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 #include <net/dst_metadata.h>
  53
  54 #define VXLAN_VERSION   "0.1"
  55
  56 #define PORT_HASH_BITS  8
  57 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  58 #define VNI_HASH_BITS   10
  59 #define VNI_HASH_SIZE   (1<<VNI_HASH_BITS)
  60 #define FDB_HASH_BITS   8
  61 #define FDB_HASH_SIZE   (1<<FDB_HASH_BITS)
  62 #define FDB_AGE_DEFAULT 300 /* 5 min */
  63 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  64
  65 /* UDP port for VXLAN traffic.
  66  * The IANA assigned port is 4789, but the Linux default is 8472
  67  * for compatibility with early adopters.
  68  */
  69 static unsigned short vxlan_port __read_mostly = 8472;
  70 module_param_named(udp_port, vxlan_port, ushort, 0444);
  71 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  72
  73 static bool log_ecn_error = true;
  74 module_param(log_ecn_error, bool, 0644);
  75 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  76
  77 static int vxlan_net_id;
  78
  79 static const u8 all_zeros_mac[ETH_ALEN];
  80
  81 /* per-network namespace private data for this module */
  82 struct vxlan_net {
  83         struct list_head  vxlan_list;
  84         struct hlist_head sock_list[PORT_HASH_SIZE];
  85         spinlock_t        sock_lock;
  86 };
  87
  88 union vxlan_addr {
  89         struct sockaddr_in sin;
  90         struct sockaddr_in6 sin6;
  91         struct sockaddr sa;
  92 };
  93
  94 struct vxlan_rdst {
  95         union vxlan_addr         remote_ip;
  96         __be16                   remote_port;
  97         u32                      remote_vni;
  98         u32                      remote_ifindex;
  99         struct list_head         list;
 100         struct rcu_head          rcu;
 101 };
 102
 103 /* Forwarding table entry */
 104 struct vxlan_fdb {
 105         struct hlist_node hlist;        /* linked list of entries */
 106         struct rcu_head   rcu;
 107         unsigned long     updated;      /* jiffies */
 108         unsigned long     used;
 109         struct list_head  remotes;
 110         u8                eth_addr[ETH_ALEN];
 111         u16               state;        /* see ndm_state */
 112         u8                flags;        /* see ndm_flags */
 113 };
 114
 115 /* Pseudo network device */
 116 struct vxlan_dev {
 117         struct hlist_node hlist;        /* vni hash table */
 118         struct list_head  next;         /* vxlan's per namespace list */
 119         struct vxlan_sock *vn_sock;     /* listening socket */
 120         struct net_device *dev;
 121         struct net        *net;         /* netns for packet i/o */
 122         struct vxlan_rdst default_dst;  /* default destination */
 123         union vxlan_addr  saddr;        /* source address */
 124         __be16            dst_port;
 125         __u16             port_min;     /* source port range */
 126         __u16             port_max;
 127         __u8              tos;          /* TOS override */
 128         __u8              ttl;
 129         u32               flags;        /* VXLAN_F_* in vxlan.h */
 130
 131         unsigned long     age_interval;
 132         struct timer_list age_timer;
 133         spinlock_t        hash_lock;
 134         unsigned int      addrcnt;
 135         unsigned int      addrmax;
 136
 137         struct hlist_head fdb_head[FDB_HASH_SIZE];
 138 };
 139
 140 /* salt for hash table */
 141 static u32 vxlan_salt __read_mostly;
 142 static struct workqueue_struct *vxlan_wq;
 143
 144 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
 145 {
 146         return vs->flags & VXLAN_F_COLLECT_METADATA;
 147 }
 148
 149 #if IS_ENABLED(CONFIG_IPV6)
 150 static inline
 151 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 152 {
 153         if (a->sa.sa_family != b->sa.sa_family)
 154                 return false;
 155         if (a->sa.sa_family == AF_INET6)
 156                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 157         else
 158                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 159 }
 160
 161 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 162 {
 163         if (ipa->sa.sa_family == AF_INET6)
 164                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 165         else
 166                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 167 }
 168
 169 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 170 {
 171         if (ipa->sa.sa_family == AF_INET6)
 172                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 173         else
 174                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 175 }
 176
 177 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 178 {
 179         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 180                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 181                 ip->sa.sa_family = AF_INET6;
 182                 return 0;
 183         } else if (nla_len(nla) >= sizeof(__be32)) {
 184                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 185                 ip->sa.sa_family = AF_INET;
 186                 return 0;
 187         } else {
 188                 return -EAFNOSUPPORT;
 189         }
 190 }
 191
 192 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 193                               const union vxlan_addr *ip)
 194 {
 195         if (ip->sa.sa_family == AF_INET6)
 196                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 197         else
 198                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 199 }
 200
 201 #else /* !CONFIG_IPV6 */
 202
 203 static inline
 204 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 205 {
 206         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 207 }
 208
 209 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 210 {
 211         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 212 }
 213
 214 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 215 {
 216         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 217 }
 218
 219 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 220 {
 221         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 222                 return -EAFNOSUPPORT;
 223         } else if (nla_len(nla) >= sizeof(__be32)) {
 224                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 225                 ip->sa.sa_family = AF_INET;
 226                 return 0;
 227         } else {
 228                 return -EAFNOSUPPORT;
 229         }
 230 }
 231
 232 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 233                               const union vxlan_addr *ip)
 234 {
 235         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 236 }
 237 #endif
 238
 239 /* Virtual Network hash table head */
 240 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
 241 {
 242         return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
 243 }
 244
 245 /* Socket hash table head */
 246 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 247 {
 248         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 249
 250         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 251 }
 252
 253 /* First remote destination for a forwarding entry.
 254  * Guaranteed to be non-NULL because remotes are never deleted.
 255  */
 256 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 257 {
 258         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 259 }
 260
 261 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 262 {
 263         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 264 }
 265
 266 /* Find VXLAN socket based on network namespace, address family and UDP port
 267  * and enabled unshareable flags.
 268  */
 269 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 270                                           __be16 port, u32 flags)
 271 {
 272         struct vxlan_sock *vs;
 273
 274         flags &= VXLAN_F_RCV_FLAGS;
 275
 276         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 277                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 278                     inet_sk(vs->sock->sk)->sk.sk_family == family &&
 279                     vs->flags == flags)
 280                         return vs;
 281         }
 282         return NULL;
 283 }
 284
 285 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
 286 {
 287         struct vxlan_dev *vxlan;
 288
 289         hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
 290                 if (vxlan->default_dst.remote_vni == id)
 291                         return vxlan;
 292         }
 293
 294         return NULL;
 295 }
 296
 297 /* Look up VNI in a per net namespace table */
 298 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id,
 299                                         sa_family_t family, __be16 port,
 300                                         u32 flags)
 301 {
 302         struct vxlan_sock *vs;
 303
 304         vs = vxlan_find_sock(net, family, port, flags);
 305         if (!vs)
 306                 return NULL;
 307
 308         return vxlan_vs_find_vni(vs, id);
 309 }
 310
 311 /* Fill in neighbour message in skbuff. */
 312 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 313                           const struct vxlan_fdb *fdb,
 314                           u32 portid, u32 seq, int type, unsigned int flags,
 315                           const struct vxlan_rdst *rdst)
 316 {
 317         unsigned long now = jiffies;
 318         struct nda_cacheinfo ci;
 319         struct nlmsghdr *nlh;
 320         struct ndmsg *ndm;
 321         bool send_ip, send_eth;
 322
 323         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 324         if (nlh == NULL)
 325                 return -EMSGSIZE;
 326
 327         ndm = nlmsg_data(nlh);
 328         memset(ndm, 0, sizeof(*ndm));
 329
 330         send_eth = send_ip = true;
 331
 332         if (type == RTM_GETNEIGH) {
 333                 ndm->ndm_family = AF_INET;
 334                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 335                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 336         } else
 337                 ndm->ndm_family = AF_BRIDGE;
 338         ndm->ndm_state = fdb->state;
 339         ndm->ndm_ifindex = vxlan->dev->ifindex;
 340         ndm->ndm_flags = fdb->flags;
 341         ndm->ndm_type = RTN_UNICAST;
 342
 343         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 344             nla_put_s32(skb, NDA_LINK_NETNSID,
 345                         peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
 346                 goto nla_put_failure;
 347
 348         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 349                 goto nla_put_failure;
 350
 351         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 352                 goto nla_put_failure;
 353
 354         if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
 355             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 356                 goto nla_put_failure;
 357         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 358             nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
 359                 goto nla_put_failure;
 360         if (rdst->remote_ifindex &&
 361             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 362                 goto nla_put_failure;
 363
 364         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 365         ci.ndm_confirmed = 0;
 366         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 367         ci.ndm_refcnt    = 0;
 368
 369         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 370                 goto nla_put_failure;
 371
 372         nlmsg_end(skb, nlh);
 373         return 0;
 374
 375 nla_put_failure:
 376         nlmsg_cancel(skb, nlh);
 377         return -EMSGSIZE;
 378 }
 379
 380 static inline size_t vxlan_nlmsg_size(void)
 381 {
 382         return NLMSG_ALIGN(sizeof(struct ndmsg))
 383                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 384                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 385                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 386                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 387                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 388                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 389                 + nla_total_size(sizeof(struct nda_cacheinfo));
 390 }
 391
 392 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 393                              struct vxlan_rdst *rd, int type)
 394 {
 395         struct net *net = dev_net(vxlan->dev);
 396         struct sk_buff *skb;
 397         int err = -ENOBUFS;
 398
 399         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 400         if (skb == NULL)
 401                 goto errout;
 402
 403         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 404         if (err < 0) {
 405                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 406                 WARN_ON(err == -EMSGSIZE);
 407                 kfree_skb(skb);
 408                 goto errout;
 409         }
 410
 411         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 412         return;
 413 errout:
 414         if (err < 0)
 415                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 416 }
 417
 418 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 419 {
 420         struct vxlan_dev *vxlan = netdev_priv(dev);
 421         struct vxlan_fdb f = {
 422                 .state = NUD_STALE,
 423         };
 424         struct vxlan_rdst remote = {
 425                 .remote_ip = *ipa, /* goes to NDA_DST */
 426                 .remote_vni = VXLAN_N_VID,
 427         };
 428
 429         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 430 }
 431
 432 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 433 {
 434         struct vxlan_fdb f = {
 435                 .state = NUD_STALE,
 436         };
 437         struct vxlan_rdst remote = { };
 438
 439         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 440
 441         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 442 }
 443
 444 /* Hash Ethernet address */
 445 static u32 eth_hash(const unsigned char *addr)
 446 {
 447         u64 value = get_unaligned((u64 *)addr);
 448
 449         /* only want 6 bytes */
 450 #ifdef __BIG_ENDIAN
 451         value >>= 16;
 452 #else
 453         value <<= 16;
 454 #endif
 455         return hash_64(value, FDB_HASH_BITS);
 456 }
 457
 458 /* Hash chain to use given mac address */
 459 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 460                                                 const u8 *mac)
 461 {
 462         return &vxlan->fdb_head[eth_hash(mac)];
 463 }
 464
 465 /* Look up Ethernet address in forwarding table */
 466 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 467                                         const u8 *mac)
 468 {
 469         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 470         struct vxlan_fdb *f;
 471
 472         hlist_for_each_entry_rcu(f, head, hlist) {
 473                 if (ether_addr_equal(mac, f->eth_addr))
 474                         return f;
 475         }
 476
 477         return NULL;
 478 }
 479
 480 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 481                                         const u8 *mac)
 482 {
 483         struct vxlan_fdb *f;
 484
 485         f = __vxlan_find_mac(vxlan, mac);
 486         if (f)
 487                 f->used = jiffies;
 488
 489         return f;
 490 }
 491
 492 /* caller should hold vxlan->hash_lock */
 493 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 494                                               union vxlan_addr *ip, __be16 port,
 495                                               __u32 vni, __u32 ifindex)
 496 {
 497         struct vxlan_rdst *rd;
 498
 499         list_for_each_entry(rd, &f->remotes, list) {
 500                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 501                     rd->remote_port == port &&
 502                     rd->remote_vni == vni &&
 503                     rd->remote_ifindex == ifindex)
 504                         return rd;
 505         }
 506
 507         return NULL;
 508 }
 509
 510 /* Replace destination of unicast mac */
 511 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 512                              union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
 513 {
 514         struct vxlan_rdst *rd;
 515
 516         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 517         if (rd)
 518                 return 0;
 519
 520         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 521         if (!rd)
 522                 return 0;
 523         rd->remote_ip = *ip;
 524         rd->remote_port = port;
 525         rd->remote_vni = vni;
 526         rd->remote_ifindex = ifindex;
 527         return 1;
 528 }
 529
 530 /* Add/update destinations for multicast */
 531 static int vxlan_fdb_append(struct vxlan_fdb *f,
 532                             union vxlan_addr *ip, __be16 port, __u32 vni,
 533                             __u32 ifindex, struct vxlan_rdst **rdp)
 534 {
 535         struct vxlan_rdst *rd;
 536
 537         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 538         if (rd)
 539                 return 0;
 540
 541         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 542         if (rd == NULL)
 543                 return -ENOBUFS;
 544         rd->remote_ip = *ip;
 545         rd->remote_port = port;
 546         rd->remote_vni = vni;
 547         rd->remote_ifindex = ifindex;
 548
 549         list_add_tail_rcu(&rd->list, &f->remotes);
 550
 551         *rdp = rd;
 552         return 1;
 553 }
 554
 555 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 556                                           unsigned int off,
 557                                           struct vxlanhdr *vh, size_t hdrlen,
 558                                           u32 data, struct gro_remcsum *grc,
 559                                           bool nopartial)
 560 {
 561         size_t start, offset, plen;
 562
 563         if (skb->remcsum_offload)
 564                 return NULL;
 565
 566         if (!NAPI_GRO_CB(skb)->csum_valid)
 567                 return NULL;
 568
 569         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
 570         offset = start + ((data & VXLAN_RCO_UDP) ?
