drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/errno.h>
  16 #include <linux/slab.h>
  17 #include <linux/udp.h>
  18 #include <linux/igmp.h>
  19 #include <linux/if_ether.h>
  20 #include <linux/ethtool.h>
  21 #include <net/arp.h>
  22 #include <net/ndisc.h>
  23 #include <net/ip.h>
  24 #include <net/icmp.h>
  25 #include <net/rtnetlink.h>
  26 #include <net/inet_ecn.h>
  27 #include <net/net_namespace.h>
  28 #include <net/netns/generic.h>
  29 #include <net/vxlan.h>
  30
  31 #if IS_ENABLED(CONFIG_IPV6)
  32 #include <net/ip6_tunnel.h>
  33 #include <net/ip6_checksum.h>
  34 #endif
  35
  36 #define VXLAN_VERSION   "0.1"
  37
  38 #define PORT_HASH_BITS  8
  39 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  40 #define FDB_AGE_DEFAULT 300 /* 5 min */
  41 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  42
  43 /* UDP port for VXLAN traffic.
  44  * The IANA assigned port is 4789, but the Linux default is 8472
  45  * for compatibility with early adopters.
  46  */
  47 static unsigned short vxlan_port __read_mostly = 8472;
  48 module_param_named(udp_port, vxlan_port, ushort, 0444);
  49 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  50
  51 static bool log_ecn_error = true;
  52 module_param(log_ecn_error, bool, 0644);
  53 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  54
  55 static int vxlan_net_id;
  56 static struct rtnl_link_ops vxlan_link_ops;
  57
  58 static const u8 all_zeros_mac[ETH_ALEN + 2];
  59
  60 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  61
  62 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
  63
  64 /* per-network namespace private data for this module */
  65 struct vxlan_net {
  66         struct list_head  vxlan_list;
  67         struct hlist_head sock_list[PORT_HASH_SIZE];
  68         spinlock_t        sock_lock;
  69 };
  70
  71 /* Forwarding table entry */
  72 struct vxlan_fdb {
  73         struct hlist_node hlist;        /* linked list of entries */
  74         struct rcu_head   rcu;
  75         unsigned long     updated;      /* jiffies */
  76         unsigned long     used;
  77         struct list_head  remotes;
  78         u8                eth_addr[ETH_ALEN];
  79         u16               state;        /* see ndm_state */
  80         u8                flags;        /* see ndm_flags */
  81 };
  82
  83 /* salt for hash table */
  84 static u32 vxlan_salt __read_mostly;
  85
  86 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  87 {
  88         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  89                ip_tunnel_collect_metadata();
  90 }
  91
  92 #if IS_ENABLED(CONFIG_IPV6)
  93 static inline
  94 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
  95 {
  96         if (a->sa.sa_family != b->sa.sa_family)
  97                 return false;
  98         if (a->sa.sa_family == AF_INET6)
  99                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 100         else
 101                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 102 }
 103
 104 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 105 {
 106         if (ipa->sa.sa_family == AF_INET6)
 107                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 108         else
 109                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 110 }
 111
 112 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 113 {
 114         if (ipa->sa.sa_family == AF_INET6)
 115                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 116         else
 117                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 118 }
 119
 120 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 121 {
 122         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 123                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 124                 ip->sa.sa_family = AF_INET6;
 125                 return 0;
 126         } else if (nla_len(nla) >= sizeof(__be32)) {
 127                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 128                 ip->sa.sa_family = AF_INET;
 129                 return 0;
 130         } else {
 131                 return -EAFNOSUPPORT;
 132         }
 133 }
 134
 135 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 136                               const union vxlan_addr *ip)
 137 {
 138         if (ip->sa.sa_family == AF_INET6)
 139                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 140         else
 141                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 142 }
 143
 144 #else /* !CONFIG_IPV6 */
 145
 146 static inline
 147 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 148 {
 149         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 150 }
 151
 152 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 153 {
 154         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 155 }
 156
 157 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 158 {
 159         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 160 }
 161
 162 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 163 {
 164         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 165                 return -EAFNOSUPPORT;
 166         } else if (nla_len(nla) >= sizeof(__be32)) {
 167                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 168                 ip->sa.sa_family = AF_INET;
 169                 return 0;
 170         } else {
 171                 return -EAFNOSUPPORT;
 172         }
 173 }
 174
 175 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 176                               const union vxlan_addr *ip)
 177 {
 178         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 179 }
 180 #endif
 181
 182 /* Virtual Network hash table head */
 183 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 184 {
 185         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 186 }
 187
 188 /* Socket hash table head */
 189 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 190 {
 191         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 192
 193         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 194 }
 195
 196 /* First remote destination for a forwarding entry.
 197  * Guaranteed to be non-NULL because remotes are never deleted.
 198  */
 199 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 200 {
 201         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 202 }
 203
 204 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 205 {
 206         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 207 }
 208
 209 /* Find VXLAN socket based on network namespace, address family and UDP port
 210  * and enabled unshareable flags.
 211  */
 212 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 213                                           __be16 port, u32 flags)
 214 {
 215         struct vxlan_sock *vs;
 216
 217         flags &= VXLAN_F_RCV_FLAGS;
 218
 219         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 220                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 221                     vxlan_get_sk_family(vs) == family &&
 222                     vs->flags == flags)
 223                         return vs;
 224         }
 225         return NULL;
 226 }
 227
 228 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
 229 {
 230         struct vxlan_dev_node *node;
 231
 232         /* For flow based devices, map all packets to VNI 0 */
 233         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 234                 vni = 0;
 235
 236         hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
 237                 if (node->vxlan->default_dst.remote_vni == vni)
 238                         return node->vxlan;
 239         }
 240
 241         return NULL;
 242 }
 243
 244 /* Look up VNI in a per net namespace table */
 245 static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
 246                                         sa_family_t family, __be16 port,
 247                                         u32 flags)
 248 {
 249         struct vxlan_sock *vs;
 250
 251         vs = vxlan_find_sock(net, family, port, flags);
 252         if (!vs)
 253                 return NULL;
 254
 255         return vxlan_vs_find_vni(vs, vni);
 256 }
 257
 258 /* Fill in neighbour message in skbuff. */
 259 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 260                           const struct vxlan_fdb *fdb,
 261                           u32 portid, u32 seq, int type, unsigned int flags,
 262                           const struct vxlan_rdst *rdst)
 263 {
 264         unsigned long now = jiffies;
 265         struct nda_cacheinfo ci;
 266         struct nlmsghdr *nlh;
 267         struct ndmsg *ndm;
 268         bool send_ip, send_eth;
 269
 270         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 271         if (nlh == NULL)
 272                 return -EMSGSIZE;
 273
 274         ndm = nlmsg_data(nlh);
 275         memset(ndm, 0, sizeof(*ndm));
 276
 277         send_eth = send_ip = true;
 278
 279         if (type == RTM_GETNEIGH) {
 280                 ndm->ndm_family = AF_INET;
 281                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 282                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 283         } else
 284                 ndm->ndm_family = AF_BRIDGE;
 285         ndm->ndm_state = fdb->state;
 286         ndm->ndm_ifindex = vxlan->dev->ifindex;
 287         ndm->ndm_flags = fdb->flags;
 288         ndm->ndm_type = RTN_UNICAST;
 289
 290         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 291             nla_put_s32(skb, NDA_LINK_NETNSID,
 292                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 293                 goto nla_put_failure;
 294
 295         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 296                 goto nla_put_failure;
 297
 298         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 299                 goto nla_put_failure;
 300
 301         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 302             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 303                 goto nla_put_failure;
 304         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 305             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 306                 goto nla_put_failure;
 307         if (rdst->remote_ifindex &&
 308             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 309                 goto nla_put_failure;
 310
 311         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 312         ci.ndm_confirmed = 0;
 313         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 314         ci.ndm_refcnt    = 0;
 315
 316         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 317                 goto nla_put_failure;
 318
 319         nlmsg_end(skb, nlh);
 320         return 0;
 321
 322 nla_put_failure:
 323         nlmsg_cancel(skb, nlh);
 324         return -EMSGSIZE;
 325 }
 326
 327 static inline size_t vxlan_nlmsg_size(void)
 328 {
 329         return NLMSG_ALIGN(sizeof(struct ndmsg))
 330                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 331                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 332                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 333                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 334                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 335                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 336                 + nla_total_size(sizeof(struct nda_cacheinfo));
 337 }
 338
 339 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 340                              struct vxlan_rdst *rd, int type)
 341 {
 342         struct net *net = dev_net(vxlan->dev);
 343         struct sk_buff *skb;
 344         int err = -ENOBUFS;
 345
 346         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 347         if (skb == NULL)
 348                 goto errout;
 349
 350         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 351         if (err < 0) {
 352                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 353                 WARN_ON(err == -EMSGSIZE);
 354                 kfree_skb(skb);
 355                 goto errout;
 356         }
 357
 358         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 359         return;
 360 errout:
 361         if (err < 0)
 362                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 363 }
 364
 365 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 366 {
 367         struct vxlan_dev *vxlan = netdev_priv(dev);
 368         struct vxlan_fdb f = {
 369                 .state = NUD_STALE,
 370         };
 371         struct vxlan_rdst remote = {
 372                 .remote_ip = *ipa, /* goes to NDA_DST */
 373                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 374         };
 375
 376         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 377 }
 378
 379 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 380 {
 381         struct vxlan_fdb f = {
 382                 .state = NUD_STALE,
 383         };
 384         struct vxlan_rdst remote = { };
 385
 386         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 387
 388         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 389 }
 390
 391 /* Hash Ethernet address */
 392 static u32 eth_hash(const unsigned char *addr)
 393 {
 394         u64 value = get_unaligned((u64 *)addr);
 395
 396         /* only want 6 bytes */
 397 #ifdef __BIG_ENDIAN
 398         value >>= 16;
 399 #else
 400         value <<= 16;
 401 #endif
 402         return hash_64(value, FDB_HASH_BITS);
 403 }
 404
 405 /* Hash chain to use given mac address */
 406 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 407                                                 const u8 *mac)
 408 {
 409         return &vxlan->fdb_head[eth_hash(mac)];
 410 }
 411
 412 /* Look up Ethernet address in forwarding table */
 413 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 414                                         const u8 *mac)
 415 {
 416         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 417         struct vxlan_fdb *f;
 418
 419         hlist_for_each_entry_rcu(f, head, hlist) {
 420                 if (ether_addr_equal(mac, f->eth_addr))
 421                         return f;
 422         }
 423
 424         return NULL;
 425 }
 426
 427 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 428                                         const u8 *mac)
 429 {
 430         struct vxlan_fdb *f;
 431
 432         f = __vxlan_find_mac(vxlan, mac);
 433         if (f)
 434                 f->used = jiffies;
 435
 436         return f;
 437 }
 438
 439 /* caller should hold vxlan->hash_lock */
 440 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 441                                               union vxlan_addr *ip, __be16 port,
 442                                               __be32 vni, __u32 ifindex)
 443 {
 444         struct vxlan_rdst *rd;
 445
 446         list_for_each_entry(rd, &f->remotes, list) {
 447                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 448                     rd->remote_port == port &&
 449                     rd->remote_vni == vni &&
 450                     rd->remote_ifindex == ifindex)
 451                         return rd;
 452         }
 453
 454         return NULL;
 455 }
 456
 457 /* Replace destination of unicast mac */
 458 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 459                              union vxlan_addr *ip, __be16 port, __be32 vni,
 460                              __u32 ifindex)
 461 {
 462         struct vxlan_rdst *rd;
 463
 464         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 465         if (rd)
 466                 return 0;
 467
 468         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 469         if (!rd)
 470                 return 0;
 471
 472         dst_cache_reset(&rd->dst_cache);
 473         rd->remote_ip = *ip;
 474         rd->remote_port = port;
 475         rd->remote_vni = vni;
 476         rd->remote_ifindex = ifindex;
 477         return 1;
 478 }
 479
 480 /* Add/update destinations for multicast */
 481 static int vxlan_fdb_append(struct vxlan_fdb *f,
 482                             union vxlan_addr *ip, __be16 port, __be32 vni,
 483                             __u32 ifindex, struct vxlan_rdst **rdp)
 484 {
 485         struct vxlan_rdst *rd;
 486
 487         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 488         if (rd)
 489                 return 0;
 490
 491         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 492         if (rd == NULL)
 493                 return -ENOBUFS;
 494
 495         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 496                 kfree(rd);
 497                 return -ENOBUFS;
 498         }
 499
 500         rd->remote_ip = *ip;
 501         rd->remote_port = port;
 502         rd->remote_vni = vni;
 503         rd->remote_ifindex = ifindex;
 504
 505         list_add_tail_rcu(&rd->list, &f->remotes);
 506
 507         *rdp = rd;
 508         return 1;
 509 }
 510
 511 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 512                                           unsigned int off,
 513                                           struct vxlanhdr *vh, size_t hdrlen,
 514                                           __be32 vni_field,
 515                                           struct gro_remcsum *grc,
 516                                           bool nopartial)
 517 {
 518         size_t start, offset;
 519
 520         if (skb->remcsum_offload)
 521                 return vh;
 522
 523         if (!NAPI_GRO_CB(skb)->csum_valid)
 524                 return NULL;
 525
 526         start = vxlan_rco_start(vni_field);
 527         offset = start + vxlan_rco_offset(vni_field);
 528
 529         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 530                                      start, offset, grc, nopartial);
 531
 532         skb->remcsum_offload = 1;
 533
 534         return vh;
 535 }
 536
 537 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
 538                                           struct sk_buff **head,
 539                                           struct sk_buff *skb)
 540 {
 541         struct sk_buff *p, **pp = NULL;
 542         struct vxlanhdr *vh, *vh2;
 543         unsigned int hlen, off_vx;
 544         int flush = 1;
 545         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 546         __be32 flags;
 547         struct gro_remcsum grc;
 548
 549         skb_gro_remcsum_init(&grc);
 550
 551         off_vx = skb_gro_offset(skb);
 552         hlen = off_vx + sizeof(*vh);
 553         vh   = skb_gro_header_fast(skb, off_vx);
 554         if (skb_gro_header_hard(skb, hlen)) {
 555                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 556                 if (unlikely(!vh))
 557                         goto out;
 558         }
 559
 560         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 561
 562         flags = vh->vx_flags;
 563
 564         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 565                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 566                                        vh->vx_vni, &grc,
 567                                        !!(vs->flags &
 568                                           VXLAN_F_REMCSUM_NOPARTIAL));
 569
 570                 if (!vh)
 571                         goto out;
 572         }
 573
 574         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 575
 576         for (p = *head; p; p = p->next) {
 577                 if (!NAPI_GRO_CB(p)->same_flow)
 578                         continue;
 579
 580                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 581                 if (vh->vx_flags != vh2->vx_flags ||
 582                     vh->vx_vni != vh2->vx_vni) {
 583                         NAPI_GRO_CB(p)->same_flow = 0;
 584                         continue;
 585                 }
 586         }
 587
 588         pp = call_gro_receive(eth_gro_receive, head, skb);
 589         flush = 0;
 590
 591 out:
 592         skb_gro_remcsum_cleanup(skb, &grc);
 593         NAPI_GRO_CB(skb)->flush |= flush;
 594
 595         return pp;
 596 }
 597
 598 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 599 {
 600         /* Sets 'skb->inner_mac_header' since we are always called with
 601          * 'skb->encapsulation' set.
