drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/types.h>
  15 #include <linux/module.h>
  16 #include <linux/errno.h>
  17 #include <linux/slab.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/rculist.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/in.h>
  22 #include <linux/ip.h>
  23 #include <linux/udp.h>
  24 #include <linux/igmp.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/hash.h>
  29 #include <linux/ethtool.h>
  30 #include <net/arp.h>
  31 #include <net/ndisc.h>
  32 #include <net/ip.h>
  33 #include <net/ip_tunnels.h>
  34 #include <net/icmp.h>
  35 #include <net/udp.h>
  36 #include <net/udp_tunnel.h>
  37 #include <net/rtnetlink.h>
  38 #include <net/route.h>
  39 #include <net/dsfield.h>
  40 #include <net/inet_ecn.h>
  41 #include <net/net_namespace.h>
  42 #include <net/netns/generic.h>
  43 #include <net/vxlan.h>
  44 #include <net/protocol.h>
  45 #include <net/udp_tunnel.h>
  46 #if IS_ENABLED(CONFIG_IPV6)
  47 #include <net/ipv6.h>
  48 #include <net/addrconf.h>
  49 #include <net/ip6_tunnel.h>
  50 #include <net/ip6_checksum.h>
  51 #endif
  52 #include <net/dst_metadata.h>
  53
  54 #define VXLAN_VERSION   "0.1"
  55
  56 #define PORT_HASH_BITS  8
  57 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  58 #define FDB_AGE_DEFAULT 300 /* 5 min */
  59 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  60
  61 /* UDP port for VXLAN traffic.
  62  * The IANA assigned port is 4789, but the Linux default is 8472
  63  * for compatibility with early adopters.
  64  */
  65 static unsigned short vxlan_port __read_mostly = 8472;
  66 module_param_named(udp_port, vxlan_port, ushort, 0444);
  67 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  68
  69 static bool log_ecn_error = true;
  70 module_param(log_ecn_error, bool, 0644);
  71 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  72
  73 static int vxlan_net_id;
  74 static struct rtnl_link_ops vxlan_link_ops;
  75
  76 static const u8 all_zeros_mac[ETH_ALEN];
  77
  78 static struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
  79                                          bool no_share, u32 flags);
  80
  81 /* per-network namespace private data for this module */
  82 struct vxlan_net {
  83         struct list_head  vxlan_list;
  84         struct hlist_head sock_list[PORT_HASH_SIZE];
  85         spinlock_t        sock_lock;
  86 };
  87
  88 /* Forwarding table entry */
  89 struct vxlan_fdb {
  90         struct hlist_node hlist;        /* linked list of entries */
  91         struct rcu_head   rcu;
  92         unsigned long     updated;      /* jiffies */
  93         unsigned long     used;
  94         struct list_head  remotes;
  95         u8                eth_addr[ETH_ALEN];
  96         u16               state;        /* see ndm_state */
  97         u8                flags;        /* see ndm_flags */
  98 };
  99
 100 /* salt for hash table */
 101 static u32 vxlan_salt __read_mostly;
 102 static struct workqueue_struct *vxlan_wq;
 103
 104 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
 105 {
 106         return vs->flags & VXLAN_F_COLLECT_METADATA ||
 107                ip_tunnel_collect_metadata();
 108 }
 109
 110 #if IS_ENABLED(CONFIG_IPV6)
 111 static inline
 112 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 113 {
 114         if (a->sa.sa_family != b->sa.sa_family)
 115                 return false;
 116         if (a->sa.sa_family == AF_INET6)
 117                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 118         else
 119                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 120 }
 121
 122 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 123 {
 124         if (ipa->sa.sa_family == AF_INET6)
 125                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 126         else
 127                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 128 }
 129
 130 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 131 {
 132         if (ipa->sa.sa_family == AF_INET6)
 133                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 134         else
 135                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 136 }
 137
 138 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 139 {
 140         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 141                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 142                 ip->sa.sa_family = AF_INET6;
 143                 return 0;
 144         } else if (nla_len(nla) >= sizeof(__be32)) {
 145                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 146                 ip->sa.sa_family = AF_INET;
 147                 return 0;
 148         } else {
 149                 return -EAFNOSUPPORT;
 150         }
 151 }
 152
 153 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 154                               const union vxlan_addr *ip)
 155 {
 156         if (ip->sa.sa_family == AF_INET6)
 157                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 158         else
 159                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 160 }
 161
 162 #else /* !CONFIG_IPV6 */
 163
 164 static inline
 165 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 166 {
 167         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 168 }
 169
 170 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 171 {
 172         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 173 }
 174
 175 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 176 {
 177         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 178 }
 179
 180 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 181 {
 182         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 183                 return -EAFNOSUPPORT;
 184         } else if (nla_len(nla) >= sizeof(__be32)) {
 185                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 186                 ip->sa.sa_family = AF_INET;
 187                 return 0;
 188         } else {
 189                 return -EAFNOSUPPORT;
 190         }
 191 }
 192
 193 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 194                               const union vxlan_addr *ip)
 195 {
 196         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 197 }
 198 #endif
 199
 200 /* Virtual Network hash table head */
 201 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
 202 {
 203         return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
 204 }
 205
 206 /* Socket hash table head */
 207 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 208 {
 209         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 210
 211         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 212 }
 213
 214 /* First remote destination for a forwarding entry.
 215  * Guaranteed to be non-NULL because remotes are never deleted.
 216  */
 217 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 218 {
 219         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 220 }
 221
 222 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 223 {
 224         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 225 }
 226
 227 /* Find VXLAN socket based on network namespace, address family and UDP port
 228  * and enabled unshareable flags.
 229  */
 230 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 231                                           __be16 port, u32 flags)
 232 {
 233         struct vxlan_sock *vs;
 234
 235         flags &= VXLAN_F_RCV_FLAGS;
 236
 237         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 238                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 239                     vxlan_get_sk_family(vs) == family &&
 240                     vs->flags == flags)
 241                         return vs;
 242         }
 243         return NULL;
 244 }
 245
 246 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
 247 {
 248         struct vxlan_dev *vxlan;
 249
 250         hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
 251                 if (vxlan->default_dst.remote_vni == id)
 252                         return vxlan;
 253         }
 254
 255         return NULL;
 256 }
 257
 258 /* Look up VNI in a per net namespace table */
 259 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id,
 260                                         sa_family_t family, __be16 port,
 261                                         u32 flags)
 262 {
 263         struct vxlan_sock *vs;
 264
 265         vs = vxlan_find_sock(net, family, port, flags);
 266         if (!vs)
 267                 return NULL;
 268
 269         return vxlan_vs_find_vni(vs, id);
 270 }
 271
 272 /* Fill in neighbour message in skbuff. */
 273 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 274                           const struct vxlan_fdb *fdb,
 275                           u32 portid, u32 seq, int type, unsigned int flags,
 276                           const struct vxlan_rdst *rdst)
 277 {
 278         unsigned long now = jiffies;
 279         struct nda_cacheinfo ci;
 280         struct nlmsghdr *nlh;
 281         struct ndmsg *ndm;
 282         bool send_ip, send_eth;
 283
 284         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 285         if (nlh == NULL)
 286                 return -EMSGSIZE;
 287
 288         ndm = nlmsg_data(nlh);
 289         memset(ndm, 0, sizeof(*ndm));
 290
 291         send_eth = send_ip = true;
 292
 293         if (type == RTM_GETNEIGH) {
 294                 ndm->ndm_family = AF_INET;
 295                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 296                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 297         } else
 298                 ndm->ndm_family = AF_BRIDGE;
 299         ndm->ndm_state = fdb->state;
 300         ndm->ndm_ifindex = vxlan->dev->ifindex;
 301         ndm->ndm_flags = fdb->flags;
 302         ndm->ndm_type = RTN_UNICAST;
 303
 304         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 305             nla_put_s32(skb, NDA_LINK_NETNSID,
 306                         peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
 307                 goto nla_put_failure;
 308
 309         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 310                 goto nla_put_failure;
 311
 312         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 313                 goto nla_put_failure;
 314
 315         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 316             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 317                 goto nla_put_failure;
 318         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 319             nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
 320                 goto nla_put_failure;
 321         if (rdst->remote_ifindex &&
 322             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 323                 goto nla_put_failure;
 324
 325         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 326         ci.ndm_confirmed = 0;
 327         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 328         ci.ndm_refcnt    = 0;
 329
 330         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 331                 goto nla_put_failure;
 332
 333         nlmsg_end(skb, nlh);
 334         return 0;
 335
 336 nla_put_failure:
 337         nlmsg_cancel(skb, nlh);
 338         return -EMSGSIZE;
 339 }
 340
 341 static inline size_t vxlan_nlmsg_size(void)
 342 {
 343         return NLMSG_ALIGN(sizeof(struct ndmsg))
 344                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 345                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 346                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 347                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 348                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 349                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 350                 + nla_total_size(sizeof(struct nda_cacheinfo));
 351 }
 352
 353 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 354                              struct vxlan_rdst *rd, int type)
 355 {
 356         struct net *net = dev_net(vxlan->dev);
 357         struct sk_buff *skb;
 358         int err = -ENOBUFS;
 359
 360         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 361         if (skb == NULL)
 362                 goto errout;
 363
 364         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 365         if (err < 0) {
 366                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 367                 WARN_ON(err == -EMSGSIZE);
 368                 kfree_skb(skb);
 369                 goto errout;
 370         }
 371
 372         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 373         return;
 374 errout:
 375         if (err < 0)
 376                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 377 }
 378
 379 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 380 {
 381         struct vxlan_dev *vxlan = netdev_priv(dev);
 382         struct vxlan_fdb f = {
 383                 .state = NUD_STALE,
 384         };
 385         struct vxlan_rdst remote = {
 386                 .remote_ip = *ipa, /* goes to NDA_DST */
 387                 .remote_vni = VXLAN_N_VID,
 388         };
 389
 390         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 391 }
 392
 393 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 394 {
 395         struct vxlan_fdb f = {
 396                 .state = NUD_STALE,
 397         };
 398         struct vxlan_rdst remote = { };
 399
 400         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 401
 402         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 403 }
 404
 405 /* Hash Ethernet address */
 406 static u32 eth_hash(const unsigned char *addr)
 407 {
 408         u64 value = get_unaligned((u64 *)addr);
 409
 410         /* only want 6 bytes */
 411 #ifdef __BIG_ENDIAN
 412         value >>= 16;
 413 #else
 414         value <<= 16;
 415 #endif
 416         return hash_64(value, FDB_HASH_BITS);
 417 }
 418
 419 /* Hash chain to use given mac address */
 420 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 421                                                 const u8 *mac)
 422 {
 423         return &vxlan->fdb_head[eth_hash(mac)];
 424 }
 425
 426 /* Look up Ethernet address in forwarding table */
 427 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 428                                         const u8 *mac)
 429 {
 430         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 431         struct vxlan_fdb *f;
 432
 433         hlist_for_each_entry_rcu(f, head, hlist) {
 434                 if (ether_addr_equal(mac, f->eth_addr))
 435                         return f;
 436         }
 437
 438         return NULL;
 439 }
 440
 441 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 442                                         const u8 *mac)
 443 {
 444         struct vxlan_fdb *f;
 445
 446         f = __vxlan_find_mac(vxlan, mac);
 447         if (f)
 448                 f->used = jiffies;
 449
 450         return f;
 451 }
 452
 453 /* caller should hold vxlan->hash_lock */
 454 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 455                                               union vxlan_addr *ip, __be16 port,
 456                                               __u32 vni, __u32 ifindex)
 457 {
 458         struct vxlan_rdst *rd;
 459
 460         list_for_each_entry(rd, &f->remotes, list) {
 461                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 462                     rd->remote_port == port &&
 463                     rd->remote_vni == vni &&
 464                     rd->remote_ifindex == ifindex)
 465                         return rd;
 466         }
 467
 468         return NULL;
 469 }
 470
 471 /* Replace destination of unicast mac */
 472 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 473                              union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
 474 {
 475         struct vxlan_rdst *rd;
 476
 477         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 478         if (rd)
 479                 return 0;
 480
 481         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 482         if (!rd)
 483                 return 0;
 484         rd->remote_ip = *ip;
 485         rd->remote_port = port;
 486         rd->remote_vni = vni;
 487         rd->remote_ifindex = ifindex;
 488         return 1;
 489 }
 490
 491 /* Add/update destinations for multicast */
 492 static int vxlan_fdb_append(struct vxlan_fdb *f,
 493                             union vxlan_addr *ip, __be16 port, __u32 vni,
 494                             __u32 ifindex, struct vxlan_rdst **rdp)
 495 {
 496         struct vxlan_rdst *rd;
 497
 498         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 499         if (rd)
 500                 return 0;
 501
 502         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 503         if (rd == NULL)
 504                 return -ENOBUFS;
 505         rd->remote_ip = *ip;
 506         rd->remote_port = port;
 507         rd->remote_vni = vni;
 508         rd->remote_ifindex = ifindex;
 509
 510         list_add_tail_rcu(&rd->list, &f->remotes);
 511
 512         *rdp = rd;
 513         return 1;
 514 }
 515
 516 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 517                                           unsigned int off,
 518                                           struct vxlanhdr *vh, size_t hdrlen,
 519                                           u32 data, struct gro_remcsum *grc,
 520                                           bool nopartial)
 521 {
 522         size_t start, offset, plen;
 523
 524         if (skb->remcsum_offload)
 525                 return NULL;
 526
 527         if (!NAPI_GRO_CB(skb)->csum_valid)
 528                 return NULL;
 529
 530         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
 531         offset = start + ((data & VXLAN_RCO_UDP) ?
