kernel/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 #include <net/udp_tunnel.h>
  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];
  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, u32 id)
 201 {
 202         return &vs->vni_list[hash_32(id, 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, u32 id)
 246 {
 247         struct vxlan_dev *vxlan;
 248
 249         hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
 250                 if (vxlan->default_dst.remote_vni == id)
 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, u32 id,
 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, id);
 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, 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 = 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                                               __u32 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, __u32 vni, __u32 ifindex)
 473 {
 474         struct vxlan_rdst *rd;
 475
 476         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 477         if (rd)
 478                 return 0;
 479
 480         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 481         if (!rd)
 482                 return 0;
 483         rd->remote_ip = *ip;
 484         rd->remote_port = port;
 485         rd->remote_vni = vni;
 486         rd->remote_ifindex = ifindex;
 487         return 1;
 488 }
 489
 490 /* Add/update destinations for multicast */
 491 static int vxlan_fdb_append(struct vxlan_fdb *f,
 492                             union vxlan_addr *ip, __be16 port, __u32 vni,
 493                             __u32 ifindex, struct vxlan_rdst **rdp)
 494 {
 495         struct vxlan_rdst *rd;
 496
 497         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 498         if (rd)
 499                 return 0;
 500
 501         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 502         if (rd == NULL)
 503                 return -ENOBUFS;
 504         rd->remote_ip = *ip;
 505         rd->remote_port = port;
 506         rd->remote_vni = vni;
 507         rd->remote_ifindex = ifindex;
 508
 509         list_add_tail_rcu(&rd->list, &f->remotes);
 510
 511         *rdp = rd;
 512         return 1;
 513 }
 514
 515 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 516                                           unsigned int off,
 517                                           struct vxlanhdr *vh, size_t hdrlen,
 518                                           u32 data, struct gro_remcsum *grc,
 519                                           bool nopartial)
 520 {
 521         size_t start, offset;
 522
 523         if (skb->remcsum_offload)
 524                 return vh;
 525
 526         if (!NAPI_GRO_CB(skb)->csum_valid)
 527                 return NULL;
 528
 529         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
 530         offset = start + ((data & VXLAN_RCO_UDP) ?
 531                           offsetof(struct udphdr, check) :
 532                           offsetof(struct tcphdr, check));
 533
 534         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 535                                      start, offset, grc, nopartial);
 536
 537         skb->remcsum_offload = 1;
 538
 539         return vh;
 540 }
 541
 542 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
 543                                           struct sk_buff *skb,
 544                                           struct udp_offload *uoff)
 545 {
 546         struct sk_buff *p, **pp = NULL;
 547         struct vxlanhdr *vh, *vh2;
 548         unsigned int hlen, off_vx;
 549         int flush = 1;
 550         struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
 551                                              udp_offloads);
 552         u32 flags;
 553         struct gro_remcsum grc;
 554
 555         skb_gro_remcsum_init(&grc);
 556
 557         off_vx = skb_gro_offset(skb);
 558         hlen = off_vx + sizeof(*vh);
 559         vh   = skb_gro_header_fast(skb, off_vx);
 560         if (skb_gro_header_hard(skb, hlen)) {
 561                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 562                 if (unlikely(!vh))
 563                         goto out;
 564         }
 565
 566         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 567
 568         flags = ntohl(vh->vx_flags);
 569
 570         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 571                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 572                                        ntohl(vh->vx_vni), &grc,
 573                                        !!(vs->flags &
 574                                           VXLAN_F_REMCSUM_NOPARTIAL));
 575
 576                 if (!vh)
 577                         goto out;
 578         }
 579
 580         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 581
 582         flush = 0;
 583
 584         for (p = *head; p; p = p->next) {
 585                 if (!NAPI_GRO_CB(p)->same_flow)
 586                         continue;
 587
 588                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 589                 if (vh->vx_flags != vh2->vx_flags ||
 590                     vh->vx_vni != vh2->vx_vni) {
 591                         NAPI_GRO_CB(p)->same_flow = 0;
 592                         continue;
 593                 }
 594         }
 595
 596         pp = eth_gro_receive(head, skb);
 597
 598 out:
 599         skb_gro_remcsum_cleanup(skb, &grc);
 600         NAPI_GRO_CB(skb)->flush |= flush;
 601
 602         return pp;
 603 }
 604
 605 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
 606                               struct udp_offload *uoff)
 607 {
 608         udp_tunnel_gro_complete(skb, nhoff);
 609
 610         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 611 }
 612
 613 /* Notify netdevs that UDP port started listening */
 614 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 615 {
 616         struct net_device *dev;
 617         struct sock *sk = vs->sock->sk;
 618         struct net *net = sock_net(sk);
 619         sa_family_t sa_family = vxlan_get_sk_family(vs);
 620         __be16 port = inet_sk(sk)->inet_sport;
 621         int err;
 622
 623         if (sa_family == AF_INET) {
 624                 err = udp_add_offload(&vs->udp_offloads);
 625                 if (err)
 626                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 627         }
 628
 629         rcu_read_lock();
 630         for_each_netdev_rcu(net, dev) {
 631                 if (dev->netdev_ops->ndo_add_vxlan_port)
 632                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 633                                                             port);
 634         }
 635         rcu_read_unlock();
 636 }
 637
 638 /* Notify netdevs that UDP port is no more listening */
 639 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 640 {
 641         struct net_device *dev;
 642         struct sock *sk = vs->sock->sk;
 643         struct net *net = sock_net(sk);
 644         sa_family_t sa_family = vxlan_get_sk_family(vs);
 645         __be16 port = inet_sk(sk)->inet_sport;
 646
 647         rcu_read_lock();
 648         for_each_netdev_rcu(net, dev) {
 649                 if (dev->netdev_ops->ndo_del_vxlan_port)
 650                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 651                                                             port);
 652         }
 653         rcu_read_unlock();
 654
 655         if (sa_family == AF_INET)
 656                 udp_del_offload(&vs->udp_offloads);
 657 }
 658
 659 /* Add new entry to forwarding table -- assumes lock held */
 660 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 661                             const u8 *mac, union vxlan_addr *ip,
 662                             __u16 state, __u16 flags,
 663                             __be16 port, __u32 vni, __u32 ifindex,
 664                             __u8 ndm_flags)
 665 {
 666         struct vxlan_rdst *rd = NULL;
 667         struct vxlan_fdb *f;
 668         int notify = 0;
 669
 670         f = __vxlan_find_mac(vxlan, mac);
 671         if (f) {
 672                 if (flags & NLM_F_EXCL) {
 673                         netdev_dbg(vxlan->dev,
 674                                    "lost race to create %pM\n", mac);
 675                         return -EEXIST;
 676                 }
 677                 if (f->state != state) {
 678                         f->state = state;
 679                         f->updated = jiffies;
 680                         notify = 1;
 681                 }
 682                 if (f->flags != ndm_flags) {
 683                         f->flags = ndm_flags;
 684                         f->updated = jiffies;
 685                         notify = 1;
 686                 }
 687                 if ((flags & NLM_F_REPLACE)) {
 688                         /* Only change unicasts */
 689                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 690                              is_zero_ether_addr(f->eth_addr))) {
 691                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 692                                                            ifindex);
 693                         } else
 694                                 return -EOPNOTSUPP;
 695                 }
 696                 if ((flags & NLM_F_APPEND) &&
 697                     (is_multicast_ether_addr(f->eth_addr) ||
 698                      is_zero_ether_addr(f->eth_addr))) {
 699                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 700                                                   &rd);
 701
 702                         if (rc < 0)
 703                                 return rc;
 704                         notify |= rc;
 705                 }
 706         } else {
 707                 if (!(flags & NLM_F_CREATE))
 708                         return -ENOENT;
 709
 710                 if (vxlan->cfg.addrmax &&
 711                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 712                         return -ENOSPC;
 713
 714                 /* Disallow replace to add a multicast entry */
 715                 if ((flags & NLM_F_REPLACE) &&
 716                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 717                         return -EOPNOTSUPP;
 718
 719                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 720                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 721                 if (!f)
 722                         return -ENOMEM;
 723
 724                 notify = 1;
 725                 f->state = state;
 726                 f->flags = ndm_flags;
 727                 f->updated = f->used = jiffies;
 728                 INIT_LIST_HEAD(&f->remotes);
 729                 memcpy(f->eth_addr, mac, ETH_ALEN);
 730
 731                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 732
 733                 ++vxlan->addrcnt;
 734                 hlist_add_head_rcu(&f->hlist,
 735                                    vxlan_fdb_head(vxlan, mac));
 736         }
 737
 738         if (notify) {
 739                 if (rd == NULL)
 740                         rd = first_remote_rtnl(f);
 741                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 742         }
 743
 744         return 0;
 745 }
 746
 747 static void vxlan_fdb_free(struct rcu_head *head)
 748 {
 749         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 750         struct vxlan_rdst *rd, *nd;
 751
 752         list_for_each_entry_safe(rd, nd, &f->remotes, list)
 753                 kfree(rd);
 754         kfree(f);
 755 }
 756
 757 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 758 {
 759         netdev_dbg(vxlan->dev,
 760                     "delete %pM\n", f->eth_addr);
 761
 762         --vxlan->addrcnt;
 763         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 764
 765         hlist_del_rcu(&f->hlist);
 766         call_rcu(&f->rcu, vxlan_fdb_free);
 767 }
 768
 769 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 770                            union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
 771 {
 772         struct net *net = dev_net(vxlan->dev);
 773         int err;
 774
 775         if (tb[NDA_DST]) {
 776                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 777                 if (err)
 778                         return err;
 779         } else {
 780                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 781                 if (remote->sa.sa_family == AF_INET) {
 782                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 783                         ip->sa.sa_family = AF_INET;
 784 #if IS_ENABLED(CONFIG_IPV6)
 785                 } else {
 786                         ip->sin6.sin6_addr = in6addr_any;
