drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/types.h>
  15 #include <linux/module.h>
  16 #include <linux/errno.h>
  17 #include <linux/slab.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/rculist.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/in.h>
  22 #include <linux/ip.h>
  23 #include <linux/udp.h>
  24 #include <linux/igmp.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/hash.h>
  29 #include <linux/ethtool.h>
  30 #include <net/arp.h>
  31 #include <net/ndisc.h>
  32 #include <net/ip.h>
  33 #include <net/ip_tunnels.h>
  34 #include <net/icmp.h>
  35 #include <net/udp.h>
  36 #include <net/udp_tunnel.h>
  37 #include <net/rtnetlink.h>
  38 #include <net/route.h>
  39 #include <net/dsfield.h>
  40 #include <net/inet_ecn.h>
  41 #include <net/net_namespace.h>
  42 #include <net/netns/generic.h>
  43 #include <net/vxlan.h>
  44 #include <net/protocol.h>
  45
  46 #if IS_ENABLED(CONFIG_IPV6)
  47 #include <net/ipv6.h>
  48 #include <net/addrconf.h>
  49 #include <net/ip6_tunnel.h>
  50 #include <net/ip6_checksum.h>
  51 #endif
  52 #include <net/dst_metadata.h>
  53
  54 #define VXLAN_VERSION   "0.1"
  55
  56 #define PORT_HASH_BITS  8
  57 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  58 #define FDB_AGE_DEFAULT 300 /* 5 min */
  59 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  60
  61 /* UDP port for VXLAN traffic.
  62  * The IANA assigned port is 4789, but the Linux default is 8472
  63  * for compatibility with early adopters.
  64  */
  65 static unsigned short vxlan_port __read_mostly = 8472;
  66 module_param_named(udp_port, vxlan_port, ushort, 0444);
  67 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  68
  69 static bool log_ecn_error = true;
  70 module_param(log_ecn_error, bool, 0644);
  71 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  72
  73 static int vxlan_net_id;
  74 static struct rtnl_link_ops vxlan_link_ops;
  75
  76 static const u8 all_zeros_mac[ETH_ALEN + 2];
  77
  78 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  79
  80 /* per-network namespace private data for this module */
  81 struct vxlan_net {
  82         struct list_head  vxlan_list;
  83         struct hlist_head sock_list[PORT_HASH_SIZE];
  84         spinlock_t        sock_lock;
  85 };
  86
  87 /* Forwarding table entry */
  88 struct vxlan_fdb {
  89         struct hlist_node hlist;        /* linked list of entries */
  90         struct rcu_head   rcu;
  91         unsigned long     updated;      /* jiffies */
  92         unsigned long     used;
  93         struct list_head  remotes;
  94         u8                eth_addr[ETH_ALEN];
  95         u16               state;        /* see ndm_state */
  96         u8                flags;        /* see ndm_flags */
  97 };
  98
  99 /* salt for hash table */
 100 static u32 vxlan_salt __read_mostly;
 101 static struct workqueue_struct *vxlan_wq;
 102
 103 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
 104 {
 105         return vs->flags & VXLAN_F_COLLECT_METADATA ||
 106                ip_tunnel_collect_metadata();
 107 }
 108
 109 #if IS_ENABLED(CONFIG_IPV6)
 110 static inline
 111 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 112 {
 113         if (a->sa.sa_family != b->sa.sa_family)
 114                 return false;
 115         if (a->sa.sa_family == AF_INET6)
 116                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 117         else
 118                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 119 }
 120
 121 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 122 {
 123         if (ipa->sa.sa_family == AF_INET6)
 124                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 125         else
 126                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 127 }
 128
 129 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 130 {
 131         if (ipa->sa.sa_family == AF_INET6)
 132                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 133         else
 134                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 135 }
 136
 137 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 138 {
 139         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 140                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 141                 ip->sa.sa_family = AF_INET6;
 142                 return 0;
 143         } else if (nla_len(nla) >= sizeof(__be32)) {
 144                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 145                 ip->sa.sa_family = AF_INET;
 146                 return 0;
 147         } else {
 148                 return -EAFNOSUPPORT;
 149         }
 150 }
 151
 152 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 153                               const union vxlan_addr *ip)
 154 {
 155         if (ip->sa.sa_family == AF_INET6)
 156                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 157         else
 158                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 159 }
 160
 161 #else /* !CONFIG_IPV6 */
 162
 163 static inline
 164 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 165 {
 166         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 167 }
 168
 169 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 170 {
 171         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 172 }
 173
 174 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 175 {
 176         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 177 }
 178
 179 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 180 {
 181         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 182                 return -EAFNOSUPPORT;
 183         } else if (nla_len(nla) >= sizeof(__be32)) {
 184                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 185                 ip->sa.sa_family = AF_INET;
 186                 return 0;
 187         } else {
 188                 return -EAFNOSUPPORT;
 189         }
 190 }
 191
 192 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 193                               const union vxlan_addr *ip)
 194 {
 195         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 196 }
 197 #endif
 198
 199 /* Virtual Network hash table head */
 200 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 201 {
 202         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 203 }
 204
 205 /* Socket hash table head */
 206 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 207 {
 208         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 209
 210         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 211 }
 212
 213 /* First remote destination for a forwarding entry.
 214  * Guaranteed to be non-NULL because remotes are never deleted.
 215  */
 216 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 217 {
 218         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 219 }
 220
 221 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 222 {
 223         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 224 }
 225
 226 /* Find VXLAN socket based on network namespace, address family and UDP port
 227  * and enabled unshareable flags.
 228  */
 229 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 230                                           __be16 port, u32 flags)
 231 {
 232         struct vxlan_sock *vs;
 233
 234         flags &= VXLAN_F_RCV_FLAGS;
 235
 236         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 237                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 238                     vxlan_get_sk_family(vs) == family &&
 239                     vs->flags == flags)
 240                         return vs;
 241         }
 242         return NULL;
 243 }
 244
 245 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
 246 {
 247         struct vxlan_dev *vxlan;
 248
 249         hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
 250                 if (vxlan->default_dst.remote_vni == vni)
 251                         return vxlan;
 252         }
 253
 254         return NULL;
 255 }
 256
 257 /* Look up VNI in a per net namespace table */
 258 static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
 259                                         sa_family_t family, __be16 port,
 260                                         u32 flags)
 261 {
 262         struct vxlan_sock *vs;
 263
 264         vs = vxlan_find_sock(net, family, port, flags);
 265         if (!vs)
 266                 return NULL;
 267
 268         return vxlan_vs_find_vni(vs, vni);
 269 }
 270
 271 /* Fill in neighbour message in skbuff. */
 272 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 273                           const struct vxlan_fdb *fdb,
 274                           u32 portid, u32 seq, int type, unsigned int flags,
 275                           const struct vxlan_rdst *rdst)
 276 {
 277         unsigned long now = jiffies;
 278         struct nda_cacheinfo ci;
 279         struct nlmsghdr *nlh;
 280         struct ndmsg *ndm;
 281         bool send_ip, send_eth;
 282
 283         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 284         if (nlh == NULL)
 285                 return -EMSGSIZE;
 286
 287         ndm = nlmsg_data(nlh);
 288         memset(ndm, 0, sizeof(*ndm));
 289
 290         send_eth = send_ip = true;
 291
 292         if (type == RTM_GETNEIGH) {
 293                 ndm->ndm_family = AF_INET;
 294                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 295                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 296         } else
 297                 ndm->ndm_family = AF_BRIDGE;
 298         ndm->ndm_state = fdb->state;
 299         ndm->ndm_ifindex = vxlan->dev->ifindex;
 300         ndm->ndm_flags = fdb->flags;
 301         ndm->ndm_type = RTN_UNICAST;
 302
 303         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 304             nla_put_s32(skb, NDA_LINK_NETNSID,
 305                         peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
 306                 goto nla_put_failure;
 307
 308         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 309                 goto nla_put_failure;
 310
 311         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 312                 goto nla_put_failure;
 313
 314         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 315             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 316                 goto nla_put_failure;
 317         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 318             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 319                 goto nla_put_failure;
 320         if (rdst->remote_ifindex &&
 321             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 322                 goto nla_put_failure;
 323
 324         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 325         ci.ndm_confirmed = 0;
 326         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 327         ci.ndm_refcnt    = 0;
 328
 329         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 330                 goto nla_put_failure;
 331
 332         nlmsg_end(skb, nlh);
 333         return 0;
 334
 335 nla_put_failure:
 336         nlmsg_cancel(skb, nlh);
 337         return -EMSGSIZE;
 338 }
 339
 340 static inline size_t vxlan_nlmsg_size(void)
 341 {
 342         return NLMSG_ALIGN(sizeof(struct ndmsg))
 343                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 344                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 345                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 346                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 347                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 348                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 349                 + nla_total_size(sizeof(struct nda_cacheinfo));
 350 }
 351
 352 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 353                              struct vxlan_rdst *rd, int type)
 354 {
 355         struct net *net = dev_net(vxlan->dev);
 356         struct sk_buff *skb;
 357         int err = -ENOBUFS;
 358
 359         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 360         if (skb == NULL)
 361                 goto errout;
 362
 363         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 364         if (err < 0) {
 365                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 366                 WARN_ON(err == -EMSGSIZE);
 367                 kfree_skb(skb);
 368                 goto errout;
 369         }
 370
 371         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 372         return;
 373 errout:
 374         if (err < 0)
 375                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 376 }
 377
 378 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 379 {
 380         struct vxlan_dev *vxlan = netdev_priv(dev);
 381         struct vxlan_fdb f = {
 382                 .state = NUD_STALE,
 383         };
 384         struct vxlan_rdst remote = {
 385                 .remote_ip = *ipa, /* goes to NDA_DST */
 386                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 387         };
 388
 389         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 390 }
 391
 392 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 393 {
 394         struct vxlan_fdb f = {
 395                 .state = NUD_STALE,
 396         };
 397         struct vxlan_rdst remote = { };
 398
 399         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 400
 401         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 402 }
 403
 404 /* Hash Ethernet address */
 405 static u32 eth_hash(const unsigned char *addr)
 406 {
 407         u64 value = get_unaligned((u64 *)addr);
 408
 409         /* only want 6 bytes */
 410 #ifdef __BIG_ENDIAN
 411         value >>= 16;
 412 #else
 413         value <<= 16;
 414 #endif
 415         return hash_64(value, FDB_HASH_BITS);
 416 }
 417
 418 /* Hash chain to use given mac address */
 419 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 420                                                 const u8 *mac)
 421 {
 422         return &vxlan->fdb_head[eth_hash(mac)];
 423 }
 424
 425 /* Look up Ethernet address in forwarding table */
 426 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 427                                         const u8 *mac)
 428 {
 429         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 430         struct vxlan_fdb *f;
 431
 432         hlist_for_each_entry_rcu(f, head, hlist) {
 433                 if (ether_addr_equal(mac, f->eth_addr))
 434                         return f;
 435         }
 436
 437         return NULL;
 438 }
 439
 440 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 441                                         const u8 *mac)
 442 {
 443         struct vxlan_fdb *f;
 444
 445         f = __vxlan_find_mac(vxlan, mac);
 446         if (f)
 447                 f->used = jiffies;
 448
 449         return f;
 450 }
 451
 452 /* caller should hold vxlan->hash_lock */
 453 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 454                                               union vxlan_addr *ip, __be16 port,
 455                                               __be32 vni, __u32 ifindex)
 456 {
 457         struct vxlan_rdst *rd;
 458
 459         list_for_each_entry(rd, &f->remotes, list) {
 460                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 461                     rd->remote_port == port &&
 462                     rd->remote_vni == vni &&
 463                     rd->remote_ifindex == ifindex)
 464                         return rd;
 465         }
 466
 467         return NULL;
 468 }
 469
 470 /* Replace destination of unicast mac */
 471 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 472                              union vxlan_addr *ip, __be16 port, __be32 vni,
 473                              __u32 ifindex)
 474 {
 475         struct vxlan_rdst *rd;
 476
 477         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 478         if (rd)
 479                 return 0;
 480
 481         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 482         if (!rd)
 483                 return 0;
 484
 485         dst_cache_reset(&rd->dst_cache);
 486         rd->remote_ip = *ip;
 487         rd->remote_port = port;
 488         rd->remote_vni = vni;
 489         rd->remote_ifindex = ifindex;
 490         return 1;
 491 }
 492
 493 /* Add/update destinations for multicast */
 494 static int vxlan_fdb_append(struct vxlan_fdb *f,
 495                             union vxlan_addr *ip, __be16 port, __be32 vni,
 496                             __u32 ifindex, struct vxlan_rdst **rdp)
 497 {
 498         struct vxlan_rdst *rd;
 499
 500         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 501         if (rd)
