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