drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/types.h>
  15 #include <linux/module.h>
  16 #include <linux/errno.h>
  17 #include <linux/slab.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/rculist.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/in.h>
  22 #include <linux/ip.h>
  23 #include <linux/udp.h>
  24 #include <linux/igmp.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/hash.h>
  29 #include <linux/ethtool.h>
  30 #include <net/arp.h>
  31 #include <net/ndisc.h>
  32 #include <net/ip.h>
  33 #include <net/ip_tunnels.h>
  34 #include <net/icmp.h>
  35 #include <net/udp.h>
  36 #include <net/udp_tunnel.h>
  37 #include <net/rtnetlink.h>
  38 #include <net/route.h>
  39 #include <net/dsfield.h>
  40 #include <net/inet_ecn.h>
  41 #include <net/net_namespace.h>
  42 #include <net/netns/generic.h>
  43 #include <net/vxlan.h>
  44 #include <net/protocol.h>
  45
  46 #if IS_ENABLED(CONFIG_IPV6)
  47 #include <net/ipv6.h>
  48 #include <net/addrconf.h>
  49 #include <net/ip6_tunnel.h>
  50 #include <net/ip6_checksum.h>
  51 #endif
  52 #include <net/dst_metadata.h>
  53
  54 #define VXLAN_VERSION   "0.1"
  55
  56 #define PORT_HASH_BITS  8
  57 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  58 #define FDB_AGE_DEFAULT 300 /* 5 min */
  59 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  60
  61 /* UDP port for VXLAN traffic.
  62  * The IANA assigned port is 4789, but the Linux default is 8472
  63  * for compatibility with early adopters.
  64  */
  65 static unsigned short vxlan_port __read_mostly = 8472;
  66 module_param_named(udp_port, vxlan_port, ushort, 0444);
  67 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  68
  69 static bool log_ecn_error = true;
  70 module_param(log_ecn_error, bool, 0644);
  71 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  72
  73 static int vxlan_net_id;
  74 static struct rtnl_link_ops vxlan_link_ops;
  75
  76 static const u8 all_zeros_mac[ETH_ALEN + 2];
  77
  78 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  79
  80 /* per-network namespace private data for this module */
  81 struct vxlan_net {
  82         struct list_head  vxlan_list;
  83         struct hlist_head sock_list[PORT_HASH_SIZE];
  84         spinlock_t        sock_lock;
  85 };
  86
  87 /* Forwarding table entry */
  88 struct vxlan_fdb {
  89         struct hlist_node hlist;        /* linked list of entries */
  90         struct rcu_head   rcu;
  91         unsigned long     updated;      /* jiffies */
  92         unsigned long     used;
  93         struct list_head  remotes;
  94         u8                eth_addr[ETH_ALEN];
  95         u16               state;        /* see ndm_state */
  96         u8                flags;        /* see ndm_flags */
  97 };
  98
  99 /* salt for hash table */
 100 static u32 vxlan_salt __read_mostly;
 101
 102 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
 103 {
 104         return vs->flags & VXLAN_F_COLLECT_METADATA ||
 105                ip_tunnel_collect_metadata();
 106 }
 107
 108 #if IS_ENABLED(CONFIG_IPV6)
 109 static inline
 110 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 111 {
 112         if (a->sa.sa_family != b->sa.sa_family)
 113                 return false;
 114         if (a->sa.sa_family == AF_INET6)
 115                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 116         else
 117                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 118 }
 119
 120 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 121 {
 122         if (ipa->sa.sa_family == AF_INET6)
 123                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 124         else
 125                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 126 }
 127
 128 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 129 {
 130         if (ipa->sa.sa_family == AF_INET6)
 131                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 132         else
 133                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 134 }
 135
 136 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 137 {
 138         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 139                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 140                 ip->sa.sa_family = AF_INET6;
 141                 return 0;
 142         } else if (nla_len(nla) >= sizeof(__be32)) {
 143                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 144                 ip->sa.sa_family = AF_INET;
 145                 return 0;
 146         } else {
 147                 return -EAFNOSUPPORT;
 148         }
 149 }
 150
 151 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 152                               const union vxlan_addr *ip)
 153 {
 154         if (ip->sa.sa_family == AF_INET6)
 155                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 156         else
 157                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 158 }
 159
 160 #else /* !CONFIG_IPV6 */
 161
 162 static inline
 163 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 164 {
 165         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 166 }
 167
 168 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 169 {
 170         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 171 }
 172
 173 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 174 {
 175         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 176 }
 177
 178 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 179 {
 180         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 181                 return -EAFNOSUPPORT;
 182         } else if (nla_len(nla) >= sizeof(__be32)) {
 183                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 184                 ip->sa.sa_family = AF_INET;
 185                 return 0;
 186         } else {
 187                 return -EAFNOSUPPORT;
 188         }
 189 }
 190
 191 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 192                               const union vxlan_addr *ip)
 193 {
 194         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 195 }
 196 #endif
 197
 198 /* Virtual Network hash table head */
 199 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 200 {
 201         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 202 }
 203
 204 /* Socket hash table head */
 205 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 206 {
 207         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 208
 209         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 210 }
 211
 212 /* First remote destination for a forwarding entry.
 213  * Guaranteed to be non-NULL because remotes are never deleted.
 214  */
 215 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 216 {
 217         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 218 }
 219
 220 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 221 {
 222         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 223 }
 224
 225 /* Find VXLAN socket based on network namespace, address family and UDP port
 226  * and enabled unshareable flags.
 227  */
 228 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 229                                           __be16 port, u32 flags)
 230 {
 231         struct vxlan_sock *vs;
 232
 233         flags &= VXLAN_F_RCV_FLAGS;
 234
 235         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 236                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 237                     vxlan_get_sk_family(vs) == family &&
 238                     vs->flags == flags)
 239                         return vs;
 240         }
 241         return NULL;
 242 }
 243
 244 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
 245 {
 246         struct vxlan_dev *vxlan;
 247
 248         /* For flow based devices, map all packets to VNI 0 */
 249         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 250                 vni = 0;
 251
 252         hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
 253                 if (vxlan->default_dst.remote_vni == vni)
 254                         return vxlan;
 255         }
 256
 257         return NULL;
 258 }
 259
 260 /* Look up VNI in a per net namespace table */
 261 static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
 262                                         sa_family_t family, __be16 port,
 263                                         u32 flags)
 264 {
 265         struct vxlan_sock *vs;
 266
 267         vs = vxlan_find_sock(net, family, port, flags);
 268         if (!vs)
 269                 return NULL;
 270
 271         return vxlan_vs_find_vni(vs, vni);
 272 }
 273
 274 /* Fill in neighbour message in skbuff. */
 275 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 276                           const struct vxlan_fdb *fdb,
 277                           u32 portid, u32 seq, int type, unsigned int flags,
 278                           const struct vxlan_rdst *rdst)
 279 {
 280         unsigned long now = jiffies;
 281         struct nda_cacheinfo ci;
 282         struct nlmsghdr *nlh;
 283         struct ndmsg *ndm;
 284         bool send_ip, send_eth;
 285
 286         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 287         if (nlh == NULL)
 288                 return -EMSGSIZE;
 289
 290         ndm = nlmsg_data(nlh);
 291         memset(ndm, 0, sizeof(*ndm));
 292
 293         send_eth = send_ip = true;
 294
 295         if (type == RTM_GETNEIGH) {
 296                 ndm->ndm_family = AF_INET;
 297                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 298                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 299         } else
 300                 ndm->ndm_family = AF_BRIDGE;
 301         ndm->ndm_state = fdb->state;
 302         ndm->ndm_ifindex = vxlan->dev->ifindex;
 303         ndm->ndm_flags = fdb->flags;
 304         ndm->ndm_type = RTN_UNICAST;
 305
 306         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 307             nla_put_s32(skb, NDA_LINK_NETNSID,
 308                         peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
 309                 goto nla_put_failure;
 310
 311         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 312                 goto nla_put_failure;
 313
 314         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 315                 goto nla_put_failure;
 316
 317         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 318             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 319                 goto nla_put_failure;
 320         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 321             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 322                 goto nla_put_failure;
 323         if (rdst->remote_ifindex &&
 324             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 325                 goto nla_put_failure;
 326
 327         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 328         ci.ndm_confirmed = 0;
 329         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 330         ci.ndm_refcnt    = 0;
 331
 332         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 333                 goto nla_put_failure;
 334
 335         nlmsg_end(skb, nlh);
 336         return 0;
 337
 338 nla_put_failure:
 339         nlmsg_cancel(skb, nlh);
 340         return -EMSGSIZE;
 341 }
 342
 343 static inline size_t vxlan_nlmsg_size(void)
 344 {
 345         return NLMSG_ALIGN(sizeof(struct ndmsg))
 346                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 347                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 348                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 349                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 350                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 351                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 352                 + nla_total_size(sizeof(struct nda_cacheinfo));
 353 }
 354
 355 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 356                              struct vxlan_rdst *rd, int type)
 357 {
 358         struct net *net = dev_net(vxlan->dev);
 359         struct sk_buff *skb;
 360         int err = -ENOBUFS;
 361
 362         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 363         if (skb == NULL)
 364                 goto errout;
 365
 366         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 367         if (err < 0) {
 368                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 369                 WARN_ON(err == -EMSGSIZE);
 370                 kfree_skb(skb);
 371                 goto errout;
 372         }
 373
 374         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 375         return;
 376 errout:
 377         if (err < 0)
 378                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 379 }
 380
 381 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 382 {
 383         struct vxlan_dev *vxlan = netdev_priv(dev);
 384         struct vxlan_fdb f = {
 385                 .state = NUD_STALE,
 386         };
 387         struct vxlan_rdst remote = {
 388                 .remote_ip = *ipa, /* goes to NDA_DST */
 389                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 390         };
 391
 392         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 393 }
 394
 395 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 396 {
 397         struct vxlan_fdb f = {
 398                 .state = NUD_STALE,
 399         };
 400         struct vxlan_rdst remote = { };
 401
 402         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 403
 404         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 405 }
 406
 407 /* Hash Ethernet address */
 408 static u32 eth_hash(const unsigned char *addr)
 409 {
 410         u64 value = get_unaligned((u64 *)addr);
 411
 412         /* only want 6 bytes */
 413 #ifdef __BIG_ENDIAN
 414         value >>= 16;
 415 #else
 416         value <<= 16;
 417 #endif
 418         return hash_64(value, FDB_HASH_BITS);
 419 }
 420
 421 /* Hash chain to use given mac address */
 422 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 423                                                 const u8 *mac)
 424 {
 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)
 431 {
 432         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 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                         return f;
 438         }
 439
 440         return NULL;
 441 }
 442
 443 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 444                                         const u8 *mac)
 445 {
 446         struct vxlan_fdb *f;
 447
 448         f = __vxlan_find_mac(vxlan, mac);
 449         if (f)
 450                 f->used = jiffies;
 451
 452         return f;
 453 }
 454
 455 /* caller should hold vxlan->hash_lock */
 456 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 457                                               union vxlan_addr *ip, __be16 port,
 458                                               __be32 vni, __u32 ifindex)
 459 {
 460         struct vxlan_rdst *rd;
 461
 462         list_for_each_entry(rd, &f->remotes, list) {
 463                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 464                     rd->remote_port == port &&
 465                     rd->remote_vni == vni &&
 466                     rd->remote_ifindex == ifindex)
 467                         return rd;
 468         }
 469
 470         return NULL;
 471 }
 472
 473 /* Replace destination of unicast mac */
 474 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 475                              union vxlan_addr *ip, __be16 port, __be32 vni,
 476                              __u32 ifindex)
 477 {
 478         struct vxlan_rdst *rd;
 479
 480         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 481         if (rd)
 482                 return 0;
 483
 484         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 485         if (!rd)
 486                 return 0;
 487
 488         dst_cache_reset(&rd->dst_cache);
 489         rd->remote_ip = *ip;
 490         rd->remote_port = port;
 491         rd->remote_vni = vni;
 492         rd->remote_ifindex = ifindex;
 493         return 1;
 494 }
 495
 496 /* Add/update destinations for multicast */
 497 static int vxlan_fdb_append(struct vxlan_fdb *f,
 498                             union vxlan_addr *ip, __be16 port, __be32 vni,
 499                             __u32 ifindex, struct vxlan_rdst **rdp)
 500 {
 501         struct vxlan_rdst *rd;
 502
 503         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 504         if (rd)
 505                 return 0;
 506
 507         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 508         if (rd == NULL)
 509                 return -ENOBUFS;
 510
 511         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 512                 kfree(rd);
 513                 return -ENOBUFS;
 514         }
 515
 516         rd->remote_ip = *ip;
 517         rd->remote_port = port;
 518         rd->remote_vni = vni;
 519         rd->remote_ifindex = ifindex;
 520
 521         list_add_tail_rcu(&rd->list, &f->remotes);
 522
 523         *rdp = rd;
 524         return 1;
 525 }
 526
 527 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 528                                           unsigned int off,
 529                                           struct vxlanhdr *vh, size_t hdrlen,
 530                                           __be32 vni_field,
 531                                           struct gro_remcsum *grc,
 532                                           bool nopartial)
 533 {
 534         size_t start, offset;
 535
 536         if (skb->remcsum_offload)
 537                 return vh;
 538
 539         if (!NAPI_GRO_CB(skb)->csum_valid)
 540                 return NULL;
 541
 542         start = vxlan_rco_start(vni_field);
 543         offset = start + vxlan_rco_offset(vni_field);
 544
 545         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 546                                      start, offset, grc, nopartial);
 547
 548         skb->remcsum_offload = 1;
 549
 550         return vh;
 551 }
 552
 553 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
 554                                           struct sk_buff **head,
 555                                           struct sk_buff *skb)
 556 {
 557         struct sk_buff *p, **pp = NULL;
 558         struct vxlanhdr *vh, *vh2;
 559         unsigned int hlen, off_vx;
 560         int flush = 1;
 561         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 562         __be32 flags;
 563         struct gro_remcsum grc;
 564
 565         skb_gro_remcsum_init(&grc);
 566
 567         off_vx = skb_gro_offset(skb);
 568         hlen = off_vx + sizeof(*vh);
 569         vh   = skb_gro_header_fast(skb, off_vx);
 570         if (skb_gro_header_hard(skb, hlen)) {
 571                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 572                 if (unlikely(!vh))
 573                         goto out;
 574         }
 575
 576         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 577
 578         flags = vh->vx_flags;
 579
 580         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 581                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 582                                        vh->vx_vni, &grc,
 583                                        !!(vs->flags &
 584                                           VXLAN_F_REMCSUM_NOPARTIAL));
 585
 586                 if (!vh)
 587                         goto out;
 588         }
 589
 590         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 591
 592         for (p = *head; p; p = p->next) {
 593                 if (!NAPI_GRO_CB(p)->same_flow)
 594                         continue;
 595
 596                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 597                 if (vh->vx_flags != vh2->vx_flags ||
 598                     vh->vx_vni != vh2->vx_vni) {
 599                         NAPI_GRO_CB(p)->same_flow = 0;
 600                         continue;
 601                 }
 602         }
 603
 604         pp = eth_gro_receive(head, skb);
 605         flush = 0;
 606
 607 out:
 608         skb_gro_remcsum_cleanup(skb, &grc);
 609         NAPI_GRO_CB(skb)->flush |= flush;
 610
 611         return pp;
 612 }
 613
 614 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 615 {
 616         /* Sets 'skb->inner_mac_header' since we are always called with
 617          * 'skb->encapsulation' set.