 502                 return 0;
 503
 504         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 505         if (rd == NULL)
 506                 return -ENOBUFS;
 507
 508         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 509                 kfree(rd);
 510                 return -ENOBUFS;
 511         }
 512
 513         rd->remote_ip = *ip;
 514         rd->remote_port = port;
 515         rd->remote_vni = vni;
 516         rd->remote_ifindex = ifindex;
 517
 518         list_add_tail_rcu(&rd->list, &f->remotes);
 519
 520         *rdp = rd;
 521         return 1;
 522 }
 523
 524 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 525                                           unsigned int off,
 526                                           struct vxlanhdr *vh, size_t hdrlen,
 527                                           __be32 vni_field,
 528                                           struct gro_remcsum *grc,
 529                                           bool nopartial)
 530 {
 531         size_t start, offset;
 532
 533         if (skb->remcsum_offload)
 534                 return vh;
 535
 536         if (!NAPI_GRO_CB(skb)->csum_valid)
 537                 return NULL;
 538
 539         start = vxlan_rco_start(vni_field);
 540         offset = start + vxlan_rco_offset(vni_field);
 541
 542         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 543                                      start, offset, grc, nopartial);
 544
 545         skb->remcsum_offload = 1;
 546
 547         return vh;
 548 }
 549
 550 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
 551                                           struct sk_buff *skb,
 552                                           struct udp_offload *uoff)
 553 {
 554         struct sk_buff *p, **pp = NULL;
 555         struct vxlanhdr *vh, *vh2;
 556         unsigned int hlen, off_vx;
 557         int flush = 1;
 558         struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
 559                                              udp_offloads);
 560         __be32 flags;
 561         struct gro_remcsum grc;
 562
 563         skb_gro_remcsum_init(&grc);
 564
 565         off_vx = skb_gro_offset(skb);
 566         hlen = off_vx + sizeof(*vh);
 567         vh   = skb_gro_header_fast(skb, off_vx);
 568         if (skb_gro_header_hard(skb, hlen)) {
 569                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 570                 if (unlikely(!vh))
 571                         goto out;
 572         }
 573
 574         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 575
 576         flags = vh->vx_flags;
 577
 578         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 579                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 580                                        vh->vx_vni, &grc,
 581                                        !!(vs->flags &
 582                                           VXLAN_F_REMCSUM_NOPARTIAL));
 583
 584                 if (!vh)
 585                         goto out;
 586         }
 587
 588         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 589
 590         flush = 0;
 591
 592         for (p = *head; p; p = p->next) {
 593                 if (!NAPI_GRO_CB(p)->same_flow)
 594                         continue;
 595
 596                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 597                 if (vh->vx_flags != vh2->vx_flags ||
 598                     vh->vx_vni != vh2->vx_vni) {
 599                         NAPI_GRO_CB(p)->same_flow = 0;
 600                         continue;
 601                 }
 602         }
 603
 604         pp = eth_gro_receive(head, skb);
 605
 606 out:
 607         skb_gro_remcsum_cleanup(skb, &grc);
 608         NAPI_GRO_CB(skb)->flush |= flush;
 609
 610         return pp;
 611 }
 612
 613 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
 614                               struct udp_offload *uoff)
 615 {
 616         udp_tunnel_gro_complete(skb, nhoff);
 617
 618         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 619 }
 620
 621 /* Notify netdevs that UDP port started listening */
 622 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 623 {
 624         struct net_device *dev;
 625         struct sock *sk = vs->sock->sk;
 626         struct net *net = sock_net(sk);
 627         sa_family_t sa_family = vxlan_get_sk_family(vs);
 628         __be16 port = inet_sk(sk)->inet_sport;
 629         int err;
 630
 631         if (sa_family == AF_INET) {
 632                 err = udp_add_offload(net, &vs->udp_offloads);
 633                 if (err)
 634                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 635         }
 636
 637         rcu_read_lock();
 638         for_each_netdev_rcu(net, dev) {
 639                 if (dev->netdev_ops->ndo_add_vxlan_port)
 640                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 641                                                             port);
 642         }
 643         rcu_read_unlock();
 644 }
 645
 646 /* Notify netdevs that UDP port is no more listening */
 647 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 648 {
 649         struct net_device *dev;
 650         struct sock *sk = vs->sock->sk;
 651         struct net *net = sock_net(sk);
 652         sa_family_t sa_family = vxlan_get_sk_family(vs);
 653         __be16 port = inet_sk(sk)->inet_sport;
 654
 655         rcu_read_lock();
 656         for_each_netdev_rcu(net, dev) {
 657                 if (dev->netdev_ops->ndo_del_vxlan_port)
 658                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 659                                                             port);
 660         }
 661         rcu_read_unlock();
 662
 663         if (sa_family == AF_INET)
 664                 udp_del_offload(&vs->udp_offloads);
 665 }
 666
 667 /* Add new entry to forwarding table -- assumes lock held */
 668 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 669                             const u8 *mac, union vxlan_addr *ip,
 670                             __u16 state, __u16 flags,
 671                             __be16 port, __be32 vni, __u32 ifindex,
 672                             __u8 ndm_flags)
 673 {
 674         struct vxlan_rdst *rd = NULL;
 675         struct vxlan_fdb *f;
 676         int notify = 0;
 677
 678         f = __vxlan_find_mac(vxlan, mac);
 679         if (f) {
 680                 if (flags & NLM_F_EXCL) {
 681                         netdev_dbg(vxlan->dev,
 682                                    "lost race to create %pM\n", mac);
 683                         return -EEXIST;
 684                 }
 685                 if (f->state != state) {
 686                         f->state = state;
 687                         f->updated = jiffies;
 688                         notify = 1;
 689                 }
 690                 if (f->flags != ndm_flags) {
 691                         f->flags = ndm_flags;
 692                         f->updated = jiffies;
 693                         notify = 1;
 694                 }
 695                 if ((flags & NLM_F_REPLACE)) {
 696                         /* Only change unicasts */
 697                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 698                              is_zero_ether_addr(f->eth_addr))) {
 699                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 700                                                            ifindex);
 701                         } else
 702                                 return -EOPNOTSUPP;
 703                 }
 704                 if ((flags & NLM_F_APPEND) &&
 705                     (is_multicast_ether_addr(f->eth_addr) ||
 706                      is_zero_ether_addr(f->eth_addr))) {
 707                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 708                                                   &rd);
 709
 710                         if (rc < 0)
 711                                 return rc;
 712                         notify |= rc;
 713                 }
 714         } else {
 715                 if (!(flags & NLM_F_CREATE))
 716                         return -ENOENT;
 717
 718                 if (vxlan->cfg.addrmax &&
 719                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 720                         return -ENOSPC;
 721
 722                 /* Disallow replace to add a multicast entry */
 723                 if ((flags & NLM_F_REPLACE) &&
 724                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 725                         return -EOPNOTSUPP;
 726
 727                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 728                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 729                 if (!f)
 730                         return -ENOMEM;
 731
 732                 notify = 1;
 733                 f->state = state;
 734                 f->flags = ndm_flags;
 735                 f->updated = f->used = jiffies;
 736                 INIT_LIST_HEAD(&f->remotes);
 737                 memcpy(f->eth_addr, mac, ETH_ALEN);
 738
 739                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 740
 741                 ++vxlan->addrcnt;
 742                 hlist_add_head_rcu(&f->hlist,
 743                                    vxlan_fdb_head(vxlan, mac));
 744         }
 745
 746         if (notify) {
 747                 if (rd == NULL)
 748                         rd = first_remote_rtnl(f);
 749                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 750         }
 751
 752         return 0;
 753 }
 754
 755 static void vxlan_fdb_free(struct rcu_head *head)
 756 {
 757         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 758         struct vxlan_rdst *rd, *nd;
 759
 760         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 761                 dst_cache_destroy(&rd->dst_cache);
 762                 kfree(rd);
 763         }
 764         kfree(f);
 765 }
 766
 767 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 768 {
 769         netdev_dbg(vxlan->dev,
 770                     "delete %pM\n", f->eth_addr);
 771
 772         --vxlan->addrcnt;
 773         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 774
 775         hlist_del_rcu(&f->hlist);
 776         call_rcu(&f->rcu, vxlan_fdb_free);
 777 }
 778
 779 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 780                            union vxlan_addr *ip, __be16 *port, __be32 *vni,
 781                            u32 *ifindex)
 782 {
 783         struct net *net = dev_net(vxlan->dev);
 784         int err;
 785
 786         if (tb[NDA_DST]) {
 787                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 788                 if (err)
 789                         return err;
 790         } else {
 791                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 792                 if (remote->sa.sa_family == AF_INET) {
 793                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 794                         ip->sa.sa_family = AF_INET;
 795 #if IS_ENABLED(CONFIG_IPV6)
 796                 } else {
 797                         ip->sin6.sin6_addr = in6addr_any;
 798                         ip->sa.sa_family = AF_INET6;
 799 #endif
 800                 }
 801         }
 802
 803         if (tb[NDA_PORT]) {
 804                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 805                         return -EINVAL;
 806                 *port = nla_get_be16(tb[NDA_PORT]);
 807         } else {
 808                 *port = vxlan->cfg.dst_port;
 809         }
 810
 811         if (tb[NDA_VNI]) {
 812                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 813                         return -EINVAL;
 814                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 815         } else {
 816                 *vni = vxlan->default_dst.remote_vni;
 817         }
 818
 819         if (tb[NDA_IFINDEX]) {
 820                 struct net_device *tdev;
 821
 822                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 823                         return -EINVAL;
 824                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 825                 tdev = __dev_get_by_index(net, *ifindex);
 826                 if (!tdev)
 827                         return -EADDRNOTAVAIL;
 828         } else {
 829                 *ifindex = 0;
 830         }
 831
 832         return 0;
 833 }
 834
 835 /* Add static entry (via netlink) */
 836 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 837                          struct net_device *dev,
 838                          const unsigned char *addr, u16 vid, u16 flags)
 839 {
 840         struct vxlan_dev *vxlan = netdev_priv(dev);
 841         /* struct net *net = dev_net(vxlan->dev); */
 842         union vxlan_addr ip;
 843         __be16 port;
 844         __be32 vni;
 845         u32 ifindex;
 846         int err;
 847
 848         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 849                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 850                         ndm->ndm_state);
 851                 return -EINVAL;
 852         }
 853
 854         if (tb[NDA_DST] == NULL)
 855                 return -EINVAL;
 856
 857         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 858         if (err)
 859                 return err;
 860
 861         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 862                 return -EAFNOSUPPORT;
 863
 864         spin_lock_bh(&vxlan->hash_lock);
 865         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 866                                port, vni, ifindex, ndm->ndm_flags);
 867         spin_unlock_bh(&vxlan->hash_lock);
 868
 869         return err;
 870 }
 871
 872 /* Delete entry (via netlink) */
 873 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 874                             struct net_device *dev,
 875                             const unsigned char *addr, u16 vid)
 876 {
 877         struct vxlan_dev *vxlan = netdev_priv(dev);
 878         struct vxlan_fdb *f;
 879         struct vxlan_rdst *rd = NULL;
 880         union vxlan_addr ip;
 881         __be16 port;
 882         __be32 vni;
 883         u32 ifindex;
 884         int err;
 885
 886         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 887         if (err)
 888                 return err;
 889
 890         err = -ENOENT;
 891
 892         spin_lock_bh(&vxlan->hash_lock);
 893         f = vxlan_find_mac(vxlan, addr);
 894         if (!f)
 895                 goto out;
 896
 897         if (!vxlan_addr_any(&ip)) {
 898                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 899                 if (!rd)
 900                         goto out;
 901         }
 902
 903         err = 0;
 904
 905         /* remove a destination if it's not the only one on the list,
 906          * otherwise destroy the fdb entry
 907          */
 908         if (rd && !list_is_singular(&f->remotes)) {
 909                 list_del_rcu(&rd->list);
 910                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 911                 kfree_rcu(rd, rcu);
 912                 goto out;
 913         }
 914
 915         vxlan_fdb_destroy(vxlan, f);
 916
 917 out:
 918         spin_unlock_bh(&vxlan->hash_lock);
 919
 920         return err;
 921 }
 922
 923 /* Dump forwarding table */
 924 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 925                           struct net_device *dev,
 926                           struct net_device *filter_dev, int idx)
 927 {
 928         struct vxlan_dev *vxlan = netdev_priv(dev);
 929         unsigned int h;
 930
 931         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 932                 struct vxlan_fdb *f;
 933                 int err;
 934
 935                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 936                         struct vxlan_rdst *rd;
 937
 938                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 939                                 if (idx < cb->args[0])
 940                                         goto skip;
 941
 942                                 err = vxlan_fdb_info(skb, vxlan, f,
 943                                                      NETLINK_CB(cb->skb).portid,
 944                                                      cb->nlh->nlmsg_seq,
 945                                                      RTM_NEWNEIGH,
 946                                                      NLM_F_MULTI, rd);
 947                                 if (err < 0)
 948                                         goto out;
 949 skip:
 950                                 ++idx;
 951                         }
 952                 }
 953         }
 954 out:
 955         return idx;
 956 }
 957
 958 /* Watch incoming packets to learn mapping between Ethernet address
 959  * and Tunnel endpoint.
