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_flush_final_remcsum(skb, pp, flush, &grc);
 627
 628         return pp;
 629 }
 630
 631 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 632 {
 633         /* Sets 'skb->inner_mac_header' since we are always called with
 634          * 'skb->encapsulation' set.
 635          */
 636         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 637 }
 638
 639 static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan,
 640                                          const u8 *mac, __u16 state,
 641                                          __be32 src_vni, __u8 ndm_flags)
 642 {
 643         struct vxlan_fdb *f;
 644
 645         f = kmalloc(sizeof(*f), GFP_ATOMIC);
 646         if (!f)
 647                 return NULL;
 648         f->state = state;
 649         f->flags = ndm_flags;
 650         f->updated = f->used = jiffies;
 651         f->vni = src_vni;
 652         INIT_LIST_HEAD(&f->remotes);
 653         memcpy(f->eth_addr, mac, ETH_ALEN);
 654
 655         return f;
 656 }
 657
 658 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 659                             const u8 *mac, union vxlan_addr *ip,
 660                             __u16 state, __be16 port, __be32 src_vni,
 661                             __be32 vni, __u32 ifindex, __u8 ndm_flags,
 662                             struct vxlan_fdb **fdb)
 663 {
 664         struct vxlan_rdst *rd = NULL;
 665         struct vxlan_fdb *f;
 666         int rc;
 667
 668         if (vxlan->cfg.addrmax &&
 669             vxlan->addrcnt >= vxlan->cfg.addrmax)
 670                 return -ENOSPC;
 671
 672         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 673         f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags);
 674         if (!f)
 675                 return -ENOMEM;
 676
 677         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 678         if (rc < 0) {
 679                 kfree(f);
 680                 return rc;
 681         }
 682
 683         ++vxlan->addrcnt;
 684         hlist_add_head_rcu(&f->hlist,
 685                            vxlan_fdb_head(vxlan, mac, src_vni));
 686
 687         *fdb = f;
 688
 689         return 0;
 690 }
 691
 692 /* Add new entry to forwarding table -- assumes lock held */
 693 static int vxlan_fdb_update(struct vxlan_dev *vxlan,
 694                             const u8 *mac, union vxlan_addr *ip,
 695                             __u16 state, __u16 flags,
 696                             __be16 port, __be32 src_vni, __be32 vni,
 697                             __u32 ifindex, __u8 ndm_flags)
 698 {
 699         struct vxlan_rdst *rd = NULL;
 700         struct vxlan_fdb *f;
 701         int notify = 0;
 702         int rc;
 703
 704         f = __vxlan_find_mac(vxlan, mac, src_vni);
 705         if (f) {
 706                 if (flags & NLM_F_EXCL) {
 707                         netdev_dbg(vxlan->dev,
 708                                    "lost race to create %pM\n", mac);
 709                         return -EEXIST;
 710                 }
 711                 if (f->state != state) {
 712                         f->state = state;
 713                         f->updated = jiffies;
 714                         notify = 1;
 715                 }
 716                 if (f->flags != ndm_flags) {
 717                         f->flags = ndm_flags;
 718                         f->updated = jiffies;
 719                         notify = 1;
 720                 }
 721                 if ((flags & NLM_F_REPLACE)) {
 722                         /* Only change unicasts */
 723                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 724                              is_zero_ether_addr(f->eth_addr))) {
 725                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 726                                                            ifindex);
 727                         } else
 728                                 return -EOPNOTSUPP;
 729                 }
 730                 if ((flags & NLM_F_APPEND) &&
 731                     (is_multicast_ether_addr(f->eth_addr) ||
 732                      is_zero_ether_addr(f->eth_addr))) {
 733                         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 734
 735                         if (rc < 0)
 736                                 return rc;
 737                         notify |= rc;
 738                 }
 739         } else {
 740                 if (!(flags & NLM_F_CREATE))
 741                         return -ENOENT;
 742
 743                 /* Disallow replace to add a multicast entry */
 744                 if ((flags & NLM_F_REPLACE) &&
 745                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 746                         return -EOPNOTSUPP;
 747
 748                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 749                 rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni,
 750                                       vni, ifindex, ndm_flags, &f);
 751                 if (rc < 0)
 752                         return rc;
 753                 notify = 1;
 754         }
 755
 756         if (notify) {
 757                 if (rd == NULL)
 758                         rd = first_remote_rtnl(f);
 759                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 760         }
 761
 762         return 0;
 763 }
 764
 765 static void vxlan_fdb_free(struct rcu_head *head)
 766 {
 767         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 768         struct vxlan_rdst *rd, *nd;
 769
 770         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 771                 dst_cache_destroy(&rd->dst_cache);
 772                 kfree(rd);
 773         }
 774         kfree(f);
 775 }
 776
 777 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 778                               bool do_notify)
 779 {
 780         netdev_dbg(vxlan->dev,
 781                     "delete %pM\n", f->eth_addr);
 782
 783         --vxlan->addrcnt;
 784         if (do_notify)
 785                 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 786
 787         hlist_del_rcu(&f->hlist);
 788         call_rcu(&f->rcu, vxlan_fdb_free);
 789 }
 790
 791 static void vxlan_dst_free(struct rcu_head *head)
 792 {
 793         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 794
 795         dst_cache_destroy(&rd->dst_cache);
 796         kfree(rd);
 797 }
 798
 799 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 800                                   struct vxlan_rdst *rd)
 801 {
 802         list_del_rcu(&rd->list);
 803         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 804         call_rcu(&rd->rcu, vxlan_dst_free);
 805 }
 806
 807 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 808                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
 809                            __be32 *vni, u32 *ifindex)
 810 {
 811         struct net *net = dev_net(vxlan->dev);
 812         int err;
 813
 814         if (tb[NDA_DST]) {
 815                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 816                 if (err)
 817                         return err;
 818         } else {
 819                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 820                 if (remote->sa.sa_family == AF_INET) {
 821                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 822                         ip->sa.sa_family = AF_INET;
 823 #if IS_ENABLED(CONFIG_IPV6)
 824                 } else {
 825                         ip->sin6.sin6_addr = in6addr_any;
 826                         ip->sa.sa_family = AF_INET6;
 827 #endif
 828                 }
 829         }
 830
 831         if (tb[NDA_PORT]) {
 832                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 833                         return -EINVAL;
 834                 *port = nla_get_be16(tb[NDA_PORT]);
 835         } else {
 836                 *port = vxlan->cfg.dst_port;
 837         }
 838
 839         if (tb[NDA_VNI]) {
 840                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 841                         return -EINVAL;
 842                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 843         } else {
 844                 *vni = vxlan->default_dst.remote_vni;
 845         }
 846
 847         if (tb[NDA_SRC_VNI]) {
 848                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
 849                         return -EINVAL;
 850                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
 851         } else {
 852                 *src_vni = vxlan->default_dst.remote_vni;
 853         }
 854
 855         if (tb[NDA_IFINDEX]) {
 856                 struct net_device *tdev;
 857
 858                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 859                         return -EINVAL;
 860                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 861                 tdev = __dev_get_by_index(net, *ifindex);
 862                 if (!tdev)
 863                         return -EADDRNOTAVAIL;
 864         } else {
 865                 *ifindex = 0;
 866         }
 867
 868         return 0;
 869 }
 870
 871 /* Add static entry (via netlink) */
 872 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 873                          struct net_device *dev,
 874                          const unsigned char *addr, u16 vid, u16 flags)
 875 {
 876         struct vxlan_dev *vxlan = netdev_priv(dev);
 877         /* struct net *net = dev_net(vxlan->dev); */
 878         union vxlan_addr ip;
 879         __be16 port;
 880         __be32 src_vni, vni;
 881         u32 ifindex;
 882         int err;
 883
 884         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 885                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 886                         ndm->ndm_state);
 887                 return -EINVAL;
 888         }
 889
 890         if (tb[NDA_DST] == NULL)
 891                 return -EINVAL;
 892
 893         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 894         if (err)
 895                 return err;
 896
 897         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 898                 return -EAFNOSUPPORT;
 899
 900         spin_lock_bh(&vxlan->hash_lock);
 901         err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags,
 902                                port, src_vni, vni, ifindex, ndm->ndm_flags);
 903         spin_unlock_bh(&vxlan->hash_lock);
 904
 905         return err;
 906 }
 907
 908 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
 909                               const unsigned char *addr, union vxlan_addr ip,
 910                               __be16 port, __be32 src_vni, __be32 vni,
 911                               u32 ifindex, u16 vid)
 912 {
 913         struct vxlan_fdb *f;
 914         struct vxlan_rdst *rd = NULL;
 915         int err = -ENOENT;
 916
 917         f = vxlan_find_mac(vxlan, addr, src_vni);
 918         if (!f)
 919                 return err;
 920
 921         if (!vxlan_addr_any(&ip)) {
 922                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 923                 if (!rd)
 924                         goto out;
 925         }
 926
 927         /* remove a destination if it's not the only one on the list,
 928          * otherwise destroy the fdb entry
 929          */
 930         if (rd && !list_is_singular(&f->remotes)) {
 931                 vxlan_fdb_dst_destroy(vxlan, f, rd);
 932                 goto out;
 933         }
 934
 935         vxlan_fdb_destroy(vxlan, f, true);
 936
 937 out:
 938         return 0;
 939 }
 940
 941 /* Delete entry (via netlink) */
 942 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 943                             struct net_device *dev,
 944                             const unsigned char *addr, u16 vid)
 945 {
 946         struct vxlan_dev *vxlan = netdev_priv(dev);
 947         union vxlan_addr ip;
 948         __be32 src_vni, vni;
 949         __be16 port;
 950         u32 ifindex;
 951         int err;
 952
 953         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 954         if (err)
 955                 return err;
 956
 957         spin_lock_bh(&vxlan->hash_lock);
 958         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
 959                                  vid);
 960         spin_unlock_bh(&vxlan->hash_lock);
 961
 962         return err;
 963 }
 964
 965 /* Dump forwarding table */
 966 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 967                           struct net_device *dev,
 968                           struct net_device *filter_dev, int *idx)
 969 {
 970         struct vxlan_dev *vxlan = netdev_priv(dev);
 971         unsigned int h;
 972         int err = 0;
 973
 974         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 975                 struct vxlan_fdb *f;
 976
 977                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 978                         struct vxlan_rdst *rd;
 979
 980                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 981                                 if (*idx < cb->args[2])
 982                                         goto skip;
 983
 984                                 err = vxlan_fdb_info(skb, vxlan, f,
 985                                                      NETLINK_CB(cb->skb).portid,
 986                                                      cb->nlh->nlmsg_seq,
 987                                                      RTM_NEWNEIGH,
 988                                                      NLM_F_MULTI, rd);
 989                                 if (err < 0)
 990                                         goto out;
 991 skip:
 992                                 *idx += 1;
 993                         }
 994                 }
 995         }
 996 out:
 997         return err;
 998 }
 999
1000 /* Watch incoming packets to learn mapping between Ethernet address
1001  * and Tunnel endpoint.
1002  * Return true if packet is bogus and should be dropped.
