drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/types.h>
  15 #include <linux/module.h>
  16 #include <linux/errno.h>
  17 #include <linux/slab.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/rculist.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/in.h>
  22 #include <linux/ip.h>
  23 #include <linux/udp.h>
  24 #include <linux/igmp.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/hash.h>
  29 #include <linux/ethtool.h>
  30 #include <net/arp.h>
  31 #include <net/ndisc.h>
  32 #include <net/ip.h>
  33 #include <net/ip_tunnels.h>
  34 #include <net/icmp.h>
  35 #include <net/udp.h>
  36 #include <net/udp_tunnel.h>
  37 #include <net/rtnetlink.h>
  38 #include <net/route.h>
  39 #include <net/dsfield.h>
  40 #include <net/inet_ecn.h>
  41 #include <net/net_namespace.h>
  42 #include <net/netns/generic.h>
  43 #include <net/vxlan.h>
  44 #include <net/protocol.h>
  45 #include <net/udp_tunnel.h>
  46 #if IS_ENABLED(CONFIG_IPV6)
  47 #include <net/ipv6.h>
  48 #include <net/addrconf.h>
  49 #include <net/ip6_tunnel.h>
  50 #include <net/ip6_checksum.h>
  51 #endif
  52
  53 #define VXLAN_VERSION   "0.1"
  54
  55 #define PORT_HASH_BITS  8
  56 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  57 #define VNI_HASH_BITS   10
  58 #define VNI_HASH_SIZE   (1<<VNI_HASH_BITS)
  59 #define FDB_HASH_BITS   8
  60 #define FDB_HASH_SIZE   (1<<FDB_HASH_BITS)
  61 #define FDB_AGE_DEFAULT 300 /* 5 min */
  62 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  63
  64 #define VXLAN_N_VID     (1u << 24)
  65 #define VXLAN_VID_MASK  (VXLAN_N_VID - 1)
  66 #define VXLAN_HLEN (sizeof(struct udphdr) + sizeof(struct vxlanhdr))
  67
  68 #define VXLAN_FLAGS 0x08000000  /* struct vxlanhdr.vx_flags required value. */
  69
  70 /* VXLAN protocol header */
  71 struct vxlanhdr {
  72         __be32 vx_flags;
  73         __be32 vx_vni;
  74 };
  75
  76 /* UDP port for VXLAN traffic.
  77  * The IANA assigned port is 4789, but the Linux default is 8472
  78  * for compatibility with early adopters.
  79  */
  80 static unsigned short vxlan_port __read_mostly = 8472;
  81 module_param_named(udp_port, vxlan_port, ushort, 0444);
  82 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  83
  84 static bool log_ecn_error = true;
  85 module_param(log_ecn_error, bool, 0644);
  86 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  87
  88 static int vxlan_net_id;
  89
  90 static const u8 all_zeros_mac[ETH_ALEN];
  91
  92 /* per-network namespace private data for this module */
  93 struct vxlan_net {
  94         struct list_head  vxlan_list;
  95         struct hlist_head sock_list[PORT_HASH_SIZE];
  96         spinlock_t        sock_lock;
  97 };
  98
  99 union vxlan_addr {
 100         struct sockaddr_in sin;
 101         struct sockaddr_in6 sin6;
 102         struct sockaddr sa;
 103 };
 104
 105 struct vxlan_rdst {
 106         union vxlan_addr         remote_ip;
 107         __be16                   remote_port;
 108         u32                      remote_vni;
 109         u32                      remote_ifindex;
 110         struct list_head         list;
 111         struct rcu_head          rcu;
 112 };
 113
 114 /* Forwarding table entry */
 115 struct vxlan_fdb {
 116         struct hlist_node hlist;        /* linked list of entries */
 117         struct rcu_head   rcu;
 118         unsigned long     updated;      /* jiffies */
 119         unsigned long     used;
 120         struct list_head  remotes;
 121         u16               state;        /* see ndm_state */
 122         u8                flags;        /* see ndm_flags */
 123         u8                eth_addr[ETH_ALEN];
 124 };
 125
 126 /* Pseudo network device */
 127 struct vxlan_dev {
 128         struct hlist_node hlist;        /* vni hash table */
 129         struct list_head  next;         /* vxlan's per namespace list */
 130         struct vxlan_sock *vn_sock;     /* listening socket */
 131         struct net_device *dev;
 132         struct net        *net;         /* netns for packet i/o */
 133         struct vxlan_rdst default_dst;  /* default destination */
 134         union vxlan_addr  saddr;        /* source address */
 135         __be16            dst_port;
 136         __u16             port_min;     /* source port range */
 137         __u16             port_max;
 138         __u8              tos;          /* TOS override */
 139         __u8              ttl;
 140         u32               flags;        /* VXLAN_F_* in vxlan.h */
 141
 142         struct work_struct sock_work;
 143         struct work_struct igmp_join;
 144         struct work_struct igmp_leave;
 145
 146         unsigned long     age_interval;
 147         struct timer_list age_timer;
 148         spinlock_t        hash_lock;
 149         unsigned int      addrcnt;
 150         unsigned int      addrmax;
 151
 152         struct hlist_head fdb_head[FDB_HASH_SIZE];
 153 };
 154
 155 /* salt for hash table */
 156 static u32 vxlan_salt __read_mostly;
 157 static struct workqueue_struct *vxlan_wq;
 158
 159 static void vxlan_sock_work(struct work_struct *work);
 160
 161 #if IS_ENABLED(CONFIG_IPV6)
 162 static inline
 163 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 164 {
 165        if (a->sa.sa_family != b->sa.sa_family)
 166                return false;
 167        if (a->sa.sa_family == AF_INET6)
 168                return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 169        else
 170                return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 171 }
 172
 173 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 174 {
 175        if (ipa->sa.sa_family == AF_INET6)
 176                return ipv6_addr_any(&ipa->sin6.sin6_addr);
 177        else
 178                return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 179 }
 180
 181 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 182 {
 183        if (ipa->sa.sa_family == AF_INET6)
 184                return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 185        else
 186                return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 187 }
 188
 189 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 190 {
 191        if (nla_len(nla) >= sizeof(struct in6_addr)) {
 192                nla_memcpy(&ip->sin6.sin6_addr, nla, sizeof(struct in6_addr));
 193                ip->sa.sa_family = AF_INET6;
 194                return 0;
 195        } else if (nla_len(nla) >= sizeof(__be32)) {
 196                ip->sin.sin_addr.s_addr = nla_get_be32(nla);
 197                ip->sa.sa_family = AF_INET;
 198                return 0;
 199        } else {
 200                return -EAFNOSUPPORT;
 201        }
 202 }
 203
 204 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 205                              const union vxlan_addr *ip)
 206 {
 207        if (ip->sa.sa_family == AF_INET6)
 208                return nla_put(skb, attr, sizeof(struct in6_addr), &ip->sin6.sin6_addr);
 209        else
 210                return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
 211 }
 212
 213 #else /* !CONFIG_IPV6 */
 214
 215 static inline
 216 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 217 {
 218        return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 219 }
 220
 221 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 222 {
 223        return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 224 }
 225
 226 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 227 {
 228        return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 229 }
 230
 231 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 232 {
 233        if (nla_len(nla) >= sizeof(struct in6_addr)) {
 234                return -EAFNOSUPPORT;
 235        } else if (nla_len(nla) >= sizeof(__be32)) {
 236                ip->sin.sin_addr.s_addr = nla_get_be32(nla);
 237                ip->sa.sa_family = AF_INET;
 238                return 0;
 239        } else {
 240                return -EAFNOSUPPORT;
 241        }
 242 }
 243
 244 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 245                              const union vxlan_addr *ip)
 246 {
 247        return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
 248 }
 249 #endif
 250
 251 /* Virtual Network hash table head */
 252 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
 253 {
 254         return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
 255 }
 256
 257 /* Socket hash table head */
 258 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 259 {
 260         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 261
 262         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 263 }
 264
 265 /* First remote destination for a forwarding entry.
 266  * Guaranteed to be non-NULL because remotes are never deleted.
