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, __be32 vni)
 230 {
 231         struct vxlan_dev *vxlan;
 232
 233         /* For flow based devices, map all packets to VNI 0 */
 234         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 235                 vni = 0;
 236
 237         hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
 238                 if (vxlan->default_dst.remote_vni == vni)
 239                         return vxlan;
 240         }
 241
 242         return NULL;
 243 }
 244
 245 /* Look up VNI in a per net namespace table */
 246 static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
 247                                         sa_family_t family, __be16 port,
 248                                         u32 flags)
 249 {
 250         struct vxlan_sock *vs;
 251
 252         vs = vxlan_find_sock(net, family, port, flags);
 253         if (!vs)
 254                 return NULL;
 255
 256         return vxlan_vs_find_vni(vs, vni);
 257 }
 258
 259 /* Fill in neighbour message in skbuff. */
 260 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 261                           const struct vxlan_fdb *fdb,
 262                           u32 portid, u32 seq, int type, unsigned int flags,
 263                           const struct vxlan_rdst *rdst)
 264 {
 265         unsigned long now = jiffies;
 266         struct nda_cacheinfo ci;
 267         struct nlmsghdr *nlh;
 268         struct ndmsg *ndm;
 269         bool send_ip, send_eth;
 270
 271         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 272         if (nlh == NULL)
 273                 return -EMSGSIZE;
 274
 275         ndm = nlmsg_data(nlh);
 276         memset(ndm, 0, sizeof(*ndm));
 277
 278         send_eth = send_ip = true;
 279
 280         if (type == RTM_GETNEIGH) {
 281                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 282                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 283                 ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
 284         } else
 285                 ndm->ndm_family = AF_BRIDGE;
 286         ndm->ndm_state = fdb->state;
 287         ndm->ndm_ifindex = vxlan->dev->ifindex;
 288         ndm->ndm_flags = fdb->flags;
 289         ndm->ndm_type = RTN_UNICAST;
 290
 291         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 292             nla_put_s32(skb, NDA_LINK_NETNSID,
 293                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 294                 goto nla_put_failure;
 295
 296         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 297                 goto nla_put_failure;
 298
 299         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 300                 goto nla_put_failure;
 301
 302         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 303             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 304                 goto nla_put_failure;
 305         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 306             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 307                 goto nla_put_failure;
 308         if ((vxlan->flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
 309             nla_put_u32(skb, NDA_SRC_VNI,
 310                         be32_to_cpu(fdb->vni)))
 311                 goto nla_put_failure;
 312         if (rdst->remote_ifindex &&
 313             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 314                 goto nla_put_failure;
 315
 316         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 317         ci.ndm_confirmed = 0;
 318         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 319         ci.ndm_refcnt    = 0;
 320
 321         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 322                 goto nla_put_failure;
 323
 324         nlmsg_end(skb, nlh);
 325         return 0;
 326
 327 nla_put_failure:
 328         nlmsg_cancel(skb, nlh);
 329         return -EMSGSIZE;
 330 }
 331
 332 static inline size_t vxlan_nlmsg_size(void)
 333 {
 334         return NLMSG_ALIGN(sizeof(struct ndmsg))
 335                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 336                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 337                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 338                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 339                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 340                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 341                 + nla_total_size(sizeof(struct nda_cacheinfo));
 342 }
 343
 344 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 345                              struct vxlan_rdst *rd, int type)
 346 {
 347         struct net *net = dev_net(vxlan->dev);
 348         struct sk_buff *skb;
 349         int err = -ENOBUFS;
 350
 351         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 352         if (skb == NULL)
 353                 goto errout;
 354
 355         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 356         if (err < 0) {
 357                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 358                 WARN_ON(err == -EMSGSIZE);
 359                 kfree_skb(skb);
 360                 goto errout;
 361         }
 362
 363         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 364         return;
 365 errout:
 366         if (err < 0)
 367                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 368 }
 369
 370 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 371 {
 372         struct vxlan_dev *vxlan = netdev_priv(dev);
 373         struct vxlan_fdb f = {
 374                 .state = NUD_STALE,
 375         };
 376         struct vxlan_rdst remote = {
 377                 .remote_ip = *ipa, /* goes to NDA_DST */
 378                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 379         };
 380
 381         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 382 }
 383
 384 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 385 {
 386         struct vxlan_fdb f = {
 387                 .state = NUD_STALE,
 388         };
 389         struct vxlan_rdst remote = { };
 390
 391         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 392
 393         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 394 }
 395
 396 /* Hash Ethernet address */
 397 static u32 eth_hash(const unsigned char *addr)
 398 {
 399         u64 value = get_unaligned((u64 *)addr);
 400
 401         /* only want 6 bytes */
 402 #ifdef __BIG_ENDIAN
 403         value >>= 16;
 404 #else
 405         value <<= 16;
 406 #endif
 407         return hash_64(value, FDB_HASH_BITS);
 408 }
 409
 410 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
 411 {
 412         /* use 1 byte of OUI and 3 bytes of NIC */
 413         u32 key = get_unaligned((u32 *)(addr + 2));
 414
 415         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
 416 }
 417
 418 /* Hash chain to use given mac address */
 419 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 420                                                 const u8 *mac, __be32 vni)
 421 {
 422         if (vxlan->flags & VXLAN_F_COLLECT_METADATA)
 423                 return &vxlan->fdb_head[eth_vni_hash(mac, vni)];
 424         else
 425                 return &vxlan->fdb_head[eth_hash(mac)];
 426 }
 427
 428 /* Look up Ethernet address in forwarding table */
 429 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 430                                           const u8 *mac, __be32 vni)
 431 {
 432         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 433         struct vxlan_fdb *f;
 434
 435         hlist_for_each_entry_rcu(f, head, hlist) {
 436                 if (ether_addr_equal(mac, f->eth_addr)) {
 437                         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
 438                                 if (vni == f->vni)
 439                                         return f;
 440                         } else {
 441                                 return f;
 442                         }
 443                 }
 444         }
 445
 446         return NULL;
 447 }
 448
 449 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 450                                         const u8 *mac, __be32 vni)
 451 {
 452         struct vxlan_fdb *f;
 453
 454         f = __vxlan_find_mac(vxlan, mac, vni);
 455         if (f)
 456                 f->used = jiffies;
 457
 458         return f;
 459 }
 460
 461 /* caller should hold vxlan->hash_lock */
 462 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 463                                               union vxlan_addr *ip, __be16 port,
 464                                               __be32 vni, __u32 ifindex)
 465 {
 466         struct vxlan_rdst *rd;
 467
 468         list_for_each_entry(rd, &f->remotes, list) {
 469                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 470                     rd->remote_port == port &&
 471                     rd->remote_vni == vni &&
 472                     rd->remote_ifindex == ifindex)
 473                         return rd;
 474         }
 475
 476         return NULL;
 477 }
 478
 479 /* Replace destination of unicast mac */
 480 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 481                              union vxlan_addr *ip, __be16 port, __be32 vni,
 482                              __u32 ifindex)
 483 {
 484         struct vxlan_rdst *rd;
 485
 486         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 487         if (rd)
 488                 return 0;
 489
 490         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 491         if (!rd)
 492                 return 0;
 493
 494         dst_cache_reset(&rd->dst_cache);
 495         rd->remote_ip = *ip;
 496         rd->remote_port = port;
 497         rd->remote_vni = vni;
 498         rd->remote_ifindex = ifindex;
 499         return 1;
 500 }
 501
 502 /* Add/update destinations for multicast */
 503 static int vxlan_fdb_append(struct vxlan_fdb *f,
 504                             union vxlan_addr *ip, __be16 port, __be32 vni,
 505                             __u32 ifindex, struct vxlan_rdst **rdp)
 506 {
 507         struct vxlan_rdst *rd;
 508
 509         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 510         if (rd)
 511                 return 0;
 512
 513         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 514         if (rd == NULL)
 515                 return -ENOBUFS;
 516
 517         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 518                 kfree(rd);
 519                 return -ENOBUFS;
 520         }
 521
 522         rd->remote_ip = *ip;
 523         rd->remote_port = port;
 524         rd->remote_vni = vni;
 525         rd->remote_ifindex = ifindex;
 526
 527         list_add_tail_rcu(&rd->list, &f->remotes);
 528
 529         *rdp = rd;
 530         return 1;
 531 }
 532
 533 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 534                                           unsigned int off,
 535                                           struct vxlanhdr *vh, size_t hdrlen,
 536                                           __be32 vni_field,
 537                                           struct gro_remcsum *grc,
 538                                           bool nopartial)
 539 {
 540         size_t start, offset;
 541
 542         if (skb->remcsum_offload)
 543                 return vh;
 544
 545         if (!NAPI_GRO_CB(skb)->csum_valid)
 546                 return NULL;
 547
 548         start = vxlan_rco_start(vni_field);
 549         offset = start + vxlan_rco_offset(vni_field);
 550
 551         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 552                                      start, offset, grc, nopartial);
 553
 554         skb->remcsum_offload = 1;
 555
 556         return vh;
 557 }
 558
 559 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
 560                                           struct sk_buff **head,
 561                                           struct sk_buff *skb)
 562 {
 563         struct sk_buff *p, **pp = NULL;
 564         struct vxlanhdr *vh, *vh2;
 565         unsigned int hlen, off_vx;
 566         int flush = 1;
 567         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 568         __be32 flags;
 569         struct gro_remcsum grc;
 570
 571         skb_gro_remcsum_init(&grc);
 572
 573         off_vx = skb_gro_offset(skb);
 574         hlen = off_vx + sizeof(*vh);
 575         vh   = skb_gro_header_fast(skb, off_vx);
 576         if (skb_gro_header_hard(skb, hlen)) {
 577                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 578                 if (unlikely(!vh))
 579                         goto out;
 580         }
 581
 582         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 583
 584         flags = vh->vx_flags;
 585
 586         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 587                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 588                                        vh->vx_vni, &grc,
 589                                        !!(vs->flags &
 590                                           VXLAN_F_REMCSUM_NOPARTIAL));
 591
 592                 if (!vh)
 593                         goto out;
 594         }
 595
 596         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 597
 598         for (p = *head; p; p = p->next) {
 599                 if (!NAPI_GRO_CB(p)->same_flow)
 600                         continue;
 601
 602                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 603                 if (vh->vx_flags != vh2->vx_flags ||
 604                     vh->vx_vni != vh2->vx_vni) {
 605                         NAPI_GRO_CB(p)->same_flow = 0;
 606                         continue;
 607                 }
 608         }
 609
 610         pp = call_gro_receive(eth_gro_receive, head, skb);
 611         flush = 0;
 612
 613 out:
 614         skb_gro_remcsum_cleanup(skb, &grc);
 615         NAPI_GRO_CB(skb)->flush |= flush;
 616
 617         return pp;
 618 }
 619
 620 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 621 {
 622         /* Sets 'skb->inner_mac_header' since we are always called with
 623          * 'skb->encapsulation' set.