1003  */
1004 static bool vxlan_snoop(struct net_device *dev,
1005                         union vxlan_addr *src_ip, const u8 *src_mac,
1006                         u32 src_ifindex, __be32 vni)
1007 {
1008         struct vxlan_dev *vxlan = netdev_priv(dev);
1009         struct vxlan_fdb *f;
1010         u32 ifindex = 0;
1011
1012 #if IS_ENABLED(CONFIG_IPV6)
1013         if (src_ip->sa.sa_family == AF_INET6 &&
1014             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1015                 ifindex = src_ifindex;
1016 #endif
1017
1018         f = vxlan_find_mac(vxlan, src_mac, vni);
1019         if (likely(f)) {
1020                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1021
1022                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1023                            rdst->remote_ifindex == ifindex))
1024                         return false;
1025
1026                 /* Don't migrate static entries, drop packets */
1027                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1028                         return true;
1029
1030                 if (net_ratelimit())
1031                         netdev_info(dev,
1032                                     "%pM migrated from %pIS to %pIS\n",
1033                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1034
1035                 rdst->remote_ip = *src_ip;
1036                 f->updated = jiffies;
1037                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1038         } else {
1039                 /* learned new entry */
1040                 spin_lock(&vxlan->hash_lock);
1041
1042                 /* close off race between vxlan_flush and incoming packets */
1043                 if (netif_running(dev))
1044                         vxlan_fdb_update(vxlan, src_mac, src_ip,
1045                                          NUD_REACHABLE,
1046                                          NLM_F_EXCL|NLM_F_CREATE,
1047                                          vxlan->cfg.dst_port,
1048                                          vni,
1049                                          vxlan->default_dst.remote_vni,
1050                                          ifindex, NTF_SELF);
1051                 spin_unlock(&vxlan->hash_lock);
1052         }
1053
1054         return false;
1055 }
1056
1057 /* See if multicast group is already in use by other ID */
1058 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1059 {
1060         struct vxlan_dev *vxlan;
1061         struct vxlan_sock *sock4;
1062 #if IS_ENABLED(CONFIG_IPV6)
1063         struct vxlan_sock *sock6;
1064 #endif
1065         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1066
1067         sock4 = rtnl_dereference(dev->vn4_sock);
1068
1069         /* The vxlan_sock is only used by dev, leaving group has
1070          * no effect on other vxlan devices.
1071          */
1072         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1073                 return false;
1074 #if IS_ENABLED(CONFIG_IPV6)
1075         sock6 = rtnl_dereference(dev->vn6_sock);
1076         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1077                 return false;
1078 #endif
1079
1080         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1081                 if (!netif_running(vxlan->dev) || vxlan == dev)
1082                         continue;
1083
1084                 if (family == AF_INET &&
1085                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1086                         continue;
1087 #if IS_ENABLED(CONFIG_IPV6)
1088                 if (family == AF_INET6 &&
1089                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1090                         continue;
1091 #endif
1092
1093                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1094                                       &dev->default_dst.remote_ip))
1095                         continue;
1096
1097                 if (vxlan->default_dst.remote_ifindex !=
1098                     dev->default_dst.remote_ifindex)
1099                         continue;
1100
1101                 return true;
1102         }
1103
1104         return false;
1105 }
1106
1107 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1108 {
1109         struct vxlan_net *vn;
1110
1111         if (!vs)
1112                 return false;
1113         if (!refcount_dec_and_test(&vs->refcnt))
1114                 return false;
1115
1116         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1117         spin_lock(&vn->sock_lock);
1118         hlist_del_rcu(&vs->hlist);
1119         udp_tunnel_notify_del_rx_port(vs->sock,
1120                                       (vs->flags & VXLAN_F_GPE) ?
1121                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1122                                       UDP_TUNNEL_TYPE_VXLAN);
1123         spin_unlock(&vn->sock_lock);
1124
1125         return true;
1126 }
1127
1128 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1129 {
1130         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1131 #if IS_ENABLED(CONFIG_IPV6)
1132         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1133
1134         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1135 #endif
1136
1137         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1138         synchronize_net();
1139
1140         vxlan_vs_del_dev(vxlan);
1141
1142         if (__vxlan_sock_release_prep(sock4)) {
1143                 udp_tunnel_sock_release(sock4->sock);
1144                 kfree(sock4);
1145         }
1146
1147 #if IS_ENABLED(CONFIG_IPV6)
1148         if (__vxlan_sock_release_prep(sock6)) {
1149                 udp_tunnel_sock_release(sock6->sock);
1150                 kfree(sock6);
1151         }
1152 #endif
1153 }
1154
1155 /* Update multicast group membership when first VNI on
1156  * multicast address is brought up
1157  */
1158 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1159 {
1160         struct sock *sk;
1161         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1162         int ifindex = vxlan->default_dst.remote_ifindex;
1163         int ret = -EINVAL;
1164
1165         if (ip->sa.sa_family == AF_INET) {
1166                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1167                 struct ip_mreqn mreq = {
1168                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1169                         .imr_ifindex            = ifindex,
1170                 };
1171
1172                 sk = sock4->sock->sk;
1173                 lock_sock(sk);
1174                 ret = ip_mc_join_group(sk, &mreq);
1175                 release_sock(sk);
1176 #if IS_ENABLED(CONFIG_IPV6)
1177         } else {
1178                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1179
1180                 sk = sock6->sock->sk;
1181                 lock_sock(sk);
1182                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1183                                                    &ip->sin6.sin6_addr);
1184                 release_sock(sk);
1185 #endif
1186         }
1187
1188         return ret;
1189 }
1190
1191 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1192 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1193 {
1194         struct sock *sk;
1195         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1196         int ifindex = vxlan->default_dst.remote_ifindex;
1197         int ret = -EINVAL;
1198
1199         if (ip->sa.sa_family == AF_INET) {
1200                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1201                 struct ip_mreqn mreq = {
1202                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1203                         .imr_ifindex            = ifindex,
1204                 };
1205
1206                 sk = sock4->sock->sk;
1207                 lock_sock(sk);
1208                 ret = ip_mc_leave_group(sk, &mreq);
1209                 release_sock(sk);
1210 #if IS_ENABLED(CONFIG_IPV6)
1211         } else {
1212                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1213
1214                 sk = sock6->sock->sk;
1215                 lock_sock(sk);
1216                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1217                                                    &ip->sin6.sin6_addr);
1218                 release_sock(sk);
1219 #endif
1220         }
1221
1222         return ret;
1223 }
1224
1225 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1226                           struct sk_buff *skb, u32 vxflags)
1227 {
1228         size_t start, offset;
1229
1230         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1231                 goto out;
1232
1233         start = vxlan_rco_start(unparsed->vx_vni);
1234         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1235
1236         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1237                 return false;
1238
1239         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1240                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1241 out:
1242         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1243         unparsed->vx_vni &= VXLAN_VNI_MASK;
1244         return true;
1245 }
1246
1247 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1248                                 struct sk_buff *skb, u32 vxflags,
1249                                 struct vxlan_metadata *md)
1250 {
1251         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1252         struct metadata_dst *tun_dst;
1253
1254         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1255                 goto out;
1256
1257         md->gbp = ntohs(gbp->policy_id);
1258
1259         tun_dst = (struct metadata_dst *)skb_dst(skb);
1260         if (tun_dst) {
1261                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1262                 tun_dst->u.tun_info.options_len = sizeof(*md);
1263         }
1264         if (gbp->dont_learn)
1265                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1266
1267         if (gbp->policy_applied)
1268                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1269
1270         /* In flow-based mode, GBP is carried in dst_metadata */
1271         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1272                 skb->mark = md->gbp;
1273 out:
1274         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1275 }
1276
1277 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1278                                 __be16 *protocol,
1279                                 struct sk_buff *skb, u32 vxflags)
1280 {
1281         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1282
1283         /* Need to have Next Protocol set for interfaces in GPE mode. */
1284         if (!gpe->np_applied)
1285                 return false;
1286         /* "The initial version is 0. If a receiver does not support the
1287          * version indicated it MUST drop the packet.
1288          */
1289         if (gpe->version != 0)
1290                 return false;
1291         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1292          * processing MUST occur." However, we don't implement OAM
1293          * processing, thus drop the packet.
1294          */
1295         if (gpe->oam_flag)
1296                 return false;
1297
1298         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1299         if (!*protocol)
1300                 return false;
1301
1302         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1303         return true;
1304 }
1305
1306 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1307                           struct vxlan_sock *vs,
1308                           struct sk_buff *skb, __be32 vni)
1309 {
1310         union vxlan_addr saddr;
1311         u32 ifindex = skb->dev->ifindex;
1312
1313         skb_reset_mac_header(skb);
1314         skb->protocol = eth_type_trans(skb, vxlan->dev);
1315         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1316
1317         /* Ignore packet loops (and multicast echo) */
1318         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1319                 return false;
1320
1321         /* Get address from the outer IP header */
1322         if (vxlan_get_sk_family(vs) == AF_INET) {
1323                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1324                 saddr.sa.sa_family = AF_INET;
1325 #if IS_ENABLED(CONFIG_IPV6)
1326         } else {
1327                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1328                 saddr.sa.sa_family = AF_INET6;
1329 #endif
1330         }
1331
1332         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1333             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1334                 return false;
1335
1336         return true;
1337 }
1338
1339 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1340                                   struct sk_buff *skb)
1341 {
1342         int err = 0;
1343
1344         if (vxlan_get_sk_family(vs) == AF_INET)
1345                 err = IP_ECN_decapsulate(oiph, skb);
1346 #if IS_ENABLED(CONFIG_IPV6)
1347         else
1348                 err = IP6_ECN_decapsulate(oiph, skb);
1349 #endif
1350
1351         if (unlikely(err) && log_ecn_error) {
1352                 if (vxlan_get_sk_family(vs) == AF_INET)
1353                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1354                                              &((struct iphdr *)oiph)->saddr,
1355                                              ((struct iphdr *)oiph)->tos);
1356                 else
1357                         net_info_ratelimited("non-ECT from %pI6\n",
1358                                              &((struct ipv6hdr *)oiph)->saddr);
1359         }
1360         return err <= 1;
1361 }
1362
1363 /* Callback from net/ipv4/udp.c to receive packets */
1364 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1365 {
1366         struct pcpu_sw_netstats *stats;
1367         struct vxlan_dev *vxlan;
1368         struct vxlan_sock *vs;
1369         struct vxlanhdr unparsed;
1370         struct vxlan_metadata _md;
1371         struct vxlan_metadata *md = &_md;
1372         __be16 protocol = htons(ETH_P_TEB);
1373         bool raw_proto = false;
1374         void *oiph;
1375         __be32 vni = 0;
1376
1377         /* Need UDP and VXLAN header to be present */
1378         if (!pskb_may_pull(skb, VXLAN_HLEN))
1379                 goto drop;
1380
1381         unparsed = *vxlan_hdr(skb);
1382         /* VNI flag always required to be set */
1383         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1384                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1385                            ntohl(vxlan_hdr(skb)->vx_flags),
1386                            ntohl(vxlan_hdr(skb)->vx_vni));
1387                 /* Return non vxlan pkt */
1388                 goto drop;
1389         }
1390         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1391         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1392
1393         vs = rcu_dereference_sk_user_data(sk);
1394         if (!vs)
1395                 goto drop;
1396
1397         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1398
1399         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1400         if (!vxlan)
1401                 goto drop;
1402
1403         /* For backwards compatibility, only allow reserved fields to be
1404          * used by VXLAN extensions if explicitly requested.
1405          */
1406         if (vs->flags & VXLAN_F_GPE) {
1407                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1408                         goto drop;
1409                 raw_proto = true;
1410         }
1411
1412         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1413                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1414                         goto drop;
1415
1416         if (vxlan_collect_metadata(vs)) {
1417                 struct metadata_dst *tun_dst;
1418
1419                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1420                                          key32_to_tunnel_id(vni), sizeof(*md));
1421
1422                 if (!tun_dst)
1423                         goto drop;
1424
1425                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1426
1427                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1428         } else {
1429                 memset(md, 0, sizeof(*md));
1430         }
1431
1432         if (vs->flags & VXLAN_F_REMCSUM_RX)
1433                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1434                         goto drop;
1435         if (vs->flags & VXLAN_F_GBP)
1436                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1437         /* Note that GBP and GPE can never be active together. This is
1438          * ensured in vxlan_dev_configure.
1439          */
1440
1441         if (unparsed.vx_flags || unparsed.vx_vni) {
1442                 /* If there are any unprocessed flags remaining treat
1443                  * this as a malformed packet. This behavior diverges from
1444                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1445                  * in reserved fields are to be ignored. The approach here
1446                  * maintains compatibility with previous stack code, and also
1447                  * is more robust and provides a little more security in
1448                  * adding extensions to VXLAN.