 267  */
 268 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 269 {
 270         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 271 }
 272
 273 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 274 {
 275         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 276 }
 277
 278 /* Find VXLAN socket based on network namespace, address family and UDP port */
 279 static struct vxlan_sock *vxlan_find_sock(struct net *net,
 280                                           sa_family_t family, __be16 port)
 281 {
 282         struct vxlan_sock *vs;
 283
 284         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 285                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 286                     inet_sk(vs->sock->sk)->sk.sk_family == family)
 287                         return vs;
 288         }
 289         return NULL;
 290 }
 291
 292 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
 293 {
 294         struct vxlan_dev *vxlan;
 295
 296         hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
 297                 if (vxlan->default_dst.remote_vni == id)
 298                         return vxlan;
 299         }
 300
 301         return NULL;
 302 }
 303
 304 /* Look up VNI in a per net namespace table */
 305 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id,
 306                                         sa_family_t family, __be16 port)
 307 {
 308         struct vxlan_sock *vs;
 309
 310         vs = vxlan_find_sock(net, family, port);
 311         if (!vs)
 312                 return NULL;
 313
 314         return vxlan_vs_find_vni(vs, id);
 315 }
 316
 317 /* Fill in neighbour message in skbuff. */
 318 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 319                           const struct vxlan_fdb *fdb,
 320                           u32 portid, u32 seq, int type, unsigned int flags,
 321                           const struct vxlan_rdst *rdst)
 322 {
 323         unsigned long now = jiffies;
 324         struct nda_cacheinfo ci;
 325         struct nlmsghdr *nlh;
 326         struct ndmsg *ndm;
 327         bool send_ip, send_eth;
 328
 329         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 330         if (nlh == NULL)
 331                 return -EMSGSIZE;
 332
 333         ndm = nlmsg_data(nlh);
 334         memset(ndm, 0, sizeof(*ndm));
 335
 336         send_eth = send_ip = true;
 337
 338         if (type == RTM_GETNEIGH) {
 339                 ndm->ndm_family = AF_INET;
 340                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 341                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 342         } else
 343                 ndm->ndm_family = AF_BRIDGE;
 344         ndm->ndm_state = fdb->state;
 345         ndm->ndm_ifindex = vxlan->dev->ifindex;
 346         ndm->ndm_flags = fdb->flags;
 347         ndm->ndm_type = RTN_UNICAST;
 348
 349         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 350                 goto nla_put_failure;
 351
 352         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 353                 goto nla_put_failure;
 354
 355         if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
 356             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 357                 goto nla_put_failure;
 358         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 359             nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
 360                 goto nla_put_failure;
 361         if (rdst->remote_ifindex &&
 362             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 363                 goto nla_put_failure;
 364
 365         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 366         ci.ndm_confirmed = 0;
 367         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 368         ci.ndm_refcnt    = 0;
 369
 370         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 371                 goto nla_put_failure;
 372
 373         return nlmsg_end(skb, nlh);
 374
 375 nla_put_failure:
 376         nlmsg_cancel(skb, nlh);
 377         return -EMSGSIZE;
 378 }
 379
 380 static inline size_t vxlan_nlmsg_size(void)
 381 {
 382         return NLMSG_ALIGN(sizeof(struct ndmsg))
 383                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 384                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 385                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 386                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 387                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 388                 + nla_total_size(sizeof(struct nda_cacheinfo));
 389 }
 390
 391 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 392                              struct vxlan_rdst *rd, int type)
 393 {
 394         struct net *net = dev_net(vxlan->dev);
 395         struct sk_buff *skb;
 396         int err = -ENOBUFS;
 397
 398         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 399         if (skb == NULL)
 400                 goto errout;
 401
 402         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 403         if (err < 0) {
 404                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 405                 WARN_ON(err == -EMSGSIZE);
 406                 kfree_skb(skb);
 407                 goto errout;
 408         }
 409
 410         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 411         return;
 412 errout:
 413         if (err < 0)
 414                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 415 }
 416
 417 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 418 {
 419         struct vxlan_dev *vxlan = netdev_priv(dev);
 420         struct vxlan_fdb f = {
 421                 .state = NUD_STALE,
 422         };
 423         struct vxlan_rdst remote = {
 424                 .remote_ip = *ipa, /* goes to NDA_DST */
 425                 .remote_vni = VXLAN_N_VID,
 426         };
 427
 428         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 429 }
 430
 431 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 432 {
 433         struct vxlan_fdb f = {
 434                 .state = NUD_STALE,
 435         };
 436         struct vxlan_rdst remote = { };
 437
 438         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 439
 440         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 441 }
 442
 443 /* Hash Ethernet address */
 444 static u32 eth_hash(const unsigned char *addr)
 445 {
 446         u64 value = get_unaligned((u64 *)addr);
 447
 448         /* only want 6 bytes */
 449 #ifdef __BIG_ENDIAN
 450         value >>= 16;
 451 #else
 452         value <<= 16;
 453 #endif
 454         return hash_64(value, FDB_HASH_BITS);
 455 }
 456
 457 /* Hash chain to use given mac address */
 458 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 459                                                 const u8 *mac)
 460 {
 461         return &vxlan->fdb_head[eth_hash(mac)];
 462 }
 463
 464 /* Look up Ethernet address in forwarding table */
 465 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 466                                         const u8 *mac)
 467 {
 468         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 469         struct vxlan_fdb *f;
 470
 471         hlist_for_each_entry_rcu(f, head, hlist) {
 472                 if (ether_addr_equal(mac, f->eth_addr))
 473                         return f;
 474         }
 475
 476         return NULL;
 477 }
 478
 479 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 480                                         const u8 *mac)
 481 {
 482         struct vxlan_fdb *f;
 483
 484         f = __vxlan_find_mac(vxlan, mac);
 485         if (f)
 486                 f->used = jiffies;
 487
 488         return f;
 489 }
 490
 491 /* caller should hold vxlan->hash_lock */
 492 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 493                                               union vxlan_addr *ip, __be16 port,
 494                                               __u32 vni, __u32 ifindex)
 495 {
 496         struct vxlan_rdst *rd;
 497
 498         list_for_each_entry(rd, &f->remotes, list) {
 499                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 500                     rd->remote_port == port &&
 501                     rd->remote_vni == vni &&
 502                     rd->remote_ifindex == ifindex)
 503                         return rd;
 504         }
 505
 506         return NULL;
 507 }
 508
 509 /* Replace destination of unicast mac */
 510 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 511                              union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
 512 {
 513         struct vxlan_rdst *rd;
 514
 515         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 516         if (rd)
 517                 return 0;
 518
 519         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 520         if (!rd)
 521                 return 0;
 522         rd->remote_ip = *ip;
 523         rd->remote_port = port;
 524         rd->remote_vni = vni;
 525         rd->remote_ifindex = ifindex;
 526         return 1;
 527 }
 528
 529 /* Add/update destinations for multicast */
 530 static int vxlan_fdb_append(struct vxlan_fdb *f,
 531                             union vxlan_addr *ip, __be16 port, __u32 vni,
 532                             __u32 ifindex, struct vxlan_rdst **rdp)
 533 {
 534         struct vxlan_rdst *rd;
 535
 536         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 537         if (rd)
 538                 return 0;
 539
 540         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 541         if (rd == NULL)
 542                 return -ENOBUFS;
 543         rd->remote_ip = *ip;
 544         rd->remote_port = port;
 545         rd->remote_vni = vni;
 546         rd->remote_ifindex = ifindex;
 547
 548         list_add_tail_rcu(&rd->list, &f->remotes);
 549
 550         *rdp = rd;
 551         return 1;
 552 }
 553
 554 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head, struct sk_buff *skb)
 555 {
 556         struct sk_buff *p, **pp = NULL;
 557         struct vxlanhdr *vh, *vh2;
 558         struct ethhdr *eh, *eh2;
 559         unsigned int hlen, off_vx, off_eth;
 560         const struct packet_offload *ptype;
 561         __be16 type;
 562         int flush = 1;
 563
 564         off_vx = skb_gro_offset(skb);
 565         hlen = off_vx + sizeof(*vh);
 566         vh   = skb_gro_header_fast(skb, off_vx);
 567         if (skb_gro_header_hard(skb, hlen)) {
 568                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 569                 if (unlikely(!vh))
 570                         goto out;
 571         }
 572         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 573         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 574
 575         off_eth = skb_gro_offset(skb);
 576         hlen = off_eth + sizeof(*eh);
 577         eh   = skb_gro_header_fast(skb, off_eth);
 578         if (skb_gro_header_hard(skb, hlen)) {
 579                 eh = skb_gro_header_slow(skb, hlen, off_eth);
 580                 if (unlikely(!eh))
 581                         goto out;
 582         }
 583
 584         flush = 0;
 585
 586         for (p = *head; p; p = p->next) {
 587                 if (!NAPI_GRO_CB(p)->same_flow)
 588                         continue;
 589
 590                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 591                 eh2 = (struct ethhdr   *)(p->data + off_eth);
 592                 if (vh->vx_vni != vh2->vx_vni || compare_ether_header(eh, eh2)) {
 593                         NAPI_GRO_CB(p)->same_flow = 0;
 594                         continue;
 595                 }
 596         }
 597
 598         type = eh->h_proto;
 599
 600         rcu_read_lock();
 601         ptype = gro_find_receive_by_type(type);
 602         if (ptype == NULL) {
 603                 flush = 1;
 604                 goto out_unlock;
 605         }
 606
 607         skb_gro_pull(skb, sizeof(*eh)); /* pull inner eth header */
 608         skb_gro_postpull_rcsum(skb, eh, sizeof(*eh));
 609         pp = ptype->callbacks.gro_receive(head, skb);
 610
 611 out_unlock:
 612         rcu_read_unlock();
 613 out:
 614         NAPI_GRO_CB(skb)->flush |= flush;
 615
 616         return pp;
 617 }
 618
 619 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff)
 620 {
 621         struct ethhdr *eh;
 622         struct packet_offload *ptype;
 623         __be16 type;
 624         int vxlan_len  = sizeof(struct vxlanhdr) + sizeof(struct ethhdr);
 625         int err = -ENOSYS;
 626
 627         udp_tunnel_gro_complete(skb, nhoff);
 628
 629         eh = (struct ethhdr *)(skb->data + nhoff + sizeof(struct vxlanhdr));
 630         type = eh->h_proto;
 631
 632         rcu_read_lock();
 633         ptype = gro_find_complete_by_type(type);
 634         if (ptype != NULL)
 635                 err = ptype->callbacks.gro_complete(skb, nhoff + vxlan_len);
 636
 637         rcu_read_unlock();
 638         return err;
 639 }
 640
 641 /* Notify netdevs that UDP port started listening */
 642 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 643 {
 644         struct net_device *dev;
 645         struct sock *sk = vs->sock->sk;
 646         struct net *net = sock_net(sk);
 647         sa_family_t sa_family = sk->sk_family;
 648         __be16 port = inet_sk(sk)->inet_sport;
 649         int err;
 650
 651         if (sa_family == AF_INET) {
 652                 err = udp_add_offload(&vs->udp_offloads);
 653                 if (err)
 654                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 655         }
 656
 657         rcu_read_lock();
 658         for_each_netdev_rcu(net, dev) {
 659                 if (dev->netdev_ops->ndo_add_vxlan_port)
 660                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 661                                                             port);
 662         }
 663         rcu_read_unlock();
 664 }
 665
 666 /* Notify netdevs that UDP port is no more listening */
 667 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 668 {
 669         struct net_device *dev;
 670         struct sock *sk = vs->sock->sk;
 671         struct net *net = sock_net(sk);
 672         sa_family_t sa_family = sk->sk_family;
 673         __be16 port = inet_sk(sk)->inet_sport;
 674
 675         rcu_read_lock();
 676         for_each_netdev_rcu(net, dev) {
 677                 if (dev->netdev_ops->ndo_del_vxlan_port)
 678                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 679                                                             port);
 680         }
 681         rcu_read_unlock();
 682
 683         if (sa_family == AF_INET)
 684                 udp_del_offload(&vs->udp_offloads);
 685 }
 686
 687 /* Add new entry to forwarding table -- assumes lock held */
