drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/errno.h>
  16 #include <linux/slab.h>
  17 #include <linux/udp.h>
  18 #include <linux/igmp.h>
  19 #include <linux/if_ether.h>
  20 #include <linux/ethtool.h>
  21 #include <net/arp.h>
  22 #include <net/ndisc.h>
  23 #include <net/ip.h>
  24 #include <net/icmp.h>
  25 #include <net/rtnetlink.h>
  26 #include <net/inet_ecn.h>
  27 #include <net/net_namespace.h>
  28 #include <net/netns/generic.h>
  29 #include <net/vxlan.h>
  30
  31 #if IS_ENABLED(CONFIG_IPV6)
  32 #include <net/ip6_tunnel.h>
  33 #include <net/ip6_checksum.h>
  34 #endif
  35
  36 #define VXLAN_VERSION   "0.1"
  37
  38 #define PORT_HASH_BITS  8
  39 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  40 #define FDB_AGE_DEFAULT 300 /* 5 min */
  41 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  42
  43 /* UDP port for VXLAN traffic.
  44  * The IANA assigned port is 4789, but the Linux default is 8472
  45  * for compatibility with early adopters.
  46  */
  47 static unsigned short vxlan_port __read_mostly = 8472;
  48 module_param_named(udp_port, vxlan_port, ushort, 0444);
  49 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  50
  51 static bool log_ecn_error = true;
  52 module_param(log_ecn_error, bool, 0644);
  53 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  54
  55 static unsigned int vxlan_net_id;
  56 static struct rtnl_link_ops vxlan_link_ops;
  57
  58 static const u8 all_zeros_mac[ETH_ALEN + 2];
  59
  60 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  61
  62 /* per-network namespace private data for this module */
  63 struct vxlan_net {
  64         struct list_head  vxlan_list;
  65         struct hlist_head sock_list[PORT_HASH_SIZE];
  66         spinlock_t        sock_lock;
  67 };
  68
  69 /* Forwarding table entry */
  70 struct vxlan_fdb {
  71         struct hlist_node hlist;        /* linked list of entries */
  72         struct rcu_head   rcu;
  73         unsigned long     updated;      /* jiffies */
  74         unsigned long     used;
  75         struct list_head  remotes;
  76         u8                eth_addr[ETH_ALEN];
  77         u16               state;        /* see ndm_state */
  78         __be32            vni;
  79         u8                flags;        /* see ndm_flags */
  80 };
  81
  82 /* salt for hash table */
  83 static u32 vxlan_salt __read_mostly;
  84
  85 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  86 {
  87         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  88                ip_tunnel_collect_metadata();
  89 }
  90
  91 #if IS_ENABLED(CONFIG_IPV6)
  92 static inline
  93 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
  94 {
  95         if (a->sa.sa_family != b->sa.sa_family)
  96                 return false;
  97         if (a->sa.sa_family == AF_INET6)
  98                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
  99         else
 100                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 101 }
 102
 103 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 104 {
 105         if (ipa->sa.sa_family == AF_INET6)
 106                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 107         else
 108                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 109 }
 110
 111 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 112 {
 113         if (ipa->sa.sa_family == AF_INET6)
 114                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 115         else
 116                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 117 }
 118
 119 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 120 {
 121         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 122                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 123                 ip->sa.sa_family = AF_INET6;
 124                 return 0;
 125         } else if (nla_len(nla) >= sizeof(__be32)) {
 126                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 127                 ip->sa.sa_family = AF_INET;
 128                 return 0;
 129         } else {
 130                 return -EAFNOSUPPORT;
 131         }
 132 }
 133
 134 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 135                               const union vxlan_addr *ip)
 136 {
 137         if (ip->sa.sa_family == AF_INET6)
 138                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 139         else
 140                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 141 }
 142
 143 #else /* !CONFIG_IPV6 */
 144
 145 static inline
 146 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 147 {
 148         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 149 }
 150
 151 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 152 {
 153         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 154 }
 155
 156 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 157 {
 158         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 159 }
 160
 161 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 162 {
 163         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 164                 return -EAFNOSUPPORT;
 165         } else if (nla_len(nla) >= sizeof(__be32)) {
 166                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 167                 ip->sa.sa_family = AF_INET;
 168                 return 0;
 169         } else {
 170                 return -EAFNOSUPPORT;
 171         }
 172 }
 173
 174 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 175                               const union vxlan_addr *ip)
 176 {
 177         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 178 }
 179 #endif
 180
 181 /* Virtual Network hash table head */
 182 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 183 {
 184         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 185 }
 186
 187 /* Socket hash table head */
 188 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 189 {
 190         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 191
 192         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 193 }
 194
 195 /* First remote destination for a forwarding entry.
 196  * Guaranteed to be non-NULL because remotes are never deleted.
 197  */
 198 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 199 {
 200         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 201 }
 202
 203 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 204 {
 205         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 206 }
 207
 208 /* Find VXLAN socket based on network namespace, address family and UDP port
 209  * and enabled unshareable flags.
 210  */
 211 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 212                                           __be16 port, u32 flags)
 213 {
 214         struct vxlan_sock *vs;
 215
 216         flags &= VXLAN_F_RCV_FLAGS;
 217
 218         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 219                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 220                     vxlan_get_sk_family(vs) == family &&
 221                     vs->flags == flags)
 222                         return vs;
 223         }
 224         return NULL;
 225 }
 226
 227 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
 228 {
 229         struct vxlan_dev *vxlan;
 230
 231         /* For flow based devices, map all packets to VNI 0 */
 232         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 233                 vni = 0;
 234
 235         hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
 236                 if (vxlan->default_dst.remote_vni == vni)
 237                         return vxlan;
 238         }
 239
 240         return NULL;
 241 }
 242
 243 /* Look up VNI in a per net namespace table */
 244 static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
 245                                         sa_family_t family, __be16 port,
 246                                         u32 flags)
 247 {
 248         struct vxlan_sock *vs;
 249
 250         vs = vxlan_find_sock(net, family, port, flags);
 251         if (!vs)
 252                 return NULL;
 253
 254         return vxlan_vs_find_vni(vs, vni);
 255 }
 256
 257 /* Fill in neighbour message in skbuff. */
 258 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 259                           const struct vxlan_fdb *fdb,
 260                           u32 portid, u32 seq, int type, unsigned int flags,
 261                           const struct vxlan_rdst *rdst)
 262 {
 263         unsigned long now = jiffies;
 264         struct nda_cacheinfo ci;
 265         struct nlmsghdr *nlh;
 266         struct ndmsg *ndm;
 267         bool send_ip, send_eth;
 268
 269         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 270         if (nlh == NULL)
 271                 return -EMSGSIZE;
 272
 273         ndm = nlmsg_data(nlh);
 274         memset(ndm, 0, sizeof(*ndm));
 275
 276         send_eth = send_ip = true;
 277
 278         if (type == RTM_GETNEIGH) {
 279                 ndm->ndm_family = AF_INET;
 280                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 281                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 282         } else
 283                 ndm->ndm_family = AF_BRIDGE;
 284         ndm->ndm_state = fdb->state;
 285         ndm->ndm_ifindex = vxlan->dev->ifindex;
 286         ndm->ndm_flags = fdb->flags;
 287         ndm->ndm_type = RTN_UNICAST;
 288
 289         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 290             nla_put_s32(skb, NDA_LINK_NETNSID,
 291                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 292                 goto nla_put_failure;
 293
 294         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 295                 goto nla_put_failure;
 296
 297         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 298                 goto nla_put_failure;
 299
 300         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 301             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 302                 goto nla_put_failure;
 303         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 304             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 305                 goto nla_put_failure;
 306         if ((vxlan->flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
 307             nla_put_u32(skb, NDA_SRC_VNI,
 308                         be32_to_cpu(fdb->vni)))
 309                 goto nla_put_failure;
 310         if (rdst->remote_ifindex &&
 311             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 312                 goto nla_put_failure;
 313
 314         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 315         ci.ndm_confirmed = 0;
 316         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 317         ci.ndm_refcnt    = 0;
 318
 319         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 320                 goto nla_put_failure;
 321
 322         nlmsg_end(skb, nlh);
 323         return 0;
 324
 325 nla_put_failure:
 326         nlmsg_cancel(skb, nlh);
 327         return -EMSGSIZE;
 328 }
 329
 330 static inline size_t vxlan_nlmsg_size(void)
 331 {
 332         return NLMSG_ALIGN(sizeof(struct ndmsg))
 333                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 334                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 335                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 336                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 337                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 338                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 339                 + nla_total_size(sizeof(struct nda_cacheinfo));
 340 }
 341
 342 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 343                              struct vxlan_rdst *rd, int type)
 344 {
 345         struct net *net = dev_net(vxlan->dev);
 346         struct sk_buff *skb;
 347         int err = -ENOBUFS;
 348
 349         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 350         if (skb == NULL)
 351                 goto errout;
 352
 353         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 354         if (err < 0) {
 355                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 356                 WARN_ON(err == -EMSGSIZE);
 357                 kfree_skb(skb);
 358                 goto errout;
 359         }
 360
 361         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 362         return;
 363 errout:
 364         if (err < 0)
 365                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 366 }
 367
 368 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 369 {
 370         struct vxlan_dev *vxlan = netdev_priv(dev);
 371         struct vxlan_fdb f = {
 372                 .state = NUD_STALE,
 373         };
 374         struct vxlan_rdst remote = {
 375                 .remote_ip = *ipa, /* goes to NDA_DST */
 376                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 377         };
 378
 379         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 380 }
 381
 382 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 383 {
 384         struct vxlan_fdb f = {
 385                 .state = NUD_STALE,
 386         };
 387         struct vxlan_rdst remote = { };
 388
 389         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 390
 391         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 392 }
 393
 394 /* Hash Ethernet address */
 395 static u32 eth_hash(const unsigned char *addr)
 396 {
 397         u64 value = get_unaligned((u64 *)addr);
 398
 399         /* only want 6 bytes */
 400 #ifdef __BIG_ENDIAN
 401         value >>= 16;
 402 #else
 403         value <<= 16;
 404 #endif
 405         return hash_64(value, FDB_HASH_BITS);
 406 }
 407
 408 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
 409 {
 410         /* use 1 byte of OUI and 3 bytes of NIC */
 411         u32 key = get_unaligned((u32 *)(addr + 2));
 412
 413         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
 414 }
 415
 416 /* Hash chain to use given mac address */
 417 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 418                                                 const u8 *mac, __be32 vni)
 419 {
 420         if (vxlan->flags & VXLAN_F_COLLECT_METADATA)
 421                 return &vxlan->fdb_head[eth_vni_hash(mac, vni)];
 422         else
 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, __be32 vni)
 429 {
 430         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 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                         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
 436                                 if (vni == f->vni)
 437                                         return f;
 438                         } else {
 439                                 return f;
 440                         }
 441                 }
 442         }
 443
 444         return NULL;
 445 }
 446
 447 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 448                                         const u8 *mac, __be32 vni)
 449 {
 450         struct vxlan_fdb *f;
 451
 452         f = __vxlan_find_mac(vxlan, mac, vni);
 453         if (f)
 454                 f->used = jiffies;
 455
 456         return f;
 457 }
 458
 459 /* caller should hold vxlan->hash_lock */
 460 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 461                                               union vxlan_addr *ip, __be16 port,
 462                                               __be32 vni, __u32 ifindex)
 463 {
 464         struct vxlan_rdst *rd;
 465
 466         list_for_each_entry(rd, &f->remotes, list) {
 467                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 468                     rd->remote_port == port &&
 469                     rd->remote_vni == vni &&
 470                     rd->remote_ifindex == ifindex)
 471                         return rd;
 472         }
 473
 474         return NULL;
 475 }
 476
 477 /* Replace destination of unicast mac */
 478 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 479                              union vxlan_addr *ip, __be16 port, __be32 vni,
 480                              __u32 ifindex)
 481 {
 482         struct vxlan_rdst *rd;
 483
 484         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 485         if (rd)
 486                 return 0;
 487
 488         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 489         if (!rd)
 490                 return 0;
 491
 492         dst_cache_reset(&rd->dst_cache);
 493         rd->remote_ip = *ip;
 494         rd->remote_port = port;
 495         rd->remote_vni = vni;
 496         rd->remote_ifindex = ifindex;
 497         return 1;
 498 }
 499
 500 /* Add/update destinations for multicast */
 501 static int vxlan_fdb_append(struct vxlan_fdb *f,
 502                             union vxlan_addr *ip, __be16 port, __be32 vni,
 503                             __u32 ifindex, struct vxlan_rdst **rdp)
 504 {
 505         struct vxlan_rdst *rd;
 506
 507         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 508         if (rd)
 509                 return 0;
 510
 511         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 512         if (rd == NULL)
 513                 return -ENOBUFS;
 514
 515         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 516                 kfree(rd);
 517                 return -ENOBUFS;
 518         }
 519
 520         rd->remote_ip = *ip;
 521         rd->remote_port = port;
 522         rd->remote_vni = vni;
 523         rd->remote_ifindex = ifindex;
 524
 525         list_add_tail_rcu(&rd->list, &f->remotes);
 526
 527         *rdp = rd;
 528         return 1;
 529 }
 530
 531 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 532                                           unsigned int off,
 533                                           struct vxlanhdr *vh, size_t hdrlen,
 534                                           __be32 vni_field,
 535                                           struct gro_remcsum *grc,
 536                                           bool nopartial)
 537 {
 538         size_t start, offset;
 539
 540         if (skb->remcsum_offload)
 541                 return vh;
 542
 543         if (!NAPI_GRO_CB(skb)->csum_valid)
 544                 return NULL;
 545
 546         start = vxlan_rco_start(vni_field);
 547         offset = start + vxlan_rco_offset(vni_field);
 548
 549         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 550                                      start, offset, grc, nopartial);
 551
 552         skb->remcsum_offload = 1;
 553
 554         return vh;
 555 }
 556
 557 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
 558                                           struct sk_buff **head,
 559                                           struct sk_buff *skb)
 560 {
 561         struct sk_buff *p, **pp = NULL;
 562         struct vxlanhdr *vh, *vh2;
 563         unsigned int hlen, off_vx;
 564         int flush = 1;
 565         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 566         __be32 flags;
 567         struct gro_remcsum grc;
 568
 569         skb_gro_remcsum_init(&grc);
 570
 571         off_vx = skb_gro_offset(skb);
 572         hlen = off_vx + sizeof(*vh);
 573         vh   = skb_gro_header_fast(skb, off_vx);
 574         if (skb_gro_header_hard(skb, hlen)) {
 575                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 576                 if (unlikely(!vh))
 577                         goto out;
 578         }
 579
 580         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 581
 582         flags = vh->vx_flags;
 583
 584         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 585                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 586                                        vh->vx_vni, &grc,
 587                                        !!(vs->flags &
 588                                           VXLAN_F_REMCSUM_NOPARTIAL));
 589
 590                 if (!vh)
 591                         goto out;
 592         }
 593
 594         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 595
 596         for (p = *head; p; p = p->next) {
 597                 if (!NAPI_GRO_CB(p)->same_flow)
 598                         continue;
 599
 600                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 601                 if (vh->vx_flags != vh2->vx_flags ||
 602                     vh->vx_vni != vh2->vx_vni) {
 603                         NAPI_GRO_CB(p)->same_flow = 0;
 604                         continue;
 605                 }
 606         }
 607
 608         pp = call_gro_receive(eth_gro_receive, head, skb);
 609         flush = 0;
 610
 611 out:
 612         skb_gro_remcsum_cleanup(skb, &grc);
 613         NAPI_GRO_CB(skb)->flush |= flush;
 614
 615         return pp;
 616 }
 617
 618 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 619 {
 620         /* Sets 'skb->inner_mac_header' since we are always called with
 621          * 'skb->encapsulation' set.
