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/vxlan.h>
  30
  31 #if IS_ENABLED(CONFIG_IPV6)
  32 #include <net/ip6_tunnel.h>
  33 #include <net/ip6_checksum.h>
  34 #endif
  35
  36 #define VXLAN_VERSION   "0.1"
  37
  38 #define PORT_HASH_BITS  8
  39 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  40 #define FDB_AGE_DEFAULT 300 /* 5 min */
  41 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  42
  43 /* UDP port for VXLAN traffic.
  44  * The IANA assigned port is 4789, but the Linux default is 8472
  45  * for compatibility with early adopters.
  46  */
  47 static unsigned short vxlan_port __read_mostly = 8472;
  48 module_param_named(udp_port, vxlan_port, ushort, 0444);
  49 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  50
  51 static bool log_ecn_error = true;
  52 module_param(log_ecn_error, bool, 0644);
  53 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  54
  55 static unsigned int vxlan_net_id;
  56 static struct rtnl_link_ops vxlan_link_ops;
  57
  58 static const u8 all_zeros_mac[ETH_ALEN + 2];
  59
  60 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  61
  62 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
  63
  64 /* per-network namespace private data for this module */
  65 struct vxlan_net {
  66         struct list_head  vxlan_list;
  67         struct hlist_head sock_list[PORT_HASH_SIZE];
  68         spinlock_t        sock_lock;
  69 };
  70
  71 /* Forwarding table entry */
  72 struct vxlan_fdb {
  73         struct hlist_node hlist;        /* linked list of entries */
  74         struct rcu_head   rcu;
  75         unsigned long     updated;      /* jiffies */
  76         unsigned long     used;
  77         struct list_head  remotes;
  78         u8                eth_addr[ETH_ALEN];
  79         u16               state;        /* see ndm_state */
  80         __be32            vni;
  81         u8                flags;        /* see ndm_flags */
  82 };
  83
  84 /* salt for hash table */
  85 static u32 vxlan_salt __read_mostly;
  86
  87 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  88 {
  89         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  90                ip_tunnel_collect_metadata();
  91 }
  92
  93 #if IS_ENABLED(CONFIG_IPV6)
  94 static inline
  95 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
  96 {
  97         if (a->sa.sa_family != b->sa.sa_family)
  98                 return false;
  99         if (a->sa.sa_family == AF_INET6)
 100                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 101         else
 102                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 103 }
 104
 105 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 106 {
 107         if (ipa->sa.sa_family == AF_INET6)
 108                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 109         else
 110                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 111 }
 112
 113 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 114 {
 115         if (ipa->sa.sa_family == AF_INET6)
 116                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 117         else
 118                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 119 }
 120
 121 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 122 {
 123         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 124                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 125                 ip->sa.sa_family = AF_INET6;
 126                 return 0;
 127         } else if (nla_len(nla) >= sizeof(__be32)) {
 128                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 129                 ip->sa.sa_family = AF_INET;
 130                 return 0;
 131         } else {
 132                 return -EAFNOSUPPORT;
 133         }
 134 }
 135
 136 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 137                               const union vxlan_addr *ip)
 138 {
 139         if (ip->sa.sa_family == AF_INET6)
 140                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 141         else
 142                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 143 }
 144
 145 #else /* !CONFIG_IPV6 */
 146
 147 static inline
 148 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 149 {
 150         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 151 }
 152
 153 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 154 {
 155         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 156 }
 157
 158 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 159 {
 160         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 161 }
 162
 163 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 164 {
 165         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 166                 return -EAFNOSUPPORT;
 167         } else if (nla_len(nla) >= sizeof(__be32)) {
 168                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 169                 ip->sa.sa_family = AF_INET;
 170                 return 0;
 171         } else {
 172                 return -EAFNOSUPPORT;
 173         }
 174 }
 175
 176 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 177                               const union vxlan_addr *ip)
 178 {
 179         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 180 }
 181 #endif
 182
 183 /* Virtual Network hash table head */
 184 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 185 {
 186         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 187 }
 188
 189 /* Socket hash table head */
 190 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 191 {
 192         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 193
 194         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 195 }
 196
 197 /* First remote destination for a forwarding entry.
 198  * Guaranteed to be non-NULL because remotes are never deleted.
 199  */
 200 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 201 {
 202         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 203 }
 204
 205 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 206 {
 207         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 208 }
 209
 210 /* Find VXLAN socket based on network namespace, address family and UDP port
 211  * and enabled unshareable flags.
 212  */
 213 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 214                                           __be16 port, u32 flags)
 215 {
 216         struct vxlan_sock *vs;
 217
 218         flags &= VXLAN_F_RCV_FLAGS;
 219
 220         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 221                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 222                     vxlan_get_sk_family(vs) == family &&
 223                     vs->flags == flags)
 224                         return vs;
 225         }
 226         return NULL;
 227 }
 228
 229 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, int ifindex,
 230                                            __be32 vni)
 231 {
 232         struct vxlan_dev_node *node;
 233
 234         /* For flow based devices, map all packets to VNI 0 */
 235         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 236                 vni = 0;
 237
 238         hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
 239                 if (node->vxlan->default_dst.remote_vni != vni)
 240                         continue;
 241
 242                 if (IS_ENABLED(CONFIG_IPV6)) {
 243                         const struct vxlan_config *cfg = &node->vxlan->cfg;
 244
 245                         if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) &&
 246                             cfg->remote_ifindex != ifindex)
 247                                 continue;
 248                 }
 249
 250                 return node->vxlan;
 251         }
 252
 253         return NULL;
 254 }
 255
 256 /* Look up VNI in a per net namespace table */
 257 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex,
 258                                         __be32 vni, sa_family_t family,
 259                                         __be16 port, u32 flags)
 260 {
 261         struct vxlan_sock *vs;
 262
 263         vs = vxlan_find_sock(net, family, port, flags);
 264         if (!vs)
 265                 return NULL;
 266
 267         return vxlan_vs_find_vni(vs, ifindex, vni);
 268 }
 269
 270 /* Fill in neighbour message in skbuff. */
 271 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 272                           const struct vxlan_fdb *fdb,
 273                           u32 portid, u32 seq, int type, unsigned int flags,
 274                           const struct vxlan_rdst *rdst)
 275 {
 276         unsigned long now = jiffies;
 277         struct nda_cacheinfo ci;
 278         struct nlmsghdr *nlh;
 279         struct ndmsg *ndm;
 280         bool send_ip, send_eth;
 281
 282         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 283         if (nlh == NULL)
 284                 return -EMSGSIZE;
 285
 286         ndm = nlmsg_data(nlh);
 287         memset(ndm, 0, sizeof(*ndm));
 288
 289         send_eth = send_ip = true;
 290
 291         if (type == RTM_GETNEIGH) {
 292                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 293                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 294                 ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
 295         } else
 296                 ndm->ndm_family = AF_BRIDGE;
 297         ndm->ndm_state = fdb->state;
 298         ndm->ndm_ifindex = vxlan->dev->ifindex;
 299         ndm->ndm_flags = fdb->flags;
 300         ndm->ndm_type = RTN_UNICAST;
 301
 302         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 303             nla_put_s32(skb, NDA_LINK_NETNSID,
 304                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 305                 goto nla_put_failure;
 306
 307         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 308                 goto nla_put_failure;
 309
 310         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 311                 goto nla_put_failure;
 312
 313         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 314             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 315                 goto nla_put_failure;
 316         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 317             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 318                 goto nla_put_failure;
 319         if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
 320             nla_put_u32(skb, NDA_SRC_VNI,
 321                         be32_to_cpu(fdb->vni)))
 322                 goto nla_put_failure;
 323         if (rdst->remote_ifindex &&
 324             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 325                 goto nla_put_failure;
 326
 327         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 328         ci.ndm_confirmed = 0;
 329         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 330         ci.ndm_refcnt    = 0;
 331
 332         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 333                 goto nla_put_failure;
 334
 335         nlmsg_end(skb, nlh);
 336         return 0;
 337
 338 nla_put_failure:
 339         nlmsg_cancel(skb, nlh);
 340         return -EMSGSIZE;
 341 }
 342
 343 static inline size_t vxlan_nlmsg_size(void)
 344 {
 345         return NLMSG_ALIGN(sizeof(struct ndmsg))
 346                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 347                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 348                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 349                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 350                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 351                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 352                 + nla_total_size(sizeof(struct nda_cacheinfo));
 353 }
 354
 355 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 356                              struct vxlan_rdst *rd, int type)
 357 {
 358         struct net *net = dev_net(vxlan->dev);
 359         struct sk_buff *skb;
 360         int err = -ENOBUFS;
 361
 362         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 363         if (skb == NULL)
 364                 goto errout;
 365
 366         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 367         if (err < 0) {
 368                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 369                 WARN_ON(err == -EMSGSIZE);
 370                 kfree_skb(skb);
 371                 goto errout;
 372         }
 373
 374         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 375         return;
 376 errout:
 377         if (err < 0)
 378                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 379 }
 380
 381 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 382 {
 383         struct vxlan_dev *vxlan = netdev_priv(dev);
 384         struct vxlan_fdb f = {
 385                 .state = NUD_STALE,
 386         };
 387         struct vxlan_rdst remote = {
 388                 .remote_ip = *ipa, /* goes to NDA_DST */
 389                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 390         };
 391
 392         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 393 }
 394
 395 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 396 {
 397         struct vxlan_fdb f = {
 398                 .state = NUD_STALE,
 399         };
 400         struct vxlan_rdst remote = { };
 401
 402         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 403
 404         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 405 }
 406
 407 /* Hash Ethernet address */
 408 static u32 eth_hash(const unsigned char *addr)
 409 {
 410         u64 value = get_unaligned((u64 *)addr);
 411
 412         /* only want 6 bytes */
 413 #ifdef __BIG_ENDIAN
 414         value >>= 16;
 415 #else
 416         value <<= 16;
 417 #endif
 418         return hash_64(value, FDB_HASH_BITS);
 419 }
 420
 421 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
 422 {
 423         /* use 1 byte of OUI and 3 bytes of NIC */
 424         u32 key = get_unaligned((u32 *)(addr + 2));
 425
 426         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
 427 }
 428
 429 /* Hash chain to use given mac address */
 430 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 431                                                 const u8 *mac, __be32 vni)
 432 {
 433         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
 434                 return &vxlan->fdb_head[eth_vni_hash(mac, vni)];
 435         else
 436                 return &vxlan->fdb_head[eth_hash(mac)];
 437 }
 438
 439 /* Look up Ethernet address in forwarding table */
 440 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 441                                           const u8 *mac, __be32 vni)
 442 {
 443         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 444         struct vxlan_fdb *f;
 445
 446         hlist_for_each_entry_rcu(f, head, hlist) {
 447                 if (ether_addr_equal(mac, f->eth_addr)) {
 448                         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
 449                                 if (vni == f->vni)
 450                                         return f;
 451                         } else {
 452                                 return f;
 453                         }
 454                 }
 455         }
 456
 457         return NULL;
 458 }
 459
 460 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 461                                         const u8 *mac, __be32 vni)
 462 {
 463         struct vxlan_fdb *f;
 464
 465         f = __vxlan_find_mac(vxlan, mac, vni);
 466         if (f)
 467                 f->used = jiffies;
 468
 469         return f;
 470 }
 471
 472 /* caller should hold vxlan->hash_lock */
 473 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 474                                               union vxlan_addr *ip, __be16 port,
 475                                               __be32 vni, __u32 ifindex)
 476 {
 477         struct vxlan_rdst *rd;
 478
 479         list_for_each_entry(rd, &f->remotes, list) {
 480                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 481                     rd->remote_port == port &&
 482                     rd->remote_vni == vni &&
 483                     rd->remote_ifindex == ifindex)
 484                         return rd;
 485         }
 486
 487         return NULL;
 488 }
 489
 490 /* Replace destination of unicast mac */
 491 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 492                              union vxlan_addr *ip, __be16 port, __be32 vni,
 493                              __u32 ifindex)
 494 {
 495         struct vxlan_rdst *rd;
 496
 497         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 498         if (rd)
 499                 return 0;
 500
 501         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 502         if (!rd)
 503                 return 0;
 504
 505         dst_cache_reset(&rd->dst_cache);
 506         rd->remote_ip = *ip;
 507         rd->remote_port = port;
 508         rd->remote_vni = vni;
 509         rd->remote_ifindex = ifindex;
 510         return 1;
 511 }
 512
 513 /* Add/update destinations for multicast */
 514 static int vxlan_fdb_append(struct vxlan_fdb *f,
 515                             union vxlan_addr *ip, __be16 port, __be32 vni,
 516                             __u32 ifindex, struct vxlan_rdst **rdp)
 517 {
 518         struct vxlan_rdst *rd;
 519
 520         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 521         if (rd)
 522                 return 0;
 523
 524         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 525         if (rd == NULL)
 526                 return -ENOBUFS;
 527
 528         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 529                 kfree(rd);
 530                 return -ENOBUFS;
 531         }
 532
 533         rd->remote_ip = *ip;
 534         rd->remote_port = port;
 535         rd->remote_vni = vni;
 536         rd->remote_ifindex = ifindex;
 537
 538         list_add_tail_rcu(&rd->list, &f->remotes);
 539
 540         *rdp = rd;
 541         return 1;
 542 }
 543
 544 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 545                                           unsigned int off,