 688 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 689                             const u8 *mac, union vxlan_addr *ip,
 690                             __u16 state, __u16 flags,
 691                             __be16 port, __u32 vni, __u32 ifindex,
 692                             __u8 ndm_flags)
 693 {
 694         struct vxlan_rdst *rd = NULL;
 695         struct vxlan_fdb *f;
 696         int notify = 0;
 697
 698         f = __vxlan_find_mac(vxlan, mac);
 699         if (f) {
 700                 if (flags & NLM_F_EXCL) {
 701                         netdev_dbg(vxlan->dev,
 702                                    "lost race to create %pM\n", mac);
 703                         return -EEXIST;
 704                 }
 705                 if (f->state != state) {
 706                         f->state = state;
 707                         f->updated = jiffies;
 708                         notify = 1;
 709                 }
 710                 if (f->flags != ndm_flags) {
 711                         f->flags = ndm_flags;
 712                         f->updated = jiffies;
 713                         notify = 1;
 714                 }
 715                 if ((flags & NLM_F_REPLACE)) {
 716                         /* Only change unicasts */
 717                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 718                              is_zero_ether_addr(f->eth_addr))) {
 719                                 int rc = vxlan_fdb_replace(f, ip, port, vni,
 720                                                            ifindex);
 721
 722                                 if (rc < 0)
 723                                         return rc;
 724                                 notify |= rc;
 725                         } else
 726                                 return -EOPNOTSUPP;
 727                 }
 728                 if ((flags & NLM_F_APPEND) &&
 729                     (is_multicast_ether_addr(f->eth_addr) ||
 730                      is_zero_ether_addr(f->eth_addr))) {
 731                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 732                                                   &rd);
 733
 734                         if (rc < 0)
 735                                 return rc;
 736                         notify |= rc;
 737                 }
 738         } else {
 739                 if (!(flags & NLM_F_CREATE))
 740                         return -ENOENT;
 741
 742                 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
 743                         return -ENOSPC;
 744
 745                 /* Disallow replace to add a multicast entry */
 746                 if ((flags & NLM_F_REPLACE) &&
 747                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 748                         return -EOPNOTSUPP;
 749
 750                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 751                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 752                 if (!f)
 753                         return -ENOMEM;
 754
 755                 notify = 1;
 756                 f->state = state;
 757                 f->flags = ndm_flags;
 758                 f->updated = f->used = jiffies;
 759                 INIT_LIST_HEAD(&f->remotes);
 760                 memcpy(f->eth_addr, mac, ETH_ALEN);
 761
 762                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 763
 764                 ++vxlan->addrcnt;
 765                 hlist_add_head_rcu(&f->hlist,
 766                                    vxlan_fdb_head(vxlan, mac));
 767         }
 768
 769         if (notify) {
 770                 if (rd == NULL)
 771                         rd = first_remote_rtnl(f);
 772                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 773         }
 774
 775         return 0;
 776 }
 777
 778 static void vxlan_fdb_free(struct rcu_head *head)
 779 {
 780         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 781         struct vxlan_rdst *rd, *nd;
 782
 783         list_for_each_entry_safe(rd, nd, &f->remotes, list)
 784                 kfree(rd);
 785         kfree(f);
 786 }
 787
 788 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 789 {
 790         netdev_dbg(vxlan->dev,
 791                     "delete %pM\n", f->eth_addr);
 792
 793         --vxlan->addrcnt;
 794         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 795
 796         hlist_del_rcu(&f->hlist);
 797         call_rcu(&f->rcu, vxlan_fdb_free);
 798 }
 799
 800 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 801                            union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
 802 {
 803         struct net *net = dev_net(vxlan->dev);
 804         int err;
 805
 806         if (tb[NDA_DST]) {
 807                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 808                 if (err)
 809                         return err;
 810         } else {
 811                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 812                 if (remote->sa.sa_family == AF_INET) {
 813                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 814                         ip->sa.sa_family = AF_INET;
 815 #if IS_ENABLED(CONFIG_IPV6)
 816                 } else {
 817                         ip->sin6.sin6_addr = in6addr_any;
 818                         ip->sa.sa_family = AF_INET6;
 819 #endif
 820                 }
 821         }
 822
 823         if (tb[NDA_PORT]) {
 824                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 825                         return -EINVAL;
 826                 *port = nla_get_be16(tb[NDA_PORT]);
 827         } else {
 828                 *port = vxlan->dst_port;
 829         }
 830
 831         if (tb[NDA_VNI]) {
 832                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 833                         return -EINVAL;
 834                 *vni = nla_get_u32(tb[NDA_VNI]);
 835         } else {
 836                 *vni = vxlan->default_dst.remote_vni;
 837         }
 838
 839         if (tb[NDA_IFINDEX]) {
 840                 struct net_device *tdev;
 841
 842                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 843                         return -EINVAL;
 844                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 845                 tdev = __dev_get_by_index(net, *ifindex);
 846                 if (!tdev)
 847                         return -EADDRNOTAVAIL;
 848         } else {
 849                 *ifindex = 0;
 850         }
 851
 852         return 0;
 853 }
 854
 855 /* Add static entry (via netlink) */
 856 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 857                          struct net_device *dev,
 858                          const unsigned char *addr, u16 flags)
 859 {
 860         struct vxlan_dev *vxlan = netdev_priv(dev);
 861         /* struct net *net = dev_net(vxlan->dev); */
 862         union vxlan_addr ip;
 863         __be16 port;
 864         u32 vni, ifindex;
 865         int err;
 866
 867         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 868                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 869                         ndm->ndm_state);
 870                 return -EINVAL;
 871         }
 872
 873         if (tb[NDA_DST] == NULL)
 874                 return -EINVAL;
 875
 876         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 877         if (err)
 878                 return err;
 879
 880         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 881                 return -EAFNOSUPPORT;
 882
 883         spin_lock_bh(&vxlan->hash_lock);
 884         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 885                                port, vni, ifindex, ndm->ndm_flags);
 886         spin_unlock_bh(&vxlan->hash_lock);
 887
 888         return err;
 889 }
 890
 891 /* Delete entry (via netlink) */
 892 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 893                             struct net_device *dev,
 894                             const unsigned char *addr)
 895 {
 896         struct vxlan_dev *vxlan = netdev_priv(dev);
 897         struct vxlan_fdb *f;
 898         struct vxlan_rdst *rd = NULL;
 899         union vxlan_addr ip;
 900         __be16 port;
 901         u32 vni, ifindex;
 902         int err;
 903
 904         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 905         if (err)
 906                 return err;
 907
 908         err = -ENOENT;
 909
 910         spin_lock_bh(&vxlan->hash_lock);
 911         f = vxlan_find_mac(vxlan, addr);
 912         if (!f)
 913                 goto out;
 914
 915         if (!vxlan_addr_any(&ip)) {
 916                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 917                 if (!rd)
 918                         goto out;
 919         }
 920
 921         err = 0;
 922
 923         /* remove a destination if it's not the only one on the list,
 924          * otherwise destroy the fdb entry
 925          */
 926         if (rd && !list_is_singular(&f->remotes)) {
 927                 list_del_rcu(&rd->list);
 928                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 929                 kfree_rcu(rd, rcu);
 930                 goto out;
 931         }
 932
 933         vxlan_fdb_destroy(vxlan, f);
 934
 935 out:
 936         spin_unlock_bh(&vxlan->hash_lock);
 937
 938         return err;
 939 }
 940
 941 /* Dump forwarding table */
 942 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 943                           struct net_device *dev,
 944                           struct net_device *filter_dev, int idx)
 945 {
 946         struct vxlan_dev *vxlan = netdev_priv(dev);
 947         unsigned int h;
 948
 949         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 950                 struct vxlan_fdb *f;
 951                 int err;
 952
 953                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 954                         struct vxlan_rdst *rd;
 955
 956                         if (idx < cb->args[0])
 957                                 goto skip;
 958
 959                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 960                                 err = vxlan_fdb_info(skb, vxlan, f,
 961                                                      NETLINK_CB(cb->skb).portid,
 962                                                      cb->nlh->nlmsg_seq,
 963                                                      RTM_NEWNEIGH,
 964                                                      NLM_F_MULTI, rd);
 965                                 if (err < 0)
 966                                         goto out;
 967                         }
 968 skip:
 969                         ++idx;
 970                 }
 971         }
 972 out:
 973         return idx;
 974 }
 975
 976 /* Watch incoming packets to learn mapping between Ethernet address
 977  * and Tunnel endpoint.
 978  * Return true if packet is bogus and should be droppped.
 979  */
 980 static bool vxlan_snoop(struct net_device *dev,
 981                         union vxlan_addr *src_ip, const u8 *src_mac)
 982 {
 983         struct vxlan_dev *vxlan = netdev_priv(dev);
 984         struct vxlan_fdb *f;
 985
 986         f = vxlan_find_mac(vxlan, src_mac);
 987         if (likely(f)) {
 988                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 989
 990                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 991                         return false;
 992
 993                 /* Don't migrate static entries, drop packets */
 994                 if (f->state & NUD_NOARP)
 995                         return true;
 996
 997                 if (net_ratelimit())
 998                         netdev_info(dev,
 999                                     "%pM migrated from %pIS to %pIS\n",
1000                                     src_mac, &rdst->remote_ip, &src_ip);
1001
1002                 rdst->remote_ip = *src_ip;
1003                 f->updated = jiffies;
1004                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1005         } else {
1006                 /* learned new entry */
1007                 spin_lock(&vxlan->hash_lock);
1008
1009                 /* close off race between vxlan_flush and incoming packets */
1010                 if (netif_running(dev))
1011                         vxlan_fdb_create(vxlan, src_mac, src_ip,
1012                                          NUD_REACHABLE,
1013                                          NLM_F_EXCL|NLM_F_CREATE,
1014                                          vxlan->dst_port,
1015                                          vxlan->default_dst.remote_vni,
1016                                          0, NTF_SELF);
1017                 spin_unlock(&vxlan->hash_lock);
1018         }
1019
1020         return false;
1021 }
1022
1023 /* See if multicast group is already in use by other ID */
1024 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1025 {
1026         struct vxlan_dev *vxlan;
1027
1028         /* The vxlan_sock is only used by dev, leaving group has
1029          * no effect on other vxlan devices.
1030          */
1031         if (atomic_read(&dev->vn_sock->refcnt) == 1)
1032                 return false;
1033
1034         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1035                 if (!netif_running(vxlan->dev) || vxlan == dev)
1036                         continue;
1037
1038                 if (vxlan->vn_sock != dev->vn_sock)
1039                         continue;
1040
1041                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1042                                       &dev->default_dst.remote_ip))
1043                         continue;
1044
1045                 if (vxlan->default_dst.remote_ifindex !=
1046                     dev->default_dst.remote_ifindex)
1047                         continue;
1048
1049                 return true;
1050         }
1051
1052         return false;
1053 }
1054
1055 static void vxlan_sock_hold(struct vxlan_sock *vs)
1056 {
1057         atomic_inc(&vs->refcnt);
1058 }
1059
1060 void vxlan_sock_release(struct vxlan_sock *vs)
1061 {
1062         struct sock *sk = vs->sock->sk;
1063         struct net *net = sock_net(sk);
1064         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1065
1066         if (!atomic_dec_and_test(&vs->refcnt))
1067                 return;
1068
1069         spin_lock(&vn->sock_lock);
1070         hlist_del_rcu(&vs->hlist);
1071         vxlan_notify_del_rx_port(vs);
1072         spin_unlock(&vn->sock_lock);
1073
1074         queue_work(vxlan_wq, &vs->del_work);
1075 }
1076 EXPORT_SYMBOL_GPL(vxlan_sock_release);
1077
1078 /* Callback to update multicast group membership when first VNI on
1079  * multicast asddress is brought up
1080  * Done as workqueue because ip_mc_join_group acquires RTNL.
