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 ip_tunnel_info *info;
1227         struct vxlan_sock *vs;
1228         struct vxlanhdr *vxh;
1229         u32 flags, vni;
1230         struct vxlan_metadata _md;
1231         struct vxlan_metadata *md = &_md;
1232
1233         /* Need Vxlan and inner Ethernet header to be present */
1234         if (!pskb_may_pull(skb, VXLAN_HLEN))
1235                 goto error;
1236
1237         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1238         flags = ntohl(vxh->vx_flags);
1239         vni = ntohl(vxh->vx_vni);
1240
1241         if (flags & VXLAN_HF_VNI) {
1242                 flags &= ~VXLAN_HF_VNI;
1243         } else {
1244                 /* VNI flag always required to be set */
1245                 goto bad_flags;
1246         }
1247
1248         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1249                 goto drop;
1250         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1251
1252         vs = rcu_dereference_sk_user_data(sk);
1253         if (!vs)
1254                 goto drop;
1255
1256         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1257                 vxh = vxlan_remcsum(skb, vxh, sizeof(struct vxlanhdr), vni,
1258                                     !!(vs->flags & VXLAN_F_REMCSUM_NOPARTIAL));
1259                 if (!vxh)
1260                         goto drop;
1261
1262                 flags &= ~VXLAN_HF_RCO;
1263                 vni &= VXLAN_VNI_MASK;
1264         }
1265
1266         if (vxlan_collect_metadata(vs)) {
1267                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1268                                          cpu_to_be64(vni >> 8), sizeof(*md));
1269
1270                 if (!tun_dst)
1271                         goto drop;
1272
1273                 info = &tun_dst->u.tun_info;
1274                 md = ip_tunnel_info_opts(info, sizeof(*md));
1275         } else {
1276                 memset(md, 0, sizeof(*md));
1277         }
1278
1279         /* For backwards compatibility, only allow reserved fields to be
1280          * used by VXLAN extensions if explicitly requested.
1281          */
1282         if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1283                 struct vxlanhdr_gbp *gbp;
1284
1285                 gbp = (struct vxlanhdr_gbp *)vxh;
1286                 md->gbp = ntohs(gbp->policy_id);
1287
1288                 if (tun_dst)
1289                         info->key.tun_flags |= TUNNEL_VXLAN_OPT;
1290
1291                 if (gbp->dont_learn)
1292                         md->gbp |= VXLAN_GBP_DONT_LEARN;
1293
1294                 if (gbp->policy_applied)
1295                         md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1296
1297                 flags &= ~VXLAN_GBP_USED_BITS;
1298         }
1299
1300         if (flags || vni & ~VXLAN_VNI_MASK) {
1301                 /* If there are any unprocessed flags remaining treat
1302                  * this as a malformed packet. This behavior diverges from
1303                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1304                  * in reserved fields are to be ignored. The approach here
1305                  * maintains compatibility with previous stack code, and also
1306                  * is more robust and provides a little more security in
1307                  * adding extensions to VXLAN.
1308                  */
1309
1310                 goto bad_flags;
1311         }
1312
1313         vxlan_rcv(vs, skb, md, vni >> 8, tun_dst);
1314         return 0;
1315
1316 drop:
1317         /* Consume bad packet */
1318         kfree_skb(skb);
1319         return 0;
1320
1321 bad_flags:
1322         netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1323                    ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1324
1325 error:
1326         if (tun_dst)
1327                 dst_release((struct dst_entry *)tun_dst);
1328
1329         /* Return non vxlan pkt */
1330         return 1;
1331 }
1332
1333 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1334 {
1335         struct vxlan_dev *vxlan = netdev_priv(dev);
1336         struct arphdr *parp;
1337         u8 *arpptr, *sha;
1338         __be32 sip, tip;
1339         struct neighbour *n;
1340
1341         if (dev->flags & IFF_NOARP)
1342                 goto out;
1343
1344         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1345                 dev->stats.tx_dropped++;
1346                 goto out;
1347         }
1348         parp = arp_hdr(skb);
1349
1350         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1351              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1352             parp->ar_pro != htons(ETH_P_IP) ||
1353             parp->ar_op != htons(ARPOP_REQUEST) ||
1354             parp->ar_hln != dev->addr_len ||
1355             parp->ar_pln != 4)
1356                 goto out;
1357         arpptr = (u8 *)parp + sizeof(struct arphdr);
1358         sha = arpptr;
1359         arpptr += dev->addr_len;        /* sha */
1360         memcpy(&sip, arpptr, sizeof(sip));
1361         arpptr += sizeof(sip);
1362         arpptr += dev->addr_len;        /* tha */
1363         memcpy(&tip, arpptr, sizeof(tip));
1364
1365         if (ipv4_is_loopback(tip) ||
1366             ipv4_is_multicast(tip))
1367                 goto out;
1368
1369         n = neigh_lookup(&arp_tbl, &tip, dev);
1370
1371         if (n) {
1372                 struct vxlan_fdb *f;
1373                 struct sk_buff  *reply;
1374
1375                 if (!(n->nud_state & NUD_CONNECTED)) {
1376                         neigh_release(n);
1377                         goto out;
1378                 }
1379
1380                 f = vxlan_find_mac(vxlan, n->ha);
1381                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1382                         /* bridge-local neighbor */
1383                         neigh_release(n);
1384                         goto out;
1385                 }
1386
1387                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1388                                 n->ha, sha);
1389
1390                 neigh_release(n);
1391
1392                 if (reply == NULL)
1393                         goto out;
1394
1395                 skb_reset_mac_header(reply);
1396                 __skb_pull(reply, skb_network_offset(reply));
1397                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1398                 reply->pkt_type = PACKET_HOST;
1399
1400                 if (netif_rx_ni(reply) == NET_RX_DROP)
1401                         dev->stats.rx_dropped++;
1402         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1403                 union vxlan_addr ipa = {
1404                         .sin.sin_addr.s_addr = tip,
1405                         .sin.sin_family = AF_INET,
1406                 };
1407
1408                 vxlan_ip_miss(dev, &ipa);
1409         }
1410 out:
1411         consume_skb(skb);
1412         return NETDEV_TX_OK;
1413 }
1414
1415 #if IS_ENABLED(CONFIG_IPV6)
1416 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1417         struct neighbour *n, bool isrouter)
1418 {
1419         struct net_device *dev = request->dev;
1420         struct sk_buff *reply;
1421         struct nd_msg *ns, *na;
1422         struct ipv6hdr *pip6;
1423         u8 *daddr;