1449                  */
1450                 goto drop;
1451         }
1452
1453         if (!raw_proto) {
1454                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1455                         goto drop;
1456         } else {
1457                 skb_reset_mac_header(skb);
1458                 skb->dev = vxlan->dev;
1459                 skb->pkt_type = PACKET_HOST;
1460         }
1461
1462         oiph = skb_network_header(skb);
1463         skb_reset_network_header(skb);
1464
1465         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1466                 ++vxlan->dev->stats.rx_frame_errors;
1467                 ++vxlan->dev->stats.rx_errors;
1468                 goto drop;
1469         }
1470
1471         stats = this_cpu_ptr(vxlan->dev->tstats);
1472         u64_stats_update_begin(&stats->syncp);
1473         stats->rx_packets++;
1474         stats->rx_bytes += skb->len;
1475         u64_stats_update_end(&stats->syncp);
1476
1477         gro_cells_receive(&vxlan->gro_cells, skb);
1478         return 0;
1479
1480 drop:
1481         /* Consume bad packet */
1482         kfree_skb(skb);
1483         return 0;
1484 }
1485
1486 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1487 {
1488         struct vxlan_dev *vxlan = netdev_priv(dev);
1489         struct arphdr *parp;
1490         u8 *arpptr, *sha;
1491         __be32 sip, tip;
1492         struct neighbour *n;
1493
1494         if (dev->flags & IFF_NOARP)
1495                 goto out;
1496
1497         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1498                 dev->stats.tx_dropped++;
1499                 goto out;
1500         }
1501         parp = arp_hdr(skb);
1502
1503         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1504              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1505             parp->ar_pro != htons(ETH_P_IP) ||
1506             parp->ar_op != htons(ARPOP_REQUEST) ||
1507             parp->ar_hln != dev->addr_len ||
1508             parp->ar_pln != 4)
1509                 goto out;
1510         arpptr = (u8 *)parp + sizeof(struct arphdr);
1511         sha = arpptr;
1512         arpptr += dev->addr_len;        /* sha */
1513         memcpy(&sip, arpptr, sizeof(sip));
1514         arpptr += sizeof(sip);
1515         arpptr += dev->addr_len;        /* tha */
1516         memcpy(&tip, arpptr, sizeof(tip));
1517
1518         if (ipv4_is_loopback(tip) ||
1519             ipv4_is_multicast(tip))
1520                 goto out;
1521
1522         n = neigh_lookup(&arp_tbl, &tip, dev);
1523
1524         if (n) {
1525                 struct vxlan_fdb *f;
1526                 struct sk_buff  *reply;
1527
1528                 if (!(n->nud_state & NUD_CONNECTED)) {
1529                         neigh_release(n);
1530                         goto out;
1531                 }
1532
1533                 f = vxlan_find_mac(vxlan, n->ha, vni);
1534                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1535                         /* bridge-local neighbor */
1536                         neigh_release(n);
1537                         goto out;
1538                 }
1539
1540                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1541                                 n->ha, sha);
1542
1543                 neigh_release(n);
1544
1545                 if (reply == NULL)
1546                         goto out;
1547
1548                 skb_reset_mac_header(reply);
1549                 __skb_pull(reply, skb_network_offset(reply));
1550                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1551                 reply->pkt_type = PACKET_HOST;
1552
1553                 if (netif_rx_ni(reply) == NET_RX_DROP)
1554                         dev->stats.rx_dropped++;
1555         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1556                 union vxlan_addr ipa = {
1557                         .sin.sin_addr.s_addr = tip,
1558                         .sin.sin_family = AF_INET,
1559                 };
1560
1561                 vxlan_ip_miss(dev, &ipa);
1562         }
1563 out:
1564         consume_skb(skb);
1565         return NETDEV_TX_OK;
1566 }
1567
1568 #if IS_ENABLED(CONFIG_IPV6)
1569 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1570         struct neighbour *n, bool isrouter)
1571 {
1572         struct net_device *dev = request->dev;
1573         struct sk_buff *reply;
1574         struct nd_msg *ns, *na;
1575         struct ipv6hdr *pip6;
1576         u8 *daddr;
1577         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1578         int ns_olen;
1579         int i, len;
1580
1581         if (dev == NULL || !pskb_may_pull(request, request->len))
1582                 return NULL;
1583
1584         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1585                 sizeof(*na) + na_olen + dev->needed_tailroom;
1586         reply = alloc_skb(len, GFP_ATOMIC);
1587         if (reply == NULL)
1588                 return NULL;
1589
1590         reply->protocol = htons(ETH_P_IPV6);
1591         reply->dev = dev;
1592         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1593         skb_push(reply, sizeof(struct ethhdr));
1594         skb_reset_mac_header(reply);
1595
1596         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1597
1598         daddr = eth_hdr(request)->h_source;
1599         ns_olen = request->len - skb_network_offset(request) -
1600                 sizeof(struct ipv6hdr) - sizeof(*ns);
1601         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1602                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1603                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1604                         break;
1605                 }
1606         }
1607
1608         /* Ethernet header */
1609         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1610         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1611         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1612         reply->protocol = htons(ETH_P_IPV6);
1613
1614         skb_pull(reply, sizeof(struct ethhdr));
1615         skb_reset_network_header(reply);
1616         skb_put(reply, sizeof(struct ipv6hdr));
1617
1618         /* IPv6 header */
1619
1620         pip6 = ipv6_hdr(reply);
1621         memset(pip6, 0, sizeof(struct ipv6hdr));
1622         pip6->version = 6;
1623         pip6->priority = ipv6_hdr(request)->priority;
1624         pip6->nexthdr = IPPROTO_ICMPV6;
1625         pip6->hop_limit = 255;
1626         pip6->daddr = ipv6_hdr(request)->saddr;
1627         pip6->saddr = *(struct in6_addr *)n->primary_key;
1628
1629         skb_pull(reply, sizeof(struct ipv6hdr));
1630         skb_reset_transport_header(reply);
1631
1632         /* Neighbor Advertisement */
1633         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1634         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1635         na->icmph.icmp6_router = isrouter;
1636         na->icmph.icmp6_override = 1;
1637         na->icmph.icmp6_solicited = 1;
1638         na->target = ns->target;
1639         ether_addr_copy(&na->opt[2], n->ha);
1640         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1641         na->opt[1] = na_olen >> 3;
1642
1643         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1644                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1645                 csum_partial(na, sizeof(*na)+na_olen, 0));
1646
1647         pip6->payload_len = htons(sizeof(*na)+na_olen);
1648
1649         skb_push(reply, sizeof(struct ipv6hdr));
1650
1651         reply->ip_summed = CHECKSUM_UNNECESSARY;
1652
1653         return reply;
1654 }
1655
1656 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1657 {
1658         struct vxlan_dev *vxlan = netdev_priv(dev);
1659         const struct in6_addr *daddr;
1660         const struct ipv6hdr *iphdr;
1661         struct inet6_dev *in6_dev;
1662         struct neighbour *n;
1663         struct nd_msg *msg;
1664
1665         in6_dev = __in6_dev_get(dev);
1666         if (!in6_dev)
1667                 goto out;
1668
1669         iphdr = ipv6_hdr(skb);
1670         daddr = &iphdr->daddr;
1671         msg = (struct nd_msg *)(iphdr + 1);
1672
1673         if (ipv6_addr_loopback(daddr) ||
1674             ipv6_addr_is_multicast(&msg->target))
1675                 goto out;
1676
1677         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1678
1679         if (n) {
1680                 struct vxlan_fdb *f;
1681                 struct sk_buff *reply;
1682
1683                 if (!(n->nud_state & NUD_CONNECTED)) {
1684                         neigh_release(n);
1685                         goto out;
1686                 }
1687
1688                 f = vxlan_find_mac(vxlan, n->ha, vni);
1689                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1690                         /* bridge-local neighbor */
1691                         neigh_release(n);
1692                         goto out;
1693                 }
1694
1695                 reply = vxlan_na_create(skb, n,
1696                                         !!(f ? f->flags & NTF_ROUTER : 0));
1697
1698                 neigh_release(n);
1699
1700                 if (reply == NULL)
1701                         goto out;
1702
1703                 if (netif_rx_ni(reply) == NET_RX_DROP)
1704                         dev->stats.rx_dropped++;
1705
1706         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1707                 union vxlan_addr ipa = {
1708                         .sin6.sin6_addr = msg->target,
1709                         .sin6.sin6_family = AF_INET6,
1710                 };
1711
1712                 vxlan_ip_miss(dev, &ipa);
1713         }
1714
1715 out:
1716         consume_skb(skb);
1717         return NETDEV_TX_OK;
1718 }
1719 #endif
1720
1721 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1722 {
1723         struct vxlan_dev *vxlan = netdev_priv(dev);
1724         struct neighbour *n;
1725
1726         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1727                 return false;
1728
1729         n = NULL;
1730         switch (ntohs(eth_hdr(skb)->h_proto)) {
1731         case ETH_P_IP:
1732         {
1733                 struct iphdr *pip;
1734
1735                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1736                         return false;
1737                 pip = ip_hdr(skb);
1738                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1739                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1740                         union vxlan_addr ipa = {
1741                                 .sin.sin_addr.s_addr = pip->daddr,
1742                                 .sin.sin_family = AF_INET,
1743                         };
1744
1745                         vxlan_ip_miss(dev, &ipa);
1746                         return false;
1747                 }
1748
1749                 break;
1750         }
1751 #if IS_ENABLED(CONFIG_IPV6)
1752         case ETH_P_IPV6:
1753         {
1754                 struct ipv6hdr *pip6;
1755
1756                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1757                         return false;
1758                 pip6 = ipv6_hdr(skb);
1759                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1760                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1761                         union vxlan_addr ipa = {
1762                                 .sin6.sin6_addr = pip6->daddr,
1763                                 .sin6.sin6_family = AF_INET6,
1764                         };
1765
1766                         vxlan_ip_miss(dev, &ipa);
1767                         return false;
1768                 }
1769
1770                 break;
1771         }
1772 #endif
1773         default:
1774                 return false;
1775         }
1776
1777         if (n) {
1778                 bool diff;
1779
1780                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1781                 if (diff) {
1782                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1783                                 dev->addr_len);
1784                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1785                 }
1786                 neigh_release(n);
1787                 return diff;
1788         }
1789
1790         return false;
1791 }
1792
1793 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1794                                 struct vxlan_metadata *md)
1795 {
1796         struct vxlanhdr_gbp *gbp;
1797
1798         if (!md->gbp)
1799                 return;
1800
1801         gbp = (struct vxlanhdr_gbp *)vxh;
1802         vxh->vx_flags |= VXLAN_HF_GBP;
1803
1804         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1805                 gbp->dont_learn = 1;
1806
1807         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1808                 gbp->policy_applied = 1;
1809
1810         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1811 }
1812
1813 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1814                                __be16 protocol)
1815 {
1816         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1817
1818         gpe->np_applied = 1;
1819         gpe->next_protocol = tun_p_from_eth_p(protocol);
1820         if (!gpe->next_protocol)
1821                 return -EPFNOSUPPORT;
1822         return 0;
1823 }
1824
1825 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1826                            int iphdr_len, __be32 vni,
1827                            struct vxlan_metadata *md, u32 vxflags,
1828                            bool udp_sum)
1829 {
1830         struct vxlanhdr *vxh;
1831         int min_headroom;
1832         int err;
1833         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1834         __be16 inner_protocol = htons(ETH_P_TEB);
1835
1836         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1837             skb->ip_summed == CHECKSUM_PARTIAL) {
1838                 int csum_start = skb_checksum_start_offset(skb);
1839
1840                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1841                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1842                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1843                      skb->csum_offset == offsetof(struct tcphdr, check)))
1844                         type |= SKB_GSO_TUNNEL_REMCSUM;
1845         }
1846
1847         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1848                         + VXLAN_HLEN + iphdr_len;
1849
1850         /* Need space for new headers (invalidates iph ptr) */
1851         err = skb_cow_head(skb, min_headroom);
1852         if (unlikely(err))
1853                 return err;
1854
1855         err = iptunnel_handle_offloads(skb, type);
1856         if (err)
1857                 return err;
1858
1859         vxh = __skb_push(skb, sizeof(*vxh));
1860         vxh->vx_flags = VXLAN_HF_VNI;
1861         vxh->vx_vni = vxlan_vni_field(vni);
1862
1863         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1864                 unsigned int start;
1865
1866                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1867                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1868                 vxh->vx_flags |= VXLAN_HF_RCO;
1869
1870                 if (!skb_is_gso(skb)) {
1871                         skb->ip_summed = CHECKSUM_NONE;
1872                         skb->encapsulation = 0;
1873                 }
1874         }
1875
1876         if (vxflags & VXLAN_F_GBP)
1877                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1878         if (vxflags & VXLAN_F_GPE) {
1879                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1880                 if (err < 0)
1881                         return err;
1882                 inner_protocol = skb->protocol;
1883         }
1884
1885         skb_set_inner_protocol(skb, inner_protocol);
1886         return 0;
1887 }
1888
1889 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1890                                       struct vxlan_sock *sock4,
1891                                       struct sk_buff *skb, int oif, u8 tos,
1892                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
1893                                       struct dst_cache *dst_cache,