1081  */
1082 static void vxlan_igmp_join(struct work_struct *work)
1083 {
1084         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_join);
1085         struct vxlan_sock *vs = vxlan->vn_sock;
1086         struct sock *sk = vs->sock->sk;
1087         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1088         int ifindex = vxlan->default_dst.remote_ifindex;
1089
1090         lock_sock(sk);
1091         if (ip->sa.sa_family == AF_INET) {
1092                 struct ip_mreqn mreq = {
1093                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1094                         .imr_ifindex            = ifindex,
1095                 };
1096
1097                 ip_mc_join_group(sk, &mreq);
1098 #if IS_ENABLED(CONFIG_IPV6)
1099         } else {
1100                 ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1101                                              &ip->sin6.sin6_addr);
1102 #endif
1103         }
1104         release_sock(sk);
1105
1106         vxlan_sock_release(vs);
1107         dev_put(vxlan->dev);
1108 }
1109
1110 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1111 static void vxlan_igmp_leave(struct work_struct *work)
1112 {
1113         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_leave);
1114         struct vxlan_sock *vs = vxlan->vn_sock;
1115         struct sock *sk = vs->sock->sk;
1116         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1117         int ifindex = vxlan->default_dst.remote_ifindex;
1118
1119         lock_sock(sk);
1120         if (ip->sa.sa_family == AF_INET) {
1121                 struct ip_mreqn mreq = {
1122                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1123                         .imr_ifindex            = ifindex,
1124                 };
1125
1126                 ip_mc_leave_group(sk, &mreq);
1127 #if IS_ENABLED(CONFIG_IPV6)
1128         } else {
1129                 ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1130                                              &ip->sin6.sin6_addr);
1131 #endif
1132         }
1133
1134         release_sock(sk);
1135
1136         vxlan_sock_release(vs);
1137         dev_put(vxlan->dev);
1138 }
1139
1140 /* Callback from net/ipv4/udp.c to receive packets */
1141 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1142 {
1143         struct vxlan_sock *vs;
1144         struct vxlanhdr *vxh;
1145
1146         /* Need Vxlan and inner Ethernet header to be present */
1147         if (!pskb_may_pull(skb, VXLAN_HLEN))
1148                 goto error;
1149
1150         /* Return packets with reserved bits set */
1151         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1152         if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
1153             (vxh->vx_vni & htonl(0xff))) {
1154                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1155                            ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1156                 goto error;
1157         }
1158
1159         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1160                 goto drop;
1161
1162         vs = rcu_dereference_sk_user_data(sk);
1163         if (!vs)
1164                 goto drop;
1165
1166         vs->rcv(vs, skb, vxh->vx_vni);
1167         return 0;
1168
1169 drop:
1170         /* Consume bad packet */
1171         kfree_skb(skb);
1172         return 0;
1173
1174 error:
1175         /* Return non vxlan pkt */
1176         return 1;
1177 }
1178
1179 static void vxlan_rcv(struct vxlan_sock *vs,
1180                       struct sk_buff *skb, __be32 vx_vni)
1181 {
1182         struct iphdr *oip = NULL;
1183         struct ipv6hdr *oip6 = NULL;
1184         struct vxlan_dev *vxlan;
1185         struct pcpu_sw_netstats *stats;
1186         union vxlan_addr saddr;
1187         __u32 vni;
1188         int err = 0;
1189         union vxlan_addr *remote_ip;
1190
1191         vni = ntohl(vx_vni) >> 8;
1192         /* Is this VNI defined? */
1193         vxlan = vxlan_vs_find_vni(vs, vni);
1194         if (!vxlan)
1195                 goto drop;
1196
1197         remote_ip = &vxlan->default_dst.remote_ip;
1198         skb_reset_mac_header(skb);
1199         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1200         skb->protocol = eth_type_trans(skb, vxlan->dev);
1201         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1202
1203         /* Ignore packet loops (and multicast echo) */
1204         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1205                 goto drop;
1206
1207         /* Re-examine inner Ethernet packet */
1208         if (remote_ip->sa.sa_family == AF_INET) {
1209                 oip = ip_hdr(skb);
1210                 saddr.sin.sin_addr.s_addr = oip->saddr;
1211                 saddr.sa.sa_family = AF_INET;
1212 #if IS_ENABLED(CONFIG_IPV6)
1213         } else {
1214                 oip6 = ipv6_hdr(skb);
1215                 saddr.sin6.sin6_addr = oip6->saddr;
1216                 saddr.sa.sa_family = AF_INET6;
1217 #endif
1218         }
1219
1220         if ((vxlan->flags & VXLAN_F_LEARN) &&
1221             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1222                 goto drop;
1223
1224         skb_reset_network_header(skb);
1225
1226         if (oip6)
1227                 err = IP6_ECN_decapsulate(oip6, skb);
1228         if (oip)
1229                 err = IP_ECN_decapsulate(oip, skb);
1230
1231         if (unlikely(err)) {
1232                 if (log_ecn_error) {
1233                         if (oip6)
1234                                 net_info_ratelimited("non-ECT from %pI6\n",
1235                                                      &oip6->saddr);
1236                         if (oip)
1237                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1238                                                      &oip->saddr, oip->tos);
1239                 }
1240                 if (err > 1) {
1241                         ++vxlan->dev->stats.rx_frame_errors;
1242                         ++vxlan->dev->stats.rx_errors;
1243                         goto drop;
1244                 }
1245         }
1246
1247         stats = this_cpu_ptr(vxlan->dev->tstats);
1248         u64_stats_update_begin(&stats->syncp);
1249         stats->rx_packets++;
1250         stats->rx_bytes += skb->len;
1251         u64_stats_update_end(&stats->syncp);
1252
1253         netif_rx(skb);
1254
1255         return;
1256 drop:
1257         /* Consume bad packet */
1258         kfree_skb(skb);
1259 }
1260
1261 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1262 {
1263         struct vxlan_dev *vxlan = netdev_priv(dev);
1264         struct arphdr *parp;
1265         u8 *arpptr, *sha;
1266         __be32 sip, tip;
1267         struct neighbour *n;
1268
1269         if (dev->flags & IFF_NOARP)
1270                 goto out;
1271
1272         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1273                 dev->stats.tx_dropped++;
1274                 goto out;
1275         }
1276         parp = arp_hdr(skb);
1277
1278         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1279              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1280             parp->ar_pro != htons(ETH_P_IP) ||
1281             parp->ar_op != htons(ARPOP_REQUEST) ||
1282             parp->ar_hln != dev->addr_len ||
1283             parp->ar_pln != 4)
1284                 goto out;
1285         arpptr = (u8 *)parp + sizeof(struct arphdr);
1286         sha = arpptr;
1287         arpptr += dev->addr_len;        /* sha */
1288         memcpy(&sip, arpptr, sizeof(sip));
1289         arpptr += sizeof(sip);
1290         arpptr += dev->addr_len;        /* tha */
1291         memcpy(&tip, arpptr, sizeof(tip));
1292
1293         if (ipv4_is_loopback(tip) ||
1294             ipv4_is_multicast(tip))
1295                 goto out;
1296
1297         n = neigh_lookup(&arp_tbl, &tip, dev);
1298
1299         if (n) {
1300                 struct vxlan_fdb *f;
1301                 struct sk_buff  *reply;
1302
1303                 if (!(n->nud_state & NUD_CONNECTED)) {
1304                         neigh_release(n);
1305                         goto out;
1306                 }
1307
1308                 f = vxlan_find_mac(vxlan, n->ha);
1309                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1310                         /* bridge-local neighbor */
1311                         neigh_release(n);
1312                         goto out;
1313                 }
1314
1315                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1316                                 n->ha, sha);
1317
1318                 neigh_release(n);
1319
1320                 if (reply == NULL)
1321                         goto out;
1322
1323                 skb_reset_mac_header(reply);
1324                 __skb_pull(reply, skb_network_offset(reply));
1325                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1326                 reply->pkt_type = PACKET_HOST;
1327
1328                 if (netif_rx_ni(reply) == NET_RX_DROP)
1329                         dev->stats.rx_dropped++;
1330         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1331                 union vxlan_addr ipa = {
1332                         .sin.sin_addr.s_addr = tip,
1333                         .sin.sin_family = AF_INET,
1334                 };
1335
1336                 vxlan_ip_miss(dev, &ipa);
1337         }
1338 out:
1339         consume_skb(skb);
1340         return NETDEV_TX_OK;
1341 }
1342
1343 #if IS_ENABLED(CONFIG_IPV6)
1344 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1345         struct neighbour *n, bool isrouter)
1346 {
1347         struct net_device *dev = request->dev;
1348         struct sk_buff *reply;
1349         struct nd_msg *ns, *na;
1350         struct ipv6hdr *pip6;
1351         u8 *daddr;
1352         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1353         int ns_olen;
1354         int i, len;
1355
1356         if (dev == NULL)
1357                 return NULL;
1358
1359         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1360                 sizeof(*na) + na_olen + dev->needed_tailroom;
1361         reply = alloc_skb(len, GFP_ATOMIC);
1362         if (reply == NULL)
1363                 return NULL;
1364
1365         reply->protocol = htons(ETH_P_IPV6);
1366         reply->dev = dev;
1367         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1368         skb_push(reply, sizeof(struct ethhdr));
1369         skb_set_mac_header(reply, 0);
1370
1371         ns = (struct nd_msg *)skb_transport_header(request);
1372
1373         daddr = eth_hdr(request)->h_source;
1374         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1375         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1376                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1377                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1378                         break;
1379                 }
1380         }
1381
1382         /* Ethernet header */
1383         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1384         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1385         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1386         reply->protocol = htons(ETH_P_IPV6);
1387
1388         skb_pull(reply, sizeof(struct ethhdr));
1389         skb_set_network_header(reply, 0);
1390         skb_put(reply, sizeof(struct ipv6hdr));
1391
1392         /* IPv6 header */
1393
1394         pip6 = ipv6_hdr(reply);
1395         memset(pip6, 0, sizeof(struct ipv6hdr));
1396         pip6->version = 6;
1397         pip6->priority = ipv6_hdr(request)->priority;
1398         pip6->nexthdr = IPPROTO_ICMPV6;
1399         pip6->hop_limit = 255;
1400         pip6->daddr = ipv6_hdr(request)->saddr;
1401         pip6->saddr = *(struct in6_addr *)n->primary_key;
1402
1403         skb_pull(reply, sizeof(struct ipv6hdr));
1404         skb_set_transport_header(reply, 0);
1405
1406         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1407
1408         /* Neighbor Advertisement */
1409         memset(na, 0, sizeof(*na)+na_olen);
1410         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1411         na->icmph.icmp6_router = isrouter;
1412         na->icmph.icmp6_override = 1;
1413         na->icmph.icmp6_solicited = 1;
1414         na->target = ns->target;
1415         ether_addr_copy(&na->opt[2], n->ha);
1416         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1417         na->opt[1] = na_olen >> 3;
1418
1419         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1420                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1421                 csum_partial(na, sizeof(*na)+na_olen, 0));
1422
1423         pip6->payload_len = htons(sizeof(*na)+na_olen);
1424
1425         skb_push(reply, sizeof(struct ipv6hdr));
1426
1427         reply->ip_summed = CHECKSUM_UNNECESSARY;
1428
1429         return reply;
1430 }
1431
1432 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1433 {
1434         struct vxlan_dev *vxlan = netdev_priv(dev);
1435         struct nd_msg *msg;
1436         const struct ipv6hdr *iphdr;
1437         const struct in6_addr *saddr, *daddr;
1438         struct neighbour *n;
1439         struct inet6_dev *in6_dev;
1440
1441         in6_dev = __in6_dev_get(dev);
1442         if (!in6_dev)
1443                 goto out;
1444
1445         iphdr = ipv6_hdr(skb);
1446         saddr = &iphdr->saddr;
1447         daddr = &iphdr->daddr;
1448
1449         msg = (struct nd_msg *)skb_transport_header(skb);
1450         if (msg->icmph.icmp6_code != 0 ||
1451             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1452                 goto out;
1453
1454         if (ipv6_addr_loopback(daddr) ||
1455             ipv6_addr_is_multicast(&msg->target))
1456                 goto out;
1457
1458         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1459
1460         if (n) {
1461                 struct vxlan_fdb *f;
1462                 struct sk_buff *reply;
1463
1464                 if (!(n->nud_state & NUD_CONNECTED)) {
1465                         neigh_release(n);
1466                         goto out;
1467                 }
1468
1469                 f = vxlan_find_mac(vxlan, n->ha);
1470                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1471                         /* bridge-local neighbor */
1472                         neigh_release(n);
1473                         goto out;
1474                 }
1475
1476                 reply = vxlan_na_create(skb, n,
1477                                         !!(f ? f->flags & NTF_ROUTER : 0));
1478
1479                 neigh_release(n);
1480
1481                 if (reply == NULL)
1482                         goto out;
1483
1484                 if (netif_rx_ni(reply) == NET_RX_DROP)
1485                         dev->stats.rx_dropped++;
1486
1487         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1488                 union vxlan_addr ipa = {