 571                           offsetof(struct udphdr, check) :
 572                           offsetof(struct tcphdr, check));
 573
 574         plen = hdrlen + offset + sizeof(u16);
 575
 576         /* Pull checksum that will be written */
 577         if (skb_gro_header_hard(skb, off + plen)) {
 578                 vh = skb_gro_header_slow(skb, off + plen, off);
 579                 if (!vh)
 580                         return NULL;
 581         }
 582
 583         skb_gro_remcsum_process(skb, (void *)vh + hdrlen,
 584                                 start, offset, grc, nopartial);
 585
 586         skb->remcsum_offload = 1;
 587
 588         return vh;
 589 }
 590
 591 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
 592                                           struct sk_buff *skb,
 593                                           struct udp_offload *uoff)
 594 {
 595         struct sk_buff *p, **pp = NULL;
 596         struct vxlanhdr *vh, *vh2;
 597         unsigned int hlen, off_vx;
 598         int flush = 1;
 599         struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
 600                                              udp_offloads);
 601         u32 flags;
 602         struct gro_remcsum grc;
 603
 604         skb_gro_remcsum_init(&grc);
 605
 606         off_vx = skb_gro_offset(skb);
 607         hlen = off_vx + sizeof(*vh);
 608         vh   = skb_gro_header_fast(skb, off_vx);
 609         if (skb_gro_header_hard(skb, hlen)) {
 610                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 611                 if (unlikely(!vh))
 612                         goto out;
 613         }
 614
 615         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 616         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 617
 618         flags = ntohl(vh->vx_flags);
 619
 620         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 621                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 622                                        ntohl(vh->vx_vni), &grc,
 623                                        !!(vs->flags &
 624                                           VXLAN_F_REMCSUM_NOPARTIAL));
 625
 626                 if (!vh)
 627                         goto out;
 628         }
 629
 630         flush = 0;
 631
 632         for (p = *head; p; p = p->next) {
 633                 if (!NAPI_GRO_CB(p)->same_flow)
 634                         continue;
 635
 636                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 637                 if (vh->vx_flags != vh2->vx_flags ||
 638                     vh->vx_vni != vh2->vx_vni) {
 639                         NAPI_GRO_CB(p)->same_flow = 0;
 640                         continue;
 641                 }
 642         }
 643
 644         pp = eth_gro_receive(head, skb);
 645
 646 out:
 647         skb_gro_remcsum_cleanup(skb, &grc);
 648         NAPI_GRO_CB(skb)->flush |= flush;
 649
 650         return pp;
 651 }
 652
 653 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
 654                               struct udp_offload *uoff)
 655 {
 656         udp_tunnel_gro_complete(skb, nhoff);
 657
 658         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 659 }
 660
 661 /* Notify netdevs that UDP port started listening */
 662 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 663 {
 664         struct net_device *dev;
 665         struct sock *sk = vs->sock->sk;
 666         struct net *net = sock_net(sk);
 667         sa_family_t sa_family = sk->sk_family;
 668         __be16 port = inet_sk(sk)->inet_sport;
 669         int err;
 670
 671         if (sa_family == AF_INET) {
 672                 err = udp_add_offload(&vs->udp_offloads);
 673                 if (err)
 674                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 675         }
 676
 677         rcu_read_lock();
 678         for_each_netdev_rcu(net, dev) {
 679                 if (dev->netdev_ops->ndo_add_vxlan_port)
 680                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 681                                                             port);
 682         }
 683         rcu_read_unlock();
 684 }
 685
 686 /* Notify netdevs that UDP port is no more listening */
 687 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 688 {
 689         struct net_device *dev;
 690         struct sock *sk = vs->sock->sk;
 691         struct net *net = sock_net(sk);
 692         sa_family_t sa_family = sk->sk_family;
 693         __be16 port = inet_sk(sk)->inet_sport;
 694
 695         rcu_read_lock();
 696         for_each_netdev_rcu(net, dev) {
 697                 if (dev->netdev_ops->ndo_del_vxlan_port)
 698                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 699                                                             port);
 700         }
 701         rcu_read_unlock();
 702
 703         if (sa_family == AF_INET)
 704                 udp_del_offload(&vs->udp_offloads);
 705 }
 706
 707 /* Add new entry to forwarding table -- assumes lock held */
 708 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 709                             const u8 *mac, union vxlan_addr *ip,
 710                             __u16 state, __u16 flags,
 711                             __be16 port, __u32 vni, __u32 ifindex,
 712                             __u8 ndm_flags)
 713 {
 714         struct vxlan_rdst *rd = NULL;
 715         struct vxlan_fdb *f;
 716         int notify = 0;
 717
 718         f = __vxlan_find_mac(vxlan, mac);
 719         if (f) {
 720                 if (flags & NLM_F_EXCL) {
 721                         netdev_dbg(vxlan->dev,
 722                                    "lost race to create %pM\n", mac);
 723                         return -EEXIST;
 724                 }
 725                 if (f->state != state) {
 726                         f->state = state;
 727                         f->updated = jiffies;
 728                         notify = 1;
 729                 }
 730                 if (f->flags != ndm_flags) {
 731                         f->flags = ndm_flags;
 732                         f->updated = jiffies;
 733                         notify = 1;
 734                 }
 735                 if ((flags & NLM_F_REPLACE)) {
 736                         /* Only change unicasts */
 737                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 738                              is_zero_ether_addr(f->eth_addr))) {
 739                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 740                                                            ifindex);
 741                         } else
 742                                 return -EOPNOTSUPP;
 743                 }
 744                 if ((flags & NLM_F_APPEND) &&
 745                     (is_multicast_ether_addr(f->eth_addr) ||
 746                      is_zero_ether_addr(f->eth_addr))) {
 747                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 748                                                   &rd);
 749
 750                         if (rc < 0)
 751                                 return rc;
 752                         notify |= rc;
 753                 }
 754         } else {
 755                 if (!(flags & NLM_F_CREATE))
 756                         return -ENOENT;
 757
 758                 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
 759                         return -ENOSPC;
 760
 761                 /* Disallow replace to add a multicast entry */
 762                 if ((flags & NLM_F_REPLACE) &&
 763                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 764                         return -EOPNOTSUPP;
 765
 766                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 767                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 768                 if (!f)
 769                         return -ENOMEM;
 770
 771                 notify = 1;
 772                 f->state = state;
 773                 f->flags = ndm_flags;
 774                 f->updated = f->used = jiffies;
 775                 INIT_LIST_HEAD(&f->remotes);
 776                 memcpy(f->eth_addr, mac, ETH_ALEN);
 777
 778                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 779
 780                 ++vxlan->addrcnt;
 781                 hlist_add_head_rcu(&f->hlist,
 782                                    vxlan_fdb_head(vxlan, mac));
 783         }
 784
 785         if (notify) {
 786                 if (rd == NULL)
 787                         rd = first_remote_rtnl(f);
 788                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 789         }
 790
 791         return 0;
 792 }
 793
 794 static void vxlan_fdb_free(struct rcu_head *head)
 795 {
 796         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 797         struct vxlan_rdst *rd, *nd;
 798
 799         list_for_each_entry_safe(rd, nd, &f->remotes, list)
 800                 kfree(rd);
 801         kfree(f);
 802 }
 803
 804 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 805 {
 806         netdev_dbg(vxlan->dev,
 807                     "delete %pM\n", f->eth_addr);
 808
 809         --vxlan->addrcnt;
 810         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 811
 812         hlist_del_rcu(&f->hlist);
 813         call_rcu(&f->rcu, vxlan_fdb_free);
 814 }
 815
 816 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 817                            union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
 818 {
 819         struct net *net = dev_net(vxlan->dev);
 820         int err;
 821
 822         if (tb[NDA_DST]) {
 823                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 824                 if (err)
 825                         return err;
 826         } else {
 827                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 828                 if (remote->sa.sa_family == AF_INET) {
 829                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 830                         ip->sa.sa_family = AF_INET;
 831 #if IS_ENABLED(CONFIG_IPV6)
 832                 } else {
 833                         ip->sin6.sin6_addr = in6addr_any;
 834                         ip->sa.sa_family = AF_INET6;
 835 #endif
 836                 }
 837         }
 838
 839         if (tb[NDA_PORT]) {
 840                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 841                         return -EINVAL;
 842                 *port = nla_get_be16(tb[NDA_PORT]);
 843         } else {
 844                 *port = vxlan->dst_port;
 845         }
 846
 847         if (tb[NDA_VNI]) {
 848                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 849                         return -EINVAL;
 850                 *vni = nla_get_u32(tb[NDA_VNI]);
 851         } else {
 852                 *vni = vxlan->default_dst.remote_vni;
 853         }
 854
 855         if (tb[NDA_IFINDEX]) {
 856                 struct net_device *tdev;
 857
 858                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 859                         return -EINVAL;
 860                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 861                 tdev = __dev_get_by_index(net, *ifindex);
 862                 if (!tdev)
 863                         return -EADDRNOTAVAIL;
 864         } else {
 865                 *ifindex = 0;
 866         }
 867
 868         return 0;
 869 }
 870
 871 /* Add static entry (via netlink) */
 872 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 873                          struct net_device *dev,
 874                          const unsigned char *addr, u16 vid, u16 flags)
 875 {
 876         struct vxlan_dev *vxlan = netdev_priv(dev);
 877         /* struct net *net = dev_net(vxlan->dev); */
 878         union vxlan_addr ip;
 879         __be16 port;
 880         u32 vni, ifindex;
 881         int err;
 882
 883         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 884                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 885                         ndm->ndm_state);
 886                 return -EINVAL;
 887         }
 888
 889         if (tb[NDA_DST] == NULL)
 890                 return -EINVAL;
 891
 892         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 893         if (err)
 894                 return err;
 895
 896         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 897                 return -EAFNOSUPPORT;
 898
 899         spin_lock_bh(&vxlan->hash_lock);
 900         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 901                                port, vni, ifindex, ndm->ndm_flags);
 902         spin_unlock_bh(&vxlan->hash_lock);
 903
 904         return err;
 905 }
 906
 907 /* Delete entry (via netlink) */
 908 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 909                             struct net_device *dev,
 910                             const unsigned char *addr, u16 vid)
 911 {
 912         struct vxlan_dev *vxlan = netdev_priv(dev);
 913         struct vxlan_fdb *f;
 914         struct vxlan_rdst *rd = NULL;
 915         union vxlan_addr ip;
 916         __be16 port;
 917         u32 vni, ifindex;
 918         int err;
 919
 920         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 921         if (err)
 922                 return err;
 923
 924         err = -ENOENT;
 925
 926         spin_lock_bh(&vxlan->hash_lock);
 927         f = vxlan_find_mac(vxlan, addr);
 928         if (!f)
 929                 goto out;
 930
 931         if (!vxlan_addr_any(&ip)) {
 932                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 933                 if (!rd)
 934                         goto out;
 935         }
 936
 937         err = 0;
 938
 939         /* remove a destination if it's not the only one on the list,
 940          * otherwise destroy the fdb entry
 941          */
 942         if (rd && !list_is_singular(&f->remotes)) {
 943                 list_del_rcu(&rd->list);
 944                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 945                 kfree_rcu(rd, rcu);
 946                 goto out;
 947         }
 948
 949         vxlan_fdb_destroy(vxlan, f);
 950
 951 out:
 952         spin_unlock_bh(&vxlan->hash_lock);
 953
 954         return err;
 955 }
 956
 957 /* Dump forwarding table */
 958 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 959                           struct net_device *dev,
 960                           struct net_device *filter_dev, int idx)
 961 {
 962         struct vxlan_dev *vxlan = netdev_priv(dev);
 963         unsigned int h;
 964
 965         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 966                 struct vxlan_fdb *f;
 967                 int err;
 968
 969                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 970                         struct vxlan_rdst *rd;
 971
 972                         if (idx < cb->args[0])
 973                                 goto skip;
 974
 975                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 976                                 err = vxlan_fdb_info(skb, vxlan, f,
 977                                                      NETLINK_CB(cb->skb).portid,
 978                                                      cb->nlh->nlmsg_seq,
 979                                                      RTM_NEWNEIGH,
 980                                                      NLM_F_MULTI, rd);
 981                                 if (err < 0)
 982                                         goto out;
 983                         }
 984 skip:
 985                         ++idx;
 986                 }
 987         }
 988 out:
 989         return idx;
 990 }
 991
 992 /* Watch incoming packets to learn mapping between Ethernet address
 993  * and Tunnel endpoint.