1424         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1425         int ns_olen;
1426         int i, len;
1427
1428         if (dev == NULL)
1429                 return NULL;
1430
1431         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1432                 sizeof(*na) + na_olen + dev->needed_tailroom;
1433         reply = alloc_skb(len, GFP_ATOMIC);
1434         if (reply == NULL)
1435                 return NULL;
1436
1437         reply->protocol = htons(ETH_P_IPV6);
1438         reply->dev = dev;
1439         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1440         skb_push(reply, sizeof(struct ethhdr));
1441         skb_set_mac_header(reply, 0);
1442
1443         ns = (struct nd_msg *)skb_transport_header(request);
1444
1445         daddr = eth_hdr(request)->h_source;
1446         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1447         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1448                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1449                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1450                         break;
1451                 }
1452         }
1453
1454         /* Ethernet header */
1455         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1456         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1457         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1458         reply->protocol = htons(ETH_P_IPV6);
1459
1460         skb_pull(reply, sizeof(struct ethhdr));
1461         skb_set_network_header(reply, 0);
1462         skb_put(reply, sizeof(struct ipv6hdr));
1463
1464         /* IPv6 header */
1465
1466         pip6 = ipv6_hdr(reply);
1467         memset(pip6, 0, sizeof(struct ipv6hdr));
1468         pip6->version = 6;
1469         pip6->priority = ipv6_hdr(request)->priority;
1470         pip6->nexthdr = IPPROTO_ICMPV6;
1471         pip6->hop_limit = 255;
1472         pip6->daddr = ipv6_hdr(request)->saddr;
1473         pip6->saddr = *(struct in6_addr *)n->primary_key;
1474
1475         skb_pull(reply, sizeof(struct ipv6hdr));
1476         skb_set_transport_header(reply, 0);
1477
1478         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1479
1480         /* Neighbor Advertisement */
1481         memset(na, 0, sizeof(*na)+na_olen);
1482         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1483         na->icmph.icmp6_router = isrouter;
1484         na->icmph.icmp6_override = 1;
1485         na->icmph.icmp6_solicited = 1;
1486         na->target = ns->target;
1487         ether_addr_copy(&na->opt[2], n->ha);
1488         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1489         na->opt[1] = na_olen >> 3;
1490
1491         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1492                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1493                 csum_partial(na, sizeof(*na)+na_olen, 0));
1494
1495         pip6->payload_len = htons(sizeof(*na)+na_olen);
1496
1497         skb_push(reply, sizeof(struct ipv6hdr));
1498
1499         reply->ip_summed = CHECKSUM_UNNECESSARY;
1500
1501         return reply;
1502 }
1503
1504 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1505 {
1506         struct vxlan_dev *vxlan = netdev_priv(dev);
1507         struct nd_msg *msg;
1508         const struct ipv6hdr *iphdr;
1509         const struct in6_addr *saddr, *daddr;
1510         struct neighbour *n;
1511         struct inet6_dev *in6_dev;
1512
1513         in6_dev = __in6_dev_get(dev);
1514         if (!in6_dev)
1515                 goto out;
1516
1517         iphdr = ipv6_hdr(skb);
1518         saddr = &iphdr->saddr;
1519         daddr = &iphdr->daddr;
1520
1521         msg = (struct nd_msg *)skb_transport_header(skb);
1522         if (msg->icmph.icmp6_code != 0 ||
1523             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1524                 goto out;
1525
1526         if (ipv6_addr_loopback(daddr) ||
1527             ipv6_addr_is_multicast(&msg->target))
1528                 goto out;
1529
1530         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1531
1532         if (n) {
1533                 struct vxlan_fdb *f;
1534                 struct sk_buff *reply;
1535
1536                 if (!(n->nud_state & NUD_CONNECTED)) {
1537                         neigh_release(n);
1538                         goto out;
1539                 }
1540
1541                 f = vxlan_find_mac(vxlan, n->ha);
1542                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1543                         /* bridge-local neighbor */
1544                         neigh_release(n);
1545                         goto out;
1546                 }
1547
1548                 reply = vxlan_na_create(skb, n,
1549                                         !!(f ? f->flags & NTF_ROUTER : 0));
1550
1551                 neigh_release(n);
1552
1553                 if (reply == NULL)
1554                         goto out;
1555
1556                 if (netif_rx_ni(reply) == NET_RX_DROP)
1557                         dev->stats.rx_dropped++;
1558
1559         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1560                 union vxlan_addr ipa = {
1561                         .sin6.sin6_addr = msg->target,
1562                         .sin6.sin6_family = AF_INET6,
1563                 };
1564
1565                 vxlan_ip_miss(dev, &ipa);
1566         }
1567
1568 out:
1569         consume_skb(skb);
1570         return NETDEV_TX_OK;
1571 }
1572 #endif
1573
1574 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1575 {
1576         struct vxlan_dev *vxlan = netdev_priv(dev);
1577         struct neighbour *n;
1578
1579         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1580                 return false;
1581
1582         n = NULL;
1583         switch (ntohs(eth_hdr(skb)->h_proto)) {
1584         case ETH_P_IP:
1585         {
1586                 struct iphdr *pip;
1587
1588                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1589                         return false;
1590                 pip = ip_hdr(skb);
1591                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1592                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1593                         union vxlan_addr ipa = {
1594                                 .sin.sin_addr.s_addr = pip->daddr,
1595                                 .sin.sin_family = AF_INET,
1596                         };
1597
1598                         vxlan_ip_miss(dev, &ipa);
1599                         return false;
1600                 }
1601
1602                 break;
1603         }
1604 #if IS_ENABLED(CONFIG_IPV6)
1605         case ETH_P_IPV6:
1606         {
1607                 struct ipv6hdr *pip6;
1608
1609                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1610                         return false;
1611                 pip6 = ipv6_hdr(skb);
1612                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1613                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1614                         union vxlan_addr ipa = {
1615                                 .sin6.sin6_addr = pip6->daddr,
1616                                 .sin6.sin6_family = AF_INET6,
1617                         };
1618
1619                         vxlan_ip_miss(dev, &ipa);
1620                         return false;