 532                           offsetof(struct udphdr, check) :
 533                           offsetof(struct tcphdr, check));
 534
 535         plen = hdrlen + offset + sizeof(u16);
 536
 537         /* Pull checksum that will be written */
 538         if (skb_gro_header_hard(skb, off + plen)) {
 539                 vh = skb_gro_header_slow(skb, off + plen, off);
 540                 if (!vh)
 541                         return NULL;
 542         }
 543
 544         skb_gro_remcsum_process(skb, (void *)vh + hdrlen,
 545                                 start, offset, grc, nopartial);
 546
 547         skb->remcsum_offload = 1;
 548
 549         return vh;
 550 }
 551
 552 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
 553                                           struct sk_buff *skb,
 554                                           struct udp_offload *uoff)
 555 {
 556         struct sk_buff *p, **pp = NULL;
 557         struct vxlanhdr *vh, *vh2;
 558         unsigned int hlen, off_vx;
 559         int flush = 1;
 560         struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
 561                                              udp_offloads);
 562         u32 flags;
 563         struct gro_remcsum grc;
 564
 565         skb_gro_remcsum_init(&grc);
 566
 567         off_vx = skb_gro_offset(skb);
 568         hlen = off_vx + sizeof(*vh);
 569         vh   = skb_gro_header_fast(skb, off_vx);
 570         if (skb_gro_header_hard(skb, hlen)) {
 571                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 572                 if (unlikely(!vh))
 573                         goto out;
 574         }
 575
 576         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 577         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 578
 579         flags = ntohl(vh->vx_flags);
 580
 581         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 582                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 583                                        ntohl(vh->vx_vni), &grc,
 584                                        !!(vs->flags &
 585                                           VXLAN_F_REMCSUM_NOPARTIAL));
 586
 587                 if (!vh)
 588                         goto out;
 589         }
 590
 591         flush = 0;
 592
 593         for (p = *head; p; p = p->next) {
 594                 if (!NAPI_GRO_CB(p)->same_flow)
 595                         continue;
 596
 597                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 598                 if (vh->vx_flags != vh2->vx_flags ||
 599                     vh->vx_vni != vh2->vx_vni) {
 600                         NAPI_GRO_CB(p)->same_flow = 0;
 601                         continue;
 602                 }
 603         }
 604
 605         pp = eth_gro_receive(head, skb);
 606
 607 out:
 608         skb_gro_remcsum_cleanup(skb, &grc);
 609         NAPI_GRO_CB(skb)->flush |= flush;
 610
 611         return pp;
 612 }
 613
 614 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
 615                               struct udp_offload *uoff)
 616 {
 617         udp_tunnel_gro_complete(skb, nhoff);
 618
 619         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 620 }
 621
 622 /* Notify netdevs that UDP port started listening */
 623 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 624 {
 625         struct net_device *dev;
 626         struct sock *sk = vs->sock->sk;
 627         struct net *net = sock_net(sk);
 628         sa_family_t sa_family = vxlan_get_sk_family(vs);
 629         __be16 port = inet_sk(sk)->inet_sport;
 630         int err;
 631
 632         if (sa_family == AF_INET) {
 633                 err = udp_add_offload(&vs->udp_offloads);
 634                 if (err)
 635                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 636         }
 637
 638         rcu_read_lock();
 639         for_each_netdev_rcu(net, dev) {
 640                 if (dev->netdev_ops->ndo_add_vxlan_port)
 641                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 642                                                             port);
 643         }
 644         rcu_read_unlock();
 645 }
 646
 647 /* Notify netdevs that UDP port is no more listening */
 648 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 649 {
 650         struct net_device *dev;
 651         struct sock *sk = vs->sock->sk;
 652         struct net *net = sock_net(sk);
 653         sa_family_t sa_family = vxlan_get_sk_family(vs);
 654         __be16 port = inet_sk(sk)->inet_sport;
 655
 656         rcu_read_lock();
 657         for_each_netdev_rcu(net, dev) {
 658                 if (dev->netdev_ops->ndo_del_vxlan_port)
 659                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 660                                                             port);
 661         }
 662         rcu_read_unlock();
 663
 664         if (sa_family == AF_INET)
 665                 udp_del_offload(&vs->udp_offloads);
 666 }
 667
 668 /* Add new entry to forwarding table -- assumes lock held */
 669 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 670                             const u8 *mac, union vxlan_addr *ip,
 671                             __u16 state, __u16 flags,
 672                             __be16 port, __u32 vni, __u32 ifindex,
 673                             __u8 ndm_flags)
 674 {
 675         struct vxlan_rdst *rd = NULL;
 676         struct vxlan_fdb *f;
 677         int notify = 0;
 678
 679         f = __vxlan_find_mac(vxlan, mac);
 680         if (f) {
 681                 if (flags & NLM_F_EXCL) {
 682                         netdev_dbg(vxlan->dev,
 683                                    "lost race to create %pM\n", mac);
 684                         return -EEXIST;
 685                 }
 686                 if (f->state != state) {
 687                         f->state = state;
 688                         f->updated = jiffies;
 689                         notify = 1;
 690                 }
 691                 if (f->flags != ndm_flags) {
 692                         f->flags = ndm_flags;
 693                         f->updated = jiffies;
 694                         notify = 1;
 695                 }
 696                 if ((flags & NLM_F_REPLACE)) {
 697                         /* Only change unicasts */
 698                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 699                              is_zero_ether_addr(f->eth_addr))) {
 700                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 701                                                            ifindex);
 702                         } else
 703                                 return -EOPNOTSUPP;
 704                 }
 705                 if ((flags & NLM_F_APPEND) &&
 706                     (is_multicast_ether_addr(f->eth_addr) ||
 707                      is_zero_ether_addr(f->eth_addr))) {
 708                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 709                                                   &rd);
 710
 711                         if (rc < 0)
 712                                 return rc;
 713                         notify |= rc;
 714                 }
 715         } else {
 716                 if (!(flags & NLM_F_CREATE))
 717                         return -ENOENT;
 718
 719                 if (vxlan->cfg.addrmax &&
 720                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 721                         return -ENOSPC;
 722
 723                 /* Disallow replace to add a multicast entry */
 724                 if ((flags & NLM_F_REPLACE) &&
 725                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 726                         return -EOPNOTSUPP;
 727
 728                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 729                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 730                 if (!f)
 731                         return -ENOMEM;
 732
 733                 notify = 1;
 734                 f->state = state;
 735                 f->flags = ndm_flags;
 736                 f->updated = f->used = jiffies;
 737                 INIT_LIST_HEAD(&f->remotes);
 738                 memcpy(f->eth_addr, mac, ETH_ALEN);
 739
 740                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 741
 742                 ++vxlan->addrcnt;
 743                 hlist_add_head_rcu(&f->hlist,
 744                                    vxlan_fdb_head(vxlan, mac));
 745         }
 746
 747         if (notify) {
 748                 if (rd == NULL)
 749                         rd = first_remote_rtnl(f);
 750                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 751         }
 752
 753         return 0;
 754 }
 755
 756 static void vxlan_fdb_free(struct rcu_head *head)
 757 {
 758         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 759         struct vxlan_rdst *rd, *nd;
 760
 761         list_for_each_entry_safe(rd, nd, &f->remotes, list)
 762                 kfree(rd);
 763         kfree(f);
 764 }
 765
 766 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 767 {
 768         netdev_dbg(vxlan->dev,
 769                     "delete %pM\n", f->eth_addr);
 770
 771         --vxlan->addrcnt;
 772         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 773
 774         hlist_del_rcu(&f->hlist);
 775         call_rcu(&f->rcu, vxlan_fdb_free);
 776 }
 777
 778 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 779                            union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
 780 {
 781         struct net *net = dev_net(vxlan->dev);
 782         int err;
 783
 784         if (tb[NDA_DST]) {
 785                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 786                 if (err)
 787                         return err;
 788         } else {
 789                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 790                 if (remote->sa.sa_family == AF_INET) {
 791                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 792                         ip->sa.sa_family = AF_INET;
 793 #if IS_ENABLED(CONFIG_IPV6)
 794                 } else {
 795                         ip->sin6.sin6_addr = in6addr_any;
 796                         ip->sa.sa_family = AF_INET6;
 797 #endif
 798                 }
 799         }
 800
 801         if (tb[NDA_PORT]) {
 802                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 803                         return -EINVAL;
 804                 *port = nla_get_be16(tb[NDA_PORT]);
 805         } else {
 806                 *port = vxlan->cfg.dst_port;
 807         }
 808
 809         if (tb[NDA_VNI]) {
 810                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 811                         return -EINVAL;
 812                 *vni = nla_get_u32(tb[NDA_VNI]);
 813         } else {
 814                 *vni = vxlan->default_dst.remote_vni;
 815         }
 816
 817         if (tb[NDA_IFINDEX]) {
 818                 struct net_device *tdev;
 819
 820                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 821                         return -EINVAL;
 822                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 823                 tdev = __dev_get_by_index(net, *ifindex);
 824                 if (!tdev)
 825                         return -EADDRNOTAVAIL;
 826         } else {
 827                 *ifindex = 0;
 828         }
 829
 830         return 0;
 831 }
 832
 833 /* Add static entry (via netlink) */
 834 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 835                          struct net_device *dev,
 836                          const unsigned char *addr, u16 vid, u16 flags)
 837 {
 838         struct vxlan_dev *vxlan = netdev_priv(dev);
 839         /* struct net *net = dev_net(vxlan->dev); */
 840         union vxlan_addr ip;
 841         __be16 port;
 842         u32 vni, ifindex;
 843         int err;
 844
 845         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 846                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 847                         ndm->ndm_state);
 848                 return -EINVAL;
 849         }
 850
 851         if (tb[NDA_DST] == NULL)
 852                 return -EINVAL;
 853
 854         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 855         if (err)
 856                 return err;
 857
 858         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 859                 return -EAFNOSUPPORT;
 860
 861         spin_lock_bh(&vxlan->hash_lock);
 862         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 863                                port, vni, ifindex, ndm->ndm_flags);
 864         spin_unlock_bh(&vxlan->hash_lock);
 865
 866         return err;
 867 }
 868
 869 /* Delete entry (via netlink) */
 870 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 871                             struct net_device *dev,
 872                             const unsigned char *addr, u16 vid)
 873 {
 874         struct vxlan_dev *vxlan = netdev_priv(dev);
 875         struct vxlan_fdb *f;
 876         struct vxlan_rdst *rd = NULL;
 877         union vxlan_addr ip;
 878         __be16 port;
 879         u32 vni, ifindex;
 880         int err;
 881
 882         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 883         if (err)
 884                 return err;
 885
 886         err = -ENOENT;
 887
 888         spin_lock_bh(&vxlan->hash_lock);
 889         f = vxlan_find_mac(vxlan, addr);
 890         if (!f)
 891                 goto out;
 892
 893         if (!vxlan_addr_any(&ip)) {
 894                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 895                 if (!rd)
 896                         goto out;
 897         }
 898
 899         err = 0;
 900
 901         /* remove a destination if it's not the only one on the list,
 902          * otherwise destroy the fdb entry
 903          */
 904         if (rd && !list_is_singular(&f->remotes)) {
 905                 list_del_rcu(&rd->list);
 906                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 907                 kfree_rcu(rd, rcu);
 908                 goto out;
 909         }
 910
 911         vxlan_fdb_destroy(vxlan, f);
 912
 913 out:
 914         spin_unlock_bh(&vxlan->hash_lock);
 915
 916         return err;
 917 }
 918
 919 /* Dump forwarding table */
 920 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 921                           struct net_device *dev,
 922                           struct net_device *filter_dev, int idx)
 923 {
 924         struct vxlan_dev *vxlan = netdev_priv(dev);
 925         unsigned int h;
 926
 927         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 928                 struct vxlan_fdb *f;
 929                 int err;
 930
 931                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 932                         struct vxlan_rdst *rd;
 933
 934                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 935                                 if (idx < cb->args[0])
 936                                         goto skip;
 937
 938                                 err = vxlan_fdb_info(skb, vxlan, f,
 939                                                      NETLINK_CB(cb->skb).portid,
 940                                                      cb->nlh->nlmsg_seq,
 941                                                      RTM_NEWNEIGH,
 942                                                      NLM_F_MULTI, rd);
 943                                 if (err < 0)
 944                                         goto out;
 945 skip:
 946                                 ++idx;
 947                         }
 948                 }
 949         }
 950 out:
 951         return idx;
 952 }
 953
 954 /* Watch incoming packets to learn mapping between Ethernet address
 955  * and Tunnel endpoint.
