drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/types.h>
  15 #include <linux/module.h>
  16 #include <linux/errno.h>
  17 #include <linux/slab.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/rculist.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/in.h>
  22 #include <linux/ip.h>
  23 #include <linux/udp.h>
  24 #include <linux/igmp.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/hash.h>
  29 #include <linux/ethtool.h>
  30 #include <net/arp.h>
  31 #include <net/ndisc.h>
  32 #include <net/ip.h>
  33 #include <net/ip_tunnels.h>
  34 #include <net/icmp.h>
  35 #include <net/udp.h>
  36 #include <net/udp_tunnel.h>
  37 #include <net/rtnetlink.h>
  38 #include <net/route.h>
  39 #include <net/dsfield.h>
  40 #include <net/inet_ecn.h>
  41 #include <net/net_namespace.h>
  42 #include <net/netns/generic.h>
  43 #include <net/vxlan.h>
  44 #include <net/protocol.h>
  45 #include <net/udp_tunnel.h>
  46 #if IS_ENABLED(CONFIG_IPV6)
  47 #include <net/ipv6.h>
  48 #include <net/addrconf.h>
  49 #include <net/ip6_tunnel.h>
  50 #include <net/ip6_checksum.h>
  51 #endif
  52
  53 #define VXLAN_VERSION   "0.1"
  54
  55 #define PORT_HASH_BITS  8
  56 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  57 #define VNI_HASH_BITS   10
  58 #define VNI_HASH_SIZE   (1<<VNI_HASH_BITS)
  59 #define FDB_HASH_BITS   8
  60 #define FDB_HASH_SIZE   (1<<FDB_HASH_BITS)
  61 #define FDB_AGE_DEFAULT 300 /* 5 min */
  62 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  63
  64 /* UDP port for VXLAN traffic.
  65  * The IANA assigned port is 4789, but the Linux default is 8472
  66  * for compatibility with early adopters.
  67  */
  68 static unsigned short vxlan_port __read_mostly = 8472;
  69 module_param_named(udp_port, vxlan_port, ushort, 0444);
  70 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  71
  72 static bool log_ecn_error = true;
  73 module_param(log_ecn_error, bool, 0644);
  74 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  75
  76 static int vxlan_net_id;
  77
  78 static const u8 all_zeros_mac[ETH_ALEN];
  79
  80 /* per-network namespace private data for this module */
  81 struct vxlan_net {
  82         struct list_head  vxlan_list;
  83         struct hlist_head sock_list[PORT_HASH_SIZE];
  84         spinlock_t        sock_lock;
  85 };
  86
  87 union vxlan_addr {
  88         struct sockaddr_in sin;
  89         struct sockaddr_in6 sin6;
  90         struct sockaddr sa;
  91 };
  92
  93 struct vxlan_rdst {
  94         union vxlan_addr         remote_ip;
  95         __be16                   remote_port;
  96         u32                      remote_vni;
  97         u32                      remote_ifindex;
  98         struct list_head         list;
  99         struct rcu_head          rcu;
 100 };
 101
 102 /* Forwarding table entry */
 103 struct vxlan_fdb {
 104         struct hlist_node hlist;        /* linked list of entries */
 105         struct rcu_head   rcu;
 106         unsigned long     updated;      /* jiffies */
 107         unsigned long     used;
 108         struct list_head  remotes;
 109         u16               state;        /* see ndm_state */
 110         u8                flags;        /* see ndm_flags */
 111         u8                eth_addr[ETH_ALEN];
 112 };
 113
 114 /* Pseudo network device */
 115 struct vxlan_dev {
 116         struct hlist_node hlist;        /* vni hash table */
 117         struct list_head  next;         /* vxlan's per namespace list */
 118         struct vxlan_sock *vn_sock;     /* listening socket */
 119         struct net_device *dev;
 120         struct net        *net;         /* netns for packet i/o */
 121         struct vxlan_rdst default_dst;  /* default destination */
 122         union vxlan_addr  saddr;        /* source address */
 123         __be16            dst_port;
 124         __u16             port_min;     /* source port range */
 125         __u16             port_max;
 126         __u8              tos;          /* TOS override */
 127         __u8              ttl;
 128         u32               flags;        /* VXLAN_F_* in vxlan.h */
 129
 130         struct work_struct sock_work;
 131         struct work_struct igmp_join;
 132         struct work_struct igmp_leave;
 133
 134         unsigned long     age_interval;
 135         struct timer_list age_timer;
 136         spinlock_t        hash_lock;
 137         unsigned int      addrcnt;
 138         unsigned int      addrmax;
 139
 140         struct hlist_head fdb_head[FDB_HASH_SIZE];
 141 };
 142
 143 /* salt for hash table */
 144 static u32 vxlan_salt __read_mostly;
 145 static struct workqueue_struct *vxlan_wq;
 146
 147 static void vxlan_sock_work(struct work_struct *work);
 148
 149 #if IS_ENABLED(CONFIG_IPV6)
 150 static inline
 151 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 152 {
 153        if (a->sa.sa_family != b->sa.sa_family)
 154                return false;
 155        if (a->sa.sa_family == AF_INET6)
 156                return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 157        else
 158                return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 159 }
 160
 161 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 162 {
 163        if (ipa->sa.sa_family == AF_INET6)
 164                return ipv6_addr_any(&ipa->sin6.sin6_addr);
 165        else
 166                return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 167 }
 168
 169 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 170 {
 171        if (ipa->sa.sa_family == AF_INET6)
 172                return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 173        else
 174                return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 175 }
 176
 177 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 178 {
 179        if (nla_len(nla) >= sizeof(struct in6_addr)) {
 180                nla_memcpy(&ip->sin6.sin6_addr, nla, sizeof(struct in6_addr));
 181                ip->sa.sa_family = AF_INET6;
 182                return 0;
 183        } else if (nla_len(nla) >= sizeof(__be32)) {
 184                ip->sin.sin_addr.s_addr = nla_get_be32(nla);
 185                ip->sa.sa_family = AF_INET;
 186                return 0;
 187        } else {
 188                return -EAFNOSUPPORT;
 189        }
 190 }
 191
 192 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 193                              const union vxlan_addr *ip)
 194 {
 195        if (ip->sa.sa_family == AF_INET6)
 196                return nla_put(skb, attr, sizeof(struct in6_addr), &ip->sin6.sin6_addr);
 197        else
 198                return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
 199 }
 200
 201 #else /* !CONFIG_IPV6 */
 202
 203 static inline
 204 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 205 {
 206        return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 207 }
 208
 209 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 210 {
 211        return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 212 }
 213
 214 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 215 {
 216        return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 217 }
 218
 219 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 220 {
 221        if (nla_len(nla) >= sizeof(struct in6_addr)) {
 222                return -EAFNOSUPPORT;
 223        } else if (nla_len(nla) >= sizeof(__be32)) {
 224                ip->sin.sin_addr.s_addr = nla_get_be32(nla);
 225                ip->sa.sa_family = AF_INET;
 226                return 0;
 227        } else {
 228                return -EAFNOSUPPORT;
 229        }
 230 }
 231
 232 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 233                              const union vxlan_addr *ip)
 234 {
 235        return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
 236 }
 237 #endif
 238
 239 /* Virtual Network hash table head */
 240 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
 241 {
 242         return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
 243 }
 244
 245 /* Socket hash table head */
 246 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 247 {
 248         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 249
 250         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 251 }
 252
 253 /* First remote destination for a forwarding entry.
 254  * Guaranteed to be non-NULL because remotes are never deleted.
 255  */
 256 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 257 {
 258         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 259 }
 260
 261 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 262 {
 263         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 264 }
 265
 266 /* Find VXLAN socket based on network namespace, address family and UDP port
 267  * and enabled unshareable flags.
 268  */
 269 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 270                                           __be16 port, u32 flags)
 271 {
 272         struct vxlan_sock *vs;
 273
 274         flags &= VXLAN_F_RCV_FLAGS;
 275
 276         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 277                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 278                     inet_sk(vs->sock->sk)->sk.sk_family == family &&
 279                     vs->flags == flags)
 280                         return vs;
 281         }
 282         return NULL;
 283 }
 284
 285 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
 286 {
 287         struct vxlan_dev *vxlan;
 288
 289         hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
 290                 if (vxlan->default_dst.remote_vni == id)
 291                         return vxlan;
 292         }
 293
 294         return NULL;
 295 }
 296
 297 /* Look up VNI in a per net namespace table */
 298 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id,
 299                                         sa_family_t family, __be16 port,
 300                                         u32 flags)
 301 {
 302         struct vxlan_sock *vs;
 303
 304         vs = vxlan_find_sock(net, family, port, flags);
 305         if (!vs)
 306                 return NULL;
 307
 308         return vxlan_vs_find_vni(vs, id);
 309 }
 310
 311 /* Fill in neighbour message in skbuff. */
 312 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 313                           const struct vxlan_fdb *fdb,
 314                           u32 portid, u32 seq, int type, unsigned int flags,
 315                           const struct vxlan_rdst *rdst)
 316 {
 317         unsigned long now = jiffies;
 318         struct nda_cacheinfo ci;
 319         struct nlmsghdr *nlh;
 320         struct ndmsg *ndm;
 321         bool send_ip, send_eth;
 322
 323         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 324         if (nlh == NULL)
 325                 return -EMSGSIZE;
 326
 327         ndm = nlmsg_data(nlh);
 328         memset(ndm, 0, sizeof(*ndm));
 329
 330         send_eth = send_ip = true;
 331
 332         if (type == RTM_GETNEIGH) {
 333                 ndm->ndm_family = AF_INET;
 334                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 335                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 336         } else
 337                 ndm->ndm_family = AF_BRIDGE;
 338         ndm->ndm_state = fdb->state;
 339         ndm->ndm_ifindex = vxlan->dev->ifindex;
 340         ndm->ndm_flags = fdb->flags;
 341         ndm->ndm_type = RTN_UNICAST;
 342
 343         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 344             nla_put_s32(skb, NDA_LINK_NETNSID,
 345                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 346                 goto nla_put_failure;
 347
 348         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 349                 goto nla_put_failure;
 350
 351         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 352                 goto nla_put_failure;
 353
 354         if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
 355             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 356                 goto nla_put_failure;
 357         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 358             nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
 359                 goto nla_put_failure;
 360         if (rdst->remote_ifindex &&
 361             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 362                 goto nla_put_failure;
 363
 364         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 365         ci.ndm_confirmed = 0;
 366         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 367         ci.ndm_refcnt    = 0;
 368
 369         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 370                 goto nla_put_failure;
 371
 372         nlmsg_end(skb, nlh);
 373         return 0;
 374
 375 nla_put_failure:
 376         nlmsg_cancel(skb, nlh);
 377         return -EMSGSIZE;
 378 }
 379
 380 static inline size_t vxlan_nlmsg_size(void)
 381 {
 382         return NLMSG_ALIGN(sizeof(struct ndmsg))
 383                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 384                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 385                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 386                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 387                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 388                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 389                 + nla_total_size(sizeof(struct nda_cacheinfo));
 390 }
 391
 392 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 393                              struct vxlan_rdst *rd, int type)
 394 {
 395         struct net *net = dev_net(vxlan->dev);
 396         struct sk_buff *skb;
 397         int err = -ENOBUFS;
 398
 399         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 400         if (skb == NULL)
 401                 goto errout;
 402
 403         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 404         if (err < 0) {
 405                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 406                 WARN_ON(err == -EMSGSIZE);
 407                 kfree_skb(skb);
 408                 goto errout;
 409         }
 410
 411         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 412         return;
 413 errout:
 414         if (err < 0)
 415                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 416 }
 417
 418 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 419 {
 420         struct vxlan_dev *vxlan = netdev_priv(dev);