 994  * Return true if packet is bogus and should be dropped.
 995  */
 996 static bool vxlan_snoop(struct net_device *dev,
 997                         union vxlan_addr *src_ip, const u8 *src_mac)
 998 {
 999         struct vxlan_dev *vxlan = netdev_priv(dev);
1000         struct vxlan_fdb *f;
1001
1002         f = vxlan_find_mac(vxlan, src_mac);
1003         if (likely(f)) {
1004                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1005
1006                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
1007                         return false;
1008
1009                 /* Don't migrate static entries, drop packets */
1010                 if (f->state & NUD_NOARP)
1011                         return true;
1012
1013                 if (net_ratelimit())
1014                         netdev_info(dev,
1015                                     "%pM migrated from %pIS to %pIS\n",
1016                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1017
1018                 rdst->remote_ip = *src_ip;
1019                 f->updated = jiffies;
1020                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1021         } else {
1022                 /* learned new entry */
1023                 spin_lock(&vxlan->hash_lock);
1024
1025                 /* close off race between vxlan_flush and incoming packets */
1026                 if (netif_running(dev))
1027                         vxlan_fdb_create(vxlan, src_mac, src_ip,
1028                                          NUD_REACHABLE,
1029                                          NLM_F_EXCL|NLM_F_CREATE,
1030                                          vxlan->dst_port,
1031                                          vxlan->default_dst.remote_vni,
1032                                          0, NTF_SELF);
1033                 spin_unlock(&vxlan->hash_lock);
1034         }
1035
1036         return false;
1037 }
1038
1039 /* See if multicast group is already in use by other ID */
1040 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1041 {
1042         struct vxlan_dev *vxlan;
1043
1044         /* The vxlan_sock is only used by dev, leaving group has
1045          * no effect on other vxlan devices.
1046          */
1047         if (atomic_read(&dev->vn_sock->refcnt) == 1)
1048                 return false;
1049
1050         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1051                 if (!netif_running(vxlan->dev) || vxlan == dev)
1052                         continue;
1053
1054                 if (vxlan->vn_sock != dev->vn_sock)
1055                         continue;
1056
1057                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1058                                       &dev->default_dst.remote_ip))
1059                         continue;
1060
1061                 if (vxlan->default_dst.remote_ifindex !=
1062                     dev->default_dst.remote_ifindex)
1063                         continue;
1064
1065                 return true;
1066         }
1067
1068         return false;
1069 }
1070
1071 void vxlan_sock_release(struct vxlan_sock *vs)
1072 {
1073         struct sock *sk = vs->sock->sk;
1074         struct net *net = sock_net(sk);
1075         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1076
1077         if (!atomic_dec_and_test(&vs->refcnt))
1078                 return;
1079
1080         spin_lock(&vn->sock_lock);
1081         hlist_del_rcu(&vs->hlist);
1082         vxlan_notify_del_rx_port(vs);
1083         spin_unlock(&vn->sock_lock);
1084
1085         queue_work(vxlan_wq, &vs->del_work);
1086 }
1087 EXPORT_SYMBOL_GPL(vxlan_sock_release);
1088
1089 /* Update multicast group membership when first VNI on
1090  * multicast address is brought up
1091  */
1092 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1093 {
1094         struct vxlan_sock *vs = vxlan->vn_sock;
1095         struct sock *sk = vs->sock->sk;
1096         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1097         int ifindex = vxlan->default_dst.remote_ifindex;
1098         int ret = -EINVAL;
1099
1100         lock_sock(sk);
1101         if (ip->sa.sa_family == AF_INET) {
1102                 struct ip_mreqn mreq = {
1103                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1104                         .imr_ifindex            = ifindex,
1105                 };
1106
1107                 ret = ip_mc_join_group(sk, &mreq);
1108 #if IS_ENABLED(CONFIG_IPV6)
1109         } else {
1110                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1111                                                    &ip->sin6.sin6_addr);
1112 #endif
1113         }
1114         release_sock(sk);
1115
1116         return ret;
1117 }
1118
1119 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1120 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1121 {
1122         struct vxlan_sock *vs = vxlan->vn_sock;
1123         struct sock *sk = vs->sock->sk;
1124         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1125         int ifindex = vxlan->default_dst.remote_ifindex;
1126         int ret = -EINVAL;
1127
1128         lock_sock(sk);
1129         if (ip->sa.sa_family == AF_INET) {
1130                 struct ip_mreqn mreq = {
1131                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1132                         .imr_ifindex            = ifindex,
1133                 };
1134
1135                 ret = ip_mc_leave_group(sk, &mreq);
1136 #if IS_ENABLED(CONFIG_IPV6)
1137         } else {
1138                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1139                                                    &ip->sin6.sin6_addr);
1140 #endif
1141         }
1142         release_sock(sk);
1143
1144         return ret;
1145 }
1146
1147 static struct vxlanhdr *vxlan_remcsum(struct sk_buff *skb, struct vxlanhdr *vh,
1148                                       size_t hdrlen, u32 data, bool nopartial)
1149 {
1150         size_t start, offset, plen;
1151
1152         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
1153         offset = start + ((data & VXLAN_RCO_UDP) ?
1154                           offsetof(struct udphdr, check) :
1155                           offsetof(struct tcphdr, check));
1156
1157         plen = hdrlen + offset + sizeof(u16);
1158
1159         if (!pskb_may_pull(skb, plen))
1160                 return NULL;
1161
1162         vh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1163
1164         skb_remcsum_process(skb, (void *)vh + hdrlen, start, offset,
1165                             nopartial);
1166
1167         return vh;
1168 }
1169
1170 /* Callback from net/ipv4/udp.c to receive packets */
1171 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1172 {
1173         struct metadata_dst *tun_dst = NULL;
1174         struct ip_tunnel_info *info;
1175         struct vxlan_sock *vs;
1176         struct vxlanhdr *vxh;
1177         u32 flags, vni;
1178         struct vxlan_metadata _md;
1179         struct vxlan_metadata *md = &_md;
1180
1181         /* Need Vxlan and inner Ethernet header to be present */
1182         if (!pskb_may_pull(skb, VXLAN_HLEN))
1183                 goto error;
1184
1185         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1186         flags = ntohl(vxh->vx_flags);
1187         vni = ntohl(vxh->vx_vni);
1188
1189         if (flags & VXLAN_HF_VNI) {
1190                 flags &= ~VXLAN_HF_VNI;
1191         } else {
1192                 /* VNI flag always required to be set */
1193                 goto bad_flags;
1194         }
1195
1196         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1197                 goto drop;
1198         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1199
1200         vs = rcu_dereference_sk_user_data(sk);
1201         if (!vs)
1202                 goto drop;
1203
1204         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1205                 vxh = vxlan_remcsum(skb, vxh, sizeof(struct vxlanhdr), vni,
1206                                     !!(vs->flags & VXLAN_F_REMCSUM_NOPARTIAL));
1207                 if (!vxh)
1208                         goto drop;
1209
1210                 flags &= ~VXLAN_HF_RCO;
1211                 vni &= VXLAN_VNI_MASK;
1212         }
1213
1214         if (vxlan_collect_metadata(vs)) {
1215                 const struct iphdr *iph = ip_hdr(skb);
1216
1217                 tun_dst = metadata_dst_alloc(sizeof(*md), GFP_ATOMIC);
1218                 if (!tun_dst)
1219                         goto drop;
1220
1221                 info = &tun_dst->u.tun_info;
1222                 info->key.ipv4_src = iph->saddr;
1223                 info->key.ipv4_dst = iph->daddr;
1224                 info->key.ipv4_tos = iph->tos;
1225                 info->key.ipv4_ttl = iph->ttl;
1226                 info->key.tp_src = udp_hdr(skb)->source;
1227                 info->key.tp_dst = udp_hdr(skb)->dest;
1228
1229                 info->mode = IP_TUNNEL_INFO_RX;
1230                 info->key.tun_flags = TUNNEL_KEY;
1231                 info->key.tun_id = cpu_to_be64(vni >> 8);
1232                 if (udp_hdr(skb)->check != 0)
1233                         info->key.tun_flags |= TUNNEL_CSUM;
1234
1235                 md = ip_tunnel_info_opts(info, sizeof(*md));
1236                 md->tun_dst = tun_dst;
1237         } else {
1238                 memset(md, 0, sizeof(*md));
1239         }
1240
1241         /* For backwards compatibility, only allow reserved fields to be
1242          * used by VXLAN extensions if explicitly requested.
1243          */
1244         if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1245                 struct vxlanhdr_gbp *gbp;
1246
1247                 gbp = (struct vxlanhdr_gbp *)vxh;
1248                 md->gbp = ntohs(gbp->policy_id);
1249
1250                 if (tun_dst)
1251                         info->key.tun_flags |= TUNNEL_VXLAN_OPT;
1252
1253                 if (gbp->dont_learn)
1254                         md->gbp |= VXLAN_GBP_DONT_LEARN;
1255
1256                 if (gbp->policy_applied)
1257                         md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1258
1259                 flags &= ~VXLAN_GBP_USED_BITS;
1260         }
1261
1262         if (flags || vni & ~VXLAN_VNI_MASK) {
1263                 /* If there are any unprocessed flags remaining treat
1264                  * this as a malformed packet. This behavior diverges from
1265                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1266                  * in reserved fields are to be ignored. The approach here
1267                  * maintains compatibility with previous stack code, and also
1268                  * is more robust and provides a little more security in
1269                  * adding extensions to VXLAN.