 960  * Return true if packet is bogus and should be dropped.
 961  */
 962 static bool vxlan_snoop(struct net_device *dev,
 963                         union vxlan_addr *src_ip, const u8 *src_mac)
 964 {
 965         struct vxlan_dev *vxlan = netdev_priv(dev);
 966         struct vxlan_fdb *f;
 967
 968         f = vxlan_find_mac(vxlan, src_mac);
 969         if (likely(f)) {
 970                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 971
 972                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 973                         return false;
 974
 975                 /* Don't migrate static entries, drop packets */
 976                 if (f->state & NUD_NOARP)
 977                         return true;
 978
 979                 if (net_ratelimit())
 980                         netdev_info(dev,
 981                                     "%pM migrated from %pIS to %pIS\n",
 982                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 983
 984                 rdst->remote_ip = *src_ip;
 985                 f->updated = jiffies;
 986                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 987         } else {
 988                 /* learned new entry */
 989                 spin_lock(&vxlan->hash_lock);
 990
 991                 /* close off race between vxlan_flush and incoming packets */
 992                 if (netif_running(dev))
 993                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 994                                          NUD_REACHABLE,
 995                                          NLM_F_EXCL|NLM_F_CREATE,
 996                                          vxlan->cfg.dst_port,
 997                                          vxlan->default_dst.remote_vni,
 998                                          0, NTF_SELF);
 999                 spin_unlock(&vxlan->hash_lock);
1000         }
1001
1002         return false;
1003 }
1004
1005 /* See if multicast group is already in use by other ID */
1006 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1007 {
1008         struct vxlan_dev *vxlan;
1009         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1010
1011         /* The vxlan_sock is only used by dev, leaving group has
1012          * no effect on other vxlan devices.
1013          */
1014         if (family == AF_INET && dev->vn4_sock &&
1015             atomic_read(&dev->vn4_sock->refcnt) == 1)
1016                 return false;
1017 #if IS_ENABLED(CONFIG_IPV6)
1018         if (family == AF_INET6 && dev->vn6_sock &&
1019             atomic_read(&dev->vn6_sock->refcnt) == 1)
1020                 return false;
1021 #endif
1022
1023         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1024                 if (!netif_running(vxlan->dev) || vxlan == dev)
1025                         continue;
1026
1027                 if (family == AF_INET && vxlan->vn4_sock != dev->vn4_sock)
1028                         continue;
1029 #if IS_ENABLED(CONFIG_IPV6)
1030                 if (family == AF_INET6 && vxlan->vn6_sock != dev->vn6_sock)
1031                         continue;
1032 #endif
1033
1034                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1035                                       &dev->default_dst.remote_ip))
1036                         continue;
1037
1038                 if (vxlan->default_dst.remote_ifindex !=
1039                     dev->default_dst.remote_ifindex)
1040                         continue;
1041
1042                 return true;
1043         }
1044
1045         return false;
1046 }
1047
1048 static void __vxlan_sock_release(struct vxlan_sock *vs)
1049 {
1050         struct vxlan_net *vn;
1051
1052         if (!vs)
1053                 return;
1054         if (!atomic_dec_and_test(&vs->refcnt))
1055                 return;
1056
1057         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1058         spin_lock(&vn->sock_lock);
1059         hlist_del_rcu(&vs->hlist);
1060         vxlan_notify_del_rx_port(vs);
1061         spin_unlock(&vn->sock_lock);
1062
1063         queue_work(vxlan_wq, &vs->del_work);
1064 }
1065
1066 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1067 {
1068         __vxlan_sock_release(vxlan->vn4_sock);
1069 #if IS_ENABLED(CONFIG_IPV6)
1070         __vxlan_sock_release(vxlan->vn6_sock);
1071 #endif
1072 }
1073
1074 /* Update multicast group membership when first VNI on
1075  * multicast address is brought up
1076  */
1077 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1078 {
1079         struct sock *sk;
1080         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1081         int ifindex = vxlan->default_dst.remote_ifindex;
1082         int ret = -EINVAL;
1083
1084         if (ip->sa.sa_family == AF_INET) {
1085                 struct ip_mreqn mreq = {
1086                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1087                         .imr_ifindex            = ifindex,
1088                 };
1089
1090                 sk = vxlan->vn4_sock->sock->sk;
1091                 lock_sock(sk);
1092                 ret = ip_mc_join_group(sk, &mreq);
1093                 release_sock(sk);
1094 #if IS_ENABLED(CONFIG_IPV6)
1095         } else {
1096                 sk = vxlan->vn6_sock->sock->sk;
1097                 lock_sock(sk);
1098                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1099                                                    &ip->sin6.sin6_addr);
1100                 release_sock(sk);
1101 #endif
1102         }
1103
1104         return ret;
1105 }
1106
1107 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1108 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1109 {
1110         struct sock *sk;
1111         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1112         int ifindex = vxlan->default_dst.remote_ifindex;
1113         int ret = -EINVAL;
1114
1115         if (ip->sa.sa_family == AF_INET) {
1116                 struct ip_mreqn mreq = {
1117                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1118                         .imr_ifindex            = ifindex,
1119                 };
1120
1121                 sk = vxlan->vn4_sock->sock->sk;
1122                 lock_sock(sk);
1123                 ret = ip_mc_leave_group(sk, &mreq);
1124                 release_sock(sk);
1125 #if IS_ENABLED(CONFIG_IPV6)
1126         } else {
1127                 sk = vxlan->vn6_sock->sock->sk;
1128                 lock_sock(sk);
1129                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1130                                                    &ip->sin6.sin6_addr);
1131                 release_sock(sk);
1132 #endif
1133         }
1134
1135         return ret;
1136 }
1137
1138 static bool vxlan_remcsum(struct sk_buff *skb, u32 vxflags, __be32 vni_field)
1139 {
1140         size_t start, offset, plen;
1141
1142         if (skb->remcsum_offload)
1143                 return true;
1144
1145         start = vxlan_rco_start(vni_field);
1146         offset = start + vxlan_rco_offset(vni_field);
1147
1148         plen = sizeof(struct vxlanhdr) + offset + sizeof(u16);
1149
1150         if (!pskb_may_pull(skb, plen))
1151                 return false;
1152
1153         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1154                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1155
1156         return true;
1157 }
1158
1159 static void vxlan_parse_gbp_hdr(struct sk_buff *skb, struct vxlan_metadata *md,
1160                                 struct metadata_dst *tun_dst)
1161 {
1162         struct vxlanhdr_gbp *gbp;
1163
1164         gbp = (struct vxlanhdr_gbp *)vxlan_hdr(skb);
1165         md->gbp = ntohs(gbp->policy_id);
1166
1167         if (tun_dst)
1168                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1169
1170         if (gbp->dont_learn)
1171                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1172
1173         if (gbp->policy_applied)
1174                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1175 }
1176
1177 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1178                       struct vxlan_metadata *md, __be32 vni,
1179                       struct metadata_dst *tun_dst)
1180 {
1181         struct iphdr *oip = NULL;
1182         struct ipv6hdr *oip6 = NULL;
1183         struct vxlan_dev *vxlan;
1184         struct pcpu_sw_netstats *stats;
1185         union vxlan_addr saddr;
1186         int err = 0;
1187
1188         /* For flow based devices, map all packets to VNI 0 */
1189         if (vs->flags & VXLAN_F_COLLECT_METADATA)
1190                 vni = 0;
1191
1192         /* Is this VNI defined? */
1193         vxlan = vxlan_vs_find_vni(vs, vni);
1194         if (!vxlan)
1195                 goto drop;
1196
1197         skb_reset_mac_header(skb);
1198         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1199         skb->protocol = eth_type_trans(skb, vxlan->dev);
1200         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1201
1202         /* Ignore packet loops (and multicast echo) */
1203         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1204                 goto drop;
1205
1206         /* Get data from the outer IP header */
1207         if (vxlan_get_sk_family(vs) == AF_INET) {
1208                 oip = ip_hdr(skb);
1209                 saddr.sin.sin_addr.s_addr = oip->saddr;
1210                 saddr.sa.sa_family = AF_INET;
1211 #if IS_ENABLED(CONFIG_IPV6)
1212         } else {
1213                 oip6 = ipv6_hdr(skb);
1214                 saddr.sin6.sin6_addr = oip6->saddr;
1215                 saddr.sa.sa_family = AF_INET6;
1216 #endif
1217         }
1218
1219         if (tun_dst) {
1220                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1221                 tun_dst = NULL;
1222         }
1223
1224         if ((vxlan->flags & VXLAN_F_LEARN) &&
1225             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1226                 goto drop;
1227
1228         skb_reset_network_header(skb);
1229         /* In flow-based mode, GBP is carried in dst_metadata */
1230         if (!(vs->flags & VXLAN_F_COLLECT_METADATA))
1231                 skb->mark = md->gbp;
1232
1233         if (oip6)
1234                 err = IP6_ECN_decapsulate(oip6, skb);
1235         if (oip)
1236                 err = IP_ECN_decapsulate(oip, skb);
1237
1238         if (unlikely(err)) {
1239                 if (log_ecn_error) {
1240                         if (oip6)
1241                                 net_info_ratelimited("non-ECT from %pI6\n",
1242                                                      &oip6->saddr);
1243                         if (oip)
1244                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1245                                                      &oip->saddr, oip->tos);
1246                 }
1247                 if (err > 1) {
1248                         ++vxlan->dev->stats.rx_frame_errors;
1249                         ++vxlan->dev->stats.rx_errors;
1250                         goto drop;
1251                 }
1252         }
1253
1254         stats = this_cpu_ptr(vxlan->dev->tstats);
1255         u64_stats_update_begin(&stats->syncp);
1256         stats->rx_packets++;
1257         stats->rx_bytes += skb->len;
1258         u64_stats_update_end(&stats->syncp);
1259
1260         gro_cells_receive(&vxlan->gro_cells, skb);
1261
1262         return;
1263 drop:
1264         if (tun_dst)
1265                 dst_release((struct dst_entry *)tun_dst);
1266
1267         /* Consume bad packet */
1268         kfree_skb(skb);
1269 }
1270
1271 /* Callback from net/ipv4/udp.c to receive packets */
1272 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1273 {
1274         struct metadata_dst *tun_dst = NULL;
1275         struct vxlan_sock *vs;
1276         __be32 flags, vni_field;
1277         struct vxlan_metadata _md;
1278         struct vxlan_metadata *md = &_md;
1279
1280         /* Need Vxlan and inner Ethernet header to be present */
1281         if (!pskb_may_pull(skb, VXLAN_HLEN))
1282                 goto error;
1283
1284         flags = vxlan_hdr(skb)->vx_flags;
1285         vni_field = vxlan_hdr(skb)->vx_vni;
1286
1287         if (flags & VXLAN_HF_VNI) {
1288                 flags &= ~VXLAN_HF_VNI;
1289         } else {
1290                 /* VNI flag always required to be set */
1291                 goto bad_flags;
1292         }
1293
1294         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1295                 goto drop;
1296
1297         vs = rcu_dereference_sk_user_data(sk);
1298         if (!vs)
1299                 goto drop;
1300
1301         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1302                 if (!vxlan_remcsum(skb, vs->flags, vni_field))
1303                         goto drop;
1304
1305                 flags &= ~VXLAN_HF_RCO;
1306                 vni_field &= VXLAN_VNI_MASK;
1307         }
1308
1309         if (vxlan_collect_metadata(vs)) {
1310                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1311                                          vxlan_vni(vni_field), sizeof(*md));
1312
1313                 if (!tun_dst)
1314                         goto drop;
1315
1316                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1317         } else {
1318                 memset(md, 0, sizeof(*md));
1319         }
1320
1321         /* For backwards compatibility, only allow reserved fields to be
1322          * used by VXLAN extensions if explicitly requested.
