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         unsigned long     age_interval;
 131         struct timer_list age_timer;
 132         spinlock_t        hash_lock;
 133         unsigned int      addrcnt;
 134         unsigned int      addrmax;
 135
 136         struct hlist_head fdb_head[FDB_HASH_SIZE];
 137 };
 138
 139 /* salt for hash table */
 140 static u32 vxlan_salt __read_mostly;
 141 static struct workqueue_struct *vxlan_wq;
 142
 143 #if IS_ENABLED(CONFIG_IPV6)
 144 static inline
 145 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 146 {
 147         if (a->sa.sa_family != b->sa.sa_family)
 148                 return false;
 149         if (a->sa.sa_family == AF_INET6)
 150                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 151         else
 152                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 153 }
 154
 155 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 156 {
 157         if (ipa->sa.sa_family == AF_INET6)
 158                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 159         else
 160                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 161 }
 162
 163 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 164 {
 165         if (ipa->sa.sa_family == AF_INET6)
 166                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 167         else
 168                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 169 }
 170
 171 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 172 {
 173         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 174                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 175                 ip->sa.sa_family = AF_INET6;
 176                 return 0;
 177         } else if (nla_len(nla) >= sizeof(__be32)) {
 178                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 179                 ip->sa.sa_family = AF_INET;
 180                 return 0;
 181         } else {
 182                 return -EAFNOSUPPORT;
 183         }
 184 }
 185
 186 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 187                               const union vxlan_addr *ip)
 188 {
 189         if (ip->sa.sa_family == AF_INET6)
 190                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 191         else
 192                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 193 }
 194
 195 #else /* !CONFIG_IPV6 */
 196
 197 static inline
 198 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 199 {
 200         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 201 }
 202
 203 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 204 {
 205         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 206 }
 207
 208 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 209 {
 210         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 211 }
 212
 213 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 214 {
 215         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 216                 return -EAFNOSUPPORT;
 217         } else if (nla_len(nla) >= sizeof(__be32)) {
 218                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 219                 ip->sa.sa_family = AF_INET;
 220                 return 0;
 221         } else {
 222                 return -EAFNOSUPPORT;
 223         }
 224 }
 225
 226 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 227                               const union vxlan_addr *ip)
 228 {
 229         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 230 }
 231 #endif
 232
 233 /* Virtual Network hash table head */
 234 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
 235 {
 236         return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
 237 }
 238
 239 /* Socket hash table head */
 240 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 241 {
 242         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 243
 244         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 245 }
 246
 247 /* First remote destination for a forwarding entry.
 248  * Guaranteed to be non-NULL because remotes are never deleted.
 249  */
 250 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 251 {
 252         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 253 }
 254
 255 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 256 {
 257         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 258 }
 259
 260 /* Find VXLAN socket based on network namespace, address family and UDP port
 261  * and enabled unshareable flags.
 262  */
 263 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 264                                           __be16 port, u32 flags)
 265 {
 266         struct vxlan_sock *vs;
 267
 268         flags &= VXLAN_F_RCV_FLAGS;
 269
 270         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 271                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 272                     inet_sk(vs->sock->sk)->sk.sk_family == family &&
 273                     vs->flags == flags)
 274                         return vs;
 275         }
 276         return NULL;
 277 }
 278
 279 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
 280 {
 281         struct vxlan_dev *vxlan;
 282
 283         hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
 284                 if (vxlan->default_dst.remote_vni == id)
 285                         return vxlan;
 286         }
 287
 288         return NULL;
 289 }
 290
 291 /* Look up VNI in a per net namespace table */
 292 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id,
 293                                         sa_family_t family, __be16 port,
 294                                         u32 flags)
 295 {
 296         struct vxlan_sock *vs;
 297
 298         vs = vxlan_find_sock(net, family, port, flags);
 299         if (!vs)
 300                 return NULL;
 301
 302         return vxlan_vs_find_vni(vs, id);
 303 }
 304
 305 /* Fill in neighbour message in skbuff. */
 306 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 307                           const struct vxlan_fdb *fdb,
 308                           u32 portid, u32 seq, int type, unsigned int flags,
 309                           const struct vxlan_rdst *rdst)
 310 {
 311         unsigned long now = jiffies;
 312         struct nda_cacheinfo ci;
 313         struct nlmsghdr *nlh;
 314         struct ndmsg *ndm;
 315         bool send_ip, send_eth;
 316
 317         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 318         if (nlh == NULL)
 319                 return -EMSGSIZE;
 320
 321         ndm = nlmsg_data(nlh);
 322         memset(ndm, 0, sizeof(*ndm));
 323
 324         send_eth = send_ip = true;
 325
 326         if (type == RTM_GETNEIGH) {
 327                 ndm->ndm_family = AF_INET;
 328                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 329                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 330         } else
 331                 ndm->ndm_family = AF_BRIDGE;
 332         ndm->ndm_state = fdb->state;
 333         ndm->ndm_ifindex = vxlan->dev->ifindex;
 334         ndm->ndm_flags = fdb->flags;
 335         ndm->ndm_type = RTN_UNICAST;
 336
 337         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 338             nla_put_s32(skb, NDA_LINK_NETNSID,
 339                         peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
 340                 goto nla_put_failure;
 341
 342         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 343                 goto nla_put_failure;
 344
 345         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 346                 goto nla_put_failure;
 347
 348         if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
 349             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 350                 goto nla_put_failure;
 351         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 352             nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
 353                 goto nla_put_failure;
 354         if (rdst->remote_ifindex &&
 355             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 356                 goto nla_put_failure;
 357
 358         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 359         ci.ndm_confirmed = 0;
 360         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 361         ci.ndm_refcnt    = 0;
 362
 363         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 364                 goto nla_put_failure;
 365
 366         nlmsg_end(skb, nlh);
 367         return 0;
 368
 369 nla_put_failure:
 370         nlmsg_cancel(skb, nlh);
 371         return -EMSGSIZE;
 372 }
 373
 374 static inline size_t vxlan_nlmsg_size(void)
 375 {
 376         return NLMSG_ALIGN(sizeof(struct ndmsg))
 377                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 378                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 379                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 380                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 381                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 382                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 383                 + nla_total_size(sizeof(struct nda_cacheinfo));
 384 }
 385
 386 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 387                              struct vxlan_rdst *rd, int type)
 388 {
 389         struct net *net = dev_net(vxlan->dev);
 390         struct sk_buff *skb;
 391         int err = -ENOBUFS;
 392
 393         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 394         if (skb == NULL)
 395                 goto errout;
 396
 397         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 398         if (err < 0) {
 399                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 400                 WARN_ON(err == -EMSGSIZE);
 401                 kfree_skb(skb);
 402                 goto errout;
 403         }
 404
 405         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 406         return;
 407 errout:
 408         if (err < 0)
 409                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 410 }
 411
 412 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 413 {
 414         struct vxlan_dev *vxlan = netdev_priv(dev);
 415         struct vxlan_fdb f = {
 416                 .state = NUD_STALE,
 417         };
 418         struct vxlan_rdst remote = {
 419                 .remote_ip = *ipa, /* goes to NDA_DST */
 420                 .remote_vni = VXLAN_N_VID,
 421         };
 422
 423         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 424 }
 425
 426 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 427 {
 428         struct vxlan_fdb f = {
 429                 .state = NUD_STALE,
 430         };
 431         struct vxlan_rdst remote = { };
 432
 433         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 434
 435         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 436 }
 437
 438 /* Hash Ethernet address */
 439 static u32 eth_hash(const unsigned char *addr)
 440 {
 441         u64 value = get_unaligned((u64 *)addr);
 442
 443         /* only want 6 bytes */
 444 #ifdef __BIG_ENDIAN
 445         value >>= 16;
 446 #else
 447         value <<= 16;
 448 #endif
 449         return hash_64(value, FDB_HASH_BITS);
 450 }
 451
 452 /* Hash chain to use given mac address */
 453 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 454                                                 const u8 *mac)
 455 {
 456         return &vxlan->fdb_head[eth_hash(mac)];
 457 }
 458
 459 /* Look up Ethernet address in forwarding table */
 460 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 461                                         const u8 *mac)
 462 {
 463         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 464         struct vxlan_fdb *f;
 465
 466         hlist_for_each_entry_rcu(f, head, hlist) {
 467                 if (ether_addr_equal(mac, f->eth_addr))
 468                         return f;
 469         }
 470
 471         return NULL;
 472 }
 473
 474 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 475                                         const u8 *mac)
 476 {
 477         struct vxlan_fdb *f;
 478
 479         f = __vxlan_find_mac(vxlan, mac);
 480         if (f)
 481                 f->used = jiffies;
 482
 483         return f;
 484 }
 485
 486 /* caller should hold vxlan->hash_lock */
 487 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 488                                               union vxlan_addr *ip, __be16 port,
 489                                               __u32 vni, __u32 ifindex)
 490 {
 491         struct vxlan_rdst *rd;
 492
 493         list_for_each_entry(rd, &f->remotes, list) {
 494                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 495                     rd->remote_port == port &&
 496                     rd->remote_vni == vni &&
 497                     rd->remote_ifindex == ifindex)
 498                         return rd;
 499         }
 500
 501         return NULL;
 502 }
 503
 504 /* Replace destination of unicast mac */
 505 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 506                              union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
 507 {
 508         struct vxlan_rdst *rd;
 509
 510         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 511         if (rd)
 512                 return 0;
 513
 514         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 515         if (!rd)
 516                 return 0;
 517         rd->remote_ip = *ip;
 518         rd->remote_port = port;
 519         rd->remote_vni = vni;
 520         rd->remote_ifindex = ifindex;
 521         return 1;
 522 }
 523
 524 /* Add/update destinations for multicast */
 525 static int vxlan_fdb_append(struct vxlan_fdb *f,
 526                             union vxlan_addr *ip, __be16 port, __u32 vni,
 527                             __u32 ifindex, struct vxlan_rdst **rdp)
 528 {
 529         struct vxlan_rdst *rd;
 530
 531         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 532         if (rd)
 533                 return 0;
 534
 535         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 536         if (rd == NULL)
 537                 return -ENOBUFS;
 538         rd->remote_ip = *ip;
 539         rd->remote_port = port;
 540         rd->remote_vni = vni;
 541         rd->remote_ifindex = ifindex;
 542
 543         list_add_tail_rcu(&rd->list, &f->remotes);
 544
 545         *rdp = rd;
 546         return 1;
 547 }
 548
 549 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 550                                           unsigned int off,
 551                                           struct vxlanhdr *vh, size_t hdrlen,
 552                                           u32 data, struct gro_remcsum *grc,
 553                                           bool nopartial)
 554 {
 555         size_t start, offset, plen;
 556
 557         if (skb->remcsum_offload)
 558                 return NULL;
 559
 560         if (!NAPI_GRO_CB(skb)->csum_valid)
 561                 return NULL;
 562
 563         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
 564         offset = start + ((data & VXLAN_RCO_UDP) ?
