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 /* UDP port for VXLAN traffic.
  65  * The IANA assigned port is 4789, but the Linux default is 8472
  66  * for compatibility with early adopters.
  67  */
  68 static unsigned short vxlan_port __read_mostly = 8472;
  69 module_param_named(udp_port, vxlan_port, ushort, 0444);
  70 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  71
  72 static bool log_ecn_error = true;
  73 module_param(log_ecn_error, bool, 0644);
  74 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  75
  76 static int vxlan_net_id;
  77
  78 static const u8 all_zeros_mac[ETH_ALEN];
  79
  80 /* per-network namespace private data for this module */
  81 struct vxlan_net {
  82         struct list_head  vxlan_list;
  83         struct hlist_head sock_list[PORT_HASH_SIZE];
  84         spinlock_t        sock_lock;
  85 };
  86
  87 union vxlan_addr {
  88         struct sockaddr_in sin;
  89         struct sockaddr_in6 sin6;
  90         struct sockaddr sa;
  91 };
  92
  93 struct vxlan_rdst {
  94         union vxlan_addr         remote_ip;
  95         __be16                   remote_port;
  96         u32                      remote_vni;
  97         u32                      remote_ifindex;
  98         struct list_head         list;
  99         struct rcu_head          rcu;
 100 };
 101
 102 /* Forwarding table entry */
 103 struct vxlan_fdb {
 104         struct hlist_node hlist;        /* linked list of entries */
 105         struct rcu_head   rcu;
 106         unsigned long     updated;      /* jiffies */
 107         unsigned long     used;
 108         struct list_head  remotes;
 109         u16               state;        /* see ndm_state */
 110         u8                flags;        /* see ndm_flags */
 111         u8                eth_addr[ETH_ALEN];
 112 };
 113
 114 /* Pseudo network device */
 115 struct vxlan_dev {
 116         struct hlist_node hlist;        /* vni hash table */
 117         struct list_head  next;         /* vxlan's per namespace list */
 118         struct vxlan_sock *vn_sock;     /* listening socket */
 119         struct net_device *dev;
 120         struct net        *net;         /* netns for packet i/o */
 121         struct vxlan_rdst default_dst;  /* default destination */
 122         union vxlan_addr  saddr;        /* source address */
 123         __be16            dst_port;
 124         __u16             port_min;     /* source port range */
 125         __u16             port_max;
 126         __u8              tos;          /* TOS override */
 127         __u8              ttl;
 128         u32               flags;        /* VXLAN_F_* in vxlan.h */
 129
 130         struct work_struct sock_work;
 131         struct work_struct igmp_join;
 132         struct work_struct igmp_leave;
 133
 134         unsigned long     age_interval;
 135         struct timer_list age_timer;
 136         spinlock_t        hash_lock;
 137         unsigned int      addrcnt;
 138         unsigned int      addrmax;
 139
 140         struct hlist_head fdb_head[FDB_HASH_SIZE];
 141 };
 142
 143 /* salt for hash table */
 144 static u32 vxlan_salt __read_mostly;
 145 static struct workqueue_struct *vxlan_wq;
 146
 147 static void vxlan_sock_work(struct work_struct *work);
 148
 149 #if IS_ENABLED(CONFIG_IPV6)
 150 static inline
 151 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 152 {
 153        if (a->sa.sa_family != b->sa.sa_family)
 154                return false;
 155        if (a->sa.sa_family == AF_INET6)
 156                return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 157        else
 158                return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 159 }
 160
 161 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 162 {
 163        if (ipa->sa.sa_family == AF_INET6)
 164                return ipv6_addr_any(&ipa->sin6.sin6_addr);
 165        else
 166                return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 167 }
 168
 169 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 170 {
 171        if (ipa->sa.sa_family == AF_INET6)
 172                return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 173        else
 174                return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 175 }
 176
 177 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 178 {
 179        if (nla_len(nla) >= sizeof(struct in6_addr)) {
 180                nla_memcpy(&ip->sin6.sin6_addr, nla, sizeof(struct in6_addr));
 181                ip->sa.sa_family = AF_INET6;
 182                return 0;
 183        } else if (nla_len(nla) >= sizeof(__be32)) {
 184                ip->sin.sin_addr.s_addr = nla_get_be32(nla);
 185                ip->sa.sa_family = AF_INET;
 186                return 0;
 187        } else {
 188                return -EAFNOSUPPORT;
 189        }
 190 }
 191
 192 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 193                              const union vxlan_addr *ip)
 194 {
 195        if (ip->sa.sa_family == AF_INET6)
 196                return nla_put(skb, attr, sizeof(struct in6_addr), &ip->sin6.sin6_addr);
 197        else
 198                return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
 199 }
 200
 201 #else /* !CONFIG_IPV6 */
 202
 203 static inline
 204 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 205 {
 206        return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 207 }
 208
 209 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 210 {
 211        return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 212 }
 213
 214 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 215 {
 216        return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 217 }
 218
 219 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 220 {
 221        if (nla_len(nla) >= sizeof(struct in6_addr)) {
 222                return -EAFNOSUPPORT;
 223        } else if (nla_len(nla) >= sizeof(__be32)) {
 224                ip->sin.sin_addr.s_addr = nla_get_be32(nla);
 225                ip->sa.sa_family = AF_INET;
 226                return 0;
 227        } else {
 228                return -EAFNOSUPPORT;
 229        }
 230 }
 231
 232 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 233                              const union vxlan_addr *ip)
 234 {
 235        return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
 236 }
 237 #endif
 238
 239 /* Virtual Network hash table head */
 240 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
 241 {
 242         return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
 243 }
 244
 245 /* Socket hash table head */
 246 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 247 {
 248         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 249
 250         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 251 }
 252
 253 /* First remote destination for a forwarding entry.
 254  * Guaranteed to be non-NULL because remotes are never deleted.
 255  */
 256 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 257 {
 258         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 259 }
 260
 261 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 262 {
 263         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 264 }
 265
 266 /* Find VXLAN socket based on network namespace, address family and UDP port
 267  * and enabled unshareable flags.
 268  */
 269 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 270                                           __be16 port, u32 flags)
 271 {
 272         struct vxlan_sock *vs;
 273         u32 match_flags = flags & VXLAN_F_UNSHAREABLE;
 274
 275         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 276                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 277                     inet_sk(vs->sock->sk)->sk.sk_family == family &&
 278                     (vs->flags & VXLAN_F_UNSHAREABLE) == match_flags)
 279                         return vs;
 280         }
 281         return NULL;
 282 }
 283
 284 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
 285 {
 286         struct vxlan_dev *vxlan;
 287
 288         hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
 289                 if (vxlan->default_dst.remote_vni == id)
 290                         return vxlan;
 291         }
 292
 293         return NULL;
 294 }
 295
 296 /* Look up VNI in a per net namespace table */
 297 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id,
 298                                         sa_family_t family, __be16 port,
 299                                         u32 flags)
 300 {
 301         struct vxlan_sock *vs;
 302
 303         vs = vxlan_find_sock(net, family, port, flags);
 304         if (!vs)
 305                 return NULL;
 306
 307         return vxlan_vs_find_vni(vs, id);
 308 }
 309
 310 /* Fill in neighbour message in skbuff. */
 311 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 312                           const struct vxlan_fdb *fdb,
 313                           u32 portid, u32 seq, int type, unsigned int flags,
 314                           const struct vxlan_rdst *rdst)
 315 {
 316         unsigned long now = jiffies;
 317         struct nda_cacheinfo ci;
 318         struct nlmsghdr *nlh;
 319         struct ndmsg *ndm;
 320         bool send_ip, send_eth;
 321
 322         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 323         if (nlh == NULL)
 324                 return -EMSGSIZE;
 325
 326         ndm = nlmsg_data(nlh);
 327         memset(ndm, 0, sizeof(*ndm));
 328
 329         send_eth = send_ip = true;
 330
 331         if (type == RTM_GETNEIGH) {
 332                 ndm->ndm_family = AF_INET;
 333                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 334                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 335         } else
 336                 ndm->ndm_family = AF_BRIDGE;
 337         ndm->ndm_state = fdb->state;
 338         ndm->ndm_ifindex = vxlan->dev->ifindex;
 339         ndm->ndm_flags = fdb->flags;
 340         ndm->ndm_type = RTN_UNICAST;
 341
 342         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 343             nla_put_s32(skb, NDA_NDM_IFINDEX_NETNSID,
 344                         peernet2id(vxlan->net, dev_net(vxlan->dev))))
 345                 goto nla_put_failure;
 346
 347         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 348                 goto nla_put_failure;
 349
 350         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 351                 goto nla_put_failure;
 352
 353         if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
 354             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 355                 goto nla_put_failure;
 356         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 357             nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
 358                 goto nla_put_failure;
 359         if (rdst->remote_ifindex &&
 360             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 361                 goto nla_put_failure;
 362
 363         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 364         ci.ndm_confirmed = 0;
 365         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 366         ci.ndm_refcnt    = 0;
 367
 368         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 369                 goto nla_put_failure;
 370
 371         nlmsg_end(skb, nlh);
 372         return 0;
 373
 374 nla_put_failure:
 375         nlmsg_cancel(skb, nlh);
 376         return -EMSGSIZE;
 377 }
 378
 379 static inline size_t vxlan_nlmsg_size(void)
 380 {
 381         return NLMSG_ALIGN(sizeof(struct ndmsg))
 382                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 383                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 384                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 385                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 386                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 387                 + nla_total_size(sizeof(struct nda_cacheinfo));
 388 }
 389
 390 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 391                              struct vxlan_rdst *rd, int type)
 392 {
 393         struct net *net = dev_net(vxlan->dev);
 394         struct sk_buff *skb;
 395         int err = -ENOBUFS;
 396
 397         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 398         if (skb == NULL)
 399                 goto errout;
 400
 401         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 402         if (err < 0) {
 403                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 404                 WARN_ON(err == -EMSGSIZE);
 405                 kfree_skb(skb);
 406                 goto errout;
 407         }
 408
 409         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 410         return;
 411 errout:
 412         if (err < 0)
 413                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 414 }
 415
 416 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 417 {
 418         struct vxlan_dev *vxlan = netdev_priv(dev);
 419         struct vxlan_fdb f = {
 420                 .state = NUD_STALE,
 421         };
 422         struct vxlan_rdst remote = {
 423                 .remote_ip = *ipa, /* goes to NDA_DST */
 424                 .remote_vni = VXLAN_N_VID,
 425         };
 426
 427         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 428 }
 429
 430 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 431 {
 432         struct vxlan_fdb f = {
 433                 .state = NUD_STALE,
 434         };
 435         struct vxlan_rdst remote = { };
 436
 437         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 438
 439         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 440 }
 441
 442 /* Hash Ethernet address */
 443 static u32 eth_hash(const unsigned char *addr)
 444 {
 445         u64 value = get_unaligned((u64 *)addr);
 446
 447         /* only want 6 bytes */
 448 #ifdef __BIG_ENDIAN
 449         value >>= 16;
 450 #else
 451         value <<= 16;
 452 #endif
 453         return hash_64(value, FDB_HASH_BITS);
 454 }
 455
 456 /* Hash chain to use given mac address */
 457 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 458                                                 const u8 *mac)
 459 {
 460         return &vxlan->fdb_head[eth_hash(mac)];
 461 }
 462
 463 /* Look up Ethernet address in forwarding table */
 464 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 465                                         const u8 *mac)
 466 {
 467         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 468         struct vxlan_fdb *f;
 469
 470         hlist_for_each_entry_rcu(f, head, hlist) {
 471                 if (ether_addr_equal(mac, f->eth_addr))
 472                         return f;
 473         }
 474
 475         return NULL;
 476 }
 477
 478 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 479                                         const u8 *mac)
 480 {
 481         struct vxlan_fdb *f;
 482
 483         f = __vxlan_find_mac(vxlan, mac);
 484         if (f)
 485                 f->used = jiffies;
 486
 487         return f;
 488 }
 489
 490 /* caller should hold vxlan->hash_lock */
 491 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 492                                               union vxlan_addr *ip, __be16 port,
 493                                               __u32 vni, __u32 ifindex)
 494 {
 495         struct vxlan_rdst *rd;
 496
 497         list_for_each_entry(rd, &f->remotes, list) {
 498                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 499                     rd->remote_port == port &&
 500                     rd->remote_vni == vni &&
 501                     rd->remote_ifindex == ifindex)
 502                         return rd;
 503         }
 504
 505         return NULL;
 506 }
 507
 508 /* Replace destination of unicast mac */
 509 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 510                              union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
 511 {
 512         struct vxlan_rdst *rd;
 513
 514         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 515         if (rd)
 516                 return 0;
 517
 518         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 519         if (!rd)
 520                 return 0;
 521         rd->remote_ip = *ip;
 522         rd->remote_port = port;
 523         rd->remote_vni = vni;
 524         rd->remote_ifindex = ifindex;
 525         return 1;
 526 }
 527
 528 /* Add/update destinations for multicast */
 529 static int vxlan_fdb_append(struct vxlan_fdb *f,
 530                             union vxlan_addr *ip, __be16 port, __u32 vni,
 531                             __u32 ifindex, struct vxlan_rdst **rdp)
 532 {
 533         struct vxlan_rdst *rd;
 534
 535         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 536         if (rd)
 537                 return 0;
 538
 539         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 540         if (rd == NULL)
 541                 return -ENOBUFS;
 542         rd->remote_ip = *ip;
 543         rd->remote_port = port;
 544         rd->remote_vni = vni;
 545         rd->remote_ifindex = ifindex;
 546
 547         list_add_tail_rcu(&rd->list, &f->remotes);
 548
 549         *rdp = rd;
 550         return 1;
 551 }
 552
 553 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 554                                           unsigned int off,
 555                                           struct vxlanhdr *vh, size_t hdrlen,
 556                                           u32 data)
 557 {
 558         size_t start, offset, plen;
 559         __wsum delta;
 560
 561         if (skb->remcsum_offload)
 562                 return vh;
 563
 564         if (!NAPI_GRO_CB(skb)->csum_valid)
 565                 return NULL;
 566
 567         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
 568         offset = start + ((data & VXLAN_RCO_UDP) ?
