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/module.h>
  15 #include <linux/errno.h>
  16 #include <linux/slab.h>
  17 #include <linux/udp.h>
  18 #include <linux/igmp.h>
  19 #include <linux/if_ether.h>
  20 #include <linux/ethtool.h>
  21 #include <net/arp.h>
  22 #include <net/ndisc.h>
  23 #include <net/ip.h>
  24 #include <net/icmp.h>
  25 #include <net/rtnetlink.h>
  26 #include <net/inet_ecn.h>
  27 #include <net/net_namespace.h>
  28 #include <net/netns/generic.h>
  29 #include <net/tun_proto.h>
  30 #include <net/vxlan.h>
  31
  32 #if IS_ENABLED(CONFIG_IPV6)
  33 #include <net/ip6_tunnel.h>
  34 #include <net/ip6_checksum.h>
  35 #endif
  36
  37 #define VXLAN_VERSION   "0.1"
  38
  39 #define PORT_HASH_BITS  8
  40 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  41 #define FDB_AGE_DEFAULT 300 /* 5 min */
  42 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  43
  44 /* UDP port for VXLAN traffic.
  45  * The IANA assigned port is 4789, but the Linux default is 8472
  46  * for compatibility with early adopters.
  47  */
  48 static unsigned short vxlan_port __read_mostly = 8472;
  49 module_param_named(udp_port, vxlan_port, ushort, 0444);
  50 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  51
  52 static bool log_ecn_error = true;
  53 module_param(log_ecn_error, bool, 0644);
  54 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  55
  56 static unsigned int vxlan_net_id;
  57 static struct rtnl_link_ops vxlan_link_ops;
  58
  59 static const u8 all_zeros_mac[ETH_ALEN + 2];
  60
  61 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  62
  63 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
  64
  65 /* per-network namespace private data for this module */
  66 struct vxlan_net {
  67         struct list_head  vxlan_list;
  68         struct hlist_head sock_list[PORT_HASH_SIZE];
  69         spinlock_t        sock_lock;
  70 };
  71
  72 /* Forwarding table entry */
  73 struct vxlan_fdb {
  74         struct hlist_node hlist;        /* linked list of entries */
  75         struct rcu_head   rcu;
  76         unsigned long     updated;      /* jiffies */
  77         unsigned long     used;
  78         struct list_head  remotes;
  79         u8                eth_addr[ETH_ALEN];
  80         u16               state;        /* see ndm_state */
  81         __be32            vni;
  82         u8                flags;        /* see ndm_flags */
  83 };
  84
  85 /* salt for hash table */
  86 static u32 vxlan_salt __read_mostly;
  87
  88 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  89 {
  90         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  91                ip_tunnel_collect_metadata();
  92 }
  93
  94 #if IS_ENABLED(CONFIG_IPV6)
  95 static inline
  96 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
  97 {
  98         if (a->sa.sa_family != b->sa.sa_family)
  99                 return false;
 100         if (a->sa.sa_family == AF_INET6)
 101                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 102         else
 103                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 104 }
 105
 106 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 107 {
 108         if (ipa->sa.sa_family == AF_INET6)
 109                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 110         else
 111                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 112 }
 113
 114 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 115 {
 116         if (ipa->sa.sa_family == AF_INET6)
 117                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 118         else
 119                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 120 }
 121
 122 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 123 {
 124         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 125                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 126                 ip->sa.sa_family = AF_INET6;
 127                 return 0;
 128         } else if (nla_len(nla) >= sizeof(__be32)) {
 129                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 130                 ip->sa.sa_family = AF_INET;
 131                 return 0;
 132         } else {
 133                 return -EAFNOSUPPORT;
 134         }
 135 }
 136
 137 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 138                               const union vxlan_addr *ip)
 139 {
 140         if (ip->sa.sa_family == AF_INET6)
 141                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 142         else
 143                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 144 }
 145
 146 #else /* !CONFIG_IPV6 */
 147
 148 static inline
 149 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 150 {
 151         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 152 }
 153
 154 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 155 {
 156         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 157 }
 158
 159 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 160 {
 161         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 162 }
 163
 164 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 165 {
 166         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 167                 return -EAFNOSUPPORT;
 168         } else if (nla_len(nla) >= sizeof(__be32)) {
 169                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 170                 ip->sa.sa_family = AF_INET;
 171                 return 0;
 172         } else {
 173                 return -EAFNOSUPPORT;
 174         }
 175 }
 176
 177 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 178                               const union vxlan_addr *ip)
 179 {
 180         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 181 }
 182 #endif
 183
 184 /* Virtual Network hash table head */
 185 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 186 {
 187         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 188 }
 189
 190 /* Socket hash table head */
 191 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 192 {
 193         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 194
 195         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 196 }
 197
 198 /* First remote destination for a forwarding entry.
 199  * Guaranteed to be non-NULL because remotes are never deleted.
 200  */
 201 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 202 {
 203         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 204 }
 205
 206 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 207 {
 208         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 209 }
 210
 211 /* Find VXLAN socket based on network namespace, address family and UDP port
 212  * and enabled unshareable flags.
 213  */
 214 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 215                                           __be16 port, u32 flags)
 216 {
 217         struct vxlan_sock *vs;
 218
 219         flags &= VXLAN_F_RCV_FLAGS;
 220
 221         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 222                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 223                     vxlan_get_sk_family(vs) == family &&
 224                     vs->flags == flags)
 225                         return vs;
 226         }
 227         return NULL;
 228 }
 229
 230 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, int ifindex,
 231                                            __be32 vni)
 232 {
 233         struct vxlan_dev_node *node;
 234
 235         /* For flow based devices, map all packets to VNI 0 */
 236         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 237                 vni = 0;
 238
 239         hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
 240                 if (node->vxlan->default_dst.remote_vni != vni)
 241                         continue;
 242
 243                 if (IS_ENABLED(CONFIG_IPV6)) {
 244                         const struct vxlan_config *cfg = &node->vxlan->cfg;
 245
 246                         if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) &&
 247                             cfg->remote_ifindex != ifindex)
 248                                 continue;
 249                 }
 250
 251                 return node->vxlan;
 252         }
 253
 254         return NULL;
 255 }
 256
 257 /* Look up VNI in a per net namespace table */
 258 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex,
 259                                         __be32 vni, sa_family_t family,
 260                                         __be16 port, u32 flags)
 261 {
 262         struct vxlan_sock *vs;
 263
 264         vs = vxlan_find_sock(net, family, port, flags);
 265         if (!vs)
 266                 return NULL;
 267
 268         return vxlan_vs_find_vni(vs, ifindex, vni);
 269 }
 270
 271 /* Fill in neighbour message in skbuff. */
 272 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 273                           const struct vxlan_fdb *fdb,
 274                           u32 portid, u32 seq, int type, unsigned int flags,
 275                           const struct vxlan_rdst *rdst)
 276 {
 277         unsigned long now = jiffies;
 278         struct nda_cacheinfo ci;
 279         struct nlmsghdr *nlh;
 280         struct ndmsg *ndm;
 281         bool send_ip, send_eth;
 282
 283         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 284         if (nlh == NULL)
 285                 return -EMSGSIZE;
 286
 287         ndm = nlmsg_data(nlh);
 288         memset(ndm, 0, sizeof(*ndm));
 289
 290         send_eth = send_ip = true;
 291
 292         if (type == RTM_GETNEIGH) {
 293                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 294                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 295                 ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
 296         } else
 297                 ndm->ndm_family = AF_BRIDGE;
 298         ndm->ndm_state = fdb->state;
 299         ndm->ndm_ifindex = vxlan->dev->ifindex;
 300         ndm->ndm_flags = fdb->flags;
 301         ndm->ndm_type = RTN_UNICAST;
 302
 303         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 304             nla_put_s32(skb, NDA_LINK_NETNSID,
 305                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 306                 goto nla_put_failure;
 307
 308         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 309                 goto nla_put_failure;
 310
 311         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 312                 goto nla_put_failure;
 313
 314         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 315             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 316                 goto nla_put_failure;
 317         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 318             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 319                 goto nla_put_failure;
 320         if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
 321             nla_put_u32(skb, NDA_SRC_VNI,
 322                         be32_to_cpu(fdb->vni)))
 323                 goto nla_put_failure;
 324         if (rdst->remote_ifindex &&
 325             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 326                 goto nla_put_failure;
 327
 328         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 329         ci.ndm_confirmed = 0;
 330         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 331         ci.ndm_refcnt    = 0;
 332
 333         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 334                 goto nla_put_failure;
 335
 336         nlmsg_end(skb, nlh);
 337         return 0;
 338
 339 nla_put_failure:
 340         nlmsg_cancel(skb, nlh);
 341         return -EMSGSIZE;
 342 }
 343
 344 static inline size_t vxlan_nlmsg_size(void)
 345 {
 346         return NLMSG_ALIGN(sizeof(struct ndmsg))
 347                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 348                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 349                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 350                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 351                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 352                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 353                 + nla_total_size(sizeof(struct nda_cacheinfo));
 354 }
 355
 356 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 357                              struct vxlan_rdst *rd, int type)
 358 {
 359         struct net *net = dev_net(vxlan->dev);
 360         struct sk_buff *skb;
 361         int err = -ENOBUFS;
 362
 363         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 364         if (skb == NULL)
 365                 goto errout;
 366
 367         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 368         if (err < 0) {
 369                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 370                 WARN_ON(err == -EMSGSIZE);
 371                 kfree_skb(skb);
 372                 goto errout;
 373         }
 374
 375         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 376         return;
 377 errout:
 378         if (err < 0)
 379                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 380 }
 381
 382 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 383 {
 384         struct vxlan_dev *vxlan = netdev_priv(dev);
 385         struct vxlan_fdb f = {
 386                 .state = NUD_STALE,
 387         };
 388         struct vxlan_rdst remote = {
 389                 .remote_ip = *ipa, /* goes to NDA_DST */
 390                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 391         };
 392
 393         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 394 }
 395
 396 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 397 {
 398         struct vxlan_fdb f = {
 399                 .state = NUD_STALE,
 400         };
 401         struct vxlan_rdst remote = { };
 402
 403         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 404
 405         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 406 }
 407
 408 /* Hash Ethernet address */
 409 static u32 eth_hash(const unsigned char *addr)
 410 {
 411         u64 value = get_unaligned((u64 *)addr);
 412
 413         /* only want 6 bytes */
 414 #ifdef __BIG_ENDIAN
 415         value >>= 16;
 416 #else
 417         value <<= 16;
 418 #endif
 419         return hash_64(value, FDB_HASH_BITS);
 420 }
 421
 422 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
 423 {
 424         /* use 1 byte of OUI and 3 bytes of NIC */
 425         u32 key = get_unaligned((u32 *)(addr + 2));
 426
 427         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
 428 }
 429
 430 /* Hash chain to use given mac address */
 431 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 432                                                 const u8 *mac, __be32 vni)
 433 {
 434         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
 435                 return &vxlan->fdb_head[eth_vni_hash(mac, vni)];
 436         else
 437                 return &vxlan->fdb_head[eth_hash(mac)];
 438 }
 439
 440 /* Look up Ethernet address in forwarding table */
 441 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 442                                           const u8 *mac, __be32 vni)
 443 {
 444         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 445         struct vxlan_fdb *f;
 446
 447         hlist_for_each_entry_rcu(f, head, hlist) {
 448                 if (ether_addr_equal(mac, f->eth_addr)) {
 449                         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
 450                                 if (vni == f->vni)
 451                                         return f;
 452                         } else {
 453                                 return f;
 454                         }
 455                 }
 456         }
 457
 458         return NULL;
 459 }
 460
 461 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 462                                         const u8 *mac, __be32 vni)
 463 {
 464         struct vxlan_fdb *f;
 465
 466         f = __vxlan_find_mac(vxlan, mac, vni);
 467         if (f)
 468                 f->used = jiffies;
 469
 470         return f;
 471 }
 472
 473 /* caller should hold vxlan->hash_lock */
 474 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 475                                               union vxlan_addr *ip, __be16 port,
 476                                               __be32 vni, __u32 ifindex)
 477 {
 478         struct vxlan_rdst *rd;
 479
 480         list_for_each_entry(rd, &f->remotes, list) {
 481                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 482                     rd->remote_port == port &&
 483                     rd->remote_vni == vni &&
 484                     rd->remote_ifindex == ifindex)
 485                         return rd;
 486         }
 487
 488         return NULL;
 489 }
 490
 491 /* Replace destination of unicast mac */
 492 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 493                              union vxlan_addr *ip, __be16 port, __be32 vni,
 494                              __u32 ifindex)
 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 = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 503         if (!rd)
 504                 return 0;
 505
 506         dst_cache_reset(&rd->dst_cache);
 507         rd->remote_ip = *ip;
 508         rd->remote_port = port;
 509         rd->remote_vni = vni;
 510         rd->remote_ifindex = ifindex;
 511         return 1;
 512 }
 513
 514 /* Add/update destinations for multicast */
 515 static int vxlan_fdb_append(struct vxlan_fdb *f,
 516                             union vxlan_addr *ip, __be16 port, __be32 vni,
 517                             __u32 ifindex, struct vxlan_rdst **rdp)
 518 {
 519         struct vxlan_rdst *rd;
 520
 521         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 522         if (rd)
 523                 return 0;
 524
 525         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 526         if (rd == NULL)
 527                 return -ENOBUFS;
 528
 529         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 530                 kfree(rd);
 531                 return -ENOBUFS;
 532         }
 533
 534         rd->remote_ip = *ip;
 535         rd->remote_port = port;
 536         rd->remote_vni = vni;
 537         rd->remote_ifindex = ifindex;
 538
 539         list_add_tail_rcu(&rd->list, &f->remotes);
 540
 541         *rdp = rd;
 542         return 1;
 543 }
 544
 545 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 546                                           unsigned int off,
 547                                           struct vxlanhdr *vh, size_t hdrlen,
 548                                           __be32 vni_field,
 549                                           struct gro_remcsum *grc,
 550                                           bool nopartial)
 551 {
 552         size_t start, offset;
 553
 554         if (skb->remcsum_offload)
 555                 return vh;
 556
 557         if (!NAPI_GRO_CB(skb)->csum_valid)
 558                 return NULL;
 559
 560         start = vxlan_rco_start(vni_field);
 561         offset = start + vxlan_rco_offset(vni_field);
 562
 563         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 564                                      start, offset, grc, nopartial);
 565
 566         skb->remcsum_offload = 1;
 567
 568         return vh;
 569 }
 570
 571 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
 572                                           struct sk_buff **head,
 573                                           struct sk_buff *skb)
 574 {
 575         struct sk_buff *p, **pp = NULL;
 576         struct vxlanhdr *vh, *vh2;
 577         unsigned int hlen, off_vx;
 578         int flush = 1;
 579         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 580         __be32 flags;
 581         struct gro_remcsum grc;
 582
 583         skb_gro_remcsum_init(&grc);
 584
 585         off_vx = skb_gro_offset(skb);
 586         hlen = off_vx + sizeof(*vh);
 587         vh   = skb_gro_header_fast(skb, off_vx);
 588         if (skb_gro_header_hard(skb, hlen)) {
 589                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 590                 if (unlikely(!vh))
 591                         goto out;
 592         }
 593
 594         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 595
 596         flags = vh->vx_flags;
 597
 598         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 599                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 600                                        vh->vx_vni, &grc,
 601                                        !!(vs->flags &
 602                                           VXLAN_F_REMCSUM_NOPARTIAL));
 603
 604                 if (!vh)
 605                         goto out;
 606         }
 607
 608         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 609
 610         for (p = *head; p; p = p->next) {
 611                 if (!NAPI_GRO_CB(p)->same_flow)
 612                         continue;
 613
 614                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 615                 if (vh->vx_flags != vh2->vx_flags ||
 616                     vh->vx_vni != vh2->vx_vni) {
 617                         NAPI_GRO_CB(p)->same_flow = 0;
 618                         continue;
 619                 }
 620         }
 621
 622         pp = call_gro_receive(eth_gro_receive, head, skb);
 623         flush = 0;
 624
 625 out:
 626         skb_gro_remcsum_cleanup(skb, &grc);
 627         skb->remcsum_offload = 0;
 628         NAPI_GRO_CB(skb)->flush |= flush;
 629
 630         return pp;
 631 }
 632
 633 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 634 {
 635         /* Sets 'skb->inner_mac_header' since we are always called with
 636          * 'skb->encapsulation' set.