 565                           offsetof(struct udphdr, check) :
 566                           offsetof(struct tcphdr, check));
 567
 568         plen = hdrlen + offset + sizeof(u16);
 569
 570         /* Pull checksum that will be written */
 571         if (skb_gro_header_hard(skb, off + plen)) {
 572                 vh = skb_gro_header_slow(skb, off + plen, off);
 573                 if (!vh)
 574                         return NULL;
 575         }
 576
 577         skb_gro_remcsum_process(skb, (void *)vh + hdrlen,
 578                                 start, offset, grc, nopartial);
 579
 580         skb->remcsum_offload = 1;
 581
 582         return vh;
 583 }
 584
 585 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
 586                                           struct sk_buff *skb,
 587                                           struct udp_offload *uoff)
 588 {
 589         struct sk_buff *p, **pp = NULL;
 590         struct vxlanhdr *vh, *vh2;
 591         unsigned int hlen, off_vx;
 592         int flush = 1;
 593         struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
 594                                              udp_offloads);
 595         u32 flags;
 596         struct gro_remcsum grc;
 597
 598         skb_gro_remcsum_init(&grc);
 599
 600         off_vx = skb_gro_offset(skb);
 601         hlen = off_vx + sizeof(*vh);
 602         vh   = skb_gro_header_fast(skb, off_vx);
 603         if (skb_gro_header_hard(skb, hlen)) {
 604                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 605                 if (unlikely(!vh))
 606                         goto out;
 607         }
 608
 609         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 610         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 611
 612         flags = ntohl(vh->vx_flags);
 613
 614         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 615                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 616                                        ntohl(vh->vx_vni), &grc,
 617                                        !!(vs->flags &
 618                                           VXLAN_F_REMCSUM_NOPARTIAL));
 619
 620                 if (!vh)
 621                         goto out;
 622         }
 623
 624         flush = 0;
 625
 626         for (p = *head; p; p = p->next) {
 627                 if (!NAPI_GRO_CB(p)->same_flow)
 628                         continue;
 629
 630                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 631                 if (vh->vx_flags != vh2->vx_flags ||
 632                     vh->vx_vni != vh2->vx_vni) {
 633                         NAPI_GRO_CB(p)->same_flow = 0;
 634                         continue;
 635                 }
 636         }
 637
 638         pp = eth_gro_receive(head, skb);
 639
 640 out:
 641         skb_gro_remcsum_cleanup(skb, &grc);
 642         NAPI_GRO_CB(skb)->flush |= flush;
 643
 644         return pp;
 645 }
 646
 647 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
 648                               struct udp_offload *uoff)
 649 {
 650         udp_tunnel_gro_complete(skb, nhoff);
 651
 652         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 653 }
 654
 655 /* Notify netdevs that UDP port started listening */
 656 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 657 {
 658         struct net_device *dev;
 659         struct sock *sk = vs->sock->sk;
 660         struct net *net = sock_net(sk);
 661         sa_family_t sa_family = sk->sk_family;
 662         __be16 port = inet_sk(sk)->inet_sport;
 663         int err;
 664
 665         if (sa_family == AF_INET) {
 666                 err = udp_add_offload(&vs->udp_offloads);
 667                 if (err)
 668                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 669         }
 670
 671         rcu_read_lock();
 672         for_each_netdev_rcu(net, dev) {
 673                 if (dev->netdev_ops->ndo_add_vxlan_port)
 674                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 675                                                             port);
 676         }
 677         rcu_read_unlock();
 678 }
 679
 680 /* Notify netdevs that UDP port is no more listening */
 681 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 682 {
 683         struct net_device *dev;
 684         struct sock *sk = vs->sock->sk;
 685         struct net *net = sock_net(sk);
 686         sa_family_t sa_family = sk->sk_family;
 687         __be16 port = inet_sk(sk)->inet_sport;
 688
 689         rcu_read_lock();
 690         for_each_netdev_rcu(net, dev) {
 691                 if (dev->netdev_ops->ndo_del_vxlan_port)
 692                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 693                                                             port);
 694         }
 695         rcu_read_unlock();
 696
 697         if (sa_family == AF_INET)
 698                 udp_del_offload(&vs->udp_offloads);
 699 }
 700
 701 /* Add new entry to forwarding table -- assumes lock held */
 702 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 703                             const u8 *mac, union vxlan_addr *ip,
 704                             __u16 state, __u16 flags,
 705                             __be16 port, __u32 vni, __u32 ifindex,
 706                             __u8 ndm_flags)
 707 {
 708         struct vxlan_rdst *rd = NULL;
 709         struct vxlan_fdb *f;
 710         int notify = 0;
 711
 712         f = __vxlan_find_mac(vxlan, mac);
 713         if (f) {
 714                 if (flags & NLM_F_EXCL) {
 715                         netdev_dbg(vxlan->dev,
 716                                    "lost race to create %pM\n", mac);
 717                         return -EEXIST;
 718                 }
 719                 if (f->state != state) {
 720                         f->state = state;
 721                         f->updated = jiffies;
 722                         notify = 1;
 723                 }
 724                 if (f->flags != ndm_flags) {
 725                         f->flags = ndm_flags;
 726                         f->updated = jiffies;
 727                         notify = 1;
 728                 }
 729                 if ((flags & NLM_F_REPLACE)) {
 730                         /* Only change unicasts */
 731                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 732                              is_zero_ether_addr(f->eth_addr))) {
 733                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 734                                                            ifindex);
 735                         } else
 736                                 return -EOPNOTSUPP;
 737                 }
 738                 if ((flags & NLM_F_APPEND) &&
 739                     (is_multicast_ether_addr(f->eth_addr) ||
 740                      is_zero_ether_addr(f->eth_addr))) {
 741                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 742                                                   &rd);
 743
 744                         if (rc < 0)
 745                                 return rc;
 746                         notify |= rc;
 747                 }
 748         } else {
 749                 if (!(flags & NLM_F_CREATE))
 750                         return -ENOENT;
 751
 752                 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
 753                         return -ENOSPC;
 754
 755                 /* Disallow replace to add a multicast entry */
 756                 if ((flags & NLM_F_REPLACE) &&
 757                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 758                         return -EOPNOTSUPP;
 759
 760                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 761                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 762                 if (!f)
 763                         return -ENOMEM;
 764
 765                 notify = 1;
 766                 f->state = state;
 767                 f->flags = ndm_flags;
 768                 f->updated = f->used = jiffies;
 769                 INIT_LIST_HEAD(&f->remotes);
 770                 memcpy(f->eth_addr, mac, ETH_ALEN);
 771
 772                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 773
 774                 ++vxlan->addrcnt;
 775                 hlist_add_head_rcu(&f->hlist,
 776                                    vxlan_fdb_head(vxlan, mac));
 777         }
 778
 779         if (notify) {
 780                 if (rd == NULL)
 781                         rd = first_remote_rtnl(f);
 782                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 783         }
 784
 785         return 0;
 786 }
 787
 788 static void vxlan_fdb_free(struct rcu_head *head)
 789 {
 790         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 791         struct vxlan_rdst *rd, *nd;
 792
 793         list_for_each_entry_safe(rd, nd, &f->remotes, list)
 794                 kfree(rd);
 795         kfree(f);
 796 }
 797
 798 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 799 {
 800         netdev_dbg(vxlan->dev,
 801                     "delete %pM\n", f->eth_addr);
 802
 803         --vxlan->addrcnt;
 804         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 805
 806         hlist_del_rcu(&f->hlist);
 807         call_rcu(&f->rcu, vxlan_fdb_free);
 808 }
 809
 810 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 811                            union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
 812 {
 813         struct net *net = dev_net(vxlan->dev);
 814         int err;
 815
 816         if (tb[NDA_DST]) {
 817                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 818                 if (err)
 819                         return err;
 820         } else {
 821                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 822                 if (remote->sa.sa_family == AF_INET) {
 823                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 824                         ip->sa.sa_family = AF_INET;
 825 #if IS_ENABLED(CONFIG_IPV6)
 826                 } else {
 827                         ip->sin6.sin6_addr = in6addr_any;
 828                         ip->sa.sa_family = AF_INET6;
 829 #endif
 830                 }
 831         }
 832
 833         if (tb[NDA_PORT]) {
 834                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 835                         return -EINVAL;
 836                 *port = nla_get_be16(tb[NDA_PORT]);
 837         } else {
 838                 *port = vxlan->dst_port;
 839         }
 840
 841         if (tb[NDA_VNI]) {
 842                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 843                         return -EINVAL;
 844                 *vni = nla_get_u32(tb[NDA_VNI]);
 845         } else {
 846                 *vni = vxlan->default_dst.remote_vni;
 847         }
 848
 849         if (tb[NDA_IFINDEX]) {
 850                 struct net_device *tdev;
 851
 852                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 853                         return -EINVAL;
 854                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 855                 tdev = __dev_get_by_index(net, *ifindex);
 856                 if (!tdev)
 857                         return -EADDRNOTAVAIL;
 858         } else {
 859                 *ifindex = 0;
 860         }
 861
 862         return 0;
 863 }
 864
 865 /* Add static entry (via netlink) */
 866 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 867                          struct net_device *dev,
 868                          const unsigned char *addr, u16 vid, u16 flags)
 869 {
 870         struct vxlan_dev *vxlan = netdev_priv(dev);
 871         /* struct net *net = dev_net(vxlan->dev); */
 872         union vxlan_addr ip;
 873         __be16 port;
 874         u32 vni, ifindex;
 875         int err;
 876
 877         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 878                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 879                         ndm->ndm_state);
 880                 return -EINVAL;
 881         }
 882
 883         if (tb[NDA_DST] == NULL)
 884                 return -EINVAL;
 885
 886         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 887         if (err)
 888                 return err;
 889
 890         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 891                 return -EAFNOSUPPORT;
 892
 893         spin_lock_bh(&vxlan->hash_lock);
 894         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 895                                port, vni, ifindex, ndm->ndm_flags);
 896         spin_unlock_bh(&vxlan->hash_lock);
 897
 898         return err;
 899 }
 900
 901 /* Delete entry (via netlink) */
 902 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 903                             struct net_device *dev,
 904                             const unsigned char *addr, u16 vid)
 905 {
 906         struct vxlan_dev *vxlan = netdev_priv(dev);
 907         struct vxlan_fdb *f;
 908         struct vxlan_rdst *rd = NULL;
 909         union vxlan_addr ip;
 910         __be16 port;
 911         u32 vni, ifindex;
 912         int err;
 913
 914         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 915         if (err)
 916                 return err;
 917
 918         err = -ENOENT;
 919
 920         spin_lock_bh(&vxlan->hash_lock);
 921         f = vxlan_find_mac(vxlan, addr);
 922         if (!f)
 923                 goto out;
 924
 925         if (!vxlan_addr_any(&ip)) {
 926                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 927                 if (!rd)
 928                         goto out;
 929         }
 930
 931         err = 0;
 932
 933         /* remove a destination if it's not the only one on the list,
 934          * otherwise destroy the fdb entry
 935          */
 936         if (rd && !list_is_singular(&f->remotes)) {
 937                 list_del_rcu(&rd->list);
 938                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 939                 kfree_rcu(rd, rcu);
 940                 goto out;
 941         }
 942
 943         vxlan_fdb_destroy(vxlan, f);
 944
 945 out:
 946         spin_unlock_bh(&vxlan->hash_lock);
 947
 948         return err;
 949 }
 950
 951 /* Dump forwarding table */
 952 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 953                           struct net_device *dev,
 954                           struct net_device *filter_dev, int idx)
 955 {
 956         struct vxlan_dev *vxlan = netdev_priv(dev);
 957         unsigned int h;
 958
 959         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 960                 struct vxlan_fdb *f;
 961                 int err;
 962
 963                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 964                         struct vxlan_rdst *rd;
 965
 966                         if (idx < cb->args[0])
 967                                 goto skip;
 968
 969                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 970                                 err = vxlan_fdb_info(skb, vxlan, f,
 971                                                      NETLINK_CB(cb->skb).portid,
 972                                                      cb->nlh->nlmsg_seq,
 973                                                      RTM_NEWNEIGH,
 974                                                      NLM_F_MULTI, rd);
 975                                 if (err < 0)
 976                                         goto out;
 977                         }
 978 skip:
 979                         ++idx;
 980                 }
 981         }
 982 out:
 983         return idx;
 984 }
 985
 986 /* Watch incoming packets to learn mapping between Ethernet address
 987  * and Tunnel endpoint.