 569                           offsetof(struct udphdr, check) :
 570                           offsetof(struct tcphdr, check));
 571
 572         plen = hdrlen + offset + sizeof(u16);
 573
 574         /* Pull checksum that will be written */
 575         if (skb_gro_header_hard(skb, off + plen)) {
 576                 vh = skb_gro_header_slow(skb, off + plen, off);
 577                 if (!vh)
 578                         return NULL;
 579         }
 580
 581         delta = remcsum_adjust((void *)vh + hdrlen,
 582                                NAPI_GRO_CB(skb)->csum, start, offset);
 583
 584         /* Adjust skb->csum since we changed the packet */
 585         skb->csum = csum_add(skb->csum, delta);
 586         NAPI_GRO_CB(skb)->csum = csum_add(NAPI_GRO_CB(skb)->csum, delta);
 587
 588         skb->remcsum_offload = 1;
 589
 590         return vh;
 591 }
 592
 593 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
 594                                           struct sk_buff *skb,
 595                                           struct udp_offload *uoff)
 596 {
 597         struct sk_buff *p, **pp = NULL;
 598         struct vxlanhdr *vh, *vh2;
 599         unsigned int hlen, off_vx;
 600         int flush = 1;
 601         struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
 602                                              udp_offloads);
 603         u32 flags;
 604
 605         off_vx = skb_gro_offset(skb);
 606         hlen = off_vx + sizeof(*vh);
 607         vh   = skb_gro_header_fast(skb, off_vx);
 608         if (skb_gro_header_hard(skb, hlen)) {
 609                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 610                 if (unlikely(!vh))
 611                         goto out;
 612         }
 613
 614         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 615         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 616
 617         flags = ntohl(vh->vx_flags);
 618
 619         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 620                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 621                                        ntohl(vh->vx_vni));
 622
 623                 if (!vh)
 624                         goto out;
 625         }
 626
 627         flush = 0;
 628
 629         for (p = *head; p; p = p->next) {
 630                 if (!NAPI_GRO_CB(p)->same_flow)
 631                         continue;
 632
 633                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 634                 if (vh->vx_flags != vh2->vx_flags ||
 635                     vh->vx_vni != vh2->vx_vni) {
 636                         NAPI_GRO_CB(p)->same_flow = 0;
 637                         continue;
 638                 }
 639         }
 640
 641         pp = eth_gro_receive(head, skb);
 642
 643 out:
 644         NAPI_GRO_CB(skb)->flush |= flush;
 645
 646         return pp;
 647 }
 648
 649 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
 650                               struct udp_offload *uoff)
 651 {
 652         udp_tunnel_gro_complete(skb, nhoff);
 653
 654         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 655 }
 656
 657 /* Notify netdevs that UDP port started listening */
 658 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 659 {
 660         struct net_device *dev;
 661         struct sock *sk = vs->sock->sk;
 662         struct net *net = sock_net(sk);
 663         sa_family_t sa_family = sk->sk_family;
 664         __be16 port = inet_sk(sk)->inet_sport;
 665         int err;
 666
 667         if (sa_family == AF_INET) {
 668                 err = udp_add_offload(&vs->udp_offloads);
 669                 if (err)
 670                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 671         }
 672
 673         rcu_read_lock();
 674         for_each_netdev_rcu(net, dev) {
 675                 if (dev->netdev_ops->ndo_add_vxlan_port)
 676                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 677                                                             port);
 678         }
 679         rcu_read_unlock();
 680 }
 681
 682 /* Notify netdevs that UDP port is no more listening */
 683 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 684 {
 685         struct net_device *dev;
 686         struct sock *sk = vs->sock->sk;
 687         struct net *net = sock_net(sk);
 688         sa_family_t sa_family = sk->sk_family;
 689         __be16 port = inet_sk(sk)->inet_sport;
 690
 691         rcu_read_lock();
 692         for_each_netdev_rcu(net, dev) {
 693                 if (dev->netdev_ops->ndo_del_vxlan_port)
 694                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 695                                                             port);
 696         }
 697         rcu_read_unlock();
 698
 699         if (sa_family == AF_INET)
 700                 udp_del_offload(&vs->udp_offloads);
 701 }
 702
 703 /* Add new entry to forwarding table -- assumes lock held */
 704 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 705                             const u8 *mac, union vxlan_addr *ip,
 706                             __u16 state, __u16 flags,
 707                             __be16 port, __u32 vni, __u32 ifindex,
 708                             __u8 ndm_flags)
 709 {
 710         struct vxlan_rdst *rd = NULL;
 711         struct vxlan_fdb *f;
 712         int notify = 0;
 713
 714         f = __vxlan_find_mac(vxlan, mac);
 715         if (f) {
 716                 if (flags & NLM_F_EXCL) {
 717                         netdev_dbg(vxlan->dev,
 718                                    "lost race to create %pM\n", mac);
 719                         return -EEXIST;
 720                 }
 721                 if (f->state != state) {
 722                         f->state = state;
 723                         f->updated = jiffies;
 724                         notify = 1;
 725                 }
 726                 if (f->flags != ndm_flags) {
 727                         f->flags = ndm_flags;
 728                         f->updated = jiffies;
 729                         notify = 1;
 730                 }
 731                 if ((flags & NLM_F_REPLACE)) {
 732                         /* Only change unicasts */
 733                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 734                              is_zero_ether_addr(f->eth_addr))) {
 735                                 int rc = vxlan_fdb_replace(f, ip, port, vni,
 736                                                            ifindex);
 737
 738                                 if (rc < 0)
 739                                         return rc;
 740                                 notify |= rc;
 741                         } else
 742                                 return -EOPNOTSUPP;
 743                 }
 744                 if ((flags & NLM_F_APPEND) &&
 745                     (is_multicast_ether_addr(f->eth_addr) ||
 746                      is_zero_ether_addr(f->eth_addr))) {
 747                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 748                                                   &rd);
 749
 750                         if (rc < 0)
 751                                 return rc;
 752                         notify |= rc;
 753                 }
 754         } else {
 755                 if (!(flags & NLM_F_CREATE))
 756                         return -ENOENT;
 757
 758                 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
 759                         return -ENOSPC;
 760
 761                 /* Disallow replace to add a multicast entry */
 762                 if ((flags & NLM_F_REPLACE) &&
 763                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 764                         return -EOPNOTSUPP;
 765
 766                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 767                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 768                 if (!f)
 769                         return -ENOMEM;
 770
 771                 notify = 1;
 772                 f->state = state;
 773                 f->flags = ndm_flags;
 774                 f->updated = f->used = jiffies;
 775                 INIT_LIST_HEAD(&f->remotes);
 776                 memcpy(f->eth_addr, mac, ETH_ALEN);
 777
 778                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 779
 780                 ++vxlan->addrcnt;
 781                 hlist_add_head_rcu(&f->hlist,
 782                                    vxlan_fdb_head(vxlan, mac));
 783         }
 784
 785         if (notify) {
 786                 if (rd == NULL)
 787                         rd = first_remote_rtnl(f);
 788                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 789         }
 790
 791         return 0;
 792 }
 793
 794 static void vxlan_fdb_free(struct rcu_head *head)
 795 {
 796         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 797         struct vxlan_rdst *rd, *nd;
 798
 799         list_for_each_entry_safe(rd, nd, &f->remotes, list)
 800                 kfree(rd);
 801         kfree(f);
 802 }
 803
 804 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 805 {
 806         netdev_dbg(vxlan->dev,
 807                     "delete %pM\n", f->eth_addr);
 808
 809         --vxlan->addrcnt;
 810         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 811
 812         hlist_del_rcu(&f->hlist);
 813         call_rcu(&f->rcu, vxlan_fdb_free);
 814 }
 815
 816 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 817                            union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
 818 {
 819         struct net *net = dev_net(vxlan->dev);
 820         int err;
 821
 822         if (tb[NDA_DST]) {
 823                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 824                 if (err)
 825                         return err;
 826         } else {
 827                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 828                 if (remote->sa.sa_family == AF_INET) {
 829                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 830                         ip->sa.sa_family = AF_INET;
 831 #if IS_ENABLED(CONFIG_IPV6)
 832                 } else {
 833                         ip->sin6.sin6_addr = in6addr_any;
 834                         ip->sa.sa_family = AF_INET6;
 835 #endif
 836                 }
 837         }
 838
 839         if (tb[NDA_PORT]) {
 840                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 841                         return -EINVAL;
 842                 *port = nla_get_be16(tb[NDA_PORT]);
 843         } else {
 844                 *port = vxlan->dst_port;
 845         }
 846
 847         if (tb[NDA_VNI]) {
 848                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 849                         return -EINVAL;
 850                 *vni = nla_get_u32(tb[NDA_VNI]);
 851         } else {
 852                 *vni = vxlan->default_dst.remote_vni;
 853         }
 854
 855         if (tb[NDA_IFINDEX]) {
 856                 struct net_device *tdev;
 857
 858                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 859                         return -EINVAL;
 860                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 861                 tdev = __dev_get_by_index(net, *ifindex);
 862                 if (!tdev)
 863                         return -EADDRNOTAVAIL;
 864         } else {
 865                 *ifindex = 0;
 866         }
 867
 868         return 0;
 869 }
 870
 871 /* Add static entry (via netlink) */
 872 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 873                          struct net_device *dev,
 874                          const unsigned char *addr, u16 vid, u16 flags)
 875 {
 876         struct vxlan_dev *vxlan = netdev_priv(dev);
 877         /* struct net *net = dev_net(vxlan->dev); */
 878         union vxlan_addr ip;
 879         __be16 port;
 880         u32 vni, ifindex;
 881         int err;
 882
 883         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 884                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 885                         ndm->ndm_state);
 886                 return -EINVAL;
 887         }
 888
 889         if (tb[NDA_DST] == NULL)
 890                 return -EINVAL;
 891
 892         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 893         if (err)
 894                 return err;
 895
 896         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 897                 return -EAFNOSUPPORT;
 898
 899         spin_lock_bh(&vxlan->hash_lock);
 900         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 901                                port, vni, ifindex, ndm->ndm_flags);
 902         spin_unlock_bh(&vxlan->hash_lock);
 903
 904         return err;
 905 }
 906
 907 /* Delete entry (via netlink) */
 908 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 909                             struct net_device *dev,
 910                             const unsigned char *addr, u16 vid)
 911 {
 912         struct vxlan_dev *vxlan = netdev_priv(dev);
 913         struct vxlan_fdb *f;
 914         struct vxlan_rdst *rd = NULL;
 915         union vxlan_addr ip;
 916         __be16 port;
 917         u32 vni, ifindex;
 918         int err;
 919
 920         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 921         if (err)
 922                 return err;
 923
 924         err = -ENOENT;
 925
 926         spin_lock_bh(&vxlan->hash_lock);
 927         f = vxlan_find_mac(vxlan, addr);
 928         if (!f)
 929                 goto out;
 930
 931         if (!vxlan_addr_any(&ip)) {
 932                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 933                 if (!rd)
 934                         goto out;
 935         }
 936
 937         err = 0;
 938
 939         /* remove a destination if it's not the only one on the list,
 940          * otherwise destroy the fdb entry
 941          */
 942         if (rd && !list_is_singular(&f->remotes)) {
 943                 list_del_rcu(&rd->list);
 944                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 945                 kfree_rcu(rd, rcu);
 946                 goto out;
 947         }
 948
 949         vxlan_fdb_destroy(vxlan, f);
 950
 951 out:
 952         spin_unlock_bh(&vxlan->hash_lock);
 953
 954         return err;
 955 }
 956
 957 /* Dump forwarding table */
 958 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 959                           struct net_device *dev,
 960                           struct net_device *filter_dev, int idx)
 961 {
 962         struct vxlan_dev *vxlan = netdev_priv(dev);
 963         unsigned int h;
 964
 965         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 966                 struct vxlan_fdb *f;
 967                 int err;
 968
 969                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 970                         struct vxlan_rdst *rd;
 971
 972                         if (idx < cb->args[0])
 973                                 goto skip;
 974
 975                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 976                                 err = vxlan_fdb_info(skb, vxlan, f,
 977                                                      NETLINK_CB(cb->skb).portid,
 978                                                      cb->nlh->nlmsg_seq,
 979                                                      RTM_NEWNEIGH,
 980                                                      NLM_F_MULTI, rd);
 981                                 if (err < 0)
 982                                         goto out;
 983                         }
 984 skip:
 985                         ++idx;
 986                 }
 987         }
 988 out:
 989         return idx;
 990 }
 991
 992 /* Watch incoming packets to learn mapping between Ethernet address
 993  * and Tunnel endpoint.