 637          */
 638         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 639 }
 640
 641 /* Add new entry to forwarding table -- assumes lock held */
 642 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 643                             const u8 *mac, union vxlan_addr *ip,
 644                             __u16 state, __u16 flags,
 645                             __be16 port, __be32 src_vni, __be32 vni,
 646                             __u32 ifindex, __u8 ndm_flags)
 647 {
 648         struct vxlan_rdst *rd = NULL;
 649         struct vxlan_fdb *f;
 650         int notify = 0;
 651         int rc;
 652
 653         f = __vxlan_find_mac(vxlan, mac, src_vni);
 654         if (f) {
 655                 if (flags & NLM_F_EXCL) {
 656                         netdev_dbg(vxlan->dev,
 657                                    "lost race to create %pM\n", mac);
 658                         return -EEXIST;
 659                 }
 660                 if (f->state != state) {
 661                         f->state = state;
 662                         f->updated = jiffies;
 663                         notify = 1;
 664                 }
 665                 if (f->flags != ndm_flags) {
 666                         f->flags = ndm_flags;
 667                         f->updated = jiffies;
 668                         notify = 1;
 669                 }
 670                 if ((flags & NLM_F_REPLACE)) {
 671                         /* Only change unicasts */
 672                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 673                              is_zero_ether_addr(f->eth_addr))) {
 674                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 675                                                            ifindex);
 676                         } else
 677                                 return -EOPNOTSUPP;
 678                 }
 679                 if ((flags & NLM_F_APPEND) &&
 680                     (is_multicast_ether_addr(f->eth_addr) ||
 681                      is_zero_ether_addr(f->eth_addr))) {
 682                         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 683
 684                         if (rc < 0)
 685                                 return rc;
 686                         notify |= rc;
 687                 }
 688         } else {
 689                 if (!(flags & NLM_F_CREATE))
 690                         return -ENOENT;
 691
 692                 if (vxlan->cfg.addrmax &&
 693                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 694                         return -ENOSPC;
 695
 696                 /* Disallow replace to add a multicast entry */
 697                 if ((flags & NLM_F_REPLACE) &&
 698                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 699                         return -EOPNOTSUPP;
 700
 701                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 702                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 703                 if (!f)
 704                         return -ENOMEM;
 705
 706                 notify = 1;
 707                 f->state = state;
 708                 f->flags = ndm_flags;
 709                 f->updated = f->used = jiffies;
 710                 f->vni = src_vni;
 711                 INIT_LIST_HEAD(&f->remotes);
 712                 memcpy(f->eth_addr, mac, ETH_ALEN);
 713
 714                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 715                 if (rc < 0) {
 716                         kfree(f);
 717                         return rc;
 718                 }
 719
 720                 ++vxlan->addrcnt;
 721                 hlist_add_head_rcu(&f->hlist,
 722                                    vxlan_fdb_head(vxlan, mac, src_vni));
 723         }
 724
 725         if (notify) {
 726                 if (rd == NULL)
 727                         rd = first_remote_rtnl(f);
 728                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 729         }
 730
 731         return 0;
 732 }
 733
 734 static void vxlan_fdb_free(struct rcu_head *head)
 735 {
 736         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 737         struct vxlan_rdst *rd, *nd;
 738
 739         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 740                 dst_cache_destroy(&rd->dst_cache);
 741                 kfree(rd);
 742         }
 743         kfree(f);
 744 }
 745
 746 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 747 {
 748         netdev_dbg(vxlan->dev,
 749                     "delete %pM\n", f->eth_addr);
 750
 751         --vxlan->addrcnt;
 752         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 753
 754         hlist_del_rcu(&f->hlist);
 755         call_rcu(&f->rcu, vxlan_fdb_free);
 756 }
 757
 758 static void vxlan_dst_free(struct rcu_head *head)
 759 {
 760         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 761
 762         dst_cache_destroy(&rd->dst_cache);
 763         kfree(rd);
 764 }
 765
 766 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 767                                   struct vxlan_rdst *rd)
 768 {
 769         list_del_rcu(&rd->list);
 770         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 771         call_rcu(&rd->rcu, vxlan_dst_free);
 772 }
 773
 774 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 775                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
 776                            __be32 *vni, u32 *ifindex)
 777 {
 778         struct net *net = dev_net(vxlan->dev);
 779         int err;
 780
 781         if (tb[NDA_DST]) {
 782                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 783                 if (err)
 784                         return err;
 785         } else {
 786                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 787                 if (remote->sa.sa_family == AF_INET) {
 788                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 789                         ip->sa.sa_family = AF_INET;
 790 #if IS_ENABLED(CONFIG_IPV6)
 791                 } else {
 792                         ip->sin6.sin6_addr = in6addr_any;
 793                         ip->sa.sa_family = AF_INET6;
 794 #endif
 795                 }
 796         }
 797
 798         if (tb[NDA_PORT]) {
 799                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 800                         return -EINVAL;
 801                 *port = nla_get_be16(tb[NDA_PORT]);
 802         } else {
 803                 *port = vxlan->cfg.dst_port;
 804         }
 805
 806         if (tb[NDA_VNI]) {
 807                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 808                         return -EINVAL;
 809                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 810         } else {
 811                 *vni = vxlan->default_dst.remote_vni;
 812         }
 813
 814         if (tb[NDA_SRC_VNI]) {
 815                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
 816                         return -EINVAL;
 817                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
 818         } else {
 819                 *src_vni = vxlan->default_dst.remote_vni;
 820         }
 821
 822         if (tb[NDA_IFINDEX]) {
 823                 struct net_device *tdev;
 824
 825                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 826                         return -EINVAL;
 827                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 828                 tdev = __dev_get_by_index(net, *ifindex);
 829                 if (!tdev)
 830                         return -EADDRNOTAVAIL;
 831         } else {
 832                 *ifindex = 0;
 833         }
 834
 835         return 0;
 836 }
 837
 838 /* Add static entry (via netlink) */
 839 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 840                          struct net_device *dev,
 841                          const unsigned char *addr, u16 vid, u16 flags)
 842 {
 843         struct vxlan_dev *vxlan = netdev_priv(dev);
 844         /* struct net *net = dev_net(vxlan->dev); */
 845         union vxlan_addr ip;
 846         __be16 port;
 847         __be32 src_vni, vni;
 848         u32 ifindex;
 849         int err;
 850
 851         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 852                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 853                         ndm->ndm_state);
 854                 return -EINVAL;
 855         }
 856
 857         if (tb[NDA_DST] == NULL)
 858                 return -EINVAL;
 859
 860         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 861         if (err)
 862                 return err;
 863
 864         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 865                 return -EAFNOSUPPORT;
 866
 867         spin_lock_bh(&vxlan->hash_lock);
 868         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 869                                port, src_vni, vni, ifindex, ndm->ndm_flags);
 870         spin_unlock_bh(&vxlan->hash_lock);
 871
 872         return err;
 873 }
 874
 875 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
 876                               const unsigned char *addr, union vxlan_addr ip,
 877                               __be16 port, __be32 src_vni, u32 vni, u32 ifindex,
 878                               u16 vid)
 879 {
 880         struct vxlan_fdb *f;
 881         struct vxlan_rdst *rd = NULL;
 882         int err = -ENOENT;
 883
 884         f = vxlan_find_mac(vxlan, addr, src_vni);
 885         if (!f)
 886                 return err;
 887
 888         if (!vxlan_addr_any(&ip)) {
 889                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 890                 if (!rd)
 891                         goto out;
 892         }
 893
 894         /* remove a destination if it's not the only one on the list,
 895          * otherwise destroy the fdb entry
 896          */
 897         if (rd && !list_is_singular(&f->remotes)) {
 898                 vxlan_fdb_dst_destroy(vxlan, f, rd);
 899                 goto out;
 900         }
 901
 902         vxlan_fdb_destroy(vxlan, f);
 903
 904 out:
 905         return 0;
 906 }
 907
 908 /* Delete entry (via netlink) */
 909 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 910                             struct net_device *dev,
 911                             const unsigned char *addr, u16 vid)
 912 {
 913         struct vxlan_dev *vxlan = netdev_priv(dev);
 914         union vxlan_addr ip;
 915         __be32 src_vni, vni;
 916         __be16 port;
 917         u32 ifindex;
 918         int err;
 919
 920         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 921         if (err)
 922                 return err;
 923
 924         spin_lock_bh(&vxlan->hash_lock);
 925         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
 926                                  vid);
 927         spin_unlock_bh(&vxlan->hash_lock);
 928
 929         return err;
 930 }
 931
 932 /* Dump forwarding table */
 933 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 934                           struct net_device *dev,
 935                           struct net_device *filter_dev, int *idx)
 936 {
 937         struct vxlan_dev *vxlan = netdev_priv(dev);
 938         unsigned int h;
 939         int err = 0;
 940
 941         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 942                 struct vxlan_fdb *f;
 943
 944                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 945                         struct vxlan_rdst *rd;
 946
 947                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 948                                 if (*idx < cb->args[2])
 949                                         goto skip;
 950
 951                                 err = vxlan_fdb_info(skb, vxlan, f,
 952                                                      NETLINK_CB(cb->skb).portid,
 953                                                      cb->nlh->nlmsg_seq,
 954                                                      RTM_NEWNEIGH,
 955                                                      NLM_F_MULTI, rd);
 956                                 if (err < 0)
 957                                         goto out;
 958 skip:
 959                                 *idx += 1;
 960                         }
 961                 }
 962         }
 963 out:
 964         return err;
 965 }
 966
 967 /* Watch incoming packets to learn mapping between Ethernet address
 968  * and Tunnel endpoint.
 969  * Return true if packet is bogus and should be dropped.
 970  */
 971 static bool vxlan_snoop(struct net_device *dev,
 972                         union vxlan_addr *src_ip, const u8 *src_mac,
 973                         u32 src_ifindex, __be32 vni)
 974 {
 975         struct vxlan_dev *vxlan = netdev_priv(dev);
 976         struct vxlan_fdb *f;
 977         u32 ifindex = 0;
 978
 979 #if IS_ENABLED(CONFIG_IPV6)
 980         if (src_ip->sa.sa_family == AF_INET6 &&
 981             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
 982                 ifindex = src_ifindex;
 983 #endif
 984
 985         f = vxlan_find_mac(vxlan, src_mac, vni);
 986         if (likely(f)) {
 987                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 988
 989                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
 990                            rdst->remote_ifindex == ifindex))
 991                         return false;
 992
 993                 /* Don't migrate static entries, drop packets */
 994                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
 995                         return true;
 996
 997                 if (net_ratelimit())
 998                         netdev_info(dev,
 999                                     "%pM migrated from %pIS to %pIS\n",
1000                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1001
1002                 rdst->remote_ip = *src_ip;
1003                 f->updated = jiffies;
1004                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1005         } else {
1006                 /* learned new entry */
1007                 spin_lock(&vxlan->hash_lock);
1008
1009                 /* close off race between vxlan_flush and incoming packets */
1010                 if (netif_running(dev))
1011                         vxlan_fdb_create(vxlan, src_mac, src_ip,
1012                                          NUD_REACHABLE,
1013                                          NLM_F_EXCL|NLM_F_CREATE,
1014                                          vxlan->cfg.dst_port,
1015                                          vni,
1016                                          vxlan->default_dst.remote_vni,
1017                                          ifindex, NTF_SELF);
1018                 spin_unlock(&vxlan->hash_lock);
1019         }
1020
1021         return false;
1022 }
1023
1024 /* See if multicast group is already in use by other ID */
1025 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1026 {
1027         struct vxlan_dev *vxlan;
1028         struct vxlan_sock *sock4;
1029 #if IS_ENABLED(CONFIG_IPV6)
1030         struct vxlan_sock *sock6;
1031 #endif
1032         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1033
1034         sock4 = rtnl_dereference(dev->vn4_sock);
1035
1036         /* The vxlan_sock is only used by dev, leaving group has
1037          * no effect on other vxlan devices.