1323          */
1324         if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1325                 vxlan_parse_gbp_hdr(skb, md, tun_dst);
1326                 flags &= ~VXLAN_GBP_USED_BITS;
1327         }
1328
1329         if (flags || vni_field & ~VXLAN_VNI_MASK) {
1330                 /* If there are any unprocessed flags remaining treat
1331                  * this as a malformed packet. This behavior diverges from
1332                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1333                  * in reserved fields are to be ignored. The approach here
1334                  * maintains compatibility with previous stack code, and also
1335                  * is more robust and provides a little more security in
1336                  * adding extensions to VXLAN.
1337                  */
1338
1339                 goto bad_flags;
1340         }
1341
1342         vxlan_rcv(vs, skb, md, vxlan_vni(vni_field), tun_dst);
1343         return 0;
1344
1345 drop:
1346         /* Consume bad packet */
1347         kfree_skb(skb);
1348         return 0;
1349
1350 bad_flags:
1351         netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1352                    ntohl(vxlan_hdr(skb)->vx_flags),
1353                    ntohl(vxlan_hdr(skb)->vx_vni));
1354
1355 error:
1356         if (tun_dst)
1357                 dst_release((struct dst_entry *)tun_dst);
1358
1359         /* Return non vxlan pkt */
1360         return 1;
1361 }
1362
1363 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1364 {
1365         struct vxlan_dev *vxlan = netdev_priv(dev);
1366         struct arphdr *parp;
1367         u8 *arpptr, *sha;
1368         __be32 sip, tip;
1369         struct neighbour *n;
1370
1371         if (dev->flags & IFF_NOARP)
1372                 goto out;
1373
1374         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1375                 dev->stats.tx_dropped++;
1376                 goto out;
1377         }
1378         parp = arp_hdr(skb);
1379
1380         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1381              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1382             parp->ar_pro != htons(ETH_P_IP) ||
1383             parp->ar_op != htons(ARPOP_REQUEST) ||
1384             parp->ar_hln != dev->addr_len ||
1385             parp->ar_pln != 4)
1386                 goto out;
1387         arpptr = (u8 *)parp + sizeof(struct arphdr);
1388         sha = arpptr;
1389         arpptr += dev->addr_len;        /* sha */
1390         memcpy(&sip, arpptr, sizeof(sip));
1391         arpptr += sizeof(sip);
1392         arpptr += dev->addr_len;        /* tha */
1393         memcpy(&tip, arpptr, sizeof(tip));
1394
1395         if (ipv4_is_loopback(tip) ||
1396             ipv4_is_multicast(tip))
1397                 goto out;
1398
1399         n = neigh_lookup(&arp_tbl, &tip, dev);
1400
1401         if (n) {
1402                 struct vxlan_fdb *f;
1403                 struct sk_buff  *reply;
1404
1405                 if (!(n->nud_state & NUD_CONNECTED)) {
1406                         neigh_release(n);
1407                         goto out;
1408                 }
1409
1410                 f = vxlan_find_mac(vxlan, n->ha);
1411                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1412                         /* bridge-local neighbor */
1413                         neigh_release(n);
1414                         goto out;
1415                 }
1416
1417                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1418                                 n->ha, sha);
1419
1420                 neigh_release(n);
1421
1422                 if (reply == NULL)
1423                         goto out;
1424
1425                 skb_reset_mac_header(reply);
1426                 __skb_pull(reply, skb_network_offset(reply));
1427                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1428                 reply->pkt_type = PACKET_HOST;
1429
1430                 if (netif_rx_ni(reply) == NET_RX_DROP)
1431                         dev->stats.rx_dropped++;
1432         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1433                 union vxlan_addr ipa = {
1434                         .sin.sin_addr.s_addr = tip,
1435                         .sin.sin_family = AF_INET,
1436                 };
1437
1438                 vxlan_ip_miss(dev, &ipa);
1439         }
1440 out:
1441         consume_skb(skb);
1442         return NETDEV_TX_OK;
1443 }
1444
1445 #if IS_ENABLED(CONFIG_IPV6)
1446 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1447         struct neighbour *n, bool isrouter)
1448 {
1449         struct net_device *dev = request->dev;
1450         struct sk_buff *reply;
1451         struct nd_msg *ns, *na;
1452         struct ipv6hdr *pip6;
1453         u8 *daddr;
1454         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1455         int ns_olen;
1456         int i, len;
1457
1458         if (dev == NULL)
1459                 return NULL;
1460
1461         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1462                 sizeof(*na) + na_olen + dev->needed_tailroom;
1463         reply = alloc_skb(len, GFP_ATOMIC);
1464         if (reply == NULL)
1465                 return NULL;
1466
1467         reply->protocol = htons(ETH_P_IPV6);
1468         reply->dev = dev;
1469         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1470         skb_push(reply, sizeof(struct ethhdr));
1471         skb_set_mac_header(reply, 0);
1472
1473         ns = (struct nd_msg *)skb_transport_header(request);
1474
1475         daddr = eth_hdr(request)->h_source;
1476         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1477         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1478                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1479                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1480                         break;
1481                 }
1482         }
1483
1484         /* Ethernet header */
1485         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1486         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1487         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1488         reply->protocol = htons(ETH_P_IPV6);
1489
1490         skb_pull(reply, sizeof(struct ethhdr));
1491         skb_set_network_header(reply, 0);
1492         skb_put(reply, sizeof(struct ipv6hdr));
1493
1494         /* IPv6 header */
1495
1496         pip6 = ipv6_hdr(reply);
1497         memset(pip6, 0, sizeof(struct ipv6hdr));
1498         pip6->version = 6;
1499         pip6->priority = ipv6_hdr(request)->priority;
1500         pip6->nexthdr = IPPROTO_ICMPV6;
1501         pip6->hop_limit = 255;
1502         pip6->daddr = ipv6_hdr(request)->saddr;
1503         pip6->saddr = *(struct in6_addr *)n->primary_key;
1504
1505         skb_pull(reply, sizeof(struct ipv6hdr));
1506         skb_set_transport_header(reply, 0);
1507
1508         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1509
1510         /* Neighbor Advertisement */
1511         memset(na, 0, sizeof(*na)+na_olen);
1512         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1513         na->icmph.icmp6_router = isrouter;
1514         na->icmph.icmp6_override = 1;
1515         na->icmph.icmp6_solicited = 1;
1516         na->target = ns->target;
1517         ether_addr_copy(&na->opt[2], n->ha);
1518         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1519         na->opt[1] = na_olen >> 3;
1520
1521         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1522                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1523                 csum_partial(na, sizeof(*na)+na_olen, 0));
1524
1525         pip6->payload_len = htons(sizeof(*na)+na_olen);
1526
1527         skb_push(reply, sizeof(struct ipv6hdr));
1528
1529         reply->ip_summed = CHECKSUM_UNNECESSARY;
1530
1531         return reply;
1532 }
1533
1534 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1535 {
1536         struct vxlan_dev *vxlan = netdev_priv(dev);
1537         struct nd_msg *msg;
1538         const struct ipv6hdr *iphdr;
1539         const struct in6_addr *saddr, *daddr;
1540         struct neighbour *n;
1541         struct inet6_dev *in6_dev;
1542
1543         in6_dev = __in6_dev_get(dev);
1544         if (!in6_dev)
1545                 goto out;
1546
1547         iphdr = ipv6_hdr(skb);
1548         saddr = &iphdr->saddr;
1549         daddr = &iphdr->daddr;
1550
1551         msg = (struct nd_msg *)skb_transport_header(skb);
1552         if (msg->icmph.icmp6_code != 0 ||
1553             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1554                 goto out;
1555
1556         if (ipv6_addr_loopback(daddr) ||
1557             ipv6_addr_is_multicast(&msg->target))
1558                 goto out;
1559
1560         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1561
1562         if (n) {
1563                 struct vxlan_fdb *f;
1564                 struct sk_buff *reply;
1565
1566                 if (!(n->nud_state & NUD_CONNECTED)) {
1567                         neigh_release(n);
1568                         goto out;
1569                 }
1570
1571                 f = vxlan_find_mac(vxlan, n->ha);
1572                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1573                         /* bridge-local neighbor */
1574                         neigh_release(n);
1575                         goto out;
1576                 }
1577
1578                 reply = vxlan_na_create(skb, n,
1579                                         !!(f ? f->flags & NTF_ROUTER : 0));
1580
1581                 neigh_release(n);
1582
1583                 if (reply == NULL)
1584                         goto out;
1585
1586                 if (netif_rx_ni(reply) == NET_RX_DROP)
1587                         dev->stats.rx_dropped++;
1588
1589         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1590                 union vxlan_addr ipa = {
1591                         .sin6.sin6_addr = msg->target,
1592                         .sin6.sin6_family = AF_INET6,
1593                 };
1594
1595                 vxlan_ip_miss(dev, &ipa);
1596         }
1597
1598 out:
1599         consume_skb(skb);
1600         return NETDEV_TX_OK;
1601 }
1602 #endif
1603
1604 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1605 {
1606         struct vxlan_dev *vxlan = netdev_priv(dev);
1607         struct neighbour *n;
1608
1609         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1610                 return false;
1611
1612         n = NULL;
1613         switch (ntohs(eth_hdr(skb)->h_proto)) {
1614         case ETH_P_IP:
1615         {
1616                 struct iphdr *pip;
1617
1618                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1619                         return false;
1620                 pip = ip_hdr(skb);
1621                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1622                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1623                         union vxlan_addr ipa = {
1624                                 .sin.sin_addr.s_addr = pip->daddr,
1625                                 .sin.sin_family = AF_INET,
1626                         };
1627
1628                         vxlan_ip_miss(dev, &ipa);
1629                         return false;
1630                 }
1631
1632                 break;
1633         }
1634 #if IS_ENABLED(CONFIG_IPV6)
1635         case ETH_P_IPV6:
1636         {
1637                 struct ipv6hdr *pip6;
1638
1639                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1640                         return false;
1641                 pip6 = ipv6_hdr(skb);
1642                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1643                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1644                         union vxlan_addr ipa = {
1645                                 .sin6.sin6_addr = pip6->daddr,
1646                                 .sin6.sin6_family = AF_INET6,
1647                         };
1648
1649                         vxlan_ip_miss(dev, &ipa);
1650                         return false;
1651                 }
1652
1653                 break;
1654         }
1655 #endif
1656         default:
1657                 return false;
1658         }
1659
1660         if (n) {
1661                 bool diff;
1662
1663                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1664                 if (diff) {
1665                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1666                                 dev->addr_len);
1667                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1668                 }
1669                 neigh_release(n);
1670                 return diff;
1671         }
1672
1673         return false;
1674 }
1675
1676 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1677                                 struct vxlan_metadata *md)
1678 {
1679         struct vxlanhdr_gbp *gbp;
1680
1681         if (!md->gbp)
1682                 return;
1683
1684         gbp = (struct vxlanhdr_gbp *)vxh;
1685         vxh->vx_flags |= VXLAN_HF_GBP;
1686
1687         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1688                 gbp->dont_learn = 1;
1689
1690         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1691                 gbp->policy_applied = 1;
1692
1693         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1694 }
1695
1696 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1697                            int iphdr_len, __be32 vni,
1698                            struct vxlan_metadata *md, u32 vxflags,
1699                            bool udp_sum)
1700 {
1701         struct vxlanhdr *vxh;
1702         int min_headroom;
1703         int err;
1704         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1705
1706         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1707             skb->ip_summed == CHECKSUM_PARTIAL) {
1708                 int csum_start = skb_checksum_start_offset(skb);
1709
1710                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1711                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1712                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1713                      skb->csum_offset == offsetof(struct tcphdr, check)))
1714                         type |= SKB_GSO_TUNNEL_REMCSUM;
1715         }
1716
1717         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1718                         + VXLAN_HLEN + iphdr_len
1719                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1720
1721         /* Need space for new headers (invalidates iph ptr) */
1722         err = skb_cow_head(skb, min_headroom);
1723         if (unlikely(err)) {
1724                 kfree_skb(skb);
1725                 return err;
1726         }
1727
1728         skb = vlan_hwaccel_push_inside(skb);
1729         if (WARN_ON(!skb))