 787                         ip->sa.sa_family = AF_INET6;
 788 #endif
 789                 }
 790         }
 791
 792         if (tb[NDA_PORT]) {
 793                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 794                         return -EINVAL;
 795                 *port = nla_get_be16(tb[NDA_PORT]);
 796         } else {
 797                 *port = vxlan->cfg.dst_port;
 798         }
 799
 800         if (tb[NDA_VNI]) {
 801                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 802                         return -EINVAL;
 803                 *vni = nla_get_u32(tb[NDA_VNI]);
 804         } else {
 805                 *vni = vxlan->default_dst.remote_vni;
 806         }
 807
 808         if (tb[NDA_IFINDEX]) {
 809                 struct net_device *tdev;
 810
 811                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 812                         return -EINVAL;
 813                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 814                 tdev = __dev_get_by_index(net, *ifindex);
 815                 if (!tdev)
 816                         return -EADDRNOTAVAIL;
 817         } else {
 818                 *ifindex = 0;
 819         }
 820
 821         return 0;
 822 }
 823
 824 /* Add static entry (via netlink) */
 825 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 826                          struct net_device *dev,
 827                          const unsigned char *addr, u16 vid, u16 flags)
 828 {
 829         struct vxlan_dev *vxlan = netdev_priv(dev);
 830         /* struct net *net = dev_net(vxlan->dev); */
 831         union vxlan_addr ip;
 832         __be16 port;
 833         u32 vni, ifindex;
 834         int err;
 835
 836         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 837                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 838                         ndm->ndm_state);
 839                 return -EINVAL;
 840         }
 841
 842         if (tb[NDA_DST] == NULL)
 843                 return -EINVAL;
 844
 845         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 846         if (err)
 847                 return err;
 848
 849         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 850                 return -EAFNOSUPPORT;
 851
 852         spin_lock_bh(&vxlan->hash_lock);
 853         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 854                                port, vni, ifindex, ndm->ndm_flags);
 855         spin_unlock_bh(&vxlan->hash_lock);
 856
 857         return err;
 858 }
 859
 860 /* Delete entry (via netlink) */
 861 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 862                             struct net_device *dev,
 863                             const unsigned char *addr, u16 vid)
 864 {
 865         struct vxlan_dev *vxlan = netdev_priv(dev);
 866         struct vxlan_fdb *f;
 867         struct vxlan_rdst *rd = NULL;
 868         union vxlan_addr ip;
 869         __be16 port;
 870         u32 vni, ifindex;
 871         int err;
 872
 873         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 874         if (err)
 875                 return err;
 876
 877         err = -ENOENT;
 878
 879         spin_lock_bh(&vxlan->hash_lock);
 880         f = vxlan_find_mac(vxlan, addr);
 881         if (!f)
 882                 goto out;
 883
 884         if (!vxlan_addr_any(&ip)) {
 885                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 886                 if (!rd)
 887                         goto out;
 888         }
 889
 890         err = 0;
 891
 892         /* remove a destination if it's not the only one on the list,
 893          * otherwise destroy the fdb entry
 894          */
 895         if (rd && !list_is_singular(&f->remotes)) {
 896                 list_del_rcu(&rd->list);
 897                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 898                 kfree_rcu(rd, rcu);
 899                 goto out;
 900         }
 901
 902         vxlan_fdb_destroy(vxlan, f);
 903
 904 out:
 905         spin_unlock_bh(&vxlan->hash_lock);
 906
 907         return err;
 908 }
 909
 910 /* Dump forwarding table */
 911 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 912                           struct net_device *dev,
 913                           struct net_device *filter_dev, int idx)
 914 {
 915         struct vxlan_dev *vxlan = netdev_priv(dev);
 916         unsigned int h;
 917
 918         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 919                 struct vxlan_fdb *f;
 920                 int err;
 921
 922                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 923                         struct vxlan_rdst *rd;
 924
 925                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 926                                 if (idx < cb->args[0])
 927                                         goto skip;
 928
 929                                 err = vxlan_fdb_info(skb, vxlan, f,
 930                                                      NETLINK_CB(cb->skb).portid,
 931                                                      cb->nlh->nlmsg_seq,
 932                                                      RTM_NEWNEIGH,
 933                                                      NLM_F_MULTI, rd);
 934                                 if (err < 0)
 935                                         goto out;
 936 skip:
 937                                 ++idx;
 938                         }
 939                 }
 940         }
 941 out:
 942         return idx;
 943 }
 944
 945 /* Watch incoming packets to learn mapping between Ethernet address
 946  * and Tunnel endpoint.
 947  * Return true if packet is bogus and should be dropped.
 948  */
 949 static bool vxlan_snoop(struct net_device *dev,
 950                         union vxlan_addr *src_ip, const u8 *src_mac)
 951 {
 952         struct vxlan_dev *vxlan = netdev_priv(dev);
 953         struct vxlan_fdb *f;
 954
 955         f = vxlan_find_mac(vxlan, src_mac);
 956         if (likely(f)) {
 957                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 958
 959                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 960                         return false;
 961
 962                 /* Don't migrate static entries, drop packets */
 963                 if (f->state & NUD_NOARP)
 964                         return true;
 965
 966                 if (net_ratelimit())
 967                         netdev_info(dev,
 968                                     "%pM migrated from %pIS to %pIS\n",
 969                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 970
 971                 rdst->remote_ip = *src_ip;
 972                 f->updated = jiffies;
 973                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 974         } else {
 975                 /* learned new entry */
 976                 spin_lock(&vxlan->hash_lock);
 977
 978                 /* close off race between vxlan_flush and incoming packets */
 979                 if (netif_running(dev))
 980                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 981                                          NUD_REACHABLE,
 982                                          NLM_F_EXCL|NLM_F_CREATE,
 983                                          vxlan->cfg.dst_port,
 984                                          vxlan->default_dst.remote_vni,
 985                                          0, NTF_SELF);
 986                 spin_unlock(&vxlan->hash_lock);
 987         }
 988
 989         return false;
 990 }
 991
 992 /* See if multicast group is already in use by other ID */
 993 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
 994 {
 995         struct vxlan_dev *vxlan;
 996         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
 997
 998         /* The vxlan_sock is only used by dev, leaving group has
 999          * no effect on other vxlan devices.
1000          */
1001         if (family == AF_INET && dev->vn4_sock &&
1002             atomic_read(&dev->vn4_sock->refcnt) == 1)
1003                 return false;
1004 #if IS_ENABLED(CONFIG_IPV6)
1005         if (family == AF_INET6 && dev->vn6_sock &&
1006             atomic_read(&dev->vn6_sock->refcnt) == 1)
1007                 return false;
1008 #endif
1009
1010         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1011                 if (!netif_running(vxlan->dev) || vxlan == dev)
1012                         continue;
1013
1014                 if (family == AF_INET && vxlan->vn4_sock != dev->vn4_sock)
1015                         continue;
1016 #if IS_ENABLED(CONFIG_IPV6)
1017                 if (family == AF_INET6 && vxlan->vn6_sock != dev->vn6_sock)
1018                         continue;
1019 #endif
1020
1021                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1022                                       &dev->default_dst.remote_ip))
1023                         continue;
1024
1025                 if (vxlan->default_dst.remote_ifindex !=
1026                     dev->default_dst.remote_ifindex)
1027                         continue;
1028
1029                 return true;
1030         }
1031
1032         return false;
1033 }
1034
1035 static void __vxlan_sock_release(struct vxlan_sock *vs)
1036 {
1037         struct vxlan_net *vn;
1038
1039         if (!vs)
1040                 return;
1041         if (!atomic_dec_and_test(&vs->refcnt))
1042                 return;
1043
1044         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1045         spin_lock(&vn->sock_lock);
1046         hlist_del_rcu(&vs->hlist);
1047         vxlan_notify_del_rx_port(vs);
1048         spin_unlock(&vn->sock_lock);
1049
1050         queue_work(vxlan_wq, &vs->del_work);
1051 }
1052
1053 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1054 {
1055         __vxlan_sock_release(vxlan->vn4_sock);
1056 #if IS_ENABLED(CONFIG_IPV6)
1057         __vxlan_sock_release(vxlan->vn6_sock);
1058 #endif
1059 }
1060
1061 /* Update multicast group membership when first VNI on
1062  * multicast address is brought up
1063  */
1064 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1065 {
1066         struct sock *sk;
1067         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1068         int ifindex = vxlan->default_dst.remote_ifindex;
1069         int ret = -EINVAL;
1070
1071         if (ip->sa.sa_family == AF_INET) {
1072                 struct ip_mreqn mreq = {
1073                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1074                         .imr_ifindex            = ifindex,
1075                 };
1076
1077                 sk = vxlan->vn4_sock->sock->sk;
1078                 lock_sock(sk);
1079                 ret = ip_mc_join_group(sk, &mreq);
1080                 release_sock(sk);
1081 #if IS_ENABLED(CONFIG_IPV6)
1082         } else {
1083                 sk = vxlan->vn6_sock->sock->sk;
1084                 lock_sock(sk);
1085                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1086                                                    &ip->sin6.sin6_addr);
1087                 release_sock(sk);
1088 #endif
1089         }
1090
1091         return ret;
1092 }
1093
1094 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1095 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1096 {
1097         struct sock *sk;
1098         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1099         int ifindex = vxlan->default_dst.remote_ifindex;
1100         int ret = -EINVAL;
1101
1102         if (ip->sa.sa_family == AF_INET) {
1103                 struct ip_mreqn mreq = {
1104                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1105                         .imr_ifindex            = ifindex,
1106                 };
1107
1108                 sk = vxlan->vn4_sock->sock->sk;
1109                 lock_sock(sk);
1110                 ret = ip_mc_leave_group(sk, &mreq);
1111                 release_sock(sk);
1112 #if IS_ENABLED(CONFIG_IPV6)
1113         } else {
1114                 sk = vxlan->vn6_sock->sock->sk;
1115                 lock_sock(sk);
1116                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1117                                                    &ip->sin6.sin6_addr);
1118                 release_sock(sk);
1119 #endif
1120         }
1121
1122         return ret;
1123 }
1124
1125 static struct vxlanhdr *vxlan_remcsum(struct sk_buff *skb, struct vxlanhdr *vh,
1126                                       size_t hdrlen, u32 data, bool nopartial)
1127 {
1128         size_t start, offset, plen;
1129
1130         if (skb->remcsum_offload)
1131                 return vh;
1132
1133         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
1134         offset = start + ((data & VXLAN_RCO_UDP) ?