1621                 }
1622
1623                 break;
1624         }
1625 #endif
1626         default:
1627                 return false;
1628         }
1629
1630         if (n) {
1631                 bool diff;
1632
1633                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1634                 if (diff) {
1635                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1636                                 dev->addr_len);
1637                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1638                 }
1639                 neigh_release(n);
1640                 return diff;
1641         }
1642
1643         return false;
1644 }
1645
1646 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1647                                 struct vxlan_metadata *md)
1648 {
1649         struct vxlanhdr_gbp *gbp;
1650
1651         if (!md->gbp)
1652                 return;
1653
1654         gbp = (struct vxlanhdr_gbp *)vxh;
1655         vxh->vx_flags |= htonl(VXLAN_HF_GBP);
1656
1657         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1658                 gbp->dont_learn = 1;
1659
1660         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1661                 gbp->policy_applied = 1;
1662
1663         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1664 }
1665
1666 #if IS_ENABLED(CONFIG_IPV6)
1667 static int vxlan6_xmit_skb(struct dst_entry *dst, struct sock *sk,
1668                            struct sk_buff *skb,
1669                            struct net_device *dev, struct in6_addr *saddr,
1670                            struct in6_addr *daddr, __u8 prio, __u8 ttl,
1671                            __be16 src_port, __be16 dst_port, __be32 vni,
1672                            struct vxlan_metadata *md, bool xnet, u32 vxflags)
1673 {
1674         struct vxlanhdr *vxh;
1675         int min_headroom;
1676         int err;
1677         bool udp_sum = !(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX);
1678         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1679         u16 hdrlen = sizeof(struct vxlanhdr);
1680
1681         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1682             skb->ip_summed == CHECKSUM_PARTIAL) {
1683                 int csum_start = skb_checksum_start_offset(skb);
1684
1685                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1686                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1687                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1688                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1689                         udp_sum = false;
1690                         type |= SKB_GSO_TUNNEL_REMCSUM;
1691                 }
1692         }
1693
1694         skb_scrub_packet(skb, xnet);
1695
1696         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1697                         + VXLAN_HLEN + sizeof(struct ipv6hdr)
1698                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1699
1700         /* Need space for new headers (invalidates iph ptr) */
1701         err = skb_cow_head(skb, min_headroom);
1702         if (unlikely(err)) {
1703                 kfree_skb(skb);
1704                 goto err;
1705         }
1706
1707         skb = vlan_hwaccel_push_inside(skb);
1708         if (WARN_ON(!skb)) {
1709                 err = -ENOMEM;
1710                 goto err;
1711         }
1712
1713         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1714         if (IS_ERR(skb)) {
1715                 err = -EINVAL;
1716                 goto err;
1717         }
1718
1719         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1720         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1721         vxh->vx_vni = vni;
1722
1723         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1724                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1725                            VXLAN_RCO_SHIFT;
1726
1727                 if (skb->csum_offset == offsetof(struct udphdr, check))
1728                         data |= VXLAN_RCO_UDP;
1729
1730                 vxh->vx_vni |= htonl(data);
1731                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1732
1733                 if (!skb_is_gso(skb)) {
1734                         skb->ip_summed = CHECKSUM_NONE;
1735                         skb->encapsulation = 0;
1736                 }
1737         }
1738
1739         if (vxflags & VXLAN_F_GBP)
1740                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1741
1742         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1743
1744         udp_tunnel6_xmit_skb(dst, sk, skb, dev, saddr, daddr, prio,
1745                              ttl, src_port, dst_port,
1746                              !!(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX));
1747         return 0;
1748 err:
1749         dst_release(dst);
1750         return err;
1751 }
1752 #endif
1753
1754 static int vxlan_xmit_skb(struct rtable *rt, struct sock *sk, struct sk_buff *skb,
1755                           __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1756                           __be16 src_port, __be16 dst_port, __be32 vni,
1757                           struct vxlan_metadata *md, bool xnet, u32 vxflags)
1758 {
1759         struct vxlanhdr *vxh;
1760         int min_headroom;
1761         int err;
1762         bool udp_sum = !!(vxflags & VXLAN_F_UDP_CSUM);
1763         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1764         u16 hdrlen = sizeof(struct vxlanhdr);
1765
1766         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1767             skb->ip_summed == CHECKSUM_PARTIAL) {
1768                 int csum_start = skb_checksum_start_offset(skb);
1769
1770                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1771                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1772                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1773                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1774                         udp_sum = false;
1775                         type |= SKB_GSO_TUNNEL_REMCSUM;
1776                 }
1777         }
1778
1779         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1780                         + VXLAN_HLEN + sizeof(struct iphdr)
1781                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1782
1783         /* Need space for new headers (invalidates iph ptr) */
1784         err = skb_cow_head(skb, min_headroom);
1785         if (unlikely(err)) {
1786                 kfree_skb(skb);
1787                 return err;
1788         }
1789
1790         skb = vlan_hwaccel_push_inside(skb);
1791         if (WARN_ON(!skb))
1792                 return -ENOMEM;
1793
1794         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1795         if (IS_ERR(skb))
1796                 return PTR_ERR(skb);
1797
1798         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1799         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1800         vxh->vx_vni = vni;
1801
1802         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1803                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1804                            VXLAN_RCO_SHIFT;
1805
1806                 if (skb->csum_offset == offsetof(struct udphdr, check))
1807                         data |= VXLAN_RCO_UDP;
