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
 274         flags &= VXLAN_F_RCV_FLAGS;
 275
 276         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 277                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 278                     inet_sk(vs->sock->sk)->sk.sk_family == family &&
 279                     vs->flags == flags)
 280                         return vs;
 281         }
 282         return NULL;
 283 }
 284
 285 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
 286 {
 287         struct vxlan_dev *vxlan;
 288
 289         hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
 290                 if (vxlan->default_dst.remote_vni == id)
 291                         return vxlan;
 292         }
 293
 294         return NULL;
 295 }
 296
 297 /* Look up VNI in a per net namespace table */
 298 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id,
 299                                         sa_family_t family, __be16 port,
 300                                         u32 flags)
 301 {
 302         struct vxlan_sock *vs;
 303
 304         vs = vxlan_find_sock(net, family, port, flags);
 305         if (!vs)
 306                 return NULL;
 307
 308         return vxlan_vs_find_vni(vs, id);
 309 }
 310
 311 /* Fill in neighbour message in skbuff. */
 312 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 313                           const struct vxlan_fdb *fdb,
 314                           u32 portid, u32 seq, int type, unsigned int flags,
 315                           const struct vxlan_rdst *rdst)
 316 {
 317         unsigned long now = jiffies;
 318         struct nda_cacheinfo ci;
 319         struct nlmsghdr *nlh;
 320         struct ndmsg *ndm;
 321         bool send_ip, send_eth;
 322
 323         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 324         if (nlh == NULL)
 325                 return -EMSGSIZE;
 326
 327         ndm = nlmsg_data(nlh);
 328         memset(ndm, 0, sizeof(*ndm));
 329
 330         send_eth = send_ip = true;
 331
 332         if (type == RTM_GETNEIGH) {
 333                 ndm->ndm_family = AF_INET;
 334                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 335                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 336         } else
 337                 ndm->ndm_family = AF_BRIDGE;
 338         ndm->ndm_state = fdb->state;
 339         ndm->ndm_ifindex = vxlan->dev->ifindex;
 340         ndm->ndm_flags = fdb->flags;
 341         ndm->ndm_type = RTN_UNICAST;
 342
 343         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 344             nla_put_s32(skb, NDA_LINK_NETNSID,
 345                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 346                 goto nla_put_failure;
 347
 348         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 349                 goto nla_put_failure;
 350
 351         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 352                 goto nla_put_failure;
 353
 354         if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
 355             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 356                 goto nla_put_failure;
 357         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 358             nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
 359                 goto nla_put_failure;
 360         if (rdst->remote_ifindex &&
 361             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 362                 goto nla_put_failure;
 363
 364         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 365         ci.ndm_confirmed = 0;
 366         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 367         ci.ndm_refcnt    = 0;
 368
 369         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 370                 goto nla_put_failure;
 371
 372         nlmsg_end(skb, nlh);
 373         return 0;
 374
 375 nla_put_failure:
 376         nlmsg_cancel(skb, nlh);
 377         return -EMSGSIZE;
 378 }
 379
 380 static inline size_t vxlan_nlmsg_size(void)
 381 {
 382         return NLMSG_ALIGN(sizeof(struct ndmsg))
 383                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 384                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 385                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 386                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 387                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 388                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 389                 + nla_total_size(sizeof(struct nda_cacheinfo));
 390 }
 391
 392 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 393                              struct vxlan_rdst *rd, int type)
 394 {
 395         struct net *net = dev_net(vxlan->dev);
 396         struct sk_buff *skb;
 397         int err = -ENOBUFS;
 398
 399         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 400         if (skb == NULL)
 401                 goto errout;
 402
 403         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 404         if (err < 0) {
 405                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 406                 WARN_ON(err == -EMSGSIZE);
 407                 kfree_skb(skb);
 408                 goto errout;
 409         }
 410
 411         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 412         return;
 413 errout:
 414         if (err < 0)
 415                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 416 }
 417
 418 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 419 {
 420         struct vxlan_dev *vxlan = netdev_priv(dev);
 421         struct vxlan_fdb f = {
 422                 .state = NUD_STALE,
 423         };
 424         struct vxlan_rdst remote = {
 425                 .remote_ip = *ipa, /* goes to NDA_DST */
 426                 .remote_vni = VXLAN_N_VID,
 427         };
 428
 429         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 430 }
 431
 432 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 433 {
 434         struct vxlan_fdb f = {
 435                 .state = NUD_STALE,
 436         };
 437         struct vxlan_rdst remote = { };
 438
 439         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 440
 441         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 442 }
 443
 444 /* Hash Ethernet address */
 445 static u32 eth_hash(const unsigned char *addr)
 446 {
 447         u64 value = get_unaligned((u64 *)addr);
 448
 449         /* only want 6 bytes */
 450 #ifdef __BIG_ENDIAN
 451         value >>= 16;
 452 #else
 453         value <<= 16;
 454 #endif
 455         return hash_64(value, FDB_HASH_BITS);
 456 }
 457
 458 /* Hash chain to use given mac address */
 459 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 460                                                 const u8 *mac)
 461 {
 462         return &vxlan->fdb_head[eth_hash(mac)];
 463 }
 464
 465 /* Look up Ethernet address in forwarding table */
 466 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 467                                         const u8 *mac)
 468 {
 469         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 470         struct vxlan_fdb *f;
 471
 472         hlist_for_each_entry_rcu(f, head, hlist) {
 473                 if (ether_addr_equal(mac, f->eth_addr))
 474                         return f;
 475         }
 476
 477         return NULL;
 478 }
 479
 480 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 481                                         const u8 *mac)
 482 {
 483         struct vxlan_fdb *f;
 484
 485         f = __vxlan_find_mac(vxlan, mac);
 486         if (f)
 487                 f->used = jiffies;
 488
 489         return f;
 490 }
 491
 492 /* caller should hold vxlan->hash_lock */
 493 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 494                                               union vxlan_addr *ip, __be16 port,
 495                                               __u32 vni, __u32 ifindex)
 496 {
 497         struct vxlan_rdst *rd;
 498
 499         list_for_each_entry(rd, &f->remotes, list) {
 500                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 501                     rd->remote_port == port &&
 502                     rd->remote_vni == vni &&
 503                     rd->remote_ifindex == ifindex)
 504                         return rd;
 505         }
 506
 507         return NULL;
 508 }
 509
 510 /* Replace destination of unicast mac */
 511 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 512                              union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
 513 {
 514         struct vxlan_rdst *rd;
 515
 516         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 517         if (rd)
 518                 return 0;
 519
 520         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 521         if (!rd)
 522                 return 0;
 523         rd->remote_ip = *ip;
 524         rd->remote_port = port;
 525         rd->remote_vni = vni;
 526         rd->remote_ifindex = ifindex;
 527         return 1;
 528 }
 529
 530 /* Add/update destinations for multicast */
 531 static int vxlan_fdb_append(struct vxlan_fdb *f,
 532                             union vxlan_addr *ip, __be16 port, __u32 vni,
 533                             __u32 ifindex, struct vxlan_rdst **rdp)
 534 {
 535         struct vxlan_rdst *rd;
 536
 537         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 538         if (rd)
 539                 return 0;
 540
 541         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 542         if (rd == NULL)
 543                 return -ENOBUFS;
 544         rd->remote_ip = *ip;
 545         rd->remote_port = port;
 546         rd->remote_vni = vni;
 547         rd->remote_ifindex = ifindex;
 548
 549         list_add_tail_rcu(&rd->list, &f->remotes);
 550
 551         *rdp = rd;
 552         return 1;
 553 }
 554
 555 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 556                                           unsigned int off,
 557                                           struct vxlanhdr *vh, size_t hdrlen,
 558                                           u32 data, struct gro_remcsum *grc,
 559                                           bool nopartial)
 560 {
 561         size_t start, offset, plen;
 562
 563         if (skb->remcsum_offload)
 564                 return NULL;
 565
 566         if (!NAPI_GRO_CB(skb)->csum_valid)
 567                 return NULL;
 568
 569         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
 570         offset = start + ((data & VXLAN_RCO_UDP) ?