1730                 return -ENOMEM;
1731
1732         skb = iptunnel_handle_offloads(skb, type);
1733         if (IS_ERR(skb))
1734                 return PTR_ERR(skb);
1735
1736         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1737         vxh->vx_flags = VXLAN_HF_VNI;
1738         vxh->vx_vni = vxlan_vni_field(vni);
1739
1740         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1741                 unsigned int start;
1742
1743                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1744                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1745                 vxh->vx_flags |= VXLAN_HF_RCO;
1746
1747                 if (!skb_is_gso(skb)) {
1748                         skb->ip_summed = CHECKSUM_NONE;
1749                         skb->encapsulation = 0;
1750                 }
1751         }
1752
1753         if (vxflags & VXLAN_F_GBP)
1754                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1755
1756         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1757         return 0;
1758 }
1759
1760 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1761                                       struct sk_buff *skb, int oif, u8 tos,
1762                                       __be32 daddr, __be32 *saddr,
1763                                       struct dst_cache *dst_cache,
1764                                       struct ip_tunnel_info *info)
1765 {
1766         struct rtable *rt = NULL;
1767         bool use_cache = false;
1768         struct flowi4 fl4;
1769
1770         /* when the ip_tunnel_info is availble, the tos used for lookup is
1771          * packet independent, so we can use the cache
1772          */
1773         if (!skb->mark && (!tos || info)) {
1774                 use_cache = true;
1775                 rt = dst_cache_get_ip4(dst_cache, saddr);
1776                 if (rt)
1777                         return rt;
1778         }
1779
1780         memset(&fl4, 0, sizeof(fl4));
1781         fl4.flowi4_oif = oif;
1782         fl4.flowi4_tos = RT_TOS(tos);
1783         fl4.flowi4_mark = skb->mark;
1784         fl4.flowi4_proto = IPPROTO_UDP;
1785         fl4.daddr = daddr;
1786         fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
1787
1788         rt = ip_route_output_key(vxlan->net, &fl4);
1789         if (!IS_ERR(rt)) {
1790                 *saddr = fl4.saddr;
1791                 if (use_cache)
1792                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1793         }
1794         return rt;
1795 }
1796
1797 #if IS_ENABLED(CONFIG_IPV6)
1798 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1799                                           struct sk_buff *skb, int oif,
1800                                           const struct in6_addr *daddr,
1801                                           struct in6_addr *saddr,
1802                                           struct dst_cache *dst_cache)
1803 {
1804         struct dst_entry *ndst;
1805         struct flowi6 fl6;
1806         int err;
1807
1808         if (!skb->mark) {
1809                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1810                 if (ndst)
1811                         return ndst;
1812         }
1813
1814         memset(&fl6, 0, sizeof(fl6));
1815         fl6.flowi6_oif = oif;
1816         fl6.daddr = *daddr;
1817         fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
1818         fl6.flowi6_mark = skb->mark;
1819         fl6.flowi6_proto = IPPROTO_UDP;
1820
1821         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1822                                          vxlan->vn6_sock->sock->sk,
1823                                          &ndst, &fl6);
1824         if (err < 0)
1825                 return ERR_PTR(err);
1826
1827         *saddr = fl6.saddr;
1828         if (!skb->mark)
1829                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1830         return ndst;
1831 }
1832 #endif
1833
1834 /* Bypass encapsulation if the destination is local */
1835 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1836                                struct vxlan_dev *dst_vxlan)
1837 {
1838         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1839         union vxlan_addr loopback;
1840         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1841         struct net_device *dev = skb->dev;
1842         int len = skb->len;
1843
1844         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1845         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1846         skb->pkt_type = PACKET_HOST;
1847         skb->encapsulation = 0;
1848         skb->dev = dst_vxlan->dev;
1849         __skb_pull(skb, skb_network_offset(skb));
1850
1851         if (remote_ip->sa.sa_family == AF_INET) {
1852                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1853                 loopback.sa.sa_family =  AF_INET;
1854 #if IS_ENABLED(CONFIG_IPV6)
1855         } else {
1856                 loopback.sin6.sin6_addr = in6addr_loopback;
1857                 loopback.sa.sa_family =  AF_INET6;
1858 #endif
1859         }
1860
1861         if (dst_vxlan->flags & VXLAN_F_LEARN)
1862                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1863
1864         u64_stats_update_begin(&tx_stats->syncp);
1865         tx_stats->tx_packets++;
1866         tx_stats->tx_bytes += len;
1867         u64_stats_update_end(&tx_stats->syncp);
1868
1869         if (netif_rx(skb) == NET_RX_SUCCESS) {
1870                 u64_stats_update_begin(&rx_stats->syncp);
1871                 rx_stats->rx_packets++;
1872                 rx_stats->rx_bytes += len;
1873                 u64_stats_update_end(&rx_stats->syncp);
1874         } else {
1875                 dev->stats.rx_dropped++;
1876         }
1877 }
1878
1879 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1880                            struct vxlan_rdst *rdst, bool did_rsc)
1881 {
1882         struct dst_cache *dst_cache;
1883         struct ip_tunnel_info *info;
1884         struct vxlan_dev *vxlan = netdev_priv(dev);
1885         struct sock *sk;
1886         struct rtable *rt = NULL;
1887         const struct iphdr *old_iph;
1888         union vxlan_addr *dst;
1889         union vxlan_addr remote_ip;
1890         struct vxlan_metadata _md;
1891         struct vxlan_metadata *md = &_md;
1892         __be16 src_port = 0, dst_port;
1893         __be32 vni;
1894         __be16 df = 0;
1895         __u8 tos, ttl;
1896         int err;
1897         u32 flags = vxlan->flags;
1898         bool udp_sum = false;
1899         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1900
1901         info = skb_tunnel_info(skb);
1902
1903         if (rdst) {
1904                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1905                 vni = rdst->remote_vni;
1906                 dst = &rdst->remote_ip;
1907                 dst_cache = &rdst->dst_cache;
1908         } else {
1909                 if (!info) {
1910                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1911                                   dev->name);
1912                         goto drop;
1913                 }
1914                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1915                 vni = vxlan_tun_id_to_vni(info->key.tun_id);
1916                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1917                 if (remote_ip.sa.sa_family == AF_INET)
1918                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1919                 else
1920                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1921                 dst = &remote_ip;
1922                 dst_cache = &info->dst_cache;
1923         }
1924
1925         if (vxlan_addr_any(dst)) {
1926                 if (did_rsc) {
1927                         /* short-circuited back to local bridge */
1928                         vxlan_encap_bypass(skb, vxlan, vxlan);
1929                         return;
1930                 }
1931                 goto drop;
1932         }
1933
1934         old_iph = ip_hdr(skb);
1935
1936         ttl = vxlan->cfg.ttl;
1937         if (!ttl && vxlan_addr_multicast(dst))
1938                 ttl = 1;
1939
1940         tos = vxlan->cfg.tos;
1941         if (tos == 1)
1942                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1943
1944         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
1945                                      vxlan->cfg.port_max, true);
1946
1947         if (info) {
1948                 ttl = info->key.ttl;
1949                 tos = info->key.tos;
1950                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
1951
1952                 if (info->options_len)
1953                         md = ip_tunnel_info_opts(info);
1954         } else {
1955                 md->gbp = skb->mark;
1956         }
1957
1958         if (dst->sa.sa_family == AF_INET) {
1959                 __be32 saddr;
1960
1961                 if (!vxlan->vn4_sock)
1962                         goto drop;
1963                 sk = vxlan->vn4_sock->sock->sk;
1964
1965                 if (info) {
1966                         if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
1967                                 df = htons(IP_DF);
1968                 } else {
1969                         udp_sum = !!(flags & VXLAN_F_UDP_CSUM);
1970                 }
1971
1972                 rt = vxlan_get_route(vxlan, skb,
1973                                      rdst ? rdst->remote_ifindex : 0, tos,
1974                                      dst->sin.sin_addr.s_addr, &saddr,
1975                                      dst_cache, info);
1976                 if (IS_ERR(rt)) {
1977                         netdev_dbg(dev, "no route to %pI4\n",
1978                                    &dst->sin.sin_addr.s_addr);
1979                         dev->stats.tx_carrier_errors++;
1980                         goto tx_error;
1981                 }
1982
1983                 if (rt->dst.dev == dev) {
1984                         netdev_dbg(dev, "circular route to %pI4\n",
1985                                    &dst->sin.sin_addr.s_addr);
1986                         dev->stats.collisions++;
1987                         goto rt_tx_error;
1988                 }
1989
1990                 /* Bypass encapsulation if the destination is local */
1991                 if (rt->rt_flags & RTCF_LOCAL &&
1992                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1993                         struct vxlan_dev *dst_vxlan;
1994
1995                         ip_rt_put(rt);
1996                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1997                                                    dst->sa.sa_family, dst_port,
1998                                                    vxlan->flags);
1999                         if (!dst_vxlan)
2000                                 goto tx_error;
2001                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2002                         return;
2003                 }
2004
2005                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2006                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2007                 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2008                                       vni, md, flags, udp_sum);
2009                 if (err < 0)
2010                         goto xmit_tx_error;
2011
2012                 udp_tunnel_xmit_skb(rt, sk, skb, saddr,
2013                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2014                                     src_port, dst_port, xnet, !udp_sum);
2015 #if IS_ENABLED(CONFIG_IPV6)
2016         } else {
2017                 struct dst_entry *ndst;
2018                 struct in6_addr saddr;
2019                 u32 rt6i_flags;
2020
2021                 if (!vxlan->vn6_sock)
2022                         goto drop;
2023                 sk = vxlan->vn6_sock->sock->sk;
2024
2025                 ndst = vxlan6_get_route(vxlan, skb,
2026                                         rdst ? rdst->remote_ifindex : 0,
2027                                         &dst->sin6.sin6_addr, &saddr,
2028                                         dst_cache);
2029                 if (IS_ERR(ndst)) {
2030                         netdev_dbg(dev, "no route to %pI6\n",
2031                                    &dst->sin6.sin6_addr);
2032                         dev->stats.tx_carrier_errors++;
2033                         goto tx_error;
2034                 }
2035
2036                 if (ndst->dev == dev) {
2037                         netdev_dbg(dev, "circular route to %pI6\n",
2038                                    &dst->sin6.sin6_addr);
2039                         dst_release(ndst);
2040                         dev->stats.collisions++;
2041                         goto tx_error;
2042                 }
2043
2044                 /* Bypass encapsulation if the destination is local */
2045                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2046                 if (rt6i_flags & RTF_LOCAL &&
2047                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2048                         struct vxlan_dev *dst_vxlan;
2049
2050                         dst_release(ndst);
2051                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2052                                                    dst->sa.sa_family, dst_port,
2053                                                    vxlan->flags);
2054                         if (!dst_vxlan)
2055                                 goto tx_error;
2056                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2057                         return;
2058                 }
2059
2060                 if (!info)
2061                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2062
2063                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2064                 skb_scrub_packet(skb, xnet);
2065                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2066                                       vni, md, flags, udp_sum);
2067                 if (err < 0) {
2068                         dst_release(ndst);
2069                         return;
2070                 }
2071                 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2072                                      &saddr, &dst->sin6.sin6_addr,
2073                                      0, ttl, src_port, dst_port, !udp_sum);
2074 #endif
2075         }
2076
2077         return;
2078
2079 drop:
2080         dev->stats.tx_dropped++;
2081         goto tx_free;
2082
2083 xmit_tx_error:
2084         /* skb is already freed. */
2085         skb = NULL;
2086 rt_tx_error:
2087         ip_rt_put(rt);
2088 tx_error:
2089         dev->stats.tx_errors++;
2090 tx_free:
2091         dev_kfree_skb(skb);
2092 }
2093
2094 /* Transmit local packets over Vxlan
2095  *
2096  * Outer IP header inherits ECN and DF from inner header.