1808
1809                 vxh->vx_vni |= htonl(data);
1810                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1811
1812                 if (!skb_is_gso(skb)) {
1813                         skb->ip_summed = CHECKSUM_NONE;
1814                         skb->encapsulation = 0;
1815                 }
1816         }
1817
1818         if (vxflags & VXLAN_F_GBP)
1819                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1820
1821         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1822
1823         return udp_tunnel_xmit_skb(rt, sk, skb, src, dst, tos,
1824                                    ttl, df, src_port, dst_port, xnet,
1825                                    !(vxflags & VXLAN_F_UDP_CSUM));
1826 }
1827
1828 /* Bypass encapsulation if the destination is local */
1829 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1830                                struct vxlan_dev *dst_vxlan)
1831 {
1832         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1833         union vxlan_addr loopback;
1834         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1835         struct net_device *dev = skb->dev;
1836         int len = skb->len;
1837
1838         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1839         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1840         skb->pkt_type = PACKET_HOST;
1841         skb->encapsulation = 0;
1842         skb->dev = dst_vxlan->dev;
1843         __skb_pull(skb, skb_network_offset(skb));
1844
1845         if (remote_ip->sa.sa_family == AF_INET) {
1846                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1847                 loopback.sa.sa_family =  AF_INET;
1848 #if IS_ENABLED(CONFIG_IPV6)
1849         } else {
1850                 loopback.sin6.sin6_addr = in6addr_loopback;
1851                 loopback.sa.sa_family =  AF_INET6;
1852 #endif
1853         }
1854
1855         if (dst_vxlan->flags & VXLAN_F_LEARN)
1856                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1857
1858         u64_stats_update_begin(&tx_stats->syncp);
1859         tx_stats->tx_packets++;
1860         tx_stats->tx_bytes += len;
1861         u64_stats_update_end(&tx_stats->syncp);
1862
1863         if (netif_rx(skb) == NET_RX_SUCCESS) {
1864                 u64_stats_update_begin(&rx_stats->syncp);
1865                 rx_stats->rx_packets++;
1866                 rx_stats->rx_bytes += len;
1867                 u64_stats_update_end(&rx_stats->syncp);
1868         } else {
1869                 dev->stats.rx_dropped++;
1870         }
1871 }
1872
1873 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1874                            struct vxlan_rdst *rdst, bool did_rsc)
1875 {
1876         struct ip_tunnel_info *info;
1877         struct vxlan_dev *vxlan = netdev_priv(dev);
1878         struct sock *sk = vxlan->vn_sock->sock->sk;
1879         unsigned short family = vxlan_get_sk_family(vxlan->vn_sock);
1880         struct rtable *rt = NULL;
1881         const struct iphdr *old_iph;
1882         struct flowi4 fl4;
1883         union vxlan_addr *dst;
1884         union vxlan_addr remote_ip;
1885         struct vxlan_metadata _md;
1886         struct vxlan_metadata *md = &_md;
1887         __be16 src_port = 0, dst_port;
1888         u32 vni;
1889         __be16 df = 0;
1890         __u8 tos, ttl;
1891         int err;
1892         u32 flags = vxlan->flags;
1893
1894         info = skb_tunnel_info(skb);
1895
1896         if (rdst) {
1897                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1898                 vni = rdst->remote_vni;
1899                 dst = &rdst->remote_ip;
1900         } else {
1901                 if (!info) {
1902                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1903                                   dev->name);
1904                         goto drop;
1905                 }
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, sizeof(*md));
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         if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2396                 dev->needed_headroom = ETH_HLEN + VXLAN6_HEADROOM;
2397         else
2398                 dev->needed_headroom = ETH_HLEN + VXLAN_HEADROOM;
2399
2400         dev->netdev_ops = &vxlan_netdev_ops;
2401         dev->destructor = free_netdev;
2402         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2403
2404         dev->features   |= NETIF_F_LLTX;
2405         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2406         dev->features   |= NETIF_F_RXCSUM;
2407         dev->features   |= NETIF_F_GSO_SOFTWARE;
2408
2409         dev->vlan_features = dev->features;
2410         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2411         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2412         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2413         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2414         netif_keep_dst(dev);
2415         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
2416
2417         INIT_LIST_HEAD(&vxlan->next);
2418         spin_lock_init(&vxlan->hash_lock);
2419
2420         init_timer_deferrable(&vxlan->age_timer);
2421         vxlan->age_timer.function = vxlan_cleanup;
2422         vxlan->age_timer.data = (unsigned long) vxlan;
2423
2424         vxlan->cfg.dst_port = htons(vxlan_port);
2425
2426         vxlan->dev = dev;
2427
2428         gro_cells_init(&vxlan->gro_cells, dev);
2429
2430         for (h = 0; h < FDB_HASH_SIZE; ++h)
2431                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2432 }
2433
2434 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2435         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2436         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2437         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2438         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2439         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2440         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2441         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2442         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2443         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2444         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2445         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2446         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2447         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2448         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2449         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2450         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2451         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2452         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2453         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2454         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2455         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2456         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2457         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2458         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2459         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2460 };
2461