 618          */
 619         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 620 }
 621
 622 /* Notify netdevs that UDP port started listening */
 623 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 624 {
 625         struct net_device *dev;
 626         struct sock *sk = vs->sock->sk;
 627         struct net *net = sock_net(sk);
 628         sa_family_t sa_family = vxlan_get_sk_family(vs);
 629         __be16 port = inet_sk(sk)->inet_sport;
 630
 631         rcu_read_lock();
 632         for_each_netdev_rcu(net, dev) {
 633                 if (dev->netdev_ops->ndo_add_vxlan_port)
 634                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 635                                                             port);
 636         }
 637         rcu_read_unlock();
 638 }
 639
 640 /* Notify netdevs that UDP port is no more listening */
 641 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 642 {
 643         struct net_device *dev;
 644         struct sock *sk = vs->sock->sk;
 645         struct net *net = sock_net(sk);
 646         sa_family_t sa_family = vxlan_get_sk_family(vs);
 647         __be16 port = inet_sk(sk)->inet_sport;
 648
 649         rcu_read_lock();
 650         for_each_netdev_rcu(net, dev) {
 651                 if (dev->netdev_ops->ndo_del_vxlan_port)
 652                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 653                                                             port);
 654         }
 655         rcu_read_unlock();
 656 }
 657
 658 /* Add new entry to forwarding table -- assumes lock held */
 659 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 660                             const u8 *mac, union vxlan_addr *ip,
 661                             __u16 state, __u16 flags,
 662                             __be16 port, __be32 vni, __u32 ifindex,
 663                             __u8 ndm_flags)
 664 {
 665         struct vxlan_rdst *rd = NULL;
 666         struct vxlan_fdb *f;
 667         int notify = 0;
 668
 669         f = __vxlan_find_mac(vxlan, mac);
 670         if (f) {
 671                 if (flags & NLM_F_EXCL) {
 672                         netdev_dbg(vxlan->dev,
 673                                    "lost race to create %pM\n", mac);
 674                         return -EEXIST;
 675                 }
 676                 if (f->state != state) {
 677                         f->state = state;
 678                         f->updated = jiffies;
 679                         notify = 1;
 680                 }
 681                 if (f->flags != ndm_flags) {
 682                         f->flags = ndm_flags;
 683                         f->updated = jiffies;
 684                         notify = 1;
 685                 }
 686                 if ((flags & NLM_F_REPLACE)) {
 687                         /* Only change unicasts */
 688                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 689                              is_zero_ether_addr(f->eth_addr))) {
 690                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 691                                                            ifindex);
 692                         } else
 693                                 return -EOPNOTSUPP;
 694                 }
 695                 if ((flags & NLM_F_APPEND) &&
 696                     (is_multicast_ether_addr(f->eth_addr) ||
 697                      is_zero_ether_addr(f->eth_addr))) {
 698                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 699                                                   &rd);
 700
 701                         if (rc < 0)
 702                                 return rc;
 703                         notify |= rc;
 704                 }
 705         } else {
 706                 if (!(flags & NLM_F_CREATE))
 707                         return -ENOENT;
 708
 709                 if (vxlan->cfg.addrmax &&
 710                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 711                         return -ENOSPC;
 712
 713                 /* Disallow replace to add a multicast entry */
 714                 if ((flags & NLM_F_REPLACE) &&
 715                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 716                         return -EOPNOTSUPP;
 717
 718                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 719                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 720                 if (!f)
 721                         return -ENOMEM;
 722
 723                 notify = 1;
 724                 f->state = state;
 725                 f->flags = ndm_flags;
 726                 f->updated = f->used = jiffies;
 727                 INIT_LIST_HEAD(&f->remotes);
 728                 memcpy(f->eth_addr, mac, ETH_ALEN);
 729
 730                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 731
 732                 ++vxlan->addrcnt;
 733                 hlist_add_head_rcu(&f->hlist,
 734                                    vxlan_fdb_head(vxlan, mac));
 735         }
 736
 737         if (notify) {
 738                 if (rd == NULL)
 739                         rd = first_remote_rtnl(f);
 740                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 741         }
 742
 743         return 0;
 744 }
 745
 746 static void vxlan_fdb_free(struct rcu_head *head)
 747 {
 748         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 749         struct vxlan_rdst *rd, *nd;
 750
 751         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 752                 dst_cache_destroy(&rd->dst_cache);
 753                 kfree(rd);
 754         }
 755         kfree(f);
 756 }
 757
 758 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 759 {
 760         netdev_dbg(vxlan->dev,
 761                     "delete %pM\n", f->eth_addr);
 762
 763         --vxlan->addrcnt;
 764         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 765
 766         hlist_del_rcu(&f->hlist);
 767         call_rcu(&f->rcu, vxlan_fdb_free);
 768 }
 769
 770 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 771                            union vxlan_addr *ip, __be16 *port, __be32 *vni,
 772                            u32 *ifindex)
 773 {
 774         struct net *net = dev_net(vxlan->dev);
 775         int err;
 776
 777         if (tb[NDA_DST]) {
 778                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 779                 if (err)
 780                         return err;
 781         } else {
 782                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 783                 if (remote->sa.sa_family == AF_INET) {
 784                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 785                         ip->sa.sa_family = AF_INET;
 786 #if IS_ENABLED(CONFIG_IPV6)
 787                 } else {
 788                         ip->sin6.sin6_addr = in6addr_any;
 789                         ip->sa.sa_family = AF_INET6;
 790 #endif
 791                 }
 792         }
 793
 794         if (tb[NDA_PORT]) {
 795                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 796                         return -EINVAL;
 797                 *port = nla_get_be16(tb[NDA_PORT]);
 798         } else {
 799                 *port = vxlan->cfg.dst_port;
 800         }
 801
 802         if (tb[NDA_VNI]) {
 803                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 804                         return -EINVAL;
 805                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 806         } else {
 807                 *vni = vxlan->default_dst.remote_vni;
 808         }
 809
 810         if (tb[NDA_IFINDEX]) {
 811                 struct net_device *tdev;
 812
 813                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 814                         return -EINVAL;
 815                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 816                 tdev = __dev_get_by_index(net, *ifindex);
 817                 if (!tdev)
 818                         return -EADDRNOTAVAIL;
 819         } else {
 820                 *ifindex = 0;
 821         }
 822
 823         return 0;
 824 }
 825
 826 /* Add static entry (via netlink) */
 827 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 828                          struct net_device *dev,
 829                          const unsigned char *addr, u16 vid, u16 flags)
 830 {
 831         struct vxlan_dev *vxlan = netdev_priv(dev);
 832         /* struct net *net = dev_net(vxlan->dev); */
 833         union vxlan_addr ip;
 834         __be16 port;
 835         __be32 vni;
 836         u32 ifindex;
 837         int err;
 838
 839         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 840                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 841                         ndm->ndm_state);
 842                 return -EINVAL;
 843         }
 844
 845         if (tb[NDA_DST] == NULL)
 846                 return -EINVAL;
 847
 848         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 849         if (err)
 850                 return err;
 851
 852         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 853                 return -EAFNOSUPPORT;
 854
 855         spin_lock_bh(&vxlan->hash_lock);
 856         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 857                                port, vni, ifindex, ndm->ndm_flags);
 858         spin_unlock_bh(&vxlan->hash_lock);
 859
 860         return err;
 861 }
 862
 863 /* Delete entry (via netlink) */
 864 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 865                             struct net_device *dev,
 866                             const unsigned char *addr, u16 vid)
 867 {
 868         struct vxlan_dev *vxlan = netdev_priv(dev);
 869         struct vxlan_fdb *f;
 870         struct vxlan_rdst *rd = NULL;
 871         union vxlan_addr ip;
 872         __be16 port;
 873         __be32 vni;
 874         u32 ifindex;
 875         int err;
 876
 877         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 878         if (err)
 879                 return err;
 880
 881         err = -ENOENT;
 882
 883         spin_lock_bh(&vxlan->hash_lock);
 884         f = vxlan_find_mac(vxlan, addr);
 885         if (!f)
 886                 goto out;
 887
 888         if (!vxlan_addr_any(&ip)) {
 889                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 890                 if (!rd)
 891                         goto out;
 892         }
 893
 894         err = 0;
 895
 896         /* remove a destination if it's not the only one on the list,
 897          * otherwise destroy the fdb entry
 898          */
 899         if (rd && !list_is_singular(&f->remotes)) {
 900                 list_del_rcu(&rd->list);
 901                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 902                 kfree_rcu(rd, rcu);
 903                 goto out;
 904         }
 905
 906         vxlan_fdb_destroy(vxlan, f);
 907
 908 out:
 909         spin_unlock_bh(&vxlan->hash_lock);
 910
 911         return err;
 912 }
 913
 914 /* Dump forwarding table */
 915 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 916                           struct net_device *dev,
 917                           struct net_device *filter_dev, int idx)
 918 {
 919         struct vxlan_dev *vxlan = netdev_priv(dev);
 920         unsigned int h;
 921
 922         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 923                 struct vxlan_fdb *f;
 924                 int err;
 925
 926                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 927                         struct vxlan_rdst *rd;
 928
 929                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 930                                 if (idx < cb->args[0])
 931                                         goto skip;
 932
 933                                 err = vxlan_fdb_info(skb, vxlan, f,
 934                                                      NETLINK_CB(cb->skb).portid,
 935                                                      cb->nlh->nlmsg_seq,
 936                                                      RTM_NEWNEIGH,
 937                                                      NLM_F_MULTI, rd);
 938                                 if (err < 0) {
 939                                         cb->args[1] = err;
 940                                         goto out;
 941                                 }
 942 skip:
 943                                 ++idx;
 944                         }
 945                 }
 946         }
 947 out:
 948         return idx;
 949 }
 950
 951 /* Watch incoming packets to learn mapping between Ethernet address
 952  * and Tunnel endpoint.