 602          */
 603         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 604 }
 605
 606 /* Add new entry to forwarding table -- assumes lock held */
 607 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 608                             const u8 *mac, union vxlan_addr *ip,
 609                             __u16 state, __u16 flags,
 610                             __be16 port, __be32 vni, __u32 ifindex,
 611                             __u8 ndm_flags)
 612 {
 613         struct vxlan_rdst *rd = NULL;
 614         struct vxlan_fdb *f;
 615         int notify = 0;
 616         int rc;
 617
 618         f = __vxlan_find_mac(vxlan, mac);
 619         if (f) {
 620                 if (flags & NLM_F_EXCL) {
 621                         netdev_dbg(vxlan->dev,
 622                                    "lost race to create %pM\n", mac);
 623                         return -EEXIST;
 624                 }
 625                 if (f->state != state) {
 626                         f->state = state;
 627                         f->updated = jiffies;
 628                         notify = 1;
 629                 }
 630                 if (f->flags != ndm_flags) {
 631                         f->flags = ndm_flags;
 632                         f->updated = jiffies;
 633                         notify = 1;
 634                 }
 635                 if ((flags & NLM_F_REPLACE)) {
 636                         /* Only change unicasts */
 637                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 638                              is_zero_ether_addr(f->eth_addr))) {
 639                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 640                                                            ifindex);
 641                         } else
 642                                 return -EOPNOTSUPP;
 643                 }
 644                 if ((flags & NLM_F_APPEND) &&
 645                     (is_multicast_ether_addr(f->eth_addr) ||
 646                      is_zero_ether_addr(f->eth_addr))) {
 647                         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 648
 649                         if (rc < 0)
 650                                 return rc;
 651                         notify |= rc;
 652                 }
 653         } else {
 654                 if (!(flags & NLM_F_CREATE))
 655                         return -ENOENT;
 656
 657                 if (vxlan->cfg.addrmax &&
 658                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 659                         return -ENOSPC;
 660
 661                 /* Disallow replace to add a multicast entry */
 662                 if ((flags & NLM_F_REPLACE) &&
 663                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 664                         return -EOPNOTSUPP;
 665
 666                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 667                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 668                 if (!f)
 669                         return -ENOMEM;
 670
 671                 notify = 1;
 672                 f->state = state;
 673                 f->flags = ndm_flags;
 674                 f->updated = f->used = jiffies;
 675                 INIT_LIST_HEAD(&f->remotes);
 676                 memcpy(f->eth_addr, mac, ETH_ALEN);
 677
 678                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 679                 if (rc < 0) {
 680                         kfree(f);
 681                         return rc;
 682                 }
 683
 684                 ++vxlan->addrcnt;
 685                 hlist_add_head_rcu(&f->hlist,
 686                                    vxlan_fdb_head(vxlan, mac));
 687         }
 688
 689         if (notify) {
 690                 if (rd == NULL)
 691                         rd = first_remote_rtnl(f);
 692                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 693         }
 694
 695         return 0;
 696 }
 697
 698 static void vxlan_fdb_free(struct rcu_head *head)
 699 {
 700         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 701         struct vxlan_rdst *rd, *nd;
 702
 703         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 704                 dst_cache_destroy(&rd->dst_cache);
 705                 kfree(rd);
 706         }
 707         kfree(f);
 708 }
 709
 710 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 711 {
 712         netdev_dbg(vxlan->dev,
 713                     "delete %pM\n", f->eth_addr);
 714
 715         --vxlan->addrcnt;
 716         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 717
 718         hlist_del_rcu(&f->hlist);
 719         call_rcu(&f->rcu, vxlan_fdb_free);
 720 }
 721
 722 static void vxlan_dst_free(struct rcu_head *head)
 723 {
 724         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 725
 726         dst_cache_destroy(&rd->dst_cache);
 727         kfree(rd);
 728 }
 729
 730 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 731                                   struct vxlan_rdst *rd)
 732 {
 733         list_del_rcu(&rd->list);
 734         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 735         call_rcu(&rd->rcu, vxlan_dst_free);
 736 }
 737
 738 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 739                            union vxlan_addr *ip, __be16 *port, __be32 *vni,
 740                            u32 *ifindex)
 741 {
 742         struct net *net = dev_net(vxlan->dev);
 743         int err;
 744
 745         if (tb[NDA_DST]) {
 746                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 747                 if (err)
 748                         return err;
 749         } else {
 750                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 751                 if (remote->sa.sa_family == AF_INET) {
 752                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 753                         ip->sa.sa_family = AF_INET;
 754 #if IS_ENABLED(CONFIG_IPV6)
 755                 } else {
 756                         ip->sin6.sin6_addr = in6addr_any;
 757                         ip->sa.sa_family = AF_INET6;
 758 #endif
 759                 }
 760         }
 761
 762         if (tb[NDA_PORT]) {
 763                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 764                         return -EINVAL;
 765                 *port = nla_get_be16(tb[NDA_PORT]);
 766         } else {
 767                 *port = vxlan->cfg.dst_port;
 768         }
 769
 770         if (tb[NDA_VNI]) {
 771                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 772                         return -EINVAL;
 773                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 774         } else {
 775                 *vni = vxlan->default_dst.remote_vni;
 776         }
 777
 778         if (tb[NDA_IFINDEX]) {
 779                 struct net_device *tdev;
 780
 781                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 782                         return -EINVAL;
 783                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 784                 tdev = __dev_get_by_index(net, *ifindex);
 785                 if (!tdev)
 786                         return -EADDRNOTAVAIL;
 787         } else {
 788                 *ifindex = 0;
 789         }
 790
 791         return 0;
 792 }
 793
 794 /* Add static entry (via netlink) */
 795 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 796                          struct net_device *dev,
 797                          const unsigned char *addr, u16 vid, u16 flags)
 798 {
 799         struct vxlan_dev *vxlan = netdev_priv(dev);
 800         /* struct net *net = dev_net(vxlan->dev); */
 801         union vxlan_addr ip;
 802         __be16 port;
 803         __be32 vni;
 804         u32 ifindex;
 805         int err;
 806
 807         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 808                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 809                         ndm->ndm_state);
 810                 return -EINVAL;
 811         }
 812
 813         if (tb[NDA_DST] == NULL)
 814                 return -EINVAL;
 815
 816         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 817         if (err)
 818                 return err;
 819
 820         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 821                 return -EAFNOSUPPORT;
 822
 823         spin_lock_bh(&vxlan->hash_lock);
 824         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 825                                port, vni, ifindex, ndm->ndm_flags);
 826         spin_unlock_bh(&vxlan->hash_lock);
 827
 828         return err;
 829 }
 830
 831 /* Delete entry (via netlink) */
 832 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 833                             struct net_device *dev,
 834                             const unsigned char *addr, u16 vid)
 835 {
 836         struct vxlan_dev *vxlan = netdev_priv(dev);
 837         struct vxlan_fdb *f;
 838         struct vxlan_rdst *rd = NULL;
 839         union vxlan_addr ip;
 840         __be16 port;
 841         __be32 vni;
 842         u32 ifindex;
 843         int err;
 844
 845         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 846         if (err)
 847                 return err;
 848
 849         err = -ENOENT;
 850
 851         spin_lock_bh(&vxlan->hash_lock);
 852         f = vxlan_find_mac(vxlan, addr);
 853         if (!f)
 854                 goto out;
 855
 856         if (!vxlan_addr_any(&ip)) {
 857                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 858                 if (!rd)
 859                         goto out;
 860         }
 861
 862         err = 0;
 863
 864         /* remove a destination if it's not the only one on the list,
 865          * otherwise destroy the fdb entry
 866          */
 867         if (rd && !list_is_singular(&f->remotes)) {
 868                 vxlan_fdb_dst_destroy(vxlan, f, rd);
 869                 goto out;
 870         }
 871
 872         vxlan_fdb_destroy(vxlan, f);
 873
 874 out:
 875         spin_unlock_bh(&vxlan->hash_lock);
 876
 877         return err;
 878 }
 879
 880 /* Dump forwarding table */
 881 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 882                           struct net_device *dev,
 883                           struct net_device *filter_dev, int *idx)
 884 {
 885         struct vxlan_dev *vxlan = netdev_priv(dev);
 886         unsigned int h;
 887         int err = 0;
 888
 889         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 890                 struct vxlan_fdb *f;
 891
 892                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 893                         struct vxlan_rdst *rd;
 894
 895                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 896                                 if (*idx < cb->args[2])
 897                                         goto skip;
 898
 899                                 err = vxlan_fdb_info(skb, vxlan, f,
 900                                                      NETLINK_CB(cb->skb).portid,
 901                                                      cb->nlh->nlmsg_seq,
 902                                                      RTM_NEWNEIGH,
 903                                                      NLM_F_MULTI, rd);
 904                                 if (err < 0)
 905                                         goto out;
 906 skip:
 907                                 *idx += 1;
 908                         }
 909                 }
 910         }
 911 out:
 912         return err;
 913 }
 914
 915 /* Watch incoming packets to learn mapping between Ethernet address
 916  * and Tunnel endpoint.
 917  * Return true if packet is bogus and should be dropped.