1270                  */
1271
1272                 goto bad_flags;
1273         }
1274
1275         md->vni = vxh->vx_vni;
1276         vs->rcv(vs, skb, md);
1277         return 0;
1278
1279 drop:
1280         /* Consume bad packet */
1281         kfree_skb(skb);
1282         return 0;
1283
1284 bad_flags:
1285         netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1286                    ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1287
1288 error:
1289         if (tun_dst)
1290                 dst_release((struct dst_entry *)tun_dst);
1291
1292         /* Return non vxlan pkt */
1293         return 1;
1294 }
1295
1296 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1297                       struct vxlan_metadata *md)
1298 {
1299         struct iphdr *oip = NULL;
1300         struct ipv6hdr *oip6 = NULL;
1301         struct vxlan_dev *vxlan;
1302         struct pcpu_sw_netstats *stats;
1303         union vxlan_addr saddr;
1304         __u32 vni;
1305         int err = 0;
1306         union vxlan_addr *remote_ip;
1307
1308         /* For flow based devices, map all packets to VNI 0 */
1309         if (vs->flags & VXLAN_F_FLOW_BASED)
1310                 vni = 0;
1311         else
1312                 vni = ntohl(md->vni) >> 8;
1313
1314         /* Is this VNI defined? */
1315         vxlan = vxlan_vs_find_vni(vs, vni);
1316         if (!vxlan)
1317                 goto drop;
1318
1319         remote_ip = &vxlan->default_dst.remote_ip;
1320         skb_reset_mac_header(skb);
1321         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1322         skb->protocol = eth_type_trans(skb, vxlan->dev);
1323         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1324
1325         /* Ignore packet loops (and multicast echo) */
1326         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1327                 goto drop;
1328
1329         /* Re-examine inner Ethernet packet */
1330         if (remote_ip->sa.sa_family == AF_INET) {
1331                 oip = ip_hdr(skb);
1332                 saddr.sin.sin_addr.s_addr = oip->saddr;
1333                 saddr.sa.sa_family = AF_INET;
1334 #if IS_ENABLED(CONFIG_IPV6)
1335         } else {
1336                 oip6 = ipv6_hdr(skb);
1337                 saddr.sin6.sin6_addr = oip6->saddr;
1338                 saddr.sa.sa_family = AF_INET6;
1339 #endif
1340         }
1341
1342         if (md->tun_dst) {
1343                 skb_dst_set(skb, (struct dst_entry *)md->tun_dst);
1344                 md->tun_dst = NULL;
1345         }
1346
1347         if ((vxlan->flags & VXLAN_F_LEARN) &&
1348             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1349                 goto drop;
1350
1351         skb_reset_network_header(skb);
1352         /* In flow-based mode, GBP is carried in dst_metadata */
1353         if (!(vs->flags & VXLAN_F_FLOW_BASED))
1354                 skb->mark = md->gbp;
1355
1356         if (oip6)
1357                 err = IP6_ECN_decapsulate(oip6, skb);
1358         if (oip)
1359                 err = IP_ECN_decapsulate(oip, skb);
1360
1361         if (unlikely(err)) {
1362                 if (log_ecn_error) {
1363                         if (oip6)
1364                                 net_info_ratelimited("non-ECT from %pI6\n",
1365                                                      &oip6->saddr);
1366                         if (oip)
1367                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1368                                                      &oip->saddr, oip->tos);
1369                 }
1370                 if (err > 1) {
1371                         ++vxlan->dev->stats.rx_frame_errors;
1372                         ++vxlan->dev->stats.rx_errors;
1373                         goto drop;
1374                 }
1375         }
1376
1377         stats = this_cpu_ptr(vxlan->dev->tstats);
1378         u64_stats_update_begin(&stats->syncp);
1379         stats->rx_packets++;
1380         stats->rx_bytes += skb->len;
1381         u64_stats_update_end(&stats->syncp);
1382
1383         netif_rx(skb);
1384
1385         return;
1386 drop:
1387         if (md->tun_dst)
1388                 dst_release((struct dst_entry *)md->tun_dst);
1389
1390         /* Consume bad packet */
1391         kfree_skb(skb);
1392 }
1393
1394 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1395 {
1396         struct vxlan_dev *vxlan = netdev_priv(dev);
1397         struct arphdr *parp;
1398         u8 *arpptr, *sha;
1399         __be32 sip, tip;
1400         struct neighbour *n;
1401
1402         if (dev->flags & IFF_NOARP)
1403                 goto out;
1404
1405         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1406                 dev->stats.tx_dropped++;
1407                 goto out;
1408         }
1409         parp = arp_hdr(skb);
1410
1411         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1412              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1413             parp->ar_pro != htons(ETH_P_IP) ||
1414             parp->ar_op != htons(ARPOP_REQUEST) ||
1415             parp->ar_hln != dev->addr_len ||
1416             parp->ar_pln != 4)
1417                 goto out;
1418         arpptr = (u8 *)parp + sizeof(struct arphdr);
1419         sha = arpptr;
1420         arpptr += dev->addr_len;        /* sha */
1421         memcpy(&sip, arpptr, sizeof(sip));
1422         arpptr += sizeof(sip);
1423         arpptr += dev->addr_len;        /* tha */
1424         memcpy(&tip, arpptr, sizeof(tip));
1425
1426         if (ipv4_is_loopback(tip) ||
1427             ipv4_is_multicast(tip))
1428                 goto out;
1429
1430         n = neigh_lookup(&arp_tbl, &tip, dev);
1431
1432         if (n) {
1433                 struct vxlan_fdb *f;
1434                 struct sk_buff  *reply;
1435
1436                 if (!(n->nud_state & NUD_CONNECTED)) {
1437                         neigh_release(n);
1438                         goto out;
1439                 }
1440
1441                 f = vxlan_find_mac(vxlan, n->ha);
1442                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1443                         /* bridge-local neighbor */
1444                         neigh_release(n);
1445                         goto out;
1446                 }
1447
1448                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1449                                 n->ha, sha);
1450
1451                 neigh_release(n);
1452
1453                 if (reply == NULL)
1454                         goto out;
1455
1456                 skb_reset_mac_header(reply);
1457                 __skb_pull(reply, skb_network_offset(reply));
1458                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1459                 reply->pkt_type = PACKET_HOST;
1460
1461                 if (netif_rx_ni(reply) == NET_RX_DROP)
1462                         dev->stats.rx_dropped++;
1463         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1464                 union vxlan_addr ipa = {
1465                         .sin.sin_addr.s_addr = tip,
1466                         .sin.sin_family = AF_INET,
1467                 };
1468
1469                 vxlan_ip_miss(dev, &ipa);
1470         }
1471 out:
1472         consume_skb(skb);
1473         return NETDEV_TX_OK;
1474 }
1475
1476 #if IS_ENABLED(CONFIG_IPV6)
1477 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1478         struct neighbour *n, bool isrouter)
1479 {
1480         struct net_device *dev = request->dev;
1481         struct sk_buff *reply;
1482         struct nd_msg *ns, *na;
1483         struct ipv6hdr *pip6;
1484         u8 *daddr;
1485         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1486         int ns_olen;
1487         int i, len;
1488
1489         if (dev == NULL)
1490                 return NULL;
1491
1492         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1493                 sizeof(*na) + na_olen + dev->needed_tailroom;
1494         reply = alloc_skb(len, GFP_ATOMIC);
1495         if (reply == NULL)
1496                 return NULL;
1497
1498         reply->protocol = htons(ETH_P_IPV6);
1499         reply->dev = dev;
1500         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1501         skb_push(reply, sizeof(struct ethhdr));
1502         skb_set_mac_header(reply, 0);
1503
1504         ns = (struct nd_msg *)skb_transport_header(request);
1505
1506         daddr = eth_hdr(request)->h_source;
1507         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1508         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1509                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1510                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1511                         break;
1512                 }
1513         }
1514
1515         /* Ethernet header */
1516         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1517         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1518         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1519         reply->protocol = htons(ETH_P_IPV6);
1520
1521         skb_pull(reply, sizeof(struct ethhdr));
1522         skb_set_network_header(reply, 0);
1523         skb_put(reply, sizeof(struct ipv6hdr));
1524
1525         /* IPv6 header */
1526
1527         pip6 = ipv6_hdr(reply);
1528         memset(pip6, 0, sizeof(struct ipv6hdr));
1529         pip6->version = 6;
1530         pip6->priority = ipv6_hdr(request)->priority;
1531         pip6->nexthdr = IPPROTO_ICMPV6;
1532         pip6->hop_limit = 255;
1533         pip6->daddr = ipv6_hdr(request)->saddr;
1534         pip6->saddr = *(struct in6_addr *)n->primary_key;
1535
1536         skb_pull(reply, sizeof(struct ipv6hdr));
1537         skb_set_transport_header(reply, 0);
1538
1539         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1540
1541         /* Neighbor Advertisement */
1542         memset(na, 0, sizeof(*na)+na_olen);
1543         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1544         na->icmph.icmp6_router = isrouter;
1545         na->icmph.icmp6_override = 1;
1546         na->icmph.icmp6_solicited = 1;
1547         na->target = ns->target;
1548         ether_addr_copy(&na->opt[2], n->ha);
1549         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1550         na->opt[1] = na_olen >> 3;
1551
1552         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1553                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1554                 csum_partial(na, sizeof(*na)+na_olen, 0));
1555
1556         pip6->payload_len = htons(sizeof(*na)+na_olen);
1557
1558         skb_push(reply, sizeof(struct ipv6hdr));
1559
1560         reply->ip_summed = CHECKSUM_UNNECESSARY;
1561
1562         return reply;
1563 }
1564
1565 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1566 {
1567         struct vxlan_dev *vxlan = netdev_priv(dev);
1568         struct nd_msg *msg;
1569         const struct ipv6hdr *iphdr;
1570         const struct in6_addr *saddr, *daddr;
1571         struct neighbour *n;
1572         struct inet6_dev *in6_dev;
1573
1574         in6_dev = __in6_dev_get(dev);
1575         if (!in6_dev)
1576                 goto out;
1577
1578         iphdr = ipv6_hdr(skb);
1579         saddr = &iphdr->saddr;
1580         daddr = &iphdr->daddr;
1581
1582         msg = (struct nd_msg *)skb_transport_header(skb);
1583         if (msg->icmph.icmp6_code != 0 ||
1584             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1585                 goto out;
1586
1587         if (ipv6_addr_loopback(daddr) ||
1588             ipv6_addr_is_multicast(&msg->target))
1589                 goto out;
1590
1591         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1592
1593         if (n) {
1594                 struct vxlan_fdb *f;
1595                 struct sk_buff *reply;
1596
1597                 if (!(n->nud_state & NUD_CONNECTED)) {
1598                         neigh_release(n);
1599                         goto out;
1600                 }
1601
1602                 f = vxlan_find_mac(vxlan, n->ha);
1603                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1604                         /* bridge-local neighbor */
1605                         neigh_release(n);
1606                         goto out;
1607                 }
1608
1609                 reply = vxlan_na_create(skb, n,
1610                                         !!(f ? f->flags & NTF_ROUTER : 0));
1611
1612                 neigh_release(n);
1613
1614                 if (reply == NULL)
1615                         goto out;
1616
1617                 if (netif_rx_ni(reply) == NET_RX_DROP)
1618                         dev->stats.rx_dropped++;
1619
1620         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1621                 union vxlan_addr ipa = {
1622                         .sin6.sin6_addr = msg->target,
1623                         .sin6.sin6_family = AF_INET6,
1624                 };
1625
1626                 vxlan_ip_miss(dev, &ipa);
1627         }
1628
1629 out:
1630         consume_skb(skb);
1631         return NETDEV_TX_OK;
1632 }
1633 #endif
1634
1635 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1636 {
1637         struct vxlan_dev *vxlan = netdev_priv(dev);
1638         struct neighbour *n;
1639
1640         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1641                 return false;
1642
1643         n = NULL;
1644         switch (ntohs(eth_hdr(skb)->h_proto)) {
1645         case ETH_P_IP:
1646         {
1647                 struct iphdr *pip;
1648
1649                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1650                         return false;
1651                 pip = ip_hdr(skb);
1652                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1653                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1654                         union vxlan_addr ipa = {
1655                                 .sin.sin_addr.s_addr = pip->daddr,
1656                                 .sin.sin_family = AF_INET,
1657                         };
1658
1659                         vxlan_ip_miss(dev, &ipa);
1660                         return false;
1661                 }
1662
1663                 break;
1664         }
1665 #if IS_ENABLED(CONFIG_IPV6)
1666         case ETH_P_IPV6:
1667         {
1668                 struct ipv6hdr *pip6;
1669
1670                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1671                         return false;
1672                 pip6 = ipv6_hdr(skb);
1673                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1674                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1675                         union vxlan_addr ipa = {
1676                                 .sin6.sin6_addr = pip6->daddr,
1677                                 .sin6.sin6_family = AF_INET6,