1489                         .sin6.sin6_addr = msg->target,
1490                         .sin6.sin6_family = AF_INET6,
1491                 };
1492
1493                 vxlan_ip_miss(dev, &ipa);
1494         }
1495
1496 out:
1497         consume_skb(skb);
1498         return NETDEV_TX_OK;
1499 }
1500 #endif
1501
1502 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1503 {
1504         struct vxlan_dev *vxlan = netdev_priv(dev);
1505         struct neighbour *n;
1506
1507         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1508                 return false;
1509
1510         n = NULL;
1511         switch (ntohs(eth_hdr(skb)->h_proto)) {
1512         case ETH_P_IP:
1513         {
1514                 struct iphdr *pip;
1515
1516                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1517                         return false;
1518                 pip = ip_hdr(skb);
1519                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1520                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1521                         union vxlan_addr ipa = {
1522                                 .sin.sin_addr.s_addr = pip->daddr,
1523                                 .sin.sin_family = AF_INET,
1524                         };
1525
1526                         vxlan_ip_miss(dev, &ipa);
1527                         return false;
1528                 }
1529
1530                 break;
1531         }
1532 #if IS_ENABLED(CONFIG_IPV6)
1533         case ETH_P_IPV6:
1534         {
1535                 struct ipv6hdr *pip6;
1536
1537                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1538                         return false;
1539                 pip6 = ipv6_hdr(skb);
1540                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1541                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1542                         union vxlan_addr ipa = {
1543                                 .sin6.sin6_addr = pip6->daddr,
1544                                 .sin6.sin6_family = AF_INET6,
1545                         };
1546
1547                         vxlan_ip_miss(dev, &ipa);
1548                         return false;
1549                 }
1550
1551                 break;
1552         }
1553 #endif
1554         default:
1555                 return false;
1556         }
1557
1558         if (n) {
1559                 bool diff;
1560
1561                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1562                 if (diff) {
1563                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1564                                 dev->addr_len);
1565                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1566                 }
1567                 neigh_release(n);
1568                 return diff;
1569         }
1570
1571         return false;
1572 }
1573
1574 #if IS_ENABLED(CONFIG_IPV6)
1575 static int vxlan6_xmit_skb(struct vxlan_sock *vs,
1576                            struct dst_entry *dst, struct sk_buff *skb,
1577                            struct net_device *dev, struct in6_addr *saddr,
1578                            struct in6_addr *daddr, __u8 prio, __u8 ttl,
1579                            __be16 src_port, __be16 dst_port, __be32 vni,
1580                            bool xnet)
1581 {
1582         struct vxlanhdr *vxh;
1583         int min_headroom;
1584         int err;
1585         bool udp_sum = !udp_get_no_check6_tx(vs->sock->sk);
1586
1587         skb = udp_tunnel_handle_offloads(skb, udp_sum);
1588         if (IS_ERR(skb))
1589                 return -EINVAL;
1590
1591         skb_scrub_packet(skb, xnet);
1592
1593         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1594                         + VXLAN_HLEN + sizeof(struct ipv6hdr)
1595                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1596
1597         /* Need space for new headers (invalidates iph ptr) */
1598         err = skb_cow_head(skb, min_headroom);
1599         if (unlikely(err))
1600                 return err;
1601
1602         if (vlan_tx_tag_present(skb)) {
1603                 if (WARN_ON(!__vlan_put_tag(skb,
1604                                             skb->vlan_proto,
1605                                             vlan_tx_tag_get(skb))))
1606                         return -ENOMEM;
1607
1608                 skb->vlan_tci = 0;
1609         }
1610
1611         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1612         vxh->vx_flags = htonl(VXLAN_FLAGS);
1613         vxh->vx_vni = vni;
1614
1615         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1616
1617         udp_tunnel6_xmit_skb(vs->sock, dst, skb, dev, saddr, daddr, prio,
1618                              ttl, src_port, dst_port);
1619         return 0;
1620 }
1621 #endif
1622
1623 int vxlan_xmit_skb(struct vxlan_sock *vs,
1624                    struct rtable *rt, struct sk_buff *skb,
1625                    __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1626                    __be16 src_port, __be16 dst_port, __be32 vni, bool xnet)
1627 {
1628         struct vxlanhdr *vxh;
1629         int min_headroom;
1630         int err;
1631         bool udp_sum = !vs->sock->sk->sk_no_check_tx;
1632
1633         skb = udp_tunnel_handle_offloads(skb, udp_sum);
1634         if (IS_ERR(skb))
1635                 return -EINVAL;
1636
1637         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1638                         + VXLAN_HLEN + sizeof(struct iphdr)
1639                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1640
1641         /* Need space for new headers (invalidates iph ptr) */
1642         err = skb_cow_head(skb, min_headroom);
1643         if (unlikely(err))
1644                 return err;
1645
1646         if (vlan_tx_tag_present(skb)) {
1647                 if (WARN_ON(!__vlan_put_tag(skb,
1648                                             skb->vlan_proto,
1649                                             vlan_tx_tag_get(skb))))
1650                         return -ENOMEM;
1651
1652                 skb->vlan_tci = 0;
1653         }
1654
1655         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1656         vxh->vx_flags = htonl(VXLAN_FLAGS);
1657         vxh->vx_vni = vni;
1658
1659         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1660
1661         return udp_tunnel_xmit_skb(vs->sock, rt, skb, src, dst, tos,
1662                                    ttl, df, src_port, dst_port, xnet);
1663 }
1664 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1665
1666 /* Bypass encapsulation if the destination is local */
1667 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1668                                struct vxlan_dev *dst_vxlan)
1669 {
1670         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1671         union vxlan_addr loopback;
1672         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1673         struct net_device *dev = skb->dev;
1674         int len = skb->len;
1675
1676         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1677         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1678         skb->pkt_type = PACKET_HOST;
1679         skb->encapsulation = 0;
1680         skb->dev = dst_vxlan->dev;
1681         __skb_pull(skb, skb_network_offset(skb));
1682
1683         if (remote_ip->sa.sa_family == AF_INET) {
1684                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1685                 loopback.sa.sa_family =  AF_INET;
1686 #if IS_ENABLED(CONFIG_IPV6)
1687         } else {
1688                 loopback.sin6.sin6_addr = in6addr_loopback;
1689                 loopback.sa.sa_family =  AF_INET6;
1690 #endif
1691         }
1692
1693         if (dst_vxlan->flags & VXLAN_F_LEARN)
1694                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1695
1696         u64_stats_update_begin(&tx_stats->syncp);
1697         tx_stats->tx_packets++;
1698         tx_stats->tx_bytes += len;
1699         u64_stats_update_end(&tx_stats->syncp);
1700
1701         if (netif_rx(skb) == NET_RX_SUCCESS) {
1702                 u64_stats_update_begin(&rx_stats->syncp);
1703                 rx_stats->rx_packets++;
1704                 rx_stats->rx_bytes += len;
1705                 u64_stats_update_end(&rx_stats->syncp);
1706         } else {
1707                 dev->stats.rx_dropped++;
1708         }
1709 }
1710
1711 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1712                            struct vxlan_rdst *rdst, bool did_rsc)
1713 {
1714         struct vxlan_dev *vxlan = netdev_priv(dev);
1715         struct rtable *rt = NULL;
1716         const struct iphdr *old_iph;
1717         struct flowi4 fl4;
1718         union vxlan_addr *dst;
1719         __be16 src_port = 0, dst_port;
1720         u32 vni;
1721         __be16 df = 0;
1722         __u8 tos, ttl;
1723         int err;
1724
1725         dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1726         vni = rdst->remote_vni;
1727         dst = &rdst->remote_ip;
1728
1729         if (vxlan_addr_any(dst)) {
1730                 if (did_rsc) {
1731                         /* short-circuited back to local bridge */
1732                         vxlan_encap_bypass(skb, vxlan, vxlan);
1733                         return;
1734                 }
1735                 goto drop;
1736         }
1737
1738         old_iph = ip_hdr(skb);
1739
1740         ttl = vxlan->ttl;
1741         if (!ttl && vxlan_addr_multicast(dst))
1742                 ttl = 1;
1743
1744         tos = vxlan->tos;
1745         if (tos == 1)
1746                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1747
1748         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->port_min,
1749                                      vxlan->port_max, true);
1750
1751         if (dst->sa.sa_family == AF_INET) {
1752                 memset(&fl4, 0, sizeof(fl4));
1753                 fl4.flowi4_oif = rdst->remote_ifindex;
1754                 fl4.flowi4_tos = RT_TOS(tos);
1755                 fl4.daddr = dst->sin.sin_addr.s_addr;
1756                 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1757
1758                 rt = ip_route_output_key(vxlan->net, &fl4);
1759                 if (IS_ERR(rt)) {
1760                         netdev_dbg(dev, "no route to %pI4\n",
1761                                    &dst->sin.sin_addr.s_addr);
1762                         dev->stats.tx_carrier_errors++;
1763                         goto tx_error;
1764                 }
1765
1766                 if (rt->dst.dev == dev) {
1767                         netdev_dbg(dev, "circular route to %pI4\n",
1768                                    &dst->sin.sin_addr.s_addr);
1769                         dev->stats.collisions++;
1770                         goto rt_tx_error;
1771                 }
1772
1773                 /* Bypass encapsulation if the destination is local */
1774                 if (rt->rt_flags & RTCF_LOCAL &&
1775                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1776                         struct vxlan_dev *dst_vxlan;
1777
1778                         ip_rt_put(rt);
1779                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1780                                                    dst->sa.sa_family, dst_port);
1781                         if (!dst_vxlan)
1782                                 goto tx_error;
1783                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1784                         return;
1785                 }
1786
1787                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1788                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1789
1790                 err = vxlan_xmit_skb(vxlan->vn_sock, rt, skb,
1791                                      fl4.saddr, dst->sin.sin_addr.s_addr,
1792                                      tos, ttl, df, src_port, dst_port,
1793                                      htonl(vni << 8),
1794                                      !net_eq(vxlan->net, dev_net(vxlan->dev)));
1795
1796                 if (err < 0)
1797                         goto rt_tx_error;
1798                 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1799 #if IS_ENABLED(CONFIG_IPV6)
1800         } else {
1801                 struct sock *sk = vxlan->vn_sock->sock->sk;
1802                 struct dst_entry *ndst;
1803                 struct flowi6 fl6;
1804                 u32 flags;
1805
1806                 memset(&fl6, 0, sizeof(fl6));
1807                 fl6.flowi6_oif = rdst->remote_ifindex;
1808                 fl6.daddr = dst->sin6.sin6_addr;
1809                 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
1810                 fl6.flowi6_proto = IPPROTO_UDP;
1811
1812                 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
1813                         netdev_dbg(dev, "no route to %pI6\n",
1814                                    &dst->sin6.sin6_addr);
1815                         dev->stats.tx_carrier_errors++;
1816                         goto tx_error;
1817                 }
1818
1819                 if (ndst->dev == dev) {
1820                         netdev_dbg(dev, "circular route to %pI6\n",
1821                                    &dst->sin6.sin6_addr);
1822                         dst_release(ndst);
1823                         dev->stats.collisions++;
1824                         goto tx_error;
1825                 }
1826
1827                 /* Bypass encapsulation if the destination is local */
1828                 flags = ((struct rt6_info *)ndst)->rt6i_flags;
1829                 if (flags & RTF_LOCAL &&
1830                     !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1831                         struct vxlan_dev *dst_vxlan;
1832
1833                         dst_release(ndst);
1834                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1835                                                    dst->sa.sa_family, dst_port);
1836                         if (!dst_vxlan)
1837                                 goto tx_error;
1838                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1839                         return;
1840                 }
1841
1842                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
1843
1844                 err = vxlan6_xmit_skb(vxlan->vn_sock, ndst, skb,
1845                                       dev, &fl6.saddr, &fl6.daddr, 0, ttl,
1846                                       src_port, dst_port, htonl(vni << 8),
1847                                       !net_eq(vxlan->net, dev_net(vxlan->dev)));
1848 #endif
1849         }
1850
1851         return;
1852
1853 drop:
1854         dev->stats.tx_dropped++;
1855         goto tx_free;
1856
1857 rt_tx_error:
1858         ip_rt_put(rt);
1859 tx_error:
1860         dev->stats.tx_errors++;
1861 tx_free:
1862         dev_kfree_skb(skb);
1863 }
1864
1865 /* Transmit local packets over Vxlan
1866  *
1867  * Outer IP header inherits ECN and DF from inner header.