2097  * Outer UDP destination is the VXLAN assigned port.
2098  *           source port is based on hash of flow
2099  */
2100 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2101 {
2102         struct vxlan_dev *vxlan = netdev_priv(dev);
2103         const struct ip_tunnel_info *info;
2104         struct ethhdr *eth;
2105         bool did_rsc = false;
2106         struct vxlan_rdst *rdst, *fdst = NULL;
2107         struct vxlan_fdb *f;
2108
2109         info = skb_tunnel_info(skb);
2110
2111         skb_reset_mac_header(skb);
2112         eth = eth_hdr(skb);
2113
2114         if ((vxlan->flags & VXLAN_F_PROXY)) {
2115                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2116                         return arp_reduce(dev, skb);
2117 #if IS_ENABLED(CONFIG_IPV6)
2118                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2119                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2120                                        + sizeof(struct nd_msg)) &&
2121                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2122                                 struct nd_msg *msg;
2123
2124                                 msg = (struct nd_msg *)skb_transport_header(skb);
2125                                 if (msg->icmph.icmp6_code == 0 &&
2126                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2127                                         return neigh_reduce(dev, skb);
2128                 }
2129                 eth = eth_hdr(skb);
2130 #endif
2131         }
2132
2133         if (vxlan->flags & VXLAN_F_COLLECT_METADATA &&
2134             info && info->mode & IP_TUNNEL_INFO_TX) {
2135                 vxlan_xmit_one(skb, dev, NULL, false);
2136                 return NETDEV_TX_OK;
2137         }
2138
2139         f = vxlan_find_mac(vxlan, eth->h_dest);
2140         did_rsc = false;
2141
2142         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2143             (ntohs(eth->h_proto) == ETH_P_IP ||
2144              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2145                 did_rsc = route_shortcircuit(dev, skb);
2146                 if (did_rsc)
2147                         f = vxlan_find_mac(vxlan, eth->h_dest);
2148         }
2149
2150         if (f == NULL) {
2151                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2152                 if (f == NULL) {
2153                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2154                             !is_multicast_ether_addr(eth->h_dest))
2155                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2156
2157                         dev->stats.tx_dropped++;
2158                         kfree_skb(skb);
2159                         return NETDEV_TX_OK;
2160                 }
2161         }
2162
2163         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2164                 struct sk_buff *skb1;
2165
2166                 if (!fdst) {
2167                         fdst = rdst;
2168                         continue;
2169                 }
2170                 skb1 = skb_clone(skb, GFP_ATOMIC);
2171                 if (skb1)
2172                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2173         }
2174
2175         if (fdst)
2176                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2177         else
2178                 kfree_skb(skb);
2179         return NETDEV_TX_OK;
2180 }
2181
2182 /* Walk the forwarding table and purge stale entries */
2183 static void vxlan_cleanup(unsigned long arg)
2184 {
2185         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2186         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2187         unsigned int h;
2188
2189         if (!netif_running(vxlan->dev))
2190                 return;
2191
2192         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2193                 struct hlist_node *p, *n;
2194
2195                 spin_lock_bh(&vxlan->hash_lock);
2196                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2197                         struct vxlan_fdb *f
2198                                 = container_of(p, struct vxlan_fdb, hlist);
2199                         unsigned long timeout;
2200
2201                         if (f->state & NUD_PERMANENT)
2202                                 continue;
2203
2204                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2205                         if (time_before_eq(timeout, jiffies)) {
2206                                 netdev_dbg(vxlan->dev,
2207                                            "garbage collect %pM\n",
2208                                            f->eth_addr);
2209                                 f->state = NUD_STALE;
2210                                 vxlan_fdb_destroy(vxlan, f);
2211                         } else if (time_before(timeout, next_timer))
2212                                 next_timer = timeout;
2213                 }
2214                 spin_unlock_bh(&vxlan->hash_lock);
2215         }
2216
2217         mod_timer(&vxlan->age_timer, next_timer);
2218 }
2219
2220 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2221 {
2222         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2223         __be32 vni = vxlan->default_dst.remote_vni;
2224
2225         spin_lock(&vn->sock_lock);
2226         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2227         spin_unlock(&vn->sock_lock);
2228 }
2229
2230 /* Setup stats when device is created */
2231 static int vxlan_init(struct net_device *dev)
2232 {
2233         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2234         if (!dev->tstats)
2235                 return -ENOMEM;
2236
2237         return 0;
2238 }
2239
2240 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2241 {
2242         struct vxlan_fdb *f;
2243
2244         spin_lock_bh(&vxlan->hash_lock);
2245         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2246         if (f)
2247                 vxlan_fdb_destroy(vxlan, f);
2248         spin_unlock_bh(&vxlan->hash_lock);
2249 }
2250
2251 static void vxlan_uninit(struct net_device *dev)
2252 {
2253         struct vxlan_dev *vxlan = netdev_priv(dev);
2254
2255         vxlan_fdb_delete_default(vxlan);
2256
2257         free_percpu(dev->tstats);
2258 }
2259
2260 /* Start ageing timer and join group when device is brought up */
2261 static int vxlan_open(struct net_device *dev)
2262 {
2263         struct vxlan_dev *vxlan = netdev_priv(dev);
2264         int ret;
2265
2266         ret = vxlan_sock_add(vxlan);
2267         if (ret < 0)
2268                 return ret;
2269
2270         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2271                 ret = vxlan_igmp_join(vxlan);
2272                 if (ret == -EADDRINUSE)
2273                         ret = 0;
2274                 if (ret) {
2275                         vxlan_sock_release(vxlan);
2276                         return ret;
2277                 }
2278         }
2279
2280         if (vxlan->cfg.age_interval)
2281                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2282
2283         return ret;
2284 }
2285
2286 /* Purge the forwarding table */
2287 static void vxlan_flush(struct vxlan_dev *vxlan)
2288 {
2289         unsigned int h;
2290
2291         spin_lock_bh(&vxlan->hash_lock);
2292         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2293                 struct hlist_node *p, *n;
2294                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2295                         struct vxlan_fdb *f
2296                                 = container_of(p, struct vxlan_fdb, hlist);
2297                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2298                         if (!is_zero_ether_addr(f->eth_addr))
2299                                 vxlan_fdb_destroy(vxlan, f);
2300                 }
2301         }
2302         spin_unlock_bh(&vxlan->hash_lock);
2303 }
2304
2305 /* Cleanup timer and forwarding table on shutdown */
2306 static int vxlan_stop(struct net_device *dev)
2307 {
2308         struct vxlan_dev *vxlan = netdev_priv(dev);
2309         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2310         int ret = 0;
2311
2312         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2313             !vxlan_group_used(vn, vxlan))
2314                 ret = vxlan_igmp_leave(vxlan);
2315
2316         del_timer_sync(&vxlan->age_timer);
2317
2318         vxlan_flush(vxlan);
2319         vxlan_sock_release(vxlan);
2320
2321         return ret;
2322 }
2323
2324 /* Stub, nothing needs to be done. */
2325 static void vxlan_set_multicast_list(struct net_device *dev)
2326 {
2327 }
2328
2329 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2330 {
2331         struct vxlan_dev *vxlan = netdev_priv(dev);
2332         struct vxlan_rdst *dst = &vxlan->default_dst;
2333         struct net_device *lowerdev;
2334         int max_mtu;
2335
2336         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2337         if (lowerdev == NULL)
2338                 return eth_change_mtu(dev, new_mtu);
2339
2340         if (dst->remote_ip.sa.sa_family == AF_INET6)
2341                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2342         else
2343                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2344
2345         if (new_mtu < 68 || new_mtu > max_mtu)
2346                 return -EINVAL;
2347
2348         dev->mtu = new_mtu;
2349         return 0;
2350 }
2351
2352 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2353 {
2354         struct vxlan_dev *vxlan = netdev_priv(dev);
2355         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2356         __be16 sport, dport;
2357
2358         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2359                                   vxlan->cfg.port_max, true);
2360         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2361
2362         if (ip_tunnel_info_af(info) == AF_INET) {
2363                 struct rtable *rt;
2364
2365                 if (!vxlan->vn4_sock)
2366                         return -EINVAL;
2367                 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2368                                      info->key.u.ipv4.dst,
2369                                      &info->key.u.ipv4.src, NULL, info);
2370                 if (IS_ERR(rt))
2371                         return PTR_ERR(rt);
2372                 ip_rt_put(rt);
2373         } else {
2374 #if IS_ENABLED(CONFIG_IPV6)
2375                 struct dst_entry *ndst;
2376
2377                 if (!vxlan->vn6_sock)
2378                         return -EINVAL;
2379                 ndst = vxlan6_get_route(vxlan, skb, 0,
2380                                         &info->key.u.ipv6.dst,
2381                                         &info->key.u.ipv6.src, NULL);
2382                 if (IS_ERR(ndst))
2383                         return PTR_ERR(ndst);
2384                 dst_release(ndst);
2385 #else /* !CONFIG_IPV6 */
2386                 return -EPFNOSUPPORT;
2387 #endif
2388         }
2389         info->key.tp_src = sport;
2390         info->key.tp_dst = dport;
2391         return 0;
2392 }
2393
2394 static const struct net_device_ops vxlan_netdev_ops = {
2395         .ndo_init               = vxlan_init,
2396         .ndo_uninit             = vxlan_uninit,
2397         .ndo_open               = vxlan_open,
2398         .ndo_stop               = vxlan_stop,
2399         .ndo_start_xmit         = vxlan_xmit,
2400         .ndo_get_stats64        = ip_tunnel_get_stats64,
2401         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2402         .ndo_change_mtu         = vxlan_change_mtu,
2403         .ndo_validate_addr      = eth_validate_addr,
2404         .ndo_set_mac_address    = eth_mac_addr,
2405         .ndo_fdb_add            = vxlan_fdb_add,
2406         .ndo_fdb_del            = vxlan_fdb_delete,
2407         .ndo_fdb_dump           = vxlan_fdb_dump,
2408         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2409 };
2410
2411 /* Info for udev, that this is a virtual tunnel endpoint */
2412 static struct device_type vxlan_type = {
2413         .name = "vxlan",
2414 };
2415
2416 /* Calls the ndo_add_vxlan_port of the caller in order to
2417  * supply the listening VXLAN udp ports. Callers are expected
2418  * to implement the ndo_add_vxlan_port.