 546                                           struct vxlanhdr *vh, size_t hdrlen,
 547                                           __be32 vni_field,
 548                                           struct gro_remcsum *grc,
 549                                           bool nopartial)
 550 {
 551         size_t start, offset;
 552
 553         if (skb->remcsum_offload)
 554                 return vh;
 555
 556         if (!NAPI_GRO_CB(skb)->csum_valid)
 557                 return NULL;
 558
 559         start = vxlan_rco_start(vni_field);
 560         offset = start + vxlan_rco_offset(vni_field);
 561
 562         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 563                                      start, offset, grc, nopartial);
 564
 565         skb->remcsum_offload = 1;
 566
 567         return vh;
 568 }
 569
 570 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
 571                                           struct sk_buff **head,
 572                                           struct sk_buff *skb)
 573 {
 574         struct sk_buff *p, **pp = NULL;
 575         struct vxlanhdr *vh, *vh2;
 576         unsigned int hlen, off_vx;
 577         int flush = 1;
 578         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 579         __be32 flags;
 580         struct gro_remcsum grc;
 581
 582         skb_gro_remcsum_init(&grc);
 583
 584         off_vx = skb_gro_offset(skb);
 585         hlen = off_vx + sizeof(*vh);
 586         vh   = skb_gro_header_fast(skb, off_vx);
 587         if (skb_gro_header_hard(skb, hlen)) {
 588                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 589                 if (unlikely(!vh))
 590                         goto out;
 591         }
 592
 593         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 594
 595         flags = vh->vx_flags;
 596
 597         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 598                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 599                                        vh->vx_vni, &grc,
 600                                        !!(vs->flags &
 601                                           VXLAN_F_REMCSUM_NOPARTIAL));
 602
 603                 if (!vh)
 604                         goto out;
 605         }
 606
 607         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 608
 609         for (p = *head; p; p = p->next) {
 610                 if (!NAPI_GRO_CB(p)->same_flow)
 611                         continue;
 612
 613                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 614                 if (vh->vx_flags != vh2->vx_flags ||
 615                     vh->vx_vni != vh2->vx_vni) {
 616                         NAPI_GRO_CB(p)->same_flow = 0;
 617                         continue;
 618                 }
 619         }
 620
 621         pp = call_gro_receive(eth_gro_receive, head, skb);
 622         flush = 0;
 623
 624 out:
 625         skb_gro_remcsum_cleanup(skb, &grc);
 626         NAPI_GRO_CB(skb)->flush |= flush;
 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 /* Add new entry to forwarding table -- assumes lock held */
 640 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 641                             const u8 *mac, union vxlan_addr *ip,
 642                             __u16 state, __u16 flags,
 643                             __be16 port, __be32 src_vni, __be32 vni,
 644                             __u32 ifindex, __u8 ndm_flags)
 645 {
 646         struct vxlan_rdst *rd = NULL;
 647         struct vxlan_fdb *f;
 648         int notify = 0;
 649         int rc;
 650
 651         f = __vxlan_find_mac(vxlan, mac, src_vni);
 652         if (f) {
 653                 if (flags & NLM_F_EXCL) {
 654                         netdev_dbg(vxlan->dev,
 655                                    "lost race to create %pM\n", mac);
 656                         return -EEXIST;
 657                 }
 658                 if (f->state != state) {
 659                         f->state = state;
 660                         f->updated = jiffies;
 661                         notify = 1;
 662                 }
 663                 if (f->flags != ndm_flags) {
 664                         f->flags = ndm_flags;
 665                         f->updated = jiffies;
 666                         notify = 1;
 667                 }
 668                 if ((flags & NLM_F_REPLACE)) {
 669                         /* Only change unicasts */
 670                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 671                              is_zero_ether_addr(f->eth_addr))) {
 672                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 673                                                            ifindex);
 674                         } else
 675                                 return -EOPNOTSUPP;
 676                 }
 677                 if ((flags & NLM_F_APPEND) &&
 678                     (is_multicast_ether_addr(f->eth_addr) ||
 679                      is_zero_ether_addr(f->eth_addr))) {
 680                         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 681
 682                         if (rc < 0)
 683                                 return rc;
 684                         notify |= rc;
 685                 }
 686         } else {
 687                 if (!(flags & NLM_F_CREATE))
 688                         return -ENOENT;
 689
 690                 if (vxlan->cfg.addrmax &&
 691                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 692                         return -ENOSPC;
 693
 694                 /* Disallow replace to add a multicast entry */
 695                 if ((flags & NLM_F_REPLACE) &&
 696                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 697                         return -EOPNOTSUPP;
 698
 699                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 700                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 701                 if (!f)
 702                         return -ENOMEM;
 703
 704                 notify = 1;
 705                 f->state = state;
 706                 f->flags = ndm_flags;
 707                 f->updated = f->used = jiffies;
 708                 f->vni = src_vni;
 709                 INIT_LIST_HEAD(&f->remotes);
 710                 memcpy(f->eth_addr, mac, ETH_ALEN);
 711
 712                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 713                 if (rc < 0) {
 714                         kfree(f);
 715                         return rc;
 716                 }
 717
 718                 ++vxlan->addrcnt;
 719                 hlist_add_head_rcu(&f->hlist,
 720                                    vxlan_fdb_head(vxlan, mac, src_vni));
 721         }
 722
 723         if (notify) {
 724                 if (rd == NULL)
 725                         rd = first_remote_rtnl(f);
 726                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 727         }
 728
 729         return 0;
 730 }
 731
 732 static void vxlan_fdb_free(struct rcu_head *head)
 733 {
 734         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 735         struct vxlan_rdst *rd, *nd;
 736
 737         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 738                 dst_cache_destroy(&rd->dst_cache);
 739                 kfree(rd);
 740         }
 741         kfree(f);
 742 }
 743
 744 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 745 {
 746         netdev_dbg(vxlan->dev,
 747                     "delete %pM\n", f->eth_addr);
 748
 749         --vxlan->addrcnt;
 750         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 751
 752         hlist_del_rcu(&f->hlist);
 753         call_rcu(&f->rcu, vxlan_fdb_free);
 754 }
 755
 756 static void vxlan_dst_free(struct rcu_head *head)
 757 {
 758         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 759
 760         dst_cache_destroy(&rd->dst_cache);
 761         kfree(rd);
 762 }
 763
 764 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 765                                   struct vxlan_rdst *rd)
 766 {
 767         list_del_rcu(&rd->list);
 768         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 769         call_rcu(&rd->rcu, vxlan_dst_free);
 770 }
 771
 772 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 773                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
 774                            __be32 *vni, u32 *ifindex)
 775 {
 776         struct net *net = dev_net(vxlan->dev);
 777         int err;
 778
 779         if (tb[NDA_DST]) {
 780                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 781                 if (err)
 782                         return err;
 783         } else {
 784                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 785                 if (remote->sa.sa_family == AF_INET) {
 786                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 787                         ip->sa.sa_family = AF_INET;
 788 #if IS_ENABLED(CONFIG_IPV6)
 789                 } else {
 790                         ip->sin6.sin6_addr = in6addr_any;
 791                         ip->sa.sa_family = AF_INET6;
 792 #endif
 793                 }
 794         }
 795
 796         if (tb[NDA_PORT]) {
 797                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 798                         return -EINVAL;
 799                 *port = nla_get_be16(tb[NDA_PORT]);
 800         } else {
 801                 *port = vxlan->cfg.dst_port;
 802         }
 803
 804         if (tb[NDA_VNI]) {
 805                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 806                         return -EINVAL;
 807                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 808         } else {
 809                 *vni = vxlan->default_dst.remote_vni;
 810         }
 811
 812         if (tb[NDA_SRC_VNI]) {
 813                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
 814                         return -EINVAL;
 815                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
 816         } else {
 817                 *src_vni = vxlan->default_dst.remote_vni;
 818         }
 819
 820         if (tb[NDA_IFINDEX]) {
 821                 struct net_device *tdev;
 822
 823                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 824                         return -EINVAL;
 825                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 826                 tdev = __dev_get_by_index(net, *ifindex);
 827                 if (!tdev)
 828                         return -EADDRNOTAVAIL;
 829         } else {
 830                 *ifindex = 0;
 831         }
 832
 833         return 0;
 834 }
 835
 836 /* Add static entry (via netlink) */
 837 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 838                          struct net_device *dev,
 839                          const unsigned char *addr, u16 vid, u16 flags)
 840 {
 841         struct vxlan_dev *vxlan = netdev_priv(dev);
 842         /* struct net *net = dev_net(vxlan->dev); */
 843         union vxlan_addr ip;
 844         __be16 port;
 845         __be32 src_vni, vni;
 846         u32 ifindex;
 847         int err;
 848
 849         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 850                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 851                         ndm->ndm_state);
 852                 return -EINVAL;
 853         }
 854
 855         if (tb[NDA_DST] == NULL)
 856                 return -EINVAL;
 857
 858         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 859         if (err)
 860                 return err;
 861
 862         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 863                 return -EAFNOSUPPORT;
 864
 865         spin_lock_bh(&vxlan->hash_lock);
 866         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 867                                port, src_vni, vni, ifindex, ndm->ndm_flags);
 868         spin_unlock_bh(&vxlan->hash_lock);
 869
 870         return err;
 871 }
 872
 873 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
 874                               const unsigned char *addr, union vxlan_addr ip,
 875                               __be16 port, __be32 src_vni, u32 vni, u32 ifindex,
 876                               u16 vid)
 877 {
 878         struct vxlan_fdb *f;
 879         struct vxlan_rdst *rd = NULL;
 880         int err = -ENOENT;
 881
 882         f = vxlan_find_mac(vxlan, addr, src_vni);
 883         if (!f)
 884                 return err;
 885
 886         if (!vxlan_addr_any(&ip)) {
 887                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 888                 if (!rd)
 889                         goto out;
 890         }
 891
 892         /* remove a destination if it's not the only one on the list,
 893          * otherwise destroy the fdb entry
 894          */
 895         if (rd && !list_is_singular(&f->remotes)) {
 896                 vxlan_fdb_dst_destroy(vxlan, f, rd);
 897                 goto out;
 898         }
 899
 900         vxlan_fdb_destroy(vxlan, f);
 901
 902 out:
 903         return 0;
 904 }
 905
 906 /* Delete entry (via netlink) */
 907 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 908                             struct net_device *dev,
 909                             const unsigned char *addr, u16 vid)
 910 {
 911         struct vxlan_dev *vxlan = netdev_priv(dev);
 912         union vxlan_addr ip;
 913         __be32 src_vni, vni;
 914         __be16 port;
 915         u32 ifindex;
 916         int err;
 917
 918         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 919         if (err)
 920                 return err;
 921
 922         spin_lock_bh(&vxlan->hash_lock);
 923         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
 924                                  vid);
 925         spin_unlock_bh(&vxlan->hash_lock);
 926
 927         return err;
 928 }
 929
 930 /* Dump forwarding table */
 931 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 932                           struct net_device *dev,
 933                           struct net_device *filter_dev, int *idx)
 934 {
 935         struct vxlan_dev *vxlan = netdev_priv(dev);
 936         unsigned int h;
 937         int err = 0;
 938
 939         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 940                 struct vxlan_fdb *f;
 941
 942                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 943                         struct vxlan_rdst *rd;
 944
 945                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 946                                 if (*idx < cb->args[2])
 947                                         goto skip;
 948
 949                                 err = vxlan_fdb_info(skb, vxlan, f,
 950                                                      NETLINK_CB(cb->skb).portid,
 951                                                      cb->nlh->nlmsg_seq,
 952                                                      RTM_NEWNEIGH,
 953                                                      NLM_F_MULTI, rd);
 954                                 if (err < 0)
 955                                         goto out;
 956 skip:
 957                                 *idx += 1;
 958                         }
 959                 }
 960         }
 961 out:
 962         return err;
 963 }
 964
 965 /* Watch incoming packets to learn mapping between Ethernet address
 966  * and Tunnel endpoint.
 967  * Return true if packet is bogus and should be dropped.
 968  */
 969 static bool vxlan_snoop(struct net_device *dev,
 970                         union vxlan_addr *src_ip, const u8 *src_mac,
 971                         u32 src_ifindex, __be32 vni)
 972 {
 973         struct vxlan_dev *vxlan = netdev_priv(dev);
 974         struct vxlan_fdb *f;
 975         u32 ifindex = 0;
 976
 977 #if IS_ENABLED(CONFIG_IPV6)
 978         if (src_ip->sa.sa_family == AF_INET6 &&
 979             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
 980                 ifindex = src_ifindex;
 981 #endif
 982
 983         f = vxlan_find_mac(vxlan, src_mac, vni);
 984         if (likely(f)) {
 985                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 986
 987                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
 988                            rdst->remote_ifindex == ifindex))
 989                         return false;
 990
 991                 /* Don't migrate static entries, drop packets */
 992                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
 993                         return true;
 994
 995                 if (net_ratelimit())
 996                         netdev_info(dev,
 997                                     "%pM migrated from %pIS to %pIS\n",
 998                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 999
1000                 rdst->remote_ip = *src_ip;
1001                 f->updated = jiffies;
1002                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1003         } else {
1004                 /* learned new entry */
1005                 spin_lock(&vxlan->hash_lock);
1006
1007                 /* close off race between vxlan_flush and incoming packets */
1008                 if (netif_running(dev))
1009                         vxlan_fdb_create(vxlan, src_mac, src_ip,
1010                                          NUD_REACHABLE,
1011                                          NLM_F_EXCL|NLM_F_CREATE,
1012                                          vxlan->cfg.dst_port,
1013                                          vni,
1014                                          vxlan->default_dst.remote_vni,
1015                                          ifindex, NTF_SELF);
1016                 spin_unlock(&vxlan->hash_lock);
1017         }
1018
1019         return false;
1020 }
1021
1022 /* See if multicast group is already in use by other ID */
1023 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1024 {
1025         struct vxlan_dev *vxlan;
1026         struct vxlan_sock *sock4;
1027 #if IS_ENABLED(CONFIG_IPV6)
1028         struct vxlan_sock *sock6;
1029 #endif
1030         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1031
1032         sock4 = rtnl_dereference(dev->vn4_sock);
1033
1034         /* The vxlan_sock is only used by dev, leaving group has
1035          * no effect on other vxlan devices.