 421         struct vxlan_fdb f = {
 422                 .state = NUD_STALE,
 423         };
 424         struct vxlan_rdst remote = {
 425                 .remote_ip = *ipa, /* goes to NDA_DST */
 426                 .remote_vni = VXLAN_N_VID,
 427         };
 428
 429         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 430 }
 431
 432 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 433 {
 434         struct vxlan_fdb f = {
 435                 .state = NUD_STALE,
 436         };
 437         struct vxlan_rdst remote = { };
 438
 439         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 440
 441         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 442 }
 443
 444 /* Hash Ethernet address */
 445 static u32 eth_hash(const unsigned char *addr)
 446 {
 447         u64 value = get_unaligned((u64 *)addr);
 448
 449         /* only want 6 bytes */
 450 #ifdef __BIG_ENDIAN
 451         value >>= 16;
 452 #else
 453         value <<= 16;
 454 #endif
 455         return hash_64(value, FDB_HASH_BITS);
 456 }
 457
 458 /* Hash chain to use given mac address */
 459 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 460                                                 const u8 *mac)
 461 {
 462         return &vxlan->fdb_head[eth_hash(mac)];
 463 }
 464
 465 /* Look up Ethernet address in forwarding table */
 466 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 467                                         const u8 *mac)
 468 {
 469         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 470         struct vxlan_fdb *f;
 471
 472         hlist_for_each_entry_rcu(f, head, hlist) {
 473                 if (ether_addr_equal(mac, f->eth_addr))
 474                         return f;
 475         }
 476
 477         return NULL;
 478 }
 479
 480 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 481                                         const u8 *mac)
 482 {
 483         struct vxlan_fdb *f;
 484
 485         f = __vxlan_find_mac(vxlan, mac);
 486         if (f)
 487                 f->used = jiffies;
 488
 489         return f;
 490 }
 491
 492 /* caller should hold vxlan->hash_lock */
 493 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 494                                               union vxlan_addr *ip, __be16 port,
 495                                               __u32 vni, __u32 ifindex)
 496 {
 497         struct vxlan_rdst *rd;
 498
 499         list_for_each_entry(rd, &f->remotes, list) {
 500                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 501                     rd->remote_port == port &&
 502                     rd->remote_vni == vni &&
 503                     rd->remote_ifindex == ifindex)
 504                         return rd;
 505         }
 506
 507         return NULL;
 508 }
 509
 510 /* Replace destination of unicast mac */
 511 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 512                              union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
 513 {
 514         struct vxlan_rdst *rd;
 515
 516         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 517         if (rd)
 518                 return 0;
 519
 520         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 521         if (!rd)
 522                 return 0;
 523         rd->remote_ip = *ip;
 524         rd->remote_port = port;
 525         rd->remote_vni = vni;
 526         rd->remote_ifindex = ifindex;
 527         return 1;
 528 }
 529
 530 /* Add/update destinations for multicast */
 531 static int vxlan_fdb_append(struct vxlan_fdb *f,
 532                             union vxlan_addr *ip, __be16 port, __u32 vni,
 533                             __u32 ifindex, struct vxlan_rdst **rdp)
 534 {
 535         struct vxlan_rdst *rd;
 536
 537         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 538         if (rd)
 539                 return 0;
 540
 541         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 542         if (rd == NULL)
 543                 return -ENOBUFS;
 544         rd->remote_ip = *ip;
 545         rd->remote_port = port;
 546         rd->remote_vni = vni;
 547         rd->remote_ifindex = ifindex;
 548
 549         list_add_tail_rcu(&rd->list, &f->remotes);
 550
 551         *rdp = rd;
 552         return 1;
 553 }
 554
 555 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 556                                           unsigned int off,
 557                                           struct vxlanhdr *vh, size_t hdrlen,
 558                                           u32 data)
 559 {
 560         size_t start, offset, plen;
 561
 562         if (skb->remcsum_offload)
 563                 return vh;
 564
 565         if (!NAPI_GRO_CB(skb)->csum_valid)
 566                 return NULL;
 567
 568         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
 569         offset = start + ((data & VXLAN_RCO_UDP) ?
 570                           offsetof(struct udphdr, check) :
 571                           offsetof(struct tcphdr, check));
 572
 573         plen = hdrlen + offset + sizeof(u16);
 574
 575         /* Pull checksum that will be written */
 576         if (skb_gro_header_hard(skb, off + plen)) {
 577                 vh = skb_gro_header_slow(skb, off + plen, off);
 578                 if (!vh)
 579                         return NULL;
 580         }
 581
 582         skb_gro_remcsum_process(skb, (void *)vh + hdrlen, start, offset);
 583
 584         skb->remcsum_offload = 1;
 585
 586         return vh;
 587 }
 588
 589 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
 590                                           struct sk_buff *skb,
 591                                           struct udp_offload *uoff)
 592 {
 593         struct sk_buff *p, **pp = NULL;
 594         struct vxlanhdr *vh, *vh2;
 595         unsigned int hlen, off_vx;
 596         int flush = 1;
 597         struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
 598                                              udp_offloads);
 599         u32 flags;
 600
 601         off_vx = skb_gro_offset(skb);
 602         hlen = off_vx + sizeof(*vh);
 603         vh   = skb_gro_header_fast(skb, off_vx);
 604         if (skb_gro_header_hard(skb, hlen)) {
 605                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 606                 if (unlikely(!vh))
 607                         goto out;
 608         }
 609
 610         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 611         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 612
 613         flags = ntohl(vh->vx_flags);
 614
 615         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 616                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 617                                        ntohl(vh->vx_vni));
 618
 619                 if (!vh)
 620                         goto out;
 621         }
 622
 623         flush = 0;
 624
 625         for (p = *head; p; p = p->next) {
 626                 if (!NAPI_GRO_CB(p)->same_flow)
 627                         continue;
 628
 629                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 630                 if (vh->vx_flags != vh2->vx_flags ||
 631                     vh->vx_vni != vh2->vx_vni) {
 632                         NAPI_GRO_CB(p)->same_flow = 0;
 633                         continue;
 634                 }
 635         }
 636
 637         pp = eth_gro_receive(head, skb);
 638
 639 out:
 640         NAPI_GRO_CB(skb)->flush |= flush;
 641
 642         return pp;
 643 }
 644
 645 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
 646                               struct udp_offload *uoff)
 647 {
 648         udp_tunnel_gro_complete(skb, nhoff);
 649
 650         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 651 }
 652
 653 /* Notify netdevs that UDP port started listening */
 654 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 655 {
 656         struct net_device *dev;
 657         struct sock *sk = vs->sock->sk;
 658         struct net *net = sock_net(sk);
 659         sa_family_t sa_family = sk->sk_family;
 660         __be16 port = inet_sk(sk)->inet_sport;
 661         int err;
 662
 663         if (sa_family == AF_INET) {
 664                 err = udp_add_offload(&vs->udp_offloads);
 665                 if (err)
 666                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 667         }
 668
 669         rcu_read_lock();
 670         for_each_netdev_rcu(net, dev) {
 671                 if (dev->netdev_ops->ndo_add_vxlan_port)
 672                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 673                                                             port);
 674         }
 675         rcu_read_unlock();
 676 }
 677
 678 /* Notify netdevs that UDP port is no more listening */
 679 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 680 {
 681         struct net_device *dev;
 682         struct sock *sk = vs->sock->sk;
 683         struct net *net = sock_net(sk);
 684         sa_family_t sa_family = sk->sk_family;
 685         __be16 port = inet_sk(sk)->inet_sport;
 686
 687         rcu_read_lock();
 688         for_each_netdev_rcu(net, dev) {
 689                 if (dev->netdev_ops->ndo_del_vxlan_port)
 690                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 691                                                             port);
 692         }
 693         rcu_read_unlock();
 694
 695         if (sa_family == AF_INET)
 696                 udp_del_offload(&vs->udp_offloads);
 697 }
 698
 699 /* Add new entry to forwarding table -- assumes lock held */
 700 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 701                             const u8 *mac, union vxlan_addr *ip,
 702                             __u16 state, __u16 flags,
 703                             __be16 port, __u32 vni, __u32 ifindex,
 704                             __u8 ndm_flags)
 705 {
 706         struct vxlan_rdst *rd = NULL;
 707         struct vxlan_fdb *f;
 708         int notify = 0;
 709
 710         f = __vxlan_find_mac(vxlan, mac);
 711         if (f) {
 712                 if (flags & NLM_F_EXCL) {
 713                         netdev_dbg(vxlan->dev,
 714                                    "lost race to create %pM\n", mac);
 715                         return -EEXIST;
 716                 }
 717                 if (f->state != state) {
 718                         f->state = state;
 719                         f->updated = jiffies;
 720                         notify = 1;
 721                 }
 722                 if (f->flags != ndm_flags) {
 723                         f->flags = ndm_flags;
 724                         f->updated = jiffies;
 725                         notify = 1;
 726                 }
 727                 if ((flags & NLM_F_REPLACE)) {
 728                         /* Only change unicasts */
 729                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 730                              is_zero_ether_addr(f->eth_addr))) {
 731                                 int rc = vxlan_fdb_replace(f, ip, port, vni,
 732                                                            ifindex);
 733
 734                                 if (rc < 0)
 735                                         return rc;
 736                                 notify |= rc;
 737                         } else
 738                                 return -EOPNOTSUPP;
 739                 }
 740                 if ((flags & NLM_F_APPEND) &&
 741                     (is_multicast_ether_addr(f->eth_addr) ||
 742                      is_zero_ether_addr(f->eth_addr))) {
 743                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 744                                                   &rd);
 745
 746                         if (rc < 0)
 747                                 return rc;
 748                         notify |= rc;
 749                 }
 750         } else {
 751                 if (!(flags & NLM_F_CREATE))
 752                         return -ENOENT;
 753
 754                 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
 755                         return -ENOSPC;
 756
 757                 /* Disallow replace to add a multicast entry */
 758                 if ((flags & NLM_F_REPLACE) &&
 759                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 760                         return -EOPNOTSUPP;
 761
 762                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 763                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 764                 if (!f)
 765                         return -ENOMEM;
 766
 767                 notify = 1;
 768                 f->state = state;
 769                 f->flags = ndm_flags;
 770                 f->updated = f->used = jiffies;
 771                 INIT_LIST_HEAD(&f->remotes);
 772                 memcpy(f->eth_addr, mac, ETH_ALEN);
 773
 774                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 775
 776                 ++vxlan->addrcnt;
 777                 hlist_add_head_rcu(&f->hlist,
 778                                    vxlan_fdb_head(vxlan, mac));
 779         }
 780
 781         if (notify) {
 782                 if (rd == NULL)
 783                         rd = first_remote_rtnl(f);
 784                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 785         }
 786
 787         return 0;
 788 }
 789
 790 static void vxlan_fdb_free(struct rcu_head *head)
 791 {
 792         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 793         struct vxlan_rdst *rd, *nd;
 794
 795         list_for_each_entry_safe(rd, nd, &f->remotes, list)
 796                 kfree(rd);
 797         kfree(f);
 798 }
 799
 800 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 801 {
 802         netdev_dbg(vxlan->dev,
 803                     "delete %pM\n", f->eth_addr);
 804
 805         --vxlan->addrcnt;
 806         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 807
 808         hlist_del_rcu(&f->hlist);
 809         call_rcu(&f->rcu, vxlan_fdb_free);
 810 }
 811
 812 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 813                            union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
 814 {
 815         struct net *net = dev_net(vxlan->dev);
 816         int err;
 817
 818         if (tb[NDA_DST]) {
 819                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 820                 if (err)
 821                         return err;
 822         } else {
