312         }
3313
3314         if (data[IFLA_VXLAN_LIMIT]) {
3315                 if (changelink)
3316                         return -EOPNOTSUPP;
3317                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3318         }
3319
3320         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3321                 if (changelink)
3322                         return -EOPNOTSUPP;
3323                 if (nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3324                         conf->flags |= VXLAN_F_COLLECT_METADATA;
3325         }
3326
3327         if (data[IFLA_VXLAN_PORT_RANGE]) {
3328                 if (!changelink) {
3329                         const struct ifla_vxlan_port_range *p
3330                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3331                         conf->port_min = ntohs(p->low);
3332                         conf->port_max = ntohs(p->high);
3333                 } else {
3334                         return -EOPNOTSUPP;
3335                 }
3336         }
3337
3338         if (data[IFLA_VXLAN_PORT]) {
3339                 if (changelink)
3340                         return -EOPNOTSUPP;
3341                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3342         }
3343
3344         if (data[IFLA_VXLAN_UDP_CSUM]) {
3345                 if (changelink)
3346                         return -EOPNOTSUPP;
3347                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3348                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3349         }
3350
3351         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3352                 if (changelink)
3353                         return -EOPNOTSUPP;
3354                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3355                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3356         }
3357
3358         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3359                 if (changelink)
3360                         return -EOPNOTSUPP;
3361                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3362                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3363         }
3364
3365         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3366                 if (changelink)
3367                         return -EOPNOTSUPP;
3368                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3369                         conf->flags |= VXLAN_F_REMCSUM_TX;
3370         }
3371
3372         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3373                 if (changelink)
3374                         return -EOPNOTSUPP;
3375                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3376                         conf->flags |= VXLAN_F_REMCSUM_RX;
3377         }
3378
3379         if (data[IFLA_VXLAN_GBP]) {
3380                 if (changelink)
3381                         return -EOPNOTSUPP;
3382                 conf->flags |= VXLAN_F_GBP;
3383         }
3384
3385         if (data[IFLA_VXLAN_GPE]) {
3386                 if (changelink)
3387                         return -EOPNOTSUPP;
3388                 conf->flags |= VXLAN_F_GPE;
3389         }
3390
3391         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3392                 if (changelink)
3393                         return -EOPNOTSUPP;
3394                 conf->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3395         }
3396
3397         if (tb[IFLA_MTU]) {
3398                 if (changelink)
3399                         return -EOPNOTSUPP;
3400                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3401         }
3402
3403         return 0;
3404 }
3405
3406 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3407                          struct nlattr *tb[], struct nlattr *data[],
3408                          struct netlink_ext_ack *extack)
3409 {
3410         struct vxlan_config conf;
3411         int err;
3412
3413         err = vxlan_nl2conf(tb, data, dev, &conf, false);
3414         if (err)
3415                 return err;
3416
3417         return __vxlan_dev_create(src_net, dev, &conf, extack);
3418 }
3419
3420 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3421                             struct nlattr *data[],
3422                             struct netlink_ext_ack *extack)
3423 {
3424         struct vxlan_dev *vxlan = netdev_priv(dev);
3425         struct vxlan_rdst *dst = &vxlan->default_dst;
3426         struct vxlan_rdst old_dst;
3427         struct vxlan_config conf;
3428         int err;
3429
3430         err = vxlan_nl2conf(tb, data,
3431                             dev, &conf, true);
3432         if (err)
3433                 return err;
3434
3435         memcpy(&old_dst, dst, sizeof(struct vxlan_rdst));
3436
3437         err = vxlan_dev_configure(vxlan->net, dev, &conf, true, extack);
3438         if (err)
3439                 return err;
3440
3441         /* handle default dst entry */
3442         if (!vxlan_addr_equal(&dst->remote_ip, &old_dst.remote_ip)) {
3443                 spin_lock_bh(&vxlan->hash_lock);
3444                 if (!vxlan_addr_any(&old_dst.remote_ip))
3445                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
3446                                            old_dst.remote_ip,
3447                                            vxlan->cfg.dst_port,
3448                                            old_dst.remote_vni,
3449                                            old_dst.remote_vni,
3450                                            old_dst.remote_ifindex, 0);
3451
3452                 if (!vxlan_addr_any(&dst->remote_ip)) {
3453                         err = vxlan_fdb_create(vxlan, all_zeros_mac,
3454                                                &dst->remote_ip,
3455                                                NUD_REACHABLE | NUD_PERMANENT,
3456                                                NLM_F_CREATE | NLM_F_APPEND,
3457                                                vxlan->cfg.dst_port,
3458                                                dst->remote_vni,
3459                                                dst->remote_vni,
3460                                                dst->remote_ifindex,
3461                                                NTF_SELF);
3462                         if (err) {
3463                                 spin_unlock_bh(&vxlan->hash_lock);
3464                                 return err;
3465                         }
3466                 }
3467                 spin_unlock_bh(&vxlan->hash_lock);
3468         }
3469
3470         return 0;
3471 }
3472
3473 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3474 {
3475         struct vxlan_dev *vxlan = netdev_priv(dev);
3476
3477         vxlan_flush(vxlan, true);
3478
3479         gro_cells_destroy(&vxlan->gro_cells);
3480         list_del(&vxlan->next);
3481         unregister_netdevice_queue(dev, head);
3482 }
3483
3484 static size_t vxlan_get_size(const struct net_device *dev)
3485 {
3486
3487         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3488                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3489                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3490                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3491                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3492                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3493                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3494                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3495                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3496                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3497                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3498                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3499                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3500                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3501                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3502                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3503                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3504                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3505                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3506                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3507                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3508                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3509                 0;