 624          */
 625         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 626 }
 627
 628 /* Add new entry to forwarding table -- assumes lock held */
 629 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 630                             const u8 *mac, union vxlan_addr *ip,
 631                             __u16 state, __u16 flags,
 632                             __be16 port, __be32 src_vni, __be32 vni,
 633                             __u32 ifindex, __u8 ndm_flags)
 634 {
 635         struct vxlan_rdst *rd = NULL;
 636         struct vxlan_fdb *f;
 637         int notify = 0;
 638         int rc;
 639
 640         f = __vxlan_find_mac(vxlan, mac, src_vni);
 641         if (f) {
 642                 if (flags & NLM_F_EXCL) {
 643                         netdev_dbg(vxlan->dev,
 644                                    "lost race to create %pM\n", mac);
 645                         return -EEXIST;
 646                 }
 647                 if (f->state != state) {
 648                         f->state = state;
 649                         f->updated = jiffies;
 650                         notify = 1;
 651                 }
 652                 if (f->flags != ndm_flags) {
 653                         f->flags = ndm_flags;
 654                         f->updated = jiffies;
 655                         notify = 1;
 656                 }
 657                 if ((flags & NLM_F_REPLACE)) {
 658                         /* Only change unicasts */
 659                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 660                              is_zero_ether_addr(f->eth_addr))) {
 661                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 662                                                            ifindex);
 663                         } else
 664                                 return -EOPNOTSUPP;
 665                 }
 666                 if ((flags & NLM_F_APPEND) &&
 667                     (is_multicast_ether_addr(f->eth_addr) ||
 668                      is_zero_ether_addr(f->eth_addr))) {
 669                         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 670
 671                         if (rc < 0)
 672                                 return rc;
 673                         notify |= rc;
 674                 }
 675         } else {
 676                 if (!(flags & NLM_F_CREATE))
 677                         return -ENOENT;
 678
 679                 if (vxlan->cfg.addrmax &&
 680                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 681                         return -ENOSPC;
 682
 683                 /* Disallow replace to add a multicast entry */
 684                 if ((flags & NLM_F_REPLACE) &&
 685                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 686                         return -EOPNOTSUPP;
 687
 688                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 689                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 690                 if (!f)
 691                         return -ENOMEM;
 692
 693                 notify = 1;
 694                 f->state = state;
 695                 f->flags = ndm_flags;
 696                 f->updated = f->used = jiffies;
 697                 f->vni = src_vni;
 698                 INIT_LIST_HEAD(&f->remotes);
 699                 memcpy(f->eth_addr, mac, ETH_ALEN);
 700
 701                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 702                 if (rc < 0) {
 703                         kfree(f);
 704                         return rc;
 705                 }
 706
 707                 ++vxlan->addrcnt;
 708                 hlist_add_head_rcu(&f->hlist,
 709                                    vxlan_fdb_head(vxlan, mac, src_vni));
 710         }
 711
 712         if (notify) {
 713                 if (rd == NULL)
 714                         rd = first_remote_rtnl(f);
 715                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 716         }
 717
 718         return 0;
 719 }
 720
 721 static void vxlan_fdb_free(struct rcu_head *head)
 722 {
 723         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 724         struct vxlan_rdst *rd, *nd;
 725
 726         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 727                 dst_cache_destroy(&rd->dst_cache);
 728                 kfree(rd);
 729         }
 730         kfree(f);
 731 }
 732
 733 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 734 {
 735         netdev_dbg(vxlan->dev,
 736                     "delete %pM\n", f->eth_addr);
 737
 738         --vxlan->addrcnt;
 739         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 740
 741         hlist_del_rcu(&f->hlist);
 742         call_rcu(&f->rcu, vxlan_fdb_free);
 743 }
 744
 745 static void vxlan_dst_free(struct rcu_head *head)
 746 {
 747         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 748
 749         dst_cache_destroy(&rd->dst_cache);
 750         kfree(rd);
 751 }
 752
 753 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 754                                   struct vxlan_rdst *rd)
 755 {
 756         list_del_rcu(&rd->list);
 757         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 758         call_rcu(&rd->rcu, vxlan_dst_free);
 759 }
 760
 761 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 762                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
 763                            __be32 *vni, u32 *ifindex)
 764 {
 765         struct net *net = dev_net(vxlan->dev);
 766         int err;
 767
 768         if (tb[NDA_DST]) {
 769                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 770                 if (err)
 771                         return err;
 772         } else {
 773                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 774                 if (remote->sa.sa_family == AF_INET) {
 775                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 776                         ip->sa.sa_family = AF_INET;
 777 #if IS_ENABLED(CONFIG_IPV6)
 778                 } else {
 779                         ip->sin6.sin6_addr = in6addr_any;
 780                         ip->sa.sa_family = AF_INET6;
 781 #endif
 782                 }
 783         }
 784
 785         if (tb[NDA_PORT]) {
 786                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 787                         return -EINVAL;
 788                 *port = nla_get_be16(tb[NDA_PORT]);
 789         } else {
 790                 *port = vxlan->cfg.dst_port;
 791         }
 792
 793         if (tb[NDA_VNI]) {
 794                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 795                         return -EINVAL;
 796                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 797         } else {
 798                 *vni = vxlan->default_dst.remote_vni;
 799         }
 800
 801         if (tb[NDA_SRC_VNI]) {
 802                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
 803                         return -EINVAL;
 804                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
 805         } else {
 806                 *src_vni = vxlan->default_dst.remote_vni;
 807         }
 808
 809         if (tb[NDA_IFINDEX]) {
 810                 struct net_device *tdev;
 811
 812                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 813                         return -EINVAL;
 814                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 815                 tdev = __dev_get_by_index(net, *ifindex);
 816                 if (!tdev)
 817                         return -EADDRNOTAVAIL;
 818         } else {
 819                 *ifindex = 0;
 820         }
 821
 822         return 0;
 823 }
 824
 825 /* Add static entry (via netlink) */
 826 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 827                          struct net_device *dev,
 828                          const unsigned char *addr, u16 vid, u16 flags)
 829 {
 830         struct vxlan_dev *vxlan = netdev_priv(dev);
 831         /* struct net *net = dev_net(vxlan->dev); */
 832         union vxlan_addr ip;
 833         __be16 port;
 834         __be32 src_vni, vni;
 835         u32 ifindex;
 836         int err;
 837
 838         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 839                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 840                         ndm->ndm_state);
 841                 return -EINVAL;
 842         }
 843
 844         if (tb[NDA_DST] == NULL)
 845                 return -EINVAL;
 846
 847         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 848         if (err)
 849                 return err;
 850
 851         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 852                 return -EAFNOSUPPORT;
 853
 854         spin_lock_bh(&vxlan->hash_lock);
 855         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 856                                port, src_vni, vni, ifindex, ndm->ndm_flags);
 857         spin_unlock_bh(&vxlan->hash_lock);
 858
 859         return err;
 860 }
 861
 862 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
 863                               const unsigned char *addr, union vxlan_addr ip,
 864                               __be16 port, __be32 src_vni, u32 vni, u32 ifindex,
 865                               u16 vid)
 866 {
 867         struct vxlan_fdb *f;
 868         struct vxlan_rdst *rd = NULL;
 869         int err = -ENOENT;
 870
 871         f = vxlan_find_mac(vxlan, addr, src_vni);
 872         if (!f)
 873                 return err;
 874
 875         if (!vxlan_addr_any(&ip)) {
 876                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 877                 if (!rd)
 878                         goto out;
 879         }
 880
 881         /* remove a destination if it's not the only one on the list,
 882          * otherwise destroy the fdb entry
 883          */
 884         if (rd && !list_is_singular(&f->remotes)) {
 885                 vxlan_fdb_dst_destroy(vxlan, f, rd);
 886                 goto out;
 887         }
 888
 889         vxlan_fdb_destroy(vxlan, f);
 890
 891 out:
 892         return 0;
 893 }
 894
 895 /* Delete entry (via netlink) */
 896 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 897                             struct net_device *dev,
 898                             const unsigned char *addr, u16 vid)
 899 {
 900         struct vxlan_dev *vxlan = netdev_priv(dev);
 901         union vxlan_addr ip;
 902         __be32 src_vni, vni;
 903         __be16 port;
 904         u32 ifindex;
 905         int err;
 906
 907         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
 908         if (err)
 909                 return err;
 910
 911         spin_lock_bh(&vxlan->hash_lock);
 912         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
 913                                  vid);
 914         spin_unlock_bh(&vxlan->hash_lock);
 915
 916         return err;
 917 }
 918
 919 /* Dump forwarding table */
 920 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 921                           struct net_device *dev,
 922                           struct net_device *filter_dev, int *idx)
 923 {
 924         struct vxlan_dev *vxlan = netdev_priv(dev);
 925         unsigned int h;
 926         int err = 0;
 927
 928         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 929                 struct vxlan_fdb *f;
 930
 931                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 932                         struct vxlan_rdst *rd;
 933
 934                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 935                                 if (*idx < cb->args[2])
 936                                         goto skip;
 937
 938                                 err = vxlan_fdb_info(skb, vxlan, f,
 939                                                      NETLINK_CB(cb->skb).portid,
 940                                                      cb->nlh->nlmsg_seq,
 941                                                      RTM_NEWNEIGH,
 942                                                      NLM_F_MULTI, rd);
 943                                 if (err < 0)
 944                                         goto out;
 945 skip:
 946                                 *idx += 1;
 947                         }
 948                 }
 949         }
 950 out:
 951         return err;
 952 }
 953
 954 /* Watch incoming packets to learn mapping between Ethernet address
 955  * and Tunnel endpoint.
 956  * Return true if packet is bogus and should be dropped.
 957  */
 958 static bool vxlan_snoop(struct net_device *dev,
 959                         union vxlan_addr *src_ip, const u8 *src_mac,
 960                         __be32 vni)
 961 {
 962         struct vxlan_dev *vxlan = netdev_priv(dev);
 963         struct vxlan_fdb *f;
 964
 965         f = vxlan_find_mac(vxlan, src_mac, vni);
 966         if (likely(f)) {
 967                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 968
 969                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 970                         return false;
 971
 972                 /* Don't migrate static entries, drop packets */
 973                 if (f->state & NUD_NOARP)
 974                         return true;
 975
 976                 if (net_ratelimit())
 977                         netdev_info(dev,
 978                                     "%pM migrated from %pIS to %pIS\n",
 979                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 980
 981                 rdst->remote_ip = *src_ip;
 982                 f->updated = jiffies;
 983                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 984         } else {
 985                 /* learned new entry */
 986                 spin_lock(&vxlan->hash_lock);
 987
 988                 /* close off race between vxlan_flush and incoming packets */
 989                 if (netif_running(dev))
 990                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 991                                          NUD_REACHABLE,
 992                                          NLM_F_EXCL|NLM_F_CREATE,
 993                                          vxlan->cfg.dst_port,
 994                                          vni,
 995                                          vxlan->default_dst.remote_vni,
 996                                          0, NTF_SELF);
 997                 spin_unlock(&vxlan->hash_lock);
 998         }
 999
1000         return false;
1001 }
1002
1003 /* See if multicast group is already in use by other ID */
1004 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1005 {
1006         struct vxlan_dev *vxlan;
1007         struct vxlan_sock *sock4;
1008 #if IS_ENABLED(CONFIG_IPV6)
1009         struct vxlan_sock *sock6;
1010 #endif
1011         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1012
1013         sock4 = rtnl_dereference(dev->vn4_sock);
1014
1015         /* The vxlan_sock is only used by dev, leaving group has
1016          * no effect on other vxlan devices.
1017          */
1018         if (family == AF_INET && sock4 && atomic_read(&sock4->refcnt) == 1)
1019                 return false;
1020 #if IS_ENABLED(CONFIG_IPV6)
1021         sock6 = rtnl_dereference(dev->vn6_sock);
1022         if (family == AF_INET6 && sock6 && atomic_read(&sock6->refcnt) == 1)
1023                 return false;
1024 #endif
1025
1026         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1027                 if (!netif_running(vxlan->dev) || vxlan == dev)
1028                         continue;
1029
1030                 if (family == AF_INET &&
1031                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1032                         continue;
1033 #if IS_ENABLED(CONFIG_IPV6)
1034                 if (family == AF_INET6 &&
1035                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1036                         continue;
1037 #endif
1038
1039                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1040                                       &dev->default_dst.remote_ip))
1041                         continue;
1042
1043                 if (vxlan->default_dst.remote_ifindex !=
1044                     dev->default_dst.remote_ifindex)
1045                         continue;
1046
1047                 return true;
1048         }
1049
1050         return false;
1051 }
1052
1053 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1054 {
1055         struct vxlan_net *vn;
1056
1057         if (!vs)
1058                 return false;
1059         if (!atomic_dec_and_test(&vs->refcnt))
1060                 return false;
1061
1062         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1063         spin_lock(&vn->sock_lock);
1064         hlist_del_rcu(&vs->hlist);
1065         udp_tunnel_notify_del_rx_port(vs->sock,
1066                                       (vs->flags & VXLAN_F_GPE) ?