 823                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 824                 if (remote->sa.sa_family == AF_INET) {
 825                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 826                         ip->sa.sa_family = AF_INET;
 827 #if IS_ENABLED(CONFIG_IPV6)
 828                 } else {
 829                         ip->sin6.sin6_addr = in6addr_any;
 830                         ip->sa.sa_family = AF_INET6;
 831 #endif
 832                 }
 833         }
 834
 835         if (tb[NDA_PORT]) {
 836                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 837                         return -EINVAL;
 838                 *port = nla_get_be16(tb[NDA_PORT]);
 839         } else {
 840                 *port = vxlan->dst_port;
 841         }
 842
 843         if (tb[NDA_VNI]) {
 844                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 845                         return -EINVAL;
 846                 *vni = nla_get_u32(tb[NDA_VNI]);
 847         } else {
 848                 *vni = vxlan->default_dst.remote_vni;
 849         }
 850
 851         if (tb[NDA_IFINDEX]) {
 852                 struct net_device *tdev;
 853
 854                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 855                         return -EINVAL;
 856                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 857                 tdev = __dev_get_by_index(net, *ifindex);
 858                 if (!tdev)
 859                         return -EADDRNOTAVAIL;
 860         } else {
 861                 *ifindex = 0;
 862         }
 863
 864         return 0;
 865 }
 866
 867 /* Add static entry (via netlink) */
 868 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 869                          struct net_device *dev,
 870                          const unsigned char *addr, u16 vid, u16 flags)
 871 {
 872         struct vxlan_dev *vxlan = netdev_priv(dev);
 873         /* struct net *net = dev_net(vxlan->dev); */
 874         union vxlan_addr ip;
 875         __be16 port;
 876         u32 vni, ifindex;
 877         int err;
 878
 879         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 880                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 881                         ndm->ndm_state);
 882                 return -EINVAL;
 883         }
 884
 885         if (tb[NDA_DST] == NULL)
 886                 return -EINVAL;
 887
 888         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 889         if (err)
 890                 return err;
 891
 892         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 893                 return -EAFNOSUPPORT;
 894
 895         spin_lock_bh(&vxlan->hash_lock);
 896         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 897                                port, vni, ifindex, ndm->ndm_flags);
 898         spin_unlock_bh(&vxlan->hash_lock);
 899
 900         return err;
 901 }
 902
 903 /* Delete entry (via netlink) */
 904 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 905                             struct net_device *dev,
 906                             const unsigned char *addr, u16 vid)
 907 {
 908         struct vxlan_dev *vxlan = netdev_priv(dev);
 909         struct vxlan_fdb *f;
 910         struct vxlan_rdst *rd = NULL;
 911         union vxlan_addr ip;
 912         __be16 port;
 913         u32 vni, ifindex;
 914         int err;
 915
 916         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 917         if (err)
 918                 return err;
 919
 920         err = -ENOENT;
 921
 922         spin_lock_bh(&vxlan->hash_lock);
 923         f = vxlan_find_mac(vxlan, addr);
 924         if (!f)
 925                 goto out;
 926
 927         if (!vxlan_addr_any(&ip)) {
 928                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 929                 if (!rd)
 930                         goto out;
 931         }
 932
 933         err = 0;
 934
 935         /* remove a destination if it's not the only one on the list,
 936          * otherwise destroy the fdb entry
 937          */
 938         if (rd && !list_is_singular(&f->remotes)) {
 939                 list_del_rcu(&rd->list);
 940                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 941                 kfree_rcu(rd, rcu);
 942                 goto out;
 943         }
 944
 945         vxlan_fdb_destroy(vxlan, f);
 946
 947 out:
 948         spin_unlock_bh(&vxlan->hash_lock);
 949
 950         return err;
 951 }
 952
 953 /* Dump forwarding table */
 954 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 955                           struct net_device *dev,
 956                           struct net_device *filter_dev, int idx)
 957 {
 958         struct vxlan_dev *vxlan = netdev_priv(dev);
 959         unsigned int h;
 960
 961         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 962                 struct vxlan_fdb *f;
 963                 int err;
 964
 965                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 966                         struct vxlan_rdst *rd;
 967
 968                         if (idx < cb->args[0])
 969                                 goto skip;
 970
 971                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 972                                 err = vxlan_fdb_info(skb, vxlan, f,
 973                                                      NETLINK_CB(cb->skb).portid,
 974                                                      cb->nlh->nlmsg_seq,
 975                                                      RTM_NEWNEIGH,
 976                                                      NLM_F_MULTI, rd);
 977                                 if (err < 0)
 978                                         goto out;
 979                         }
 980 skip:
 981                         ++idx;
 982                 }
 983         }
 984 out:
 985         return idx;
 986 }
 987
 988 /* Watch incoming packets to learn mapping between Ethernet address
 989  * and Tunnel endpoint.
 990  * Return true if packet is bogus and should be droppped.
 991  */
 992 static bool vxlan_snoop(struct net_device *dev,
 993                         union vxlan_addr *src_ip, const u8 *src_mac)
 994 {
 995         struct vxlan_dev *vxlan = netdev_priv(dev);
 996         struct vxlan_fdb *f;
 997
 998         f = vxlan_find_mac(vxlan, src_mac);
 999         if (likely(f)) {
1000                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1001
1002                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
1003                         return false;
1004
1005                 /* Don't migrate static entries, drop packets */
1006                 if (f->state & NUD_NOARP)
1007                         return true;
1008
1009                 if (net_ratelimit())
1010                         netdev_info(dev,
1011                                     "%pM migrated from %pIS to %pIS\n",
1012                                     src_mac, &rdst->remote_ip, &src_ip);
1013
1014                 rdst->remote_ip = *src_ip;
1015                 f->updated = jiffies;
1016                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1017         } else {
1018                 /* learned new entry */
1019                 spin_lock(&vxlan->hash_lock);
1020
1021                 /* close off race between vxlan_flush and incoming packets */
1022                 if (netif_running(dev))
1023                         vxlan_fdb_create(vxlan, src_mac, src_ip,
1024                                          NUD_REACHABLE,
1025                                          NLM_F_EXCL|NLM_F_CREATE,
1026                                          vxlan->dst_port,
1027                                          vxlan->default_dst.remote_vni,
1028                                          0, NTF_SELF);
1029                 spin_unlock(&vxlan->hash_lock);
1030         }
1031
1032         return false;
1033 }
1034
1035 /* See if multicast group is already in use by other ID */
1036 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1037 {
1038         struct vxlan_dev *vxlan;
1039
1040         /* The vxlan_sock is only used by dev, leaving group has
1041          * no effect on other vxlan devices.
1042          */
1043         if (atomic_read(&dev->vn_sock->refcnt) == 1)
1044                 return false;
1045
1046         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1047                 if (!netif_running(vxlan->dev) || vxlan == dev)
1048                         continue;
1049
1050                 if (vxlan->vn_sock != dev->vn_sock)
1051                         continue;
1052
1053                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1054                                       &dev->default_dst.remote_ip))
1055                         continue;
1056
1057                 if (vxlan->default_dst.remote_ifindex !=
1058                     dev->default_dst.remote_ifindex)
1059                         continue;
1060
1061                 return true;
1062         }
1063
1064         return false;
1065 }
1066
1067 static void vxlan_sock_hold(struct vxlan_sock *vs)
1068 {
1069         atomic_inc(&vs->refcnt);
1070 }
1071
1072 void vxlan_sock_release(struct vxlan_sock *vs)
1073 {
1074         struct sock *sk = vs->sock->sk;
1075         struct net *net = sock_net(sk);
1076         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1077
1078         if (!atomic_dec_and_test(&vs->refcnt))
1079                 return;
1080
1081         spin_lock(&vn->sock_lock);
1082         hlist_del_rcu(&vs->hlist);
1083         vxlan_notify_del_rx_port(vs);
1084         spin_unlock(&vn->sock_lock);
1085
1086         queue_work(vxlan_wq, &vs->del_work);
1087 }
1088 EXPORT_SYMBOL_GPL(vxlan_sock_release);
1089
1090 /* Callback to update multicast group membership when first VNI on
1091  * multicast asddress is brought up
1092  * Done as workqueue because ip_mc_join_group acquires RTNL.
1093  */
1094 static void vxlan_igmp_join(struct work_struct *work)
1095 {
1096         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_join);
1097         struct vxlan_sock *vs = vxlan->vn_sock;
1098         struct sock *sk = vs->sock->sk;
1099         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1100         int ifindex = vxlan->default_dst.remote_ifindex;
1101
1102         lock_sock(sk);
1103         if (ip->sa.sa_family == AF_INET) {
1104                 struct ip_mreqn mreq = {
1105                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1106                         .imr_ifindex            = ifindex,
1107                 };
1108
1109                 ip_mc_join_group(sk, &mreq);
1110 #if IS_ENABLED(CONFIG_IPV6)
1111         } else {
1112                 ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1113                                              &ip->sin6.sin6_addr);
1114 #endif
1115         }
1116         release_sock(sk);
1117
1118         vxlan_sock_release(vs);
1119         dev_put(vxlan->dev);
1120 }
1121
1122 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1123 static void vxlan_igmp_leave(struct work_struct *work)
1124 {
1125         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_leave);
1126         struct vxlan_sock *vs = vxlan->vn_sock;
1127         struct sock *sk = vs->sock->sk;
1128         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1129         int ifindex = vxlan->default_dst.remote_ifindex;
1130
1131         lock_sock(sk);
1132         if (ip->sa.sa_family == AF_INET) {
1133                 struct ip_mreqn mreq = {
1134                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1135                         .imr_ifindex            = ifindex,
1136                 };
1137
1138                 ip_mc_leave_group(sk, &mreq);
1139 #if IS_ENABLED(CONFIG_IPV6)
1140         } else {
1141                 ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1142                                              &ip->sin6.sin6_addr);
1143 #endif
1144         }
1145
1146         release_sock(sk);
1147
1148         vxlan_sock_release(vs);
1149         dev_put(vxlan->dev);
1150 }
1151
1152 static struct vxlanhdr *vxlan_remcsum(struct sk_buff *skb, struct vxlanhdr *vh,
1153                                       size_t hdrlen, u32 data)
1154 {
1155         size_t start, offset, plen;
1156
1157         if (skb->remcsum_offload) {
1158                 /* Already processed in GRO path */
1159                 skb->remcsum_offload = 0;
1160                 return vh;
1161         }
1162
1163         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
1164         offset = start + ((data & VXLAN_RCO_UDP) ?
1165                           offsetof(struct udphdr, check) :
1166                           offsetof(struct tcphdr, check));
1167
1168         plen = hdrlen + offset + sizeof(u16);
1169
1170         if (!pskb_may_pull(skb, plen))
1171                 return NULL;
1172
1173         vh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1174
1175         skb_remcsum_process(skb, (void *)vh + hdrlen, start, offset);
1176
1177         return vh;
1178 }
1179
1180 /* Callback from net/ipv4/udp.c to receive packets */
1181 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1182 {
1183         struct vxlan_sock *vs;
1184         struct vxlanhdr *vxh;
1185         u32 flags, vni;
1186         struct vxlan_metadata md = {0};
1187
1188         /* Need Vxlan and inner Ethernet header to be present */
1189         if (!pskb_may_pull(skb, VXLAN_HLEN))
1190                 goto error;
1191
1192         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1193         flags = ntohl(vxh->vx_flags);
1194         vni = ntohl(vxh->vx_vni);
1195
1196         if (flags & VXLAN_HF_VNI) {
1197                 flags &= ~VXLAN_HF_VNI;
1198         } else {
1199                 /* VNI flag always required to be set */
1200                 goto bad_flags;
1201         }
1202
1203         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1204                 goto drop;
1205         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1206
1207         vs = rcu_dereference_sk_user_data(sk);
1208         if (!vs)
1209                 goto drop;
1210
1211         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1212                 vxh = vxlan_remcsum(skb, vxh, sizeof(struct vxlanhdr), vni);
1213                 if (!vxh)
1214                         goto drop;
1215
1216                 flags &= ~VXLAN_HF_RCO;
1217                 vni &= VXLAN_VID_MASK;
1218         }
1219
1220         /* For backwards compatibility, only allow reserved fields to be
1221          * used by VXLAN extensions if explicitly requested.