1868  * Outer UDP destination is the VXLAN assigned port.
1869  *           source port is based on hash of flow
1870  */
1871 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1872 {
1873         struct vxlan_dev *vxlan = netdev_priv(dev);
1874         struct ethhdr *eth;
1875         bool did_rsc = false;
1876         struct vxlan_rdst *rdst, *fdst = NULL;
1877         struct vxlan_fdb *f;
1878
1879         skb_reset_mac_header(skb);
1880         eth = eth_hdr(skb);
1881
1882         if ((vxlan->flags & VXLAN_F_PROXY)) {
1883                 if (ntohs(eth->h_proto) == ETH_P_ARP)
1884                         return arp_reduce(dev, skb);
1885 #if IS_ENABLED(CONFIG_IPV6)
1886                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
1887                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
1888                                        + sizeof(struct nd_msg)) &&
1889                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
1890                                 struct nd_msg *msg;
1891
1892                                 msg = (struct nd_msg *)skb_transport_header(skb);
1893                                 if (msg->icmph.icmp6_code == 0 &&
1894                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
1895                                         return neigh_reduce(dev, skb);
1896                 }
1897                 eth = eth_hdr(skb);
1898 #endif
1899         }
1900
1901         f = vxlan_find_mac(vxlan, eth->h_dest);
1902         did_rsc = false;
1903
1904         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
1905             (ntohs(eth->h_proto) == ETH_P_IP ||
1906              ntohs(eth->h_proto) == ETH_P_IPV6)) {
1907                 did_rsc = route_shortcircuit(dev, skb);
1908                 if (did_rsc)
1909                         f = vxlan_find_mac(vxlan, eth->h_dest);
1910         }
1911
1912         if (f == NULL) {
1913                 f = vxlan_find_mac(vxlan, all_zeros_mac);
1914                 if (f == NULL) {
1915                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
1916                             !is_multicast_ether_addr(eth->h_dest))
1917                                 vxlan_fdb_miss(vxlan, eth->h_dest);
1918
1919                         dev->stats.tx_dropped++;
1920                         kfree_skb(skb);
1921                         return NETDEV_TX_OK;
1922                 }
1923         }
1924
1925         list_for_each_entry_rcu(rdst, &f->remotes, list) {
1926                 struct sk_buff *skb1;
1927
1928                 if (!fdst) {
1929                         fdst = rdst;
1930                         continue;
1931                 }
1932                 skb1 = skb_clone(skb, GFP_ATOMIC);
1933                 if (skb1)
1934                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1935         }
1936
1937         if (fdst)
1938                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
1939         else
1940                 kfree_skb(skb);
1941         return NETDEV_TX_OK;
1942 }
1943
1944 /* Walk the forwarding table and purge stale entries */
1945 static void vxlan_cleanup(unsigned long arg)
1946 {
1947         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1948         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1949         unsigned int h;
1950
1951         if (!netif_running(vxlan->dev))
1952                 return;
1953
1954         spin_lock_bh(&vxlan->hash_lock);
1955         for (h = 0; h < FDB_HASH_SIZE; ++h) {
1956                 struct hlist_node *p, *n;
1957                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1958                         struct vxlan_fdb *f
1959                                 = container_of(p, struct vxlan_fdb, hlist);
1960                         unsigned long timeout;
1961
1962                         if (f->state & NUD_PERMANENT)
1963                                 continue;
1964
1965                         timeout = f->used + vxlan->age_interval * HZ;
1966                         if (time_before_eq(timeout, jiffies)) {
1967                                 netdev_dbg(vxlan->dev,
1968                                            "garbage collect %pM\n",
1969                                            f->eth_addr);
1970                                 f->state = NUD_STALE;
1971                                 vxlan_fdb_destroy(vxlan, f);
1972                         } else if (time_before(timeout, next_timer))
1973                                 next_timer = timeout;
1974                 }
1975         }
1976         spin_unlock_bh(&vxlan->hash_lock);
1977
1978         mod_timer(&vxlan->age_timer, next_timer);
1979 }
1980
1981 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
1982 {
1983         __u32 vni = vxlan->default_dst.remote_vni;
1984
1985         vxlan->vn_sock = vs;
1986         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
1987 }
1988
1989 /* Setup stats when device is created */
1990 static int vxlan_init(struct net_device *dev)
1991 {
1992         struct vxlan_dev *vxlan = netdev_priv(dev);
1993         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
1994         struct vxlan_sock *vs;
1995         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
1996
1997         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1998         if (!dev->tstats)
1999                 return -ENOMEM;
2000
2001         spin_lock(&vn->sock_lock);
2002         vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2003                              vxlan->dst_port);
2004         if (vs) {
2005                 /* If we have a socket with same port already, reuse it */
2006                 atomic_inc(&vs->refcnt);
2007                 vxlan_vs_add_dev(vs, vxlan);
2008         } else {
2009                 /* otherwise make new socket outside of RTNL */
2010                 dev_hold(dev);
2011                 queue_work(vxlan_wq, &vxlan->sock_work);
2012         }
2013         spin_unlock(&vn->sock_lock);
2014
2015         return 0;
2016 }
2017
2018 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2019 {
2020         struct vxlan_fdb *f;
2021
2022         spin_lock_bh(&vxlan->hash_lock);
2023         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2024         if (f)
2025                 vxlan_fdb_destroy(vxlan, f);
2026         spin_unlock_bh(&vxlan->hash_lock);
2027 }
2028
2029 static void vxlan_uninit(struct net_device *dev)
2030 {
2031         struct vxlan_dev *vxlan = netdev_priv(dev);
2032         struct vxlan_sock *vs = vxlan->vn_sock;
2033
2034         vxlan_fdb_delete_default(vxlan);
2035
2036         if (vs)
2037                 vxlan_sock_release(vs);
2038         free_percpu(dev->tstats);
2039 }
2040
2041 /* Start ageing timer and join group when device is brought up */
2042 static int vxlan_open(struct net_device *dev)
2043 {
2044         struct vxlan_dev *vxlan = netdev_priv(dev);
2045         struct vxlan_sock *vs = vxlan->vn_sock;
2046
2047         /* socket hasn't been created */
2048         if (!vs)
2049                 return -ENOTCONN;
2050
2051         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2052                 vxlan_sock_hold(vs);
2053                 dev_hold(dev);
2054                 queue_work(vxlan_wq, &vxlan->igmp_join);
2055         }
2056
2057         if (vxlan->age_interval)
2058                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2059
2060         return 0;
2061 }
2062
2063 /* Purge the forwarding table */
2064 static void vxlan_flush(struct vxlan_dev *vxlan)
2065 {
2066         unsigned int h;
2067
2068         spin_lock_bh(&vxlan->hash_lock);
2069         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2070                 struct hlist_node *p, *n;
2071                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2072                         struct vxlan_fdb *f
2073                                 = container_of(p, struct vxlan_fdb, hlist);
2074                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2075                         if (!is_zero_ether_addr(f->eth_addr))
2076                                 vxlan_fdb_destroy(vxlan, f);
2077                 }
2078         }
2079         spin_unlock_bh(&vxlan->hash_lock);
2080 }
2081
2082 /* Cleanup timer and forwarding table on shutdown */
2083 static int vxlan_stop(struct net_device *dev)
2084 {
2085         struct vxlan_dev *vxlan = netdev_priv(dev);
2086         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2087         struct vxlan_sock *vs = vxlan->vn_sock;
2088
2089         if (vs && vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2090             !vxlan_group_used(vn, vxlan)) {
2091                 vxlan_sock_hold(vs);
2092                 dev_hold(dev);
2093                 queue_work(vxlan_wq, &vxlan->igmp_leave);
2094         }
2095
2096         del_timer_sync(&vxlan->age_timer);
2097
2098         vxlan_flush(vxlan);
2099
2100         return 0;
2101 }
2102
2103 /* Stub, nothing needs to be done. */
2104 static void vxlan_set_multicast_list(struct net_device *dev)
2105 {
2106 }
2107
2108 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2109 {
2110         struct vxlan_dev *vxlan = netdev_priv(dev);
2111         struct vxlan_rdst *dst = &vxlan->default_dst;
2112         struct net_device *lowerdev;
2113         int max_mtu;
2114
2115         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2116         if (lowerdev == NULL)
2117                 return eth_change_mtu(dev, new_mtu);
2118
2119         if (dst->remote_ip.sa.sa_family == AF_INET6)
2120                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2121         else
2122                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2123
2124         if (new_mtu < 68 || new_mtu > max_mtu)
2125                 return -EINVAL;
2126
2127         dev->mtu = new_mtu;
2128         return 0;
2129 }
2130
2131 static const struct net_device_ops vxlan_netdev_ops = {
2132         .ndo_init               = vxlan_init,
2133         .ndo_uninit             = vxlan_uninit,
2134         .ndo_open               = vxlan_open,
2135         .ndo_stop               = vxlan_stop,
2136         .ndo_start_xmit         = vxlan_xmit,
2137         .ndo_get_stats64        = ip_tunnel_get_stats64,
2138         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2139         .ndo_change_mtu         = vxlan_change_mtu,
2140         .ndo_validate_addr      = eth_validate_addr,
2141         .ndo_set_mac_address    = eth_mac_addr,
2142         .ndo_fdb_add            = vxlan_fdb_add,
2143         .ndo_fdb_del            = vxlan_fdb_delete,
2144         .ndo_fdb_dump           = vxlan_fdb_dump,
2145 };
2146
2147 /* Info for udev, that this is a virtual tunnel endpoint */
2148 static struct device_type vxlan_type = {
2149         .name = "vxlan",
2150 };
2151
2152 /* Calls the ndo_add_vxlan_port of the caller in order to
2153  * supply the listening VXLAN udp ports. Callers are expected
2154  * to implement the ndo_add_vxlan_port.