 994  * Return true if packet is bogus and should be droppped.
 995  */
 996 static bool vxlan_snoop(struct net_device *dev,
 997                         union vxlan_addr *src_ip, const u8 *src_mac)
 998 {
 999         struct vxlan_dev *vxlan = netdev_priv(dev);
1000         struct vxlan_fdb *f;
1001
1002         f = vxlan_find_mac(vxlan, src_mac);
1003         if (likely(f)) {
1004                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1005
1006                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
1007                         return false;
1008
1009                 /* Don't migrate static entries, drop packets */
1010                 if (f->state & NUD_NOARP)
1011                         return true;
1012
1013                 if (net_ratelimit())
1014                         netdev_info(dev,
1015                                     "%pM migrated from %pIS to %pIS\n",
1016                                     src_mac, &rdst->remote_ip, &src_ip);
1017
1018                 rdst->remote_ip = *src_ip;
1019                 f->updated = jiffies;
1020                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1021         } else {
1022                 /* learned new entry */
1023                 spin_lock(&vxlan->hash_lock);
1024
1025                 /* close off race between vxlan_flush and incoming packets */
1026                 if (netif_running(dev))
1027                         vxlan_fdb_create(vxlan, src_mac, src_ip,
1028                                          NUD_REACHABLE,
1029                                          NLM_F_EXCL|NLM_F_CREATE,
1030                                          vxlan->dst_port,
1031                                          vxlan->default_dst.remote_vni,
1032                                          0, NTF_SELF);
1033                 spin_unlock(&vxlan->hash_lock);
1034         }
1035
1036         return false;
1037 }
1038
1039 /* See if multicast group is already in use by other ID */
1040 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1041 {
1042         struct vxlan_dev *vxlan;
1043
1044         /* The vxlan_sock is only used by dev, leaving group has
1045          * no effect on other vxlan devices.
1046          */
1047         if (atomic_read(&dev->vn_sock->refcnt) == 1)
1048                 return false;
1049
1050         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1051                 if (!netif_running(vxlan->dev) || vxlan == dev)
1052                         continue;
1053
1054                 if (vxlan->vn_sock != dev->vn_sock)
1055                         continue;
1056
1057                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1058                                       &dev->default_dst.remote_ip))
1059                         continue;
1060
1061                 if (vxlan->default_dst.remote_ifindex !=
1062                     dev->default_dst.remote_ifindex)
1063                         continue;
1064
1065                 return true;
1066         }
1067
1068         return false;
1069 }
1070
1071 static void vxlan_sock_hold(struct vxlan_sock *vs)
1072 {
1073         atomic_inc(&vs->refcnt);
1074 }
1075
1076 void vxlan_sock_release(struct vxlan_sock *vs)
1077 {
1078         struct sock *sk = vs->sock->sk;
1079         struct net *net = sock_net(sk);
1080         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1081
1082         if (!atomic_dec_and_test(&vs->refcnt))
1083                 return;
1084
1085         spin_lock(&vn->sock_lock);
1086         hlist_del_rcu(&vs->hlist);
1087         vxlan_notify_del_rx_port(vs);
1088         spin_unlock(&vn->sock_lock);
1089
1090         queue_work(vxlan_wq, &vs->del_work);
1091 }
1092 EXPORT_SYMBOL_GPL(vxlan_sock_release);
1093
1094 /* Callback to update multicast group membership when first VNI on
1095  * multicast asddress is brought up
1096  * Done as workqueue because ip_mc_join_group acquires RTNL.
1097  */
1098 static void vxlan_igmp_join(struct work_struct *work)
1099 {
1100         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_join);
1101         struct vxlan_sock *vs = vxlan->vn_sock;
1102         struct sock *sk = vs->sock->sk;
1103         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1104         int ifindex = vxlan->default_dst.remote_ifindex;
1105
1106         lock_sock(sk);
1107         if (ip->sa.sa_family == AF_INET) {
1108                 struct ip_mreqn mreq = {
1109                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1110                         .imr_ifindex            = ifindex,
1111                 };
1112
1113                 ip_mc_join_group(sk, &mreq);
1114 #if IS_ENABLED(CONFIG_IPV6)
1115         } else {
1116                 ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1117                                              &ip->sin6.sin6_addr);
1118 #endif
1119         }
1120         release_sock(sk);
1121
1122         vxlan_sock_release(vs);
1123         dev_put(vxlan->dev);
1124 }
1125
1126 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1127 static void vxlan_igmp_leave(struct work_struct *work)
1128 {
1129         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_leave);
1130         struct vxlan_sock *vs = vxlan->vn_sock;
1131         struct sock *sk = vs->sock->sk;
1132         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1133         int ifindex = vxlan->default_dst.remote_ifindex;
1134
1135         lock_sock(sk);
1136         if (ip->sa.sa_family == AF_INET) {
1137                 struct ip_mreqn mreq = {
1138                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1139                         .imr_ifindex            = ifindex,
1140                 };
1141
1142                 ip_mc_leave_group(sk, &mreq);
1143 #if IS_ENABLED(CONFIG_IPV6)
1144         } else {
1145                 ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1146                                              &ip->sin6.sin6_addr);
1147 #endif
1148         }
1149
1150         release_sock(sk);
1151
1152         vxlan_sock_release(vs);
1153         dev_put(vxlan->dev);
1154 }
1155
1156 static struct vxlanhdr *vxlan_remcsum(struct sk_buff *skb, struct vxlanhdr *vh,
1157                                       size_t hdrlen, u32 data)
1158 {
1159         size_t start, offset, plen;
1160         __wsum delta;
1161
1162         if (skb->remcsum_offload) {
1163                 /* Already processed in GRO path */
1164                 skb->remcsum_offload = 0;
1165                 return vh;
1166         }
1167
1168         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
1169         offset = start + ((data & VXLAN_RCO_UDP) ?
1170                           offsetof(struct udphdr, check) :
1171                           offsetof(struct tcphdr, check));
1172
1173         plen = hdrlen + offset + sizeof(u16);
1174
1175         if (!pskb_may_pull(skb, plen))
1176                 return NULL;
1177
1178         vh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1179
1180         if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE))
1181                 __skb_checksum_complete(skb);
1182
1183         delta = remcsum_adjust((void *)vh + hdrlen,
1184                                skb->csum, start, offset);
1185
1186         /* Adjust skb->csum since we changed the packet */
1187         skb->csum = csum_add(skb->csum, delta);
1188
1189         return vh;
1190 }
1191
1192 /* Callback from net/ipv4/udp.c to receive packets */
1193 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1194 {
1195         struct vxlan_sock *vs;
1196         struct vxlanhdr *vxh;
1197         u32 flags, vni;
1198         struct vxlan_metadata md = {0};
1199
1200         /* Need Vxlan and inner Ethernet header to be present */
1201         if (!pskb_may_pull(skb, VXLAN_HLEN))
1202                 goto error;
1203
1204         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1205         flags = ntohl(vxh->vx_flags);
1206         vni = ntohl(vxh->vx_vni);
1207
1208         if (flags & VXLAN_HF_VNI) {
1209                 flags &= ~VXLAN_HF_VNI;
1210         } else {
1211                 /* VNI flag always required to be set */
1212                 goto bad_flags;
1213         }
1214
1215         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1216                 goto drop;
1217         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1218
1219         vs = rcu_dereference_sk_user_data(sk);
1220         if (!vs)
1221                 goto drop;
1222
1223         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1224                 vxh = vxlan_remcsum(skb, vxh, sizeof(struct vxlanhdr), vni);
1225                 if (!vxh)
1226                         goto drop;
1227
1228                 flags &= ~VXLAN_HF_RCO;
1229                 vni &= VXLAN_VID_MASK;
1230         }
1231
1232         /* For backwards compatibility, only allow reserved fields to be
1233          * used by VXLAN extensions if explicitly requested.
1234          */
1235         if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1236                 struct vxlanhdr_gbp *gbp;
1237
1238                 gbp = (struct vxlanhdr_gbp *)vxh;
1239                 md.gbp = ntohs(gbp->policy_id);
1240
1241                 if (gbp->dont_learn)
1242                         md.gbp |= VXLAN_GBP_DONT_LEARN;
1243
1244                 if (gbp->policy_applied)
1245                         md.gbp |= VXLAN_GBP_POLICY_APPLIED;
1246
1247                 flags &= ~VXLAN_GBP_USED_BITS;
1248         }
1249
1250         if (flags || (vni & ~VXLAN_VID_MASK)) {
1251                 /* If there are any unprocessed flags remaining treat
1252                  * this as a malformed packet. This behavior diverges from
1253                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1254                  * in reserved fields are to be ignored. The approach here
1255                  * maintains compatbility with previous stack code, and also
1256                  * is more robust and provides a little more security in
1257                  * adding extensions to VXLAN.