1222          */
1223         if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1224                 struct vxlanhdr_gbp *gbp;
1225
1226                 gbp = (struct vxlanhdr_gbp *)vxh;
1227                 md.gbp = ntohs(gbp->policy_id);
1228
1229                 if (gbp->dont_learn)
1230                         md.gbp |= VXLAN_GBP_DONT_LEARN;
1231
1232                 if (gbp->policy_applied)
1233                         md.gbp |= VXLAN_GBP_POLICY_APPLIED;
1234
1235                 flags &= ~VXLAN_GBP_USED_BITS;
1236         }
1237
1238         if (flags || (vni & ~VXLAN_VID_MASK)) {
1239                 /* If there are any unprocessed flags remaining treat
1240                  * this as a malformed packet. This behavior diverges from
1241                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1242                  * in reserved fields are to be ignored. The approach here
1243                  * maintains compatbility with previous stack code, and also
1244                  * is more robust and provides a little more security in
1245                  * adding extensions to VXLAN.
1246                  */
1247
1248                 goto bad_flags;
1249         }
1250
1251         md.vni = vxh->vx_vni;
1252         vs->rcv(vs, skb, &md);
1253         return 0;
1254
1255 drop:
1256         /* Consume bad packet */
1257         kfree_skb(skb);
1258         return 0;
1259
1260 bad_flags:
1261         netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1262                    ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1263
1264 error:
1265         /* Return non vxlan pkt */
1266         return 1;
1267 }
1268
1269 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1270                       struct vxlan_metadata *md)
1271 {
1272         struct iphdr *oip = NULL;
1273         struct ipv6hdr *oip6 = NULL;
1274         struct vxlan_dev *vxlan;
1275         struct pcpu_sw_netstats *stats;
1276         union vxlan_addr saddr;
1277         __u32 vni;
1278         int err = 0;
1279         union vxlan_addr *remote_ip;
1280
1281         vni = ntohl(md->vni) >> 8;
1282         /* Is this VNI defined? */
1283         vxlan = vxlan_vs_find_vni(vs, vni);
1284         if (!vxlan)
1285                 goto drop;
1286
1287         remote_ip = &vxlan->default_dst.remote_ip;
1288         skb_reset_mac_header(skb);
1289         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1290         skb->protocol = eth_type_trans(skb, vxlan->dev);
1291         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1292
1293         /* Ignore packet loops (and multicast echo) */
1294         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1295                 goto drop;
1296
1297         /* Re-examine inner Ethernet packet */
1298         if (remote_ip->sa.sa_family == AF_INET) {
1299                 oip = ip_hdr(skb);
1300                 saddr.sin.sin_addr.s_addr = oip->saddr;
1301                 saddr.sa.sa_family = AF_INET;
1302 #if IS_ENABLED(CONFIG_IPV6)
1303         } else {
1304                 oip6 = ipv6_hdr(skb);
1305                 saddr.sin6.sin6_addr = oip6->saddr;
1306                 saddr.sa.sa_family = AF_INET6;
1307 #endif
1308         }
1309
1310         if ((vxlan->flags & VXLAN_F_LEARN) &&
1311             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1312                 goto drop;
1313
1314         skb_reset_network_header(skb);
1315         skb->mark = md->gbp;
1316
1317         if (oip6)
1318                 err = IP6_ECN_decapsulate(oip6, skb);
1319         if (oip)
1320                 err = IP_ECN_decapsulate(oip, skb);
1321
1322         if (unlikely(err)) {
1323                 if (log_ecn_error) {
1324                         if (oip6)
1325                                 net_info_ratelimited("non-ECT from %pI6\n",
1326                                                      &oip6->saddr);
1327                         if (oip)
1328                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1329                                                      &oip->saddr, oip->tos);
1330                 }
1331                 if (err > 1) {
1332                         ++vxlan->dev->stats.rx_frame_errors;
1333                         ++vxlan->dev->stats.rx_errors;
1334                         goto drop;
1335                 }
1336         }
1337
1338         stats = this_cpu_ptr(vxlan->dev->tstats);
1339         u64_stats_update_begin(&stats->syncp);
1340         stats->rx_packets++;
1341         stats->rx_bytes += skb->len;
1342         u64_stats_update_end(&stats->syncp);
1343
1344         netif_rx(skb);
1345
1346         return;
1347 drop:
1348         /* Consume bad packet */
1349         kfree_skb(skb);
1350 }
1351
1352 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1353 {
1354         struct vxlan_dev *vxlan = netdev_priv(dev);
1355         struct arphdr *parp;
1356         u8 *arpptr, *sha;
1357         __be32 sip, tip;
1358         struct neighbour *n;
1359
1360         if (dev->flags & IFF_NOARP)
1361                 goto out;
1362
1363         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1364                 dev->stats.tx_dropped++;
1365                 goto out;
1366         }
1367         parp = arp_hdr(skb);
1368
1369         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1370              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1371             parp->ar_pro != htons(ETH_P_IP) ||
1372             parp->ar_op != htons(ARPOP_REQUEST) ||
1373             parp->ar_hln != dev->addr_len ||
1374             parp->ar_pln != 4)
1375                 goto out;
1376         arpptr = (u8 *)parp + sizeof(struct arphdr);
1377         sha = arpptr;
1378         arpptr += dev->addr_len;        /* sha */
1379         memcpy(&sip, arpptr, sizeof(sip));
1380         arpptr += sizeof(sip);
1381         arpptr += dev->addr_len;        /* tha */
1382         memcpy(&tip, arpptr, sizeof(tip));
1383
1384         if (ipv4_is_loopback(tip) ||
1385             ipv4_is_multicast(tip))
1386                 goto out;
1387
1388         n = neigh_lookup(&arp_tbl, &tip, dev);
1389
1390         if (n) {
1391                 struct vxlan_fdb *f;
1392                 struct sk_buff  *reply;
1393
1394                 if (!(n->nud_state & NUD_CONNECTED)) {
1395                         neigh_release(n);
1396                         goto out;
1397                 }
1398
1399                 f = vxlan_find_mac(vxlan, n->ha);
1400                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1401                         /* bridge-local neighbor */
1402                         neigh_release(n);
1403                         goto out;
1404                 }
1405
1406                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1407                                 n->ha, sha);
1408
1409                 neigh_release(n);
1410
1411                 if (reply == NULL)
1412                         goto out;
1413
1414                 skb_reset_mac_header(reply);
1415                 __skb_pull(reply, skb_network_offset(reply));
1416                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1417                 reply->pkt_type = PACKET_HOST;
1418
1419                 if (netif_rx_ni(reply) == NET_RX_DROP)
1420                         dev->stats.rx_dropped++;
1421         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1422                 union vxlan_addr ipa = {
1423                         .sin.sin_addr.s_addr = tip,
1424                         .sin.sin_family = AF_INET,
1425                 };
1426
1427                 vxlan_ip_miss(dev, &ipa);
1428         }
1429 out:
1430         consume_skb(skb);
1431         return NETDEV_TX_OK;
1432 }
1433
1434 #if IS_ENABLED(CONFIG_IPV6)
1435 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1436         struct neighbour *n, bool isrouter)
1437 {
1438         struct net_device *dev = request->dev;
1439         struct sk_buff *reply;
1440         struct nd_msg *ns, *na;
1441         struct ipv6hdr *pip6;
1442         u8 *daddr;
1443         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1444         int ns_olen;
1445         int i, len;
1446
1447         if (dev == NULL)
1448                 return NULL;
1449
1450         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1451                 sizeof(*na) + na_olen + dev->needed_tailroom;
1452         reply = alloc_skb(len, GFP_ATOMIC);
1453         if (reply == NULL)
1454                 return NULL;
1455
1456         reply->protocol = htons(ETH_P_IPV6);
1457         reply->dev = dev;
1458         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1459         skb_push(reply, sizeof(struct ethhdr));
1460         skb_set_mac_header(reply, 0);
1461
1462         ns = (struct nd_msg *)skb_transport_header(request);
1463
1464         daddr = eth_hdr(request)->h_source;
1465         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1466         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1467                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1468                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1469                         break;
1470                 }
1471         }
1472
1473         /* Ethernet header */
1474         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1475         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1476         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1477         reply->protocol = htons(ETH_P_IPV6);
1478
1479         skb_pull(reply, sizeof(struct ethhdr));
1480         skb_set_network_header(reply, 0);
1481         skb_put(reply, sizeof(struct ipv6hdr));
1482
1483         /* IPv6 header */
1484
1485         pip6 = ipv6_hdr(reply);
1486         memset(pip6, 0, sizeof(struct ipv6hdr));
1487         pip6->version = 6;
1488         pip6->priority = ipv6_hdr(request)->priority;
1489         pip6->nexthdr = IPPROTO_ICMPV6;
1490         pip6->hop_limit = 255;
1491         pip6->daddr = ipv6_hdr(request)->saddr;
1492         pip6->saddr = *(struct in6_addr *)n->primary_key;
1493
1494         skb_pull(reply, sizeof(struct ipv6hdr));
1495         skb_set_transport_header(reply, 0);
1496
1497         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1498
1499         /* Neighbor Advertisement */
1500         memset(na, 0, sizeof(*na)+na_olen);
1501         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1502         na->icmph.icmp6_router = isrouter;
1503         na->icmph.icmp6_override = 1;
1504         na->icmph.icmp6_solicited = 1;
1505         na->target = ns->target;
1506         ether_addr_copy(&na->opt[2], n->ha);
1507         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1508         na->opt[1] = na_olen >> 3;
1509
1510         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1511                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1512                 csum_partial(na, sizeof(*na)+na_olen, 0));
1513
1514         pip6->payload_len = htons(sizeof(*na)+na_olen);
1515
1516         skb_push(reply, sizeof(struct ipv6hdr));
1517
1518         reply->ip_summed = CHECKSUM_UNNECESSARY;
1519
1520         return reply;
1521 }
1522
1523 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1524 {
1525         struct vxlan_dev *vxlan = netdev_priv(dev);
1526         struct nd_msg *msg;
1527         const struct ipv6hdr *iphdr;
1528         const struct in6_addr *saddr, *daddr;
1529         struct neighbour *n;
1530         struct inet6_dev *in6_dev;
1531
1532         in6_dev = __in6_dev_get(dev);
1533         if (!in6_dev)
1534                 goto out;
1535
1536         iphdr = ipv6_hdr(skb);
1537         saddr = &iphdr->saddr;
1538         daddr = &iphdr->daddr;
1539
1540         msg = (struct nd_msg *)skb_transport_header(skb);
1541         if (msg->icmph.icmp6_code != 0 ||
1542             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1543                 goto out;
1544
1545         if (ipv6_addr_loopback(daddr) ||
1546             ipv6_addr_is_multicast(&msg->target))
1547                 goto out;
1548
1549         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1550
1551         if (n) {
1552                 struct vxlan_fdb *f;
1553                 struct sk_buff *reply;
1554
1555                 if (!(n->nud_state & NUD_CONNECTED)) {
1556                         neigh_release(n);
1557                         goto out;
1558                 }
1559
1560                 f = vxlan_find_mac(vxlan, n->ha);
1561                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1562                         /* bridge-local neighbor */
1563                         neigh_release(n);
1564                         goto out;
1565                 }
1566
1567                 reply = vxlan_na_create(skb, n,
1568                                         !!(f ? f->flags & NTF_ROUTER : 0));
1569
1570                 neigh_release(n);
1571
1572                 if (reply == NULL)
1573                         goto out;
1574
1575                 if (netif_rx_ni(reply) == NET_RX_DROP)
1576                         dev->stats.rx_dropped++;
1577
1578         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1579                 union vxlan_addr ipa = {
1580                         .sin6.sin6_addr = msg->target,
1581                         .sin6.sin6_family = AF_INET6,
1582                 };
1583
1584                 vxlan_ip_miss(dev, &ipa);
1585         }
1586
1587 out:
1588         consume_skb(skb);
1589         return NETDEV_TX_OK;
1590 }
1591 #endif
1592
1593 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1594 {
1595         struct vxlan_dev *vxlan = netdev_priv(dev);
1596         struct neighbour *n;
1597
1598         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1599                 return false;
1600
1601         n = NULL;
1602         switch (ntohs(eth_hdr(skb)->h_proto)) {
1603         case ETH_P_IP:
1604         {
1605                 struct iphdr *pip;
1606
1607                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1608                         return false;
1609                 pip = ip_hdr(skb);
1610                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1611                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1612                         union vxlan_addr ipa = {
1613                                 .sin.sin_addr.s_addr = pip->daddr,
1614                                 .sin.sin_family = AF_INET,
1615                         };
1616
1617                         vxlan_ip_miss(dev, &ipa);
1618                         return false;
1619                 }
1620
1621                 break;
1622         }
1623 #if IS_ENABLED(CONFIG_IPV6)
1624         case ETH_P_IPV6:
1625         {
1626                 struct ipv6hdr *pip6;
1627
1628                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1629                         return false;
1630                 pip6 = ipv6_hdr(skb);
1631                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1632                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1633                         union vxlan_addr ipa = {
1634                                 .sin6.sin6_addr = pip6->daddr,
1635                                 .sin6.sin6_family = AF_INET6,
1636                         };
1637
1638                         vxlan_ip_miss(dev, &ipa);
1639                         return false;
1640                 }
1641
1642                 break;
1643         }
1644 #endif
1645         default:
1646                 return false;
1647         }
1648
1649         if (n) {
1650                 bool diff;
1651
1652                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1653                 if (diff) {
1654                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1655                                 dev->addr_len);
1656                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1657                 }
1658                 neigh_release(n);
1659                 return diff;
1660         }
1661
1662         return false;
1663 }
1664
1665 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1666                                 struct vxlan_metadata *md)