 622          */
 623         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 624 }
 625
 626 /* Add new entry to forwarding table -- assumes lock held */
 627 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 628                             const u8 *mac, union vxlan_addr *ip,
 629                             __u16 state, __u16 flags,
 630                             __be16 port, __be32 src_vni, __be32 vni,
 631                             __u32 ifindex, __u8 ndm_flags)
 632 {
 633         struct vxlan_rdst *rd = NULL;
 634         struct vxlan_fdb *f;
 635         int notify = 0;
 636         int rc;
 637
 638         f = __vxlan_find_mac(vxlan, mac, src_vni);
 639         if (f) {
 640                 if (flags & NLM_F_EXCL) {
 641                         netdev_dbg(vxlan->dev,
 642                                    "lost race to create %pM\n", mac);
 643                         return -EEXIST;
 644                 }
 645                 if (f->state != state) {
 646                         f->state = state;
 647                         f->updated = jiffies;
 648                         notify = 1;
 649                 }
 650                 if (f->flags != ndm_flags) {
 651                         f->flags = ndm_flags;
 652                         f->updated = jiffies;
 653                         notify = 1;
 654                 }
 655                 if ((flags & NLM_F_REPLACE)) {
 656                         /* Only change unicasts */
 657                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 658                              is_zero_ether_addr(f->eth_addr))) {
 659                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 660                                                            ifindex);
 661                         } else
 662                                 return -EOPNOTSUPP;
 663                 }
 664                 if ((flags & NLM_F_APPEND) &&
 665                     (is_multicast_ether_addr(f->eth_addr) ||
 666                      is_zero_ether_addr(f->eth_addr))) {
 667                         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 668
 669                         if (rc < 0)
 670                                 return rc;
 671                         notify |= rc;
 672                 }
 673         } else {
 674                 if (!(flags & NLM_F_CREATE))
 675                         return -ENOENT;
 676
 677                 if (vxlan->cfg.addrmax &&
 678                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 679                         return -ENOSPC;
 680
 681                 /* Disallow replace to add a multicast entry */
 682                 if ((flags & NLM_F_REPLACE) &&
 683                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 684                         return -EOPNOTSUPP;
 685
 686                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 687                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 688                 if (!f)
 689                         return -ENOMEM;
 690
 691                 notify = 1;
 692                 f->state = state;
 693                 f->flags = ndm_flags;
 694                 f->updated = f->used = jiffies;
 695                 f->vni = src_vni;
 696                 INIT_LIST_HEAD(&f->remotes);
 697                 memcpy(f->eth_addr, mac, ETH_ALEN);
 698
 699                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 700                 if (rc < 0) {
 701                         kfree(f);
 702                         return rc;
 703                 }
 704
 705                 ++vxlan->addrcnt;
 706                 hlist_add_head_rcu(&f->hlist,
 707                                    vxlan_fdb_head(vxlan, mac, src_vni));
 708         }
 709
 710         if (notify) {
 711                 if (rd == NULL)
 712                         rd = first_remote_rtnl(f);
 713                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 714         }
 715
 716         return 0;
 717 }
 718
 719 static void vxlan_fdb_free(struct rcu_head *head)
 720 {
 721         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 722         struct vxlan_rdst *rd, *nd;
 723
 724         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 725                 dst_cache_destroy(&rd->dst_cache);
 726                 kfree(rd);
 727         }
 728         kfree(f);
 729 }
 730
 731 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 732 {
 733         netdev_dbg(vxlan->dev,
 734                     "delete %pM\n", f->eth_addr);
 735
 736         --vxlan->addrcnt;
 737         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 738
 739         hlist_del_rcu(&f->hlist);
 740         call_rcu(&f->rcu, vxlan_fdb_free);
 741 }
 742
 743 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 744                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
 745                            __be32 *vni, u32 *ifindex)
 746 {
 747         struct net *net = dev_net(vxlan->dev);
 748         int err;
 749
 750         if (tb[NDA_DST]) {
 751                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 752                 if (err)
 753                         return err;
 754         } else {
 755                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 756                 if (remote->sa.sa_family == AF_INET) {
 757                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 758                         ip->sa.sa_family = AF_INET;
 759 #if IS_ENABLED(CONFIG_IPV6)
 760                 } else {
 761                         ip->sin6.sin6_addr = in6addr_any;
 762                         ip->sa.sa_family = AF_INET6;
 763 #endif
 764                 }
 765         }
 766
 767         if (tb[NDA_PORT]) {
 768                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 769                         return -EINVAL;
 770                 *port = nla_get_be16(tb[NDA_PORT]);
 771         } else {
 772                 *port = vxlan->cfg.dst_port;
 773         }
 774
 775         if (tb[NDA_VNI]) {
 776                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 777                         return -EINVAL;
 778                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 779         } else {
 780                 *vni = vxlan->default_dst.remote_vni;
 781         }
 782
 783         if (tb[NDA_SRC_VNI]) {
 784                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
 785                         return -EINVAL;
 786                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
 787         } else {
 788                 *src_vni = vxlan->default_dst.remote_vni;
 789         }
 790
 791         if (tb[NDA_IFINDEX]) {
 792                 struct net_device *tdev;
 793
 794                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 795                         return -EINVAL;
 796                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 797                 tdev = __dev_get_by_index(net, *ifindex);
 798                 if (!tdev)
 799                         return -EADDRNOTAVAIL;
 800         } else {
 801                 *ifindex = 0;
 802         }
 803
 804         return 0;
 805 }
 806
 807 /* Add static entry (via netlink) */
 808 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 809                          struct net_device *dev,
 810                          const unsigned char *addr, u16 vid, u16 flags)
 811 {
 812         struct vxlan_dev *vxlan = netdev_priv(dev);
 813         /* struct net *net = dev_net(vxlan->dev); */
 814         union vxlan_addr ip;
 815         __be16 port;
 816         __be32 src_vni, vni;
 817         u32 ifindex;
 818         int err;
 819
 820         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 821                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 822                         ndm->ndm_state);
 823                 return -EINVAL;
 824         }
 825
 826         if (tb[NDA_DST] == NULL)
 827                 return -EINVAL;
 828
 829         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 830         if (err)
 831                 return err;
 832
 833         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 834                 return -EAFNOSUPPORT;
 835
 836         spin_lock_bh(&vxlan->hash_lock);
 837         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 838                                port, src_vni, vni, ifindex, ndm->ndm_flags);
 839         spin_unlock_bh(&vxlan->hash_lock);
 840
 841         return err;
 842 }
 843
 844 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
 845                               const unsigned char *addr, union vxlan_addr ip,
 846                               __be16 port, __be32 src_vni, u32 vni, u32 ifindex,
 847                               u16 vid)
 848 {
 849         struct vxlan_fdb *f;
 850         struct vxlan_rdst *rd = NULL;
 851         int err = -ENOENT;
 852
 853         f = vxlan_find_mac(vxlan, addr, src_vni);
 854         if (!f)
 855                 return err;
 856
 857         if (!vxlan_addr_any(&ip)) {
 858                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 859                 if (!rd)
 860                         goto out;
 861         }
 862
 863         /* remove a destination if it's not the only one on the list,
 864          * otherwise destroy the fdb entry
 865          */
 866         if (rd && !list_is_singular(&f->remotes)) {
 867                 list_del_rcu(&rd->list);
 868                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 869                 kfree_rcu(rd, rcu);
 870                 goto out;
 871         }
 872
 873         vxlan_fdb_destroy(vxlan, f);
 874
 875 out:
 876         return 0;
 877 }
 878
 879 /* Delete entry (via netlink) */
 880 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 881                             struct net_device *dev,
 882                             const unsigned char *addr, u16 vid)
 883 {
 884         struct vxlan_dev *vxlan = netdev_priv(dev);
 885         union vxlan_addr ip;
 886         __be32 src_vni, vni;
 887         __be16 port;
 888         u32 ifindex;
 889         int err;
 890
 891         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 892         if (err)
 893                 return err;
 894
 895         spin_lock_bh(&vxlan->hash_lock);
 896         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
 897                                  vid);
 898         spin_unlock_bh(&vxlan->hash_lock);
 899
 900         return err;
 901 }
 902
 903 /* Dump forwarding table */
 904 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 905                           struct net_device *dev,
 906                           struct net_device *filter_dev, int *idx)
 907 {
 908         struct vxlan_dev *vxlan = netdev_priv(dev);
 909         unsigned int h;
 910         int err = 0;
 911
 912         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 913                 struct vxlan_fdb *f;
 914
 915                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 916                         struct vxlan_rdst *rd;
 917
 918                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 919                                 if (*idx < cb->args[2])
 920                                         goto skip;
 921
 922                                 err = vxlan_fdb_info(skb, vxlan, f,
 923                                                      NETLINK_CB(cb->skb).portid,
 924                                                      cb->nlh->nlmsg_seq,
 925                                                      RTM_NEWNEIGH,
 926                                                      NLM_F_MULTI, rd);
 927                                 if (err < 0)
 928                                         goto out;
 929 skip:
 930                                 *idx += 1;
 931                         }
 932                 }
 933         }
 934 out:
 935         return err;
 936 }
 937
 938 /* Watch incoming packets to learn mapping between Ethernet address
 939  * and Tunnel endpoint.
 940  * Return true if packet is bogus and should be dropped.
 941  */
 942 static bool vxlan_snoop(struct net_device *dev,
 943                         union vxlan_addr *src_ip, const u8 *src_mac,
 944                         __be32 vni)
 945 {
 946         struct vxlan_dev *vxlan = netdev_priv(dev);
 947         struct vxlan_fdb *f;
 948
 949         f = vxlan_find_mac(vxlan, src_mac, vni);
 950         if (likely(f)) {
 951                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 952
 953                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 954                         return false;
 955
 956                 /* Don't migrate static entries, drop packets */
 957                 if (f->state & NUD_NOARP)
 958                         return true;
 959
 960                 if (net_ratelimit())
 961                         netdev_info(dev,
 962                                     "%pM migrated from %pIS to %pIS\n",
 963                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 964
 965                 rdst->remote_ip = *src_ip;
 966                 f->updated = jiffies;
 967                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 968         } else {
 969                 /* learned new entry */
 970                 spin_lock(&vxlan->hash_lock);
 971
 972                 /* close off race between vxlan_flush and incoming packets */
 973                 if (netif_running(dev))
 974                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 975                                          NUD_REACHABLE,
 976                                          NLM_F_EXCL|NLM_F_CREATE,
 977                                          vxlan->cfg.dst_port,
 978                                          vni,
 979                                          vxlan->default_dst.remote_vni,
 980                                          0, NTF_SELF);
 981                 spin_unlock(&vxlan->hash_lock);
 982         }
 983
 984         return false;
 985 }
 986
 987 /* See if multicast group is already in use by other ID */
 988 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
 989 {
 990         struct vxlan_dev *vxlan;
 991         struct vxlan_sock *sock4;
 992 #if IS_ENABLED(CONFIG_IPV6)
 993         struct vxlan_sock *sock6;
 994 #endif
 995         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
 996
 997         sock4 = rtnl_dereference(dev->vn4_sock);
 998
 999         /* The vxlan_sock is only used by dev, leaving group has
1000          * no effect on other vxlan devices.