1067                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1068                                       UDP_TUNNEL_TYPE_VXLAN);
1069         spin_unlock(&vn->sock_lock);
1070
1071         return true;
1072 }
1073
1074 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1075 {
1076         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1077 #if IS_ENABLED(CONFIG_IPV6)
1078         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1079
1080         rcu_assign_pointer(vxlan->vn6_sock, NULL);
1081 #endif
1082
1083         rcu_assign_pointer(vxlan->vn4_sock, NULL);
1084         synchronize_net();
1085
1086         vxlan_vs_del_dev(vxlan);
1087
1088         if (__vxlan_sock_release_prep(sock4)) {
1089                 udp_tunnel_sock_release(sock4->sock);
1090                 kfree(sock4);
1091         }
1092
1093 #if IS_ENABLED(CONFIG_IPV6)
1094         if (__vxlan_sock_release_prep(sock6)) {
1095                 udp_tunnel_sock_release(sock6->sock);
1096                 kfree(sock6);
1097         }
1098 #endif
1099 }
1100
1101 /* Update multicast group membership when first VNI on
1102  * multicast address is brought up
1103  */
1104 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1105 {
1106         struct sock *sk;
1107         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1108         int ifindex = vxlan->default_dst.remote_ifindex;
1109         int ret = -EINVAL;
1110
1111         if (ip->sa.sa_family == AF_INET) {
1112                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1113                 struct ip_mreqn mreq = {
1114                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1115                         .imr_ifindex            = ifindex,
1116                 };
1117
1118                 sk = sock4->sock->sk;
1119                 lock_sock(sk);
1120                 ret = ip_mc_join_group(sk, &mreq);
1121                 release_sock(sk);
1122 #if IS_ENABLED(CONFIG_IPV6)
1123         } else {
1124                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1125
1126                 sk = sock6->sock->sk;
1127                 lock_sock(sk);
1128                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1129                                                    &ip->sin6.sin6_addr);
1130                 release_sock(sk);
1131 #endif
1132         }
1133
1134         return ret;
1135 }
1136
1137 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1138 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1139 {
1140         struct sock *sk;
1141         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1142         int ifindex = vxlan->default_dst.remote_ifindex;
1143         int ret = -EINVAL;
1144
1145         if (ip->sa.sa_family == AF_INET) {
1146                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1147                 struct ip_mreqn mreq = {
1148                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1149                         .imr_ifindex            = ifindex,
1150                 };
1151
1152                 sk = sock4->sock->sk;
1153                 lock_sock(sk);
1154                 ret = ip_mc_leave_group(sk, &mreq);
1155                 release_sock(sk);
1156 #if IS_ENABLED(CONFIG_IPV6)
1157         } else {
1158                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1159
1160                 sk = sock6->sock->sk;
1161                 lock_sock(sk);
1162                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1163                                                    &ip->sin6.sin6_addr);
1164                 release_sock(sk);
1165 #endif
1166         }
1167
1168         return ret;
1169 }
1170
1171 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1172                           struct sk_buff *skb, u32 vxflags)
1173 {
1174         size_t start, offset;
1175
1176         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1177                 goto out;
1178
1179         start = vxlan_rco_start(unparsed->vx_vni);
1180         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1181
1182         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1183                 return false;
1184
1185         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1186                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1187 out:
1188         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1189         unparsed->vx_vni &= VXLAN_VNI_MASK;
1190         return true;
1191 }
1192
1193 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1194                                 struct sk_buff *skb, u32 vxflags,
1195                                 struct vxlan_metadata *md)
1196 {
1197         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1198         struct metadata_dst *tun_dst;
1199
1200         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1201                 goto out;
1202
1203         md->gbp = ntohs(gbp->policy_id);
1204
1205         tun_dst = (struct metadata_dst *)skb_dst(skb);
1206         if (tun_dst) {
1207                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1208                 tun_dst->u.tun_info.options_len = sizeof(*md);
1209         }
1210         if (gbp->dont_learn)
1211                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1212
1213         if (gbp->policy_applied)
1214                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1215
1216         /* In flow-based mode, GBP is carried in dst_metadata */
1217         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1218                 skb->mark = md->gbp;
1219 out:
1220         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1221 }
1222
1223 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1224                                 __be16 *protocol,
1225                                 struct sk_buff *skb, u32 vxflags)
1226 {
1227         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1228
1229         /* Need to have Next Protocol set for interfaces in GPE mode. */
1230         if (!gpe->np_applied)
1231                 return false;
1232         /* "The initial version is 0. If a receiver does not support the
1233          * version indicated it MUST drop the packet.
1234          */
1235         if (gpe->version != 0)
1236                 return false;
1237         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1238          * processing MUST occur." However, we don't implement OAM
1239          * processing, thus drop the packet.
1240          */
1241         if (gpe->oam_flag)
1242                 return false;
1243
1244         switch (gpe->next_protocol) {
1245         case VXLAN_GPE_NP_IPV4:
1246                 *protocol = htons(ETH_P_IP);
1247                 break;
1248         case VXLAN_GPE_NP_IPV6:
1249                 *protocol = htons(ETH_P_IPV6);
1250                 break;
1251         case VXLAN_GPE_NP_ETHERNET:
1252                 *protocol = htons(ETH_P_TEB);
1253                 break;
1254         default:
1255                 return false;
1256         }
1257
1258         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1259         return true;
1260 }
1261
1262 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1263                           struct vxlan_sock *vs,
1264                           struct sk_buff *skb, __be32 vni)
1265 {
1266         union vxlan_addr saddr;
1267
1268         skb_reset_mac_header(skb);
1269         skb->protocol = eth_type_trans(skb, vxlan->dev);
1270         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1271
1272         /* Ignore packet loops (and multicast echo) */
1273         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1274                 return false;
1275
1276         /* Get address from the outer IP header */
1277         if (vxlan_get_sk_family(vs) == AF_INET) {
1278                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1279                 saddr.sa.sa_family = AF_INET;
1280 #if IS_ENABLED(CONFIG_IPV6)
1281         } else {
1282                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1283                 saddr.sa.sa_family = AF_INET6;
1284 #endif
1285         }
1286
1287         if ((vxlan->flags & VXLAN_F_LEARN) &&
1288             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, vni))
1289                 return false;
1290
1291         return true;
1292 }
1293
1294 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1295                                   struct sk_buff *skb)
1296 {
1297         int err = 0;
1298
1299         if (vxlan_get_sk_family(vs) == AF_INET)
1300                 err = IP_ECN_decapsulate(oiph, skb);
1301 #if IS_ENABLED(CONFIG_IPV6)
1302         else
1303                 err = IP6_ECN_decapsulate(oiph, skb);
1304 #endif
1305
1306         if (unlikely(err) && log_ecn_error) {
1307                 if (vxlan_get_sk_family(vs) == AF_INET)
1308                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1309                                              &((struct iphdr *)oiph)->saddr,
1310                                              ((struct iphdr *)oiph)->tos);
1311                 else
1312                         net_info_ratelimited("non-ECT from %pI6\n",
1313                                              &((struct ipv6hdr *)oiph)->saddr);
1314         }
1315         return err <= 1;
1316 }
1317
1318 /* Callback from net/ipv4/udp.c to receive packets */
1319 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1320 {
1321         struct pcpu_sw_netstats *stats;
1322         struct vxlan_dev *vxlan;
1323         struct vxlan_sock *vs;
1324         struct vxlanhdr unparsed;
1325         struct vxlan_metadata _md;
1326         struct vxlan_metadata *md = &_md;
1327         __be16 protocol = htons(ETH_P_TEB);
1328         bool raw_proto = false;
1329         void *oiph;
1330         __be32 vni = 0;
1331
1332         /* Need UDP and VXLAN header to be present */
1333         if (!pskb_may_pull(skb, VXLAN_HLEN))
1334                 goto drop;
1335
1336         unparsed = *vxlan_hdr(skb);
1337         /* VNI flag always required to be set */
1338         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1339                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1340                            ntohl(vxlan_hdr(skb)->vx_flags),
1341                            ntohl(vxlan_hdr(skb)->vx_vni));
1342                 /* Return non vxlan pkt */
1343                 goto drop;
1344         }
1345         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1346         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1347
1348         vs = rcu_dereference_sk_user_data(sk);
1349         if (!vs)
1350                 goto drop;
1351
1352         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1353
1354         vxlan = vxlan_vs_find_vni(vs, vni);
1355         if (!vxlan)
1356                 goto drop;
1357
1358         /* For backwards compatibility, only allow reserved fields to be
1359          * used by VXLAN extensions if explicitly requested.
1360          */
1361         if (vs->flags & VXLAN_F_GPE) {
1362                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1363                         goto drop;
1364                 raw_proto = true;
1365         }
1366
1367         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1368                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1369                         goto drop;
1370
1371         if (vxlan_collect_metadata(vs)) {
1372                 struct metadata_dst *tun_dst;
1373
1374                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1375                                          key32_to_tunnel_id(vni), sizeof(*md));
1376
1377                 if (!tun_dst)
1378                         goto drop;
1379
1380                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1381
1382                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1383         } else {
1384                 memset(md, 0, sizeof(*md));
1385         }
1386
1387         if (vs->flags & VXLAN_F_REMCSUM_RX)
1388                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1389                         goto drop;
1390         if (vs->flags & VXLAN_F_GBP)
1391                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1392         /* Note that GBP and GPE can never be active together. This is
1393          * ensured in vxlan_dev_configure.
1394          */
1395
1396         if (unparsed.vx_flags || unparsed.vx_vni) {
1397                 /* If there are any unprocessed flags remaining treat
1398                  * this as a malformed packet. This behavior diverges from
1399                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1400                  * in reserved fields are to be ignored. The approach here
1401                  * maintains compatibility with previous stack code, and also
1402                  * is more robust and provides a little more security in
1403                  * adding extensions to VXLAN.
1404                  */
1405                 goto drop;
1406         }
1407
1408         if (!raw_proto) {
1409                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1410                         goto drop;
1411         } else {
1412                 skb_reset_mac_header(skb);
1413                 skb->dev = vxlan->dev;
1414                 skb->pkt_type = PACKET_HOST;
1415         }
1416
1417         oiph = skb_network_header(skb);
1418         skb_reset_network_header(skb);
1419
1420         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1421                 ++vxlan->dev->stats.rx_frame_errors;
1422                 ++vxlan->dev->stats.rx_errors;
1423                 goto drop;
1424         }
1425
1426         stats = this_cpu_ptr(vxlan->dev->tstats);
1427         u64_stats_update_begin(&stats->syncp);
1428         stats->rx_packets++;
1429         stats->rx_bytes += skb->len;
1430         u64_stats_update_end(&stats->syncp);
1431
1432         gro_cells_receive(&vxlan->gro_cells, skb);
1433         return 0;
1434
1435 drop:
1436         /* Consume bad packet */
1437         kfree_skb(skb);
1438         return 0;
1439 }
1440
1441 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1442 {
1443         struct vxlan_dev *vxlan = netdev_priv(dev);
1444         struct arphdr *parp;
1445         u8 *arpptr, *sha;
1446         __be32 sip, tip;
1447         struct neighbour *n;
1448
1449         if (dev->flags & IFF_NOARP)
1450                 goto out;
1451
1452         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1453                 dev->stats.tx_dropped++;
1454                 goto out;
1455         }
1456         parp = arp_hdr(skb);
1457
1458         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1459              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1460             parp->ar_pro != htons(ETH_P_IP) ||
1461             parp->ar_op != htons(ARPOP_REQUEST) ||
1462             parp->ar_hln != dev->addr_len ||
1463             parp->ar_pln != 4)
1464                 goto out;
1465         arpptr = (u8 *)parp + sizeof(struct arphdr);
1466         sha = arpptr;
1467         arpptr += dev->addr_len;        /* sha */
1468         memcpy(&sip, arpptr, sizeof(sip));
1469         arpptr += sizeof(sip);
1470         arpptr += dev->addr_len;        /* tha */
1471         memcpy(&tip, arpptr, sizeof(tip));
1472
1473         if (ipv4_is_loopback(tip) ||
1474             ipv4_is_multicast(tip))
1475                 goto out;
1476
1477         n = neigh_lookup(&arp_tbl, &tip, dev);
1478
1479         if (n) {
1480                 struct vxlan_fdb *f;
1481                 struct sk_buff  *reply;
1482
1483                 if (!(n->nud_state & NUD_CONNECTED)) {
1484                         neigh_release(n);
1485                         goto out;
1486                 }
1487
1488                 f = vxlan_find_mac(vxlan, n->ha, vni);
1489                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1490                         /* bridge-local neighbor */
1491                         neigh_release(n);
1492                         goto out;
1493                 }
1494
1495                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1496                                 n->ha, sha);
1497
1498                 neigh_release(n);
1499
1500                 if (reply == NULL)
1501                         goto out;
1502
1503                 skb_reset_mac_header(reply);
1504                 __skb_pull(reply, skb_network_offset(reply));
1505                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1506                 reply->pkt_type = PACKET_HOST;
1507
1508                 if (netif_rx_ni(reply) == NET_RX_DROP)
1509                         dev->stats.rx_dropped++;
1510         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1511                 union vxlan_addr ipa = {
1512                         .sin.sin_addr.s_addr = tip,
1513                         .sin.sin_family = AF_INET,
1514                 };
1515
