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/types.h>
  15 #include <linux/module.h>
  16 #include <linux/errno.h>
  17 #include <linux/slab.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/rculist.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/in.h>
  22 #include <linux/ip.h>
  23 #include <linux/udp.h>
  24 #include <linux/igmp.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/hash.h>
  29 #include <linux/ethtool.h>
  30 #include <net/arp.h>
  31 #include <net/ndisc.h>
  32 #include <net/ip.h>
  33 #include <net/ip_tunnels.h>
  34 #include <net/icmp.h>
  35 #include <net/udp.h>
  36 #include <net/udp_tunnel.h>
  37 #include <net/rtnetlink.h>
  38 #include <net/route.h>
  39 #include <net/dsfield.h>
  40 #include <net/inet_ecn.h>
  41 #include <net/net_namespace.h>
  42 #include <net/netns/generic.h>
  43 #include <net/vxlan.h>
  44 #include <net/protocol.h>
  45 #include <net/udp_tunnel.h>
  46 #if IS_ENABLED(CONFIG_IPV6)
  47 #include <net/ipv6.h>
  48 #include <net/addrconf.h>
  49 #include <net/ip6_tunnel.h>
  50 #include <net/ip6_checksum.h>
  51 #endif
  52 #include <net/dst_metadata.h>
  53
  54 #define VXLAN_VERSION   "0.1"
  55
  56 #define PORT_HASH_BITS  8
  57 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  58 #define FDB_AGE_DEFAULT 300 /* 5 min */
  59 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  60
  61 /* UDP port for VXLAN traffic.
  62  * The IANA assigned port is 4789, but the Linux default is 8472
  63  * for compatibility with early adopters.
  64  */
  65 static unsigned short vxlan_port __read_mostly = 8472;
  66 module_param_named(udp_port, vxlan_port, ushort, 0444);
  67 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  68
  69 static bool log_ecn_error = true;
  70 module_param(log_ecn_error, bool, 0644);
  71 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  72
  73 static int vxlan_net_id;
  74 static struct rtnl_link_ops vxlan_link_ops;
  75
  76 static const u8 all_zeros_mac[ETH_ALEN];
  77
  78 static struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
  79                                          bool no_share, u32 flags);
  80
  81 /* per-network namespace private data for this module */
  82 struct vxlan_net {
  83         struct list_head  vxlan_list;
  84         struct hlist_head sock_list[PORT_HASH_SIZE];
  85         spinlock_t        sock_lock;
  86 };
  87
  88 /* Forwarding table entry */
  89 struct vxlan_fdb {
  90         struct hlist_node hlist;        /* linked list of entries */
  91         struct rcu_head   rcu;
  92         unsigned long     updated;      /* jiffies */
  93         unsigned long     used;
  94         struct list_head  remotes;
  95         u8                eth_addr[ETH_ALEN];
  96         u16               state;        /* see ndm_state */
  97         u8                flags;        /* see ndm_flags */
  98 };
  99
 100 /* salt for hash table */
 101 static u32 vxlan_salt __read_mostly;
 102 static struct workqueue_struct *vxlan_wq;
 103
 104 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
 105 {
 106         return vs->flags & VXLAN_F_COLLECT_METADATA ||
 107                ip_tunnel_collect_metadata();
 108 }
 109
 110 #if IS_ENABLED(CONFIG_IPV6)
 111 static inline
 112 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 113 {
 114         if (a->sa.sa_family != b->sa.sa_family)
 115                 return false;
 116         if (a->sa.sa_family == AF_INET6)
 117                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 118         else
 119                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 120 }
 121
 122 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 123 {
 124         if (ipa->sa.sa_family == AF_INET6)
 125                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 126         else
 127                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 128 }
 129
 130 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 131 {
 132         if (ipa->sa.sa_family == AF_INET6)
 133                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 134         else
 135                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 136 }
 137
 138 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 139 {
 140         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 141                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 142                 ip->sa.sa_family = AF_INET6;
 143                 return 0;
 144         } else if (nla_len(nla) >= sizeof(__be32)) {
 145                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 146                 ip->sa.sa_family = AF_INET;
 147                 return 0;
 148         } else {
 149                 return -EAFNOSUPPORT;
 150         }
 151 }
 152
 153 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 154                               const union vxlan_addr *ip)
 155 {
 156         if (ip->sa.sa_family == AF_INET6)
 157                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 158         else
 159                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 160 }
 161
 162 #else /* !CONFIG_IPV6 */
 163
 164 static inline
 165 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 166 {
 167         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 168 }
 169
 170 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 171 {
 172         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 173 }
 174
 175 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 176 {
 177         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 178 }
 179
 180 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 181 {
 182         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 183                 return -EAFNOSUPPORT;
 184         } else if (nla_len(nla) >= sizeof(__be32)) {
 185                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 186                 ip->sa.sa_family = AF_INET;
 187                 return 0;
 188         } else {
 189                 return -EAFNOSUPPORT;
 190         }
 191 }
 192
 193 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 194                               const union vxlan_addr *ip)
 195 {
 196         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 197 }
 198 #endif
 199
 200 /* Virtual Network hash table head */
 201 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
 202 {
 203         return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
 204 }
 205
 206 /* Socket hash table head */
 207 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 208 {
 209         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 210
 211         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 212 }
 213
 214 /* First remote destination for a forwarding entry.
 215  * Guaranteed to be non-NULL because remotes are never deleted.
 216  */
 217 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 218 {
 219         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 220 }
 221
 222 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 223 {
 224         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 225 }
 226
 227 /* Find VXLAN socket based on network namespace, address family and UDP port
 228  * and enabled unshareable flags.
 229  */
 230 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 231                                           __be16 port, u32 flags)
 232 {
 233         struct vxlan_sock *vs;
 234
 235         flags &= VXLAN_F_RCV_FLAGS;
 236
 237         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 238                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 239                     vxlan_get_sk_family(vs) == family &&
 240                     vs->flags == flags)
 241                         return vs;
 242         }
 243         return NULL;
 244 }
 245
 246 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
 247 {
 248         struct vxlan_dev *vxlan;
 249
 250         hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
 251                 if (vxlan->default_dst.remote_vni == id)
 252                         return vxlan;
 253         }
 254
 255         return NULL;
 256 }
 257
 258 /* Look up VNI in a per net namespace table */
 259 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id,
 260                                         sa_family_t family, __be16 port,
 261                                         u32 flags)
 262 {
 263         struct vxlan_sock *vs;
 264
 265         vs = vxlan_find_sock(net, family, port, flags);
 266         if (!vs)
 267                 return NULL;
 268
 269         return vxlan_vs_find_vni(vs, id);
 270 }
 271
 272 /* Fill in neighbour message in skbuff. */
 273 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 274                           const struct vxlan_fdb *fdb,
 275                           u32 portid, u32 seq, int type, unsigned int flags,
 276                           const struct vxlan_rdst *rdst)
 277 {
 278         unsigned long now = jiffies;
 279         struct nda_cacheinfo ci;
 280         struct nlmsghdr *nlh;
 281         struct ndmsg *ndm;
 282         bool send_ip, send_eth;
 283
 284         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 285         if (nlh == NULL)
 286                 return -EMSGSIZE;
 287
 288         ndm = nlmsg_data(nlh);
 289         memset(ndm, 0, sizeof(*ndm));
 290
 291         send_eth = send_ip = true;
 292
 293         if (type == RTM_GETNEIGH) {
 294                 ndm->ndm_family = AF_INET;
 295                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 296                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 297         } else
 298                 ndm->ndm_family = AF_BRIDGE;
 299         ndm->ndm_state = fdb->state;
 300         ndm->ndm_ifindex = vxlan->dev->ifindex;
 301         ndm->ndm_flags = fdb->flags;
 302         ndm->ndm_type = RTN_UNICAST;
 303
 304         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 305             nla_put_s32(skb, NDA_LINK_NETNSID,
 306                         peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
 307                 goto nla_put_failure;
 308
 309         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 310                 goto nla_put_failure;
 311
 312         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 313                 goto nla_put_failure;
 314
 315         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 316             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 317                 goto nla_put_failure;
 318         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 319             nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
 320                 goto nla_put_failure;
 321         if (rdst->remote_ifindex &&
 322             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 323                 goto nla_put_failure;
 324
 325         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 326         ci.ndm_confirmed = 0;
 327         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 328         ci.ndm_refcnt    = 0;
 329
 330         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 331                 goto nla_put_failure;
 332
 333         nlmsg_end(skb, nlh);
 334         return 0;
 335
 336 nla_put_failure:
 337         nlmsg_cancel(skb, nlh);
 338         return -EMSGSIZE;
 339 }
 340
 341 static inline size_t vxlan_nlmsg_size(void)
 342 {
 343         return NLMSG_ALIGN(sizeof(struct ndmsg))
 344                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 345                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 346                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 347                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 348                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 349                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 350                 + nla_total_size(sizeof(struct nda_cacheinfo));
 351 }
 352
 353 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 354                              struct vxlan_rdst *rd, int type)
 355 {
 356         struct net *net = dev_net(vxlan->dev);
 357         struct sk_buff *skb;
 358         int err = -ENOBUFS;
 359
 360         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 361         if (skb == NULL)
 362                 goto errout;
 363
 364         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 365         if (err < 0) {
 366                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 367                 WARN_ON(err == -EMSGSIZE);
 368                 kfree_skb(skb);
 369                 goto errout;
 370         }
 371
 372         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 373         return;
 374 errout:
 375         if (err < 0)
 376                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 377 }
 378
 379 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 380 {
 381         struct vxlan_dev *vxlan = netdev_priv(dev);
 382         struct vxlan_fdb f = {
 383                 .state = NUD_STALE,
 384         };
 385         struct vxlan_rdst remote = {
 386                 .remote_ip = *ipa, /* goes to NDA_DST */
 387                 .remote_vni = VXLAN_N_VID,
 388         };
 389
 390         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 391 }
 392
 393 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 394 {
 395         struct vxlan_fdb f = {
 396                 .state = NUD_STALE,
 397         };
 398         struct vxlan_rdst remote = { };
 399
 400         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 401
 402         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 403 }
 404
 405 /* Hash Ethernet address */
 406 static u32 eth_hash(const unsigned char *addr)
 407 {
 408         u64 value = get_unaligned((u64 *)addr);
 409
 410         /* only want 6 bytes */
 411 #ifdef __BIG_ENDIAN
 412         value >>= 16;
 413 #else
 414         value <<= 16;
 415 #endif
 416         return hash_64(value, FDB_HASH_BITS);
 417 }
 418
 419 /* Hash chain to use given mac address */
 420 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 421                                                 const u8 *mac)
 422 {
 423         return &vxlan->fdb_head[eth_hash(mac)];
 424 }
 425
 426 /* Look up Ethernet address in forwarding table */
 427 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 428                                         const u8 *mac)
 429 {
 430         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 431         struct vxlan_fdb *f;
 432
 433         hlist_for_each_entry_rcu(f, head, hlist) {
 434                 if (ether_addr_equal(mac, f->eth_addr))
 435                         return f;
 436         }
 437
 438         return NULL;
 439 }
 440
 441 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 442                                         const u8 *mac)
 443 {
 444         struct vxlan_fdb *f;
 445
 446         f = __vxlan_find_mac(vxlan, mac);
 447         if (f)
 448                 f->used = jiffies;
 449
 450         return f;
 451 }
 452
 453 /* caller should hold vxlan->hash_lock */
 454 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 455                                               union vxlan_addr *ip, __be16 port,
 456                                               __u32 vni, __u32 ifindex)
 457 {
 458         struct vxlan_rdst *rd;
 459
 460         list_for_each_entry(rd, &f->remotes, list) {
 461                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 462                     rd->remote_port == port &&
 463                     rd->remote_vni == vni &&
 464                     rd->remote_ifindex == ifindex)
 465                         return rd;
 466         }
 467
 468         return NULL;
 469 }
 470
 471 /* Replace destination of unicast mac */
 472 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 473                              union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
 474 {
 475         struct vxlan_rdst *rd;
 476
 477         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 478         if (rd)
 479                 return 0;
 480
 481         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 482         if (!rd)
 483                 return 0;
 484         rd->remote_ip = *ip;
 485         rd->remote_port = port;
 486         rd->remote_vni = vni;
 487         rd->remote_ifindex = ifindex;
 488         return 1;
 489 }
 490
 491 /* Add/update destinations for multicast */
 492 static int vxlan_fdb_append(struct vxlan_fdb *f,
 493                             union vxlan_addr *ip, __be16 port, __u32 vni,
 494                             __u32 ifindex, struct vxlan_rdst **rdp)
 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 = kmalloc(sizeof(*rd), GFP_ATOMIC);
 503         if (rd == NULL)
 504                 return -ENOBUFS;
 505         rd->remote_ip = *ip;
 506         rd->remote_port = port;
 507         rd->remote_vni = vni;
 508         rd->remote_ifindex = ifindex;
 509
 510         list_add_tail_rcu(&rd->list, &f->remotes);
 511
 512         *rdp = rd;
 513         return 1;
 514 }
 515
 516 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 517                                           unsigned int off,
 518                                           struct vxlanhdr *vh, size_t hdrlen,
 519                                           u32 data, struct gro_remcsum *grc,
 520                                           bool nopartial)
 521 {
 522         size_t start, offset;
 523
 524         if (skb->remcsum_offload)
 525                 return vh;
 526
 527         if (!NAPI_GRO_CB(skb)->csum_valid)
 528                 return NULL;
 529
 530         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
 531         offset = start + ((data & VXLAN_RCO_UDP) ?