1135                           offsetof(struct udphdr, check) :
1136                           offsetof(struct tcphdr, check));
1137
1138         plen = hdrlen + offset + sizeof(u16);
1139
1140         if (!pskb_may_pull(skb, plen))
1141                 return NULL;
1142
1143         vh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1144
1145         skb_remcsum_process(skb, (void *)vh + hdrlen, start, offset,
1146                             nopartial);
1147
1148         return vh;
1149 }
1150
1151 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1152                       struct vxlan_metadata *md, u32 vni,
1153                       struct metadata_dst *tun_dst)
1154 {
1155         struct iphdr *oip = NULL;
1156         struct ipv6hdr *oip6 = NULL;
1157         struct vxlan_dev *vxlan;
1158         struct pcpu_sw_netstats *stats;
1159         union vxlan_addr saddr;
1160         int err = 0;
1161
1162         /* For flow based devices, map all packets to VNI 0 */
1163         if (vs->flags & VXLAN_F_COLLECT_METADATA)
1164                 vni = 0;
1165
1166         /* Is this VNI defined? */
1167         vxlan = vxlan_vs_find_vni(vs, vni);
1168         if (!vxlan)
1169                 goto drop;
1170
1171         skb_reset_mac_header(skb);
1172         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1173         skb->protocol = eth_type_trans(skb, vxlan->dev);
1174         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1175
1176         /* Ignore packet loops (and multicast echo) */
1177         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1178                 goto drop;
1179
1180         /* Get data from the outer IP header */
1181         if (vxlan_get_sk_family(vs) == AF_INET) {
1182                 oip = ip_hdr(skb);
1183                 saddr.sin.sin_addr.s_addr = oip->saddr;
1184                 saddr.sa.sa_family = AF_INET;
1185 #if IS_ENABLED(CONFIG_IPV6)
1186         } else {
1187                 oip6 = ipv6_hdr(skb);
1188                 saddr.sin6.sin6_addr = oip6->saddr;
1189                 saddr.sa.sa_family = AF_INET6;
1190 #endif
1191         }
1192
1193         if (tun_dst) {
1194                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1195                 tun_dst = NULL;
1196         }
1197
1198         if ((vxlan->flags & VXLAN_F_LEARN) &&
1199             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1200                 goto drop;
1201
1202         skb_reset_network_header(skb);
1203         /* In flow-based mode, GBP is carried in dst_metadata */
1204         if (!(vs->flags & VXLAN_F_COLLECT_METADATA))
1205                 skb->mark = md->gbp;
1206
1207         if (oip6)
1208                 err = IP6_ECN_decapsulate(oip6, skb);
1209         if (oip)
1210                 err = IP_ECN_decapsulate(oip, skb);
1211
1212         if (unlikely(err)) {
1213                 if (log_ecn_error) {
1214                         if (oip6)
1215                                 net_info_ratelimited("non-ECT from %pI6\n",
1216                                                      &oip6->saddr);
1217                         if (oip)
1218                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1219                                                      &oip->saddr, oip->tos);
1220                 }
1221                 if (err > 1) {
1222                         ++vxlan->dev->stats.rx_frame_errors;
1223                         ++vxlan->dev->stats.rx_errors;
1224                         goto drop;
1225                 }
1226         }
1227
1228         stats = this_cpu_ptr(vxlan->dev->tstats);
1229         u64_stats_update_begin(&stats->syncp);
1230         stats->rx_packets++;
1231         stats->rx_bytes += skb->len;
1232         u64_stats_update_end(&stats->syncp);
1233
1234         gro_cells_receive(&vxlan->gro_cells, skb);
1235
1236         return;
1237 drop:
1238         if (tun_dst)
1239                 dst_release((struct dst_entry *)tun_dst);
1240
1241         /* Consume bad packet */
1242         kfree_skb(skb);
1243 }
1244
1245 /* Callback from net/ipv4/udp.c to receive packets */
1246 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1247 {
1248         struct metadata_dst *tun_dst = NULL;
1249         struct vxlan_sock *vs;
1250         struct vxlanhdr *vxh;
1251         u32 flags, vni;
1252         struct vxlan_metadata _md;
1253         struct vxlan_metadata *md = &_md;
1254
1255         /* Need Vxlan and inner Ethernet header to be present */
1256         if (!pskb_may_pull(skb, VXLAN_HLEN))
1257                 goto error;
1258
1259         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1260         flags = ntohl(vxh->vx_flags);
1261         vni = ntohl(vxh->vx_vni);
1262
1263         if (flags & VXLAN_HF_VNI) {
1264                 flags &= ~VXLAN_HF_VNI;
1265         } else {
1266                 /* VNI flag always required to be set */
1267                 goto bad_flags;
1268         }
1269
1270         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1271                 goto drop;
1272         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1273
1274         vs = rcu_dereference_sk_user_data(sk);
1275         if (!vs)
1276                 goto drop;
1277
1278         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1279                 vxh = vxlan_remcsum(skb, vxh, sizeof(struct vxlanhdr), vni,
1280                                     !!(vs->flags & VXLAN_F_REMCSUM_NOPARTIAL));
1281                 if (!vxh)
1282                         goto drop;
1283
1284                 flags &= ~VXLAN_HF_RCO;
1285                 vni &= VXLAN_VNI_MASK;
1286         }
1287
1288         if (vxlan_collect_metadata(vs)) {
1289                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1290                                          cpu_to_be64(vni >> 8), sizeof(*md));
1291
1292                 if (!tun_dst)
1293                         goto drop;
1294
1295                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1296         } else {
1297                 memset(md, 0, sizeof(*md));
1298         }
1299
1300         /* For backwards compatibility, only allow reserved fields to be
1301          * used by VXLAN extensions if explicitly requested.
1302          */
1303         if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1304                 struct vxlanhdr_gbp *gbp;
1305
1306                 gbp = (struct vxlanhdr_gbp *)vxh;
1307                 md->gbp = ntohs(gbp->policy_id);
1308
1309                 if (tun_dst)
1310                         tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1311
1312                 if (gbp->dont_learn)
1313                         md->gbp |= VXLAN_GBP_DONT_LEARN;
1314
1315                 if (gbp->policy_applied)
1316                         md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1317
1318                 flags &= ~VXLAN_GBP_USED_BITS;
1319         }
1320
1321         if (flags || vni & ~VXLAN_VNI_MASK) {
1322                 /* If there are any unprocessed flags remaining treat
1323                  * this as a malformed packet. This behavior diverges from
1324                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1325                  * in reserved fields are to be ignored. The approach here
1326                  * maintains compatibility with previous stack code, and also
1327                  * is more robust and provides a little more security in
1328                  * adding extensions to VXLAN.