1001          */
1002         if (family == AF_INET && sock4 && atomic_read(&sock4->refcnt) == 1)
1003                 return false;
1004 #if IS_ENABLED(CONFIG_IPV6)
1005         sock6 = rtnl_dereference(dev->vn6_sock);
1006         if (family == AF_INET6 && sock6 && atomic_read(&sock6->refcnt) == 1)
1007                 return false;
1008 #endif
1009
1010         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1011                 if (!netif_running(vxlan->dev) || vxlan == dev)
1012                         continue;
1013
1014                 if (family == AF_INET &&
1015                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1016                         continue;
1017 #if IS_ENABLED(CONFIG_IPV6)
1018                 if (family == AF_INET6 &&
1019                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1020                         continue;
1021 #endif
1022
1023                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1024                                       &dev->default_dst.remote_ip))
1025                         continue;
1026
1027                 if (vxlan->default_dst.remote_ifindex !=
1028                     dev->default_dst.remote_ifindex)
1029                         continue;
1030
1031                 return true;
1032         }
1033
1034         return false;
1035 }
1036
1037 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1038 {
1039         struct vxlan_net *vn;
1040
1041         if (!vs)
1042                 return false;
1043         if (!atomic_dec_and_test(&vs->refcnt))
1044                 return false;
1045
1046         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1047         spin_lock(&vn->sock_lock);
1048         hlist_del_rcu(&vs->hlist);
1049         udp_tunnel_notify_del_rx_port(vs->sock,
1050                                       (vs->flags & VXLAN_F_GPE) ?
1051                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1052                                       UDP_TUNNEL_TYPE_VXLAN);
1053         spin_unlock(&vn->sock_lock);
1054
1055         return true;
1056 }
1057
1058 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1059 {
1060         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1061 #if IS_ENABLED(CONFIG_IPV6)
1062         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1063
1064         rcu_assign_pointer(vxlan->vn6_sock, NULL);
1065 #endif
1066
1067         rcu_assign_pointer(vxlan->vn4_sock, NULL);
1068         synchronize_net();
1069
1070         if (__vxlan_sock_release_prep(sock4)) {
1071                 udp_tunnel_sock_release(sock4->sock);
1072                 kfree(sock4);
1073         }
1074
1075 #if IS_ENABLED(CONFIG_IPV6)
1076         if (__vxlan_sock_release_prep(sock6)) {
1077                 udp_tunnel_sock_release(sock6->sock);
1078                 kfree(sock6);
1079         }
1080 #endif
1081 }
1082
1083 /* Update multicast group membership when first VNI on
1084  * multicast address is brought up
1085  */
1086 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1087 {
1088         struct sock *sk;
1089         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1090         int ifindex = vxlan->default_dst.remote_ifindex;
1091         int ret = -EINVAL;
1092
1093         if (ip->sa.sa_family == AF_INET) {
1094                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1095                 struct ip_mreqn mreq = {
1096                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1097                         .imr_ifindex            = ifindex,
1098                 };
1099
1100                 sk = sock4->sock->sk;
1101                 lock_sock(sk);
1102                 ret = ip_mc_join_group(sk, &mreq);
1103                 release_sock(sk);
1104 #if IS_ENABLED(CONFIG_IPV6)
1105         } else {
1106                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1107
1108                 sk = sock6->sock->sk;
1109                 lock_sock(sk);
1110                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1111                                                    &ip->sin6.sin6_addr);
1112                 release_sock(sk);
1113 #endif
1114         }
1115
1116         return ret;
1117 }
1118
1119 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1120 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1121 {
1122         struct sock *sk;
1123         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1124         int ifindex = vxlan->default_dst.remote_ifindex;
1125         int ret = -EINVAL;
1126
1127         if (ip->sa.sa_family == AF_INET) {
1128                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1129                 struct ip_mreqn mreq = {
1130                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1131                         .imr_ifindex            = ifindex,
1132                 };
1133
1134                 sk = sock4->sock->sk;
1135                 lock_sock(sk);
1136                 ret = ip_mc_leave_group(sk, &mreq);
1137                 release_sock(sk);
1138 #if IS_ENABLED(CONFIG_IPV6)
1139         } else {
1140                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1141
1142                 sk = sock6->sock->sk;
1143                 lock_sock(sk);
1144                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1145                                                    &ip->sin6.sin6_addr);
1146                 release_sock(sk);
1147 #endif
1148         }
1149
1150         return ret;
1151 }
1152
1153 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1154                           struct sk_buff *skb, u32 vxflags)
1155 {
1156         size_t start, offset;
1157
1158         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1159                 goto out;
1160
1161         start = vxlan_rco_start(unparsed->vx_vni);
1162         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1163
1164         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1165                 return false;
1166
1167         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1168                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1169 out:
1170         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1171         unparsed->vx_vni &= VXLAN_VNI_MASK;
1172         return true;
1173 }
1174
1175 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1176                                 struct sk_buff *skb, u32 vxflags,
1177                                 struct vxlan_metadata *md)
1178 {
1179         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1180         struct metadata_dst *tun_dst;
1181
1182         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1183                 goto out;
1184
1185         md->gbp = ntohs(gbp->policy_id);
1186
1187         tun_dst = (struct metadata_dst *)skb_dst(skb);
1188         if (tun_dst) {
1189                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1190                 tun_dst->u.tun_info.options_len = sizeof(*md);
1191         }
1192         if (gbp->dont_learn)
1193                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1194
1195         if (gbp->policy_applied)
1196                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1197
1198         /* In flow-based mode, GBP is carried in dst_metadata */
1199         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1200                 skb->mark = md->gbp;
1201 out:
1202         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1203 }
1204
1205 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1206                                 __be16 *protocol,
1207                                 struct sk_buff *skb, u32 vxflags)
1208 {
1209         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1210
1211         /* Need to have Next Protocol set for interfaces in GPE mode. */
1212         if (!gpe->np_applied)
1213                 return false;
1214         /* "The initial version is 0. If a receiver does not support the
1215          * version indicated it MUST drop the packet.
1216          */
1217         if (gpe->version != 0)
1218                 return false;
1219         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1220          * processing MUST occur." However, we don't implement OAM
1221          * processing, thus drop the packet.
1222          */
1223         if (gpe->oam_flag)
1224                 return false;
1225
1226         switch (gpe->next_protocol) {
1227         case VXLAN_GPE_NP_IPV4:
1228                 *protocol = htons(ETH_P_IP);
1229                 break;
1230         case VXLAN_GPE_NP_IPV6:
1231                 *protocol = htons(ETH_P_IPV6);
1232                 break;
1233         case VXLAN_GPE_NP_ETHERNET:
1234                 *protocol = htons(ETH_P_TEB);
1235                 break;
1236         default:
1237                 return false;
1238         }
1239
1240         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1241         return true;
1242 }
1243
1244 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1245                           struct vxlan_sock *vs,
1246                           struct sk_buff *skb, __be32 vni)
1247 {
1248         union vxlan_addr saddr;
1249
1250         skb_reset_mac_header(skb);
1251         skb->protocol = eth_type_trans(skb, vxlan->dev);
1252         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1253
1254         /* Ignore packet loops (and multicast echo) */
1255         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1256                 return false;
1257
1258         /* Get address from the outer IP header */
1259         if (vxlan_get_sk_family(vs) == AF_INET) {
1260                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1261                 saddr.sa.sa_family = AF_INET;
1262 #if IS_ENABLED(CONFIG_IPV6)
1263         } else {
1264                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1265                 saddr.sa.sa_family = AF_INET6;
1266 #endif
1267         }
1268
1269         if ((vxlan->flags & VXLAN_F_LEARN) &&
1270             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, vni))
1271                 return false;
1272
1273         return true;
1274 }
1275
1276 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1277                                   struct sk_buff *skb)
1278 {
1279         int err = 0;
1280
1281         if (vxlan_get_sk_family(vs) == AF_INET)
1282                 err = IP_ECN_decapsulate(oiph, skb);
1283 #if IS_ENABLED(CONFIG_IPV6)
1284         else
1285                 err = IP6_ECN_decapsulate(oiph, skb);
1286 #endif
1287
1288         if (unlikely(err) && log_ecn_error) {
1289                 if (vxlan_get_sk_family(vs) == AF_INET)
1290                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1291                                              &((struct iphdr *)oiph)->saddr,
1292                                              ((struct iphdr *)oiph)->tos);
1293                 else
1294                         net_info_ratelimited("non-ECT from %pI6\n",
1295                                              &((struct ipv6hdr *)oiph)->saddr);
1296         }
1297         return err <= 1;
1298 }
1299
1300 /* Callback from net/ipv4/udp.c to receive packets */
1301 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1302 {
1303         struct pcpu_sw_netstats *stats;
1304         struct vxlan_dev *vxlan;
1305         struct vxlan_sock *vs;
1306         struct vxlanhdr unparsed;
1307         struct vxlan_metadata _md;
1308         struct vxlan_metadata *md = &_md;
1309         __be16 protocol = htons(ETH_P_TEB);
1310         bool raw_proto = false;
1311         void *oiph;
1312         __be32 vni = 0;
1313
1314         /* Need UDP and VXLAN header to be present */
1315         if (!pskb_may_pull(skb, VXLAN_HLEN))
1316                 goto drop;
1317
1318         unparsed = *vxlan_hdr(skb);
1319         /* VNI flag always required to be set */
1320         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1321                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1322                            ntohl(vxlan_hdr(skb)->vx_flags),
1323                            ntohl(vxlan_hdr(skb)->vx_vni));
1324                 /* Return non vxlan pkt */
1325                 goto drop;
1326         }
1327         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1328         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1329
1330         vs = rcu_dereference_sk_user_data(sk);
1331         if (!vs)
1332                 goto drop;
1333
1334         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1335
1336         vxlan = vxlan_vs_find_vni(vs, vni);
1337         if (!vxlan)
1338                 goto drop;
1339
1340         /* For backwards compatibility, only allow reserved fields to be
1341          * used by VXLAN extensions if explicitly requested.
1342          */
1343         if (vs->flags & VXLAN_F_GPE) {
1344                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1345                         goto drop;
1346                 raw_proto = true;
1347         }
1348
1349         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1350                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1351                         goto drop;
1352
1353         if (vxlan_collect_metadata(vs)) {
1354                 struct metadata_dst *tun_dst;
1355
1356                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1357                                          key32_to_tunnel_id(vni), sizeof(*md));
1358
1359                 if (!tun_dst)
1360                         goto drop;
1361
1362                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1363
1364                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1365         } else {
1366                 memset(md, 0, sizeof(*md));
1367         }
1368
1369         if (vs->flags & VXLAN_F_REMCSUM_RX)
1370                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1371                         goto drop;
1372         if (vs->flags & VXLAN_F_GBP)
1373                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1374         /* Note that GBP and GPE can never be active together. This is
1375          * ensured in vxlan_dev_configure.
1376          */
1377
1378         if (unparsed.vx_flags || unparsed.vx_vni) {
1379                 /* If there are any unprocessed flags remaining treat
1380                  * this as a malformed packet. This behavior diverges from
1381                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1382                  * in reserved fields are to be ignored. The approach here
1383                  * maintains compatibility with previous stack code, and also
1384                  * is more robust and provides a little more security in
1385                  * adding extensions to VXLAN.