 532                           offsetof(struct udphdr, check) :
 533                           offsetof(struct tcphdr, check));
 534
 535         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 536                                      start, offset, grc, nopartial);
 537
 538         skb->remcsum_offload = 1;
 539
 540         return vh;
 541 }
 542
 543 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
 544                                           struct sk_buff *skb,
 545                                           struct udp_offload *uoff)
 546 {
 547         struct sk_buff *p, **pp = NULL;
 548         struct vxlanhdr *vh, *vh2;
 549         unsigned int hlen, off_vx;
 550         int flush = 1;
 551         struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
 552                                              udp_offloads);
 553         u32 flags;
 554         struct gro_remcsum grc;
 555
 556         skb_gro_remcsum_init(&grc);
 557
 558         off_vx = skb_gro_offset(skb);
 559         hlen = off_vx + sizeof(*vh);
 560         vh   = skb_gro_header_fast(skb, off_vx);
 561         if (skb_gro_header_hard(skb, hlen)) {
 562                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 563                 if (unlikely(!vh))
 564                         goto out;
 565         }
 566
 567         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 568
 569         flags = ntohl(vh->vx_flags);
 570
 571         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 572                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 573                                        ntohl(vh->vx_vni), &grc,
 574                                        !!(vs->flags &
 575                                           VXLAN_F_REMCSUM_NOPARTIAL));
 576
 577                 if (!vh)
 578                         goto out;
 579         }
 580
 581         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 582
 583         flush = 0;
 584
 585         for (p = *head; p; p = p->next) {
 586                 if (!NAPI_GRO_CB(p)->same_flow)
 587                         continue;
 588
 589                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 590                 if (vh->vx_flags != vh2->vx_flags ||
 591                     vh->vx_vni != vh2->vx_vni) {
 592                         NAPI_GRO_CB(p)->same_flow = 0;
 593                         continue;
 594                 }
 595         }
 596
 597         pp = eth_gro_receive(head, skb);
 598
 599 out:
 600         skb_gro_remcsum_cleanup(skb, &grc);
 601         NAPI_GRO_CB(skb)->flush |= flush;
 602
 603         return pp;
 604 }
 605
 606 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
 607                               struct udp_offload *uoff)
 608 {
 609         udp_tunnel_gro_complete(skb, nhoff);
 610
 611         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 612 }
 613
 614 /* Notify netdevs that UDP port started listening */
 615 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 616 {
 617         struct net_device *dev;
 618         struct sock *sk = vs->sock->sk;
 619         struct net *net = sock_net(sk);
 620         sa_family_t sa_family = vxlan_get_sk_family(vs);
 621         __be16 port = inet_sk(sk)->inet_sport;
 622         int err;
 623
 624         if (sa_family == AF_INET) {
 625                 err = udp_add_offload(&vs->udp_offloads);
 626                 if (err)
 627                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 628         }
 629
 630         rcu_read_lock();
 631         for_each_netdev_rcu(net, dev) {
 632                 if (dev->netdev_ops->ndo_add_vxlan_port)
 633                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 634                                                             port);
 635         }
 636         rcu_read_unlock();
 637 }
 638
 639 /* Notify netdevs that UDP port is no more listening */
 640 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 641 {
 642         struct net_device *dev;
 643         struct sock *sk = vs->sock->sk;
 644         struct net *net = sock_net(sk);
 645         sa_family_t sa_family = vxlan_get_sk_family(vs);
 646         __be16 port = inet_sk(sk)->inet_sport;
 647
 648         rcu_read_lock();
 649         for_each_netdev_rcu(net, dev) {
 650                 if (dev->netdev_ops->ndo_del_vxlan_port)
 651                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 652                                                             port);
 653         }
 654         rcu_read_unlock();
 655
 656         if (sa_family == AF_INET)
 657                 udp_del_offload(&vs->udp_offloads);
 658 }
 659
 660 /* Add new entry to forwarding table -- assumes lock held */
 661 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 662                             const u8 *mac, union vxlan_addr *ip,
 663                             __u16 state, __u16 flags,
 664                             __be16 port, __u32 vni, __u32 ifindex,
 665                             __u8 ndm_flags)
 666 {
 667         struct vxlan_rdst *rd = NULL;
 668         struct vxlan_fdb *f;
 669         int notify = 0;
 670
 671         f = __vxlan_find_mac(vxlan, mac);
 672         if (f) {
 673                 if (flags & NLM_F_EXCL) {
 674                         netdev_dbg(vxlan->dev,
 675                                    "lost race to create %pM\n", mac);
 676                         return -EEXIST;
 677                 }
 678                 if (f->state != state) {
 679                         f->state = state;
 680                         f->updated = jiffies;
 681                         notify = 1;
 682                 }
 683                 if (f->flags != ndm_flags) {
 684                         f->flags = ndm_flags;
 685                         f->updated = jiffies;
 686                         notify = 1;
 687                 }
 688                 if ((flags & NLM_F_REPLACE)) {
 689                         /* Only change unicasts */
 690                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 691                              is_zero_ether_addr(f->eth_addr))) {
 692                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 693                                                            ifindex);
 694                         } else
 695                                 return -EOPNOTSUPP;
 696                 }
 697                 if ((flags & NLM_F_APPEND) &&
 698                     (is_multicast_ether_addr(f->eth_addr) ||
 699                      is_zero_ether_addr(f->eth_addr))) {
 700                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 701                                                   &rd);
 702
 703                         if (rc < 0)
 704                                 return rc;
 705                         notify |= rc;
 706                 }
 707         } else {
 708                 if (!(flags & NLM_F_CREATE))
 709                         return -ENOENT;
 710
 711                 if (vxlan->cfg.addrmax &&
 712                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 713                         return -ENOSPC;
 714
 715                 /* Disallow replace to add a multicast entry */
 716                 if ((flags & NLM_F_REPLACE) &&
 717                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 718                         return -EOPNOTSUPP;
 719
 720                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 721                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 722                 if (!f)
 723                         return -ENOMEM;
 724
 725                 notify = 1;
 726                 f->state = state;
 727                 f->flags = ndm_flags;
 728                 f->updated = f->used = jiffies;
 729                 INIT_LIST_HEAD(&f->remotes);
 730                 memcpy(f->eth_addr, mac, ETH_ALEN);
 731
 732                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 733
 734                 ++vxlan->addrcnt;
 735                 hlist_add_head_rcu(&f->hlist,
 736                                    vxlan_fdb_head(vxlan, mac));
 737         }
 738
 739         if (notify) {
 740                 if (rd == NULL)
 741                         rd = first_remote_rtnl(f);
 742                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 743         }
 744
 745         return 0;
 746 }
 747
 748 static void vxlan_fdb_free(struct rcu_head *head)
 749 {
 750         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 751         struct vxlan_rdst *rd, *nd;
 752
 753         list_for_each_entry_safe(rd, nd, &f->remotes, list)
 754                 kfree(rd);
 755         kfree(f);
 756 }
 757
 758 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 759 {
 760         netdev_dbg(vxlan->dev,
 761                     "delete %pM\n", f->eth_addr);
 762
 763         --vxlan->addrcnt;
 764         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 765
 766         hlist_del_rcu(&f->hlist);
 767         call_rcu(&f->rcu, vxlan_fdb_free);
 768 }
 769
 770 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 771                            union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
 772 {
 773         struct net *net = dev_net(vxlan->dev);
 774         int err;
 775
 776         if (tb[NDA_DST]) {
 777                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 778                 if (err)
 779                         return err;
 780         } else {
 781                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 782                 if (remote->sa.sa_family == AF_INET) {
 783                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 784                         ip->sa.sa_family = AF_INET;
 785 #if IS_ENABLED(CONFIG_IPV6)
 786                 } else {
 787                         ip->sin6.sin6_addr = in6addr_any;
 788                         ip->sa.sa_family = AF_INET6;
 789 #endif
 790                 }
 791         }
 792
 793         if (tb[NDA_PORT]) {
 794                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 795                         return -EINVAL;
 796                 *port = nla_get_be16(tb[NDA_PORT]);
 797         } else {
 798                 *port = vxlan->cfg.dst_port;
 799         }
 800
 801         if (tb[NDA_VNI]) {
 802                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 803                         return -EINVAL;
 804                 *vni = nla_get_u32(tb[NDA_VNI]);
 805         } else {
 806                 *vni = vxlan->default_dst.remote_vni;
 807         }
 808
 809         if (tb[NDA_IFINDEX]) {
 810                 struct net_device *tdev;
 811
 812                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 813                         return -EINVAL;
 814                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 815                 tdev = __dev_get_by_index(net, *ifindex);
 816                 if (!tdev)
 817                         return -EADDRNOTAVAIL;
 818         } else {
 819                 *ifindex = 0;
 820         }
 821
 822         return 0;
 823 }
 824
 825 /* Add static entry (via netlink) */
 826 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 827                          struct net_device *dev,
 828                          const unsigned char *addr, u16 vid, u16 flags)
 829 {
 830         struct vxlan_dev *vxlan = netdev_priv(dev);
 831         /* struct net *net = dev_net(vxlan->dev); */
 832         union vxlan_addr ip;
 833         __be16 port;
 834         u32 vni, ifindex;
 835         int err;
 836
 837         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 838                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 839                         ndm->ndm_state);
 840                 return -EINVAL;
 841         }
 842
 843         if (tb[NDA_DST] == NULL)
 844                 return -EINVAL;
 845
 846         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 847         if (err)
 848                 return err;
 849
 850         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 851                 return -EAFNOSUPPORT;
 852
 853         spin_lock_bh(&vxlan->hash_lock);
 854         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 855                                port, vni, ifindex, ndm->ndm_flags);
 856         spin_unlock_bh(&vxlan->hash_lock);
 857
 858         return err;
 859 }
 860
 861 /* Delete entry (via netlink) */
 862 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 863                             struct net_device *dev,
 864                             const unsigned char *addr, u16 vid)
 865 {
 866         struct vxlan_dev *vxlan = netdev_priv(dev);
 867         struct vxlan_fdb *f;
 868         struct vxlan_rdst *rd = NULL;
 869         union vxlan_addr ip;
 870         __be16 port;
 871         u32 vni, ifindex;
 872         int err;
 873
 874         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 875         if (err)
 876                 return err;
 877
 878         err = -ENOENT;
 879
 880         spin_lock_bh(&vxlan->hash_lock);
 881         f = vxlan_find_mac(vxlan, addr);
 882         if (!f)
 883                 goto out;
 884
 885         if (!vxlan_addr_any(&ip)) {
 886                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 887                 if (!rd)
 888                         goto out;
 889         }
 890
 891         err = 0;
 892
 893         /* remove a destination if it's not the only one on the list,
 894          * otherwise destroy the fdb entry
 895          */
 896         if (rd && !list_is_singular(&f->remotes)) {
 897                 list_del_rcu(&rd->list);
 898                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 899                 kfree_rcu(rd, rcu);
 900                 goto out;
 901         }
 902
 903         vxlan_fdb_destroy(vxlan, f);
 904
 905 out:
 906         spin_unlock_bh(&vxlan->hash_lock);
 907
 908         return err;
 909 }
 910
 911 /* Dump forwarding table */
 912 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 913                           struct net_device *dev,
 914                           struct net_device *filter_dev, int idx)
 915 {
 916         struct vxlan_dev *vxlan = netdev_priv(dev);
 917         unsigned int h;
 918
 919         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 920                 struct vxlan_fdb *f;
 921                 int err;
 922
 923                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 924                         struct vxlan_rdst *rd;
 925
 926                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 927                                 if (idx < cb->args[0])
 928                                         goto skip;
 929
 930                                 err = vxlan_fdb_info(skb, vxlan, f,
 931                                                      NETLINK_CB(cb->skb).portid,
 932                                                      cb->nlh->nlmsg_seq,
 933                                                      RTM_NEWNEIGH,
 934                                                      NLM_F_MULTI, rd);
 935                                 if (err < 0)
 936                                         goto out;
 937 skip:
 938                                 ++idx;
 939                         }
 940                 }
 941         }
 942 out:
 943         return idx;
 944 }
 945
 946 /* Watch incoming packets to learn mapping between Ethernet address
 947  * and Tunnel endpoint.