1894                                       const struct ip_tunnel_info *info)
1895 {
1896         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1897         struct rtable *rt = NULL;
1898         struct flowi4 fl4;
1899
1900         if (!sock4)
1901                 return ERR_PTR(-EIO);
1902
1903         if (tos && !info)
1904                 use_cache = false;
1905         if (use_cache) {
1906                 rt = dst_cache_get_ip4(dst_cache, saddr);
1907                 if (rt)
1908                         return rt;
1909         }
1910
1911         memset(&fl4, 0, sizeof(fl4));
1912         fl4.flowi4_oif = oif;
1913         fl4.flowi4_tos = RT_TOS(tos);
1914         fl4.flowi4_mark = skb->mark;
1915         fl4.flowi4_proto = IPPROTO_UDP;
1916         fl4.daddr = daddr;
1917         fl4.saddr = *saddr;
1918         fl4.fl4_dport = dport;
1919         fl4.fl4_sport = sport;
1920
1921         rt = ip_route_output_key(vxlan->net, &fl4);
1922         if (likely(!IS_ERR(rt))) {
1923                 if (rt->dst.dev == dev) {
1924                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1925                         ip_rt_put(rt);
1926                         return ERR_PTR(-ELOOP);
1927                 }
1928
1929                 *saddr = fl4.saddr;
1930                 if (use_cache)
1931                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1932         } else {
1933                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1934                 return ERR_PTR(-ENETUNREACH);
1935         }
1936         return rt;
1937 }
1938
1939 #if IS_ENABLED(CONFIG_IPV6)
1940 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1941                                           struct net_device *dev,
1942                                           struct vxlan_sock *sock6,
1943                                           struct sk_buff *skb, int oif, u8 tos,
1944                                           __be32 label,
1945                                           const struct in6_addr *daddr,
1946                                           struct in6_addr *saddr,
1947                                           __be16 dport, __be16 sport,
1948                                           struct dst_cache *dst_cache,
1949                                           const struct ip_tunnel_info *info)
1950 {
1951         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1952         struct dst_entry *ndst;
1953         struct flowi6 fl6;
1954         int err;
1955
1956         if (!sock6)
1957                 return ERR_PTR(-EIO);
1958
1959         if (tos && !info)
1960                 use_cache = false;
1961         if (use_cache) {
1962                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1963                 if (ndst)
1964                         return ndst;
1965         }
1966
1967         memset(&fl6, 0, sizeof(fl6));
1968         fl6.flowi6_oif = oif;
1969         fl6.daddr = *daddr;
1970         fl6.saddr = *saddr;
1971         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1972         fl6.flowi6_mark = skb->mark;
1973         fl6.flowi6_proto = IPPROTO_UDP;
1974         fl6.fl6_dport = dport;
1975         fl6.fl6_sport = sport;
1976
1977         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1978                                          sock6->sock->sk,
1979                                          &ndst, &fl6);
1980         if (unlikely(err < 0)) {
1981                 netdev_dbg(dev, "no route to %pI6\n", daddr);
1982                 return ERR_PTR(-ENETUNREACH);
1983         }
1984
1985         if (unlikely(ndst->dev == dev)) {
1986                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
1987                 dst_release(ndst);
1988                 return ERR_PTR(-ELOOP);
1989         }
1990
1991         *saddr = fl6.saddr;
1992         if (use_cache)
1993                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1994         return ndst;
1995 }
1996 #endif
1997
1998 /* Bypass encapsulation if the destination is local */
1999 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2000                                struct vxlan_dev *dst_vxlan, __be32 vni)
2001 {
2002         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2003         union vxlan_addr loopback;
2004         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2005         struct net_device *dev = skb->dev;
2006         int len = skb->len;
2007
2008         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2009         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2010         skb->pkt_type = PACKET_HOST;
2011         skb->encapsulation = 0;
2012         skb->dev = dst_vxlan->dev;
2013         __skb_pull(skb, skb_network_offset(skb));
2014
2015         if (remote_ip->sa.sa_family == AF_INET) {
2016                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2017                 loopback.sa.sa_family =  AF_INET;
2018 #if IS_ENABLED(CONFIG_IPV6)
2019         } else {
2020                 loopback.sin6.sin6_addr = in6addr_loopback;
2021                 loopback.sa.sa_family =  AF_INET6;
2022 #endif
2023         }
2024
2025         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
2026                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source, 0,
2027                             vni);
2028
2029         u64_stats_update_begin(&tx_stats->syncp);
2030         tx_stats->tx_packets++;
2031         tx_stats->tx_bytes += len;
2032         u64_stats_update_end(&tx_stats->syncp);
2033
2034         if (netif_rx(skb) == NET_RX_SUCCESS) {
2035                 u64_stats_update_begin(&rx_stats->syncp);
2036                 rx_stats->rx_packets++;
2037                 rx_stats->rx_bytes += len;
2038                 u64_stats_update_end(&rx_stats->syncp);
2039         } else {
2040                 dev->stats.rx_dropped++;
2041         }
2042 }
2043
2044 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2045                                  struct vxlan_dev *vxlan,
2046                                  union vxlan_addr *daddr,
2047                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2048                                  struct dst_entry *dst,
2049                                  u32 rt_flags)
2050 {
2051 #if IS_ENABLED(CONFIG_IPV6)
2052         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2053          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2054          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2055          */
2056         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2057 #endif
2058         /* Bypass encapsulation if the destination is local */
2059         if (rt_flags & RTCF_LOCAL &&
2060             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2061                 struct vxlan_dev *dst_vxlan;
2062
2063                 dst_release(dst);
2064                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2065                                            daddr->sa.sa_family, dst_port,
2066                                            vxlan->cfg.flags);
2067                 if (!dst_vxlan) {
2068                         dev->stats.tx_errors++;
2069                         kfree_skb(skb);
2070
2071                         return -ENOENT;
2072                 }
2073                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2074                 return 1;
2075         }
2076
2077         return 0;
2078 }
2079
2080 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2081                            __be32 default_vni, struct vxlan_rdst *rdst,
2082                            bool did_rsc)
2083 {
2084         struct dst_cache *dst_cache;
2085         struct ip_tunnel_info *info;
2086         struct vxlan_dev *vxlan = netdev_priv(dev);
2087         const struct iphdr *old_iph = ip_hdr(skb);
2088         union vxlan_addr *dst;
2089         union vxlan_addr remote_ip, local_ip;
2090         struct vxlan_metadata _md;
2091         struct vxlan_metadata *md = &_md;
2092         __be16 src_port = 0, dst_port;
2093         struct dst_entry *ndst = NULL;
2094         __be32 vni, label;
2095         __u8 tos, ttl;
2096         int ifindex;
2097         int err;
2098         u32 flags = vxlan->cfg.flags;
2099         bool udp_sum = false;
2100         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2101
2102         info = skb_tunnel_info(skb);
2103
2104         if (rdst) {
2105                 dst = &rdst->remote_ip;
2106                 if (vxlan_addr_any(dst)) {
2107                         if (did_rsc) {
2108                                 /* short-circuited back to local bridge */
2109                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2110                                 return;
2111                         }
2112                         goto drop;
2113                 }
2114
2115                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2116                 vni = (rdst->remote_vni) ? : default_vni;
2117                 ifindex = rdst->remote_ifindex;
2118                 local_ip = vxlan->cfg.saddr;
2119                 dst_cache = &rdst->dst_cache;
2120                 md->gbp = skb->mark;
2121                 if (flags & VXLAN_F_TTL_INHERIT) {
2122                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2123                 } else {
2124                         ttl = vxlan->cfg.ttl;
2125                         if (!ttl && vxlan_addr_multicast(dst))
2126                                 ttl = 1;
2127                 }
2128
2129                 tos = vxlan->cfg.tos;
2130                 if (tos == 1)
2131                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2132
2133                 if (dst->sa.sa_family == AF_INET)
2134                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2135                 else
2136                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2137                 label = vxlan->cfg.label;
2138         } else {
2139                 if (!info) {
2140                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2141                                   dev->name);
2142                         goto drop;
2143                 }
2144                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2145                 if (remote_ip.sa.sa_family == AF_INET) {
2146                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2147                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2148                 } else {
2149                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2150                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2151                 }
2152                 dst = &remote_ip;
2153                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2154                 vni = tunnel_id_to_key32(info->key.tun_id);
2155                 ifindex = 0;
2156                 dst_cache = &info->dst_cache;
2157                 if (info->options_len)
2158                         md = ip_tunnel_info_opts(info);
2159                 ttl = info->key.ttl;
2160                 tos = info->key.tos;
2161                 label = info->key.label;
2162                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2163         }
2164         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2165                                      vxlan->cfg.port_max, true);
2166
2167         rcu_read_lock();
2168         if (dst->sa.sa_family == AF_INET) {
2169                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2170                 struct rtable *rt;
2171                 __be16 df = 0;
2172
2173                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2174                                      dst->sin.sin_addr.s_addr,
2175                                      &local_ip.sin.sin_addr.s_addr,
2176                                      dst_port, src_port,
2177                                      dst_cache, info);
2178                 if (IS_ERR(rt)) {
2179                         err = PTR_ERR(rt);
2180                         goto tx_error;
2181                 }
2182
2183                 /* Bypass encapsulation if the destination is local */
2184                 if (!info) {
2185                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2186                                                     dst_port, ifindex, vni,
2187                                                     &rt->dst, rt->rt_flags);
2188                         if (err)
2189                                 goto out_unlock;
2190                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2191                         df = htons(IP_DF);
2192                 }
2193
2194                 ndst = &rt->dst;
2195                 if (skb_dst(skb)) {
2196                         int mtu = dst_mtu(ndst) - VXLAN_HEADROOM;
2197
2198                         skb_dst_update_pmtu(skb, mtu);
2199                 }
2200
2201                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2202                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2203                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2204                                       vni, md, flags, udp_sum);
2205                 if (err < 0)
2206                         goto tx_error;
2207
2208                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2209                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2210                                     src_port, dst_port, xnet, !udp_sum);
2211 #if IS_ENABLED(CONFIG_IPV6)
2212         } else {
2213                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2214
2215                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2216                                         label, &dst->sin6.sin6_addr,
2217                                         &local_ip.sin6.sin6_addr,
2218                                         dst_port, src_port,
2219                                         dst_cache, info);
2220                 if (IS_ERR(ndst)) {
2221                         err = PTR_ERR(ndst);
2222                         ndst = NULL;
2223                         goto tx_error;
2224                 }
2225
2226                 if (!info) {
2227                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2228
2229                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2230                                                     dst_port, ifindex, vni,
2231                                                     ndst, rt6i_flags);
2232                         if (err)
2233                                 goto out_unlock;
2234                 }
2235
2236                 if (skb_dst(skb)) {
2237                         int mtu = dst_mtu(ndst) - VXLAN6_HEADROOM;
2238
2239                         skb_dst_update_pmtu(skb, mtu);
2240                 }
2241
2242                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2243                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2244                 skb_scrub_packet(skb, xnet);
2245                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2246                                       vni, md, flags, udp_sum);
2247                 if (err < 0)
2248                         goto tx_error;
2249
2250                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2251                                      &local_ip.sin6.sin6_addr,
2252                                      &dst->sin6.sin6_addr, tos, ttl,
2253                                      label, src_port, dst_port, !udp_sum);
2254 #endif
2255         }
2256 out_unlock:
2257         rcu_read_unlock();
2258         return;
2259
2260 drop:
2261         dev->stats.tx_dropped++;
2262         dev_kfree_skb(skb);
2263         return;
2264
2265 tx_error:
2266         rcu_read_unlock();
2267         if (err == -ELOOP)
2268                 dev->stats.collisions++;
2269         else if (err == -ENETUNREACH)
2270                 dev->stats.tx_carrier_errors++;
2271         dst_release(ndst);
2272         dev->stats.tx_errors++;
2273         kfree_skb(skb);
2274 }
2275
2276 /* Transmit local packets over Vxlan
2277  *
2278  * Outer IP header inherits ECN and DF from inner header.