1516                 vxlan_ip_miss(dev, &ipa);
1517         }
1518 out:
1519         consume_skb(skb);
1520         return NETDEV_TX_OK;
1521 }
1522
1523 #if IS_ENABLED(CONFIG_IPV6)
1524 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1525         struct neighbour *n, bool isrouter)
1526 {
1527         struct net_device *dev = request->dev;
1528         struct sk_buff *reply;
1529         struct nd_msg *ns, *na;
1530         struct ipv6hdr *pip6;
1531         u8 *daddr;
1532         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1533         int ns_olen;
1534         int i, len;
1535
1536         if (dev == NULL || !pskb_may_pull(request, request->len))
1537                 return NULL;
1538
1539         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1540                 sizeof(*na) + na_olen + dev->needed_tailroom;
1541         reply = alloc_skb(len, GFP_ATOMIC);
1542         if (reply == NULL)
1543                 return NULL;
1544
1545         reply->protocol = htons(ETH_P_IPV6);
1546         reply->dev = dev;
1547         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1548         skb_push(reply, sizeof(struct ethhdr));
1549         skb_reset_mac_header(reply);
1550
1551         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1552
1553         daddr = eth_hdr(request)->h_source;
1554         ns_olen = request->len - skb_network_offset(request) -
1555                 sizeof(struct ipv6hdr) - sizeof(*ns);
1556         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1557                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1558                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1559                         break;
1560                 }
1561         }
1562
1563         /* Ethernet header */
1564         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1565         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1566         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1567         reply->protocol = htons(ETH_P_IPV6);
1568
1569         skb_pull(reply, sizeof(struct ethhdr));
1570         skb_reset_network_header(reply);
1571         skb_put(reply, sizeof(struct ipv6hdr));
1572
1573         /* IPv6 header */
1574
1575         pip6 = ipv6_hdr(reply);
1576         memset(pip6, 0, sizeof(struct ipv6hdr));
1577         pip6->version = 6;
1578         pip6->priority = ipv6_hdr(request)->priority;
1579         pip6->nexthdr = IPPROTO_ICMPV6;
1580         pip6->hop_limit = 255;
1581         pip6->daddr = ipv6_hdr(request)->saddr;
1582         pip6->saddr = *(struct in6_addr *)n->primary_key;
1583
1584         skb_pull(reply, sizeof(struct ipv6hdr));
1585         skb_reset_transport_header(reply);
1586
1587         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1588
1589         /* Neighbor Advertisement */
1590         memset(na, 0, sizeof(*na)+na_olen);
1591         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1592         na->icmph.icmp6_router = isrouter;
1593         na->icmph.icmp6_override = 1;
1594         na->icmph.icmp6_solicited = 1;
1595         na->target = ns->target;
1596         ether_addr_copy(&na->opt[2], n->ha);
1597         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1598         na->opt[1] = na_olen >> 3;
1599
1600         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1601                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1602                 csum_partial(na, sizeof(*na)+na_olen, 0));
1603
1604         pip6->payload_len = htons(sizeof(*na)+na_olen);
1605
1606         skb_push(reply, sizeof(struct ipv6hdr));
1607
1608         reply->ip_summed = CHECKSUM_UNNECESSARY;
1609
1610         return reply;
1611 }
1612
1613 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1614 {
1615         struct vxlan_dev *vxlan = netdev_priv(dev);
1616         struct nd_msg *msg;
1617         const struct ipv6hdr *iphdr;
1618         const struct in6_addr *daddr;
1619         struct neighbour *n;
1620         struct inet6_dev *in6_dev;
1621
1622         in6_dev = __in6_dev_get(dev);
1623         if (!in6_dev)
1624                 goto out;
1625
1626         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr) + sizeof(struct nd_msg)))
1627                 goto out;
1628
1629         iphdr = ipv6_hdr(skb);
1630         daddr = &iphdr->daddr;
1631
1632         msg = (struct nd_msg *)(iphdr + 1);
1633         if (msg->icmph.icmp6_code != 0 ||
1634             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1635                 goto out;
1636
1637         if (ipv6_addr_loopback(daddr) ||
1638             ipv6_addr_is_multicast(&msg->target))
1639                 goto out;
1640
1641         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1642
1643         if (n) {
1644                 struct vxlan_fdb *f;
1645                 struct sk_buff *reply;
1646
1647                 if (!(n->nud_state & NUD_CONNECTED)) {
1648                         neigh_release(n);
1649                         goto out;
1650                 }
1651
1652                 f = vxlan_find_mac(vxlan, n->ha, vni);
1653                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1654                         /* bridge-local neighbor */
1655                         neigh_release(n);
1656                         goto out;
1657                 }
1658
1659                 reply = vxlan_na_create(skb, n,
1660                                         !!(f ? f->flags & NTF_ROUTER : 0));
1661
1662                 neigh_release(n);
1663
1664                 if (reply == NULL)
1665                         goto out;
1666
1667                 if (netif_rx_ni(reply) == NET_RX_DROP)
1668                         dev->stats.rx_dropped++;
1669
1670         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1671                 union vxlan_addr ipa = {
1672                         .sin6.sin6_addr = msg->target,
1673                         .sin6.sin6_family = AF_INET6,
1674                 };
1675
1676                 vxlan_ip_miss(dev, &ipa);
1677         }
1678
1679 out:
1680         consume_skb(skb);
1681         return NETDEV_TX_OK;
1682 }
1683 #endif
1684
1685 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1686 {
1687         struct vxlan_dev *vxlan = netdev_priv(dev);
1688         struct neighbour *n;
1689
1690         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1691                 return false;
1692
1693         n = NULL;
1694         switch (ntohs(eth_hdr(skb)->h_proto)) {
1695         case ETH_P_IP:
1696         {
1697                 struct iphdr *pip;
1698
1699                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1700                         return false;
1701                 pip = ip_hdr(skb);
1702                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1703                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1704                         union vxlan_addr ipa = {
1705                                 .sin.sin_addr.s_addr = pip->daddr,
1706                                 .sin.sin_family = AF_INET,
1707                         };
1708
1709                         vxlan_ip_miss(dev, &ipa);
1710                         return false;
1711                 }
1712
1713                 break;
1714         }
1715 #if IS_ENABLED(CONFIG_IPV6)
1716         case ETH_P_IPV6:
1717         {
1718                 struct ipv6hdr *pip6;
1719
1720                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1721                         return false;
1722                 pip6 = ipv6_hdr(skb);
1723                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1724                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1725                         union vxlan_addr ipa = {
1726                                 .sin6.sin6_addr = pip6->daddr,
1727                                 .sin6.sin6_family = AF_INET6,
1728                         };
1729
1730                         vxlan_ip_miss(dev, &ipa);
1731                         return false;
1732                 }
1733
1734                 break;
1735         }
1736 #endif
1737         default:
1738                 return false;
1739         }
1740
1741         if (n) {
1742                 bool diff;
1743
1744                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1745                 if (diff) {
1746                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1747                                 dev->addr_len);
1748                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1749                 }
1750                 neigh_release(n);
1751                 return diff;
1752         }
1753
1754         return false;
1755 }
1756
1757 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1758                                 struct vxlan_metadata *md)
1759 {
1760         struct vxlanhdr_gbp *gbp;
1761
1762         if (!md->gbp)
1763                 return;
1764
1765         gbp = (struct vxlanhdr_gbp *)vxh;
1766         vxh->vx_flags |= VXLAN_HF_GBP;
1767
1768         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1769                 gbp->dont_learn = 1;
1770
1771         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1772                 gbp->policy_applied = 1;
1773
1774         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1775 }
1776
1777 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1778                                __be16 protocol)
1779 {
1780         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1781
1782         gpe->np_applied = 1;
1783
1784         switch (protocol) {
1785         case htons(ETH_P_IP):
1786                 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1787                 return 0;
1788         case htons(ETH_P_IPV6):
1789                 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1790                 return 0;
1791         case htons(ETH_P_TEB):
1792                 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1793                 return 0;
1794         }
1795         return -EPFNOSUPPORT;
1796 }
1797
1798 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1799                            int iphdr_len, __be32 vni,
1800                            struct vxlan_metadata *md, u32 vxflags,
1801                            bool udp_sum)
1802 {
1803         struct vxlanhdr *vxh;
1804         int min_headroom;
1805         int err;
1806         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1807         __be16 inner_protocol = htons(ETH_P_TEB);
1808
1809         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1810             skb->ip_summed == CHECKSUM_PARTIAL) {
1811                 int csum_start = skb_checksum_start_offset(skb);
1812
1813                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1814                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1815                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1816                      skb->csum_offset == offsetof(struct tcphdr, check)))
1817                         type |= SKB_GSO_TUNNEL_REMCSUM;
1818         }
1819
1820         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1821                         + VXLAN_HLEN + iphdr_len;
1822
1823         /* Need space for new headers (invalidates iph ptr) */
1824         err = skb_cow_head(skb, min_headroom);
1825         if (unlikely(err))
1826                 return err;
1827
1828         err = iptunnel_handle_offloads(skb, type);
1829         if (err)
1830                 return err;
1831
1832         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1833         vxh->vx_flags = VXLAN_HF_VNI;
1834         vxh->vx_vni = vxlan_vni_field(vni);
1835
1836         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1837                 unsigned int start;
1838
1839                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1840                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1841                 vxh->vx_flags |= VXLAN_HF_RCO;
1842
1843                 if (!skb_is_gso(skb)) {
1844                         skb->ip_summed = CHECKSUM_NONE;
1845                         skb->encapsulation = 0;
1846                 }
1847         }
1848
1849         if (vxflags & VXLAN_F_GBP)
1850                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1851         if (vxflags & VXLAN_F_GPE) {
1852                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1853                 if (err < 0)
1854                         return err;
1855                 inner_protocol = skb->protocol;
1856         }
1857
1858         skb_set_inner_protocol(skb, inner_protocol);
1859         return 0;
1860 }
1861
1862 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1863                                       struct vxlan_sock *sock4,
1864                                       struct sk_buff *skb, int oif, u8 tos,
1865                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
1866                                       struct dst_cache *dst_cache,
1867                                       const struct ip_tunnel_info *info)
1868 {
1869         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1870         struct rtable *rt = NULL;
1871         struct flowi4 fl4;
1872
1873         if (!sock4)
1874                 return ERR_PTR(-EIO);
1875
1876         if (tos && !info)
1877                 use_cache = false;
1878         if (use_cache) {
1879                 rt = dst_cache_get_ip4(dst_cache, saddr);
1880                 if (rt)
1881                         return rt;
1882         }
1883
1884         memset(&fl4, 0, sizeof(fl4));
1885         fl4.flowi4_oif = oif;
1886         fl4.flowi4_tos = RT_TOS(tos);
1887         fl4.flowi4_mark = skb->mark;
1888         fl4.flowi4_proto = IPPROTO_UDP;
1889         fl4.daddr = daddr;
1890         fl4.saddr = *saddr;
1891         fl4.fl4_dport = dport;
1892         fl4.fl4_sport = sport;
1893
1894         rt = ip_route_output_key(vxlan->net, &fl4);
1895         if (likely(!IS_ERR(rt))) {
1896                 if (rt->dst.dev == dev) {
1897                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1898                         ip_rt_put(rt);
1899                         return ERR_PTR(-ELOOP);
1900                 }
1901
1902                 *saddr = fl4.saddr;
1903                 if (use_cache)
1904                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1905         } else {
1906                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1907                 return ERR_PTR(-ENETUNREACH);
1908         }
1909         return rt;
1910 }
1911
1912 #if IS_ENABLED(CONFIG_IPV6)
1913 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1914                                           struct net_device *dev,
1915                                           struct vxlan_sock *sock6,
1916                                           struct sk_buff *skb, int oif, u8 tos,
1917                                           __be32 label,
1918                                           const struct in6_addr *daddr,
1919                                           struct in6_addr *saddr,
1920                                           __be16 dport, __be16 sport,
1921                                           struct dst_cache *dst_cache,
1922                                           const struct ip_tunnel_info *info)
1923 {
1924         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1925         struct dst_entry *ndst;
1926         struct flowi6 fl6;
1927         int err;
1928
1929         if (!sock6)
1930                 return ERR_PTR(-EIO);
1931
1932         if (tos && !info)
1933                 use_cache = false;
1934         if (use_cache) {
1935                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1936                 if (ndst)
1937                         return ndst;
1938         }
1939
1940         memset(&fl6, 0, sizeof(fl6));
1941         fl6.flowi6_oif = oif;
1942         fl6.daddr = *daddr;
1943         fl6.saddr = *saddr;
1944         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1945         fl6.flowi6_mark = skb->mark;
1946         fl6.flowi6_proto = IPPROTO_UDP;
1947         fl6.fl6_dport = dport;
1948         fl6.fl6_sport = sport;
1949
1950         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1951                                          sock6->sock->sk,
1952                                          &ndst, &fl6);
1953         if (unlikely(err < 0)) {
1954                 netdev_dbg(dev, "no route to %pI6\n", daddr);
1955                 return ERR_PTR(-ENETUNREACH);
1956         }
1957
1958         if (unlikely(ndst->dev == dev)) {
1959                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
1960                 dst_release(ndst);
1961                 return ERR_PTR(-ELOOP);
1962         }
1963
1964         *saddr = fl6.saddr;
1965         if (use_cache)
1966                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1967         return ndst;
1968 }
1969 #endif
1970
1971 /* Bypass encapsulation if the destination is local */