 988  * Return true if packet is bogus and should be dropped.
 989  */
 990 static bool vxlan_snoop(struct net_device *dev,
 991                         union vxlan_addr *src_ip, const u8 *src_mac)
 992 {
 993         struct vxlan_dev *vxlan = netdev_priv(dev);
 994         struct vxlan_fdb *f;
 995
 996         f = vxlan_find_mac(vxlan, src_mac);
 997         if (likely(f)) {
 998                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 999
1000                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
1001                         return false;
1002
1003                 /* Don't migrate static entries, drop packets */
1004                 if (f->state & NUD_NOARP)
1005                         return true;
1006
1007                 if (net_ratelimit())
1008                         netdev_info(dev,
1009                                     "%pM migrated from %pIS to %pIS\n",
1010                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1011
1012                 rdst->remote_ip = *src_ip;
1013                 f->updated = jiffies;
1014                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1015         } else {
1016                 /* learned new entry */
1017                 spin_lock(&vxlan->hash_lock);
1018
1019                 /* close off race between vxlan_flush and incoming packets */
1020                 if (netif_running(dev))
1021                         vxlan_fdb_create(vxlan, src_mac, src_ip,
1022                                          NUD_REACHABLE,
1023                                          NLM_F_EXCL|NLM_F_CREATE,
1024                                          vxlan->dst_port,
1025                                          vxlan->default_dst.remote_vni,
1026                                          0, NTF_SELF);
1027                 spin_unlock(&vxlan->hash_lock);
1028         }
1029
1030         return false;
1031 }
1032
1033 /* See if multicast group is already in use by other ID */
1034 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1035 {
1036         struct vxlan_dev *vxlan;
1037
1038         /* The vxlan_sock is only used by dev, leaving group has
1039          * no effect on other vxlan devices.
1040          */
1041         if (atomic_read(&dev->vn_sock->refcnt) == 1)
1042                 return false;
1043
1044         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1045                 if (!netif_running(vxlan->dev) || vxlan == dev)
1046                         continue;
1047
1048                 if (vxlan->vn_sock != dev->vn_sock)
1049                         continue;
1050
1051                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1052                                       &dev->default_dst.remote_ip))
1053                         continue;
1054
1055                 if (vxlan->default_dst.remote_ifindex !=
1056                     dev->default_dst.remote_ifindex)
1057                         continue;
1058
1059                 return true;
1060         }
1061
1062         return false;
1063 }
1064
1065 void vxlan_sock_release(struct vxlan_sock *vs)
1066 {
1067         struct sock *sk = vs->sock->sk;
1068         struct net *net = sock_net(sk);
1069         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1070
1071         if (!atomic_dec_and_test(&vs->refcnt))
1072                 return;
1073
1074         spin_lock(&vn->sock_lock);
1075         hlist_del_rcu(&vs->hlist);
1076         vxlan_notify_del_rx_port(vs);
1077         spin_unlock(&vn->sock_lock);
1078
1079         queue_work(vxlan_wq, &vs->del_work);
1080 }
1081 EXPORT_SYMBOL_GPL(vxlan_sock_release);
1082
1083 /* Update multicast group membership when first VNI on
1084  * multicast address is brought up
1085  */
1086 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1087 {
1088         struct vxlan_sock *vs = vxlan->vn_sock;
1089         struct sock *sk = vs->sock->sk;
1090         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1091         int ifindex = vxlan->default_dst.remote_ifindex;
1092         int ret = -EINVAL;
1093
1094         lock_sock(sk);
1095         if (ip->sa.sa_family == AF_INET) {
1096                 struct ip_mreqn mreq = {
1097                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1098                         .imr_ifindex            = ifindex,
1099                 };
1100
1101                 ret = ip_mc_join_group(sk, &mreq);
1102 #if IS_ENABLED(CONFIG_IPV6)
1103         } else {
1104                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1105                                                    &ip->sin6.sin6_addr);
1106 #endif
1107         }
1108         release_sock(sk);
1109
1110         return ret;
1111 }
1112
1113 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1114 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1115 {
1116         struct vxlan_sock *vs = vxlan->vn_sock;
1117         struct sock *sk = vs->sock->sk;
1118         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1119         int ifindex = vxlan->default_dst.remote_ifindex;
1120         int ret = -EINVAL;
1121
1122         lock_sock(sk);
1123         if (ip->sa.sa_family == AF_INET) {
1124                 struct ip_mreqn mreq = {
1125                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1126                         .imr_ifindex            = ifindex,
1127                 };
1128
1129                 ret = ip_mc_leave_group(sk, &mreq);
1130 #if IS_ENABLED(CONFIG_IPV6)
1131         } else {
1132                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1133                                                    &ip->sin6.sin6_addr);
1134 #endif
1135         }
1136         release_sock(sk);
1137
1138         return ret;
1139 }
1140
1141 static struct vxlanhdr *vxlan_remcsum(struct sk_buff *skb, struct vxlanhdr *vh,
1142                                       size_t hdrlen, u32 data, bool nopartial)
1143 {
1144         size_t start, offset, plen;
1145
1146         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
1147         offset = start + ((data & VXLAN_RCO_UDP) ?
1148                           offsetof(struct udphdr, check) :
1149                           offsetof(struct tcphdr, check));
1150
1151         plen = hdrlen + offset + sizeof(u16);
1152
1153         if (!pskb_may_pull(skb, plen))
1154                 return NULL;
1155
1156         vh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1157
1158         skb_remcsum_process(skb, (void *)vh + hdrlen, start, offset,
1159                             nopartial);
1160
1161         return vh;
1162 }
1163
1164 /* Callback from net/ipv4/udp.c to receive packets */
1165 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1166 {
1167         struct vxlan_sock *vs;
1168         struct vxlanhdr *vxh;
1169         u32 flags, vni;
1170         struct vxlan_metadata md = {0};
1171
1172         /* Need Vxlan and inner Ethernet header to be present */
1173         if (!pskb_may_pull(skb, VXLAN_HLEN))
1174                 goto error;
1175
1176         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1177         flags = ntohl(vxh->vx_flags);
1178         vni = ntohl(vxh->vx_vni);
1179
1180         if (flags & VXLAN_HF_VNI) {
1181                 flags &= ~VXLAN_HF_VNI;
1182         } else {
1183                 /* VNI flag always required to be set */
1184                 goto bad_flags;
1185         }
1186
1187         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1188                 goto drop;
1189         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1190
1191         vs = rcu_dereference_sk_user_data(sk);
1192         if (!vs)
1193                 goto drop;
1194
1195         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1196                 vxh = vxlan_remcsum(skb, vxh, sizeof(struct vxlanhdr), vni,
1197                                     !!(vs->flags & VXLAN_F_REMCSUM_NOPARTIAL));
1198                 if (!vxh)
1199                         goto drop;
1200
1201                 flags &= ~VXLAN_HF_RCO;
1202                 vni &= VXLAN_VNI_MASK;
1203         }
1204
1205         /* For backwards compatibility, only allow reserved fields to be
1206          * used by VXLAN extensions if explicitly requested.
1207          */
1208         if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1209                 struct vxlanhdr_gbp *gbp;
1210
1211                 gbp = (struct vxlanhdr_gbp *)vxh;
1212                 md.gbp = ntohs(gbp->policy_id);
1213
1214                 if (gbp->dont_learn)
1215                         md.gbp |= VXLAN_GBP_DONT_LEARN;
1216
1217                 if (gbp->policy_applied)
1218                         md.gbp |= VXLAN_GBP_POLICY_APPLIED;
1219
1220                 flags &= ~VXLAN_GBP_USED_BITS;
1221         }
1222
1223         if (flags || vni & ~VXLAN_VNI_MASK) {
1224                 /* If there are any unprocessed flags remaining treat
1225                  * this as a malformed packet. This behavior diverges from
1226                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1227                  * in reserved fields are to be ignored. The approach here
1228                  * maintains compatibility with previous stack code, and also
1229                  * is more robust and provides a little more security in
1230                  * adding extensions to VXLAN.