2279  * Outer UDP destination is the VXLAN assigned port.
2280  *           source port is based on hash of flow
2281  */
2282 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2283 {
2284         struct vxlan_dev *vxlan = netdev_priv(dev);
2285         struct vxlan_rdst *rdst, *fdst = NULL;
2286         const struct ip_tunnel_info *info;
2287         bool did_rsc = false;
2288         struct vxlan_fdb *f;
2289         struct ethhdr *eth;
2290         __be32 vni = 0;
2291
2292         info = skb_tunnel_info(skb);
2293
2294         skb_reset_mac_header(skb);
2295
2296         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2297                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2298                     info->mode & IP_TUNNEL_INFO_TX) {
2299                         vni = tunnel_id_to_key32(info->key.tun_id);
2300                 } else {
2301                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2302                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2303                         else
2304                                 kfree_skb(skb);
2305                         return NETDEV_TX_OK;
2306                 }
2307         }
2308
2309         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2310                 eth = eth_hdr(skb);
2311                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2312                         return arp_reduce(dev, skb, vni);
2313 #if IS_ENABLED(CONFIG_IPV6)
2314                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2315                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2316                                             sizeof(struct nd_msg)) &&
2317                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2318                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2319
2320                         if (m->icmph.icmp6_code == 0 &&
2321                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2322                                 return neigh_reduce(dev, skb, vni);
2323                 }
2324 #endif
2325         }
2326
2327         eth = eth_hdr(skb);
2328         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2329         did_rsc = false;
2330
2331         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2332             (ntohs(eth->h_proto) == ETH_P_IP ||
2333              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2334                 did_rsc = route_shortcircuit(dev, skb);
2335                 if (did_rsc)
2336                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2337         }
2338
2339         if (f == NULL) {
2340                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2341                 if (f == NULL) {
2342                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2343                             !is_multicast_ether_addr(eth->h_dest))
2344                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2345
2346                         dev->stats.tx_dropped++;
2347                         kfree_skb(skb);
2348                         return NETDEV_TX_OK;
2349                 }
2350         }
2351
2352         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2353                 struct sk_buff *skb1;
2354
2355                 if (!fdst) {
2356                         fdst = rdst;
2357                         continue;
2358                 }
2359                 skb1 = skb_clone(skb, GFP_ATOMIC);
2360                 if (skb1)
2361                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2362         }
2363
2364         if (fdst)
2365                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2366         else
2367                 kfree_skb(skb);
2368         return NETDEV_TX_OK;
2369 }
2370
2371 /* Walk the forwarding table and purge stale entries */
2372 static void vxlan_cleanup(struct timer_list *t)
2373 {
2374         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2375         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2376         unsigned int h;
2377
2378         if (!netif_running(vxlan->dev))
2379                 return;
2380
2381         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2382                 struct hlist_node *p, *n;
2383
2384                 spin_lock_bh(&vxlan->hash_lock);
2385                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2386                         struct vxlan_fdb *f
2387                                 = container_of(p, struct vxlan_fdb, hlist);
2388                         unsigned long timeout;
2389
2390                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2391                                 continue;
2392
2393                         if (f->flags & NTF_EXT_LEARNED)
2394                                 continue;
2395
2396                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2397                         if (time_before_eq(timeout, jiffies)) {
2398                                 netdev_dbg(vxlan->dev,
2399                                            "garbage collect %pM\n",
2400                                            f->eth_addr);
2401                                 f->state = NUD_STALE;
2402                                 vxlan_fdb_destroy(vxlan, f, true);
2403                         } else if (time_before(timeout, next_timer))
2404                                 next_timer = timeout;
2405                 }
2406                 spin_unlock_bh(&vxlan->hash_lock);
2407         }
2408
2409         mod_timer(&vxlan->age_timer, next_timer);
2410 }
2411
2412 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2413 {
2414         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2415
2416         spin_lock(&vn->sock_lock);
2417         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2418 #if IS_ENABLED(CONFIG_IPV6)
2419         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2420 #endif
2421         spin_unlock(&vn->sock_lock);
2422 }
2423
2424 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2425                              struct vxlan_dev_node *node)
2426 {
2427         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2428         __be32 vni = vxlan->default_dst.remote_vni;
2429
2430         node->vxlan = vxlan;
2431         spin_lock(&vn->sock_lock);
2432         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2433         spin_unlock(&vn->sock_lock);
2434 }
2435
2436 /* Setup stats when device is created */
2437 static int vxlan_init(struct net_device *dev)
2438 {
2439         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2440         if (!dev->tstats)
2441                 return -ENOMEM;
2442
2443         return 0;
2444 }
2445
2446 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2447 {
2448         struct vxlan_fdb *f;
2449
2450         spin_lock_bh(&vxlan->hash_lock);
2451         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2452         if (f)
2453                 vxlan_fdb_destroy(vxlan, f, true);
2454         spin_unlock_bh(&vxlan->hash_lock);
2455 }
2456
2457 static void vxlan_uninit(struct net_device *dev)
2458 {
2459         struct vxlan_dev *vxlan = netdev_priv(dev);
2460
2461         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2462
2463         free_percpu(dev->tstats);
2464 }
2465
2466 /* Start ageing timer and join group when device is brought up */
2467 static int vxlan_open(struct net_device *dev)
2468 {
2469         struct vxlan_dev *vxlan = netdev_priv(dev);
2470         int ret;
2471
2472         ret = vxlan_sock_add(vxlan);
2473         if (ret < 0)
2474                 return ret;
2475
2476         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2477                 ret = vxlan_igmp_join(vxlan);
2478                 if (ret == -EADDRINUSE)
2479                         ret = 0;
2480                 if (ret) {
2481                         vxlan_sock_release(vxlan);
2482                         return ret;
2483                 }
2484         }
2485
2486         if (vxlan->cfg.age_interval)
2487                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2488
2489         return ret;
2490 }
2491
2492 /* Purge the forwarding table */
2493 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2494 {
2495         unsigned int h;
2496
2497         spin_lock_bh(&vxlan->hash_lock);
2498         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2499                 struct hlist_node *p, *n;
2500                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2501                         struct vxlan_fdb *f
2502                                 = container_of(p, struct vxlan_fdb, hlist);
2503                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2504                                 continue;
2505                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2506                         if (!is_zero_ether_addr(f->eth_addr))
2507                                 vxlan_fdb_destroy(vxlan, f, true);
2508                 }
2509         }
2510         spin_unlock_bh(&vxlan->hash_lock);
2511 }
2512
2513 /* Cleanup timer and forwarding table on shutdown */
2514 static int vxlan_stop(struct net_device *dev)
2515 {
2516         struct vxlan_dev *vxlan = netdev_priv(dev);
2517         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2518         int ret = 0;
2519
2520         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2521             !vxlan_group_used(vn, vxlan))
2522                 ret = vxlan_igmp_leave(vxlan);
2523
2524         del_timer_sync(&vxlan->age_timer);
2525
2526         vxlan_flush(vxlan, false);
2527         vxlan_sock_release(vxlan);
2528
2529         return ret;
2530 }
2531
2532 /* Stub, nothing needs to be done. */
2533 static void vxlan_set_multicast_list(struct net_device *dev)
2534 {
2535 }
2536
2537 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2538 {
2539         struct vxlan_dev *vxlan = netdev_priv(dev);
2540         struct vxlan_rdst *dst = &vxlan->default_dst;
2541         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2542                                                          dst->remote_ifindex);
2543         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
2544
2545         /* This check is different than dev->max_mtu, because it looks at
2546          * the lowerdev->mtu, rather than the static dev->max_mtu
2547          */
2548         if (lowerdev) {
2549                 int max_mtu = lowerdev->mtu -
2550                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2551                 if (new_mtu > max_mtu)
2552                         return -EINVAL;
2553         }
2554
2555         dev->mtu = new_mtu;
2556         return 0;
2557 }
2558
2559 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2560 {
2561         struct vxlan_dev *vxlan = netdev_priv(dev);
2562         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2563         __be16 sport, dport;
2564
2565         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2566                                   vxlan->cfg.port_max, true);
2567         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2568
2569         if (ip_tunnel_info_af(info) == AF_INET) {
2570                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2571                 struct rtable *rt;
2572
2573                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2574                                      info->key.u.ipv4.dst,
2575                                      &info->key.u.ipv4.src, dport, sport,
2576                                      &info->dst_cache, info);
2577                 if (IS_ERR(rt))
2578                         return PTR_ERR(rt);
2579                 ip_rt_put(rt);
2580         } else {
2581 #if IS_ENABLED(CONFIG_IPV6)
2582                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2583                 struct dst_entry *ndst;
2584
2585                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2586                                         info->key.label, &info->key.u.ipv6.dst,
2587                                         &info->key.u.ipv6.src, dport, sport,
2588                                         &info->dst_cache, info);
2589                 if (IS_ERR(ndst))
2590                         return PTR_ERR(ndst);
2591                 dst_release(ndst);
2592 #else /* !CONFIG_IPV6 */
2593                 return -EPFNOSUPPORT;
2594 #endif
2595         }
2596         info->key.tp_src = sport;
2597         info->key.tp_dst = dport;
2598         return 0;
2599 }
2600
2601 static const struct net_device_ops vxlan_netdev_ether_ops = {
2602         .ndo_init               = vxlan_init,
2603         .ndo_uninit             = vxlan_uninit,
2604         .ndo_open               = vxlan_open,
2605         .ndo_stop               = vxlan_stop,
2606         .ndo_start_xmit         = vxlan_xmit,
2607         .ndo_get_stats64        = ip_tunnel_get_stats64,
2608         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2609         .ndo_change_mtu         = vxlan_change_mtu,
2610         .ndo_validate_addr      = eth_validate_addr,
2611         .ndo_set_mac_address    = eth_mac_addr,
2612         .ndo_fdb_add            = vxlan_fdb_add,
2613         .ndo_fdb_del            = vxlan_fdb_delete,
2614         .ndo_fdb_dump           = vxlan_fdb_dump,
2615         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2616 };
2617
2618 static const struct net_device_ops vxlan_netdev_raw_ops = {
2619         .ndo_init               = vxlan_init,
2620         .ndo_uninit             = vxlan_uninit,
2621         .ndo_open               = vxlan_open,
2622         .ndo_stop               = vxlan_stop,
2623         .ndo_start_xmit         = vxlan_xmit,
2624         .ndo_get_stats64        = ip_tunnel_get_stats64,
2625         .ndo_change_mtu         = vxlan_change_mtu,
2626         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2627 };
2628
2629 /* Info for udev, that this is a virtual tunnel endpoint */
2630 static struct device_type vxlan_type = {
2631         .name = "vxlan",
2632 };
2633
2634 /* Calls the ndo_udp_tunnel_add of the caller in order to
2635  * supply the listening VXLAN udp ports. Callers are expected
2636  * to implement the ndo_udp_tunnel_add.