 948  * Return true if packet is bogus and should be dropped.
 949  */
 950 static bool vxlan_snoop(struct net_device *dev,
 951                         union vxlan_addr *src_ip, const u8 *src_mac)
 952 {
 953         struct vxlan_dev *vxlan = netdev_priv(dev);
 954         struct vxlan_fdb *f;
 955
 956         f = vxlan_find_mac(vxlan, src_mac);
 957         if (likely(f)) {
 958                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 959
 960                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 961                         return false;
 962
 963                 /* Don't migrate static entries, drop packets */
 964                 if (f->state & NUD_NOARP)
 965                         return true;
 966
 967                 if (net_ratelimit())
 968                         netdev_info(dev,
 969                                     "%pM migrated from %pIS to %pIS\n",
 970                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 971
 972                 rdst->remote_ip = *src_ip;
 973                 f->updated = jiffies;
 974                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 975         } else {
 976                 /* learned new entry */
 977                 spin_lock(&vxlan->hash_lock);
 978
 979                 /* close off race between vxlan_flush and incoming packets */
 980                 if (netif_running(dev))
 981                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 982                                          NUD_REACHABLE,
 983                                          NLM_F_EXCL|NLM_F_CREATE,
 984                                          vxlan->cfg.dst_port,
 985                                          vxlan->default_dst.remote_vni,
 986                                          0, NTF_SELF);
 987                 spin_unlock(&vxlan->hash_lock);
 988         }
 989
 990         return false;
 991 }
 992
 993 /* See if multicast group is already in use by other ID */
 994 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
 995 {
 996         struct vxlan_dev *vxlan;
 997
 998         /* The vxlan_sock is only used by dev, leaving group has
 999          * no effect on other vxlan devices.
1000          */
1001         if (atomic_read(&dev->vn_sock->refcnt) == 1)
1002                 return false;
1003
1004         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1005                 if (!netif_running(vxlan->dev) || vxlan == dev)
1006                         continue;
1007
1008                 if (vxlan->vn_sock != dev->vn_sock)
1009                         continue;
1010
1011                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1012                                       &dev->default_dst.remote_ip))
1013                         continue;
1014
1015                 if (vxlan->default_dst.remote_ifindex !=
1016                     dev->default_dst.remote_ifindex)
1017                         continue;
1018
1019                 return true;
1020         }
1021
1022         return false;
1023 }
1024
1025 static void vxlan_sock_release(struct vxlan_sock *vs)
1026 {
1027         struct sock *sk = vs->sock->sk;
1028         struct net *net = sock_net(sk);
1029         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1030
1031         if (!atomic_dec_and_test(&vs->refcnt))
1032                 return;
1033
1034         spin_lock(&vn->sock_lock);
1035         hlist_del_rcu(&vs->hlist);
1036         vxlan_notify_del_rx_port(vs);
1037         spin_unlock(&vn->sock_lock);
1038
1039         queue_work(vxlan_wq, &vs->del_work);
1040 }
1041
1042 /* Update multicast group membership when first VNI on
1043  * multicast address is brought up
1044  */
1045 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1046 {
1047         struct vxlan_sock *vs = vxlan->vn_sock;
1048         struct sock *sk = vs->sock->sk;
1049         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1050         int ifindex = vxlan->default_dst.remote_ifindex;
1051         int ret = -EINVAL;
1052
1053         lock_sock(sk);
1054         if (ip->sa.sa_family == AF_INET) {
1055                 struct ip_mreqn mreq = {
1056                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1057                         .imr_ifindex            = ifindex,
1058                 };
1059
1060                 ret = ip_mc_join_group(sk, &mreq);
1061 #if IS_ENABLED(CONFIG_IPV6)
1062         } else {
1063                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1064                                                    &ip->sin6.sin6_addr);
1065 #endif
1066         }
1067         release_sock(sk);
1068
1069         return ret;
1070 }
1071
1072 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1073 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1074 {
1075         struct vxlan_sock *vs = vxlan->vn_sock;
1076         struct sock *sk = vs->sock->sk;
1077         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1078         int ifindex = vxlan->default_dst.remote_ifindex;
1079         int ret = -EINVAL;
1080
1081         lock_sock(sk);
1082         if (ip->sa.sa_family == AF_INET) {
1083                 struct ip_mreqn mreq = {
1084                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1085                         .imr_ifindex            = ifindex,
1086                 };
1087
1088                 ret = ip_mc_leave_group(sk, &mreq);
1089 #if IS_ENABLED(CONFIG_IPV6)
1090         } else {
1091                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1092                                                    &ip->sin6.sin6_addr);
1093 #endif
1094         }
1095         release_sock(sk);
1096
1097         return ret;
1098 }
1099
1100 static struct vxlanhdr *vxlan_remcsum(struct sk_buff *skb, struct vxlanhdr *vh,
1101                                       size_t hdrlen, u32 data, bool nopartial)
1102 {
1103         size_t start, offset, plen;
1104
1105         if (skb->remcsum_offload)
1106                 return vh;
1107
1108         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
1109         offset = start + ((data & VXLAN_RCO_UDP) ?
1110                           offsetof(struct udphdr, check) :
1111                           offsetof(struct tcphdr, check));
1112
1113         plen = hdrlen + offset + sizeof(u16);
1114
1115         if (!pskb_may_pull(skb, plen))
1116                 return NULL;
1117
1118         vh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1119
1120         skb_remcsum_process(skb, (void *)vh + hdrlen, start, offset,
1121                             nopartial);
1122
1123         return vh;
1124 }
1125
1126 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1127                       struct vxlan_metadata *md, u32 vni,
1128                       struct metadata_dst *tun_dst)
1129 {
1130         struct iphdr *oip = NULL;
1131         struct ipv6hdr *oip6 = NULL;
1132         struct vxlan_dev *vxlan;
1133         struct pcpu_sw_netstats *stats;
1134         union vxlan_addr saddr;
1135         int err = 0;
1136         union vxlan_addr *remote_ip;
1137
1138         /* For flow based devices, map all packets to VNI 0 */
1139         if (vs->flags & VXLAN_F_COLLECT_METADATA)
1140                 vni = 0;
1141
1142         /* Is this VNI defined? */
1143         vxlan = vxlan_vs_find_vni(vs, vni);
1144         if (!vxlan)
1145                 goto drop;
1146
1147         remote_ip = &vxlan->default_dst.remote_ip;
1148         skb_reset_mac_header(skb);
1149         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1150         skb->protocol = eth_type_trans(skb, vxlan->dev);
1151         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1152
1153         /* Ignore packet loops (and multicast echo) */
1154         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1155                 goto drop;
1156
1157         /* Re-examine inner Ethernet packet */
1158         if (remote_ip->sa.sa_family == AF_INET) {
1159                 oip = ip_hdr(skb);
1160                 saddr.sin.sin_addr.s_addr = oip->saddr;
1161                 saddr.sa.sa_family = AF_INET;
1162 #if IS_ENABLED(CONFIG_IPV6)
1163         } else {
1164                 oip6 = ipv6_hdr(skb);
1165                 saddr.sin6.sin6_addr = oip6->saddr;
1166                 saddr.sa.sa_family = AF_INET6;
1167 #endif
1168         }
1169
1170         if (tun_dst) {
1171                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1172                 tun_dst = NULL;
1173         }
1174
1175         if ((vxlan->flags & VXLAN_F_LEARN) &&
1176             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1177                 goto drop;
1178
1179         skb_reset_network_header(skb);
1180         /* In flow-based mode, GBP is carried in dst_metadata */
1181         if (!(vs->flags & VXLAN_F_COLLECT_METADATA))
1182                 skb->mark = md->gbp;
1183
1184         if (oip6)
1185                 err = IP6_ECN_decapsulate(oip6, skb);
1186         if (oip)
1187                 err = IP_ECN_decapsulate(oip, skb);
1188
1189         if (unlikely(err)) {
1190                 if (log_ecn_error) {
1191                         if (oip6)
1192                                 net_info_ratelimited("non-ECT from %pI6\n",
1193                                                      &oip6->saddr);
1194                         if (oip)
1195                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1196                                                      &oip->saddr, oip->tos);
1197                 }
1198                 if (err > 1) {
1199                         ++vxlan->dev->stats.rx_frame_errors;
1200                         ++vxlan->dev->stats.rx_errors;
1201                         goto drop;
1202                 }
1203         }
1204
1205         stats = this_cpu_ptr(vxlan->dev->tstats);
1206         u64_stats_update_begin(&stats->syncp);
1207         stats->rx_packets++;
1208         stats->rx_bytes += skb->len;
1209         u64_stats_update_end(&stats->syncp);
1210
1211         gro_cells_receive(&vxlan->gro_cells, skb);
1212
1213         return;
1214 drop:
1215         if (tun_dst)
1216                 dst_release((struct dst_entry *)tun_dst);
1217
1218         /* Consume bad packet */
1219         kfree_skb(skb);
1220 }
1221
1222 /* Callback from net/ipv4/udp.c to receive packets */
1223 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1224 {
1225         struct metadata_dst *tun_dst = NULL;
1226         struct vxlan_sock *vs;
1227         struct vxlanhdr *vxh;
1228         u32 flags, vni;
1229         struct vxlan_metadata _md;
1230         struct vxlan_metadata *md = &_md;
1231
1232         /* Need Vxlan and inner Ethernet header to be present */
1233         if (!pskb_may_pull(skb, VXLAN_HLEN))
1234                 goto error;
1235
1236         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1237         flags = ntohl(vxh->vx_flags);
1238         vni = ntohl(vxh->vx_vni);
1239
1240         if (flags & VXLAN_HF_VNI) {
1241                 flags &= ~VXLAN_HF_VNI;
1242         } else {
1243                 /* VNI flag always required to be set */
1244                 goto bad_flags;
1245         }
1246
1247         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1248                 goto drop;
1249         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1250
1251         vs = rcu_dereference_sk_user_data(sk);
1252         if (!vs)
1253                 goto drop;
1254
1255         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1256                 vxh = vxlan_remcsum(skb, vxh, sizeof(struct vxlanhdr), vni,
1257                                     !!(vs->flags & VXLAN_F_REMCSUM_NOPARTIAL));
1258                 if (!vxh)
1259                         goto drop;
1260
1261                 flags &= ~VXLAN_HF_RCO;
1262                 vni &= VXLAN_VNI_MASK;
1263         }
1264
1265         if (vxlan_collect_metadata(vs)) {
1266                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1267                                          cpu_to_be64(vni >> 8), sizeof(*md));
1268
1269                 if (!tun_dst)
1270                         goto drop;
1271
1272                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1273         } else {
1274                 memset(md, 0, sizeof(*md));
1275         }
1276
1277         /* For backwards compatibility, only allow reserved fields to be
1278          * used by VXLAN extensions if explicitly requested.