1329                  */
1330
1331                 goto bad_flags;
1332         }
1333
1334         vxlan_rcv(vs, skb, md, vni >> 8, tun_dst);
1335         return 0;
1336
1337 drop:
1338         /* Consume bad packet */
1339         kfree_skb(skb);
1340         return 0;
1341
1342 bad_flags:
1343         netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1344                    ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1345
1346 error:
1347         if (tun_dst)
1348                 dst_release((struct dst_entry *)tun_dst);
1349
1350         /* Return non vxlan pkt */
1351         return 1;
1352 }
1353
1354 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1355 {
1356         struct vxlan_dev *vxlan = netdev_priv(dev);
1357         struct arphdr *parp;
1358         u8 *arpptr, *sha;
1359         __be32 sip, tip;
1360         struct neighbour *n;
1361
1362         if (dev->flags & IFF_NOARP)
1363                 goto out;
1364
1365         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1366                 dev->stats.tx_dropped++;
1367                 goto out;
1368         }
1369         parp = arp_hdr(skb);
1370
1371         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1372              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1373             parp->ar_pro != htons(ETH_P_IP) ||
1374             parp->ar_op != htons(ARPOP_REQUEST) ||
1375             parp->ar_hln != dev->addr_len ||
1376             parp->ar_pln != 4)
1377                 goto out;
1378         arpptr = (u8 *)parp + sizeof(struct arphdr);
1379         sha = arpptr;
1380         arpptr += dev->addr_len;        /* sha */
1381         memcpy(&sip, arpptr, sizeof(sip));
1382         arpptr += sizeof(sip);
1383         arpptr += dev->addr_len;        /* tha */
1384         memcpy(&tip, arpptr, sizeof(tip));
1385
1386         if (ipv4_is_loopback(tip) ||
1387             ipv4_is_multicast(tip))
1388                 goto out;
1389
1390         n = neigh_lookup(&arp_tbl, &tip, dev);
1391
1392         if (n) {
1393                 struct vxlan_fdb *f;
1394                 struct sk_buff  *reply;
1395
1396                 if (!(n->nud_state & NUD_CONNECTED)) {
1397                         neigh_release(n);
1398                         goto out;
1399                 }
1400
1401                 f = vxlan_find_mac(vxlan, n->ha);
1402                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1403                         /* bridge-local neighbor */
1404                         neigh_release(n);
1405                         goto out;
1406                 }
1407
1408                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1409                                 n->ha, sha);
1410
1411                 neigh_release(n);
1412
1413                 if (reply == NULL)
1414                         goto out;
1415
1416                 skb_reset_mac_header(reply);
1417                 __skb_pull(reply, skb_network_offset(reply));
1418                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1419                 reply->pkt_type = PACKET_HOST;
1420
1421                 if (netif_rx_ni(reply) == NET_RX_DROP)
1422                         dev->stats.rx_dropped++;
1423         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1424                 union vxlan_addr ipa = {
1425                         .sin.sin_addr.s_addr = tip,
1426                         .sin.sin_family = AF_INET,
1427                 };
1428
1429                 vxlan_ip_miss(dev, &ipa);
1430         }
1431 out:
1432         consume_skb(skb);
1433         return NETDEV_TX_OK;
1434 }
1435
1436 #if IS_ENABLED(CONFIG_IPV6)
1437 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1438         struct neighbour *n, bool isrouter)
1439 {
1440         struct net_device *dev = request->dev;
1441         struct sk_buff *reply;
1442         struct nd_msg *ns, *na;
1443         struct ipv6hdr *pip6;
1444         u8 *daddr;
1445         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1446         int ns_olen;
1447         int i, len;
1448
1449         if (dev == NULL)
1450                 return NULL;
1451
1452         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1453                 sizeof(*na) + na_olen + dev->needed_tailroom;
1454         reply = alloc_skb(len, GFP_ATOMIC);
1455         if (reply == NULL)
1456                 return NULL;
1457
1458         reply->protocol = htons(ETH_P_IPV6);
1459         reply->dev = dev;
1460         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1461         skb_push(reply, sizeof(struct ethhdr));
1462         skb_set_mac_header(reply, 0);
1463
1464         ns = (struct nd_msg *)skb_transport_header(request);
1465
1466         daddr = eth_hdr(request)->h_source;
1467         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1468         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1469                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1470                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1471                         break;
1472                 }
1473         }
1474
1475         /* Ethernet header */
1476         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1477         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1478         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1479         reply->protocol = htons(ETH_P_IPV6);
1480
1481         skb_pull(reply, sizeof(struct ethhdr));
1482         skb_set_network_header(reply, 0);
1483         skb_put(reply, sizeof(struct ipv6hdr));
1484
1485         /* IPv6 header */
1486
1487         pip6 = ipv6_hdr(reply);
1488         memset(pip6, 0, sizeof(struct ipv6hdr));
1489         pip6->version = 6;
1490         pip6->priority = ipv6_hdr(request)->priority;
1491         pip6->nexthdr = IPPROTO_ICMPV6;
1492         pip6->hop_limit = 255;
1493         pip6->daddr = ipv6_hdr(request)->saddr;
1494         pip6->saddr = *(struct in6_addr *)n->primary_key;
1495
1496         skb_pull(reply, sizeof(struct ipv6hdr));
1497         skb_set_transport_header(reply, 0);
1498
1499         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1500
1501         /* Neighbor Advertisement */
1502         memset(na, 0, sizeof(*na)+na_olen);
1503         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1504         na->icmph.icmp6_router = isrouter;
1505         na->icmph.icmp6_override = 1;
1506         na->icmph.icmp6_solicited = 1;
1507         na->target = ns->target;
1508         ether_addr_copy(&na->opt[2], n->ha);
1509         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1510         na->opt[1] = na_olen >> 3;
1511
1512         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1513                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1514                 csum_partial(na, sizeof(*na)+na_olen, 0));
1515
1516         pip6->payload_len = htons(sizeof(*na)+na_olen);
1517
1518         skb_push(reply, sizeof(struct ipv6hdr));
1519
1520         reply->ip_summed = CHECKSUM_UNNECESSARY;
1521
1522         return reply;
1523 }
1524
1525 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1526 {
1527         struct vxlan_dev *vxlan = netdev_priv(dev);
1528         struct nd_msg *msg;
1529         const struct ipv6hdr *iphdr;
1530         const struct in6_addr *saddr, *daddr;
1531         struct neighbour *n;
1532         struct inet6_dev *in6_dev;
1533
1534         in6_dev = __in6_dev_get(dev);
1535         if (!in6_dev)
1536                 goto out;
1537
1538         iphdr = ipv6_hdr(skb);
1539         saddr = &iphdr->saddr;
1540         daddr = &iphdr->daddr;
1541
1542         msg = (struct nd_msg *)skb_transport_header(skb);
1543         if (msg->icmph.icmp6_code != 0 ||
1544             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1545                 goto out;
1546
1547         if (ipv6_addr_loopback(daddr) ||
1548             ipv6_addr_is_multicast(&msg->target))
1549                 goto out;
1550
1551         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1552
1553         if (n) {
1554                 struct vxlan_fdb *f;
1555                 struct sk_buff *reply;
1556
1557                 if (!(n->nud_state & NUD_CONNECTED)) {
1558                         neigh_release(n);
1559                         goto out;
1560                 }
1561
1562                 f = vxlan_find_mac(vxlan, n->ha);
1563                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1564                         /* bridge-local neighbor */
1565                         neigh_release(n);
1566                         goto out;
1567                 }
1568
1569                 reply = vxlan_na_create(skb, n,
1570                                         !!(f ? f->flags & NTF_ROUTER : 0));
1571
1572                 neigh_release(n);
1573
1574                 if (reply == NULL)
1575                         goto out;
1576
1577                 if (netif_rx_ni(reply) == NET_RX_DROP)
1578                         dev->stats.rx_dropped++;
1579
1580         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1581                 union vxlan_addr ipa = {
1582                         .sin6.sin6_addr = msg->target,
1583                         .sin6.sin6_family = AF_INET6,
1584                 };
1585
1586                 vxlan_ip_miss(dev, &ipa);
1587         }
1588
1589 out:
1590         consume_skb(skb);
1591         return NETDEV_TX_OK;
1592 }
1593 #endif
1594
1595 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1596 {
1597         struct vxlan_dev *vxlan = netdev_priv(dev);
1598         struct neighbour *n;
1599
1600         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1601                 return false;
1602
1603         n = NULL;
1604         switch (ntohs(eth_hdr(skb)->h_proto)) {
1605         case ETH_P_IP:
1606         {
1607                 struct iphdr *pip;
1608
1609                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1610                         return false;
1611                 pip = ip_hdr(skb);
1612                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1613                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1614                         union vxlan_addr ipa = {
1615                                 .sin.sin_addr.s_addr = pip->daddr,
1616                                 .sin.sin_family = AF_INET,
1617                         };
1618
1619                         vxlan_ip_miss(dev, &ipa);
1620                         return false;
1621                 }
1622
1623                 break;
1624         }
1625 #if IS_ENABLED(CONFIG_IPV6)
1626         case ETH_P_IPV6:
1627         {
1628                 struct ipv6hdr *pip6;
1629
1630                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1631                         return false;
1632                 pip6 = ipv6_hdr(skb);
1633                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1634                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1635                         union vxlan_addr ipa = {
1636                                 .sin6.sin6_addr = pip6->daddr,
1637                                 .sin6.sin6_family = AF_INET6,
1638                         };
1639
1640                         vxlan_ip_miss(dev, &ipa);
1641                         return false;
1642                 }
1643
1644                 break;
1645         }
1646 #endif
1647         default:
1648                 return false;
1649         }
1650
1651         if (n) {
1652                 bool diff;
1653
1654                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1655                 if (diff) {
1656                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1657                                 dev->addr_len);
1658                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1659                 }
1660                 neigh_release(n);
1661                 return diff;
1662         }
1663
1664         return false;
1665 }
1666
1667 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1668                                 struct vxlan_metadata *md)
1669 {
1670         struct vxlanhdr_gbp *gbp;
1671
1672         if (!md->gbp)
1673                 return;
1674
1675         gbp = (struct vxlanhdr_gbp *)vxh;
1676         vxh->vx_flags |= htonl(VXLAN_HF_GBP);
1677
1678         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1679                 gbp->dont_learn = 1;
1680
1681         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1682                 gbp->policy_applied = 1;
1683
1684         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1685 }
1686
1687 #if IS_ENABLED(CONFIG_IPV6)
1688 static int vxlan6_xmit_skb(struct dst_entry *dst, struct sock *sk,
1689                            struct sk_buff *skb,
1690                            struct net_device *dev, struct in6_addr *saddr,
1691                            struct in6_addr *daddr, __u8 prio, __u8 ttl,
1692                            __be16 src_port, __be16 dst_port, __be32 vni,