2637  */
2638 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
2639 {
2640         struct vxlan_sock *vs;
2641         struct net *net = dev_net(dev);
2642         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2643         unsigned int i;
2644
2645         spin_lock(&vn->sock_lock);
2646         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2647                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2648                         unsigned short type;
2649
2650                         if (vs->flags & VXLAN_F_GPE)
2651                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
2652                         else
2653                                 type = UDP_TUNNEL_TYPE_VXLAN;
2654
2655                         if (push)
2656                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
2657                         else
2658                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
2659                 }
2660         }
2661         spin_unlock(&vn->sock_lock);
2662 }
2663
2664 /* Initialize the device structure. */
2665 static void vxlan_setup(struct net_device *dev)
2666 {
2667         struct vxlan_dev *vxlan = netdev_priv(dev);
2668         unsigned int h;
2669
2670         eth_hw_addr_random(dev);
2671         ether_setup(dev);
2672
2673         dev->needs_free_netdev = true;
2674         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2675
2676         dev->features   |= NETIF_F_LLTX;
2677         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2678         dev->features   |= NETIF_F_RXCSUM;
2679         dev->features   |= NETIF_F_GSO_SOFTWARE;
2680
2681         dev->vlan_features = dev->features;
2682         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2683         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2684         netif_keep_dst(dev);
2685         dev->priv_flags |= IFF_NO_QUEUE;
2686
2687         /* MTU range: 68 - 65535 */
2688         dev->min_mtu = ETH_MIN_MTU;
2689         dev->max_mtu = ETH_MAX_MTU;
2690
2691         INIT_LIST_HEAD(&vxlan->next);
2692         spin_lock_init(&vxlan->hash_lock);
2693
2694         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
2695
2696         vxlan->dev = dev;
2697
2698         gro_cells_init(&vxlan->gro_cells, dev);
2699
2700         for (h = 0; h < FDB_HASH_SIZE; ++h)
2701                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2702 }
2703
2704 static void vxlan_ether_setup(struct net_device *dev)
2705 {
2706         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2707         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2708         dev->netdev_ops = &vxlan_netdev_ether_ops;
2709 }
2710
2711 static void vxlan_raw_setup(struct net_device *dev)
2712 {
2713         dev->header_ops = NULL;
2714         dev->type = ARPHRD_NONE;
2715         dev->hard_header_len = 0;
2716         dev->addr_len = 0;
2717         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2718         dev->netdev_ops = &vxlan_netdev_raw_ops;
2719 }
2720
2721 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2722         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2723         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2724         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2725         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2726         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2727         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2728         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2729         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2730         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2731         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2732         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2733         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2734         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2735         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2736         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2737         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2738         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2739         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2740         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2741         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2742         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2743         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2744         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2745         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2746         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2747         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2748         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2749         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
2750 };
2751
2752 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
2753                           struct netlink_ext_ack *extack)
2754 {
2755         if (tb[IFLA_ADDRESS]) {
2756                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2757                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2758                                             "Provided link layer address is not Ethernet");
2759                         return -EINVAL;
2760                 }
2761
2762                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2763                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
2764                                             "Provided Ethernet address is not unicast");
2765                         return -EADDRNOTAVAIL;
2766                 }
2767         }
2768
2769         if (tb[IFLA_MTU]) {
2770                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
2771
2772                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
2773                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
2774                                             "MTU must be between 68 and 65535");
2775                         return -EINVAL;
2776                 }
2777         }
2778
2779         if (!data) {
2780                 NL_SET_ERR_MSG(extack,
2781                                "Required attributes not provided to perform the operation");
2782                 return -EINVAL;
2783         }
2784
2785         if (data[IFLA_VXLAN_ID]) {
2786                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2787
2788                 if (id >= VXLAN_N_VID) {
2789                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID],
2790                                             "VXLAN ID must be lower than 16777216");
2791                         return -ERANGE;
2792                 }
2793         }
2794
2795         if (data[IFLA_VXLAN_PORT_RANGE]) {
2796                 const struct ifla_vxlan_port_range *p
2797                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2798
2799                 if (ntohs(p->high) < ntohs(p->low)) {
2800                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
2801                                             "Invalid source port range");
2802                         return -EINVAL;
2803                 }
2804         }
2805
2806         return 0;
2807 }
2808
2809 static void vxlan_get_drvinfo(struct net_device *netdev,
2810                               struct ethtool_drvinfo *drvinfo)
2811 {
2812         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2813         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2814 }
2815
2816 static const struct ethtool_ops vxlan_ethtool_ops = {
2817         .get_drvinfo    = vxlan_get_drvinfo,
2818         .get_link       = ethtool_op_get_link,
2819 };
2820
2821 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2822                                         __be16 port, u32 flags)
2823 {
2824         struct socket *sock;
2825         struct udp_port_cfg udp_conf;
2826         int err;
2827
2828         memset(&udp_conf, 0, sizeof(udp_conf));
2829
2830         if (ipv6) {
2831                 udp_conf.family = AF_INET6;
2832                 udp_conf.use_udp6_rx_checksums =
2833                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2834                 udp_conf.ipv6_v6only = 1;
2835         } else {
2836                 udp_conf.family = AF_INET;
2837         }
2838
2839         udp_conf.local_udp_port = port;
2840
2841         /* Open UDP socket */
2842         err = udp_sock_create(net, &udp_conf, &sock);
2843         if (err < 0)
2844                 return ERR_PTR(err);
2845
2846         return sock;
2847 }
2848
2849 /* Create new listen socket if needed */
2850 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2851                                               __be16 port, u32 flags)
2852 {
2853         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2854         struct vxlan_sock *vs;
2855         struct socket *sock;
2856         unsigned int h;
2857         struct udp_tunnel_sock_cfg tunnel_cfg;
2858
2859         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2860         if (!vs)
2861                 return ERR_PTR(-ENOMEM);
2862
2863         for (h = 0; h < VNI_HASH_SIZE; ++h)
2864                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2865
2866         sock = vxlan_create_sock(net, ipv6, port, flags);
2867         if (IS_ERR(sock)) {
2868                 kfree(vs);
2869                 return ERR_CAST(sock);
2870         }
2871
2872         vs->sock = sock;
2873         refcount_set(&vs->refcnt, 1);
2874         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2875
2876         spin_lock(&vn->sock_lock);
2877         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2878         udp_tunnel_notify_add_rx_port(sock,
2879                                       (vs->flags & VXLAN_F_GPE) ?
2880                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2881                                       UDP_TUNNEL_TYPE_VXLAN);
2882         spin_unlock(&vn->sock_lock);
2883
2884         /* Mark socket as an encapsulation socket. */
2885         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2886         tunnel_cfg.sk_user_data = vs;
2887         tunnel_cfg.encap_type = 1;
2888         tunnel_cfg.encap_rcv = vxlan_rcv;
2889         tunnel_cfg.encap_destroy = NULL;
2890         tunnel_cfg.gro_receive = vxlan_gro_receive;
2891         tunnel_cfg.gro_complete = vxlan_gro_complete;
2892
2893         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2894
2895         return vs;
2896 }
2897
2898 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2899 {
2900         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2901         struct vxlan_sock *vs = NULL;
2902         struct vxlan_dev_node *node;
2903
2904         if (!vxlan->cfg.no_share) {
2905                 spin_lock(&vn->sock_lock);
2906                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2907                                      vxlan->cfg.dst_port, vxlan->cfg.flags);
2908                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
2909                         spin_unlock(&vn->sock_lock);
2910                         return -EBUSY;
2911                 }
2912                 spin_unlock(&vn->sock_lock);
2913         }
2914         if (!vs)
2915                 vs = vxlan_socket_create(vxlan->net, ipv6,
2916                                          vxlan->cfg.dst_port, vxlan->cfg.flags);
2917         if (IS_ERR(vs))
2918                 return PTR_ERR(vs);
2919 #if IS_ENABLED(CONFIG_IPV6)
2920         if (ipv6) {
2921                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2922                 node = &vxlan->hlist6;
2923         } else
2924 #endif
2925         {
2926                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2927                 node = &vxlan->hlist4;
2928         }
2929         vxlan_vs_add_dev(vs, vxlan, node);
2930         return 0;
2931 }
2932
2933 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2934 {
2935         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
2936         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
2937         bool ipv4 = !ipv6 || metadata;
2938         int ret = 0;
2939
2940         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2941 #if IS_ENABLED(CONFIG_IPV6)
2942         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2943         if (ipv6) {
2944                 ret = __vxlan_sock_add(vxlan, true);
2945                 if (ret < 0 && ret != -EAFNOSUPPORT)
2946                         ipv4 = false;
2947         }
2948 #endif
2949         if (ipv4)
2950                 ret = __vxlan_sock_add(vxlan, false);
2951         if (ret < 0)
2952                 vxlan_sock_release(vxlan);
2953         return ret;
2954 }
2955
2956 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
2957                                  struct net_device **lower,
2958                                  struct vxlan_dev *old,
2959                                  struct netlink_ext_ack *extack)
2960 {
2961         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2962         struct vxlan_dev *tmp;
2963         bool use_ipv6 = false;
2964
2965         if (conf->flags & VXLAN_F_GPE) {
2966                 /* For now, allow GPE only together with
2967                  * COLLECT_METADATA. This can be relaxed later; in such
2968                  * case, the other side of the PtP link will have to be
2969                  * provided.