1972 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1973                                struct vxlan_dev *dst_vxlan, __be32 vni)
1974 {
1975         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1976         union vxlan_addr loopback;
1977         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1978         struct net_device *dev = skb->dev;
1979         int len = skb->len;
1980
1981         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1982         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1983         skb->pkt_type = PACKET_HOST;
1984         skb->encapsulation = 0;
1985         skb->dev = dst_vxlan->dev;
1986         __skb_pull(skb, skb_network_offset(skb));
1987
1988         if (remote_ip->sa.sa_family == AF_INET) {
1989                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1990                 loopback.sa.sa_family =  AF_INET;
1991 #if IS_ENABLED(CONFIG_IPV6)
1992         } else {
1993                 loopback.sin6.sin6_addr = in6addr_loopback;
1994                 loopback.sa.sa_family =  AF_INET6;
1995 #endif
1996         }
1997
1998         if (dst_vxlan->flags & VXLAN_F_LEARN)
1999                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source, vni);
2000
2001         u64_stats_update_begin(&tx_stats->syncp);
2002         tx_stats->tx_packets++;
2003         tx_stats->tx_bytes += len;
2004         u64_stats_update_end(&tx_stats->syncp);
2005
2006         if (netif_rx(skb) == NET_RX_SUCCESS) {
2007                 u64_stats_update_begin(&rx_stats->syncp);
2008                 rx_stats->rx_packets++;
2009                 rx_stats->rx_bytes += len;
2010                 u64_stats_update_end(&rx_stats->syncp);
2011         } else {
2012                 dev->stats.rx_dropped++;
2013         }
2014 }
2015
2016 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2017                                  struct vxlan_dev *vxlan, union vxlan_addr *daddr,
2018                                  __be16 dst_port, __be32 vni, struct dst_entry *dst,
2019                                  u32 rt_flags)
2020 {
2021 #if IS_ENABLED(CONFIG_IPV6)
2022         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2023          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2024          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2025          */
2026         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2027 #endif
2028         /* Bypass encapsulation if the destination is local */
2029         if (rt_flags & RTCF_LOCAL &&
2030             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2031                 struct vxlan_dev *dst_vxlan;
2032
2033                 dst_release(dst);
2034                 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2035                                            daddr->sa.sa_family, dst_port,
2036                                            vxlan->flags);
2037                 if (!dst_vxlan) {
2038                         dev->stats.tx_errors++;
2039                         kfree_skb(skb);
2040
2041                         return -ENOENT;
2042                 }
2043                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2044                 return 1;
2045         }
2046
2047         return 0;
2048 }
2049
2050 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2051                            __be32 default_vni, struct vxlan_rdst *rdst,
2052                            bool did_rsc)
2053 {
2054         struct dst_cache *dst_cache;
2055         struct ip_tunnel_info *info;
2056         struct vxlan_dev *vxlan = netdev_priv(dev);
2057         const struct iphdr *old_iph = ip_hdr(skb);
2058         union vxlan_addr *dst;
2059         union vxlan_addr remote_ip, local_ip;
2060         struct vxlan_metadata _md;
2061         struct vxlan_metadata *md = &_md;
2062         __be16 src_port = 0, dst_port;
2063         struct dst_entry *ndst = NULL;
2064         __be32 vni, label;
2065         __u8 tos, ttl;
2066         int err;
2067         u32 flags = vxlan->flags;
2068         bool udp_sum = false;
2069         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2070
2071         info = skb_tunnel_info(skb);
2072
2073         if (rdst) {
2074                 dst = &rdst->remote_ip;
2075                 if (vxlan_addr_any(dst)) {
2076                         if (did_rsc) {
2077                                 /* short-circuited back to local bridge */
2078                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2079                                 return;
2080                         }
2081                         goto drop;
2082                 }
2083
2084                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2085                 vni = (rdst->remote_vni) ? : default_vni;
2086                 local_ip = vxlan->cfg.saddr;
2087                 dst_cache = &rdst->dst_cache;
2088                 md->gbp = skb->mark;
2089                 ttl = vxlan->cfg.ttl;
2090                 if (!ttl && vxlan_addr_multicast(dst))
2091                         ttl = 1;
2092
2093                 tos = vxlan->cfg.tos;
2094                 if (tos == 1)
2095                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2096
2097                 if (dst->sa.sa_family == AF_INET)
2098                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2099                 else
2100                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2101                 label = vxlan->cfg.label;
2102         } else {
2103                 if (!info) {
2104                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2105                                   dev->name);
2106                         goto drop;
2107                 }
2108                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2109                 if (remote_ip.sa.sa_family == AF_INET) {
2110                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2111                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2112                 } else {
2113                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2114                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2115                 }
2116                 dst = &remote_ip;
2117                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2118                 vni = tunnel_id_to_key32(info->key.tun_id);
2119                 dst_cache = &info->dst_cache;
2120                 if (info->options_len)
2121                         md = ip_tunnel_info_opts(info);
2122                 ttl = info->key.ttl;
2123                 tos = info->key.tos;
2124                 label = info->key.label;
2125                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2126         }
2127         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2128                                      vxlan->cfg.port_max, true);
2129
2130         rcu_read_lock();
2131         if (dst->sa.sa_family == AF_INET) {
2132                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2133                 struct rtable *rt;
2134                 __be16 df = 0;
2135
2136                 rt = vxlan_get_route(vxlan, dev, sock4, skb,
2137                                      rdst ? rdst->remote_ifindex : 0, tos,
2138                                      dst->sin.sin_addr.s_addr,
2139                                      &local_ip.sin.sin_addr.s_addr,
2140                                      dst_port, src_port,
2141                                      dst_cache, info);
2142                 if (IS_ERR(rt)) {
2143                         err = PTR_ERR(rt);
2144                         goto tx_error;
2145                 }
2146
2147                 /* Bypass encapsulation if the destination is local */
2148                 if (!info) {
2149                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2150                                                     dst_port, vni, &rt->dst,
2151                                                     rt->rt_flags);
2152                         if (err)
2153                                 goto out_unlock;
2154                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2155                         df = htons(IP_DF);
2156                 }
2157
2158                 ndst = &rt->dst;
2159                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2160                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2161                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2162                                       vni, md, flags, udp_sum);
2163                 if (err < 0)
2164                         goto tx_error;
2165
2166                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2167                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2168                                     src_port, dst_port, xnet, !udp_sum);
2169 #if IS_ENABLED(CONFIG_IPV6)
2170         } else {
2171                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2172
2173                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb,
2174                                         rdst ? rdst->remote_ifindex : 0, tos,
2175                                         label, &dst->sin6.sin6_addr,
2176                                         &local_ip.sin6.sin6_addr,
2177                                         dst_port, src_port,
2178                                         dst_cache, info);
2179                 if (IS_ERR(ndst)) {
2180                         err = PTR_ERR(ndst);
2181                         ndst = NULL;
2182                         goto tx_error;
2183                 }
2184
2185                 if (!info) {
2186                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2187
2188                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2189                                                     dst_port, vni, ndst,
2190                                                     rt6i_flags);
2191                         if (err)
2192                                 goto out_unlock;
2193                 }
2194
2195                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2196                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2197                 skb_scrub_packet(skb, xnet);
2198                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2199                                       vni, md, flags, udp_sum);
2200                 if (err < 0)
2201                         goto tx_error;
2202
2203                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2204                                      &local_ip.sin6.sin6_addr,
2205                                      &dst->sin6.sin6_addr, tos, ttl,
2206                                      label, src_port, dst_port, !udp_sum);
2207 #endif
2208         }
2209 out_unlock:
2210         rcu_read_unlock();
2211         return;
2212
2213 drop:
2214         dev->stats.tx_dropped++;
2215         dev_kfree_skb(skb);
2216         return;
2217
2218 tx_error:
2219         rcu_read_unlock();
2220         if (err == -ELOOP)
2221                 dev->stats.collisions++;
2222         else if (err == -ENETUNREACH)
2223                 dev->stats.tx_carrier_errors++;
2224         dst_release(ndst);
2225         dev->stats.tx_errors++;
2226         kfree_skb(skb);
2227 }
2228
2229 /* Transmit local packets over Vxlan
2230  *
2231  * Outer IP header inherits ECN and DF from inner header.
2232  * Outer UDP destination is the VXLAN assigned port.
2233  *           source port is based on hash of flow
2234  */
2235 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2236 {
2237         struct vxlan_dev *vxlan = netdev_priv(dev);
2238         const struct ip_tunnel_info *info;
2239         struct ethhdr *eth;
2240         bool did_rsc = false;
2241         struct vxlan_rdst *rdst, *fdst = NULL;
2242         struct vxlan_fdb *f;
2243         __be32 vni = 0;
2244
2245         info = skb_tunnel_info(skb);
2246
2247         skb_reset_mac_header(skb);
2248
2249         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2250                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2251                     info->mode & IP_TUNNEL_INFO_TX) {
2252                         vni = tunnel_id_to_key32(info->key.tun_id);
2253                 } else {
2254                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2255                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2256                         else
2257                                 kfree_skb(skb);
2258                         return NETDEV_TX_OK;
2259                 }
2260         }
2261
2262         if (vxlan->flags & VXLAN_F_PROXY) {
2263                 eth = eth_hdr(skb);
2264                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2265                         return arp_reduce(dev, skb, vni);
2266 #if IS_ENABLED(CONFIG_IPV6)
2267                 else if (ntohs(eth->h_proto) == ETH_P_IPV6) {
2268                         struct ipv6hdr *hdr, _hdr;
2269                         if ((hdr = skb_header_pointer(skb,
2270                                                       skb_network_offset(skb),
2271                                                       sizeof(_hdr), &_hdr)) &&
2272                             hdr->nexthdr == IPPROTO_ICMPV6)
2273                                 return neigh_reduce(dev, skb, vni);
2274                 }
2275 #endif
2276         }
2277
2278         eth = eth_hdr(skb);
2279         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2280         did_rsc = false;
2281
2282         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2283             (ntohs(eth->h_proto) == ETH_P_IP ||
2284              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2285                 did_rsc = route_shortcircuit(dev, skb);
2286                 if (did_rsc)
2287                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2288         }
2289
2290         if (f == NULL) {
2291                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2292                 if (f == NULL) {
2293                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2294                             !is_multicast_ether_addr(eth->h_dest))
2295                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2296
2297                         dev->stats.tx_dropped++;
2298                         kfree_skb(skb);
2299                         return NETDEV_TX_OK;
2300                 }
2301         }
2302
2303         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2304                 struct sk_buff *skb1;
2305
2306                 if (!fdst) {
2307                         fdst = rdst;
2308                         continue;
2309                 }
2310                 skb1 = skb_clone(skb, GFP_ATOMIC);
2311                 if (skb1)
2312                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2313         }
2314
2315         if (fdst)
2316                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2317         else
2318                 kfree_skb(skb);
2319         return NETDEV_TX_OK;
2320 }
2321
2322 /* Walk the forwarding table and purge stale entries */
2323 static void vxlan_cleanup(unsigned long arg)
2324 {
2325         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2326         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2327         unsigned int h;
2328
2329         if (!netif_running(vxlan->dev))
2330                 return;
2331
2332         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2333                 struct hlist_node *p, *n;
2334
2335                 spin_lock_bh(&vxlan->hash_lock);
2336                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2337                         struct vxlan_fdb *f
2338                                 = container_of(p, struct vxlan_fdb, hlist);
2339                         unsigned long timeout;
2340
2341                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2342                                 continue;
2343
2344                         if (f->flags & NTF_EXT_LEARNED)
2345                                 continue;
2346
2347                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2348                         if (time_before_eq(timeout, jiffies)) {
2349                                 netdev_dbg(vxlan->dev,
2350                                            "garbage collect %pM\n",
2351                                            f->eth_addr);
2352                                 f->state = NUD_STALE;
2353                                 vxlan_fdb_destroy(vxlan, f);
2354                         } else if (time_before(timeout, next_timer))
2355                                 next_timer = timeout;
2356                 }
2357                 spin_unlock_bh(&vxlan->hash_lock);
2358         }
2359
2360         mod_timer(&vxlan->age_timer, next_timer);
2361 }
2362
2363 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2364 {