1693                            struct vxlan_metadata *md, bool xnet, u32 vxflags)
1694 {
1695         struct vxlanhdr *vxh;
1696         int min_headroom;
1697         int err;
1698         bool udp_sum = !(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX);
1699         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1700         u16 hdrlen = sizeof(struct vxlanhdr);
1701
1702         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1703             skb->ip_summed == CHECKSUM_PARTIAL) {
1704                 int csum_start = skb_checksum_start_offset(skb);
1705
1706                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1707                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1708                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1709                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1710                         udp_sum = false;
1711                         type |= SKB_GSO_TUNNEL_REMCSUM;
1712                 }
1713         }
1714
1715         skb_scrub_packet(skb, xnet);
1716
1717         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1718                         + VXLAN_HLEN + sizeof(struct ipv6hdr)
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                 goto err;
1726         }
1727
1728         skb = vlan_hwaccel_push_inside(skb);
1729         if (WARN_ON(!skb)) {
1730                 err = -ENOMEM;
1731                 goto err;
1732         }
1733
1734         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1735         if (IS_ERR(skb)) {
1736                 err = -EINVAL;
1737                 goto err;
1738         }
1739
1740         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1741         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1742         vxh->vx_vni = vni;
1743
1744         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1745                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1746                            VXLAN_RCO_SHIFT;
1747
1748                 if (skb->csum_offset == offsetof(struct udphdr, check))
1749                         data |= VXLAN_RCO_UDP;
1750
1751                 vxh->vx_vni |= htonl(data);
1752                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1753
1754                 if (!skb_is_gso(skb)) {
1755                         skb->ip_summed = CHECKSUM_NONE;
1756                         skb->encapsulation = 0;
1757                 }
1758         }
1759
1760         if (vxflags & VXLAN_F_GBP)
1761                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1762
1763         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1764
1765         udp_tunnel6_xmit_skb(dst, sk, skb, dev, saddr, daddr, prio,
1766                              ttl, src_port, dst_port,
1767                              !!(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX));
1768         return 0;
1769 err:
1770         dst_release(dst);
1771         return err;
1772 }
1773 #endif
1774
1775 static int vxlan_xmit_skb(struct rtable *rt, struct sock *sk, struct sk_buff *skb,
1776                           __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1777                           __be16 src_port, __be16 dst_port, __be32 vni,
1778                           struct vxlan_metadata *md, bool xnet, u32 vxflags)
1779 {
1780         struct vxlanhdr *vxh;
1781         int min_headroom;
1782         int err;
1783         bool udp_sum = !!(vxflags & VXLAN_F_UDP_CSUM);
1784         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1785         u16 hdrlen = sizeof(struct vxlanhdr);
1786
1787         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1788             skb->ip_summed == CHECKSUM_PARTIAL) {
1789                 int csum_start = skb_checksum_start_offset(skb);
1790
1791                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1792                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1793                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1794                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1795                         udp_sum = false;
1796                         type |= SKB_GSO_TUNNEL_REMCSUM;
1797                 }
1798         }
1799
1800         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1801                         + VXLAN_HLEN + sizeof(struct iphdr)
1802                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1803
1804         /* Need space for new headers (invalidates iph ptr) */
1805         err = skb_cow_head(skb, min_headroom);
1806         if (unlikely(err)) {
1807                 kfree_skb(skb);
1808                 return err;
1809         }
1810
1811         skb = vlan_hwaccel_push_inside(skb);
1812         if (WARN_ON(!skb))
1813                 return -ENOMEM;
1814
1815         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1816         if (IS_ERR(skb))
1817                 return PTR_ERR(skb);
1818
1819         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1820         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1821         vxh->vx_vni = vni;
1822
1823         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1824                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1825                            VXLAN_RCO_SHIFT;
1826
1827                 if (skb->csum_offset == offsetof(struct udphdr, check))
1828                         data |= VXLAN_RCO_UDP;
1829
1830                 vxh->vx_vni |= htonl(data);
1831                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1832
1833                 if (!skb_is_gso(skb)) {
1834                         skb->ip_summed = CHECKSUM_NONE;
1835                         skb->encapsulation = 0;
1836                 }
1837         }
1838
1839         if (vxflags & VXLAN_F_GBP)
1840                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1841
1842         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1843
1844         return udp_tunnel_xmit_skb(rt, sk, skb, src, dst, tos,
1845                                    ttl, df, src_port, dst_port, xnet,
1846                                    !(vxflags & VXLAN_F_UDP_CSUM));
1847 }
1848
1849 #if IS_ENABLED(CONFIG_IPV6)
1850 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1851                                           struct sk_buff *skb, int oif,
1852                                           const struct in6_addr *daddr,
1853                                           struct in6_addr *saddr)
1854 {
1855         struct dst_entry *ndst;
1856         struct flowi6 fl6;
1857         int err;
1858
1859         memset(&fl6, 0, sizeof(fl6));
1860         fl6.flowi6_oif = oif;
1861         fl6.daddr = *daddr;
1862         fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
1863         fl6.flowi6_mark = skb->mark;
1864         fl6.flowi6_proto = IPPROTO_UDP;
1865
1866         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1867                                          vxlan->vn6_sock->sock->sk,
1868                                          &ndst, &fl6);
1869         if (err < 0)
1870                 return ERR_PTR(err);
1871
1872         *saddr = fl6.saddr;
1873         return ndst;
1874 }
1875 #endif
1876
1877 /* Bypass encapsulation if the destination is local */
1878 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1879                                struct vxlan_dev *dst_vxlan)
1880 {
1881         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1882         union vxlan_addr loopback;
1883         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1884         struct net_device *dev = skb->dev;
1885         int len = skb->len;
1886
1887         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1888         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1889         skb->pkt_type = PACKET_HOST;
1890         skb->encapsulation = 0;
1891         skb->dev = dst_vxlan->dev;
1892         __skb_pull(skb, skb_network_offset(skb));
1893
1894         if (remote_ip->sa.sa_family == AF_INET) {
1895                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1896                 loopback.sa.sa_family =  AF_INET;
1897 #if IS_ENABLED(CONFIG_IPV6)
1898         } else {
1899                 loopback.sin6.sin6_addr = in6addr_loopback;
1900                 loopback.sa.sa_family =  AF_INET6;
1901 #endif
1902         }
1903
1904         if (dst_vxlan->flags & VXLAN_F_LEARN)
1905                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1906
1907         u64_stats_update_begin(&tx_stats->syncp);
1908         tx_stats->tx_packets++;
1909         tx_stats->tx_bytes += len;
1910         u64_stats_update_end(&tx_stats->syncp);
1911
1912         if (netif_rx(skb) == NET_RX_SUCCESS) {
1913                 u64_stats_update_begin(&rx_stats->syncp);
1914                 rx_stats->rx_packets++;
1915                 rx_stats->rx_bytes += len;
1916                 u64_stats_update_end(&rx_stats->syncp);
1917         } else {
1918                 dev->stats.rx_dropped++;
1919         }
1920 }
1921
1922 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1923                            struct vxlan_rdst *rdst, bool did_rsc)
1924 {
1925         struct ip_tunnel_info *info;
1926         struct vxlan_dev *vxlan = netdev_priv(dev);
1927         struct sock *sk;
1928         struct rtable *rt = NULL;
1929         const struct iphdr *old_iph;
1930         struct flowi4 fl4;
1931         union vxlan_addr *dst;
1932         union vxlan_addr remote_ip;
1933         struct vxlan_metadata _md;
1934         struct vxlan_metadata *md = &_md;
1935         __be16 src_port = 0, dst_port;
1936         u32 vni;
1937         __be16 df = 0;
1938         __u8 tos, ttl;
1939         int err;
1940         u32 flags = vxlan->flags;
1941
1942         info = skb_tunnel_info(skb);
1943
1944         if (rdst) {
1945                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1946                 vni = rdst->remote_vni;
1947                 dst = &rdst->remote_ip;
1948         } else {
1949                 if (!info) {
1950                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1951                                   dev->name);
1952                         goto drop;
1953                 }
1954                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1955                 vni = be64_to_cpu(info->key.tun_id);
1956                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1957                 if (remote_ip.sa.sa_family == AF_INET)
1958                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1959                 else
1960                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1961                 dst = &remote_ip;
1962         }
1963
1964         if (vxlan_addr_any(dst)) {
1965                 if (did_rsc) {
1966                         /* short-circuited back to local bridge */
1967                         vxlan_encap_bypass(skb, vxlan, vxlan);
1968                         return;
1969                 }
1970                 goto drop;
1971         }
1972
1973         old_iph = ip_hdr(skb);
1974
1975         ttl = vxlan->cfg.ttl;
1976         if (!ttl && vxlan_addr_multicast(dst))
1977                 ttl = 1;
1978
1979         tos = vxlan->cfg.tos;
1980         if (tos == 1)
1981                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1982
1983         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
1984                                      vxlan->cfg.port_max, true);
1985
1986         if (info) {
1987                 ttl = info->key.ttl;
1988                 tos = info->key.tos;
1989
1990                 if (info->options_len)
1991                         md = ip_tunnel_info_opts(info);
1992         } else {
1993                 md->gbp = skb->mark;
1994         }
1995
1996         if (dst->sa.sa_family == AF_INET) {
1997                 if (!vxlan->vn4_sock)
1998                         goto drop;
1999                 sk = vxlan->vn4_sock->sock->sk;
2000
2001                 if (info) {
2002                         if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2003                                 df = htons(IP_DF);
2004
2005                         if (info->key.tun_flags & TUNNEL_CSUM)
2006                                 flags |= VXLAN_F_UDP_CSUM;
2007                         else
2008                                 flags &= ~VXLAN_F_UDP_CSUM;
2009                 }
2010
2011                 memset(&fl4, 0, sizeof(fl4));
2012                 fl4.flowi4_oif = rdst ? rdst->remote_ifindex : 0;
2013                 fl4.flowi4_tos = RT_TOS(tos);
2014                 fl4.flowi4_mark = skb->mark;
2015                 fl4.flowi4_proto = IPPROTO_UDP;
2016                 fl4.daddr = dst->sin.sin_addr.s_addr;
2017                 fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
2018
2019                 rt = ip_route_output_key(vxlan->net, &fl4);
2020                 if (IS_ERR(rt)) {