 571                           offsetof(struct udphdr, check) :
 572                           offsetof(struct tcphdr, check));
 573
 574         plen = hdrlen + offset + sizeof(u16);
 575
 576         /* Pull checksum that will be written */
 577         if (skb_gro_header_hard(skb, off + plen)) {
 578                 vh = skb_gro_header_slow(skb, off + plen, off);
 579                 if (!vh)
 580                         return NULL;
 581         }
 582
 583         skb_gro_remcsum_process(skb, (void *)vh + hdrlen,
 584                                 start, offset, grc, nopartial);
 585
 586         skb->remcsum_offload = 1;
 587
 588         return vh;
 589 }
 590
 591 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
 592                                           struct sk_buff *skb,
 593                                           struct udp_offload *uoff)
 594 {
 595         struct sk_buff *p, **pp = NULL;
 596         struct vxlanhdr *vh, *vh2;
 597         unsigned int hlen, off_vx;
 598         int flush = 1;
 599         struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
 600                                              udp_offloads);
 601         u32 flags;
 602         struct gro_remcsum grc;
 603
 604         skb_gro_remcsum_init(&grc);
 605
 606         off_vx = skb_gro_offset(skb);
 607         hlen = off_vx + sizeof(*vh);
 608         vh   = skb_gro_header_fast(skb, off_vx);
 609         if (skb_gro_header_hard(skb, hlen)) {
 610                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 611                 if (unlikely(!vh))
 612                         goto out;
 613         }
 614
 615         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 616         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 617
 618         flags = ntohl(vh->vx_flags);
 619
 620         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 621                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 622                                        ntohl(vh->vx_vni), &grc,
 623                                        !!(vs->flags &
 624                                           VXLAN_F_REMCSUM_NOPARTIAL));
 625
 626                 if (!vh)
 627                         goto out;
 628         }
 629
 630         flush = 0;
 631
 632         for (p = *head; p; p = p->next) {
 633                 if (!NAPI_GRO_CB(p)->same_flow)
 634                         continue;
 635
 636                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 637                 if (vh->vx_flags != vh2->vx_flags ||
 638                     vh->vx_vni != vh2->vx_vni) {
 639                         NAPI_GRO_CB(p)->same_flow = 0;
 640                         continue;
 641                 }
 642         }
 643
 644         pp = eth_gro_receive(head, skb);
 645
 646 out:
 647         skb_gro_remcsum_cleanup(skb, &grc);
 648         NAPI_GRO_CB(skb)->flush |= flush;
 649
 650         return pp;
 651 }
 652
 653 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
 654                               struct udp_offload *uoff)
 655 {
 656         udp_tunnel_gro_complete(skb, nhoff);
 657
 658         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 659 }
 660
 661 /* Notify netdevs that UDP port started listening */
 662 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 663 {
 664         struct net_device *dev;
 665         struct sock *sk = vs->sock->sk;
 666         struct net *net = sock_net(sk);
 667         sa_family_t sa_family = sk->sk_family;
 668         __be16 port = inet_sk(sk)->inet_sport;
 669         int err;
 670
 671         if (sa_family == AF_INET) {
 672                 err = udp_add_offload(&vs->udp_offloads);
 673                 if (err)
 674                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 675         }
 676
 677         rcu_read_lock();
 678         for_each_netdev_rcu(net, dev) {
 679                 if (dev->netdev_ops->ndo_add_vxlan_port)
 680                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 681                                                             port);
 682         }
 683         rcu_read_unlock();
 684 }
 685
 686 /* Notify netdevs that UDP port is no more listening */
 687 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 688 {
 689         struct net_device *dev;
 690         struct sock *sk = vs->sock->sk;
 691         struct net *net = sock_net(sk);
 692         sa_family_t sa_family = sk->sk_family;
 693         __be16 port = inet_sk(sk)->inet_sport;
 694
 695         rcu_read_lock();
 696         for_each_netdev_rcu(net, dev) {
 697                 if (dev->netdev_ops->ndo_del_vxlan_port)
 698                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 699                                                             port);
 700         }
 701         rcu_read_unlock();
 702
 703         if (sa_family == AF_INET)
 704                 udp_del_offload(&vs->udp_offloads);
 705 }
 706
 707 /* Add new entry to forwarding table -- assumes lock held */
 708 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 709                             const u8 *mac, union vxlan_addr *ip,
 710                             __u16 state, __u16 flags,
 711                             __be16 port, __u32 vni, __u32 ifindex,
 712                             __u8 ndm_flags)
 713 {
 714         struct vxlan_rdst *rd = NULL;
 715         struct vxlan_fdb *f;
 716         int notify = 0;
 717
 718         f = __vxlan_find_mac(vxlan, mac);
 719         if (f) {
 720                 if (flags & NLM_F_EXCL) {
 721                         netdev_dbg(vxlan->dev,
 722                                    "lost race to create %pM\n", mac);
 723                         return -EEXIST;
 724                 }
 725                 if (f->state != state) {
 726                         f->state = state;
 727                         f->updated = jiffies;
 728                         notify = 1;
 729                 }
 730                 if (f->flags != ndm_flags) {
 731                         f->flags = ndm_flags;
 732                         f->updated = jiffies;
 733                         notify = 1;
 734                 }
 735                 if ((flags & NLM_F_REPLACE)) {
 736                         /* Only change unicasts */
 737                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 738                              is_zero_ether_addr(f->eth_addr))) {
 739                                 int rc = vxlan_fdb_replace(f, ip, port, vni,
 740                                                            ifindex);
 741
 742                                 if (rc < 0)
 743                                         return rc;
 744                                 notify |= rc;
 745                         } else
 746                                 return -EOPNOTSUPP;
 747                 }
 748                 if ((flags & NLM_F_APPEND) &&
 749                     (is_multicast_ether_addr(f->eth_addr) ||
 750                      is_zero_ether_addr(f->eth_addr))) {
 751                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 752                                                   &rd);
 753
 754                         if (rc < 0)
 755                                 return rc;
 756                         notify |= rc;
 757                 }
 758         } else {
 759                 if (!(flags & NLM_F_CREATE))
 760                         return -ENOENT;
 761
 762                 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
 763                         return -ENOSPC;
 764
 765                 /* Disallow replace to add a multicast entry */
 766                 if ((flags & NLM_F_REPLACE) &&
 767                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 768                         return -EOPNOTSUPP;
 769
 770                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 771                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 772                 if (!f)
 773                         return -ENOMEM;
 774
 775                 notify = 1;
 776                 f->state = state;
 777                 f->flags = ndm_flags;
 778                 f->updated = f->used = jiffies;
 779                 INIT_LIST_HEAD(&f->remotes);
 780                 memcpy(f->eth_addr, mac, ETH_ALEN);
 781
 782                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 783
 784                 ++vxlan->addrcnt;
 785                 hlist_add_head_rcu(&f->hlist,
 786                                    vxlan_fdb_head(vxlan, mac));
 787         }
 788
 789         if (notify) {
 790                 if (rd == NULL)
 791                         rd = first_remote_rtnl(f);
 792                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 793         }
 794
 795         return 0;
 796 }
 797
 798 static void vxlan_fdb_free(struct rcu_head *head)
 799 {
 800         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 801         struct vxlan_rdst *rd, *nd;
 802
 803         list_for_each_entry_safe(rd, nd, &f->remotes, list)
 804                 kfree(rd);
 805         kfree(f);
 806 }
 807
 808 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 809 {
 810         netdev_dbg(vxlan->dev,
 811                     "delete %pM\n", f->eth_addr);
 812
 813         --vxlan->addrcnt;
 814         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 815
 816         hlist_del_rcu(&f->hlist);
 817         call_rcu(&f->rcu, vxlan_fdb_free);
 818 }
 819
 820 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 821                            union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
 822 {
 823         struct net *net = dev_net(vxlan->dev);
 824         int err;
 825
 826         if (tb[NDA_DST]) {
 827                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 828                 if (err)
 829                         return err;
 830         } else {
 831                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 832                 if (remote->sa.sa_family == AF_INET) {
 833                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 834                         ip->sa.sa_family = AF_INET;
 835 #if IS_ENABLED(CONFIG_IPV6)
 836                 } else {
 837                         ip->sin6.sin6_addr = in6addr_any;
 838                         ip->sa.sa_family = AF_INET6;
 839 #endif
 840                 }
 841         }
 842
 843         if (tb[NDA_PORT]) {
 844                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 845                         return -EINVAL;
 846                 *port = nla_get_be16(tb[NDA_PORT]);
 847         } else {
 848                 *port = vxlan->dst_port;
 849         }
 850
 851         if (tb[NDA_VNI]) {
 852                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 853                         return -EINVAL;
 854                 *vni = nla_get_u32(tb[NDA_VNI]);
 855         } else {
 856                 *vni = vxlan->default_dst.remote_vni;
 857         }
 858
 859         if (tb[NDA_IFINDEX]) {
 860                 struct net_device *tdev;
 861
 862                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 863                         return -EINVAL;
 864                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 865                 tdev = __dev_get_by_index(net, *ifindex);
 866                 if (!tdev)
 867                         return -EADDRNOTAVAIL;
 868         } else {
 869                 *ifindex = 0;
 870         }
 871
 872         return 0;
 873 }
 874
 875 /* Add static entry (via netlink) */
 876 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 877                          struct net_device *dev,
 878                          const unsigned char *addr, u16 vid, u16 flags)
 879 {
 880         struct vxlan_dev *vxlan = netdev_priv(dev);
 881         /* struct net *net = dev_net(vxlan->dev); */
 882         union vxlan_addr ip;
 883         __be16 port;
 884         u32 vni, ifindex;
 885         int err;
 886
 887         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 888                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 889                         ndm->ndm_state);
 890                 return -EINVAL;
 891         }
 892
 893         if (tb[NDA_DST] == NULL)
 894                 return -EINVAL;
 895
 896         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 897         if (err)
 898                 return err;
 899
 900         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 901                 return -EAFNOSUPPORT;
 902
 903         spin_lock_bh(&vxlan->hash_lock);
 904         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 905                                port, vni, ifindex, ndm->ndm_flags);
 906         spin_unlock_bh(&vxlan->hash_lock);
 907
 908         return err;
 909 }
 910
 911 /* Delete entry (via netlink) */
 912 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 913                             struct net_device *dev,
 914                             const unsigned char *addr, u16 vid)
 915 {
 916         struct vxlan_dev *vxlan = netdev_priv(dev);
 917         struct vxlan_fdb *f;
 918         struct vxlan_rdst *rd = NULL;
 919         union vxlan_addr ip;
 920         __be16 port;
 921         u32 vni, ifindex;
 922         int err;
 923
 924         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 925         if (err)
 926                 return err;
 927
 928         err = -ENOENT;
 929
 930         spin_lock_bh(&vxlan->hash_lock);
 931         f = vxlan_find_mac(vxlan, addr);
 932         if (!f)
 933                 goto out;
 934
 935         if (!vxlan_addr_any(&ip)) {
 936                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 937                 if (!rd)
 938                         goto out;
 939         }
 940
 941         err = 0;
 942
 943         /* remove a destination if it's not the only one on the list,
 944          * otherwise destroy the fdb entry
 945          */
 946         if (rd && !list_is_singular(&f->remotes)) {
 947                 list_del_rcu(&rd->list);
 948                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 949                 kfree_rcu(rd, rcu);
 950                 goto out;
 951         }
 952
 953         vxlan_fdb_destroy(vxlan, f);
 954
 955 out:
 956         spin_unlock_bh(&vxlan->hash_lock);
 957
 958         return err;
 959 }
 960
 961 /* Dump forwarding table */
 962 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 963                           struct net_device *dev,
 964                           struct net_device *filter_dev, int idx)
 965 {
 966         struct vxlan_dev *vxlan = netdev_priv(dev);
 967         unsigned int h;
 968
 969         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 970                 struct vxlan_fdb *f;
 971                 int err;
 972
 973                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 974                         struct vxlan_rdst *rd;
 975
 976                         if (idx < cb->args[0])
 977                                 goto skip;
 978
 979                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 980                                 err = vxlan_fdb_info(skb, vxlan, f,
 981                                                      NETLINK_CB(cb->skb).portid,
 982                                                      cb->nlh->nlmsg_seq,
 983                                                      RTM_NEWNEIGH,
 984                                                      NLM_F_MULTI, rd);
 985                                 if (err < 0)
 986                                         goto out;
 987                         }
 988 skip:
 989                         ++idx;
 990                 }
 991         }
 992 out:
 993         return idx;
 994 }
 995
 996 /* Watch incoming packets to learn mapping between Ethernet address
 997  * and Tunnel endpoint.
 998  * Return true if packet is bogus and should be droppped.