1231                  */
1232
1233                 goto bad_flags;
1234         }
1235
1236         md.vni = vxh->vx_vni;
1237         vs->rcv(vs, skb, &md);
1238         return 0;
1239
1240 drop:
1241         /* Consume bad packet */
1242         kfree_skb(skb);
1243         return 0;
1244
1245 bad_flags:
1246         netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1247                    ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1248
1249 error:
1250         /* Return non vxlan pkt */
1251         return 1;
1252 }
1253
1254 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1255                       struct vxlan_metadata *md)
1256 {
1257         struct iphdr *oip = NULL;
1258         struct ipv6hdr *oip6 = NULL;
1259         struct vxlan_dev *vxlan;
1260         struct pcpu_sw_netstats *stats;
1261         union vxlan_addr saddr;
1262         __u32 vni;
1263         int err = 0;
1264         union vxlan_addr *remote_ip;
1265
1266         vni = ntohl(md->vni) >> 8;
1267         /* Is this VNI defined? */
1268         vxlan = vxlan_vs_find_vni(vs, vni);
1269         if (!vxlan)
1270                 goto drop;
1271
1272         remote_ip = &vxlan->default_dst.remote_ip;
1273         skb_reset_mac_header(skb);
1274         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1275         skb->protocol = eth_type_trans(skb, vxlan->dev);
1276         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1277
1278         /* Ignore packet loops (and multicast echo) */
1279         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1280                 goto drop;
1281
1282         /* Re-examine inner Ethernet packet */
1283         if (remote_ip->sa.sa_family == AF_INET) {
1284                 oip = ip_hdr(skb);
1285                 saddr.sin.sin_addr.s_addr = oip->saddr;
1286                 saddr.sa.sa_family = AF_INET;
1287 #if IS_ENABLED(CONFIG_IPV6)
1288         } else {
1289                 oip6 = ipv6_hdr(skb);
1290                 saddr.sin6.sin6_addr = oip6->saddr;
1291                 saddr.sa.sa_family = AF_INET6;
1292 #endif
1293         }
1294
1295         if ((vxlan->flags & VXLAN_F_LEARN) &&
1296             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1297                 goto drop;
1298
1299         skb_reset_network_header(skb);
1300         skb->mark = md->gbp;
1301
1302         if (oip6)
1303                 err = IP6_ECN_decapsulate(oip6, skb);
1304         if (oip)
1305                 err = IP_ECN_decapsulate(oip, skb);
1306
1307         if (unlikely(err)) {
1308                 if (log_ecn_error) {
1309                         if (oip6)
1310                                 net_info_ratelimited("non-ECT from %pI6\n",
1311                                                      &oip6->saddr);
1312                         if (oip)
1313                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1314                                                      &oip->saddr, oip->tos);
1315                 }
1316                 if (err > 1) {
1317                         ++vxlan->dev->stats.rx_frame_errors;
1318                         ++vxlan->dev->stats.rx_errors;
1319                         goto drop;
1320                 }
1321         }
1322
1323         stats = this_cpu_ptr(vxlan->dev->tstats);
1324         u64_stats_update_begin(&stats->syncp);
1325         stats->rx_packets++;
1326         stats->rx_bytes += skb->len;
1327         u64_stats_update_end(&stats->syncp);
1328
1329         netif_rx(skb);
1330
1331         return;
1332 drop:
1333         /* Consume bad packet */
1334         kfree_skb(skb);
1335 }
1336
1337 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1338 {
1339         struct vxlan_dev *vxlan = netdev_priv(dev);
1340         struct arphdr *parp;
1341         u8 *arpptr, *sha;
1342         __be32 sip, tip;
1343         struct neighbour *n;
1344
1345         if (dev->flags & IFF_NOARP)
1346                 goto out;
1347
1348         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1349                 dev->stats.tx_dropped++;
1350                 goto out;
1351         }
1352         parp = arp_hdr(skb);
1353
1354         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1355              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1356             parp->ar_pro != htons(ETH_P_IP) ||
1357             parp->ar_op != htons(ARPOP_REQUEST) ||
1358             parp->ar_hln != dev->addr_len ||
1359             parp->ar_pln != 4)
1360                 goto out;
1361         arpptr = (u8 *)parp + sizeof(struct arphdr);
1362         sha = arpptr;
1363         arpptr += dev->addr_len;        /* sha */
1364         memcpy(&sip, arpptr, sizeof(sip));
1365         arpptr += sizeof(sip);
1366         arpptr += dev->addr_len;        /* tha */
1367         memcpy(&tip, arpptr, sizeof(tip));
1368
1369         if (ipv4_is_loopback(tip) ||
1370             ipv4_is_multicast(tip))
1371                 goto out;
1372
1373         n = neigh_lookup(&arp_tbl, &tip, dev);
1374
1375         if (n) {
1376                 struct vxlan_fdb *f;
1377                 struct sk_buff  *reply;
1378
1379                 if (!(n->nud_state & NUD_CONNECTED)) {
1380                         neigh_release(n);
1381                         goto out;
1382                 }
1383
1384                 f = vxlan_find_mac(vxlan, n->ha);
1385                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1386                         /* bridge-local neighbor */
1387                         neigh_release(n);
1388                         goto out;
1389                 }
1390
1391                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1392                                 n->ha, sha);
1393
1394                 neigh_release(n);
1395
1396                 if (reply == NULL)
1397                         goto out;
1398
1399                 skb_reset_mac_header(reply);
1400                 __skb_pull(reply, skb_network_offset(reply));
1401                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1402                 reply->pkt_type = PACKET_HOST;
1403
1404                 if (netif_rx_ni(reply) == NET_RX_DROP)
1405                         dev->stats.rx_dropped++;
1406         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1407                 union vxlan_addr ipa = {
1408                         .sin.sin_addr.s_addr = tip,
1409                         .sin.sin_family = AF_INET,
1410                 };
1411
1412                 vxlan_ip_miss(dev, &ipa);
1413         }
1414 out:
1415         consume_skb(skb);
1416         return NETDEV_TX_OK;
1417 }
1418
1419 #if IS_ENABLED(CONFIG_IPV6)
1420 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1421         struct neighbour *n, bool isrouter)
1422 {
1423         struct net_device *dev = request->dev;
1424         struct sk_buff *reply;
1425         struct nd_msg *ns, *na;
1426         struct ipv6hdr *pip6;
1427         u8 *daddr;
1428         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1429         int ns_olen;
1430         int i, len;
1431
1432         if (dev == NULL)
1433                 return NULL;
1434
1435         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1436                 sizeof(*na) + na_olen + dev->needed_tailroom;
1437         reply = alloc_skb(len, GFP_ATOMIC);
1438         if (reply == NULL)
1439                 return NULL;
1440
1441         reply->protocol = htons(ETH_P_IPV6);
1442         reply->dev = dev;
1443         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1444         skb_push(reply, sizeof(struct ethhdr));
1445         skb_set_mac_header(reply, 0);
1446
1447         ns = (struct nd_msg *)skb_transport_header(request);
1448
1449         daddr = eth_hdr(request)->h_source;
1450         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1451         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1452                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1453                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1454                         break;
1455                 }
1456         }
1457
1458         /* Ethernet header */
1459         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1460         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1461         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1462         reply->protocol = htons(ETH_P_IPV6);
1463
1464         skb_pull(reply, sizeof(struct ethhdr));
1465         skb_set_network_header(reply, 0);
1466         skb_put(reply, sizeof(struct ipv6hdr));
1467
1468         /* IPv6 header */
1469
1470         pip6 = ipv6_hdr(reply);
1471         memset(pip6, 0, sizeof(struct ipv6hdr));
1472         pip6->version = 6;
1473         pip6->priority = ipv6_hdr(request)->priority;
1474         pip6->nexthdr = IPPROTO_ICMPV6;
1475         pip6->hop_limit = 255;
1476         pip6->daddr = ipv6_hdr(request)->saddr;
1477         pip6->saddr = *(struct in6_addr *)n->primary_key;
1478
1479         skb_pull(reply, sizeof(struct ipv6hdr));
1480         skb_set_transport_header(reply, 0);
1481
1482         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1483
1484         /* Neighbor Advertisement */
1485         memset(na, 0, sizeof(*na)+na_olen);
1486         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1487         na->icmph.icmp6_router = isrouter;
1488         na->icmph.icmp6_override = 1;
1489         na->icmph.icmp6_solicited = 1;
1490         na->target = ns->target;
1491         ether_addr_copy(&na->opt[2], n->ha);
1492         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1493         na->opt[1] = na_olen >> 3;
1494
1495         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1496                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1497                 csum_partial(na, sizeof(*na)+na_olen, 0));
1498
1499         pip6->payload_len = htons(sizeof(*na)+na_olen);
1500
1501         skb_push(reply, sizeof(struct ipv6hdr));
1502
1503         reply->ip_summed = CHECKSUM_UNNECESSARY;
1504
1505         return reply;
1506 }
1507
1508 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1509 {
1510         struct vxlan_dev *vxlan = netdev_priv(dev);
1511         struct nd_msg *msg;
1512         const struct ipv6hdr *iphdr;
1513         const struct in6_addr *saddr, *daddr;
1514         struct neighbour *n;
1515         struct inet6_dev *in6_dev;
1516
1517         in6_dev = __in6_dev_get(dev);
1518         if (!in6_dev)
1519                 goto out;
1520
1521         iphdr = ipv6_hdr(skb);
1522         saddr = &iphdr->saddr;
1523         daddr = &iphdr->daddr;
1524
1525         msg = (struct nd_msg *)skb_transport_header(skb);
1526         if (msg->icmph.icmp6_code != 0 ||
1527             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1528                 goto out;
1529
1530         if (ipv6_addr_loopback(daddr) ||
1531             ipv6_addr_is_multicast(&msg->target))
1532                 goto out;
1533
1534         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1535
1536         if (n) {
1537                 struct vxlan_fdb *f;
1538                 struct sk_buff *reply;
1539
1540                 if (!(n->nud_state & NUD_CONNECTED)) {
1541                         neigh_release(n);
1542                         goto out;
1543                 }
1544
1545                 f = vxlan_find_mac(vxlan, n->ha);
1546                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1547                         /* bridge-local neighbor */
1548                         neigh_release(n);
1549                         goto out;
1550                 }
1551
1552                 reply = vxlan_na_create(skb, n,
1553                                         !!(f ? f->flags & NTF_ROUTER : 0));
1554
1555                 neigh_release(n);
1556
1557                 if (reply == NULL)
1558                         goto out;
1559
1560                 if (netif_rx_ni(reply) == NET_RX_DROP)
1561                         dev->stats.rx_dropped++;
1562
1563         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1564                 union vxlan_addr ipa = {
1565                         .sin6.sin6_addr = msg->target,
1566                         .sin6.sin6_family = AF_INET6,
1567                 };
1568
1569                 vxlan_ip_miss(dev, &ipa);
1570         }
1571
1572 out:
1573         consume_skb(skb);
1574         return NETDEV_TX_OK;
1575 }
1576 #endif
1577
1578 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1579 {
1580         struct vxlan_dev *vxlan = netdev_priv(dev);
1581         struct neighbour *n;
1582
1583         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1584                 return false;
1585
1586         n = NULL;
1587         switch (ntohs(eth_hdr(skb)->h_proto)) {
1588         case ETH_P_IP:
1589         {
1590                 struct iphdr *pip;
1591
1592                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1593                         return false;
1594                 pip = ip_hdr(skb);
1595                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1596                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1597                         union vxlan_addr ipa = {
1598                                 .sin.sin_addr.s_addr = pip->daddr,
1599                                 .sin.sin_family = AF_INET,
1600                         };
1601
1602                         vxlan_ip_miss(dev, &ipa);
1603                         return false;
1604                 }
1605
1606                 break;
1607         }
1608 #if IS_ENABLED(CONFIG_IPV6)
1609         case ETH_P_IPV6:
1610         {
1611                 struct ipv6hdr *pip6;
1612
1613                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1614                         return false;
1615                 pip6 = ipv6_hdr(skb);
1616                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1617                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1618                         union vxlan_addr ipa = {
1619                                 .sin6.sin6_addr = pip6->daddr,
1620                                 .sin6.sin6_family = AF_INET6,
1621                         };
1622
1623                         vxlan_ip_miss(dev, &ipa);
1624                         return false;
1625                 }
1626
1627                 break;
1628         }
1629 #endif
1630         default:
1631                 return false;
1632         }
1633
1634         if (n) {
1635                 bool diff;
1636
1637                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1638                 if (diff) {
1639                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1640                                 dev->addr_len);
1641                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1642                 }
1643                 neigh_release(n);
1644                 return diff;
1645         }
1646
1647         return false;
1648 }
1649
1650 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1651                                 struct vxlan_metadata *md)
1652 {
1653         struct vxlanhdr_gbp *gbp;
1654
1655         if (!md->gbp)
1656                 return;
1657
1658         gbp = (struct vxlanhdr_gbp *)vxh;
1659         vxh->vx_flags |= htonl(VXLAN_HF_GBP);
1660
1661         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1662                 gbp->dont_learn = 1;
1663
1664         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1665                 gbp->policy_applied = 1;
1666
1667         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1668 }
1669
1670 #if IS_ENABLED(CONFIG_IPV6)
1671 static int vxlan6_xmit_skb(struct dst_entry *dst, struct sock *sk,