1386                  */
1387                 goto drop;
1388         }
1389
1390         if (!raw_proto) {
1391                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1392                         goto drop;
1393         } else {
1394                 skb_reset_mac_header(skb);
1395                 skb->dev = vxlan->dev;
1396                 skb->pkt_type = PACKET_HOST;
1397         }
1398
1399         oiph = skb_network_header(skb);
1400         skb_reset_network_header(skb);
1401
1402         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1403                 ++vxlan->dev->stats.rx_frame_errors;
1404                 ++vxlan->dev->stats.rx_errors;
1405                 goto drop;
1406         }
1407
1408         stats = this_cpu_ptr(vxlan->dev->tstats);
1409         u64_stats_update_begin(&stats->syncp);
1410         stats->rx_packets++;
1411         stats->rx_bytes += skb->len;
1412         u64_stats_update_end(&stats->syncp);
1413
1414         gro_cells_receive(&vxlan->gro_cells, skb);
1415         return 0;
1416
1417 drop:
1418         /* Consume bad packet */
1419         kfree_skb(skb);
1420         return 0;
1421 }
1422
1423 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1424 {
1425         struct vxlan_dev *vxlan = netdev_priv(dev);
1426         struct arphdr *parp;
1427         u8 *arpptr, *sha;
1428         __be32 sip, tip;
1429         struct neighbour *n;
1430
1431         if (dev->flags & IFF_NOARP)
1432                 goto out;
1433
1434         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1435                 dev->stats.tx_dropped++;
1436                 goto out;
1437         }
1438         parp = arp_hdr(skb);
1439
1440         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1441              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1442             parp->ar_pro != htons(ETH_P_IP) ||
1443             parp->ar_op != htons(ARPOP_REQUEST) ||
1444             parp->ar_hln != dev->addr_len ||
1445             parp->ar_pln != 4)
1446                 goto out;
1447         arpptr = (u8 *)parp + sizeof(struct arphdr);
1448         sha = arpptr;
1449         arpptr += dev->addr_len;        /* sha */
1450         memcpy(&sip, arpptr, sizeof(sip));
1451         arpptr += sizeof(sip);
1452         arpptr += dev->addr_len;        /* tha */
1453         memcpy(&tip, arpptr, sizeof(tip));
1454
1455         if (ipv4_is_loopback(tip) ||
1456             ipv4_is_multicast(tip))
1457                 goto out;
1458
1459         n = neigh_lookup(&arp_tbl, &tip, dev);
1460
1461         if (n) {
1462                 struct vxlan_fdb *f;
1463                 struct sk_buff  *reply;
1464
1465                 if (!(n->nud_state & NUD_CONNECTED)) {
1466                         neigh_release(n);
1467                         goto out;
1468                 }
1469
1470                 f = vxlan_find_mac(vxlan, n->ha, vni);
1471                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1472                         /* bridge-local neighbor */
1473                         neigh_release(n);
1474                         goto out;
1475                 }
1476
1477                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1478                                 n->ha, sha);
1479
1480                 neigh_release(n);
1481
1482                 if (reply == NULL)
1483                         goto out;
1484
1485                 skb_reset_mac_header(reply);
1486                 __skb_pull(reply, skb_network_offset(reply));
1487                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1488                 reply->pkt_type = PACKET_HOST;
1489
1490                 if (netif_rx_ni(reply) == NET_RX_DROP)
1491                         dev->stats.rx_dropped++;
1492         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1493                 union vxlan_addr ipa = {
1494                         .sin.sin_addr.s_addr = tip,
1495                         .sin.sin_family = AF_INET,
1496                 };
1497
1498                 vxlan_ip_miss(dev, &ipa);
1499         }
1500 out:
1501         consume_skb(skb);
1502         return NETDEV_TX_OK;
1503 }
1504
1505 #if IS_ENABLED(CONFIG_IPV6)
1506 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1507         struct neighbour *n, bool isrouter)
1508 {
1509         struct net_device *dev = request->dev;
1510         struct sk_buff *reply;
1511         struct nd_msg *ns, *na;
1512         struct ipv6hdr *pip6;
1513         u8 *daddr;
1514         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1515         int ns_olen;
1516         int i, len;
1517
1518         if (dev == NULL)
1519                 return NULL;
1520
1521         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1522                 sizeof(*na) + na_olen + dev->needed_tailroom;
1523         reply = alloc_skb(len, GFP_ATOMIC);
1524         if (reply == NULL)
1525                 return NULL;
1526
1527         reply->protocol = htons(ETH_P_IPV6);
1528         reply->dev = dev;
1529         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1530         skb_push(reply, sizeof(struct ethhdr));
1531         skb_reset_mac_header(reply);
1532
1533         ns = (struct nd_msg *)skb_transport_header(request);
1534
1535         daddr = eth_hdr(request)->h_source;
1536         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1537         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1538                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1539                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1540                         break;
1541                 }
1542         }
1543
1544         /* Ethernet header */
1545         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1546         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1547         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1548         reply->protocol = htons(ETH_P_IPV6);
1549
1550         skb_pull(reply, sizeof(struct ethhdr));
1551         skb_reset_network_header(reply);
1552         skb_put(reply, sizeof(struct ipv6hdr));
1553
1554         /* IPv6 header */
1555
1556         pip6 = ipv6_hdr(reply);
1557         memset(pip6, 0, sizeof(struct ipv6hdr));
1558         pip6->version = 6;
1559         pip6->priority = ipv6_hdr(request)->priority;
1560         pip6->nexthdr = IPPROTO_ICMPV6;
1561         pip6->hop_limit = 255;
1562         pip6->daddr = ipv6_hdr(request)->saddr;
1563         pip6->saddr = *(struct in6_addr *)n->primary_key;
1564
1565         skb_pull(reply, sizeof(struct ipv6hdr));
1566         skb_reset_transport_header(reply);
1567
1568         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1569
1570         /* Neighbor Advertisement */
1571         memset(na, 0, sizeof(*na)+na_olen);
1572         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1573         na->icmph.icmp6_router = isrouter;
1574         na->icmph.icmp6_override = 1;
1575         na->icmph.icmp6_solicited = 1;
1576         na->target = ns->target;
1577         ether_addr_copy(&na->opt[2], n->ha);
1578         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1579         na->opt[1] = na_olen >> 3;
1580
1581         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1582                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1583                 csum_partial(na, sizeof(*na)+na_olen, 0));
1584
1585         pip6->payload_len = htons(sizeof(*na)+na_olen);
1586
1587         skb_push(reply, sizeof(struct ipv6hdr));
1588
1589         reply->ip_summed = CHECKSUM_UNNECESSARY;
1590
1591         return reply;
1592 }
1593
1594 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1595 {
1596         struct vxlan_dev *vxlan = netdev_priv(dev);
1597         struct nd_msg *msg;
1598         const struct ipv6hdr *iphdr;
1599         const struct in6_addr *daddr;
1600         struct neighbour *n;
1601         struct inet6_dev *in6_dev;
1602
1603         in6_dev = __in6_dev_get(dev);
1604         if (!in6_dev)
1605                 goto out;
1606
1607         iphdr = ipv6_hdr(skb);
1608         daddr = &iphdr->daddr;
1609
1610         msg = (struct nd_msg *)skb_transport_header(skb);
1611         if (msg->icmph.icmp6_code != 0 ||
1612             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1613                 goto out;
1614
1615         if (ipv6_addr_loopback(daddr) ||
1616             ipv6_addr_is_multicast(&msg->target))
1617                 goto out;
1618
1619         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1620
1621         if (n) {
1622                 struct vxlan_fdb *f;
1623                 struct sk_buff *reply;
1624
1625                 if (!(n->nud_state & NUD_CONNECTED)) {
1626                         neigh_release(n);
1627                         goto out;
1628                 }
1629
1630                 f = vxlan_find_mac(vxlan, n->ha, vni);
1631                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1632                         /* bridge-local neighbor */
1633                         neigh_release(n);
1634                         goto out;
1635                 }
1636
1637                 reply = vxlan_na_create(skb, n,
1638                                         !!(f ? f->flags & NTF_ROUTER : 0));
1639
1640                 neigh_release(n);
1641
1642                 if (reply == NULL)
1643                         goto out;
1644
1645                 if (netif_rx_ni(reply) == NET_RX_DROP)
1646                         dev->stats.rx_dropped++;
1647
1648         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1649                 union vxlan_addr ipa = {
1650                         .sin6.sin6_addr = msg->target,
1651                         .sin6.sin6_family = AF_INET6,
1652                 };
1653
1654                 vxlan_ip_miss(dev, &ipa);
1655         }
1656
1657 out:
1658         consume_skb(skb);
1659         return NETDEV_TX_OK;
1660 }
1661 #endif
1662
1663 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1664 {
1665         struct vxlan_dev *vxlan = netdev_priv(dev);
1666         struct neighbour *n;
1667
1668         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1669                 return false;
1670
1671         n = NULL;
1672         switch (ntohs(eth_hdr(skb)->h_proto)) {
1673         case ETH_P_IP:
1674         {
1675                 struct iphdr *pip;
1676
1677                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1678                         return false;
1679                 pip = ip_hdr(skb);
1680                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1681                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1682                         union vxlan_addr ipa = {
1683                                 .sin.sin_addr.s_addr = pip->daddr,
1684                                 .sin.sin_family = AF_INET,
1685                         };
1686
1687                         vxlan_ip_miss(dev, &ipa);
1688                         return false;
1689                 }
1690
1691                 break;
1692         }
1693 #if IS_ENABLED(CONFIG_IPV6)
1694         case ETH_P_IPV6:
1695         {
1696                 struct ipv6hdr *pip6;
1697
1698                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1699                         return false;
1700                 pip6 = ipv6_hdr(skb);
1701                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1702                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1703                         union vxlan_addr ipa = {
1704                                 .sin6.sin6_addr = pip6->daddr,
1705                                 .sin6.sin6_family = AF_INET6,
1706                         };
1707
1708                         vxlan_ip_miss(dev, &ipa);
1709                         return false;
1710                 }
1711
1712                 break;
1713         }
1714 #endif
1715         default:
1716                 return false;
1717         }
1718
1719         if (n) {
1720                 bool diff;
1721
1722                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1723                 if (diff) {
1724                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1725                                 dev->addr_len);
1726                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1727                 }
1728                 neigh_release(n);
1729                 return diff;
1730         }
1731
1732         return false;
1733 }
1734
1735 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1736                                 struct vxlan_metadata *md)
1737 {
1738         struct vxlanhdr_gbp *gbp;
1739
1740         if (!md->gbp)
1741                 return;
1742
1743         gbp = (struct vxlanhdr_gbp *)vxh;
1744         vxh->vx_flags |= VXLAN_HF_GBP;
1745
1746         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1747                 gbp->dont_learn = 1;
1748
1749         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1750                 gbp->policy_applied = 1;
1751
1752         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1753 }
1754
1755 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1756                                __be16 protocol)
1757 {
1758         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1759
1760         gpe->np_applied = 1;
1761
1762         switch (protocol) {
1763         case htons(ETH_P_IP):
1764                 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1765                 return 0;
1766         case htons(ETH_P_IPV6):
1767                 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1768                 return 0;
1769         case htons(ETH_P_TEB):
1770                 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1771                 return 0;
1772         }
1773         return -EPFNOSUPPORT;
1774 }
1775
1776 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1777                            int iphdr_len, __be32 vni,
1778                            struct vxlan_metadata *md, u32 vxflags,
1779                            bool udp_sum)
1780 {
1781         struct vxlanhdr *vxh;
1782         int min_headroom;
1783         int err;
1784         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1785         __be16 inner_protocol = htons(ETH_P_TEB);
1786
1787         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1788             skb->ip_summed == CHECKSUM_PARTIAL) {
1789                 int csum_start = skb_checksum_start_offset(skb);
1790
1791                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1792                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1793                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1794                      skb->csum_offset == offsetof(struct tcphdr, check)))
1795                         type |= SKB_GSO_TUNNEL_REMCSUM;
1796         }
1797
1798         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1799                         + VXLAN_HLEN + iphdr_len;
1800
1801         /* Need space for new headers (invalidates iph ptr) */
1802         err = skb_cow_head(skb, min_headroom);
1803         if (unlikely(err))
1804                 return err;
1805
1806         err = iptunnel_handle_offloads(skb, type);
1807         if (err)
1808                 return err;
1809
1810         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1811         vxh->vx_flags = VXLAN_HF_VNI;
1812         vxh->vx_vni = vxlan_vni_field(vni);
1813
1814         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1815                 unsigned int start;
1816
1817                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1818                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1819                 vxh->vx_flags |= VXLAN_HF_RCO;
1820
1821                 if (!skb_is_gso(skb)) {
1822                         skb->ip_summed = CHECKSUM_NONE;
1823                         skb->encapsulation = 0;
1824                 }
1825         }
1826
1827         if (vxflags & VXLAN_F_GBP)
1828                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1829         if (vxflags & VXLAN_F_GPE) {
1830                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1831                 if (err < 0)