3510 }
3511
3512 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3513 {
3514         const struct vxlan_dev *vxlan = netdev_priv(dev);
3515         const struct vxlan_rdst *dst = &vxlan->default_dst;
3516         struct ifla_vxlan_port_range ports = {
3517                 .low =  htons(vxlan->cfg.port_min),
3518                 .high = htons(vxlan->cfg.port_max),
3519         };
3520
3521         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3522                 goto nla_put_failure;
3523
3524         if (!vxlan_addr_any(&dst->remote_ip)) {
3525                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3526                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3527                                             dst->remote_ip.sin.sin_addr.s_addr))
3528                                 goto nla_put_failure;
3529 #if IS_ENABLED(CONFIG_IPV6)
3530                 } else {
3531                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3532                                              &dst->remote_ip.sin6.sin6_addr))
3533                                 goto nla_put_failure;
3534 #endif
3535                 }
3536         }
3537
3538         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3539                 goto nla_put_failure;
3540
3541         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3542                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3543                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3544                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3545                                 goto nla_put_failure;
3546 #if IS_ENABLED(CONFIG_IPV6)
3547                 } else {
3548                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3549                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3550                                 goto nla_put_failure;
3551 #endif
3552                 }
3553         }
3554
3555         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3556             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3557             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3558             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3559                         !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
3560             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3561                         !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
3562             nla_put_u8(skb, IFLA_VXLAN_RSC,
3563                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
3564             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3565                         !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
3566             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3567                         !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
3568             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3569                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
3570             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3571             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3572             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3573             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3574                         !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3575             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3576                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3577             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3578                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3579             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3580                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
3581             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3582                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
3583                 goto nla_put_failure;
3584
3585         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3586                 goto nla_put_failure;
3587
3588         if (vxlan->cfg.flags & VXLAN_F_GBP &&
3589             nla_put_flag(skb, IFLA_VXLAN_GBP))
3590                 goto nla_put_failure;
3591
3592         if (vxlan->cfg.flags & VXLAN_F_GPE &&
3593             nla_put_flag(skb, IFLA_VXLAN_GPE))
3594                 goto nla_put_failure;
3595
3596         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3597             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3598                 goto nla_put_failure;
3599
3600         return 0;
3601
3602 nla_put_failure:
3603         return -EMSGSIZE;
3604 }
3605
3606 static struct net *vxlan_get_link_net(const struct net_device *dev)
3607 {
3608         struct vxlan_dev *vxlan = netdev_priv(dev);
3609
3610         return vxlan->net;
3611 }
3612
3613 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3614         .kind           = "vxlan",
3615         .maxtype        = IFLA_VXLAN_MAX,
3616         .policy         = vxlan_policy,
3617         .priv_size      = sizeof(struct vxlan_dev),
3618         .setup          = vxlan_setup,
3619         .validate       = vxlan_validate,
3620         .newlink        = vxlan_newlink,
3621         .changelink     = vxlan_changelink,
3622         .dellink        = vxlan_dellink,
3623         .get_size       = vxlan_get_size,
3624         .fill_info      = vxlan_fill_info,
3625         .get_link_net   = vxlan_get_link_net,
3626 };
3627
3628 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3629                                     u8 name_assign_type,
3630                                     struct vxlan_config *conf)
3631 {
3632         struct nlattr *tb[IFLA_MAX + 1];
3633         struct net_device *dev;
3634         int err;
3635
3636         memset(&tb, 0, sizeof(tb));
3637
3638         dev = rtnl_create_link(net, name, name_assign_type,
3639                                &vxlan_link_ops, tb);
3640         if (IS_ERR(dev))
3641                 return dev;
3642
3643         err = __vxlan_dev_create(net, dev, conf, NULL);
3644         if (err < 0) {
3645                 free_netdev(dev);
3646                 return ERR_PTR(err);
3647         }
3648
3649         err = rtnl_configure_link(dev, NULL);
3650         if (err < 0) {
3651                 LIST_HEAD(list_kill);
3652
3653                 vxlan_dellink(dev, &list_kill);
3654                 unregister_netdevice_many(&list_kill);
3655                 return ERR_PTR(err);
3656         }
3657
3658         return dev;
3659 }
3660 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3661
3662 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3663                                              struct net_device *dev)
3664 {
3665         struct vxlan_dev *vxlan, *next;
3666         LIST_HEAD(list_kill);
3667
3668         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3669                 struct vxlan_rdst *dst = &vxlan->default_dst;
3670
3671                 /* In case we created vxlan device with carrier
3672                  * and we loose the carrier due to module unload
3673                  * we also need to remove vxlan device. In other
3674                  * cases, it's not necessary and remote_ifindex
3675                  * is 0 here, so no matches.