1038          */
1039         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1040                 return false;
1041 #if IS_ENABLED(CONFIG_IPV6)
1042         sock6 = rtnl_dereference(dev->vn6_sock);
1043         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1044                 return false;
1045 #endif
1046
1047         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1048                 if (!netif_running(vxlan->dev) || vxlan == dev)
1049                         continue;
1050
1051                 if (family == AF_INET &&
1052                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1053                         continue;
1054 #if IS_ENABLED(CONFIG_IPV6)
1055                 if (family == AF_INET6 &&
1056                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1057                         continue;
1058 #endif
1059
1060                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1061                                       &dev->default_dst.remote_ip))
1062                         continue;
1063
1064                 if (vxlan->default_dst.remote_ifindex !=
1065                     dev->default_dst.remote_ifindex)
1066                         continue;
1067
1068                 return true;
1069         }
1070
1071         return false;
1072 }
1073
1074 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1075 {
1076         struct vxlan_net *vn;
1077
1078         if (!vs)
1079                 return false;
1080         if (!refcount_dec_and_test(&vs->refcnt))
1081                 return false;
1082
1083         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1084         spin_lock(&vn->sock_lock);
1085         hlist_del_rcu(&vs->hlist);
1086         udp_tunnel_notify_del_rx_port(vs->sock,
1087                                       (vs->flags & VXLAN_F_GPE) ?
1088                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1089                                       UDP_TUNNEL_TYPE_VXLAN);
1090         spin_unlock(&vn->sock_lock);
1091
1092         return true;
1093 }
1094
1095 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1096 {
1097         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1098 #if IS_ENABLED(CONFIG_IPV6)
1099         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1100
1101         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1102 #endif
1103
1104         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1105         synchronize_net();
1106
1107         vxlan_vs_del_dev(vxlan);
1108
1109         if (__vxlan_sock_release_prep(sock4)) {
1110                 udp_tunnel_sock_release(sock4->sock);
1111                 kfree(sock4);
1112         }
1113
1114 #if IS_ENABLED(CONFIG_IPV6)
1115         if (__vxlan_sock_release_prep(sock6)) {
1116                 udp_tunnel_sock_release(sock6->sock);
1117                 kfree(sock6);
1118         }
1119 #endif
1120 }
1121
1122 /* Update multicast group membership when first VNI on
1123  * multicast address is brought up
1124  */
1125 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1126 {
1127         struct sock *sk;
1128         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1129         int ifindex = vxlan->default_dst.remote_ifindex;
1130         int ret = -EINVAL;
1131
1132         if (ip->sa.sa_family == AF_INET) {
1133                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1134                 struct ip_mreqn mreq = {
1135                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1136                         .imr_ifindex            = ifindex,
1137                 };
1138
1139                 sk = sock4->sock->sk;
1140                 lock_sock(sk);
1141                 ret = ip_mc_join_group(sk, &mreq);
1142                 release_sock(sk);
1143 #if IS_ENABLED(CONFIG_IPV6)
1144         } else {
1145                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1146
1147                 sk = sock6->sock->sk;
1148                 lock_sock(sk);
1149                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1150                                                    &ip->sin6.sin6_addr);
1151                 release_sock(sk);
1152 #endif
1153         }
1154
1155         return ret;
1156 }
1157
1158 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1159 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1160 {
1161         struct sock *sk;
1162         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1163         int ifindex = vxlan->default_dst.remote_ifindex;
1164         int ret = -EINVAL;
1165
1166         if (ip->sa.sa_family == AF_INET) {
1167                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1168                 struct ip_mreqn mreq = {
1169                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1170                         .imr_ifindex            = ifindex,
1171                 };
1172
1173                 sk = sock4->sock->sk;
1174                 lock_sock(sk);
1175                 ret = ip_mc_leave_group(sk, &mreq);
1176                 release_sock(sk);
1177 #if IS_ENABLED(CONFIG_IPV6)
1178         } else {
1179                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1180
1181                 sk = sock6->sock->sk;
1182                 lock_sock(sk);
1183                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1184                                                    &ip->sin6.sin6_addr);
1185                 release_sock(sk);
1186 #endif
1187         }
1188
1189         return ret;
1190 }
1191
1192 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1193                           struct sk_buff *skb, u32 vxflags)
1194 {
1195         size_t start, offset;
1196
1197         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1198                 goto out;
1199
1200         start = vxlan_rco_start(unparsed->vx_vni);
1201         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1202
1203         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1204                 return false;
1205
1206         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1207                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1208 out:
1209         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1210         unparsed->vx_vni &= VXLAN_VNI_MASK;
1211         return true;
1212 }
1213
1214 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1215                                 struct sk_buff *skb, u32 vxflags,
1216                                 struct vxlan_metadata *md)
1217 {
1218         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1219         struct metadata_dst *tun_dst;
1220
1221         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1222                 goto out;
1223
1224         md->gbp = ntohs(gbp->policy_id);
1225
1226         tun_dst = (struct metadata_dst *)skb_dst(skb);
1227         if (tun_dst) {
1228                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1229                 tun_dst->u.tun_info.options_len = sizeof(*md);
1230         }
1231         if (gbp->dont_learn)
1232                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1233
1234         if (gbp->policy_applied)
1235                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1236
1237         /* In flow-based mode, GBP is carried in dst_metadata */
1238         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1239                 skb->mark = md->gbp;
1240 out:
1241         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1242 }
1243
1244 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1245                                 __be16 *protocol,
1246                                 struct sk_buff *skb, u32 vxflags)
1247 {
1248         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1249
1250         /* Need to have Next Protocol set for interfaces in GPE mode. */
1251         if (!gpe->np_applied)
1252                 return false;
1253         /* "The initial version is 0. If a receiver does not support the
1254          * version indicated it MUST drop the packet.
1255          */
1256         if (gpe->version != 0)
1257                 return false;
1258         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1259          * processing MUST occur." However, we don't implement OAM
1260          * processing, thus drop the packet.
1261          */
1262         if (gpe->oam_flag)
1263                 return false;
1264
1265         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1266         if (!*protocol)
1267                 return false;
1268
1269         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1270         return true;
1271 }
1272
1273 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1274                           struct vxlan_sock *vs,
1275                           struct sk_buff *skb, __be32 vni)
1276 {
1277         union vxlan_addr saddr;
1278         u32 ifindex = skb->dev->ifindex;
1279
1280         skb_reset_mac_header(skb);
1281         skb->protocol = eth_type_trans(skb, vxlan->dev);
1282         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1283
1284         /* Ignore packet loops (and multicast echo) */
1285         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1286                 return false;
1287
1288         /* Get address from the outer IP header */
1289         if (vxlan_get_sk_family(vs) == AF_INET) {
1290                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1291                 saddr.sa.sa_family = AF_INET;
1292 #if IS_ENABLED(CONFIG_IPV6)
1293         } else {
1294                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1295                 saddr.sa.sa_family = AF_INET6;
1296 #endif
1297         }
1298
1299         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1300             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1301                 return false;
1302
1303         return true;
1304 }
1305
1306 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1307                                   struct sk_buff *skb)
1308 {
1309         int err = 0;
1310
1311         if (vxlan_get_sk_family(vs) == AF_INET)
1312                 err = IP_ECN_decapsulate(oiph, skb);
1313 #if IS_ENABLED(CONFIG_IPV6)
1314         else
1315                 err = IP6_ECN_decapsulate(oiph, skb);
1316 #endif
1317
1318         if (unlikely(err) && log_ecn_error) {
1319                 if (vxlan_get_sk_family(vs) == AF_INET)
1320                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1321                                              &((struct iphdr *)oiph)->saddr,
1322                                              ((struct iphdr *)oiph)->tos);
1323                 else
1324                         net_info_ratelimited("non-ECT from %pI6\n",
1325                                              &((struct ipv6hdr *)oiph)->saddr);
1326         }
1327         return err <= 1;
1328 }
1329
1330 /* Callback from net/ipv4/udp.c to receive packets */
1331 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1332 {
1333         struct pcpu_sw_netstats *stats;
1334         struct vxlan_dev *vxlan;
1335         struct vxlan_sock *vs;
1336         struct vxlanhdr unparsed;
1337         struct vxlan_metadata _md;
1338         struct vxlan_metadata *md = &_md;
1339         __be16 protocol = htons(ETH_P_TEB);
1340         bool raw_proto = false;
1341         void *oiph;
1342         __be32 vni = 0;
1343
1344         /* Need UDP and VXLAN header to be present */
1345         if (!pskb_may_pull(skb, VXLAN_HLEN))
1346                 goto drop;
1347
1348         unparsed = *vxlan_hdr(skb);
1349         /* VNI flag always required to be set */
1350         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1351                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1352                            ntohl(vxlan_hdr(skb)->vx_flags),
1353                            ntohl(vxlan_hdr(skb)->vx_vni));
1354                 /* Return non vxlan pkt */
1355                 goto drop;
1356         }
1357         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1358         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1359
1360         vs = rcu_dereference_sk_user_data(sk);
1361         if (!vs)
1362                 goto drop;
1363
1364         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1365
1366         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1367         if (!vxlan)
1368                 goto drop;
1369
1370         /* For backwards compatibility, only allow reserved fields to be
1371          * used by VXLAN extensions if explicitly requested.
1372          */
1373         if (vs->flags & VXLAN_F_GPE) {
1374                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1375                         goto drop;
1376                 raw_proto = true;
1377         }
1378
1379         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1380                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1381                         goto drop;
1382
1383         if (vxlan_collect_metadata(vs)) {
1384                 struct metadata_dst *tun_dst;
1385
1386                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1387                                          key32_to_tunnel_id(vni), sizeof(*md));
1388
1389                 if (!tun_dst)
1390                         goto drop;
1391
1392                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1393
1394                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1395         } else {
1396                 memset(md, 0, sizeof(*md));
1397         }
1398
1399         if (vs->flags & VXLAN_F_REMCSUM_RX)
1400                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1401                         goto drop;
1402         if (vs->flags & VXLAN_F_GBP)
1403                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1404         /* Note that GBP and GPE can never be active together. This is
1405          * ensured in vxlan_dev_configure.
1406          */
1407
1408         if (unparsed.vx_flags || unparsed.vx_vni) {
1409                 /* If there are any unprocessed flags remaining treat
1410                  * this as a malformed packet. This behavior diverges from
1411                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1412                  * in reserved fields are to be ignored. The approach here
1413                  * maintains compatibility with previous stack code, and also
1414                  * is more robust and provides a little more security in
1415                  * adding extensions to VXLAN.