 918  */
 919 static bool vxlan_snoop(struct net_device *dev,
 920                         union vxlan_addr *src_ip, const u8 *src_mac)
 921 {
 922         struct vxlan_dev *vxlan = netdev_priv(dev);
 923         struct vxlan_fdb *f;
 924
 925         f = vxlan_find_mac(vxlan, src_mac);
 926         if (likely(f)) {
 927                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 928
 929                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 930                         return false;
 931
 932                 /* Don't migrate static entries, drop packets */
 933                 if (f->state & NUD_NOARP)
 934                         return true;
 935
 936                 if (net_ratelimit())
 937                         netdev_info(dev,
 938                                     "%pM migrated from %pIS to %pIS\n",
 939                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 940
 941                 rdst->remote_ip = *src_ip;
 942                 f->updated = jiffies;
 943                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 944         } else {
 945                 /* learned new entry */
 946                 spin_lock(&vxlan->hash_lock);
 947
 948                 /* close off race between vxlan_flush and incoming packets */
 949                 if (netif_running(dev))
 950                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 951                                          NUD_REACHABLE,
 952                                          NLM_F_EXCL|NLM_F_CREATE,
 953                                          vxlan->cfg.dst_port,
 954                                          vxlan->default_dst.remote_vni,
 955                                          0, NTF_SELF);
 956                 spin_unlock(&vxlan->hash_lock);
 957         }
 958
 959         return false;
 960 }
 961
 962 /* See if multicast group is already in use by other ID */
 963 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
 964 {
 965         struct vxlan_dev *vxlan;
 966         struct vxlan_sock *sock4;
 967 #if IS_ENABLED(CONFIG_IPV6)
 968         struct vxlan_sock *sock6;
 969 #endif
 970         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
 971
 972         sock4 = rtnl_dereference(dev->vn4_sock);
 973
 974         /* The vxlan_sock is only used by dev, leaving group has
 975          * no effect on other vxlan devices.
 976          */
 977         if (family == AF_INET && sock4 && atomic_read(&sock4->refcnt) == 1)
 978                 return false;
 979 #if IS_ENABLED(CONFIG_IPV6)
 980         sock6 = rtnl_dereference(dev->vn6_sock);
 981         if (family == AF_INET6 && sock6 && atomic_read(&sock6->refcnt) == 1)
 982                 return false;
 983 #endif
 984
 985         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
 986                 if (!netif_running(vxlan->dev) || vxlan == dev)
 987                         continue;
 988
 989                 if (family == AF_INET &&
 990                     rtnl_dereference(vxlan->vn4_sock) != sock4)
 991                         continue;
 992 #if IS_ENABLED(CONFIG_IPV6)
 993                 if (family == AF_INET6 &&
 994                     rtnl_dereference(vxlan->vn6_sock) != sock6)
 995                         continue;
 996 #endif
 997
 998                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
 999                                       &dev->default_dst.remote_ip))
1000                         continue;
1001
1002                 if (vxlan->default_dst.remote_ifindex !=
1003                     dev->default_dst.remote_ifindex)
1004                         continue;
1005
1006                 return true;
1007         }
1008
1009         return false;
1010 }
1011
1012 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1013 {
1014         struct vxlan_net *vn;
1015
1016         if (!vs)
1017                 return false;
1018         if (!atomic_dec_and_test(&vs->refcnt))
1019                 return false;
1020
1021         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1022         spin_lock(&vn->sock_lock);
1023         hlist_del_rcu(&vs->hlist);
1024         udp_tunnel_notify_del_rx_port(vs->sock,
1025                                       (vs->flags & VXLAN_F_GPE) ?
1026                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1027                                       UDP_TUNNEL_TYPE_VXLAN);
1028         spin_unlock(&vn->sock_lock);
1029
1030         return true;
1031 }
1032
1033 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1034 {
1035         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1036 #if IS_ENABLED(CONFIG_IPV6)
1037         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1038
1039         rcu_assign_pointer(vxlan->vn6_sock, NULL);
1040 #endif
1041
1042         rcu_assign_pointer(vxlan->vn4_sock, NULL);
1043         synchronize_net();
1044
1045         vxlan_vs_del_dev(vxlan);
1046
1047         if (__vxlan_sock_release_prep(sock4)) {
1048                 udp_tunnel_sock_release(sock4->sock);
1049                 kfree(sock4);
1050         }
1051
1052 #if IS_ENABLED(CONFIG_IPV6)
1053         if (__vxlan_sock_release_prep(sock6)) {
1054                 udp_tunnel_sock_release(sock6->sock);
1055                 kfree(sock6);
1056         }
1057 #endif
1058 }
1059
1060 /* Update multicast group membership when first VNI on
1061  * multicast address is brought up
1062  */
1063 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1064 {
1065         struct sock *sk;
1066         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1067         int ifindex = vxlan->default_dst.remote_ifindex;
1068         int ret = -EINVAL;
1069
1070         if (ip->sa.sa_family == AF_INET) {
1071                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1072                 struct ip_mreqn mreq = {
1073                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1074                         .imr_ifindex            = ifindex,
1075                 };
1076
1077                 sk = sock4->sock->sk;
1078                 lock_sock(sk);
1079                 ret = ip_mc_join_group(sk, &mreq);
1080                 release_sock(sk);
1081 #if IS_ENABLED(CONFIG_IPV6)
1082         } else {
1083                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1084
1085                 sk = sock6->sock->sk;
1086                 lock_sock(sk);
1087                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1088                                                    &ip->sin6.sin6_addr);
1089                 release_sock(sk);
1090 #endif
1091         }
1092
1093         return ret;
1094 }
1095
1096 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1097 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1098 {
1099         struct sock *sk;
1100         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1101         int ifindex = vxlan->default_dst.remote_ifindex;
1102         int ret = -EINVAL;
1103
1104         if (ip->sa.sa_family == AF_INET) {
1105                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1106                 struct ip_mreqn mreq = {
1107                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1108                         .imr_ifindex            = ifindex,
1109                 };
1110
1111                 sk = sock4->sock->sk;
1112                 lock_sock(sk);
1113                 ret = ip_mc_leave_group(sk, &mreq);
1114                 release_sock(sk);
1115 #if IS_ENABLED(CONFIG_IPV6)
1116         } else {
1117                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1118
1119                 sk = sock6->sock->sk;
1120                 lock_sock(sk);
1121                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1122                                                    &ip->sin6.sin6_addr);
1123                 release_sock(sk);
1124 #endif
1125         }
1126
1127         return ret;
1128 }
1129
1130 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1131                           struct sk_buff *skb, u32 vxflags)
1132 {
1133         size_t start, offset;
1134
1135         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1136                 goto out;
1137
1138         start = vxlan_rco_start(unparsed->vx_vni);
1139         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1140
1141         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1142                 return false;
1143
1144         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1145                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1146 out:
1147         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1148         unparsed->vx_vni &= VXLAN_VNI_MASK;
1149         return true;
1150 }
1151
1152 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1153                                 struct sk_buff *skb, u32 vxflags,
1154                                 struct vxlan_metadata *md)
1155 {
1156         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1157         struct metadata_dst *tun_dst;
1158
1159         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1160                 goto out;
1161
1162         md->gbp = ntohs(gbp->policy_id);
1163
1164         tun_dst = (struct metadata_dst *)skb_dst(skb);
1165         if (tun_dst) {
1166                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1167                 tun_dst->u.tun_info.options_len = sizeof(*md);
1168         }
1169         if (gbp->dont_learn)
1170                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1171
1172         if (gbp->policy_applied)
1173                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1174
1175         /* In flow-based mode, GBP is carried in dst_metadata */
1176         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1177                 skb->mark = md->gbp;
1178 out:
1179         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1180 }
1181
1182 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1183                                 __be16 *protocol,
1184                                 struct sk_buff *skb, u32 vxflags)
1185 {
1186         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1187
1188         /* Need to have Next Protocol set for interfaces in GPE mode. */
1189         if (!gpe->np_applied)
1190                 return false;
1191         /* "The initial version is 0. If a receiver does not support the
1192          * version indicated it MUST drop the packet.
1193          */
1194         if (gpe->version != 0)
1195                 return false;
1196         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1197          * processing MUST occur." However, we don't implement OAM
1198          * processing, thus drop the packet.
1199          */
1200         if (gpe->oam_flag)
1201                 return false;
1202
1203         switch (gpe->next_protocol) {
1204         case VXLAN_GPE_NP_IPV4:
1205                 *protocol = htons(ETH_P_IP);
1206                 break;
1207         case VXLAN_GPE_NP_IPV6:
1208                 *protocol = htons(ETH_P_IPV6);
1209                 break;
1210         case VXLAN_GPE_NP_ETHERNET:
1211                 *protocol = htons(ETH_P_TEB);
1212                 break;
1213         default:
1214                 return false;
1215         }
1216
1217         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1218         return true;
1219 }
1220
1221 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1222                           struct vxlan_sock *vs,
1223                           struct sk_buff *skb)
1224 {
1225         union vxlan_addr saddr;
1226
1227         skb_reset_mac_header(skb);
1228         skb->protocol = eth_type_trans(skb, vxlan->dev);
1229         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1230
1231         /* Ignore packet loops (and multicast echo) */
1232         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1233                 return false;
1234
1235         /* Get address from the outer IP header */
1236         if (vxlan_get_sk_family(vs) == AF_INET) {
1237                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1238                 saddr.sa.sa_family = AF_INET;
1239 #if IS_ENABLED(CONFIG_IPV6)
1240         } else {
1241                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1242                 saddr.sa.sa_family = AF_INET6;
1243 #endif
1244         }
1245
1246         if ((vxlan->flags & VXLAN_F_LEARN) &&
1247             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1248                 return false;
1249
1250         return true;
1251 }
1252
1253 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1254                                   struct sk_buff *skb)
1255 {
1256         int err = 0;
1257
1258         if (vxlan_get_sk_family(vs) == AF_INET)
1259                 err = IP_ECN_decapsulate(oiph, skb);
1260 #if IS_ENABLED(CONFIG_IPV6)
1261         else
1262                 err = IP6_ECN_decapsulate(oiph, skb);
1263 #endif
1264
1265         if (unlikely(err) && log_ecn_error) {
1266                 if (vxlan_get_sk_family(vs) == AF_INET)
1267                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1268                                              &((struct iphdr *)oiph)->saddr,
1269                                              ((struct iphdr *)oiph)->tos);
1270                 else
1271                         net_info_ratelimited("non-ECT from %pI6\n",
1272                                              &((struct ipv6hdr *)oiph)->saddr);
1273         }
1274         return err <= 1;
1275 }
1276
1277 /* Callback from net/ipv4/udp.c to receive packets */
1278 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1279 {
1280         struct pcpu_sw_netstats *stats;
1281         struct vxlan_dev *vxlan;
1282         struct vxlan_sock *vs;
1283         struct vxlanhdr unparsed;
1284         struct vxlan_metadata _md;
1285         struct vxlan_metadata *md = &_md;
1286         __be16 protocol = htons(ETH_P_TEB);
1287         bool raw_proto = false;
1288         void *oiph;
1289
1290         /* Need UDP and VXLAN header to be present */
1291         if (!pskb_may_pull(skb, VXLAN_HLEN))
1292                 goto drop;
1293
1294         unparsed = *vxlan_hdr(skb);
1295         /* VNI flag always required to be set */
1296         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1297                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1298                            ntohl(vxlan_hdr(skb)->vx_flags),
1299                            ntohl(vxlan_hdr(skb)->vx_vni));
1300                 /* Return non vxlan pkt */
1301                 goto drop;
1302         }
1303         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1304         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1305
1306         vs = rcu_dereference_sk_user_data(sk);
1307         if (!vs)
1308                 goto drop;
1309
1310         vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1311         if (!vxlan)
1312                 goto drop;
1313
1314         /* For backwards compatibility, only allow reserved fields to be
1315          * used by VXLAN extensions if explicitly requested.
1316          */
1317         if (vs->flags & VXLAN_F_GPE) {
1318                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1319                         goto drop;
1320                 raw_proto = true;
1321         }
1322
1323         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1324                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1325                         goto drop;
1326
1327         if (vxlan_collect_metadata(vs)) {
1328                 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1329                 struct metadata_dst *tun_dst;
1330
1331                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1332                                          key32_to_tunnel_id(vni), sizeof(*md));
1333
1334                 if (!tun_dst)
1335                         goto drop;
1336
1337                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1338
1339                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1340         } else {
1341                 memset(md, 0, sizeof(*md));
1342         }
1343
1344         if (vs->flags & VXLAN_F_REMCSUM_RX)
1345                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1346                         goto drop;
1347         if (vs->flags & VXLAN_F_GBP)
1348                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1349         /* Note that GBP and GPE can never be active together. This is
1350          * ensured in vxlan_dev_configure.