2155  */
2156 void vxlan_get_rx_port(struct net_device *dev)
2157 {
2158         struct vxlan_sock *vs;
2159         struct net *net = dev_net(dev);
2160         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2161         sa_family_t sa_family;
2162         __be16 port;
2163         unsigned int i;
2164
2165         spin_lock(&vn->sock_lock);
2166         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2167                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2168                         port = inet_sk(vs->sock->sk)->inet_sport;
2169                         sa_family = vs->sock->sk->sk_family;
2170                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2171                                                             port);
2172                 }
2173         }
2174         spin_unlock(&vn->sock_lock);
2175 }
2176 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2177
2178 /* Initialize the device structure. */
2179 static void vxlan_setup(struct net_device *dev)
2180 {
2181         struct vxlan_dev *vxlan = netdev_priv(dev);
2182         unsigned int h;
2183
2184         eth_hw_addr_random(dev);
2185         ether_setup(dev);
2186         if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2187                 dev->needed_headroom = ETH_HLEN + VXLAN6_HEADROOM;
2188         else
2189                 dev->needed_headroom = ETH_HLEN + VXLAN_HEADROOM;
2190
2191         dev->netdev_ops = &vxlan_netdev_ops;
2192         dev->destructor = free_netdev;
2193         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2194
2195         dev->tx_queue_len = 0;
2196         dev->features   |= NETIF_F_LLTX;
2197         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2198         dev->features   |= NETIF_F_RXCSUM;
2199         dev->features   |= NETIF_F_GSO_SOFTWARE;
2200
2201         dev->vlan_features = dev->features;
2202         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2203         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2204         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2205         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2206         netif_keep_dst(dev);
2207         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2208
2209         INIT_LIST_HEAD(&vxlan->next);
2210         spin_lock_init(&vxlan->hash_lock);
2211         INIT_WORK(&vxlan->igmp_join, vxlan_igmp_join);
2212         INIT_WORK(&vxlan->igmp_leave, vxlan_igmp_leave);
2213         INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
2214
2215         init_timer_deferrable(&vxlan->age_timer);
2216         vxlan->age_timer.function = vxlan_cleanup;
2217         vxlan->age_timer.data = (unsigned long) vxlan;
2218
2219         vxlan->dst_port = htons(vxlan_port);
2220
2221         vxlan->dev = dev;
2222
2223         for (h = 0; h < FDB_HASH_SIZE; ++h)
2224                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2225 }
2226
2227 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2228         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2229         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2230         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2231         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2232         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2233         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2234         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2235         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2236         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2237         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2238         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2239         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2240         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2241         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2242         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2243         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2244         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2245         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2246         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2247         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2248 };
2249
2250 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2251 {
2252         if (tb[IFLA_ADDRESS]) {
2253                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2254                         pr_debug("invalid link address (not ethernet)\n");
2255                         return -EINVAL;
2256                 }
2257
2258                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2259                         pr_debug("invalid all zero ethernet address\n");
2260                         return -EADDRNOTAVAIL;
2261                 }
2262         }
2263
2264         if (!data)
2265                 return -EINVAL;
2266
2267         if (data[IFLA_VXLAN_ID]) {
2268                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2269                 if (id >= VXLAN_VID_MASK)
2270                         return -ERANGE;
2271         }
2272
2273         if (data[IFLA_VXLAN_PORT_RANGE]) {
2274                 const struct ifla_vxlan_port_range *p
2275                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2276
2277                 if (ntohs(p->high) < ntohs(p->low)) {
2278                         pr_debug("port range %u .. %u not valid\n",
2279                                  ntohs(p->low), ntohs(p->high));
2280                         return -EINVAL;
2281                 }
2282         }
2283
2284         return 0;
2285 }
2286
2287 static void vxlan_get_drvinfo(struct net_device *netdev,
2288                               struct ethtool_drvinfo *drvinfo)
2289 {
2290         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2291         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2292 }
2293
2294 static const struct ethtool_ops vxlan_ethtool_ops = {
2295         .get_drvinfo    = vxlan_get_drvinfo,
2296         .get_link       = ethtool_op_get_link,
2297 };
2298
2299 static void vxlan_del_work(struct work_struct *work)
2300 {
2301         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2302         udp_tunnel_sock_release(vs->sock);
2303         kfree_rcu(vs, rcu);
2304 }
2305
2306 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2307                                         __be16 port, u32 flags)
2308 {
2309         struct socket *sock;
2310         struct udp_port_cfg udp_conf;
2311         int err;
2312
2313         memset(&udp_conf, 0, sizeof(udp_conf));
2314
2315         if (ipv6) {
2316                 udp_conf.family = AF_INET6;
2317                 udp_conf.use_udp6_tx_checksums =
2318                     !!(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2319                 udp_conf.use_udp6_rx_checksums =
2320                     !!(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2321         } else {
2322                 udp_conf.family = AF_INET;
2323                 udp_conf.local_ip.s_addr = INADDR_ANY;
2324                 udp_conf.use_udp_checksums =
2325                     !!(flags & VXLAN_F_UDP_CSUM);
2326         }
2327
2328         udp_conf.local_udp_port = port;
2329
2330         /* Open UDP socket */
2331         err = udp_sock_create(net, &udp_conf, &sock);
2332         if (err < 0)
2333                 return ERR_PTR(err);
2334
2335         return sock;
2336 }
2337
2338 /* Create new listen socket if needed */
2339 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2340                                               vxlan_rcv_t *rcv, void *data,
2341                                               u32 flags)
2342 {
2343         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2344         struct vxlan_sock *vs;
2345         struct socket *sock;
2346         unsigned int h;
2347         bool ipv6 = !!(flags & VXLAN_F_IPV6);
2348         struct udp_tunnel_sock_cfg tunnel_cfg;
2349
2350         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2351         if (!vs)
2352                 return ERR_PTR(-ENOMEM);
2353
2354         for (h = 0; h < VNI_HASH_SIZE; ++h)
2355                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2356
2357         INIT_WORK(&vs->del_work, vxlan_del_work);
2358
2359         sock = vxlan_create_sock(net, ipv6, port, flags);
2360         if (IS_ERR(sock)) {
2361                 kfree(vs);
2362                 return ERR_CAST(sock);
2363         }
2364
2365         vs->sock = sock;
2366         atomic_set(&vs->refcnt, 1);
2367         vs->rcv = rcv;
2368         vs->data = data;
2369
2370         /* Initialize the vxlan udp offloads structure */
2371         vs->udp_offloads.port = port;
2372         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2373         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2374
2375         spin_lock(&vn->sock_lock);
2376         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2377         vxlan_notify_add_rx_port(vs);
2378         spin_unlock(&vn->sock_lock);
2379
2380         /* Mark socket as an encapsulation socket. */
2381         tunnel_cfg.sk_user_data = vs;
2382         tunnel_cfg.encap_type = 1;
2383         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2384         tunnel_cfg.encap_destroy = NULL;
2385
2386         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2387
2388         return vs;
2389 }
2390
2391 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2392                                   vxlan_rcv_t *rcv, void *data,
2393                                   bool no_share, u32 flags)
2394 {
2395         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2396         struct vxlan_sock *vs;
2397         bool ipv6 = flags & VXLAN_F_IPV6;
2398
2399         vs = vxlan_socket_create(net, port, rcv, data, flags);
2400         if (!IS_ERR(vs))
2401                 return vs;
2402
2403         if (no_share)   /* Return error if sharing is not allowed. */
2404                 return vs;
2405
2406         spin_lock(&vn->sock_lock);
2407         vs = vxlan_find_sock(net, ipv6 ? AF_INET6 : AF_INET, port);
2408         if (vs) {
2409                 if (vs->rcv == rcv)
2410                         atomic_inc(&vs->refcnt);
2411                 else
2412                         vs = ERR_PTR(-EBUSY);
2413         }
2414         spin_unlock(&vn->sock_lock);
2415
2416         if (!vs)
2417                 vs = ERR_PTR(-EINVAL);
2418
2419         return vs;
2420 }
2421 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2422
2423 /* Scheduled at device creation to bind to a socket */
2424 static void vxlan_sock_work(struct work_struct *work)
2425 {
2426         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, sock_work);
2427         struct net *net = vxlan->net;
2428         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2429         __be16 port = vxlan->dst_port;
2430         struct vxlan_sock *nvs;
2431
2432         nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags);
2433         spin_lock(&vn->sock_lock);
2434         if (!IS_ERR(nvs))
2435                 vxlan_vs_add_dev(nvs, vxlan);
2436         spin_unlock(&vn->sock_lock);
2437
2438         dev_put(vxlan->dev);
2439 }
2440
2441 static int vxlan_newlink(struct net *net, struct net_device *dev,
2442                          struct nlattr *tb[], struct nlattr *data[])
2443 {
2444         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2445         struct vxlan_dev *vxlan = netdev_priv(dev);
2446         struct vxlan_rdst *dst = &vxlan->default_dst;
2447         __u32 vni;
2448         int err;
2449         bool use_ipv6 = false;
2450
2451         if (!data[IFLA_VXLAN_ID])
2452                 return -EINVAL;
2453
2454         vxlan->net = dev_net(dev);
2455
2456         vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2457         dst->remote_vni = vni;
2458
2459         /* Unless IPv6 is explicitly requested, assume IPv4 */
2460         dst->remote_ip.sa.sa_family = AF_INET;
2461         if (data[IFLA_VXLAN_GROUP]) {
2462                 dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
2463         } else if (data[IFLA_VXLAN_GROUP6]) {
2464                 if (!IS_ENABLED(CONFIG_IPV6))
2465                         return -EPFNOSUPPORT;
2466
2467                 nla_memcpy(&dst->remote_ip.sin6.sin6_addr, data[IFLA_VXLAN_GROUP6],
2468                            sizeof(struct in6_addr));
2469                 dst->remote_ip.sa.sa_family = AF_INET6;
2470                 use_ipv6 = true;
2471         }
2472
2473         if (data[IFLA_VXLAN_LOCAL]) {
2474                 vxlan->saddr.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
2475                 vxlan->saddr.sa.sa_family = AF_INET;
2476         } else if (data[IFLA_VXLAN_LOCAL6]) {
2477                 if (!IS_ENABLED(CONFIG_IPV6))
2478                         return -EPFNOSUPPORT;
2479
2480                 /* TODO: respect scope id */
2481                 nla_memcpy(&vxlan->saddr.sin6.sin6_addr, data[IFLA_VXLAN_LOCAL6],
2482                            sizeof(struct in6_addr));
2483                 vxlan->saddr.sa.sa_family = AF_INET6;
2484                 use_ipv6 = true;
2485         }
2486
2487         if (data[IFLA_VXLAN_LINK] &&
2488             (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2489                 struct net_device *lowerdev
2490                          = __dev_get_by_index(net, dst->remote_ifindex);
2491
2492                 if (!lowerdev) {
2493                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2494                         return -ENODEV;
2495                 }
2496
2497 #if IS_ENABLED(CONFIG_IPV6)
2498                 if (use_ipv6) {
2499                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2500                         if (idev && idev->cnf.disable_ipv6) {
2501                                 pr_info("IPv6 is disabled via sysctl\n");
2502                                 return -EPERM;
2503                         }
2504                         vxlan->flags |= VXLAN_F_IPV6;
2505                 }
2506 #endif
2507
2508                 if (!tb[IFLA_MTU])
2509                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2510
2511                 dev->needed_headroom = lowerdev->hard_header_len +