1672                            struct sk_buff *skb,
1673                            struct net_device *dev, struct in6_addr *saddr,
1674                            struct in6_addr *daddr, __u8 prio, __u8 ttl,
1675                            __be16 src_port, __be16 dst_port,
1676                            struct vxlan_metadata *md, bool xnet, u32 vxflags)
1677 {
1678         struct vxlanhdr *vxh;
1679         int min_headroom;
1680         int err;
1681         bool udp_sum = !(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX);
1682         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1683         u16 hdrlen = sizeof(struct vxlanhdr);
1684
1685         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1686             skb->ip_summed == CHECKSUM_PARTIAL) {
1687                 int csum_start = skb_checksum_start_offset(skb);
1688
1689                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1690                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1691                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1692                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1693                         udp_sum = false;
1694                         type |= SKB_GSO_TUNNEL_REMCSUM;
1695                 }
1696         }
1697
1698         skb_scrub_packet(skb, xnet);
1699
1700         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1701                         + VXLAN_HLEN + sizeof(struct ipv6hdr)
1702                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1703
1704         /* Need space for new headers (invalidates iph ptr) */
1705         err = skb_cow_head(skb, min_headroom);
1706         if (unlikely(err)) {
1707                 kfree_skb(skb);
1708                 goto err;
1709         }
1710
1711         skb = vlan_hwaccel_push_inside(skb);
1712         if (WARN_ON(!skb)) {
1713                 err = -ENOMEM;
1714                 goto err;
1715         }
1716
1717         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1718         if (IS_ERR(skb)) {
1719                 err = -EINVAL;
1720                 goto err;
1721         }
1722
1723         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1724         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1725         vxh->vx_vni = md->vni;
1726
1727         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1728                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1729                            VXLAN_RCO_SHIFT;
1730
1731                 if (skb->csum_offset == offsetof(struct udphdr, check))
1732                         data |= VXLAN_RCO_UDP;
1733
1734                 vxh->vx_vni |= htonl(data);
1735                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1736
1737                 if (!skb_is_gso(skb)) {
1738                         skb->ip_summed = CHECKSUM_NONE;
1739                         skb->encapsulation = 0;
1740                 }
1741         }
1742
1743         if (vxflags & VXLAN_F_GBP)
1744                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1745
1746         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1747
1748         udp_tunnel6_xmit_skb(dst, sk, skb, dev, saddr, daddr, prio,
1749                              ttl, src_port, dst_port,
1750                              !!(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX));
1751         return 0;
1752 err:
1753         dst_release(dst);
1754         return err;
1755 }
1756 #endif
1757
1758 int vxlan_xmit_skb(struct rtable *rt, struct sock *sk, struct sk_buff *skb,
1759                    __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1760                    __be16 src_port, __be16 dst_port,
1761                    struct vxlan_metadata *md, bool xnet, u32 vxflags)
1762 {
1763         struct vxlanhdr *vxh;
1764         int min_headroom;
1765         int err;
1766         bool udp_sum = !!(vxflags & VXLAN_F_UDP_CSUM);
1767         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1768         u16 hdrlen = sizeof(struct vxlanhdr);
1769
1770         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1771             skb->ip_summed == CHECKSUM_PARTIAL) {
1772                 int csum_start = skb_checksum_start_offset(skb);
1773
1774                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1775                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1776                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1777                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1778                         udp_sum = false;
1779                         type |= SKB_GSO_TUNNEL_REMCSUM;
1780                 }
1781         }
1782
1783         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1784                         + VXLAN_HLEN + sizeof(struct iphdr)
1785                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1786
1787         /* Need space for new headers (invalidates iph ptr) */
1788         err = skb_cow_head(skb, min_headroom);
1789         if (unlikely(err)) {
1790                 kfree_skb(skb);
1791                 return err;
1792         }
1793
1794         skb = vlan_hwaccel_push_inside(skb);
1795         if (WARN_ON(!skb))
1796                 return -ENOMEM;
1797
1798         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1799         if (IS_ERR(skb))
1800                 return PTR_ERR(skb);
1801
1802         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1803         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1804         vxh->vx_vni = md->vni;
1805
1806         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1807                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1808                            VXLAN_RCO_SHIFT;
1809
1810                 if (skb->csum_offset == offsetof(struct udphdr, check))
1811                         data |= VXLAN_RCO_UDP;
1812
1813                 vxh->vx_vni |= htonl(data);
1814                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1815
1816                 if (!skb_is_gso(skb)) {
1817                         skb->ip_summed = CHECKSUM_NONE;
1818                         skb->encapsulation = 0;
1819                 }
1820         }
1821
1822         if (vxflags & VXLAN_F_GBP)
1823                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1824
1825         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1826
1827         return udp_tunnel_xmit_skb(rt, sk, skb, src, dst, tos,
1828                                    ttl, df, src_port, dst_port, xnet,
1829                                    !(vxflags & VXLAN_F_UDP_CSUM));
1830 }
1831 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1832
1833 /* Bypass encapsulation if the destination is local */
1834 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1835                                struct vxlan_dev *dst_vxlan)
1836 {
1837         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1838         union vxlan_addr loopback;
1839         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1840         struct net_device *dev = skb->dev;
1841         int len = skb->len;
1842
1843         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1844         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1845         skb->pkt_type = PACKET_HOST;
1846         skb->encapsulation = 0;
1847         skb->dev = dst_vxlan->dev;
1848         __skb_pull(skb, skb_network_offset(skb));
1849
1850         if (remote_ip->sa.sa_family == AF_INET) {
1851                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1852                 loopback.sa.sa_family =  AF_INET;
1853 #if IS_ENABLED(CONFIG_IPV6)
1854         } else {
1855                 loopback.sin6.sin6_addr = in6addr_loopback;
1856                 loopback.sa.sa_family =  AF_INET6;
1857 #endif
1858         }
1859
1860         if (dst_vxlan->flags & VXLAN_F_LEARN)
1861                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1862
1863         u64_stats_update_begin(&tx_stats->syncp);
1864         tx_stats->tx_packets++;
1865         tx_stats->tx_bytes += len;
1866         u64_stats_update_end(&tx_stats->syncp);
1867
1868         if (netif_rx(skb) == NET_RX_SUCCESS) {
1869                 u64_stats_update_begin(&rx_stats->syncp);
1870                 rx_stats->rx_packets++;
1871                 rx_stats->rx_bytes += len;
1872                 u64_stats_update_end(&rx_stats->syncp);
1873         } else {
1874                 dev->stats.rx_dropped++;
1875         }
1876 }
1877
1878 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1879                            struct vxlan_rdst *rdst, bool did_rsc)
1880 {
1881         struct vxlan_dev *vxlan = netdev_priv(dev);
1882         struct sock *sk = vxlan->vn_sock->sock->sk;
1883         struct rtable *rt = NULL;
1884         const struct iphdr *old_iph;
1885         struct flowi4 fl4;
1886         union vxlan_addr *dst;
1887         struct vxlan_metadata md;
1888         __be16 src_port = 0, dst_port;
1889         u32 vni;
1890         __be16 df = 0;
1891         __u8 tos, ttl;
1892         int err;
1893
1894         dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1895         vni = rdst->remote_vni;
1896         dst = &rdst->remote_ip;
1897
1898         if (vxlan_addr_any(dst)) {
1899                 if (did_rsc) {
1900                         /* short-circuited back to local bridge */
1901                         vxlan_encap_bypass(skb, vxlan, vxlan);
1902                         return;
1903                 }
1904                 goto drop;
1905         }
1906
1907         old_iph = ip_hdr(skb);
1908
1909         ttl = vxlan->ttl;
1910         if (!ttl && vxlan_addr_multicast(dst))
1911                 ttl = 1;
1912
1913         tos = vxlan->tos;
1914         if (tos == 1)
1915                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1916
1917         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->port_min,
1918                                      vxlan->port_max, true);
1919
1920         if (dst->sa.sa_family == AF_INET) {
1921                 memset(&fl4, 0, sizeof(fl4));
1922                 fl4.flowi4_oif = rdst->remote_ifindex;
1923                 fl4.flowi4_tos = RT_TOS(tos);
1924                 fl4.flowi4_mark = skb->mark;
1925                 fl4.flowi4_proto = IPPROTO_UDP;
1926                 fl4.daddr = dst->sin.sin_addr.s_addr;
1927                 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1928
1929                 rt = ip_route_output_key(vxlan->net, &fl4);
1930                 if (IS_ERR(rt)) {
1931                         netdev_dbg(dev, "no route to %pI4\n",
1932                                    &dst->sin.sin_addr.s_addr);
1933                         dev->stats.tx_carrier_errors++;
1934                         goto tx_error;
1935                 }
1936
1937                 if (rt->dst.dev == dev) {
1938                         netdev_dbg(dev, "circular route to %pI4\n",
1939                                    &dst->sin.sin_addr.s_addr);
1940                         dev->stats.collisions++;
1941                         goto rt_tx_error;
1942                 }
1943
1944                 /* Bypass encapsulation if the destination is local */
1945                 if (rt->rt_flags & RTCF_LOCAL &&
1946                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1947                         struct vxlan_dev *dst_vxlan;
1948
1949                         ip_rt_put(rt);
1950                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1951                                                    dst->sa.sa_family, dst_port,
1952                                                    vxlan->flags);
1953                         if (!dst_vxlan)
1954                                 goto tx_error;
1955                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1956                         return;
1957                 }
1958
1959                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1960                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1961                 md.vni = htonl(vni << 8);
1962                 md.gbp = skb->mark;
1963
1964                 err = vxlan_xmit_skb(rt, sk, skb, fl4.saddr,
1965                                      dst->sin.sin_addr.s_addr, tos, ttl, df,
1966                                      src_port, dst_port, &md,
1967                                      !net_eq(vxlan->net, dev_net(vxlan->dev)),
1968                                      vxlan->flags);
1969                 if (err < 0) {
1970                         /* skb is already freed. */
1971                         skb = NULL;
1972                         goto rt_tx_error;
1973                 }
1974
1975                 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1976 #if IS_ENABLED(CONFIG_IPV6)
1977         } else {
1978                 struct dst_entry *ndst;
1979                 struct flowi6 fl6;
1980                 u32 flags;
1981
1982                 memset(&fl6, 0, sizeof(fl6));
1983                 fl6.flowi6_oif = rdst->remote_ifindex;
1984                 fl6.daddr = dst->sin6.sin6_addr;
1985                 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
1986                 fl6.flowi6_mark = skb->mark;
1987                 fl6.flowi6_proto = IPPROTO_UDP;
1988
1989                 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
1990                         netdev_dbg(dev, "no route to %pI6\n",
1991                                    &dst->sin6.sin6_addr);
1992                         dev->stats.tx_carrier_errors++;
1993                         goto tx_error;
1994                 }
1995
1996                 if (ndst->dev == dev) {
1997                         netdev_dbg(dev, "circular route to %pI6\n",
1998                                    &dst->sin6.sin6_addr);
1999                         dst_release(ndst);
2000                         dev->stats.collisions++;
2001                         goto tx_error;
2002                 }
2003
2004                 /* Bypass encapsulation if the destination is local */
2005                 flags = ((struct rt6_info *)ndst)->rt6i_flags;
2006                 if (flags & RTF_LOCAL &&
2007                     !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2008                         struct vxlan_dev *dst_vxlan;
2009
2010                         dst_release(ndst);
2011                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2012                                                    dst->sa.sa_family, dst_port,
2013                                                    vxlan->flags);
2014                         if (!dst_vxlan)
2015                                 goto tx_error;
2016                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2017                         return;
2018                 }
2019
2020                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2021                 md.vni = htonl(vni << 8);
2022                 md.gbp = skb->mark;
2023
2024                 err = vxlan6_xmit_skb(ndst, sk, skb, dev, &fl6.saddr, &fl6.daddr,
2025                                       0, ttl, src_port, dst_port, &md,
2026                                       !net_eq(vxlan->net, dev_net(vxlan->dev)),
2027                                       vxlan->flags);
2028 #endif
2029         }
2030
2031         return;
2032
2033 drop:
2034         dev->stats.tx_dropped++;
2035         goto tx_free;
2036
2037 rt_tx_error:
2038         ip_rt_put(rt);
2039 tx_error:
2040         dev->stats.tx_errors++;
2041 tx_free:
2042         dev_kfree_skb(skb);
2043 }
2044
2045 /* Transmit local packets over Vxlan
2046  *
2047  * Outer IP header inherits ECN and DF from inner header.