1678                         };
1679
1680                         vxlan_ip_miss(dev, &ipa);
1681                         return false;
1682                 }
1683
1684                 break;
1685         }
1686 #endif
1687         default:
1688                 return false;
1689         }
1690
1691         if (n) {
1692                 bool diff;
1693
1694                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1695                 if (diff) {
1696                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1697                                 dev->addr_len);
1698                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1699                 }
1700                 neigh_release(n);
1701                 return diff;
1702         }
1703
1704         return false;
1705 }
1706
1707 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1708                                 struct vxlan_metadata *md)
1709 {
1710         struct vxlanhdr_gbp *gbp;
1711
1712         if (!md->gbp)
1713                 return;
1714
1715         gbp = (struct vxlanhdr_gbp *)vxh;
1716         vxh->vx_flags |= htonl(VXLAN_HF_GBP);
1717
1718         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1719                 gbp->dont_learn = 1;
1720
1721         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1722                 gbp->policy_applied = 1;
1723
1724         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1725 }
1726
1727 #if IS_ENABLED(CONFIG_IPV6)
1728 static int vxlan6_xmit_skb(struct dst_entry *dst, struct sock *sk,
1729                            struct sk_buff *skb,
1730                            struct net_device *dev, struct in6_addr *saddr,
1731                            struct in6_addr *daddr, __u8 prio, __u8 ttl,
1732                            __be16 src_port, __be16 dst_port,
1733                            struct vxlan_metadata *md, bool xnet, u32 vxflags)
1734 {
1735         struct vxlanhdr *vxh;
1736         int min_headroom;
1737         int err;
1738         bool udp_sum = !(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX);
1739         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1740         u16 hdrlen = sizeof(struct vxlanhdr);
1741
1742         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1743             skb->ip_summed == CHECKSUM_PARTIAL) {
1744                 int csum_start = skb_checksum_start_offset(skb);
1745
1746                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1747                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1748                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1749                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1750                         udp_sum = false;
1751                         type |= SKB_GSO_TUNNEL_REMCSUM;
1752                 }
1753         }
1754
1755         skb_scrub_packet(skb, xnet);
1756
1757         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1758                         + VXLAN_HLEN + sizeof(struct ipv6hdr)
1759                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1760
1761         /* Need space for new headers (invalidates iph ptr) */
1762         err = skb_cow_head(skb, min_headroom);
1763         if (unlikely(err)) {
1764                 kfree_skb(skb);
1765                 goto err;
1766         }
1767
1768         skb = vlan_hwaccel_push_inside(skb);
1769         if (WARN_ON(!skb)) {
1770                 err = -ENOMEM;
1771                 goto err;
1772         }
1773
1774         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1775         if (IS_ERR(skb)) {
1776                 err = -EINVAL;
1777                 goto err;
1778         }
1779
1780         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1781         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1782         vxh->vx_vni = md->vni;
1783
1784         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1785                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1786                            VXLAN_RCO_SHIFT;
1787
1788                 if (skb->csum_offset == offsetof(struct udphdr, check))
1789                         data |= VXLAN_RCO_UDP;
1790
1791                 vxh->vx_vni |= htonl(data);
1792                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1793
1794                 if (!skb_is_gso(skb)) {
1795                         skb->ip_summed = CHECKSUM_NONE;
1796                         skb->encapsulation = 0;
1797                 }
1798         }
1799
1800         if (vxflags & VXLAN_F_GBP)
1801                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1802
1803         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1804
1805         udp_tunnel6_xmit_skb(dst, sk, skb, dev, saddr, daddr, prio,
1806                              ttl, src_port, dst_port,
1807                              !!(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX));
1808         return 0;
1809 err:
1810         dst_release(dst);
1811         return err;
1812 }
1813 #endif
1814
1815 int vxlan_xmit_skb(struct rtable *rt, struct sock *sk, struct sk_buff *skb,
1816                    __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1817                    __be16 src_port, __be16 dst_port,
1818                    struct vxlan_metadata *md, bool xnet, u32 vxflags)
1819 {
1820         struct vxlanhdr *vxh;
1821         int min_headroom;
1822         int err;
1823         bool udp_sum = !!(vxflags & VXLAN_F_UDP_CSUM);
1824         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1825         u16 hdrlen = sizeof(struct vxlanhdr);
1826
1827         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1828             skb->ip_summed == CHECKSUM_PARTIAL) {
1829                 int csum_start = skb_checksum_start_offset(skb);
1830
1831                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1832                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1833                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1834                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1835                         udp_sum = false;
1836                         type |= SKB_GSO_TUNNEL_REMCSUM;
1837                 }
1838         }
1839
1840         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1841                         + VXLAN_HLEN + sizeof(struct iphdr)
1842                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1843
1844         /* Need space for new headers (invalidates iph ptr) */
1845         err = skb_cow_head(skb, min_headroom);
1846         if (unlikely(err)) {
1847                 kfree_skb(skb);
1848                 return err;
1849         }
1850
1851         skb = vlan_hwaccel_push_inside(skb);
1852         if (WARN_ON(!skb))
1853                 return -ENOMEM;
1854
1855         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1856         if (IS_ERR(skb))
1857                 return PTR_ERR(skb);
1858
1859         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1860         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1861         vxh->vx_vni = md->vni;
1862
1863         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1864                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1865                            VXLAN_RCO_SHIFT;
1866
1867                 if (skb->csum_offset == offsetof(struct udphdr, check))
1868                         data |= VXLAN_RCO_UDP;
1869
1870                 vxh->vx_vni |= htonl(data);
1871                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1872
1873                 if (!skb_is_gso(skb)) {
1874                         skb->ip_summed = CHECKSUM_NONE;
1875                         skb->encapsulation = 0;
1876                 }
1877         }
1878
1879         if (vxflags & VXLAN_F_GBP)
1880                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1881
1882         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1883
1884         return udp_tunnel_xmit_skb(rt, sk, skb, src, dst, tos,
1885                                    ttl, df, src_port, dst_port, xnet,
1886                                    !(vxflags & VXLAN_F_UDP_CSUM));
1887 }
1888 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1889
1890 /* Bypass encapsulation if the destination is local */
1891 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1892                                struct vxlan_dev *dst_vxlan)
1893 {
1894         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1895         union vxlan_addr loopback;
1896         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1897         struct net_device *dev = skb->dev;
1898         int len = skb->len;
1899
1900         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1901         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1902         skb->pkt_type = PACKET_HOST;
1903         skb->encapsulation = 0;
1904         skb->dev = dst_vxlan->dev;
1905         __skb_pull(skb, skb_network_offset(skb));
1906
1907         if (remote_ip->sa.sa_family == AF_INET) {
1908                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1909                 loopback.sa.sa_family =  AF_INET;
1910 #if IS_ENABLED(CONFIG_IPV6)
1911         } else {
1912                 loopback.sin6.sin6_addr = in6addr_loopback;
1913                 loopback.sa.sa_family =  AF_INET6;
1914 #endif
1915         }
1916
1917         if (dst_vxlan->flags & VXLAN_F_LEARN)
1918                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1919
1920         u64_stats_update_begin(&tx_stats->syncp);
1921         tx_stats->tx_packets++;
1922         tx_stats->tx_bytes += len;
1923         u64_stats_update_end(&tx_stats->syncp);
1924
1925         if (netif_rx(skb) == NET_RX_SUCCESS) {
1926                 u64_stats_update_begin(&rx_stats->syncp);
1927                 rx_stats->rx_packets++;
1928                 rx_stats->rx_bytes += len;
1929                 u64_stats_update_end(&rx_stats->syncp);
1930         } else {
1931                 dev->stats.rx_dropped++;
1932         }
1933 }
1934
1935 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1936                            struct vxlan_rdst *rdst, bool did_rsc)
1937 {
1938         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1939         struct vxlan_dev *vxlan = netdev_priv(dev);
1940         struct sock *sk = vxlan->vn_sock->sock->sk;
1941         struct rtable *rt = NULL;
1942         const struct iphdr *old_iph;
1943         struct flowi4 fl4;
1944         union vxlan_addr *dst;
1945         union vxlan_addr remote_ip;
1946         struct vxlan_metadata _md;
1947         struct vxlan_metadata *md = &_md;
1948         __be16 src_port = 0, dst_port;
1949         u32 vni;
1950         __be16 df = 0;
1951         __u8 tos, ttl;
1952         int err;
1953         u32 flags = vxlan->flags;
1954
1955         if (rdst) {
1956                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1957                 vni = rdst->remote_vni;
1958                 dst = &rdst->remote_ip;
1959         } else {
1960                 if (!info) {
1961                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1962                                   dev->name);
1963                         goto drop;
1964                 }
1965
1966                 dst_port = info->key.tp_dst ? : vxlan->dst_port;
1967                 vni = be64_to_cpu(info->key.tun_id);
1968                 remote_ip.sin.sin_family = AF_INET;
1969                 remote_ip.sin.sin_addr.s_addr = info->key.ipv4_dst;
1970                 dst = &remote_ip;
1971         }
1972
1973         if (vxlan_addr_any(dst)) {
1974                 if (did_rsc) {
1975                         /* short-circuited back to local bridge */
1976                         vxlan_encap_bypass(skb, vxlan, vxlan);
1977                         return;
1978                 }
1979                 goto drop;
1980         }
1981
1982         old_iph = ip_hdr(skb);
1983
1984         ttl = vxlan->ttl;
1985         if (!ttl && vxlan_addr_multicast(dst))
1986                 ttl = 1;
1987
1988         tos = vxlan->tos;
1989         if (tos == 1)
1990                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1991
1992         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->port_min,
1993                                      vxlan->port_max, true);
1994
1995         if (dst->sa.sa_family == AF_INET) {
1996                 if (info) {
1997                         if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
1998                                 df = htons(IP_DF);
1999                         if (info->key.tun_flags & TUNNEL_CSUM)
2000                                 flags |= VXLAN_F_UDP_CSUM;
2001                         else
2002                                 flags &= ~VXLAN_F_UDP_CSUM;
2003
2004                         ttl = info->key.ipv4_ttl;
2005                         tos = info->key.ipv4_tos;
2006
2007                         if (info->options_len)
2008                                 md = ip_tunnel_info_opts(info, sizeof(*md));
2009                 } else {
2010                         md->gbp = skb->mark;
2011                 }
2012
2013                 memset(&fl4, 0, sizeof(fl4));
2014                 fl4.flowi4_oif = rdst ? rdst->remote_ifindex : 0;
2015                 fl4.flowi4_tos = RT_TOS(tos);
2016                 fl4.flowi4_mark = skb->mark;
2017                 fl4.flowi4_proto = IPPROTO_UDP;
2018                 fl4.daddr = dst->sin.sin_addr.s_addr;
2019                 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
2020
2021                 rt = ip_route_output_key(vxlan->net, &fl4);
2022                 if (IS_ERR(rt)) {
2023                         netdev_dbg(dev, "no route to %pI4\n",
2024                                    &dst->sin.sin_addr.s_addr);
2025                         dev->stats.tx_carrier_errors++;
2026                         goto tx_error;
2027                 }
2028
2029                 if (rt->dst.dev == dev) {
2030                         netdev_dbg(dev, "circular route to %pI4\n",
2031                                    &dst->sin.sin_addr.s_addr);
2032                         dev->stats.collisions++;
2033                         goto rt_tx_error;
2034                 }
2035
2036                 /* Bypass encapsulation if the destination is local */
2037                 if (rt->rt_flags & RTCF_LOCAL &&
2038                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2039                         struct vxlan_dev *dst_vxlan;
2040
2041                         ip_rt_put(rt);
2042                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2043                                                    dst->sa.sa_family, dst_port,