2021                         netdev_dbg(dev, "no route to %pI4\n",
2022                                    &dst->sin.sin_addr.s_addr);
2023                         dev->stats.tx_carrier_errors++;
2024                         goto tx_error;
2025                 }
2026
2027                 if (rt->dst.dev == dev) {
2028                         netdev_dbg(dev, "circular route to %pI4\n",
2029                                    &dst->sin.sin_addr.s_addr);
2030                         dev->stats.collisions++;
2031                         goto rt_tx_error;
2032                 }
2033
2034                 /* Bypass encapsulation if the destination is local */
2035                 if (rt->rt_flags & RTCF_LOCAL &&
2036                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2037                         struct vxlan_dev *dst_vxlan;
2038
2039                         ip_rt_put(rt);
2040                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2041                                                    dst->sa.sa_family, dst_port,
2042                                                    vxlan->flags);
2043                         if (!dst_vxlan)
2044                                 goto tx_error;
2045                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2046                         return;
2047                 }
2048
2049                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2050                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2051                 err = vxlan_xmit_skb(rt, sk, skb, fl4.saddr,
2052                                      dst->sin.sin_addr.s_addr, tos, ttl, df,
2053                                      src_port, dst_port, htonl(vni << 8), md,
2054                                      !net_eq(vxlan->net, dev_net(vxlan->dev)),
2055                                      flags);
2056                 if (err < 0) {
2057                         /* skb is already freed. */
2058                         skb = NULL;
2059                         goto rt_tx_error;
2060                 }
2061
2062                 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
2063 #if IS_ENABLED(CONFIG_IPV6)
2064         } else {
2065                 struct dst_entry *ndst;
2066                 struct in6_addr saddr;
2067                 u32 rt6i_flags;
2068
2069                 if (!vxlan->vn6_sock)
2070                         goto drop;
2071                 sk = vxlan->vn6_sock->sock->sk;
2072
2073                 ndst = vxlan6_get_route(vxlan, skb,
2074                                         rdst ? rdst->remote_ifindex : 0,
2075                                         &dst->sin6.sin6_addr, &saddr);
2076                 if (IS_ERR(ndst)) {
2077                         netdev_dbg(dev, "no route to %pI6\n",
2078                                    &dst->sin6.sin6_addr);
2079                         dev->stats.tx_carrier_errors++;
2080                         goto tx_error;
2081                 }
2082
2083                 if (ndst->dev == dev) {
2084                         netdev_dbg(dev, "circular route to %pI6\n",
2085                                    &dst->sin6.sin6_addr);
2086                         dst_release(ndst);
2087                         dev->stats.collisions++;
2088                         goto tx_error;
2089                 }
2090
2091                 /* Bypass encapsulation if the destination is local */
2092                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2093                 if (rt6i_flags & RTF_LOCAL &&
2094                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2095                         struct vxlan_dev *dst_vxlan;
2096
2097                         dst_release(ndst);
2098                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2099                                                    dst->sa.sa_family, dst_port,
2100                                                    vxlan->flags);
2101                         if (!dst_vxlan)
2102                                 goto tx_error;
2103                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2104                         return;
2105                 }
2106
2107                 if (info) {
2108                         if (info->key.tun_flags & TUNNEL_CSUM)
2109                                 flags &= ~VXLAN_F_UDP_ZERO_CSUM6_TX;
2110                         else
2111                                 flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2112                 }
2113
2114                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2115                 err = vxlan6_xmit_skb(ndst, sk, skb, dev, &saddr, &dst->sin6.sin6_addr,
2116                                       0, ttl, src_port, dst_port, htonl(vni << 8), md,
2117                                       !net_eq(vxlan->net, dev_net(vxlan->dev)),
2118                                       flags);
2119 #endif
2120         }
2121
2122         return;
2123
2124 drop:
2125         dev->stats.tx_dropped++;
2126         goto tx_free;
2127
2128 rt_tx_error:
2129         ip_rt_put(rt);
2130 tx_error:
2131         dev->stats.tx_errors++;
2132 tx_free:
2133         dev_kfree_skb(skb);
2134 }
2135
2136 /* Transmit local packets over Vxlan
2137  *
2138  * Outer IP header inherits ECN and DF from inner header.
2139  * Outer UDP destination is the VXLAN assigned port.
2140  *           source port is based on hash of flow
2141  */
2142 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2143 {
2144         struct vxlan_dev *vxlan = netdev_priv(dev);
2145         const struct ip_tunnel_info *info;
2146         struct ethhdr *eth;
2147         bool did_rsc = false;
2148         struct vxlan_rdst *rdst, *fdst = NULL;
2149         struct vxlan_fdb *f;
2150
2151         info = skb_tunnel_info(skb);
2152
2153         skb_reset_mac_header(skb);
2154         eth = eth_hdr(skb);
2155
2156         if ((vxlan->flags & VXLAN_F_PROXY)) {
2157                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2158                         return arp_reduce(dev, skb);
2159 #if IS_ENABLED(CONFIG_IPV6)
2160                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2161                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2162                                        + sizeof(struct nd_msg)) &&
2163                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2164                                 struct nd_msg *msg;
2165
2166                                 msg = (struct nd_msg *)skb_transport_header(skb);
2167                                 if (msg->icmph.icmp6_code == 0 &&
2168                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2169                                         return neigh_reduce(dev, skb);
2170                 }
2171                 eth = eth_hdr(skb);
2172 #endif
2173         }
2174
2175         if (vxlan->flags & VXLAN_F_COLLECT_METADATA &&
2176             info && info->mode & IP_TUNNEL_INFO_TX) {
2177                 vxlan_xmit_one(skb, dev, NULL, false);
2178                 return NETDEV_TX_OK;
2179         }
2180
2181         f = vxlan_find_mac(vxlan, eth->h_dest);
2182         did_rsc = false;
2183
2184         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2185             (ntohs(eth->h_proto) == ETH_P_IP ||
2186              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2187                 did_rsc = route_shortcircuit(dev, skb);
2188                 if (did_rsc)
2189                         f = vxlan_find_mac(vxlan, eth->h_dest);
2190         }
2191
2192         if (f == NULL) {
2193                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2194                 if (f == NULL) {
2195                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2196                             !is_multicast_ether_addr(eth->h_dest))
2197                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2198
2199                         dev->stats.tx_dropped++;
2200                         kfree_skb(skb);
2201                         return NETDEV_TX_OK;
2202                 }
2203         }
2204
2205         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2206                 struct sk_buff *skb1;
2207
2208                 if (!fdst) {
2209                         fdst = rdst;
2210                         continue;
2211                 }
2212                 skb1 = skb_clone(skb, GFP_ATOMIC);
2213                 if (skb1)
2214                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2215         }
2216
2217         if (fdst)
2218                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2219         else
2220                 kfree_skb(skb);
2221         return NETDEV_TX_OK;
2222 }
2223
2224 /* Walk the forwarding table and purge stale entries */
2225 static void vxlan_cleanup(unsigned long arg)
2226 {
2227         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2228         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2229         unsigned int h;
2230
2231         if (!netif_running(vxlan->dev))
2232                 return;
2233
2234         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2235                 struct hlist_node *p, *n;
2236
2237                 spin_lock_bh(&vxlan->hash_lock);
2238                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2239                         struct vxlan_fdb *f
2240                                 = container_of(p, struct vxlan_fdb, hlist);
2241                         unsigned long timeout;
2242
2243                         if (f->state & NUD_PERMANENT)
2244                                 continue;
2245
2246                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2247                         if (time_before_eq(timeout, jiffies)) {
2248                                 netdev_dbg(vxlan->dev,
2249                                            "garbage collect %pM\n",
2250                                            f->eth_addr);
2251                                 f->state = NUD_STALE;
2252                                 vxlan_fdb_destroy(vxlan, f);
2253                         } else if (time_before(timeout, next_timer))
2254                                 next_timer = timeout;
2255                 }
2256                 spin_unlock_bh(&vxlan->hash_lock);
2257         }
2258
2259         mod_timer(&vxlan->age_timer, next_timer);
2260 }
2261
2262 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2263 {
2264         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2265         __u32 vni = vxlan->default_dst.remote_vni;
2266
2267         spin_lock(&vn->sock_lock);
2268         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2269         spin_unlock(&vn->sock_lock);
2270 }
2271
2272 /* Setup stats when device is created */
2273 static int vxlan_init(struct net_device *dev)
2274 {
2275         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2276         if (!dev->tstats)
2277                 return -ENOMEM;
2278
2279         return 0;
2280 }
2281
2282 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2283 {
2284         struct vxlan_fdb *f;
2285
2286         spin_lock_bh(&vxlan->hash_lock);
2287         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2288         if (f)
2289                 vxlan_fdb_destroy(vxlan, f);
2290         spin_unlock_bh(&vxlan->hash_lock);
2291 }
2292
2293 static void vxlan_uninit(struct net_device *dev)
2294 {
2295         struct vxlan_dev *vxlan = netdev_priv(dev);
2296
2297         vxlan_fdb_delete_default(vxlan);
2298
2299         free_percpu(dev->tstats);
2300 }
2301
2302 /* Start ageing timer and join group when device is brought up */
2303 static int vxlan_open(struct net_device *dev)
2304 {
2305         struct vxlan_dev *vxlan = netdev_priv(dev);
2306         int ret;
2307
2308         ret = vxlan_sock_add(vxlan);
2309         if (ret < 0)
2310                 return ret;
2311
2312         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2313                 ret = vxlan_igmp_join(vxlan);
2314                 if (ret == -EADDRINUSE)
2315                         ret = 0;
2316                 if (ret) {
2317                         vxlan_sock_release(vxlan);
2318                         return ret;
2319                 }
2320         }
2321
2322         if (vxlan->cfg.age_interval)
2323                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2324
2325         return ret;
2326 }
2327
2328 /* Purge the forwarding table */
2329 static void vxlan_flush(struct vxlan_dev *vxlan)
2330 {
2331         unsigned int h;
2332
2333         spin_lock_bh(&vxlan->hash_lock);
2334         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2335                 struct hlist_node *p, *n;
2336                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2337                         struct vxlan_fdb *f
2338                                 = container_of(p, struct vxlan_fdb, hlist);
2339                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2340                         if (!is_zero_ether_addr(f->eth_addr))