2970                  */
2971                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2972                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2973                         NL_SET_ERR_MSG(extack,
2974                                        "VXLAN GPE does not support this combination of attributes");
2975                         return -EINVAL;
2976                 }
2977         }
2978
2979         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
2980                 /* Unless IPv6 is explicitly requested, assume IPv4 */
2981                 conf->remote_ip.sa.sa_family = AF_INET;
2982                 conf->saddr.sa.sa_family = AF_INET;
2983         } else if (!conf->remote_ip.sa.sa_family) {
2984                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
2985         } else if (!conf->saddr.sa.sa_family) {
2986                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
2987         }
2988
2989         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
2990                 NL_SET_ERR_MSG(extack,
2991                                "Local and remote address must be from the same family");
2992                 return -EINVAL;
2993         }
2994
2995         if (vxlan_addr_multicast(&conf->saddr)) {
2996                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
2997                 return -EINVAL;
2998         }
2999
3000         if (conf->saddr.sa.sa_family == AF_INET6) {
3001                 if (!IS_ENABLED(CONFIG_IPV6)) {
3002                         NL_SET_ERR_MSG(extack,
3003                                        "IPv6 support not enabled in the kernel");
3004                         return -EPFNOSUPPORT;
3005                 }
3006                 use_ipv6 = true;
3007                 conf->flags |= VXLAN_F_IPV6;
3008
3009                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3010                         int local_type =
3011                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3012                         int remote_type =
3013                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3014
3015                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3016                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3017                                     (remote_type != IPV6_ADDR_ANY)) {
3018                                         NL_SET_ERR_MSG(extack,
3019                                                        "Invalid combination of local and remote address scopes");
3020                                         return -EINVAL;
3021                                 }
3022
3023                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3024                         } else {
3025                                 if (remote_type ==
3026                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3027                                         NL_SET_ERR_MSG(extack,
3028                                                        "Invalid combination of local and remote address scopes");
3029                                         return -EINVAL;
3030                                 }
3031
3032                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3033                         }
3034                 }
3035         }
3036
3037         if (conf->label && !use_ipv6) {
3038                 NL_SET_ERR_MSG(extack,
3039                                "Label attribute only applies to IPv6 VXLAN devices");
3040                 return -EINVAL;
3041         }
3042
3043         if (conf->remote_ifindex) {
3044                 struct net_device *lowerdev;
3045
3046                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3047                 if (!lowerdev) {
3048                         NL_SET_ERR_MSG(extack,
3049                                        "Invalid local interface, device not found");
3050                         return -ENODEV;
3051                 }
3052
3053 #if IS_ENABLED(CONFIG_IPV6)
3054                 if (use_ipv6) {
3055                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3056                         if (idev && idev->cnf.disable_ipv6) {
3057                                 NL_SET_ERR_MSG(extack,
3058                                                "IPv6 support disabled by administrator");
3059                                 return -EPERM;
3060                         }
3061                 }
3062 #endif
3063
3064                 *lower = lowerdev;
3065         } else {
3066                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3067                         NL_SET_ERR_MSG(extack,
3068                                        "Local interface required for multicast remote destination");
3069
3070                         return -EINVAL;
3071                 }
3072
3073 #if IS_ENABLED(CONFIG_IPV6)
3074                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3075                         NL_SET_ERR_MSG(extack,
3076                                        "Local interface required for link-local local/remote addresses");
3077                         return -EINVAL;
3078                 }
3079 #endif
3080
3081                 *lower = NULL;
3082         }
3083
3084         if (!conf->dst_port) {
3085                 if (conf->flags & VXLAN_F_GPE)
3086                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3087                 else
3088                         conf->dst_port = htons(vxlan_port);
3089         }
3090
3091         if (!conf->age_interval)
3092                 conf->age_interval = FDB_AGE_DEFAULT;
3093
3094         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3095                 if (tmp == old)
3096                         continue;
3097
3098                 if (tmp->cfg.vni != conf->vni)
3099                         continue;
3100                 if (tmp->cfg.dst_port != conf->dst_port)
3101                         continue;
3102                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3103                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3104                         continue;
3105
3106                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3107                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3108                         continue;
3109
3110                 NL_SET_ERR_MSG(extack,
3111                                "A VXLAN device with the specified VNI already exists");
3112                 return -EEXIST;
3113         }
3114
3115         return 0;
3116 }
3117
3118 static void vxlan_config_apply(struct net_device *dev,
3119                                struct vxlan_config *conf,
3120                                struct net_device *lowerdev,
3121                                struct net *src_net,
3122                                bool changelink)
3123 {
3124         struct vxlan_dev *vxlan = netdev_priv(dev);
3125         struct vxlan_rdst *dst = &vxlan->default_dst;
3126         unsigned short needed_headroom = ETH_HLEN;
3127         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3128         int max_mtu = ETH_MAX_MTU;
3129
3130         if (!changelink) {
3131                 if (conf->flags & VXLAN_F_GPE)
3132                         vxlan_raw_setup(dev);
3133                 else
3134                         vxlan_ether_setup(dev);
3135
3136                 if (conf->mtu)
3137                         dev->mtu = conf->mtu;
3138
3139                 vxlan->net = src_net;
3140         }
3141
3142         dst->remote_vni = conf->vni;
3143
3144         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3145
3146         if (lowerdev) {
3147                 dst->remote_ifindex = conf->remote_ifindex;
3148
3149                 dev->gso_max_size = lowerdev->gso_max_size;
3150                 dev->gso_max_segs = lowerdev->gso_max_segs;
3151
3152                 needed_headroom = lowerdev->hard_header_len;
3153
3154                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3155                                            VXLAN_HEADROOM);
3156                 if (max_mtu < ETH_MIN_MTU)
3157                         max_mtu = ETH_MIN_MTU;
3158
3159                 if (!changelink && !conf->mtu)
3160                         dev->mtu = max_mtu;
3161         }
3162
3163         if (dev->mtu > max_mtu)
3164                 dev->mtu = max_mtu;
3165
3166         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3167                 needed_headroom += VXLAN6_HEADROOM;
3168         else
3169                 needed_headroom += VXLAN_HEADROOM;
3170         dev->needed_headroom = needed_headroom;
3171
3172         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3173 }
3174
3175 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3176                                struct vxlan_config *conf, bool changelink,
3177                                struct netlink_ext_ack *extack)
3178 {
3179         struct vxlan_dev *vxlan = netdev_priv(dev);
3180         struct net_device *lowerdev;
3181         int ret;
3182
3183         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3184         if (ret)
3185                 return ret;
3186
3187         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3188
3189         return 0;
3190 }
3191
3192 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3193                               struct vxlan_config *conf,
3194                               struct netlink_ext_ack *extack)
3195 {
3196         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3197         struct vxlan_dev *vxlan = netdev_priv(dev);
3198         struct vxlan_fdb *f = NULL;
3199         int err;
3200
3201         err = vxlan_dev_configure(net, dev, conf, false, extack);
3202         if (err)
3203                 return err;
3204
3205         dev->ethtool_ops = &vxlan_ethtool_ops;
3206
3207         /* create an fdb entry for a valid default destination */
3208         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
3209                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3210                                        &vxlan->default_dst.remote_ip,
3211                                        NUD_REACHABLE | NUD_PERMANENT,
3212                                        vxlan->cfg.dst_port,
3213                                        vxlan->default_dst.remote_vni,
3214                                        vxlan->default_dst.remote_vni,
3215                                        vxlan->default_dst.remote_ifindex,
3216                                        NTF_SELF, &f);
3217                 if (err)
3218                         return err;
3219         }
3220
3221         err = register_netdevice(dev);
3222         if (err)
3223                 goto errout;
3224
3225         err = rtnl_configure_link(dev, NULL);
3226         if (err) {
3227                 unregister_netdevice(dev);
3228                 goto errout;
3229         }
3230
3231         /* notify default fdb entry */
3232         if (f)
3233                 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH);
3234
3235         list_add(&vxlan->next, &vn->vxlan_list);
3236         return 0;
3237 errout:
3238         if (f)
3239                 vxlan_fdb_destroy(vxlan, f, false);
3240         return err;
3241 }
3242
3243 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3244                          struct net_device *dev, struct vxlan_config *conf,
3245                          bool changelink)
3246 {
3247         struct vxlan_dev *vxlan = netdev_priv(dev);
3248
3249         memset(conf, 0, sizeof(*conf));
3250
3251         /* if changelink operation, start with old existing cfg */
3252         if (changelink)
3253                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3254
3255         if (data[IFLA_VXLAN_ID]) {
3256                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3257
3258                 if (changelink && (vni != conf->vni))
3259                         return -EOPNOTSUPP;
3260                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3261         }
3262
3263         if (data[IFLA_VXLAN_GROUP]) {
3264                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET))
3265                         return -EOPNOTSUPP;
3266
3267                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3268                 conf->remote_ip.sa.sa_family = AF_INET;
3269         } else if (data[IFLA_VXLAN_GROUP6]) {
3270                 if (!IS_ENABLED(CONFIG_IPV6))
3271                         return -EPFNOSUPPORT;
3272
3273                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6))
3274                         return -EOPNOTSUPP;
3275
3276                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3277                 conf->remote_ip.sa.sa_family = AF_INET6;
3278         }
3279
3280         if (data[IFLA_VXLAN_LOCAL]) {
3281                 if (changelink && (conf->saddr.sa.sa_family != AF_INET))
3282                         return -EOPNOTSUPP;
3283
3284                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3285                 conf->saddr.sa.sa_family = AF_INET;
3286         } else if (data[IFLA_VXLAN_LOCAL6]) {
3287                 if (!IS_ENABLED(CONFIG_IPV6))
3288                         return -EPFNOSUPPORT;
3289
3290                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6))
3291                         return -EOPNOTSUPP;
3292
3293                 /* TODO: respect scope id */
3294                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3295                 conf->saddr.sa.sa_family = AF_INET6;
3296         }
3297
3298         if (data[IFLA_VXLAN_LINK])
3299                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3300
3301         if (data[IFLA_VXLAN_TOS])
3302                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3303
3304         if (data[IFLA_VXLAN_TTL])
3305                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3306
3307         if (data[IFLA_VXLAN_TTL_INHERIT]) {
3308                 if (changelink)
3309                         return -EOPNOTSUPP;
3310                 conf->flags |= VXLAN_F_TTL_INHERIT;
3311         }
3312
3313         if (data[IFLA_VXLAN_LABEL])
3314                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3315                              IPV6_FLOWLABEL_MASK;
3316
3317         if (data[IFLA_VXLAN_LEARNING]) {
3318                 if (nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3319                         conf->flags |= VXLAN_F_LEARN;
3320                 else
3321                         conf->flags &= ~VXLAN_F_LEARN;
3322         } else if (!changelink) {
3323                 /* default to learn on a new device */
3324                 conf->flags |= VXLAN_F_LEARN;
3325         }
3326
3327         if (data[IFLA_VXLAN_AGEING]) {
3328                 if (changelink)
3329                         return -EOPNOTSUPP;
3330                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3331         }
3332
3333         if (data[IFLA_VXLAN_PROXY]) {
3334                 if (changelink)
3335                         return -EOPNOTSUPP;
3336                 if (nla_get_u8(data[IFLA_VXLAN_PROXY]))
3337                         conf->flags |= VXLAN_F_PROXY;
3338         }
3339
3340         if (data[IFLA_VXLAN_RSC]) {
3341                 if (changelink)
3342                         return -EOPNOTSUPP;
3343                 if (nla_get_u8(data[IFLA_VXLAN_RSC]))
3344                         conf->flags |= VXLAN_F_RSC;
3345         }
3346
3347         if (data[IFLA_VXLAN_L2MISS]) {
3348                 if (changelink)
3349                         return -EOPNOTSUPP;
3350                 if (nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3351                         conf->flags |= VXLAN_F_L2MISS;
3352         }
3353
3354         if (data[IFLA_VXLAN_L3MISS]) {
3355                 if (changelink)
3356                         return -EOPNOTSUPP;
3357                 if (nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3358                         conf->flags |= VXLAN_F_L3MISS;
3359         }
3360
3361         if (data[IFLA_VXLAN_LIMIT]) {
3362                 if (changelink)
3363                         return -EOPNOTSUPP;
3364                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3365         }
3366
3367         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3368                 if (changelink)
3369                         return -EOPNOTSUPP;
3370                 if (nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3371                         conf->flags |= VXLAN_F_COLLECT_METADATA;
3372         }
3373
3374         if (data[IFLA_VXLAN_PORT_RANGE]) {
3375                 if (!changelink) {
3376                         const struct ifla_vxlan_port_range *p
3377                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3378                         conf->port_min = ntohs(p->low);
3379                         conf->port_max = ntohs(p->high);
3380                 } else {
3381                         return -EOPNOTSUPP;
3382                 }
3383         }
3384
3385         if (data[IFLA_VXLAN_PORT]) {
3386                 if (changelink)
3387                         return -EOPNOTSUPP;
3388                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3389         }
3390
3391         if (data[IFLA_VXLAN_UDP_CSUM]) {
3392                 if (changelink)
3393                         return -EOPNOTSUPP;
3394                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3395                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3396         }
3397
3398         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3399                 if (changelink)
3400                         return -EOPNOTSUPP;
3401                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3402                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3403         }
3404
3405         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3406                 if (changelink)
3407                         return -EOPNOTSUPP;
3408                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3409                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3410         }
3411
3412         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3413                 if (changelink)