2044                                                    vxlan->flags);
2045                         if (!dst_vxlan)
2046                                 goto tx_error;
2047                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2048                         return;
2049                 }
2050
2051                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2052                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2053                 md->vni = htonl(vni << 8);
2054                 err = vxlan_xmit_skb(rt, sk, skb, fl4.saddr,
2055                                      dst->sin.sin_addr.s_addr, tos, ttl, df,
2056                                      src_port, dst_port, md,
2057                                      !net_eq(vxlan->net, dev_net(vxlan->dev)),
2058                                      flags);
2059                 if (err < 0) {
2060                         /* skb is already freed. */
2061                         skb = NULL;
2062                         goto rt_tx_error;
2063                 }
2064
2065                 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
2066 #if IS_ENABLED(CONFIG_IPV6)
2067         } else {
2068                 struct dst_entry *ndst;
2069                 struct flowi6 fl6;
2070                 u32 flags;
2071
2072                 memset(&fl6, 0, sizeof(fl6));
2073                 fl6.flowi6_oif = rdst ? rdst->remote_ifindex : 0;
2074                 fl6.daddr = dst->sin6.sin6_addr;
2075                 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
2076                 fl6.flowi6_mark = skb->mark;
2077                 fl6.flowi6_proto = IPPROTO_UDP;
2078
2079                 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
2080                         netdev_dbg(dev, "no route to %pI6\n",
2081                                    &dst->sin6.sin6_addr);
2082                         dev->stats.tx_carrier_errors++;
2083                         goto tx_error;
2084                 }
2085
2086                 if (ndst->dev == dev) {
2087                         netdev_dbg(dev, "circular route to %pI6\n",
2088                                    &dst->sin6.sin6_addr);
2089                         dst_release(ndst);
2090                         dev->stats.collisions++;
2091                         goto tx_error;
2092                 }
2093
2094                 /* Bypass encapsulation if the destination is local */
2095                 flags = ((struct rt6_info *)ndst)->rt6i_flags;
2096                 if (flags & RTF_LOCAL &&
2097                     !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2098                         struct vxlan_dev *dst_vxlan;
2099
2100                         dst_release(ndst);
2101                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2102                                                    dst->sa.sa_family, dst_port,
2103                                                    vxlan->flags);
2104                         if (!dst_vxlan)
2105                                 goto tx_error;
2106                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2107                         return;
2108                 }
2109
2110                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2111                 md->vni = htonl(vni << 8);
2112                 md->gbp = skb->mark;
2113
2114                 err = vxlan6_xmit_skb(ndst, sk, skb, dev, &fl6.saddr, &fl6.daddr,
2115                                       0, ttl, src_port, dst_port, md,
2116                                       !net_eq(vxlan->net, dev_net(vxlan->dev)),
2117                                       vxlan->flags);
2118 #endif
2119         }
2120
2121         return;
2122
2123 drop:
2124         dev->stats.tx_dropped++;
2125         goto tx_free;
2126
2127 rt_tx_error:
2128         ip_rt_put(rt);
2129 tx_error:
2130         dev->stats.tx_errors++;
2131 tx_free:
2132         dev_kfree_skb(skb);
2133 }
2134
2135 /* Transmit local packets over Vxlan
2136  *
2137  * Outer IP header inherits ECN and DF from inner header.
2138  * Outer UDP destination is the VXLAN assigned port.
2139  *           source port is based on hash of flow
2140  */
2141 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2142 {
2143         struct vxlan_dev *vxlan = netdev_priv(dev);
2144         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2145         struct ethhdr *eth;
2146         bool did_rsc = false;
2147         struct vxlan_rdst *rdst, *fdst = NULL;
2148         struct vxlan_fdb *f;
2149
2150         skb_reset_mac_header(skb);
2151         eth = eth_hdr(skb);
2152
2153         if ((vxlan->flags & VXLAN_F_PROXY)) {
2154                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2155                         return arp_reduce(dev, skb);
2156 #if IS_ENABLED(CONFIG_IPV6)
2157                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2158                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2159                                        + sizeof(struct nd_msg)) &&
2160                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2161                                 struct nd_msg *msg;
2162
2163                                 msg = (struct nd_msg *)skb_transport_header(skb);
2164                                 if (msg->icmph.icmp6_code == 0 &&
2165                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2166                                         return neigh_reduce(dev, skb);
2167                 }
2168                 eth = eth_hdr(skb);
2169 #endif
2170         }
2171
2172         if (vxlan->flags & VXLAN_F_FLOW_BASED &&
2173             info && info->mode == IP_TUNNEL_INFO_TX) {
2174                 vxlan_xmit_one(skb, dev, NULL, false);
2175                 return NETDEV_TX_OK;
2176         }
2177
2178         f = vxlan_find_mac(vxlan, eth->h_dest);
2179         did_rsc = false;
2180
2181         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2182             (ntohs(eth->h_proto) == ETH_P_IP ||
2183              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2184                 did_rsc = route_shortcircuit(dev, skb);
2185                 if (did_rsc)
2186                         f = vxlan_find_mac(vxlan, eth->h_dest);
2187         }
2188
2189         if (f == NULL) {
2190                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2191                 if (f == NULL) {
2192                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2193                             !is_multicast_ether_addr(eth->h_dest))
2194                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2195
2196                         dev->stats.tx_dropped++;
2197                         kfree_skb(skb);
2198                         return NETDEV_TX_OK;
2199                 }
2200         }
2201
2202         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2203                 struct sk_buff *skb1;
2204
2205                 if (!fdst) {
2206                         fdst = rdst;
2207                         continue;
2208                 }
2209                 skb1 = skb_clone(skb, GFP_ATOMIC);
2210                 if (skb1)
2211                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2212         }
2213
2214         if (fdst)
2215                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2216         else
2217                 kfree_skb(skb);
2218         return NETDEV_TX_OK;
2219 }
2220
2221 /* Walk the forwarding table and purge stale entries */
2222 static void vxlan_cleanup(unsigned long arg)
2223 {
2224         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2225         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2226         unsigned int h;
2227
2228         if (!netif_running(vxlan->dev))
2229                 return;
2230
2231         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2232                 struct hlist_node *p, *n;
2233
2234                 spin_lock_bh(&vxlan->hash_lock);
2235                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2236                         struct vxlan_fdb *f
2237                                 = container_of(p, struct vxlan_fdb, hlist);
2238                         unsigned long timeout;
2239
2240                         if (f->state & NUD_PERMANENT)
2241                                 continue;
2242
2243                         timeout = f->used + vxlan->age_interval * HZ;
2244                         if (time_before_eq(timeout, jiffies)) {
2245                                 netdev_dbg(vxlan->dev,
2246                                            "garbage collect %pM\n",
2247                                            f->eth_addr);
2248                                 f->state = NUD_STALE;
2249                                 vxlan_fdb_destroy(vxlan, f);
2250                         } else if (time_before(timeout, next_timer))
2251                                 next_timer = timeout;
2252                 }
2253                 spin_unlock_bh(&vxlan->hash_lock);
2254         }
2255
2256         mod_timer(&vxlan->age_timer, next_timer);
2257 }
2258
2259 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2260 {
2261         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2262         __u32 vni = vxlan->default_dst.remote_vni;
2263
2264         vxlan->vn_sock = vs;
2265         spin_lock(&vn->sock_lock);
2266         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2267         spin_unlock(&vn->sock_lock);
2268 }
2269
2270 /* Setup stats when device is created */
2271 static int vxlan_init(struct net_device *dev)
2272 {
2273         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2274         if (!dev->tstats)
2275                 return -ENOMEM;
2276
2277         return 0;
2278 }
2279
2280 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2281 {
2282         struct vxlan_fdb *f;
2283
2284         spin_lock_bh(&vxlan->hash_lock);
2285         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2286         if (f)
2287                 vxlan_fdb_destroy(vxlan, f);
2288         spin_unlock_bh(&vxlan->hash_lock);
2289 }
2290
2291 static void vxlan_uninit(struct net_device *dev)
2292 {
2293         struct vxlan_dev *vxlan = netdev_priv(dev);
2294
2295         vxlan_fdb_delete_default(vxlan);
2296
2297         free_percpu(dev->tstats);
2298 }
2299
2300 /* Start ageing timer and join group when device is brought up */
2301 static int vxlan_open(struct net_device *dev)
2302 {
2303         struct vxlan_dev *vxlan = netdev_priv(dev);
2304         struct vxlan_sock *vs;
2305         int ret = 0;
2306
2307         vs = vxlan_sock_add(vxlan->net, vxlan->dst_port, vxlan_rcv, NULL,
2308                             false, vxlan->flags);
2309         if (IS_ERR(vs))
2310                 return PTR_ERR(vs);
2311
2312         vxlan_vs_add_dev(vs, vxlan);
2313
2314         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2315                 ret = vxlan_igmp_join(vxlan);
2316                 if (ret) {
2317                         vxlan_sock_release(vs);
2318                         return ret;
2319                 }
2320         }
2321
2322         if (vxlan->age_interval)
2323                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2324
2325         return ret;
2326 }
2327
2328 /* Purge the forwarding table */
2329 static void vxlan_flush(struct vxlan_dev *vxlan)
2330 {
2331         unsigned int h;
2332
2333         spin_lock_bh(&vxlan->hash_lock);
2334         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2335                 struct hlist_node *p, *n;
2336                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2337                         struct vxlan_fdb *f
2338                                 = container_of(p, struct vxlan_fdb, hlist);
2339                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2340                         if (!is_zero_ether_addr(f->eth_addr))
2341                                 vxlan_fdb_destroy(vxlan, f);
2342                 }
2343         }
2344         spin_unlock_bh(&vxlan->hash_lock);
2345 }
2346
2347 /* Cleanup timer and forwarding table on shutdown */
2348 static int vxlan_stop(struct net_device *dev)
2349 {
2350         struct vxlan_dev *vxlan = netdev_priv(dev);
2351         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2352         struct vxlan_sock *vs = vxlan->vn_sock;
2353         int ret = 0;
2354
2355         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2356             !vxlan_group_used(vn, vxlan))
2357                 ret = vxlan_igmp_leave(vxlan);
2358
2359         del_timer_sync(&vxlan->age_timer);
2360
2361         vxlan_flush(vxlan);
2362         vxlan_sock_release(vs);
2363
2364         return ret;
2365 }
2366
2367 /* Stub, nothing needs to be done. */
2368 static void vxlan_set_multicast_list(struct net_device *dev)
2369 {
2370 }
2371
2372 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2373 {
2374         struct vxlan_dev *vxlan = netdev_priv(dev);
2375         struct vxlan_rdst *dst = &vxlan->default_dst;
2376         struct net_device *lowerdev;
2377         int max_mtu;
2378
2379         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2380         if (lowerdev == NULL)
2381                 return eth_change_mtu(dev, new_mtu);
2382
2383         if (dst->remote_ip.sa.sa_family == AF_INET6)
2384                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2385         else
2386                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2387
2388         if (new_mtu < 68 || new_mtu > max_mtu)
2389                 return -EINVAL;
2390
2391         dev->mtu = new_mtu;
2392         return 0;
2393 }
2394
2395 static const struct net_device_ops vxlan_netdev_ops = {
2396         .ndo_init               = vxlan_init,
2397         .ndo_uninit             = vxlan_uninit,
2398         .ndo_open               = vxlan_open,
2399         .ndo_stop               = vxlan_stop,
2400         .ndo_start_xmit         = vxlan_xmit,
2401         .ndo_get_stats64        = ip_tunnel_get_stats64,
2402         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2403         .ndo_change_mtu         = vxlan_change_mtu,
2404         .ndo_validate_addr      = eth_validate_addr,
2405         .ndo_set_mac_address    = eth_mac_addr,
2406         .ndo_fdb_add            = vxlan_fdb_add,
2407         .ndo_fdb_del            = vxlan_fdb_delete,
2408         .ndo_fdb_dump           = vxlan_fdb_dump,
2409 };
2410
2411 /* Info for udev, that this is a virtual tunnel endpoint */
2412 static struct device_type vxlan_type = {
2413         .name = "vxlan",
2414 };
2415
2416 /* Calls the ndo_add_vxlan_port of the caller in order to
2417  * supply the listening VXLAN udp ports. Callers are expected
2418  * to implement the ndo_add_vxlan_port.