3676                  */
3677                 if (dst->remote_ifindex == dev->ifindex)
3678                         vxlan_dellink(vxlan->dev, &list_kill);
3679         }
3680
3681         unregister_netdevice_many(&list_kill);
3682 }
3683
3684 static int vxlan_netdevice_event(struct notifier_block *unused,
3685                                  unsigned long event, void *ptr)
3686 {
3687         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3688         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3689
3690         if (event == NETDEV_UNREGISTER) {
3691                 vxlan_offload_rx_ports(dev, false);
3692                 vxlan_handle_lowerdev_unregister(vn, dev);
3693         } else if (event == NETDEV_REGISTER) {
3694                 vxlan_offload_rx_ports(dev, true);
3695         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
3696                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
3697                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
3698         }
3699
3700         return NOTIFY_DONE;
3701 }
3702
3703 static struct notifier_block vxlan_notifier_block __read_mostly = {
3704         .notifier_call = vxlan_netdevice_event,
3705 };
3706
3707 static __net_init int vxlan_init_net(struct net *net)
3708 {
3709         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3710         unsigned int h;
3711
3712         INIT_LIST_HEAD(&vn->vxlan_list);
3713         spin_lock_init(&vn->sock_lock);
3714
3715         for (h = 0; h < PORT_HASH_SIZE; ++h)
3716                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3717
3718         return 0;
3719 }
3720
3721 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
3722 {
3723         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3724         struct vxlan_dev *vxlan, *next;
3725         struct net_device *dev, *aux;
3726         unsigned int h;
3727
3728         for_each_netdev_safe(net, dev, aux)
3729                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3730                         unregister_netdevice_queue(dev, head);
3731
3732         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3733                 /* If vxlan->dev is in the same netns, it has already been added
3734                  * to the list by the previous loop.
3735                  */
3736                 if (!net_eq(dev_net(vxlan->dev), net)) {
3737                         gro_cells_destroy(&vxlan->gro_cells);
3738                         unregister_netdevice_queue(vxlan->dev, head);
3739                 }
3740         }
3741
3742         for (h = 0; h < PORT_HASH_SIZE; ++h)
3743                 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
3744 }
3745
3746 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
3747 {
3748         struct net *net;
3749         LIST_HEAD(list);
3750
3751         rtnl_lock();
3752         list_for_each_entry(net, net_list, exit_list)
3753                 vxlan_destroy_tunnels(net, &list);
3754
3755         unregister_netdevice_many(&list);
3756         rtnl_unlock();
3757 }
3758
3759 static struct pernet_operations vxlan_net_ops = {
3760         .init = vxlan_init_net,
3761         .exit_batch = vxlan_exit_batch_net,
3762         .id   = &vxlan_net_id,
3763         .size = sizeof(struct vxlan_net),
3764 };
3765
3766 static int __init vxlan_init_module(void)
3767 {
3768         int rc;
3769
3770         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3771
3772         rc = register_pernet_subsys(&vxlan_net_ops);
3773         if (rc)
3774                 goto out1;
3775
3776         rc = register_netdevice_notifier(&vxlan_notifier_block);
3777         if (rc)
3778                 goto out2;
3779
3780         rc = rtnl_link_register(&vxlan_link_ops);
3781         if (rc)
3782                 goto out3;
3783
3784         return 0;
3785 out3:
3786         unregister_netdevice_notifier(&vxlan_notifier_block);
3787 out2:
3788         unregister_pernet_subsys(&vxlan_net_ops);
3789 out1:
3790         return rc;
3791 }
3792 late_initcall(vxlan_init_module);
3793
3794 static void __exit vxlan_cleanup_module(void)
3795 {
3796         rtnl_link_unregister(&vxlan_link_ops);
3797         unregister_netdevice_notifier(&vxlan_notifier_block);
3798         unregister_pernet_subsys(&vxlan_net_ops);
3799         /* rcu_barrier() is called by netns */
3800 }
3801 module_exit(vxlan_cleanup_module);
3802
3803 MODULE_LICENSE("GPL");
3804 MODULE_VERSION(VXLAN_VERSION);
3805 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3806 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3807 MODULE_ALIAS_RTNL_LINK("vxlan");