1667 {
1668         struct vxlanhdr_gbp *gbp;
1669
1670         gbp = (struct vxlanhdr_gbp *)vxh;
1671         vxh->vx_flags |= htonl(VXLAN_HF_GBP);
1672
1673         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1674                 gbp->dont_learn = 1;
1675
1676         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1677                 gbp->policy_applied = 1;
1678
1679         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1680 }
1681
1682 #if IS_ENABLED(CONFIG_IPV6)
1683 static int vxlan6_xmit_skb(struct dst_entry *dst, struct sk_buff *skb,
1684                            struct net_device *dev, struct in6_addr *saddr,
1685                            struct in6_addr *daddr, __u8 prio, __u8 ttl,
1686                            __be16 src_port, __be16 dst_port,
1687                            struct vxlan_metadata *md, bool xnet, u32 vxflags)
1688 {
1689         struct vxlanhdr *vxh;
1690         int min_headroom;
1691         int err;
1692         bool udp_sum = !(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX);
1693         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1694         u16 hdrlen = sizeof(struct vxlanhdr);
1695
1696         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1697             skb->ip_summed == CHECKSUM_PARTIAL) {
1698                 int csum_start = skb_checksum_start_offset(skb);
1699
1700                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1701                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1702                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1703                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1704                         udp_sum = false;
1705                         type |= SKB_GSO_TUNNEL_REMCSUM;
1706                 }
1707         }
1708
1709         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1710         if (IS_ERR(skb)) {
1711                 err = -EINVAL;
1712                 goto err;
1713         }
1714
1715         skb_scrub_packet(skb, xnet);
1716
1717         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1718                         + VXLAN_HLEN + sizeof(struct ipv6hdr)
1719                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1720
1721         /* Need space for new headers (invalidates iph ptr) */
1722         err = skb_cow_head(skb, min_headroom);
1723         if (unlikely(err)) {
1724                 kfree_skb(skb);
1725                 goto err;
1726         }
1727
1728         skb = vlan_hwaccel_push_inside(skb);
1729         if (WARN_ON(!skb)) {
1730                 err = -ENOMEM;
1731                 goto err;
1732         }
1733
1734         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1735         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1736         vxh->vx_vni = md->vni;
1737
1738         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1739                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1740                            VXLAN_RCO_SHIFT;
1741
1742                 if (skb->csum_offset == offsetof(struct udphdr, check))
1743                         data |= VXLAN_RCO_UDP;
1744
1745                 vxh->vx_vni |= htonl(data);
1746                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1747
1748                 if (!skb_is_gso(skb)) {
1749                         skb->ip_summed = CHECKSUM_NONE;
1750                         skb->encapsulation = 0;
1751                 }
1752         }
1753
1754         if (vxflags & VXLAN_F_GBP)
1755                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1756
1757         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1758
1759         udp_tunnel6_xmit_skb(dst, skb, dev, saddr, daddr, prio,
1760                              ttl, src_port, dst_port,
1761                              !!(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX));
1762         return 0;
1763 err:
1764         dst_release(dst);
1765         return err;
1766 }
1767 #endif
1768
1769 int vxlan_xmit_skb(struct rtable *rt, struct sk_buff *skb,
1770                    __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1771                    __be16 src_port, __be16 dst_port,
1772                    struct vxlan_metadata *md, bool xnet, u32 vxflags)
1773 {
1774         struct vxlanhdr *vxh;
1775         int min_headroom;
1776         int err;
1777         bool udp_sum = !!(vxflags & VXLAN_F_UDP_CSUM);
1778         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1779         u16 hdrlen = sizeof(struct vxlanhdr);
1780
1781         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1782             skb->ip_summed == CHECKSUM_PARTIAL) {
1783                 int csum_start = skb_checksum_start_offset(skb);
1784
1785                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1786                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1787                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1788                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1789                         udp_sum = false;
1790                         type |= SKB_GSO_TUNNEL_REMCSUM;
1791                 }
1792         }
1793
1794         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1795         if (IS_ERR(skb))
1796                 return PTR_ERR(skb);
1797
1798         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1799                         + VXLAN_HLEN + sizeof(struct iphdr)
1800                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1801
1802         /* Need space for new headers (invalidates iph ptr) */
1803         err = skb_cow_head(skb, min_headroom);
1804         if (unlikely(err)) {
1805                 kfree_skb(skb);
1806                 return err;
1807         }
1808
1809         skb = vlan_hwaccel_push_inside(skb);
1810         if (WARN_ON(!skb))
1811                 return -ENOMEM;
1812
1813         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1814         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1815         vxh->vx_vni = md->vni;
1816
1817         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1818                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1819                            VXLAN_RCO_SHIFT;
1820
1821                 if (skb->csum_offset == offsetof(struct udphdr, check))
1822                         data |= VXLAN_RCO_UDP;
1823
1824                 vxh->vx_vni |= htonl(data);
1825                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1826
1827                 if (!skb_is_gso(skb)) {
1828                         skb->ip_summed = CHECKSUM_NONE;
1829                         skb->encapsulation = 0;
1830                 }
1831         }
1832
1833         if (vxflags & VXLAN_F_GBP)
1834                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1835
1836         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1837
1838         return udp_tunnel_xmit_skb(rt, skb, src, dst, tos,
1839                                    ttl, df, src_port, dst_port, xnet,
1840                                    !(vxflags & VXLAN_F_UDP_CSUM));
1841 }
1842 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1843
1844 /* Bypass encapsulation if the destination is local */
1845 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1846                                struct vxlan_dev *dst_vxlan)
1847 {
1848         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1849         union vxlan_addr loopback;
1850         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1851         struct net_device *dev = skb->dev;
1852         int len = skb->len;
1853
1854         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1855         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1856         skb->pkt_type = PACKET_HOST;
1857         skb->encapsulation = 0;
1858         skb->dev = dst_vxlan->dev;
1859         __skb_pull(skb, skb_network_offset(skb));
1860
1861         if (remote_ip->sa.sa_family == AF_INET) {
1862                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1863                 loopback.sa.sa_family =  AF_INET;
1864 #if IS_ENABLED(CONFIG_IPV6)
1865         } else {
1866                 loopback.sin6.sin6_addr = in6addr_loopback;
1867                 loopback.sa.sa_family =  AF_INET6;
1868 #endif
1869         }
1870
1871         if (dst_vxlan->flags & VXLAN_F_LEARN)
1872                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1873
1874         u64_stats_update_begin(&tx_stats->syncp);
1875         tx_stats->tx_packets++;
1876         tx_stats->tx_bytes += len;
1877         u64_stats_update_end(&tx_stats->syncp);
1878
1879         if (netif_rx(skb) == NET_RX_SUCCESS) {
1880                 u64_stats_update_begin(&rx_stats->syncp);
1881                 rx_stats->rx_packets++;
1882                 rx_stats->rx_bytes += len;
1883                 u64_stats_update_end(&rx_stats->syncp);
1884         } else {
1885                 dev->stats.rx_dropped++;
1886         }
1887 }
1888
1889 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1890                            struct vxlan_rdst *rdst, bool did_rsc)
1891 {
1892         struct vxlan_dev *vxlan = netdev_priv(dev);
1893         struct rtable *rt = NULL;
1894         const struct iphdr *old_iph;
1895         struct flowi4 fl4;
1896         union vxlan_addr *dst;
1897         struct vxlan_metadata md;
1898         __be16 src_port = 0, dst_port;
1899         u32 vni;
1900         __be16 df = 0;
1901         __u8 tos, ttl;
1902         int err;
1903
1904         dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1905         vni = rdst->remote_vni;
1906         dst = &rdst->remote_ip;
1907
1908         if (vxlan_addr_any(dst)) {
1909                 if (did_rsc) {
1910                         /* short-circuited back to local bridge */
1911                         vxlan_encap_bypass(skb, vxlan, vxlan);
1912                         return;
1913                 }
1914                 goto drop;
1915         }
1916
1917         old_iph = ip_hdr(skb);
1918
1919         ttl = vxlan->ttl;
1920         if (!ttl && vxlan_addr_multicast(dst))
1921                 ttl = 1;
1922
1923         tos = vxlan->tos;
1924         if (tos == 1)
1925                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1926
1927         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->port_min,
1928                                      vxlan->port_max, true);
1929
1930         if (dst->sa.sa_family == AF_INET) {
1931                 memset(&fl4, 0, sizeof(fl4));
1932                 fl4.flowi4_oif = rdst->remote_ifindex;
1933                 fl4.flowi4_tos = RT_TOS(tos);
1934                 fl4.daddr = dst->sin.sin_addr.s_addr;
1935                 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1936
1937                 rt = ip_route_output_key(vxlan->net, &fl4);
1938                 if (IS_ERR(rt)) {
1939                         netdev_dbg(dev, "no route to %pI4\n",
1940                                    &dst->sin.sin_addr.s_addr);
1941                         dev->stats.tx_carrier_errors++;
1942                         goto tx_error;
1943                 }
1944
1945                 if (rt->dst.dev == dev) {
1946                         netdev_dbg(dev, "circular route to %pI4\n",
1947                                    &dst->sin.sin_addr.s_addr);
1948                         dev->stats.collisions++;
1949                         goto rt_tx_error;
1950                 }
1951
1952                 /* Bypass encapsulation if the destination is local */
1953                 if (rt->rt_flags & RTCF_LOCAL &&
1954                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1955                         struct vxlan_dev *dst_vxlan;
1956
1957                         ip_rt_put(rt);
1958                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1959                                                    dst->sa.sa_family, dst_port,
1960                                                    vxlan->flags);
1961                         if (!dst_vxlan)
1962                                 goto tx_error;
1963                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1964                         return;
1965                 }
1966
1967                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1968                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1969                 md.vni = htonl(vni << 8);
1970                 md.gbp = skb->mark;
1971
1972                 err = vxlan_xmit_skb(rt, skb, fl4.saddr,
1973                                      dst->sin.sin_addr.s_addr, tos, ttl, df,
1974                                      src_port, dst_port, &md,
1975                                      !net_eq(vxlan->net, dev_net(vxlan->dev)),
1976                                      vxlan->flags);
1977                 if (err < 0) {
1978                         /* skb is already freed. */
1979                         skb = NULL;
1980                         goto rt_tx_error;
1981                 }
1982
1983                 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1984 #if IS_ENABLED(CONFIG_IPV6)
1985         } else {
1986                 struct sock *sk = vxlan->vn_sock->sock->sk;
1987                 struct dst_entry *ndst;
1988                 struct flowi6 fl6;
1989                 u32 flags;
1990
1991                 memset(&fl6, 0, sizeof(fl6));
1992                 fl6.flowi6_oif = rdst->remote_ifindex;
1993                 fl6.daddr = dst->sin6.sin6_addr;
1994                 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
1995                 fl6.flowi6_proto = IPPROTO_UDP;
1996
1997                 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
1998                         netdev_dbg(dev, "no route to %pI6\n",
1999                                    &dst->sin6.sin6_addr);
2000                         dev->stats.tx_carrier_errors++;
2001                         goto tx_error;
2002                 }
2003
2004                 if (ndst->dev == dev) {
2005                         netdev_dbg(dev, "circular route to %pI6\n",
2006                                    &dst->sin6.sin6_addr);
2007                         dst_release(ndst);
2008                         dev->stats.collisions++;
2009                         goto tx_error;
2010                 }
2011
2012                 /* Bypass encapsulation if the destination is local */
2013                 flags = ((struct rt6_info *)ndst)->rt6i_flags;
2014                 if (flags & RTF_LOCAL &&
2015                     !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2016                         struct vxlan_dev *dst_vxlan;
2017
2018                         dst_release(ndst);
2019                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2020                                                    dst->sa.sa_family, dst_port,
2021                                                    vxlan->flags);
2022                         if (!dst_vxlan)
2023                                 goto tx_error;
2024                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2025                         return;
2026                 }
2027
2028                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2029                 md.vni = htonl(vni << 8);
2030                 md.gbp = skb->mark;
2031
2032                 err = vxlan6_xmit_skb(ndst, skb, dev, &fl6.saddr, &fl6.daddr,
2033                                       0, ttl, src_port, dst_port, &md,
2034                                       !net_eq(vxlan->net, dev_net(vxlan->dev)),
2035                                       vxlan->flags);
2036 #endif
2037         }
2038
2039         return;
2040
2041 drop:
2042         dev->stats.tx_dropped++;
2043         goto tx_free;
2044
2045 rt_tx_error:
2046         ip_rt_put(rt);
2047 tx_error:
2048         dev->stats.tx_errors++;
2049 tx_free:
2050         dev_kfree_skb(skb);
2051 }
2052
2053 /* Transmit local packets over Vxlan
2054  *
2055  * Outer IP header inherits ECN and DF from inner header.