1258                  */
1259
1260                 goto bad_flags;
1261         }
1262
1263         md.vni = vxh->vx_vni;
1264         vs->rcv(vs, skb, &md);
1265         return 0;
1266
1267 drop:
1268         /* Consume bad packet */
1269         kfree_skb(skb);
1270         return 0;
1271
1272 bad_flags:
1273         netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1274                    ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1275
1276 error:
1277         /* Return non vxlan pkt */
1278         return 1;
1279 }
1280
1281 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1282                       struct vxlan_metadata *md)
1283 {
1284         struct iphdr *oip = NULL;
1285         struct ipv6hdr *oip6 = NULL;
1286         struct vxlan_dev *vxlan;
1287         struct pcpu_sw_netstats *stats;
1288         union vxlan_addr saddr;
1289         __u32 vni;
1290         int err = 0;
1291         union vxlan_addr *remote_ip;
1292
1293         vni = ntohl(md->vni) >> 8;
1294         /* Is this VNI defined? */
1295         vxlan = vxlan_vs_find_vni(vs, vni);
1296         if (!vxlan)
1297                 goto drop;
1298
1299         remote_ip = &vxlan->default_dst.remote_ip;
1300         skb_reset_mac_header(skb);
1301         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1302         skb->protocol = eth_type_trans(skb, vxlan->dev);
1303         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1304
1305         /* Ignore packet loops (and multicast echo) */
1306         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1307                 goto drop;
1308
1309         /* Re-examine inner Ethernet packet */
1310         if (remote_ip->sa.sa_family == AF_INET) {
1311                 oip = ip_hdr(skb);
1312                 saddr.sin.sin_addr.s_addr = oip->saddr;
1313                 saddr.sa.sa_family = AF_INET;
1314 #if IS_ENABLED(CONFIG_IPV6)
1315         } else {
1316                 oip6 = ipv6_hdr(skb);
1317                 saddr.sin6.sin6_addr = oip6->saddr;
1318                 saddr.sa.sa_family = AF_INET6;
1319 #endif
1320         }
1321
1322         if ((vxlan->flags & VXLAN_F_LEARN) &&
1323             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1324                 goto drop;
1325
1326         skb_reset_network_header(skb);
1327         skb->mark = md->gbp;
1328
1329         if (oip6)
1330                 err = IP6_ECN_decapsulate(oip6, skb);
1331         if (oip)
1332                 err = IP_ECN_decapsulate(oip, skb);
1333
1334         if (unlikely(err)) {
1335                 if (log_ecn_error) {
1336                         if (oip6)
1337                                 net_info_ratelimited("non-ECT from %pI6\n",
1338                                                      &oip6->saddr);
1339                         if (oip)
1340                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1341                                                      &oip->saddr, oip->tos);
1342                 }
1343                 if (err > 1) {
1344                         ++vxlan->dev->stats.rx_frame_errors;
1345                         ++vxlan->dev->stats.rx_errors;
1346                         goto drop;
1347                 }
1348         }
1349
1350         stats = this_cpu_ptr(vxlan->dev->tstats);
1351         u64_stats_update_begin(&stats->syncp);
1352         stats->rx_packets++;
1353         stats->rx_bytes += skb->len;
1354         u64_stats_update_end(&stats->syncp);
1355
1356         netif_rx(skb);
1357
1358         return;
1359 drop:
1360         /* Consume bad packet */
1361         kfree_skb(skb);
1362 }
1363
1364 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1365 {
1366         struct vxlan_dev *vxlan = netdev_priv(dev);
1367         struct arphdr *parp;
1368         u8 *arpptr, *sha;
1369         __be32 sip, tip;
1370         struct neighbour *n;
1371
1372         if (dev->flags & IFF_NOARP)
1373                 goto out;
1374
1375         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1376                 dev->stats.tx_dropped++;
1377                 goto out;
1378         }
1379         parp = arp_hdr(skb);
1380
1381         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1382              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1383             parp->ar_pro != htons(ETH_P_IP) ||
1384             parp->ar_op != htons(ARPOP_REQUEST) ||
1385             parp->ar_hln != dev->addr_len ||
1386             parp->ar_pln != 4)
1387                 goto out;
1388         arpptr = (u8 *)parp + sizeof(struct arphdr);
1389         sha = arpptr;
1390         arpptr += dev->addr_len;        /* sha */
1391         memcpy(&sip, arpptr, sizeof(sip));
1392         arpptr += sizeof(sip);
1393         arpptr += dev->addr_len;        /* tha */
1394         memcpy(&tip, arpptr, sizeof(tip));
1395
1396         if (ipv4_is_loopback(tip) ||
1397             ipv4_is_multicast(tip))
1398                 goto out;
1399
1400         n = neigh_lookup(&arp_tbl, &tip, dev);
1401
1402         if (n) {
1403                 struct vxlan_fdb *f;
1404                 struct sk_buff  *reply;
1405
1406                 if (!(n->nud_state & NUD_CONNECTED)) {
1407                         neigh_release(n);
1408                         goto out;
1409                 }
1410
1411                 f = vxlan_find_mac(vxlan, n->ha);
1412                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1413                         /* bridge-local neighbor */
1414                         neigh_release(n);
1415                         goto out;
1416                 }
1417
1418                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1419                                 n->ha, sha);
1420
1421                 neigh_release(n);
1422
1423                 if (reply == NULL)
1424                         goto out;
1425
1426                 skb_reset_mac_header(reply);
1427                 __skb_pull(reply, skb_network_offset(reply));
1428                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1429                 reply->pkt_type = PACKET_HOST;
1430
1431                 if (netif_rx_ni(reply) == NET_RX_DROP)
1432                         dev->stats.rx_dropped++;
1433         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1434                 union vxlan_addr ipa = {
1435                         .sin.sin_addr.s_addr = tip,
1436                         .sin.sin_family = AF_INET,
1437                 };
1438
1439                 vxlan_ip_miss(dev, &ipa);
1440         }
1441 out:
1442         consume_skb(skb);
1443         return NETDEV_TX_OK;
1444 }
1445
1446 #if IS_ENABLED(CONFIG_IPV6)
1447 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1448         struct neighbour *n, bool isrouter)
1449 {
1450         struct net_device *dev = request->dev;
1451         struct sk_buff *reply;
1452         struct nd_msg *ns, *na;
1453         struct ipv6hdr *pip6;
1454         u8 *daddr;
1455         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1456         int ns_olen;
1457         int i, len;
1458
1459         if (dev == NULL)
1460                 return NULL;
1461
1462         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1463                 sizeof(*na) + na_olen + dev->needed_tailroom;
1464         reply = alloc_skb(len, GFP_ATOMIC);
1465         if (reply == NULL)
1466                 return NULL;
1467
1468         reply->protocol = htons(ETH_P_IPV6);
1469         reply->dev = dev;
1470         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1471         skb_push(reply, sizeof(struct ethhdr));
1472         skb_set_mac_header(reply, 0);
1473
1474         ns = (struct nd_msg *)skb_transport_header(request);
1475
1476         daddr = eth_hdr(request)->h_source;
1477         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1478         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1479                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1480                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1481                         break;
1482                 }
1483         }
1484
1485         /* Ethernet header */
1486         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1487         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1488         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1489         reply->protocol = htons(ETH_P_IPV6);
1490
1491         skb_pull(reply, sizeof(struct ethhdr));
1492         skb_set_network_header(reply, 0);
1493         skb_put(reply, sizeof(struct ipv6hdr));
1494
1495         /* IPv6 header */
1496
1497         pip6 = ipv6_hdr(reply);
1498         memset(pip6, 0, sizeof(struct ipv6hdr));
1499         pip6->version = 6;
1500         pip6->priority = ipv6_hdr(request)->priority;
1501         pip6->nexthdr = IPPROTO_ICMPV6;
1502         pip6->hop_limit = 255;
1503         pip6->daddr = ipv6_hdr(request)->saddr;
1504         pip6->saddr = *(struct in6_addr *)n->primary_key;
1505
1506         skb_pull(reply, sizeof(struct ipv6hdr));
1507         skb_set_transport_header(reply, 0);
1508
1509         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1510
1511         /* Neighbor Advertisement */
1512         memset(na, 0, sizeof(*na)+na_olen);
1513         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1514         na->icmph.icmp6_router = isrouter;
1515         na->icmph.icmp6_override = 1;
1516         na->icmph.icmp6_solicited = 1;
1517         na->target = ns->target;
1518         ether_addr_copy(&na->opt[2], n->ha);
1519         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1520         na->opt[1] = na_olen >> 3;
1521
1522         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1523                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1524                 csum_partial(na, sizeof(*na)+na_olen, 0));
1525
1526         pip6->payload_len = htons(sizeof(*na)+na_olen);
1527
1528         skb_push(reply, sizeof(struct ipv6hdr));
1529
1530         reply->ip_summed = CHECKSUM_UNNECESSARY;
1531
1532         return reply;
1533 }
1534
1535 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1536 {
1537         struct vxlan_dev *vxlan = netdev_priv(dev);
1538         struct nd_msg *msg;
1539         const struct ipv6hdr *iphdr;
1540         const struct in6_addr *saddr, *daddr;
1541         struct neighbour *n;
1542         struct inet6_dev *in6_dev;
1543
1544         in6_dev = __in6_dev_get(dev);
1545         if (!in6_dev)
1546                 goto out;
1547
1548         iphdr = ipv6_hdr(skb);
1549         saddr = &iphdr->saddr;
1550         daddr = &iphdr->daddr;
1551
1552         msg = (struct nd_msg *)skb_transport_header(skb);
1553         if (msg->icmph.icmp6_code != 0 ||
1554             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1555                 goto out;
1556
1557         if (ipv6_addr_loopback(daddr) ||
1558             ipv6_addr_is_multicast(&msg->target))
1559                 goto out;
1560
1561         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1562
1563         if (n) {
1564                 struct vxlan_fdb *f;
1565                 struct sk_buff *reply;
1566
1567                 if (!(n->nud_state & NUD_CONNECTED)) {
1568                         neigh_release(n);
1569                         goto out;
1570                 }
1571
1572                 f = vxlan_find_mac(vxlan, n->ha);
1573                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1574                         /* bridge-local neighbor */
1575                         neigh_release(n);
1576                         goto out;
1577                 }
1578
1579                 reply = vxlan_na_create(skb, n,
1580                                         !!(f ? f->flags & NTF_ROUTER : 0));
1581
1582                 neigh_release(n);
1583
1584                 if (reply == NULL)
1585                         goto out;
1586
1587                 if (netif_rx_ni(reply) == NET_RX_DROP)
1588                         dev->stats.rx_dropped++;
1589
1590         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1591                 union vxlan_addr ipa = {
1592                         .sin6.sin6_addr = msg->target,
1593                         .sin6.sin6_family = AF_INET6,
1594                 };
1595
1596                 vxlan_ip_miss(dev, &ipa);
1597         }
1598
1599 out:
1600         consume_skb(skb);
1601         return NETDEV_TX_OK;
1602 }
1603 #endif
1604
1605 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1606 {
1607         struct vxlan_dev *vxlan = netdev_priv(dev);
1608         struct neighbour *n;
1609
1610         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1611                 return false;
1612
1613         n = NULL;
1614         switch (ntohs(eth_hdr(skb)->h_proto)) {
1615         case ETH_P_IP:
1616         {
1617                 struct iphdr *pip;
1618
1619                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1620                         return false;
1621                 pip = ip_hdr(skb);
1622                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1623                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1624                         union vxlan_addr ipa = {
1625                                 .sin.sin_addr.s_addr = pip->daddr,
1626                                 .sin.sin_family = AF_INET,
1627                         };
1628
1629                         vxlan_ip_miss(dev, &ipa);
1630                         return false;
1631                 }
1632
1633                 break;
1634         }
1635 #if IS_ENABLED(CONFIG_IPV6)
1636         case ETH_P_IPV6:
1637         {
1638                 struct ipv6hdr *pip6;
1639
1640                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1641                         return false;
1642                 pip6 = ipv6_hdr(skb);
1643                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1644                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1645                         union vxlan_addr ipa = {
1646                                 .sin6.sin6_addr = pip6->daddr,
1647                                 .sin6.sin6_family = AF_INET6,
1648                         };
1649
1650                         vxlan_ip_miss(dev, &ipa);
1651                         return false;
1652                 }
1653
1654                 break;
1655         }
1656 #endif
1657         default:
1658                 return false;
1659         }
1660
1661         if (n) {
1662                 bool diff;
1663
1664                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1665                 if (diff) {
1666                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1667                                 dev->addr_len);
1668                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1669                 }
1670                 neigh_release(n);
1671                 return diff;
1672         }
1673
1674         return false;
1675 }
1676
1677 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, struct vxlan_sock *vs,
1678                                 struct vxlan_metadata *md)
1679 {
1680         struct vxlanhdr_gbp *gbp;
1681
1682         gbp = (struct vxlanhdr_gbp *)vxh;
1683         vxh->vx_flags |= htonl(VXLAN_HF_GBP);
1684
1685         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1686                 gbp->dont_learn = 1;
1687
1688         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1689                 gbp->policy_applied = 1;
1690