1832                         return err;
1833                 inner_protocol = skb->protocol;
1834         }
1835
1836         skb_set_inner_protocol(skb, inner_protocol);
1837         return 0;
1838 }
1839
1840 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1841                                       struct vxlan_sock *sock4,
1842                                       struct sk_buff *skb, int oif, u8 tos,
1843                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
1844                                       struct dst_cache *dst_cache,
1845                                       const struct ip_tunnel_info *info)
1846 {
1847         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1848         struct rtable *rt = NULL;
1849         struct flowi4 fl4;
1850
1851         if (!sock4)
1852                 return ERR_PTR(-EIO);
1853
1854         if (tos && !info)
1855                 use_cache = false;
1856         if (use_cache) {
1857                 rt = dst_cache_get_ip4(dst_cache, saddr);
1858                 if (rt)
1859                         return rt;
1860         }
1861
1862         memset(&fl4, 0, sizeof(fl4));
1863         fl4.flowi4_oif = oif;
1864         fl4.flowi4_tos = RT_TOS(tos);
1865         fl4.flowi4_mark = skb->mark;
1866         fl4.flowi4_proto = IPPROTO_UDP;
1867         fl4.daddr = daddr;
1868         fl4.saddr = *saddr;
1869         fl4.fl4_dport = dport;
1870         fl4.fl4_sport = sport;
1871
1872         rt = ip_route_output_key(vxlan->net, &fl4);
1873         if (likely(!IS_ERR(rt))) {
1874                 if (rt->dst.dev == dev) {
1875                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1876                         ip_rt_put(rt);
1877                         return ERR_PTR(-ELOOP);
1878                 }
1879
1880                 *saddr = fl4.saddr;
1881                 if (use_cache)
1882                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1883         } else {
1884                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1885                 return ERR_PTR(-ENETUNREACH);
1886         }
1887         return rt;
1888 }
1889
1890 #if IS_ENABLED(CONFIG_IPV6)
1891 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1892                                           struct net_device *dev,
1893                                           struct vxlan_sock *sock6,
1894                                           struct sk_buff *skb, int oif, u8 tos,
1895                                           __be32 label,
1896                                           const struct in6_addr *daddr,
1897                                           struct in6_addr *saddr,
1898                                           __be16 dport, __be16 sport,
1899                                           struct dst_cache *dst_cache,
1900                                           const struct ip_tunnel_info *info)
1901 {
1902         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1903         struct dst_entry *ndst;
1904         struct flowi6 fl6;
1905         int err;
1906
1907         if (!sock6)
1908                 return ERR_PTR(-EIO);
1909
1910         if (tos && !info)
1911                 use_cache = false;
1912         if (use_cache) {
1913                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1914                 if (ndst)
1915                         return ndst;
1916         }
1917
1918         memset(&fl6, 0, sizeof(fl6));
1919         fl6.flowi6_oif = oif;
1920         fl6.daddr = *daddr;
1921         fl6.saddr = *saddr;
1922         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1923         fl6.flowi6_mark = skb->mark;
1924         fl6.flowi6_proto = IPPROTO_UDP;
1925         fl6.fl6_dport = dport;
1926         fl6.fl6_sport = sport;
1927
1928         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1929                                          sock6->sock->sk,
1930                                          &ndst, &fl6);
1931         if (unlikely(err < 0)) {
1932                 netdev_dbg(dev, "no route to %pI6\n", daddr);
1933                 return ERR_PTR(-ENETUNREACH);
1934         }
1935
1936         if (unlikely(ndst->dev == dev)) {
1937                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
1938                 dst_release(ndst);
1939                 return ERR_PTR(-ELOOP);
1940         }
1941
1942         *saddr = fl6.saddr;
1943         if (use_cache)
1944                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1945         return ndst;
1946 }
1947 #endif
1948
1949 /* Bypass encapsulation if the destination is local */
1950 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1951                                struct vxlan_dev *dst_vxlan, __be32 vni)
1952 {
1953         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1954         union vxlan_addr loopback;
1955         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1956         struct net_device *dev = skb->dev;
1957         int len = skb->len;
1958
1959         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1960         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1961         skb->pkt_type = PACKET_HOST;
1962         skb->encapsulation = 0;
1963         skb->dev = dst_vxlan->dev;
1964         __skb_pull(skb, skb_network_offset(skb));
1965
1966         if (remote_ip->sa.sa_family == AF_INET) {
1967                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1968                 loopback.sa.sa_family =  AF_INET;
1969 #if IS_ENABLED(CONFIG_IPV6)
1970         } else {
1971                 loopback.sin6.sin6_addr = in6addr_loopback;
1972                 loopback.sa.sa_family =  AF_INET6;
1973 #endif
1974         }
1975
1976         if (dst_vxlan->flags & VXLAN_F_LEARN)
1977                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source, vni);
1978
1979         u64_stats_update_begin(&tx_stats->syncp);
1980         tx_stats->tx_packets++;
1981         tx_stats->tx_bytes += len;
1982         u64_stats_update_end(&tx_stats->syncp);
1983
1984         if (netif_rx(skb) == NET_RX_SUCCESS) {
1985                 u64_stats_update_begin(&rx_stats->syncp);
1986                 rx_stats->rx_packets++;
1987                 rx_stats->rx_bytes += len;
1988                 u64_stats_update_end(&rx_stats->syncp);
1989         } else {
1990                 dev->stats.rx_dropped++;
1991         }
1992 }
1993
1994 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
1995                                  struct vxlan_dev *vxlan, union vxlan_addr *daddr,
1996                                  __be16 dst_port, __be32 vni, struct dst_entry *dst,
1997                                  u32 rt_flags)
1998 {
1999 #if IS_ENABLED(CONFIG_IPV6)
2000         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2001          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2002          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2003          */
2004         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2005 #endif
2006         /* Bypass encapsulation if the destination is local */
2007         if (rt_flags & RTCF_LOCAL &&
2008             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2009                 struct vxlan_dev *dst_vxlan;
2010
2011                 dst_release(dst);
2012                 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2013                                            daddr->sa.sa_family, dst_port,
2014                                            vxlan->flags);
2015                 if (!dst_vxlan) {
2016                         dev->stats.tx_errors++;
2017                         kfree_skb(skb);
2018
2019                         return -ENOENT;
2020                 }
2021                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2022                 return 1;
2023         }
2024
2025         return 0;
2026 }
2027
2028 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2029                            __be32 default_vni, struct vxlan_rdst *rdst,
2030                            bool did_rsc)
2031 {
2032         struct dst_cache *dst_cache;
2033         struct ip_tunnel_info *info;
2034         struct vxlan_dev *vxlan = netdev_priv(dev);
2035         const struct iphdr *old_iph = ip_hdr(skb);
2036         union vxlan_addr *dst;
2037         union vxlan_addr remote_ip, local_ip;
2038         struct vxlan_metadata _md;
2039         struct vxlan_metadata *md = &_md;
2040         __be16 src_port = 0, dst_port;
2041         struct dst_entry *ndst = NULL;
2042         __be32 vni, label;
2043         __u8 tos, ttl;
2044         int err;
2045         u32 flags = vxlan->flags;
2046         bool udp_sum = false;
2047         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2048
2049         info = skb_tunnel_info(skb);
2050
2051         if (rdst) {
2052                 dst = &rdst->remote_ip;
2053                 if (vxlan_addr_any(dst)) {
2054                         if (did_rsc) {
2055                                 /* short-circuited back to local bridge */
2056                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2057                                 return;
2058                         }
2059                         goto drop;
2060                 }
2061
2062                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2063                 vni = (rdst->remote_vni) ? : default_vni;
2064                 local_ip = vxlan->cfg.saddr;
2065                 dst_cache = &rdst->dst_cache;
2066                 md->gbp = skb->mark;
2067                 ttl = vxlan->cfg.ttl;
2068                 if (!ttl && vxlan_addr_multicast(dst))
2069                         ttl = 1;
2070
2071                 tos = vxlan->cfg.tos;
2072                 if (tos == 1)
2073                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2074
2075                 if (dst->sa.sa_family == AF_INET)
2076                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2077                 else
2078                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2079                 label = vxlan->cfg.label;
2080         } else {
2081                 if (!info) {
2082                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2083                                   dev->name);
2084                         goto drop;
2085                 }
2086                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2087                 if (remote_ip.sa.sa_family == AF_INET) {
2088                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2089                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2090                 } else {
2091                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2092                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2093                 }
2094                 dst = &remote_ip;
2095                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2096                 vni = tunnel_id_to_key32(info->key.tun_id);
2097                 dst_cache = &info->dst_cache;
2098                 if (info->options_len)
2099                         md = ip_tunnel_info_opts(info);
2100                 ttl = info->key.ttl;
2101                 tos = info->key.tos;
2102                 label = info->key.label;
2103                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2104         }
2105         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2106                                      vxlan->cfg.port_max, true);
2107
2108         rcu_read_lock();
2109         if (dst->sa.sa_family == AF_INET) {
2110                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2111                 struct rtable *rt;
2112                 __be16 df = 0;
2113
2114                 rt = vxlan_get_route(vxlan, dev, sock4, skb,
2115                                      rdst ? rdst->remote_ifindex : 0, tos,
2116                                      dst->sin.sin_addr.s_addr,
2117                                      &local_ip.sin.sin_addr.s_addr,
2118                                      dst_port, src_port,
2119                                      dst_cache, info);
2120                 if (IS_ERR(rt)) {
2121                         err = PTR_ERR(rt);
2122                         goto tx_error;
2123                 }
2124
2125                 /* Bypass encapsulation if the destination is local */
2126                 if (!info) {
2127                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2128                                                     dst_port, vni, &rt->dst,
2129                                                     rt->rt_flags);
2130                         if (err)
2131                                 goto out_unlock;
2132                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2133                         df = htons(IP_DF);
2134                 }
2135
2136                 ndst = &rt->dst;
2137                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2138                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2139                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2140                                       vni, md, flags, udp_sum);
2141                 if (err < 0)
2142                         goto tx_error;
2143
2144                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2145                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2146                                     src_port, dst_port, xnet, !udp_sum);
2147 #if IS_ENABLED(CONFIG_IPV6)
2148         } else {
2149                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2150
2151                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb,
2152                                         rdst ? rdst->remote_ifindex : 0, tos,
2153                                         label, &dst->sin6.sin6_addr,
2154                                         &local_ip.sin6.sin6_addr,
2155                                         dst_port, src_port,
2156                                         dst_cache, info);
2157                 if (IS_ERR(ndst)) {
2158                         err = PTR_ERR(ndst);
2159                         ndst = NULL;
2160                         goto tx_error;
2161                 }
2162
2163                 if (!info) {
2164                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2165
2166                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2167                                                     dst_port, vni, ndst,
2168                                                     rt6i_flags);
2169                         if (err)
2170                                 goto out_unlock;
2171                 }
2172
2173                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2174                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2175                 skb_scrub_packet(skb, xnet);
2176                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2177                                       vni, md, flags, udp_sum);
2178                 if (err < 0)
2179                         goto tx_error;
2180
2181                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2182                                      &local_ip.sin6.sin6_addr,
2183                                      &dst->sin6.sin6_addr, tos, ttl,
2184                                      label, src_port, dst_port, !udp_sum);
2185 #endif
2186         }
2187 out_unlock:
2188         rcu_read_unlock();
2189         return;
2190
2191 drop:
2192         dev->stats.tx_dropped++;
2193         dev_kfree_skb(skb);
2194         return;
2195
2196 tx_error:
2197         rcu_read_unlock();
2198         if (err == -ELOOP)
2199                 dev->stats.collisions++;
2200         else if (err == -ENETUNREACH)
2201                 dev->stats.tx_carrier_errors++;
2202         dst_release(ndst);
2203         dev->stats.tx_errors++;
2204         kfree_skb(skb);
2205 }
2206
2207 /* Transmit local packets over Vxlan
2208  *
2209  * Outer IP header inherits ECN and DF from inner header.
2210  * Outer UDP destination is the VXLAN assigned port.