2462 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2463 {
2464         if (tb[IFLA_ADDRESS]) {
2465                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2466                         pr_debug("invalid link address (not ethernet)\n");
2467                         return -EINVAL;
2468                 }
2469
2470                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2471                         pr_debug("invalid all zero ethernet address\n");
2472                         return -EADDRNOTAVAIL;
2473                 }
2474         }
2475
2476         if (!data)
2477                 return -EINVAL;
2478
2479         if (data[IFLA_VXLAN_ID]) {
2480                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2481                 if (id >= VXLAN_VID_MASK)
2482                         return -ERANGE;
2483         }
2484
2485         if (data[IFLA_VXLAN_PORT_RANGE]) {
2486                 const struct ifla_vxlan_port_range *p
2487                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2488
2489                 if (ntohs(p->high) < ntohs(p->low)) {
2490                         pr_debug("port range %u .. %u not valid\n",
2491                                  ntohs(p->low), ntohs(p->high));
2492                         return -EINVAL;
2493                 }
2494         }
2495
2496         return 0;
2497 }
2498
2499 static void vxlan_get_drvinfo(struct net_device *netdev,
2500                               struct ethtool_drvinfo *drvinfo)
2501 {
2502         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2503         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2504 }
2505
2506 static const struct ethtool_ops vxlan_ethtool_ops = {
2507         .get_drvinfo    = vxlan_get_drvinfo,
2508         .get_link       = ethtool_op_get_link,
2509 };
2510
2511 static void vxlan_del_work(struct work_struct *work)
2512 {
2513         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2514         udp_tunnel_sock_release(vs->sock);
2515         kfree_rcu(vs, rcu);
2516 }
2517
2518 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2519                                         __be16 port, u32 flags)
2520 {
2521         struct socket *sock;
2522         struct udp_port_cfg udp_conf;
2523         int err;
2524
2525         memset(&udp_conf, 0, sizeof(udp_conf));
2526
2527         if (ipv6) {
2528                 udp_conf.family = AF_INET6;
2529                 udp_conf.use_udp6_rx_checksums =
2530                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2531         } else {
2532                 udp_conf.family = AF_INET;
2533         }
2534
2535         udp_conf.local_udp_port = port;
2536
2537         /* Open UDP socket */
2538         err = udp_sock_create(net, &udp_conf, &sock);
2539         if (err < 0)
2540                 return ERR_PTR(err);
2541
2542         return sock;
2543 }
2544
2545 /* Create new listen socket if needed */
2546 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2547                                               u32 flags)
2548 {
2549         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2550         struct vxlan_sock *vs;
2551         struct socket *sock;
2552         unsigned int h;
2553         bool ipv6 = !!(flags & VXLAN_F_IPV6);
2554         struct udp_tunnel_sock_cfg tunnel_cfg;
2555
2556         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2557         if (!vs)
2558                 return ERR_PTR(-ENOMEM);
2559
2560         for (h = 0; h < VNI_HASH_SIZE; ++h)
2561                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2562
2563         INIT_WORK(&vs->del_work, vxlan_del_work);
2564
2565         sock = vxlan_create_sock(net, ipv6, port, flags);
2566         if (IS_ERR(sock)) {
2567                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2568                         PTR_ERR(sock));
2569                 kfree(vs);
2570                 return ERR_CAST(sock);
2571         }
2572
2573         vs->sock = sock;
2574         atomic_set(&vs->refcnt, 1);
2575         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2576
2577         /* Initialize the vxlan udp offloads structure */
2578         vs->udp_offloads.port = port;
2579         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2580         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2581
2582         spin_lock(&vn->sock_lock);
2583         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2584         vxlan_notify_add_rx_port(vs);
2585         spin_unlock(&vn->sock_lock);
2586
2587         /* Mark socket as an encapsulation socket. */
2588         tunnel_cfg.sk_user_data = vs;
2589         tunnel_cfg.encap_type = 1;
2590         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2591         tunnel_cfg.encap_destroy = NULL;
2592
2593         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2594
2595         return vs;
2596 }
2597
2598 static struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2599                                          bool no_share, u32 flags)
2600 {
2601         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2602         struct vxlan_sock *vs;
2603         bool ipv6 = flags & VXLAN_F_IPV6;
2604
2605         if (!no_share) {
2606                 spin_lock(&vn->sock_lock);
2607                 vs = vxlan_find_sock(net, ipv6 ? AF_INET6 : AF_INET, port,
2608                                      flags);
2609                 if (vs) {
2610                         if (!atomic_add_unless(&vs->refcnt, 1, 0))
2611                                 vs = ERR_PTR(-EBUSY);
2612                         spin_unlock(&vn->sock_lock);
2613                         return vs;
2614                 }
2615                 spin_unlock(&vn->sock_lock);
2616         }
2617
2618         return vxlan_socket_create(net, port, flags);
2619 }
2620
2621 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2622                                struct vxlan_config *conf)
2623 {
2624         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2625         struct vxlan_dev *vxlan = netdev_priv(dev);
2626         struct vxlan_rdst *dst = &vxlan->default_dst;
2627         int err;
2628         bool use_ipv6 = false;
2629         __be16 default_port = vxlan->cfg.dst_port;
2630
2631         vxlan->net = src_net;
2632
2633         dst->remote_vni = conf->vni;
2634
2635         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2636
2637         /* Unless IPv6 is explicitly requested, assume IPv4 */
2638         if (!dst->remote_ip.sa.sa_family)
2639                 dst->remote_ip.sa.sa_family = AF_INET;
2640
2641         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2642             vxlan->cfg.saddr.sa.sa_family == AF_INET6)
2643                 use_ipv6 = true;
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
2675         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2676         if (!vxlan->cfg.dst_port)
2677                 vxlan->cfg.dst_port = default_port;
2678         vxlan->flags |= conf->flags;
2679
2680         if (!vxlan->cfg.age_interval)
2681                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2682
2683         if (vxlan_find_vni(src_net, conf->vni, use_ipv6 ? AF_INET6 : AF_INET,
2684                            vxlan->cfg.dst_port, vxlan->flags))
2685                 return -EEXIST;
2686
2687         dev->ethtool_ops = &vxlan_ethtool_ops;