 953  * Return true if packet is bogus and should be dropped.
 954  */
 955 static bool vxlan_snoop(struct net_device *dev,
 956                         union vxlan_addr *src_ip, const u8 *src_mac)
 957 {
 958         struct vxlan_dev *vxlan = netdev_priv(dev);
 959         struct vxlan_fdb *f;
 960
 961         f = vxlan_find_mac(vxlan, src_mac);
 962         if (likely(f)) {
 963                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 964
 965                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 966                         return false;
 967
 968                 /* Don't migrate static entries, drop packets */
 969                 if (f->state & NUD_NOARP)
 970                         return true;
 971
 972                 if (net_ratelimit())
 973                         netdev_info(dev,
 974                                     "%pM migrated from %pIS to %pIS\n",
 975                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 976
 977                 rdst->remote_ip = *src_ip;
 978                 f->updated = jiffies;
 979                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 980         } else {
 981                 /* learned new entry */
 982                 spin_lock(&vxlan->hash_lock);
 983
 984                 /* close off race between vxlan_flush and incoming packets */
 985                 if (netif_running(dev))
 986                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 987                                          NUD_REACHABLE,
 988                                          NLM_F_EXCL|NLM_F_CREATE,
 989                                          vxlan->cfg.dst_port,
 990                                          vxlan->default_dst.remote_vni,
 991                                          0, NTF_SELF);
 992                 spin_unlock(&vxlan->hash_lock);
 993         }
 994
 995         return false;
 996 }
 997
 998 /* See if multicast group is already in use by other ID */
 999 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1000 {
1001         struct vxlan_dev *vxlan;
1002         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1003
1004         /* The vxlan_sock is only used by dev, leaving group has
1005          * no effect on other vxlan devices.
1006          */
1007         if (family == AF_INET && dev->vn4_sock &&
1008             atomic_read(&dev->vn4_sock->refcnt) == 1)
1009                 return false;
1010 #if IS_ENABLED(CONFIG_IPV6)
1011         if (family == AF_INET6 && dev->vn6_sock &&
1012             atomic_read(&dev->vn6_sock->refcnt) == 1)
1013                 return false;
1014 #endif
1015
1016         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1017                 if (!netif_running(vxlan->dev) || vxlan == dev)
1018                         continue;
1019
1020                 if (family == AF_INET && vxlan->vn4_sock != dev->vn4_sock)
1021                         continue;
1022 #if IS_ENABLED(CONFIG_IPV6)
1023                 if (family == AF_INET6 && vxlan->vn6_sock != dev->vn6_sock)
1024                         continue;
1025 #endif
1026
1027                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1028                                       &dev->default_dst.remote_ip))
1029                         continue;
1030
1031                 if (vxlan->default_dst.remote_ifindex !=
1032                     dev->default_dst.remote_ifindex)
1033                         continue;
1034
1035                 return true;
1036         }
1037
1038         return false;
1039 }
1040
1041 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1042 {
1043         struct vxlan_net *vn;
1044
1045         if (!vs)
1046                 return false;
1047         if (!atomic_dec_and_test(&vs->refcnt))
1048                 return false;
1049
1050         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1051         spin_lock(&vn->sock_lock);
1052         hlist_del_rcu(&vs->hlist);
1053         vxlan_notify_del_rx_port(vs);
1054         spin_unlock(&vn->sock_lock);
1055
1056         return true;
1057 }
1058
1059 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1060 {
1061         bool ipv4 = __vxlan_sock_release_prep(vxlan->vn4_sock);
1062 #if IS_ENABLED(CONFIG_IPV6)
1063         bool ipv6 = __vxlan_sock_release_prep(vxlan->vn6_sock);
1064 #endif
1065
1066         synchronize_net();
1067
1068         if (ipv4) {
1069                 udp_tunnel_sock_release(vxlan->vn4_sock->sock);
1070                 kfree(vxlan->vn4_sock);
1071         }
1072
1073 #if IS_ENABLED(CONFIG_IPV6)
1074         if (ipv6) {
1075                 udp_tunnel_sock_release(vxlan->vn6_sock->sock);
1076                 kfree(vxlan->vn6_sock);
1077         }
1078 #endif
1079 }
1080
1081 /* Update multicast group membership when first VNI on
1082  * multicast address is brought up
1083  */
1084 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1085 {
1086         struct sock *sk;
1087         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1088         int ifindex = vxlan->default_dst.remote_ifindex;
1089         int ret = -EINVAL;
1090
1091         if (ip->sa.sa_family == AF_INET) {
1092                 struct ip_mreqn mreq = {
1093                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1094                         .imr_ifindex            = ifindex,
1095                 };
1096
1097                 sk = vxlan->vn4_sock->sock->sk;
1098                 lock_sock(sk);
1099                 ret = ip_mc_join_group(sk, &mreq);
1100                 release_sock(sk);
1101 #if IS_ENABLED(CONFIG_IPV6)
1102         } else {
1103                 sk = vxlan->vn6_sock->sock->sk;
1104                 lock_sock(sk);
1105                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1106                                                    &ip->sin6.sin6_addr);
1107                 release_sock(sk);
1108 #endif
1109         }
1110
1111         return ret;
1112 }
1113
1114 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1115 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1116 {
1117         struct sock *sk;
1118         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1119         int ifindex = vxlan->default_dst.remote_ifindex;
1120         int ret = -EINVAL;
1121
1122         if (ip->sa.sa_family == AF_INET) {
1123                 struct ip_mreqn mreq = {
1124                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1125                         .imr_ifindex            = ifindex,
1126                 };
1127
1128                 sk = vxlan->vn4_sock->sock->sk;
1129                 lock_sock(sk);
1130                 ret = ip_mc_leave_group(sk, &mreq);
1131                 release_sock(sk);
1132 #if IS_ENABLED(CONFIG_IPV6)
1133         } else {
1134                 sk = vxlan->vn6_sock->sock->sk;
1135                 lock_sock(sk);
1136                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1137                                                    &ip->sin6.sin6_addr);
1138                 release_sock(sk);
1139 #endif
1140         }
1141
1142         return ret;
1143 }
1144
1145 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1146                           struct sk_buff *skb, u32 vxflags)
1147 {
1148         size_t start, offset;
1149
1150         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1151                 goto out;
1152
1153         start = vxlan_rco_start(unparsed->vx_vni);
1154         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1155
1156         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1157                 return false;
1158
1159         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1160                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1161 out:
1162         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1163         unparsed->vx_vni &= VXLAN_VNI_MASK;
1164         return true;
1165 }
1166
1167 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1168                                 struct sk_buff *skb, u32 vxflags,
1169                                 struct vxlan_metadata *md)
1170 {
1171         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1172         struct metadata_dst *tun_dst;
1173
1174         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1175                 goto out;
1176
1177         md->gbp = ntohs(gbp->policy_id);
1178
1179         tun_dst = (struct metadata_dst *)skb_dst(skb);
1180         if (tun_dst) {
1181                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1182                 tun_dst->u.tun_info.options_len = sizeof(*md);
1183         }
1184         if (gbp->dont_learn)
1185                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1186
1187         if (gbp->policy_applied)
1188                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1189
1190         /* In flow-based mode, GBP is carried in dst_metadata */
1191         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1192                 skb->mark = md->gbp;
1193 out:
1194         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1195 }
1196
1197 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1198                                 __be16 *protocol,
1199                                 struct sk_buff *skb, u32 vxflags)
1200 {
1201         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1202
1203         /* Need to have Next Protocol set for interfaces in GPE mode. */
1204         if (!gpe->np_applied)
1205                 return false;
1206         /* "The initial version is 0. If a receiver does not support the
1207          * version indicated it MUST drop the packet.
1208          */
1209         if (gpe->version != 0)
1210                 return false;
1211         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1212          * processing MUST occur." However, we don't implement OAM
1213          * processing, thus drop the packet.
1214          */
1215         if (gpe->oam_flag)
1216                 return false;
1217
1218         switch (gpe->next_protocol) {
1219         case VXLAN_GPE_NP_IPV4:
1220                 *protocol = htons(ETH_P_IP);
1221                 break;
1222         case VXLAN_GPE_NP_IPV6:
1223                 *protocol = htons(ETH_P_IPV6);
1224                 break;
1225         case VXLAN_GPE_NP_ETHERNET:
1226                 *protocol = htons(ETH_P_TEB);
1227                 break;
1228         default:
1229                 return false;
1230         }
1231
1232         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1233         return true;
1234 }
1235
1236 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1237                           struct vxlan_sock *vs,
1238                           struct sk_buff *skb)
1239 {
1240         union vxlan_addr saddr;
1241
1242         skb_reset_mac_header(skb);
1243         skb->protocol = eth_type_trans(skb, vxlan->dev);
1244         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1245
1246         /* Ignore packet loops (and multicast echo) */
1247         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1248                 return false;
1249
1250         /* Get address from the outer IP header */
1251         if (vxlan_get_sk_family(vs) == AF_INET) {
1252                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1253                 saddr.sa.sa_family = AF_INET;
1254 #if IS_ENABLED(CONFIG_IPV6)
1255         } else {
1256                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1257                 saddr.sa.sa_family = AF_INET6;
1258 #endif
1259         }
1260
1261         if ((vxlan->flags & VXLAN_F_LEARN) &&
1262             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1263                 return false;
1264
1265         return true;
1266 }
1267
1268 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1269                                   struct sk_buff *skb)
1270 {
1271         int err = 0;
1272
1273         if (vxlan_get_sk_family(vs) == AF_INET)
1274                 err = IP_ECN_decapsulate(oiph, skb);
1275 #if IS_ENABLED(CONFIG_IPV6)
1276         else
1277                 err = IP6_ECN_decapsulate(oiph, skb);
1278 #endif
1279
1280         if (unlikely(err) && log_ecn_error) {
1281                 if (vxlan_get_sk_family(vs) == AF_INET)
1282                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1283                                              &((struct iphdr *)oiph)->saddr,
1284                                              ((struct iphdr *)oiph)->tos);
1285                 else
1286                         net_info_ratelimited("non-ECT from %pI6\n",
1287                                              &((struct ipv6hdr *)oiph)->saddr);
1288         }
1289         return err <= 1;
1290 }
1291
1292 /* Callback from net/ipv4/udp.c to receive packets */
1293 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1294 {
1295         struct pcpu_sw_netstats *stats;
1296         struct vxlan_dev *vxlan;
1297         struct vxlan_sock *vs;
1298         struct vxlanhdr unparsed;
1299         struct vxlan_metadata _md;
1300         struct vxlan_metadata *md = &_md;
1301         __be16 protocol = htons(ETH_P_TEB);
1302         bool raw_proto = false;
1303         void *oiph;
1304
1305         /* Need UDP and VXLAN header to be present */
1306         if (!pskb_may_pull(skb, VXLAN_HLEN))
1307                 goto drop;
1308
1309         unparsed = *vxlan_hdr(skb);
1310         /* VNI flag always required to be set */
1311         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1312                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1313                            ntohl(vxlan_hdr(skb)->vx_flags),
1314                            ntohl(vxlan_hdr(skb)->vx_vni));
1315                 /* Return non vxlan pkt */
1316                 goto drop;
1317         }
1318         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1319         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1320
1321         vs = rcu_dereference_sk_user_data(sk);
1322         if (!vs)
1323                 goto drop;
1324
1325         vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1326         if (!vxlan)
1327                 goto drop;
1328
1329         /* For backwards compatibility, only allow reserved fields to be
1330          * used by VXLAN extensions if explicitly requested.