1279          */
1280         if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1281                 struct vxlanhdr_gbp *gbp;
1282
1283                 gbp = (struct vxlanhdr_gbp *)vxh;
1284                 md->gbp = ntohs(gbp->policy_id);
1285
1286                 if (tun_dst)
1287                         tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1288
1289                 if (gbp->dont_learn)
1290                         md->gbp |= VXLAN_GBP_DONT_LEARN;
1291
1292                 if (gbp->policy_applied)
1293                         md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1294
1295                 flags &= ~VXLAN_GBP_USED_BITS;
1296         }
1297
1298         if (flags || vni & ~VXLAN_VNI_MASK) {
1299                 /* If there are any unprocessed flags remaining treat
1300                  * this as a malformed packet. This behavior diverges from
1301                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1302                  * in reserved fields are to be ignored. The approach here
1303                  * maintains compatibility with previous stack code, and also
1304                  * is more robust and provides a little more security in
1305                  * adding extensions to VXLAN.
1306                  */
1307
1308                 goto bad_flags;
1309         }
1310
1311         vxlan_rcv(vs, skb, md, vni >> 8, tun_dst);
1312         return 0;
1313
1314 drop:
1315         /* Consume bad packet */
1316         kfree_skb(skb);
1317         return 0;
1318
1319 bad_flags:
1320         netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1321                    ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1322
1323 error:
1324         if (tun_dst)
1325                 dst_release((struct dst_entry *)tun_dst);
1326
1327         /* Return non vxlan pkt */
1328         return 1;
1329 }
1330
1331 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1332 {
1333         struct vxlan_dev *vxlan = netdev_priv(dev);
1334         struct arphdr *parp;
1335         u8 *arpptr, *sha;
1336         __be32 sip, tip;
1337         struct neighbour *n;
1338
1339         if (dev->flags & IFF_NOARP)
1340                 goto out;
1341
1342         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1343                 dev->stats.tx_dropped++;
1344                 goto out;
1345         }
1346         parp = arp_hdr(skb);
1347
1348         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1349              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1350             parp->ar_pro != htons(ETH_P_IP) ||
1351             parp->ar_op != htons(ARPOP_REQUEST) ||
1352             parp->ar_hln != dev->addr_len ||
1353             parp->ar_pln != 4)
1354                 goto out;
1355         arpptr = (u8 *)parp + sizeof(struct arphdr);
1356         sha = arpptr;
1357         arpptr += dev->addr_len;        /* sha */
1358         memcpy(&sip, arpptr, sizeof(sip));
1359         arpptr += sizeof(sip);
1360         arpptr += dev->addr_len;        /* tha */
1361         memcpy(&tip, arpptr, sizeof(tip));
1362
1363         if (ipv4_is_loopback(tip) ||
1364             ipv4_is_multicast(tip))
1365                 goto out;
1366
1367         n = neigh_lookup(&arp_tbl, &tip, dev);
1368
1369         if (n) {
1370                 struct vxlan_fdb *f;
1371                 struct sk_buff  *reply;
1372
1373                 if (!(n->nud_state & NUD_CONNECTED)) {
1374                         neigh_release(n);
1375                         goto out;
1376                 }
1377
1378                 f = vxlan_find_mac(vxlan, n->ha);
1379                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1380                         /* bridge-local neighbor */
1381                         neigh_release(n);
1382                         goto out;
1383                 }
1384
1385                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1386                                 n->ha, sha);
1387
1388                 neigh_release(n);
1389
1390                 if (reply == NULL)
1391                         goto out;
1392
1393                 skb_reset_mac_header(reply);
1394                 __skb_pull(reply, skb_network_offset(reply));
1395                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1396                 reply->pkt_type = PACKET_HOST;
1397
1398                 if (netif_rx_ni(reply) == NET_RX_DROP)
1399                         dev->stats.rx_dropped++;
1400         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1401                 union vxlan_addr ipa = {
1402                         .sin.sin_addr.s_addr = tip,
1403                         .sin.sin_family = AF_INET,
1404                 };
1405
1406                 vxlan_ip_miss(dev, &ipa);
1407         }
1408 out:
1409         consume_skb(skb);
1410         return NETDEV_TX_OK;
1411 }
1412
1413 #if IS_ENABLED(CONFIG_IPV6)
1414 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1415         struct neighbour *n, bool isrouter)
1416 {
1417         struct net_device *dev = request->dev;
1418         struct sk_buff *reply;
1419         struct nd_msg *ns, *na;
1420         struct ipv6hdr *pip6;
1421         u8 *daddr;
1422         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1423         int ns_olen;
1424         int i, len;
1425
1426         if (dev == NULL)
1427                 return NULL;
1428
1429         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1430                 sizeof(*na) + na_olen + dev->needed_tailroom;
1431         reply = alloc_skb(len, GFP_ATOMIC);
1432         if (reply == NULL)
1433                 return NULL;
1434
1435         reply->protocol = htons(ETH_P_IPV6);
1436         reply->dev = dev;
1437         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1438         skb_push(reply, sizeof(struct ethhdr));
1439         skb_set_mac_header(reply, 0);
1440
1441         ns = (struct nd_msg *)skb_transport_header(request);
1442
1443         daddr = eth_hdr(request)->h_source;
1444         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1445         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1446                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1447                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1448                         break;
1449                 }
1450         }
1451
1452         /* Ethernet header */
1453         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1454         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1455         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1456         reply->protocol = htons(ETH_P_IPV6);
1457
1458         skb_pull(reply, sizeof(struct ethhdr));
1459         skb_set_network_header(reply, 0);
1460         skb_put(reply, sizeof(struct ipv6hdr));
1461
1462         /* IPv6 header */
1463
1464         pip6 = ipv6_hdr(reply);
1465         memset(pip6, 0, sizeof(struct ipv6hdr));
1466         pip6->version = 6;
1467         pip6->priority = ipv6_hdr(request)->priority;
1468         pip6->nexthdr = IPPROTO_ICMPV6;
1469         pip6->hop_limit = 255;
1470         pip6->daddr = ipv6_hdr(request)->saddr;
1471         pip6->saddr = *(struct in6_addr *)n->primary_key;
1472
1473         skb_pull(reply, sizeof(struct ipv6hdr));
1474         skb_set_transport_header(reply, 0);
1475
1476         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1477
1478         /* Neighbor Advertisement */
1479         memset(na, 0, sizeof(*na)+na_olen);
1480         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1481         na->icmph.icmp6_router = isrouter;
1482         na->icmph.icmp6_override = 1;
1483         na->icmph.icmp6_solicited = 1;
1484         na->target = ns->target;
1485         ether_addr_copy(&na->opt[2], n->ha);
1486         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1487         na->opt[1] = na_olen >> 3;
1488
1489         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1490                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1491                 csum_partial(na, sizeof(*na)+na_olen, 0));
1492
1493         pip6->payload_len = htons(sizeof(*na)+na_olen);
1494
1495         skb_push(reply, sizeof(struct ipv6hdr));
1496
1497         reply->ip_summed = CHECKSUM_UNNECESSARY;
1498
1499         return reply;
1500 }
1501
1502 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1503 {
1504         struct vxlan_dev *vxlan = netdev_priv(dev);
1505         struct nd_msg *msg;
1506         const struct ipv6hdr *iphdr;
1507         const struct in6_addr *saddr, *daddr;
1508         struct neighbour *n;
1509         struct inet6_dev *in6_dev;
1510
1511         in6_dev = __in6_dev_get(dev);
1512         if (!in6_dev)
1513                 goto out;
1514
1515         iphdr = ipv6_hdr(skb);
1516         saddr = &iphdr->saddr;
1517         daddr = &iphdr->daddr;
1518
1519         msg = (struct nd_msg *)skb_transport_header(skb);
1520         if (msg->icmph.icmp6_code != 0 ||
1521             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1522                 goto out;
1523
1524         if (ipv6_addr_loopback(daddr) ||
1525             ipv6_addr_is_multicast(&msg->target))
1526                 goto out;
1527
1528         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1529
1530         if (n) {
1531                 struct vxlan_fdb *f;
1532                 struct sk_buff *reply;
1533
1534                 if (!(n->nud_state & NUD_CONNECTED)) {
1535                         neigh_release(n);
1536                         goto out;
1537                 }
1538
1539                 f = vxlan_find_mac(vxlan, n->ha);
1540                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1541                         /* bridge-local neighbor */
1542                         neigh_release(n);
1543                         goto out;
1544                 }
1545
1546                 reply = vxlan_na_create(skb, n,
1547                                         !!(f ? f->flags & NTF_ROUTER : 0));
1548
1549                 neigh_release(n);
1550
1551                 if (reply == NULL)
1552                         goto out;
1553
1554                 if (netif_rx_ni(reply) == NET_RX_DROP)
1555                         dev->stats.rx_dropped++;
1556
1557         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1558                 union vxlan_addr ipa = {
1559                         .sin6.sin6_addr = msg->target,
1560                         .sin6.sin6_family = AF_INET6,
1561                 };
1562
1563                 vxlan_ip_miss(dev, &ipa);
1564         }
1565
1566 out:
1567         consume_skb(skb);
1568         return NETDEV_TX_OK;
1569 }
1570 #endif
1571
1572 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1573 {
1574         struct vxlan_dev *vxlan = netdev_priv(dev);
1575         struct neighbour *n;
1576
1577         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1578                 return false;
1579
1580         n = NULL;
1581         switch (ntohs(eth_hdr(skb)->h_proto)) {
1582         case ETH_P_IP:
1583         {
1584                 struct iphdr *pip;
1585
1586                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1587                         return false;
1588                 pip = ip_hdr(skb);
1589                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1590                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1591                         union vxlan_addr ipa = {
1592                                 .sin.sin_addr.s_addr = pip->daddr,
1593                                 .sin.sin_family = AF_INET,
1594                         };
1595
1596                         vxlan_ip_miss(dev, &ipa);
1597                         return false;
1598                 }
1599
1600                 break;
1601         }
1602 #if IS_ENABLED(CONFIG_IPV6)
1603         case ETH_P_IPV6:
1604         {
1605                 struct ipv6hdr *pip6;
1606
1607                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1608                         return false;
1609                 pip6 = ipv6_hdr(skb);
1610                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1611                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1612                         union vxlan_addr ipa = {
1613                                 .sin6.sin6_addr = pip6->daddr,
1614                                 .sin6.sin6_family = AF_INET6,
1615                         };
1616
1617                         vxlan_ip_miss(dev, &ipa);
1618                         return false;
1619                 }
1620
1621                 break;
1622         }
1623 #endif
1624         default:
1625                 return false;
1626         }
1627
1628         if (n) {
1629                 bool diff;
1630
1631                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1632                 if (diff) {
1633                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1634                                 dev->addr_len);
1635                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1636                 }
1637                 neigh_release(n);
1638                 return diff;
1639         }
1640
1641         return false;
1642 }
1643
1644 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1645                                 struct vxlan_metadata *md)
1646 {
1647         struct vxlanhdr_gbp *gbp;
1648
1649         if (!md->gbp)
1650                 return;
1651
1652         gbp = (struct vxlanhdr_gbp *)vxh;
1653         vxh->vx_flags |= htonl(VXLAN_HF_GBP);
1654
1655         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1656                 gbp->dont_learn = 1;
1657
1658         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1659                 gbp->policy_applied = 1;
1660
1661         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1662 }
1663
1664 #if IS_ENABLED(CONFIG_IPV6)
1665 static int vxlan6_xmit_skb(struct dst_entry *dst, struct sock *sk,
1666                            struct sk_buff *skb,
1667                            struct net_device *dev, struct in6_addr *saddr,
1668                            struct in6_addr *daddr, __u8 prio, __u8 ttl,
1669                            __be16 src_port, __be16 dst_port, __be32 vni,
1670                            struct vxlan_metadata *md, bool xnet, u32 vxflags)
1671 {
1672         struct vxlanhdr *vxh;
1673         int min_headroom;
1674         int err;
1675         bool udp_sum = !(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX);
1676         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1677         u16 hdrlen = sizeof(struct vxlanhdr);
1678
1679         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1680             skb->ip_summed == CHECKSUM_PARTIAL) {
1681                 int csum_start = skb_checksum_start_offset(skb);
1682
1683                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1684                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1685                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1686                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1687                         udp_sum = false;
1688                         type |= SKB_GSO_TUNNEL_REMCSUM;
1689                 }
1690         }
1691
1692         skb_scrub_packet(skb, xnet);
1693
1694         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1695                         + VXLAN_HLEN + sizeof(struct ipv6hdr)
1696                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1697
1698         /* Need space for new headers (invalidates iph ptr) */
1699         err = skb_cow_head(skb, min_headroom);
1700         if (unlikely(err)) {