2365         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2366
2367         spin_lock(&vn->sock_lock);
2368         hlist_del_init_rcu(&vxlan->hlist);
2369         spin_unlock(&vn->sock_lock);
2370 }
2371
2372 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2373 {
2374         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2375         __be32 vni = vxlan->default_dst.remote_vni;
2376
2377         spin_lock(&vn->sock_lock);
2378         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2379         spin_unlock(&vn->sock_lock);
2380 }
2381
2382 /* Setup stats when device is created */
2383 static int vxlan_init(struct net_device *dev)
2384 {
2385         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2386         if (!dev->tstats)
2387                 return -ENOMEM;
2388
2389         return 0;
2390 }
2391
2392 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2393 {
2394         struct vxlan_fdb *f;
2395
2396         spin_lock_bh(&vxlan->hash_lock);
2397         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2398         if (f)
2399                 vxlan_fdb_destroy(vxlan, f);
2400         spin_unlock_bh(&vxlan->hash_lock);
2401 }
2402
2403 static void vxlan_uninit(struct net_device *dev)
2404 {
2405         struct vxlan_dev *vxlan = netdev_priv(dev);
2406
2407         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2408
2409         free_percpu(dev->tstats);
2410 }
2411
2412 /* Start ageing timer and join group when device is brought up */
2413 static int vxlan_open(struct net_device *dev)
2414 {
2415         struct vxlan_dev *vxlan = netdev_priv(dev);
2416         int ret;
2417
2418         ret = vxlan_sock_add(vxlan);
2419         if (ret < 0)
2420                 return ret;
2421
2422         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2423                 ret = vxlan_igmp_join(vxlan);
2424                 if (ret == -EADDRINUSE)
2425                         ret = 0;
2426                 if (ret) {
2427                         vxlan_sock_release(vxlan);
2428                         return ret;
2429                 }
2430         }
2431
2432         if (vxlan->cfg.age_interval)
2433                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2434
2435         return ret;
2436 }
2437
2438 /* Purge the forwarding table */
2439 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2440 {
2441         unsigned int h;
2442
2443         spin_lock_bh(&vxlan->hash_lock);
2444         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2445                 struct hlist_node *p, *n;
2446                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2447                         struct vxlan_fdb *f
2448                                 = container_of(p, struct vxlan_fdb, hlist);
2449                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2450                                 continue;
2451                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2452                         if (!is_zero_ether_addr(f->eth_addr))
2453                                 vxlan_fdb_destroy(vxlan, f);
2454                 }
2455         }
2456         spin_unlock_bh(&vxlan->hash_lock);
2457 }
2458
2459 /* Cleanup timer and forwarding table on shutdown */
2460 static int vxlan_stop(struct net_device *dev)
2461 {
2462         struct vxlan_dev *vxlan = netdev_priv(dev);
2463         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2464         int ret = 0;
2465
2466         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2467             !vxlan_group_used(vn, vxlan))
2468                 ret = vxlan_igmp_leave(vxlan);
2469
2470         del_timer_sync(&vxlan->age_timer);
2471
2472         vxlan_flush(vxlan, false);
2473         vxlan_sock_release(vxlan);
2474
2475         return ret;
2476 }
2477
2478 /* Stub, nothing needs to be done. */
2479 static void vxlan_set_multicast_list(struct net_device *dev)
2480 {
2481 }
2482
2483 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2484 {
2485         struct vxlan_dev *vxlan = netdev_priv(dev);
2486         struct vxlan_rdst *dst = &vxlan->default_dst;
2487         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2488                                                          dst->remote_ifindex);
2489         bool use_ipv6 = false;
2490
2491         if (dst->remote_ip.sa.sa_family == AF_INET6)
2492                 use_ipv6 = true;
2493
2494         /* This check is different than dev->max_mtu, because it looks at
2495          * the lowerdev->mtu, rather than the static dev->max_mtu
2496          */
2497         if (lowerdev) {
2498                 int max_mtu = lowerdev->mtu -
2499                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2500                 if (new_mtu > max_mtu)
2501                         return -EINVAL;
2502         }
2503
2504         dev->mtu = new_mtu;
2505         return 0;
2506 }
2507
2508 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2509 {
2510         struct vxlan_dev *vxlan = netdev_priv(dev);
2511         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2512         __be16 sport, dport;
2513
2514         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2515                                   vxlan->cfg.port_max, true);
2516         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2517
2518         if (ip_tunnel_info_af(info) == AF_INET) {
2519                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2520                 struct rtable *rt;
2521
2522                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2523                                      info->key.u.ipv4.dst,
2524                                      &info->key.u.ipv4.src, dport, sport,
2525                                      &info->dst_cache, info);
2526                 if (IS_ERR(rt))
2527                         return PTR_ERR(rt);
2528                 ip_rt_put(rt);
2529         } else {
2530 #if IS_ENABLED(CONFIG_IPV6)
2531                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2532                 struct dst_entry *ndst;
2533
2534                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2535                                         info->key.label, &info->key.u.ipv6.dst,
2536                                         &info->key.u.ipv6.src, dport, sport,
2537                                         &info->dst_cache, info);
2538                 if (IS_ERR(ndst))
2539                         return PTR_ERR(ndst);
2540                 dst_release(ndst);
2541 #else /* !CONFIG_IPV6 */
2542                 return -EPFNOSUPPORT;
2543 #endif
2544         }
2545         info->key.tp_src = sport;
2546         info->key.tp_dst = dport;
2547         return 0;
2548 }
2549
2550 static const struct net_device_ops vxlan_netdev_ether_ops = {
2551         .ndo_init               = vxlan_init,
2552         .ndo_uninit             = vxlan_uninit,
2553         .ndo_open               = vxlan_open,
2554         .ndo_stop               = vxlan_stop,
2555         .ndo_start_xmit         = vxlan_xmit,
2556         .ndo_get_stats64        = ip_tunnel_get_stats64,
2557         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2558         .ndo_change_mtu         = vxlan_change_mtu,
2559         .ndo_validate_addr      = eth_validate_addr,
2560         .ndo_set_mac_address    = eth_mac_addr,
2561         .ndo_fdb_add            = vxlan_fdb_add,
2562         .ndo_fdb_del            = vxlan_fdb_delete,
2563         .ndo_fdb_dump           = vxlan_fdb_dump,
2564         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2565 };
2566
2567 static const struct net_device_ops vxlan_netdev_raw_ops = {
2568         .ndo_init               = vxlan_init,
2569         .ndo_uninit             = vxlan_uninit,
2570         .ndo_open               = vxlan_open,
2571         .ndo_stop               = vxlan_stop,
2572         .ndo_start_xmit         = vxlan_xmit,
2573         .ndo_get_stats64        = ip_tunnel_get_stats64,
2574         .ndo_change_mtu         = vxlan_change_mtu,
2575         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2576 };
2577
2578 /* Info for udev, that this is a virtual tunnel endpoint */
2579 static struct device_type vxlan_type = {
2580         .name = "vxlan",
2581 };
2582
2583 /* Calls the ndo_udp_tunnel_add of the caller in order to
2584  * supply the listening VXLAN udp ports. Callers are expected
2585  * to implement the ndo_udp_tunnel_add.
2586  */
2587 static void vxlan_push_rx_ports(struct net_device *dev)
2588 {
2589         struct vxlan_sock *vs;
2590         struct net *net = dev_net(dev);
2591         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2592         unsigned int i;
2593
2594         spin_lock(&vn->sock_lock);
2595         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2596                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2597                         udp_tunnel_push_rx_port(dev, vs->sock,
2598                                                 (vs->flags & VXLAN_F_GPE) ?
2599                                                 UDP_TUNNEL_TYPE_VXLAN_GPE :
2600                                                 UDP_TUNNEL_TYPE_VXLAN);
2601         }
2602         spin_unlock(&vn->sock_lock);
2603 }
2604
2605 /* Initialize the device structure. */
2606 static void vxlan_setup(struct net_device *dev)
2607 {
2608         struct vxlan_dev *vxlan = netdev_priv(dev);
2609         unsigned int h;
2610
2611         eth_hw_addr_random(dev);
2612         ether_setup(dev);
2613
2614         dev->needs_free_netdev = true;
2615         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2616
2617         dev->features   |= NETIF_F_LLTX;
2618         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2619         dev->features   |= NETIF_F_RXCSUM;
2620         dev->features   |= NETIF_F_GSO_SOFTWARE;
2621
2622         dev->vlan_features = dev->features;
2623         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2624         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2625         netif_keep_dst(dev);
2626         dev->priv_flags |= IFF_NO_QUEUE;
2627
2628         INIT_LIST_HEAD(&vxlan->next);
2629         spin_lock_init(&vxlan->hash_lock);
2630
2631         init_timer_deferrable(&vxlan->age_timer);
2632         vxlan->age_timer.function = vxlan_cleanup;
2633         vxlan->age_timer.data = (unsigned long) vxlan;
2634
2635         vxlan->cfg.dst_port = htons(vxlan_port);
2636
2637         vxlan->dev = dev;
2638
2639         gro_cells_init(&vxlan->gro_cells, dev);
2640
2641         for (h = 0; h < FDB_HASH_SIZE; ++h)
2642                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2643 }
2644
2645 static void vxlan_ether_setup(struct net_device *dev)
2646 {
2647         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2648         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2649         dev->netdev_ops = &vxlan_netdev_ether_ops;
2650 }
2651
2652 static void vxlan_raw_setup(struct net_device *dev)
2653 {
2654         dev->header_ops = NULL;
2655         dev->type = ARPHRD_NONE;
2656         dev->hard_header_len = 0;
2657         dev->addr_len = 0;
2658         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2659         dev->netdev_ops = &vxlan_netdev_raw_ops;
2660 }
2661
2662 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2663         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2664         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2665         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2666         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2667         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2668         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2669         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2670         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2671         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2672         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2673         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2674         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2675         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2676         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2677         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2678         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2679         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2680         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2681         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2682         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2683         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2684         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2685         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2686         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2687         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2688         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2689         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2690 };
2691
2692 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2693 {
2694         if (tb[IFLA_ADDRESS]) {
2695                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2696                         pr_debug("invalid link address (not ethernet)\n");
2697                         return -EINVAL;
2698                 }
2699
2700                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2701                         pr_debug("invalid all zero ethernet address\n");
2702                         return -EADDRNOTAVAIL;
2703                 }
2704         }
2705
2706         if (!data)
2707                 return -EINVAL;
2708
2709         if (data[IFLA_VXLAN_ID]) {
2710                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2711                 if (id >= VXLAN_N_VID)
2712                         return -ERANGE;
2713         }
2714
2715         if (data[IFLA_VXLAN_PORT_RANGE]) {
2716                 const struct ifla_vxlan_port_range *p
2717                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2718
2719                 if (ntohs(p->high) < ntohs(p->low)) {
2720                         pr_debug("port range %u .. %u not valid\n",
2721                                  ntohs(p->low), ntohs(p->high));
2722                         return -EINVAL;
2723                 }
2724         }
2725
2726         return 0;
2727 }
2728
2729 static void vxlan_get_drvinfo(struct net_device *netdev,
2730                               struct ethtool_drvinfo *drvinfo)
2731 {
2732         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2733         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2734 }
2735
2736 static const struct ethtool_ops vxlan_ethtool_ops = {
2737         .get_drvinfo    = vxlan_get_drvinfo,
2738         .get_link       = ethtool_op_get_link,
2739 };
2740
2741 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2742                                         __be16 port, u32 flags)
2743 {
2744         struct socket *sock;
2745         struct udp_port_cfg udp_conf;
2746         int err;
2747
2748         memset(&udp_conf, 0, sizeof(udp_conf));
2749
2750         if (ipv6) {
2751                 udp_conf.family = AF_INET6;
2752                 udp_conf.use_udp6_rx_checksums =
2753                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2754                 udp_conf.ipv6_v6only = 1;
2755         } else {
2756                 udp_conf.family = AF_INET;
2757         }
2758
2759         udp_conf.local_udp_port = port;
2760
2761         /* Open UDP socket */
2762         err = udp_sock_create(net, &udp_conf, &sock);
2763         if (err < 0)
2764                 return ERR_PTR(err);
2765
2766         return sock;
2767 }
2768
2769 /* Create new listen socket if needed */
2770 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2771                                               __be16 port, u32 flags)
2772 {
2773         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2774         struct vxlan_sock *vs;
2775         struct socket *sock;
2776         unsigned int h;
2777         struct udp_tunnel_sock_cfg tunnel_cfg;
2778
2779         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2780         if (!vs)
2781                 return ERR_PTR(-ENOMEM);
2782
2783         for (h = 0; h < VNI_HASH_SIZE; ++h)
2784                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2785
2786         sock = vxlan_create_sock(net, ipv6, port, flags);
2787         if (IS_ERR(sock)) {
2788                 kfree(vs);
2789                 return ERR_CAST(sock);
2790         }
2791
2792         vs->sock = sock;
2793         atomic_set(&vs->refcnt, 1);
2794         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2795
2796         spin_lock(&vn->sock_lock);
2797         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2798         udp_tunnel_notify_add_rx_port(sock,
2799                                       (vs->flags & VXLAN_F_GPE) ?