1416                  */
1417                 goto drop;
1418         }
1419
1420         if (!raw_proto) {
1421                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1422                         goto drop;
1423         } else {
1424                 skb_reset_mac_header(skb);
1425                 skb->dev = vxlan->dev;
1426                 skb->pkt_type = PACKET_HOST;
1427         }
1428
1429         oiph = skb_network_header(skb);
1430         skb_reset_network_header(skb);
1431
1432         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1433                 ++vxlan->dev->stats.rx_frame_errors;
1434                 ++vxlan->dev->stats.rx_errors;
1435                 goto drop;
1436         }
1437
1438         stats = this_cpu_ptr(vxlan->dev->tstats);
1439         u64_stats_update_begin(&stats->syncp);
1440         stats->rx_packets++;
1441         stats->rx_bytes += skb->len;
1442         u64_stats_update_end(&stats->syncp);
1443
1444         gro_cells_receive(&vxlan->gro_cells, skb);
1445         return 0;
1446
1447 drop:
1448         /* Consume bad packet */
1449         kfree_skb(skb);
1450         return 0;
1451 }
1452
1453 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1454 {
1455         struct vxlan_dev *vxlan = netdev_priv(dev);
1456         struct arphdr *parp;
1457         u8 *arpptr, *sha;
1458         __be32 sip, tip;
1459         struct neighbour *n;
1460
1461         if (dev->flags & IFF_NOARP)
1462                 goto out;
1463
1464         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1465                 dev->stats.tx_dropped++;
1466                 goto out;
1467         }
1468         parp = arp_hdr(skb);
1469
1470         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1471              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1472             parp->ar_pro != htons(ETH_P_IP) ||
1473             parp->ar_op != htons(ARPOP_REQUEST) ||
1474             parp->ar_hln != dev->addr_len ||
1475             parp->ar_pln != 4)
1476                 goto out;
1477         arpptr = (u8 *)parp + sizeof(struct arphdr);
1478         sha = arpptr;
1479         arpptr += dev->addr_len;        /* sha */
1480         memcpy(&sip, arpptr, sizeof(sip));
1481         arpptr += sizeof(sip);
1482         arpptr += dev->addr_len;        /* tha */
1483         memcpy(&tip, arpptr, sizeof(tip));
1484
1485         if (ipv4_is_loopback(tip) ||
1486             ipv4_is_multicast(tip))
1487                 goto out;
1488
1489         n = neigh_lookup(&arp_tbl, &tip, dev);
1490
1491         if (n) {
1492                 struct vxlan_fdb *f;
1493                 struct sk_buff  *reply;
1494
1495                 if (!(n->nud_state & NUD_CONNECTED)) {
1496                         neigh_release(n);
1497                         goto out;
1498                 }
1499
1500                 f = vxlan_find_mac(vxlan, n->ha, vni);
1501                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1502                         /* bridge-local neighbor */
1503                         neigh_release(n);
1504                         goto out;
1505                 }
1506
1507                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1508                                 n->ha, sha);
1509
1510                 neigh_release(n);
1511
1512                 if (reply == NULL)
1513                         goto out;
1514
1515                 skb_reset_mac_header(reply);
1516                 __skb_pull(reply, skb_network_offset(reply));
1517                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1518                 reply->pkt_type = PACKET_HOST;
1519
1520                 if (netif_rx_ni(reply) == NET_RX_DROP)
1521                         dev->stats.rx_dropped++;
1522         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1523                 union vxlan_addr ipa = {
1524                         .sin.sin_addr.s_addr = tip,
1525                         .sin.sin_family = AF_INET,
1526                 };
1527
1528                 vxlan_ip_miss(dev, &ipa);
1529         }
1530 out:
1531         consume_skb(skb);
1532         return NETDEV_TX_OK;
1533 }
1534
1535 #if IS_ENABLED(CONFIG_IPV6)
1536 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1537         struct neighbour *n, bool isrouter)
1538 {
1539         struct net_device *dev = request->dev;
1540         struct sk_buff *reply;
1541         struct nd_msg *ns, *na;
1542         struct ipv6hdr *pip6;
1543         u8 *daddr;
1544         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1545         int ns_olen;
1546         int i, len;
1547
1548         if (dev == NULL || !pskb_may_pull(request, request->len))
1549                 return NULL;
1550
1551         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1552                 sizeof(*na) + na_olen + dev->needed_tailroom;
1553         reply = alloc_skb(len, GFP_ATOMIC);
1554         if (reply == NULL)
1555                 return NULL;
1556
1557         reply->protocol = htons(ETH_P_IPV6);
1558         reply->dev = dev;
1559         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1560         skb_push(reply, sizeof(struct ethhdr));
1561         skb_reset_mac_header(reply);
1562
1563         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1564
1565         daddr = eth_hdr(request)->h_source;
1566         ns_olen = request->len - skb_network_offset(request) -
1567                 sizeof(struct ipv6hdr) - sizeof(*ns);
1568         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1569                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1570                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1571                         break;
1572                 }
1573         }
1574
1575         /* Ethernet header */
1576         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1577         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1578         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1579         reply->protocol = htons(ETH_P_IPV6);
1580
1581         skb_pull(reply, sizeof(struct ethhdr));
1582         skb_reset_network_header(reply);
1583         skb_put(reply, sizeof(struct ipv6hdr));
1584
1585         /* IPv6 header */
1586
1587         pip6 = ipv6_hdr(reply);
1588         memset(pip6, 0, sizeof(struct ipv6hdr));
1589         pip6->version = 6;
1590         pip6->priority = ipv6_hdr(request)->priority;
1591         pip6->nexthdr = IPPROTO_ICMPV6;
1592         pip6->hop_limit = 255;
1593         pip6->daddr = ipv6_hdr(request)->saddr;
1594         pip6->saddr = *(struct in6_addr *)n->primary_key;
1595
1596         skb_pull(reply, sizeof(struct ipv6hdr));
1597         skb_reset_transport_header(reply);
1598
1599         /* Neighbor Advertisement */
1600         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1601         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1602         na->icmph.icmp6_router = isrouter;
1603         na->icmph.icmp6_override = 1;
1604         na->icmph.icmp6_solicited = 1;
1605         na->target = ns->target;
1606         ether_addr_copy(&na->opt[2], n->ha);
1607         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1608         na->opt[1] = na_olen >> 3;
1609
1610         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1611                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1612                 csum_partial(na, sizeof(*na)+na_olen, 0));
1613
1614         pip6->payload_len = htons(sizeof(*na)+na_olen);
1615
1616         skb_push(reply, sizeof(struct ipv6hdr));
1617
1618         reply->ip_summed = CHECKSUM_UNNECESSARY;
1619
1620         return reply;
1621 }
1622
1623 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1624 {
1625         struct vxlan_dev *vxlan = netdev_priv(dev);
1626         struct nd_msg *msg;
1627         const struct ipv6hdr *iphdr;
1628         const struct in6_addr *daddr;
1629         struct neighbour *n;
1630         struct inet6_dev *in6_dev;
1631
1632         in6_dev = __in6_dev_get(dev);
1633         if (!in6_dev)
1634                 goto out;
1635
1636         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr) + sizeof(struct nd_msg)))
1637                 goto out;
1638
1639         iphdr = ipv6_hdr(skb);
1640         daddr = &iphdr->daddr;
1641
1642         msg = (struct nd_msg *)(iphdr + 1);
1643         if (msg->icmph.icmp6_code != 0 ||
1644             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1645                 goto out;
1646
1647         if (ipv6_addr_loopback(daddr) ||
1648             ipv6_addr_is_multicast(&msg->target))
1649                 goto out;
1650
1651         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1652
1653         if (n) {
1654                 struct vxlan_fdb *f;
1655                 struct sk_buff *reply;
1656
1657                 if (!(n->nud_state & NUD_CONNECTED)) {
1658                         neigh_release(n);
1659                         goto out;
1660                 }
1661
1662                 f = vxlan_find_mac(vxlan, n->ha, vni);
1663                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1664                         /* bridge-local neighbor */
1665                         neigh_release(n);
1666                         goto out;
1667                 }
1668
1669                 reply = vxlan_na_create(skb, n,
1670                                         !!(f ? f->flags & NTF_ROUTER : 0));
1671
1672                 neigh_release(n);
1673
1674                 if (reply == NULL)
1675                         goto out;
1676
1677                 if (netif_rx_ni(reply) == NET_RX_DROP)
1678                         dev->stats.rx_dropped++;
1679
1680         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1681                 union vxlan_addr ipa = {
1682                         .sin6.sin6_addr = msg->target,
1683                         .sin6.sin6_family = AF_INET6,
1684                 };
1685
1686                 vxlan_ip_miss(dev, &ipa);
1687         }
1688
1689 out:
1690         consume_skb(skb);
1691         return NETDEV_TX_OK;
1692 }
1693 #endif
1694
1695 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1696 {
1697         struct vxlan_dev *vxlan = netdev_priv(dev);
1698         struct neighbour *n;
1699
1700         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1701                 return false;
1702
1703         n = NULL;
1704         switch (ntohs(eth_hdr(skb)->h_proto)) {
1705         case ETH_P_IP:
1706         {
1707                 struct iphdr *pip;
1708
1709                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1710                         return false;
1711                 pip = ip_hdr(skb);
1712                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1713                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1714                         union vxlan_addr ipa = {
1715                                 .sin.sin_addr.s_addr = pip->daddr,
1716                                 .sin.sin_family = AF_INET,
1717                         };
1718
1719                         vxlan_ip_miss(dev, &ipa);
1720                         return false;
1721                 }
1722
1723                 break;
1724         }
1725 #if IS_ENABLED(CONFIG_IPV6)
1726         case ETH_P_IPV6:
1727         {
1728                 struct ipv6hdr *pip6;
1729
1730                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1731                         return false;
1732                 pip6 = ipv6_hdr(skb);
1733                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1734                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1735                         union vxlan_addr ipa = {
1736                                 .sin6.sin6_addr = pip6->daddr,
1737                                 .sin6.sin6_family = AF_INET6,
1738                         };
1739
1740                         vxlan_ip_miss(dev, &ipa);
1741                         return false;
1742                 }
1743
1744                 break;
1745         }
1746 #endif
1747         default:
1748                 return false;
1749         }
1750
1751         if (n) {
1752                 bool diff;
1753
1754                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1755                 if (diff) {
1756                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1757                                 dev->addr_len);
1758                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1759                 }
1760                 neigh_release(n);
1761                 return diff;
1762         }
1763
1764         return false;
1765 }
1766
1767 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1768                                 struct vxlan_metadata *md)
1769 {
1770         struct vxlanhdr_gbp *gbp;
1771
1772         if (!md->gbp)
1773                 return;
1774
1775         gbp = (struct vxlanhdr_gbp *)vxh;
1776         vxh->vx_flags |= VXLAN_HF_GBP;
1777
1778         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1779                 gbp->dont_learn = 1;
1780
1781         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1782                 gbp->policy_applied = 1;
1783
1784         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1785 }
1786
1787 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1788                                __be16 protocol)
1789 {
1790         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1791
1792         gpe->np_applied = 1;
1793         gpe->next_protocol = tun_p_from_eth_p(protocol);
1794         if (!gpe->next_protocol)
1795                 return -EPFNOSUPPORT;
1796         return 0;
1797 }
1798
1799 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1800                            int iphdr_len, __be32 vni,
1801                            struct vxlan_metadata *md, u32 vxflags,
1802                            bool udp_sum)
1803 {
1804         struct vxlanhdr *vxh;
1805         int min_headroom;
1806         int err;
1807         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1808         __be16 inner_protocol = htons(ETH_P_TEB);
1809
1810         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1811             skb->ip_summed == CHECKSUM_PARTIAL) {
1812                 int csum_start = skb_checksum_start_offset(skb);
1813
1814                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1815                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1816                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1817                      skb->csum_offset == offsetof(struct tcphdr, check)))
1818                         type |= SKB_GSO_TUNNEL_REMCSUM;
1819         }
1820
1821         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1822                         + VXLAN_HLEN + iphdr_len;
1823
1824         /* Need space for new headers (invalidates iph ptr) */
1825         err = skb_cow_head(skb, min_headroom);
1826         if (unlikely(err))
1827                 return err;
1828
1829         err = iptunnel_handle_offloads(skb, type);
1830         if (err)
1831                 return err;
1832
1833         vxh = __skb_push(skb, sizeof(*vxh));
1834         vxh->vx_flags = VXLAN_HF_VNI;
1835         vxh->vx_vni = vxlan_vni_field(vni);
1836
1837         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1838                 unsigned int start;
1839
1840                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1841                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1842                 vxh->vx_flags |= VXLAN_HF_RCO;
1843
1844                 if (!skb_is_gso(skb)) {
1845                         skb->ip_summed = CHECKSUM_NONE;
1846                         skb->encapsulation = 0;
1847                 }
1848         }
1849
1850         if (vxflags & VXLAN_F_GBP)
1851                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1852         if (vxflags & VXLAN_F_GPE) {
1853                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1854                 if (err < 0)
1855                         return err;
1856                 inner_protocol = skb->protocol;
1857         }
1858
1859         skb_set_inner_protocol(skb, inner_protocol);
1860         return 0;
1861 }
1862
1863 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1864                                       struct vxlan_sock *sock4,
1865                                       struct sk_buff *skb, int oif, u8 tos,
1866                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
1867                                       struct dst_cache *dst_cache,
1868                                       const struct ip_tunnel_info *info)
1869 {
1870         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1871         struct rtable *rt = NULL;
1872         struct flowi4 fl4;
1873
1874         if (!sock4)
1875                 return ERR_PTR(-EIO);
1876
1877         if (tos && !info)
1878                 use_cache = false;
1879         if (use_cache) {
1880                 rt = dst_cache_get_ip4(dst_cache, saddr);
1881                 if (rt)
1882                         return rt;
1883         }
1884
1885         memset(&fl4, 0, sizeof(fl4));
1886         fl4.flowi4_oif = oif;
1887         fl4.flowi4_tos = RT_TOS(tos);
1888         fl4.flowi4_mark = skb->mark;
1889         fl4.flowi4_proto = IPPROTO_UDP;
1890         fl4.daddr = daddr;
1891         fl4.saddr = *saddr;
1892         fl4.fl4_dport = dport;
1893         fl4.fl4_sport = sport;
1894
1895         rt = ip_route_output_key(vxlan->net, &fl4);
1896         if (likely(!IS_ERR(rt))) {
1897                 if (rt->dst.dev == dev) {
1898                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1899                         ip_rt_put(rt);
1900                         return ERR_PTR(-ELOOP);
1901                 }
1902
1903                 *saddr = fl4.saddr;
1904                 if (use_cache)
1905                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1906         } else {
1907                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1908                 return ERR_PTR(-ENETUNREACH);
1909         }
1910         return rt;
1911 }
1912
1913 #if IS_ENABLED(CONFIG_IPV6)
1914 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1915                                           struct net_device *dev,
1916                                           struct vxlan_sock *sock6,
1917                                           struct sk_buff *skb, int oif, u8 tos,
1918                                           __be32 label,
1919                                           const struct in6_addr *daddr,
1920                                           struct in6_addr *saddr,
1921                                           __be16 dport, __be16 sport,