 999  */
1000 static bool vxlan_snoop(struct net_device *dev,
1001                         union vxlan_addr *src_ip, const u8 *src_mac)
1002 {
1003         struct vxlan_dev *vxlan = netdev_priv(dev);
1004         struct vxlan_fdb *f;
1005
1006         f = vxlan_find_mac(vxlan, src_mac);
1007         if (likely(f)) {
1008                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1009
1010                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
1011                         return false;
1012
1013                 /* Don't migrate static entries, drop packets */
1014                 if (f->state & NUD_NOARP)
1015                         return true;
1016
1017                 if (net_ratelimit())
1018                         netdev_info(dev,
1019                                     "%pM migrated from %pIS to %pIS\n",
1020                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1021
1022                 rdst->remote_ip = *src_ip;
1023                 f->updated = jiffies;
1024                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1025         } else {
1026                 /* learned new entry */
1027                 spin_lock(&vxlan->hash_lock);
1028
1029                 /* close off race between vxlan_flush and incoming packets */
1030                 if (netif_running(dev))
1031                         vxlan_fdb_create(vxlan, src_mac, src_ip,
1032                                          NUD_REACHABLE,
1033                                          NLM_F_EXCL|NLM_F_CREATE,
1034                                          vxlan->dst_port,
1035                                          vxlan->default_dst.remote_vni,
1036                                          0, NTF_SELF);
1037                 spin_unlock(&vxlan->hash_lock);
1038         }
1039
1040         return false;
1041 }
1042
1043 /* See if multicast group is already in use by other ID */
1044 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1045 {
1046         struct vxlan_dev *vxlan;
1047
1048         /* The vxlan_sock is only used by dev, leaving group has
1049          * no effect on other vxlan devices.
1050          */
1051         if (atomic_read(&dev->vn_sock->refcnt) == 1)
1052                 return false;
1053
1054         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1055                 if (!netif_running(vxlan->dev) || vxlan == dev)
1056                         continue;
1057
1058                 if (vxlan->vn_sock != dev->vn_sock)
1059                         continue;
1060
1061                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1062                                       &dev->default_dst.remote_ip))
1063                         continue;
1064
1065                 if (vxlan->default_dst.remote_ifindex !=
1066                     dev->default_dst.remote_ifindex)
1067                         continue;
1068
1069                 return true;
1070         }
1071
1072         return false;
1073 }
1074
1075 static void vxlan_sock_hold(struct vxlan_sock *vs)
1076 {
1077         atomic_inc(&vs->refcnt);
1078 }
1079
1080 void vxlan_sock_release(struct vxlan_sock *vs)
1081 {
1082         struct sock *sk = vs->sock->sk;
1083         struct net *net = sock_net(sk);
1084         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1085
1086         if (!atomic_dec_and_test(&vs->refcnt))
1087                 return;
1088
1089         spin_lock(&vn->sock_lock);
1090         hlist_del_rcu(&vs->hlist);
1091         vxlan_notify_del_rx_port(vs);
1092         spin_unlock(&vn->sock_lock);
1093
1094         queue_work(vxlan_wq, &vs->del_work);
1095 }
1096 EXPORT_SYMBOL_GPL(vxlan_sock_release);
1097
1098 /* Callback to update multicast group membership when first VNI on
1099  * multicast asddress is brought up
1100  */
1101 static void vxlan_igmp_join(struct work_struct *work)
1102 {
1103         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_join);
1104         struct vxlan_sock *vs = vxlan->vn_sock;
1105         struct sock *sk = vs->sock->sk;
1106         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1107         int ifindex = vxlan->default_dst.remote_ifindex;
1108
1109         rtnl_lock();
1110         lock_sock(sk);
1111         if (ip->sa.sa_family == AF_INET) {
1112                 struct ip_mreqn mreq = {
1113                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1114                         .imr_ifindex            = ifindex,
1115                 };
1116
1117                 ip_mc_join_group(sk, &mreq);
1118 #if IS_ENABLED(CONFIG_IPV6)
1119         } else {
1120                 ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1121                                              &ip->sin6.sin6_addr);
1122 #endif
1123         }
1124         release_sock(sk);
1125         rtnl_unlock();
1126
1127         vxlan_sock_release(vs);
1128         dev_put(vxlan->dev);
1129 }
1130
1131 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1132 static void vxlan_igmp_leave(struct work_struct *work)
1133 {
1134         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_leave);
1135         struct vxlan_sock *vs = vxlan->vn_sock;
1136         struct sock *sk = vs->sock->sk;
1137         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1138         int ifindex = vxlan->default_dst.remote_ifindex;
1139
1140         rtnl_lock();
1141         lock_sock(sk);
1142         if (ip->sa.sa_family == AF_INET) {
1143                 struct ip_mreqn mreq = {
1144                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1145                         .imr_ifindex            = ifindex,
1146                 };
1147
1148                 ip_mc_leave_group(sk, &mreq);
1149 #if IS_ENABLED(CONFIG_IPV6)
1150         } else {
1151                 ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1152                                              &ip->sin6.sin6_addr);
1153 #endif
1154         }
1155
1156         release_sock(sk);
1157         rtnl_unlock();
1158
1159         vxlan_sock_release(vs);
1160         dev_put(vxlan->dev);
1161 }
1162
1163 static struct vxlanhdr *vxlan_remcsum(struct sk_buff *skb, struct vxlanhdr *vh,
1164                                       size_t hdrlen, u32 data, bool nopartial)
1165 {
1166         size_t start, offset, plen;
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         skb_remcsum_process(skb, (void *)vh + hdrlen, start, offset,
1181                             nopartial);
1182
1183         return vh;
1184 }
1185
1186 /* Callback from net/ipv4/udp.c to receive packets */
1187 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1188 {
1189         struct vxlan_sock *vs;
1190         struct vxlanhdr *vxh;
1191         u32 flags, vni;
1192         struct vxlan_metadata md = {0};
1193
1194         /* Need Vxlan and inner Ethernet header to be present */
1195         if (!pskb_may_pull(skb, VXLAN_HLEN))
1196                 goto error;
1197
1198         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1199         flags = ntohl(vxh->vx_flags);
1200         vni = ntohl(vxh->vx_vni);
1201
1202         if (flags & VXLAN_HF_VNI) {
1203                 flags &= ~VXLAN_HF_VNI;
1204         } else {
1205                 /* VNI flag always required to be set */
1206                 goto bad_flags;
1207         }
1208
1209         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1210                 goto drop;
1211         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1212
1213         vs = rcu_dereference_sk_user_data(sk);
1214         if (!vs)
1215                 goto drop;
1216
1217         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1218                 vxh = vxlan_remcsum(skb, vxh, sizeof(struct vxlanhdr), vni,
1219                                     !!(vs->flags & VXLAN_F_REMCSUM_NOPARTIAL));
1220                 if (!vxh)
1221                         goto drop;
1222
1223                 flags &= ~VXLAN_HF_RCO;
1224                 vni &= VXLAN_VID_MASK;
1225         }
1226
1227         /* For backwards compatibility, only allow reserved fields to be
1228          * used by VXLAN extensions if explicitly requested.
1229          */
1230         if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1231                 struct vxlanhdr_gbp *gbp;
1232
1233                 gbp = (struct vxlanhdr_gbp *)vxh;
1234                 md.gbp = ntohs(gbp->policy_id);
1235
1236                 if (gbp->dont_learn)
1237                         md.gbp |= VXLAN_GBP_DONT_LEARN;
1238
1239                 if (gbp->policy_applied)
1240                         md.gbp |= VXLAN_GBP_POLICY_APPLIED;
1241
1242                 flags &= ~VXLAN_GBP_USED_BITS;
1243         }
1244
1245         if (flags || (vni & ~VXLAN_VID_MASK)) {
1246                 /* If there are any unprocessed flags remaining treat
1247                  * this as a malformed packet. This behavior diverges from
1248                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1249                  * in reserved fields are to be ignored. The approach here
1250                  * maintains compatbility with previous stack code, and also
1251                  * is more robust and provides a little more security in
1252                  * adding extensions to VXLAN.