2048  * Outer UDP destination is the VXLAN assigned port.
2049  *           source port is based on hash of flow
2050  */
2051 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2052 {
2053         struct vxlan_dev *vxlan = netdev_priv(dev);
2054         struct ethhdr *eth;
2055         bool did_rsc = false;
2056         struct vxlan_rdst *rdst, *fdst = NULL;
2057         struct vxlan_fdb *f;
2058
2059         skb_reset_mac_header(skb);
2060         eth = eth_hdr(skb);
2061
2062         if ((vxlan->flags & VXLAN_F_PROXY)) {
2063                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2064                         return arp_reduce(dev, skb);
2065 #if IS_ENABLED(CONFIG_IPV6)
2066                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2067                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2068                                        + sizeof(struct nd_msg)) &&
2069                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2070                                 struct nd_msg *msg;
2071
2072                                 msg = (struct nd_msg *)skb_transport_header(skb);
2073                                 if (msg->icmph.icmp6_code == 0 &&
2074                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2075                                         return neigh_reduce(dev, skb);
2076                 }
2077                 eth = eth_hdr(skb);
2078 #endif
2079         }
2080
2081         f = vxlan_find_mac(vxlan, eth->h_dest);
2082         did_rsc = false;
2083
2084         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2085             (ntohs(eth->h_proto) == ETH_P_IP ||
2086              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2087                 did_rsc = route_shortcircuit(dev, skb);
2088                 if (did_rsc)
2089                         f = vxlan_find_mac(vxlan, eth->h_dest);
2090         }
2091
2092         if (f == NULL) {
2093                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2094                 if (f == NULL) {
2095                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2096                             !is_multicast_ether_addr(eth->h_dest))
2097                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2098
2099                         dev->stats.tx_dropped++;
2100                         kfree_skb(skb);
2101                         return NETDEV_TX_OK;
2102                 }
2103         }
2104
2105         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2106                 struct sk_buff *skb1;
2107
2108                 if (!fdst) {
2109                         fdst = rdst;
2110                         continue;
2111                 }
2112                 skb1 = skb_clone(skb, GFP_ATOMIC);
2113                 if (skb1)
2114                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2115         }
2116
2117         if (fdst)
2118                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2119         else
2120                 kfree_skb(skb);
2121         return NETDEV_TX_OK;
2122 }
2123
2124 /* Walk the forwarding table and purge stale entries */
2125 static void vxlan_cleanup(unsigned long arg)
2126 {
2127         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2128         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2129         unsigned int h;
2130
2131         if (!netif_running(vxlan->dev))
2132                 return;
2133
2134         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2135                 struct hlist_node *p, *n;
2136
2137                 spin_lock_bh(&vxlan->hash_lock);
2138                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2139                         struct vxlan_fdb *f
2140                                 = container_of(p, struct vxlan_fdb, hlist);
2141                         unsigned long timeout;
2142
2143                         if (f->state & NUD_PERMANENT)
2144                                 continue;
2145
2146                         timeout = f->used + vxlan->age_interval * HZ;
2147                         if (time_before_eq(timeout, jiffies)) {
2148                                 netdev_dbg(vxlan->dev,
2149                                            "garbage collect %pM\n",
2150                                            f->eth_addr);
2151                                 f->state = NUD_STALE;
2152                                 vxlan_fdb_destroy(vxlan, f);
2153                         } else if (time_before(timeout, next_timer))
2154                                 next_timer = timeout;
2155                 }
2156                 spin_unlock_bh(&vxlan->hash_lock);
2157         }
2158
2159         mod_timer(&vxlan->age_timer, next_timer);
2160 }
2161
2162 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2163 {
2164         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2165         __u32 vni = vxlan->default_dst.remote_vni;
2166
2167         vxlan->vn_sock = vs;
2168         spin_lock(&vn->sock_lock);
2169         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2170         spin_unlock(&vn->sock_lock);
2171 }
2172
2173 /* Setup stats when device is created */
2174 static int vxlan_init(struct net_device *dev)
2175 {
2176         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2177         if (!dev->tstats)
2178                 return -ENOMEM;
2179
2180         return 0;
2181 }
2182
2183 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2184 {
2185         struct vxlan_fdb *f;
2186
2187         spin_lock_bh(&vxlan->hash_lock);
2188         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2189         if (f)
2190                 vxlan_fdb_destroy(vxlan, f);
2191         spin_unlock_bh(&vxlan->hash_lock);
2192 }
2193
2194 static void vxlan_uninit(struct net_device *dev)
2195 {
2196         struct vxlan_dev *vxlan = netdev_priv(dev);
2197
2198         vxlan_fdb_delete_default(vxlan);
2199
2200         free_percpu(dev->tstats);
2201 }
2202
2203 /* Start ageing timer and join group when device is brought up */
2204 static int vxlan_open(struct net_device *dev)
2205 {
2206         struct vxlan_dev *vxlan = netdev_priv(dev);
2207         struct vxlan_sock *vs;
2208         int ret = 0;
2209
2210         vs = vxlan_sock_add(vxlan->net, vxlan->dst_port, vxlan_rcv, NULL,
2211                             false, vxlan->flags);
2212         if (IS_ERR(vs))
2213                 return PTR_ERR(vs);
2214
2215         vxlan_vs_add_dev(vs, vxlan);
2216
2217         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2218                 ret = vxlan_igmp_join(vxlan);
2219                 if (ret == -EADDRINUSE)
2220                         ret = 0;
2221                 if (ret) {
2222                         vxlan_sock_release(vs);
2223                         return ret;
2224                 }
2225         }
2226
2227         if (vxlan->age_interval)
2228                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2229
2230         return ret;
2231 }
2232
2233 /* Purge the forwarding table */
2234 static void vxlan_flush(struct vxlan_dev *vxlan)
2235 {
2236         unsigned int h;
2237
2238         spin_lock_bh(&vxlan->hash_lock);
2239         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2240                 struct hlist_node *p, *n;
2241                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2242                         struct vxlan_fdb *f
2243                                 = container_of(p, struct vxlan_fdb, hlist);
2244                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2245                         if (!is_zero_ether_addr(f->eth_addr))
2246                                 vxlan_fdb_destroy(vxlan, f);
2247                 }
2248         }
2249         spin_unlock_bh(&vxlan->hash_lock);
2250 }
2251
2252 /* Cleanup timer and forwarding table on shutdown */
2253 static int vxlan_stop(struct net_device *dev)
2254 {
2255         struct vxlan_dev *vxlan = netdev_priv(dev);
2256         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2257         struct vxlan_sock *vs = vxlan->vn_sock;
2258         int ret = 0;
2259
2260         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2261             !vxlan_group_used(vn, vxlan))
2262                 ret = vxlan_igmp_leave(vxlan);
2263
2264         del_timer_sync(&vxlan->age_timer);
2265
2266         vxlan_flush(vxlan);
2267         vxlan_sock_release(vs);
2268
2269         return ret;
2270 }
2271
2272 /* Stub, nothing needs to be done. */
2273 static void vxlan_set_multicast_list(struct net_device *dev)
2274 {
2275 }
2276
2277 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2278 {
2279         struct vxlan_dev *vxlan = netdev_priv(dev);
2280         struct vxlan_rdst *dst = &vxlan->default_dst;
2281         struct net_device *lowerdev;
2282         int max_mtu;
2283
2284         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2285         if (lowerdev == NULL)
2286                 return eth_change_mtu(dev, new_mtu);
2287
2288         if (dst->remote_ip.sa.sa_family == AF_INET6)
2289                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2290         else
2291                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2292
2293         if (new_mtu < 68 || new_mtu > max_mtu)
2294                 return -EINVAL;
2295
2296         dev->mtu = new_mtu;
2297         return 0;
2298 }
2299
2300 static const struct net_device_ops vxlan_netdev_ops = {
2301         .ndo_init               = vxlan_init,
2302         .ndo_uninit             = vxlan_uninit,
2303         .ndo_open               = vxlan_open,
2304         .ndo_stop               = vxlan_stop,
2305         .ndo_start_xmit         = vxlan_xmit,
2306         .ndo_get_stats64        = ip_tunnel_get_stats64,
2307         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2308         .ndo_change_mtu         = vxlan_change_mtu,
2309         .ndo_validate_addr      = eth_validate_addr,
2310         .ndo_set_mac_address    = eth_mac_addr,
2311         .ndo_fdb_add            = vxlan_fdb_add,
2312         .ndo_fdb_del            = vxlan_fdb_delete,
2313         .ndo_fdb_dump           = vxlan_fdb_dump,
2314 };
2315
2316 /* Info for udev, that this is a virtual tunnel endpoint */
2317 static struct device_type vxlan_type = {
2318         .name = "vxlan",
2319 };
2320
2321 /* Calls the ndo_add_vxlan_port of the caller in order to
2322  * supply the listening VXLAN udp ports. Callers are expected
2323  * to implement the ndo_add_vxlan_port.