2056  * Outer UDP destination is the VXLAN assigned port.
2057  *           source port is based on hash of flow
2058  */
2059 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2060 {
2061         struct vxlan_dev *vxlan = netdev_priv(dev);
2062         struct ethhdr *eth;
2063         bool did_rsc = false;
2064         struct vxlan_rdst *rdst, *fdst = NULL;
2065         struct vxlan_fdb *f;
2066
2067         skb_reset_mac_header(skb);
2068         eth = eth_hdr(skb);
2069
2070         if ((vxlan->flags & VXLAN_F_PROXY)) {
2071                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2072                         return arp_reduce(dev, skb);
2073 #if IS_ENABLED(CONFIG_IPV6)
2074                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2075                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2076                                        + sizeof(struct nd_msg)) &&
2077                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2078                                 struct nd_msg *msg;
2079
2080                                 msg = (struct nd_msg *)skb_transport_header(skb);
2081                                 if (msg->icmph.icmp6_code == 0 &&
2082                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2083                                         return neigh_reduce(dev, skb);
2084                 }
2085                 eth = eth_hdr(skb);
2086 #endif
2087         }
2088
2089         f = vxlan_find_mac(vxlan, eth->h_dest);
2090         did_rsc = false;
2091
2092         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2093             (ntohs(eth->h_proto) == ETH_P_IP ||
2094              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2095                 did_rsc = route_shortcircuit(dev, skb);
2096                 if (did_rsc)
2097                         f = vxlan_find_mac(vxlan, eth->h_dest);
2098         }
2099
2100         if (f == NULL) {
2101                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2102                 if (f == NULL) {
2103                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2104                             !is_multicast_ether_addr(eth->h_dest))
2105                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2106
2107                         dev->stats.tx_dropped++;
2108                         kfree_skb(skb);
2109                         return NETDEV_TX_OK;
2110                 }
2111         }
2112
2113         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2114                 struct sk_buff *skb1;
2115
2116                 if (!fdst) {
2117                         fdst = rdst;
2118                         continue;
2119                 }
2120                 skb1 = skb_clone(skb, GFP_ATOMIC);
2121                 if (skb1)
2122                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2123         }
2124
2125         if (fdst)
2126                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2127         else
2128                 kfree_skb(skb);
2129         return NETDEV_TX_OK;
2130 }
2131
2132 /* Walk the forwarding table and purge stale entries */
2133 static void vxlan_cleanup(unsigned long arg)
2134 {
2135         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2136         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2137         unsigned int h;
2138
2139         if (!netif_running(vxlan->dev))
2140                 return;
2141
2142         spin_lock_bh(&vxlan->hash_lock);
2143         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2144                 struct hlist_node *p, *n;
2145                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2146                         struct vxlan_fdb *f
2147                                 = container_of(p, struct vxlan_fdb, hlist);
2148                         unsigned long timeout;
2149
2150                         if (f->state & NUD_PERMANENT)
2151                                 continue;
2152
2153                         timeout = f->used + vxlan->age_interval * HZ;
2154                         if (time_before_eq(timeout, jiffies)) {
2155                                 netdev_dbg(vxlan->dev,
2156                                            "garbage collect %pM\n",
2157                                            f->eth_addr);
2158                                 f->state = NUD_STALE;
2159                                 vxlan_fdb_destroy(vxlan, f);
2160                         } else if (time_before(timeout, next_timer))
2161                                 next_timer = timeout;
2162                 }
2163         }
2164         spin_unlock_bh(&vxlan->hash_lock);
2165
2166         mod_timer(&vxlan->age_timer, next_timer);
2167 }
2168
2169 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2170 {
2171         __u32 vni = vxlan->default_dst.remote_vni;
2172
2173         vxlan->vn_sock = vs;
2174         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2175 }
2176
2177 /* Setup stats when device is created */
2178 static int vxlan_init(struct net_device *dev)
2179 {
2180         struct vxlan_dev *vxlan = netdev_priv(dev);
2181         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2182         struct vxlan_sock *vs;
2183         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2184
2185         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2186         if (!dev->tstats)
2187                 return -ENOMEM;
2188
2189         spin_lock(&vn->sock_lock);
2190         vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2191                              vxlan->dst_port, vxlan->flags);
2192         if (vs && atomic_add_unless(&vs->refcnt, 1, 0)) {
2193                 /* If we have a socket with same port already, reuse it */
2194                 vxlan_vs_add_dev(vs, vxlan);
2195         } else {
2196                 /* otherwise make new socket outside of RTNL */
2197                 dev_hold(dev);
2198                 queue_work(vxlan_wq, &vxlan->sock_work);
2199         }
2200         spin_unlock(&vn->sock_lock);
2201
2202         return 0;
2203 }
2204
2205 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2206 {
2207         struct vxlan_fdb *f;
2208
2209         spin_lock_bh(&vxlan->hash_lock);
2210         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2211         if (f)
2212                 vxlan_fdb_destroy(vxlan, f);
2213         spin_unlock_bh(&vxlan->hash_lock);
2214 }
2215
2216 static void vxlan_uninit(struct net_device *dev)
2217 {
2218         struct vxlan_dev *vxlan = netdev_priv(dev);
2219         struct vxlan_sock *vs = vxlan->vn_sock;
2220
2221         vxlan_fdb_delete_default(vxlan);
2222
2223         if (vs)
2224                 vxlan_sock_release(vs);
2225         free_percpu(dev->tstats);
2226 }
2227
2228 /* Start ageing timer and join group when device is brought up */
2229 static int vxlan_open(struct net_device *dev)
2230 {
2231         struct vxlan_dev *vxlan = netdev_priv(dev);
2232         struct vxlan_sock *vs = vxlan->vn_sock;
2233
2234         /* socket hasn't been created */
2235         if (!vs)
2236                 return -ENOTCONN;
2237
2238         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2239                 vxlan_sock_hold(vs);
2240                 dev_hold(dev);
2241                 queue_work(vxlan_wq, &vxlan->igmp_join);
2242         }
2243
2244         if (vxlan->age_interval)
2245                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2246
2247         return 0;
2248 }
2249
2250 /* Purge the forwarding table */
2251 static void vxlan_flush(struct vxlan_dev *vxlan)
2252 {
2253         unsigned int h;
2254
2255         spin_lock_bh(&vxlan->hash_lock);
2256         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2257                 struct hlist_node *p, *n;
2258                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2259                         struct vxlan_fdb *f
2260                                 = container_of(p, struct vxlan_fdb, hlist);
2261                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2262                         if (!is_zero_ether_addr(f->eth_addr))
2263                                 vxlan_fdb_destroy(vxlan, f);
2264                 }
2265         }
2266         spin_unlock_bh(&vxlan->hash_lock);
2267 }
2268
2269 /* Cleanup timer and forwarding table on shutdown */
2270 static int vxlan_stop(struct net_device *dev)
2271 {
2272         struct vxlan_dev *vxlan = netdev_priv(dev);
2273         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2274         struct vxlan_sock *vs = vxlan->vn_sock;
2275
2276         if (vs && vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2277             !vxlan_group_used(vn, vxlan)) {
2278                 vxlan_sock_hold(vs);
2279                 dev_hold(dev);
2280                 queue_work(vxlan_wq, &vxlan->igmp_leave);
2281         }
2282
2283         del_timer_sync(&vxlan->age_timer);
2284
2285         vxlan_flush(vxlan);
2286
2287         return 0;
2288 }
2289
2290 /* Stub, nothing needs to be done. */
2291 static void vxlan_set_multicast_list(struct net_device *dev)
2292 {
2293 }
2294
2295 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2296 {
2297         struct vxlan_dev *vxlan = netdev_priv(dev);
2298         struct vxlan_rdst *dst = &vxlan->default_dst;
2299         struct net_device *lowerdev;
2300         int max_mtu;
2301
2302         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2303         if (lowerdev == NULL)
2304                 return eth_change_mtu(dev, new_mtu);
2305
2306         if (dst->remote_ip.sa.sa_family == AF_INET6)
2307                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2308         else
2309                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2310
2311         if (new_mtu < 68 || new_mtu > max_mtu)
2312                 return -EINVAL;
2313
2314         dev->mtu = new_mtu;
2315         return 0;
2316 }
2317
2318 static const struct net_device_ops vxlan_netdev_ops = {
2319         .ndo_init               = vxlan_init,
2320         .ndo_uninit             = vxlan_uninit,
2321         .ndo_open               = vxlan_open,
2322         .ndo_stop               = vxlan_stop,
2323         .ndo_start_xmit         = vxlan_xmit,
2324         .ndo_get_stats64        = ip_tunnel_get_stats64,
2325         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2326         .ndo_change_mtu         = vxlan_change_mtu,
2327         .ndo_validate_addr      = eth_validate_addr,
2328         .ndo_set_mac_address    = eth_mac_addr,
2329         .ndo_fdb_add            = vxlan_fdb_add,
2330         .ndo_fdb_del            = vxlan_fdb_delete,
2331         .ndo_fdb_dump           = vxlan_fdb_dump,
2332 };
2333
2334 /* Info for udev, that this is a virtual tunnel endpoint */
2335 static struct device_type vxlan_type = {
2336         .name = "vxlan",
2337 };
2338
2339 /* Calls the ndo_add_vxlan_port of the caller in order to
2340  * supply the listening VXLAN udp ports. Callers are expected
2341  * to implement the ndo_add_vxlan_port.