 956  * Return true if packet is bogus and should be dropped.
 957  */
 958 static bool vxlan_snoop(struct net_device *dev,
 959                         union vxlan_addr *src_ip, const u8 *src_mac)
 960 {
 961         struct vxlan_dev *vxlan = netdev_priv(dev);
 962         struct vxlan_fdb *f;
 963
 964         f = vxlan_find_mac(vxlan, src_mac);
 965         if (likely(f)) {
 966                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 967
 968                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 969                         return false;
 970
 971                 /* Don't migrate static entries, drop packets */
 972                 if (f->state & NUD_NOARP)
 973                         return true;
 974
 975                 if (net_ratelimit())
 976                         netdev_info(dev,
 977                                     "%pM migrated from %pIS to %pIS\n",
 978                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 979
 980                 rdst->remote_ip = *src_ip;
 981                 f->updated = jiffies;
 982                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 983         } else {
 984                 /* learned new entry */
 985                 spin_lock(&vxlan->hash_lock);
 986
 987                 /* close off race between vxlan_flush and incoming packets */
 988                 if (netif_running(dev))
 989                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 990                                          NUD_REACHABLE,
 991                                          NLM_F_EXCL|NLM_F_CREATE,
 992                                          vxlan->cfg.dst_port,
 993                                          vxlan->default_dst.remote_vni,
 994                                          0, NTF_SELF);
 995                 spin_unlock(&vxlan->hash_lock);
 996         }
 997
 998         return false;
 999 }
1000
1001 /* See if multicast group is already in use by other ID */
1002 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1003 {
1004         struct vxlan_dev *vxlan;
1005
1006         /* The vxlan_sock is only used by dev, leaving group has
1007          * no effect on other vxlan devices.
1008          */
1009         if (atomic_read(&dev->vn_sock->refcnt) == 1)
1010                 return false;
1011
1012         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1013                 if (!netif_running(vxlan->dev) || vxlan == dev)
1014                         continue;
1015
1016                 if (vxlan->vn_sock != dev->vn_sock)
1017                         continue;
1018
1019                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1020                                       &dev->default_dst.remote_ip))
1021                         continue;
1022
1023                 if (vxlan->default_dst.remote_ifindex !=
1024                     dev->default_dst.remote_ifindex)
1025                         continue;
1026
1027                 return true;
1028         }
1029
1030         return false;
1031 }
1032
1033 static void vxlan_sock_release(struct vxlan_sock *vs)
1034 {
1035         struct sock *sk = vs->sock->sk;
1036         struct net *net = sock_net(sk);
1037         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1038
1039         if (!atomic_dec_and_test(&vs->refcnt))
1040                 return;
1041
1042         spin_lock(&vn->sock_lock);
1043         hlist_del_rcu(&vs->hlist);
1044         vxlan_notify_del_rx_port(vs);
1045         spin_unlock(&vn->sock_lock);
1046
1047         queue_work(vxlan_wq, &vs->del_work);
1048 }
1049
1050 /* Update multicast group membership when first VNI on
1051  * multicast address is brought up
1052  */
1053 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1054 {
1055         struct vxlan_sock *vs = vxlan->vn_sock;
1056         struct sock *sk = vs->sock->sk;
1057         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1058         int ifindex = vxlan->default_dst.remote_ifindex;
1059         int ret = -EINVAL;
1060
1061         lock_sock(sk);
1062         if (ip->sa.sa_family == AF_INET) {
1063                 struct ip_mreqn mreq = {
1064                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1065                         .imr_ifindex            = ifindex,
1066                 };
1067
1068                 ret = ip_mc_join_group(sk, &mreq);
1069 #if IS_ENABLED(CONFIG_IPV6)
1070         } else {
1071                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1072                                                    &ip->sin6.sin6_addr);
1073 #endif
1074         }
1075         release_sock(sk);
1076
1077         return ret;
1078 }
1079
1080 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1081 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1082 {
1083         struct vxlan_sock *vs = vxlan->vn_sock;
1084         struct sock *sk = vs->sock->sk;
1085         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1086         int ifindex = vxlan->default_dst.remote_ifindex;
1087         int ret = -EINVAL;
1088
1089         lock_sock(sk);
1090         if (ip->sa.sa_family == AF_INET) {
1091                 struct ip_mreqn mreq = {
1092                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1093                         .imr_ifindex            = ifindex,
1094                 };
1095
1096                 ret = ip_mc_leave_group(sk, &mreq);
1097 #if IS_ENABLED(CONFIG_IPV6)
1098         } else {
1099                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1100                                                    &ip->sin6.sin6_addr);
1101 #endif
1102         }
1103         release_sock(sk);
1104
1105         return ret;
1106 }
1107
1108 static struct vxlanhdr *vxlan_remcsum(struct sk_buff *skb, struct vxlanhdr *vh,
1109                                       size_t hdrlen, u32 data, bool nopartial)
1110 {
1111         size_t start, offset, plen;
1112
1113         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
1114         offset = start + ((data & VXLAN_RCO_UDP) ?
1115                           offsetof(struct udphdr, check) :
1116                           offsetof(struct tcphdr, check));
1117
1118         plen = hdrlen + offset + sizeof(u16);
1119
1120         if (!pskb_may_pull(skb, plen))
1121                 return NULL;
1122
1123         vh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1124
1125         skb_remcsum_process(skb, (void *)vh + hdrlen, start, offset,
1126                             nopartial);
1127
1128         return vh;
1129 }
1130
1131 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1132                       struct vxlan_metadata *md, u32 vni,
1133                       struct metadata_dst *tun_dst)
1134 {
1135         struct iphdr *oip = NULL;
1136         struct ipv6hdr *oip6 = NULL;
1137         struct vxlan_dev *vxlan;
1138         struct pcpu_sw_netstats *stats;
1139         union vxlan_addr saddr;
1140         int err = 0;
1141         union vxlan_addr *remote_ip;
1142
1143         /* For flow based devices, map all packets to VNI 0 */
1144         if (vs->flags & VXLAN_F_COLLECT_METADATA)
1145                 vni = 0;
1146
1147         /* Is this VNI defined? */
1148         vxlan = vxlan_vs_find_vni(vs, vni);
1149         if (!vxlan)
1150                 goto drop;
1151
1152         remote_ip = &vxlan->default_dst.remote_ip;
1153         skb_reset_mac_header(skb);
1154         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1155         skb->protocol = eth_type_trans(skb, vxlan->dev);
1156         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1157
1158         /* Ignore packet loops (and multicast echo) */
1159         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1160                 goto drop;
1161
1162         /* Re-examine inner Ethernet packet */
1163         if (remote_ip->sa.sa_family == AF_INET) {
1164                 oip = ip_hdr(skb);
1165                 saddr.sin.sin_addr.s_addr = oip->saddr;
1166                 saddr.sa.sa_family = AF_INET;
1167 #if IS_ENABLED(CONFIG_IPV6)
1168         } else {
1169                 oip6 = ipv6_hdr(skb);
1170                 saddr.sin6.sin6_addr = oip6->saddr;
1171                 saddr.sa.sa_family = AF_INET6;
1172 #endif
1173         }
1174
1175         if (tun_dst) {
1176                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1177                 tun_dst = NULL;
1178         }
1179
1180         if ((vxlan->flags & VXLAN_F_LEARN) &&
1181             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1182                 goto drop;
1183
1184         skb_reset_network_header(skb);
1185         /* In flow-based mode, GBP is carried in dst_metadata */
1186         if (!(vs->flags & VXLAN_F_COLLECT_METADATA))
1187                 skb->mark = md->gbp;
1188
1189         if (oip6)
1190                 err = IP6_ECN_decapsulate(oip6, skb);
1191         if (oip)
1192                 err = IP_ECN_decapsulate(oip, skb);
1193
1194         if (unlikely(err)) {
1195                 if (log_ecn_error) {
1196                         if (oip6)
1197                                 net_info_ratelimited("non-ECT from %pI6\n",
1198                                                      &oip6->saddr);
1199                         if (oip)
1200                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1201                                                      &oip->saddr, oip->tos);
1202                 }
1203                 if (err > 1) {
1204                         ++vxlan->dev->stats.rx_frame_errors;
1205                         ++vxlan->dev->stats.rx_errors;
1206                         goto drop;
1207                 }
1208         }
1209
1210         stats = this_cpu_ptr(vxlan->dev->tstats);
1211         u64_stats_update_begin(&stats->syncp);
1212         stats->rx_packets++;
1213         stats->rx_bytes += skb->len;
1214         u64_stats_update_end(&stats->syncp);
1215
1216         netif_rx(skb);
1217
1218         return;
1219 drop:
1220         if (tun_dst)
1221                 dst_release((struct dst_entry *)tun_dst);
1222
1223         /* Consume bad packet */
1224         kfree_skb(skb);
1225 }
1226
1227 /* Callback from net/ipv4/udp.c to receive packets */
1228 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1229 {
1230         struct metadata_dst *tun_dst = NULL;
1231         struct ip_tunnel_info *info;
1232         struct vxlan_sock *vs;
1233         struct vxlanhdr *vxh;
1234         u32 flags, vni;
1235         struct vxlan_metadata _md;
1236         struct vxlan_metadata *md = &_md;
1237
1238         /* Need Vxlan and inner Ethernet header to be present */
1239         if (!pskb_may_pull(skb, VXLAN_HLEN))
1240                 goto error;
1241
1242         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1243         flags = ntohl(vxh->vx_flags);
1244         vni = ntohl(vxh->vx_vni);
1245
1246         if (flags & VXLAN_HF_VNI) {
1247                 flags &= ~VXLAN_HF_VNI;
1248         } else {
1249                 /* VNI flag always required to be set */
1250                 goto bad_flags;
1251         }
1252
1253         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1254                 goto drop;
1255         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1256
1257         vs = rcu_dereference_sk_user_data(sk);
1258         if (!vs)
1259                 goto drop;
1260
1261         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1262                 vxh = vxlan_remcsum(skb, vxh, sizeof(struct vxlanhdr), vni,
1263                                     !!(vs->flags & VXLAN_F_REMCSUM_NOPARTIAL));
1264                 if (!vxh)
1265                         goto drop;
1266
1267                 flags &= ~VXLAN_HF_RCO;
1268                 vni &= VXLAN_VNI_MASK;
1269         }
1270
1271         if (vxlan_collect_metadata(vs)) {
1272                 tun_dst = metadata_dst_alloc(sizeof(*md), GFP_ATOMIC);
1273                 if (!tun_dst)
1274                         goto drop;
1275
1276                 info = &tun_dst->u.tun_info;
1277                 if (vxlan_get_sk_family(vs) == AF_INET) {
1278                         const struct iphdr *iph = ip_hdr(skb);
1279
1280                         info->key.u.ipv4.src = iph->saddr;
1281                         info->key.u.ipv4.dst = iph->daddr;
1282                         info->key.tos = iph->tos;
1283                         info->key.ttl = iph->ttl;
1284                 } else {
1285                         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
1286
1287                         info->key.u.ipv6.src = ip6h->saddr;
1288                         info->key.u.ipv6.dst = ip6h->daddr;
1289                         info->key.tos = ipv6_get_dsfield(ip6h);
1290                         info->key.ttl = ip6h->hop_limit;
1291                 }
1292
1293                 info->key.tp_src = udp_hdr(skb)->source;
1294                 info->key.tp_dst = udp_hdr(skb)->dest;
1295
1296                 info->mode = IP_TUNNEL_INFO_RX;
1297                 info->key.tun_flags = TUNNEL_KEY;
1298                 info->key.tun_id = cpu_to_be64(vni >> 8);
1299                 if (udp_hdr(skb)->check != 0)
1300                         info->key.tun_flags |= TUNNEL_CSUM;
1301
1302                 md = ip_tunnel_info_opts(info, sizeof(*md));
1303         } else {
1304                 memset(md, 0, sizeof(*md));
1305         }
1306
1307         /* For backwards compatibility, only allow reserved fields to be
1308          * used by VXLAN extensions if explicitly requested.