1351          */
1352
1353         if (unparsed.vx_flags || unparsed.vx_vni) {
1354                 /* If there are any unprocessed flags remaining treat
1355                  * this as a malformed packet. This behavior diverges from
1356                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1357                  * in reserved fields are to be ignored. The approach here
1358                  * maintains compatibility with previous stack code, and also
1359                  * is more robust and provides a little more security in
1360                  * adding extensions to VXLAN.
1361                  */
1362                 goto drop;
1363         }
1364
1365         if (!raw_proto) {
1366                 if (!vxlan_set_mac(vxlan, vs, skb))
1367                         goto drop;
1368         } else {
1369                 skb_reset_mac_header(skb);
1370                 skb->dev = vxlan->dev;
1371                 skb->pkt_type = PACKET_HOST;
1372         }
1373
1374         oiph = skb_network_header(skb);
1375         skb_reset_network_header(skb);
1376
1377         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1378                 ++vxlan->dev->stats.rx_frame_errors;
1379                 ++vxlan->dev->stats.rx_errors;
1380                 goto drop;
1381         }
1382
1383         stats = this_cpu_ptr(vxlan->dev->tstats);
1384         u64_stats_update_begin(&stats->syncp);
1385         stats->rx_packets++;
1386         stats->rx_bytes += skb->len;
1387         u64_stats_update_end(&stats->syncp);
1388
1389         gro_cells_receive(&vxlan->gro_cells, skb);
1390         return 0;
1391
1392 drop:
1393         /* Consume bad packet */
1394         kfree_skb(skb);
1395         return 0;
1396 }
1397
1398 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1399 {
1400         struct vxlan_dev *vxlan = netdev_priv(dev);
1401         struct arphdr *parp;
1402         u8 *arpptr, *sha;
1403         __be32 sip, tip;
1404         struct neighbour *n;
1405
1406         if (dev->flags & IFF_NOARP)
1407                 goto out;
1408
1409         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1410                 dev->stats.tx_dropped++;
1411                 goto out;
1412         }
1413         parp = arp_hdr(skb);
1414
1415         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1416              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1417             parp->ar_pro != htons(ETH_P_IP) ||
1418             parp->ar_op != htons(ARPOP_REQUEST) ||
1419             parp->ar_hln != dev->addr_len ||
1420             parp->ar_pln != 4)
1421                 goto out;
1422         arpptr = (u8 *)parp + sizeof(struct arphdr);
1423         sha = arpptr;
1424         arpptr += dev->addr_len;        /* sha */
1425         memcpy(&sip, arpptr, sizeof(sip));
1426         arpptr += sizeof(sip);
1427         arpptr += dev->addr_len;        /* tha */
1428         memcpy(&tip, arpptr, sizeof(tip));
1429
1430         if (ipv4_is_loopback(tip) ||
1431             ipv4_is_multicast(tip))
1432                 goto out;
1433
1434         n = neigh_lookup(&arp_tbl, &tip, dev);
1435
1436         if (n) {
1437                 struct vxlan_fdb *f;
1438                 struct sk_buff  *reply;
1439
1440                 if (!(n->nud_state & NUD_CONNECTED)) {
1441                         neigh_release(n);
1442                         goto out;
1443                 }
1444
1445                 f = vxlan_find_mac(vxlan, n->ha);
1446                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1447                         /* bridge-local neighbor */
1448                         neigh_release(n);
1449                         goto out;
1450                 }
1451
1452                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1453                                 n->ha, sha);
1454
1455                 neigh_release(n);
1456
1457                 if (reply == NULL)
1458                         goto out;
1459
1460                 skb_reset_mac_header(reply);
1461                 __skb_pull(reply, skb_network_offset(reply));
1462                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1463                 reply->pkt_type = PACKET_HOST;
1464
1465                 if (netif_rx_ni(reply) == NET_RX_DROP)
1466                         dev->stats.rx_dropped++;
1467         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1468                 union vxlan_addr ipa = {
1469                         .sin.sin_addr.s_addr = tip,
1470                         .sin.sin_family = AF_INET,
1471                 };
1472
1473                 vxlan_ip_miss(dev, &ipa);
1474         }
1475 out:
1476         consume_skb(skb);
1477         return NETDEV_TX_OK;
1478 }
1479
1480 #if IS_ENABLED(CONFIG_IPV6)
1481 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1482         struct neighbour *n, bool isrouter)
1483 {
1484         struct net_device *dev = request->dev;
1485         struct sk_buff *reply;
1486         struct nd_msg *ns, *na;
1487         struct ipv6hdr *pip6;
1488         u8 *daddr;
1489         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1490         int ns_olen;
1491         int i, len;
1492
1493         if (dev == NULL)
1494                 return NULL;
1495
1496         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1497                 sizeof(*na) + na_olen + dev->needed_tailroom;
1498         reply = alloc_skb(len, GFP_ATOMIC);
1499         if (reply == NULL)
1500                 return NULL;
1501
1502         reply->protocol = htons(ETH_P_IPV6);
1503         reply->dev = dev;
1504         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1505         skb_push(reply, sizeof(struct ethhdr));
1506         skb_reset_mac_header(reply);
1507
1508         ns = (struct nd_msg *)skb_transport_header(request);
1509
1510         daddr = eth_hdr(request)->h_source;
1511         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1512         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1513                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1514                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1515                         break;
1516                 }
1517         }
1518
1519         /* Ethernet header */
1520         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1521         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1522         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1523         reply->protocol = htons(ETH_P_IPV6);
1524
1525         skb_pull(reply, sizeof(struct ethhdr));
1526         skb_reset_network_header(reply);
1527         skb_put(reply, sizeof(struct ipv6hdr));
1528
1529         /* IPv6 header */
1530
1531         pip6 = ipv6_hdr(reply);
1532         memset(pip6, 0, sizeof(struct ipv6hdr));
1533         pip6->version = 6;
1534         pip6->priority = ipv6_hdr(request)->priority;
1535         pip6->nexthdr = IPPROTO_ICMPV6;
1536         pip6->hop_limit = 255;
1537         pip6->daddr = ipv6_hdr(request)->saddr;
1538         pip6->saddr = *(struct in6_addr *)n->primary_key;
1539
1540         skb_pull(reply, sizeof(struct ipv6hdr));
1541         skb_reset_transport_header(reply);
1542
1543         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1544
1545         /* Neighbor Advertisement */
1546         memset(na, 0, sizeof(*na)+na_olen);
1547         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1548         na->icmph.icmp6_router = isrouter;
1549         na->icmph.icmp6_override = 1;
1550         na->icmph.icmp6_solicited = 1;
1551         na->target = ns->target;
1552         ether_addr_copy(&na->opt[2], n->ha);
1553         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1554         na->opt[1] = na_olen >> 3;
1555
1556         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1557                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1558                 csum_partial(na, sizeof(*na)+na_olen, 0));
1559
1560         pip6->payload_len = htons(sizeof(*na)+na_olen);
1561
1562         skb_push(reply, sizeof(struct ipv6hdr));
1563
1564         reply->ip_summed = CHECKSUM_UNNECESSARY;
1565
1566         return reply;
1567 }
1568
1569 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1570 {
1571         struct vxlan_dev *vxlan = netdev_priv(dev);
1572         struct nd_msg *msg;
1573         const struct ipv6hdr *iphdr;
1574         const struct in6_addr *saddr, *daddr;
1575         struct neighbour *n;
1576         struct inet6_dev *in6_dev;
1577
1578         in6_dev = __in6_dev_get(dev);
1579         if (!in6_dev)
1580                 goto out;
1581
1582         iphdr = ipv6_hdr(skb);
1583         saddr = &iphdr->saddr;
1584         daddr = &iphdr->daddr;
1585
1586         msg = (struct nd_msg *)skb_transport_header(skb);
1587         if (msg->icmph.icmp6_code != 0 ||
1588             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1589                 goto out;
1590
1591         if (ipv6_addr_loopback(daddr) ||
1592             ipv6_addr_is_multicast(&msg->target))
1593                 goto out;
1594
1595         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1596
1597         if (n) {
1598                 struct vxlan_fdb *f;
1599                 struct sk_buff *reply;
1600
1601                 if (!(n->nud_state & NUD_CONNECTED)) {
1602                         neigh_release(n);
1603                         goto out;
1604                 }
1605
1606                 f = vxlan_find_mac(vxlan, n->ha);
1607                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1608                         /* bridge-local neighbor */
1609                         neigh_release(n);
1610                         goto out;
1611                 }
1612
1613                 reply = vxlan_na_create(skb, n,
1614                                         !!(f ? f->flags & NTF_ROUTER : 0));
1615
1616                 neigh_release(n);
1617
1618                 if (reply == NULL)
1619                         goto out;
1620
1621                 if (netif_rx_ni(reply) == NET_RX_DROP)
1622                         dev->stats.rx_dropped++;
1623
1624         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1625                 union vxlan_addr ipa = {
1626                         .sin6.sin6_addr = msg->target,
1627                         .sin6.sin6_family = AF_INET6,
1628                 };
1629
1630                 vxlan_ip_miss(dev, &ipa);
1631         }
1632
1633 out:
1634         consume_skb(skb);
1635         return NETDEV_TX_OK;
1636 }
1637 #endif
1638
1639 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1640 {
1641         struct vxlan_dev *vxlan = netdev_priv(dev);
1642         struct neighbour *n;
1643
1644         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1645                 return false;
1646
1647         n = NULL;
1648         switch (ntohs(eth_hdr(skb)->h_proto)) {
1649         case ETH_P_IP:
1650         {
1651                 struct iphdr *pip;
1652
1653                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1654                         return false;
1655                 pip = ip_hdr(skb);
1656                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1657                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1658                         union vxlan_addr ipa = {
1659                                 .sin.sin_addr.s_addr = pip->daddr,
1660                                 .sin.sin_family = AF_INET,
1661                         };
1662
1663                         vxlan_ip_miss(dev, &ipa);
1664                         return false;
1665                 }
1666
1667                 break;
1668         }
1669 #if IS_ENABLED(CONFIG_IPV6)
1670         case ETH_P_IPV6:
1671         {
1672                 struct ipv6hdr *pip6;
1673
1674                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1675                         return false;
1676                 pip6 = ipv6_hdr(skb);
1677                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1678                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1679                         union vxlan_addr ipa = {
1680                                 .sin6.sin6_addr = pip6->daddr,
1681                                 .sin6.sin6_family = AF_INET6,
1682                         };
1683
1684                         vxlan_ip_miss(dev, &ipa);
1685                         return false;
1686                 }
1687
1688                 break;
1689         }
1690 #endif
1691         default:
1692                 return false;
1693         }
1694
1695         if (n) {
1696                 bool diff;
1697
1698                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1699                 if (diff) {
1700                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1701                                 dev->addr_len);
1702                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1703                 }
1704                 neigh_release(n);
1705                 return diff;
1706         }
1707
1708         return false;
1709 }
1710
1711 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1712                                 struct vxlan_metadata *md)
1713 {
1714         struct vxlanhdr_gbp *gbp;
1715
1716         if (!md->gbp)
1717                 return;
1718
1719         gbp = (struct vxlanhdr_gbp *)vxh;
1720         vxh->vx_flags |= VXLAN_HF_GBP;
1721
1722         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1723                 gbp->dont_learn = 1;
1724
1725         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1726                 gbp->policy_applied = 1;
1727
1728         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1729 }
1730
1731 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1732                                __be16 protocol)
1733 {
1734         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1735
1736         gpe->np_applied = 1;
1737
1738         switch (protocol) {
1739         case htons(ETH_P_IP):
1740                 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1741                 return 0;
1742         case htons(ETH_P_IPV6):
1743                 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1744                 return 0;
1745         case htons(ETH_P_TEB):
1746                 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1747                 return 0;
1748         }
1749         return -EPFNOSUPPORT;
1750 }
1751
1752 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1753                            int iphdr_len, __be32 vni,
1754                            struct vxlan_metadata *md, u32 vxflags,
1755                            bool udp_sum)
1756 {
1757         struct vxlanhdr *vxh;
1758         int min_headroom;
1759         int err;
1760         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1761         __be16 inner_protocol = htons(ETH_P_TEB);
1762
1763         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1764             skb->ip_summed == CHECKSUM_PARTIAL) {
1765                 int csum_start = skb_checksum_start_offset(skb);
1766
1767                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1768                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1769                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1770                      skb->csum_offset == offsetof(struct tcphdr, check)))
1771                         type |= SKB_GSO_TUNNEL_REMCSUM;
1772         }
1773
1774         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1775                         + VXLAN_HLEN + iphdr_len
1776                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1777
1778         /* Need space for new headers (invalidates iph ptr) */
1779         err = skb_cow_head(skb, min_headroom);
1780         if (unlikely(err))
1781                 goto out_free;
1782
1783         skb = vlan_hwaccel_push_inside(skb);
1784         if (WARN_ON(!skb))
1785                 return -ENOMEM;
1786
1787         err = iptunnel_handle_offloads(skb, type);
1788         if (err)
1789                 goto out_free;
1790