1922                                           struct dst_cache *dst_cache,
1923                                           const struct ip_tunnel_info *info)
1924 {
1925         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1926         struct dst_entry *ndst;
1927         struct flowi6 fl6;
1928         int err;
1929
1930         if (!sock6)
1931                 return ERR_PTR(-EIO);
1932
1933         if (tos && !info)
1934                 use_cache = false;
1935         if (use_cache) {
1936                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1937                 if (ndst)
1938                         return ndst;
1939         }
1940
1941         memset(&fl6, 0, sizeof(fl6));
1942         fl6.flowi6_oif = oif;
1943         fl6.daddr = *daddr;
1944         fl6.saddr = *saddr;
1945         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1946         fl6.flowi6_mark = skb->mark;
1947         fl6.flowi6_proto = IPPROTO_UDP;
1948         fl6.fl6_dport = dport;
1949         fl6.fl6_sport = sport;
1950
1951         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1952                                          sock6->sock->sk,
1953                                          &ndst, &fl6);
1954         if (unlikely(err < 0)) {
1955                 netdev_dbg(dev, "no route to %pI6\n", daddr);
1956                 return ERR_PTR(-ENETUNREACH);
1957         }
1958
1959         if (unlikely(ndst->dev == dev)) {
1960                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
1961                 dst_release(ndst);
1962                 return ERR_PTR(-ELOOP);
1963         }
1964
1965         *saddr = fl6.saddr;
1966         if (use_cache)
1967                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1968         return ndst;
1969 }
1970 #endif
1971
1972 /* Bypass encapsulation if the destination is local */
1973 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1974                                struct vxlan_dev *dst_vxlan, __be32 vni)
1975 {
1976         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1977         union vxlan_addr loopback;
1978         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1979         struct net_device *dev = skb->dev;
1980         int len = skb->len;
1981
1982         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1983         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1984         skb->pkt_type = PACKET_HOST;
1985         skb->encapsulation = 0;
1986         skb->dev = dst_vxlan->dev;
1987         __skb_pull(skb, skb_network_offset(skb));
1988
1989         if (remote_ip->sa.sa_family == AF_INET) {
1990                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1991                 loopback.sa.sa_family =  AF_INET;
1992 #if IS_ENABLED(CONFIG_IPV6)
1993         } else {
1994                 loopback.sin6.sin6_addr = in6addr_loopback;
1995                 loopback.sa.sa_family =  AF_INET6;
1996 #endif
1997         }
1998
1999         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
2000                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source, 0,
2001                             vni);
2002
2003         u64_stats_update_begin(&tx_stats->syncp);
2004         tx_stats->tx_packets++;
2005         tx_stats->tx_bytes += len;
2006         u64_stats_update_end(&tx_stats->syncp);
2007
2008         if (netif_rx(skb) == NET_RX_SUCCESS) {
2009                 u64_stats_update_begin(&rx_stats->syncp);
2010                 rx_stats->rx_packets++;
2011                 rx_stats->rx_bytes += len;
2012                 u64_stats_update_end(&rx_stats->syncp);
2013         } else {
2014                 dev->stats.rx_dropped++;
2015         }
2016 }
2017
2018 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2019                                  struct vxlan_dev *vxlan,
2020                                  union vxlan_addr *daddr,
2021                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2022                                  struct dst_entry *dst,
2023                                  u32 rt_flags)
2024 {
2025 #if IS_ENABLED(CONFIG_IPV6)
2026         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2027          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2028          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2029          */
2030         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2031 #endif
2032         /* Bypass encapsulation if the destination is local */
2033         if (rt_flags & RTCF_LOCAL &&
2034             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2035                 struct vxlan_dev *dst_vxlan;
2036
2037                 dst_release(dst);
2038                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2039                                            daddr->sa.sa_family, dst_port,
2040                                            vxlan->cfg.flags);
2041                 if (!dst_vxlan) {
2042                         dev->stats.tx_errors++;
2043                         kfree_skb(skb);
2044
2045                         return -ENOENT;
2046                 }
2047                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2048                 return 1;
2049         }
2050
2051         return 0;
2052 }
2053
2054 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2055                            __be32 default_vni, struct vxlan_rdst *rdst,
2056                            bool did_rsc)
2057 {
2058         struct dst_cache *dst_cache;
2059         struct ip_tunnel_info *info;
2060         struct vxlan_dev *vxlan = netdev_priv(dev);
2061         const struct iphdr *old_iph = ip_hdr(skb);
2062         union vxlan_addr *dst;
2063         union vxlan_addr remote_ip, local_ip;
2064         struct vxlan_metadata _md;
2065         struct vxlan_metadata *md = &_md;
2066         __be16 src_port = 0, dst_port;
2067         struct dst_entry *ndst = NULL;
2068         __be32 vni, label;
2069         __u8 tos, ttl;
2070         int ifindex;
2071         int err;
2072         u32 flags = vxlan->cfg.flags;
2073         bool udp_sum = false;
2074         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2075
2076         info = skb_tunnel_info(skb);
2077
2078         if (rdst) {
2079                 dst = &rdst->remote_ip;
2080                 if (vxlan_addr_any(dst)) {
2081                         if (did_rsc) {
2082                                 /* short-circuited back to local bridge */
2083                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2084                                 return;
2085                         }
2086                         goto drop;
2087                 }
2088
2089                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2090                 vni = (rdst->remote_vni) ? : default_vni;
2091                 ifindex = rdst->remote_ifindex;
2092                 local_ip = vxlan->cfg.saddr;
2093                 dst_cache = &rdst->dst_cache;
2094                 md->gbp = skb->mark;
2095                 ttl = vxlan->cfg.ttl;
2096                 if (!ttl && vxlan_addr_multicast(dst))
2097                         ttl = 1;
2098
2099                 tos = vxlan->cfg.tos;
2100                 if (tos == 1)
2101                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2102
2103                 if (dst->sa.sa_family == AF_INET)
2104                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2105                 else
2106                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2107                 label = vxlan->cfg.label;
2108         } else {
2109                 if (!info) {
2110                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2111                                   dev->name);
2112                         goto drop;
2113                 }
2114                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2115                 if (remote_ip.sa.sa_family == AF_INET) {
2116                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2117                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2118                 } else {
2119                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2120                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2121                 }
2122                 dst = &remote_ip;
2123                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2124                 vni = tunnel_id_to_key32(info->key.tun_id);
2125                 ifindex = 0;
2126                 dst_cache = &info->dst_cache;
2127                 if (info->options_len)
2128                         md = ip_tunnel_info_opts(info);
2129                 ttl = info->key.ttl;
2130                 tos = info->key.tos;
2131                 label = info->key.label;
2132                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2133         }
2134         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2135                                      vxlan->cfg.port_max, true);
2136
2137         rcu_read_lock();
2138         if (dst->sa.sa_family == AF_INET) {
2139                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2140                 struct rtable *rt;
2141                 __be16 df = 0;
2142
2143                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2144                                      dst->sin.sin_addr.s_addr,
2145                                      &local_ip.sin.sin_addr.s_addr,
2146                                      dst_port, src_port,
2147                                      dst_cache, info);
2148                 if (IS_ERR(rt)) {
2149                         err = PTR_ERR(rt);
2150                         goto tx_error;
2151                 }
2152
2153                 /* Bypass encapsulation if the destination is local */
2154                 if (!info) {
2155                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2156                                                     dst_port, ifindex, vni,
2157                                                     &rt->dst, rt->rt_flags);
2158                         if (err)
2159                                 goto out_unlock;
2160                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2161                         df = htons(IP_DF);
2162                 }
2163
2164                 ndst = &rt->dst;
2165                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2166                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2167                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2168                                       vni, md, flags, udp_sum);
2169                 if (err < 0)
2170                         goto tx_error;
2171
2172                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2173                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2174                                     src_port, dst_port, xnet, !udp_sum);
2175 #if IS_ENABLED(CONFIG_IPV6)
2176         } else {
2177                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2178
2179                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2180                                         label, &dst->sin6.sin6_addr,
2181                                         &local_ip.sin6.sin6_addr,
2182                                         dst_port, src_port,
2183                                         dst_cache, info);
2184                 if (IS_ERR(ndst)) {
2185                         err = PTR_ERR(ndst);
2186                         ndst = NULL;
2187                         goto tx_error;
2188                 }
2189
2190                 if (!info) {
2191                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2192
2193                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2194                                                     dst_port, ifindex, vni,
2195                                                     ndst, rt6i_flags);
2196                         if (err)
2197                                 goto out_unlock;
2198                 }
2199
2200                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2201                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2202                 skb_scrub_packet(skb, xnet);
2203                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2204                                       vni, md, flags, udp_sum);
2205                 if (err < 0)
2206                         goto tx_error;
2207
2208                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2209                                      &local_ip.sin6.sin6_addr,
2210                                      &dst->sin6.sin6_addr, tos, ttl,
2211                                      label, src_port, dst_port, !udp_sum);
2212 #endif
2213         }
2214 out_unlock:
2215         rcu_read_unlock();
2216         return;
2217
2218 drop:
2219         dev->stats.tx_dropped++;
2220         dev_kfree_skb(skb);
2221         return;
2222
2223 tx_error:
2224         rcu_read_unlock();
2225         if (err == -ELOOP)
2226                 dev->stats.collisions++;
2227         else if (err == -ENETUNREACH)
2228                 dev->stats.tx_carrier_errors++;
2229         dst_release(ndst);
2230         dev->stats.tx_errors++;
2231         kfree_skb(skb);
2232 }
2233
2234 /* Transmit local packets over Vxlan
2235  *
2236  * Outer IP header inherits ECN and DF from inner header.
2237  * Outer UDP destination is the VXLAN assigned port.
2238  *           source port is based on hash of flow
2239  */
2240 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2241 {
2242         struct vxlan_dev *vxlan = netdev_priv(dev);
2243         const struct ip_tunnel_info *info;
2244         struct ethhdr *eth;
2245         bool did_rsc = false;
2246         struct vxlan_rdst *rdst, *fdst = NULL;
2247         struct vxlan_fdb *f;
2248         __be32 vni = 0;
2249
2250         info = skb_tunnel_info(skb);
2251
2252         skb_reset_mac_header(skb);
2253
2254         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2255                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2256                     info->mode & IP_TUNNEL_INFO_TX) {
2257                         vni = tunnel_id_to_key32(info->key.tun_id);
2258                 } else {
2259                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2260                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2261                         else
2262                                 kfree_skb(skb);
2263                         return NETDEV_TX_OK;
2264                 }
2265         }
2266
2267         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2268                 eth = eth_hdr(skb);
2269                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2270                         return arp_reduce(dev, skb, vni);
2271 #if IS_ENABLED(CONFIG_IPV6)
2272                 else if (ntohs(eth->h_proto) == ETH_P_IPV6) {
2273                         struct ipv6hdr *hdr, _hdr;
2274                         if ((hdr = skb_header_pointer(skb,
2275                                                       skb_network_offset(skb),
2276                                                       sizeof(_hdr), &_hdr)) &&
2277                             hdr->nexthdr == IPPROTO_ICMPV6)
2278                                 return neigh_reduce(dev, skb, vni);
2279                 }
2280 #endif
2281         }
2282
2283         eth = eth_hdr(skb);
2284         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2285         did_rsc = false;
2286
2287         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2288             (ntohs(eth->h_proto) == ETH_P_IP ||
2289              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2290                 did_rsc = route_shortcircuit(dev, skb);
2291                 if (did_rsc)
2292                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2293         }
2294
2295         if (f == NULL) {
2296                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2297                 if (f == NULL) {
2298                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2299                             !is_multicast_ether_addr(eth->h_dest))
2300                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2301
2302                         dev->stats.tx_dropped++;
2303                         kfree_skb(skb);
2304                         return NETDEV_TX_OK;
2305                 }
2306         }
2307
2308         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2309                 struct sk_buff *skb1;
2310
2311                 if (!fdst) {
2312                         fdst = rdst;
2313                         continue;
2314                 }
2315                 skb1 = skb_clone(skb, GFP_ATOMIC);
2316                 if (skb1)
2317                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2318         }
2319
2320         if (fdst)
2321                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2322         else
2323                 kfree_skb(skb);
2324         return NETDEV_TX_OK;
2325 }
2326
2327 /* Walk the forwarding table and purge stale entries */
2328 static void vxlan_cleanup(unsigned long arg)
2329 {
2330         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2331         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2332         unsigned int h;
2333
2334         if (!netif_running(vxlan->dev))
2335                 return;
2336
2337         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2338                 struct hlist_node *p, *n;
2339
2340                 spin_lock_bh(&vxlan->hash_lock);
2341                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2342                         struct vxlan_fdb *f
2343                                 = container_of(p, struct vxlan_fdb, hlist);
2344                         unsigned long timeout;
2345
2346                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2347                                 continue;
2348
2349                         if (f->flags & NTF_EXT_LEARNED)
2350                                 continue;
2351
2352                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2353                         if (time_before_eq(timeout, jiffies)) {
2354                                 netdev_dbg(vxlan->dev,
2355                                            "garbage collect %pM\n",
2356                                            f->eth_addr);
2357                                 f->state = NUD_STALE;
2358                                 vxlan_fdb_destroy(vxlan, f);
2359                         } else if (time_before(timeout, next_timer))
2360                                 next_timer = timeout;