1691         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1692 }
1693
1694 #if IS_ENABLED(CONFIG_IPV6)
1695 static int vxlan6_xmit_skb(struct vxlan_sock *vs,
1696                            struct dst_entry *dst, struct sk_buff *skb,
1697                            struct net_device *dev, struct in6_addr *saddr,
1698                            struct in6_addr *daddr, __u8 prio, __u8 ttl,
1699                            __be16 src_port, __be16 dst_port,
1700                            struct vxlan_metadata *md, bool xnet)
1701 {
1702         struct vxlanhdr *vxh;
1703         int min_headroom;
1704         int err;
1705         bool udp_sum = !udp_get_no_check6_tx(vs->sock->sk);
1706         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1707         u16 hdrlen = sizeof(struct vxlanhdr);
1708
1709         if ((vs->flags & VXLAN_F_REMCSUM_TX) &&
1710             skb->ip_summed == CHECKSUM_PARTIAL) {
1711                 int csum_start = skb_checksum_start_offset(skb);
1712
1713                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1714                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1715                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1716                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1717                         udp_sum = false;
1718                         type |= SKB_GSO_TUNNEL_REMCSUM;
1719                 }
1720         }
1721
1722         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1723         if (IS_ERR(skb)) {
1724                 err = -EINVAL;
1725                 goto err;
1726         }
1727
1728         skb_scrub_packet(skb, xnet);
1729
1730         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1731                         + VXLAN_HLEN + sizeof(struct ipv6hdr)
1732                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1733
1734         /* Need space for new headers (invalidates iph ptr) */
1735         err = skb_cow_head(skb, min_headroom);
1736         if (unlikely(err)) {
1737                 kfree_skb(skb);
1738                 goto err;
1739         }
1740
1741         skb = vlan_hwaccel_push_inside(skb);
1742         if (WARN_ON(!skb)) {
1743                 err = -ENOMEM;
1744                 goto err;
1745         }
1746
1747         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1748         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1749         vxh->vx_vni = md->vni;
1750
1751         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1752                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1753                            VXLAN_RCO_SHIFT;
1754
1755                 if (skb->csum_offset == offsetof(struct udphdr, check))
1756                         data |= VXLAN_RCO_UDP;
1757
1758                 vxh->vx_vni |= htonl(data);
1759                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1760
1761                 if (!skb_is_gso(skb)) {
1762                         skb->ip_summed = CHECKSUM_NONE;
1763                         skb->encapsulation = 0;
1764                 }
1765         }
1766
1767         if (vs->flags & VXLAN_F_GBP)
1768                 vxlan_build_gbp_hdr(vxh, vs, md);
1769
1770         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1771
1772         udp_tunnel6_xmit_skb(vs->sock, dst, skb, dev, saddr, daddr, prio,
1773                              ttl, src_port, dst_port);
1774         return 0;
1775 err:
1776         dst_release(dst);
1777         return err;
1778 }
1779 #endif
1780
1781 int vxlan_xmit_skb(struct vxlan_sock *vs,
1782                    struct rtable *rt, struct sk_buff *skb,
1783                    __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1784                    __be16 src_port, __be16 dst_port,
1785                    struct vxlan_metadata *md, bool xnet)
1786 {
1787         struct vxlanhdr *vxh;
1788         int min_headroom;
1789         int err;
1790         bool udp_sum = !vs->sock->sk->sk_no_check_tx;
1791         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1792         u16 hdrlen = sizeof(struct vxlanhdr);
1793
1794         if ((vs->flags & VXLAN_F_REMCSUM_TX) &&
1795             skb->ip_summed == CHECKSUM_PARTIAL) {
1796                 int csum_start = skb_checksum_start_offset(skb);
1797
1798                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1799                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1800                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1801                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1802                         udp_sum = false;
1803                         type |= SKB_GSO_TUNNEL_REMCSUM;
1804                 }
1805         }
1806
1807         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1808         if (IS_ERR(skb))
1809                 return PTR_ERR(skb);
1810
1811         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1812                         + VXLAN_HLEN + sizeof(struct iphdr)
1813                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1814
1815         /* Need space for new headers (invalidates iph ptr) */
1816         err = skb_cow_head(skb, min_headroom);
1817         if (unlikely(err)) {
1818                 kfree_skb(skb);
1819                 return err;
1820         }
1821
1822         skb = vlan_hwaccel_push_inside(skb);
1823         if (WARN_ON(!skb))
1824                 return -ENOMEM;
1825
1826         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1827         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1828         vxh->vx_vni = md->vni;
1829
1830         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1831                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1832                            VXLAN_RCO_SHIFT;
1833
1834                 if (skb->csum_offset == offsetof(struct udphdr, check))
1835                         data |= VXLAN_RCO_UDP;
1836
1837                 vxh->vx_vni |= htonl(data);
1838                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1839
1840                 if (!skb_is_gso(skb)) {
1841                         skb->ip_summed = CHECKSUM_NONE;
1842                         skb->encapsulation = 0;
1843                 }
1844         }
1845
1846         if (vs->flags & VXLAN_F_GBP)
1847                 vxlan_build_gbp_hdr(vxh, vs, md);
1848
1849         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1850
1851         return udp_tunnel_xmit_skb(vs->sock, rt, skb, src, dst, tos,
1852                                    ttl, df, src_port, dst_port, xnet);
1853 }
1854 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1855
1856 /* Bypass encapsulation if the destination is local */
1857 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1858                                struct vxlan_dev *dst_vxlan)
1859 {
1860         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1861         union vxlan_addr loopback;
1862         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1863         struct net_device *dev = skb->dev;
1864         int len = skb->len;
1865
1866         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1867         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1868         skb->pkt_type = PACKET_HOST;
1869         skb->encapsulation = 0;
1870         skb->dev = dst_vxlan->dev;
1871         __skb_pull(skb, skb_network_offset(skb));
1872
1873         if (remote_ip->sa.sa_family == AF_INET) {
1874                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1875                 loopback.sa.sa_family =  AF_INET;
1876 #if IS_ENABLED(CONFIG_IPV6)
1877         } else {
1878                 loopback.sin6.sin6_addr = in6addr_loopback;
1879                 loopback.sa.sa_family =  AF_INET6;
1880 #endif
1881         }
1882
1883         if (dst_vxlan->flags & VXLAN_F_LEARN)
1884                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1885
1886         u64_stats_update_begin(&tx_stats->syncp);
1887         tx_stats->tx_packets++;
1888         tx_stats->tx_bytes += len;
1889         u64_stats_update_end(&tx_stats->syncp);
1890
1891         if (netif_rx(skb) == NET_RX_SUCCESS) {
1892                 u64_stats_update_begin(&rx_stats->syncp);
1893                 rx_stats->rx_packets++;
1894                 rx_stats->rx_bytes += len;
1895                 u64_stats_update_end(&rx_stats->syncp);
1896         } else {
1897                 dev->stats.rx_dropped++;
1898         }
1899 }
1900
1901 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1902                            struct vxlan_rdst *rdst, bool did_rsc)
1903 {
1904         struct vxlan_dev *vxlan = netdev_priv(dev);
1905         struct rtable *rt = NULL;
1906         const struct iphdr *old_iph;
1907         struct flowi4 fl4;
1908         union vxlan_addr *dst;
1909         struct vxlan_metadata md;
1910         __be16 src_port = 0, dst_port;
1911         u32 vni;
1912         __be16 df = 0;
1913         __u8 tos, ttl;
1914         int err;
1915
1916         dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1917         vni = rdst->remote_vni;
1918         dst = &rdst->remote_ip;
1919
1920         if (vxlan_addr_any(dst)) {
1921                 if (did_rsc) {
1922                         /* short-circuited back to local bridge */
1923                         vxlan_encap_bypass(skb, vxlan, vxlan);
1924                         return;
1925                 }
1926                 goto drop;
1927         }
1928
1929         old_iph = ip_hdr(skb);
1930
1931         ttl = vxlan->ttl;
1932         if (!ttl && vxlan_addr_multicast(dst))
1933                 ttl = 1;
1934
1935         tos = vxlan->tos;
1936         if (tos == 1)
1937                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1938
1939         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->port_min,
1940                                      vxlan->port_max, true);
1941
1942         if (dst->sa.sa_family == AF_INET) {
1943                 memset(&fl4, 0, sizeof(fl4));
1944                 fl4.flowi4_oif = rdst->remote_ifindex;
1945                 fl4.flowi4_tos = RT_TOS(tos);
1946                 fl4.daddr = dst->sin.sin_addr.s_addr;
1947                 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1948
1949                 rt = ip_route_output_key(vxlan->net, &fl4);
1950                 if (IS_ERR(rt)) {
1951                         netdev_dbg(dev, "no route to %pI4\n",
1952                                    &dst->sin.sin_addr.s_addr);
1953                         dev->stats.tx_carrier_errors++;
1954                         goto tx_error;
1955                 }
1956
1957                 if (rt->dst.dev == dev) {
1958                         netdev_dbg(dev, "circular route to %pI4\n",
1959                                    &dst->sin.sin_addr.s_addr);
1960                         dev->stats.collisions++;
1961                         goto rt_tx_error;
1962                 }
1963
1964                 /* Bypass encapsulation if the destination is local */
1965                 if (rt->rt_flags & RTCF_LOCAL &&
1966                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1967                         struct vxlan_dev *dst_vxlan;
1968
1969                         ip_rt_put(rt);
1970                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1971                                                    dst->sa.sa_family, dst_port,
1972                                                    vxlan->flags);
1973                         if (!dst_vxlan)
1974                                 goto tx_error;
1975                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1976                         return;
1977                 }
1978
1979                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1980                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1981                 md.vni = htonl(vni << 8);
1982                 md.gbp = skb->mark;
1983
1984                 err = vxlan_xmit_skb(vxlan->vn_sock, rt, skb,
1985                                      fl4.saddr, dst->sin.sin_addr.s_addr,
1986                                      tos, ttl, df, src_port, dst_port, &md,
1987                                      !net_eq(vxlan->net, dev_net(vxlan->dev)));
1988                 if (err < 0) {
1989                         /* skb is already freed. */
1990                         skb = NULL;
1991                         goto rt_tx_error;
1992                 }
1993
1994                 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1995 #if IS_ENABLED(CONFIG_IPV6)
1996         } else {
1997                 struct sock *sk = vxlan->vn_sock->sock->sk;
1998                 struct dst_entry *ndst;
1999                 struct flowi6 fl6;
2000                 u32 flags;
2001
2002                 memset(&fl6, 0, sizeof(fl6));
2003                 fl6.flowi6_oif = rdst->remote_ifindex;
2004                 fl6.daddr = dst->sin6.sin6_addr;
2005                 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
2006                 fl6.flowi6_proto = IPPROTO_UDP;
2007
2008                 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
2009                         netdev_dbg(dev, "no route to %pI6\n",
2010                                    &dst->sin6.sin6_addr);
2011                         dev->stats.tx_carrier_errors++;
2012                         goto tx_error;
2013                 }
2014
2015                 if (ndst->dev == dev) {
2016                         netdev_dbg(dev, "circular route to %pI6\n",
2017                                    &dst->sin6.sin6_addr);
2018                         dst_release(ndst);
2019                         dev->stats.collisions++;
2020                         goto tx_error;
2021                 }
2022
2023                 /* Bypass encapsulation if the destination is local */
2024                 flags = ((struct rt6_info *)ndst)->rt6i_flags;
2025                 if (flags & RTF_LOCAL &&
2026                     !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2027                         struct vxlan_dev *dst_vxlan;
2028
2029                         dst_release(ndst);
2030                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2031                                                    dst->sa.sa_family, dst_port,
2032                                                    vxlan->flags);
2033                         if (!dst_vxlan)
2034                                 goto tx_error;
2035                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2036                         return;
2037                 }
2038
2039                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2040                 md.vni = htonl(vni << 8);
2041                 md.gbp = skb->mark;
2042
2043                 err = vxlan6_xmit_skb(vxlan->vn_sock, ndst, skb,
2044                                       dev, &fl6.saddr, &fl6.daddr, 0, ttl,
2045                                       src_port, dst_port, &md,
2046                                       !net_eq(vxlan->net, dev_net(vxlan->dev)));
2047 #endif
2048         }
2049
2050         return;
2051
2052 drop:
2053         dev->stats.tx_dropped++;
2054         goto tx_free;
2055
2056 rt_tx_error:
2057         ip_rt_put(rt);
2058 tx_error:
2059         dev->stats.tx_errors++;
2060 tx_free:
2061         dev_kfree_skb(skb);
2062 }
2063
2064 /* Transmit local packets over Vxlan
2065  *
2066  * Outer IP header inherits ECN and DF from inner header.