1309          */
1310         if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1311                 struct vxlanhdr_gbp *gbp;
1312
1313                 gbp = (struct vxlanhdr_gbp *)vxh;
1314                 md->gbp = ntohs(gbp->policy_id);
1315
1316                 if (tun_dst)
1317                         info->key.tun_flags |= TUNNEL_VXLAN_OPT;
1318
1319                 if (gbp->dont_learn)
1320                         md->gbp |= VXLAN_GBP_DONT_LEARN;
1321
1322                 if (gbp->policy_applied)
1323                         md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1324
1325                 flags &= ~VXLAN_GBP_USED_BITS;
1326         }
1327
1328         if (flags || vni & ~VXLAN_VNI_MASK) {
1329                 /* If there are any unprocessed flags remaining treat
1330                  * this as a malformed packet. This behavior diverges from
1331                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1332                  * in reserved fields are to be ignored. The approach here
1333                  * maintains compatibility with previous stack code, and also
1334                  * is more robust and provides a little more security in
1335                  * adding extensions to VXLAN.
1336                  */
1337
1338                 goto bad_flags;
1339         }
1340
1341         vxlan_rcv(vs, skb, md, vni >> 8, tun_dst);
1342         return 0;
1343
1344 drop:
1345         /* Consume bad packet */
1346         kfree_skb(skb);
1347         return 0;
1348
1349 bad_flags:
1350         netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1351                    ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1352
1353 error:
1354         if (tun_dst)
1355                 dst_release((struct dst_entry *)tun_dst);
1356
1357         /* Return non vxlan pkt */
1358         return 1;
1359 }
1360
1361 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1362 {
1363         struct vxlan_dev *vxlan = netdev_priv(dev);
1364         struct arphdr *parp;
1365         u8 *arpptr, *sha;
1366         __be32 sip, tip;
1367         struct neighbour *n;
1368
1369         if (dev->flags & IFF_NOARP)
1370                 goto out;
1371
1372         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1373                 dev->stats.tx_dropped++;
1374                 goto out;
1375         }
1376         parp = arp_hdr(skb);
1377
1378         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1379              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1380             parp->ar_pro != htons(ETH_P_IP) ||
1381             parp->ar_op != htons(ARPOP_REQUEST) ||
1382             parp->ar_hln != dev->addr_len ||
1383             parp->ar_pln != 4)
1384                 goto out;
1385         arpptr = (u8 *)parp + sizeof(struct arphdr);
1386         sha = arpptr;
1387         arpptr += dev->addr_len;        /* sha */
1388         memcpy(&sip, arpptr, sizeof(sip));
1389         arpptr += sizeof(sip);
1390         arpptr += dev->addr_len;        /* tha */
1391         memcpy(&tip, arpptr, sizeof(tip));
1392
1393         if (ipv4_is_loopback(tip) ||
1394             ipv4_is_multicast(tip))
1395                 goto out;
1396
1397         n = neigh_lookup(&arp_tbl, &tip, dev);
1398
1399         if (n) {
1400                 struct vxlan_fdb *f;
1401                 struct sk_buff  *reply;
1402
1403                 if (!(n->nud_state & NUD_CONNECTED)) {
1404                         neigh_release(n);
1405                         goto out;
1406                 }
1407
1408                 f = vxlan_find_mac(vxlan, n->ha);
1409                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1410                         /* bridge-local neighbor */
1411                         neigh_release(n);
1412                         goto out;
1413                 }
1414
1415                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1416                                 n->ha, sha);
1417
1418                 neigh_release(n);
1419
1420                 if (reply == NULL)
1421                         goto out;
1422
1423                 skb_reset_mac_header(reply);
1424                 __skb_pull(reply, skb_network_offset(reply));
1425                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1426                 reply->pkt_type = PACKET_HOST;
1427
1428                 if (netif_rx_ni(reply) == NET_RX_DROP)
1429                         dev->stats.rx_dropped++;
1430         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1431                 union vxlan_addr ipa = {
1432                         .sin.sin_addr.s_addr = tip,
1433                         .sin.sin_family = AF_INET,
1434                 };
1435
1436                 vxlan_ip_miss(dev, &ipa);
1437         }
1438 out:
1439         consume_skb(skb);
1440         return NETDEV_TX_OK;
1441 }
1442
1443 #if IS_ENABLED(CONFIG_IPV6)
1444 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1445         struct neighbour *n, bool isrouter)
1446 {
1447         struct net_device *dev = request->dev;
1448         struct sk_buff *reply;
1449         struct nd_msg *ns, *na;
1450         struct ipv6hdr *pip6;
1451         u8 *daddr;
1452         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1453         int ns_olen;
1454         int i, len;
1455
1456         if (dev == NULL)
1457                 return NULL;
1458
1459         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1460                 sizeof(*na) + na_olen + dev->needed_tailroom;
1461         reply = alloc_skb(len, GFP_ATOMIC);
1462         if (reply == NULL)
1463                 return NULL;
1464
1465         reply->protocol = htons(ETH_P_IPV6);
1466         reply->dev = dev;
1467         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1468         skb_push(reply, sizeof(struct ethhdr));
1469         skb_set_mac_header(reply, 0);
1470
1471         ns = (struct nd_msg *)skb_transport_header(request);
1472
1473         daddr = eth_hdr(request)->h_source;
1474         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1475         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1476                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1477                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1478                         break;
1479                 }
1480         }
1481
1482         /* Ethernet header */
1483         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1484         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1485         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1486         reply->protocol = htons(ETH_P_IPV6);
1487
1488         skb_pull(reply, sizeof(struct ethhdr));
1489         skb_set_network_header(reply, 0);
1490         skb_put(reply, sizeof(struct ipv6hdr));
1491
1492         /* IPv6 header */
1493
1494         pip6 = ipv6_hdr(reply);
1495         memset(pip6, 0, sizeof(struct ipv6hdr));
1496         pip6->version = 6;
1497         pip6->priority = ipv6_hdr(request)->priority;
1498         pip6->nexthdr = IPPROTO_ICMPV6;
1499         pip6->hop_limit = 255;
1500         pip6->daddr = ipv6_hdr(request)->saddr;
1501         pip6->saddr = *(struct in6_addr *)n->primary_key;
1502
1503         skb_pull(reply, sizeof(struct ipv6hdr));
1504         skb_set_transport_header(reply, 0);
1505
1506         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1507
1508         /* Neighbor Advertisement */
1509         memset(na, 0, sizeof(*na)+na_olen);
1510         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1511         na->icmph.icmp6_router = isrouter;
1512         na->icmph.icmp6_override = 1;
1513         na->icmph.icmp6_solicited = 1;
1514         na->target = ns->target;
1515         ether_addr_copy(&na->opt[2], n->ha);
1516         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1517         na->opt[1] = na_olen >> 3;
1518
1519         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1520                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1521                 csum_partial(na, sizeof(*na)+na_olen, 0));
1522
1523         pip6->payload_len = htons(sizeof(*na)+na_olen);
1524
1525         skb_push(reply, sizeof(struct ipv6hdr));
1526
1527         reply->ip_summed = CHECKSUM_UNNECESSARY;
1528
1529         return reply;
1530 }
1531
1532 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1533 {
1534         struct vxlan_dev *vxlan = netdev_priv(dev);
1535         struct nd_msg *msg;
1536         const struct ipv6hdr *iphdr;
1537         const struct in6_addr *saddr, *daddr;
1538         struct neighbour *n;
1539         struct inet6_dev *in6_dev;
1540
1541         in6_dev = __in6_dev_get(dev);
1542         if (!in6_dev)
1543                 goto out;
1544
1545         iphdr = ipv6_hdr(skb);
1546         saddr = &iphdr->saddr;
1547         daddr = &iphdr->daddr;
1548
1549         msg = (struct nd_msg *)skb_transport_header(skb);
1550         if (msg->icmph.icmp6_code != 0 ||
1551             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1552                 goto out;
1553
1554         if (ipv6_addr_loopback(daddr) ||
1555             ipv6_addr_is_multicast(&msg->target))
1556                 goto out;
1557
1558         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1559
1560         if (n) {
1561                 struct vxlan_fdb *f;
1562                 struct sk_buff *reply;
1563
1564                 if (!(n->nud_state & NUD_CONNECTED)) {
1565                         neigh_release(n);
1566                         goto out;
1567                 }
1568
1569                 f = vxlan_find_mac(vxlan, n->ha);
1570                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1571                         /* bridge-local neighbor */
1572                         neigh_release(n);
1573                         goto out;
1574                 }
1575
1576                 reply = vxlan_na_create(skb, n,
1577                                         !!(f ? f->flags & NTF_ROUTER : 0));
1578
1579                 neigh_release(n);
1580
1581                 if (reply == NULL)
1582                         goto out;
1583
1584                 if (netif_rx_ni(reply) == NET_RX_DROP)
1585                         dev->stats.rx_dropped++;
1586
1587         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1588                 union vxlan_addr ipa = {
1589                         .sin6.sin6_addr = msg->target,
1590                         .sin6.sin6_family = AF_INET6,
1591                 };
1592
1593                 vxlan_ip_miss(dev, &ipa);
1594         }
1595
1596 out:
1597         consume_skb(skb);
1598         return NETDEV_TX_OK;
1599 }
1600 #endif
1601
1602 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1603 {
1604         struct vxlan_dev *vxlan = netdev_priv(dev);
1605         struct neighbour *n;
1606
1607         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1608                 return false;
1609
1610         n = NULL;
1611         switch (ntohs(eth_hdr(skb)->h_proto)) {
1612         case ETH_P_IP:
1613         {
1614                 struct iphdr *pip;
1615
1616                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1617                         return false;
1618                 pip = ip_hdr(skb);
1619                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1620                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1621                         union vxlan_addr ipa = {
1622                                 .sin.sin_addr.s_addr = pip->daddr,
1623                                 .sin.sin_family = AF_INET,
1624                         };
1625
1626                         vxlan_ip_miss(dev, &ipa);
1627                         return false;
1628                 }
1629
1630                 break;
1631         }
1632 #if IS_ENABLED(CONFIG_IPV6)
1633         case ETH_P_IPV6:
1634         {
1635                 struct ipv6hdr *pip6;
1636
1637                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1638                         return false;
1639                 pip6 = ipv6_hdr(skb);
1640                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1641                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1642                         union vxlan_addr ipa = {
1643                                 .sin6.sin6_addr = pip6->daddr,
1644                                 .sin6.sin6_family = AF_INET6,
1645                         };
1646
1647                         vxlan_ip_miss(dev, &ipa);
1648                         return false;
1649                 }
1650
1651                 break;
1652         }
1653 #endif
1654         default:
1655                 return false;
1656         }
1657
1658         if (n) {
1659                 bool diff;
1660
1661                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1662                 if (diff) {
1663                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1664                                 dev->addr_len);
1665                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1666                 }
1667                 neigh_release(n);
1668                 return diff;
1669         }
1670
1671         return false;
1672 }
1673
1674 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1675                                 struct vxlan_metadata *md)
1676 {
1677         struct vxlanhdr_gbp *gbp;
1678
1679         if (!md->gbp)
1680                 return;
1681
1682         gbp = (struct vxlanhdr_gbp *)vxh;
1683         vxh->vx_flags |= htonl(VXLAN_HF_GBP);
1684
1685         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1686                 gbp->dont_learn = 1;
1687
1688         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1689                 gbp->policy_applied = 1;
1690
1691         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1692 }
1693
1694 #if IS_ENABLED(CONFIG_IPV6)
1695 static int vxlan6_xmit_skb(struct dst_entry *dst, struct sock *sk,
1696                            struct sk_buff *skb,