1036          */
1037         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1038                 return false;
1039 #if IS_ENABLED(CONFIG_IPV6)
1040         sock6 = rtnl_dereference(dev->vn6_sock);
1041         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1042                 return false;
1043 #endif
1044
1045         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1046                 if (!netif_running(vxlan->dev) || vxlan == dev)
1047                         continue;
1048
1049                 if (family == AF_INET &&
1050                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1051                         continue;
1052 #if IS_ENABLED(CONFIG_IPV6)
1053                 if (family == AF_INET6 &&
1054                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1055                         continue;
1056 #endif
1057
1058                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1059                                       &dev->default_dst.remote_ip))
1060                         continue;
1061
1062                 if (vxlan->default_dst.remote_ifindex !=
1063                     dev->default_dst.remote_ifindex)
1064                         continue;
1065
1066                 return true;
1067         }
1068
1069         return false;
1070 }
1071
1072 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1073 {
1074         struct vxlan_net *vn;
1075
1076         if (!vs)
1077                 return false;
1078         if (!refcount_dec_and_test(&vs->refcnt))
1079                 return false;
1080
1081         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1082         spin_lock(&vn->sock_lock);
1083         hlist_del_rcu(&vs->hlist);
1084         udp_tunnel_notify_del_rx_port(vs->sock,
1085                                       (vs->flags & VXLAN_F_GPE) ?
1086                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1087                                       UDP_TUNNEL_TYPE_VXLAN);
1088         spin_unlock(&vn->sock_lock);
1089
1090         return true;
1091 }
1092
1093 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1094 {
1095         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1096 #if IS_ENABLED(CONFIG_IPV6)
1097         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1098
1099         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1100 #endif
1101
1102         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1103         synchronize_net();
1104
1105         vxlan_vs_del_dev(vxlan);
1106
1107         if (__vxlan_sock_release_prep(sock4)) {
1108                 udp_tunnel_sock_release(sock4->sock);
1109                 kfree(sock4);
1110         }
1111
1112 #if IS_ENABLED(CONFIG_IPV6)
1113         if (__vxlan_sock_release_prep(sock6)) {
1114                 udp_tunnel_sock_release(sock6->sock);
1115                 kfree(sock6);
1116         }
1117 #endif
1118 }
1119
1120 /* Update multicast group membership when first VNI on
1121  * multicast address is brought up
1122  */
1123 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1124 {
1125         struct sock *sk;
1126         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1127         int ifindex = vxlan->default_dst.remote_ifindex;
1128         int ret = -EINVAL;
1129
1130         if (ip->sa.sa_family == AF_INET) {
1131                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1132                 struct ip_mreqn mreq = {
1133                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1134                         .imr_ifindex            = ifindex,
1135                 };
1136
1137                 sk = sock4->sock->sk;
1138                 lock_sock(sk);
1139                 ret = ip_mc_join_group(sk, &mreq);
1140                 release_sock(sk);
1141 #if IS_ENABLED(CONFIG_IPV6)
1142         } else {
1143                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1144
1145                 sk = sock6->sock->sk;
1146                 lock_sock(sk);
1147                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1148                                                    &ip->sin6.sin6_addr);
1149                 release_sock(sk);
1150 #endif
1151         }
1152
1153         return ret;
1154 }
1155
1156 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1157 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1158 {
1159         struct sock *sk;
1160         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1161         int ifindex = vxlan->default_dst.remote_ifindex;
1162         int ret = -EINVAL;
1163
1164         if (ip->sa.sa_family == AF_INET) {
1165                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1166                 struct ip_mreqn mreq = {
1167                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1168                         .imr_ifindex            = ifindex,
1169                 };
1170
1171                 sk = sock4->sock->sk;
1172                 lock_sock(sk);
1173                 ret = ip_mc_leave_group(sk, &mreq);
1174                 release_sock(sk);
1175 #if IS_ENABLED(CONFIG_IPV6)
1176         } else {
1177                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1178
1179                 sk = sock6->sock->sk;
1180                 lock_sock(sk);
1181                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1182                                                    &ip->sin6.sin6_addr);
1183                 release_sock(sk);
1184 #endif
1185         }
1186
1187         return ret;
1188 }
1189
1190 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1191                           struct sk_buff *skb, u32 vxflags)
1192 {
1193         size_t start, offset;
1194
1195         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1196                 goto out;
1197
1198         start = vxlan_rco_start(unparsed->vx_vni);
1199         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1200
1201         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1202                 return false;
1203
1204         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1205                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1206 out:
1207         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1208         unparsed->vx_vni &= VXLAN_VNI_MASK;
1209         return true;
1210 }
1211
1212 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1213                                 struct sk_buff *skb, u32 vxflags,
1214                                 struct vxlan_metadata *md)
1215 {
1216         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1217         struct metadata_dst *tun_dst;
1218
1219         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1220                 goto out;
1221
1222         md->gbp = ntohs(gbp->policy_id);
1223
1224         tun_dst = (struct metadata_dst *)skb_dst(skb);
1225         if (tun_dst) {
1226                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1227                 tun_dst->u.tun_info.options_len = sizeof(*md);
1228         }
1229         if (gbp->dont_learn)
1230                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1231
1232         if (gbp->policy_applied)
1233                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1234
1235         /* In flow-based mode, GBP is carried in dst_metadata */
1236         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1237                 skb->mark = md->gbp;
1238 out:
1239         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1240 }
1241
1242 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1243                                 __be16 *protocol,
1244                                 struct sk_buff *skb, u32 vxflags)
1245 {
1246         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1247
1248         /* Need to have Next Protocol set for interfaces in GPE mode. */
1249         if (!gpe->np_applied)
1250                 return false;
1251         /* "The initial version is 0. If a receiver does not support the
1252          * version indicated it MUST drop the packet.
1253          */
1254         if (gpe->version != 0)
1255                 return false;
1256         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1257          * processing MUST occur." However, we don't implement OAM
1258          * processing, thus drop the packet.
1259          */
1260         if (gpe->oam_flag)
1261                 return false;
1262
1263         switch (gpe->next_protocol) {
1264         case VXLAN_GPE_NP_IPV4:
1265                 *protocol = htons(ETH_P_IP);
1266                 break;
1267         case VXLAN_GPE_NP_IPV6:
1268                 *protocol = htons(ETH_P_IPV6);
1269                 break;
1270         case VXLAN_GPE_NP_ETHERNET:
1271                 *protocol = htons(ETH_P_TEB);
1272                 break;
1273         default:
1274                 return false;
1275         }
1276
1277         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1278         return true;
1279 }
1280
1281 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1282                           struct vxlan_sock *vs,
1283                           struct sk_buff *skb, __be32 vni)
1284 {
1285         union vxlan_addr saddr;
1286         u32 ifindex = skb->dev->ifindex;
1287
1288         skb_reset_mac_header(skb);
1289         skb->protocol = eth_type_trans(skb, vxlan->dev);
1290         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1291
1292         /* Ignore packet loops (and multicast echo) */
1293         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1294                 return false;
1295
1296         /* Get address from the outer IP header */
1297         if (vxlan_get_sk_family(vs) == AF_INET) {
1298                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1299                 saddr.sa.sa_family = AF_INET;
1300 #if IS_ENABLED(CONFIG_IPV6)
1301         } else {
1302                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1303                 saddr.sa.sa_family = AF_INET6;
1304 #endif
1305         }
1306
1307         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1308             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1309                 return false;
1310
1311         return true;
1312 }
1313
1314 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1315                                   struct sk_buff *skb)
1316 {
1317         int err = 0;
1318
1319         if (vxlan_get_sk_family(vs) == AF_INET)
1320                 err = IP_ECN_decapsulate(oiph, skb);
1321 #if IS_ENABLED(CONFIG_IPV6)
1322         else
1323                 err = IP6_ECN_decapsulate(oiph, skb);
1324 #endif
1325
1326         if (unlikely(err) && log_ecn_error) {
1327                 if (vxlan_get_sk_family(vs) == AF_INET)
1328                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1329                                              &((struct iphdr *)oiph)->saddr,
1330                                              ((struct iphdr *)oiph)->tos);
1331                 else
1332                         net_info_ratelimited("non-ECT from %pI6\n",
1333                                              &((struct ipv6hdr *)oiph)->saddr);
1334         }
1335         return err <= 1;
1336 }
1337
1338 /* Callback from net/ipv4/udp.c to receive packets */
1339 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1340 {
1341         struct pcpu_sw_netstats *stats;
1342         struct vxlan_dev *vxlan;
1343         struct vxlan_sock *vs;
1344         struct vxlanhdr unparsed;
1345         struct vxlan_metadata _md;
1346         struct vxlan_metadata *md = &_md;
1347         __be16 protocol = htons(ETH_P_TEB);
1348         bool raw_proto = false;
1349         void *oiph;
1350         __be32 vni = 0;
1351
1352         /* Need UDP and VXLAN header to be present */
1353         if (!pskb_may_pull(skb, VXLAN_HLEN))
1354                 goto drop;
1355
1356         unparsed = *vxlan_hdr(skb);
1357         /* VNI flag always required to be set */
1358         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1359                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1360                            ntohl(vxlan_hdr(skb)->vx_flags),
1361                            ntohl(vxlan_hdr(skb)->vx_vni));
1362                 /* Return non vxlan pkt */
1363                 goto drop;
1364         }
1365         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1366         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1367
1368         vs = rcu_dereference_sk_user_data(sk);
1369         if (!vs)
1370                 goto drop;
1371
1372         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1373
1374         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1375         if (!vxlan)
1376                 goto drop;
1377
1378         /* For backwards compatibility, only allow reserved fields to be
1379          * used by VXLAN extensions if explicitly requested.
1380          */
1381         if (vs->flags & VXLAN_F_GPE) {
1382                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1383                         goto drop;
1384                 raw_proto = true;
1385         }
1386
1387         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1388                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1389                         goto drop;
1390
1391         if (vxlan_collect_metadata(vs)) {
1392                 struct metadata_dst *tun_dst;
1393
1394                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1395                                          key32_to_tunnel_id(vni), sizeof(*md));
1396
1397                 if (!tun_dst)
1398                         goto drop;
1399
1400                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1401
1402                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1403         } else {
1404                 memset(md, 0, sizeof(*md));
1405         }
1406
1407         if (vs->flags & VXLAN_F_REMCSUM_RX)
1408                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1409                         goto drop;
1410         if (vs->flags & VXLAN_F_GBP)
1411                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1412         /* Note that GBP and GPE can never be active together. This is
1413          * ensured in vxlan_dev_configure.
1414          */
1415
1416         if (unparsed.vx_flags || unparsed.vx_vni) {
1417                 /* If there are any unprocessed flags remaining treat
1418                  * this as a malformed packet. This behavior diverges from
1419                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1420                  * in reserved fields are to be ignored. The approach here
1421                  * maintains compatibility with previous stack code, and also
1422                  * is more robust and provides a little more security in
1423                  * adding extensions to VXLAN.