2067  * Outer UDP destination is the VXLAN assigned port.
2068  *           source port is based on hash of flow
2069  */
2070 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2071 {
2072         struct vxlan_dev *vxlan = netdev_priv(dev);
2073         struct ethhdr *eth;
2074         bool did_rsc = false;
2075         struct vxlan_rdst *rdst, *fdst = NULL;
2076         struct vxlan_fdb *f;
2077
2078         skb_reset_mac_header(skb);
2079         eth = eth_hdr(skb);
2080
2081         if ((vxlan->flags & VXLAN_F_PROXY)) {
2082                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2083                         return arp_reduce(dev, skb);
2084 #if IS_ENABLED(CONFIG_IPV6)
2085                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2086                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2087                                        + sizeof(struct nd_msg)) &&
2088                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2089                                 struct nd_msg *msg;
2090
2091                                 msg = (struct nd_msg *)skb_transport_header(skb);
2092                                 if (msg->icmph.icmp6_code == 0 &&
2093                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2094                                         return neigh_reduce(dev, skb);
2095                 }
2096                 eth = eth_hdr(skb);
2097 #endif
2098         }
2099
2100         f = vxlan_find_mac(vxlan, eth->h_dest);
2101         did_rsc = false;
2102
2103         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2104             (ntohs(eth->h_proto) == ETH_P_IP ||
2105              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2106                 did_rsc = route_shortcircuit(dev, skb);
2107                 if (did_rsc)
2108                         f = vxlan_find_mac(vxlan, eth->h_dest);
2109         }
2110
2111         if (f == NULL) {
2112                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2113                 if (f == NULL) {
2114                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2115                             !is_multicast_ether_addr(eth->h_dest))
2116                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2117
2118                         dev->stats.tx_dropped++;
2119                         kfree_skb(skb);
2120                         return NETDEV_TX_OK;
2121                 }
2122         }
2123
2124         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2125                 struct sk_buff *skb1;
2126
2127                 if (!fdst) {
2128                         fdst = rdst;
2129                         continue;
2130                 }
2131                 skb1 = skb_clone(skb, GFP_ATOMIC);
2132                 if (skb1)
2133                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2134         }
2135
2136         if (fdst)
2137                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2138         else
2139                 kfree_skb(skb);
2140         return NETDEV_TX_OK;
2141 }
2142
2143 /* Walk the forwarding table and purge stale entries */
2144 static void vxlan_cleanup(unsigned long arg)
2145 {
2146         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2147         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2148         unsigned int h;
2149
2150         if (!netif_running(vxlan->dev))
2151                 return;
2152
2153         spin_lock_bh(&vxlan->hash_lock);
2154         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2155                 struct hlist_node *p, *n;
2156                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2157                         struct vxlan_fdb *f
2158                                 = container_of(p, struct vxlan_fdb, hlist);
2159                         unsigned long timeout;
2160
2161                         if (f->state & NUD_PERMANENT)
2162                                 continue;
2163
2164                         timeout = f->used + vxlan->age_interval * HZ;
2165                         if (time_before_eq(timeout, jiffies)) {
2166                                 netdev_dbg(vxlan->dev,
2167                                            "garbage collect %pM\n",
2168                                            f->eth_addr);
2169                                 f->state = NUD_STALE;
2170                                 vxlan_fdb_destroy(vxlan, f);
2171                         } else if (time_before(timeout, next_timer))
2172                                 next_timer = timeout;
2173                 }
2174         }
2175         spin_unlock_bh(&vxlan->hash_lock);
2176
2177         mod_timer(&vxlan->age_timer, next_timer);
2178 }
2179
2180 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2181 {
2182         __u32 vni = vxlan->default_dst.remote_vni;
2183
2184         vxlan->vn_sock = vs;
2185         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2186 }
2187
2188 /* Setup stats when device is created */
2189 static int vxlan_init(struct net_device *dev)
2190 {
2191         struct vxlan_dev *vxlan = netdev_priv(dev);
2192         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2193         struct vxlan_sock *vs;
2194         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2195
2196         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2197         if (!dev->tstats)
2198                 return -ENOMEM;
2199
2200         spin_lock(&vn->sock_lock);
2201         vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2202                              vxlan->dst_port, vxlan->flags);
2203         if (vs && atomic_add_unless(&vs->refcnt, 1, 0)) {
2204                 /* If we have a socket with same port already, reuse it */
2205                 vxlan_vs_add_dev(vs, vxlan);
2206         } else {
2207                 /* otherwise make new socket outside of RTNL */
2208                 dev_hold(dev);
2209                 queue_work(vxlan_wq, &vxlan->sock_work);
2210         }
2211         spin_unlock(&vn->sock_lock);
2212
2213         return 0;
2214 }
2215
2216 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2217 {
2218         struct vxlan_fdb *f;
2219
2220         spin_lock_bh(&vxlan->hash_lock);
2221         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2222         if (f)
2223                 vxlan_fdb_destroy(vxlan, f);
2224         spin_unlock_bh(&vxlan->hash_lock);
2225 }
2226
2227 static void vxlan_uninit(struct net_device *dev)
2228 {
2229         struct vxlan_dev *vxlan = netdev_priv(dev);
2230         struct vxlan_sock *vs = vxlan->vn_sock;
2231
2232         vxlan_fdb_delete_default(vxlan);
2233
2234         if (vs)
2235                 vxlan_sock_release(vs);
2236         free_percpu(dev->tstats);
2237 }
2238
2239 /* Start ageing timer and join group when device is brought up */
2240 static int vxlan_open(struct net_device *dev)
2241 {
2242         struct vxlan_dev *vxlan = netdev_priv(dev);
2243         struct vxlan_sock *vs = vxlan->vn_sock;
2244
2245         /* socket hasn't been created */
2246         if (!vs)
2247                 return -ENOTCONN;
2248
2249         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2250                 vxlan_sock_hold(vs);
2251                 dev_hold(dev);
2252                 queue_work(vxlan_wq, &vxlan->igmp_join);
2253         }
2254
2255         if (vxlan->age_interval)
2256                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2257
2258         return 0;
2259 }
2260
2261 /* Purge the forwarding table */
2262 static void vxlan_flush(struct vxlan_dev *vxlan)
2263 {
2264         unsigned int h;
2265
2266         spin_lock_bh(&vxlan->hash_lock);
2267         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2268                 struct hlist_node *p, *n;
2269                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2270                         struct vxlan_fdb *f
2271                                 = container_of(p, struct vxlan_fdb, hlist);
2272                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2273                         if (!is_zero_ether_addr(f->eth_addr))
2274                                 vxlan_fdb_destroy(vxlan, f);
2275                 }
2276         }
2277         spin_unlock_bh(&vxlan->hash_lock);
2278 }
2279
2280 /* Cleanup timer and forwarding table on shutdown */
2281 static int vxlan_stop(struct net_device *dev)
2282 {
2283         struct vxlan_dev *vxlan = netdev_priv(dev);
2284         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2285         struct vxlan_sock *vs = vxlan->vn_sock;
2286
2287         if (vs && vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2288             !vxlan_group_used(vn, vxlan)) {
2289                 vxlan_sock_hold(vs);
2290                 dev_hold(dev);
2291                 queue_work(vxlan_wq, &vxlan->igmp_leave);
2292         }
2293
2294         del_timer_sync(&vxlan->age_timer);
2295
2296         vxlan_flush(vxlan);
2297
2298         return 0;
2299 }
2300
2301 /* Stub, nothing needs to be done. */
2302 static void vxlan_set_multicast_list(struct net_device *dev)
2303 {
2304 }
2305
2306 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2307 {
2308         struct vxlan_dev *vxlan = netdev_priv(dev);
2309         struct vxlan_rdst *dst = &vxlan->default_dst;
2310         struct net_device *lowerdev;
2311         int max_mtu;
2312
2313         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2314         if (lowerdev == NULL)
2315                 return eth_change_mtu(dev, new_mtu);
2316
2317         if (dst->remote_ip.sa.sa_family == AF_INET6)
2318                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2319         else
2320                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2321
2322         if (new_mtu < 68 || new_mtu > max_mtu)
2323                 return -EINVAL;
2324
2325         dev->mtu = new_mtu;
2326         return 0;
2327 }
2328
2329 static const struct net_device_ops vxlan_netdev_ops = {
2330         .ndo_init               = vxlan_init,
2331         .ndo_uninit             = vxlan_uninit,
2332         .ndo_open               = vxlan_open,
2333         .ndo_stop               = vxlan_stop,
2334         .ndo_start_xmit         = vxlan_xmit,
2335         .ndo_get_stats64        = ip_tunnel_get_stats64,
2336         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2337         .ndo_change_mtu         = vxlan_change_mtu,
2338         .ndo_validate_addr      = eth_validate_addr,
2339         .ndo_set_mac_address    = eth_mac_addr,
2340         .ndo_fdb_add            = vxlan_fdb_add,
2341         .ndo_fdb_del            = vxlan_fdb_delete,
2342         .ndo_fdb_dump           = vxlan_fdb_dump,
2343 };
2344
2345 /* Info for udev, that this is a virtual tunnel endpoint */
2346 static struct device_type vxlan_type = {
2347         .name = "vxlan",
2348 };
2349
2350 /* Calls the ndo_add_vxlan_port of the caller in order to
2351  * supply the listening VXLAN udp ports. Callers are expected
2352  * to implement the ndo_add_vxlan_port.