2324  */
2325 void vxlan_get_rx_port(struct net_device *dev)
2326 {
2327         struct vxlan_sock *vs;
2328         struct net *net = dev_net(dev);
2329         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2330         sa_family_t sa_family;
2331         __be16 port;
2332         unsigned int i;
2333
2334         spin_lock(&vn->sock_lock);
2335         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2336                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2337                         port = inet_sk(vs->sock->sk)->inet_sport;
2338                         sa_family = vs->sock->sk->sk_family;
2339                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2340                                                             port);
2341                 }
2342         }
2343         spin_unlock(&vn->sock_lock);
2344 }
2345 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2346
2347 /* Initialize the device structure. */
2348 static void vxlan_setup(struct net_device *dev)
2349 {
2350         struct vxlan_dev *vxlan = netdev_priv(dev);
2351         unsigned int h;
2352
2353         eth_hw_addr_random(dev);
2354         ether_setup(dev);
2355         if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2356                 dev->needed_headroom = ETH_HLEN + VXLAN6_HEADROOM;
2357         else
2358                 dev->needed_headroom = ETH_HLEN + VXLAN_HEADROOM;
2359
2360         dev->netdev_ops = &vxlan_netdev_ops;
2361         dev->destructor = free_netdev;
2362         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2363
2364         dev->tx_queue_len = 0;
2365         dev->features   |= NETIF_F_LLTX;
2366         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2367         dev->features   |= NETIF_F_RXCSUM;
2368         dev->features   |= NETIF_F_GSO_SOFTWARE;
2369
2370         dev->vlan_features = dev->features;
2371         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2372         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2373         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2374         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2375         netif_keep_dst(dev);
2376         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2377
2378         INIT_LIST_HEAD(&vxlan->next);
2379         spin_lock_init(&vxlan->hash_lock);
2380
2381         init_timer_deferrable(&vxlan->age_timer);
2382         vxlan->age_timer.function = vxlan_cleanup;
2383         vxlan->age_timer.data = (unsigned long) vxlan;
2384
2385         vxlan->dst_port = htons(vxlan_port);
2386
2387         vxlan->dev = dev;
2388
2389         for (h = 0; h < FDB_HASH_SIZE; ++h)
2390                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2391 }
2392
2393 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2394         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2395         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2396         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2397         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2398         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2399         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2400         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2401         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2402         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2403         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2404         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2405         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2406         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2407         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2408         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2409         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2410         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2411         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2412         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2413         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2414         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2415         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2416         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2417         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2418 };
2419
2420 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2421 {
2422         if (tb[IFLA_ADDRESS]) {
2423                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2424                         pr_debug("invalid link address (not ethernet)\n");
2425                         return -EINVAL;
2426                 }
2427
2428                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2429                         pr_debug("invalid all zero ethernet address\n");
2430                         return -EADDRNOTAVAIL;
2431                 }
2432         }
2433
2434         if (!data)
2435                 return -EINVAL;
2436
2437         if (data[IFLA_VXLAN_ID]) {
2438                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2439                 if (id >= VXLAN_VID_MASK)
2440                         return -ERANGE;
2441         }
2442
2443         if (data[IFLA_VXLAN_PORT_RANGE]) {
2444                 const struct ifla_vxlan_port_range *p
2445                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2446
2447                 if (ntohs(p->high) < ntohs(p->low)) {
2448                         pr_debug("port range %u .. %u not valid\n",
2449                                  ntohs(p->low), ntohs(p->high));
2450                         return -EINVAL;
2451                 }
2452         }
2453
2454         return 0;
2455 }
2456
2457 static void vxlan_get_drvinfo(struct net_device *netdev,
2458                               struct ethtool_drvinfo *drvinfo)
2459 {
2460         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2461         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2462 }
2463
2464 static const struct ethtool_ops vxlan_ethtool_ops = {
2465         .get_drvinfo    = vxlan_get_drvinfo,
2466         .get_link       = ethtool_op_get_link,
2467 };
2468
2469 static void vxlan_del_work(struct work_struct *work)
2470 {
2471         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2472         udp_tunnel_sock_release(vs->sock);
2473         kfree_rcu(vs, rcu);
2474 }
2475
2476 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2477                                         __be16 port, u32 flags)
2478 {
2479         struct socket *sock;
2480         struct udp_port_cfg udp_conf;
2481         int err;
2482
2483         memset(&udp_conf, 0, sizeof(udp_conf));
2484
2485         if (ipv6) {
2486                 udp_conf.family = AF_INET6;
2487                 udp_conf.use_udp6_rx_checksums =
2488                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2489         } else {
2490                 udp_conf.family = AF_INET;
2491         }
2492
2493         udp_conf.local_udp_port = port;
2494
2495         /* Open UDP socket */
2496         err = udp_sock_create(net, &udp_conf, &sock);
2497         if (err < 0)
2498                 return ERR_PTR(err);
2499
2500         return sock;
2501 }
2502
2503 /* Create new listen socket if needed */
2504 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2505                                               vxlan_rcv_t *rcv, void *data,
2506                                               u32 flags)
2507 {
2508         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2509         struct vxlan_sock *vs;
2510         struct socket *sock;
2511         unsigned int h;
2512         bool ipv6 = !!(flags & VXLAN_F_IPV6);
2513         struct udp_tunnel_sock_cfg tunnel_cfg;
2514
2515         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2516         if (!vs)
2517                 return ERR_PTR(-ENOMEM);
2518
2519         for (h = 0; h < VNI_HASH_SIZE; ++h)
2520                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2521
2522         INIT_WORK(&vs->del_work, vxlan_del_work);
2523
2524         sock = vxlan_create_sock(net, ipv6, port, flags);
2525         if (IS_ERR(sock)) {
2526                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2527                         PTR_ERR(sock));
2528                 kfree(vs);
2529                 return ERR_CAST(sock);
2530         }
2531
2532         vs->sock = sock;
2533         atomic_set(&vs->refcnt, 1);
2534         vs->rcv = rcv;
2535         vs->data = data;
2536         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2537
2538         /* Initialize the vxlan udp offloads structure */
2539         vs->udp_offloads.port = port;
2540         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2541         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2542
2543         spin_lock(&vn->sock_lock);
2544         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2545         vxlan_notify_add_rx_port(vs);
2546         spin_unlock(&vn->sock_lock);
2547
2548         /* Mark socket as an encapsulation socket. */
2549         tunnel_cfg.sk_user_data = vs;
2550         tunnel_cfg.encap_type = 1;
2551         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2552         tunnel_cfg.encap_destroy = NULL;
2553
2554         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2555
2556         return vs;
2557 }
2558
2559 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2560                                   vxlan_rcv_t *rcv, void *data,
2561                                   bool no_share, u32 flags)
2562 {
2563         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2564         struct vxlan_sock *vs;
2565         bool ipv6 = flags & VXLAN_F_IPV6;
2566
2567         if (!no_share) {
2568                 spin_lock(&vn->sock_lock);
2569                 vs = vxlan_find_sock(net, ipv6 ? AF_INET6 : AF_INET, port,
2570                                      flags);
2571                 if (vs && vs->rcv == rcv) {
2572                         if (!atomic_add_unless(&vs->refcnt, 1, 0))
2573                                 vs = ERR_PTR(-EBUSY);
2574                         spin_unlock(&vn->sock_lock);
2575                         return vs;
2576                 }
2577                 spin_unlock(&vn->sock_lock);
2578         }
2579
2580         return vxlan_socket_create(net, port, rcv, data, flags);
2581 }
2582 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2583
2584 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2585                          struct nlattr *tb[], struct nlattr *data[])
2586 {
2587         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2588         struct vxlan_dev *vxlan = netdev_priv(dev);
2589         struct vxlan_rdst *dst = &vxlan->default_dst;
2590         __u32 vni;
2591         int err;
2592         bool use_ipv6 = false;
2593
2594         if (!data[IFLA_VXLAN_ID])
2595                 return -EINVAL;
2596
2597         vxlan->net = src_net;
2598
2599         vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2600         dst->remote_vni = vni;
2601
2602         /* Unless IPv6 is explicitly requested, assume IPv4 */
2603         dst->remote_ip.sa.sa_family = AF_INET;
2604         if (data[IFLA_VXLAN_GROUP]) {
2605                 dst->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2606         } else if (data[IFLA_VXLAN_GROUP6]) {
2607                 if (!IS_ENABLED(CONFIG_IPV6))
2608                         return -EPFNOSUPPORT;
2609
2610                 dst->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2611                 dst->remote_ip.sa.sa_family = AF_INET6;
2612                 use_ipv6 = true;
2613         }
2614
2615         if (data[IFLA_VXLAN_LOCAL]) {
2616                 vxlan->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2617                 vxlan->saddr.sa.sa_family = AF_INET;
2618         } else if (data[IFLA_VXLAN_LOCAL6]) {
2619                 if (!IS_ENABLED(CONFIG_IPV6))
2620                         return -EPFNOSUPPORT;
2621
2622                 /* TODO: respect scope id */
2623                 vxlan->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2624                 vxlan->saddr.sa.sa_family = AF_INET6;
2625                 use_ipv6 = true;
2626         }
2627
2628         if (data[IFLA_VXLAN_LINK] &&
2629             (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2630                 struct net_device *lowerdev
2631                          = __dev_get_by_index(src_net, dst->remote_ifindex);
2632
2633                 if (!lowerdev) {
2634                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2635                         return -ENODEV;
2636                 }
2637
2638 #if IS_ENABLED(CONFIG_IPV6)
2639                 if (use_ipv6) {
2640                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2641                         if (idev && idev->cnf.disable_ipv6) {
2642                                 pr_info("IPv6 is disabled via sysctl\n");
2643                                 return -EPERM;
2644                         }
2645                         vxlan->flags |= VXLAN_F_IPV6;
2646                 }
2647 #endif
2648
2649                 if (!tb[IFLA_MTU])
2650                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2651
2652                 dev->needed_headroom = lowerdev->hard_header_len +
2653                                        (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2654         } else if (use_ipv6)
2655                 vxlan->flags |= VXLAN_F_IPV6;
2656
2657         if (data[IFLA_VXLAN_TOS])
2658                 vxlan->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2659
2660         if (data[IFLA_VXLAN_TTL])
2661                 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2662
2663         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2664                 vxlan->flags |= VXLAN_F_LEARN;
2665
2666         if (data[IFLA_VXLAN_AGEING])
2667                 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2668         else
2669                 vxlan->age_interval = FDB_AGE_DEFAULT;
2670
2671         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2672                 vxlan->flags |= VXLAN_F_PROXY;
2673
2674         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2675                 vxlan->flags |= VXLAN_F_RSC;
2676
2677         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2678                 vxlan->flags |= VXLAN_F_L2MISS;
2679
2680         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2681                 vxlan->flags |= VXLAN_F_L3MISS;
2682
2683         if (data[IFLA_VXLAN_LIMIT])