1701                 kfree_skb(skb);
1702                 goto err;
1703         }
1704
1705         skb = vlan_hwaccel_push_inside(skb);
1706         if (WARN_ON(!skb)) {
1707                 err = -ENOMEM;
1708                 goto err;
1709         }
1710
1711         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1712         if (IS_ERR(skb)) {
1713                 err = -EINVAL;
1714                 goto err;
1715         }
1716
1717         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1718         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1719         vxh->vx_vni = vni;
1720
1721         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1722                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1723                            VXLAN_RCO_SHIFT;
1724
1725                 if (skb->csum_offset == offsetof(struct udphdr, check))
1726                         data |= VXLAN_RCO_UDP;
1727
1728                 vxh->vx_vni |= htonl(data);
1729                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1730
1731                 if (!skb_is_gso(skb)) {
1732                         skb->ip_summed = CHECKSUM_NONE;
1733                         skb->encapsulation = 0;
1734                 }
1735         }
1736
1737         if (vxflags & VXLAN_F_GBP)
1738                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1739
1740         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1741
1742         udp_tunnel6_xmit_skb(dst, sk, skb, dev, saddr, daddr, prio,
1743                              ttl, src_port, dst_port,
1744                              !!(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX));
1745         return 0;
1746 err:
1747         dst_release(dst);
1748         return err;
1749 }
1750 #endif
1751
1752 static int vxlan_xmit_skb(struct rtable *rt, struct sock *sk, struct sk_buff *skb,
1753                           __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1754                           __be16 src_port, __be16 dst_port, __be32 vni,
1755                           struct vxlan_metadata *md, bool xnet, u32 vxflags)
1756 {
1757         struct vxlanhdr *vxh;
1758         int min_headroom;
1759         int err;
1760         bool udp_sum = !!(vxflags & VXLAN_F_UDP_CSUM);
1761         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1762         u16 hdrlen = sizeof(struct vxlanhdr);
1763
1764         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1765             skb->ip_summed == CHECKSUM_PARTIAL) {
1766                 int csum_start = skb_checksum_start_offset(skb);
1767
1768                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1769                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1770                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1771                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1772                         udp_sum = false;
1773                         type |= SKB_GSO_TUNNEL_REMCSUM;
1774                 }
1775         }
1776
1777         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1778                         + VXLAN_HLEN + sizeof(struct iphdr)
1779                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1780
1781         /* Need space for new headers (invalidates iph ptr) */
1782         err = skb_cow_head(skb, min_headroom);
1783         if (unlikely(err)) {
1784                 kfree_skb(skb);
1785                 return err;
1786         }
1787
1788         skb = vlan_hwaccel_push_inside(skb);
1789         if (WARN_ON(!skb))
1790                 return -ENOMEM;
1791
1792         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1793         if (IS_ERR(skb))
1794                 return PTR_ERR(skb);
1795
1796         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1797         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1798         vxh->vx_vni = vni;
1799
1800         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1801                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1802                            VXLAN_RCO_SHIFT;
1803
1804                 if (skb->csum_offset == offsetof(struct udphdr, check))
1805                         data |= VXLAN_RCO_UDP;
1806
1807                 vxh->vx_vni |= htonl(data);
1808                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1809
1810                 if (!skb_is_gso(skb)) {
1811                         skb->ip_summed = CHECKSUM_NONE;
1812                         skb->encapsulation = 0;
1813                 }
1814         }
1815
1816         if (vxflags & VXLAN_F_GBP)
1817                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1818
1819         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1820
1821         return udp_tunnel_xmit_skb(rt, sk, skb, src, dst, tos,
1822                                    ttl, df, src_port, dst_port, xnet,
1823                                    !(vxflags & VXLAN_F_UDP_CSUM));
1824 }
1825
1826 /* Bypass encapsulation if the destination is local */
1827 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1828                                struct vxlan_dev *dst_vxlan)
1829 {
1830         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1831         union vxlan_addr loopback;
1832         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1833         struct net_device *dev = skb->dev;
1834         int len = skb->len;
1835
1836         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1837         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1838         skb->pkt_type = PACKET_HOST;
1839         skb->encapsulation = 0;
1840         skb->dev = dst_vxlan->dev;
1841         __skb_pull(skb, skb_network_offset(skb));
1842
1843         if (remote_ip->sa.sa_family == AF_INET) {
1844                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1845                 loopback.sa.sa_family =  AF_INET;
1846 #if IS_ENABLED(CONFIG_IPV6)
1847         } else {
1848                 loopback.sin6.sin6_addr = in6addr_loopback;
1849                 loopback.sa.sa_family =  AF_INET6;
1850 #endif
1851         }
1852
1853         if (dst_vxlan->flags & VXLAN_F_LEARN)
1854                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1855
1856         u64_stats_update_begin(&tx_stats->syncp);
1857         tx_stats->tx_packets++;
1858         tx_stats->tx_bytes += len;
1859         u64_stats_update_end(&tx_stats->syncp);
1860
1861         if (netif_rx(skb) == NET_RX_SUCCESS) {
1862                 u64_stats_update_begin(&rx_stats->syncp);
1863                 rx_stats->rx_packets++;
1864                 rx_stats->rx_bytes += len;
1865                 u64_stats_update_end(&rx_stats->syncp);
1866         } else {
1867                 dev->stats.rx_dropped++;
1868         }
1869 }
1870
1871 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1872                            struct vxlan_rdst *rdst, bool did_rsc)
1873 {
1874         struct ip_tunnel_info *info;
1875         struct vxlan_dev *vxlan = netdev_priv(dev);
1876         struct sock *sk = vxlan->vn_sock->sock->sk;
1877         unsigned short family = vxlan_get_sk_family(vxlan->vn_sock);
1878         struct rtable *rt = NULL;
1879         const struct iphdr *old_iph;
1880         struct flowi4 fl4;
1881         union vxlan_addr *dst;
1882         union vxlan_addr remote_ip;
1883         struct vxlan_metadata _md;
1884         struct vxlan_metadata *md = &_md;
1885         __be16 src_port = 0, dst_port;
1886         u32 vni;
1887         __be16 df = 0;
1888         __u8 tos, ttl;
1889         int err;
1890         u32 flags = vxlan->flags;
1891
1892         info = skb_tunnel_info(skb);
1893
1894         if (rdst) {
1895                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1896                 vni = rdst->remote_vni;
1897                 dst = &rdst->remote_ip;
1898         } else {
1899                 if (!info) {
1900                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1901                                   dev->name);
1902                         goto drop;
1903                 }
1904                 if (family != ip_tunnel_info_af(info))
1905                         goto drop;
1906
1907                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1908                 vni = be64_to_cpu(info->key.tun_id);
1909                 remote_ip.sa.sa_family = family;
1910                 if (family == AF_INET)
1911                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1912                 else
1913                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1914                 dst = &remote_ip;
1915         }
1916
1917         if (vxlan_addr_any(dst)) {
1918                 if (did_rsc) {
1919                         /* short-circuited back to local bridge */
1920                         vxlan_encap_bypass(skb, vxlan, vxlan);
1921                         return;
1922                 }
1923                 goto drop;
1924         }
1925
1926         old_iph = ip_hdr(skb);
1927
1928         ttl = vxlan->cfg.ttl;
1929         if (!ttl && vxlan_addr_multicast(dst))
1930                 ttl = 1;
1931
1932         tos = vxlan->cfg.tos;
1933         if (tos == 1)
1934                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1935
1936         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
1937                                      vxlan->cfg.port_max, true);
1938
1939         if (info) {
1940                 if (info->key.tun_flags & TUNNEL_CSUM)
1941                         flags |= VXLAN_F_UDP_CSUM;
1942                 else
1943                         flags &= ~VXLAN_F_UDP_CSUM;
1944
1945                 ttl = info->key.ttl;
1946                 tos = info->key.tos;
1947
1948                 if (info->options_len)
1949                         md = ip_tunnel_info_opts(info);
1950         } else {
1951                 md->gbp = skb->mark;
1952         }
1953
1954         if (dst->sa.sa_family == AF_INET) {
1955                 if (info && (info->key.tun_flags & TUNNEL_DONT_FRAGMENT))
1956                         df = htons(IP_DF);
1957
1958                 memset(&fl4, 0, sizeof(fl4));
1959                 fl4.flowi4_oif = rdst ? rdst->remote_ifindex : 0;
1960                 fl4.flowi4_tos = RT_TOS(tos);
1961                 fl4.flowi4_mark = skb->mark;
1962                 fl4.flowi4_proto = IPPROTO_UDP;
1963                 fl4.daddr = dst->sin.sin_addr.s_addr;
1964                 fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
1965
1966                 rt = ip_route_output_key(vxlan->net, &fl4);
1967                 if (IS_ERR(rt)) {
1968                         netdev_dbg(dev, "no route to %pI4\n",
1969                                    &dst->sin.sin_addr.s_addr);
1970                         dev->stats.tx_carrier_errors++;
1971                         goto tx_error;
1972                 }
1973
1974                 if (rt->dst.dev == dev) {
1975                         netdev_dbg(dev, "circular route to %pI4\n",
1976                                    &dst->sin.sin_addr.s_addr);
1977                         dev->stats.collisions++;
1978                         goto rt_tx_error;
1979                 }
1980
1981                 /* Bypass encapsulation if the destination is local */
1982                 if (rt->rt_flags & RTCF_LOCAL &&
1983                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1984                         struct vxlan_dev *dst_vxlan;
1985
1986                         ip_rt_put(rt);
1987                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1988                                                    dst->sa.sa_family, dst_port,
1989                                                    vxlan->flags);
1990                         if (!dst_vxlan)
1991                                 goto tx_error;
1992                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1993                         return;
1994                 }
1995
1996                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1997                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1998                 err = vxlan_xmit_skb(rt, sk, skb, fl4.saddr,
1999                                      dst->sin.sin_addr.s_addr, tos, ttl, df,
2000                                      src_port, dst_port, htonl(vni << 8), md,
2001                                      !net_eq(vxlan->net, dev_net(vxlan->dev)),
2002                                      flags);
2003                 if (err < 0) {
2004                         /* skb is already freed. */
2005                         skb = NULL;
2006                         goto rt_tx_error;
2007                 }
2008
2009                 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
2010 #if IS_ENABLED(CONFIG_IPV6)
2011         } else {
2012                 struct dst_entry *ndst;
2013                 struct flowi6 fl6;
2014                 u32 rt6i_flags;
2015
2016                 memset(&fl6, 0, sizeof(fl6));
2017                 fl6.flowi6_oif = rdst ? rdst->remote_ifindex : 0;
2018                 fl6.daddr = dst->sin6.sin6_addr;
2019                 fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
2020                 fl6.flowi6_mark = skb->mark;
2021                 fl6.flowi6_proto = IPPROTO_UDP;
2022
2023                 if (ipv6_stub->ipv6_dst_lookup(vxlan->net, sk, &ndst, &fl6)) {
2024                         netdev_dbg(dev, "no route to %pI6\n",
2025                                    &dst->sin6.sin6_addr);
2026                         dev->stats.tx_carrier_errors++;
2027                         goto tx_error;
2028                 }
2029
2030                 if (ndst->dev == dev) {
2031                         netdev_dbg(dev, "circular route to %pI6\n",
2032                                    &dst->sin6.sin6_addr);
2033                         dst_release(ndst);
2034                         dev->stats.collisions++;
2035                         goto tx_error;
2036                 }
2037
2038                 /* Bypass encapsulation if the destination is local */
2039                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2040                 if (rt6i_flags & RTF_LOCAL &&
2041                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2042                         struct vxlan_dev *dst_vxlan;
2043
2044                         dst_release(ndst);
2045                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2046                                                    dst->sa.sa_family, dst_port,
2047                                                    vxlan->flags);
2048                         if (!dst_vxlan)
2049                                 goto tx_error;
2050                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2051                         return;
2052                 }
2053
2054                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2055                 err = vxlan6_xmit_skb(ndst, sk, skb, dev, &fl6.saddr, &fl6.daddr,
2056                                       0, ttl, src_port, dst_port, htonl(vni << 8), md,
2057                                       !net_eq(vxlan->net, dev_net(vxlan->dev)),
2058                                       flags);
2059 #endif
2060         }
2061
2062         return;
2063
2064 drop:
2065         dev->stats.tx_dropped++;
2066         goto tx_free;
2067
2068 rt_tx_error:
2069         ip_rt_put(rt);
2070 tx_error:
2071         dev->stats.tx_errors++;
2072 tx_free:
2073         dev_kfree_skb(skb);
2074 }
2075
2076 /* Transmit local packets over Vxlan
2077  *
2078  * Outer IP header inherits ECN and DF from inner header.