2353  */
2354 void vxlan_get_rx_port(struct net_device *dev)
2355 {
2356         struct vxlan_sock *vs;
2357         struct net *net = dev_net(dev);
2358         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2359         sa_family_t sa_family;
2360         __be16 port;
2361         unsigned int i;
2362
2363         spin_lock(&vn->sock_lock);
2364         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2365                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2366                         port = inet_sk(vs->sock->sk)->inet_sport;
2367                         sa_family = vs->sock->sk->sk_family;
2368                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2369                                                             port);
2370                 }
2371         }
2372         spin_unlock(&vn->sock_lock);
2373 }
2374 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2375
2376 /* Initialize the device structure. */
2377 static void vxlan_setup(struct net_device *dev)
2378 {
2379         struct vxlan_dev *vxlan = netdev_priv(dev);
2380         unsigned int h;
2381
2382         eth_hw_addr_random(dev);
2383         ether_setup(dev);
2384         if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2385                 dev->needed_headroom = ETH_HLEN + VXLAN6_HEADROOM;
2386         else
2387                 dev->needed_headroom = ETH_HLEN + VXLAN_HEADROOM;
2388
2389         dev->netdev_ops = &vxlan_netdev_ops;
2390         dev->destructor = free_netdev;
2391         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2392
2393         dev->tx_queue_len = 0;
2394         dev->features   |= NETIF_F_LLTX;
2395         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2396         dev->features   |= NETIF_F_RXCSUM;
2397         dev->features   |= NETIF_F_GSO_SOFTWARE;
2398
2399         dev->vlan_features = dev->features;
2400         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2401         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2402         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2403         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2404         netif_keep_dst(dev);
2405         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2406
2407         INIT_LIST_HEAD(&vxlan->next);
2408         spin_lock_init(&vxlan->hash_lock);
2409         INIT_WORK(&vxlan->igmp_join, vxlan_igmp_join);
2410         INIT_WORK(&vxlan->igmp_leave, vxlan_igmp_leave);
2411         INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
2412
2413         init_timer_deferrable(&vxlan->age_timer);
2414         vxlan->age_timer.function = vxlan_cleanup;
2415         vxlan->age_timer.data = (unsigned long) vxlan;
2416
2417         vxlan->dst_port = htons(vxlan_port);
2418
2419         vxlan->dev = dev;
2420
2421         for (h = 0; h < FDB_HASH_SIZE; ++h)
2422                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2423 }
2424
2425 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2426         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2427         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2428         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2429         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2430         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2431         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2432         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2433         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2434         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2435         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2436         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2437         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2438         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2439         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2440         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2441         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2442         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2443         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2444         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2445         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2446         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2447         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2448         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2449 };
2450
2451 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2452 {
2453         if (tb[IFLA_ADDRESS]) {
2454                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2455                         pr_debug("invalid link address (not ethernet)\n");
2456                         return -EINVAL;
2457                 }
2458
2459                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2460                         pr_debug("invalid all zero ethernet address\n");
2461                         return -EADDRNOTAVAIL;
2462                 }
2463         }
2464
2465         if (!data)
2466                 return -EINVAL;
2467
2468         if (data[IFLA_VXLAN_ID]) {
2469                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2470                 if (id >= VXLAN_VID_MASK)
2471                         return -ERANGE;
2472         }
2473
2474         if (data[IFLA_VXLAN_PORT_RANGE]) {
2475                 const struct ifla_vxlan_port_range *p
2476                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2477
2478                 if (ntohs(p->high) < ntohs(p->low)) {
2479                         pr_debug("port range %u .. %u not valid\n",
2480                                  ntohs(p->low), ntohs(p->high));
2481                         return -EINVAL;
2482                 }
2483         }
2484
2485         return 0;
2486 }
2487
2488 static void vxlan_get_drvinfo(struct net_device *netdev,
2489                               struct ethtool_drvinfo *drvinfo)
2490 {
2491         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2492         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2493 }
2494
2495 static const struct ethtool_ops vxlan_ethtool_ops = {
2496         .get_drvinfo    = vxlan_get_drvinfo,
2497         .get_link       = ethtool_op_get_link,
2498 };
2499
2500 static void vxlan_del_work(struct work_struct *work)
2501 {
2502         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2503         udp_tunnel_sock_release(vs->sock);
2504         kfree_rcu(vs, rcu);
2505 }
2506
2507 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2508                                         __be16 port, u32 flags)
2509 {
2510         struct socket *sock;
2511         struct udp_port_cfg udp_conf;
2512         int err;
2513
2514         memset(&udp_conf, 0, sizeof(udp_conf));
2515
2516         if (ipv6) {
2517                 udp_conf.family = AF_INET6;
2518                 udp_conf.use_udp6_tx_checksums =
2519                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2520                 udp_conf.use_udp6_rx_checksums =
2521                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2522         } else {
2523                 udp_conf.family = AF_INET;
2524                 udp_conf.local_ip.s_addr = INADDR_ANY;
2525                 udp_conf.use_udp_checksums =
2526                     !!(flags & VXLAN_F_UDP_CSUM);
2527         }
2528
2529         udp_conf.local_udp_port = port;
2530
2531         /* Open UDP socket */
2532         err = udp_sock_create(net, &udp_conf, &sock);
2533         if (err < 0)
2534                 return ERR_PTR(err);
2535
2536         return sock;
2537 }
2538
2539 /* Create new listen socket if needed */
2540 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2541                                               vxlan_rcv_t *rcv, void *data,
2542                                               u32 flags)
2543 {
2544         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2545         struct vxlan_sock *vs;
2546         struct socket *sock;
2547         unsigned int h;
2548         bool ipv6 = !!(flags & VXLAN_F_IPV6);
2549         struct udp_tunnel_sock_cfg tunnel_cfg;
2550
2551         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2552         if (!vs)
2553                 return ERR_PTR(-ENOMEM);
2554
2555         for (h = 0; h < VNI_HASH_SIZE; ++h)
2556                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2557
2558         INIT_WORK(&vs->del_work, vxlan_del_work);
2559
2560         sock = vxlan_create_sock(net, ipv6, port, flags);
2561         if (IS_ERR(sock)) {
2562                 kfree(vs);
2563                 return ERR_CAST(sock);
2564         }
2565
2566         vs->sock = sock;
2567         atomic_set(&vs->refcnt, 1);
2568         vs->rcv = rcv;
2569         vs->data = data;
2570         vs->flags = flags;
2571
2572         /* Initialize the vxlan udp offloads structure */
2573         vs->udp_offloads.port = port;
2574         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2575         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2576
2577         spin_lock(&vn->sock_lock);
2578         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2579         vxlan_notify_add_rx_port(vs);
2580         spin_unlock(&vn->sock_lock);
2581
2582         /* Mark socket as an encapsulation socket. */
2583         tunnel_cfg.sk_user_data = vs;
2584         tunnel_cfg.encap_type = 1;
2585         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2586         tunnel_cfg.encap_destroy = NULL;
2587
2588         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2589
2590         return vs;
2591 }
2592
2593 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2594                                   vxlan_rcv_t *rcv, void *data,
2595                                   bool no_share, u32 flags)
2596 {
2597         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2598         struct vxlan_sock *vs;
2599         bool ipv6 = flags & VXLAN_F_IPV6;
2600
2601         vs = vxlan_socket_create(net, port, rcv, data, flags);
2602         if (!IS_ERR(vs))
2603                 return vs;
2604
2605         if (no_share)   /* Return error if sharing is not allowed. */
2606                 return vs;
2607
2608         spin_lock(&vn->sock_lock);
2609         vs = vxlan_find_sock(net, ipv6 ? AF_INET6 : AF_INET, port, flags);
2610         if (vs && ((vs->rcv != rcv) ||
2611                    !atomic_add_unless(&vs->refcnt, 1, 0)))
2612                         vs = ERR_PTR(-EBUSY);
2613         spin_unlock(&vn->sock_lock);
2614
2615         if (!vs)
2616                 vs = ERR_PTR(-EINVAL);
2617
2618         return vs;
2619 }
2620 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2621
2622 /* Scheduled at device creation to bind to a socket */
2623 static void vxlan_sock_work(struct work_struct *work)
2624 {
2625         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, sock_work);
2626         struct net *net = vxlan->net;
2627         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2628         __be16 port = vxlan->dst_port;
2629         struct vxlan_sock *nvs;
2630
2631         nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags);
2632         spin_lock(&vn->sock_lock);
2633         if (!IS_ERR(nvs))
2634                 vxlan_vs_add_dev(nvs, vxlan);
2635         spin_unlock(&vn->sock_lock);
2636
2637         dev_put(vxlan->dev);
2638 }
2639
2640 static int vxlan_newlink(struct net *net, struct net_device *dev,
2641                          struct nlattr *tb[], struct nlattr *data[])
2642 {
2643         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2644         struct vxlan_dev *vxlan = netdev_priv(dev);
2645         struct vxlan_rdst *dst = &vxlan->default_dst;
2646         __u32 vni;
2647         int err;
2648         bool use_ipv6 = false;
2649
2650         if (!data[IFLA_VXLAN_ID])
2651                 return -EINVAL;
2652
2653         vxlan->net = dev_net(dev);
2654
2655         vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2656         dst->remote_vni = vni;
2657
2658         /* Unless IPv6 is explicitly requested, assume IPv4 */
2659         dst->remote_ip.sa.sa_family = AF_INET;
2660         if (data[IFLA_VXLAN_GROUP]) {
2661                 dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
2662         } else if (data[IFLA_VXLAN_GROUP6]) {
2663                 if (!IS_ENABLED(CONFIG_IPV6))
2664                         return -EPFNOSUPPORT;
2665
2666                 nla_memcpy(&dst->remote_ip.sin6.sin6_addr, data[IFLA_VXLAN_GROUP6],
2667                            sizeof(struct in6_addr));
2668                 dst->remote_ip.sa.sa_family = AF_INET6;
2669                 use_ipv6 = true;
2670         }
2671
2672         if (data[IFLA_VXLAN_LOCAL]) {
2673                 vxlan->saddr.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
2674                 vxlan->saddr.sa.sa_family = AF_INET;
2675         } else if (data[IFLA_VXLAN_LOCAL6]) {
2676                 if (!IS_ENABLED(CONFIG_IPV6))
2677                         return -EPFNOSUPPORT;
2678
2679                 /* TODO: respect scope id */
2680                 nla_memcpy(&vxlan->saddr.sin6.sin6_addr, data[IFLA_VXLAN_LOCAL6],
2681                            sizeof(struct in6_addr));
2682                 vxlan->saddr.sa.sa_family = AF_INET6;
2683                 use_ipv6 = true;
2684         }
2685
2686         if (data[IFLA_VXLAN_LINK] &&
2687             (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2688                 struct net_device *lowerdev
2689                          = __dev_get_by_index(net, dst->remote_ifindex);
2690
2691                 if (!lowerdev) {
2692                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2693                         return -ENODEV;
2694                 }
2695
2696 #if IS_ENABLED(CONFIG_IPV6)
2697                 if (use_ipv6) {
2698                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2699                         if (idev && idev->cnf.disable_ipv6) {
2700                                 pr_info("IPv6 is disabled via sysctl\n");
2701                                 return -EPERM;
2702                         }
2703                         vxlan->flags |= VXLAN_F_IPV6;
2704                 }
2705 #endif
2706
2707                 if (!tb[IFLA_MTU])
2708                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2709
2710                 dev->needed_headroom = lowerdev->hard_header_len +
2711                                        (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2712         } else if (use_ipv6)
2713                 vxlan->flags |= VXLAN_F_IPV6;
2714
2715         if (data[IFLA_VXLAN_TOS])
2716                 vxlan->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2717
2718         if (data[IFLA_VXLAN_TTL])
2719                 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2720
2721         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2722                 vxlan->flags |= VXLAN_F_LEARN;