2688
2689         /* create an fdb entry for a valid default destination */
2690         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2691                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2692                                        &vxlan->default_dst.remote_ip,
2693                                        NUD_REACHABLE|NUD_PERMANENT,
2694                                        NLM_F_EXCL|NLM_F_CREATE,
2695                                        vxlan->cfg.dst_port,
2696                                        vxlan->default_dst.remote_vni,
2697                                        vxlan->default_dst.remote_ifindex,
2698                                        NTF_SELF);
2699                 if (err)
2700                         return err;
2701         }
2702
2703         err = register_netdevice(dev);
2704         if (err) {
2705                 vxlan_fdb_delete_default(vxlan);
2706                 return err;
2707         }
2708
2709         list_add(&vxlan->next, &vn->vxlan_list);
2710
2711         return 0;
2712 }
2713
2714 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2715                                     u8 name_assign_type, struct vxlan_config *conf)
2716 {
2717         struct nlattr *tb[IFLA_MAX+1];
2718         struct net_device *dev;
2719         int err;
2720
2721         memset(&tb, 0, sizeof(tb));
2722
2723         dev = rtnl_create_link(net, name, name_assign_type,
2724                                &vxlan_link_ops, tb);
2725         if (IS_ERR(dev))
2726                 return dev;
2727
2728         err = vxlan_dev_configure(net, dev, conf);
2729         if (err < 0) {
2730                 free_netdev(dev);
2731                 return ERR_PTR(err);
2732         }
2733
2734         return dev;
2735 }
2736 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2737
2738 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2739                          struct nlattr *tb[], struct nlattr *data[])
2740 {
2741         struct vxlan_config conf;
2742         int err;
2743
2744         if (!data[IFLA_VXLAN_ID])
2745                 return -EINVAL;
2746
2747         memset(&conf, 0, sizeof(conf));
2748         conf.vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2749
2750         if (data[IFLA_VXLAN_GROUP]) {
2751                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2752         } else if (data[IFLA_VXLAN_GROUP6]) {
2753                 if (!IS_ENABLED(CONFIG_IPV6))
2754                         return -EPFNOSUPPORT;
2755
2756                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2757                 conf.remote_ip.sa.sa_family = AF_INET6;
2758         }
2759
2760         if (data[IFLA_VXLAN_LOCAL]) {
2761                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2762                 conf.saddr.sa.sa_family = AF_INET;
2763         } else if (data[IFLA_VXLAN_LOCAL6]) {
2764                 if (!IS_ENABLED(CONFIG_IPV6))
2765                         return -EPFNOSUPPORT;
2766
2767                 /* TODO: respect scope id */
2768                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2769                 conf.saddr.sa.sa_family = AF_INET6;
2770         }
2771
2772         if (data[IFLA_VXLAN_LINK])
2773                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2774
2775         if (data[IFLA_VXLAN_TOS])
2776                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2777
2778         if (data[IFLA_VXLAN_TTL])
2779                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2780
2781         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2782                 conf.flags |= VXLAN_F_LEARN;
2783
2784         if (data[IFLA_VXLAN_AGEING])
2785                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2786
2787         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2788                 conf.flags |= VXLAN_F_PROXY;
2789
2790         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2791                 conf.flags |= VXLAN_F_RSC;
2792
2793         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2794                 conf.flags |= VXLAN_F_L2MISS;
2795
2796         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2797                 conf.flags |= VXLAN_F_L3MISS;
2798
2799         if (data[IFLA_VXLAN_LIMIT])
2800                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2801
2802         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2803             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2804                 conf.flags |= VXLAN_F_COLLECT_METADATA;
2805
2806         if (data[IFLA_VXLAN_PORT_RANGE]) {
2807                 const struct ifla_vxlan_port_range *p
2808                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2809                 conf.port_min = ntohs(p->low);
2810                 conf.port_max = ntohs(p->high);
2811         }
2812
2813         if (data[IFLA_VXLAN_PORT])
2814                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2815
2816         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2817                 conf.flags |= VXLAN_F_UDP_CSUM;
2818
2819         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2820             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2821                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2822
2823         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2824             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2825                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2826
2827         if (data[IFLA_VXLAN_REMCSUM_TX] &&
2828             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2829                 conf.flags |= VXLAN_F_REMCSUM_TX;
2830
2831         if (data[IFLA_VXLAN_REMCSUM_RX] &&
2832             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2833                 conf.flags |= VXLAN_F_REMCSUM_RX;
2834
2835         if (data[IFLA_VXLAN_GBP])
2836                 conf.flags |= VXLAN_F_GBP;
2837
2838         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2839                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2840
2841         err = vxlan_dev_configure(src_net, dev, &conf);
2842         switch (err) {
2843         case -ENODEV:
2844                 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
2845                 break;
2846
2847         case -EPERM:
2848                 pr_info("IPv6 is disabled via sysctl\n");
2849                 break;
2850
2851         case -EEXIST:
2852                 pr_info("duplicate VNI %u\n", conf.vni);
2853                 break;
2854         }
2855
2856         return err;
2857 }
2858
2859 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2860 {
2861         struct vxlan_dev *vxlan = netdev_priv(dev);
2862         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2863
2864         spin_lock(&vn->sock_lock);
2865         if (!hlist_unhashed(&vxlan->hlist))
2866                 hlist_del_rcu(&vxlan->hlist);
2867         spin_unlock(&vn->sock_lock);
2868
2869         gro_cells_destroy(&vxlan->gro_cells);
2870         list_del(&vxlan->next);
2871         unregister_netdevice_queue(dev, head);
2872 }
2873
2874 static size_t vxlan_get_size(const struct net_device *dev)
2875 {
2876