2361                 }
2362                 spin_unlock_bh(&vxlan->hash_lock);
2363         }
2364
2365         mod_timer(&vxlan->age_timer, next_timer);
2366 }
2367
2368 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2369 {
2370         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2371
2372         spin_lock(&vn->sock_lock);
2373         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2374 #if IS_ENABLED(CONFIG_IPV6)
2375         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2376 #endif
2377         spin_unlock(&vn->sock_lock);
2378 }
2379
2380 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2381                              struct vxlan_dev_node *node)
2382 {
2383         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2384         __be32 vni = vxlan->default_dst.remote_vni;
2385
2386         node->vxlan = vxlan;
2387         spin_lock(&vn->sock_lock);
2388         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2389         spin_unlock(&vn->sock_lock);
2390 }
2391
2392 /* Setup stats when device is created */
2393 static int vxlan_init(struct net_device *dev)
2394 {
2395         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2396         if (!dev->tstats)
2397                 return -ENOMEM;
2398
2399         return 0;
2400 }
2401
2402 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2403 {
2404         struct vxlan_fdb *f;
2405
2406         spin_lock_bh(&vxlan->hash_lock);
2407         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2408         if (f)
2409                 vxlan_fdb_destroy(vxlan, f);
2410         spin_unlock_bh(&vxlan->hash_lock);
2411 }
2412
2413 static void vxlan_uninit(struct net_device *dev)
2414 {
2415         struct vxlan_dev *vxlan = netdev_priv(dev);
2416
2417         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2418
2419         free_percpu(dev->tstats);
2420 }
2421
2422 /* Start ageing timer and join group when device is brought up */
2423 static int vxlan_open(struct net_device *dev)
2424 {
2425         struct vxlan_dev *vxlan = netdev_priv(dev);
2426         int ret;
2427
2428         ret = vxlan_sock_add(vxlan);
2429         if (ret < 0)
2430                 return ret;
2431
2432         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2433                 ret = vxlan_igmp_join(vxlan);
2434                 if (ret == -EADDRINUSE)
2435                         ret = 0;
2436                 if (ret) {
2437                         vxlan_sock_release(vxlan);
2438                         return ret;
2439                 }
2440         }
2441
2442         if (vxlan->cfg.age_interval)
2443                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2444
2445         return ret;
2446 }
2447
2448 /* Purge the forwarding table */
2449 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2450 {
2451         unsigned int h;
2452
2453         spin_lock_bh(&vxlan->hash_lock);
2454         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2455                 struct hlist_node *p, *n;
2456                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2457                         struct vxlan_fdb *f
2458                                 = container_of(p, struct vxlan_fdb, hlist);
2459                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2460                                 continue;
2461                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2462                         if (!is_zero_ether_addr(f->eth_addr))
2463                                 vxlan_fdb_destroy(vxlan, f);
2464                 }
2465         }
2466         spin_unlock_bh(&vxlan->hash_lock);
2467 }
2468
2469 /* Cleanup timer and forwarding table on shutdown */
2470 static int vxlan_stop(struct net_device *dev)
2471 {
2472         struct vxlan_dev *vxlan = netdev_priv(dev);
2473         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2474         int ret = 0;
2475
2476         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2477             !vxlan_group_used(vn, vxlan))
2478                 ret = vxlan_igmp_leave(vxlan);
2479
2480         del_timer_sync(&vxlan->age_timer);
2481
2482         vxlan_flush(vxlan, false);
2483         vxlan_sock_release(vxlan);
2484
2485         return ret;
2486 }
2487
2488 /* Stub, nothing needs to be done. */
2489 static void vxlan_set_multicast_list(struct net_device *dev)
2490 {
2491 }
2492
2493 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2494 {
2495         struct vxlan_dev *vxlan = netdev_priv(dev);
2496         struct vxlan_rdst *dst = &vxlan->default_dst;
2497         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2498                                                          dst->remote_ifindex);
2499         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
2500
2501         /* This check is different than dev->max_mtu, because it looks at
2502          * the lowerdev->mtu, rather than the static dev->max_mtu
2503          */
2504         if (lowerdev) {
2505                 int max_mtu = lowerdev->mtu -
2506                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2507                 if (new_mtu > max_mtu)
2508                         return -EINVAL;
2509         }
2510
2511         dev->mtu = new_mtu;
2512         return 0;
2513 }
2514
2515 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2516 {
2517         struct vxlan_dev *vxlan = netdev_priv(dev);
2518         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2519         __be16 sport, dport;
2520
2521         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2522                                   vxlan->cfg.port_max, true);
2523         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2524
2525         if (ip_tunnel_info_af(info) == AF_INET) {
2526                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2527                 struct rtable *rt;
2528
2529                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2530                                      info->key.u.ipv4.dst,
2531                                      &info->key.u.ipv4.src, dport, sport,
2532                                      &info->dst_cache, info);
2533                 if (IS_ERR(rt))
2534                         return PTR_ERR(rt);
2535                 ip_rt_put(rt);
2536         } else {
2537 #if IS_ENABLED(CONFIG_IPV6)
2538                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2539                 struct dst_entry *ndst;
2540
2541                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2542                                         info->key.label, &info->key.u.ipv6.dst,
2543                                         &info->key.u.ipv6.src, dport, sport,
2544                                         &info->dst_cache, info);
2545                 if (IS_ERR(ndst))
2546                         return PTR_ERR(ndst);
2547                 dst_release(ndst);
2548 #else /* !CONFIG_IPV6 */
2549                 return -EPFNOSUPPORT;
2550 #endif
2551         }
2552         info->key.tp_src = sport;
2553         info->key.tp_dst = dport;
2554         return 0;
2555 }
2556
2557 static const struct net_device_ops vxlan_netdev_ether_ops = {
2558         .ndo_init               = vxlan_init,
2559         .ndo_uninit             = vxlan_uninit,
2560         .ndo_open               = vxlan_open,
2561         .ndo_stop               = vxlan_stop,
2562         .ndo_start_xmit         = vxlan_xmit,
2563         .ndo_get_stats64        = ip_tunnel_get_stats64,
2564         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2565         .ndo_change_mtu         = vxlan_change_mtu,
2566         .ndo_validate_addr      = eth_validate_addr,
2567         .ndo_set_mac_address    = eth_mac_addr,
2568         .ndo_fdb_add            = vxlan_fdb_add,
2569         .ndo_fdb_del            = vxlan_fdb_delete,
2570         .ndo_fdb_dump           = vxlan_fdb_dump,
2571         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2572 };
2573
2574 static const struct net_device_ops vxlan_netdev_raw_ops = {
2575         .ndo_init               = vxlan_init,
2576         .ndo_uninit             = vxlan_uninit,
2577         .ndo_open               = vxlan_open,
2578         .ndo_stop               = vxlan_stop,
2579         .ndo_start_xmit         = vxlan_xmit,
2580         .ndo_get_stats64        = ip_tunnel_get_stats64,
2581         .ndo_change_mtu         = vxlan_change_mtu,
2582         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2583 };
2584
2585 /* Info for udev, that this is a virtual tunnel endpoint */
2586 static struct device_type vxlan_type = {
2587         .name = "vxlan",
2588 };
2589
2590 /* Calls the ndo_udp_tunnel_add of the caller in order to
2591  * supply the listening VXLAN udp ports. Callers are expected
2592  * to implement the ndo_udp_tunnel_add.
2593  */
2594 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
2595 {
2596         struct vxlan_sock *vs;
2597         struct net *net = dev_net(dev);
2598         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2599         unsigned int i;
2600
2601         spin_lock(&vn->sock_lock);
2602         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2603                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2604                         unsigned short type;
2605
2606                         if (vs->flags & VXLAN_F_GPE)
2607                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
2608                         else
2609                                 type = UDP_TUNNEL_TYPE_VXLAN;
2610
2611                         if (push)
2612                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
2613                         else
2614                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
2615                 }
2616         }
2617         spin_unlock(&vn->sock_lock);
2618 }
2619
2620 /* Initialize the device structure. */
2621 static void vxlan_setup(struct net_device *dev)
2622 {
2623         struct vxlan_dev *vxlan = netdev_priv(dev);
2624         unsigned int h;
2625
2626         eth_hw_addr_random(dev);
2627         ether_setup(dev);
2628
2629         dev->needs_free_netdev = true;
2630         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2631
2632         dev->features   |= NETIF_F_LLTX;
2633         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2634         dev->features   |= NETIF_F_RXCSUM;
2635         dev->features   |= NETIF_F_GSO_SOFTWARE;
2636
2637         dev->vlan_features = dev->features;
2638         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2639         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2640         netif_keep_dst(dev);
2641         dev->priv_flags |= IFF_NO_QUEUE;
2642
2643         /* MTU range: 68 - 65535 */
2644         dev->min_mtu = ETH_MIN_MTU;
2645         dev->max_mtu = ETH_MAX_MTU;
2646
2647         INIT_LIST_HEAD(&vxlan->next);
2648         spin_lock_init(&vxlan->hash_lock);
2649
2650         init_timer_deferrable(&vxlan->age_timer);
2651         vxlan->age_timer.function = vxlan_cleanup;
2652         vxlan->age_timer.data = (unsigned long) vxlan;
2653
2654         vxlan->dev = dev;
2655
2656         gro_cells_init(&vxlan->gro_cells, dev);
2657
2658         for (h = 0; h < FDB_HASH_SIZE; ++h)
2659                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2660 }
2661
2662 static void vxlan_ether_setup(struct net_device *dev)
2663 {
2664         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2665         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2666         dev->netdev_ops = &vxlan_netdev_ether_ops;
2667 }
2668
2669 static void vxlan_raw_setup(struct net_device *dev)
2670 {
2671         dev->header_ops = NULL;
2672         dev->type = ARPHRD_NONE;
2673         dev->hard_header_len = 0;
2674         dev->addr_len = 0;
2675         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2676         dev->netdev_ops = &vxlan_netdev_raw_ops;
2677 }
2678
2679 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2680         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2681         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2682         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2683         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2684         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2685         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2686         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2687         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2688         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2689         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2690         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2691         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2692         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2693         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2694         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2695         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2696         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2697         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2698         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2699         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2700         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2701         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2702         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2703         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2704         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2705         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2706         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2707 };
2708
2709 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
2710                           struct netlink_ext_ack *extack)
2711 {
2712         if (tb[IFLA_ADDRESS]) {
2713                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2714                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2715                                             "Provided link layer address is not Ethernet");
2716                         return -EINVAL;
2717                 }
2718
2719                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2720                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2721                                             "Provided Ethernet address is not unicast");
2722                         return -EADDRNOTAVAIL;
2723                 }
2724         }
2725
2726         if (tb[IFLA_MTU]) {
2727                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
2728
2729                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
2730                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
2731                                             "MTU must be between 68 and 65535");
2732                         return -EINVAL;
2733                 }
2734         }
2735
2736         if (!data) {
2737                 NL_SET_ERR_MSG(extack,
2738                                "Required attributes not provided to perform the operation");
2739                 return -EINVAL;
2740         }
2741
2742         if (data[IFLA_VXLAN_ID]) {
2743                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2744
2745                 if (id >= VXLAN_N_VID) {
2746                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID],
2747                                             "VXLAN ID must be lower than 16777216");
2748                         return -ERANGE;
2749                 }
2750         }
2751
2752         if (data[IFLA_VXLAN_PORT_RANGE]) {
2753                 const struct ifla_vxlan_port_range *p
2754                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2755
2756                 if (ntohs(p->high) < ntohs(p->low)) {
2757                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
2758                                             "Invalid source port range");
2759                         return -EINVAL;
2760                 }
2761         }
2762
2763         return 0;
2764 }
2765
2766 static void vxlan_get_drvinfo(struct net_device *netdev,
2767                               struct ethtool_drvinfo *drvinfo)
2768 {
2769         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2770         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2771 }
2772
2773 static const struct ethtool_ops vxlan_ethtool_ops = {
2774         .get_drvinfo    = vxlan_get_drvinfo,
2775         .get_link       = ethtool_op_get_link,
2776 };
2777
2778 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2779                                         __be16 port, u32 flags)
2780 {
2781         struct socket *sock;
2782         struct udp_port_cfg udp_conf;
2783         int err;
2784
2785         memset(&udp_conf, 0, sizeof(udp_conf));
2786
2787         if (ipv6) {
2788                 udp_conf.family = AF_INET6;
2789                 udp_conf.use_udp6_rx_checksums =
2790                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2791                 udp_conf.ipv6_v6only = 1;
2792         } else {
2793                 udp_conf.family = AF_INET;
2794         }
2795
2796         udp_conf.local_udp_port = port;
2797
2798         /* Open UDP socket */
2799         err = udp_sock_create(net, &udp_conf, &sock);
2800         if (err < 0)
2801                 return ERR_PTR(err);
2802
2803         return sock;
2804 }
2805
2806 /* Create new listen socket if needed */
2807 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2808                                               __be16 port, u32 flags)
2809 {
2810         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2811         struct vxlan_sock *vs;
2812         struct socket *sock;
2813         unsigned int h;
2814         struct udp_tunnel_sock_cfg tunnel_cfg;
2815
2816         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2817         if (!vs)
2818                 return ERR_PTR(-ENOMEM);
2819
2820         for (h = 0; h < VNI_HASH_SIZE; ++h)
2821                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2822
2823         sock = vxlan_create_sock(net, ipv6, port, flags);
2824         if (IS_ERR(sock)) {
2825                 kfree(vs);
2826                 return ERR_CAST(sock);
2827         }
2828
2829         vs->sock = sock;
2830         refcount_set(&vs->refcnt, 1);
2831         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2832
2833         spin_lock(&vn->sock_lock);
2834         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2835         udp_tunnel_notify_add_rx_port(sock,
2836                                       (vs->flags & VXLAN_F_GPE) ?