2342  */
2343 void vxlan_get_rx_port(struct net_device *dev)
2344 {
2345         struct vxlan_sock *vs;
2346         struct net *net = dev_net(dev);
2347         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2348         sa_family_t sa_family;
2349         __be16 port;
2350         unsigned int i;
2351
2352         spin_lock(&vn->sock_lock);
2353         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2354                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2355                         port = inet_sk(vs->sock->sk)->inet_sport;
2356                         sa_family = vs->sock->sk->sk_family;
2357                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2358                                                             port);
2359                 }
2360         }
2361         spin_unlock(&vn->sock_lock);
2362 }
2363 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2364
2365 /* Initialize the device structure. */
2366 static void vxlan_setup(struct net_device *dev)
2367 {
2368         struct vxlan_dev *vxlan = netdev_priv(dev);
2369         unsigned int h;
2370
2371         eth_hw_addr_random(dev);
2372         ether_setup(dev);
2373         if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2374                 dev->needed_headroom = ETH_HLEN + VXLAN6_HEADROOM;
2375         else
2376                 dev->needed_headroom = ETH_HLEN + VXLAN_HEADROOM;
2377
2378         dev->netdev_ops = &vxlan_netdev_ops;
2379         dev->destructor = free_netdev;
2380         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2381
2382         dev->tx_queue_len = 0;
2383         dev->features   |= NETIF_F_LLTX;
2384         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2385         dev->features   |= NETIF_F_RXCSUM;
2386         dev->features   |= NETIF_F_GSO_SOFTWARE;
2387
2388         dev->vlan_features = dev->features;
2389         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2390         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2391         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2392         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2393         netif_keep_dst(dev);
2394         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2395
2396         INIT_LIST_HEAD(&vxlan->next);
2397         spin_lock_init(&vxlan->hash_lock);
2398         INIT_WORK(&vxlan->igmp_join, vxlan_igmp_join);
2399         INIT_WORK(&vxlan->igmp_leave, vxlan_igmp_leave);
2400         INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
2401
2402         init_timer_deferrable(&vxlan->age_timer);
2403         vxlan->age_timer.function = vxlan_cleanup;
2404         vxlan->age_timer.data = (unsigned long) vxlan;
2405
2406         vxlan->dst_port = htons(vxlan_port);
2407
2408         vxlan->dev = dev;
2409
2410         for (h = 0; h < FDB_HASH_SIZE; ++h)
2411                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2412 }
2413
2414 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2415         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2416         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2417         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2418         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2419         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2420         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2421         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2422         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2423         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2424         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2425         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2426         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2427         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2428         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2429         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2430         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2431         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2432         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2433         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2434         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2435         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2436         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2437         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2438 };
2439
2440 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2441 {
2442         if (tb[IFLA_ADDRESS]) {
2443                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2444                         pr_debug("invalid link address (not ethernet)\n");
2445                         return -EINVAL;
2446                 }
2447
2448                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2449                         pr_debug("invalid all zero ethernet address\n");
2450                         return -EADDRNOTAVAIL;
2451                 }
2452         }
2453
2454         if (!data)
2455                 return -EINVAL;
2456
2457         if (data[IFLA_VXLAN_ID]) {
2458                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2459                 if (id >= VXLAN_VID_MASK)
2460                         return -ERANGE;
2461         }
2462
2463         if (data[IFLA_VXLAN_PORT_RANGE]) {
2464                 const struct ifla_vxlan_port_range *p
2465                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2466
2467                 if (ntohs(p->high) < ntohs(p->low)) {
2468                         pr_debug("port range %u .. %u not valid\n",
2469                                  ntohs(p->low), ntohs(p->high));
2470                         return -EINVAL;
2471                 }
2472         }
2473
2474         return 0;
2475 }
2476
2477 static void vxlan_get_drvinfo(struct net_device *netdev,
2478                               struct ethtool_drvinfo *drvinfo)
2479 {
2480         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2481         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2482 }
2483
2484 static const struct ethtool_ops vxlan_ethtool_ops = {
2485         .get_drvinfo    = vxlan_get_drvinfo,
2486         .get_link       = ethtool_op_get_link,
2487 };
2488
2489 static void vxlan_del_work(struct work_struct *work)
2490 {
2491         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2492         udp_tunnel_sock_release(vs->sock);
2493         kfree_rcu(vs, rcu);
2494 }
2495
2496 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2497                                         __be16 port, u32 flags)
2498 {
2499         struct socket *sock;
2500         struct udp_port_cfg udp_conf;
2501         int err;
2502
2503         memset(&udp_conf, 0, sizeof(udp_conf));
2504
2505         if (ipv6) {
2506                 udp_conf.family = AF_INET6;
2507                 udp_conf.use_udp6_rx_checksums =
2508                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2509         } else {
2510                 udp_conf.family = AF_INET;
2511                 udp_conf.local_ip.s_addr = INADDR_ANY;
2512         }
2513
2514         udp_conf.local_udp_port = port;
2515
2516         /* Open UDP socket */
2517         err = udp_sock_create(net, &udp_conf, &sock);
2518         if (err < 0)
2519                 return ERR_PTR(err);
2520
2521         return sock;
2522 }
2523
2524 /* Create new listen socket if needed */
2525 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2526                                               vxlan_rcv_t *rcv, void *data,
2527                                               u32 flags)
2528 {
2529         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2530         struct vxlan_sock *vs;
2531         struct socket *sock;
2532         unsigned int h;
2533         bool ipv6 = !!(flags & VXLAN_F_IPV6);
2534         struct udp_tunnel_sock_cfg tunnel_cfg;
2535
2536         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2537         if (!vs)
2538                 return ERR_PTR(-ENOMEM);
2539
2540         for (h = 0; h < VNI_HASH_SIZE; ++h)
2541                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2542
2543         INIT_WORK(&vs->del_work, vxlan_del_work);
2544
2545         sock = vxlan_create_sock(net, ipv6, port, flags);
2546         if (IS_ERR(sock)) {
2547                 kfree(vs);
2548                 return ERR_CAST(sock);
2549         }
2550
2551         vs->sock = sock;
2552         atomic_set(&vs->refcnt, 1);
2553         vs->rcv = rcv;
2554         vs->data = data;
2555         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2556
2557         /* Initialize the vxlan udp offloads structure */
2558         vs->udp_offloads.port = port;
2559         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2560         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2561
2562         spin_lock(&vn->sock_lock);
2563         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2564         vxlan_notify_add_rx_port(vs);
2565         spin_unlock(&vn->sock_lock);
2566
2567         /* Mark socket as an encapsulation socket. */
2568         tunnel_cfg.sk_user_data = vs;
2569         tunnel_cfg.encap_type = 1;
2570         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2571         tunnel_cfg.encap_destroy = NULL;
2572
2573         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2574
2575         return vs;
2576 }
2577
2578 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2579                                   vxlan_rcv_t *rcv, void *data,
2580                                   bool no_share, u32 flags)
2581 {
2582         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2583         struct vxlan_sock *vs;
2584         bool ipv6 = flags & VXLAN_F_IPV6;
2585
2586         vs = vxlan_socket_create(net, port, rcv, data, flags);
2587         if (!IS_ERR(vs))
2588                 return vs;
2589
2590         if (no_share)   /* Return error if sharing is not allowed. */
2591                 return vs;
2592
2593         spin_lock(&vn->sock_lock);
2594         vs = vxlan_find_sock(net, ipv6 ? AF_INET6 : AF_INET, port, flags);
2595         if (vs && ((vs->rcv != rcv) ||
2596                    !atomic_add_unless(&vs->refcnt, 1, 0)))
2597                         vs = ERR_PTR(-EBUSY);
2598         spin_unlock(&vn->sock_lock);
2599
2600         if (!vs)
2601                 vs = ERR_PTR(-EINVAL);
2602
2603         return vs;
2604 }
2605 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2606
2607 /* Scheduled at device creation to bind to a socket */
2608 static void vxlan_sock_work(struct work_struct *work)
2609 {
2610         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, sock_work);
2611         struct net *net = vxlan->net;
2612         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2613         __be16 port = vxlan->dst_port;
2614         struct vxlan_sock *nvs;
2615
2616         nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags);
2617         spin_lock(&vn->sock_lock);
2618         if (!IS_ERR(nvs))
2619                 vxlan_vs_add_dev(nvs, vxlan);
2620         spin_unlock(&vn->sock_lock);
2621
2622         dev_put(vxlan->dev);
2623 }
2624
2625 static int vxlan_newlink(struct net *net, struct net_device *dev,
2626                          struct nlattr *tb[], struct nlattr *data[])
2627 {
2628         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2629         struct vxlan_dev *vxlan = netdev_priv(dev);
2630         struct vxlan_rdst *dst = &vxlan->default_dst;
2631         __u32 vni;
2632         int err;
2633         bool use_ipv6 = false;
2634
2635         if (!data[IFLA_VXLAN_ID])
2636                 return -EINVAL;
2637
2638         vxlan->net = dev_net(dev);
2639
2640         vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2641         dst->remote_vni = vni;
2642
2643         /* Unless IPv6 is explicitly requested, assume IPv4 */
2644         dst->remote_ip.sa.sa_family = AF_INET;
2645         if (data[IFLA_VXLAN_GROUP]) {
2646                 dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
2647         } else if (data[IFLA_VXLAN_GROUP6]) {
2648                 if (!IS_ENABLED(CONFIG_IPV6))
2649                         return -EPFNOSUPPORT;
2650
2651                 nla_memcpy(&dst->remote_ip.sin6.sin6_addr, data[IFLA_VXLAN_GROUP6],
2652                            sizeof(struct in6_addr));
2653                 dst->remote_ip.sa.sa_family = AF_INET6;
2654                 use_ipv6 = true;
2655         }
2656
2657         if (data[IFLA_VXLAN_LOCAL]) {
2658                 vxlan->saddr.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
2659                 vxlan->saddr.sa.sa_family = AF_INET;
2660         } else if (data[IFLA_VXLAN_LOCAL6]) {
2661                 if (!IS_ENABLED(CONFIG_IPV6))
2662                         return -EPFNOSUPPORT;
2663
2664                 /* TODO: respect scope id */
2665                 nla_memcpy(&vxlan->saddr.sin6.sin6_addr, data[IFLA_VXLAN_LOCAL6],
2666                            sizeof(struct in6_addr));
2667                 vxlan->saddr.sa.sa_family = AF_INET6;
2668                 use_ipv6 = true;
2669         }
2670
2671         if (data[IFLA_VXLAN_LINK] &&
2672             (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2673                 struct net_device *lowerdev
2674                          = __dev_get_by_index(net, dst->remote_ifindex);
2675
2676                 if (!lowerdev) {
2677                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2678                         return -ENODEV;
2679                 }
2680
2681 #if IS_ENABLED(CONFIG_IPV6)
2682                 if (use_ipv6) {
2683                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2684                         if (idev && idev->cnf.disable_ipv6) {
2685                                 pr_info("IPv6 is disabled via sysctl\n");
2686                                 return -EPERM;
2687                         }
2688                         vxlan->flags |= VXLAN_F_IPV6;
2689                 }
2690 #endif
2691
2692                 if (!tb[IFLA_MTU])
2693                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2694
2695                 dev->needed_headroom = lowerdev->hard_header_len +
2696                                        (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2697         } else if (use_ipv6)
2698                 vxlan->flags |= VXLAN_F_IPV6;
2699
2700         if (data[IFLA_VXLAN_TOS])
2701                 vxlan->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2702
2703         if (data[IFLA_VXLAN_TTL])
2704                 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2705
2706         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2707                 vxlan->flags |= VXLAN_F_LEARN;
2708
2709         if (data[IFLA_VXLAN_AGEING])