1697                            struct net_device *dev, struct in6_addr *saddr,
1698                            struct in6_addr *daddr, __u8 prio, __u8 ttl,
1699                            __be16 src_port, __be16 dst_port, __be32 vni,
1700                            struct vxlan_metadata *md, bool xnet, u32 vxflags)
1701 {
1702         struct vxlanhdr *vxh;
1703         int min_headroom;
1704         int err;
1705         bool udp_sum = !(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX);
1706         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1707         u16 hdrlen = sizeof(struct vxlanhdr);
1708
1709         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1710             skb->ip_summed == CHECKSUM_PARTIAL) {
1711                 int csum_start = skb_checksum_start_offset(skb);
1712
1713                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1714                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1715                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1716                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1717                         udp_sum = false;
1718                         type |= SKB_GSO_TUNNEL_REMCSUM;
1719                 }
1720         }
1721
1722         skb_scrub_packet(skb, xnet);
1723
1724         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1725                         + VXLAN_HLEN + sizeof(struct ipv6hdr)
1726                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1727
1728         /* Need space for new headers (invalidates iph ptr) */
1729         err = skb_cow_head(skb, min_headroom);
1730         if (unlikely(err)) {
1731                 kfree_skb(skb);
1732                 goto err;
1733         }
1734
1735         skb = vlan_hwaccel_push_inside(skb);
1736         if (WARN_ON(!skb)) {
1737                 err = -ENOMEM;
1738                 goto err;
1739         }
1740
1741         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1742         if (IS_ERR(skb)) {
1743                 err = -EINVAL;
1744                 goto err;
1745         }
1746
1747         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1748         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1749         vxh->vx_vni = vni;
1750
1751         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1752                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1753                            VXLAN_RCO_SHIFT;
1754
1755                 if (skb->csum_offset == offsetof(struct udphdr, check))
1756                         data |= VXLAN_RCO_UDP;
1757
1758                 vxh->vx_vni |= htonl(data);
1759                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1760
1761                 if (!skb_is_gso(skb)) {
1762                         skb->ip_summed = CHECKSUM_NONE;
1763                         skb->encapsulation = 0;
1764                 }
1765         }
1766
1767         if (vxflags & VXLAN_F_GBP)
1768                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1769
1770         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1771
1772         udp_tunnel6_xmit_skb(dst, sk, skb, dev, saddr, daddr, prio,
1773                              ttl, src_port, dst_port,
1774                              !!(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX));
1775         return 0;
1776 err:
1777         dst_release(dst);
1778         return err;
1779 }
1780 #endif
1781
1782 static int vxlan_xmit_skb(struct rtable *rt, struct sock *sk, struct sk_buff *skb,
1783                           __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1784                           __be16 src_port, __be16 dst_port, __be32 vni,
1785                           struct vxlan_metadata *md, bool xnet, u32 vxflags)
1786 {
1787         struct vxlanhdr *vxh;
1788         int min_headroom;
1789         int err;
1790         bool udp_sum = !!(vxflags & VXLAN_F_UDP_CSUM);
1791         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1792         u16 hdrlen = sizeof(struct vxlanhdr);
1793
1794         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1795             skb->ip_summed == CHECKSUM_PARTIAL) {
1796                 int csum_start = skb_checksum_start_offset(skb);
1797
1798                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1799                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1800                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1801                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1802                         udp_sum = false;
1803                         type |= SKB_GSO_TUNNEL_REMCSUM;
1804                 }
1805         }
1806
1807         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1808                         + VXLAN_HLEN + sizeof(struct iphdr)
1809                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1810
1811         /* Need space for new headers (invalidates iph ptr) */
1812         err = skb_cow_head(skb, min_headroom);
1813         if (unlikely(err)) {
1814                 kfree_skb(skb);
1815                 return err;
1816         }
1817
1818         skb = vlan_hwaccel_push_inside(skb);
1819         if (WARN_ON(!skb))
1820                 return -ENOMEM;
1821
1822         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1823         if (IS_ERR(skb))
1824                 return PTR_ERR(skb);
1825
1826         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1827         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1828         vxh->vx_vni = vni;
1829
1830         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1831                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1832                            VXLAN_RCO_SHIFT;
1833
1834                 if (skb->csum_offset == offsetof(struct udphdr, check))
1835                         data |= VXLAN_RCO_UDP;
1836
1837                 vxh->vx_vni |= htonl(data);
1838                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1839
1840                 if (!skb_is_gso(skb)) {
1841                         skb->ip_summed = CHECKSUM_NONE;
1842                         skb->encapsulation = 0;
1843                 }
1844         }
1845
1846         if (vxflags & VXLAN_F_GBP)
1847                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1848
1849         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1850
1851         return udp_tunnel_xmit_skb(rt, sk, skb, src, dst, tos,
1852                                    ttl, df, src_port, dst_port, xnet,
1853                                    !(vxflags & VXLAN_F_UDP_CSUM));
1854 }
1855
1856 /* Bypass encapsulation if the destination is local */
1857 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1858                                struct vxlan_dev *dst_vxlan)
1859 {
1860         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1861         union vxlan_addr loopback;
1862         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1863         struct net_device *dev = skb->dev;
1864         int len = skb->len;
1865
1866         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1867         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1868         skb->pkt_type = PACKET_HOST;
1869         skb->encapsulation = 0;
1870         skb->dev = dst_vxlan->dev;
1871         __skb_pull(skb, skb_network_offset(skb));
1872
1873         if (remote_ip->sa.sa_family == AF_INET) {
1874                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1875                 loopback.sa.sa_family =  AF_INET;
1876 #if IS_ENABLED(CONFIG_IPV6)
1877         } else {
1878                 loopback.sin6.sin6_addr = in6addr_loopback;
1879                 loopback.sa.sa_family =  AF_INET6;
1880 #endif
1881         }
1882
1883         if (dst_vxlan->flags & VXLAN_F_LEARN)
1884                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1885
1886         u64_stats_update_begin(&tx_stats->syncp);
1887         tx_stats->tx_packets++;
1888         tx_stats->tx_bytes += len;
1889         u64_stats_update_end(&tx_stats->syncp);
1890
1891         if (netif_rx(skb) == NET_RX_SUCCESS) {
1892                 u64_stats_update_begin(&rx_stats->syncp);
1893                 rx_stats->rx_packets++;
1894                 rx_stats->rx_bytes += len;
1895                 u64_stats_update_end(&rx_stats->syncp);
1896         } else {
1897                 dev->stats.rx_dropped++;
1898         }
1899 }
1900
1901 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1902                            struct vxlan_rdst *rdst, bool did_rsc)
1903 {
1904         struct ip_tunnel_info *info;
1905         struct vxlan_dev *vxlan = netdev_priv(dev);
1906         struct sock *sk = vxlan->vn_sock->sock->sk;
1907         unsigned short family = vxlan_get_sk_family(vxlan->vn_sock);
1908         struct rtable *rt = NULL;
1909         const struct iphdr *old_iph;
1910         struct flowi4 fl4;
1911         union vxlan_addr *dst;
1912         union vxlan_addr remote_ip;
1913         struct vxlan_metadata _md;
1914         struct vxlan_metadata *md = &_md;
1915         __be16 src_port = 0, dst_port;
1916         u32 vni;
1917         __be16 df = 0;
1918         __u8 tos, ttl;
1919         int err;
1920         u32 flags = vxlan->flags;
1921
1922         info = skb_tunnel_info(skb);
1923
1924         if (rdst) {
1925                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1926                 vni = rdst->remote_vni;
1927                 dst = &rdst->remote_ip;
1928         } else {
1929                 if (!info) {
1930                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1931                                   dev->name);
1932                         goto drop;
1933                 }
1934
1935                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1936                 vni = be64_to_cpu(info->key.tun_id);
1937                 remote_ip.sa.sa_family = family;
1938                 if (family == AF_INET)
1939                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1940                 else
1941                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1942                 dst = &remote_ip;
1943         }
1944
1945         if (vxlan_addr_any(dst)) {
1946                 if (did_rsc) {
1947                         /* short-circuited back to local bridge */
1948                         vxlan_encap_bypass(skb, vxlan, vxlan);
1949                         return;
1950                 }
1951                 goto drop;
1952         }
1953
1954         old_iph = ip_hdr(skb);
1955
1956         ttl = vxlan->cfg.ttl;
1957         if (!ttl && vxlan_addr_multicast(dst))
1958                 ttl = 1;
1959
1960         tos = vxlan->cfg.tos;
1961         if (tos == 1)
1962                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1963
1964         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
1965                                      vxlan->cfg.port_max, true);
1966
1967         if (info) {
1968                 if (info->key.tun_flags & TUNNEL_CSUM)
1969                         flags |= VXLAN_F_UDP_CSUM;
1970                 else
1971                         flags &= ~VXLAN_F_UDP_CSUM;
1972
1973                 ttl = info->key.ttl;
1974                 tos = info->key.tos;
1975
1976                 if (info->options_len)
1977                         md = ip_tunnel_info_opts(info, sizeof(*md));
1978         } else {
1979                 md->gbp = skb->mark;
1980         }
1981
1982         if (dst->sa.sa_family == AF_INET) {
1983                 if (info && (info->key.tun_flags & TUNNEL_DONT_FRAGMENT))
1984                         df = htons(IP_DF);
1985
1986                 memset(&fl4, 0, sizeof(fl4));
1987                 fl4.flowi4_oif = rdst ? rdst->remote_ifindex : 0;
1988                 fl4.flowi4_tos = RT_TOS(tos);
1989                 fl4.flowi4_mark = skb->mark;
1990                 fl4.flowi4_proto = IPPROTO_UDP;
1991                 fl4.daddr = dst->sin.sin_addr.s_addr;
1992                 fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
1993
1994                 rt = ip_route_output_key(vxlan->net, &fl4);
1995                 if (IS_ERR(rt)) {
1996                         netdev_dbg(dev, "no route to %pI4\n",
1997                                    &dst->sin.sin_addr.s_addr);
1998                         dev->stats.tx_carrier_errors++;
1999                         goto tx_error;
2000                 }
2001
2002                 if (rt->dst.dev == dev) {
2003                         netdev_dbg(dev, "circular route to %pI4\n",
2004                                    &dst->sin.sin_addr.s_addr);
2005                         dev->stats.collisions++;
2006                         goto rt_tx_error;
2007                 }
2008
2009                 /* Bypass encapsulation if the destination is local */
2010                 if (rt->rt_flags & RTCF_LOCAL &&
2011                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2012                         struct vxlan_dev *dst_vxlan;
2013
2014                         ip_rt_put(rt);
2015                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2016                                                    dst->sa.sa_family, dst_port,
2017                                                    vxlan->flags);
2018                         if (!dst_vxlan)
2019                                 goto tx_error;
2020                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2021                         return;
2022                 }
2023
2024                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2025                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2026                 err = vxlan_xmit_skb(rt, sk, skb, fl4.saddr,
2027                                      dst->sin.sin_addr.s_addr, tos, ttl, df,
2028                                      src_port, dst_port, htonl(vni << 8), md,
2029                                      !net_eq(vxlan->net, dev_net(vxlan->dev)),