2837                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2838                                       UDP_TUNNEL_TYPE_VXLAN);
2839         spin_unlock(&vn->sock_lock);
2840
2841         /* Mark socket as an encapsulation socket. */
2842         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2843         tunnel_cfg.sk_user_data = vs;
2844         tunnel_cfg.encap_type = 1;
2845         tunnel_cfg.encap_rcv = vxlan_rcv;
2846         tunnel_cfg.encap_destroy = NULL;
2847         tunnel_cfg.gro_receive = vxlan_gro_receive;
2848         tunnel_cfg.gro_complete = vxlan_gro_complete;
2849
2850         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2851
2852         return vs;
2853 }
2854
2855 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2856 {
2857         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2858         struct vxlan_sock *vs = NULL;
2859         struct vxlan_dev_node *node;
2860
2861         if (!vxlan->cfg.no_share) {
2862                 spin_lock(&vn->sock_lock);
2863                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2864                                      vxlan->cfg.dst_port, vxlan->cfg.flags);
2865                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
2866                         spin_unlock(&vn->sock_lock);
2867                         return -EBUSY;
2868                 }
2869                 spin_unlock(&vn->sock_lock);
2870         }
2871         if (!vs)
2872                 vs = vxlan_socket_create(vxlan->net, ipv6,
2873                                          vxlan->cfg.dst_port, vxlan->cfg.flags);
2874         if (IS_ERR(vs))
2875                 return PTR_ERR(vs);
2876 #if IS_ENABLED(CONFIG_IPV6)
2877         if (ipv6) {
2878                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2879                 node = &vxlan->hlist6;
2880         } else
2881 #endif
2882         {
2883                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2884                 node = &vxlan->hlist4;
2885         }
2886         vxlan_vs_add_dev(vs, vxlan, node);
2887         return 0;
2888 }
2889
2890 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2891 {
2892         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
2893         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
2894         bool ipv4 = !ipv6 || metadata;
2895         int ret = 0;
2896
2897         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2898 #if IS_ENABLED(CONFIG_IPV6)
2899         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2900         if (ipv6) {
2901                 ret = __vxlan_sock_add(vxlan, true);
2902                 if (ret < 0 && ret != -EAFNOSUPPORT)
2903                         ipv4 = false;
2904         }
2905 #endif
2906         if (ipv4)
2907                 ret = __vxlan_sock_add(vxlan, false);
2908         if (ret < 0)
2909                 vxlan_sock_release(vxlan);
2910         return ret;
2911 }
2912
2913 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
2914                                  struct net_device **lower,
2915                                  struct vxlan_dev *old,
2916                                  struct netlink_ext_ack *extack)
2917 {
2918         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2919         struct vxlan_dev *tmp;
2920         bool use_ipv6 = false;
2921
2922         if (conf->flags & VXLAN_F_GPE) {
2923                 /* For now, allow GPE only together with
2924                  * COLLECT_METADATA. This can be relaxed later; in such
2925                  * case, the other side of the PtP link will have to be
2926                  * provided.
2927                  */
2928                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2929                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2930                         NL_SET_ERR_MSG(extack,
2931                                        "VXLAN GPE does not support this combination of attributes");
2932                         return -EINVAL;
2933                 }
2934         }
2935
2936         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
2937                 /* Unless IPv6 is explicitly requested, assume IPv4 */
2938                 conf->remote_ip.sa.sa_family = AF_INET;
2939                 conf->saddr.sa.sa_family = AF_INET;
2940         } else if (!conf->remote_ip.sa.sa_family) {
2941                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
2942         } else if (!conf->saddr.sa.sa_family) {
2943                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
2944         }
2945
2946         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
2947                 NL_SET_ERR_MSG(extack,
2948                                "Local and remote address must be from the same family");
2949                 return -EINVAL;
2950         }
2951
2952         if (vxlan_addr_multicast(&conf->saddr)) {
2953                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
2954                 return -EINVAL;
2955         }
2956
2957         if (conf->saddr.sa.sa_family == AF_INET6) {
2958                 if (!IS_ENABLED(CONFIG_IPV6)) {
2959                         NL_SET_ERR_MSG(extack,
2960                                        "IPv6 support not enabled in the kernel");
2961                         return -EPFNOSUPPORT;
2962                 }
2963                 use_ipv6 = true;
2964                 conf->flags |= VXLAN_F_IPV6;
2965
2966                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2967                         int local_type =
2968                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
2969                         int remote_type =
2970                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
2971
2972                         if (local_type & IPV6_ADDR_LINKLOCAL) {
2973                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
2974                                     (remote_type != IPV6_ADDR_ANY)) {
2975                                         NL_SET_ERR_MSG(extack,
2976                                                        "Invalid combination of local and remote address scopes");
2977                                         return -EINVAL;
2978                                 }
2979
2980                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
2981                         } else {
2982                                 if (remote_type ==
2983                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
2984                                         NL_SET_ERR_MSG(extack,
2985                                                        "Invalid combination of local and remote address scopes");
2986                                         return -EINVAL;
2987                                 }
2988
2989                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
2990                         }
2991                 }
2992         }
2993
2994         if (conf->label && !use_ipv6) {
2995                 NL_SET_ERR_MSG(extack,
2996                                "Label attribute only applies to IPv6 VXLAN devices");
2997                 return -EINVAL;
2998         }
2999
3000         if (conf->remote_ifindex) {
3001                 struct net_device *lowerdev;
3002
3003                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3004                 if (!lowerdev) {
3005                         NL_SET_ERR_MSG(extack,
3006                                        "Invalid local interface, device not found");
3007                         return -ENODEV;
3008                 }
3009
3010 #if IS_ENABLED(CONFIG_IPV6)
3011                 if (use_ipv6) {
3012                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3013                         if (idev && idev->cnf.disable_ipv6) {
3014                                 NL_SET_ERR_MSG(extack,
3015                                                "IPv6 support disabled by administrator");
3016                                 return -EPERM;
3017                         }
3018                 }
3019 #endif
3020
3021                 *lower = lowerdev;
3022         } else {
3023                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3024                         NL_SET_ERR_MSG(extack,
3025                                        "Local interface required for multicast remote destination");
3026
3027                         return -EINVAL;
3028                 }
3029
3030 #if IS_ENABLED(CONFIG_IPV6)
3031                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3032                         NL_SET_ERR_MSG(extack,
3033                                        "Local interface required for link-local local/remote addresses");
3034                         return -EINVAL;
3035                 }
3036 #endif
3037
3038                 *lower = NULL;
3039         }
3040
3041         if (!conf->dst_port) {
3042                 if (conf->flags & VXLAN_F_GPE)
3043                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3044                 else
3045                         conf->dst_port = htons(vxlan_port);
3046         }
3047
3048         if (!conf->age_interval)
3049                 conf->age_interval = FDB_AGE_DEFAULT;
3050
3051         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3052                 if (tmp == old)
3053                         continue;
3054
3055                 if (tmp->cfg.vni != conf->vni)
3056                         continue;
3057                 if (tmp->cfg.dst_port != conf->dst_port)
3058                         continue;
3059                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3060                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3061                         continue;
3062
3063                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3064                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3065                         continue;
3066
3067                 NL_SET_ERR_MSG(extack,
3068                                "A VXLAN device with the specified VNI already exists");
3069                 return -EEXIST;
3070         }
3071
3072         return 0;
3073 }
3074
3075 static void vxlan_config_apply(struct net_device *dev,
3076                                struct vxlan_config *conf,
3077                                struct net_device *lowerdev,
3078                                struct net *src_net,
3079                                bool changelink)
3080 {
3081         struct vxlan_dev *vxlan = netdev_priv(dev);
3082         struct vxlan_rdst *dst = &vxlan->default_dst;
3083         unsigned short needed_headroom = ETH_HLEN;
3084         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3085         int max_mtu = ETH_MAX_MTU;
3086
3087         if (!changelink) {
3088                 if (conf->flags & VXLAN_F_GPE)
3089                         vxlan_raw_setup(dev);
3090                 else
3091                         vxlan_ether_setup(dev);
3092
3093                 if (conf->mtu)
3094                         dev->mtu = conf->mtu;
3095
3096                 vxlan->net = src_net;
3097         }
3098
3099         dst->remote_vni = conf->vni;
3100
3101         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3102
3103         if (lowerdev) {
3104                 dst->remote_ifindex = conf->remote_ifindex;
3105
3106                 dev->gso_max_size = lowerdev->gso_max_size;
3107                 dev->gso_max_segs = lowerdev->gso_max_segs;
3108
3109                 needed_headroom = lowerdev->hard_header_len;
3110
3111                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3112                                            VXLAN_HEADROOM);
3113         }
3114
3115         if (dev->mtu > max_mtu)
3116                 dev->mtu = max_mtu;
3117
3118         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3119                 needed_headroom += VXLAN6_HEADROOM;
3120         else
3121                 needed_headroom += VXLAN_HEADROOM;
3122         dev->needed_headroom = needed_headroom;
3123
3124         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3125 }
3126
3127 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3128                                struct vxlan_config *conf, bool changelink,
3129                                struct netlink_ext_ack *extack)
3130 {
3131         struct vxlan_dev *vxlan = netdev_priv(dev);
3132         struct net_device *lowerdev;
3133         int ret;
3134
3135         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3136         if (ret)
3137                 return ret;
3138
3139         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3140
3141         return 0;
3142 }
3143
3144 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3145                               struct vxlan_config *conf,
3146                               struct netlink_ext_ack *extack)
3147 {
3148         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3149         struct vxlan_dev *vxlan = netdev_priv(dev);
3150         int err;
3151
3152         err = vxlan_dev_configure(net, dev, conf, false, extack);
3153         if (err)
3154                 return err;
3155
3156         dev->ethtool_ops = &vxlan_ethtool_ops;
3157
3158         /* create an fdb entry for a valid default destination */
3159         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
3160                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3161                                        &vxlan->default_dst.remote_ip,
3162                                        NUD_REACHABLE | NUD_PERMANENT,
3163                                        NLM_F_EXCL | NLM_F_CREATE,
3164                                        vxlan->cfg.dst_port,
3165                                        vxlan->default_dst.remote_vni,
3166                                        vxlan->default_dst.remote_vni,
3167                                        vxlan->default_dst.remote_ifindex,
3168                                        NTF_SELF);
3169                 if (err)
3170                         return err;
3171         }
3172
3173         err = register_netdevice(dev);
3174         if (err) {
3175                 vxlan_fdb_delete_default(vxlan, vxlan->default_dst.remote_vni);
3176                 return err;
3177         }
3178
3179         list_add(&vxlan->next, &vn->vxlan_list);
3180         return 0;
3181 }
3182
3183 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3184                          struct net_device *dev, struct vxlan_config *conf,
3185                          bool changelink)
3186 {
3187         struct vxlan_dev *vxlan = netdev_priv(dev);
3188
3189         memset(conf, 0, sizeof(*conf));
3190
3191         /* if changelink operation, start with old existing cfg */
3192         if (changelink)
3193                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3194
3195         if (data[IFLA_VXLAN_ID]) {
3196                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3197
3198                 if (changelink && (vni != conf->vni))
3199                         return -EOPNOTSUPP;
3200                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3201         }
3202
3203         if (data[IFLA_VXLAN_GROUP]) {
3204                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET))
3205                         return -EOPNOTSUPP;
3206
3207                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3208                 conf->remote_ip.sa.sa_family = AF_INET;
3209         } else if (data[IFLA_VXLAN_GROUP6]) {
3210                 if (!IS_ENABLED(CONFIG_IPV6))
3211                         return -EPFNOSUPPORT;
3212
3213                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6))
3214                         return -EOPNOTSUPP;
3215
3216                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3217                 conf->remote_ip.sa.sa_family = AF_INET6;
3218         }
3219
3220         if (data[IFLA_VXLAN_LOCAL]) {
3221                 if (changelink && (conf->saddr.sa.sa_family != AF_INET))
3222                         return -EOPNOTSUPP;
3223
3224                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3225                 conf->saddr.sa.sa_family = AF_INET;
3226         } else if (data[IFLA_VXLAN_LOCAL6]) {
3227                 if (!IS_ENABLED(CONFIG_IPV6))
3228                         return -EPFNOSUPPORT;
3229
3230                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6))
3231                         return -EOPNOTSUPP;
3232
3233                 /* TODO: respect scope id */
3234                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3235                 conf->saddr.sa.sa_family = AF_INET6;
3236         }
3237
3238         if (data[IFLA_VXLAN_LINK])
3239                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3240
3241         if (data[IFLA_VXLAN_TOS])
3242                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3243
3244         if (data[IFLA_VXLAN_TTL])
3245                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3246
3247         if (data[IFLA_VXLAN_LABEL])
3248                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3249                              IPV6_FLOWLABEL_MASK;
3250
3251         if (data[IFLA_VXLAN_LEARNING]) {
3252                 if (nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3253                         conf->flags |= VXLAN_F_LEARN;
3254                 else
3255                         conf->flags &= ~VXLAN_F_LEARN;
3256         } else if (!changelink) {
3257                 /* default to learn on a new device */
3258                 conf->flags |= VXLAN_F_LEARN;
3259         }
3260
3261         if (data[IFLA_VXLAN_AGEING]) {
3262                 if (changelink)
3263                         return -EOPNOTSUPP;
3264                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3265         }
3266
3267         if (data[IFLA_VXLAN_PROXY]) {
3268                 if (changelink)
3269                         return -EOPNOTSUPP;
3270                 if (nla_get_u8(data[IFLA_VXLAN_PROXY]))
3271                         conf->flags |= VXLAN_F_PROXY;
3272         }
3273
3274         if (data[IFLA_VXLAN_RSC]) {
3275                 if (changelink)
3276                         return -EOPNOTSUPP;
3277                 if (nla_get_u8(data[IFLA_VXLAN_RSC]))
3278                         conf->flags |= VXLAN_F_RSC;
3279         }
3280
3281         if (data[IFLA_VXLAN_L2MISS]) {
3282                 if (changelink)
3283                         return -EOPNOTSUPP;
3284                 if (nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3285                         conf->flags |= VXLAN_F_L2MISS;
3286         }
3287
3288         if (data[IFLA_VXLAN_L3MISS]) {
3289                 if (changelink)
3290                         return -EOPNOTSUPP;
3291                 if (nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3292                         conf->flags |= VXLAN_F_L3MISS;
3293         }
3294
3295         if (data[IFLA_VXLAN_LIMIT]) {
3296                 if (changelink)
3297                         return -EOPNOTSUPP;
3298                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3299         }
3300
3301         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3302                 if (changelink)
3303                         return -EOPNOTSUPP;
3304                 if (nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3305                         conf->flags |= VXLAN_F_COLLECT_METADATA;
3306         }
3307