1253                  */
1254
1255                 goto bad_flags;
1256         }
1257
1258         md.vni = vxh->vx_vni;
1259         vs->rcv(vs, skb, &md);
1260         return 0;
1261
1262 drop:
1263         /* Consume bad packet */
1264         kfree_skb(skb);
1265         return 0;
1266
1267 bad_flags:
1268         netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1269                    ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1270
1271 error:
1272         /* Return non vxlan pkt */
1273         return 1;
1274 }
1275
1276 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1277                       struct vxlan_metadata *md)
1278 {
1279         struct iphdr *oip = NULL;
1280         struct ipv6hdr *oip6 = NULL;
1281         struct vxlan_dev *vxlan;
1282         struct pcpu_sw_netstats *stats;
1283         union vxlan_addr saddr;
1284         __u32 vni;
1285         int err = 0;
1286         union vxlan_addr *remote_ip;
1287
1288         vni = ntohl(md->vni) >> 8;
1289         /* Is this VNI defined? */
1290         vxlan = vxlan_vs_find_vni(vs, vni);
1291         if (!vxlan)
1292                 goto drop;
1293
1294         remote_ip = &vxlan->default_dst.remote_ip;
1295         skb_reset_mac_header(skb);
1296         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1297         skb->protocol = eth_type_trans(skb, vxlan->dev);
1298         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1299
1300         /* Ignore packet loops (and multicast echo) */
1301         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1302                 goto drop;
1303
1304         /* Re-examine inner Ethernet packet */
1305         if (remote_ip->sa.sa_family == AF_INET) {
1306                 oip = ip_hdr(skb);
1307                 saddr.sin.sin_addr.s_addr = oip->saddr;
1308                 saddr.sa.sa_family = AF_INET;
1309 #if IS_ENABLED(CONFIG_IPV6)
1310         } else {
1311                 oip6 = ipv6_hdr(skb);
1312                 saddr.sin6.sin6_addr = oip6->saddr;
1313                 saddr.sa.sa_family = AF_INET6;
1314 #endif
1315         }
1316
1317         if ((vxlan->flags & VXLAN_F_LEARN) &&
1318             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1319                 goto drop;
1320
1321         skb_reset_network_header(skb);
1322         skb->mark = md->gbp;
1323
1324         if (oip6)
1325                 err = IP6_ECN_decapsulate(oip6, skb);
1326         if (oip)
1327                 err = IP_ECN_decapsulate(oip, skb);
1328
1329         if (unlikely(err)) {
1330                 if (log_ecn_error) {
1331                         if (oip6)
1332                                 net_info_ratelimited("non-ECT from %pI6\n",
1333                                                      &oip6->saddr);
1334                         if (oip)
1335                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1336                                                      &oip->saddr, oip->tos);
1337                 }
1338                 if (err > 1) {
1339                         ++vxlan->dev->stats.rx_frame_errors;
1340                         ++vxlan->dev->stats.rx_errors;
1341                         goto drop;
1342                 }
1343         }
1344
1345         stats = this_cpu_ptr(vxlan->dev->tstats);
1346         u64_stats_update_begin(&stats->syncp);
1347         stats->rx_packets++;
1348         stats->rx_bytes += skb->len;
1349         u64_stats_update_end(&stats->syncp);
1350
1351         netif_rx(skb);
1352
1353         return;
1354 drop:
1355         /* Consume bad packet */
1356         kfree_skb(skb);
1357 }
1358
1359 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1360 {
1361         struct vxlan_dev *vxlan = netdev_priv(dev);
1362         struct arphdr *parp;
1363         u8 *arpptr, *sha;
1364         __be32 sip, tip;
1365         struct neighbour *n;
1366
1367         if (dev->flags & IFF_NOARP)
1368                 goto out;
1369
1370         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1371                 dev->stats.tx_dropped++;
1372                 goto out;
1373         }
1374         parp = arp_hdr(skb);
1375
1376         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1377              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1378             parp->ar_pro != htons(ETH_P_IP) ||
1379             parp->ar_op != htons(ARPOP_REQUEST) ||
1380             parp->ar_hln != dev->addr_len ||
1381             parp->ar_pln != 4)
1382                 goto out;
1383         arpptr = (u8 *)parp + sizeof(struct arphdr);
1384         sha = arpptr;
1385         arpptr += dev->addr_len;        /* sha */
1386         memcpy(&sip, arpptr, sizeof(sip));
1387         arpptr += sizeof(sip);
1388         arpptr += dev->addr_len;        /* tha */
1389         memcpy(&tip, arpptr, sizeof(tip));
1390
1391         if (ipv4_is_loopback(tip) ||
1392             ipv4_is_multicast(tip))
1393                 goto out;
1394
1395         n = neigh_lookup(&arp_tbl, &tip, dev);
1396
1397         if (n) {
1398                 struct vxlan_fdb *f;
1399                 struct sk_buff  *reply;
1400
1401                 if (!(n->nud_state & NUD_CONNECTED)) {
1402                         neigh_release(n);
1403                         goto out;
1404                 }
1405
1406                 f = vxlan_find_mac(vxlan, n->ha);
1407                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1408                         /* bridge-local neighbor */
1409                         neigh_release(n);
1410                         goto out;
1411                 }
1412
1413                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1414                                 n->ha, sha);
1415
1416                 neigh_release(n);
1417
1418                 if (reply == NULL)
1419                         goto out;
1420
1421                 skb_reset_mac_header(reply);
1422                 __skb_pull(reply, skb_network_offset(reply));
1423                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1424                 reply->pkt_type = PACKET_HOST;
1425
1426                 if (netif_rx_ni(reply) == NET_RX_DROP)
1427                         dev->stats.rx_dropped++;
1428         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1429                 union vxlan_addr ipa = {
1430                         .sin.sin_addr.s_addr = tip,
1431                         .sin.sin_family = AF_INET,
1432                 };
1433
1434                 vxlan_ip_miss(dev, &ipa);
1435         }
1436 out:
1437         consume_skb(skb);
1438         return NETDEV_TX_OK;
1439 }
1440
1441 #if IS_ENABLED(CONFIG_IPV6)
1442 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1443         struct neighbour *n, bool isrouter)
1444 {
1445         struct net_device *dev = request->dev;
1446         struct sk_buff *reply;
1447         struct nd_msg *ns, *na;
1448         struct ipv6hdr *pip6;
1449         u8 *daddr;
1450         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1451         int ns_olen;
1452         int i, len;
1453
1454         if (dev == NULL)
1455                 return NULL;
1456
1457         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1458                 sizeof(*na) + na_olen + dev->needed_tailroom;
1459         reply = alloc_skb(len, GFP_ATOMIC);
1460         if (reply == NULL)
1461                 return NULL;
1462
1463         reply->protocol = htons(ETH_P_IPV6);
1464         reply->dev = dev;
1465         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1466         skb_push(reply, sizeof(struct ethhdr));
1467         skb_set_mac_header(reply, 0);
1468
1469         ns = (struct nd_msg *)skb_transport_header(request);
1470
1471         daddr = eth_hdr(request)->h_source;
1472         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1473         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1474                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1475                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1476                         break;
1477                 }
1478         }
1479
1480         /* Ethernet header */
1481         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1482         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1483         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1484         reply->protocol = htons(ETH_P_IPV6);
1485
1486         skb_pull(reply, sizeof(struct ethhdr));
1487         skb_set_network_header(reply, 0);
1488         skb_put(reply, sizeof(struct ipv6hdr));
1489
1490         /* IPv6 header */
1491
1492         pip6 = ipv6_hdr(reply);
1493         memset(pip6, 0, sizeof(struct ipv6hdr));
1494         pip6->version = 6;
1495         pip6->priority = ipv6_hdr(request)->priority;
1496         pip6->nexthdr = IPPROTO_ICMPV6;
1497         pip6->hop_limit = 255;
1498         pip6->daddr = ipv6_hdr(request)->saddr;
1499         pip6->saddr = *(struct in6_addr *)n->primary_key;
1500
1501         skb_pull(reply, sizeof(struct ipv6hdr));
1502         skb_set_transport_header(reply, 0);
1503
1504         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1505
1506         /* Neighbor Advertisement */
1507         memset(na, 0, sizeof(*na)+na_olen);
1508         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1509         na->icmph.icmp6_router = isrouter;
1510         na->icmph.icmp6_override = 1;
1511         na->icmph.icmp6_solicited = 1;
1512         na->target = ns->target;
1513         ether_addr_copy(&na->opt[2], n->ha);
1514         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1515         na->opt[1] = na_olen >> 3;
1516
1517         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1518                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1519                 csum_partial(na, sizeof(*na)+na_olen, 0));
1520
1521         pip6->payload_len = htons(sizeof(*na)+na_olen);
1522
1523         skb_push(reply, sizeof(struct ipv6hdr));
1524
1525         reply->ip_summed = CHECKSUM_UNNECESSARY;
1526
1527         return reply;
1528 }
1529
1530 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1531 {
1532         struct vxlan_dev *vxlan = netdev_priv(dev);
1533         struct nd_msg *msg;
1534         const struct ipv6hdr *iphdr;
1535         const struct in6_addr *saddr, *daddr;
1536         struct neighbour *n;
1537         struct inet6_dev *in6_dev;
1538
1539         in6_dev = __in6_dev_get(dev);
1540         if (!in6_dev)
1541                 goto out;
1542
1543         iphdr = ipv6_hdr(skb);
1544         saddr = &iphdr->saddr;
1545         daddr = &iphdr->daddr;
1546
1547         msg = (struct nd_msg *)skb_transport_header(skb);
1548         if (msg->icmph.icmp6_code != 0 ||
1549             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1550                 goto out;
1551
1552         if (ipv6_addr_loopback(daddr) ||
1553             ipv6_addr_is_multicast(&msg->target))
1554                 goto out;
1555
1556         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1557
1558         if (n) {
1559                 struct vxlan_fdb *f;
1560                 struct sk_buff *reply;
1561
1562                 if (!(n->nud_state & NUD_CONNECTED)) {
1563                         neigh_release(n);
1564                         goto out;
1565                 }
1566
1567                 f = vxlan_find_mac(vxlan, n->ha);
1568                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1569                         /* bridge-local neighbor */
1570                         neigh_release(n);
1571                         goto out;
1572                 }
1573
1574                 reply = vxlan_na_create(skb, n,
1575                                         !!(f ? f->flags & NTF_ROUTER : 0));
1576
1577                 neigh_release(n);
1578
1579                 if (reply == NULL)
1580                         goto out;
1581
1582                 if (netif_rx_ni(reply) == NET_RX_DROP)
1583                         dev->stats.rx_dropped++;
1584
1585         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1586                 union vxlan_addr ipa = {
1587                         .sin6.sin6_addr = msg->target,
1588                         .sin6.sin6_family = AF_INET6,
1589                 };
1590
1591                 vxlan_ip_miss(dev, &ipa);
1592         }
1593
1594 out:
1595         consume_skb(skb);
1596         return NETDEV_TX_OK;
1597 }
1598 #endif
1599
1600 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1601 {
1602         struct vxlan_dev *vxlan = netdev_priv(dev);
1603         struct neighbour *n;
1604
1605         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1606                 return false;
1607
1608         n = NULL;
1609         switch (ntohs(eth_hdr(skb)->h_proto)) {
1610         case ETH_P_IP:
1611         {
1612                 struct iphdr *pip;
1613
1614                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1615                         return false;
1616                 pip = ip_hdr(skb);
1617                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1618                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1619                         union vxlan_addr ipa = {
1620                                 .sin.sin_addr.s_addr = pip->daddr,
1621                                 .sin.sin_family = AF_INET,
1622                         };
1623
1624                         vxlan_ip_miss(dev, &ipa);
1625                         return false;
1626                 }
1627
1628                 break;
1629         }
1630 #if IS_ENABLED(CONFIG_IPV6)
1631         case ETH_P_IPV6:
1632         {
1633                 struct ipv6hdr *pip6;
1634
1635                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1636                         return false;
1637                 pip6 = ipv6_hdr(skb);
1638                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1639                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1640                         union vxlan_addr ipa = {
1641                                 .sin6.sin6_addr = pip6->daddr,
1642                                 .sin6.sin6_family = AF_INET6,
1643                         };
1644
1645                         vxlan_ip_miss(dev, &ipa);
1646                         return false;
1647                 }
1648
1649                 break;
1650         }
1651 #endif
1652         default:
1653                 return false;
1654         }
1655
1656         if (n) {
1657                 bool diff;
1658
1659                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1660                 if (diff) {
1661                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1662                                 dev->addr_len);
1663                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1664                 }
1665                 neigh_release(n);
1666                 return diff;
1667         }
1668
1669         return false;
1670 }
1671
1672 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1673                                 struct vxlan_metadata *md)
1674 {
1675         struct vxlanhdr_gbp *gbp;
1676
1677         if (!md->gbp)
1678                 return;
1679
1680         gbp = (struct vxlanhdr_gbp *)vxh;
1681         vxh->vx_flags |= htonl(VXLAN_HF_GBP);
1682
1683         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1684                 gbp->dont_learn = 1;
1685
1686         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1687                 gbp->policy_applied = 1;
1688
1689         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1690 }
1691