1331          */
1332         if (vs->flags & VXLAN_F_GPE) {
1333                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1334                         goto drop;
1335                 raw_proto = true;
1336         }
1337
1338         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1339                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1340                         goto drop;
1341
1342         if (vxlan_collect_metadata(vs)) {
1343                 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1344                 struct metadata_dst *tun_dst;
1345
1346                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1347                                          vxlan_vni_to_tun_id(vni), sizeof(*md));
1348
1349                 if (!tun_dst)
1350                         goto drop;
1351
1352                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1353
1354                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1355         } else {
1356                 memset(md, 0, sizeof(*md));
1357         }
1358
1359         if (vs->flags & VXLAN_F_REMCSUM_RX)
1360                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1361                         goto drop;
1362         if (vs->flags & VXLAN_F_GBP)
1363                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1364         /* Note that GBP and GPE can never be active together. This is
1365          * ensured in vxlan_dev_configure.
1366          */
1367
1368         if (unparsed.vx_flags || unparsed.vx_vni) {
1369                 /* If there are any unprocessed flags remaining treat
1370                  * this as a malformed packet. This behavior diverges from
1371                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1372                  * in reserved fields are to be ignored. The approach here
1373                  * maintains compatibility with previous stack code, and also
1374                  * is more robust and provides a little more security in
1375                  * adding extensions to VXLAN.
1376                  */
1377                 goto drop;
1378         }
1379
1380         if (!raw_proto) {
1381                 if (!vxlan_set_mac(vxlan, vs, skb))
1382                         goto drop;
1383         } else {
1384                 skb_reset_mac_header(skb);
1385                 skb->dev = vxlan->dev;
1386                 skb->pkt_type = PACKET_HOST;
1387         }
1388
1389         oiph = skb_network_header(skb);
1390         skb_reset_network_header(skb);
1391
1392         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1393                 ++vxlan->dev->stats.rx_frame_errors;
1394                 ++vxlan->dev->stats.rx_errors;
1395                 goto drop;
1396         }
1397
1398         stats = this_cpu_ptr(vxlan->dev->tstats);
1399         u64_stats_update_begin(&stats->syncp);
1400         stats->rx_packets++;
1401         stats->rx_bytes += skb->len;
1402         u64_stats_update_end(&stats->syncp);
1403
1404         gro_cells_receive(&vxlan->gro_cells, skb);
1405         return 0;
1406
1407 drop:
1408         /* Consume bad packet */
1409         kfree_skb(skb);
1410         return 0;
1411 }
1412
1413 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1414 {
1415         struct vxlan_dev *vxlan = netdev_priv(dev);
1416         struct arphdr *parp;
1417         u8 *arpptr, *sha;
1418         __be32 sip, tip;
1419         struct neighbour *n;
1420
1421         if (dev->flags & IFF_NOARP)
1422                 goto out;
1423
1424         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1425                 dev->stats.tx_dropped++;
1426                 goto out;
1427         }
1428         parp = arp_hdr(skb);
1429
1430         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1431              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1432             parp->ar_pro != htons(ETH_P_IP) ||
1433             parp->ar_op != htons(ARPOP_REQUEST) ||
1434             parp->ar_hln != dev->addr_len ||
1435             parp->ar_pln != 4)
1436                 goto out;
1437         arpptr = (u8 *)parp + sizeof(struct arphdr);
1438         sha = arpptr;
1439         arpptr += dev->addr_len;        /* sha */
1440         memcpy(&sip, arpptr, sizeof(sip));
1441         arpptr += sizeof(sip);
1442         arpptr += dev->addr_len;        /* tha */
1443         memcpy(&tip, arpptr, sizeof(tip));
1444
1445         if (ipv4_is_loopback(tip) ||
1446             ipv4_is_multicast(tip))
1447                 goto out;
1448
1449         n = neigh_lookup(&arp_tbl, &tip, dev);
1450
1451         if (n) {
1452                 struct vxlan_fdb *f;
1453                 struct sk_buff  *reply;
1454
1455                 if (!(n->nud_state & NUD_CONNECTED)) {
1456                         neigh_release(n);
1457                         goto out;
1458                 }
1459
1460                 f = vxlan_find_mac(vxlan, n->ha);
1461                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1462                         /* bridge-local neighbor */
1463                         neigh_release(n);
1464                         goto out;
1465                 }
1466
1467                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1468                                 n->ha, sha);
1469
1470                 neigh_release(n);
1471
1472                 if (reply == NULL)
1473                         goto out;
1474
1475                 skb_reset_mac_header(reply);
1476                 __skb_pull(reply, skb_network_offset(reply));
1477                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1478                 reply->pkt_type = PACKET_HOST;
1479
1480                 if (netif_rx_ni(reply) == NET_RX_DROP)
1481                         dev->stats.rx_dropped++;
1482         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1483                 union vxlan_addr ipa = {
1484                         .sin.sin_addr.s_addr = tip,
1485                         .sin.sin_family = AF_INET,
1486                 };
1487
1488                 vxlan_ip_miss(dev, &ipa);
1489         }
1490 out:
1491         consume_skb(skb);
1492         return NETDEV_TX_OK;
1493 }
1494
1495 #if IS_ENABLED(CONFIG_IPV6)
1496 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1497         struct neighbour *n, bool isrouter)
1498 {
1499         struct net_device *dev = request->dev;
1500         struct sk_buff *reply;
1501         struct nd_msg *ns, *na;
1502         struct ipv6hdr *pip6;
1503         u8 *daddr;
1504         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1505         int ns_olen;
1506         int i, len;
1507
1508         if (dev == NULL)
1509                 return NULL;
1510
1511         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1512                 sizeof(*na) + na_olen + dev->needed_tailroom;
1513         reply = alloc_skb(len, GFP_ATOMIC);
1514         if (reply == NULL)
1515                 return NULL;
1516
1517         reply->protocol = htons(ETH_P_IPV6);
1518         reply->dev = dev;
1519         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1520         skb_push(reply, sizeof(struct ethhdr));
1521         skb_reset_mac_header(reply);
1522
1523         ns = (struct nd_msg *)skb_transport_header(request);
1524
1525         daddr = eth_hdr(request)->h_source;
1526         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1527         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1528                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1529                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1530                         break;
1531                 }
1532         }
1533
1534         /* Ethernet header */
1535         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1536         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1537         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1538         reply->protocol = htons(ETH_P_IPV6);
1539
1540         skb_pull(reply, sizeof(struct ethhdr));
1541         skb_reset_network_header(reply);
1542         skb_put(reply, sizeof(struct ipv6hdr));
1543
1544         /* IPv6 header */
1545
1546         pip6 = ipv6_hdr(reply);
1547         memset(pip6, 0, sizeof(struct ipv6hdr));
1548         pip6->version = 6;
1549         pip6->priority = ipv6_hdr(request)->priority;
1550         pip6->nexthdr = IPPROTO_ICMPV6;
1551         pip6->hop_limit = 255;
1552         pip6->daddr = ipv6_hdr(request)->saddr;
1553         pip6->saddr = *(struct in6_addr *)n->primary_key;
1554
1555         skb_pull(reply, sizeof(struct ipv6hdr));
1556         skb_reset_transport_header(reply);
1557
1558         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1559
1560         /* Neighbor Advertisement */
1561         memset(na, 0, sizeof(*na)+na_olen);
1562         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1563         na->icmph.icmp6_router = isrouter;
1564         na->icmph.icmp6_override = 1;
1565         na->icmph.icmp6_solicited = 1;
1566         na->target = ns->target;
1567         ether_addr_copy(&na->opt[2], n->ha);
1568         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1569         na->opt[1] = na_olen >> 3;
1570
1571         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1572                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1573                 csum_partial(na, sizeof(*na)+na_olen, 0));
1574
1575         pip6->payload_len = htons(sizeof(*na)+na_olen);
1576
1577         skb_push(reply, sizeof(struct ipv6hdr));
1578
1579         reply->ip_summed = CHECKSUM_UNNECESSARY;
1580
1581         return reply;
1582 }
1583
1584 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1585 {
1586         struct vxlan_dev *vxlan = netdev_priv(dev);
1587         struct nd_msg *msg;
1588         const struct ipv6hdr *iphdr;
1589         const struct in6_addr *saddr, *daddr;
1590         struct neighbour *n;
1591         struct inet6_dev *in6_dev;
1592
1593         in6_dev = __in6_dev_get(dev);
1594         if (!in6_dev)
1595                 goto out;
1596
1597         iphdr = ipv6_hdr(skb);
1598         saddr = &iphdr->saddr;
1599         daddr = &iphdr->daddr;
1600
1601         msg = (struct nd_msg *)skb_transport_header(skb);
1602         if (msg->icmph.icmp6_code != 0 ||
1603             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1604                 goto out;
1605
1606         if (ipv6_addr_loopback(daddr) ||
1607             ipv6_addr_is_multicast(&msg->target))
1608                 goto out;
1609
1610         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1611
1612         if (n) {
1613                 struct vxlan_fdb *f;
1614                 struct sk_buff *reply;
1615
1616                 if (!(n->nud_state & NUD_CONNECTED)) {
1617                         neigh_release(n);
1618                         goto out;
1619                 }
1620
1621                 f = vxlan_find_mac(vxlan, n->ha);
1622                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1623                         /* bridge-local neighbor */
1624                         neigh_release(n);
1625                         goto out;
1626                 }
1627
1628                 reply = vxlan_na_create(skb, n,
1629                                         !!(f ? f->flags & NTF_ROUTER : 0));
1630
1631                 neigh_release(n);
1632
1633                 if (reply == NULL)
1634                         goto out;
1635
1636                 if (netif_rx_ni(reply) == NET_RX_DROP)
1637                         dev->stats.rx_dropped++;
1638
1639         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1640                 union vxlan_addr ipa = {
1641                         .sin6.sin6_addr = msg->target,
1642                         .sin6.sin6_family = AF_INET6,
1643                 };
1644
1645                 vxlan_ip_miss(dev, &ipa);
1646         }
1647
1648 out:
1649         consume_skb(skb);
1650         return NETDEV_TX_OK;
1651 }
1652 #endif
1653
1654 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1655 {
1656         struct vxlan_dev *vxlan = netdev_priv(dev);
1657         struct neighbour *n;
1658
1659         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1660                 return false;
1661
1662         n = NULL;
1663         switch (ntohs(eth_hdr(skb)->h_proto)) {
1664         case ETH_P_IP:
1665         {
1666                 struct iphdr *pip;
1667
1668                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1669                         return false;
1670                 pip = ip_hdr(skb);
1671                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1672                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1673                         union vxlan_addr ipa = {
1674                                 .sin.sin_addr.s_addr = pip->daddr,
1675                                 .sin.sin_family = AF_INET,
1676                         };
1677
1678                         vxlan_ip_miss(dev, &ipa);
1679                         return false;
1680                 }
1681
1682                 break;
1683         }
1684 #if IS_ENABLED(CONFIG_IPV6)
1685         case ETH_P_IPV6:
1686         {
1687                 struct ipv6hdr *pip6;
1688
1689                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1690                         return false;
1691                 pip6 = ipv6_hdr(skb);
1692                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1693                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1694                         union vxlan_addr ipa = {
1695                                 .sin6.sin6_addr = pip6->daddr,
1696                                 .sin6.sin6_family = AF_INET6,
1697                         };
1698
1699                         vxlan_ip_miss(dev, &ipa);
1700                         return false;
1701                 }
1702
1703                 break;
1704         }
1705 #endif
1706         default:
1707                 return false;
1708         }
1709
1710         if (n) {
1711                 bool diff;
1712
1713                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1714                 if (diff) {
1715                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1716                                 dev->addr_len);
1717                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1718                 }
1719                 neigh_release(n);
1720                 return diff;
1721         }
1722
1723         return false;
1724 }
1725
1726 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1727                                 struct vxlan_metadata *md)
1728 {
1729         struct vxlanhdr_gbp *gbp;
1730
1731         if (!md->gbp)
1732                 return;
1733
1734         gbp = (struct vxlanhdr_gbp *)vxh;
1735         vxh->vx_flags |= VXLAN_HF_GBP;
1736
1737         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1738                 gbp->dont_learn = 1;
1739
1740         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1741                 gbp->policy_applied = 1;
1742
1743         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1744 }
1745
1746 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1747                                __be16 protocol)
1748 {
1749         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1750
1751         gpe->np_applied = 1;
1752
1753         switch (protocol) {
1754         case htons(ETH_P_IP):
1755                 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1756                 return 0;
1757         case htons(ETH_P_IPV6):
1758                 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1759                 return 0;
1760         case htons(ETH_P_TEB):
1761                 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1762                 return 0;
1763         }
1764         return -EPFNOSUPPORT;
1765 }
1766
1767 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1768                            int iphdr_len, __be32 vni,
1769                            struct vxlan_metadata *md, u32 vxflags,
1770                            bool udp_sum)
1771 {
1772         struct vxlanhdr *vxh;
1773         int min_headroom;
1774         int err;
1775         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1776         __be16 inner_protocol = htons(ETH_P_TEB);
1777
1778         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1779             skb->ip_summed == CHECKSUM_PARTIAL) {