1424                  */
1425                 goto drop;
1426         }
1427
1428         if (!raw_proto) {
1429                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1430                         goto drop;
1431         } else {
1432                 skb_reset_mac_header(skb);
1433                 skb->dev = vxlan->dev;
1434                 skb->pkt_type = PACKET_HOST;
1435         }
1436
1437         oiph = skb_network_header(skb);
1438         skb_reset_network_header(skb);
1439
1440         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1441                 ++vxlan->dev->stats.rx_frame_errors;
1442                 ++vxlan->dev->stats.rx_errors;
1443                 goto drop;
1444         }
1445
1446         stats = this_cpu_ptr(vxlan->dev->tstats);
1447         u64_stats_update_begin(&stats->syncp);
1448         stats->rx_packets++;
1449         stats->rx_bytes += skb->len;
1450         u64_stats_update_end(&stats->syncp);
1451
1452         gro_cells_receive(&vxlan->gro_cells, skb);
1453         return 0;
1454
1455 drop:
1456         /* Consume bad packet */
1457         kfree_skb(skb);
1458         return 0;
1459 }
1460
1461 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1462 {
1463         struct vxlan_dev *vxlan = netdev_priv(dev);
1464         struct arphdr *parp;
1465         u8 *arpptr, *sha;
1466         __be32 sip, tip;
1467         struct neighbour *n;
1468
1469         if (dev->flags & IFF_NOARP)
1470                 goto out;
1471
1472         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1473                 dev->stats.tx_dropped++;
1474                 goto out;
1475         }
1476         parp = arp_hdr(skb);
1477
1478         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1479              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1480             parp->ar_pro != htons(ETH_P_IP) ||
1481             parp->ar_op != htons(ARPOP_REQUEST) ||
1482             parp->ar_hln != dev->addr_len ||
1483             parp->ar_pln != 4)
1484                 goto out;
1485         arpptr = (u8 *)parp + sizeof(struct arphdr);
1486         sha = arpptr;
1487         arpptr += dev->addr_len;        /* sha */
1488         memcpy(&sip, arpptr, sizeof(sip));
1489         arpptr += sizeof(sip);
1490         arpptr += dev->addr_len;        /* tha */
1491         memcpy(&tip, arpptr, sizeof(tip));
1492
1493         if (ipv4_is_loopback(tip) ||
1494             ipv4_is_multicast(tip))
1495                 goto out;
1496
1497         n = neigh_lookup(&arp_tbl, &tip, dev);
1498
1499         if (n) {
1500                 struct vxlan_fdb *f;
1501                 struct sk_buff  *reply;
1502
1503                 if (!(n->nud_state & NUD_CONNECTED)) {
1504                         neigh_release(n);
1505                         goto out;
1506                 }
1507
1508                 f = vxlan_find_mac(vxlan, n->ha, vni);
1509                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1510                         /* bridge-local neighbor */
1511                         neigh_release(n);
1512                         goto out;
1513                 }
1514
1515                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1516                                 n->ha, sha);
1517
1518                 neigh_release(n);
1519
1520                 if (reply == NULL)
1521                         goto out;
1522
1523                 skb_reset_mac_header(reply);
1524                 __skb_pull(reply, skb_network_offset(reply));
1525                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1526                 reply->pkt_type = PACKET_HOST;
1527
1528                 if (netif_rx_ni(reply) == NET_RX_DROP)
1529                         dev->stats.rx_dropped++;
1530         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1531                 union vxlan_addr ipa = {
1532                         .sin.sin_addr.s_addr = tip,
1533                         .sin.sin_family = AF_INET,
1534                 };
1535
1536                 vxlan_ip_miss(dev, &ipa);
1537         }
1538 out:
1539         consume_skb(skb);
1540         return NETDEV_TX_OK;
1541 }
1542
1543 #if IS_ENABLED(CONFIG_IPV6)
1544 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1545         struct neighbour *n, bool isrouter)
1546 {
1547         struct net_device *dev = request->dev;
1548         struct sk_buff *reply;
1549         struct nd_msg *ns, *na;
1550         struct ipv6hdr *pip6;
1551         u8 *daddr;
1552         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1553         int ns_olen;
1554         int i, len;
1555
1556         if (dev == NULL || !pskb_may_pull(request, request->len))
1557                 return NULL;
1558
1559         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1560                 sizeof(*na) + na_olen + dev->needed_tailroom;
1561         reply = alloc_skb(len, GFP_ATOMIC);
1562         if (reply == NULL)
1563                 return NULL;
1564
1565         reply->protocol = htons(ETH_P_IPV6);
1566         reply->dev = dev;
1567         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1568         skb_push(reply, sizeof(struct ethhdr));
1569         skb_reset_mac_header(reply);
1570
1571         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1572
1573         daddr = eth_hdr(request)->h_source;
1574         ns_olen = request->len - skb_network_offset(request) -
1575                 sizeof(struct ipv6hdr) - sizeof(*ns);
1576         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1577                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1578                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1579                         break;
1580                 }
1581         }
1582
1583         /* Ethernet header */
1584         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1585         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1586         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1587         reply->protocol = htons(ETH_P_IPV6);
1588
1589         skb_pull(reply, sizeof(struct ethhdr));
1590         skb_reset_network_header(reply);
1591         skb_put(reply, sizeof(struct ipv6hdr));
1592
1593         /* IPv6 header */
1594
1595         pip6 = ipv6_hdr(reply);
1596         memset(pip6, 0, sizeof(struct ipv6hdr));
1597         pip6->version = 6;
1598         pip6->priority = ipv6_hdr(request)->priority;
1599         pip6->nexthdr = IPPROTO_ICMPV6;
1600         pip6->hop_limit = 255;
1601         pip6->daddr = ipv6_hdr(request)->saddr;
1602         pip6->saddr = *(struct in6_addr *)n->primary_key;
1603
1604         skb_pull(reply, sizeof(struct ipv6hdr));
1605         skb_reset_transport_header(reply);
1606
1607         /* Neighbor Advertisement */
1608         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1609         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1610         na->icmph.icmp6_router = isrouter;
1611         na->icmph.icmp6_override = 1;
1612         na->icmph.icmp6_solicited = 1;
1613         na->target = ns->target;
1614         ether_addr_copy(&na->opt[2], n->ha);
1615         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1616         na->opt[1] = na_olen >> 3;
1617
1618         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1619                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1620                 csum_partial(na, sizeof(*na)+na_olen, 0));
1621
1622         pip6->payload_len = htons(sizeof(*na)+na_olen);
1623
1624         skb_push(reply, sizeof(struct ipv6hdr));
1625
1626         reply->ip_summed = CHECKSUM_UNNECESSARY;
1627
1628         return reply;
1629 }
1630
1631 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1632 {
1633         struct vxlan_dev *vxlan = netdev_priv(dev);
1634         struct nd_msg *msg;
1635         const struct ipv6hdr *iphdr;
1636         const struct in6_addr *daddr;
1637         struct neighbour *n;
1638         struct inet6_dev *in6_dev;
1639
1640         in6_dev = __in6_dev_get(dev);
1641         if (!in6_dev)
1642                 goto out;
1643
1644         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr) + sizeof(struct nd_msg)))
1645                 goto out;
1646
1647         iphdr = ipv6_hdr(skb);
1648         daddr = &iphdr->daddr;
1649
1650         msg = (struct nd_msg *)(iphdr + 1);
1651         if (msg->icmph.icmp6_code != 0 ||
1652             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1653                 goto out;
1654
1655         if (ipv6_addr_loopback(daddr) ||
1656             ipv6_addr_is_multicast(&msg->target))
1657                 goto out;
1658
1659         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1660
1661         if (n) {
1662                 struct vxlan_fdb *f;
1663                 struct sk_buff *reply;
1664
1665                 if (!(n->nud_state & NUD_CONNECTED)) {
1666                         neigh_release(n);
1667                         goto out;
1668                 }
1669
1670                 f = vxlan_find_mac(vxlan, n->ha, vni);
1671                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1672                         /* bridge-local neighbor */
1673                         neigh_release(n);
1674                         goto out;
1675                 }
1676
1677                 reply = vxlan_na_create(skb, n,
1678                                         !!(f ? f->flags & NTF_ROUTER : 0));
1679
1680                 neigh_release(n);
1681
1682                 if (reply == NULL)
1683                         goto out;
1684
1685                 if (netif_rx_ni(reply) == NET_RX_DROP)
1686                         dev->stats.rx_dropped++;
1687
1688         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1689                 union vxlan_addr ipa = {
1690                         .sin6.sin6_addr = msg->target,
1691                         .sin6.sin6_family = AF_INET6,
1692                 };
1693
1694                 vxlan_ip_miss(dev, &ipa);
1695         }
1696
1697 out:
1698         consume_skb(skb);
1699         return NETDEV_TX_OK;
1700 }
1701 #endif
1702
1703 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1704 {
1705         struct vxlan_dev *vxlan = netdev_priv(dev);
1706         struct neighbour *n;
1707
1708         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1709                 return false;
1710
1711         n = NULL;
1712         switch (ntohs(eth_hdr(skb)->h_proto)) {
1713         case ETH_P_IP:
1714         {
1715                 struct iphdr *pip;
1716
1717                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1718                         return false;
1719                 pip = ip_hdr(skb);
1720                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1721                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1722                         union vxlan_addr ipa = {
1723                                 .sin.sin_addr.s_addr = pip->daddr,
1724                                 .sin.sin_family = AF_INET,
1725                         };
1726
1727                         vxlan_ip_miss(dev, &ipa);
1728                         return false;
1729                 }
1730
1731                 break;
1732         }
1733 #if IS_ENABLED(CONFIG_IPV6)
1734         case ETH_P_IPV6:
1735         {
1736                 struct ipv6hdr *pip6;
1737
1738                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1739                         return false;
1740                 pip6 = ipv6_hdr(skb);
1741                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1742                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
1743                         union vxlan_addr ipa = {
1744                                 .sin6.sin6_addr = pip6->daddr,
1745                                 .sin6.sin6_family = AF_INET6,
1746                         };
1747
1748                         vxlan_ip_miss(dev, &ipa);
1749                         return false;
1750                 }
1751
1752                 break;
1753         }
1754 #endif
1755         default:
1756                 return false;
1757         }
1758
1759         if (n) {
1760                 bool diff;
1761
1762                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1763                 if (diff) {
1764                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1765                                 dev->addr_len);
1766                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1767                 }
1768                 neigh_release(n);
1769                 return diff;
1770         }
1771
1772         return false;
1773 }
1774
1775 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1776                                 struct vxlan_metadata *md)
1777 {
1778         struct vxlanhdr_gbp *gbp;
1779
1780         if (!md->gbp)
1781                 return;
1782
1783         gbp = (struct vxlanhdr_gbp *)vxh;
1784         vxh->vx_flags |= VXLAN_HF_GBP;
1785
1786         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1787                 gbp->dont_learn = 1;
1788
1789         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1790                 gbp->policy_applied = 1;
1791
1792         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1793 }
1794
1795 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1796                                __be16 protocol)
1797 {
1798         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1799
1800         gpe->np_applied = 1;
1801
1802         switch (protocol) {
1803         case htons(ETH_P_IP):
1804                 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1805                 return 0;
1806         case htons(ETH_P_IPV6):
1807                 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1808                 return 0;
1809         case htons(ETH_P_TEB):
1810                 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1811                 return 0;
1812         }
1813         return -EPFNOSUPPORT;
1814 }
1815
1816 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1817                            int iphdr_len, __be32 vni,
1818                            struct vxlan_metadata *md, u32 vxflags,
1819                            bool udp_sum)
1820 {
1821         struct vxlanhdr *vxh;
1822         int min_headroom;
1823         int err;
1824         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1825         __be16 inner_protocol = htons(ETH_P_TEB);
1826
1827         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1828             skb->ip_summed == CHECKSUM_PARTIAL) {
1829                 int csum_start = skb_checksum_start_offset(skb);
1830
1831                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1832                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1833                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1834                      skb->csum_offset == offsetof(struct tcphdr, check)))
1835                         type |= SKB_GSO_TUNNEL_REMCSUM;
1836         }
1837
1838         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1839                         + VXLAN_HLEN + iphdr_len;
1840
1841         /* Need space for new headers (invalidates iph ptr) */
1842         err = skb_cow_head(skb, min_headroom);
1843         if (unlikely(err))
1844                 return err;
1845
1846         err = iptunnel_handle_offloads(skb, type);
1847         if (err)
1848                 return err;
1849
1850         vxh = __skb_push(skb, sizeof(*vxh));
1851         vxh->vx_flags = VXLAN_HF_VNI;
1852         vxh->vx_vni = vxlan_vni_field(vni);
1853
1854         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1855                 unsigned int start;
1856
1857                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1858                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1859                 vxh->vx_flags |= VXLAN_HF_RCO;
1860
1861                 if (!skb_is_gso(skb)) {
1862                         skb->ip_summed = CHECKSUM_NONE;
1863                         skb->encapsulation = 0;
1864                 }
1865         }
1866
1867         if (vxflags & VXLAN_F_GBP)
1868                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1869         if (vxflags & VXLAN_F_GPE) {