1692 #if IS_ENABLED(CONFIG_IPV6)
1693 static int vxlan6_xmit_skb(struct dst_entry *dst, struct sk_buff *skb,
1694                            struct net_device *dev, struct in6_addr *saddr,
1695                            struct in6_addr *daddr, __u8 prio, __u8 ttl,
1696                            __be16 src_port, __be16 dst_port,
1697                            struct vxlan_metadata *md, bool xnet, u32 vxflags)
1698 {
1699         struct vxlanhdr *vxh;
1700         int min_headroom;
1701         int err;
1702         bool udp_sum = !(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX);
1703         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1704         u16 hdrlen = sizeof(struct vxlanhdr);
1705
1706         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1707             skb->ip_summed == CHECKSUM_PARTIAL) {
1708                 int csum_start = skb_checksum_start_offset(skb);
1709
1710                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1711                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1712                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1713                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1714                         udp_sum = false;
1715                         type |= SKB_GSO_TUNNEL_REMCSUM;
1716                 }
1717         }
1718
1719         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1720         if (IS_ERR(skb)) {
1721                 err = -EINVAL;
1722                 goto err;
1723         }
1724
1725         skb_scrub_packet(skb, xnet);
1726
1727         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1728                         + VXLAN_HLEN + sizeof(struct ipv6hdr)
1729                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1730
1731         /* Need space for new headers (invalidates iph ptr) */
1732         err = skb_cow_head(skb, min_headroom);
1733         if (unlikely(err)) {
1734                 kfree_skb(skb);
1735                 goto err;
1736         }
1737
1738         skb = vlan_hwaccel_push_inside(skb);
1739         if (WARN_ON(!skb)) {
1740                 err = -ENOMEM;
1741                 goto err;
1742         }
1743
1744         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1745         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1746         vxh->vx_vni = md->vni;
1747
1748         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1749                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1750                            VXLAN_RCO_SHIFT;
1751
1752                 if (skb->csum_offset == offsetof(struct udphdr, check))
1753                         data |= VXLAN_RCO_UDP;
1754
1755                 vxh->vx_vni |= htonl(data);
1756                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1757
1758                 if (!skb_is_gso(skb)) {
1759                         skb->ip_summed = CHECKSUM_NONE;
1760                         skb->encapsulation = 0;
1761                 }
1762         }
1763
1764         if (vxflags & VXLAN_F_GBP)
1765                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1766
1767         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1768
1769         udp_tunnel6_xmit_skb(dst, skb, dev, saddr, daddr, prio,
1770                              ttl, src_port, dst_port,
1771                              !!(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX));
1772         return 0;
1773 err:
1774         dst_release(dst);
1775         return err;
1776 }
1777 #endif
1778
1779 int vxlan_xmit_skb(struct rtable *rt, struct sk_buff *skb,
1780                    __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1781                    __be16 src_port, __be16 dst_port,
1782                    struct vxlan_metadata *md, bool xnet, u32 vxflags)
1783 {
1784         struct vxlanhdr *vxh;
1785         int min_headroom;
1786         int err;
1787         bool udp_sum = !!(vxflags & VXLAN_F_UDP_CSUM);
1788         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1789         u16 hdrlen = sizeof(struct vxlanhdr);
1790
1791         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1792             skb->ip_summed == CHECKSUM_PARTIAL) {
1793                 int csum_start = skb_checksum_start_offset(skb);
1794
1795                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1796                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1797                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1798                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1799                         udp_sum = false;
1800                         type |= SKB_GSO_TUNNEL_REMCSUM;
1801                 }
1802         }
1803
1804         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1805         if (IS_ERR(skb))
1806                 return PTR_ERR(skb);
1807
1808         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1809                         + VXLAN_HLEN + sizeof(struct iphdr)
1810                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1811
1812         /* Need space for new headers (invalidates iph ptr) */
1813         err = skb_cow_head(skb, min_headroom);
1814         if (unlikely(err)) {
1815                 kfree_skb(skb);
1816                 return err;
1817         }
1818
1819         skb = vlan_hwaccel_push_inside(skb);
1820         if (WARN_ON(!skb))
1821                 return -ENOMEM;
1822
1823         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1824         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1825         vxh->vx_vni = md->vni;
1826
1827         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1828                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1829                            VXLAN_RCO_SHIFT;
1830
1831                 if (skb->csum_offset == offsetof(struct udphdr, check))
1832                         data |= VXLAN_RCO_UDP;
1833
1834                 vxh->vx_vni |= htonl(data);
1835                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1836
1837                 if (!skb_is_gso(skb)) {
1838                         skb->ip_summed = CHECKSUM_NONE;
1839                         skb->encapsulation = 0;
1840                 }
1841         }
1842
1843         if (vxflags & VXLAN_F_GBP)
1844                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1845
1846         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1847
1848         return udp_tunnel_xmit_skb(rt, skb, src, dst, tos,
1849                                    ttl, df, src_port, dst_port, xnet,
1850                                    !(vxflags & VXLAN_F_UDP_CSUM));
1851 }
1852 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1853
1854 /* Bypass encapsulation if the destination is local */
1855 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1856                                struct vxlan_dev *dst_vxlan)
1857 {
1858         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1859         union vxlan_addr loopback;
1860         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1861         struct net_device *dev = skb->dev;
1862         int len = skb->len;
1863
1864         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1865         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1866         skb->pkt_type = PACKET_HOST;
1867         skb->encapsulation = 0;
1868         skb->dev = dst_vxlan->dev;
1869         __skb_pull(skb, skb_network_offset(skb));
1870
1871         if (remote_ip->sa.sa_family == AF_INET) {
1872                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1873                 loopback.sa.sa_family =  AF_INET;
1874 #if IS_ENABLED(CONFIG_IPV6)
1875         } else {
1876                 loopback.sin6.sin6_addr = in6addr_loopback;
1877                 loopback.sa.sa_family =  AF_INET6;
1878 #endif
1879         }
1880
1881         if (dst_vxlan->flags & VXLAN_F_LEARN)
1882                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1883
1884         u64_stats_update_begin(&tx_stats->syncp);
1885         tx_stats->tx_packets++;
1886         tx_stats->tx_bytes += len;
1887         u64_stats_update_end(&tx_stats->syncp);
1888
1889         if (netif_rx(skb) == NET_RX_SUCCESS) {
1890                 u64_stats_update_begin(&rx_stats->syncp);
1891                 rx_stats->rx_packets++;
1892                 rx_stats->rx_bytes += len;
1893                 u64_stats_update_end(&rx_stats->syncp);
1894         } else {
1895                 dev->stats.rx_dropped++;
1896         }
1897 }
1898
1899 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1900                            struct vxlan_rdst *rdst, bool did_rsc)
1901 {
1902         struct vxlan_dev *vxlan = netdev_priv(dev);
1903         struct rtable *rt = NULL;
1904         const struct iphdr *old_iph;
1905         struct flowi4 fl4;
1906         union vxlan_addr *dst;
1907         struct vxlan_metadata md;
1908         __be16 src_port = 0, dst_port;
1909         u32 vni;
1910         __be16 df = 0;
1911         __u8 tos, ttl;
1912         int err;
1913
1914         dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1915         vni = rdst->remote_vni;
1916         dst = &rdst->remote_ip;
1917
1918         if (vxlan_addr_any(dst)) {
1919                 if (did_rsc) {
1920                         /* short-circuited back to local bridge */
1921                         vxlan_encap_bypass(skb, vxlan, vxlan);
1922                         return;
1923                 }
1924                 goto drop;
1925         }
1926
1927         old_iph = ip_hdr(skb);
1928
1929         ttl = vxlan->ttl;
1930         if (!ttl && vxlan_addr_multicast(dst))
1931                 ttl = 1;
1932
1933         tos = vxlan->tos;
1934         if (tos == 1)
1935                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1936
1937         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->port_min,
1938                                      vxlan->port_max, true);
1939
1940         if (dst->sa.sa_family == AF_INET) {
1941                 memset(&fl4, 0, sizeof(fl4));
1942                 fl4.flowi4_oif = rdst->remote_ifindex;
1943                 fl4.flowi4_tos = RT_TOS(tos);
1944                 fl4.daddr = dst->sin.sin_addr.s_addr;
1945                 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1946
1947                 rt = ip_route_output_key(vxlan->net, &fl4);
1948                 if (IS_ERR(rt)) {
1949                         netdev_dbg(dev, "no route to %pI4\n",
1950                                    &dst->sin.sin_addr.s_addr);
1951                         dev->stats.tx_carrier_errors++;
1952                         goto tx_error;
1953                 }
1954
1955                 if (rt->dst.dev == dev) {
1956                         netdev_dbg(dev, "circular route to %pI4\n",
1957                                    &dst->sin.sin_addr.s_addr);
1958                         dev->stats.collisions++;
1959                         goto rt_tx_error;
1960                 }
1961
1962                 /* Bypass encapsulation if the destination is local */
1963                 if (rt->rt_flags & RTCF_LOCAL &&
1964                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1965                         struct vxlan_dev *dst_vxlan;
1966
1967                         ip_rt_put(rt);
1968                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1969                                                    dst->sa.sa_family, dst_port,
1970                                                    vxlan->flags);
1971                         if (!dst_vxlan)
1972                                 goto tx_error;
1973                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1974                         return;
1975                 }
1976
1977                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1978                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1979                 md.vni = htonl(vni << 8);
1980                 md.gbp = skb->mark;
1981
1982                 err = vxlan_xmit_skb(rt, skb, fl4.saddr,
1983                                      dst->sin.sin_addr.s_addr, tos, ttl, df,
1984                                      src_port, dst_port, &md,
1985                                      !net_eq(vxlan->net, dev_net(vxlan->dev)),
1986                                      vxlan->flags);
1987                 if (err < 0) {
1988                         /* skb is already freed. */
1989                         skb = NULL;
1990                         goto rt_tx_error;
1991                 }
1992
1993                 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1994 #if IS_ENABLED(CONFIG_IPV6)
1995         } else {
1996                 struct sock *sk = vxlan->vn_sock->sock->sk;
1997                 struct dst_entry *ndst;
1998                 struct flowi6 fl6;
1999                 u32 flags;
2000
2001                 memset(&fl6, 0, sizeof(fl6));
2002                 fl6.flowi6_oif = rdst->remote_ifindex;
2003                 fl6.daddr = dst->sin6.sin6_addr;
2004                 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
2005                 fl6.flowi6_proto = IPPROTO_UDP;
2006
2007                 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
2008                         netdev_dbg(dev, "no route to %pI6\n",
2009                                    &dst->sin6.sin6_addr);
2010                         dev->stats.tx_carrier_errors++;
2011                         goto tx_error;
2012                 }
2013
2014                 if (ndst->dev == dev) {
2015                         netdev_dbg(dev, "circular route to %pI6\n",
2016                                    &dst->sin6.sin6_addr);
2017                         dst_release(ndst);
2018                         dev->stats.collisions++;
2019                         goto tx_error;
2020                 }
2021
2022                 /* Bypass encapsulation if the destination is local */
2023                 flags = ((struct rt6_info *)ndst)->rt6i_flags;
2024                 if (flags & RTF_LOCAL &&
2025                     !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2026                         struct vxlan_dev *dst_vxlan;
2027
2028                         dst_release(ndst);
2029                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2030                                                    dst->sa.sa_family, dst_port,
2031                                                    vxlan->flags);
2032                         if (!dst_vxlan)
2033                                 goto tx_error;
2034                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2035                         return;
2036                 }
2037
2038                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2039                 md.vni = htonl(vni << 8);
2040                 md.gbp = skb->mark;
2041
2042                 err = vxlan6_xmit_skb(ndst, skb, dev, &fl6.saddr, &fl6.daddr,
2043                                       0, ttl, src_port, dst_port, &md,
2044                                       !net_eq(vxlan->net, dev_net(vxlan->dev)),
2045                                       vxlan->flags);
2046 #endif
2047         }
2048
2049         return;
2050
2051 drop:
2052         dev->stats.tx_dropped++;
2053         goto tx_free;
2054
2055 rt_tx_error:
2056         ip_rt_put(rt);
2057 tx_error:
2058         dev->stats.tx_errors++;
2059 tx_free:
2060         dev_kfree_skb(skb);
2061 }
2062
2063 /* Transmit local packets over Vxlan
2064  *
2065  * Outer IP header inherits ECN and DF from inner header.
