drivers/net/vxlan.c

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