1870                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1871                 if (err < 0)
1872                         return err;
1873                 inner_protocol = skb->protocol;
1874         }
1875
1876         skb_set_inner_protocol(skb, inner_protocol);
1877         return 0;
1878 }
1879
1880 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1881                                       struct vxlan_sock *sock4,
1882                                       struct sk_buff *skb, int oif, u8 tos,
1883                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
1884                                       struct dst_cache *dst_cache,
1885                                       const struct ip_tunnel_info *info)
1886 {
1887         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1888         struct rtable *rt = NULL;
1889         struct flowi4 fl4;
1890
1891         if (!sock4)
1892                 return ERR_PTR(-EIO);
1893
1894         if (tos && !info)
1895                 use_cache = false;
1896         if (use_cache) {
1897                 rt = dst_cache_get_ip4(dst_cache, saddr);
1898                 if (rt)
1899                         return rt;
1900         }
1901
1902         memset(&fl4, 0, sizeof(fl4));
1903         fl4.flowi4_oif = oif;
1904         fl4.flowi4_tos = RT_TOS(tos);
1905         fl4.flowi4_mark = skb->mark;
1906         fl4.flowi4_proto = IPPROTO_UDP;
1907         fl4.daddr = daddr;
1908         fl4.saddr = *saddr;
1909         fl4.fl4_dport = dport;
1910         fl4.fl4_sport = sport;
1911
1912         rt = ip_route_output_key(vxlan->net, &fl4);
1913         if (likely(!IS_ERR(rt))) {
1914                 if (rt->dst.dev == dev) {
1915                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1916                         ip_rt_put(rt);
1917                         return ERR_PTR(-ELOOP);
1918                 }
1919
1920                 *saddr = fl4.saddr;
1921                 if (use_cache)
1922                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1923         } else {
1924                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1925                 return ERR_PTR(-ENETUNREACH);
1926         }
1927         return rt;
1928 }
1929
1930 #if IS_ENABLED(CONFIG_IPV6)
1931 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1932                                           struct net_device *dev,
1933                                           struct vxlan_sock *sock6,
1934                                           struct sk_buff *skb, int oif, u8 tos,
1935                                           __be32 label,
1936                                           const struct in6_addr *daddr,
1937                                           struct in6_addr *saddr,
1938                                           __be16 dport, __be16 sport,
1939                                           struct dst_cache *dst_cache,
1940                                           const struct ip_tunnel_info *info)
1941 {
1942         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1943         struct dst_entry *ndst;
1944         struct flowi6 fl6;
1945         int err;
1946
1947         if (!sock6)
1948                 return ERR_PTR(-EIO);
1949
1950         if (tos && !info)
1951                 use_cache = false;
1952         if (use_cache) {
1953                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1954                 if (ndst)
1955                         return ndst;
1956         }
1957
1958         memset(&fl6, 0, sizeof(fl6));
1959         fl6.flowi6_oif = oif;
1960         fl6.daddr = *daddr;
1961         fl6.saddr = *saddr;
1962         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1963         fl6.flowi6_mark = skb->mark;
1964         fl6.flowi6_proto = IPPROTO_UDP;
1965         fl6.fl6_dport = dport;
1966         fl6.fl6_sport = sport;
1967
1968         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1969                                          sock6->sock->sk,
1970                                          &ndst, &fl6);
1971         if (unlikely(err < 0)) {
1972                 netdev_dbg(dev, "no route to %pI6\n", daddr);
1973                 return ERR_PTR(-ENETUNREACH);
1974         }
1975
1976         if (unlikely(ndst->dev == dev)) {
1977                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
1978                 dst_release(ndst);
1979                 return ERR_PTR(-ELOOP);
1980         }
1981
1982         *saddr = fl6.saddr;
1983         if (use_cache)
1984                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1985         return ndst;
1986 }
1987 #endif
1988
1989 /* Bypass encapsulation if the destination is local */
1990 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1991                                struct vxlan_dev *dst_vxlan, __be32 vni)
1992 {
1993         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1994         union vxlan_addr loopback;
1995         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1996         struct net_device *dev = skb->dev;
1997         int len = skb->len;
1998
1999         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2000         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2001         skb->pkt_type = PACKET_HOST;
2002         skb->encapsulation = 0;
2003         skb->dev = dst_vxlan->dev;
2004         __skb_pull(skb, skb_network_offset(skb));
2005
2006         if (remote_ip->sa.sa_family == AF_INET) {
2007                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2008                 loopback.sa.sa_family =  AF_INET;
2009 #if IS_ENABLED(CONFIG_IPV6)
2010         } else {
2011                 loopback.sin6.sin6_addr = in6addr_loopback;
2012                 loopback.sa.sa_family =  AF_INET6;
2013 #endif
2014         }
2015
2016         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
2017                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source, 0,
2018                             vni);
2019
2020         u64_stats_update_begin(&tx_stats->syncp);
2021         tx_stats->tx_packets++;
2022         tx_stats->tx_bytes += len;
2023         u64_stats_update_end(&tx_stats->syncp);
2024
2025         if (netif_rx(skb) == NET_RX_SUCCESS) {
2026                 u64_stats_update_begin(&rx_stats->syncp);
2027                 rx_stats->rx_packets++;
2028                 rx_stats->rx_bytes += len;
2029                 u64_stats_update_end(&rx_stats->syncp);
2030         } else {
2031                 dev->stats.rx_dropped++;
2032         }
2033 }
2034
2035 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2036                                  struct vxlan_dev *vxlan,
2037                                  union vxlan_addr *daddr,
2038                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2039                                  struct dst_entry *dst,
2040                                  u32 rt_flags)
2041 {
2042 #if IS_ENABLED(CONFIG_IPV6)
2043         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2044          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2045          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2046          */
2047         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2048 #endif
2049         /* Bypass encapsulation if the destination is local */
2050         if (rt_flags & RTCF_LOCAL &&
2051             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2052                 struct vxlan_dev *dst_vxlan;
2053
2054                 dst_release(dst);
2055                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2056                                            daddr->sa.sa_family, dst_port,
2057                                            vxlan->cfg.flags);
2058                 if (!dst_vxlan) {
2059                         dev->stats.tx_errors++;
2060                         kfree_skb(skb);
2061
2062                         return -ENOENT;
2063                 }
2064                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2065                 return 1;
2066         }
2067
2068         return 0;
2069 }
2070
2071 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2072                            __be32 default_vni, struct vxlan_rdst *rdst,
2073                            bool did_rsc)
2074 {
2075         struct dst_cache *dst_cache;
2076         struct ip_tunnel_info *info;
2077         struct vxlan_dev *vxlan = netdev_priv(dev);
2078         const struct iphdr *old_iph = ip_hdr(skb);
2079         union vxlan_addr *dst;
2080         union vxlan_addr remote_ip, local_ip;
2081         struct vxlan_metadata _md;
2082         struct vxlan_metadata *md = &_md;
2083         __be16 src_port = 0, dst_port;
2084         struct dst_entry *ndst = NULL;
2085         __be32 vni, label;
2086         __u8 tos, ttl;
2087         int ifindex;
2088         int err;
2089         u32 flags = vxlan->cfg.flags;
2090         bool udp_sum = false;
2091         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2092
2093         info = skb_tunnel_info(skb);
2094
2095         if (rdst) {
2096                 dst = &rdst->remote_ip;
2097                 if (vxlan_addr_any(dst)) {
2098                         if (did_rsc) {
2099                                 /* short-circuited back to local bridge */
2100                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2101                                 return;
2102                         }
2103                         goto drop;
2104                 }
2105
2106                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2107                 vni = (rdst->remote_vni) ? : default_vni;
2108                 ifindex = rdst->remote_ifindex;
2109                 local_ip = vxlan->cfg.saddr;
2110                 dst_cache = &rdst->dst_cache;
2111                 md->gbp = skb->mark;
2112                 ttl = vxlan->cfg.ttl;
2113                 if (!ttl && vxlan_addr_multicast(dst))
2114                         ttl = 1;
2115
2116                 tos = vxlan->cfg.tos;
2117                 if (tos == 1)
2118                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2119
2120                 if (dst->sa.sa_family == AF_INET)
2121                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2122                 else
2123                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2124                 label = vxlan->cfg.label;
2125         } else {
2126                 if (!info) {
2127                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2128                                   dev->name);
2129                         goto drop;
2130                 }
2131                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2132                 if (remote_ip.sa.sa_family == AF_INET) {
2133                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2134                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2135                 } else {
2136                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2137                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2138                 }
2139                 dst = &remote_ip;
2140                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2141                 vni = tunnel_id_to_key32(info->key.tun_id);
2142                 ifindex = 0;
2143                 dst_cache = &info->dst_cache;
2144                 if (info->options_len)
2145                         md = ip_tunnel_info_opts(info);
2146                 ttl = info->key.ttl;
2147                 tos = info->key.tos;
2148                 label = info->key.label;
2149                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2150         }
2151         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2152                                      vxlan->cfg.port_max, true);
2153
2154         rcu_read_lock();
2155         if (dst->sa.sa_family == AF_INET) {
2156                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2157                 struct rtable *rt;
2158                 __be16 df = 0;
2159
2160                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2161                                      dst->sin.sin_addr.s_addr,
2162                                      &local_ip.sin.sin_addr.s_addr,
2163                                      dst_port, src_port,
2164                                      dst_cache, info);
2165                 if (IS_ERR(rt)) {
2166                         err = PTR_ERR(rt);
2167                         goto tx_error;
2168                 }
2169
2170                 /* Bypass encapsulation if the destination is local */
2171                 if (!info) {
2172                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2173                                                     dst_port, ifindex, vni,
2174                                                     &rt->dst, rt->rt_flags);
2175                         if (err)
2176                                 goto out_unlock;
2177                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2178                         df = htons(IP_DF);
2179                 }
2180
2181                 ndst = &rt->dst;
2182                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2183                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2184                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2185                                       vni, md, flags, udp_sum);
2186                 if (err < 0)
2187                         goto tx_error;
2188
2189                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2190                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2191                                     src_port, dst_port, xnet, !udp_sum);
2192 #if IS_ENABLED(CONFIG_IPV6)
2193         } else {
2194                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2195
2196                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2197                                         label, &dst->sin6.sin6_addr,
2198                                         &local_ip.sin6.sin6_addr,
2199                                         dst_port, src_port,
2200                                         dst_cache, info);
2201                 if (IS_ERR(ndst)) {
2202                         err = PTR_ERR(ndst);
2203                         ndst = NULL;
2204                         goto tx_error;
2205                 }
2206
2207                 if (!info) {
2208                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2209
2210                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2211                                                     dst_port, ifindex, vni,
2212                                                     ndst, rt6i_flags);
2213                         if (err)
2214                                 goto out_unlock;
2215                 }
2216
2217                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2218                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2219                 skb_scrub_packet(skb, xnet);
2220                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2221                                       vni, md, flags, udp_sum);
2222                 if (err < 0)
2223                         goto tx_error;
2224
2225                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2226                                      &local_ip.sin6.sin6_addr,
2227                                      &dst->sin6.sin6_addr, tos, ttl,
2228                                      label, src_port, dst_port, !udp_sum);
2229 #endif
2230         }
2231 out_unlock:
2232         rcu_read_unlock();
2233         return;
2234
2235 drop:
2236         dev->stats.tx_dropped++;
2237         dev_kfree_skb(skb);
2238         return;
2239
2240 tx_error:
2241         rcu_read_unlock();
2242         if (err == -ELOOP)
2243                 dev->stats.collisions++;
2244         else if (err == -ENETUNREACH)
2245                 dev->stats.tx_carrier_errors++;
2246         dst_release(ndst);
2247         dev->stats.tx_errors++;
2248         kfree_skb(skb);
2249 }
2250
2251 /* Transmit local packets over Vxlan
2252  *
2253  * Outer IP header inherits ECN and DF from inner header.
2254  * Outer UDP destination is the VXLAN assigned port.