2066  * Outer UDP destination is the VXLAN assigned port.
2067  *           source port is based on hash of flow
2068  */
2069 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2070 {
2071         struct vxlan_dev *vxlan = netdev_priv(dev);
2072         struct ethhdr *eth;
2073         bool did_rsc = false;
2074         struct vxlan_rdst *rdst, *fdst = NULL;
2075         struct vxlan_fdb *f;
2076
2077         skb_reset_mac_header(skb);
2078         eth = eth_hdr(skb);
2079
2080         if ((vxlan->flags & VXLAN_F_PROXY)) {
2081                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2082                         return arp_reduce(dev, skb);
2083 #if IS_ENABLED(CONFIG_IPV6)
2084                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2085                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2086                                        + sizeof(struct nd_msg)) &&
2087                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2088                                 struct nd_msg *msg;
2089
2090                                 msg = (struct nd_msg *)skb_transport_header(skb);
2091                                 if (msg->icmph.icmp6_code == 0 &&
2092                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2093                                         return neigh_reduce(dev, skb);
2094                 }
2095                 eth = eth_hdr(skb);
2096 #endif
2097         }
2098
2099         f = vxlan_find_mac(vxlan, eth->h_dest);
2100         did_rsc = false;
2101
2102         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2103             (ntohs(eth->h_proto) == ETH_P_IP ||
2104              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2105                 did_rsc = route_shortcircuit(dev, skb);
2106                 if (did_rsc)
2107                         f = vxlan_find_mac(vxlan, eth->h_dest);
2108         }
2109
2110         if (f == NULL) {
2111                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2112                 if (f == NULL) {
2113                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2114                             !is_multicast_ether_addr(eth->h_dest))
2115                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2116
2117                         dev->stats.tx_dropped++;
2118                         kfree_skb(skb);
2119                         return NETDEV_TX_OK;
2120                 }
2121         }
2122
2123         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2124                 struct sk_buff *skb1;
2125
2126                 if (!fdst) {
2127                         fdst = rdst;
2128                         continue;
2129                 }
2130                 skb1 = skb_clone(skb, GFP_ATOMIC);
2131                 if (skb1)
2132                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2133         }
2134
2135         if (fdst)
2136                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2137         else
2138                 kfree_skb(skb);
2139         return NETDEV_TX_OK;
2140 }
2141
2142 /* Walk the forwarding table and purge stale entries */
2143 static void vxlan_cleanup(unsigned long arg)
2144 {
2145         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2146         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2147         unsigned int h;
2148
2149         if (!netif_running(vxlan->dev))
2150                 return;
2151
2152         spin_lock_bh(&vxlan->hash_lock);
2153         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2154                 struct hlist_node *p, *n;
2155                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2156                         struct vxlan_fdb *f
2157                                 = container_of(p, struct vxlan_fdb, hlist);
2158                         unsigned long timeout;
2159
2160                         if (f->state & NUD_PERMANENT)
2161                                 continue;
2162
2163                         timeout = f->used + vxlan->age_interval * HZ;
2164                         if (time_before_eq(timeout, jiffies)) {
2165                                 netdev_dbg(vxlan->dev,
2166                                            "garbage collect %pM\n",
2167                                            f->eth_addr);
2168                                 f->state = NUD_STALE;
2169                                 vxlan_fdb_destroy(vxlan, f);
2170                         } else if (time_before(timeout, next_timer))
2171                                 next_timer = timeout;
2172                 }
2173         }
2174         spin_unlock_bh(&vxlan->hash_lock);
2175
2176         mod_timer(&vxlan->age_timer, next_timer);
2177 }
2178
2179 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2180 {
2181         __u32 vni = vxlan->default_dst.remote_vni;
2182
2183         vxlan->vn_sock = vs;
2184         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2185 }
2186
2187 /* Setup stats when device is created */
2188 static int vxlan_init(struct net_device *dev)
2189 {
2190         struct vxlan_dev *vxlan = netdev_priv(dev);
2191         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2192         struct vxlan_sock *vs;
2193         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2194
2195         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2196         if (!dev->tstats)
2197                 return -ENOMEM;
2198
2199         spin_lock(&vn->sock_lock);
2200         vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2201                              vxlan->dst_port, vxlan->flags);
2202         if (vs && atomic_add_unless(&vs->refcnt, 1, 0)) {
2203                 /* If we have a socket with same port already, reuse it */
2204                 vxlan_vs_add_dev(vs, vxlan);
2205         } else {
2206                 /* otherwise make new socket outside of RTNL */
2207                 dev_hold(dev);
2208                 queue_work(vxlan_wq, &vxlan->sock_work);
2209         }
2210         spin_unlock(&vn->sock_lock);
2211
2212         return 0;
2213 }
2214
2215 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2216 {
2217         struct vxlan_fdb *f;
2218
2219         spin_lock_bh(&vxlan->hash_lock);
2220         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2221         if (f)
2222                 vxlan_fdb_destroy(vxlan, f);
2223         spin_unlock_bh(&vxlan->hash_lock);
2224 }
2225
2226 static void vxlan_uninit(struct net_device *dev)
2227 {
2228         struct vxlan_dev *vxlan = netdev_priv(dev);
2229         struct vxlan_sock *vs = vxlan->vn_sock;
2230
2231         vxlan_fdb_delete_default(vxlan);
2232
2233         if (vs)
2234                 vxlan_sock_release(vs);
2235         free_percpu(dev->tstats);
2236 }
2237
2238 /* Start ageing timer and join group when device is brought up */
2239 static int vxlan_open(struct net_device *dev)
2240 {
2241         struct vxlan_dev *vxlan = netdev_priv(dev);
2242         struct vxlan_sock *vs = vxlan->vn_sock;
2243
2244         /* socket hasn't been created */
2245         if (!vs)
2246                 return -ENOTCONN;
2247
2248         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2249                 vxlan_sock_hold(vs);
2250                 dev_hold(dev);
2251                 queue_work(vxlan_wq, &vxlan->igmp_join);
2252         }
2253
2254         if (vxlan->age_interval)
2255                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2256
2257         return 0;
2258 }
2259
2260 /* Purge the forwarding table */
2261 static void vxlan_flush(struct vxlan_dev *vxlan)
2262 {
2263         unsigned int h;
2264
2265         spin_lock_bh(&vxlan->hash_lock);
2266         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2267                 struct hlist_node *p, *n;
2268                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2269                         struct vxlan_fdb *f
2270                                 = container_of(p, struct vxlan_fdb, hlist);
2271                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2272                         if (!is_zero_ether_addr(f->eth_addr))
2273                                 vxlan_fdb_destroy(vxlan, f);
2274                 }
2275         }
2276         spin_unlock_bh(&vxlan->hash_lock);
2277 }
2278
2279 /* Cleanup timer and forwarding table on shutdown */
2280 static int vxlan_stop(struct net_device *dev)
2281 {
2282         struct vxlan_dev *vxlan = netdev_priv(dev);
2283         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2284         struct vxlan_sock *vs = vxlan->vn_sock;
2285
2286         if (vs && vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2287             !vxlan_group_used(vn, vxlan)) {
2288                 vxlan_sock_hold(vs);
2289                 dev_hold(dev);
2290                 queue_work(vxlan_wq, &vxlan->igmp_leave);
2291         }
2292
2293         del_timer_sync(&vxlan->age_timer);
2294
2295         vxlan_flush(vxlan);
2296
2297         return 0;
2298 }
2299
2300 /* Stub, nothing needs to be done. */
2301 static void vxlan_set_multicast_list(struct net_device *dev)
2302 {
2303 }
2304
2305 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2306 {
2307         struct vxlan_dev *vxlan = netdev_priv(dev);
2308         struct vxlan_rdst *dst = &vxlan->default_dst;
2309         struct net_device *lowerdev;
2310         int max_mtu;
2311
2312         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2313         if (lowerdev == NULL)
2314                 return eth_change_mtu(dev, new_mtu);
2315
2316         if (dst->remote_ip.sa.sa_family == AF_INET6)
2317                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2318         else
2319                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2320
2321         if (new_mtu < 68 || new_mtu > max_mtu)
2322                 return -EINVAL;
2323
2324         dev->mtu = new_mtu;
2325         return 0;
2326 }
2327
2328 static const struct net_device_ops vxlan_netdev_ops = {
2329         .ndo_init               = vxlan_init,
2330         .ndo_uninit             = vxlan_uninit,
2331         .ndo_open               = vxlan_open,
2332         .ndo_stop               = vxlan_stop,
2333         .ndo_start_xmit         = vxlan_xmit,
2334         .ndo_get_stats64        = ip_tunnel_get_stats64,
2335         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2336         .ndo_change_mtu         = vxlan_change_mtu,
2337         .ndo_validate_addr      = eth_validate_addr,
2338         .ndo_set_mac_address    = eth_mac_addr,
2339         .ndo_fdb_add            = vxlan_fdb_add,
2340         .ndo_fdb_del            = vxlan_fdb_delete,
2341         .ndo_fdb_dump           = vxlan_fdb_dump,
2342 };
2343
2344 /* Info for udev, that this is a virtual tunnel endpoint */
2345 static struct device_type vxlan_type = {
2346         .name = "vxlan",
2347 };
2348
2349 /* Calls the ndo_add_vxlan_port of the caller in order to
2350  * supply the listening VXLAN udp ports. Callers are expected
2351  * to implement the ndo_add_vxlan_port.