2419  */
2420 void vxlan_get_rx_port(struct net_device *dev)
2421 {
2422         struct vxlan_sock *vs;
2423         struct net *net = dev_net(dev);
2424         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2425         sa_family_t sa_family;
2426         __be16 port;
2427         unsigned int i;
2428
2429         spin_lock(&vn->sock_lock);
2430         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2431                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2432                         port = inet_sk(vs->sock->sk)->inet_sport;
2433                         sa_family = vxlan_get_sk_family(vs);
2434                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2435                                                             port);
2436                 }
2437         }
2438         spin_unlock(&vn->sock_lock);
2439 }
2440 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2441
2442 /* Initialize the device structure. */
2443 static void vxlan_setup(struct net_device *dev)
2444 {
2445         struct vxlan_dev *vxlan = netdev_priv(dev);
2446         unsigned int h;
2447
2448         eth_hw_addr_random(dev);
2449         ether_setup(dev);
2450
2451         dev->netdev_ops = &vxlan_netdev_ops;
2452         dev->destructor = free_netdev;
2453         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2454
2455         dev->features   |= NETIF_F_LLTX;
2456         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2457         dev->features   |= NETIF_F_RXCSUM;
2458         dev->features   |= NETIF_F_GSO_SOFTWARE;
2459
2460         dev->vlan_features = dev->features;
2461         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2462         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2463         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2464         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2465         netif_keep_dst(dev);
2466         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
2467
2468         INIT_LIST_HEAD(&vxlan->next);
2469         spin_lock_init(&vxlan->hash_lock);
2470
2471         init_timer_deferrable(&vxlan->age_timer);
2472         vxlan->age_timer.function = vxlan_cleanup;
2473         vxlan->age_timer.data = (unsigned long) vxlan;
2474
2475         vxlan->cfg.dst_port = htons(vxlan_port);
2476
2477         vxlan->dev = dev;
2478
2479         gro_cells_init(&vxlan->gro_cells, dev);
2480
2481         for (h = 0; h < FDB_HASH_SIZE; ++h)
2482                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2483 }
2484
2485 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2486         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2487         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2488         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2489         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2490         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2491         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2492         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2493         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2494         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2495         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2496         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2497         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2498         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2499         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2500         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2501         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2502         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2503         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2504         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2505         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2506         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2507         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2508         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2509         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2510         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2511 };
2512
2513 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2514 {
2515         if (tb[IFLA_ADDRESS]) {
2516                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2517                         pr_debug("invalid link address (not ethernet)\n");
2518                         return -EINVAL;
2519                 }
2520
2521                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2522                         pr_debug("invalid all zero ethernet address\n");
2523                         return -EADDRNOTAVAIL;
2524                 }
2525         }
2526
2527         if (!data)
2528                 return -EINVAL;
2529
2530         if (data[IFLA_VXLAN_ID]) {
2531                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2532                 if (id >= VXLAN_VID_MASK)
2533                         return -ERANGE;
2534         }
2535
2536         if (data[IFLA_VXLAN_PORT_RANGE]) {
2537                 const struct ifla_vxlan_port_range *p
2538                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2539
2540                 if (ntohs(p->high) < ntohs(p->low)) {
2541                         pr_debug("port range %u .. %u not valid\n",
2542                                  ntohs(p->low), ntohs(p->high));
2543                         return -EINVAL;
2544                 }
2545         }
2546
2547         return 0;
2548 }
2549
2550 static void vxlan_get_drvinfo(struct net_device *netdev,
2551                               struct ethtool_drvinfo *drvinfo)
2552 {
2553         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2554         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2555 }
2556
2557 static const struct ethtool_ops vxlan_ethtool_ops = {
2558         .get_drvinfo    = vxlan_get_drvinfo,
2559         .get_link       = ethtool_op_get_link,
2560 };
2561
2562 static void vxlan_del_work(struct work_struct *work)
2563 {
2564         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2565         udp_tunnel_sock_release(vs->sock);
2566         kfree_rcu(vs, rcu);
2567 }
2568
2569 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2570                                         __be16 port, u32 flags)
2571 {
2572         struct socket *sock;
2573         struct udp_port_cfg udp_conf;
2574         int err;
2575
2576         memset(&udp_conf, 0, sizeof(udp_conf));
2577
2578         if (ipv6) {
2579                 udp_conf.family = AF_INET6;
2580                 udp_conf.use_udp6_rx_checksums =
2581                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2582                 udp_conf.ipv6_v6only = 1;
2583         } else {
2584                 udp_conf.family = AF_INET;
2585         }
2586
2587         udp_conf.local_udp_port = port;
2588
2589         /* Open UDP socket */
2590         err = udp_sock_create(net, &udp_conf, &sock);
2591         if (err < 0)
2592                 return ERR_PTR(err);
2593
2594         return sock;
2595 }
2596
2597 /* Create new listen socket if needed */
2598 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2599                                               __be16 port, u32 flags)
2600 {
2601         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2602         struct vxlan_sock *vs;
2603         struct socket *sock;
2604         unsigned int h;
2605         struct udp_tunnel_sock_cfg tunnel_cfg;
2606
2607         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2608         if (!vs)
2609                 return ERR_PTR(-ENOMEM);
2610
2611         for (h = 0; h < VNI_HASH_SIZE; ++h)
2612                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2613
2614         INIT_WORK(&vs->del_work, vxlan_del_work);
2615
2616         sock = vxlan_create_sock(net, ipv6, port, flags);
2617         if (IS_ERR(sock)) {
2618                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2619                         PTR_ERR(sock));
2620                 kfree(vs);
2621                 return ERR_CAST(sock);
2622         }
2623
2624         vs->sock = sock;
2625         atomic_set(&vs->refcnt, 1);
2626         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2627
2628         /* Initialize the vxlan udp offloads structure */
2629         vs->udp_offloads.port = port;
2630         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2631         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2632
2633         spin_lock(&vn->sock_lock);
2634         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2635         vxlan_notify_add_rx_port(vs);
2636         spin_unlock(&vn->sock_lock);
2637
2638         /* Mark socket as an encapsulation socket. */
2639         tunnel_cfg.sk_user_data = vs;
2640         tunnel_cfg.encap_type = 1;
2641         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2642         tunnel_cfg.encap_destroy = NULL;
2643
2644         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2645
2646         return vs;
2647 }
2648
2649 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2650 {
2651         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2652         struct vxlan_sock *vs = NULL;
2653
2654         if (!vxlan->cfg.no_share) {
2655                 spin_lock(&vn->sock_lock);
2656                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2657                                      vxlan->cfg.dst_port, vxlan->flags);
2658                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2659                         spin_unlock(&vn->sock_lock);
2660                         return -EBUSY;
2661                 }
2662                 spin_unlock(&vn->sock_lock);
2663         }
2664         if (!vs)
2665                 vs = vxlan_socket_create(vxlan->net, ipv6,
2666                                          vxlan->cfg.dst_port, vxlan->flags);
2667         if (IS_ERR(vs))
2668                 return PTR_ERR(vs);
2669 #if IS_ENABLED(CONFIG_IPV6)
2670         if (ipv6)
2671                 vxlan->vn6_sock = vs;
2672         else
2673 #endif
2674                 vxlan->vn4_sock = vs;
2675         vxlan_vs_add_dev(vs, vxlan);
2676         return 0;
2677 }
2678
2679 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2680 {
2681         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2682         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2683         int ret = 0;
2684
2685         vxlan->vn4_sock = NULL;
2686 #if IS_ENABLED(CONFIG_IPV6)
2687         vxlan->vn6_sock = NULL;
2688         if (ipv6 || metadata)
2689                 ret = __vxlan_sock_add(vxlan, true);
2690 #endif
2691         if (!ret && (!ipv6 || metadata))
2692                 ret = __vxlan_sock_add(vxlan, false);
2693         if (ret < 0)
2694                 vxlan_sock_release(vxlan);
2695         return ret;
2696 }
2697
2698 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2699                                struct vxlan_config *conf)
2700 {
2701         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2702         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2703         struct vxlan_rdst *dst = &vxlan->default_dst;
2704         unsigned short needed_headroom = ETH_HLEN;
2705         int err;
2706         bool use_ipv6 = false;
2707         __be16 default_port = vxlan->cfg.dst_port;
2708
2709         vxlan->net = src_net;
2710
2711         dst->remote_vni = conf->vni;
2712
2713         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2714
2715         /* Unless IPv6 is explicitly requested, assume IPv4 */
2716         if (!dst->remote_ip.sa.sa_family)
2717                 dst->remote_ip.sa.sa_family = AF_INET;
2718
2719         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2720             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2721                 if (!IS_ENABLED(CONFIG_IPV6))
2722                         return -EPFNOSUPPORT;
2723                 use_ipv6 = true;
2724                 vxlan->flags |= VXLAN_F_IPV6;
2725         }
2726
2727         if (conf->remote_ifindex) {
2728                 struct net_device *lowerdev
2729                          = __dev_get_by_index(src_net, conf->remote_ifindex);
2730
2731                 dst->remote_ifindex = conf->remote_ifindex;
2732
2733                 if (!lowerdev) {
2734                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2735                         return -ENODEV;
2736                 }
2737
2738 #if IS_ENABLED(CONFIG_IPV6)
2739                 if (use_ipv6) {
2740                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2741                         if (idev && idev->cnf.disable_ipv6) {
2742                                 pr_info("IPv6 is disabled via sysctl\n");
2743                                 return -EPERM;
2744                         }
2745                 }
2746 #endif
2747
2748                 if (!conf->mtu)
2749                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2750
2751                 needed_headroom = lowerdev->hard_header_len;
2752         }
2753
2754         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2755                 needed_headroom += VXLAN6_HEADROOM;
2756         else
2757                 needed_headroom += VXLAN_HEADROOM;
2758         dev->needed_headroom = needed_headroom;
2759
2760         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2761         if (!vxlan->cfg.dst_port)
2762                 vxlan->cfg.dst_port = default_port;
2763         vxlan->flags |= conf->flags;
2764
2765         if (!vxlan->cfg.age_interval)
2766                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2767
2768         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2769                 if (tmp->cfg.vni == conf->vni &&
2770                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2771                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2772                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2773                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2774                     (vxlan->flags & VXLAN_F_RCV_FLAGS))
2775                 return -EEXIST;
2776         }
2777
2778         dev->ethtool_ops = &vxlan_ethtool_ops;
2779
2780         /* create an fdb entry for a valid default destination */
2781         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2782                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2783                                        &vxlan->default_dst.remote_ip,
2784                                        NUD_REACHABLE|NUD_PERMANENT,
2785                                        NLM_F_EXCL|NLM_F_CREATE,
2786                                        vxlan->cfg.dst_port,
2787                                        vxlan->default_dst.remote_vni,
2788                                        vxlan->default_dst.remote_ifindex,
2789                                        NTF_SELF);
2790                 if (err)
2791                         return err;
2792         }
2793
2794         err = register_netdevice(dev);
2795         if (err) {
2796                 vxlan_fdb_delete_default(vxlan);
2797                 return err;
2798         }
2799
2800         list_add(&vxlan->next, &vn->vxlan_list);
2801
2802         return 0;
2803 }
2804
2805 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2806                                     u8 name_assign_type, struct vxlan_config *conf)
2807 {
2808         struct nlattr *tb[IFLA_MAX+1];
2809         struct net_device *dev;
2810         int err;
2811
2812         memset(&tb, 0, sizeof(tb));
2813
2814         dev = rtnl_create_link(net, name, name_assign_type,
2815                                &vxlan_link_ops, tb);
2816         if (IS_ERR(dev))
2817                 return dev;
2818
2819         err = vxlan_dev_configure(net, dev, conf);
2820         if (err < 0) {
2821                 free_netdev(dev);
2822                 return ERR_PTR(err);
2823         }
2824
2825         return dev;
2826 }
2827 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2828
2829 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2830                          struct nlattr *tb[], struct nlattr *data[])
2831 {
2832         struct vxlan_config conf;
2833         int err;
2834
2835         memset(&conf, 0, sizeof(conf));
2836
2837         if (data[IFLA_VXLAN_ID])