2030                                      flags);
2031                 if (err < 0) {
2032                         /* skb is already freed. */
2033                         skb = NULL;
2034                         goto rt_tx_error;
2035                 }
2036
2037                 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
2038 #if IS_ENABLED(CONFIG_IPV6)
2039         } else {
2040                 struct dst_entry *ndst;
2041                 struct flowi6 fl6;
2042                 u32 rt6i_flags;
2043
2044                 memset(&fl6, 0, sizeof(fl6));
2045                 fl6.flowi6_oif = rdst ? rdst->remote_ifindex : 0;
2046                 fl6.daddr = dst->sin6.sin6_addr;
2047                 fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
2048                 fl6.flowi6_mark = skb->mark;
2049                 fl6.flowi6_proto = IPPROTO_UDP;
2050
2051                 if (ipv6_stub->ipv6_dst_lookup(vxlan->net, sk, &ndst, &fl6)) {
2052                         netdev_dbg(dev, "no route to %pI6\n",
2053                                    &dst->sin6.sin6_addr);
2054                         dev->stats.tx_carrier_errors++;
2055                         goto tx_error;
2056                 }
2057
2058                 if (ndst->dev == dev) {
2059                         netdev_dbg(dev, "circular route to %pI6\n",
2060                                    &dst->sin6.sin6_addr);
2061                         dst_release(ndst);
2062                         dev->stats.collisions++;
2063                         goto tx_error;
2064                 }
2065
2066                 /* Bypass encapsulation if the destination is local */
2067                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2068                 if (rt6i_flags & RTF_LOCAL &&
2069                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2070                         struct vxlan_dev *dst_vxlan;
2071
2072                         dst_release(ndst);
2073                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2074                                                    dst->sa.sa_family, dst_port,
2075                                                    vxlan->flags);
2076                         if (!dst_vxlan)
2077                                 goto tx_error;
2078                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2079                         return;
2080                 }
2081
2082                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2083                 err = vxlan6_xmit_skb(ndst, sk, skb, dev, &fl6.saddr, &fl6.daddr,
2084                                       0, ttl, src_port, dst_port, htonl(vni << 8), md,
2085                                       !net_eq(vxlan->net, dev_net(vxlan->dev)),
2086                                       flags);
2087 #endif
2088         }
2089
2090         return;
2091
2092 drop:
2093         dev->stats.tx_dropped++;
2094         goto tx_free;
2095
2096 rt_tx_error:
2097         ip_rt_put(rt);
2098 tx_error:
2099         dev->stats.tx_errors++;
2100 tx_free:
2101         dev_kfree_skb(skb);
2102 }
2103
2104 /* Transmit local packets over Vxlan
2105  *
2106  * Outer IP header inherits ECN and DF from inner header.
2107  * Outer UDP destination is the VXLAN assigned port.
2108  *           source port is based on hash of flow
2109  */
2110 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2111 {
2112         struct vxlan_dev *vxlan = netdev_priv(dev);
2113         const struct ip_tunnel_info *info;
2114         struct ethhdr *eth;
2115         bool did_rsc = false;
2116         struct vxlan_rdst *rdst, *fdst = NULL;
2117         struct vxlan_fdb *f;
2118
2119         info = skb_tunnel_info(skb);
2120
2121         skb_reset_mac_header(skb);
2122         eth = eth_hdr(skb);
2123
2124         if ((vxlan->flags & VXLAN_F_PROXY)) {
2125                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2126                         return arp_reduce(dev, skb);
2127 #if IS_ENABLED(CONFIG_IPV6)
2128                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2129                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2130                                        + sizeof(struct nd_msg)) &&
2131                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2132                                 struct nd_msg *msg;
2133
2134                                 msg = (struct nd_msg *)skb_transport_header(skb);
2135                                 if (msg->icmph.icmp6_code == 0 &&
2136                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2137                                         return neigh_reduce(dev, skb);
2138                 }
2139                 eth = eth_hdr(skb);
2140 #endif
2141         }
2142
2143         if (vxlan->flags & VXLAN_F_COLLECT_METADATA &&
2144             info && info->mode == IP_TUNNEL_INFO_TX) {
2145                 vxlan_xmit_one(skb, dev, NULL, false);
2146                 return NETDEV_TX_OK;
2147         }
2148
2149         f = vxlan_find_mac(vxlan, eth->h_dest);
2150         did_rsc = false;
2151
2152         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2153             (ntohs(eth->h_proto) == ETH_P_IP ||
2154              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2155                 did_rsc = route_shortcircuit(dev, skb);
2156                 if (did_rsc)
2157                         f = vxlan_find_mac(vxlan, eth->h_dest);
2158         }
2159
2160         if (f == NULL) {
2161                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2162                 if (f == NULL) {
2163                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2164                             !is_multicast_ether_addr(eth->h_dest))
2165                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2166
2167                         dev->stats.tx_dropped++;
2168                         kfree_skb(skb);
2169                         return NETDEV_TX_OK;
2170                 }
2171         }
2172
2173         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2174                 struct sk_buff *skb1;
2175
2176                 if (!fdst) {
2177                         fdst = rdst;
2178                         continue;
2179                 }
2180                 skb1 = skb_clone(skb, GFP_ATOMIC);
2181                 if (skb1)
2182                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2183         }
2184
2185         if (fdst)
2186                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2187         else
2188                 kfree_skb(skb);
2189         return NETDEV_TX_OK;
2190 }
2191
2192 /* Walk the forwarding table and purge stale entries */
2193 static void vxlan_cleanup(unsigned long arg)
2194 {
2195         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2196         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2197         unsigned int h;
2198
2199         if (!netif_running(vxlan->dev))
2200                 return;
2201
2202         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2203                 struct hlist_node *p, *n;
2204
2205                 spin_lock_bh(&vxlan->hash_lock);
2206                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2207                         struct vxlan_fdb *f
2208                                 = container_of(p, struct vxlan_fdb, hlist);
2209                         unsigned long timeout;
2210
2211                         if (f->state & NUD_PERMANENT)
2212                                 continue;
2213
2214                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2215                         if (time_before_eq(timeout, jiffies)) {
2216                                 netdev_dbg(vxlan->dev,
2217                                            "garbage collect %pM\n",
2218                                            f->eth_addr);
2219                                 f->state = NUD_STALE;
2220                                 vxlan_fdb_destroy(vxlan, f);
2221                         } else if (time_before(timeout, next_timer))
2222                                 next_timer = timeout;
2223                 }
2224                 spin_unlock_bh(&vxlan->hash_lock);
2225         }
2226
2227         mod_timer(&vxlan->age_timer, next_timer);
2228 }
2229
2230 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2231 {
2232         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2233         __u32 vni = vxlan->default_dst.remote_vni;
2234
2235         vxlan->vn_sock = vs;
2236         spin_lock(&vn->sock_lock);
2237         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2238         spin_unlock(&vn->sock_lock);
2239 }
2240
2241 /* Setup stats when device is created */
2242 static int vxlan_init(struct net_device *dev)
2243 {
2244         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2245         if (!dev->tstats)
2246                 return -ENOMEM;
2247
2248         return 0;
2249 }
2250
2251 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2252 {
2253         struct vxlan_fdb *f;
2254
2255         spin_lock_bh(&vxlan->hash_lock);
2256         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2257         if (f)
2258                 vxlan_fdb_destroy(vxlan, f);
2259         spin_unlock_bh(&vxlan->hash_lock);
2260 }
2261
2262 static void vxlan_uninit(struct net_device *dev)
2263 {
2264         struct vxlan_dev *vxlan = netdev_priv(dev);
2265
2266         vxlan_fdb_delete_default(vxlan);
2267
2268         free_percpu(dev->tstats);
2269 }
2270
2271 /* Start ageing timer and join group when device is brought up */
2272 static int vxlan_open(struct net_device *dev)
2273 {
2274         struct vxlan_dev *vxlan = netdev_priv(dev);
2275         struct vxlan_sock *vs;
2276         int ret = 0;
2277
2278         vs = vxlan_sock_add(vxlan->net, vxlan->cfg.dst_port,
2279                             vxlan->cfg.no_share, vxlan->flags);
2280         if (IS_ERR(vs))
2281                 return PTR_ERR(vs);
2282
2283         vxlan_vs_add_dev(vs, vxlan);
2284
2285         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2286                 ret = vxlan_igmp_join(vxlan);
2287                 if (ret) {
2288                         vxlan_sock_release(vs);
2289                         return ret;
2290                 }
2291         }
2292
2293         if (vxlan->cfg.age_interval)
2294                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2295
2296         return ret;
2297 }
2298
2299 /* Purge the forwarding table */
2300 static void vxlan_flush(struct vxlan_dev *vxlan)
2301 {
2302         unsigned int h;
2303
2304         spin_lock_bh(&vxlan->hash_lock);
2305         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2306                 struct hlist_node *p, *n;
2307                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2308                         struct vxlan_fdb *f
2309                                 = container_of(p, struct vxlan_fdb, hlist);
2310                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2311                         if (!is_zero_ether_addr(f->eth_addr))
2312                                 vxlan_fdb_destroy(vxlan, f);
2313                 }
2314         }
2315         spin_unlock_bh(&vxlan->hash_lock);
2316 }
2317
2318 /* Cleanup timer and forwarding table on shutdown */
2319 static int vxlan_stop(struct net_device *dev)
2320 {
2321         struct vxlan_dev *vxlan = netdev_priv(dev);
2322         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2323         struct vxlan_sock *vs = vxlan->vn_sock;
2324         int ret = 0;
2325
2326         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2327             !vxlan_group_used(vn, vxlan))
2328                 ret = vxlan_igmp_leave(vxlan);
2329
2330         del_timer_sync(&vxlan->age_timer);
2331
2332         vxlan_flush(vxlan);
2333         vxlan_sock_release(vs);
2334
2335         return ret;
2336 }
2337
2338 /* Stub, nothing needs to be done. */
2339 static void vxlan_set_multicast_list(struct net_device *dev)
2340 {
2341 }
2342
2343 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2344 {
2345         struct vxlan_dev *vxlan = netdev_priv(dev);
2346         struct vxlan_rdst *dst = &vxlan->default_dst;
2347         struct net_device *lowerdev;
2348         int max_mtu;
2349
2350         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2351         if (lowerdev == NULL)
2352                 return eth_change_mtu(dev, new_mtu);
2353
2354         if (dst->remote_ip.sa.sa_family == AF_INET6)
2355                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2356         else
2357                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2358
2359         if (new_mtu < 68 || new_mtu > max_mtu)
2360                 return -EINVAL;
2361
2362         dev->mtu = new_mtu;
2363         return 0;
2364 }
2365
2366 static const struct net_device_ops vxlan_netdev_ops = {
2367         .ndo_init               = vxlan_init,
2368         .ndo_uninit             = vxlan_uninit,
2369         .ndo_open               = vxlan_open,
2370         .ndo_stop               = vxlan_stop,
2371         .ndo_start_xmit         = vxlan_xmit,
2372         .ndo_get_stats64        = ip_tunnel_get_stats64,
2373         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2374         .ndo_change_mtu         = vxlan_change_mtu,
2375         .ndo_validate_addr      = eth_validate_addr,
2376         .ndo_set_mac_address    = eth_mac_addr,
2377         .ndo_fdb_add            = vxlan_fdb_add,
2378         .ndo_fdb_del            = vxlan_fdb_delete,
2379         .ndo_fdb_dump           = vxlan_fdb_dump,
2380 };
2381
2382 /* Info for udev, that this is a virtual tunnel endpoint */
2383 static struct device_type vxlan_type = {
2384         .name = "vxlan",
2385 };
2386
2387 /* Calls the ndo_add_vxlan_port of the caller in order to
2388  * supply the listening VXLAN udp ports. Callers are expected
2389  * to implement the ndo_add_vxlan_port.