2723
2724         if (data[IFLA_VXLAN_AGEING])
2725                 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2726         else
2727                 vxlan->age_interval = FDB_AGE_DEFAULT;
2728
2729         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2730                 vxlan->flags |= VXLAN_F_PROXY;
2731
2732         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2733                 vxlan->flags |= VXLAN_F_RSC;
2734
2735         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2736                 vxlan->flags |= VXLAN_F_L2MISS;
2737
2738         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2739                 vxlan->flags |= VXLAN_F_L3MISS;
2740
2741         if (data[IFLA_VXLAN_LIMIT])
2742                 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2743
2744         if (data[IFLA_VXLAN_PORT_RANGE]) {
2745                 const struct ifla_vxlan_port_range *p
2746                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2747                 vxlan->port_min = ntohs(p->low);
2748                 vxlan->port_max = ntohs(p->high);
2749         }
2750
2751         if (data[IFLA_VXLAN_PORT])
2752                 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2753
2754         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2755                 vxlan->flags |= VXLAN_F_UDP_CSUM;
2756
2757         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2758             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2759                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2760
2761         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2762             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2763                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2764
2765         if (data[IFLA_VXLAN_REMCSUM_TX] &&
2766             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2767                 vxlan->flags |= VXLAN_F_REMCSUM_TX;
2768
2769         if (data[IFLA_VXLAN_REMCSUM_RX] &&
2770             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2771                 vxlan->flags |= VXLAN_F_REMCSUM_RX;
2772
2773         if (data[IFLA_VXLAN_GBP])
2774                 vxlan->flags |= VXLAN_F_GBP;
2775
2776         if (vxlan_find_vni(net, vni, use_ipv6 ? AF_INET6 : AF_INET,
2777                            vxlan->dst_port, vxlan->flags)) {
2778                 pr_info("duplicate VNI %u\n", vni);
2779                 return -EEXIST;
2780         }
2781
2782         dev->ethtool_ops = &vxlan_ethtool_ops;
2783
2784         /* create an fdb entry for a valid default destination */
2785         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2786                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2787                                        &vxlan->default_dst.remote_ip,
2788                                        NUD_REACHABLE|NUD_PERMANENT,
2789                                        NLM_F_EXCL|NLM_F_CREATE,
2790                                        vxlan->dst_port,
2791                                        vxlan->default_dst.remote_vni,
2792                                        vxlan->default_dst.remote_ifindex,
2793                                        NTF_SELF);
2794                 if (err)
2795                         return err;
2796         }
2797
2798         err = register_netdevice(dev);
2799         if (err) {
2800                 vxlan_fdb_delete_default(vxlan);
2801                 return err;
2802         }
2803
2804         list_add(&vxlan->next, &vn->vxlan_list);
2805
2806         return 0;
2807 }
2808
2809 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2810 {
2811         struct vxlan_dev *vxlan = netdev_priv(dev);
2812         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2813
2814         spin_lock(&vn->sock_lock);
2815         if (!hlist_unhashed(&vxlan->hlist))
2816                 hlist_del_rcu(&vxlan->hlist);
2817         spin_unlock(&vn->sock_lock);
2818
2819         list_del(&vxlan->next);
2820         unregister_netdevice_queue(dev, head);
2821 }
2822
2823 static size_t vxlan_get_size(const struct net_device *dev)
2824 {
2825
2826         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2827                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2828                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2829                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2830                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2831                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2832                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2833                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2834                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2835                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2836                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2837                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2838                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2839                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2840                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2841                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2842                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2843                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2844                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
2845                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
2846                 0;
2847 }
2848
2849 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2850 {
2851         const struct vxlan_dev *vxlan = netdev_priv(dev);
2852         const struct vxlan_rdst *dst = &vxlan->default_dst;
2853         struct ifla_vxlan_port_range ports = {
2854                 .low =  htons(vxlan->port_min),
2855                 .high = htons(vxlan->port_max),
2856         };
2857
2858         if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2859                 goto nla_put_failure;
2860
2861         if (!vxlan_addr_any(&dst->remote_ip)) {
2862                 if (dst->remote_ip.sa.sa_family == AF_INET) {
2863                         if (nla_put_be32(skb, IFLA_VXLAN_GROUP,
2864                                          dst->remote_ip.sin.sin_addr.s_addr))
2865                                 goto nla_put_failure;
2866 #if IS_ENABLED(CONFIG_IPV6)
2867                 } else {
2868                         if (nla_put(skb, IFLA_VXLAN_GROUP6, sizeof(struct in6_addr),
2869                                     &dst->remote_ip.sin6.sin6_addr))
2870                                 goto nla_put_failure;
2871 #endif
2872                 }
2873         }
2874
2875         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2876                 goto nla_put_failure;
2877
2878         if (!vxlan_addr_any(&vxlan->saddr)) {
2879                 if (vxlan->saddr.sa.sa_family == AF_INET) {
2880                         if (nla_put_be32(skb, IFLA_VXLAN_LOCAL,
2881                                          vxlan->saddr.sin.sin_addr.s_addr))
2882                                 goto nla_put_failure;
2883 #if IS_ENABLED(CONFIG_IPV6)
2884                 } else {
2885                         if (nla_put(skb, IFLA_VXLAN_LOCAL6, sizeof(struct in6_addr),
2886                                     &vxlan->saddr.sin6.sin6_addr))
2887                                 goto nla_put_failure;
2888 #endif
2889                 }
2890         }
2891
2892         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2893             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2894             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2895                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
2896             nla_put_u8(skb, IFLA_VXLAN_PROXY,
2897                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
2898             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2899             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2900                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2901             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2902                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2903             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2904             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2905             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port) ||
2906             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2907                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2908             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2909                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2910             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2911                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
2912             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
2913                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
2914             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
2915                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
2916                 goto nla_put_failure;
2917
2918         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2919                 goto nla_put_failure;
2920
2921         if (vxlan->flags & VXLAN_F_GBP &&
2922             nla_put_flag(skb, IFLA_VXLAN_GBP))
2923                 goto nla_put_failure;
2924
2925         return 0;
2926
2927 nla_put_failure:
2928         return -EMSGSIZE;
2929 }
2930
2931 static struct net *vxlan_get_link_net(const struct net_device *dev)
2932 {
2933         struct vxlan_dev *vxlan = netdev_priv(dev);
2934
2935         return vxlan->net;
2936 }
2937
2938 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2939         .kind           = "vxlan",
2940         .maxtype        = IFLA_VXLAN_MAX,
2941         .policy         = vxlan_policy,
2942         .priv_size      = sizeof(struct vxlan_dev),
2943         .setup          = vxlan_setup,
2944         .validate       = vxlan_validate,
2945         .newlink        = vxlan_newlink,
2946         .dellink        = vxlan_dellink,
2947         .get_size       = vxlan_get_size,
2948         .fill_info      = vxlan_fill_info,
2949         .get_link_net   = vxlan_get_link_net,
2950 };
2951
2952 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
2953                                              struct net_device *dev)
2954 {
2955         struct vxlan_dev *vxlan, *next;
2956         LIST_HEAD(list_kill);
2957
2958         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2959                 struct vxlan_rdst *dst = &vxlan->default_dst;
2960
2961                 /* In case we created vxlan device with carrier
2962                  * and we loose the carrier due to module unload
2963                  * we also need to remove vxlan device. In other
2964                  * cases, it's not necessary and remote_ifindex
2965                  * is 0 here, so no matches.
2966                  */
2967                 if (dst->remote_ifindex == dev->ifindex)
2968                         vxlan_dellink(vxlan->dev, &list_kill);
2969         }
2970
2971         unregister_netdevice_many(&list_kill);
2972 }
2973
2974 static int vxlan_lowerdev_event(struct notifier_block *unused,
2975                                 unsigned long event, void *ptr)
2976 {
2977         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2978         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2979
2980         if (event == NETDEV_UNREGISTER)
2981                 vxlan_handle_lowerdev_unregister(vn, dev);
2982
2983         return NOTIFY_DONE;
2984 }
2985
2986 static struct notifier_block vxlan_notifier_block __read_mostly = {
2987         .notifier_call = vxlan_lowerdev_event,
2988 };
2989
2990 static __net_init int vxlan_init_net(struct net *net)
2991 {
2992         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2993         unsigned int h;
2994
2995         INIT_LIST_HEAD(&vn->vxlan_list);
2996         spin_lock_init(&vn->sock_lock);
2997
2998         for (h = 0; h < PORT_HASH_SIZE; ++h)
2999                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3000
3001         return 0;
3002 }
3003
3004 static void __net_exit vxlan_exit_net(struct net *net)
3005 {
3006         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3007         struct vxlan_dev *vxlan, *next;
3008         struct net_device *dev, *aux;
3009         LIST_HEAD(list);
3010
3011         rtnl_lock();
3012         for_each_netdev_safe(net, dev, aux)
3013                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3014                         unregister_netdevice_queue(dev, &list);
3015
3016         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3017                 /* If vxlan->dev is in the same netns, it has already been added
3018                  * to the list by the previous loop.
3019                  */
3020                 if (!net_eq(dev_net(vxlan->dev), net))
3021                         unregister_netdevice_queue(dev, &list);
3022         }
3023
3024         unregister_netdevice_many(&list);
3025         rtnl_unlock();
3026 }
3027
3028 static struct pernet_operations vxlan_net_ops = {
3029         .init = vxlan_init_net,
3030         .exit = vxlan_exit_net,
3031         .id   = &vxlan_net_id,
3032         .size = sizeof(struct vxlan_net),
3033 };
3034
3035 static int __init vxlan_init_module(void)
3036 {
3037         int rc;
3038
3039         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3040         if (!vxlan_wq)
3041                 return -ENOMEM;
3042
3043         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3044
3045         rc = register_pernet_subsys(&vxlan_net_ops);
3046         if (rc)
3047                 goto out1;
3048
3049         rc = register_netdevice_notifier(&vxlan_notifier_block);
3050         if (rc)
3051                 goto out2;
3052
3053         rc = rtnl_link_register(&vxlan_link_ops);
3054         if (rc)
3055                 goto out3;
3056
3057         return 0;
3058 out3:
3059         unregister_netdevice_notifier(&vxlan_notifier_block);
3060 out2:
3061         unregister_pernet_subsys(&vxlan_net_ops);
3062 out1:
3063         destroy_workqueue(vxlan_wq);
3064         return rc;
3065 }
3066 late_initcall(vxlan_init_module);
3067
3068 static void __exit vxlan_cleanup_module(void)
3069 {
3070         rtnl_link_unregister(&vxlan_link_ops);
3071         unregister_netdevice_notifier(&vxlan_notifier_block);
3072         destroy_workqueue(vxlan_wq);
3073         unregister_pernet_subsys(&vxlan_net_ops);
3074         /* rcu_barrier() is called by netns */
3075 }
3076 module_exit(vxlan_cleanup_module);
3077
3078 MODULE_LICENSE("GPL");
3079 MODULE_VERSION(VXLAN_VERSION);
3080 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3081 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3082 MODULE_ALIAS_RTNL_LINK("vxlan");