1780                 int csum_start = skb_checksum_start_offset(skb);
1781
1782                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1783                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1784                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1785                      skb->csum_offset == offsetof(struct tcphdr, check)))
1786                         type |= SKB_GSO_TUNNEL_REMCSUM;
1787         }
1788
1789         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1790                         + VXLAN_HLEN + iphdr_len
1791                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1792
1793         /* Need space for new headers (invalidates iph ptr) */
1794         err = skb_cow_head(skb, min_headroom);
1795         if (unlikely(err))
1796                 goto out_free;
1797
1798         skb = vlan_hwaccel_push_inside(skb);
1799         if (WARN_ON(!skb))
1800                 return -ENOMEM;
1801
1802         err = iptunnel_handle_offloads(skb, type);
1803         if (err)
1804                 goto out_free;
1805
1806         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1807         vxh->vx_flags = VXLAN_HF_VNI;
1808         vxh->vx_vni = vxlan_vni_field(vni);
1809
1810         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1811                 unsigned int start;
1812
1813                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1814                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1815                 vxh->vx_flags |= VXLAN_HF_RCO;
1816
1817                 if (!skb_is_gso(skb)) {
1818                         skb->ip_summed = CHECKSUM_NONE;
1819                         skb->encapsulation = 0;
1820                 }
1821         }
1822
1823         if (vxflags & VXLAN_F_GBP)
1824                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1825         if (vxflags & VXLAN_F_GPE) {
1826                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1827                 if (err < 0)
1828                         goto out_free;
1829                 inner_protocol = skb->protocol;
1830         }
1831
1832         skb_set_inner_protocol(skb, inner_protocol);
1833         return 0;
1834
1835 out_free:
1836         kfree_skb(skb);
1837         return err;
1838 }
1839
1840 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1841                                       struct sk_buff *skb, int oif, u8 tos,
1842                                       __be32 daddr, __be32 *saddr,
1843                                       struct dst_cache *dst_cache,
1844                                       const struct ip_tunnel_info *info)
1845 {
1846         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1847         struct rtable *rt = NULL;
1848         struct flowi4 fl4;
1849
1850         if (tos && !info)
1851                 use_cache = false;
1852         if (use_cache) {
1853                 rt = dst_cache_get_ip4(dst_cache, saddr);
1854                 if (rt)
1855                         return rt;
1856         }
1857
1858         memset(&fl4, 0, sizeof(fl4));
1859         fl4.flowi4_oif = oif;
1860         fl4.flowi4_tos = RT_TOS(tos);
1861         fl4.flowi4_mark = skb->mark;
1862         fl4.flowi4_proto = IPPROTO_UDP;
1863         fl4.daddr = daddr;
1864         fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
1865
1866         rt = ip_route_output_key(vxlan->net, &fl4);
1867         if (!IS_ERR(rt)) {
1868                 *saddr = fl4.saddr;
1869                 if (use_cache)
1870                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1871         }
1872         return rt;
1873 }
1874
1875 #if IS_ENABLED(CONFIG_IPV6)
1876 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1877                                           struct sk_buff *skb, int oif, u8 tos,
1878                                           __be32 label,
1879                                           const struct in6_addr *daddr,
1880                                           struct in6_addr *saddr,
1881                                           struct dst_cache *dst_cache,
1882                                           const struct ip_tunnel_info *info)
1883 {
1884         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1885         struct dst_entry *ndst;
1886         struct flowi6 fl6;
1887         int err;
1888
1889         if (tos && !info)
1890                 use_cache = false;
1891         if (use_cache) {
1892                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1893                 if (ndst)
1894                         return ndst;
1895         }
1896
1897         memset(&fl6, 0, sizeof(fl6));
1898         fl6.flowi6_oif = oif;
1899         fl6.daddr = *daddr;
1900         fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
1901         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1902         fl6.flowi6_mark = skb->mark;
1903         fl6.flowi6_proto = IPPROTO_UDP;
1904
1905         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1906                                          vxlan->vn6_sock->sock->sk,
1907                                          &ndst, &fl6);
1908         if (err < 0)
1909                 return ERR_PTR(err);
1910
1911         *saddr = fl6.saddr;
1912         if (use_cache)
1913                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1914         return ndst;
1915 }
1916 #endif
1917
1918 /* Bypass encapsulation if the destination is local */
1919 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1920                                struct vxlan_dev *dst_vxlan)
1921 {
1922         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1923         union vxlan_addr loopback;
1924         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1925         struct net_device *dev = skb->dev;
1926         int len = skb->len;
1927
1928         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1929         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1930         skb->pkt_type = PACKET_HOST;
1931         skb->encapsulation = 0;
1932         skb->dev = dst_vxlan->dev;
1933         __skb_pull(skb, skb_network_offset(skb));
1934
1935         if (remote_ip->sa.sa_family == AF_INET) {
1936                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1937                 loopback.sa.sa_family =  AF_INET;
1938 #if IS_ENABLED(CONFIG_IPV6)
1939         } else {
1940                 loopback.sin6.sin6_addr = in6addr_loopback;
1941                 loopback.sa.sa_family =  AF_INET6;
1942 #endif
1943         }
1944
1945         if (dst_vxlan->flags & VXLAN_F_LEARN)
1946                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1947
1948         u64_stats_update_begin(&tx_stats->syncp);
1949         tx_stats->tx_packets++;
1950         tx_stats->tx_bytes += len;
1951         u64_stats_update_end(&tx_stats->syncp);
1952
1953         if (netif_rx(skb) == NET_RX_SUCCESS) {
1954                 u64_stats_update_begin(&rx_stats->syncp);
1955                 rx_stats->rx_packets++;
1956                 rx_stats->rx_bytes += len;
1957                 u64_stats_update_end(&rx_stats->syncp);
1958         } else {
1959                 dev->stats.rx_dropped++;
1960         }
1961 }
1962
1963 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1964                            struct vxlan_rdst *rdst, bool did_rsc)
1965 {
1966         struct dst_cache *dst_cache;
1967         struct ip_tunnel_info *info;
1968         struct vxlan_dev *vxlan = netdev_priv(dev);
1969         struct sock *sk;
1970         struct rtable *rt = NULL;
1971         const struct iphdr *old_iph;
1972         union vxlan_addr *dst;
1973         union vxlan_addr remote_ip;
1974         struct vxlan_metadata _md;
1975         struct vxlan_metadata *md = &_md;
1976         __be16 src_port = 0, dst_port;
1977         __be32 vni, label;
1978         __be16 df = 0;
1979         __u8 tos, ttl;
1980         int err;
1981         u32 flags = vxlan->flags;
1982         bool udp_sum = false;
1983         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1984
1985         info = skb_tunnel_info(skb);
1986
1987         if (rdst) {
1988                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1989                 vni = rdst->remote_vni;
1990                 dst = &rdst->remote_ip;
1991                 dst_cache = &rdst->dst_cache;
1992         } else {
1993                 if (!info) {
1994                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1995                                   dev->name);
1996                         goto drop;
1997                 }
1998                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1999                 vni = vxlan_tun_id_to_vni(info->key.tun_id);
2000                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2001                 if (remote_ip.sa.sa_family == AF_INET)
2002                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2003                 else
2004                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2005                 dst = &remote_ip;
2006                 dst_cache = &info->dst_cache;
2007         }
2008
2009         if (vxlan_addr_any(dst)) {
2010                 if (did_rsc) {
2011                         /* short-circuited back to local bridge */
2012                         vxlan_encap_bypass(skb, vxlan, vxlan);
2013                         return;
2014                 }
2015                 goto drop;
2016         }
2017
2018         old_iph = ip_hdr(skb);
2019
2020         ttl = vxlan->cfg.ttl;
2021         if (!ttl && vxlan_addr_multicast(dst))
2022                 ttl = 1;
2023
2024         tos = vxlan->cfg.tos;
2025         if (tos == 1)
2026                 tos = ip_tunnel_get_dsfield(old_iph, skb);
2027
2028         label = vxlan->cfg.label;
2029         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2030                                      vxlan->cfg.port_max, true);
2031
2032         if (info) {
2033                 ttl = info->key.ttl;
2034                 tos = info->key.tos;
2035                 label = info->key.label;
2036                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2037
2038                 if (info->options_len)
2039                         md = ip_tunnel_info_opts(info);
2040         } else {
2041                 md->gbp = skb->mark;
2042         }
2043
2044         if (dst->sa.sa_family == AF_INET) {
2045                 __be32 saddr;
2046
2047                 if (!vxlan->vn4_sock)
2048                         goto drop;
2049                 sk = vxlan->vn4_sock->sock->sk;
2050
2051                 rt = vxlan_get_route(vxlan, skb,
2052                                      rdst ? rdst->remote_ifindex : 0, tos,
2053                                      dst->sin.sin_addr.s_addr, &saddr,
2054                                      dst_cache, info);
2055                 if (IS_ERR(rt)) {
2056                         netdev_dbg(dev, "no route to %pI4\n",
2057                                    &dst->sin.sin_addr.s_addr);
2058                         dev->stats.tx_carrier_errors++;
2059                         goto tx_error;
2060                 }
2061
2062                 if (rt->dst.dev == dev) {
2063                         netdev_dbg(dev, "circular route to %pI4\n",
2064                                    &dst->sin.sin_addr.s_addr);
2065                         dev->stats.collisions++;
2066                         goto rt_tx_error;
2067                 }
2068
2069                 /* Bypass encapsulation if the destination is local */
2070                 if (rt->rt_flags & RTCF_LOCAL &&
2071                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2072                         struct vxlan_dev *dst_vxlan;
2073
2074                         ip_rt_put(rt);
2075                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2076                                                    dst->sa.sa_family, dst_port,
2077                                                    vxlan->flags);
2078                         if (!dst_vxlan)
2079                                 goto tx_error;
2080                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2081                         return;
2082                 }
2083
2084                 if (!info)
2085                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2086                 else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2087                         df = htons(IP_DF);
2088
2089                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2090                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2091                 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2092                                       vni, md, flags, udp_sum);
2093                 if (err < 0)
2094                         goto xmit_tx_error;
2095
2096                 udp_tunnel_xmit_skb(rt, sk, skb, saddr,
2097                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2098                                     src_port, dst_port, xnet, !udp_sum);
2099 #if IS_ENABLED(CONFIG_IPV6)
2100         } else {
2101                 struct dst_entry *ndst;
2102                 struct in6_addr saddr;
2103                 u32 rt6i_flags;
2104
2105                 if (!vxlan->vn6_sock)
2106                         goto drop;
2107                 sk = vxlan->vn6_sock->sock->sk;
2108
2109                 ndst = vxlan6_get_route(vxlan, skb,
2110                                         rdst ? rdst->remote_ifindex : 0, tos,
2111                                         label, &dst->sin6.sin6_addr, &saddr,
2112                                         dst_cache, info);
2113                 if (IS_ERR(ndst)) {
2114                         netdev_dbg(dev, "no route to %pI6\n",
2115                                    &dst->sin6.sin6_addr);
2116                         dev->stats.tx_carrier_errors++;
2117                         goto tx_error;
2118                 }
2119
2120                 if (ndst->dev == dev) {
2121                         netdev_dbg(dev, "circular route to %pI6\n",
2122                                    &dst->sin6.sin6_addr);
2123                         dst_release(ndst);
2124                         dev->stats.collisions++;
2125                         goto tx_error;
2126                 }
2127
2128                 /* Bypass encapsulation if the destination is local */
2129                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2130                 if (rt6i_flags & RTF_LOCAL &&
2131                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2132                         struct vxlan_dev *dst_vxlan;
2133
2134                         dst_release(ndst);
2135                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2136                                                    dst->sa.sa_family, dst_port,
2137                                                    vxlan->flags);
2138                         if (!dst_vxlan)
2139                                 goto tx_error;
2140                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2141                         return;
2142                 }
2143
2144                 if (!info)
2145                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2146
2147                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2148                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2149                 skb_scrub_packet(skb, xnet);
2150                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2151                                       vni, md, flags, udp_sum);
2152                 if (err < 0) {
2153                         dst_release(ndst);
2154                         return;
2155                 }
2156                 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2157                                      &saddr, &dst->sin6.sin6_addr, tos, ttl,
2158                                      label, src_port, dst_port, !udp_sum);
2159 #endif
2160         }
2161
2162         return;
2163
2164 drop:
2165         dev->stats.tx_dropped++;
2166         goto tx_free;
2167
2168 xmit_tx_error:
2169         /* skb is already freed. */
2170         skb = NULL;
2171 rt_tx_error:
2172         ip_rt_put(rt);
2173 tx_error:
2174         dev->stats.tx_errors++;
2175 tx_free:
2176         dev_kfree_skb(skb);
2177 }
2178
2179 /* Transmit local packets over Vxlan
2180  *
2181  * Outer IP header inherits ECN and DF from inner header.