1791         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1792         vxh->vx_flags = VXLAN_HF_VNI;
1793         vxh->vx_vni = vxlan_vni_field(vni);
1794
1795         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1796                 unsigned int start;
1797
1798                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1799                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1800                 vxh->vx_flags |= VXLAN_HF_RCO;
1801
1802                 if (!skb_is_gso(skb)) {
1803                         skb->ip_summed = CHECKSUM_NONE;
1804                         skb->encapsulation = 0;
1805                 }
1806         }
1807
1808         if (vxflags & VXLAN_F_GBP)
1809                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1810         if (vxflags & VXLAN_F_GPE) {
1811                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1812                 if (err < 0)
1813                         goto out_free;
1814                 inner_protocol = skb->protocol;
1815         }
1816
1817         skb_set_inner_protocol(skb, inner_protocol);
1818         return 0;
1819
1820 out_free:
1821         kfree_skb(skb);
1822         return err;
1823 }
1824
1825 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1826                                       struct sk_buff *skb, int oif, u8 tos,
1827                                       __be32 daddr, __be32 *saddr,
1828                                       struct dst_cache *dst_cache,
1829                                       const struct ip_tunnel_info *info)
1830 {
1831         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1832         struct rtable *rt = NULL;
1833         struct flowi4 fl4;
1834
1835         if (tos && !info)
1836                 use_cache = false;
1837         if (use_cache) {
1838                 rt = dst_cache_get_ip4(dst_cache, saddr);
1839                 if (rt)
1840                         return rt;
1841         }
1842
1843         memset(&fl4, 0, sizeof(fl4));
1844         fl4.flowi4_oif = oif;
1845         fl4.flowi4_tos = RT_TOS(tos);
1846         fl4.flowi4_mark = skb->mark;
1847         fl4.flowi4_proto = IPPROTO_UDP;
1848         fl4.daddr = daddr;
1849         fl4.saddr = *saddr;
1850
1851         rt = ip_route_output_key(vxlan->net, &fl4);
1852         if (!IS_ERR(rt)) {
1853                 *saddr = fl4.saddr;
1854                 if (use_cache)
1855                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1856         }
1857         return rt;
1858 }
1859
1860 #if IS_ENABLED(CONFIG_IPV6)
1861 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1862                                           struct sk_buff *skb, int oif, u8 tos,
1863                                           __be32 label,
1864                                           const struct in6_addr *daddr,
1865                                           struct in6_addr *saddr,
1866                                           struct dst_cache *dst_cache,
1867                                           const struct ip_tunnel_info *info)
1868 {
1869         struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
1870         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1871         struct dst_entry *ndst;
1872         struct flowi6 fl6;
1873         int err;
1874
1875         if (!sock6)
1876                 return ERR_PTR(-EIO);
1877
1878         if (tos && !info)
1879                 use_cache = false;
1880         if (use_cache) {
1881                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1882                 if (ndst)
1883                         return ndst;
1884         }
1885
1886         memset(&fl6, 0, sizeof(fl6));
1887         fl6.flowi6_oif = oif;
1888         fl6.daddr = *daddr;
1889         fl6.saddr = *saddr;
1890         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1891         fl6.flowi6_mark = skb->mark;
1892         fl6.flowi6_proto = IPPROTO_UDP;
1893
1894         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1895                                          sock6->sock->sk,
1896                                          &ndst, &fl6);
1897         if (err < 0)
1898                 return ERR_PTR(err);
1899
1900         *saddr = fl6.saddr;
1901         if (use_cache)
1902                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1903         return ndst;
1904 }
1905 #endif
1906
1907 /* Bypass encapsulation if the destination is local */
1908 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1909                                struct vxlan_dev *dst_vxlan)
1910 {
1911         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1912         union vxlan_addr loopback;
1913         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1914         struct net_device *dev = skb->dev;
1915         int len = skb->len;
1916
1917         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1918         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1919         skb->pkt_type = PACKET_HOST;
1920         skb->encapsulation = 0;
1921         skb->dev = dst_vxlan->dev;
1922         __skb_pull(skb, skb_network_offset(skb));
1923
1924         if (remote_ip->sa.sa_family == AF_INET) {
1925                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1926                 loopback.sa.sa_family =  AF_INET;
1927 #if IS_ENABLED(CONFIG_IPV6)
1928         } else {
1929                 loopback.sin6.sin6_addr = in6addr_loopback;
1930                 loopback.sa.sa_family =  AF_INET6;
1931 #endif
1932         }
1933
1934         if (dst_vxlan->flags & VXLAN_F_LEARN)
1935                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1936
1937         u64_stats_update_begin(&tx_stats->syncp);
1938         tx_stats->tx_packets++;
1939         tx_stats->tx_bytes += len;
1940         u64_stats_update_end(&tx_stats->syncp);
1941
1942         if (netif_rx(skb) == NET_RX_SUCCESS) {
1943                 u64_stats_update_begin(&rx_stats->syncp);
1944                 rx_stats->rx_packets++;
1945                 rx_stats->rx_bytes += len;
1946                 u64_stats_update_end(&rx_stats->syncp);
1947         } else {
1948                 dev->stats.rx_dropped++;
1949         }
1950 }
1951
1952 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1953                            struct vxlan_rdst *rdst, bool did_rsc)
1954 {
1955         struct dst_cache *dst_cache;
1956         struct ip_tunnel_info *info;
1957         struct vxlan_dev *vxlan = netdev_priv(dev);
1958         struct sock *sk;
1959         struct rtable *rt = NULL;
1960         const struct iphdr *old_iph;
1961         union vxlan_addr *dst;
1962         union vxlan_addr remote_ip, local_ip;
1963         struct vxlan_metadata _md;
1964         struct vxlan_metadata *md = &_md;
1965         __be16 src_port = 0, dst_port;
1966         __be32 vni, label;
1967         __be16 df = 0;
1968         __u8 tos, ttl;
1969         int err;
1970         u32 flags = vxlan->flags;
1971         bool udp_sum = false;
1972         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1973
1974         info = skb_tunnel_info(skb);
1975
1976         rcu_read_lock();
1977         if (rdst) {
1978                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1979                 vni = rdst->remote_vni;
1980                 dst = &rdst->remote_ip;
1981                 local_ip = vxlan->cfg.saddr;
1982                 dst_cache = &rdst->dst_cache;
1983         } else {
1984                 if (!info) {
1985                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1986                                   dev->name);
1987                         goto drop;
1988                 }
1989                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1990                 vni = tunnel_id_to_key32(info->key.tun_id);
1991                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1992                 if (remote_ip.sa.sa_family == AF_INET) {
1993                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1994                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
1995                 } else {
1996                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1997                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
1998                 }
1999                 dst = &remote_ip;
2000                 dst_cache = &info->dst_cache;
2001         }
2002
2003         if (vxlan_addr_any(dst)) {
2004                 if (did_rsc) {
2005                         /* short-circuited back to local bridge */
2006                         vxlan_encap_bypass(skb, vxlan, vxlan);
2007                         goto out_unlock;
2008                 }
2009                 goto drop;
2010         }
2011
2012         old_iph = ip_hdr(skb);
2013
2014         ttl = vxlan->cfg.ttl;
2015         if (!ttl && vxlan_addr_multicast(dst))
2016                 ttl = 1;
2017
2018         tos = vxlan->cfg.tos;
2019         if (tos == 1)
2020                 tos = ip_tunnel_get_dsfield(old_iph, skb);
2021
2022         label = vxlan->cfg.label;
2023         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2024                                      vxlan->cfg.port_max, true);
2025
2026         if (info) {
2027                 ttl = info->key.ttl;
2028                 tos = info->key.tos;
2029                 label = info->key.label;
2030                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2031
2032                 if (info->options_len)
2033                         md = ip_tunnel_info_opts(info);
2034         } else {
2035                 md->gbp = skb->mark;
2036         }
2037
2038         if (dst->sa.sa_family == AF_INET) {
2039                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2040
2041                 if (!sock4)
2042                         goto drop;
2043                 sk = sock4->sock->sk;
2044
2045                 rt = vxlan_get_route(vxlan, skb,
2046                                      rdst ? rdst->remote_ifindex : 0, tos,
2047                                      dst->sin.sin_addr.s_addr,
2048                                      &local_ip.sin.sin_addr.s_addr,
2049                                      dst_cache, info);
2050                 if (IS_ERR(rt)) {
2051                         netdev_dbg(dev, "no route to %pI4\n",
2052                                    &dst->sin.sin_addr.s_addr);
2053                         dev->stats.tx_carrier_errors++;
2054                         goto tx_error;
2055                 }
2056
2057                 if (rt->dst.dev == dev) {
2058                         netdev_dbg(dev, "circular route to %pI4\n",
2059                                    &dst->sin.sin_addr.s_addr);
2060                         dev->stats.collisions++;
2061                         goto rt_tx_error;
2062                 }
2063
2064                 /* Bypass encapsulation if the destination is local */
2065                 if (!info && rt->rt_flags & RTCF_LOCAL &&
2066                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2067                         struct vxlan_dev *dst_vxlan;
2068
2069                         ip_rt_put(rt);
2070                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2071                                                    dst->sa.sa_family, dst_port,
2072                                                    vxlan->flags);
2073                         if (!dst_vxlan)
2074                                 goto tx_error;
2075                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2076                         goto out_unlock;
2077                 }
2078
2079                 if (!info)
2080                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2081                 else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2082                         df = htons(IP_DF);
2083
2084                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2085                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2086                 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2087                                       vni, md, flags, udp_sum);
2088                 if (err < 0)
2089                         goto xmit_tx_error;
2090
2091                 udp_tunnel_xmit_skb(rt, sk, skb, local_ip.sin.sin_addr.s_addr,
2092                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2093                                     src_port, dst_port, xnet, !udp_sum);
2094 #if IS_ENABLED(CONFIG_IPV6)
2095         } else {
2096                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2097                 struct dst_entry *ndst;
2098                 u32 rt6i_flags;
2099
2100                 if (!sock6)
2101                         goto drop;
2102                 sk = sock6->sock->sk;
2103
2104                 ndst = vxlan6_get_route(vxlan, skb,
2105                                         rdst ? rdst->remote_ifindex : 0, tos,
2106                                         label, &dst->sin6.sin6_addr,
2107                                         &local_ip.sin6.sin6_addr,
2108                                         dst_cache, info);
2109                 if (IS_ERR(ndst)) {
2110                         netdev_dbg(dev, "no route to %pI6\n",
2111                                    &dst->sin6.sin6_addr);
2112                         dev->stats.tx_carrier_errors++;
2113                         goto tx_error;
2114                 }
2115
2116                 if (ndst->dev == dev) {
2117                         netdev_dbg(dev, "circular route to %pI6\n",
2118                                    &dst->sin6.sin6_addr);
2119                         dst_release(ndst);
2120                         dev->stats.collisions++;
2121                         goto tx_error;
2122                 }
2123
2124                 /* Bypass encapsulation if the destination is local */
2125                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2126                 if (!info && rt6i_flags & RTF_LOCAL &&
2127                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2128                         struct vxlan_dev *dst_vxlan;
2129
2130                         dst_release(ndst);
2131                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2132                                                    dst->sa.sa_family, dst_port,
2133                                                    vxlan->flags);
2134                         if (!dst_vxlan)
2135                                 goto tx_error;
2136                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2137                         goto out_unlock;
2138                 }
2139
2140                 if (!info)
2141                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2142
2143                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2144                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2145                 skb_scrub_packet(skb, xnet);
2146                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2147                                       vni, md, flags, udp_sum);
2148                 if (err < 0) {
2149                         dst_release(ndst);
2150                         dev->stats.tx_errors++;
2151                         goto out_unlock;
2152                 }
2153                 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2154                                      &local_ip.sin6.sin6_addr,
2155                                      &dst->sin6.sin6_addr, tos, ttl,
2156                                      label, src_port, dst_port, !udp_sum);
2157 #endif
2158         }
2159 out_unlock:
2160         rcu_read_unlock();
2161         return;
2162
2163 drop:
2164         dev->stats.tx_dropped++;
2165         goto tx_free;
2166
2167 xmit_tx_error:
2168         /* skb is already freed. */
2169         skb = NULL;
2170 rt_tx_error:
2171         ip_rt_put(rt);
2172 tx_error:
2173         dev->stats.tx_errors++;
2174 tx_free:
2175         dev_kfree_skb(skb);
2176         rcu_read_unlock();
2177 }
2178
2179 /* Transmit local packets over Vxlan
2180  *
2181  * Outer IP header inherits ECN and DF from inner header.