2255  *           source port is based on hash of flow
2256  */
2257 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2258 {
2259         struct vxlan_dev *vxlan = netdev_priv(dev);
2260         const struct ip_tunnel_info *info;
2261         struct ethhdr *eth;
2262         bool did_rsc = false;
2263         struct vxlan_rdst *rdst, *fdst = NULL;
2264         struct vxlan_fdb *f;
2265         __be32 vni = 0;
2266
2267         info = skb_tunnel_info(skb);
2268
2269         skb_reset_mac_header(skb);
2270
2271         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2272                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2273                     info->mode & IP_TUNNEL_INFO_TX) {
2274                         vni = tunnel_id_to_key32(info->key.tun_id);
2275                 } else {
2276                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2277                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2278                         else
2279                                 kfree_skb(skb);
2280                         return NETDEV_TX_OK;
2281                 }
2282         }
2283
2284         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2285                 eth = eth_hdr(skb);
2286                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2287                         return arp_reduce(dev, skb, vni);
2288 #if IS_ENABLED(CONFIG_IPV6)
2289                 else if (ntohs(eth->h_proto) == ETH_P_IPV6) {
2290                         struct ipv6hdr *hdr, _hdr;
2291                         if ((hdr = skb_header_pointer(skb,
2292                                                       skb_network_offset(skb),
2293                                                       sizeof(_hdr), &_hdr)) &&
2294                             hdr->nexthdr == IPPROTO_ICMPV6)
2295                                 return neigh_reduce(dev, skb, vni);
2296                 }
2297 #endif
2298         }
2299
2300         eth = eth_hdr(skb);
2301         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2302         did_rsc = false;
2303
2304         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2305             (ntohs(eth->h_proto) == ETH_P_IP ||
2306              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2307                 did_rsc = route_shortcircuit(dev, skb);
2308                 if (did_rsc)
2309                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2310         }
2311
2312         if (f == NULL) {
2313                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2314                 if (f == NULL) {
2315                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2316                             !is_multicast_ether_addr(eth->h_dest))
2317                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2318
2319                         dev->stats.tx_dropped++;
2320                         kfree_skb(skb);
2321                         return NETDEV_TX_OK;
2322                 }
2323         }
2324
2325         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2326                 struct sk_buff *skb1;
2327
2328                 if (!fdst) {
2329                         fdst = rdst;
2330                         continue;
2331                 }
2332                 skb1 = skb_clone(skb, GFP_ATOMIC);
2333                 if (skb1)
2334                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2335         }
2336
2337         if (fdst)
2338                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2339         else
2340                 kfree_skb(skb);
2341         return NETDEV_TX_OK;
2342 }
2343
2344 /* Walk the forwarding table and purge stale entries */
2345 static void vxlan_cleanup(unsigned long arg)
2346 {
2347         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2348         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2349         unsigned int h;
2350
2351         if (!netif_running(vxlan->dev))
2352                 return;
2353
2354         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2355                 struct hlist_node *p, *n;
2356
2357                 spin_lock_bh(&vxlan->hash_lock);
2358                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2359                         struct vxlan_fdb *f
2360                                 = container_of(p, struct vxlan_fdb, hlist);
2361                         unsigned long timeout;
2362
2363                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2364                                 continue;
2365
2366                         if (f->flags & NTF_EXT_LEARNED)
2367                                 continue;
2368
2369                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2370                         if (time_before_eq(timeout, jiffies)) {
2371                                 netdev_dbg(vxlan->dev,
2372                                            "garbage collect %pM\n",
2373                                            f->eth_addr);
2374                                 f->state = NUD_STALE;
2375                                 vxlan_fdb_destroy(vxlan, f);
2376                         } else if (time_before(timeout, next_timer))
2377                                 next_timer = timeout;
2378                 }
2379                 spin_unlock_bh(&vxlan->hash_lock);
2380         }
2381
2382         mod_timer(&vxlan->age_timer, next_timer);
2383 }
2384
2385 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2386 {
2387         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2388
2389         spin_lock(&vn->sock_lock);
2390         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2391 #if IS_ENABLED(CONFIG_IPV6)
2392         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2393 #endif
2394         spin_unlock(&vn->sock_lock);
2395 }
2396
2397 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2398                              struct vxlan_dev_node *node)
2399 {
2400         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2401         __be32 vni = vxlan->default_dst.remote_vni;
2402
2403         node->vxlan = vxlan;
2404         spin_lock(&vn->sock_lock);
2405         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2406         spin_unlock(&vn->sock_lock);
2407 }
2408
2409 /* Setup stats when device is created */
2410 static int vxlan_init(struct net_device *dev)
2411 {
2412         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2413         if (!dev->tstats)
2414                 return -ENOMEM;
2415
2416         return 0;
2417 }
2418
2419 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2420 {
2421         struct vxlan_fdb *f;
2422
2423         spin_lock_bh(&vxlan->hash_lock);
2424         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2425         if (f)
2426                 vxlan_fdb_destroy(vxlan, f);
2427         spin_unlock_bh(&vxlan->hash_lock);
2428 }
2429
2430 static void vxlan_uninit(struct net_device *dev)
2431 {
2432         struct vxlan_dev *vxlan = netdev_priv(dev);
2433
2434         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2435
2436         free_percpu(dev->tstats);
2437 }
2438
2439 /* Start ageing timer and join group when device is brought up */
2440 static int vxlan_open(struct net_device *dev)
2441 {
2442         struct vxlan_dev *vxlan = netdev_priv(dev);
2443         int ret;
2444
2445         ret = vxlan_sock_add(vxlan);
2446         if (ret < 0)
2447                 return ret;
2448
2449         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2450                 ret = vxlan_igmp_join(vxlan);
2451                 if (ret == -EADDRINUSE)
2452                         ret = 0;
2453                 if (ret) {
2454                         vxlan_sock_release(vxlan);
2455                         return ret;
2456                 }
2457         }
2458
2459         if (vxlan->cfg.age_interval)
2460                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2461
2462         return ret;
2463 }
2464
2465 /* Purge the forwarding table */
2466 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2467 {
2468         unsigned int h;
2469
2470         spin_lock_bh(&vxlan->hash_lock);
2471         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2472                 struct hlist_node *p, *n;
2473                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2474                         struct vxlan_fdb *f
2475                                 = container_of(p, struct vxlan_fdb, hlist);
2476                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2477                                 continue;
2478                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2479                         if (!is_zero_ether_addr(f->eth_addr))
2480                                 vxlan_fdb_destroy(vxlan, f);
2481                 }
2482         }
2483         spin_unlock_bh(&vxlan->hash_lock);
2484 }
2485
2486 /* Cleanup timer and forwarding table on shutdown */
2487 static int vxlan_stop(struct net_device *dev)
2488 {
2489         struct vxlan_dev *vxlan = netdev_priv(dev);
2490         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2491         int ret = 0;
2492
2493         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2494             !vxlan_group_used(vn, vxlan))
2495                 ret = vxlan_igmp_leave(vxlan);
2496
2497         del_timer_sync(&vxlan->age_timer);
2498
2499         vxlan_flush(vxlan, false);
2500         vxlan_sock_release(vxlan);
2501
2502         return ret;
2503 }
2504
2505 /* Stub, nothing needs to be done. */
2506 static void vxlan_set_multicast_list(struct net_device *dev)
2507 {
2508 }
2509
2510 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2511 {
2512         struct vxlan_dev *vxlan = netdev_priv(dev);
2513         struct vxlan_rdst *dst = &vxlan->default_dst;
2514         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2515                                                          dst->remote_ifindex);
2516         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
2517
2518         /* This check is different than dev->max_mtu, because it looks at
2519          * the lowerdev->mtu, rather than the static dev->max_mtu
2520          */
2521         if (lowerdev) {
2522                 int max_mtu = lowerdev->mtu -
2523                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2524                 if (new_mtu > max_mtu)
2525                         return -EINVAL;
2526         }
2527
2528         dev->mtu = new_mtu;
2529         return 0;
2530 }
2531
2532 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2533 {
2534         struct vxlan_dev *vxlan = netdev_priv(dev);
2535         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2536         __be16 sport, dport;
2537
2538         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2539                                   vxlan->cfg.port_max, true);
2540         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2541
2542         if (ip_tunnel_info_af(info) == AF_INET) {
2543                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2544                 struct rtable *rt;
2545
2546                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2547                                      info->key.u.ipv4.dst,
2548                                      &info->key.u.ipv4.src, dport, sport,
2549                                      &info->dst_cache, info);
2550                 if (IS_ERR(rt))
2551                         return PTR_ERR(rt);
2552                 ip_rt_put(rt);
2553         } else {
2554 #if IS_ENABLED(CONFIG_IPV6)
2555                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2556                 struct dst_entry *ndst;
2557
2558                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2559                                         info->key.label, &info->key.u.ipv6.dst,
2560                                         &info->key.u.ipv6.src, dport, sport,
2561                                         &info->dst_cache, info);
2562                 if (IS_ERR(ndst))
2563                         return PTR_ERR(ndst);
2564                 dst_release(ndst);
2565 #else /* !CONFIG_IPV6 */
2566                 return -EPFNOSUPPORT;
2567 #endif
2568         }
2569         info->key.tp_src = sport;
2570         info->key.tp_dst = dport;
2571         return 0;
2572 }
2573
2574 static const struct net_device_ops vxlan_netdev_ether_ops = {
2575         .ndo_init               = vxlan_init,
2576         .ndo_uninit             = vxlan_uninit,
2577         .ndo_open               = vxlan_open,
2578         .ndo_stop               = vxlan_stop,
2579         .ndo_start_xmit         = vxlan_xmit,
2580         .ndo_get_stats64        = ip_tunnel_get_stats64,
2581         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2582         .ndo_change_mtu         = vxlan_change_mtu,
2583         .ndo_validate_addr      = eth_validate_addr,
2584         .ndo_set_mac_address    = eth_mac_addr,
2585         .ndo_fdb_add            = vxlan_fdb_add,
2586         .ndo_fdb_del            = vxlan_fdb_delete,
2587         .ndo_fdb_dump           = vxlan_fdb_dump,
2588         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2589 };
2590
2591 static const struct net_device_ops vxlan_netdev_raw_ops = {
2592         .ndo_init               = vxlan_init,
2593         .ndo_uninit             = vxlan_uninit,
2594         .ndo_open               = vxlan_open,
2595         .ndo_stop               = vxlan_stop,
2596         .ndo_start_xmit         = vxlan_xmit,
2597         .ndo_get_stats64        = ip_tunnel_get_stats64,
2598         .ndo_change_mtu         = vxlan_change_mtu,
2599         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2600 };
2601
2602 /* Info for udev, that this is a virtual tunnel endpoint */
2603 static struct device_type vxlan_type = {
2604         .name = "vxlan",
2605 };
2606
2607 /* Calls the ndo_udp_tunnel_add of the caller in order to
2608  * supply the listening VXLAN udp ports. Callers are expected
2609  * to implement the ndo_udp_tunnel_add.
2610  */
2611 static void vxlan_push_rx_ports(struct net_device *dev)
2612 {
2613         struct vxlan_sock *vs;
2614         struct net *net = dev_net(dev);
2615         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2616         unsigned int i;
2617
2618         spin_lock(&vn->sock_lock);
2619         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2620                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2621                         udp_tunnel_push_rx_port(dev, vs->sock,
2622                                                 (vs->flags & VXLAN_F_GPE) ?
2623                                                 UDP_TUNNEL_TYPE_VXLAN_GPE :
2624                                                 UDP_TUNNEL_TYPE_VXLAN);
2625         }
2626         spin_unlock(&vn->sock_lock);
2627 }
2628
2629 /* Initialize the device structure. */
2630 static void vxlan_setup(struct net_device *dev)
2631 {
2632         struct vxlan_dev *vxlan = netdev_priv(dev);
2633         unsigned int h;
2634
2635         eth_hw_addr_random(dev);
2636         ether_setup(dev);
2637
2638         dev->needs_free_netdev = true;
2639         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2640
2641         dev->features   |= NETIF_F_LLTX;
2642         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2643         dev->features   |= NETIF_F_RXCSUM;
2644         dev->features   |= NETIF_F_GSO_SOFTWARE;
2645
2646         dev->vlan_features = dev->features;
2647         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2648         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2649         netif_keep_dst(dev);
2650         dev->priv_flags |= IFF_NO_QUEUE;
2651
2652         /* MTU range: 68 - 65535 */
2653         dev->min_mtu = ETH_MIN_MTU;
2654         dev->max_mtu = ETH_MAX_MTU;
2655
2656         INIT_LIST_HEAD(&vxlan->next);
2657         spin_lock_init(&vxlan->hash_lock);
2658
2659         init_timer_deferrable(&vxlan->age_timer);
2660         vxlan->age_timer.function = vxlan_cleanup;
2661         vxlan->age_timer.data = (unsigned long) vxlan;
2662
2663         vxlan->dev = dev;
2664
2665         gro_cells_init(&vxlan->gro_cells, dev);
2666
2667         for (h = 0; h < FDB_HASH_SIZE; ++h)
2668                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2669 }
2670
2671 static void vxlan_ether_setup(struct net_device *dev)
2672 {
2673         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2674         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2675         dev->netdev_ops = &vxlan_netdev_ether_ops;
2676 }
2677
2678 static void vxlan_raw_setup(struct net_device *dev)
2679 {
2680         dev->header_ops = NULL;
2681         dev->type = ARPHRD_NONE;
2682         dev->hard_header_len = 0;
2683         dev->addr_len = 0;
2684         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2685         dev->netdev_ops = &vxlan_netdev_raw_ops;
2686 }
2687
2688 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2689         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2690         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2691         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2692         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2693         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2694         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2695         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2696         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2697         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2698         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2699         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2700         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2701         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2702         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2703         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2704         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2705         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2706         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2707         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2708         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2709         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2710         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2711         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2712         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2713         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2714         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2715         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2716 };
2717
2718 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
2719                           struct netlink_ext_ack *extack)
2720 {
2721         if (tb[IFLA_ADDRESS]) {
2722                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2723                         pr_debug("invalid link address (not ethernet)\n");
2724                         return -EINVAL;
2725                 }
2726
2727                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2728                         pr_debug("invalid all zero ethernet address\n");
2729                         return -EADDRNOTAVAIL;
2730                 }
2731         }
2732
2733         if (tb[IFLA_MTU]) {
2734                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
2735
2736                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU)
2737                         return -EINVAL;
2738         }
2739
2740         if (!data)
2741                 return -EINVAL;
2742
2743         if (data[IFLA_VXLAN_ID]) {
2744                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2745
2746                 if (id >= VXLAN_N_VID)
2747                         return -ERANGE;
2748         }
2749
2750         if (data[IFLA_VXLAN_PORT_RANGE]) {
2751                 const struct ifla_vxlan_port_range *p
2752                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2753
2754                 if (ntohs(p->high) < ntohs(p->low)) {
2755                         pr_debug("port range %u .. %u not valid\n",
2756                                  ntohs(p->low), ntohs(p->high));
2757                         return -EINVAL;
2758                 }
2759         }
2760
2761         return 0;
2762 }
2763
2764 static void vxlan_get_drvinfo(struct net_device *netdev,
2765                               struct ethtool_drvinfo *drvinfo)
2766 {
2767         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2768         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2769 }
2770
2771 static const struct ethtool_ops vxlan_ethtool_ops = {
2772         .get_drvinfo    = vxlan_get_drvinfo,
2773         .get_link       = ethtool_op_get_link,
2774 };
2775
2776 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2777                                         __be16 port, u32 flags)
2778 {
2779         struct socket *sock;
2780         struct udp_port_cfg udp_conf;
2781         int err;
2782
2783         memset(&udp_conf, 0, sizeof(udp_conf));
2784
2785         if (ipv6) {
2786                 udp_conf.family = AF_INET6;
2787                 udp_conf.use_udp6_rx_checksums =
2788                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2789                 udp_conf.ipv6_v6only = 1;
2790         } else {
2791                 udp_conf.family = AF_INET;
2792         }
2793
2794         udp_conf.local_udp_port = port;
2795
2796         /* Open UDP socket */
2797         err = udp_sock_create(net, &udp_conf, &sock);
2798         if (err < 0)
2799                 return ERR_PTR(err);
2800
2801         return sock;
2802 }
2803
2804 /* Create new listen socket if needed */
2805 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2806                                               __be16 port, u32 flags)
2807 {
2808         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2809         struct vxlan_sock *vs;
2810         struct socket *sock;
2811         unsigned int h;
2812         struct udp_tunnel_sock_cfg tunnel_cfg;
2813
2814         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2815         if (!vs)
2816                 return ERR_PTR(-ENOMEM);
2817
2818         for (h = 0; h < VNI_HASH_SIZE; ++h)
2819                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2820
2821         sock = vxlan_create_sock(net, ipv6, port, flags);
2822         if (IS_ERR(sock)) {
2823                 kfree(vs);
2824                 return ERR_CAST(sock);
2825         }
2826
2827         vs->sock = sock;
2828         refcount_set(&vs->refcnt, 1);
2829         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2830
2831         spin_lock(&vn->sock_lock);
2832         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2833         udp_tunnel_notify_add_rx_port(sock,
2834                                       (vs->flags & VXLAN_F_GPE) ?