2182  * Outer UDP destination is the VXLAN assigned port.
2183  *           source port is based on hash of flow
2184  */
2185 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2186 {
2187         struct vxlan_dev *vxlan = netdev_priv(dev);
2188         const struct ip_tunnel_info *info;
2189         struct ethhdr *eth;
2190         bool did_rsc = false;
2191         struct vxlan_rdst *rdst, *fdst = NULL;
2192         struct vxlan_fdb *f;
2193
2194         info = skb_tunnel_info(skb);
2195
2196         skb_reset_mac_header(skb);
2197
2198         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2199                 if (info && info->mode & IP_TUNNEL_INFO_TX)
2200                         vxlan_xmit_one(skb, dev, NULL, false);
2201                 else
2202                         kfree_skb(skb);
2203                 return NETDEV_TX_OK;
2204         }
2205
2206         if (vxlan->flags & VXLAN_F_PROXY) {
2207                 eth = eth_hdr(skb);
2208                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2209                         return arp_reduce(dev, skb);
2210 #if IS_ENABLED(CONFIG_IPV6)
2211                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2212                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2213                                        + sizeof(struct nd_msg)) &&
2214                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2215                                 struct nd_msg *msg;
2216
2217                                 msg = (struct nd_msg *)skb_transport_header(skb);
2218                                 if (msg->icmph.icmp6_code == 0 &&
2219                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2220                                         return neigh_reduce(dev, skb);
2221                 }
2222 #endif
2223         }
2224
2225         eth = eth_hdr(skb);
2226         f = vxlan_find_mac(vxlan, eth->h_dest);
2227         did_rsc = false;
2228
2229         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2230             (ntohs(eth->h_proto) == ETH_P_IP ||
2231              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2232                 did_rsc = route_shortcircuit(dev, skb);
2233                 if (did_rsc)
2234                         f = vxlan_find_mac(vxlan, eth->h_dest);
2235         }
2236
2237         if (f == NULL) {
2238                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2239                 if (f == NULL) {
2240                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2241                             !is_multicast_ether_addr(eth->h_dest))
2242                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2243
2244                         dev->stats.tx_dropped++;
2245                         kfree_skb(skb);
2246                         return NETDEV_TX_OK;
2247                 }
2248         }
2249
2250         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2251                 struct sk_buff *skb1;
2252
2253                 if (!fdst) {
2254                         fdst = rdst;
2255                         continue;
2256                 }
2257                 skb1 = skb_clone(skb, GFP_ATOMIC);
2258                 if (skb1)
2259                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2260         }
2261
2262         if (fdst)
2263                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2264         else
2265                 kfree_skb(skb);
2266         return NETDEV_TX_OK;
2267 }
2268
2269 /* Walk the forwarding table and purge stale entries */
2270 static void vxlan_cleanup(unsigned long arg)
2271 {
2272         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2273         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2274         unsigned int h;
2275
2276         if (!netif_running(vxlan->dev))
2277                 return;
2278
2279         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2280                 struct hlist_node *p, *n;
2281
2282                 spin_lock_bh(&vxlan->hash_lock);
2283                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2284                         struct vxlan_fdb *f
2285                                 = container_of(p, struct vxlan_fdb, hlist);
2286                         unsigned long timeout;
2287
2288                         if (f->state & NUD_PERMANENT)
2289                                 continue;
2290
2291                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2292                         if (time_before_eq(timeout, jiffies)) {
2293                                 netdev_dbg(vxlan->dev,
2294                                            "garbage collect %pM\n",
2295                                            f->eth_addr);
2296                                 f->state = NUD_STALE;
2297                                 vxlan_fdb_destroy(vxlan, f);
2298                         } else if (time_before(timeout, next_timer))
2299                                 next_timer = timeout;
2300                 }
2301                 spin_unlock_bh(&vxlan->hash_lock);
2302         }
2303
2304         mod_timer(&vxlan->age_timer, next_timer);
2305 }
2306
2307 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2308 {
2309         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2310         __be32 vni = vxlan->default_dst.remote_vni;
2311
2312         spin_lock(&vn->sock_lock);
2313         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2314         spin_unlock(&vn->sock_lock);
2315 }
2316
2317 /* Setup stats when device is created */
2318 static int vxlan_init(struct net_device *dev)
2319 {
2320         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2321         if (!dev->tstats)
2322                 return -ENOMEM;
2323
2324         return 0;
2325 }
2326
2327 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2328 {
2329         struct vxlan_fdb *f;
2330
2331         spin_lock_bh(&vxlan->hash_lock);
2332         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2333         if (f)
2334                 vxlan_fdb_destroy(vxlan, f);
2335         spin_unlock_bh(&vxlan->hash_lock);
2336 }
2337
2338 static void vxlan_uninit(struct net_device *dev)
2339 {
2340         struct vxlan_dev *vxlan = netdev_priv(dev);
2341
2342         vxlan_fdb_delete_default(vxlan);
2343
2344         free_percpu(dev->tstats);
2345 }
2346
2347 /* Start ageing timer and join group when device is brought up */
2348 static int vxlan_open(struct net_device *dev)
2349 {
2350         struct vxlan_dev *vxlan = netdev_priv(dev);
2351         int ret;
2352
2353         ret = vxlan_sock_add(vxlan);
2354         if (ret < 0)
2355                 return ret;
2356
2357         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2358                 ret = vxlan_igmp_join(vxlan);
2359                 if (ret == -EADDRINUSE)
2360                         ret = 0;
2361                 if (ret) {
2362                         vxlan_sock_release(vxlan);
2363                         return ret;
2364                 }
2365         }
2366
2367         if (vxlan->cfg.age_interval)
2368                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2369
2370         return ret;
2371 }
2372
2373 /* Purge the forwarding table */
2374 static void vxlan_flush(struct vxlan_dev *vxlan)
2375 {
2376         unsigned int h;
2377
2378         spin_lock_bh(&vxlan->hash_lock);
2379         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2380                 struct hlist_node *p, *n;
2381                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2382                         struct vxlan_fdb *f
2383                                 = container_of(p, struct vxlan_fdb, hlist);
2384                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2385                         if (!is_zero_ether_addr(f->eth_addr))
2386                                 vxlan_fdb_destroy(vxlan, f);
2387                 }
2388         }
2389         spin_unlock_bh(&vxlan->hash_lock);
2390 }
2391
2392 /* Cleanup timer and forwarding table on shutdown */
2393 static int vxlan_stop(struct net_device *dev)
2394 {
2395         struct vxlan_dev *vxlan = netdev_priv(dev);
2396         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2397         int ret = 0;
2398
2399         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2400             !vxlan_group_used(vn, vxlan))
2401                 ret = vxlan_igmp_leave(vxlan);
2402
2403         del_timer_sync(&vxlan->age_timer);
2404
2405         vxlan_flush(vxlan);
2406         vxlan_sock_release(vxlan);
2407
2408         return ret;
2409 }
2410
2411 /* Stub, nothing needs to be done. */
2412 static void vxlan_set_multicast_list(struct net_device *dev)
2413 {
2414 }
2415
2416 static int __vxlan_change_mtu(struct net_device *dev,
2417                               struct net_device *lowerdev,
2418                               struct vxlan_rdst *dst, int new_mtu, bool strict)
2419 {
2420         int max_mtu = IP_MAX_MTU;
2421
2422         if (lowerdev)
2423                 max_mtu = lowerdev->mtu;
2424
2425         if (dst->remote_ip.sa.sa_family == AF_INET6)
2426                 max_mtu -= VXLAN6_HEADROOM;
2427         else
2428                 max_mtu -= VXLAN_HEADROOM;
2429
2430         if (new_mtu < 68)
2431                 return -EINVAL;
2432
2433         if (new_mtu > max_mtu) {
2434                 if (strict)
2435                         return -EINVAL;
2436
2437                 new_mtu = max_mtu;
2438         }
2439
2440         dev->mtu = new_mtu;
2441         return 0;
2442 }
2443
2444 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2445 {
2446         struct vxlan_dev *vxlan = netdev_priv(dev);
2447         struct vxlan_rdst *dst = &vxlan->default_dst;
2448         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2449                                                          dst->remote_ifindex);
2450         return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2451 }
2452
2453 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2454 {
2455         struct vxlan_dev *vxlan = netdev_priv(dev);
2456         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2457         __be16 sport, dport;
2458
2459         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2460                                   vxlan->cfg.port_max, true);
2461         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2462
2463         if (ip_tunnel_info_af(info) == AF_INET) {
2464                 struct rtable *rt;
2465
2466                 if (!vxlan->vn4_sock)
2467                         return -EINVAL;
2468                 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2469                                      info->key.u.ipv4.dst,
2470                                      &info->key.u.ipv4.src, NULL, info);
2471                 if (IS_ERR(rt))
2472                         return PTR_ERR(rt);
2473                 ip_rt_put(rt);
2474         } else {
2475 #if IS_ENABLED(CONFIG_IPV6)
2476                 struct dst_entry *ndst;
2477
2478                 if (!vxlan->vn6_sock)
2479                         return -EINVAL;
2480                 ndst = vxlan6_get_route(vxlan, skb, 0, info->key.tos,
2481                                         info->key.label, &info->key.u.ipv6.dst,
2482                                         &info->key.u.ipv6.src, NULL, info);
2483                 if (IS_ERR(ndst))
2484                         return PTR_ERR(ndst);
2485                 dst_release(ndst);
2486 #else /* !CONFIG_IPV6 */
2487                 return -EPFNOSUPPORT;
2488 #endif
2489         }
2490         info->key.tp_src = sport;
2491         info->key.tp_dst = dport;
2492         return 0;
2493 }
2494
2495 static const struct net_device_ops vxlan_netdev_ether_ops = {
2496         .ndo_init               = vxlan_init,
2497         .ndo_uninit             = vxlan_uninit,
2498         .ndo_open               = vxlan_open,
2499         .ndo_stop               = vxlan_stop,
2500         .ndo_start_xmit         = vxlan_xmit,
2501         .ndo_get_stats64        = ip_tunnel_get_stats64,
2502         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2503         .ndo_change_mtu         = vxlan_change_mtu,
2504         .ndo_validate_addr      = eth_validate_addr,
2505         .ndo_set_mac_address    = eth_mac_addr,
2506         .ndo_fdb_add            = vxlan_fdb_add,
2507         .ndo_fdb_del            = vxlan_fdb_delete,
2508         .ndo_fdb_dump           = vxlan_fdb_dump,
2509         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2510 };
2511
2512 static const struct net_device_ops vxlan_netdev_raw_ops = {
2513         .ndo_init               = vxlan_init,
2514         .ndo_uninit             = vxlan_uninit,
2515         .ndo_open               = vxlan_open,
2516         .ndo_stop               = vxlan_stop,
2517         .ndo_start_xmit         = vxlan_xmit,
2518         .ndo_get_stats64        = ip_tunnel_get_stats64,
2519         .ndo_change_mtu         = vxlan_change_mtu,
2520         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2521 };
2522
2523 /* Info for udev, that this is a virtual tunnel endpoint */
2524 static struct device_type vxlan_type = {
2525         .name = "vxlan",
2526 };
2527
2528 /* Calls the ndo_add_vxlan_port of the caller in order to
2529  * supply the listening VXLAN udp ports. Callers are expected
2530  * to implement the ndo_add_vxlan_port.