2182  * Outer UDP destination is the VXLAN assigned port.
2183  *           source port is based on hash of flow
2184  */
2185 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2186 {
2187         struct vxlan_dev *vxlan = netdev_priv(dev);
2188         const struct ip_tunnel_info *info;
2189         struct ethhdr *eth;
2190         bool did_rsc = false;
2191         struct vxlan_rdst *rdst, *fdst = NULL;
2192         struct vxlan_fdb *f;
2193
2194         info = skb_tunnel_info(skb);
2195
2196         skb_reset_mac_header(skb);
2197
2198         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2199                 if (info && info->mode & IP_TUNNEL_INFO_TX)
2200                         vxlan_xmit_one(skb, dev, NULL, false);
2201                 else
2202                         kfree_skb(skb);
2203                 return NETDEV_TX_OK;
2204         }
2205
2206         if (vxlan->flags & VXLAN_F_PROXY) {
2207                 eth = eth_hdr(skb);
2208                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2209                         return arp_reduce(dev, skb);
2210 #if IS_ENABLED(CONFIG_IPV6)
2211                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2212                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2213                                        + sizeof(struct nd_msg)) &&
2214                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2215                                 struct nd_msg *msg;
2216
2217                                 msg = (struct nd_msg *)skb_transport_header(skb);
2218                                 if (msg->icmph.icmp6_code == 0 &&
2219                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2220                                         return neigh_reduce(dev, skb);
2221                 }
2222 #endif
2223         }
2224
2225         eth = eth_hdr(skb);
2226         f = vxlan_find_mac(vxlan, eth->h_dest);
2227         did_rsc = false;
2228
2229         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2230             (ntohs(eth->h_proto) == ETH_P_IP ||
2231              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2232                 did_rsc = route_shortcircuit(dev, skb);
2233                 if (did_rsc)
2234                         f = vxlan_find_mac(vxlan, eth->h_dest);
2235         }
2236
2237         if (f == NULL) {
2238                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2239                 if (f == NULL) {
2240                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2241                             !is_multicast_ether_addr(eth->h_dest))
2242                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2243
2244                         dev->stats.tx_dropped++;
2245                         kfree_skb(skb);
2246                         return NETDEV_TX_OK;
2247                 }
2248         }
2249
2250         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2251                 struct sk_buff *skb1;
2252
2253                 if (!fdst) {
2254                         fdst = rdst;
2255                         continue;
2256                 }
2257                 skb1 = skb_clone(skb, GFP_ATOMIC);
2258                 if (skb1)
2259                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2260         }
2261
2262         if (fdst)
2263                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2264         else
2265                 kfree_skb(skb);
2266         return NETDEV_TX_OK;
2267 }
2268
2269 /* Walk the forwarding table and purge stale entries */
2270 static void vxlan_cleanup(unsigned long arg)
2271 {
2272         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2273         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2274         unsigned int h;
2275
2276         if (!netif_running(vxlan->dev))
2277                 return;
2278
2279         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2280                 struct hlist_node *p, *n;
2281
2282                 spin_lock_bh(&vxlan->hash_lock);
2283                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2284                         struct vxlan_fdb *f
2285                                 = container_of(p, struct vxlan_fdb, hlist);
2286                         unsigned long timeout;
2287
2288                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2289                                 continue;
2290
2291                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2292                         if (time_before_eq(timeout, jiffies)) {
2293                                 netdev_dbg(vxlan->dev,
2294                                            "garbage collect %pM\n",
2295                                            f->eth_addr);
2296                                 f->state = NUD_STALE;
2297                                 vxlan_fdb_destroy(vxlan, f);
2298                         } else if (time_before(timeout, next_timer))
2299                                 next_timer = timeout;
2300                 }
2301                 spin_unlock_bh(&vxlan->hash_lock);
2302         }
2303
2304         mod_timer(&vxlan->age_timer, next_timer);
2305 }
2306
2307 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2308 {
2309         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2310
2311         spin_lock(&vn->sock_lock);
2312         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2313 #if IS_ENABLED(CONFIG_IPV6)
2314         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2315 #endif
2316         spin_unlock(&vn->sock_lock);
2317 }
2318
2319 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2320                              struct vxlan_dev_node *node)
2321 {
2322         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2323         __be32 vni = vxlan->default_dst.remote_vni;
2324
2325         node->vxlan = vxlan;
2326         spin_lock(&vn->sock_lock);
2327         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2328         spin_unlock(&vn->sock_lock);
2329 }
2330
2331 /* Setup stats when device is created */
2332 static int vxlan_init(struct net_device *dev)
2333 {
2334         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2335         if (!dev->tstats)
2336                 return -ENOMEM;
2337
2338         return 0;
2339 }
2340
2341 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2342 {
2343         struct vxlan_fdb *f;
2344
2345         spin_lock_bh(&vxlan->hash_lock);
2346         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2347         if (f)
2348                 vxlan_fdb_destroy(vxlan, f);
2349         spin_unlock_bh(&vxlan->hash_lock);
2350 }
2351
2352 static void vxlan_uninit(struct net_device *dev)
2353 {
2354         struct vxlan_dev *vxlan = netdev_priv(dev);
2355
2356         vxlan_fdb_delete_default(vxlan);
2357
2358         free_percpu(dev->tstats);
2359 }
2360
2361 /* Start ageing timer and join group when device is brought up */
2362 static int vxlan_open(struct net_device *dev)
2363 {
2364         struct vxlan_dev *vxlan = netdev_priv(dev);
2365         int ret;
2366
2367         ret = vxlan_sock_add(vxlan);
2368         if (ret < 0)
2369                 return ret;
2370
2371         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2372                 ret = vxlan_igmp_join(vxlan);
2373                 if (ret == -EADDRINUSE)
2374                         ret = 0;
2375                 if (ret) {
2376                         vxlan_sock_release(vxlan);
2377                         return ret;
2378                 }
2379         }
2380
2381         if (vxlan->cfg.age_interval)
2382                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2383
2384         return ret;
2385 }
2386
2387 /* Purge the forwarding table */
2388 static void vxlan_flush(struct vxlan_dev *vxlan)
2389 {
2390         unsigned int h;
2391
2392         spin_lock_bh(&vxlan->hash_lock);
2393         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2394                 struct hlist_node *p, *n;
2395                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2396                         struct vxlan_fdb *f
2397                                 = container_of(p, struct vxlan_fdb, hlist);
2398                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2399                         if (!is_zero_ether_addr(f->eth_addr))
2400                                 vxlan_fdb_destroy(vxlan, f);
2401                 }
2402         }
2403         spin_unlock_bh(&vxlan->hash_lock);
2404 }
2405
2406 /* Cleanup timer and forwarding table on shutdown */
2407 static int vxlan_stop(struct net_device *dev)
2408 {
2409         struct vxlan_dev *vxlan = netdev_priv(dev);
2410         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2411         int ret = 0;
2412
2413         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2414             !vxlan_group_used(vn, vxlan))
2415                 ret = vxlan_igmp_leave(vxlan);
2416
2417         del_timer_sync(&vxlan->age_timer);
2418
2419         vxlan_flush(vxlan);
2420         vxlan_sock_release(vxlan);
2421
2422         return ret;
2423 }
2424
2425 /* Stub, nothing needs to be done. */
2426 static void vxlan_set_multicast_list(struct net_device *dev)
2427 {
2428 }
2429
2430 static int __vxlan_change_mtu(struct net_device *dev,
2431                               struct net_device *lowerdev,
2432                               struct vxlan_rdst *dst, int new_mtu, bool strict)
2433 {
2434         int max_mtu = IP_MAX_MTU;
2435
2436         if (lowerdev)
2437                 max_mtu = lowerdev->mtu;
2438
2439         if (dst->remote_ip.sa.sa_family == AF_INET6)
2440                 max_mtu -= VXLAN6_HEADROOM;
2441         else
2442                 max_mtu -= VXLAN_HEADROOM;
2443
2444         if (new_mtu < 68)
2445                 return -EINVAL;
2446
2447         if (new_mtu > max_mtu) {
2448                 if (strict)
2449                         return -EINVAL;
2450
2451                 new_mtu = max_mtu;
2452         }
2453
2454         dev->mtu = new_mtu;
2455         return 0;
2456 }
2457
2458 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2459 {
2460         struct vxlan_dev *vxlan = netdev_priv(dev);
2461         struct vxlan_rdst *dst = &vxlan->default_dst;
2462         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2463                                                          dst->remote_ifindex);
2464         return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2465 }
2466
2467 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2468 {
2469         struct vxlan_dev *vxlan = netdev_priv(dev);
2470         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2471         __be16 sport, dport;
2472
2473         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2474                                   vxlan->cfg.port_max, true);
2475         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2476
2477         if (ip_tunnel_info_af(info) == AF_INET) {
2478                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2479                 struct rtable *rt;
2480
2481                 if (!sock4)
2482                         return -EINVAL;
2483                 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2484                                      info->key.u.ipv4.dst,
2485                                      &info->key.u.ipv4.src,
2486                                      &info->dst_cache, info);
2487                 if (IS_ERR(rt))
2488                         return PTR_ERR(rt);
2489                 ip_rt_put(rt);
2490         } else {
2491 #if IS_ENABLED(CONFIG_IPV6)
2492                 struct dst_entry *ndst;
2493
2494                 ndst = vxlan6_get_route(vxlan, skb, 0, info->key.tos,
2495                                         info->key.label, &info->key.u.ipv6.dst,
2496                                         &info->key.u.ipv6.src,
2497                                         &info->dst_cache, info);
2498                 if (IS_ERR(ndst))
2499                         return PTR_ERR(ndst);
2500                 dst_release(ndst);
2501 #else /* !CONFIG_IPV6 */
2502                 return -EPFNOSUPPORT;
2503 #endif
2504         }
2505         info->key.tp_src = sport;
2506         info->key.tp_dst = dport;
2507         return 0;
2508 }
2509
2510 static const struct net_device_ops vxlan_netdev_ether_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_set_rx_mode        = vxlan_set_multicast_list,
2518         .ndo_change_mtu         = vxlan_change_mtu,
2519         .ndo_validate_addr      = eth_validate_addr,
2520         .ndo_set_mac_address    = eth_mac_addr,
2521         .ndo_fdb_add            = vxlan_fdb_add,
2522         .ndo_fdb_del            = vxlan_fdb_delete,
2523         .ndo_fdb_dump           = vxlan_fdb_dump,
2524         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2525 };
2526
2527 static const struct net_device_ops vxlan_netdev_raw_ops = {
2528         .ndo_init               = vxlan_init,
2529         .ndo_uninit             = vxlan_uninit,
2530         .ndo_open               = vxlan_open,
2531         .ndo_stop               = vxlan_stop,
2532         .ndo_start_xmit         = vxlan_xmit,
2533         .ndo_get_stats64        = ip_tunnel_get_stats64,
2534         .ndo_change_mtu         = vxlan_change_mtu,
2535         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2536 };
2537
2538 /* Info for udev, that this is a virtual tunnel endpoint */
2539 static struct device_type vxlan_type = {
2540         .name = "vxlan",
2541 };
2542
2543 /* Calls the ndo_udp_tunnel_add of the caller in order to
2544  * supply the listening VXLAN udp ports. Callers are expected
2545  * to implement the ndo_udp_tunnel_add.