2211  *           source port is based on hash of flow
2212  */
2213 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2214 {
2215         struct vxlan_dev *vxlan = netdev_priv(dev);
2216         const struct ip_tunnel_info *info;
2217         struct ethhdr *eth;
2218         bool did_rsc = false;
2219         struct vxlan_rdst *rdst, *fdst = NULL;
2220         struct vxlan_fdb *f;
2221         __be32 vni = 0;
2222
2223         info = skb_tunnel_info(skb);
2224
2225         skb_reset_mac_header(skb);
2226
2227         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2228                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2229                     info->mode & IP_TUNNEL_INFO_TX) {
2230                         vni = tunnel_id_to_key32(info->key.tun_id);
2231                 } else {
2232                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2233                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2234                         else
2235                                 kfree_skb(skb);
2236                         return NETDEV_TX_OK;
2237                 }
2238         }
2239
2240         if (vxlan->flags & VXLAN_F_PROXY) {
2241                 eth = eth_hdr(skb);
2242                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2243                         return arp_reduce(dev, skb, vni);
2244 #if IS_ENABLED(CONFIG_IPV6)
2245                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2246                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2247                                        + sizeof(struct nd_msg)) &&
2248                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2249                                 struct nd_msg *msg;
2250
2251                                 msg = (struct nd_msg *)skb_transport_header(skb);
2252                                 if (msg->icmph.icmp6_code == 0 &&
2253                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2254                                         return neigh_reduce(dev, skb, vni);
2255                 }
2256 #endif
2257         }
2258
2259         eth = eth_hdr(skb);
2260         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2261         did_rsc = false;
2262
2263         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2264             (ntohs(eth->h_proto) == ETH_P_IP ||
2265              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2266                 did_rsc = route_shortcircuit(dev, skb);
2267                 if (did_rsc)
2268                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2269         }
2270
2271         if (f == NULL) {
2272                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2273                 if (f == NULL) {
2274                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2275                             !is_multicast_ether_addr(eth->h_dest))
2276                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2277
2278                         dev->stats.tx_dropped++;
2279                         kfree_skb(skb);
2280                         return NETDEV_TX_OK;
2281                 }
2282         }
2283
2284         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2285                 struct sk_buff *skb1;
2286
2287                 if (!fdst) {
2288                         fdst = rdst;
2289                         continue;
2290                 }
2291                 skb1 = skb_clone(skb, GFP_ATOMIC);
2292                 if (skb1)
2293                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2294         }
2295
2296         if (fdst)
2297                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2298         else
2299                 kfree_skb(skb);
2300         return NETDEV_TX_OK;
2301 }
2302
2303 /* Walk the forwarding table and purge stale entries */
2304 static void vxlan_cleanup(unsigned long arg)
2305 {
2306         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2307         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2308         unsigned int h;
2309
2310         if (!netif_running(vxlan->dev))
2311                 return;
2312
2313         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2314                 struct hlist_node *p, *n;
2315
2316                 spin_lock_bh(&vxlan->hash_lock);
2317                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2318                         struct vxlan_fdb *f
2319                                 = container_of(p, struct vxlan_fdb, hlist);
2320                         unsigned long timeout;
2321
2322                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2323                                 continue;
2324
2325                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2326                         if (time_before_eq(timeout, jiffies)) {
2327                                 netdev_dbg(vxlan->dev,
2328                                            "garbage collect %pM\n",
2329                                            f->eth_addr);
2330                                 f->state = NUD_STALE;
2331                                 vxlan_fdb_destroy(vxlan, f);
2332                         } else if (time_before(timeout, next_timer))
2333                                 next_timer = timeout;
2334                 }
2335                 spin_unlock_bh(&vxlan->hash_lock);
2336         }
2337
2338         mod_timer(&vxlan->age_timer, next_timer);
2339 }
2340
2341 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2342 {
2343         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2344         __be32 vni = vxlan->default_dst.remote_vni;
2345
2346         spin_lock(&vn->sock_lock);
2347         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2348         spin_unlock(&vn->sock_lock);
2349 }
2350
2351 /* Setup stats when device is created */
2352 static int vxlan_init(struct net_device *dev)
2353 {
2354         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2355         if (!dev->tstats)
2356                 return -ENOMEM;
2357
2358         return 0;
2359 }
2360
2361 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2362 {
2363         struct vxlan_fdb *f;
2364
2365         spin_lock_bh(&vxlan->hash_lock);
2366         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2367         if (f)
2368                 vxlan_fdb_destroy(vxlan, f);
2369         spin_unlock_bh(&vxlan->hash_lock);
2370 }
2371
2372 static void vxlan_uninit(struct net_device *dev)
2373 {
2374         struct vxlan_dev *vxlan = netdev_priv(dev);
2375
2376         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2377
2378         free_percpu(dev->tstats);
2379 }
2380
2381 /* Start ageing timer and join group when device is brought up */
2382 static int vxlan_open(struct net_device *dev)
2383 {
2384         struct vxlan_dev *vxlan = netdev_priv(dev);
2385         int ret;
2386
2387         ret = vxlan_sock_add(vxlan);
2388         if (ret < 0)
2389                 return ret;
2390
2391         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2392                 ret = vxlan_igmp_join(vxlan);
2393                 if (ret == -EADDRINUSE)
2394                         ret = 0;
2395                 if (ret) {
2396                         vxlan_sock_release(vxlan);
2397                         return ret;
2398                 }
2399         }
2400
2401         if (vxlan->cfg.age_interval)
2402                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2403
2404         return ret;
2405 }
2406
2407 /* Purge the forwarding table */
2408 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2409 {
2410         unsigned int h;
2411
2412         spin_lock_bh(&vxlan->hash_lock);
2413         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2414                 struct hlist_node *p, *n;
2415                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2416                         struct vxlan_fdb *f
2417                                 = container_of(p, struct vxlan_fdb, hlist);
2418                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2419                                 continue;
2420                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2421                         if (!is_zero_ether_addr(f->eth_addr))
2422                                 vxlan_fdb_destroy(vxlan, f);
2423                 }
2424         }
2425         spin_unlock_bh(&vxlan->hash_lock);
2426 }
2427
2428 /* Cleanup timer and forwarding table on shutdown */
2429 static int vxlan_stop(struct net_device *dev)
2430 {
2431         struct vxlan_dev *vxlan = netdev_priv(dev);
2432         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2433         int ret = 0;
2434
2435         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2436             !vxlan_group_used(vn, vxlan))
2437                 ret = vxlan_igmp_leave(vxlan);
2438
2439         del_timer_sync(&vxlan->age_timer);
2440
2441         vxlan_flush(vxlan, false);
2442         vxlan_sock_release(vxlan);
2443
2444         return ret;
2445 }
2446
2447 /* Stub, nothing needs to be done. */
2448 static void vxlan_set_multicast_list(struct net_device *dev)
2449 {
2450 }
2451
2452 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2453 {
2454         struct vxlan_dev *vxlan = netdev_priv(dev);
2455         struct vxlan_rdst *dst = &vxlan->default_dst;
2456         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2457                                                          dst->remote_ifindex);
2458         bool use_ipv6 = false;
2459
2460         if (dst->remote_ip.sa.sa_family == AF_INET6)
2461                 use_ipv6 = true;
2462
2463         /* This check is different than dev->max_mtu, because it looks at
2464          * the lowerdev->mtu, rather than the static dev->max_mtu
2465          */
2466         if (lowerdev) {
2467                 int max_mtu = lowerdev->mtu -
2468                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2469                 if (new_mtu > max_mtu)
2470                         return -EINVAL;
2471         }
2472
2473         dev->mtu = new_mtu;
2474         return 0;
2475 }
2476
2477 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2478 {
2479         struct vxlan_dev *vxlan = netdev_priv(dev);
2480         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2481         __be16 sport, dport;
2482
2483         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2484                                   vxlan->cfg.port_max, true);
2485         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2486
2487         if (ip_tunnel_info_af(info) == AF_INET) {
2488                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2489                 struct rtable *rt;
2490
2491                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2492                                      info->key.u.ipv4.dst,
2493                                      &info->key.u.ipv4.src, dport, sport,
2494                                      &info->dst_cache, info);
2495                 if (IS_ERR(rt))
2496                         return PTR_ERR(rt);
2497                 ip_rt_put(rt);
2498         } else {
2499 #if IS_ENABLED(CONFIG_IPV6)
2500                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2501                 struct dst_entry *ndst;
2502
2503                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2504                                         info->key.label, &info->key.u.ipv6.dst,
2505                                         &info->key.u.ipv6.src, dport, sport,
2506                                         &info->dst_cache, info);
2507                 if (IS_ERR(ndst))
2508                         return PTR_ERR(ndst);
2509                 dst_release(ndst);
2510 #else /* !CONFIG_IPV6 */
2511                 return -EPFNOSUPPORT;
2512 #endif
2513         }
2514         info->key.tp_src = sport;
2515         info->key.tp_dst = dport;
2516         return 0;
2517 }
2518
2519 static const struct net_device_ops vxlan_netdev_ether_ops = {
2520         .ndo_init               = vxlan_init,
2521         .ndo_uninit             = vxlan_uninit,
2522         .ndo_open               = vxlan_open,
2523         .ndo_stop               = vxlan_stop,
2524         .ndo_start_xmit         = vxlan_xmit,
2525         .ndo_get_stats64        = ip_tunnel_get_stats64,
2526         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2527         .ndo_change_mtu         = vxlan_change_mtu,
2528         .ndo_validate_addr      = eth_validate_addr,
2529         .ndo_set_mac_address    = eth_mac_addr,
2530         .ndo_fdb_add            = vxlan_fdb_add,
2531         .ndo_fdb_del            = vxlan_fdb_delete,
2532         .ndo_fdb_dump           = vxlan_fdb_dump,
2533         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2534 };
2535
2536 static const struct net_device_ops vxlan_netdev_raw_ops = {
2537         .ndo_init               = vxlan_init,
2538         .ndo_uninit             = vxlan_uninit,
2539         .ndo_open               = vxlan_open,
2540         .ndo_stop               = vxlan_stop,
2541         .ndo_start_xmit         = vxlan_xmit,
2542         .ndo_get_stats64        = ip_tunnel_get_stats64,
2543         .ndo_change_mtu         = vxlan_change_mtu,
2544         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2545 };
2546
2547 /* Info for udev, that this is a virtual tunnel endpoint */
2548 static struct device_type vxlan_type = {
2549         .name = "vxlan",
2550 };
2551
2552 /* Calls the ndo_udp_tunnel_add of the caller in order to
2553  * supply the listening VXLAN udp ports. Callers are expected
2554  * to implement the ndo_udp_tunnel_add.
2555  */
2556 static void vxlan_push_rx_ports(struct net_device *dev)
2557 {
2558         struct vxlan_sock *vs;
2559         struct net *net = dev_net(dev);
2560         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2561         unsigned int i;
2562
2563         spin_lock(&vn->sock_lock);
2564         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2565                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2566                         udp_tunnel_push_rx_port(dev, vs->sock,
2567                                                 (vs->flags & VXLAN_F_GPE) ?
2568                                                 UDP_TUNNEL_TYPE_VXLAN_GPE :
2569                                                 UDP_TUNNEL_TYPE_VXLAN);
2570         }
2571         spin_unlock(&vn->sock_lock);
2572 }
2573
2574 /* Initialize the device structure. */
2575 static void vxlan_setup(struct net_device *dev)
2576 {
2577         struct vxlan_dev *vxlan = netdev_priv(dev);
2578         unsigned int h;
2579
2580         eth_hw_addr_random(dev);
2581         ether_setup(dev);
2582
2583         dev->destructor = free_netdev;
2584         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2585
2586         dev->features   |= NETIF_F_LLTX;
2587         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2588         dev->features   |= NETIF_F_RXCSUM;
2589         dev->features   |= NETIF_F_GSO_SOFTWARE;
2590
2591         dev->vlan_features = dev->features;
2592         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2593         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2594         netif_keep_dst(dev);
2595         dev->priv_flags |= IFF_NO_QUEUE;
2596
2597         INIT_LIST_HEAD(&vxlan->next);
2598         spin_lock_init(&vxlan->hash_lock);
2599
2600         init_timer_deferrable(&vxlan->age_timer);
2601         vxlan->age_timer.function = vxlan_cleanup;
2602         vxlan->age_timer.data = (unsigned long) vxlan;
2603
2604         vxlan->cfg.dst_port = htons(vxlan_port);
2605
2606         vxlan->dev = dev;
2607
2608         gro_cells_init(&vxlan->gro_cells, dev);
2609
2610         for (h = 0; h < FDB_HASH_SIZE; ++h)
2611                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2612 }
2613
2614 static void vxlan_ether_setup(struct net_device *dev)
2615 {
2616         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2617         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2618         dev->netdev_ops = &vxlan_netdev_ether_ops;
2619 }
2620
2621 static void vxlan_raw_setup(struct net_device *dev)
2622 {
2623         dev->header_ops = NULL;
2624         dev->type = ARPHRD_NONE;
2625         dev->hard_header_len = 0;
2626         dev->addr_len = 0;
2627         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2628         dev->netdev_ops = &vxlan_netdev_raw_ops;
2629 }
2630
2631 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2632         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2633         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2634         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2635         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2636         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2637         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2638         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2639         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2640         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2641         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2642         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2643         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2644         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2645         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2646         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2647         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2648         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2649         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2650         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2651         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2652         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2653         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2654         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2655         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2656         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2657         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2658         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2659 };
2660
2661 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2662 {
2663         if (tb[IFLA_ADDRESS]) {
2664                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2665                         pr_debug("invalid link address (not ethernet)\n");
2666                         return -EINVAL;
2667                 }
2668
2669                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2670                         pr_debug("invalid all zero ethernet address\n");
2671                         return -EADDRNOTAVAIL;
2672                 }
2673         }
2674
2675         if (!data)
2676                 return -EINVAL;
2677
2678         if (data[IFLA_VXLAN_ID]) {
2679                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2680                 if (id >= VXLAN_N_VID)
2681                         return -ERANGE;
2682         }
2683
2684         if (data[IFLA_VXLAN_PORT_RANGE]) {
2685                 const struct ifla_vxlan_port_range *p
2686                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2687
2688                 if (ntohs(p->high) < ntohs(p->low)) {
2689                         pr_debug("port range %u .. %u not valid\n",
2690                                  ntohs(p->low), ntohs(p->high));
2691                         return -EINVAL;
2692                 }
2693         }
2694
2695         return 0;
2696 }
2697
2698 static void vxlan_get_drvinfo(struct net_device *netdev,
2699                               struct ethtool_drvinfo *drvinfo)
2700 {
2701         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2702         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2703 }
2704
2705 static const struct ethtool_ops vxlan_ethtool_ops = {
2706         .get_drvinfo    = vxlan_get_drvinfo,
2707         .get_link       = ethtool_op_get_link,
2708 };
2709
2710 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2711                                         __be16 port, u32 flags)
2712 {
2713         struct socket *sock;
2714         struct udp_port_cfg udp_conf;
2715         int err;
2716
2717         memset(&udp_conf, 0, sizeof(udp_conf));
2718
2719         if (ipv6) {
2720                 udp_conf.family = AF_INET6;
2721                 udp_conf.use_udp6_rx_checksums =
2722                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2723                 udp_conf.ipv6_v6only = 1;
2724         } else {
2725                 udp_conf.family = AF_INET;
2726         }
2727
2728         udp_conf.local_udp_port = port;
2729
2730         /* Open UDP socket */
2731         err = udp_sock_create(net, &udp_conf, &sock);
2732         if (err < 0)
2733                 return ERR_PTR(err);
2734
2735         return sock;
2736 }
2737
2738 /* Create new listen socket if needed */
2739 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2740                                               __be16 port, u32 flags)
2741 {
2742         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2743         struct vxlan_sock *vs;
2744         struct socket *sock;
2745         unsigned int h;
2746         struct udp_tunnel_sock_cfg tunnel_cfg;
2747
2748         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2749         if (!vs)
2750                 return ERR_PTR(-ENOMEM);
2751
2752         for (h = 0; h < VNI_HASH_SIZE; ++h)
2753                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2754
2755         sock = vxlan_create_sock(net, ipv6, port, flags);
2756         if (IS_ERR(sock)) {
2757                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2758                         PTR_ERR(sock));
2759                 kfree(vs);
2760                 return ERR_CAST(sock);
2761         }
2762
2763         vs->sock = sock;
2764         atomic_set(&vs->refcnt, 1);
2765         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2766
2767         spin_lock(&vn->sock_lock);
2768         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2769         udp_tunnel_notify_add_rx_port(sock,
2770                                       (vs->flags & VXLAN_F_GPE) ?