2341                                 vxlan_fdb_destroy(vxlan, f);
2342                 }
2343         }
2344         spin_unlock_bh(&vxlan->hash_lock);
2345 }
2346
2347 /* Cleanup timer and forwarding table on shutdown */
2348 static int vxlan_stop(struct net_device *dev)
2349 {
2350         struct vxlan_dev *vxlan = netdev_priv(dev);
2351         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2352         int ret = 0;
2353
2354         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2355             !vxlan_group_used(vn, vxlan))
2356                 ret = vxlan_igmp_leave(vxlan);
2357
2358         del_timer_sync(&vxlan->age_timer);
2359
2360         vxlan_flush(vxlan);
2361         vxlan_sock_release(vxlan);
2362
2363         return ret;
2364 }
2365
2366 /* Stub, nothing needs to be done. */
2367 static void vxlan_set_multicast_list(struct net_device *dev)
2368 {
2369 }
2370
2371 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2372 {
2373         struct vxlan_dev *vxlan = netdev_priv(dev);
2374         struct vxlan_rdst *dst = &vxlan->default_dst;
2375         struct net_device *lowerdev;
2376         int max_mtu;
2377
2378         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2379         if (lowerdev == NULL)
2380                 return eth_change_mtu(dev, new_mtu);
2381
2382         if (dst->remote_ip.sa.sa_family == AF_INET6)
2383                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2384         else
2385                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2386
2387         if (new_mtu < 68 || new_mtu > max_mtu)
2388                 return -EINVAL;
2389
2390         dev->mtu = new_mtu;
2391         return 0;
2392 }
2393
2394 static int egress_ipv4_tun_info(struct net_device *dev, struct sk_buff *skb,
2395                                 struct ip_tunnel_info *info,
2396                                 __be16 sport, __be16 dport)
2397 {
2398         struct vxlan_dev *vxlan = netdev_priv(dev);
2399         struct rtable *rt;
2400         struct flowi4 fl4;
2401
2402         memset(&fl4, 0, sizeof(fl4));
2403         fl4.flowi4_tos = RT_TOS(info->key.tos);
2404         fl4.flowi4_mark = skb->mark;
2405         fl4.flowi4_proto = IPPROTO_UDP;
2406         fl4.daddr = info->key.u.ipv4.dst;
2407
2408         rt = ip_route_output_key(vxlan->net, &fl4);
2409         if (IS_ERR(rt))
2410                 return PTR_ERR(rt);
2411         ip_rt_put(rt);
2412
2413         info->key.u.ipv4.src = fl4.saddr;
2414         info->key.tp_src = sport;
2415         info->key.tp_dst = dport;
2416         return 0;
2417 }
2418
2419 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2420 {
2421         struct vxlan_dev *vxlan = netdev_priv(dev);
2422         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2423         __be16 sport, dport;
2424
2425         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2426                                   vxlan->cfg.port_max, true);
2427         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2428
2429         if (ip_tunnel_info_af(info) == AF_INET) {
2430                 if (!vxlan->vn4_sock)
2431                         return -EINVAL;
2432                 return egress_ipv4_tun_info(dev, skb, info, sport, dport);
2433         } else {
2434 #if IS_ENABLED(CONFIG_IPV6)
2435                 struct dst_entry *ndst;
2436
2437                 if (!vxlan->vn6_sock)
2438                         return -EINVAL;
2439                 ndst = vxlan6_get_route(vxlan, skb, 0,
2440                                         &info->key.u.ipv6.dst,
2441                                         &info->key.u.ipv6.src);
2442                 if (IS_ERR(ndst))
2443                         return PTR_ERR(ndst);
2444                 dst_release(ndst);
2445
2446                 info->key.tp_src = sport;
2447                 info->key.tp_dst = dport;
2448 #else /* !CONFIG_IPV6 */
2449                 return -EPFNOSUPPORT;
2450 #endif
2451         }
2452         return 0;
2453 }
2454
2455 static const struct net_device_ops vxlan_netdev_ops = {
2456         .ndo_init               = vxlan_init,
2457         .ndo_uninit             = vxlan_uninit,
2458         .ndo_open               = vxlan_open,
2459         .ndo_stop               = vxlan_stop,
2460         .ndo_start_xmit         = vxlan_xmit,
2461         .ndo_get_stats64        = ip_tunnel_get_stats64,
2462         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2463         .ndo_change_mtu         = vxlan_change_mtu,
2464         .ndo_validate_addr      = eth_validate_addr,
2465         .ndo_set_mac_address    = eth_mac_addr,
2466         .ndo_fdb_add            = vxlan_fdb_add,
2467         .ndo_fdb_del            = vxlan_fdb_delete,
2468         .ndo_fdb_dump           = vxlan_fdb_dump,
2469         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2470 };
2471
2472 /* Info for udev, that this is a virtual tunnel endpoint */
2473 static struct device_type vxlan_type = {
2474         .name = "vxlan",
2475 };
2476
2477 /* Calls the ndo_add_vxlan_port of the caller in order to
2478  * supply the listening VXLAN udp ports. Callers are expected
2479  * to implement the ndo_add_vxlan_port.
2480  */
2481 void vxlan_get_rx_port(struct net_device *dev)
2482 {
2483         struct vxlan_sock *vs;
2484         struct net *net = dev_net(dev);
2485         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2486         sa_family_t sa_family;
2487         __be16 port;
2488         unsigned int i;
2489
2490         spin_lock(&vn->sock_lock);
2491         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2492                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2493                         port = inet_sk(vs->sock->sk)->inet_sport;
2494                         sa_family = vxlan_get_sk_family(vs);
2495                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2496                                                             port);
2497                 }
2498         }
2499         spin_unlock(&vn->sock_lock);
2500 }
2501 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2502
2503 /* Initialize the device structure. */
2504 static void vxlan_setup(struct net_device *dev)
2505 {
2506         struct vxlan_dev *vxlan = netdev_priv(dev);
2507         unsigned int h;
2508
2509         eth_hw_addr_random(dev);
2510         ether_setup(dev);
2511
2512         dev->netdev_ops = &vxlan_netdev_ops;
2513         dev->destructor = free_netdev;
2514         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2515
2516         dev->features   |= NETIF_F_LLTX;
2517         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2518         dev->features   |= NETIF_F_RXCSUM;
2519         dev->features   |= NETIF_F_GSO_SOFTWARE;
2520
2521         dev->vlan_features = dev->features;
2522         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2523         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2524         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2525         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2526         netif_keep_dst(dev);
2527         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
2528
2529         INIT_LIST_HEAD(&vxlan->next);
2530         spin_lock_init(&vxlan->hash_lock);
2531
2532         init_timer_deferrable(&vxlan->age_timer);
2533         vxlan->age_timer.function = vxlan_cleanup;
2534         vxlan->age_timer.data = (unsigned long) vxlan;
2535
2536         vxlan->cfg.dst_port = htons(vxlan_port);
2537
2538         vxlan->dev = dev;
2539
2540         gro_cells_init(&vxlan->gro_cells, dev);
2541
2542         for (h = 0; h < FDB_HASH_SIZE; ++h)
2543                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2544 }
2545
2546 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2547         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2548         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2549         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2550         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2551         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2552         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2553         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2554         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2555         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2556         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2557         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2558         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2559         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2560         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2561         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2562         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2563         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2564         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2565         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2566         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2567         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2568         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2569         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2570         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2571         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2572 };
2573
2574 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2575 {
2576         if (tb[IFLA_ADDRESS]) {
2577                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2578                         pr_debug("invalid link address (not ethernet)\n");
2579                         return -EINVAL;
2580                 }
2581
2582                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2583                         pr_debug("invalid all zero ethernet address\n");
2584                         return -EADDRNOTAVAIL;
2585                 }
2586         }
2587
2588         if (!data)
2589                 return -EINVAL;
2590
2591         if (data[IFLA_VXLAN_ID]) {
2592                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2593                 if (id >= VXLAN_VID_MASK)
2594                         return -ERANGE;
2595         }
2596
2597         if (data[IFLA_VXLAN_PORT_RANGE]) {
2598                 const struct ifla_vxlan_port_range *p
2599                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2600
2601                 if (ntohs(p->high) < ntohs(p->low)) {
2602                         pr_debug("port range %u .. %u not valid\n",
2603                                  ntohs(p->low), ntohs(p->high));
2604                         return -EINVAL;
2605                 }
2606         }
2607
2608         return 0;
2609 }
2610
2611 static void vxlan_get_drvinfo(struct net_device *netdev,
2612                               struct ethtool_drvinfo *drvinfo)
2613 {
2614         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2615         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2616 }
2617
2618 static const struct ethtool_ops vxlan_ethtool_ops = {
2619         .get_drvinfo    = vxlan_get_drvinfo,
2620         .get_link       = ethtool_op_get_link,
2621 };
2622
2623 static void vxlan_del_work(struct work_struct *work)
2624 {
2625         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2626         udp_tunnel_sock_release(vs->sock);
2627         kfree_rcu(vs, rcu);
2628 }
2629
2630 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2631                                         __be16 port, u32 flags)
2632 {
2633         struct socket *sock;
2634         struct udp_port_cfg udp_conf;
2635         int err;
2636
2637         memset(&udp_conf, 0, sizeof(udp_conf));
2638
2639         if (ipv6) {
2640                 udp_conf.family = AF_INET6;
2641                 udp_conf.use_udp6_rx_checksums =
2642                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2643                 udp_conf.ipv6_v6only = 1;
2644         } else {
2645                 udp_conf.family = AF_INET;
2646         }
2647
2648         udp_conf.local_udp_port = port;
2649
2650         /* Open UDP socket */
2651         err = udp_sock_create(net, &udp_conf, &sock);
2652         if (err < 0)
2653                 return ERR_PTR(err);
2654
2655         return sock;
2656 }
2657
2658 /* Create new listen socket if needed */
2659 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2660                                               __be16 port, u32 flags)
2661 {
2662         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2663         struct vxlan_sock *vs;
2664         struct socket *sock;
2665         unsigned int h;
2666         struct udp_tunnel_sock_cfg tunnel_cfg;
2667
2668         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2669         if (!vs)
2670                 return ERR_PTR(-ENOMEM);
2671
2672         for (h = 0; h < VNI_HASH_SIZE; ++h)