2800                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2801                                       UDP_TUNNEL_TYPE_VXLAN);
2802         spin_unlock(&vn->sock_lock);
2803
2804         /* Mark socket as an encapsulation socket. */
2805         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2806         tunnel_cfg.sk_user_data = vs;
2807         tunnel_cfg.encap_type = 1;
2808         tunnel_cfg.encap_rcv = vxlan_rcv;
2809         tunnel_cfg.encap_destroy = NULL;
2810         tunnel_cfg.gro_receive = vxlan_gro_receive;
2811         tunnel_cfg.gro_complete = vxlan_gro_complete;
2812
2813         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2814
2815         return vs;
2816 }
2817
2818 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2819 {
2820         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2821         struct vxlan_sock *vs = NULL;
2822
2823         if (!vxlan->cfg.no_share) {
2824                 spin_lock(&vn->sock_lock);
2825                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2826                                      vxlan->cfg.dst_port, vxlan->flags);
2827                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2828                         spin_unlock(&vn->sock_lock);
2829                         return -EBUSY;
2830                 }
2831                 spin_unlock(&vn->sock_lock);
2832         }
2833         if (!vs)
2834                 vs = vxlan_socket_create(vxlan->net, ipv6,
2835                                          vxlan->cfg.dst_port, vxlan->flags);
2836         if (IS_ERR(vs))
2837                 return PTR_ERR(vs);
2838 #if IS_ENABLED(CONFIG_IPV6)
2839         if (ipv6)
2840                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2841         else
2842 #endif
2843                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2844         vxlan_vs_add_dev(vs, vxlan);
2845         return 0;
2846 }
2847
2848 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2849 {
2850         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2851         bool ipv6 = vxlan->flags & VXLAN_F_IPV6 || metadata;
2852         bool ipv4 = !ipv6 || metadata;
2853         int ret = 0;
2854
2855         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2856 #if IS_ENABLED(CONFIG_IPV6)
2857         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2858         if (ipv6) {
2859                 ret = __vxlan_sock_add(vxlan, true);
2860                 if (ret < 0 && ret != -EAFNOSUPPORT)
2861                         ipv4 = false;
2862         }
2863 #endif
2864         if (ipv4)
2865                 ret = __vxlan_sock_add(vxlan, false);
2866         if (ret < 0)
2867                 vxlan_sock_release(vxlan);
2868         return ret;
2869 }
2870
2871 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2872                                struct vxlan_config *conf,
2873                                bool changelink)
2874 {
2875         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2876         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2877         struct vxlan_rdst *dst = &vxlan->default_dst;
2878         unsigned short needed_headroom = ETH_HLEN;
2879         bool use_ipv6 = false;
2880         __be16 default_port = vxlan->cfg.dst_port;
2881         struct net_device *lowerdev = NULL;
2882
2883         if (!changelink) {
2884                 if (conf->flags & VXLAN_F_GPE) {
2885                         /* For now, allow GPE only together with
2886                          * COLLECT_METADATA. This can be relaxed later; in such
2887                          * case, the other side of the PtP link will have to be
2888                          * provided.
2889                          */
2890                         if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2891                             !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2892                                 pr_info("unsupported combination of extensions\n");
2893                                 return -EINVAL;
2894                         }
2895                         vxlan_raw_setup(dev);
2896                 } else {
2897                         vxlan_ether_setup(dev);
2898                 }
2899
2900                 /* MTU range: 68 - 65535 */
2901                 dev->min_mtu = ETH_MIN_MTU;
2902                 dev->max_mtu = ETH_MAX_MTU;
2903                 vxlan->net = src_net;
2904         }
2905
2906         dst->remote_vni = conf->vni;
2907
2908         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2909
2910         /* Unless IPv6 is explicitly requested, assume IPv4 */
2911         if (!dst->remote_ip.sa.sa_family)
2912                 dst->remote_ip.sa.sa_family = AF_INET;
2913
2914         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2915             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2916                 if (!IS_ENABLED(CONFIG_IPV6))
2917                         return -EPFNOSUPPORT;
2918                 use_ipv6 = true;
2919                 vxlan->flags |= VXLAN_F_IPV6;
2920         }
2921
2922         if (conf->label && !use_ipv6) {
2923                 pr_info("label only supported in use with IPv6\n");
2924                 return -EINVAL;
2925         }
2926
2927         if (conf->remote_ifindex &&
2928             conf->remote_ifindex != vxlan->cfg.remote_ifindex) {
2929                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2930                 dst->remote_ifindex = conf->remote_ifindex;
2931
2932                 if (!lowerdev) {
2933                         pr_info("ifindex %d does not exist\n",
2934                                 dst->remote_ifindex);
2935                         return -ENODEV;
2936                 }
2937
2938 #if IS_ENABLED(CONFIG_IPV6)
2939                 if (use_ipv6) {
2940                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2941                         if (idev && idev->cnf.disable_ipv6) {
2942                                 pr_info("IPv6 is disabled via sysctl\n");
2943                                 return -EPERM;
2944                         }
2945                 }
2946 #endif
2947
2948                 if (!conf->mtu)
2949                         dev->mtu = lowerdev->mtu -
2950                                    (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2951
2952                 needed_headroom = lowerdev->hard_header_len;
2953         } else if (!conf->remote_ifindex &&
2954                    vxlan_addr_multicast(&dst->remote_ip)) {
2955                 pr_info("multicast destination requires interface to be specified\n");
2956                 return -EINVAL;
2957         }
2958
2959         if (lowerdev) {
2960                 dev->gso_max_size = lowerdev->gso_max_size;
2961                 dev->gso_max_segs = lowerdev->gso_max_segs;
2962         }
2963
2964         if (conf->mtu) {
2965                 int max_mtu = ETH_MAX_MTU;
2966
2967                 if (lowerdev)
2968                         max_mtu = lowerdev->mtu;
2969
2970                 max_mtu -= (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2971
2972                 if (conf->mtu < dev->min_mtu || conf->mtu > dev->max_mtu)
2973                         return -EINVAL;
2974
2975                 dev->mtu = conf->mtu;
2976
2977                 if (conf->mtu > max_mtu)
2978                         dev->mtu = max_mtu;
2979         }
2980
2981         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2982                 needed_headroom += VXLAN6_HEADROOM;
2983         else
2984                 needed_headroom += VXLAN_HEADROOM;
2985         dev->needed_headroom = needed_headroom;
2986
2987         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2988         if (!vxlan->cfg.dst_port) {
2989                 if (conf->flags & VXLAN_F_GPE)
2990                         vxlan->cfg.dst_port = htons(4790); /* IANA VXLAN-GPE port */
2991                 else
2992                         vxlan->cfg.dst_port = default_port;
2993         }
2994         vxlan->flags |= conf->flags;
2995
2996         if (!vxlan->cfg.age_interval)
2997                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2998
2999         if (changelink)
3000                 return 0;
3001
3002         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3003                 if (tmp->cfg.vni == conf->vni &&
3004                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
3005                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
3006                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
3007                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
3008                     (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
3009                         pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
3010                         return -EEXIST;
3011                 }
3012         }
3013
3014         return 0;
3015 }
3016
3017 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3018                               struct vxlan_config *conf)
3019 {
3020         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3021         struct vxlan_dev *vxlan = netdev_priv(dev);
3022         int err;
3023
3024         err = vxlan_dev_configure(net, dev, conf, false);
3025         if (err)
3026                 return err;
3027
3028         dev->ethtool_ops = &vxlan_ethtool_ops;
3029
3030         /* create an fdb entry for a valid default destination */
3031         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
3032                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3033                                        &vxlan->default_dst.remote_ip,
3034                                        NUD_REACHABLE | NUD_PERMANENT,
3035                                        NLM_F_EXCL | NLM_F_CREATE,
3036                                        vxlan->cfg.dst_port,
3037                                        vxlan->default_dst.remote_vni,
3038                                        vxlan->default_dst.remote_vni,
3039                                        vxlan->default_dst.remote_ifindex,
3040                                        NTF_SELF);
3041                 if (err)
3042                         return err;
3043         }
3044
3045         err = register_netdevice(dev);
3046         if (err) {
3047                 vxlan_fdb_delete_default(vxlan, vxlan->default_dst.remote_vni);
3048                 return err;
3049         }
3050
3051         list_add(&vxlan->next, &vn->vxlan_list);
3052         return 0;
3053 }
3054
3055 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3056                          struct net_device *dev, struct vxlan_config *conf,
3057                          bool changelink)
3058 {
3059         struct vxlan_dev *vxlan = netdev_priv(dev);
3060
3061         memset(conf, 0, sizeof(*conf));
3062
3063         /* if changelink operation, start with old existing cfg */
3064         if (changelink)
3065                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3066
3067         if (data[IFLA_VXLAN_ID]) {
3068                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3069
3070                 if (changelink && (vni != conf->vni))
3071                         return -EOPNOTSUPP;
3072                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3073         }
3074
3075         if (data[IFLA_VXLAN_GROUP]) {
3076                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3077         } else if (data[IFLA_VXLAN_GROUP6]) {
3078                 if (!IS_ENABLED(CONFIG_IPV6))
3079                         return -EPFNOSUPPORT;
3080
3081                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3082                 conf->remote_ip.sa.sa_family = AF_INET6;
3083         }
3084
3085         if (data[IFLA_VXLAN_LOCAL]) {
3086                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3087                 conf->saddr.sa.sa_family = AF_INET;
3088         } else if (data[IFLA_VXLAN_LOCAL6]) {
3089                 if (!IS_ENABLED(CONFIG_IPV6))
3090                         return -EPFNOSUPPORT;
3091
3092                 /* TODO: respect scope id */
3093                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3094                 conf->saddr.sa.sa_family = AF_INET6;
3095         }
3096
3097         if (data[IFLA_VXLAN_LINK])
3098                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3099
3100         if (data[IFLA_VXLAN_TOS])
3101                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3102
3103         if (data[IFLA_VXLAN_TTL])
3104                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3105
3106         if (data[IFLA_VXLAN_LABEL])
3107                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3108                              IPV6_FLOWLABEL_MASK;
3109
3110         if (data[IFLA_VXLAN_LEARNING]) {
3111                 if (nla_get_u8(data[IFLA_VXLAN_LEARNING])) {
3112                         conf->flags |= VXLAN_F_LEARN;
3113                 } else {
3114                         conf->flags &= ~VXLAN_F_LEARN;
3115                         vxlan->flags &= ~VXLAN_F_LEARN;
3116                 }
3117         } else if (!changelink) {
3118                 /* default to learn on a new device */
3119                 conf->flags |= VXLAN_F_LEARN;
3120         }
3121
3122         if (data[IFLA_VXLAN_AGEING]) {
3123                 if (changelink)
3124                         return -EOPNOTSUPP;
3125                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3126         }
3127
3128         if (data[IFLA_VXLAN_PROXY]) {
3129                 if (changelink)
3130                         return -EOPNOTSUPP;
3131                 if (nla_get_u8(data[IFLA_VXLAN_PROXY]))
3132                         conf->flags |= VXLAN_F_PROXY;
3133         }
3134
3135         if (data[IFLA_VXLAN_RSC]) {
3136                 if (changelink)
3137                         return -EOPNOTSUPP;
3138                 if (nla_get_u8(data[IFLA_VXLAN_RSC]))
3139                         conf->flags |= VXLAN_F_RSC;
3140         }
3141
3142         if (data[IFLA_VXLAN_L2MISS]) {
3143                 if (changelink)
3144                         return -EOPNOTSUPP;
3145                 if (nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3146                         conf->flags |= VXLAN_F_L2MISS;
3147         }
3148
3149         if (data[IFLA_VXLAN_L3MISS]) {
3150                 if (changelink)
3151                         return -EOPNOTSUPP;
3152                 if (nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3153                         conf->flags |= VXLAN_F_L3MISS;
3154         }
3155
3156         if (data[IFLA_VXLAN_LIMIT]) {
3157                 if (changelink)
3158                         return -EOPNOTSUPP;
3159                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3160         }
3161
3162         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3163                 if (changelink)
3164                         return -EOPNOTSUPP;
3165                 if (nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3166                         conf->flags |= VXLAN_F_COLLECT_METADATA;
3167         }
3168
3169         if (data[IFLA_VXLAN_PORT_RANGE]) {
3170                 if (!changelink) {
3171                         const struct ifla_vxlan_port_range *p
3172                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3173                         conf->port_min = ntohs(p->low);
3174                         conf->port_max = ntohs(p->high);
3175                 } else {
3176                         return -EOPNOTSUPP;
3177                 }
3178         }
3179
3180         if (data[IFLA_VXLAN_PORT]) {
3181                 if (changelink)
3182                         return -EOPNOTSUPP;
3183                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3184         }
3185
3186         if (data[IFLA_VXLAN_UDP_CSUM]) {
3187                 if (changelink)
3188                         return -EOPNOTSUPP;
3189                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3190                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3191         }
3192
3193         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3194                 if (changelink)
3195                         return -EOPNOTSUPP;
3196                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3197                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3198         }
3199
3200         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3201                 if (changelink)
3202                         return -EOPNOTSUPP;
3203                 if (nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3204                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3205         }
3206
3207         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3208                 if (changelink)
3209                         return -EOPNOTSUPP;
3210                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3211                         conf->flags |= VXLAN_F_REMCSUM_TX;
3212         }