2877         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2878                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2879                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2880                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2881                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2882                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2883                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2884                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2885                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2886                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2887                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2888                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
2889                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2890                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2891                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2892                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2893                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2894                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2895                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2896                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
2897                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
2898                 0;
2899 }
2900
2901 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2902 {
2903         const struct vxlan_dev *vxlan = netdev_priv(dev);
2904         const struct vxlan_rdst *dst = &vxlan->default_dst;
2905         struct ifla_vxlan_port_range ports = {
2906                 .low =  htons(vxlan->cfg.port_min),
2907                 .high = htons(vxlan->cfg.port_max),
2908         };
2909
2910         if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2911                 goto nla_put_failure;
2912
2913         if (!vxlan_addr_any(&dst->remote_ip)) {
2914                 if (dst->remote_ip.sa.sa_family == AF_INET) {
2915                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
2916                                             dst->remote_ip.sin.sin_addr.s_addr))
2917                                 goto nla_put_failure;
2918 #if IS_ENABLED(CONFIG_IPV6)
2919                 } else {
2920                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
2921                                              &dst->remote_ip.sin6.sin6_addr))
2922                                 goto nla_put_failure;
2923 #endif
2924                 }
2925         }
2926
2927         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2928                 goto nla_put_failure;
2929
2930         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
2931                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
2932                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
2933                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
2934                                 goto nla_put_failure;
2935 #if IS_ENABLED(CONFIG_IPV6)
2936                 } else {
2937                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
2938                                              &vxlan->cfg.saddr.sin6.sin6_addr))
2939                                 goto nla_put_failure;
2940 #endif
2941                 }
2942         }
2943
2944         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
2945             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
2946             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2947                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
2948             nla_put_u8(skb, IFLA_VXLAN_PROXY,
2949                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
2950             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2951             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2952                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2953             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2954                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2955             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
2956                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
2957             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
2958             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
2959             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
2960             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2961                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2962             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2963                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2964             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2965                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
2966             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
2967                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
2968             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
2969                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
2970                 goto nla_put_failure;
2971
2972         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2973                 goto nla_put_failure;
2974
2975         if (vxlan->flags & VXLAN_F_GBP &&
2976             nla_put_flag(skb, IFLA_VXLAN_GBP))
2977                 goto nla_put_failure;
2978
2979         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
2980             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
2981                 goto nla_put_failure;
2982
2983         return 0;
2984
2985 nla_put_failure:
2986         return -EMSGSIZE;
2987 }
2988
2989 static struct net *vxlan_get_link_net(const struct net_device *dev)
2990 {
2991         struct vxlan_dev *vxlan = netdev_priv(dev);
2992
2993         return vxlan->net;
2994 }
2995
2996 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2997         .kind           = "vxlan",
2998         .maxtype        = IFLA_VXLAN_MAX,
2999         .policy         = vxlan_policy,
3000         .priv_size      = sizeof(struct vxlan_dev),
3001         .setup          = vxlan_setup,
3002         .validate       = vxlan_validate,
3003         .newlink        = vxlan_newlink,
3004         .dellink        = vxlan_dellink,
3005         .get_size       = vxlan_get_size,
3006         .fill_info      = vxlan_fill_info,
3007         .get_link_net   = vxlan_get_link_net,
3008 };
3009
3010 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3011                                              struct net_device *dev)
3012 {
3013         struct vxlan_dev *vxlan, *next;
3014         LIST_HEAD(list_kill);
3015
3016         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3017                 struct vxlan_rdst *dst = &vxlan->default_dst;
3018
3019                 /* In case we created vxlan device with carrier
3020                  * and we loose the carrier due to module unload
3021                  * we also need to remove vxlan device. In other
3022                  * cases, it's not necessary and remote_ifindex
3023                  * is 0 here, so no matches.