2352  */
2353 void vxlan_get_rx_port(struct net_device *dev)
2354 {
2355         struct vxlan_sock *vs;
2356         struct net *net = dev_net(dev);
2357         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2358         sa_family_t sa_family;
2359         __be16 port;
2360         unsigned int i;
2361
2362         spin_lock(&vn->sock_lock);
2363         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2364                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2365                         port = inet_sk(vs->sock->sk)->inet_sport;
2366                         sa_family = vs->sock->sk->sk_family;
2367                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2368                                                             port);
2369                 }
2370         }
2371         spin_unlock(&vn->sock_lock);
2372 }
2373 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2374
2375 /* Initialize the device structure. */
2376 static void vxlan_setup(struct net_device *dev)
2377 {
2378         struct vxlan_dev *vxlan = netdev_priv(dev);
2379         unsigned int h;
2380
2381         eth_hw_addr_random(dev);
2382         ether_setup(dev);
2383         if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2384                 dev->needed_headroom = ETH_HLEN + VXLAN6_HEADROOM;
2385         else
2386                 dev->needed_headroom = ETH_HLEN + VXLAN_HEADROOM;
2387
2388         dev->netdev_ops = &vxlan_netdev_ops;
2389         dev->destructor = free_netdev;
2390         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2391
2392         dev->tx_queue_len = 0;
2393         dev->features   |= NETIF_F_LLTX;
2394         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2395         dev->features   |= NETIF_F_RXCSUM;
2396         dev->features   |= NETIF_F_GSO_SOFTWARE;
2397
2398         dev->vlan_features = dev->features;
2399         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2400         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2401         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2402         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2403         netif_keep_dst(dev);
2404         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2405
2406         INIT_LIST_HEAD(&vxlan->next);
2407         spin_lock_init(&vxlan->hash_lock);
2408         INIT_WORK(&vxlan->igmp_join, vxlan_igmp_join);
2409         INIT_WORK(&vxlan->igmp_leave, vxlan_igmp_leave);
2410         INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
2411
2412         init_timer_deferrable(&vxlan->age_timer);
2413         vxlan->age_timer.function = vxlan_cleanup;
2414         vxlan->age_timer.data = (unsigned long) vxlan;
2415
2416         vxlan->dst_port = htons(vxlan_port);
2417
2418         vxlan->dev = dev;
2419
2420         for (h = 0; h < FDB_HASH_SIZE; ++h)
2421                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2422 }
2423
2424 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2425         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2426         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2427         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2428         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2429         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2430         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2431         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2432         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2433         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2434         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2435         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2436         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2437         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2438         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2439         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2440         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2441         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2442         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2443         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2444         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2445         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2446         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2447         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2448         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .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_rx_checksums =
2519                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2520         } else {
2521                 udp_conf.family = AF_INET;
2522         }
2523
2524         udp_conf.local_udp_port = port;
2525
2526         /* Open UDP socket */
2527         err = udp_sock_create(net, &udp_conf, &sock);
2528         if (err < 0)
2529                 return ERR_PTR(err);
2530
2531         return sock;
2532 }
2533
2534 /* Create new listen socket if needed */
2535 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2536                                               vxlan_rcv_t *rcv, void *data,
2537                                               u32 flags)
2538 {
2539         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2540         struct vxlan_sock *vs;
2541         struct socket *sock;
2542         unsigned int h;
2543         bool ipv6 = !!(flags & VXLAN_F_IPV6);
2544         struct udp_tunnel_sock_cfg tunnel_cfg;
2545
2546         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2547         if (!vs)
2548                 return ERR_PTR(-ENOMEM);
2549
2550         for (h = 0; h < VNI_HASH_SIZE; ++h)
2551                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2552
2553         INIT_WORK(&vs->del_work, vxlan_del_work);
2554
2555         sock = vxlan_create_sock(net, ipv6, port, flags);
2556         if (IS_ERR(sock)) {
2557                 kfree(vs);
2558                 return ERR_CAST(sock);
2559         }
2560
2561         vs->sock = sock;
2562         atomic_set(&vs->refcnt, 1);
2563         vs->rcv = rcv;
2564         vs->data = data;
2565         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2566
2567         /* Initialize the vxlan udp offloads structure */
2568         vs->udp_offloads.port = port;
2569         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2570         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2571
2572         spin_lock(&vn->sock_lock);
2573         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2574         vxlan_notify_add_rx_port(vs);
2575         spin_unlock(&vn->sock_lock);
2576
2577         /* Mark socket as an encapsulation socket. */
2578         tunnel_cfg.sk_user_data = vs;
2579         tunnel_cfg.encap_type = 1;
2580         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2581         tunnel_cfg.encap_destroy = NULL;
2582
2583         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2584
2585         return vs;
2586 }
2587
2588 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2589                                   vxlan_rcv_t *rcv, void *data,
2590                                   bool no_share, u32 flags)
2591 {
2592         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2593         struct vxlan_sock *vs;
2594         bool ipv6 = flags & VXLAN_F_IPV6;
2595
2596         vs = vxlan_socket_create(net, port, rcv, data, flags);
2597         if (!IS_ERR(vs))
2598                 return vs;
2599
2600         if (no_share)   /* Return error if sharing is not allowed. */
2601                 return vs;
2602
2603         spin_lock(&vn->sock_lock);
2604         vs = vxlan_find_sock(net, ipv6 ? AF_INET6 : AF_INET, port, flags);
2605         if (vs && ((vs->rcv != rcv) ||
2606                    !atomic_add_unless(&vs->refcnt, 1, 0)))
2607                         vs = ERR_PTR(-EBUSY);
2608         spin_unlock(&vn->sock_lock);
2609
2610         if (!vs)
2611                 vs = ERR_PTR(-EINVAL);
2612
2613         return vs;
2614 }
2615 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2616
2617 /* Scheduled at device creation to bind to a socket */
2618 static void vxlan_sock_work(struct work_struct *work)
2619 {
2620         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, sock_work);
2621         struct net *net = vxlan->net;
2622         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2623         __be16 port = vxlan->dst_port;
2624         struct vxlan_sock *nvs;
2625
2626         nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags);
2627         spin_lock(&vn->sock_lock);
2628         if (!IS_ERR(nvs))
2629                 vxlan_vs_add_dev(nvs, vxlan);
2630         spin_unlock(&vn->sock_lock);
2631
2632         dev_put(vxlan->dev);
2633 }
2634
2635 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2636                          struct nlattr *tb[], struct nlattr *data[])
2637 {
2638         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2639         struct vxlan_dev *vxlan = netdev_priv(dev);
2640         struct vxlan_rdst *dst = &vxlan->default_dst;
2641         __u32 vni;
2642         int err;
2643         bool use_ipv6 = false;
2644
2645         if (!data[IFLA_VXLAN_ID])
2646                 return -EINVAL;
2647
2648         vxlan->net = src_net;
2649
2650         vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2651         dst->remote_vni = vni;
2652
2653         /* Unless IPv6 is explicitly requested, assume IPv4 */
2654         dst->remote_ip.sa.sa_family = AF_INET;
2655         if (data[IFLA_VXLAN_GROUP]) {
2656                 dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
2657         } else if (data[IFLA_VXLAN_GROUP6]) {
2658                 if (!IS_ENABLED(CONFIG_IPV6))
2659                         return -EPFNOSUPPORT;
2660
2661                 nla_memcpy(&dst->remote_ip.sin6.sin6_addr, data[IFLA_VXLAN_GROUP6],
2662                            sizeof(struct in6_addr));
2663                 dst->remote_ip.sa.sa_family = AF_INET6;
2664                 use_ipv6 = true;
2665         }
2666
2667         if (data[IFLA_VXLAN_LOCAL]) {
2668                 vxlan->saddr.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
2669                 vxlan->saddr.sa.sa_family = AF_INET;
2670         } else if (data[IFLA_VXLAN_LOCAL6]) {
2671                 if (!IS_ENABLED(CONFIG_IPV6))
2672                         return -EPFNOSUPPORT;
2673
2674                 /* TODO: respect scope id */
2675                 nla_memcpy(&vxlan->saddr.sin6.sin6_addr, data[IFLA_VXLAN_LOCAL6],
2676                            sizeof(struct in6_addr));
2677                 vxlan->saddr.sa.sa_family = AF_INET6;
2678                 use_ipv6 = true;
2679         }
2680
2681         if (data[IFLA_VXLAN_LINK] &&
2682             (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2683                 struct net_device *lowerdev
2684                          = __dev_get_by_index(src_net, dst->remote_ifindex);
2685
2686                 if (!lowerdev) {
2687                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2688                         return -ENODEV;
2689                 }
2690
2691 #if IS_ENABLED(CONFIG_IPV6)
2692                 if (use_ipv6) {
2693                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2694                         if (idev && idev->cnf.disable_ipv6) {
2695                                 pr_info("IPv6 is disabled via sysctl\n");
2696                                 return -EPERM;
2697                         }
2698                         vxlan->flags |= VXLAN_F_IPV6;
2699                 }
2700 #endif
2701
2702                 if (!tb[IFLA_MTU])
2703                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2704
2705                 dev->needed_headroom = lowerdev->hard_header_len +
2706                                        (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2707         } else if (use_ipv6)
2708                 vxlan->flags |= VXLAN_F_IPV6;
2709
2710         if (data[IFLA_VXLAN_TOS])
2711                 vxlan->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2712
2713         if (data[IFLA_VXLAN_TTL])
2714                 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2715
2716         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2717                 vxlan->flags |= VXLAN_F_LEARN;
2718
2719         if (data[IFLA_VXLAN_AGEING])
2720                 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2721         else
2722                 vxlan->age_interval = FDB_AGE_DEFAULT;
2723
2724         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2725                 vxlan->flags |= VXLAN_F_PROXY;
2726
2727         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2728                 vxlan->flags |= VXLAN_F_RSC;
2729
2730         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2731                 vxlan->flags |= VXLAN_F_L2MISS;
2732
2733         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2734                 vxlan->flags |= VXLAN_F_L3MISS;
2735
2736         if (data[IFLA_VXLAN_LIMIT])
2737                 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2738
2739         if (data[IFLA_VXLAN_PORT_RANGE]) {
2740                 const struct ifla_vxlan_port_range *p
2741                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2742                 vxlan->port_min = ntohs(p->low);
2743                 vxlan->port_max = ntohs(p->high);
2744         }
2745
2746         if (data[IFLA_VXLAN_PORT])
2747                 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2748
2749         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2750                 vxlan->flags |= VXLAN_F_UDP_CSUM;
2751
2752         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2753             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2754                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2755