3414                         return -EOPNOTSUPP;
3415                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3416                         conf->flags |= VXLAN_F_REMCSUM_TX;
3417         }
3418
3419         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3420                 if (changelink)
3421                         return -EOPNOTSUPP;
3422                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3423                         conf->flags |= VXLAN_F_REMCSUM_RX;
3424         }
3425
3426         if (data[IFLA_VXLAN_GBP]) {
3427                 if (changelink)
3428                         return -EOPNOTSUPP;
3429                 conf->flags |= VXLAN_F_GBP;
3430         }
3431
3432         if (data[IFLA_VXLAN_GPE]) {
3433                 if (changelink)
3434                         return -EOPNOTSUPP;
3435                 conf->flags |= VXLAN_F_GPE;
3436         }
3437
3438         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3439                 if (changelink)
3440                         return -EOPNOTSUPP;
3441                 conf->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3442         }
3443
3444         if (tb[IFLA_MTU]) {
3445                 if (changelink)
3446                         return -EOPNOTSUPP;
3447                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3448         }
3449
3450         return 0;
3451 }
3452
3453 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3454                          struct nlattr *tb[], struct nlattr *data[],
3455                          struct netlink_ext_ack *extack)
3456 {
3457         struct vxlan_config conf;
3458         int err;
3459
3460         err = vxlan_nl2conf(tb, data, dev, &conf, false);
3461         if (err)
3462                 return err;
3463
3464         return __vxlan_dev_create(src_net, dev, &conf, extack);
3465 }
3466
3467 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3468                             struct nlattr *data[],
3469                             struct netlink_ext_ack *extack)
3470 {
3471         struct vxlan_dev *vxlan = netdev_priv(dev);
3472         struct vxlan_rdst *dst = &vxlan->default_dst;
3473         struct vxlan_rdst old_dst;
3474         struct vxlan_config conf;
3475         struct vxlan_fdb *f = NULL;
3476         int err;
3477
3478         err = vxlan_nl2conf(tb, data,
3479                             dev, &conf, true);
3480         if (err)
3481                 return err;
3482
3483         memcpy(&old_dst, dst, sizeof(struct vxlan_rdst));
3484
3485         err = vxlan_dev_configure(vxlan->net, dev, &conf, true, extack);
3486         if (err)
3487                 return err;
3488
3489         /* handle default dst entry */
3490         if (!vxlan_addr_equal(&dst->remote_ip, &old_dst.remote_ip)) {
3491                 spin_lock_bh(&vxlan->hash_lock);
3492                 if (!vxlan_addr_any(&old_dst.remote_ip))
3493                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
3494                                            old_dst.remote_ip,
3495                                            vxlan->cfg.dst_port,
3496                                            old_dst.remote_vni,
3497                                            old_dst.remote_vni,
3498                                            old_dst.remote_ifindex, 0);
3499
3500                 if (!vxlan_addr_any(&dst->remote_ip)) {
3501                         err = vxlan_fdb_create(vxlan, all_zeros_mac,
3502                                                &dst->remote_ip,
3503                                                NUD_REACHABLE | NUD_PERMANENT,
3504                                                vxlan->cfg.dst_port,
3505                                                dst->remote_vni,
3506                                                dst->remote_vni,
3507                                                dst->remote_ifindex,
3508                                                NTF_SELF, &f);
3509                         if (err) {
3510                                 spin_unlock_bh(&vxlan->hash_lock);
3511                                 return err;
3512                         }
3513                         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH);
3514                 }
3515                 spin_unlock_bh(&vxlan->hash_lock);
3516         }
3517
3518         return 0;
3519 }
3520
3521 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3522 {
3523         struct vxlan_dev *vxlan = netdev_priv(dev);
3524
3525         vxlan_flush(vxlan, true);
3526
3527         gro_cells_destroy(&vxlan->gro_cells);
3528         list_del(&vxlan->next);
3529         unregister_netdevice_queue(dev, head);
3530 }
3531
3532 static size_t vxlan_get_size(const struct net_device *dev)
3533 {
3534
3535         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3536                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3537                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3538                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3539                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3540                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3541                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3542                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3543                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3544                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3545                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3546                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3547                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3548                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3549                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3550                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3551                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3552                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3553                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3554                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3555                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3556                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3557                 0;
3558 }
3559
3560 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3561 {
3562         const struct vxlan_dev *vxlan = netdev_priv(dev);
3563         const struct vxlan_rdst *dst = &vxlan->default_dst;
3564         struct ifla_vxlan_port_range ports = {
3565                 .low =  htons(vxlan->cfg.port_min),
3566                 .high = htons(vxlan->cfg.port_max),
3567         };
3568
3569         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3570                 goto nla_put_failure;
3571
3572         if (!vxlan_addr_any(&dst->remote_ip)) {
3573                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3574                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3575                                             dst->remote_ip.sin.sin_addr.s_addr))
3576                                 goto nla_put_failure;
3577 #if IS_ENABLED(CONFIG_IPV6)
3578                 } else {
3579                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3580                                              &dst->remote_ip.sin6.sin6_addr))
3581                                 goto nla_put_failure;
3582 #endif
3583                 }
3584         }
3585
3586         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3587                 goto nla_put_failure;
3588
3589         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3590                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3591                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3592                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3593                                 goto nla_put_failure;
3594 #if IS_ENABLED(CONFIG_IPV6)
3595                 } else {
3596                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3597                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3598                                 goto nla_put_failure;
3599 #endif
3600                 }
3601         }
3602
3603         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3604             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3605             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3606             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3607                         !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
3608             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3609                         !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
3610             nla_put_u8(skb, IFLA_VXLAN_RSC,
3611                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
3612             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3613                         !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
3614             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3615                         !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
3616             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3617                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
3618             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3619             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3620             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3621             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3622                         !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3623             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3624                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3625             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3626                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3627             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3628                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
3629             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3630                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
3631                 goto nla_put_failure;
3632
3633         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3634                 goto nla_put_failure;
3635
3636         if (vxlan->cfg.flags & VXLAN_F_GBP &&
3637             nla_put_flag(skb, IFLA_VXLAN_GBP))
3638                 goto nla_put_failure;
3639
3640         if (vxlan->cfg.flags & VXLAN_F_GPE &&
3641             nla_put_flag(skb, IFLA_VXLAN_GPE))
3642                 goto nla_put_failure;
3643
3644         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3645             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3646                 goto nla_put_failure;
3647
3648         return 0;
3649
3650 nla_put_failure:
3651         return -EMSGSIZE;
3652 }
3653
3654 static struct net *vxlan_get_link_net(const struct net_device *dev)
3655 {
3656         struct vxlan_dev *vxlan = netdev_priv(dev);
3657
3658         return vxlan->net;
3659 }
3660
3661 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3662         .kind           = "vxlan",
3663         .maxtype        = IFLA_VXLAN_MAX,
3664         .policy         = vxlan_policy,
3665         .priv_size      = sizeof(struct vxlan_dev),
3666         .setup          = vxlan_setup,
3667         .validate       = vxlan_validate,
3668         .newlink        = vxlan_newlink,
3669         .changelink     = vxlan_changelink,
3670         .dellink        = vxlan_dellink,
3671         .get_size       = vxlan_get_size,
3672         .fill_info      = vxlan_fill_info,
3673         .get_link_net   = vxlan_get_link_net,
3674 };
3675
3676 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3677                                     u8 name_assign_type,
3678                                     struct vxlan_config *conf)
3679 {
3680         struct nlattr *tb[IFLA_MAX + 1];
3681         struct net_device *dev;
3682         int err;
3683
3684         memset(&tb, 0, sizeof(tb));
3685
3686         dev = rtnl_create_link(net, name, name_assign_type,
3687                                &vxlan_link_ops, tb);
3688         if (IS_ERR(dev))
3689                 return dev;
3690
3691         err = __vxlan_dev_create(net, dev, conf, NULL);
3692         if (err < 0) {
3693                 free_netdev(dev);
3694                 return ERR_PTR(err);
3695         }
3696
3697         err = rtnl_configure_link(dev, NULL);
3698         if (err < 0) {
3699                 LIST_HEAD(list_kill);
3700
3701                 vxlan_dellink(dev, &list_kill);
3702                 unregister_netdevice_many(&list_kill);
3703                 return ERR_PTR(err);
3704         }
3705
3706         return dev;
3707 }
3708 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3709
3710 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3711                                              struct net_device *dev)
3712 {
3713         struct vxlan_dev *vxlan, *next;
3714         LIST_HEAD(list_kill);
3715
3716         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3717                 struct vxlan_rdst *dst = &vxlan->default_dst;
3718
3719                 /* In case we created vxlan device with carrier
3720                  * and we loose the carrier due to module unload
3721                  * we also need to remove vxlan device. In other
3722                  * cases, it's not necessary and remote_ifindex
3723                  * is 0 here, so no matches.
3724                  */
3725                 if (dst->remote_ifindex == dev->ifindex)
3726                         vxlan_dellink(vxlan->dev, &list_kill);
3727         }
3728
3729         unregister_netdevice_many(&list_kill);
3730 }
3731
3732 static int vxlan_netdevice_event(struct notifier_block *unused,
3733                                  unsigned long event, void *ptr)
3734 {
3735         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3736         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3737
3738         if (event == NETDEV_UNREGISTER) {
3739                 vxlan_offload_rx_ports(dev, false);
3740                 vxlan_handle_lowerdev_unregister(vn, dev);
3741         } else if (event == NETDEV_REGISTER) {
3742                 vxlan_offload_rx_ports(dev, true);
3743         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
3744                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
3745                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
3746         }
3747
3748         return NOTIFY_DONE;
3749 }
3750
3751 static struct notifier_block vxlan_notifier_block __read_mostly = {
3752         .notifier_call = vxlan_netdevice_event,
3753 };
3754
3755 static __net_init int vxlan_init_net(struct net *net)
3756 {
3757         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3758         unsigned int h;
3759
3760         INIT_LIST_HEAD(&vn->vxlan_list);
3761         spin_lock_init(&vn->sock_lock);
3762
3763         for (h = 0; h < PORT_HASH_SIZE; ++h)
3764                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3765
3766         return 0;
3767 }
3768
3769 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
3770 {
3771         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3772         struct vxlan_dev *vxlan, *next;
3773         struct net_device *dev, *aux;
3774         unsigned int h;
3775
3776         for_each_netdev_safe(net, dev, aux)
3777                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3778                         unregister_netdevice_queue(dev, head);
3779
3780         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3781                 /* If vxlan->dev is in the same netns, it has already been added
3782                  * to the list by the previous loop.
3783                  */
3784                 if (!net_eq(dev_net(vxlan->dev), net)) {
3785                         gro_cells_destroy(&vxlan->gro_cells);
3786                         unregister_netdevice_queue(vxlan->dev, head);
3787                 }
3788         }
3789
3790         for (h = 0; h < PORT_HASH_SIZE; ++h)
3791                 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
3792 }
3793
3794 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
3795 {
3796         struct net *net;
3797         LIST_HEAD(list);
3798
3799         rtnl_lock();
3800         list_for_each_entry(net, net_list, exit_list)
3801                 vxlan_destroy_tunnels(net, &list);
3802
3803         unregister_netdevice_many(&list);
3804         rtnl_unlock();
3805 }
3806
3807 static struct pernet_operations vxlan_net_ops = {
3808         .init = vxlan_init_net,
3809         .exit_batch = vxlan_exit_batch_net,
3810         .id   = &vxlan_net_id,
3811         .size = sizeof(struct vxlan_net),
3812 };
3813
3814 static int __init vxlan_init_module(void)
3815 {
3816         int rc;
3817
3818         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3819
3820         rc = register_pernet_subsys(&vxlan_net_ops);
3821         if (rc)
3822                 goto out1;
3823
3824         rc = register_netdevice_notifier(&vxlan_notifier_block);
3825         if (rc)
3826                 goto out2;
3827
3828         rc = rtnl_link_register(&vxlan_link_ops);
3829         if (rc)
3830                 goto out3;
3831
3832         return 0;
3833 out3:
3834         unregister_netdevice_notifier(&vxlan_notifier_block);
3835 out2:
3836         unregister_pernet_subsys(&vxlan_net_ops);
3837 out1:
3838         return rc;
3839 }
3840 late_initcall(vxlan_init_module);
3841
3842 static void __exit vxlan_cleanup_module(void)
3843 {
3844         rtnl_link_unregister(&vxlan_link_ops);
3845         unregister_netdevice_notifier(&vxlan_notifier_block);
3846         unregister_pernet_subsys(&vxlan_net_ops);
3847         /* rcu_barrier() is called by netns */
3848 }
3849 module_exit(vxlan_cleanup_module);
3850
3851 MODULE_LICENSE("GPL");
3852 MODULE_VERSION(VXLAN_VERSION);
3853 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3854 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3855 MODULE_ALIAS_RTNL_LINK("vxlan");