2531  */
2532 static void vxlan_push_rx_ports(struct net_device *dev)
2533 {
2534         struct vxlan_sock *vs;
2535         struct net *net = dev_net(dev);
2536         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2537         sa_family_t sa_family;
2538         __be16 port;
2539         unsigned int i;
2540
2541         if (!dev->netdev_ops->ndo_add_vxlan_port)
2542                 return;
2543
2544         spin_lock(&vn->sock_lock);
2545         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2546                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2547                         port = inet_sk(vs->sock->sk)->inet_sport;
2548                         sa_family = vxlan_get_sk_family(vs);
2549                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2550                                                             port);
2551                 }
2552         }
2553         spin_unlock(&vn->sock_lock);
2554 }
2555
2556 /* Initialize the device structure. */
2557 static void vxlan_setup(struct net_device *dev)
2558 {
2559         struct vxlan_dev *vxlan = netdev_priv(dev);
2560         unsigned int h;
2561
2562         eth_hw_addr_random(dev);
2563         ether_setup(dev);
2564
2565         dev->destructor = free_netdev;
2566         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2567
2568         dev->features   |= NETIF_F_LLTX;
2569         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2570         dev->features   |= NETIF_F_RXCSUM;
2571         dev->features   |= NETIF_F_GSO_SOFTWARE;
2572
2573         dev->vlan_features = dev->features;
2574         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2575         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2576         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2577         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2578         netif_keep_dst(dev);
2579         dev->priv_flags |= IFF_NO_QUEUE;
2580
2581         INIT_LIST_HEAD(&vxlan->next);
2582         spin_lock_init(&vxlan->hash_lock);
2583
2584         init_timer_deferrable(&vxlan->age_timer);
2585         vxlan->age_timer.function = vxlan_cleanup;
2586         vxlan->age_timer.data = (unsigned long) vxlan;
2587
2588         vxlan->cfg.dst_port = htons(vxlan_port);
2589
2590         vxlan->dev = dev;
2591
2592         gro_cells_init(&vxlan->gro_cells, dev);
2593
2594         for (h = 0; h < FDB_HASH_SIZE; ++h)
2595                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2596 }
2597
2598 static void vxlan_ether_setup(struct net_device *dev)
2599 {
2600         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2601         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2602         dev->netdev_ops = &vxlan_netdev_ether_ops;
2603 }
2604
2605 static void vxlan_raw_setup(struct net_device *dev)
2606 {
2607         dev->header_ops = NULL;
2608         dev->type = ARPHRD_NONE;
2609         dev->hard_header_len = 0;
2610         dev->addr_len = 0;
2611         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2612         dev->netdev_ops = &vxlan_netdev_raw_ops;
2613 }
2614
2615 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2616         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2617         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2618         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2619         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2620         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2621         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2622         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2623         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2624         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2625         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2626         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2627         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2628         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2629         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2630         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2631         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2632         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2633         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2634         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2635         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2636         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2637         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2638         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2639         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2640         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2641         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2642         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2643 };
2644
2645 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2646 {
2647         if (tb[IFLA_ADDRESS]) {
2648                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2649                         pr_debug("invalid link address (not ethernet)\n");
2650                         return -EINVAL;
2651                 }
2652
2653                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2654                         pr_debug("invalid all zero ethernet address\n");
2655                         return -EADDRNOTAVAIL;
2656                 }
2657         }
2658
2659         if (!data)
2660                 return -EINVAL;
2661
2662         if (data[IFLA_VXLAN_ID]) {
2663                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2664                 if (id >= VXLAN_VID_MASK)
2665                         return -ERANGE;
2666         }
2667
2668         if (data[IFLA_VXLAN_PORT_RANGE]) {
2669                 const struct ifla_vxlan_port_range *p
2670                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2671
2672                 if (ntohs(p->high) < ntohs(p->low)) {
2673                         pr_debug("port range %u .. %u not valid\n",
2674                                  ntohs(p->low), ntohs(p->high));
2675                         return -EINVAL;
2676                 }
2677         }
2678
2679         return 0;
2680 }
2681
2682 static void vxlan_get_drvinfo(struct net_device *netdev,
2683                               struct ethtool_drvinfo *drvinfo)
2684 {
2685         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2686         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2687 }
2688
2689 static const struct ethtool_ops vxlan_ethtool_ops = {
2690         .get_drvinfo    = vxlan_get_drvinfo,
2691         .get_link       = ethtool_op_get_link,
2692 };
2693
2694 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2695                                         __be16 port, u32 flags)
2696 {
2697         struct socket *sock;
2698         struct udp_port_cfg udp_conf;
2699         int err;
2700
2701         memset(&udp_conf, 0, sizeof(udp_conf));
2702
2703         if (ipv6) {
2704                 udp_conf.family = AF_INET6;
2705                 udp_conf.use_udp6_rx_checksums =
2706                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2707                 udp_conf.ipv6_v6only = 1;
2708         } else {
2709                 udp_conf.family = AF_INET;
2710         }
2711
2712         udp_conf.local_udp_port = port;
2713
2714         /* Open UDP socket */
2715         err = udp_sock_create(net, &udp_conf, &sock);
2716         if (err < 0)
2717                 return ERR_PTR(err);
2718
2719         return sock;
2720 }
2721
2722 /* Create new listen socket if needed */
2723 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2724                                               __be16 port, u32 flags)
2725 {
2726         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2727         struct vxlan_sock *vs;
2728         struct socket *sock;
2729         unsigned int h;
2730         struct udp_tunnel_sock_cfg tunnel_cfg;
2731
2732         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2733         if (!vs)
2734                 return ERR_PTR(-ENOMEM);
2735
2736         for (h = 0; h < VNI_HASH_SIZE; ++h)
2737                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2738
2739         sock = vxlan_create_sock(net, ipv6, port, flags);
2740         if (IS_ERR(sock)) {
2741                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2742                         PTR_ERR(sock));
2743                 kfree(vs);
2744                 return ERR_CAST(sock);
2745         }
2746
2747         vs->sock = sock;
2748         atomic_set(&vs->refcnt, 1);
2749         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2750
2751         spin_lock(&vn->sock_lock);
2752         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2753         vxlan_notify_add_rx_port(vs);
2754         spin_unlock(&vn->sock_lock);
2755
2756         /* Mark socket as an encapsulation socket. */
2757         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2758         tunnel_cfg.sk_user_data = vs;
2759         tunnel_cfg.encap_type = 1;
2760         tunnel_cfg.encap_rcv = vxlan_rcv;
2761         tunnel_cfg.encap_destroy = NULL;
2762         tunnel_cfg.gro_receive = vxlan_gro_receive;
2763         tunnel_cfg.gro_complete = vxlan_gro_complete;
2764
2765         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2766
2767         return vs;
2768 }
2769
2770 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2771 {
2772         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2773         struct vxlan_sock *vs = NULL;
2774
2775         if (!vxlan->cfg.no_share) {
2776                 spin_lock(&vn->sock_lock);
2777                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2778                                      vxlan->cfg.dst_port, vxlan->flags);
2779                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2780                         spin_unlock(&vn->sock_lock);
2781                         return -EBUSY;
2782                 }
2783                 spin_unlock(&vn->sock_lock);
2784         }
2785         if (!vs)
2786                 vs = vxlan_socket_create(vxlan->net, ipv6,
2787                                          vxlan->cfg.dst_port, vxlan->flags);
2788         if (IS_ERR(vs))
2789                 return PTR_ERR(vs);
2790 #if IS_ENABLED(CONFIG_IPV6)
2791         if (ipv6)
2792                 vxlan->vn6_sock = vs;
2793         else
2794 #endif
2795                 vxlan->vn4_sock = vs;
2796         vxlan_vs_add_dev(vs, vxlan);
2797         return 0;
2798 }
2799
2800 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2801 {
2802         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2803         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2804         int ret = 0;
2805
2806         vxlan->vn4_sock = NULL;
2807 #if IS_ENABLED(CONFIG_IPV6)
2808         vxlan->vn6_sock = NULL;
2809         if (ipv6 || metadata)
2810                 ret = __vxlan_sock_add(vxlan, true);
2811 #endif
2812         if (!ret && (!ipv6 || metadata))
2813                 ret = __vxlan_sock_add(vxlan, false);
2814         if (ret < 0)
2815                 vxlan_sock_release(vxlan);
2816         return ret;
2817 }
2818
2819 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2820                                struct vxlan_config *conf)
2821 {
2822         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2823         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2824         struct vxlan_rdst *dst = &vxlan->default_dst;
2825         unsigned short needed_headroom = ETH_HLEN;
2826         int err;
2827         bool use_ipv6 = false;
2828         __be16 default_port = vxlan->cfg.dst_port;
2829         struct net_device *lowerdev = NULL;
2830
2831         if (conf->flags & VXLAN_F_GPE) {
2832                 if (conf->flags & ~VXLAN_F_ALLOWED_GPE)
2833                         return -EINVAL;
2834                 /* For now, allow GPE only together with COLLECT_METADATA.
2835                  * This can be relaxed later; in such case, the other side
2836                  * of the PtP link will have to be provided.