2684                 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2685
2686         if (data[IFLA_VXLAN_PORT_RANGE]) {
2687                 const struct ifla_vxlan_port_range *p
2688                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2689                 vxlan->port_min = ntohs(p->low);
2690                 vxlan->port_max = ntohs(p->high);
2691         }
2692
2693         if (data[IFLA_VXLAN_PORT])
2694                 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2695
2696         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2697                 vxlan->flags |= VXLAN_F_UDP_CSUM;
2698
2699         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2700             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2701                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2702
2703         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2704             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2705                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2706
2707         if (data[IFLA_VXLAN_REMCSUM_TX] &&
2708             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2709                 vxlan->flags |= VXLAN_F_REMCSUM_TX;
2710
2711         if (data[IFLA_VXLAN_REMCSUM_RX] &&
2712             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2713                 vxlan->flags |= VXLAN_F_REMCSUM_RX;
2714
2715         if (data[IFLA_VXLAN_GBP])
2716                 vxlan->flags |= VXLAN_F_GBP;
2717
2718         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2719                 vxlan->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2720
2721         if (vxlan_find_vni(src_net, vni, use_ipv6 ? AF_INET6 : AF_INET,
2722                            vxlan->dst_port, vxlan->flags)) {
2723                 pr_info("duplicate VNI %u\n", vni);
2724                 return -EEXIST;
2725         }
2726
2727         dev->ethtool_ops = &vxlan_ethtool_ops;
2728
2729         /* create an fdb entry for a valid default destination */
2730         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2731                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2732                                        &vxlan->default_dst.remote_ip,
2733                                        NUD_REACHABLE|NUD_PERMANENT,
2734                                        NLM_F_EXCL|NLM_F_CREATE,
2735                                        vxlan->dst_port,
2736                                        vxlan->default_dst.remote_vni,
2737                                        vxlan->default_dst.remote_ifindex,
2738                                        NTF_SELF);
2739                 if (err)
2740                         return err;
2741         }
2742
2743         err = register_netdevice(dev);
2744         if (err) {
2745                 vxlan_fdb_delete_default(vxlan);
2746                 return err;
2747         }
2748
2749         list_add(&vxlan->next, &vn->vxlan_list);
2750
2751         return 0;
2752 }
2753
2754 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2755 {
2756         struct vxlan_dev *vxlan = netdev_priv(dev);
2757         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2758
2759         spin_lock(&vn->sock_lock);
2760         if (!hlist_unhashed(&vxlan->hlist))
2761                 hlist_del_rcu(&vxlan->hlist);
2762         spin_unlock(&vn->sock_lock);
2763
2764         list_del(&vxlan->next);
2765         unregister_netdevice_queue(dev, head);
2766 }
2767
2768 static size_t vxlan_get_size(const struct net_device *dev)
2769 {
2770
2771         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2772                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2773                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2774                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2775                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2776                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2777                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2778                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2779                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2780                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2781                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2782                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2783                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2784                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2785                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2786                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2787                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2788                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2789                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
2790                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
2791                 0;
2792 }
2793
2794 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2795 {
2796         const struct vxlan_dev *vxlan = netdev_priv(dev);
2797         const struct vxlan_rdst *dst = &vxlan->default_dst;
2798         struct ifla_vxlan_port_range ports = {
2799                 .low =  htons(vxlan->port_min),
2800                 .high = htons(vxlan->port_max),
2801         };
2802
2803         if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2804                 goto nla_put_failure;
2805
2806         if (!vxlan_addr_any(&dst->remote_ip)) {
2807                 if (dst->remote_ip.sa.sa_family == AF_INET) {
2808                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
2809                                             dst->remote_ip.sin.sin_addr.s_addr))
2810                                 goto nla_put_failure;
2811 #if IS_ENABLED(CONFIG_IPV6)
2812                 } else {
2813                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
2814                                              &dst->remote_ip.sin6.sin6_addr))
2815                                 goto nla_put_failure;
2816 #endif
2817                 }
2818         }
2819
2820         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2821                 goto nla_put_failure;
2822
2823         if (!vxlan_addr_any(&vxlan->saddr)) {
2824                 if (vxlan->saddr.sa.sa_family == AF_INET) {
2825                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
2826                                             vxlan->saddr.sin.sin_addr.s_addr))
2827                                 goto nla_put_failure;
2828 #if IS_ENABLED(CONFIG_IPV6)
2829                 } else {
2830                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
2831                                              &vxlan->saddr.sin6.sin6_addr))
2832                                 goto nla_put_failure;
2833 #endif
2834                 }
2835         }
2836
2837         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2838             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2839             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2840                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
2841             nla_put_u8(skb, IFLA_VXLAN_PROXY,
2842                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
2843             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2844             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2845                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2846             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2847                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2848             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2849             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2850             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port) ||
2851             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2852                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2853             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2854                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2855             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2856                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
2857             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
2858                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
2859             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
2860                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
2861                 goto nla_put_failure;
2862
2863         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2864                 goto nla_put_failure;
2865
2866         if (vxlan->flags & VXLAN_F_GBP &&
2867             nla_put_flag(skb, IFLA_VXLAN_GBP))
2868                 goto nla_put_failure;
2869
2870         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
2871             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
2872                 goto nla_put_failure;
2873
2874         return 0;
2875
2876 nla_put_failure:
2877         return -EMSGSIZE;
2878 }
2879
2880 static struct net *vxlan_get_link_net(const struct net_device *dev)
2881 {
2882         struct vxlan_dev *vxlan = netdev_priv(dev);
2883
2884         return vxlan->net;
2885 }
2886
2887 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2888         .kind           = "vxlan",
2889         .maxtype        = IFLA_VXLAN_MAX,
2890         .policy         = vxlan_policy,
2891         .priv_size      = sizeof(struct vxlan_dev),
2892         .setup          = vxlan_setup,
2893         .validate       = vxlan_validate,
2894         .newlink        = vxlan_newlink,
2895         .dellink        = vxlan_dellink,
2896         .get_size       = vxlan_get_size,
2897         .fill_info      = vxlan_fill_info,
2898         .get_link_net   = vxlan_get_link_net,
2899 };
2900
2901 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
2902                                              struct net_device *dev)
2903 {
2904         struct vxlan_dev *vxlan, *next;
2905         LIST_HEAD(list_kill);
2906
2907         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2908                 struct vxlan_rdst *dst = &vxlan->default_dst;
2909
2910                 /* In case we created vxlan device with carrier
2911                  * and we loose the carrier due to module unload
2912                  * we also need to remove vxlan device. In other
2913                  * cases, it's not necessary and remote_ifindex
2914                  * is 0 here, so no matches.
2915                  */
2916                 if (dst->remote_ifindex == dev->ifindex)
2917                         vxlan_dellink(vxlan->dev, &list_kill);
2918         }
2919
2920         unregister_netdevice_many(&list_kill);
2921 }
2922
2923 static int vxlan_lowerdev_event(struct notifier_block *unused,
2924                                 unsigned long event, void *ptr)
2925 {
2926         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2927         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2928
2929         if (event == NETDEV_UNREGISTER)
2930                 vxlan_handle_lowerdev_unregister(vn, dev);
2931
2932         return NOTIFY_DONE;
2933 }
2934
2935 static struct notifier_block vxlan_notifier_block __read_mostly = {
2936         .notifier_call = vxlan_lowerdev_event,
2937 };
2938
2939 static __net_init int vxlan_init_net(struct net *net)
2940 {
2941         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2942         unsigned int h;
2943
2944         INIT_LIST_HEAD(&vn->vxlan_list);
2945         spin_lock_init(&vn->sock_lock);
2946
2947         for (h = 0; h < PORT_HASH_SIZE; ++h)
2948                 INIT_HLIST_HEAD(&vn->sock_list[h]);
2949
2950         return 0;
2951 }
2952
2953 static void __net_exit vxlan_exit_net(struct net *net)
2954 {
2955         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2956         struct vxlan_dev *vxlan, *next;
2957         struct net_device *dev, *aux;
2958         LIST_HEAD(list);
2959
2960         rtnl_lock();
2961         for_each_netdev_safe(net, dev, aux)
2962                 if (dev->rtnl_link_ops == &vxlan_link_ops)
2963                         unregister_netdevice_queue(dev, &list);
2964
2965         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2966                 /* If vxlan->dev is in the same netns, it has already been added
2967                  * to the list by the previous loop.
2968                  */
2969                 if (!net_eq(dev_net(vxlan->dev), net))
2970                         unregister_netdevice_queue(vxlan->dev, &list);
2971         }
2972
2973         unregister_netdevice_many(&list);
2974         rtnl_unlock();
2975 }
2976
2977 static struct pernet_operations vxlan_net_ops = {
2978         .init = vxlan_init_net,
2979         .exit = vxlan_exit_net,
2980         .id   = &vxlan_net_id,
2981         .size = sizeof(struct vxlan_net),
2982 };
2983
2984 static int __init vxlan_init_module(void)
2985 {
2986         int rc;
2987
2988         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
2989         if (!vxlan_wq)
2990                 return -ENOMEM;
2991
2992         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
2993
2994         rc = register_pernet_subsys(&vxlan_net_ops);
2995         if (rc)
2996                 goto out1;
2997
2998         rc = register_netdevice_notifier(&vxlan_notifier_block);
2999         if (rc)
3000                 goto out2;
3001
3002         rc = rtnl_link_register(&vxlan_link_ops);
3003         if (rc)
3004                 goto out3;
3005
3006         return 0;
3007 out3:
3008         unregister_netdevice_notifier(&vxlan_notifier_block);
3009 out2:
3010         unregister_pernet_subsys(&vxlan_net_ops);
3011 out1:
3012         destroy_workqueue(vxlan_wq);
3013         return rc;
3014 }
3015 late_initcall(vxlan_init_module);
3016
3017 static void __exit vxlan_cleanup_module(void)
3018 {
3019         rtnl_link_unregister(&vxlan_link_ops);
3020         unregister_netdevice_notifier(&vxlan_notifier_block);
3021         destroy_workqueue(vxlan_wq);
3022         unregister_pernet_subsys(&vxlan_net_ops);
3023         /* rcu_barrier() is called by netns */
3024 }
3025 module_exit(vxlan_cleanup_module);
3026
3027 MODULE_LICENSE("GPL");
3028 MODULE_VERSION(VXLAN_VERSION);
3029 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3030 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3031 MODULE_ALIAS_RTNL_LINK("vxlan");