2756         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2757             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2758                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2759
2760         if (data[IFLA_VXLAN_REMCSUM_TX] &&
2761             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2762                 vxlan->flags |= VXLAN_F_REMCSUM_TX;
2763
2764         if (data[IFLA_VXLAN_REMCSUM_RX] &&
2765             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2766                 vxlan->flags |= VXLAN_F_REMCSUM_RX;
2767
2768         if (data[IFLA_VXLAN_GBP])
2769                 vxlan->flags |= VXLAN_F_GBP;
2770
2771         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2772                 vxlan->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2773
2774         if (vxlan_find_vni(src_net, vni, use_ipv6 ? AF_INET6 : AF_INET,
2775                            vxlan->dst_port, vxlan->flags)) {
2776                 pr_info("duplicate VNI %u\n", vni);
2777                 return -EEXIST;
2778         }
2779
2780         dev->ethtool_ops = &vxlan_ethtool_ops;
2781
2782         /* create an fdb entry for a valid default destination */
2783         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2784                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2785                                        &vxlan->default_dst.remote_ip,
2786                                        NUD_REACHABLE|NUD_PERMANENT,
2787                                        NLM_F_EXCL|NLM_F_CREATE,
2788                                        vxlan->dst_port,
2789                                        vxlan->default_dst.remote_vni,
2790                                        vxlan->default_dst.remote_ifindex,
2791                                        NTF_SELF);
2792                 if (err)
2793                         return err;
2794         }
2795
2796         err = register_netdevice(dev);
2797         if (err) {
2798                 vxlan_fdb_delete_default(vxlan);
2799                 return err;
2800         }
2801
2802         list_add(&vxlan->next, &vn->vxlan_list);
2803
2804         return 0;
2805 }
2806
2807 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2808 {
2809         struct vxlan_dev *vxlan = netdev_priv(dev);
2810         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2811
2812         spin_lock(&vn->sock_lock);
2813         if (!hlist_unhashed(&vxlan->hlist))
2814                 hlist_del_rcu(&vxlan->hlist);
2815         spin_unlock(&vn->sock_lock);
2816
2817         list_del(&vxlan->next);
2818         unregister_netdevice_queue(dev, head);
2819 }
2820
2821 static size_t vxlan_get_size(const struct net_device *dev)
2822 {
2823
2824         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2825                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2826                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2827                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2828                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2829                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2830                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2831                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2832                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2833                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2834                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2835                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2836                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2837                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2838                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2839                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2840                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2841                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2842                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
2843                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
2844                 0;
2845 }
2846
2847 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2848 {
2849         const struct vxlan_dev *vxlan = netdev_priv(dev);
2850         const struct vxlan_rdst *dst = &vxlan->default_dst;
2851         struct ifla_vxlan_port_range ports = {
2852                 .low =  htons(vxlan->port_min),
2853                 .high = htons(vxlan->port_max),
2854         };
2855
2856         if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2857                 goto nla_put_failure;
2858
2859         if (!vxlan_addr_any(&dst->remote_ip)) {
2860                 if (dst->remote_ip.sa.sa_family == AF_INET) {
2861                         if (nla_put_be32(skb, IFLA_VXLAN_GROUP,
2862                                          dst->remote_ip.sin.sin_addr.s_addr))
2863                                 goto nla_put_failure;
2864 #if IS_ENABLED(CONFIG_IPV6)
2865                 } else {
2866                         if (nla_put(skb, IFLA_VXLAN_GROUP6, sizeof(struct in6_addr),
2867                                     &dst->remote_ip.sin6.sin6_addr))
2868                                 goto nla_put_failure;
2869 #endif
2870                 }
2871         }
2872
2873         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2874                 goto nla_put_failure;
2875
2876         if (!vxlan_addr_any(&vxlan->saddr)) {
2877                 if (vxlan->saddr.sa.sa_family == AF_INET) {
2878                         if (nla_put_be32(skb, IFLA_VXLAN_LOCAL,
2879                                          vxlan->saddr.sin.sin_addr.s_addr))
2880                                 goto nla_put_failure;
2881 #if IS_ENABLED(CONFIG_IPV6)
2882                 } else {
2883                         if (nla_put(skb, IFLA_VXLAN_LOCAL6, sizeof(struct in6_addr),
2884                                     &vxlan->saddr.sin6.sin6_addr))
2885                                 goto nla_put_failure;
2886 #endif
2887                 }
2888         }
2889
2890         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2891             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2892             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2893                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
2894             nla_put_u8(skb, IFLA_VXLAN_PROXY,
2895                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
2896             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2897             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2898                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2899             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2900                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2901             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2902             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2903             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port) ||
2904             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2905                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2906             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2907                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2908             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2909                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
2910             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
2911                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
2912             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
2913                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
2914                 goto nla_put_failure;
2915
2916         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2917                 goto nla_put_failure;
2918
2919         if (vxlan->flags & VXLAN_F_GBP &&
2920             nla_put_flag(skb, IFLA_VXLAN_GBP))
2921                 goto nla_put_failure;
2922
2923         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
2924             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
2925                 goto nla_put_failure;
2926
2927         return 0;
2928
2929 nla_put_failure:
2930         return -EMSGSIZE;
2931 }
2932
2933 static struct net *vxlan_get_link_net(const struct net_device *dev)
2934 {
2935         struct vxlan_dev *vxlan = netdev_priv(dev);
2936
2937         return vxlan->net;
2938 }
2939
2940 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2941         .kind           = "vxlan",
2942         .maxtype        = IFLA_VXLAN_MAX,
2943         .policy         = vxlan_policy,
2944         .priv_size      = sizeof(struct vxlan_dev),
2945         .setup          = vxlan_setup,
2946         .validate       = vxlan_validate,
2947         .newlink        = vxlan_newlink,
2948         .dellink        = vxlan_dellink,
2949         .get_size       = vxlan_get_size,
2950         .fill_info      = vxlan_fill_info,
2951         .get_link_net   = vxlan_get_link_net,
2952 };
2953
2954 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
2955                                              struct net_device *dev)
2956 {
2957         struct vxlan_dev *vxlan, *next;
2958         LIST_HEAD(list_kill);
2959
2960         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2961                 struct vxlan_rdst *dst = &vxlan->default_dst;
2962
2963                 /* In case we created vxlan device with carrier
2964                  * and we loose the carrier due to module unload
2965                  * we also need to remove vxlan device. In other
2966                  * cases, it's not necessary and remote_ifindex
2967                  * is 0 here, so no matches.
2968                  */
2969                 if (dst->remote_ifindex == dev->ifindex)
2970                         vxlan_dellink(vxlan->dev, &list_kill);
2971         }
2972
2973         unregister_netdevice_many(&list_kill);
2974 }
2975
2976 static int vxlan_lowerdev_event(struct notifier_block *unused,
2977                                 unsigned long event, void *ptr)
2978 {
2979         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2980         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2981
2982         if (event == NETDEV_UNREGISTER)
2983                 vxlan_handle_lowerdev_unregister(vn, dev);
2984
2985         return NOTIFY_DONE;
2986 }
2987
2988 static struct notifier_block vxlan_notifier_block __read_mostly = {
2989         .notifier_call = vxlan_lowerdev_event,
2990 };
2991
2992 static __net_init int vxlan_init_net(struct net *net)
2993 {
2994         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2995         unsigned int h;
2996
2997         INIT_LIST_HEAD(&vn->vxlan_list);
2998         spin_lock_init(&vn->sock_lock);
2999
3000         for (h = 0; h < PORT_HASH_SIZE; ++h)
3001                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3002
3003         return 0;
3004 }
3005
3006 static void __net_exit vxlan_exit_net(struct net *net)
3007 {
3008         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3009         struct vxlan_dev *vxlan, *next;
3010         struct net_device *dev, *aux;
3011         LIST_HEAD(list);
3012
3013         rtnl_lock();
3014         for_each_netdev_safe(net, dev, aux)
3015                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3016                         unregister_netdevice_queue(dev, &list);
3017
3018         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3019                 /* If vxlan->dev is in the same netns, it has already been added
3020                  * to the list by the previous loop.
3021                  */
3022                 if (!net_eq(dev_net(vxlan->dev), net))
3023                         unregister_netdevice_queue(dev, &list);
3024         }
3025
3026         unregister_netdevice_many(&list);
3027         rtnl_unlock();
3028 }
3029
3030 static struct pernet_operations vxlan_net_ops = {
3031         .init = vxlan_init_net,
3032         .exit = vxlan_exit_net,
3033         .id   = &vxlan_net_id,
3034         .size = sizeof(struct vxlan_net),
3035 };
3036
3037 static int __init vxlan_init_module(void)
3038 {
3039         int rc;
3040
3041         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3042         if (!vxlan_wq)
3043                 return -ENOMEM;
3044
3045         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3046
3047         rc = register_pernet_subsys(&vxlan_net_ops);
3048         if (rc)
3049                 goto out1;
3050
3051         rc = register_netdevice_notifier(&vxlan_notifier_block);
3052         if (rc)
3053                 goto out2;
3054
3055         rc = rtnl_link_register(&vxlan_link_ops);
3056         if (rc)
3057                 goto out3;
3058
3059         return 0;
3060 out3:
3061         unregister_netdevice_notifier(&vxlan_notifier_block);
3062 out2:
3063         unregister_pernet_subsys(&vxlan_net_ops);
3064 out1:
3065         destroy_workqueue(vxlan_wq);
3066         return rc;
3067 }
3068 late_initcall(vxlan_init_module);
3069
3070 static void __exit vxlan_cleanup_module(void)
3071 {
3072         rtnl_link_unregister(&vxlan_link_ops);
3073         unregister_netdevice_notifier(&vxlan_notifier_block);
3074         destroy_workqueue(vxlan_wq);
3075         unregister_pernet_subsys(&vxlan_net_ops);
3076         /* rcu_barrier() is called by netns */
3077 }
3078 module_exit(vxlan_cleanup_module);
3079
3080 MODULE_LICENSE("GPL");
3081 MODULE_VERSION(VXLAN_VERSION);
3082 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3083 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3084 MODULE_ALIAS_RTNL_LINK("vxlan");