2079  * Outer UDP destination is the VXLAN assigned port.
2080  *           source port is based on hash of flow
2081  */
2082 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2083 {
2084         struct vxlan_dev *vxlan = netdev_priv(dev);
2085         const struct ip_tunnel_info *info;
2086         struct ethhdr *eth;
2087         bool did_rsc = false;
2088         struct vxlan_rdst *rdst, *fdst = NULL;
2089         struct vxlan_fdb *f;
2090
2091         info = skb_tunnel_info(skb);
2092
2093         skb_reset_mac_header(skb);
2094         eth = eth_hdr(skb);
2095
2096         if ((vxlan->flags & VXLAN_F_PROXY)) {
2097                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2098                         return arp_reduce(dev, skb);
2099 #if IS_ENABLED(CONFIG_IPV6)
2100                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2101                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2102                                        + sizeof(struct nd_msg)) &&
2103                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2104                                 struct nd_msg *msg;
2105
2106                                 msg = (struct nd_msg *)skb_transport_header(skb);
2107                                 if (msg->icmph.icmp6_code == 0 &&
2108                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2109                                         return neigh_reduce(dev, skb);
2110                 }
2111                 eth = eth_hdr(skb);
2112 #endif
2113         }
2114
2115         if (vxlan->flags & VXLAN_F_COLLECT_METADATA &&
2116             info && info->mode & IP_TUNNEL_INFO_TX) {
2117                 vxlan_xmit_one(skb, dev, NULL, false);
2118                 return NETDEV_TX_OK;
2119         }
2120
2121         f = vxlan_find_mac(vxlan, eth->h_dest);
2122         did_rsc = false;
2123
2124         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2125             (ntohs(eth->h_proto) == ETH_P_IP ||
2126              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2127                 did_rsc = route_shortcircuit(dev, skb);
2128                 if (did_rsc)
2129                         f = vxlan_find_mac(vxlan, eth->h_dest);
2130         }
2131
2132         if (f == NULL) {
2133                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2134                 if (f == NULL) {
2135                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2136                             !is_multicast_ether_addr(eth->h_dest))
2137                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2138
2139                         dev->stats.tx_dropped++;
2140                         kfree_skb(skb);
2141                         return NETDEV_TX_OK;
2142                 }
2143         }
2144
2145         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2146                 struct sk_buff *skb1;
2147
2148                 if (!fdst) {
2149                         fdst = rdst;
2150                         continue;
2151                 }
2152                 skb1 = skb_clone(skb, GFP_ATOMIC);
2153                 if (skb1)
2154                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2155         }
2156
2157         if (fdst)
2158                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2159         else
2160                 kfree_skb(skb);
2161         return NETDEV_TX_OK;
2162 }
2163
2164 /* Walk the forwarding table and purge stale entries */
2165 static void vxlan_cleanup(unsigned long arg)
2166 {
2167         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2168         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2169         unsigned int h;
2170
2171         if (!netif_running(vxlan->dev))
2172                 return;
2173
2174         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2175                 struct hlist_node *p, *n;
2176
2177                 spin_lock_bh(&vxlan->hash_lock);
2178                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2179                         struct vxlan_fdb *f
2180                                 = container_of(p, struct vxlan_fdb, hlist);
2181                         unsigned long timeout;
2182
2183                         if (f->state & NUD_PERMANENT)
2184                                 continue;
2185
2186                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2187                         if (time_before_eq(timeout, jiffies)) {
2188                                 netdev_dbg(vxlan->dev,
2189                                            "garbage collect %pM\n",
2190                                            f->eth_addr);
2191                                 f->state = NUD_STALE;
2192                                 vxlan_fdb_destroy(vxlan, f);
2193                         } else if (time_before(timeout, next_timer))
2194                                 next_timer = timeout;
2195                 }
2196                 spin_unlock_bh(&vxlan->hash_lock);
2197         }
2198
2199         mod_timer(&vxlan->age_timer, next_timer);
2200 }
2201
2202 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2203 {
2204         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2205         __u32 vni = vxlan->default_dst.remote_vni;
2206
2207         vxlan->vn_sock = vs;
2208         spin_lock(&vn->sock_lock);
2209         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2210         spin_unlock(&vn->sock_lock);
2211 }
2212
2213 /* Setup stats when device is created */
2214 static int vxlan_init(struct net_device *dev)
2215 {
2216         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2217         if (!dev->tstats)
2218                 return -ENOMEM;
2219
2220         return 0;
2221 }
2222
2223 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2224 {
2225         struct vxlan_fdb *f;
2226
2227         spin_lock_bh(&vxlan->hash_lock);
2228         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2229         if (f)
2230                 vxlan_fdb_destroy(vxlan, f);
2231         spin_unlock_bh(&vxlan->hash_lock);
2232 }
2233
2234 static void vxlan_uninit(struct net_device *dev)
2235 {
2236         struct vxlan_dev *vxlan = netdev_priv(dev);
2237
2238         vxlan_fdb_delete_default(vxlan);
2239
2240         free_percpu(dev->tstats);
2241 }
2242
2243 /* Start ageing timer and join group when device is brought up */
2244 static int vxlan_open(struct net_device *dev)
2245 {
2246         struct vxlan_dev *vxlan = netdev_priv(dev);
2247         struct vxlan_sock *vs;
2248         int ret = 0;
2249
2250         vs = vxlan_sock_add(vxlan->net, vxlan->cfg.dst_port,
2251                             vxlan->cfg.no_share, vxlan->flags);
2252         if (IS_ERR(vs))
2253                 return PTR_ERR(vs);
2254
2255         vxlan_vs_add_dev(vs, vxlan);
2256
2257         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2258                 ret = vxlan_igmp_join(vxlan);
2259                 if (ret == -EADDRINUSE)
2260                         ret = 0;
2261                 if (ret) {
2262                         vxlan_sock_release(vs);
2263                         return ret;
2264                 }
2265         }
2266
2267         if (vxlan->cfg.age_interval)
2268                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2269
2270         return ret;
2271 }
2272
2273 /* Purge the forwarding table */
2274 static void vxlan_flush(struct vxlan_dev *vxlan)
2275 {
2276         unsigned int h;
2277
2278         spin_lock_bh(&vxlan->hash_lock);
2279         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2280                 struct hlist_node *p, *n;
2281                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2282                         struct vxlan_fdb *f
2283                                 = container_of(p, struct vxlan_fdb, hlist);
2284                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2285                         if (!is_zero_ether_addr(f->eth_addr))
2286                                 vxlan_fdb_destroy(vxlan, f);
2287                 }
2288         }
2289         spin_unlock_bh(&vxlan->hash_lock);
2290 }
2291
2292 /* Cleanup timer and forwarding table on shutdown */
2293 static int vxlan_stop(struct net_device *dev)
2294 {
2295         struct vxlan_dev *vxlan = netdev_priv(dev);
2296         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2297         struct vxlan_sock *vs = vxlan->vn_sock;
2298         int ret = 0;
2299
2300         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2301             !vxlan_group_used(vn, vxlan))
2302                 ret = vxlan_igmp_leave(vxlan);
2303
2304         del_timer_sync(&vxlan->age_timer);
2305
2306         vxlan_flush(vxlan);
2307         vxlan_sock_release(vs);
2308
2309         return ret;
2310 }
2311
2312 /* Stub, nothing needs to be done. */
2313 static void vxlan_set_multicast_list(struct net_device *dev)
2314 {
2315 }
2316
2317 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2318 {
2319         struct vxlan_dev *vxlan = netdev_priv(dev);
2320         struct vxlan_rdst *dst = &vxlan->default_dst;
2321         struct net_device *lowerdev;
2322         int max_mtu;
2323
2324         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2325         if (lowerdev == NULL)
2326                 return eth_change_mtu(dev, new_mtu);
2327
2328         if (dst->remote_ip.sa.sa_family == AF_INET6)
2329                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2330         else
2331                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2332
2333         if (new_mtu < 68 || new_mtu > max_mtu)
2334                 return -EINVAL;
2335
2336         dev->mtu = new_mtu;
2337         return 0;
2338 }
2339
2340 static const struct net_device_ops vxlan_netdev_ops = {
2341         .ndo_init               = vxlan_init,
2342         .ndo_uninit             = vxlan_uninit,
2343         .ndo_open               = vxlan_open,
2344         .ndo_stop               = vxlan_stop,
2345         .ndo_start_xmit         = vxlan_xmit,
2346         .ndo_get_stats64        = ip_tunnel_get_stats64,
2347         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2348         .ndo_change_mtu         = vxlan_change_mtu,
2349         .ndo_validate_addr      = eth_validate_addr,
2350         .ndo_set_mac_address    = eth_mac_addr,
2351         .ndo_fdb_add            = vxlan_fdb_add,
2352         .ndo_fdb_del            = vxlan_fdb_delete,
2353         .ndo_fdb_dump           = vxlan_fdb_dump,
2354 };
2355
2356 /* Info for udev, that this is a virtual tunnel endpoint */
2357 static struct device_type vxlan_type = {
2358         .name = "vxlan",
2359 };
2360
2361 /* Calls the ndo_add_vxlan_port of the caller in order to
2362  * supply the listening VXLAN udp ports. Callers are expected
2363  * to implement the ndo_add_vxlan_port.