3308         if (data[IFLA_VXLAN_PORT_RANGE]) {
3309                 if (!changelink) {
3310                         const struct ifla_vxlan_port_range *p
3311                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3312                         conf->port_min = ntohs(p->low);
3313                         conf->port_max = ntohs(p->high);
3314                 } else {
3315                         return -EOPNOTSUPP;
3316                 }
3317         }
3318
3319         if (data[IFLA_VXLAN_PORT]) {
3320                 if (changelink)
3321                         return -EOPNOTSUPP;
3322                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3323         }
3324
3325         if (data[IFLA_VXLAN_UDP_CSUM]) {
3326                 if (changelink)
3327                         return -EOPNOTSUPP;
3328                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3329                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3330         }
3331
3332         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3333                 if (changelink)
3334                         return -EOPNOTSUPP;
3335                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3336                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3337         }
3338
3339         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3340                 if (changelink)
3341                         return -EOPNOTSUPP;
3342                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3343                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3344         }
3345
3346         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3347                 if (changelink)
3348                         return -EOPNOTSUPP;
3349                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3350                         conf->flags |= VXLAN_F_REMCSUM_TX;
3351         }
3352
3353         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3354                 if (changelink)
3355                         return -EOPNOTSUPP;
3356                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3357                         conf->flags |= VXLAN_F_REMCSUM_RX;
3358         }
3359
3360         if (data[IFLA_VXLAN_GBP]) {
3361                 if (changelink)
3362                         return -EOPNOTSUPP;
3363                 conf->flags |= VXLAN_F_GBP;
3364         }
3365
3366         if (data[IFLA_VXLAN_GPE]) {
3367                 if (changelink)
3368                         return -EOPNOTSUPP;
3369                 conf->flags |= VXLAN_F_GPE;
3370         }
3371
3372         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3373                 if (changelink)
3374                         return -EOPNOTSUPP;
3375                 conf->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3376         }
3377
3378         if (tb[IFLA_MTU]) {
3379                 if (changelink)
3380                         return -EOPNOTSUPP;
3381                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3382         }
3383
3384         return 0;
3385 }
3386
3387 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3388                          struct nlattr *tb[], struct nlattr *data[],
3389                          struct netlink_ext_ack *extack)
3390 {
3391         struct vxlan_config conf;
3392         int err;
3393
3394         err = vxlan_nl2conf(tb, data, dev, &conf, false);
3395         if (err)
3396                 return err;
3397
3398         return __vxlan_dev_create(src_net, dev, &conf, extack);
3399 }
3400
3401 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3402                             struct nlattr *data[],
3403                             struct netlink_ext_ack *extack)
3404 {
3405         struct vxlan_dev *vxlan = netdev_priv(dev);
3406         struct vxlan_rdst *dst = &vxlan->default_dst;
3407         struct vxlan_rdst old_dst;
3408         struct vxlan_config conf;
3409         int err;
3410
3411         err = vxlan_nl2conf(tb, data,
3412                             dev, &conf, true);
3413         if (err)
3414                 return err;
3415
3416         memcpy(&old_dst, dst, sizeof(struct vxlan_rdst));
3417
3418         err = vxlan_dev_configure(vxlan->net, dev, &conf, true, extack);
3419         if (err)
3420                 return err;
3421
3422         /* handle default dst entry */
3423         if (!vxlan_addr_equal(&dst->remote_ip, &old_dst.remote_ip)) {
3424                 spin_lock_bh(&vxlan->hash_lock);
3425                 if (!vxlan_addr_any(&old_dst.remote_ip))
3426                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
3427                                            old_dst.remote_ip,
3428                                            vxlan->cfg.dst_port,
3429                                            old_dst.remote_vni,
3430                                            old_dst.remote_vni,
3431                                            old_dst.remote_ifindex, 0);
3432
3433                 if (!vxlan_addr_any(&dst->remote_ip)) {
3434                         err = vxlan_fdb_create(vxlan, all_zeros_mac,
3435                                                &dst->remote_ip,
3436                                                NUD_REACHABLE | NUD_PERMANENT,
3437                                                NLM_F_CREATE | NLM_F_APPEND,
3438                                                vxlan->cfg.dst_port,
3439                                                dst->remote_vni,
3440                                                dst->remote_vni,
3441                                                dst->remote_ifindex,
3442                                                NTF_SELF);
3443                         if (err) {
3444                                 spin_unlock_bh(&vxlan->hash_lock);
3445                                 return err;
3446                         }
3447                 }
3448                 spin_unlock_bh(&vxlan->hash_lock);
3449         }
3450
3451         return 0;
3452 }
3453
3454 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3455 {
3456         struct vxlan_dev *vxlan = netdev_priv(dev);
3457
3458         vxlan_flush(vxlan, true);
3459
3460         gro_cells_destroy(&vxlan->gro_cells);
3461         list_del(&vxlan->next);
3462         unregister_netdevice_queue(dev, head);
3463 }
3464
3465 static size_t vxlan_get_size(const struct net_device *dev)
3466 {
3467
3468         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3469                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3470                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3471                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3472                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3473                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3474                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3475                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3476                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3477                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3478                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3479                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3480                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3481                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3482                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3483                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3484                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3485                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3486                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3487                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3488                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3489                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3490                 0;
3491 }
3492
3493 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3494 {
3495         const struct vxlan_dev *vxlan = netdev_priv(dev);
3496         const struct vxlan_rdst *dst = &vxlan->default_dst;
3497         struct ifla_vxlan_port_range ports = {
3498                 .low =  htons(vxlan->cfg.port_min),
3499                 .high = htons(vxlan->cfg.port_max),
3500         };
3501
3502         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3503                 goto nla_put_failure;
3504
3505         if (!vxlan_addr_any(&dst->remote_ip)) {
3506                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3507                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3508                                             dst->remote_ip.sin.sin_addr.s_addr))
3509                                 goto nla_put_failure;
3510 #if IS_ENABLED(CONFIG_IPV6)
3511                 } else {
3512                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3513                                              &dst->remote_ip.sin6.sin6_addr))
3514                                 goto nla_put_failure;
3515 #endif
3516                 }
3517         }
3518
3519         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3520                 goto nla_put_failure;
3521
3522         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3523                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3524                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3525                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3526                                 goto nla_put_failure;
3527 #if IS_ENABLED(CONFIG_IPV6)
3528                 } else {
3529                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3530                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3531                                 goto nla_put_failure;
3532 #endif
3533                 }
3534         }
3535
3536         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3537             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3538             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3539             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3540                         !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
3541             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3542                         !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
3543             nla_put_u8(skb, IFLA_VXLAN_RSC,
3544                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
3545             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3546                         !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
3547             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3548                         !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
3549             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3550                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
3551             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3552             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3553             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3554             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3555                         !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3556             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3557                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3558             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3559                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3560             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3561                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
3562             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3563                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
3564                 goto nla_put_failure;
3565
3566         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3567                 goto nla_put_failure;
3568
3569         if (vxlan->cfg.flags & VXLAN_F_GBP &&
3570             nla_put_flag(skb, IFLA_VXLAN_GBP))
3571                 goto nla_put_failure;
3572
3573         if (vxlan->cfg.flags & VXLAN_F_GPE &&
3574             nla_put_flag(skb, IFLA_VXLAN_GPE))
3575                 goto nla_put_failure;
3576
3577         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3578             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3579                 goto nla_put_failure;
3580
3581         return 0;
3582
3583 nla_put_failure:
3584         return -EMSGSIZE;
3585 }
3586
3587 static struct net *vxlan_get_link_net(const struct net_device *dev)
3588 {
3589         struct vxlan_dev *vxlan = netdev_priv(dev);
3590
3591         return vxlan->net;
3592 }
3593
3594 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3595         .kind           = "vxlan",
3596         .maxtype        = IFLA_VXLAN_MAX,
3597         .policy         = vxlan_policy,
3598         .priv_size      = sizeof(struct vxlan_dev),
3599         .setup          = vxlan_setup,
3600         .validate       = vxlan_validate,
3601         .newlink        = vxlan_newlink,
3602         .changelink     = vxlan_changelink,
3603         .dellink        = vxlan_dellink,
3604         .get_size       = vxlan_get_size,
3605         .fill_info      = vxlan_fill_info,
3606         .get_link_net   = vxlan_get_link_net,
3607 };
3608
3609 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3610                                     u8 name_assign_type,
3611                                     struct vxlan_config *conf)
3612 {
3613         struct nlattr *tb[IFLA_MAX + 1];
3614         struct net_device *dev;
3615         int err;
3616
3617         memset(&tb, 0, sizeof(tb));
3618
3619         dev = rtnl_create_link(net, name, name_assign_type,
3620                                &vxlan_link_ops, tb);
3621         if (IS_ERR(dev))
3622                 return dev;
3623
3624         err = __vxlan_dev_create(net, dev, conf, NULL);
3625         if (err < 0) {
3626                 free_netdev(dev);
3627                 return ERR_PTR(err);
3628         }
3629
3630         err = rtnl_configure_link(dev, NULL);
3631         if (err < 0) {
3632                 LIST_HEAD(list_kill);
3633
3634                 vxlan_dellink(dev, &list_kill);
3635                 unregister_netdevice_many(&list_kill);
3636                 return ERR_PTR(err);
3637         }
3638
3639         return dev;
3640 }
3641 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3642
3643 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3644                                              struct net_device *dev)
3645 {
3646         struct vxlan_dev *vxlan, *next;
3647         LIST_HEAD(list_kill);
3648
3649         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3650                 struct vxlan_rdst *dst = &vxlan->default_dst;
3651
3652                 /* In case we created vxlan device with carrier
3653                  * and we loose the carrier due to module unload
3654                  * we also need to remove vxlan device. In other
3655                  * cases, it's not necessary and remote_ifindex
3656                  * is 0 here, so no matches.
3657                  */
3658                 if (dst->remote_ifindex == dev->ifindex)
3659                         vxlan_dellink(vxlan->dev, &list_kill);
3660         }
3661
3662         unregister_netdevice_many(&list_kill);
3663 }
3664
3665 static int vxlan_netdevice_event(struct notifier_block *unused,
3666                                  unsigned long event, void *ptr)
3667 {
3668         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3669         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3670
3671         if (event == NETDEV_UNREGISTER) {
3672                 vxlan_offload_rx_ports(dev, false);
3673                 vxlan_handle_lowerdev_unregister(vn, dev);
3674         } else if (event == NETDEV_REGISTER) {
3675                 vxlan_offload_rx_ports(dev, true);
3676         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
3677                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
3678                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
3679         }
3680
3681         return NOTIFY_DONE;
3682 }
3683
3684 static struct notifier_block vxlan_notifier_block __read_mostly = {
3685         .notifier_call = vxlan_netdevice_event,
3686 };
3687
3688 static __net_init int vxlan_init_net(struct net *net)
3689 {
3690         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3691         unsigned int h;
3692
3693         INIT_LIST_HEAD(&vn->vxlan_list);
3694         spin_lock_init(&vn->sock_lock);
3695
3696         for (h = 0; h < PORT_HASH_SIZE; ++h)
3697                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3698
3699         return 0;
3700 }
3701
3702 static void __net_exit vxlan_exit_net(struct net *net)
3703 {
3704         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3705         struct vxlan_dev *vxlan, *next;
3706         struct net_device *dev, *aux;
3707         LIST_HEAD(list);
3708
3709         rtnl_lock();
3710         for_each_netdev_safe(net, dev, aux)
3711                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3712                         unregister_netdevice_queue(dev, &list);
3713
3714         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3715                 /* If vxlan->dev is in the same netns, it has already been added
3716                  * to the list by the previous loop.
3717                  */
3718                 if (!net_eq(dev_net(vxlan->dev), net)) {
3719                         gro_cells_destroy(&vxlan->gro_cells);
3720                         unregister_netdevice_queue(vxlan->dev, &list);
3721                 }
3722         }
3723
3724         unregister_netdevice_many(&list);
3725         rtnl_unlock();
3726 }
3727
3728 static struct pernet_operations vxlan_net_ops = {
3729         .init = vxlan_init_net,
3730         .exit = vxlan_exit_net,
3731         .id   = &vxlan_net_id,
3732         .size = sizeof(struct vxlan_net),
3733 };
3734
3735 static int __init vxlan_init_module(void)
3736 {
3737         int rc;
3738
3739         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3740
3741         rc = register_pernet_subsys(&vxlan_net_ops);
3742         if (rc)
3743                 goto out1;
3744
3745         rc = register_netdevice_notifier(&vxlan_notifier_block);
3746         if (rc)
3747                 goto out2;
3748
3749         rc = rtnl_link_register(&vxlan_link_ops);
3750         if (rc)
3751                 goto out3;
3752
3753         return 0;
3754 out3:
3755         unregister_netdevice_notifier(&vxlan_notifier_block);
3756 out2:
3757         unregister_pernet_subsys(&vxlan_net_ops);
3758 out1:
3759         return rc;
3760 }
3761 late_initcall(vxlan_init_module);
3762
3763 static void __exit vxlan_cleanup_module(void)
3764 {
3765         rtnl_link_unregister(&vxlan_link_ops);
3766         unregister_netdevice_notifier(&vxlan_notifier_block);
3767         unregister_pernet_subsys(&vxlan_net_ops);
3768         /* rcu_barrier() is called by netns */
3769 }
3770 module_exit(vxlan_cleanup_module);
3771
3772 MODULE_LICENSE("GPL");
3773 MODULE_VERSION(VXLAN_VERSION);
3774 MODULE_AUTHOR("Stephen Hemminger <[email protected]>");
3775 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3776 MODULE_ALIAS_RTNL_LINK("vxlan");