2419  */
2420 void vxlan_get_rx_port(struct net_device *dev)
2421 {
2422         struct vxlan_sock *vs;
2423         struct net *net = dev_net(dev);
2424         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2425         sa_family_t sa_family;
2426         __be16 port;
2427         unsigned int i;
2428
2429         spin_lock(&vn->sock_lock);
2430         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2431                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2432                         port = inet_sk(vs->sock->sk)->inet_sport;
2433                         sa_family = vs->sock->sk->sk_family;
2434                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2435                                                             port);
2436                 }
2437         }
2438         spin_unlock(&vn->sock_lock);
2439 }
2440 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2441
2442 /* Initialize the device structure. */
2443 static void vxlan_setup(struct net_device *dev)
2444 {
2445         struct vxlan_dev *vxlan = netdev_priv(dev);
2446         unsigned int h;
2447
2448         eth_hw_addr_random(dev);
2449         ether_setup(dev);
2450         if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2451                 dev->needed_headroom = ETH_HLEN + VXLAN6_HEADROOM;
2452         else
2453                 dev->needed_headroom = ETH_HLEN + VXLAN_HEADROOM;
2454
2455         dev->netdev_ops = &vxlan_netdev_ops;
2456         dev->destructor = free_netdev;
2457         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2458
2459         dev->tx_queue_len = 0;
2460         dev->features   |= NETIF_F_LLTX;
2461         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2462         dev->features   |= NETIF_F_RXCSUM;
2463         dev->features   |= NETIF_F_GSO_SOFTWARE;
2464
2465         dev->vlan_features = dev->features;
2466         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2467         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2468         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2469         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2470         netif_keep_dst(dev);
2471         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2472
2473         INIT_LIST_HEAD(&vxlan->next);
2474         spin_lock_init(&vxlan->hash_lock);
2475
2476         init_timer_deferrable(&vxlan->age_timer);
2477         vxlan->age_timer.function = vxlan_cleanup;
2478         vxlan->age_timer.data = (unsigned long) vxlan;
2479
2480         vxlan->dst_port = htons(vxlan_port);
2481
2482         vxlan->dev = dev;
2483
2484         for (h = 0; h < FDB_HASH_SIZE; ++h)
2485                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2486 }
2487
2488 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2489         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2490         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2491         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2492         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2493         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2494         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2495         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2496         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2497         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2498         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2499         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2500         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2501         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2502         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2503         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2504         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2505         [IFLA_VXLAN_FLOWBASED]  = { .type = NLA_U8 },
2506         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2507         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2508         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2509         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2510         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2511         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2512         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2513         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2514 };
2515
2516 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2517 {
2518         if (tb[IFLA_ADDRESS]) {
2519                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2520                         pr_debug("invalid link address (not ethernet)\n");
2521                         return -EINVAL;
2522                 }
2523
2524                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2525                         pr_debug("invalid all zero ethernet address\n");
2526                         return -EADDRNOTAVAIL;
2527                 }
2528         }
2529
2530         if (!data)
2531                 return -EINVAL;
2532
2533         if (data[IFLA_VXLAN_ID]) {
2534                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2535                 if (id >= VXLAN_VID_MASK)
2536                         return -ERANGE;
2537         }
2538
2539         if (data[IFLA_VXLAN_PORT_RANGE]) {
2540                 const struct ifla_vxlan_port_range *p
2541                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2542
2543                 if (ntohs(p->high) < ntohs(p->low)) {
2544                         pr_debug("port range %u .. %u not valid\n",
2545                                  ntohs(p->low), ntohs(p->high));
2546                         return -EINVAL;
2547                 }
2548         }
2549
2550         return 0;
2551 }
2552
2553 static void vxlan_get_drvinfo(struct net_device *netdev,
2554                               struct ethtool_drvinfo *drvinfo)
2555 {
2556         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2557         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2558 }
2559
2560 static const struct ethtool_ops vxlan_ethtool_ops = {
2561         .get_drvinfo    = vxlan_get_drvinfo,
2562         .get_link       = ethtool_op_get_link,
2563 };
2564
2565 static void vxlan_del_work(struct work_struct *work)
2566 {
2567         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2568         udp_tunnel_sock_release(vs->sock);
2569         kfree_rcu(vs, rcu);
2570 }
2571
2572 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2573                                         __be16 port, u32 flags)
2574 {
2575         struct socket *sock;
2576         struct udp_port_cfg udp_conf;
2577         int err;
2578
2579         memset(&udp_conf, 0, sizeof(udp_conf));
2580
2581         if (ipv6) {
2582                 udp_conf.family = AF_INET6;
2583                 udp_conf.use_udp6_rx_checksums =
2584                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2585         } else {
2586                 udp_conf.family = AF_INET;
2587         }
2588
2589         udp_conf.local_udp_port = port;
2590
2591         /* Open UDP socket */
2592         err = udp_sock_create(net, &udp_conf, &sock);
2593         if (err < 0)
2594                 return ERR_PTR(err);
2595
2596         return sock;
2597 }
2598
2599 /* Create new listen socket if needed */
2600 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2601                                               vxlan_rcv_t *rcv, void *data,
2602                                               u32 flags)
2603 {
2604         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2605         struct vxlan_sock *vs;
2606         struct socket *sock;
2607         unsigned int h;
2608         bool ipv6 = !!(flags & VXLAN_F_IPV6);
2609         struct udp_tunnel_sock_cfg tunnel_cfg;
2610
2611         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2612         if (!vs)
2613                 return ERR_PTR(-ENOMEM);
2614
2615         for (h = 0; h < VNI_HASH_SIZE; ++h)
2616                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2617
2618         INIT_WORK(&vs->del_work, vxlan_del_work);
2619
2620         sock = vxlan_create_sock(net, ipv6, port, flags);
2621         if (IS_ERR(sock)) {
2622                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2623                         PTR_ERR(sock));
2624                 kfree(vs);
2625                 return ERR_CAST(sock);
2626         }
2627
2628         vs->sock = sock;
2629         atomic_set(&vs->refcnt, 1);
2630         vs->rcv = rcv;
2631         vs->data = data;
2632         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2633
2634         /* Initialize the vxlan udp offloads structure */
2635         vs->udp_offloads.port = port;
2636         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2637         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2638
2639         spin_lock(&vn->sock_lock);
2640         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2641         vxlan_notify_add_rx_port(vs);
2642         spin_unlock(&vn->sock_lock);
2643
2644         /* Mark socket as an encapsulation socket. */
2645         tunnel_cfg.sk_user_data = vs;
2646         tunnel_cfg.encap_type = 1;
2647         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2648         tunnel_cfg.encap_destroy = NULL;
2649
2650         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2651
2652         return vs;
2653 }
2654
2655 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2656                                   vxlan_rcv_t *rcv, void *data,
2657                                   bool no_share, u32 flags)
2658 {
2659         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2660         struct vxlan_sock *vs;
2661         bool ipv6 = flags & VXLAN_F_IPV6;
2662
2663         if (!no_share) {
2664                 spin_lock(&vn->sock_lock);
2665                 vs = vxlan_find_sock(net, ipv6 ? AF_INET6 : AF_INET, port,
2666                                      flags);
2667                 if (vs && vs->rcv == rcv) {
2668                         if (!atomic_add_unless(&vs->refcnt, 1, 0))
2669                                 vs = ERR_PTR(-EBUSY);
2670                         spin_unlock(&vn->sock_lock);
2671                         return vs;
2672                 }
2673                 spin_unlock(&vn->sock_lock);
2674         }
2675
2676         return vxlan_socket_create(net, port, rcv, data, flags);
2677 }
2678 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2679
2680 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2681                          struct nlattr *tb[], struct nlattr *data[])
2682 {
2683         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2684         struct vxlan_dev *vxlan = netdev_priv(dev);
2685         struct vxlan_rdst *dst = &vxlan->default_dst;
2686         __u32 vni;
2687         int err;
2688         bool use_ipv6 = false;
2689
2690         if (!data[IFLA_VXLAN_ID])
2691                 return -EINVAL;
2692
2693         vxlan->net = src_net;
2694
2695         vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2696         dst->remote_vni = vni;
2697
2698         /* Unless IPv6 is explicitly requested, assume IPv4 */
2699         dst->remote_ip.sa.sa_family = AF_INET;
2700         if (data[IFLA_VXLAN_GROUP]) {
2701                 dst->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2702         } else if (data[IFLA_VXLAN_GROUP6]) {
2703                 if (!IS_ENABLED(CONFIG_IPV6))
2704                         return -EPFNOSUPPORT;
2705
2706                 dst->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2707                 dst->remote_ip.sa.sa_family = AF_INET6;
2708                 use_ipv6 = true;
2709         }
2710
2711         if (data[IFLA_VXLAN_LOCAL]) {
2712                 vxlan->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2713                 vxlan->saddr.sa.sa_family = AF_INET;
2714         } else if (data[IFLA_VXLAN_LOCAL6]) {
2715                 if (!IS_ENABLED(CONFIG_IPV6))
2716                         return -EPFNOSUPPORT;
2717
2718                 /* TODO: respect scope id */
2719                 vxlan->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2720                 vxlan->saddr.sa.sa_family = AF_INET6;
2721                 use_ipv6 = true;
2722         }
2723
2724         if (data[IFLA_VXLAN_LINK] &&
2725             (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2726                 struct net_device *lowerdev
2727                          = __dev_get_by_index(src_net, dst->remote_ifindex);
2728
2729                 if (!lowerdev) {
2730                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2731                         return -ENODEV;
2732                 }
2733
2734 #if IS_ENABLED(CONFIG_IPV6)
2735                 if (use_ipv6) {
2736                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2737                         if (idev && idev->cnf.disable_ipv6) {
2738                                 pr_info("IPv6 is disabled via sysctl\n");
2739                                 return -EPERM;
2740                         }
2741                         vxlan->flags |= VXLAN_F_IPV6;
2742                 }
2743 #endif
2744
2745                 if (!tb[IFLA_MTU])
2746                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2747
2748                 dev->needed_headroom = lowerdev->hard_header_len +
2749                                        (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2750         } else if (use_ipv6)
2751                 vxlan->flags |= VXLAN_F_IPV6;
2752
2753         if (data[IFLA_VXLAN_TOS])
2754                 vxlan->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2755
2756         if (data[IFLA_VXLAN_TTL])
2757                 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2758
2759         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2760                 vxlan->flags |= VXLAN_F_LEARN;
2761
2762         if (data[IFLA_VXLAN_AGEING])
2763                 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2764         else
2765                 vxlan->age_interval = FDB_AGE_DEFAULT;
2766
2767         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2768                 vxlan->flags |= VXLAN_F_PROXY;
2769
2770         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2771                 vxlan->flags |= VXLAN_F_RSC;
2772
2773         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2774                 vxlan->flags |= VXLAN_F_L2MISS;
2775
2776         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2777                 vxlan->flags |= VXLAN_F_L3MISS;
2778
2779         if (data[IFLA_VXLAN_LIMIT])
2780                 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2781
2782         if (data[IFLA_VXLAN_FLOWBASED] &&