2364  */
2365 void vxlan_get_rx_port(struct net_device *dev)
2366 {
2367         struct vxlan_sock *vs;
2368         struct net *net = dev_net(dev);
2369         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2370         sa_family_t sa_family;
2371         __be16 port;
2372         unsigned int i;
2373
2374         spin_lock(&vn->sock_lock);
2375         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2376                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2377                         port = inet_sk(vs->sock->sk)->inet_sport;
2378                         sa_family = vxlan_get_sk_family(vs);
2379                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2380                                                             port);
2381                 }
2382         }
2383         spin_unlock(&vn->sock_lock);
2384 }
2385 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2386
2387 /* Initialize the device structure. */
2388 static void vxlan_setup(struct net_device *dev)
2389 {
2390         struct vxlan_dev *vxlan = netdev_priv(dev);
2391         unsigned int h;
2392
2393         eth_hw_addr_random(dev);
2394         ether_setup(dev);
2395
2396         dev->netdev_ops = &vxlan_netdev_ops;
2397         dev->destructor = free_netdev;
2398         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2399
2400         dev->features   |= NETIF_F_LLTX;
2401         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2402         dev->features   |= NETIF_F_RXCSUM;
2403         dev->features   |= NETIF_F_GSO_SOFTWARE;
2404
2405         dev->vlan_features = dev->features;
2406         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2407         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2408         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2409         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2410         netif_keep_dst(dev);
2411         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
2412
2413         INIT_LIST_HEAD(&vxlan->next);
2414         spin_lock_init(&vxlan->hash_lock);
2415
2416         init_timer_deferrable(&vxlan->age_timer);
2417         vxlan->age_timer.function = vxlan_cleanup;
2418         vxlan->age_timer.data = (unsigned long) vxlan;
2419
2420         vxlan->cfg.dst_port = htons(vxlan_port);
2421
2422         vxlan->dev = dev;
2423
2424         gro_cells_init(&vxlan->gro_cells, dev);
2425
2426         for (h = 0; h < FDB_HASH_SIZE; ++h)
2427                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2428 }
2429
2430 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2431         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2432         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2433         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2434         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2435         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2436         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2437         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2438         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2439         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2440         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2441         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2442         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2443         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2444         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2445         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2446         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2447         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2448         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2449         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2450         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2451         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2452         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2453         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2454         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2455         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2456 };
2457
2458 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2459 {
2460         if (tb[IFLA_ADDRESS]) {
2461                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2462                         pr_debug("invalid link address (not ethernet)\n");
2463                         return -EINVAL;
2464                 }
2465
2466                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2467                         pr_debug("invalid all zero ethernet address\n");
2468                         return -EADDRNOTAVAIL;
2469                 }
2470         }
2471
2472         if (!data)
2473                 return -EINVAL;
2474
2475         if (data[IFLA_VXLAN_ID]) {
2476                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2477                 if (id >= VXLAN_VID_MASK)
2478                         return -ERANGE;
2479         }
2480
2481         if (data[IFLA_VXLAN_PORT_RANGE]) {
2482                 const struct ifla_vxlan_port_range *p
2483                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2484
2485                 if (ntohs(p->high) < ntohs(p->low)) {
2486                         pr_debug("port range %u .. %u not valid\n",
2487                                  ntohs(p->low), ntohs(p->high));
2488                         return -EINVAL;
2489                 }
2490         }
2491
2492         return 0;
2493 }
2494
2495 static void vxlan_get_drvinfo(struct net_device *netdev,
2496                               struct ethtool_drvinfo *drvinfo)
2497 {
2498         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2499         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2500 }
2501
2502 static const struct ethtool_ops vxlan_ethtool_ops = {
2503         .get_drvinfo    = vxlan_get_drvinfo,
2504         .get_link       = ethtool_op_get_link,
2505 };
2506
2507 static void vxlan_del_work(struct work_struct *work)
2508 {
2509         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2510         udp_tunnel_sock_release(vs->sock);
2511         kfree_rcu(vs, rcu);
2512 }
2513
2514 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2515                                         __be16 port, u32 flags)
2516 {
2517         struct socket *sock;
2518         struct udp_port_cfg udp_conf;
2519         int err;
2520
2521         memset(&udp_conf, 0, sizeof(udp_conf));
2522
2523         if (ipv6) {
2524                 udp_conf.family = AF_INET6;
2525                 udp_conf.use_udp6_rx_checksums =
2526                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2527                 udp_conf.ipv6_v6only = 1;
2528         } else {
2529                 udp_conf.family = AF_INET;
2530         }
2531
2532         udp_conf.local_udp_port = port;
2533
2534         /* Open UDP socket */
2535         err = udp_sock_create(net, &udp_conf, &sock);
2536         if (err < 0)
2537                 return ERR_PTR(err);
2538
2539         return sock;
2540 }
2541
2542 /* Create new listen socket if needed */
2543 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2544                                               u32 flags)
2545 {
2546         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2547         struct vxlan_sock *vs;
2548         struct socket *sock;
2549         unsigned int h;
2550         bool ipv6 = !!(flags & VXLAN_F_IPV6);
2551         struct udp_tunnel_sock_cfg tunnel_cfg;
2552
2553         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2554         if (!vs)
2555                 return ERR_PTR(-ENOMEM);
2556
2557         for (h = 0; h < VNI_HASH_SIZE; ++h)
2558                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2559
2560         INIT_WORK(&vs->del_work, vxlan_del_work);
2561
2562         sock = vxlan_create_sock(net, ipv6, port, flags);
2563         if (IS_ERR(sock)) {
2564                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2565                         PTR_ERR(sock));
2566                 kfree(vs);
2567                 return ERR_CAST(sock);
2568         }
2569
2570         vs->sock = sock;
2571         atomic_set(&vs->refcnt, 1);
2572         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2573
2574         /* Initialize the vxlan udp offloads structure */
2575         vs->udp_offloads.port = port;
2576         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2577         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2578
2579         spin_lock(&vn->sock_lock);
2580         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2581         vxlan_notify_add_rx_port(vs);
2582         spin_unlock(&vn->sock_lock);
2583
2584         /* Mark socket as an encapsulation socket. */
2585         tunnel_cfg.sk_user_data = vs;
2586         tunnel_cfg.encap_type = 1;
2587         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2588         tunnel_cfg.encap_destroy = NULL;
2589
2590         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2591
2592         return vs;
2593 }
2594
2595 static struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2596                                          bool no_share, u32 flags)
2597 {
2598         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2599         struct vxlan_sock *vs;
2600         bool ipv6 = flags & VXLAN_F_IPV6;
2601
2602         if (!no_share) {
2603                 spin_lock(&vn->sock_lock);
2604                 vs = vxlan_find_sock(net, ipv6 ? AF_INET6 : AF_INET, port,
2605                                      flags);
2606                 if (vs) {
2607                         if (!atomic_add_unless(&vs->refcnt, 1, 0))
2608                                 vs = ERR_PTR(-EBUSY);
2609                         spin_unlock(&vn->sock_lock);
2610                         return vs;
2611                 }
2612                 spin_unlock(&vn->sock_lock);
2613         }
2614
2615         return vxlan_socket_create(net, port, flags);
2616 }
2617
2618 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2619                                struct vxlan_config *conf)
2620 {
2621         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2622         struct vxlan_dev *vxlan = netdev_priv(dev);
2623         struct vxlan_rdst *dst = &vxlan->default_dst;
2624         int err;
2625         bool use_ipv6 = false;
2626         __be16 default_port = vxlan->cfg.dst_port;
2627
2628         vxlan->net = src_net;
2629
2630         dst->remote_vni = conf->vni;
2631
2632         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2633
2634         /* Unless IPv6 is explicitly requested, assume IPv4 */
2635         if (!dst->remote_ip.sa.sa_family)
2636                 dst->remote_ip.sa.sa_family = AF_INET;
2637
2638         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2639             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2640                 if (!IS_ENABLED(CONFIG_IPV6))
2641                         return -EPFNOSUPPORT;
2642                 use_ipv6 = true;
2643         }
2644
2645         if (conf->remote_ifindex) {
2646                 struct net_device *lowerdev
2647                          = __dev_get_by_index(src_net, conf->remote_ifindex);
2648
2649                 dst->remote_ifindex = conf->remote_ifindex;
2650
2651                 if (!lowerdev) {
2652                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2653                         return -ENODEV;
2654                 }
2655
2656 #if IS_ENABLED(CONFIG_IPV6)
2657                 if (use_ipv6) {
2658                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2659                         if (idev && idev->cnf.disable_ipv6) {
2660                                 pr_info("IPv6 is disabled via sysctl\n");
2661                                 return -EPERM;
2662                         }
2663                         vxlan->flags |= VXLAN_F_IPV6;
2664                 }
2665 #endif
2666
2667                 if (!conf->mtu)
2668                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2669
2670                 dev->needed_headroom = lowerdev->hard_header_len +
2671                                        (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2672         } else if (use_ipv6) {
2673                 vxlan->flags |= VXLAN_F_IPV6;
2674                 dev->needed_headroom = ETH_HLEN + VXLAN6_HEADROOM;
2675         } else {
2676                 dev->needed_headroom = ETH_HLEN + VXLAN_HEADROOM;
2677         }
2678
2679         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2680         if (!vxlan->cfg.dst_port)
2681                 vxlan->cfg.dst_port = default_port;
2682         vxlan->flags |= conf->flags;
2683
2684         if (!vxlan->cfg.age_interval)
2685                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2686
2687         if (vxlan_find_vni(src_net, conf->vni, use_ipv6 ? AF_INET6 : AF_INET,
2688                            vxlan->cfg.dst_port, vxlan->flags))
2689                 return -EEXIST;
2690
2691         dev->ethtool_ops = &vxlan_ethtool_ops;
2692
2693         /* create an fdb entry for a valid default destination */
2694         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2695                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2696                                        &vxlan->default_dst.remote_ip,
2697                                        NUD_REACHABLE|NUD_PERMANENT,
2698                                        NLM_F_EXCL|NLM_F_CREATE,
2699                                        vxlan->cfg.dst_port,
2700                                        vxlan->default_dst.remote_vni,
2701                                        vxlan->default_dst.remote_ifindex,
2702                                        NTF_SELF);
2703                 if (err)
2704                         return err;
2705         }
2706
2707         err = register_netdevice(dev);
2708         if (err) {
2709                 vxlan_fdb_delete_default(vxlan);
2710                 return err;
2711         }
2712
2713         list_add(&vxlan->next, &vn->vxlan_list);
2714
2715         return 0;
2716 }
2717
2718 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2719                                     u8 name_assign_type, struct vxlan_config *conf)
2720 {
2721         struct nlattr *tb[IFLA_MAX+1];
2722         struct net_device *dev;
2723         int err;
2724
2725         memset(&tb, 0, sizeof(tb));
2726
2727         dev = rtnl_create_link(net, name, name_assign_type,
2728                                &vxlan_link_ops, tb);
2729         if (IS_ERR(dev))
2730                 return dev;
2731
2732         err = vxlan_dev_configure(net, dev, conf);
2733         if (err < 0) {
2734                 free_netdev(dev);
2735                 return ERR_PTR(err);
2736         }
2737
2738         return dev;
2739 }
2740 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2741
2742 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2743                          struct nlattr *tb[], struct nlattr *data[])
2744 {
2745         struct vxlan_config conf;
2746         int err;
2747
2748         if (!data[IFLA_VXLAN_ID])
2749                 return -EINVAL;
2750
2751         memset(&conf, 0, sizeof(conf));
2752         conf.vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2753
2754         if (data[IFLA_VXLAN_GROUP]) {
2755                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2756         } else if (data[IFLA_VXLAN_GROUP6]) {
2757                 if (!IS_ENABLED(CONFIG_IPV6))
2758                         return -EPFNOSUPPORT;
2759
2760                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2761                 conf.remote_ip.sa.sa_family = AF_INET6;
2762         }
2763
2764         if (data[IFLA_VXLAN_LOCAL]) {