2390  */
2391 void vxlan_get_rx_port(struct net_device *dev)
2392 {
2393         struct vxlan_sock *vs;
2394         struct net *net = dev_net(dev);
2395         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2396         sa_family_t sa_family;
2397         __be16 port;
2398         unsigned int i;
2399
2400         spin_lock(&vn->sock_lock);
2401         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2402                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2403                         port = inet_sk(vs->sock->sk)->inet_sport;
2404                         sa_family = vxlan_get_sk_family(vs);
2405                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2406                                                             port);
2407                 }
2408         }
2409         spin_unlock(&vn->sock_lock);
2410 }
2411 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2412
2413 /* Initialize the device structure. */
2414 static void vxlan_setup(struct net_device *dev)
2415 {
2416         struct vxlan_dev *vxlan = netdev_priv(dev);
2417         unsigned int h;
2418
2419         eth_hw_addr_random(dev);
2420         ether_setup(dev);
2421         if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2422                 dev->needed_headroom = ETH_HLEN + VXLAN6_HEADROOM;
2423         else
2424                 dev->needed_headroom = ETH_HLEN + VXLAN_HEADROOM;
2425
2426         dev->netdev_ops = &vxlan_netdev_ops;
2427         dev->destructor = free_netdev;
2428         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2429
2430         dev->features   |= NETIF_F_LLTX;
2431         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2432         dev->features   |= NETIF_F_RXCSUM;
2433         dev->features   |= NETIF_F_GSO_SOFTWARE;
2434
2435         dev->vlan_features = dev->features;
2436         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2437         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2438         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2439         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2440         netif_keep_dst(dev);
2441         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
2442
2443         INIT_LIST_HEAD(&vxlan->next);
2444         spin_lock_init(&vxlan->hash_lock);
2445
2446         init_timer_deferrable(&vxlan->age_timer);
2447         vxlan->age_timer.function = vxlan_cleanup;
2448         vxlan->age_timer.data = (unsigned long) vxlan;
2449
2450         vxlan->cfg.dst_port = htons(vxlan_port);
2451
2452         vxlan->dev = dev;
2453
2454         for (h = 0; h < FDB_HASH_SIZE; ++h)
2455                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2456 }
2457
2458 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2459         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2460         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2461         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2462         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2463         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2464         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2465         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2466         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2467         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2468         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2469         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2470         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2471         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2472         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2473         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2474         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2475         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2476         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2477         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2478         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2479         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2480         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2481         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2482         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2483         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2484 };
2485
2486 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2487 {
2488         if (tb[IFLA_ADDRESS]) {
2489                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2490                         pr_debug("invalid link address (not ethernet)\n");
2491                         return -EINVAL;
2492                 }
2493
2494                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2495                         pr_debug("invalid all zero ethernet address\n");
2496                         return -EADDRNOTAVAIL;
2497                 }
2498         }
2499
2500         if (!data)
2501                 return -EINVAL;
2502
2503         if (data[IFLA_VXLAN_ID]) {
2504                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2505                 if (id >= VXLAN_VID_MASK)
2506                         return -ERANGE;
2507         }
2508
2509         if (data[IFLA_VXLAN_PORT_RANGE]) {
2510                 const struct ifla_vxlan_port_range *p
2511                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2512
2513                 if (ntohs(p->high) < ntohs(p->low)) {
2514                         pr_debug("port range %u .. %u not valid\n",
2515                                  ntohs(p->low), ntohs(p->high));
2516                         return -EINVAL;
2517                 }
2518         }
2519
2520         return 0;
2521 }
2522
2523 static void vxlan_get_drvinfo(struct net_device *netdev,
2524                               struct ethtool_drvinfo *drvinfo)
2525 {
2526         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2527         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2528 }
2529
2530 static const struct ethtool_ops vxlan_ethtool_ops = {
2531         .get_drvinfo    = vxlan_get_drvinfo,
2532         .get_link       = ethtool_op_get_link,
2533 };
2534
2535 static void vxlan_del_work(struct work_struct *work)
2536 {
2537         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2538         udp_tunnel_sock_release(vs->sock);
2539         kfree_rcu(vs, rcu);
2540 }
2541
2542 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2543                                         __be16 port, u32 flags)
2544 {
2545         struct socket *sock;
2546         struct udp_port_cfg udp_conf;
2547         int err;
2548
2549         memset(&udp_conf, 0, sizeof(udp_conf));
2550
2551         if (ipv6) {
2552                 udp_conf.family = AF_INET6;
2553                 udp_conf.use_udp6_rx_checksums =
2554                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2555         } else {
2556                 udp_conf.family = AF_INET;
2557         }
2558
2559         udp_conf.local_udp_port = port;
2560
2561         /* Open UDP socket */
2562         err = udp_sock_create(net, &udp_conf, &sock);
2563         if (err < 0)
2564                 return ERR_PTR(err);
2565
2566         return sock;
2567 }
2568
2569 /* Create new listen socket if needed */
2570 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2571                                               u32 flags)
2572 {
2573         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2574         struct vxlan_sock *vs;
2575         struct socket *sock;
2576         unsigned int h;
2577         bool ipv6 = !!(flags & VXLAN_F_IPV6);
2578         struct udp_tunnel_sock_cfg tunnel_cfg;
2579
2580         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2581         if (!vs)
2582                 return ERR_PTR(-ENOMEM);
2583
2584         for (h = 0; h < VNI_HASH_SIZE; ++h)
2585                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2586
2587         INIT_WORK(&vs->del_work, vxlan_del_work);
2588
2589         sock = vxlan_create_sock(net, ipv6, port, flags);
2590         if (IS_ERR(sock)) {
2591                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2592                         PTR_ERR(sock));
2593                 kfree(vs);
2594                 return ERR_CAST(sock);
2595         }
2596
2597         vs->sock = sock;
2598         atomic_set(&vs->refcnt, 1);
2599         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2600
2601         /* Initialize the vxlan udp offloads structure */
2602         vs->udp_offloads.port = port;
2603         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2604         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2605
2606         spin_lock(&vn->sock_lock);
2607         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2608         vxlan_notify_add_rx_port(vs);
2609         spin_unlock(&vn->sock_lock);
2610
2611         /* Mark socket as an encapsulation socket. */
2612         tunnel_cfg.sk_user_data = vs;
2613         tunnel_cfg.encap_type = 1;
2614         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2615         tunnel_cfg.encap_destroy = NULL;
2616
2617         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2618
2619         return vs;
2620 }
2621
2622 static struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2623                                          bool no_share, u32 flags)
2624 {
2625         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2626         struct vxlan_sock *vs;
2627         bool ipv6 = flags & VXLAN_F_IPV6;
2628
2629         if (!no_share) {
2630                 spin_lock(&vn->sock_lock);
2631                 vs = vxlan_find_sock(net, ipv6 ? AF_INET6 : AF_INET, port,
2632                                      flags);
2633                 if (vs) {
2634                         if (!atomic_add_unless(&vs->refcnt, 1, 0))
2635                                 vs = ERR_PTR(-EBUSY);
2636                         spin_unlock(&vn->sock_lock);
2637                         return vs;
2638                 }
2639                 spin_unlock(&vn->sock_lock);
2640         }
2641
2642         return vxlan_socket_create(net, port, flags);
2643 }
2644
2645 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2646                                struct vxlan_config *conf)
2647 {
2648         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2649         struct vxlan_dev *vxlan = netdev_priv(dev);
2650         struct vxlan_rdst *dst = &vxlan->default_dst;
2651         int err;
2652         bool use_ipv6 = false;
2653         __be16 default_port = vxlan->cfg.dst_port;
2654
2655         vxlan->net = src_net;
2656
2657         dst->remote_vni = conf->vni;
2658
2659         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2660
2661         /* Unless IPv6 is explicitly requested, assume IPv4 */
2662         if (!dst->remote_ip.sa.sa_family)
2663                 dst->remote_ip.sa.sa_family = AF_INET;
2664
2665         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2666             vxlan->cfg.saddr.sa.sa_family == AF_INET6)
2667                 use_ipv6 = true;
2668
2669         if (conf->remote_ifindex) {
2670                 struct net_device *lowerdev
2671                          = __dev_get_by_index(src_net, conf->remote_ifindex);
2672
2673                 dst->remote_ifindex = conf->remote_ifindex;
2674
2675                 if (!lowerdev) {
2676                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2677                         return -ENODEV;
2678                 }
2679
2680 #if IS_ENABLED(CONFIG_IPV6)
2681                 if (use_ipv6) {
2682                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2683                         if (idev && idev->cnf.disable_ipv6) {
2684                                 pr_info("IPv6 is disabled via sysctl\n");
2685                                 return -EPERM;
2686                         }
2687                         vxlan->flags |= VXLAN_F_IPV6;
2688                 }
2689 #endif
2690
2691                 if (!conf->mtu)
2692                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2693
2694                 dev->needed_headroom = lowerdev->hard_header_len +
2695                                        (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2696         } else if (use_ipv6)
2697                 vxlan->flags |= VXLAN_F_IPV6;
2698
2699         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2700         if (!vxlan->cfg.dst_port)
2701                 vxlan->cfg.dst_port = default_port;
2702         vxlan->flags |= conf->flags;
2703
2704         if (!vxlan->cfg.age_interval)
2705                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2706
2707         if (vxlan_find_vni(src_net, conf->vni, use_ipv6 ? AF_INET6 : AF_INET,
2708                            vxlan->cfg.dst_port, vxlan->flags))
2709                 return -EEXIST;
2710
2711         dev->ethtool_ops = &vxlan_ethtool_ops;
2712
2713         /* create an fdb entry for a valid default destination */
2714         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2715                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2716                                        &vxlan->default_dst.remote_ip,
2717                                        NUD_REACHABLE|NUD_PERMANENT,
2718                                        NLM_F_EXCL|NLM_F_CREATE,
2719                                        vxlan->cfg.dst_port,
2720                                        vxlan->default_dst.remote_vni,
2721                                        vxlan->default_dst.remote_ifindex,
2722                                        NTF_SELF);
2723                 if (err)
2724                         return err;
2725         }
2726
2727         err = register_netdevice(dev);
2728         if (err) {
2729                 vxlan_fdb_delete_default(vxlan);
2730                 return err;
2731         }
2732
2733         list_add(&vxlan->next, &vn->vxlan_list);
2734
2735         return 0;
2736 }
2737
2738 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2739                                     u8 name_assign_type, struct vxlan_config *conf)
2740 {
2741         struct nlattr *tb[IFLA_MAX+1];
2742         struct net_device *dev;
2743         int err;
2744
2745         memset(&tb, 0, sizeof(tb));
2746
2747         dev = rtnl_create_link(net, name, name_assign_type,
2748                                &vxlan_link_ops, tb);
2749         if (IS_ERR(dev))
2750                 return dev;
2751
2752         err = vxlan_dev_configure(net, dev, conf);
2753         if (err < 0) {
2754                 free_netdev(dev);
2755                 return ERR_PTR(err);
2756         }
2757
2758         return dev;
2759 }
2760 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2761
2762 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2763                          struct nlattr *tb[], struct nlattr *data[])
2764 {
2765         struct vxlan_config conf;
2766         int err;
2767
2768         if (!data[IFLA_VXLAN_ID])
2769                 return -EINVAL;
2770
2771         memset(&conf, 0, sizeof(conf));
2772         conf.vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2773
2774         if (data[IFLA_VXLAN_GROUP]) {
2775                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2776         } else if (data[IFLA_VXLAN_GROUP6]) {
2777                 if (!IS_ENABLED(CONFIG_IPV6))
2778                         return -EPFNOSUPPORT;
2779
2780                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2781                 conf.remote_ip.sa.sa_family = AF_INET6;
2782         }
2783
2784         if (data[IFLA_VXLAN_LOCAL]) {