2835                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2836                                       UDP_TUNNEL_TYPE_VXLAN);
2837         spin_unlock(&vn->sock_lock);
2838
2839         /* Mark socket as an encapsulation socket. */
2840         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2841         tunnel_cfg.sk_user_data = vs;
2842         tunnel_cfg.encap_type = 1;
2843         tunnel_cfg.encap_rcv = vxlan_rcv;
2844         tunnel_cfg.encap_destroy = NULL;
2845         tunnel_cfg.gro_receive = vxlan_gro_receive;
2846         tunnel_cfg.gro_complete = vxlan_gro_complete;
2847
2848         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2849
2850         return vs;
2851 }
2852
2853 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2854 {
2855         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2856         struct vxlan_sock *vs = NULL;
2857         struct vxlan_dev_node *node;
2858
2859         if (!vxlan->cfg.no_share) {
2860                 spin_lock(&vn->sock_lock);
2861                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2862                                      vxlan->cfg.dst_port, vxlan->cfg.flags);
2863                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
2864                         spin_unlock(&vn->sock_lock);
2865                         return -EBUSY;
2866                 }
2867                 spin_unlock(&vn->sock_lock);
2868         }
2869         if (!vs)
2870                 vs = vxlan_socket_create(vxlan->net, ipv6,
2871                                          vxlan->cfg.dst_port, vxlan->cfg.flags);
2872         if (IS_ERR(vs))
2873                 return PTR_ERR(vs);
2874 #if IS_ENABLED(CONFIG_IPV6)
2875         if (ipv6) {
2876                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2877                 node = &vxlan->hlist6;
2878         } else
2879 #endif
2880         {
2881                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2882                 node = &vxlan->hlist4;
2883         }
2884         vxlan_vs_add_dev(vs, vxlan, node);
2885         return 0;
2886 }
2887
2888 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2889 {
2890         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
2891         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
2892         bool ipv4 = !ipv6 || metadata;
2893         int ret = 0;
2894
2895         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2896 #if IS_ENABLED(CONFIG_IPV6)
2897         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2898         if (ipv6) {
2899                 ret = __vxlan_sock_add(vxlan, true);
2900                 if (ret < 0 && ret != -EAFNOSUPPORT)
2901                         ipv4 = false;
2902         }
2903 #endif
2904         if (ipv4)
2905                 ret = __vxlan_sock_add(vxlan, false);
2906         if (ret < 0)
2907                 vxlan_sock_release(vxlan);
2908         return ret;
2909 }
2910
2911 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
2912                                  struct net_device **lower,
2913                                  struct vxlan_dev *old)
2914 {
2915         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2916         struct vxlan_dev *tmp;
2917         bool use_ipv6 = false;
2918
2919         if (conf->flags & VXLAN_F_GPE) {
2920                 /* For now, allow GPE only together with
2921                  * COLLECT_METADATA. This can be relaxed later; in such
2922                  * case, the other side of the PtP link will have to be
2923                  * provided.
2924                  */
2925                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2926                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2927                         return -EINVAL;
2928                 }
2929         }
2930
2931         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
2932                 /* Unless IPv6 is explicitly requested, assume IPv4 */
2933                 conf->remote_ip.sa.sa_family = AF_INET;
2934                 conf->saddr.sa.sa_family = AF_INET;
2935         } else if (!conf->remote_ip.sa.sa_family) {
2936                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
2937         } else if (!conf->saddr.sa.sa_family) {
2938                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
2939         }
2940
2941         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family)
2942                 return -EINVAL;
2943
2944         if (vxlan_addr_multicast(&conf->saddr))
2945                 return -EINVAL;
2946
2947         if (conf->saddr.sa.sa_family == AF_INET6) {
2948                 if (!IS_ENABLED(CONFIG_IPV6))
2949                         return -EPFNOSUPPORT;
2950                 use_ipv6 = true;
2951                 conf->flags |= VXLAN_F_IPV6;
2952
2953                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2954                         int local_type =
2955                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
2956                         int remote_type =
2957                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
2958
2959                         if (local_type & IPV6_ADDR_LINKLOCAL) {
2960                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
2961                                     (remote_type != IPV6_ADDR_ANY))
2962                                         return -EINVAL;
2963
2964                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
2965                         } else {
2966                                 if (remote_type ==
2967                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL))
2968                                         return -EINVAL;
2969
2970                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
2971                         }
2972                 }
2973         }
2974
2975         if (conf->label && !use_ipv6)
2976                 return -EINVAL;
2977
2978         if (conf->remote_ifindex) {
2979                 struct net_device *lowerdev;
2980
2981                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2982                 if (!lowerdev)
2983                         return -ENODEV;
2984
2985 #if IS_ENABLED(CONFIG_IPV6)
2986                 if (use_ipv6) {
2987                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2988                         if (idev && idev->cnf.disable_ipv6)
2989                                 return -EPERM;
2990                 }
2991 #endif
2992
2993                 *lower = lowerdev;
2994         } else {
2995                 if (vxlan_addr_multicast(&conf->remote_ip))
2996                         return -EINVAL;
2997
2998 #if IS_ENABLED(CONFIG_IPV6)
2999                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL)
3000                         return -EINVAL;
3001 #endif
3002
3003                 *lower = NULL;
3004         }
3005
3006         if (!conf->dst_port) {
3007                 if (conf->flags & VXLAN_F_GPE)
3008                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3009                 else
3010                         conf->dst_port = htons(vxlan_port);
3011         }
3012
3013         if (!conf->age_interval)
3014                 conf->age_interval = FDB_AGE_DEFAULT;
3015
3016         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3017                 if (tmp == old)
3018                         continue;
3019
3020                 if (tmp->cfg.vni != conf->vni)
3021                         continue;
3022                 if (tmp->cfg.dst_port != conf->dst_port)
3023                         continue;
3024                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3025                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3026                         continue;
3027
3028                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3029                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3030                         continue;
3031
3032                 return -EEXIST;
3033         }
3034
3035         return 0;
3036 }
3037
3038 static void vxlan_config_apply(struct net_device *dev,
3039                                struct vxlan_config *conf,
3040                                struct net_device *lowerdev,
3041                                struct net *src_net,
3042                                bool changelink)
3043 {
3044         struct vxlan_dev *vxlan = netdev_priv(dev);
3045         struct vxlan_rdst *dst = &vxlan->default_dst;
3046         unsigned short needed_headroom = ETH_HLEN;
3047         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3048         int max_mtu = ETH_MAX_MTU;
3049
3050         if (!changelink) {
3051                 if (conf->flags & VXLAN_F_GPE)
3052                         vxlan_raw_setup(dev);
3053                 else
3054                         vxlan_ether_setup(dev);
3055
3056                 if (conf->mtu)
3057                         dev->mtu = conf->mtu;
3058
3059                 vxlan->net = src_net;
3060         }
3061
3062         dst->remote_vni = conf->vni;
3063
3064         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3065
3066         if (lowerdev) {
3067                 dst->remote_ifindex = conf->remote_ifindex;
3068
3069                 dev->gso_max_size = lowerdev->gso_max_size;
3070                 dev->gso_max_segs = lowerdev->gso_max_segs;
3071
3072                 needed_headroom = lowerdev->hard_header_len;
3073
3074                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3075                                            VXLAN_HEADROOM);
3076         }
3077
3078         if (dev->mtu > max_mtu)
3079                 dev->mtu = max_mtu;
3080
3081         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3082                 needed_headroom += VXLAN6_HEADROOM;
3083         else
3084                 needed_headroom += VXLAN_HEADROOM;
3085         dev->needed_headroom = needed_headroom;
3086
3087         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3088 }
3089
3090 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3091                                struct vxlan_config *conf,
3092                                bool changelink)
3093 {
3094         struct vxlan_dev *vxlan = netdev_priv(dev);
3095         struct net_device *lowerdev;
3096         int ret;
3097
3098         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan);
3099         if (ret)
3100                 return ret;
3101
3102         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3103
3104         return 0;
3105 }
3106
3107 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3108                               struct vxlan_config *conf)
3109 {
3110         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3111         struct vxlan_dev *vxlan = netdev_priv(dev);
3112         int err;
3113
3114         err = vxlan_dev_configure(net, dev, conf, false);
3115         if (err)
3116                 return err;
3117
3118         dev->ethtool_ops = &vxlan_ethtool_ops;
3119
3120         /* create an fdb entry for a valid default destination */
3121         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
3122                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3123                                        &vxlan->default_dst.remote_ip,
3124                                        NUD_REACHABLE | NUD_PERMANENT,
3125                                        NLM_F_EXCL | NLM_F_CREATE,
3126                                        vxlan->cfg.dst_port,
3127                                        vxlan->default_dst.remote_vni,
3128                                        vxlan->default_dst.remote_vni,
3129                                        vxlan->default_dst.remote_ifindex,
3130                                        NTF_SELF);
3131                 if (err)
3132                         return err;
3133         }
3134
3135         err = register_netdevice(dev);
3136         if (err) {
3137                 vxlan_fdb_delete_default(vxlan, vxlan->default_dst.remote_vni);
3138                 return err;
3139         }
3140
3141         list_add(&vxlan->next, &vn->vxlan_list);
3142         return 0;
3143 }
3144
3145 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3146                          struct net_device *dev, struct vxlan_config *conf,
3147                          bool changelink)
3148 {
3149         struct vxlan_dev *vxlan = netdev_priv(dev);
3150
3151         memset(conf, 0, sizeof(*conf));
3152
3153         /* if changelink operation, start with old existing cfg */
3154         if (changelink)
3155                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3156
3157         if (data[IFLA_VXLAN_ID]) {
3158                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3159
3160                 if (changelink && (vni != conf->vni))
3161                         return -EOPNOTSUPP;
3162                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3163         }
3164
3165         if (data[IFLA_VXLAN_GROUP]) {
3166                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET))
3167                         return -EOPNOTSUPP;
3168
3169                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3170                 conf->remote_ip.sa.sa_family = AF_INET;
3171         } else if (data[IFLA_VXLAN_GROUP6]) {
3172                 if (!IS_ENABLED(CONFIG_IPV6))
3173                         return -EPFNOSUPPORT;
3174
3175                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6))
3176                         return -EOPNOTSUPP;
3177
3178                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3179                 conf->remote_ip.sa.sa_family = AF_INET6;
3180         }
3181
3182         if (data[IFLA_VXLAN_LOCAL]) {
3183                 if (changelink && (conf->saddr.sa.sa_family != AF_INET))
3184                         return -EOPNOTSUPP;
3185
3186                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3187                 conf->saddr.sa.sa_family = AF_INET;
3188         } else if (data[IFLA_VXLAN_LOCAL6]) {
3189                 if (!IS_ENABLED(CONFIG_IPV6))
3190                         return -EPFNOSUPPORT;
3191
3192                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6))
3193                         return -EOPNOTSUPP;
3194
3195                 /* TODO: respect scope id */
3196                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3197                 conf->saddr.sa.sa_family = AF_INET6;
3198         }
3199
3200         if (data[IFLA_VXLAN_LINK])
3201                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3202
3203         if (data[IFLA_VXLAN_TOS])
3204                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3205
3206         if (data[IFLA_VXLAN_TTL])
3207                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3208
3209         if (data[IFLA_VXLAN_LABEL])
3210                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3211                              IPV6_FLOWLABEL_MASK;
3212
3213         if (data[IFLA_VXLAN_LEARNING]) {
3214                 if (nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3215                         conf->flags |= VXLAN_F_LEARN;
3216                 else
3217                         conf->flags &= ~VXLAN_F_LEARN;
3218         } else if (!changelink) {
3219                 /* default to learn on a new device */
3220                 conf->flags |= VXLAN_F_LEARN;
3221         }
3222
3223         if (data[IFLA_VXLAN_AGEING]) {
3224                 if (changelink)
3225                         return -EOPNOTSUPP;
3226                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3227         }
3228
3229         if (data[IFLA_VXLAN_PROXY]) {
3230                 if (changelink)
3231                         return -EOPNOTSUPP;
3232                 if (nla_get_u8(data[IFLA_VXLAN_PROXY]))
3233                         conf->flags |= VXLAN_F_PROXY;
3234         }
3235
3236         if (data[IFLA_VXLAN_RSC]) {
3237                 if (changelink)
3238                         return -EOPNOTSUPP;
3239                 if (nla_get_u8(data[IFLA_VXLAN_RSC]))
3240                         conf->flags |= VXLAN_F_RSC;
3241         }
3242
3243         if (data[IFLA_VXLAN_L2MISS]) {
3244                 if (changelink)
3245                         return -EOPNOTSUPP;
3246                 if (nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3247                         conf->flags |= VXLAN_F_L2MISS;
3248         }
3249
3250         if (data[IFLA_VXLAN_L3MISS]) {
3251                 if (changelink)
3252                         return -EOPNOTSUPP;
3253                 if (nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3254                         conf->flags |= VXLAN_F_L3MISS;
3255         }
3256
3257         if (data[IFLA_VXLAN_LIMIT]) {
3258                 if (changelink)
3259                         return -EOPNOTSUPP;
3260                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3261         }
3262
3263         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3264                 if (changelink)
3265                         return -EOPNOTSUPP;
3266                 if (nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3267                         conf->flags |= VXLAN_F_COLLECT_METADATA;
3268         }
3269
3270         if (data[IFLA_VXLAN_PORT_RANGE]) {
3271                 if (!changelink) {
3272                         const struct ifla_vxlan_port_range *p
3273                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3274                         conf->port_min = ntohs(p->low);
3275                         conf->port_max = ntohs(p->high);
3276                 } else {
3277                         return -EOPNOTSUPP;
3278                 }
3279         }
3280
3281         if (data[IFLA_VXLAN_PORT]) {
3282                 if (changelink)
3283                         return -EOPNOTSUPP;
3284                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3285         }
3286
3287         if (data[IFLA_VXLAN_UDP_CSUM]) {
3288                 if (changelink)
3289                         return -EOPNOTSUPP;
3290                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3291                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3292         }