2765                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2766                 conf.saddr.sa.sa_family = AF_INET;
2767         } else if (data[IFLA_VXLAN_LOCAL6]) {
2768                 if (!IS_ENABLED(CONFIG_IPV6))
2769                         return -EPFNOSUPPORT;
2770
2771                 /* TODO: respect scope id */
2772                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2773                 conf.saddr.sa.sa_family = AF_INET6;
2774         }
2775
2776         if (data[IFLA_VXLAN_LINK])
2777                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2778
2779         if (data[IFLA_VXLAN_TOS])
2780                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2781
2782         if (data[IFLA_VXLAN_TTL])
2783                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2784
2785         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2786                 conf.flags |= VXLAN_F_LEARN;
2787
2788         if (data[IFLA_VXLAN_AGEING])
2789                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2790
2791         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2792                 conf.flags |= VXLAN_F_PROXY;
2793
2794         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2795                 conf.flags |= VXLAN_F_RSC;
2796
2797         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2798                 conf.flags |= VXLAN_F_L2MISS;
2799
2800         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2801                 conf.flags |= VXLAN_F_L3MISS;
2802
2803         if (data[IFLA_VXLAN_LIMIT])
2804                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2805
2806         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2807             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2808                 conf.flags |= VXLAN_F_COLLECT_METADATA;
2809
2810         if (data[IFLA_VXLAN_PORT_RANGE]) {
2811                 const struct ifla_vxlan_port_range *p
2812                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2813                 conf.port_min = ntohs(p->low);
2814                 conf.port_max = ntohs(p->high);
2815         }
2816
2817         if (data[IFLA_VXLAN_PORT])
2818                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2819
2820         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2821                 conf.flags |= VXLAN_F_UDP_CSUM;
2822
2823         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2824             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2825                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2826
2827         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2828             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2829                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2830
2831         if (data[IFLA_VXLAN_REMCSUM_TX] &&
2832             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2833                 conf.flags |= VXLAN_F_REMCSUM_TX;
2834
2835         if (data[IFLA_VXLAN_REMCSUM_RX] &&
2836             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2837                 conf.flags |= VXLAN_F_REMCSUM_RX;
2838
2839         if (data[IFLA_VXLAN_GBP])
2840                 conf.flags |= VXLAN_F_GBP;
2841
2842         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2843                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2844
2845         err = vxlan_dev_configure(src_net, dev, &conf);
2846         switch (err) {
2847         case -ENODEV:
2848                 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
2849                 break;
2850
2851         case -EPERM:
2852                 pr_info("IPv6 is disabled via sysctl\n");
2853                 break;
2854
2855         case -EEXIST:
2856                 pr_info("duplicate VNI %u\n", conf.vni);
2857                 break;
2858         }
2859
2860         return err;
2861 }
2862
2863 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2864 {
2865         struct vxlan_dev *vxlan = netdev_priv(dev);
2866         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2867
2868         spin_lock(&vn->sock_lock);
2869         if (!hlist_unhashed(&vxlan->hlist))
2870                 hlist_del_rcu(&vxlan->hlist);
2871         spin_unlock(&vn->sock_lock);
2872
2873         gro_cells_destroy(&vxlan->gro_cells);
2874         list_del(&vxlan->next);
2875         unregister_netdevice_queue(dev, head);
2876 }
2877
2878 static size_t vxlan_get_size(const struct net_device *dev)
2879 {
2880
2881         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2882                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2883                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2884                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2885                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2886                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2887                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2888                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2889                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2890                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2891                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2892                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
2893                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2894                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2895                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2896                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2897                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2898                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2899                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2900                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
2901                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
2902                 0;
2903 }
2904
2905 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2906 {
2907         const struct vxlan_dev *vxlan = netdev_priv(dev);
2908         const struct vxlan_rdst *dst = &vxlan->default_dst;
2909         struct ifla_vxlan_port_range ports = {
2910                 .low =  htons(vxlan->cfg.port_min),
2911                 .high = htons(vxlan->cfg.port_max),
2912         };
2913
2914         if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2915                 goto nla_put_failure;
2916
2917         if (!vxlan_addr_any(&dst->remote_ip)) {
2918                 if (dst->remote_ip.sa.sa_family == AF_INET) {
2919                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
2920                                             dst->remote_ip.sin.sin_addr.s_addr))
2921                                 goto nla_put_failure;
2922 #if IS_ENABLED(CONFIG_IPV6)
2923                 } else {
2924                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
2925                                              &dst->remote_ip.sin6.sin6_addr))
2926                                 goto nla_put_failure;
2927 #endif
2928                 }
2929         }
2930
2931         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2932                 goto nla_put_failure;
2933
2934         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
2935                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
2936                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
2937                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
2938                                 goto nla_put_failure;
2939 #if IS_ENABLED(CONFIG_IPV6)
2940                 } else {
2941                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
2942                                              &vxlan->cfg.saddr.sin6.sin6_addr))
2943                                 goto nla_put_failure;
2944 #endif
2945                 }
2946         }
2947
2948         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
2949             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
2950             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2951                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
2952             nla_put_u8(skb, IFLA_VXLAN_PROXY,
2953                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
2954             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2955             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2956                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2957             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2958                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2959             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
2960                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
2961             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
2962             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
2963             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
2964             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2965                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2966             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2967                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2968             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2969                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
2970             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
2971                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
2972             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
2973                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
2974                 goto nla_put_failure;
2975
2976         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2977                 goto nla_put_failure;
2978
2979         if (vxlan->flags & VXLAN_F_GBP &&
2980             nla_put_flag(skb, IFLA_VXLAN_GBP))
2981                 goto nla_put_failure;
2982
2983         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
2984             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
2985                 goto nla_put_failure;
2986
2987         return 0;
2988
2989 nla_put_failure:
2990         return -EMSGSIZE;
2991 }
2992
2993 static struct net *vxlan_get_link_net(const struct net_device *dev)
2994 {
2995         struct vxlan_dev *vxlan = netdev_priv(dev);
2996
2997         return vxlan->net;
2998 }
2999
3000 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3001         .kind           = "vxlan",
3002         .maxtype        = IFLA_VXLAN_MAX,
3003         .policy         = vxlan_policy,
3004         .priv_size      = sizeof(struct vxlan_dev),
3005         .setup          = vxlan_setup,
3006         .validate       = vxlan_validate,
3007         .newlink        = vxlan_newlink,
3008         .dellink        = vxlan_dellink,
3009         .get_size       = vxlan_get_size,
3010         .fill_info      = vxlan_fill_info,
3011         .get_link_net   = vxlan_get_link_net,
3012 };
3013
3014 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3015                                              struct net_device *dev)
3016 {
3017         struct vxlan_dev *vxlan, *next;
3018         LIST_HEAD(list_kill);
3019
3020         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3021                 struct vxlan_rdst *dst = &vxlan->default_dst;
3022
3023                 /* In case we created vxlan device with carrier
3024                  * and we loose the carrier due to module unload
3025                  * we also need to remove vxlan device. In other
3026                  * cases, it's not necessary and remote_ifindex
3027                  * is 0 here, so no matches.
3028                  */
3029                 if (dst->remote_ifindex == dev->ifindex)
3030                         vxlan_dellink(vxlan->dev, &list_kill);
3031         }
3032
3033         unregister_netdevice_many(&list_kill);
3034 }
3035
3036 static int vxlan_lowerdev_event(struct notifier_block *unused,
3037                                 unsigned long event, void *ptr)
3038 {
3039         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3040         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3041
3042         if (event == NETDEV_UNREGISTER)
3043                 vxlan_handle_lowerdev_unregister(vn, dev);
3044
3045         return NOTIFY_DONE;
3046 }
3047
3048 static struct notifier_block vxlan_notifier_block __read_mostly = {
3049         .notifier_call = vxlan_lowerdev_event,
3050 };
3051
3052 static __net_init int vxlan_init_net(struct net *net)
3053 {
3054         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3055         unsigned int h;
3056
3057         INIT_LIST_HEAD(&vn->vxlan_list);
3058         spin_lock_init(&vn->sock_lock);
3059
3060         for (h = 0; h < PORT_HASH_SIZE; ++h)
3061                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3062
3063         return 0;
3064 }
3065
3066 static void __net_exit vxlan_exit_net(struct net *net)
3067 {
3068         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3069         struct vxlan_dev *vxlan, *next;
3070         struct net_device *dev, *aux;
3071         LIST_HEAD(list);
3072
3073         rtnl_lock();
3074         for_each_netdev_safe(net, dev, aux)
3075                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3076                         unregister_netdevice_queue(dev, &list);
3077
3078         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3079                 /* If vxlan->dev is in the same netns, it has already been added
3080                  * to the list by the previous loop.
3081                  */
3082                 if (!net_eq(dev_net(vxlan->dev), net)) {
3083                         gro_cells_destroy(&vxlan->gro_cells);
3084                         unregister_netdevice_queue(vxlan->dev, &list);
3085                 }
3086         }
3087
3088         unregister_netdevice_many(&list);
3089         rtnl_unlock();
3090 }
3091
3092 static struct pernet_operations vxlan_net_ops = {
3093         .init = vxlan_init_net,
3094         .exit = vxlan_exit_net,
3095         .id   = &vxlan_net_id,
3096         .size = sizeof(struct vxlan_net),
3097 };
3098
3099 static int __init vxlan_init_module(void)
3100 {
3101         int rc;
3102
3103         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3104         if (!vxlan_wq)
3105                 return -ENOMEM;
3106
3107         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3108
3109         rc = register_pernet_subsys(&vxlan_net_ops);
3110         if (rc)
3111                 goto out1;
3112
3113         rc = register_netdevice_notifier(&vxlan_notifier_block);
3114         if (rc)
3115                 goto out2;
3116
3117         rc = rtnl_link_register(&vxlan_link_ops);
3118         if (rc)
3119                 goto out3;
3120
3121         return 0;
3122 out3:
3123         unregister_netdevice_notifier(&vxlan_notifier_block);
3124 out2:
3125         unregister_pernet_subsys(&vxlan_net_ops);
3126 out1:
3127         destroy_workqueue(vxlan_wq);
3128         return rc;
3129 }
3130 late_initcall(vxlan_init_module);
3131
3132 static void __exit vxlan_cleanup_module(void)
3133 {
3134         rtnl_link_unregister(&vxlan_link_ops);
3135         unregister_netdevice_notifier(&vxlan_notifier_block);
3136         destroy_workqueue(vxlan_wq);
3137         unregister_pernet_subsys(&vxlan_net_ops);
3138         /* rcu_barrier() is called by netns */
3139 }
3140 module_exit(vxlan_cleanup_module);
3141
3142 MODULE_LICENSE("GPL");
3143 MODULE_VERSION(VXLAN_VERSION);
3144 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3145 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3146 MODULE_ALIAS_RTNL_LINK("vxlan");