2710                 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2711         else
2712                 vxlan->age_interval = FDB_AGE_DEFAULT;
2713
2714         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2715                 vxlan->flags |= VXLAN_F_PROXY;
2716
2717         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2718                 vxlan->flags |= VXLAN_F_RSC;
2719
2720         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2721                 vxlan->flags |= VXLAN_F_L2MISS;
2722
2723         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2724                 vxlan->flags |= VXLAN_F_L3MISS;
2725
2726         if (data[IFLA_VXLAN_LIMIT])
2727                 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2728
2729         if (data[IFLA_VXLAN_PORT_RANGE]) {
2730                 const struct ifla_vxlan_port_range *p
2731                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2732                 vxlan->port_min = ntohs(p->low);
2733                 vxlan->port_max = ntohs(p->high);
2734         }
2735
2736         if (data[IFLA_VXLAN_PORT])
2737                 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2738
2739         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2740                 vxlan->flags |= VXLAN_F_UDP_CSUM;
2741
2742         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2743             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2744                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2745
2746         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2747             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2748                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2749
2750         if (data[IFLA_VXLAN_REMCSUM_TX] &&
2751             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2752                 vxlan->flags |= VXLAN_F_REMCSUM_TX;
2753
2754         if (data[IFLA_VXLAN_REMCSUM_RX] &&
2755             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2756                 vxlan->flags |= VXLAN_F_REMCSUM_RX;
2757
2758         if (data[IFLA_VXLAN_GBP])
2759                 vxlan->flags |= VXLAN_F_GBP;
2760
2761         if (vxlan_find_vni(net, vni, use_ipv6 ? AF_INET6 : AF_INET,
2762                            vxlan->dst_port, vxlan->flags)) {
2763                 pr_info("duplicate VNI %u\n", vni);
2764                 return -EEXIST;
2765         }
2766
2767         dev->ethtool_ops = &vxlan_ethtool_ops;
2768
2769         /* create an fdb entry for a valid default destination */
2770         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2771                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2772                                        &vxlan->default_dst.remote_ip,
2773                                        NUD_REACHABLE|NUD_PERMANENT,
2774                                        NLM_F_EXCL|NLM_F_CREATE,
2775                                        vxlan->dst_port,
2776                                        vxlan->default_dst.remote_vni,
2777                                        vxlan->default_dst.remote_ifindex,
2778                                        NTF_SELF);
2779                 if (err)
2780                         return err;
2781         }
2782
2783         err = register_netdevice(dev);
2784         if (err) {
2785                 vxlan_fdb_delete_default(vxlan);
2786                 return err;
2787         }
2788
2789         list_add(&vxlan->next, &vn->vxlan_list);
2790
2791         return 0;
2792 }
2793
2794 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2795 {
2796         struct vxlan_dev *vxlan = netdev_priv(dev);
2797         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2798
2799         spin_lock(&vn->sock_lock);
2800         if (!hlist_unhashed(&vxlan->hlist))
2801                 hlist_del_rcu(&vxlan->hlist);
2802         spin_unlock(&vn->sock_lock);
2803
2804         list_del(&vxlan->next);
2805         unregister_netdevice_queue(dev, head);
2806 }
2807
2808 static size_t vxlan_get_size(const struct net_device *dev)
2809 {
2810
2811         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2812                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2813                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2814                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2815                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2816                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2817                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2818                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2819                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2820                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2821                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2822                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2823                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2824                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2825                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2826                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2827                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2828                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2829                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
2830                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
2831                 0;
2832 }
2833
2834 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2835 {
2836         const struct vxlan_dev *vxlan = netdev_priv(dev);
2837         const struct vxlan_rdst *dst = &vxlan->default_dst;
2838         struct ifla_vxlan_port_range ports = {
2839                 .low =  htons(vxlan->port_min),
2840                 .high = htons(vxlan->port_max),
2841         };
2842
2843         if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2844                 goto nla_put_failure;
2845
2846         if (!vxlan_addr_any(&dst->remote_ip)) {
2847                 if (dst->remote_ip.sa.sa_family == AF_INET) {
2848                         if (nla_put_be32(skb, IFLA_VXLAN_GROUP,
2849                                          dst->remote_ip.sin.sin_addr.s_addr))
2850                                 goto nla_put_failure;
2851 #if IS_ENABLED(CONFIG_IPV6)
2852                 } else {
2853                         if (nla_put(skb, IFLA_VXLAN_GROUP6, sizeof(struct in6_addr),
2854                                     &dst->remote_ip.sin6.sin6_addr))
2855                                 goto nla_put_failure;
2856 #endif
2857                 }
2858         }
2859
2860         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2861                 goto nla_put_failure;
2862
2863         if (!vxlan_addr_any(&vxlan->saddr)) {
2864                 if (vxlan->saddr.sa.sa_family == AF_INET) {
2865                         if (nla_put_be32(skb, IFLA_VXLAN_LOCAL,
2866                                          vxlan->saddr.sin.sin_addr.s_addr))
2867                                 goto nla_put_failure;
2868 #if IS_ENABLED(CONFIG_IPV6)
2869                 } else {
2870                         if (nla_put(skb, IFLA_VXLAN_LOCAL6, sizeof(struct in6_addr),
2871                                     &vxlan->saddr.sin6.sin6_addr))
2872                                 goto nla_put_failure;
2873 #endif
2874                 }
2875         }
2876
2877         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2878             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2879             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2880                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
2881             nla_put_u8(skb, IFLA_VXLAN_PROXY,
2882                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
2883             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2884             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2885                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2886             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2887                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2888             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2889             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2890             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port) ||
2891             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2892                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2893             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2894                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2895             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2896                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
2897             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
2898                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
2899             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
2900                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
2901                 goto nla_put_failure;
2902
2903         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2904                 goto nla_put_failure;
2905
2906         if (vxlan->flags & VXLAN_F_GBP &&
2907             nla_put_flag(skb, IFLA_VXLAN_GBP))
2908                 goto nla_put_failure;
2909
2910         return 0;
2911
2912 nla_put_failure:
2913         return -EMSGSIZE;
2914 }
2915
2916 static struct net *vxlan_get_link_net(const struct net_device *dev)
2917 {
2918         struct vxlan_dev *vxlan = netdev_priv(dev);
2919
2920         return vxlan->net;
2921 }
2922
2923 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2924         .kind           = "vxlan",
2925         .maxtype        = IFLA_VXLAN_MAX,
2926         .policy         = vxlan_policy,
2927         .priv_size      = sizeof(struct vxlan_dev),
2928         .setup          = vxlan_setup,
2929         .validate       = vxlan_validate,
2930         .newlink        = vxlan_newlink,
2931         .dellink        = vxlan_dellink,
2932         .get_size       = vxlan_get_size,
2933         .fill_info      = vxlan_fill_info,
2934         .get_link_net   = vxlan_get_link_net,
2935 };
2936
2937 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
2938                                              struct net_device *dev)
2939 {
2940         struct vxlan_dev *vxlan, *next;
2941         LIST_HEAD(list_kill);
2942
2943         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2944                 struct vxlan_rdst *dst = &vxlan->default_dst;
2945
2946                 /* In case we created vxlan device with carrier
2947                  * and we loose the carrier due to module unload
2948                  * we also need to remove vxlan device. In other
2949                  * cases, it's not necessary and remote_ifindex
2950                  * is 0 here, so no matches.
2951                  */
2952                 if (dst->remote_ifindex == dev->ifindex)
2953                         vxlan_dellink(vxlan->dev, &list_kill);
2954         }
2955
2956         unregister_netdevice_many(&list_kill);
2957 }
2958
2959 static int vxlan_lowerdev_event(struct notifier_block *unused,
2960                                 unsigned long event, void *ptr)
2961 {
2962         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2963         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2964
2965         if (event == NETDEV_UNREGISTER)
2966                 vxlan_handle_lowerdev_unregister(vn, dev);
2967
2968         return NOTIFY_DONE;
2969 }
2970
2971 static struct notifier_block vxlan_notifier_block __read_mostly = {
2972         .notifier_call = vxlan_lowerdev_event,
2973 };
2974
2975 static __net_init int vxlan_init_net(struct net *net)
2976 {
2977         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2978         unsigned int h;
2979
2980         INIT_LIST_HEAD(&vn->vxlan_list);
2981         spin_lock_init(&vn->sock_lock);
2982
2983         for (h = 0; h < PORT_HASH_SIZE; ++h)
2984                 INIT_HLIST_HEAD(&vn->sock_list[h]);
2985
2986         return 0;
2987 }
2988
2989 static void __net_exit vxlan_exit_net(struct net *net)
2990 {
2991         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2992         struct vxlan_dev *vxlan, *next;
2993         struct net_device *dev, *aux;
2994         LIST_HEAD(list);
2995
2996         rtnl_lock();
2997         for_each_netdev_safe(net, dev, aux)
2998                 if (dev->rtnl_link_ops == &vxlan_link_ops)
2999                         unregister_netdevice_queue(dev, &list);
3000
3001         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3002                 /* If vxlan->dev is in the same netns, it has already been added
3003                  * to the list by the previous loop.
3004                  */
3005                 if (!net_eq(dev_net(vxlan->dev), net))
3006                         unregister_netdevice_queue(dev, &list);
3007         }
3008
3009         unregister_netdevice_many(&list);
3010         rtnl_unlock();
3011 }
3012
3013 static struct pernet_operations vxlan_net_ops = {
3014         .init = vxlan_init_net,
3015         .exit = vxlan_exit_net,
3016         .id   = &vxlan_net_id,
3017         .size = sizeof(struct vxlan_net),
3018 };
3019
3020 static int __init vxlan_init_module(void)
3021 {
3022         int rc;
3023
3024         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3025         if (!vxlan_wq)
3026                 return -ENOMEM;
3027
3028         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3029
3030         rc = register_pernet_subsys(&vxlan_net_ops);
3031         if (rc)
3032                 goto out1;
3033
3034         rc = register_netdevice_notifier(&vxlan_notifier_block);
3035         if (rc)
3036                 goto out2;
3037
3038         rc = rtnl_link_register(&vxlan_link_ops);
3039         if (rc)
3040                 goto out3;
3041
3042         return 0;
3043 out3:
3044         unregister_netdevice_notifier(&vxlan_notifier_block);
3045 out2:
3046         unregister_pernet_subsys(&vxlan_net_ops);
3047 out1:
3048         destroy_workqueue(vxlan_wq);
3049         return rc;
3050 }
3051 late_initcall(vxlan_init_module);
3052
3053 static void __exit vxlan_cleanup_module(void)
3054 {
3055         rtnl_link_unregister(&vxlan_link_ops);
3056         unregister_netdevice_notifier(&vxlan_notifier_block);
3057         destroy_workqueue(vxlan_wq);
3058         unregister_pernet_subsys(&vxlan_net_ops);
3059         /* rcu_barrier() is called by netns */
3060 }
3061 module_exit(vxlan_cleanup_module);
3062
3063 MODULE_LICENSE("GPL");
3064 MODULE_VERSION(VXLAN_VERSION);
3065 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3066 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3067 MODULE_ALIAS_RTNL_LINK("vxlan");