2546  */
2547 static void vxlan_push_rx_ports(struct net_device *dev)
2548 {
2549         struct vxlan_sock *vs;
2550         struct net *net = dev_net(dev);
2551         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2552         unsigned int i;
2553
2554         spin_lock(&vn->sock_lock);
2555         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2556                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2557                         udp_tunnel_push_rx_port(dev, vs->sock,
2558                                                 (vs->flags & VXLAN_F_GPE) ?
2559                                                 UDP_TUNNEL_TYPE_VXLAN_GPE :
2560                                                 UDP_TUNNEL_TYPE_VXLAN);
2561         }
2562         spin_unlock(&vn->sock_lock);
2563 }
2564
2565 /* Initialize the device structure. */
2566 static void vxlan_setup(struct net_device *dev)
2567 {
2568         struct vxlan_dev *vxlan = netdev_priv(dev);
2569         unsigned int h;
2570
2571         eth_hw_addr_random(dev);
2572         ether_setup(dev);
2573
2574         dev->destructor = free_netdev;
2575         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2576
2577         dev->features   |= NETIF_F_LLTX;
2578         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2579         dev->features   |= NETIF_F_RXCSUM;
2580         dev->features   |= NETIF_F_GSO_SOFTWARE;
2581
2582         dev->vlan_features = dev->features;
2583         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2584         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2585         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2586         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2587         netif_keep_dst(dev);
2588         dev->priv_flags |= IFF_NO_QUEUE;
2589
2590         INIT_LIST_HEAD(&vxlan->next);
2591         spin_lock_init(&vxlan->hash_lock);
2592
2593         init_timer_deferrable(&vxlan->age_timer);
2594         vxlan->age_timer.function = vxlan_cleanup;
2595         vxlan->age_timer.data = (unsigned long) vxlan;
2596
2597         vxlan->cfg.dst_port = htons(vxlan_port);
2598
2599         vxlan->dev = dev;
2600
2601         gro_cells_init(&vxlan->gro_cells, dev);
2602
2603         for (h = 0; h < FDB_HASH_SIZE; ++h)
2604                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2605 }
2606
2607 static void vxlan_ether_setup(struct net_device *dev)
2608 {
2609         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2610         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2611         dev->netdev_ops = &vxlan_netdev_ether_ops;
2612 }
2613
2614 static void vxlan_raw_setup(struct net_device *dev)
2615 {
2616         dev->header_ops = NULL;
2617         dev->type = ARPHRD_NONE;
2618         dev->hard_header_len = 0;
2619         dev->addr_len = 0;
2620         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2621         dev->netdev_ops = &vxlan_netdev_raw_ops;
2622 }
2623
2624 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2625         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2626         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2627         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2628         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2629         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2630         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2631         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2632         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2633         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2634         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2635         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2636         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2637         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2638         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2639         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2640         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2641         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2642         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2643         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2644         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2645         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2646         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2647         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2648         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2649         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2650         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2651         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2652 };
2653
2654 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2655 {
2656         if (tb[IFLA_ADDRESS]) {
2657                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2658                         pr_debug("invalid link address (not ethernet)\n");
2659                         return -EINVAL;
2660                 }
2661
2662                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2663                         pr_debug("invalid all zero ethernet address\n");
2664                         return -EADDRNOTAVAIL;
2665                 }
2666         }
2667
2668         if (!data)
2669                 return -EINVAL;
2670
2671         if (data[IFLA_VXLAN_ID]) {
2672                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2673                 if (id >= VXLAN_N_VID)
2674                         return -ERANGE;
2675         }
2676
2677         if (data[IFLA_VXLAN_PORT_RANGE]) {
2678                 const struct ifla_vxlan_port_range *p
2679                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2680
2681                 if (ntohs(p->high) < ntohs(p->low)) {
2682                         pr_debug("port range %u .. %u not valid\n",
2683                                  ntohs(p->low), ntohs(p->high));
2684                         return -EINVAL;
2685                 }
2686         }
2687
2688         return 0;
2689 }
2690
2691 static void vxlan_get_drvinfo(struct net_device *netdev,
2692                               struct ethtool_drvinfo *drvinfo)
2693 {
2694         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2695         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2696 }
2697
2698 static const struct ethtool_ops vxlan_ethtool_ops = {
2699         .get_drvinfo    = vxlan_get_drvinfo,
2700         .get_link       = ethtool_op_get_link,
2701 };
2702
2703 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2704                                         __be16 port, u32 flags)
2705 {
2706         struct socket *sock;
2707         struct udp_port_cfg udp_conf;
2708         int err;
2709
2710         memset(&udp_conf, 0, sizeof(udp_conf));
2711
2712         if (ipv6) {
2713                 udp_conf.family = AF_INET6;
2714                 udp_conf.use_udp6_rx_checksums =
2715                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2716                 udp_conf.ipv6_v6only = 1;
2717         } else {
2718                 udp_conf.family = AF_INET;
2719         }
2720
2721         udp_conf.local_udp_port = port;
2722
2723         /* Open UDP socket */
2724         err = udp_sock_create(net, &udp_conf, &sock);
2725         if (err < 0)
2726                 return ERR_PTR(err);
2727
2728         return sock;
2729 }
2730
2731 /* Create new listen socket if needed */
2732 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2733                                               __be16 port, u32 flags)
2734 {
2735         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2736         struct vxlan_sock *vs;
2737         struct socket *sock;
2738         unsigned int h;
2739         struct udp_tunnel_sock_cfg tunnel_cfg;
2740
2741         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2742         if (!vs)
2743                 return ERR_PTR(-ENOMEM);
2744
2745         for (h = 0; h < VNI_HASH_SIZE; ++h)
2746                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2747
2748         sock = vxlan_create_sock(net, ipv6, port, flags);
2749         if (IS_ERR(sock)) {
2750                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2751                         PTR_ERR(sock));
2752                 kfree(vs);
2753                 return ERR_CAST(sock);
2754         }
2755
2756         vs->sock = sock;
2757         atomic_set(&vs->refcnt, 1);
2758         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2759
2760         spin_lock(&vn->sock_lock);
2761         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2762         udp_tunnel_notify_add_rx_port(sock,
2763                                       (vs->flags & VXLAN_F_GPE) ?
2764                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2765                                       UDP_TUNNEL_TYPE_VXLAN);
2766         spin_unlock(&vn->sock_lock);
2767
2768         /* Mark socket as an encapsulation socket. */
2769         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2770         tunnel_cfg.sk_user_data = vs;
2771         tunnel_cfg.encap_type = 1;
2772         tunnel_cfg.encap_rcv = vxlan_rcv;
2773         tunnel_cfg.encap_destroy = NULL;
2774         tunnel_cfg.gro_receive = vxlan_gro_receive;
2775         tunnel_cfg.gro_complete = vxlan_gro_complete;
2776
2777         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2778
2779         return vs;
2780 }
2781
2782 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2783 {
2784         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2785         struct vxlan_sock *vs = NULL;
2786         struct vxlan_dev_node *node;
2787
2788         if (!vxlan->cfg.no_share) {
2789                 spin_lock(&vn->sock_lock);
2790                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2791                                      vxlan->cfg.dst_port, vxlan->flags);
2792                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2793                         spin_unlock(&vn->sock_lock);
2794                         return -EBUSY;
2795                 }
2796                 spin_unlock(&vn->sock_lock);
2797         }
2798         if (!vs)
2799                 vs = vxlan_socket_create(vxlan->net, ipv6,
2800                                          vxlan->cfg.dst_port, vxlan->flags);
2801         if (IS_ERR(vs))
2802                 return PTR_ERR(vs);
2803 #if IS_ENABLED(CONFIG_IPV6)
2804         if (ipv6) {
2805                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2806                 node = &vxlan->hlist6;
2807         } else
2808 #endif
2809         {
2810                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2811                 node = &vxlan->hlist4;
2812         }
2813         vxlan_vs_add_dev(vs, vxlan, node);
2814         return 0;
2815 }
2816
2817 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2818 {
2819         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2820         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2821         int ret = 0;
2822
2823         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2824 #if IS_ENABLED(CONFIG_IPV6)
2825         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2826         if (ipv6 || metadata)
2827                 ret = __vxlan_sock_add(vxlan, true);
2828 #endif
2829         if (!ret && (!ipv6 || metadata))
2830                 ret = __vxlan_sock_add(vxlan, false);
2831         if (ret < 0)
2832                 vxlan_sock_release(vxlan);
2833         return ret;
2834 }
2835
2836 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2837                                struct vxlan_config *conf)
2838 {
2839         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2840         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2841         struct vxlan_rdst *dst = &vxlan->default_dst;
2842         unsigned short needed_headroom = ETH_HLEN;
2843         int err;
2844         bool use_ipv6 = false;
2845         __be16 default_port = vxlan->cfg.dst_port;
2846         struct net_device *lowerdev = NULL;
2847
2848         if (conf->flags & VXLAN_F_GPE) {
2849                 /* For now, allow GPE only together with COLLECT_METADATA.
2850                  * This can be relaxed later; in such case, the other side
2851                  * of the PtP link will have to be provided.