2785                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2786                 conf.saddr.sa.sa_family = AF_INET;
2787         } else if (data[IFLA_VXLAN_LOCAL6]) {
2788                 if (!IS_ENABLED(CONFIG_IPV6))
2789                         return -EPFNOSUPPORT;
2790
2791                 /* TODO: respect scope id */
2792                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2793                 conf.saddr.sa.sa_family = AF_INET6;
2794         }
2795
2796         if (data[IFLA_VXLAN_LINK])
2797                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2798
2799         if (data[IFLA_VXLAN_TOS])
2800                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2801
2802         if (data[IFLA_VXLAN_TTL])
2803                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2804
2805         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2806                 conf.flags |= VXLAN_F_LEARN;
2807
2808         if (data[IFLA_VXLAN_AGEING])
2809                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2810
2811         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2812                 conf.flags |= VXLAN_F_PROXY;
2813
2814         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2815                 conf.flags |= VXLAN_F_RSC;
2816
2817         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2818                 conf.flags |= VXLAN_F_L2MISS;
2819
2820         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2821                 conf.flags |= VXLAN_F_L3MISS;
2822
2823         if (data[IFLA_VXLAN_LIMIT])
2824                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2825
2826         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2827             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2828                 conf.flags |= VXLAN_F_COLLECT_METADATA;
2829
2830         if (data[IFLA_VXLAN_PORT_RANGE]) {
2831                 const struct ifla_vxlan_port_range *p
2832                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2833                 conf.port_min = ntohs(p->low);
2834                 conf.port_max = ntohs(p->high);
2835         }
2836
2837         if (data[IFLA_VXLAN_PORT])
2838                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2839
2840         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2841                 conf.flags |= VXLAN_F_UDP_CSUM;
2842
2843         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2844             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2845                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2846
2847         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2848             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2849                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2850
2851         if (data[IFLA_VXLAN_REMCSUM_TX] &&
2852             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2853                 conf.flags |= VXLAN_F_REMCSUM_TX;
2854
2855         if (data[IFLA_VXLAN_REMCSUM_RX] &&
2856             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2857                 conf.flags |= VXLAN_F_REMCSUM_RX;
2858
2859         if (data[IFLA_VXLAN_GBP])
2860                 conf.flags |= VXLAN_F_GBP;
2861
2862         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2863                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2864
2865         err = vxlan_dev_configure(src_net, dev, &conf);
2866         switch (err) {
2867         case -ENODEV:
2868                 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
2869                 break;
2870
2871         case -EPERM:
2872                 pr_info("IPv6 is disabled via sysctl\n");
2873                 break;
2874
2875         case -EEXIST:
2876                 pr_info("duplicate VNI %u\n", conf.vni);
2877                 break;
2878         }
2879
2880         return err;
2881 }
2882
2883 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2884 {
2885         struct vxlan_dev *vxlan = netdev_priv(dev);
2886         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2887
2888         spin_lock(&vn->sock_lock);
2889         if (!hlist_unhashed(&vxlan->hlist))
2890                 hlist_del_rcu(&vxlan->hlist);
2891         spin_unlock(&vn->sock_lock);
2892
2893         list_del(&vxlan->next);
2894         unregister_netdevice_queue(dev, head);
2895 }
2896
2897 static size_t vxlan_get_size(const struct net_device *dev)
2898 {
2899
2900         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2901                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2902                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2903                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2904                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2905                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2906                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2907                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2908                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2909                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2910                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2911                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
2912                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2913                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2914                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2915                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2916                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2917                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2918                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2919                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
2920                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
2921                 0;
2922 }
2923
2924 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2925 {
2926         const struct vxlan_dev *vxlan = netdev_priv(dev);
2927         const struct vxlan_rdst *dst = &vxlan->default_dst;
2928         struct ifla_vxlan_port_range ports = {
2929                 .low =  htons(vxlan->cfg.port_min),
2930                 .high = htons(vxlan->cfg.port_max),
2931         };
2932
2933         if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2934                 goto nla_put_failure;
2935
2936         if (!vxlan_addr_any(&dst->remote_ip)) {
2937                 if (dst->remote_ip.sa.sa_family == AF_INET) {
2938                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
2939                                             dst->remote_ip.sin.sin_addr.s_addr))
2940                                 goto nla_put_failure;
2941 #if IS_ENABLED(CONFIG_IPV6)
2942                 } else {
2943                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
2944                                              &dst->remote_ip.sin6.sin6_addr))
2945                                 goto nla_put_failure;
2946 #endif
2947                 }
2948         }
2949
2950         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2951                 goto nla_put_failure;
2952
2953         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
2954                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
2955                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
2956                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
2957                                 goto nla_put_failure;
2958 #if IS_ENABLED(CONFIG_IPV6)
2959                 } else {
2960                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
2961                                              &vxlan->cfg.saddr.sin6.sin6_addr))
2962                                 goto nla_put_failure;
2963 #endif
2964                 }
2965         }
2966
2967         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
2968             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
2969             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2970                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
2971             nla_put_u8(skb, IFLA_VXLAN_PROXY,
2972                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
2973             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2974             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2975                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2976             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2977                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2978             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
2979                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
2980             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
2981             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
2982             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
2983             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2984                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2985             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2986                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2987             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2988                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
2989             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
2990                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
2991             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
2992                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
2993                 goto nla_put_failure;
2994
2995         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2996                 goto nla_put_failure;
2997
2998         if (vxlan->flags & VXLAN_F_GBP &&
2999             nla_put_flag(skb, IFLA_VXLAN_GBP))
3000                 goto nla_put_failure;
3001
3002         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3003             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3004                 goto nla_put_failure;
3005
3006         return 0;
3007
3008 nla_put_failure:
3009         return -EMSGSIZE;
3010 }
3011
3012 static struct net *vxlan_get_link_net(const struct net_device *dev)
3013 {
3014         struct vxlan_dev *vxlan = netdev_priv(dev);
3015
3016         return vxlan->net;
3017 }
3018
3019 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3020         .kind           = "vxlan",
3021         .maxtype        = IFLA_VXLAN_MAX,
3022         .policy         = vxlan_policy,
3023         .priv_size      = sizeof(struct vxlan_dev),
3024         .setup          = vxlan_setup,
3025         .validate       = vxlan_validate,
3026         .newlink        = vxlan_newlink,
3027         .dellink        = vxlan_dellink,
3028         .get_size       = vxlan_get_size,
3029         .fill_info      = vxlan_fill_info,
3030         .get_link_net   = vxlan_get_link_net,
3031 };
3032
3033 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3034                                              struct net_device *dev)
3035 {
3036         struct vxlan_dev *vxlan, *next;
3037         LIST_HEAD(list_kill);
3038
3039         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3040                 struct vxlan_rdst *dst = &vxlan->default_dst;
3041
3042                 /* In case we created vxlan device with carrier
3043                  * and we loose the carrier due to module unload
3044                  * we also need to remove vxlan device. In other
3045                  * cases, it's not necessary and remote_ifindex
3046                  * is 0 here, so no matches.
3047                  */
3048                 if (dst->remote_ifindex == dev->ifindex)
3049                         vxlan_dellink(vxlan->dev, &list_kill);
3050         }
3051
3052         unregister_netdevice_many(&list_kill);
3053 }
3054
3055 static int vxlan_lowerdev_event(struct notifier_block *unused,
3056                                 unsigned long event, void *ptr)
3057 {
3058         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3059         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3060
3061         if (event == NETDEV_UNREGISTER)
3062                 vxlan_handle_lowerdev_unregister(vn, dev);
3063
3064         return NOTIFY_DONE;
3065 }
3066
3067 static struct notifier_block vxlan_notifier_block __read_mostly = {
3068         .notifier_call = vxlan_lowerdev_event,
3069 };
3070
3071 static __net_init int vxlan_init_net(struct net *net)
3072 {
3073         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3074         unsigned int h;
3075
3076         INIT_LIST_HEAD(&vn->vxlan_list);
3077         spin_lock_init(&vn->sock_lock);
3078
3079         for (h = 0; h < PORT_HASH_SIZE; ++h)
3080                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3081
3082         return 0;
3083 }
3084
3085 static void __net_exit vxlan_exit_net(struct net *net)
3086 {
3087         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3088         struct vxlan_dev *vxlan, *next;
3089         struct net_device *dev, *aux;
3090         LIST_HEAD(list);
3091
3092         rtnl_lock();
3093         for_each_netdev_safe(net, dev, aux)
3094                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3095                         unregister_netdevice_queue(dev, &list);
3096
3097         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3098                 /* If vxlan->dev is in the same netns, it has already been added
3099                  * to the list by the previous loop.
3100                  */
3101                 if (!net_eq(dev_net(vxlan->dev), net))
3102                         unregister_netdevice_queue(vxlan->dev, &list);
3103         }
3104
3105         unregister_netdevice_many(&list);
3106         rtnl_unlock();
3107 }
3108
3109 static struct pernet_operations vxlan_net_ops = {
3110         .init = vxlan_init_net,
3111         .exit = vxlan_exit_net,
3112         .id   = &vxlan_net_id,
3113         .size = sizeof(struct vxlan_net),
3114 };
3115
3116 static int __init vxlan_init_module(void)
3117 {
3118         int rc;
3119
3120         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3121         if (!vxlan_wq)
3122                 return -ENOMEM;
3123
3124         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3125
3126         rc = register_pernet_subsys(&vxlan_net_ops);
3127         if (rc)
3128                 goto out1;
3129
3130         rc = register_netdevice_notifier(&vxlan_notifier_block);
3131         if (rc)
3132                 goto out2;
3133
3134         rc = rtnl_link_register(&vxlan_link_ops);
3135         if (rc)
3136                 goto out3;
3137
3138         return 0;
3139 out3:
3140         unregister_netdevice_notifier(&vxlan_notifier_block);
3141 out2:
3142         unregister_pernet_subsys(&vxlan_net_ops);
3143 out1:
3144         destroy_workqueue(vxlan_wq);
3145         return rc;
3146 }
3147 late_initcall(vxlan_init_module);
3148
3149 static void __exit vxlan_cleanup_module(void)
3150 {
3151         rtnl_link_unregister(&vxlan_link_ops);
3152         unregister_netdevice_notifier(&vxlan_notifier_block);
3153         destroy_workqueue(vxlan_wq);
3154         unregister_pernet_subsys(&vxlan_net_ops);
3155         /* rcu_barrier() is called by netns */
3156 }
3157 module_exit(vxlan_cleanup_module);
3158
3159 MODULE_LICENSE("GPL");
3160 MODULE_VERSION(VXLAN_VERSION);
3161 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3162 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3163 MODULE_ALIAS_RTNL_LINK("vxlan");