2512                                        (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2513         } else if (use_ipv6)
2514                 vxlan->flags |= VXLAN_F_IPV6;
2515
2516         if (data[IFLA_VXLAN_TOS])
2517                 vxlan->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2518
2519         if (data[IFLA_VXLAN_TTL])
2520                 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2521
2522         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2523                 vxlan->flags |= VXLAN_F_LEARN;
2524
2525         if (data[IFLA_VXLAN_AGEING])
2526                 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2527         else
2528                 vxlan->age_interval = FDB_AGE_DEFAULT;
2529
2530         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2531                 vxlan->flags |= VXLAN_F_PROXY;
2532
2533         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2534                 vxlan->flags |= VXLAN_F_RSC;
2535
2536         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2537                 vxlan->flags |= VXLAN_F_L2MISS;
2538
2539         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2540                 vxlan->flags |= VXLAN_F_L3MISS;
2541
2542         if (data[IFLA_VXLAN_LIMIT])
2543                 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2544
2545         if (data[IFLA_VXLAN_PORT_RANGE]) {
2546                 const struct ifla_vxlan_port_range *p
2547                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2548                 vxlan->port_min = ntohs(p->low);
2549                 vxlan->port_max = ntohs(p->high);
2550         }
2551
2552         if (data[IFLA_VXLAN_PORT])
2553                 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2554
2555         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2556                 vxlan->flags |= VXLAN_F_UDP_CSUM;
2557
2558         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2559             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2560                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2561
2562         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2563             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2564                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2565
2566         if (vxlan_find_vni(net, vni, use_ipv6 ? AF_INET6 : AF_INET,
2567                            vxlan->dst_port)) {
2568                 pr_info("duplicate VNI %u\n", vni);
2569                 return -EEXIST;
2570         }
2571
2572         dev->ethtool_ops = &vxlan_ethtool_ops;
2573
2574         /* create an fdb entry for a valid default destination */
2575         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2576                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2577                                        &vxlan->default_dst.remote_ip,
2578                                        NUD_REACHABLE|NUD_PERMANENT,
2579                                        NLM_F_EXCL|NLM_F_CREATE,
2580                                        vxlan->dst_port,
2581                                        vxlan->default_dst.remote_vni,
2582                                        vxlan->default_dst.remote_ifindex,
2583                                        NTF_SELF);
2584                 if (err)
2585                         return err;
2586         }
2587
2588         err = register_netdevice(dev);
2589         if (err) {
2590                 vxlan_fdb_delete_default(vxlan);
2591                 return err;
2592         }
2593
2594         list_add(&vxlan->next, &vn->vxlan_list);
2595
2596         return 0;
2597 }
2598
2599 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2600 {
2601         struct vxlan_dev *vxlan = netdev_priv(dev);
2602         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2603
2604         spin_lock(&vn->sock_lock);
2605         if (!hlist_unhashed(&vxlan->hlist))
2606                 hlist_del_rcu(&vxlan->hlist);
2607         spin_unlock(&vn->sock_lock);
2608
2609         list_del(&vxlan->next);
2610         unregister_netdevice_queue(dev, head);
2611 }
2612
2613 static size_t vxlan_get_size(const struct net_device *dev)
2614 {
2615
2616         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2617                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2618                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2619                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2620                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2621                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2622                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2623                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2624                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2625                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2626                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2627                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2628                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2629                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2630                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2631                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2632                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2633                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2634                 0;
2635 }
2636
2637 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2638 {
2639         const struct vxlan_dev *vxlan = netdev_priv(dev);
2640         const struct vxlan_rdst *dst = &vxlan->default_dst;
2641         struct ifla_vxlan_port_range ports = {
2642                 .low =  htons(vxlan->port_min),
2643                 .high = htons(vxlan->port_max),
2644         };
2645
2646         if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2647                 goto nla_put_failure;
2648
2649         if (!vxlan_addr_any(&dst->remote_ip)) {
2650                 if (dst->remote_ip.sa.sa_family == AF_INET) {
2651                         if (nla_put_be32(skb, IFLA_VXLAN_GROUP,
2652                                          dst->remote_ip.sin.sin_addr.s_addr))
2653                                 goto nla_put_failure;
2654 #if IS_ENABLED(CONFIG_IPV6)
2655                 } else {
2656                         if (nla_put(skb, IFLA_VXLAN_GROUP6, sizeof(struct in6_addr),
2657                                     &dst->remote_ip.sin6.sin6_addr))
2658                                 goto nla_put_failure;
2659 #endif
2660                 }
2661         }
2662
2663         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2664                 goto nla_put_failure;
2665
2666         if (!vxlan_addr_any(&vxlan->saddr)) {
2667                 if (vxlan->saddr.sa.sa_family == AF_INET) {
2668                         if (nla_put_be32(skb, IFLA_VXLAN_LOCAL,
2669                                          vxlan->saddr.sin.sin_addr.s_addr))
2670                                 goto nla_put_failure;
2671 #if IS_ENABLED(CONFIG_IPV6)
2672                 } else {
2673                         if (nla_put(skb, IFLA_VXLAN_LOCAL6, sizeof(struct in6_addr),
2674                                     &vxlan->saddr.sin6.sin6_addr))
2675                                 goto nla_put_failure;
2676 #endif
2677                 }
2678         }
2679
2680         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2681             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2682             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2683                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
2684             nla_put_u8(skb, IFLA_VXLAN_PROXY,
2685                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
2686             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2687             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2688                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2689             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2690                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2691             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2692             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2693             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port) ||
2694             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2695                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2696             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2697                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2698             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2699                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)))
2700                 goto nla_put_failure;
2701
2702         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2703                 goto nla_put_failure;
2704
2705         return 0;
2706
2707 nla_put_failure:
2708         return -EMSGSIZE;
2709 }
2710
2711 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2712         .kind           = "vxlan",
2713         .maxtype        = IFLA_VXLAN_MAX,
2714         .policy         = vxlan_policy,
2715         .priv_size      = sizeof(struct vxlan_dev),
2716         .setup          = vxlan_setup,
2717         .validate       = vxlan_validate,
2718         .newlink        = vxlan_newlink,
2719         .dellink        = vxlan_dellink,
2720         .get_size       = vxlan_get_size,
2721         .fill_info      = vxlan_fill_info,
2722 };
2723
2724 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
2725                                              struct net_device *dev)
2726 {
2727         struct vxlan_dev *vxlan, *next;
2728         LIST_HEAD(list_kill);
2729
2730         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2731                 struct vxlan_rdst *dst = &vxlan->default_dst;
2732
2733                 /* In case we created vxlan device with carrier
2734                  * and we loose the carrier due to module unload
2735                  * we also need to remove vxlan device. In other
2736                  * cases, it's not necessary and remote_ifindex
2737                  * is 0 here, so no matches.
2738                  */
2739                 if (dst->remote_ifindex == dev->ifindex)
2740                         vxlan_dellink(vxlan->dev, &list_kill);
2741         }
2742
2743         unregister_netdevice_many(&list_kill);
2744 }
2745
2746 static int vxlan_lowerdev_event(struct notifier_block *unused,
2747                                 unsigned long event, void *ptr)
2748 {
2749         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2750         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2751
2752         if (event == NETDEV_UNREGISTER)
2753                 vxlan_handle_lowerdev_unregister(vn, dev);
2754
2755         return NOTIFY_DONE;
2756 }
2757
2758 static struct notifier_block vxlan_notifier_block __read_mostly = {
2759         .notifier_call = vxlan_lowerdev_event,
2760 };
2761
2762 static __net_init int vxlan_init_net(struct net *net)
2763 {
2764         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2765         unsigned int h;
2766
2767         INIT_LIST_HEAD(&vn->vxlan_list);
2768         spin_lock_init(&vn->sock_lock);
2769
2770         for (h = 0; h < PORT_HASH_SIZE; ++h)
2771                 INIT_HLIST_HEAD(&vn->sock_list[h]);
2772
2773         return 0;
2774 }
2775
2776 static void __net_exit vxlan_exit_net(struct net *net)
2777 {
2778         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2779         struct vxlan_dev *vxlan, *next;
2780         struct net_device *dev, *aux;
2781         LIST_HEAD(list);
2782
2783         rtnl_lock();
2784         for_each_netdev_safe(net, dev, aux)
2785                 if (dev->rtnl_link_ops == &vxlan_link_ops)
2786                         unregister_netdevice_queue(dev, &list);
2787
2788         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2789                 /* If vxlan->dev is in the same netns, it has already been added
2790                  * to the list by the previous loop.
2791                  */
2792                 if (!net_eq(dev_net(vxlan->dev), net))
2793                         unregister_netdevice_queue(dev, &list);
2794         }
2795
2796         unregister_netdevice_many(&list);
2797         rtnl_unlock();
2798 }
2799
2800 static struct pernet_operations vxlan_net_ops = {
2801         .init = vxlan_init_net,
2802         .exit = vxlan_exit_net,
2803         .id   = &vxlan_net_id,
2804         .size = sizeof(struct vxlan_net),
2805 };
2806
2807 static int __init vxlan_init_module(void)
2808 {
2809         int rc;
2810
2811         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
2812         if (!vxlan_wq)
2813                 return -ENOMEM;
2814
2815         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
2816
2817         rc = register_pernet_subsys(&vxlan_net_ops);
2818         if (rc)
2819                 goto out1;
2820
2821         rc = register_netdevice_notifier(&vxlan_notifier_block);
2822         if (rc)
2823                 goto out2;
2824
2825         rc = rtnl_link_register(&vxlan_link_ops);
2826         if (rc)
2827                 goto out3;
2828
2829         return 0;
2830 out3:
2831         unregister_netdevice_notifier(&vxlan_notifier_block);
2832 out2:
2833         unregister_pernet_subsys(&vxlan_net_ops);
2834 out1:
2835         destroy_workqueue(vxlan_wq);
2836         return rc;
2837 }
2838 late_initcall(vxlan_init_module);
2839
2840 static void __exit vxlan_cleanup_module(void)
2841 {
2842         rtnl_link_unregister(&vxlan_link_ops);
2843         unregister_netdevice_notifier(&vxlan_notifier_block);
2844         destroy_workqueue(vxlan_wq);
2845         unregister_pernet_subsys(&vxlan_net_ops);
2846         /* rcu_barrier() is called by netns */
2847 }
2848 module_exit(vxlan_cleanup_module);
2849
2850 MODULE_LICENSE("GPL");
2851 MODULE_VERSION(VXLAN_VERSION);
2852 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
2853 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
2854 MODULE_ALIAS_RTNL_LINK("vxlan");