2838                 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2839
2840         if (data[IFLA_VXLAN_GROUP]) {
2841                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2842         } else if (data[IFLA_VXLAN_GROUP6]) {
2843                 if (!IS_ENABLED(CONFIG_IPV6))
2844                         return -EPFNOSUPPORT;
2845
2846                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2847                 conf.remote_ip.sa.sa_family = AF_INET6;
2848         }
2849
2850         if (data[IFLA_VXLAN_LOCAL]) {
2851                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2852                 conf.saddr.sa.sa_family = AF_INET;
2853         } else if (data[IFLA_VXLAN_LOCAL6]) {
2854                 if (!IS_ENABLED(CONFIG_IPV6))
2855                         return -EPFNOSUPPORT;
2856
2857                 /* TODO: respect scope id */
2858                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2859                 conf.saddr.sa.sa_family = AF_INET6;
2860         }
2861
2862         if (data[IFLA_VXLAN_LINK])
2863                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2864
2865         if (data[IFLA_VXLAN_TOS])
2866                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2867
2868         if (data[IFLA_VXLAN_TTL])
2869                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2870
2871         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2872                 conf.flags |= VXLAN_F_LEARN;
2873
2874         if (data[IFLA_VXLAN_AGEING])
2875                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2876
2877         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2878                 conf.flags |= VXLAN_F_PROXY;
2879
2880         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2881                 conf.flags |= VXLAN_F_RSC;
2882
2883         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2884                 conf.flags |= VXLAN_F_L2MISS;
2885
2886         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2887                 conf.flags |= VXLAN_F_L3MISS;
2888
2889         if (data[IFLA_VXLAN_LIMIT])
2890                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2891
2892         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2893             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2894                 conf.flags |= VXLAN_F_COLLECT_METADATA;
2895
2896         if (data[IFLA_VXLAN_PORT_RANGE]) {
2897                 const struct ifla_vxlan_port_range *p
2898                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2899                 conf.port_min = ntohs(p->low);
2900                 conf.port_max = ntohs(p->high);
2901         }
2902
2903         if (data[IFLA_VXLAN_PORT])
2904                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2905
2906         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2907                 conf.flags |= VXLAN_F_UDP_CSUM;
2908
2909         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2910             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2911                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2912
2913         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2914             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2915                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2916
2917         if (data[IFLA_VXLAN_REMCSUM_TX] &&
2918             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2919                 conf.flags |= VXLAN_F_REMCSUM_TX;
2920
2921         if (data[IFLA_VXLAN_REMCSUM_RX] &&
2922             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2923                 conf.flags |= VXLAN_F_REMCSUM_RX;
2924
2925         if (data[IFLA_VXLAN_GBP])
2926                 conf.flags |= VXLAN_F_GBP;
2927
2928         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2929                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2930
2931         err = vxlan_dev_configure(src_net, dev, &conf);
2932         switch (err) {
2933         case -ENODEV:
2934                 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
2935                 break;
2936
2937         case -EPERM:
2938                 pr_info("IPv6 is disabled via sysctl\n");
2939                 break;
2940
2941         case -EEXIST:
2942                 pr_info("duplicate VNI %u\n", be32_to_cpu(conf.vni));
2943                 break;
2944         }
2945
2946         return err;
2947 }
2948
2949 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2950 {
2951         struct vxlan_dev *vxlan = netdev_priv(dev);
2952         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2953
2954         spin_lock(&vn->sock_lock);
2955         if (!hlist_unhashed(&vxlan->hlist))
2956                 hlist_del_rcu(&vxlan->hlist);
2957         spin_unlock(&vn->sock_lock);
2958
2959         gro_cells_destroy(&vxlan->gro_cells);
2960         list_del(&vxlan->next);
2961         unregister_netdevice_queue(dev, head);
2962 }
2963
2964 static size_t vxlan_get_size(const struct net_device *dev)
2965 {
2966
2967         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2968                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2969                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2970                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2971                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2972                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2973                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2974                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2975                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2976                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2977                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2978                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
2979                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2980                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2981                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2982                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2983                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2984                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2985                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2986                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
2987                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
2988                 0;
2989 }
2990
2991 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2992 {
2993         const struct vxlan_dev *vxlan = netdev_priv(dev);
2994         const struct vxlan_rdst *dst = &vxlan->default_dst;
2995         struct ifla_vxlan_port_range ports = {
2996                 .low =  htons(vxlan->cfg.port_min),
2997                 .high = htons(vxlan->cfg.port_max),
2998         };
2999
3000         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3001                 goto nla_put_failure;
3002
3003         if (!vxlan_addr_any(&dst->remote_ip)) {
3004                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3005                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3006                                             dst->remote_ip.sin.sin_addr.s_addr))
3007                                 goto nla_put_failure;
3008 #if IS_ENABLED(CONFIG_IPV6)
3009                 } else {
3010                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3011                                              &dst->remote_ip.sin6.sin6_addr))
3012                                 goto nla_put_failure;
3013 #endif
3014                 }
3015         }
3016
3017         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3018                 goto nla_put_failure;
3019
3020         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3021                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3022                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3023                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3024                                 goto nla_put_failure;
3025 #if IS_ENABLED(CONFIG_IPV6)
3026                 } else {
3027                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3028                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3029                                 goto nla_put_failure;
3030 #endif
3031                 }
3032         }
3033
3034         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3035             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3036             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3037                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3038             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3039                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3040             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3041             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3042                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3043             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3044                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3045             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3046                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3047             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3048             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3049             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3050             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3051                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
3052             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3053                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3054             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3055                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3056             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3057                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3058             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3059                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3060                 goto nla_put_failure;
3061
3062         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3063                 goto nla_put_failure;
3064
3065         if (vxlan->flags & VXLAN_F_GBP &&
3066             nla_put_flag(skb, IFLA_VXLAN_GBP))
3067                 goto nla_put_failure;
3068
3069         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3070             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3071                 goto nla_put_failure;
3072
3073         return 0;
3074
3075 nla_put_failure:
3076         return -EMSGSIZE;
3077 }
3078
3079 static struct net *vxlan_get_link_net(const struct net_device *dev)
3080 {
3081         struct vxlan_dev *vxlan = netdev_priv(dev);
3082
3083         return vxlan->net;
3084 }
3085
3086 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3087         .kind           = "vxlan",
3088         .maxtype        = IFLA_VXLAN_MAX,
3089         .policy         = vxlan_policy,
3090         .priv_size      = sizeof(struct vxlan_dev),
3091         .setup          = vxlan_setup,
3092         .validate       = vxlan_validate,
3093         .newlink        = vxlan_newlink,
3094         .dellink        = vxlan_dellink,
3095         .get_size       = vxlan_get_size,
3096         .fill_info      = vxlan_fill_info,
3097         .get_link_net   = vxlan_get_link_net,
3098 };
3099
3100 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3101                                              struct net_device *dev)
3102 {
3103         struct vxlan_dev *vxlan, *next;
3104         LIST_HEAD(list_kill);
3105
3106         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3107                 struct vxlan_rdst *dst = &vxlan->default_dst;
3108
3109                 /* In case we created vxlan device with carrier
3110                  * and we loose the carrier due to module unload
3111                  * we also need to remove vxlan device. In other
3112                  * cases, it's not necessary and remote_ifindex
3113                  * is 0 here, so no matches.
3114                  */
3115                 if (dst->remote_ifindex == dev->ifindex)
3116                         vxlan_dellink(vxlan->dev, &list_kill);
3117         }
3118
3119         unregister_netdevice_many(&list_kill);
3120 }
3121
3122 static int vxlan_lowerdev_event(struct notifier_block *unused,
3123                                 unsigned long event, void *ptr)
3124 {
3125         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3126         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3127
3128         if (event == NETDEV_UNREGISTER)
3129                 vxlan_handle_lowerdev_unregister(vn, dev);
3130
3131         return NOTIFY_DONE;
3132 }
3133
3134 static struct notifier_block vxlan_notifier_block __read_mostly = {
3135         .notifier_call = vxlan_lowerdev_event,
3136 };
3137
3138 static __net_init int vxlan_init_net(struct net *net)
3139 {
3140         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3141         unsigned int h;
3142
3143         INIT_LIST_HEAD(&vn->vxlan_list);
3144         spin_lock_init(&vn->sock_lock);
3145
3146         for (h = 0; h < PORT_HASH_SIZE; ++h)
3147                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3148
3149         return 0;
3150 }
3151
3152 static void __net_exit vxlan_exit_net(struct net *net)
3153 {
3154         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3155         struct vxlan_dev *vxlan, *next;
3156         struct net_device *dev, *aux;
3157         LIST_HEAD(list);
3158
3159         rtnl_lock();
3160         for_each_netdev_safe(net, dev, aux)
3161                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3162                         unregister_netdevice_queue(dev, &list);
3163
3164         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3165                 /* If vxlan->dev is in the same netns, it has already been added
3166                  * to the list by the previous loop.
3167                  */
3168                 if (!net_eq(dev_net(vxlan->dev), net)) {
3169                         gro_cells_destroy(&vxlan->gro_cells);
3170                         unregister_netdevice_queue(vxlan->dev, &list);
3171                 }
3172         }
3173
3174         unregister_netdevice_many(&list);
3175         rtnl_unlock();
3176 }
3177
3178 static struct pernet_operations vxlan_net_ops = {
3179         .init = vxlan_init_net,
3180         .exit = vxlan_exit_net,
3181         .id   = &vxlan_net_id,
3182         .size = sizeof(struct vxlan_net),
3183 };
3184
3185 static int __init vxlan_init_module(void)
3186 {
3187         int rc;
3188
3189         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3190         if (!vxlan_wq)
3191                 return -ENOMEM;
3192
3193         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3194
3195         rc = register_pernet_subsys(&vxlan_net_ops);
3196         if (rc)
3197                 goto out1;
3198
3199         rc = register_netdevice_notifier(&vxlan_notifier_block);
3200         if (rc)
3201                 goto out2;
3202
3203         rc = rtnl_link_register(&vxlan_link_ops);
3204         if (rc)
3205                 goto out3;
3206
3207         return 0;
3208 out3:
3209         unregister_netdevice_notifier(&vxlan_notifier_block);
3210 out2:
3211         unregister_pernet_subsys(&vxlan_net_ops);
3212 out1:
3213         destroy_workqueue(vxlan_wq);
3214         return rc;
3215 }
3216 late_initcall(vxlan_init_module);
3217
3218 static void __exit vxlan_cleanup_module(void)
3219 {
3220         rtnl_link_unregister(&vxlan_link_ops);
3221         unregister_netdevice_notifier(&vxlan_notifier_block);
3222         destroy_workqueue(vxlan_wq);
3223         unregister_pernet_subsys(&vxlan_net_ops);
3224         /* rcu_barrier() is called by netns */
3225 }
3226 module_exit(vxlan_cleanup_module);
3227
3228 MODULE_LICENSE("GPL");
3229 MODULE_VERSION(VXLAN_VERSION);
3230 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3231 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3232 MODULE_ALIAS_RTNL_LINK("vxlan");