2771                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2772                                       UDP_TUNNEL_TYPE_VXLAN);
2773         spin_unlock(&vn->sock_lock);
2774
2775         /* Mark socket as an encapsulation socket. */
2776         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2777         tunnel_cfg.sk_user_data = vs;
2778         tunnel_cfg.encap_type = 1;
2779         tunnel_cfg.encap_rcv = vxlan_rcv;
2780         tunnel_cfg.encap_destroy = NULL;
2781         tunnel_cfg.gro_receive = vxlan_gro_receive;
2782         tunnel_cfg.gro_complete = vxlan_gro_complete;
2783
2784         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2785
2786         return vs;
2787 }
2788
2789 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2790 {
2791         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2792         struct vxlan_sock *vs = NULL;
2793
2794         if (!vxlan->cfg.no_share) {
2795                 spin_lock(&vn->sock_lock);
2796                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2797                                      vxlan->cfg.dst_port, vxlan->flags);
2798                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2799                         spin_unlock(&vn->sock_lock);
2800                         return -EBUSY;
2801                 }
2802                 spin_unlock(&vn->sock_lock);
2803         }
2804         if (!vs)
2805                 vs = vxlan_socket_create(vxlan->net, ipv6,
2806                                          vxlan->cfg.dst_port, vxlan->flags);
2807         if (IS_ERR(vs))
2808                 return PTR_ERR(vs);
2809 #if IS_ENABLED(CONFIG_IPV6)
2810         if (ipv6)
2811                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2812         else
2813 #endif
2814                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2815         vxlan_vs_add_dev(vs, vxlan);
2816         return 0;
2817 }
2818
2819 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2820 {
2821         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2822         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2823         int ret = 0;
2824
2825         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2826 #if IS_ENABLED(CONFIG_IPV6)
2827         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2828         if (ipv6 || metadata)
2829                 ret = __vxlan_sock_add(vxlan, true);
2830 #endif
2831         if (!ret && (!ipv6 || metadata))
2832                 ret = __vxlan_sock_add(vxlan, false);
2833         if (ret < 0)
2834                 vxlan_sock_release(vxlan);
2835         return ret;
2836 }
2837
2838 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2839                                struct vxlan_config *conf,
2840                                bool changelink)
2841 {
2842         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2843         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2844         struct vxlan_rdst *dst = &vxlan->default_dst;
2845         unsigned short needed_headroom = ETH_HLEN;
2846         bool use_ipv6 = false;
2847         __be16 default_port = vxlan->cfg.dst_port;
2848         struct net_device *lowerdev = NULL;
2849
2850         if (!changelink) {
2851                 if (conf->flags & VXLAN_F_GPE) {
2852                         /* For now, allow GPE only together with
2853                          * COLLECT_METADATA. This can be relaxed later; in such
2854                          * case, the other side of the PtP link will have to be
2855                          * provided.
2856                          */
2857                         if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2858                             !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2859                                 pr_info("unsupported combination of extensions\n");
2860                                 return -EINVAL;
2861                         }
2862                         vxlan_raw_setup(dev);
2863                 } else {
2864                         vxlan_ether_setup(dev);
2865                 }
2866
2867                 /* MTU range: 68 - 65535 */
2868                 dev->min_mtu = ETH_MIN_MTU;
2869                 dev->max_mtu = ETH_MAX_MTU;
2870                 vxlan->net = src_net;
2871         }
2872
2873         dst->remote_vni = conf->vni;
2874
2875         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2876
2877         /* Unless IPv6 is explicitly requested, assume IPv4 */
2878         if (!dst->remote_ip.sa.sa_family)
2879                 dst->remote_ip.sa.sa_family = AF_INET;
2880
2881         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2882             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2883                 if (!IS_ENABLED(CONFIG_IPV6))
2884                         return -EPFNOSUPPORT;
2885                 use_ipv6 = true;
2886                 vxlan->flags |= VXLAN_F_IPV6;
2887         }
2888
2889         if (conf->label && !use_ipv6) {
2890                 pr_info("label only supported in use with IPv6\n");
2891                 return -EINVAL;
2892         }
2893
2894         if (conf->remote_ifindex &&
2895             conf->remote_ifindex != vxlan->cfg.remote_ifindex) {
2896                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2897                 dst->remote_ifindex = conf->remote_ifindex;
2898
2899                 if (!lowerdev) {
2900                         pr_info("ifindex %d does not exist\n",
2901                                 dst->remote_ifindex);
2902                         return -ENODEV;
2903                 }
2904
2905 #if IS_ENABLED(CONFIG_IPV6)
2906                 if (use_ipv6) {
2907                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2908                         if (idev && idev->cnf.disable_ipv6) {
2909                                 pr_info("IPv6 is disabled via sysctl\n");
2910                                 return -EPERM;
2911                         }
2912                 }
2913 #endif
2914
2915                 if (!conf->mtu)
2916                         dev->mtu = lowerdev->mtu -
2917                                    (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2918
2919                 needed_headroom = lowerdev->hard_header_len;
2920         } else if (!conf->remote_ifindex &&
2921                    vxlan_addr_multicast(&dst->remote_ip)) {
2922                 pr_info("multicast destination requires interface to be specified\n");
2923                 return -EINVAL;
2924         }
2925
2926         if (conf->mtu) {
2927                 int max_mtu = ETH_MAX_MTU;
2928
2929                 if (lowerdev)
2930                         max_mtu = lowerdev->mtu;
2931
2932                 max_mtu -= (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2933
2934                 if (conf->mtu < dev->min_mtu || conf->mtu > dev->max_mtu)
2935                         return -EINVAL;
2936
2937                 dev->mtu = conf->mtu;
2938
2939                 if (conf->mtu > max_mtu)
2940                         dev->mtu = max_mtu;
2941         }
2942
2943         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2944                 needed_headroom += VXLAN6_HEADROOM;
2945         else
2946                 needed_headroom += VXLAN_HEADROOM;
2947         dev->needed_headroom = needed_headroom;
2948
2949         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2950         if (!vxlan->cfg.dst_port) {
2951                 if (conf->flags & VXLAN_F_GPE)
2952                         vxlan->cfg.dst_port = htons(4790); /* IANA VXLAN-GPE port */
2953                 else
2954                         vxlan->cfg.dst_port = default_port;
2955         }
2956         vxlan->flags |= conf->flags;
2957
2958         if (!vxlan->cfg.age_interval)
2959                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2960
2961         if (changelink)
2962                 return 0;
2963
2964         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2965                 if (tmp->cfg.vni == conf->vni &&
2966                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2967                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2968                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2969                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2970                     (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
2971                         pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
2972                         return -EEXIST;
2973                 }
2974         }
2975
2976         return 0;
2977 }
2978
2979 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
2980                               struct vxlan_config *conf)
2981 {
2982         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2983         struct vxlan_dev *vxlan = netdev_priv(dev);
2984         int err;
2985
2986         err = vxlan_dev_configure(net, dev, conf, false);
2987         if (err)
2988                 return err;
2989
2990         dev->ethtool_ops = &vxlan_ethtool_ops;
2991
2992         /* create an fdb entry for a valid default destination */
2993         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2994                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2995                                        &vxlan->default_dst.remote_ip,
2996                                        NUD_REACHABLE | NUD_PERMANENT,
2997                                        NLM_F_EXCL | NLM_F_CREATE,
2998                                        vxlan->cfg.dst_port,
2999                                        vxlan->default_dst.remote_vni,
3000                                        vxlan->default_dst.remote_vni,
3001                                        vxlan->default_dst.remote_ifindex,
3002                                        NTF_SELF);
3003                 if (err)
3004                         return err;
3005         }
3006
3007         err = register_netdevice(dev);
3008         if (err) {
3009                 vxlan_fdb_delete_default(vxlan, vxlan->default_dst.remote_vni);
3010                 return err;
3011         }
3012
3013         list_add(&vxlan->next, &vn->vxlan_list);
3014         return 0;
3015 }
3016
3017 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3018                          struct net_device *dev, struct vxlan_config *conf,
3019                          bool changelink)
3020 {
3021         struct vxlan_dev *vxlan = netdev_priv(dev);
3022
3023         memset(conf, 0, sizeof(*conf));
3024
3025         /* if changelink operation, start with old existing cfg */
3026         if (changelink)
3027                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3028
3029         if (data[IFLA_VXLAN_ID]) {
3030                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3031
3032                 if (changelink && (vni != conf->vni))
3033                         return -EOPNOTSUPP;
3034                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3035         }
3036
3037         if (data[IFLA_VXLAN_GROUP]) {
3038                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3039         } else if (data[IFLA_VXLAN_GROUP6]) {
3040                 if (!IS_ENABLED(CONFIG_IPV6))
3041                         return -EPFNOSUPPORT;
3042
3043                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3044                 conf->remote_ip.sa.sa_family = AF_INET6;
3045         }
3046
3047         if (data[IFLA_VXLAN_LOCAL]) {
3048                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3049                 conf->saddr.sa.sa_family = AF_INET;
3050         } else if (data[IFLA_VXLAN_LOCAL6]) {
3051                 if (!IS_ENABLED(CONFIG_IPV6))
3052                         return -EPFNOSUPPORT;
3053
3054                 /* TODO: respect scope id */
3055                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3056                 conf->saddr.sa.sa_family = AF_INET6;
3057         }
3058
3059         if (data[IFLA_VXLAN_LINK])
3060                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3061
3062         if (data[IFLA_VXLAN_TOS])
3063                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3064
3065         if (data[IFLA_VXLAN_TTL])
3066                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3067
3068         if (data[IFLA_VXLAN_LABEL])
3069                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3070                              IPV6_FLOWLABEL_MASK;
3071
3072         if (data[IFLA_VXLAN_LEARNING]) {
3073                 if (nla_get_u8(data[IFLA_VXLAN_LEARNING])) {
3074                         conf->flags |= VXLAN_F_LEARN;
3075                 } else {
3076                         conf->flags &= ~VXLAN_F_LEARN;
3077                         vxlan->flags &= ~VXLAN_F_LEARN;
3078                 }
3079         } else if (!changelink) {
3080                 /* default to learn on a new device */
3081                 conf->flags |= VXLAN_F_LEARN;
3082         }
3083
3084         if (data[IFLA_VXLAN_AGEING]) {
3085                 if (changelink)
3086                         return -EOPNOTSUPP;
3087                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3088         }
3089
3090         if (data[IFLA_VXLAN_PROXY]) {
3091                 if (changelink)
3092                         return -EOPNOTSUPP;
3093                 if (nla_get_u8(data[IFLA_VXLAN_PROXY]))
3094                         conf->flags |= VXLAN_F_PROXY;
3095         }
3096
3097         if (data[IFLA_VXLAN_RSC]) {
3098                 if (changelink)
3099                         return -EOPNOTSUPP;
3100                 if (nla_get_u8(data[IFLA_VXLAN_RSC]))
3101                         conf->flags |= VXLAN_F_RSC;
3102         }
3103
3104         if (data[IFLA_VXLAN_L2MISS]) {
3105                 if (changelink)
3106                         return -EOPNOTSUPP;
3107                 if (nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3108                         conf->flags |= VXLAN_F_L2MISS;
3109         }
3110
3111         if (data[IFLA_VXLAN_L3MISS]) {
3112                 if (changelink)
3113                         return -EOPNOTSUPP;
3114                 if (nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3115                         conf->flags |= VXLAN_F_L3MISS;
3116         }
3117
3118         if (data[IFLA_VXLAN_LIMIT]) {
3119                 if (changelink)
3120                         return -EOPNOTSUPP;
3121                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3122         }
3123
3124         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3125                 if (changelink)
3126                         return -EOPNOTSUPP;
3127                 if (nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3128                         conf->flags |= VXLAN_F_COLLECT_METADATA;
3129         }
3130
3131         if (data[IFLA_VXLAN_PORT_RANGE]) {
3132                 if (!changelink) {
3133                         const struct ifla_vxlan_port_range *p
3134                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3135                         conf->port_min = ntohs(p->low);
3136                         conf->port_max = ntohs(p->high);
3137                 } else {
3138                         return -EOPNOTSUPP;
3139                 }
3140         }
3141
3142         if (data[IFLA_VXLAN_PORT]) {
3143                 if (changelink)
3144                         return -EOPNOTSUPP;
3145                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3146         }
3147
3148         if (data[IFLA_VXLAN_UDP_CSUM]) {
3149                 if (changelink)
3150                         return -EOPNOTSUPP;
3151                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3152                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3153         }
3154
3155         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3156                 if (changelink)
3157                         return -EOPNOTSUPP;
3158                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3159                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3160         }
3161
3162         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3163                 if (changelink)
3164                         return -EOPNOTSUPP;
3165                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3166                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3167         }
3168
3169         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3170                 if (changelink)
3171                         return -EOPNOTSUPP;
3172                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3173                         conf->flags |= VXLAN_F_REMCSUM_TX;
3174         }
3175
3176         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3177                 if (changelink)
3178                         return -EOPNOTSUPP;
3179                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3180                         conf->flags |= VXLAN_F_REMCSUM_RX;