2783             nla_get_u8(data[IFLA_VXLAN_FLOWBASED]))
2784                 vxlan->flags |= VXLAN_F_FLOW_BASED;
2785
2786         if (data[IFLA_VXLAN_PORT_RANGE]) {
2787                 const struct ifla_vxlan_port_range *p
2788                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2789                 vxlan->port_min = ntohs(p->low);
2790                 vxlan->port_max = ntohs(p->high);
2791         }
2792
2793         if (data[IFLA_VXLAN_PORT])
2794                 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2795
2796         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2797                 vxlan->flags |= VXLAN_F_UDP_CSUM;
2798
2799         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2800             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2801                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2802
2803         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2804             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2805                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2806
2807         if (data[IFLA_VXLAN_REMCSUM_TX] &&
2808             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2809                 vxlan->flags |= VXLAN_F_REMCSUM_TX;
2810
2811         if (data[IFLA_VXLAN_REMCSUM_RX] &&
2812             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2813                 vxlan->flags |= VXLAN_F_REMCSUM_RX;
2814
2815         if (data[IFLA_VXLAN_GBP])
2816                 vxlan->flags |= VXLAN_F_GBP;
2817
2818         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2819                 vxlan->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2820
2821         if (vxlan_find_vni(src_net, vni, use_ipv6 ? AF_INET6 : AF_INET,
2822                            vxlan->dst_port, vxlan->flags)) {
2823                 pr_info("duplicate VNI %u\n", vni);
2824                 return -EEXIST;
2825         }
2826
2827         dev->ethtool_ops = &vxlan_ethtool_ops;
2828
2829         /* create an fdb entry for a valid default destination */
2830         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2831                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2832                                        &vxlan->default_dst.remote_ip,
2833                                        NUD_REACHABLE|NUD_PERMANENT,
2834                                        NLM_F_EXCL|NLM_F_CREATE,
2835                                        vxlan->dst_port,
2836                                        vxlan->default_dst.remote_vni,
2837                                        vxlan->default_dst.remote_ifindex,
2838                                        NTF_SELF);
2839                 if (err)
2840                         return err;
2841         }
2842
2843         err = register_netdevice(dev);
2844         if (err) {
2845                 vxlan_fdb_delete_default(vxlan);
2846                 return err;
2847         }
2848
2849         list_add(&vxlan->next, &vn->vxlan_list);
2850
2851         return 0;
2852 }
2853
2854 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2855 {
2856         struct vxlan_dev *vxlan = netdev_priv(dev);
2857         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2858
2859         spin_lock(&vn->sock_lock);
2860         if (!hlist_unhashed(&vxlan->hlist))
2861                 hlist_del_rcu(&vxlan->hlist);
2862         spin_unlock(&vn->sock_lock);
2863
2864         list_del(&vxlan->next);
2865         unregister_netdevice_queue(dev, head);
2866 }
2867
2868 static size_t vxlan_get_size(const struct net_device *dev)
2869 {
2870
2871         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2872                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2873                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2874                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2875                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2876                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2877                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2878                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2879                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2880                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2881                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2882                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_FLOWBASED */
2883                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2884                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2885                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2886                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2887                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2888                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2889                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2890                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
2891                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
2892                 0;
2893 }
2894
2895 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2896 {
2897         const struct vxlan_dev *vxlan = netdev_priv(dev);
2898         const struct vxlan_rdst *dst = &vxlan->default_dst;
2899         struct ifla_vxlan_port_range ports = {
2900                 .low =  htons(vxlan->port_min),
2901                 .high = htons(vxlan->port_max),
2902         };
2903
2904         if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2905                 goto nla_put_failure;
2906
2907         if (!vxlan_addr_any(&dst->remote_ip)) {
2908                 if (dst->remote_ip.sa.sa_family == AF_INET) {
2909                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
2910                                             dst->remote_ip.sin.sin_addr.s_addr))
2911                                 goto nla_put_failure;
2912 #if IS_ENABLED(CONFIG_IPV6)
2913                 } else {
2914                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
2915                                              &dst->remote_ip.sin6.sin6_addr))
2916                                 goto nla_put_failure;
2917 #endif
2918                 }
2919         }
2920
2921         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2922                 goto nla_put_failure;
2923
2924         if (!vxlan_addr_any(&vxlan->saddr)) {
2925                 if (vxlan->saddr.sa.sa_family == AF_INET) {
2926                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
2927                                             vxlan->saddr.sin.sin_addr.s_addr))
2928                                 goto nla_put_failure;
2929 #if IS_ENABLED(CONFIG_IPV6)
2930                 } else {
2931                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
2932                                              &vxlan->saddr.sin6.sin6_addr))
2933                                 goto nla_put_failure;
2934 #endif
2935                 }
2936         }
2937
2938         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2939             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2940             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2941                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
2942             nla_put_u8(skb, IFLA_VXLAN_PROXY,
2943                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
2944             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2945             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2946                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2947             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2948                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2949             nla_put_u8(skb, IFLA_VXLAN_FLOWBASED,
2950                        !!(vxlan->flags & VXLAN_F_FLOW_BASED)) ||
2951             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2952             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2953             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port) ||
2954             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2955                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2956             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2957                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2958             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2959                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
2960             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
2961                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
2962             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
2963                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
2964                 goto nla_put_failure;
2965
2966         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2967                 goto nla_put_failure;
2968
2969         if (vxlan->flags & VXLAN_F_GBP &&
2970             nla_put_flag(skb, IFLA_VXLAN_GBP))
2971                 goto nla_put_failure;
2972
2973         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
2974             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
2975                 goto nla_put_failure;
2976
2977         return 0;
2978
2979 nla_put_failure:
2980         return -EMSGSIZE;
2981 }
2982
2983 static struct net *vxlan_get_link_net(const struct net_device *dev)
2984 {
2985         struct vxlan_dev *vxlan = netdev_priv(dev);
2986
2987         return vxlan->net;
2988 }
2989
2990 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2991         .kind           = "vxlan",
2992         .maxtype        = IFLA_VXLAN_MAX,
2993         .policy         = vxlan_policy,
2994         .priv_size      = sizeof(struct vxlan_dev),
2995         .setup          = vxlan_setup,
2996         .validate       = vxlan_validate,
2997         .newlink        = vxlan_newlink,
2998         .dellink        = vxlan_dellink,
2999         .get_size       = vxlan_get_size,
3000         .fill_info      = vxlan_fill_info,
3001         .get_link_net   = vxlan_get_link_net,
3002 };
3003
3004 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3005                                              struct net_device *dev)
3006 {
3007         struct vxlan_dev *vxlan, *next;
3008         LIST_HEAD(list_kill);
3009
3010         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3011                 struct vxlan_rdst *dst = &vxlan->default_dst;
3012
3013                 /* In case we created vxlan device with carrier
3014                  * and we loose the carrier due to module unload
3015                  * we also need to remove vxlan device. In other
3016                  * cases, it's not necessary and remote_ifindex
3017                  * is 0 here, so no matches.
3018                  */
3019                 if (dst->remote_ifindex == dev->ifindex)
3020                         vxlan_dellink(vxlan->dev, &list_kill);
3021         }
3022
3023         unregister_netdevice_many(&list_kill);
3024 }
3025
3026 static int vxlan_lowerdev_event(struct notifier_block *unused,
3027                                 unsigned long event, void *ptr)
3028 {
3029         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3030         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3031
3032         if (event == NETDEV_UNREGISTER)
3033                 vxlan_handle_lowerdev_unregister(vn, dev);
3034
3035         return NOTIFY_DONE;
3036 }
3037
3038 static struct notifier_block vxlan_notifier_block __read_mostly = {
3039         .notifier_call = vxlan_lowerdev_event,
3040 };
3041
3042 static __net_init int vxlan_init_net(struct net *net)
3043 {
3044         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3045         unsigned int h;
3046
3047         INIT_LIST_HEAD(&vn->vxlan_list);
3048         spin_lock_init(&vn->sock_lock);
3049
3050         for (h = 0; h < PORT_HASH_SIZE; ++h)
3051                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3052
3053         return 0;
3054 }
3055
3056 static void __net_exit vxlan_exit_net(struct net *net)
3057 {
3058         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3059         struct vxlan_dev *vxlan, *next;
3060         struct net_device *dev, *aux;
3061         LIST_HEAD(list);
3062
3063         rtnl_lock();
3064         for_each_netdev_safe(net, dev, aux)
3065                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3066                         unregister_netdevice_queue(dev, &list);
3067
3068         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3069                 /* If vxlan->dev is in the same netns, it has already been added
3070                  * to the list by the previous loop.
3071                  */
3072                 if (!net_eq(dev_net(vxlan->dev), net))
3073                         unregister_netdevice_queue(vxlan->dev, &list);
3074         }
3075
3076         unregister_netdevice_many(&list);
3077         rtnl_unlock();
3078 }
3079
3080 static struct pernet_operations vxlan_net_ops = {
3081         .init = vxlan_init_net,
3082         .exit = vxlan_exit_net,
3083         .id   = &vxlan_net_id,
3084         .size = sizeof(struct vxlan_net),
3085 };
3086
3087 static int __init vxlan_init_module(void)
3088 {
3089         int rc;
3090
3091         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3092         if (!vxlan_wq)
3093                 return -ENOMEM;
3094
3095         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3096
3097         rc = register_pernet_subsys(&vxlan_net_ops);
3098         if (rc)
3099                 goto out1;
3100
3101         rc = register_netdevice_notifier(&vxlan_notifier_block);
3102         if (rc)
3103                 goto out2;
3104
3105         rc = rtnl_link_register(&vxlan_link_ops);
3106         if (rc)
3107                 goto out3;
3108
3109         return 0;
3110 out3:
3111         unregister_netdevice_notifier(&vxlan_notifier_block);
3112 out2:
3113         unregister_pernet_subsys(&vxlan_net_ops);
3114 out1:
3115         destroy_workqueue(vxlan_wq);
3116         return rc;
3117 }
3118 late_initcall(vxlan_init_module);
3119
3120 static void __exit vxlan_cleanup_module(void)
3121 {
3122         rtnl_link_unregister(&vxlan_link_ops);
3123         unregister_netdevice_notifier(&vxlan_notifier_block);
3124         destroy_workqueue(vxlan_wq);
3125         unregister_pernet_subsys(&vxlan_net_ops);
3126         /* rcu_barrier() is called by netns */
3127 }
3128 module_exit(vxlan_cleanup_module);
3129
3130 MODULE_LICENSE("GPL");
3131 MODULE_VERSION(VXLAN_VERSION);
3132 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3133 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3134 MODULE_ALIAS_RTNL_LINK("vxlan");