2837                  */
2838                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA))
2839                         return -EINVAL;
2840
2841                 vxlan_raw_setup(dev);
2842         } else {
2843                 vxlan_ether_setup(dev);
2844         }
2845
2846         vxlan->net = src_net;
2847
2848         dst->remote_vni = conf->vni;
2849
2850         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2851
2852         /* Unless IPv6 is explicitly requested, assume IPv4 */
2853         if (!dst->remote_ip.sa.sa_family)
2854                 dst->remote_ip.sa.sa_family = AF_INET;
2855
2856         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2857             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2858                 if (!IS_ENABLED(CONFIG_IPV6))
2859                         return -EPFNOSUPPORT;
2860                 use_ipv6 = true;
2861                 vxlan->flags |= VXLAN_F_IPV6;
2862         }
2863
2864         if (conf->label && !use_ipv6) {
2865                 pr_info("label only supported in use with IPv6\n");
2866                 return -EINVAL;
2867         }
2868
2869         if (conf->remote_ifindex) {
2870                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2871                 dst->remote_ifindex = conf->remote_ifindex;
2872
2873                 if (!lowerdev) {
2874                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2875                         return -ENODEV;
2876                 }
2877
2878 #if IS_ENABLED(CONFIG_IPV6)
2879                 if (use_ipv6) {
2880                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2881                         if (idev && idev->cnf.disable_ipv6) {
2882                                 pr_info("IPv6 is disabled via sysctl\n");
2883                                 return -EPERM;
2884                         }
2885                 }
2886 #endif
2887
2888                 if (!conf->mtu)
2889                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2890
2891                 needed_headroom = lowerdev->hard_header_len;
2892         }
2893
2894         if (conf->mtu) {
2895                 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2896                 if (err)
2897                         return err;
2898         }
2899
2900         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2901                 needed_headroom += VXLAN6_HEADROOM;
2902         else
2903                 needed_headroom += VXLAN_HEADROOM;
2904         dev->needed_headroom = needed_headroom;
2905
2906         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2907         if (!vxlan->cfg.dst_port) {
2908                 if (conf->flags & VXLAN_F_GPE)
2909                         vxlan->cfg.dst_port = 4790; /* IANA assigned VXLAN-GPE port */
2910                 else
2911                         vxlan->cfg.dst_port = default_port;
2912         }
2913         vxlan->flags |= conf->flags;
2914
2915         if (!vxlan->cfg.age_interval)
2916                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2917
2918         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2919                 if (tmp->cfg.vni == conf->vni &&
2920                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2921                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2922                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2923                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2924                     (vxlan->flags & VXLAN_F_RCV_FLAGS))
2925                 return -EEXIST;
2926         }
2927
2928         dev->ethtool_ops = &vxlan_ethtool_ops;
2929
2930         /* create an fdb entry for a valid default destination */
2931         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2932                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2933                                        &vxlan->default_dst.remote_ip,
2934                                        NUD_REACHABLE|NUD_PERMANENT,
2935                                        NLM_F_EXCL|NLM_F_CREATE,
2936                                        vxlan->cfg.dst_port,
2937                                        vxlan->default_dst.remote_vni,
2938                                        vxlan->default_dst.remote_ifindex,
2939                                        NTF_SELF);
2940                 if (err)
2941                         return err;
2942         }
2943
2944         err = register_netdevice(dev);
2945         if (err) {
2946                 vxlan_fdb_delete_default(vxlan);
2947                 return err;
2948         }
2949
2950         list_add(&vxlan->next, &vn->vxlan_list);
2951
2952         return 0;
2953 }
2954
2955 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2956                                     u8 name_assign_type, struct vxlan_config *conf)
2957 {
2958         struct nlattr *tb[IFLA_MAX+1];
2959         struct net_device *dev;
2960         int err;
2961
2962         memset(&tb, 0, sizeof(tb));
2963
2964         dev = rtnl_create_link(net, name, name_assign_type,
2965                                &vxlan_link_ops, tb);
2966         if (IS_ERR(dev))
2967                 return dev;
2968
2969         err = vxlan_dev_configure(net, dev, conf);
2970         if (err < 0) {
2971                 free_netdev(dev);
2972                 return ERR_PTR(err);
2973         }
2974
2975         return dev;
2976 }
2977 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2978
2979 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2980                          struct nlattr *tb[], struct nlattr *data[])
2981 {
2982         struct vxlan_config conf;
2983         int err;
2984
2985         memset(&conf, 0, sizeof(conf));
2986
2987         if (data[IFLA_VXLAN_ID])
2988                 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2989
2990         if (data[IFLA_VXLAN_GROUP]) {
2991                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2992         } else if (data[IFLA_VXLAN_GROUP6]) {
2993                 if (!IS_ENABLED(CONFIG_IPV6))
2994                         return -EPFNOSUPPORT;
2995
2996                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2997                 conf.remote_ip.sa.sa_family = AF_INET6;
2998         }
2999
3000         if (data[IFLA_VXLAN_LOCAL]) {
3001                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3002                 conf.saddr.sa.sa_family = AF_INET;
3003         } else if (data[IFLA_VXLAN_LOCAL6]) {
3004                 if (!IS_ENABLED(CONFIG_IPV6))
3005                         return -EPFNOSUPPORT;
3006
3007                 /* TODO: respect scope id */
3008                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3009                 conf.saddr.sa.sa_family = AF_INET6;
3010         }
3011
3012         if (data[IFLA_VXLAN_LINK])
3013                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3014
3015         if (data[IFLA_VXLAN_TOS])
3016                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3017
3018         if (data[IFLA_VXLAN_TTL])
3019                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3020
3021         if (data[IFLA_VXLAN_LABEL])
3022                 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3023                              IPV6_FLOWLABEL_MASK;
3024
3025         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3026                 conf.flags |= VXLAN_F_LEARN;
3027
3028         if (data[IFLA_VXLAN_AGEING])
3029                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3030
3031         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
3032                 conf.flags |= VXLAN_F_PROXY;
3033
3034         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
3035                 conf.flags |= VXLAN_F_RSC;
3036
3037         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3038                 conf.flags |= VXLAN_F_L2MISS;
3039
3040         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3041                 conf.flags |= VXLAN_F_L3MISS;
3042
3043         if (data[IFLA_VXLAN_LIMIT])
3044                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3045
3046         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
3047             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3048                 conf.flags |= VXLAN_F_COLLECT_METADATA;
3049
3050         if (data[IFLA_VXLAN_PORT_RANGE]) {
3051                 const struct ifla_vxlan_port_range *p
3052                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3053                 conf.port_min = ntohs(p->low);
3054                 conf.port_max = ntohs(p->high);
3055         }
3056
3057         if (data[IFLA_VXLAN_PORT])
3058                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3059
3060         if (data[IFLA_VXLAN_UDP_CSUM] &&
3061             !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3062                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3063
3064         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
3065             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3066                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3067
3068         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3069             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3070                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3071
3072         if (data[IFLA_VXLAN_REMCSUM_TX] &&
3073             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3074                 conf.flags |= VXLAN_F_REMCSUM_TX;
3075
3076         if (data[IFLA_VXLAN_REMCSUM_RX] &&
3077             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3078                 conf.flags |= VXLAN_F_REMCSUM_RX;
3079
3080         if (data[IFLA_VXLAN_GBP])
3081                 conf.flags |= VXLAN_F_GBP;
3082
3083         if (data[IFLA_VXLAN_GPE])
3084                 conf.flags |= VXLAN_F_GPE;
3085
3086         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3087                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3088
3089         err = vxlan_dev_configure(src_net, dev, &conf);
3090         switch (err) {
3091         case -ENODEV:
3092                 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
3093                 break;
3094
3095         case -EPERM:
3096                 pr_info("IPv6 is disabled via sysctl\n");
3097                 break;
3098
3099         case -EEXIST:
3100                 pr_info("duplicate VNI %u\n", be32_to_cpu(conf.vni));
3101                 break;
3102
3103         case -EINVAL:
3104                 pr_info("unsupported combination of extensions\n");
3105                 break;
3106         }
3107
3108         return err;
3109 }
3110
3111 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3112 {
3113         struct vxlan_dev *vxlan = netdev_priv(dev);
3114         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3115
3116         spin_lock(&vn->sock_lock);
3117         if (!hlist_unhashed(&vxlan->hlist))
3118                 hlist_del_rcu(&vxlan->hlist);
3119         spin_unlock(&vn->sock_lock);
3120
3121         gro_cells_destroy(&vxlan->gro_cells);
3122         list_del(&vxlan->next);
3123         unregister_netdevice_queue(dev, head);
3124 }
3125
3126 static size_t vxlan_get_size(const struct net_device *dev)
3127 {
3128
3129         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3130                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3131                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3132                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3133                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3134                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3135                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3136                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3137                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3138                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3139                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3140                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3141                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3142                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3143                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3144                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3145                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3146                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3147                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3148                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3149                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3150                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3151                 0;
3152 }
3153
3154 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3155 {
3156         const struct vxlan_dev *vxlan = netdev_priv(dev);
3157         const struct vxlan_rdst *dst = &vxlan->default_dst;
3158         struct ifla_vxlan_port_range ports = {
3159                 .low =  htons(vxlan->cfg.port_min),
3160                 .high = htons(vxlan->cfg.port_max),
3161         };
3162
3163         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3164                 goto nla_put_failure;
3165
3166         if (!vxlan_addr_any(&dst->remote_ip)) {
3167                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3168                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3169                                             dst->remote_ip.sin.sin_addr.s_addr))
3170                                 goto nla_put_failure;
3171 #if IS_ENABLED(CONFIG_IPV6)
3172                 } else {
3173                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3174                                              &dst->remote_ip.sin6.sin6_addr))
3175                                 goto nla_put_failure;
3176 #endif
3177                 }
3178         }
3179
3180         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3181                 goto nla_put_failure;
3182
3183         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3184                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3185                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3186                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3187                                 goto nla_put_failure;
3188 #if IS_ENABLED(CONFIG_IPV6)
3189                 } else {
3190                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3191                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3192                                 goto nla_put_failure;
3193 #endif
3194                 }
3195         }
3196
3197         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3198             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3199             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3200             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3201                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3202             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3203                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3204             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3205             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3206                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3207             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3208                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3209             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3210                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3211             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3212             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3213             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3214             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3215                         !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3216             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3217                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3218             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3219                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3220             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3221                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3222             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3223                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3224                 goto nla_put_failure;
3225
3226         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3227                 goto nla_put_failure;
3228
3229         if (vxlan->flags & VXLAN_F_GBP &&
3230             nla_put_flag(skb, IFLA_VXLAN_GBP))
3231                 goto nla_put_failure;
3232
3233         if (vxlan->flags & VXLAN_F_GPE &&
3234             nla_put_flag(skb, IFLA_VXLAN_GPE))
3235                 goto nla_put_failure;
3236
3237         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3238             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3239                 goto nla_put_failure;
3240
3241         return 0;
3242
3243 nla_put_failure:
3244         return -EMSGSIZE;
3245 }
3246
3247 static struct net *vxlan_get_link_net(const struct net_device *dev)
3248 {
3249         struct vxlan_dev *vxlan = netdev_priv(dev);
3250
3251         return vxlan->net;
3252 }
3253
3254 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3255         .kind           = "vxlan",
3256         .maxtype        = IFLA_VXLAN_MAX,
3257         .policy         = vxlan_policy,
3258         .priv_size      = sizeof(struct vxlan_dev),
3259         .setup          = vxlan_setup,
3260         .validate       = vxlan_validate,
3261         .newlink        = vxlan_newlink,
3262         .dellink        = vxlan_dellink,
3263         .get_size       = vxlan_get_size,
3264         .fill_info      = vxlan_fill_info,
3265         .get_link_net   = vxlan_get_link_net,
3266 };
3267
3268 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3269                                              struct net_device *dev)
3270 {
3271         struct vxlan_dev *vxlan, *next;
3272         LIST_HEAD(list_kill);
3273
3274         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3275                 struct vxlan_rdst *dst = &vxlan->default_dst;
3276
3277                 /* In case we created vxlan device with carrier
3278                  * and we loose the carrier due to module unload
3279                  * we also need to remove vxlan device. In other
3280                  * cases, it's not necessary and remote_ifindex
3281                  * is 0 here, so no matches.
3282                  */
3283                 if (dst->remote_ifindex == dev->ifindex)
3284                         vxlan_dellink(vxlan->dev, &list_kill);
3285         }
3286
3287         unregister_netdevice_many(&list_kill);
3288 }
3289
3290 static int vxlan_netdevice_event(struct notifier_block *unused,
3291                                  unsigned long event, void *ptr)
3292 {
3293         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3294         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3295
3296         if (event == NETDEV_UNREGISTER)
3297                 vxlan_handle_lowerdev_unregister(vn, dev);
3298         else if (event == NETDEV_OFFLOAD_PUSH_VXLAN)
3299                 vxlan_push_rx_ports(dev);
3300
3301         return NOTIFY_DONE;
3302 }
3303
3304 static struct notifier_block vxlan_notifier_block __read_mostly = {
3305         .notifier_call = vxlan_netdevice_event,
3306 };
3307
3308 static __net_init int vxlan_init_net(struct net *net)
3309 {
3310         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3311         unsigned int h;
3312
3313         INIT_LIST_HEAD(&vn->vxlan_list);
3314         spin_lock_init(&vn->sock_lock);
3315
3316         for (h = 0; h < PORT_HASH_SIZE; ++h)
3317                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3318
3319         return 0;
3320 }
3321
3322 static void __net_exit vxlan_exit_net(struct net *net)
3323 {
3324         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3325         struct vxlan_dev *vxlan, *next;
3326         struct net_device *dev, *aux;
3327         LIST_HEAD(list);
3328
3329         rtnl_lock();
3330         for_each_netdev_safe(net, dev, aux)
3331                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3332                         unregister_netdevice_queue(dev, &list);
3333
3334         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3335                 /* If vxlan->dev is in the same netns, it has already been added
3336                  * to the list by the previous loop.
3337                  */
3338                 if (!net_eq(dev_net(vxlan->dev), net)) {
3339                         gro_cells_destroy(&vxlan->gro_cells);
3340                         unregister_netdevice_queue(vxlan->dev, &list);
3341                 }
3342         }
3343
3344         unregister_netdevice_many(&list);
3345         rtnl_unlock();
3346 }
3347
3348 static struct pernet_operations vxlan_net_ops = {
3349         .init = vxlan_init_net,
3350         .exit = vxlan_exit_net,
3351         .id   = &vxlan_net_id,
3352         .size = sizeof(struct vxlan_net),
3353 };
3354
3355 static int __init vxlan_init_module(void)
3356 {
3357         int rc;
3358
3359         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3360
3361         rc = register_pernet_subsys(&vxlan_net_ops);
3362         if (rc)
3363                 goto out1;
3364
3365         rc = register_netdevice_notifier(&vxlan_notifier_block);
3366         if (rc)
3367                 goto out2;
3368
3369         rc = rtnl_link_register(&vxlan_link_ops);
3370         if (rc)
3371                 goto out3;
3372
3373         return 0;
3374 out3:
3375         unregister_netdevice_notifier(&vxlan_notifier_block);
3376 out2:
3377         unregister_pernet_subsys(&vxlan_net_ops);
3378 out1:
3379         return rc;
3380 }
3381 late_initcall(vxlan_init_module);
3382
3383 static void __exit vxlan_cleanup_module(void)
3384 {
3385         rtnl_link_unregister(&vxlan_link_ops);
3386         unregister_netdevice_notifier(&vxlan_notifier_block);
3387         unregister_pernet_subsys(&vxlan_net_ops);
3388         /* rcu_barrier() is called by netns */
3389 }
3390 module_exit(vxlan_cleanup_module);
3391
3392 MODULE_LICENSE("GPL");
3393 MODULE_VERSION(VXLAN_VERSION);
3394 MODULE_AUTHOR("Stephen Hemminger <[email protected]>");
3395 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3396 MODULE_ALIAS_RTNL_LINK("vxlan");