3181         }
3182
3183         if (data[IFLA_VXLAN_GBP]) {
3184                 if (changelink)
3185                         return -EOPNOTSUPP;
3186                 conf->flags |= VXLAN_F_GBP;
3187         }
3188
3189         if (data[IFLA_VXLAN_GPE]) {
3190                 if (changelink)
3191                         return -EOPNOTSUPP;
3192                 conf->flags |= VXLAN_F_GPE;
3193         }
3194
3195         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3196                 if (changelink)
3197                         return -EOPNOTSUPP;
3198                 conf->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3199         }
3200
3201         if (tb[IFLA_MTU]) {
3202                 if (changelink)
3203                         return -EOPNOTSUPP;
3204                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3205         }
3206
3207         return 0;
3208 }
3209
3210 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3211                          struct nlattr *tb[], struct nlattr *data[])
3212 {
3213         struct vxlan_config conf;
3214         int err;
3215
3216         err = vxlan_nl2conf(tb, data, dev, &conf, false);
3217         if (err)
3218                 return err;
3219
3220         return __vxlan_dev_create(src_net, dev, &conf);
3221 }
3222
3223 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3224                             struct nlattr *data[])
3225 {
3226         struct vxlan_dev *vxlan = netdev_priv(dev);
3227         struct vxlan_rdst *dst = &vxlan->default_dst;
3228         struct vxlan_rdst old_dst;
3229         struct vxlan_config conf;
3230         int err;
3231
3232         err = vxlan_nl2conf(tb, data,
3233                             dev, &conf, true);
3234         if (err)
3235                 return err;
3236
3237         memcpy(&old_dst, dst, sizeof(struct vxlan_rdst));
3238
3239         err = vxlan_dev_configure(vxlan->net, dev, &conf, true);
3240         if (err)
3241                 return err;
3242
3243         /* handle default dst entry */
3244         if (!vxlan_addr_equal(&dst->remote_ip, &old_dst.remote_ip)) {
3245                 spin_lock_bh(&vxlan->hash_lock);
3246                 if (!vxlan_addr_any(&old_dst.remote_ip))
3247                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
3248                                            old_dst.remote_ip,
3249                                            vxlan->cfg.dst_port,
3250                                            old_dst.remote_vni,
3251                                            old_dst.remote_vni,
3252                                            old_dst.remote_ifindex, 0);
3253
3254                 if (!vxlan_addr_any(&dst->remote_ip)) {
3255                         err = vxlan_fdb_create(vxlan, all_zeros_mac,
3256                                                &dst->remote_ip,
3257                                                NUD_REACHABLE | NUD_PERMANENT,
3258                                                NLM_F_CREATE | NLM_F_APPEND,
3259                                                vxlan->cfg.dst_port,
3260                                                dst->remote_vni,
3261                                                dst->remote_vni,
3262                                                dst->remote_ifindex,
3263                                                NTF_SELF);
3264                         if (err) {
3265                                 spin_unlock_bh(&vxlan->hash_lock);
3266                                 return err;
3267                         }
3268                 }
3269                 spin_unlock_bh(&vxlan->hash_lock);
3270         }
3271
3272         return 0;
3273 }
3274
3275 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3276 {
3277         struct vxlan_dev *vxlan = netdev_priv(dev);
3278         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3279
3280         vxlan_flush(vxlan, true);
3281
3282         spin_lock(&vn->sock_lock);
3283         if (!hlist_unhashed(&vxlan->hlist))
3284                 hlist_del_rcu(&vxlan->hlist);
3285         spin_unlock(&vn->sock_lock);
3286
3287         gro_cells_destroy(&vxlan->gro_cells);
3288         list_del(&vxlan->next);
3289         unregister_netdevice_queue(dev, head);
3290 }
3291
3292 static size_t vxlan_get_size(const struct net_device *dev)
3293 {
3294
3295         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3296                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3297                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3298                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3299                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3300                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3301                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3302                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3303                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3304                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3305                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3306                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3307                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3308                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3309                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3310                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3311                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3312                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3313                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3314                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3315                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3316                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3317                 0;
3318 }
3319
3320 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3321 {
3322         const struct vxlan_dev *vxlan = netdev_priv(dev);
3323         const struct vxlan_rdst *dst = &vxlan->default_dst;
3324         struct ifla_vxlan_port_range ports = {
3325                 .low =  htons(vxlan->cfg.port_min),
3326                 .high = htons(vxlan->cfg.port_max),
3327         };
3328
3329         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3330                 goto nla_put_failure;
3331
3332         if (!vxlan_addr_any(&dst->remote_ip)) {
3333                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3334                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3335                                             dst->remote_ip.sin.sin_addr.s_addr))
3336                                 goto nla_put_failure;
3337 #if IS_ENABLED(CONFIG_IPV6)
3338                 } else {
3339                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3340                                              &dst->remote_ip.sin6.sin6_addr))
3341                                 goto nla_put_failure;
3342 #endif
3343                 }
3344         }
3345
3346         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3347                 goto nla_put_failure;
3348
3349         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3350                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3351                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3352                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3353                                 goto nla_put_failure;
3354 #if IS_ENABLED(CONFIG_IPV6)
3355                 } else {
3356                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3357                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3358                                 goto nla_put_failure;
3359 #endif
3360                 }
3361         }
3362
3363         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3364             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3365             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3366             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3367                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3368             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3369                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3370             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3371             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3372                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3373             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3374                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3375             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3376                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3377             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3378             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3379             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3380             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3381                         !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3382             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3383                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3384             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3385                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3386             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3387                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3388             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3389                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3390                 goto nla_put_failure;
3391
3392         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3393                 goto nla_put_failure;
3394
3395         if (vxlan->flags & VXLAN_F_GBP &&
3396             nla_put_flag(skb, IFLA_VXLAN_GBP))
3397                 goto nla_put_failure;
3398
3399         if (vxlan->flags & VXLAN_F_GPE &&
3400             nla_put_flag(skb, IFLA_VXLAN_GPE))
3401                 goto nla_put_failure;
3402
3403         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3404             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3405                 goto nla_put_failure;
3406
3407         return 0;
3408
3409 nla_put_failure:
3410         return -EMSGSIZE;
3411 }
3412
3413 static struct net *vxlan_get_link_net(const struct net_device *dev)
3414 {
3415         struct vxlan_dev *vxlan = netdev_priv(dev);
3416
3417         return vxlan->net;
3418 }
3419
3420 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3421         .kind           = "vxlan",
3422         .maxtype        = IFLA_VXLAN_MAX,
3423         .policy         = vxlan_policy,
3424         .priv_size      = sizeof(struct vxlan_dev),
3425         .setup          = vxlan_setup,
3426         .validate       = vxlan_validate,
3427         .newlink        = vxlan_newlink,
3428         .changelink     = vxlan_changelink,
3429         .dellink        = vxlan_dellink,
3430         .get_size       = vxlan_get_size,
3431         .fill_info      = vxlan_fill_info,
3432         .get_link_net   = vxlan_get_link_net,
3433 };
3434
3435 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3436                                     u8 name_assign_type,
3437                                     struct vxlan_config *conf)
3438 {
3439         struct nlattr *tb[IFLA_MAX + 1];
3440         struct net_device *dev;
3441         int err;
3442
3443         memset(&tb, 0, sizeof(tb));
3444
3445         dev = rtnl_create_link(net, name, name_assign_type,
3446                                &vxlan_link_ops, tb);
3447         if (IS_ERR(dev))
3448                 return dev;
3449
3450         err = __vxlan_dev_create(net, dev, conf);
3451         if (err < 0) {
3452                 free_netdev(dev);
3453                 return ERR_PTR(err);
3454         }
3455
3456         err = rtnl_configure_link(dev, NULL);
3457         if (err < 0) {
3458                 LIST_HEAD(list_kill);
3459
3460                 vxlan_dellink(dev, &list_kill);
3461                 unregister_netdevice_many(&list_kill);
3462                 return ERR_PTR(err);
3463         }
3464
3465         return dev;
3466 }
3467 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3468
3469 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3470                                              struct net_device *dev)
3471 {
3472         struct vxlan_dev *vxlan, *next;
3473         LIST_HEAD(list_kill);
3474
3475         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3476                 struct vxlan_rdst *dst = &vxlan->default_dst;
3477
3478                 /* In case we created vxlan device with carrier
3479                  * and we loose the carrier due to module unload
3480                  * we also need to remove vxlan device. In other
3481                  * cases, it's not necessary and remote_ifindex
3482                  * is 0 here, so no matches.
3483                  */
3484                 if (dst->remote_ifindex == dev->ifindex)
3485                         vxlan_dellink(vxlan->dev, &list_kill);
3486         }
3487
3488         unregister_netdevice_many(&list_kill);
3489 }
3490
3491 static int vxlan_netdevice_event(struct notifier_block *unused,
3492                                  unsigned long event, void *ptr)
3493 {
3494         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3495         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3496
3497         if (event == NETDEV_UNREGISTER)
3498                 vxlan_handle_lowerdev_unregister(vn, dev);
3499         else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3500                 vxlan_push_rx_ports(dev);
3501
3502         return NOTIFY_DONE;
3503 }
3504
3505 static struct notifier_block vxlan_notifier_block __read_mostly = {
3506         .notifier_call = vxlan_netdevice_event,
3507 };
3508
3509 static __net_init int vxlan_init_net(struct net *net)
3510 {
3511         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3512         unsigned int h;
3513
3514         INIT_LIST_HEAD(&vn->vxlan_list);
3515         spin_lock_init(&vn->sock_lock);
3516
3517         for (h = 0; h < PORT_HASH_SIZE; ++h)
3518                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3519
3520         return 0;
3521 }
3522
3523 static void __net_exit vxlan_exit_net(struct net *net)
3524 {
3525         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3526         struct vxlan_dev *vxlan, *next;
3527         struct net_device *dev, *aux;
3528         LIST_HEAD(list);
3529
3530         rtnl_lock();
3531         for_each_netdev_safe(net, dev, aux)
3532                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3533                         unregister_netdevice_queue(dev, &list);
3534
3535         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3536                 /* If vxlan->dev is in the same netns, it has already been added
3537                  * to the list by the previous loop.
3538                  */
3539                 if (!net_eq(dev_net(vxlan->dev), net)) {
3540                         gro_cells_destroy(&vxlan->gro_cells);
3541                         unregister_netdevice_queue(vxlan->dev, &list);
3542                 }
3543         }
3544
3545         unregister_netdevice_many(&list);
3546         rtnl_unlock();
3547 }
3548
3549 static struct pernet_operations vxlan_net_ops = {
3550         .init = vxlan_init_net,
3551         .exit = vxlan_exit_net,
3552         .id   = &vxlan_net_id,
3553         .size = sizeof(struct vxlan_net),
3554 };
3555
3556 static int __init vxlan_init_module(void)
3557 {
3558         int rc;
3559
3560         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3561
3562         rc = register_pernet_subsys(&vxlan_net_ops);
3563         if (rc)
3564                 goto out1;
3565
3566         rc = register_netdevice_notifier(&vxlan_notifier_block);
3567         if (rc)
3568                 goto out2;
3569
3570         rc = rtnl_link_register(&vxlan_link_ops);
3571         if (rc)
3572                 goto out3;
3573
3574         return 0;
3575 out3:
3576         unregister_netdevice_notifier(&vxlan_notifier_block);
3577 out2:
3578         unregister_pernet_subsys(&vxlan_net_ops);
3579 out1:
3580         return rc;
3581 }
3582 late_initcall(vxlan_init_module);
3583
3584 static void __exit vxlan_cleanup_module(void)
3585 {
3586         rtnl_link_unregister(&vxlan_link_ops);
3587         unregister_netdevice_notifier(&vxlan_notifier_block);
3588         unregister_pernet_subsys(&vxlan_net_ops);
3589         /* rcu_barrier() is called by netns */
3590 }
3591 module_exit(vxlan_cleanup_module);
3592
3593 MODULE_LICENSE("GPL");
3594 MODULE_VERSION(VXLAN_VERSION);
3595 MODULE_AUTHOR("Stephen Hemminger <[email protected]>");
3596 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3597 MODULE_ALIAS_RTNL_LINK("vxlan");