3293
3294         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3295                 if (changelink)
3296                         return -EOPNOTSUPP;
3297                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3298                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3299         }
3300
3301         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3302                 if (changelink)
3303                         return -EOPNOTSUPP;
3304                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3305                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3306         }
3307
3308         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3309                 if (changelink)
3310                         return -EOPNOTSUPP;
3311                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3312                         conf->flags |= VXLAN_F_REMCSUM_TX;
3313         }
3314
3315         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3316                 if (changelink)
3317                         return -EOPNOTSUPP;
3318                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3319                         conf->flags |= VXLAN_F_REMCSUM_RX;
3320         }
3321
3322         if (data[IFLA_VXLAN_GBP]) {
3323                 if (changelink)
3324                         return -EOPNOTSUPP;
3325                 conf->flags |= VXLAN_F_GBP;
3326         }
3327
3328         if (data[IFLA_VXLAN_GPE]) {
3329                 if (changelink)
3330                         return -EOPNOTSUPP;
3331                 conf->flags |= VXLAN_F_GPE;
3332         }
3333
3334         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3335                 if (changelink)
3336                         return -EOPNOTSUPP;
3337                 conf->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3338         }
3339
3340         if (tb[IFLA_MTU]) {
3341                 if (changelink)
3342                         return -EOPNOTSUPP;
3343                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3344         }
3345
3346         return 0;
3347 }
3348
3349 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3350                          struct nlattr *tb[], struct nlattr *data[],
3351                          struct netlink_ext_ack *extack)
3352 {
3353         struct vxlan_config conf;
3354         int err;
3355
3356         err = vxlan_nl2conf(tb, data, dev, &conf, false);
3357         if (err)
3358                 return err;
3359
3360         return __vxlan_dev_create(src_net, dev, &conf);
3361 }
3362
3363 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3364                             struct nlattr *data[],
3365                             struct netlink_ext_ack *extack)
3366 {
3367         struct vxlan_dev *vxlan = netdev_priv(dev);
3368         struct vxlan_rdst *dst = &vxlan->default_dst;
3369         struct vxlan_rdst old_dst;
3370         struct vxlan_config conf;
3371         int err;
3372
3373         err = vxlan_nl2conf(tb, data,
3374                             dev, &conf, true);
3375         if (err)
3376                 return err;
3377
3378         memcpy(&old_dst, dst, sizeof(struct vxlan_rdst));
3379
3380         err = vxlan_dev_configure(vxlan->net, dev, &conf, true);
3381         if (err)
3382                 return err;
3383
3384         /* handle default dst entry */
3385         if (!vxlan_addr_equal(&dst->remote_ip, &old_dst.remote_ip)) {
3386                 spin_lock_bh(&vxlan->hash_lock);
3387                 if (!vxlan_addr_any(&old_dst.remote_ip))
3388                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
3389                                            old_dst.remote_ip,
3390                                            vxlan->cfg.dst_port,
3391                                            old_dst.remote_vni,
3392                                            old_dst.remote_vni,
3393                                            old_dst.remote_ifindex, 0);
3394
3395                 if (!vxlan_addr_any(&dst->remote_ip)) {
3396                         err = vxlan_fdb_create(vxlan, all_zeros_mac,
3397                                                &dst->remote_ip,
3398                                                NUD_REACHABLE | NUD_PERMANENT,
3399                                                NLM_F_CREATE | NLM_F_APPEND,
3400                                                vxlan->cfg.dst_port,
3401                                                dst->remote_vni,
3402                                                dst->remote_vni,
3403                                                dst->remote_ifindex,
3404                                                NTF_SELF);
3405                         if (err) {
3406                                 spin_unlock_bh(&vxlan->hash_lock);
3407                                 return err;
3408                         }
3409                 }
3410                 spin_unlock_bh(&vxlan->hash_lock);
3411         }
3412
3413         return 0;
3414 }
3415
3416 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3417 {
3418         struct vxlan_dev *vxlan = netdev_priv(dev);
3419
3420         vxlan_flush(vxlan, true);
3421
3422         gro_cells_destroy(&vxlan->gro_cells);
3423         list_del(&vxlan->next);
3424         unregister_netdevice_queue(dev, head);
3425 }
3426
3427 static size_t vxlan_get_size(const struct net_device *dev)
3428 {
3429
3430         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3431                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3432                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3433                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3434                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3435                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3436                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3437                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3438                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3439                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3440                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3441                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3442                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3443                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3444                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3445                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3446                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3447                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3448                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3449                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3450                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3451                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3452                 0;
3453 }
3454
3455 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3456 {
3457         const struct vxlan_dev *vxlan = netdev_priv(dev);
3458         const struct vxlan_rdst *dst = &vxlan->default_dst;
3459         struct ifla_vxlan_port_range ports = {
3460                 .low =  htons(vxlan->cfg.port_min),
3461                 .high = htons(vxlan->cfg.port_max),
3462         };
3463
3464         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3465                 goto nla_put_failure;
3466
3467         if (!vxlan_addr_any(&dst->remote_ip)) {
3468                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3469                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3470                                             dst->remote_ip.sin.sin_addr.s_addr))
3471                                 goto nla_put_failure;
3472 #if IS_ENABLED(CONFIG_IPV6)
3473                 } else {
3474                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3475                                              &dst->remote_ip.sin6.sin6_addr))
3476                                 goto nla_put_failure;
3477 #endif
3478                 }
3479         }
3480
3481         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3482                 goto nla_put_failure;
3483
3484         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3485                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3486                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3487                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3488                                 goto nla_put_failure;
3489 #if IS_ENABLED(CONFIG_IPV6)
3490                 } else {
3491                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3492                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3493                                 goto nla_put_failure;
3494 #endif
3495                 }
3496         }
3497
3498         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3499             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3500             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3501             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3502                         !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
3503             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3504                         !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
3505             nla_put_u8(skb, IFLA_VXLAN_RSC,
3506                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
3507             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3508                         !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
3509             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3510                         !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
3511             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3512                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
3513             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3514             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3515             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3516             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3517                         !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3518             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3519                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3520             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3521                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3522             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3523                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
3524             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3525                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
3526                 goto nla_put_failure;
3527
3528         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3529                 goto nla_put_failure;
3530
3531         if (vxlan->cfg.flags & VXLAN_F_GBP &&
3532             nla_put_flag(skb, IFLA_VXLAN_GBP))
3533                 goto nla_put_failure;
3534
3535         if (vxlan->cfg.flags & VXLAN_F_GPE &&
3536             nla_put_flag(skb, IFLA_VXLAN_GPE))
3537                 goto nla_put_failure;
3538
3539         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3540             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3541                 goto nla_put_failure;
3542
3543         return 0;
3544
3545 nla_put_failure:
3546         return -EMSGSIZE;
3547 }
3548
3549 static struct net *vxlan_get_link_net(const struct net_device *dev)
3550 {
3551         struct vxlan_dev *vxlan = netdev_priv(dev);
3552
3553         return vxlan->net;
3554 }
3555
3556 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3557         .kind           = "vxlan",
3558         .maxtype        = IFLA_VXLAN_MAX,
3559         .policy         = vxlan_policy,
3560         .priv_size      = sizeof(struct vxlan_dev),
3561         .setup          = vxlan_setup,
3562         .validate       = vxlan_validate,
3563         .newlink        = vxlan_newlink,
3564         .changelink     = vxlan_changelink,
3565         .dellink        = vxlan_dellink,
3566         .get_size       = vxlan_get_size,
3567         .fill_info      = vxlan_fill_info,
3568         .get_link_net   = vxlan_get_link_net,
3569 };
3570
3571 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3572                                     u8 name_assign_type,
3573                                     struct vxlan_config *conf)
3574 {
3575         struct nlattr *tb[IFLA_MAX + 1];
3576         struct net_device *dev;
3577         int err;
3578
3579         memset(&tb, 0, sizeof(tb));
3580
3581         dev = rtnl_create_link(net, name, name_assign_type,
3582                                &vxlan_link_ops, tb);
3583         if (IS_ERR(dev))
3584                 return dev;
3585
3586         err = __vxlan_dev_create(net, dev, conf);
3587         if (err < 0) {
3588                 free_netdev(dev);
3589                 return ERR_PTR(err);
3590         }
3591
3592         err = rtnl_configure_link(dev, NULL);
3593         if (err < 0) {
3594                 LIST_HEAD(list_kill);
3595
3596                 vxlan_dellink(dev, &list_kill);
3597                 unregister_netdevice_many(&list_kill);
3598                 return ERR_PTR(err);
3599         }
3600
3601         return dev;
3602 }
3603 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3604
3605 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3606                                              struct net_device *dev)
3607 {
3608         struct vxlan_dev *vxlan, *next;
3609         LIST_HEAD(list_kill);
3610
3611         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3612                 struct vxlan_rdst *dst = &vxlan->default_dst;
3613
3614                 /* In case we created vxlan device with carrier
3615                  * and we loose the carrier due to module unload
3616                  * we also need to remove vxlan device. In other
3617                  * cases, it's not necessary and remote_ifindex
3618                  * is 0 here, so no matches.
3619                  */
3620                 if (dst->remote_ifindex == dev->ifindex)
3621                         vxlan_dellink(vxlan->dev, &list_kill);
3622         }
3623
3624         unregister_netdevice_many(&list_kill);
3625 }
3626
3627 static int vxlan_netdevice_event(struct notifier_block *unused,
3628                                  unsigned long event, void *ptr)
3629 {
3630         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3631         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3632
3633         if (event == NETDEV_UNREGISTER)
3634                 vxlan_handle_lowerdev_unregister(vn, dev);
3635         else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3636                 vxlan_push_rx_ports(dev);
3637
3638         return NOTIFY_DONE;
3639 }
3640
3641 static struct notifier_block vxlan_notifier_block __read_mostly = {
3642         .notifier_call = vxlan_netdevice_event,
3643 };
3644
3645 static __net_init int vxlan_init_net(struct net *net)
3646 {
3647         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3648         unsigned int h;
3649
3650         INIT_LIST_HEAD(&vn->vxlan_list);
3651         spin_lock_init(&vn->sock_lock);
3652
3653         for (h = 0; h < PORT_HASH_SIZE; ++h)
3654                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3655
3656         return 0;
3657 }
3658
3659 static void __net_exit vxlan_exit_net(struct net *net)
3660 {
3661         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3662         struct vxlan_dev *vxlan, *next;
3663         struct net_device *dev, *aux;
3664         LIST_HEAD(list);
3665
3666         rtnl_lock();
3667         for_each_netdev_safe(net, dev, aux)
3668                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3669                         unregister_netdevice_queue(dev, &list);
3670
3671         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3672                 /* If vxlan->dev is in the same netns, it has already been added
3673                  * to the list by the previous loop.
3674                  */
3675                 if (!net_eq(dev_net(vxlan->dev), net)) {
3676                         gro_cells_destroy(&vxlan->gro_cells);
3677                         unregister_netdevice_queue(vxlan->dev, &list);
3678                 }
3679         }
3680
3681         unregister_netdevice_many(&list);
3682         rtnl_unlock();
3683 }
3684
3685 static struct pernet_operations vxlan_net_ops = {
3686         .init = vxlan_init_net,
3687         .exit = vxlan_exit_net,
3688         .id   = &vxlan_net_id,
3689         .size = sizeof(struct vxlan_net),
3690 };
3691
3692 static int __init vxlan_init_module(void)
3693 {
3694         int rc;
3695
3696         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3697
3698         rc = register_pernet_subsys(&vxlan_net_ops);
3699         if (rc)
3700                 goto out1;
3701
3702         rc = register_netdevice_notifier(&vxlan_notifier_block);
3703         if (rc)
3704                 goto out2;
3705
3706         rc = rtnl_link_register(&vxlan_link_ops);
3707         if (rc)
3708                 goto out3;
3709
3710         return 0;
3711 out3:
3712         unregister_netdevice_notifier(&vxlan_notifier_block);
3713 out2:
3714         unregister_pernet_subsys(&vxlan_net_ops);
3715 out1:
3716         return rc;
3717 }
3718 late_initcall(vxlan_init_module);
3719
3720 static void __exit vxlan_cleanup_module(void)
3721 {
3722         rtnl_link_unregister(&vxlan_link_ops);
3723         unregister_netdevice_notifier(&vxlan_notifier_block);
3724         unregister_pernet_subsys(&vxlan_net_ops);
3725         /* rcu_barrier() is called by netns */
3726 }
3727 module_exit(vxlan_cleanup_module);
3728
3729 MODULE_LICENSE("GPL");
3730 MODULE_VERSION(VXLAN_VERSION);
3731 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3732 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3733 MODULE_ALIAS_RTNL_LINK("vxlan");