3024                  */
3025                 if (dst->remote_ifindex == dev->ifindex)
3026                         vxlan_dellink(vxlan->dev, &list_kill);
3027         }
3028
3029         unregister_netdevice_many(&list_kill);
3030 }
3031
3032 static int vxlan_lowerdev_event(struct notifier_block *unused,
3033                                 unsigned long event, void *ptr)
3034 {
3035         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3036         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3037
3038         if (event == NETDEV_UNREGISTER)
3039                 vxlan_handle_lowerdev_unregister(vn, dev);
3040
3041         return NOTIFY_DONE;
3042 }
3043
3044 static struct notifier_block vxlan_notifier_block __read_mostly = {
3045         .notifier_call = vxlan_lowerdev_event,
3046 };
3047
3048 static __net_init int vxlan_init_net(struct net *net)
3049 {
3050         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3051         unsigned int h;
3052
3053         INIT_LIST_HEAD(&vn->vxlan_list);
3054         spin_lock_init(&vn->sock_lock);
3055
3056         for (h = 0; h < PORT_HASH_SIZE; ++h)
3057                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3058
3059         return 0;
3060 }
3061
3062 static void __net_exit vxlan_exit_net(struct net *net)
3063 {
3064         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3065         struct vxlan_dev *vxlan, *next;
3066         struct net_device *dev, *aux;
3067         LIST_HEAD(list);
3068
3069         rtnl_lock();
3070         for_each_netdev_safe(net, dev, aux)
3071                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3072                         unregister_netdevice_queue(dev, &list);
3073
3074         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3075                 /* If vxlan->dev is in the same netns, it has already been added
3076                  * to the list by the previous loop.
3077                  */
3078                 if (!net_eq(dev_net(vxlan->dev), net)) {
3079                         gro_cells_destroy(&vxlan->gro_cells);
3080                         unregister_netdevice_queue(vxlan->dev, &list);
3081                 }
3082         }
3083
3084         unregister_netdevice_many(&list);
3085         rtnl_unlock();
3086 }
3087
3088 static struct pernet_operations vxlan_net_ops = {
3089         .init = vxlan_init_net,
3090         .exit = vxlan_exit_net,
3091         .id   = &vxlan_net_id,
3092         .size = sizeof(struct vxlan_net),
3093 };
3094
3095 static int __init vxlan_init_module(void)
3096 {
3097         int rc;
3098
3099         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3100         if (!vxlan_wq)
3101                 return -ENOMEM;
3102
3103         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3104
3105         rc = register_pernet_subsys(&vxlan_net_ops);
3106         if (rc)
3107                 goto out1;
3108
3109         rc = register_netdevice_notifier(&vxlan_notifier_block);
3110         if (rc)
3111                 goto out2;
3112
3113         rc = rtnl_link_register(&vxlan_link_ops);
3114         if (rc)
3115                 goto out3;
3116
3117         return 0;
3118 out3:
3119         unregister_netdevice_notifier(&vxlan_notifier_block);
3120 out2:
3121         unregister_pernet_subsys(&vxlan_net_ops);
3122 out1:
3123         destroy_workqueue(vxlan_wq);
3124         return rc;
3125 }
3126 late_initcall(vxlan_init_module);
3127
3128 static void __exit vxlan_cleanup_module(void)
3129 {
3130         rtnl_link_unregister(&vxlan_link_ops);
3131         unregister_netdevice_notifier(&vxlan_notifier_block);
3132         destroy_workqueue(vxlan_wq);
3133         unregister_pernet_subsys(&vxlan_net_ops);
3134         /* rcu_barrier() is called by netns */
3135 }
3136 module_exit(vxlan_cleanup_module);
3137
3138 MODULE_LICENSE("GPL");
3139 MODULE_VERSION(VXLAN_VERSION);
3140 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3141 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3142 MODULE_ALIAS_RTNL_LINK("vxlan");