2852                  */
2853                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2854                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2855                         pr_info("unsupported combination of extensions\n");
2856                         return -EINVAL;
2857                 }
2858
2859                 vxlan_raw_setup(dev);
2860         } else {
2861                 vxlan_ether_setup(dev);
2862         }
2863
2864         vxlan->net = src_net;
2865
2866         dst->remote_vni = conf->vni;
2867
2868         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2869
2870         /* Unless IPv6 is explicitly requested, assume IPv4 */
2871         if (!dst->remote_ip.sa.sa_family)
2872                 dst->remote_ip.sa.sa_family = AF_INET;
2873
2874         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2875             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2876                 if (!IS_ENABLED(CONFIG_IPV6))
2877                         return -EPFNOSUPPORT;
2878                 use_ipv6 = true;
2879                 vxlan->flags |= VXLAN_F_IPV6;
2880         }
2881
2882         if (conf->label && !use_ipv6) {
2883                 pr_info("label only supported in use with IPv6\n");
2884                 return -EINVAL;
2885         }
2886
2887         if (conf->remote_ifindex) {
2888                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2889                 dst->remote_ifindex = conf->remote_ifindex;
2890
2891                 if (!lowerdev) {
2892                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2893                         return -ENODEV;
2894                 }
2895
2896 #if IS_ENABLED(CONFIG_IPV6)
2897                 if (use_ipv6) {
2898                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2899                         if (idev && idev->cnf.disable_ipv6) {
2900                                 pr_info("IPv6 is disabled via sysctl\n");
2901                                 return -EPERM;
2902                         }
2903                 }
2904 #endif
2905
2906                 if (!conf->mtu)
2907                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2908
2909                 needed_headroom = lowerdev->hard_header_len;
2910         } else if (vxlan_addr_multicast(&dst->remote_ip)) {
2911                 pr_info("multicast destination requires interface to be specified\n");
2912                 return -EINVAL;
2913         }
2914
2915         if (conf->mtu) {
2916                 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2917                 if (err)
2918                         return err;
2919         }
2920
2921         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2922                 needed_headroom += VXLAN6_HEADROOM;
2923         else
2924                 needed_headroom += VXLAN_HEADROOM;
2925         dev->needed_headroom = needed_headroom;
2926
2927         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2928         if (!vxlan->cfg.dst_port) {
2929                 if (conf->flags & VXLAN_F_GPE)
2930                         vxlan->cfg.dst_port = htons(4790); /* IANA VXLAN-GPE port */
2931                 else
2932                         vxlan->cfg.dst_port = default_port;
2933         }
2934         vxlan->flags |= conf->flags;
2935
2936         if (!vxlan->cfg.age_interval)
2937                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2938
2939         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2940                 if (tmp->cfg.vni == conf->vni &&
2941                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2942                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2943                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2944                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2945                     (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
2946                         pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
2947                         return -EEXIST;
2948                 }
2949         }
2950
2951         dev->ethtool_ops = &vxlan_ethtool_ops;
2952
2953         /* create an fdb entry for a valid default destination */
2954         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2955                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2956                                        &vxlan->default_dst.remote_ip,
2957                                        NUD_REACHABLE|NUD_PERMANENT,
2958                                        NLM_F_EXCL|NLM_F_CREATE,
2959                                        vxlan->cfg.dst_port,
2960                                        vxlan->default_dst.remote_vni,
2961                                        vxlan->default_dst.remote_ifindex,
2962                                        NTF_SELF);
2963                 if (err)
2964                         return err;
2965         }
2966
2967         err = register_netdevice(dev);
2968         if (err) {
2969                 vxlan_fdb_delete_default(vxlan);
2970                 return err;
2971         }
2972
2973         list_add(&vxlan->next, &vn->vxlan_list);
2974
2975         return 0;
2976 }
2977
2978 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2979                          struct nlattr *tb[], struct nlattr *data[])
2980 {
2981         struct vxlan_config conf;
2982
2983         memset(&conf, 0, sizeof(conf));
2984
2985         if (data[IFLA_VXLAN_ID])
2986                 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2987
2988         if (data[IFLA_VXLAN_GROUP]) {
2989                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2990         } else if (data[IFLA_VXLAN_GROUP6]) {
2991                 if (!IS_ENABLED(CONFIG_IPV6))
2992                         return -EPFNOSUPPORT;
2993
2994                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2995                 conf.remote_ip.sa.sa_family = AF_INET6;
2996         }
2997
2998         if (data[IFLA_VXLAN_LOCAL]) {
2999                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3000                 conf.saddr.sa.sa_family = AF_INET;
3001         } else if (data[IFLA_VXLAN_LOCAL6]) {
3002                 if (!IS_ENABLED(CONFIG_IPV6))
3003                         return -EPFNOSUPPORT;
3004
3005                 /* TODO: respect scope id */
3006                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3007                 conf.saddr.sa.sa_family = AF_INET6;
3008         }
3009
3010         if (data[IFLA_VXLAN_LINK])
3011                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3012
3013         if (data[IFLA_VXLAN_TOS])
3014                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3015
3016         if (data[IFLA_VXLAN_TTL])
3017                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3018
3019         if (data[IFLA_VXLAN_LABEL])
3020                 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3021                              IPV6_FLOWLABEL_MASK;
3022
3023         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3024                 conf.flags |= VXLAN_F_LEARN;
3025
3026         if (data[IFLA_VXLAN_AGEING])
3027                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3028
3029         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
3030                 conf.flags |= VXLAN_F_PROXY;
3031
3032         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
3033                 conf.flags |= VXLAN_F_RSC;
3034
3035         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3036                 conf.flags |= VXLAN_F_L2MISS;
3037
3038         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3039                 conf.flags |= VXLAN_F_L3MISS;
3040
3041         if (data[IFLA_VXLAN_LIMIT])
3042                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3043
3044         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
3045             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3046                 conf.flags |= VXLAN_F_COLLECT_METADATA;
3047
3048         if (data[IFLA_VXLAN_PORT_RANGE]) {
3049                 const struct ifla_vxlan_port_range *p
3050                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3051                 conf.port_min = ntohs(p->low);
3052                 conf.port_max = ntohs(p->high);
3053         }
3054
3055         if (data[IFLA_VXLAN_PORT])
3056                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3057
3058         if (data[IFLA_VXLAN_UDP_CSUM] &&
3059             !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3060                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3061
3062         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
3063             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3064                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3065
3066         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3067             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3068                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3069
3070         if (data[IFLA_VXLAN_REMCSUM_TX] &&
3071             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3072                 conf.flags |= VXLAN_F_REMCSUM_TX;
3073
3074         if (data[IFLA_VXLAN_REMCSUM_RX] &&
3075             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3076                 conf.flags |= VXLAN_F_REMCSUM_RX;
3077
3078         if (data[IFLA_VXLAN_GBP])
3079                 conf.flags |= VXLAN_F_GBP;
3080
3081         if (data[IFLA_VXLAN_GPE])
3082                 conf.flags |= VXLAN_F_GPE;
3083
3084         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3085                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3086
3087         if (tb[IFLA_MTU])
3088                 conf.mtu = nla_get_u32(tb[IFLA_MTU]);
3089
3090         return vxlan_dev_configure(src_net, dev, &conf);
3091 }
3092
3093 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3094 {
3095         struct vxlan_dev *vxlan = netdev_priv(dev);
3096
3097         gro_cells_destroy(&vxlan->gro_cells);
3098         list_del(&vxlan->next);
3099         unregister_netdevice_queue(dev, head);
3100 }
3101
3102 static size_t vxlan_get_size(const struct net_device *dev)
3103 {
3104
3105         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3106                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3107                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3108                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3109                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3110                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3111                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3112                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3113                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3114                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3115                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3116                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3117                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3118                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3119                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3120                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3121                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3122                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3123                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3124                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3125                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3126                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3127                 0;
3128 }
3129
3130 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3131 {
3132         const struct vxlan_dev *vxlan = netdev_priv(dev);
3133         const struct vxlan_rdst *dst = &vxlan->default_dst;
3134         struct ifla_vxlan_port_range ports = {
3135                 .low =  htons(vxlan->cfg.port_min),
3136                 .high = htons(vxlan->cfg.port_max),
3137         };
3138
3139         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3140                 goto nla_put_failure;
3141
3142         if (!vxlan_addr_any(&dst->remote_ip)) {
3143                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3144                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3145                                             dst->remote_ip.sin.sin_addr.s_addr))
3146                                 goto nla_put_failure;
3147 #if IS_ENABLED(CONFIG_IPV6)
3148                 } else {
3149                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3150                                              &dst->remote_ip.sin6.sin6_addr))
3151                                 goto nla_put_failure;
3152 #endif
3153                 }
3154         }
3155
3156         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3157                 goto nla_put_failure;
3158
3159         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3160                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3161                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3162                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3163                                 goto nla_put_failure;
3164 #if IS_ENABLED(CONFIG_IPV6)
3165                 } else {
3166                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3167                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3168                                 goto nla_put_failure;
3169 #endif
3170                 }
3171         }
3172
3173         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3174             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3175             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3176             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3177                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3178             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3179                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3180             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3181             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3182                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3183             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3184                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3185             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3186                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3187             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3188             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3189             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3190             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3191                         !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3192             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3193                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3194             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3195                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3196             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3197                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3198             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3199                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3200                 goto nla_put_failure;
3201
3202         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3203                 goto nla_put_failure;
3204
3205         if (vxlan->flags & VXLAN_F_GBP &&
3206             nla_put_flag(skb, IFLA_VXLAN_GBP))
3207                 goto nla_put_failure;
3208
3209         if (vxlan->flags & VXLAN_F_GPE &&
3210             nla_put_flag(skb, IFLA_VXLAN_GPE))
3211                 goto nla_put_failure;
3212
3213         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3214             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3215                 goto nla_put_failure;
3216
3217         return 0;
3218
3219 nla_put_failure:
3220         return -EMSGSIZE;
3221 }
3222
3223 static struct net *vxlan_get_link_net(const struct net_device *dev)
3224 {
3225         struct vxlan_dev *vxlan = netdev_priv(dev);
3226
3227         return vxlan->net;
3228 }
3229
3230 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3231         .kind           = "vxlan",
3232         .maxtype        = IFLA_VXLAN_MAX,
3233         .policy         = vxlan_policy,
3234         .priv_size      = sizeof(struct vxlan_dev),
3235         .setup          = vxlan_setup,
3236         .validate       = vxlan_validate,
3237         .newlink        = vxlan_newlink,
3238         .dellink        = vxlan_dellink,
3239         .get_size       = vxlan_get_size,
3240         .fill_info      = vxlan_fill_info,
3241         .get_link_net   = vxlan_get_link_net,
3242 };
3243
3244 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3245                                     u8 name_assign_type,
3246                                     struct vxlan_config *conf)
3247 {
3248         struct nlattr *tb[IFLA_MAX + 1];
3249         struct net_device *dev;
3250         int err;
3251
3252         memset(&tb, 0, sizeof(tb));
3253
3254         dev = rtnl_create_link(net, name, name_assign_type,
3255                                &vxlan_link_ops, tb);
3256         if (IS_ERR(dev))
3257                 return dev;
3258
3259         err = vxlan_dev_configure(net, dev, conf);
3260         if (err < 0) {
3261                 free_netdev(dev);
3262                 return ERR_PTR(err);
3263         }
3264
3265         err = rtnl_configure_link(dev, NULL);
3266         if (err < 0) {
3267                 LIST_HEAD(list_kill);
3268
3269                 vxlan_dellink(dev, &list_kill);
3270                 unregister_netdevice_many(&list_kill);
3271                 return ERR_PTR(err);
3272         }
3273
3274         return dev;
3275 }
3276 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3277
3278 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3279                                              struct net_device *dev)
3280 {
3281         struct vxlan_dev *vxlan, *next;
3282         LIST_HEAD(list_kill);
3283
3284         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3285                 struct vxlan_rdst *dst = &vxlan->default_dst;
3286
3287                 /* In case we created vxlan device with carrier
3288                  * and we loose the carrier due to module unload
3289                  * we also need to remove vxlan device. In other
3290                  * cases, it's not necessary and remote_ifindex
3291                  * is 0 here, so no matches.
3292                  */
3293                 if (dst->remote_ifindex == dev->ifindex)
3294                         vxlan_dellink(vxlan->dev, &list_kill);
3295         }
3296
3297         unregister_netdevice_many(&list_kill);
3298 }
3299
3300 static int vxlan_netdevice_event(struct notifier_block *unused,
3301                                  unsigned long event, void *ptr)
3302 {
3303         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3304         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3305
3306         if (event == NETDEV_UNREGISTER)
3307                 vxlan_handle_lowerdev_unregister(vn, dev);
3308         else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3309                 vxlan_push_rx_ports(dev);
3310
3311         return NOTIFY_DONE;
3312 }
3313
3314 static struct notifier_block vxlan_notifier_block __read_mostly = {
3315         .notifier_call = vxlan_netdevice_event,
3316 };
3317
3318 static __net_init int vxlan_init_net(struct net *net)
3319 {
3320         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3321         unsigned int h;
3322
3323         INIT_LIST_HEAD(&vn->vxlan_list);
3324         spin_lock_init(&vn->sock_lock);
3325
3326         for (h = 0; h < PORT_HASH_SIZE; ++h)
3327                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3328
3329         return 0;
3330 }
3331
3332 static void __net_exit vxlan_exit_net(struct net *net)
3333 {
3334         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3335         struct vxlan_dev *vxlan, *next;
3336         struct net_device *dev, *aux;
3337         LIST_HEAD(list);
3338
3339         rtnl_lock();
3340         for_each_netdev_safe(net, dev, aux)
3341                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3342                         unregister_netdevice_queue(dev, &list);
3343
3344         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3345                 /* If vxlan->dev is in the same netns, it has already been added
3346                  * to the list by the previous loop.
3347                  */
3348                 if (!net_eq(dev_net(vxlan->dev), net)) {
3349                         gro_cells_destroy(&vxlan->gro_cells);
3350                         unregister_netdevice_queue(vxlan->dev, &list);
3351                 }
3352         }
3353
3354         unregister_netdevice_many(&list);
3355         rtnl_unlock();
3356 }
3357
3358 static struct pernet_operations vxlan_net_ops = {
3359         .init = vxlan_init_net,
3360         .exit = vxlan_exit_net,
3361         .id   = &vxlan_net_id,
3362         .size = sizeof(struct vxlan_net),
3363 };
3364
3365 static int __init vxlan_init_module(void)
3366 {
3367         int rc;
3368
3369         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3370
3371         rc = register_pernet_subsys(&vxlan_net_ops);
3372         if (rc)
3373                 goto out1;
3374
3375         rc = register_netdevice_notifier(&vxlan_notifier_block);
3376         if (rc)
3377                 goto out2;
3378
3379         rc = rtnl_link_register(&vxlan_link_ops);
3380         if (rc)
3381                 goto out3;
3382
3383         return 0;
3384 out3:
3385         unregister_netdevice_notifier(&vxlan_notifier_block);
3386 out2:
3387         unregister_pernet_subsys(&vxlan_net_ops);
3388 out1:
3389         return rc;
3390 }
3391 late_initcall(vxlan_init_module);
3392
3393 static void __exit vxlan_cleanup_module(void)
3394 {
3395         rtnl_link_unregister(&vxlan_link_ops);
3396         unregister_netdevice_notifier(&vxlan_notifier_block);
3397         unregister_pernet_subsys(&vxlan_net_ops);
3398         /* rcu_barrier() is called by netns */
3399 }
3400 module_exit(vxlan_cleanup_module);
3401
3402 MODULE_LICENSE("GPL");
3403 MODULE_VERSION(VXLAN_VERSION);
3404 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3405 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3406 MODULE_ALIAS_RTNL_LINK("vxlan");