3213
3214         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3215                 if (changelink)
3216                         return -EOPNOTSUPP;
3217                 if (nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3218                         conf->flags |= VXLAN_F_REMCSUM_RX;
3219         }
3220
3221         if (data[IFLA_VXLAN_GBP]) {
3222                 if (changelink)
3223                         return -EOPNOTSUPP;
3224                 conf->flags |= VXLAN_F_GBP;
3225         }
3226
3227         if (data[IFLA_VXLAN_GPE]) {
3228                 if (changelink)
3229                         return -EOPNOTSUPP;
3230                 conf->flags |= VXLAN_F_GPE;
3231         }
3232
3233         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3234                 if (changelink)
3235                         return -EOPNOTSUPP;
3236                 conf->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3237         }
3238
3239         if (tb[IFLA_MTU]) {
3240                 if (changelink)
3241                         return -EOPNOTSUPP;
3242                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3243         }
3244
3245         return 0;
3246 }
3247
3248 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3249                          struct nlattr *tb[], struct nlattr *data[])
3250 {
3251         struct vxlan_config conf;
3252         int err;
3253
3254         err = vxlan_nl2conf(tb, data, dev, &conf, false);
3255         if (err)
3256                 return err;
3257
3258         return __vxlan_dev_create(src_net, dev, &conf);
3259 }
3260
3261 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3262                             struct nlattr *data[])
3263 {
3264         struct vxlan_dev *vxlan = netdev_priv(dev);
3265         struct vxlan_rdst *dst = &vxlan->default_dst;
3266         struct vxlan_rdst old_dst;
3267         struct vxlan_config conf;
3268         int err;
3269
3270         err = vxlan_nl2conf(tb, data,
3271                             dev, &conf, true);
3272         if (err)
3273                 return err;
3274
3275         memcpy(&old_dst, dst, sizeof(struct vxlan_rdst));
3276
3277         err = vxlan_dev_configure(vxlan->net, dev, &conf, true);
3278         if (err)
3279                 return err;
3280
3281         /* handle default dst entry */
3282         if (!vxlan_addr_equal(&dst->remote_ip, &old_dst.remote_ip)) {
3283                 spin_lock_bh(&vxlan->hash_lock);
3284                 if (!vxlan_addr_any(&old_dst.remote_ip))
3285                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
3286                                            old_dst.remote_ip,
3287                                            vxlan->cfg.dst_port,
3288                                            old_dst.remote_vni,
3289                                            old_dst.remote_vni,
3290                                            old_dst.remote_ifindex, 0);
3291
3292                 if (!vxlan_addr_any(&dst->remote_ip)) {
3293                         err = vxlan_fdb_create(vxlan, all_zeros_mac,
3294                                                &dst->remote_ip,
3295                                                NUD_REACHABLE | NUD_PERMANENT,
3296                                                NLM_F_CREATE | NLM_F_APPEND,
3297                                                vxlan->cfg.dst_port,
3298                                                dst->remote_vni,
3299                                                dst->remote_vni,
3300                                                dst->remote_ifindex,
3301                                                NTF_SELF);
3302                         if (err) {
3303                                 spin_unlock_bh(&vxlan->hash_lock);
3304                                 return err;
3305                         }
3306                 }
3307                 spin_unlock_bh(&vxlan->hash_lock);
3308         }
3309
3310         return 0;
3311 }
3312
3313 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3314 {
3315         struct vxlan_dev *vxlan = netdev_priv(dev);
3316
3317         vxlan_flush(vxlan, true);
3318
3319         gro_cells_destroy(&vxlan->gro_cells);
3320         list_del(&vxlan->next);
3321         unregister_netdevice_queue(dev, head);
3322 }
3323
3324 static size_t vxlan_get_size(const struct net_device *dev)
3325 {
3326
3327         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3328                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3329                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3330                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3331                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3332                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3333                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3334                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3335                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3336                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3337                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3338                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3339                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3340                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3341                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3342                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3343                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3344                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3345                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3346                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3347                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3348                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3349                 0;
3350 }
3351
3352 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3353 {
3354         const struct vxlan_dev *vxlan = netdev_priv(dev);
3355         const struct vxlan_rdst *dst = &vxlan->default_dst;
3356         struct ifla_vxlan_port_range ports = {
3357                 .low =  htons(vxlan->cfg.port_min),
3358                 .high = htons(vxlan->cfg.port_max),
3359         };
3360
3361         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3362                 goto nla_put_failure;
3363
3364         if (!vxlan_addr_any(&dst->remote_ip)) {
3365                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3366                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3367                                             dst->remote_ip.sin.sin_addr.s_addr))
3368                                 goto nla_put_failure;
3369 #if IS_ENABLED(CONFIG_IPV6)
3370                 } else {
3371                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3372                                              &dst->remote_ip.sin6.sin6_addr))
3373                                 goto nla_put_failure;
3374 #endif
3375                 }
3376         }
3377
3378         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3379                 goto nla_put_failure;
3380
3381         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3382                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3383                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3384                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3385                                 goto nla_put_failure;
3386 #if IS_ENABLED(CONFIG_IPV6)
3387                 } else {
3388                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3389                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3390                                 goto nla_put_failure;
3391 #endif
3392                 }
3393         }
3394
3395         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3396             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3397             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3398             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3399                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3400             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3401                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3402             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3403             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3404                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3405             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3406                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3407             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3408                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3409             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3410             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3411             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3412             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3413                         !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3414             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3415                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3416             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3417                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3418             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3419                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3420             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3421                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3422                 goto nla_put_failure;
3423
3424         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3425                 goto nla_put_failure;
3426
3427         if (vxlan->flags & VXLAN_F_GBP &&
3428             nla_put_flag(skb, IFLA_VXLAN_GBP))
3429                 goto nla_put_failure;
3430
3431         if (vxlan->flags & VXLAN_F_GPE &&
3432             nla_put_flag(skb, IFLA_VXLAN_GPE))
3433                 goto nla_put_failure;
3434
3435         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3436             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3437                 goto nla_put_failure;
3438
3439         return 0;
3440
3441 nla_put_failure:
3442         return -EMSGSIZE;
3443 }
3444
3445 static struct net *vxlan_get_link_net(const struct net_device *dev)
3446 {
3447         struct vxlan_dev *vxlan = netdev_priv(dev);
3448
3449         return vxlan->net;
3450 }
3451
3452 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3453         .kind           = "vxlan",
3454         .maxtype        = IFLA_VXLAN_MAX,
3455         .policy         = vxlan_policy,
3456         .priv_size      = sizeof(struct vxlan_dev),
3457         .setup          = vxlan_setup,
3458         .validate       = vxlan_validate,
3459         .newlink        = vxlan_newlink,
3460         .changelink     = vxlan_changelink,
3461         .dellink        = vxlan_dellink,
3462         .get_size       = vxlan_get_size,
3463         .fill_info      = vxlan_fill_info,
3464         .get_link_net   = vxlan_get_link_net,
3465 };
3466
3467 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3468                                     u8 name_assign_type,
3469                                     struct vxlan_config *conf)
3470 {
3471         struct nlattr *tb[IFLA_MAX + 1];
3472         struct net_device *dev;
3473         int err;
3474
3475         memset(&tb, 0, sizeof(tb));
3476
3477         dev = rtnl_create_link(net, name, name_assign_type,
3478                                &vxlan_link_ops, tb);
3479         if (IS_ERR(dev))
3480                 return dev;
3481
3482         err = __vxlan_dev_create(net, dev, conf);
3483         if (err < 0) {
3484                 free_netdev(dev);
3485                 return ERR_PTR(err);
3486         }
3487
3488         err = rtnl_configure_link(dev, NULL);
3489         if (err < 0) {
3490                 LIST_HEAD(list_kill);
3491
3492                 vxlan_dellink(dev, &list_kill);
3493                 unregister_netdevice_many(&list_kill);
3494                 return ERR_PTR(err);
3495         }
3496
3497         return dev;
3498 }
3499 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3500
3501 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3502                                              struct net_device *dev)
3503 {
3504         struct vxlan_dev *vxlan, *next;
3505         LIST_HEAD(list_kill);
3506
3507         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3508                 struct vxlan_rdst *dst = &vxlan->default_dst;
3509
3510                 /* In case we created vxlan device with carrier
3511                  * and we loose the carrier due to module unload
3512                  * we also need to remove vxlan device. In other
3513                  * cases, it's not necessary and remote_ifindex
3514                  * is 0 here, so no matches.
3515                  */
3516                 if (dst->remote_ifindex == dev->ifindex)
3517                         vxlan_dellink(vxlan->dev, &list_kill);
3518         }
3519
3520         unregister_netdevice_many(&list_kill);
3521 }
3522
3523 static int vxlan_netdevice_event(struct notifier_block *unused,
3524                                  unsigned long event, void *ptr)
3525 {
3526         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3527         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3528
3529         if (event == NETDEV_UNREGISTER)
3530                 vxlan_handle_lowerdev_unregister(vn, dev);
3531         else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3532                 vxlan_push_rx_ports(dev);
3533
3534         return NOTIFY_DONE;
3535 }
3536
3537 static struct notifier_block vxlan_notifier_block __read_mostly = {
3538         .notifier_call = vxlan_netdevice_event,
3539 };
3540
3541 static __net_init int vxlan_init_net(struct net *net)
3542 {
3543         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3544         unsigned int h;
3545
3546         INIT_LIST_HEAD(&vn->vxlan_list);
3547         spin_lock_init(&vn->sock_lock);
3548
3549         for (h = 0; h < PORT_HASH_SIZE; ++h)
3550                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3551
3552         return 0;
3553 }
3554
3555 static void __net_exit vxlan_exit_net(struct net *net)
3556 {
3557         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3558         struct vxlan_dev *vxlan, *next;
3559         struct net_device *dev, *aux;
3560         LIST_HEAD(list);
3561
3562         rtnl_lock();
3563         for_each_netdev_safe(net, dev, aux)
3564                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3565                         unregister_netdevice_queue(dev, &list);
3566
3567         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3568                 /* If vxlan->dev is in the same netns, it has already been added
3569                  * to the list by the previous loop.
3570                  */
3571                 if (!net_eq(dev_net(vxlan->dev), net)) {
3572                         gro_cells_destroy(&vxlan->gro_cells);
3573                         unregister_netdevice_queue(vxlan->dev, &list);
3574                 }
3575         }
3576
3577         unregister_netdevice_many(&list);
3578         rtnl_unlock();
3579 }
3580
3581 static struct pernet_operations vxlan_net_ops = {
3582         .init = vxlan_init_net,
3583         .exit = vxlan_exit_net,
3584         .id   = &vxlan_net_id,
3585         .size = sizeof(struct vxlan_net),
3586 };
3587
3588 static int __init vxlan_init_module(void)
3589 {
3590         int rc;
3591
3592         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3593
3594         rc = register_pernet_subsys(&vxlan_net_ops);
3595         if (rc)
3596                 goto out1;
3597
3598         rc = register_netdevice_notifier(&vxlan_notifier_block);
3599         if (rc)
3600                 goto out2;
3601
3602         rc = rtnl_link_register(&vxlan_link_ops);
3603         if (rc)
3604                 goto out3;
3605
3606         return 0;
3607 out3:
3608         unregister_netdevice_notifier(&vxlan_notifier_block);
3609 out2:
3610         unregister_pernet_subsys(&vxlan_net_ops);
3611 out1:
3612         return rc;
3613 }
3614 late_initcall(vxlan_init_module);
3615
3616 static void __exit vxlan_cleanup_module(void)
3617 {
3618         rtnl_link_unregister(&vxlan_link_ops);
3619         unregister_netdevice_notifier(&vxlan_notifier_block);
3620         unregister_pernet_subsys(&vxlan_net_ops);
3621         /* rcu_barrier() is called by netns */
3622 }
3623 module_exit(vxlan_cleanup_module);
3624
3625 MODULE_LICENSE("GPL");
3626 MODULE_VERSION(VXLAN_VERSION);
3627 MODULE_AUTHOR("Stephen Hemminger <[email protected]>");
3628 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3629 MODULE_ALIAS_RTNL_LINK("vxlan");