2673                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2674
2675         INIT_WORK(&vs->del_work, vxlan_del_work);
2676
2677         sock = vxlan_create_sock(net, ipv6, port, flags);
2678         if (IS_ERR(sock)) {
2679                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2680                         PTR_ERR(sock));
2681                 kfree(vs);
2682                 return ERR_CAST(sock);
2683         }
2684
2685         vs->sock = sock;
2686         atomic_set(&vs->refcnt, 1);
2687         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2688
2689         /* Initialize the vxlan udp offloads structure */
2690         vs->udp_offloads.port = port;
2691         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2692         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2693
2694         spin_lock(&vn->sock_lock);
2695         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2696         vxlan_notify_add_rx_port(vs);
2697         spin_unlock(&vn->sock_lock);
2698
2699         /* Mark socket as an encapsulation socket. */
2700         tunnel_cfg.sk_user_data = vs;
2701         tunnel_cfg.encap_type = 1;
2702         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2703         tunnel_cfg.encap_destroy = NULL;
2704
2705         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2706
2707         return vs;
2708 }
2709
2710 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2711 {
2712         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2713         struct vxlan_sock *vs = NULL;
2714
2715         if (!vxlan->cfg.no_share) {
2716                 spin_lock(&vn->sock_lock);
2717                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2718                                      vxlan->cfg.dst_port, vxlan->flags);
2719                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2720                         spin_unlock(&vn->sock_lock);
2721                         return -EBUSY;
2722                 }
2723                 spin_unlock(&vn->sock_lock);
2724         }
2725         if (!vs)
2726                 vs = vxlan_socket_create(vxlan->net, ipv6,
2727                                          vxlan->cfg.dst_port, vxlan->flags);
2728         if (IS_ERR(vs))
2729                 return PTR_ERR(vs);
2730 #if IS_ENABLED(CONFIG_IPV6)
2731         if (ipv6)
2732                 vxlan->vn6_sock = vs;
2733         else
2734 #endif
2735                 vxlan->vn4_sock = vs;
2736         vxlan_vs_add_dev(vs, vxlan);
2737         return 0;
2738 }
2739
2740 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2741 {
2742         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2743         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2744         int ret = 0;
2745
2746         vxlan->vn4_sock = NULL;
2747 #if IS_ENABLED(CONFIG_IPV6)
2748         vxlan->vn6_sock = NULL;
2749         if (ipv6 || metadata)
2750                 ret = __vxlan_sock_add(vxlan, true);
2751 #endif
2752         if (!ret && (!ipv6 || metadata))
2753                 ret = __vxlan_sock_add(vxlan, false);
2754         if (ret < 0)
2755                 vxlan_sock_release(vxlan);
2756         return ret;
2757 }
2758
2759 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2760                                struct vxlan_config *conf)
2761 {
2762         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2763         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2764         struct vxlan_rdst *dst = &vxlan->default_dst;
2765         unsigned short needed_headroom = ETH_HLEN;
2766         int err;
2767         bool use_ipv6 = false;
2768         __be16 default_port = vxlan->cfg.dst_port;
2769
2770         vxlan->net = src_net;
2771
2772         dst->remote_vni = conf->vni;
2773
2774         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2775
2776         /* Unless IPv6 is explicitly requested, assume IPv4 */
2777         if (!dst->remote_ip.sa.sa_family)
2778                 dst->remote_ip.sa.sa_family = AF_INET;
2779
2780         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2781             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2782                 if (!IS_ENABLED(CONFIG_IPV6))
2783                         return -EPFNOSUPPORT;
2784                 use_ipv6 = true;
2785                 vxlan->flags |= VXLAN_F_IPV6;
2786         }
2787
2788         if (conf->remote_ifindex) {
2789                 struct net_device *lowerdev
2790                          = __dev_get_by_index(src_net, conf->remote_ifindex);
2791
2792                 dst->remote_ifindex = conf->remote_ifindex;
2793
2794                 if (!lowerdev) {
2795                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2796                         return -ENODEV;
2797                 }
2798
2799 #if IS_ENABLED(CONFIG_IPV6)
2800                 if (use_ipv6) {
2801                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2802                         if (idev && idev->cnf.disable_ipv6) {
2803                                 pr_info("IPv6 is disabled via sysctl\n");
2804                                 return -EPERM;
2805                         }
2806                 }
2807 #endif
2808
2809                 if (!conf->mtu)
2810                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2811
2812                 needed_headroom = lowerdev->hard_header_len;
2813         }
2814
2815         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2816                 needed_headroom += VXLAN6_HEADROOM;
2817         else
2818                 needed_headroom += VXLAN_HEADROOM;
2819         dev->needed_headroom = needed_headroom;
2820
2821         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2822         if (!vxlan->cfg.dst_port)
2823                 vxlan->cfg.dst_port = default_port;
2824         vxlan->flags |= conf->flags;
2825
2826         if (!vxlan->cfg.age_interval)
2827                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2828
2829         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2830                 if (tmp->cfg.vni == conf->vni &&
2831                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2832                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2833                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2834                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2835                     (vxlan->flags & VXLAN_F_RCV_FLAGS))
2836                 return -EEXIST;
2837         }
2838
2839         dev->ethtool_ops = &vxlan_ethtool_ops;
2840
2841         /* create an fdb entry for a valid default destination */
2842         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2843                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2844                                        &vxlan->default_dst.remote_ip,
2845                                        NUD_REACHABLE|NUD_PERMANENT,
2846                                        NLM_F_EXCL|NLM_F_CREATE,
2847                                        vxlan->cfg.dst_port,
2848                                        vxlan->default_dst.remote_vni,
2849                                        vxlan->default_dst.remote_ifindex,
2850                                        NTF_SELF);
2851                 if (err)
2852                         return err;
2853         }
2854
2855         err = register_netdevice(dev);
2856         if (err) {
2857                 vxlan_fdb_delete_default(vxlan);
2858                 return err;
2859         }
2860
2861         list_add(&vxlan->next, &vn->vxlan_list);
2862
2863         return 0;
2864 }
2865
2866 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2867                                     u8 name_assign_type, struct vxlan_config *conf)
2868 {
2869         struct nlattr *tb[IFLA_MAX+1];
2870         struct net_device *dev;
2871         int err;
2872
2873         memset(&tb, 0, sizeof(tb));
2874
2875         dev = rtnl_create_link(net, name, name_assign_type,
2876                                &vxlan_link_ops, tb);
2877         if (IS_ERR(dev))
2878                 return dev;
2879
2880         err = vxlan_dev_configure(net, dev, conf);
2881         if (err < 0) {
2882                 free_netdev(dev);
2883                 return ERR_PTR(err);
2884         }
2885
2886         return dev;
2887 }
2888 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2889
2890 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2891                          struct nlattr *tb[], struct nlattr *data[])
2892 {
2893         struct vxlan_config conf;
2894         int err;
2895
2896         memset(&conf, 0, sizeof(conf));
2897
2898         if (data[IFLA_VXLAN_ID])
2899                 conf.vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2900
2901         if (data[IFLA_VXLAN_GROUP]) {
2902                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2903         } else if (data[IFLA_VXLAN_GROUP6]) {
2904                 if (!IS_ENABLED(CONFIG_IPV6))
2905                         return -EPFNOSUPPORT;
2906
2907                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2908                 conf.remote_ip.sa.sa_family = AF_INET6;
2909         }
2910
2911         if (data[IFLA_VXLAN_LOCAL]) {
2912                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2913                 conf.saddr.sa.sa_family = AF_INET;
2914         } else if (data[IFLA_VXLAN_LOCAL6]) {
2915                 if (!IS_ENABLED(CONFIG_IPV6))
2916                         return -EPFNOSUPPORT;
2917
2918                 /* TODO: respect scope id */
2919                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2920                 conf.saddr.sa.sa_family = AF_INET6;
2921         }
2922
2923         if (data[IFLA_VXLAN_LINK])
2924                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2925
2926         if (data[IFLA_VXLAN_TOS])
2927                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2928
2929         if (data[IFLA_VXLAN_TTL])
2930                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2931
2932         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2933                 conf.flags |= VXLAN_F_LEARN;
2934
2935         if (data[IFLA_VXLAN_AGEING])
2936                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2937
2938         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2939                 conf.flags |= VXLAN_F_PROXY;
2940
2941         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2942                 conf.flags |= VXLAN_F_RSC;
2943
2944         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2945                 conf.flags |= VXLAN_F_L2MISS;
2946
2947         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2948                 conf.flags |= VXLAN_F_L3MISS;
2949
2950         if (data[IFLA_VXLAN_LIMIT])
2951                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2952
2953         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2954             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2955                 conf.flags |= VXLAN_F_COLLECT_METADATA;
2956
2957         if (data[IFLA_VXLAN_PORT_RANGE]) {
2958                 const struct ifla_vxlan_port_range *p
2959                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2960                 conf.port_min = ntohs(p->low);
2961                 conf.port_max = ntohs(p->high);
2962         }
2963
2964         if (data[IFLA_VXLAN_PORT])
2965                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2966
2967         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2968                 conf.flags |= VXLAN_F_UDP_CSUM;
2969
2970         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2971             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2972                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2973
2974         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2975             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2976                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2977
2978         if (data[IFLA_VXLAN_REMCSUM_TX] &&
2979             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2980                 conf.flags |= VXLAN_F_REMCSUM_TX;
2981
2982         if (data[IFLA_VXLAN_REMCSUM_RX] &&
2983             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2984                 conf.flags |= VXLAN_F_REMCSUM_RX;
2985
2986         if (data[IFLA_VXLAN_GBP])
2987                 conf.flags |= VXLAN_F_GBP;
2988
2989         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2990                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2991
2992         err = vxlan_dev_configure(src_net, dev, &conf);
2993         switch (err) {
2994         case -ENODEV:
2995                 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
2996                 break;
2997
2998         case -EPERM:
2999                 pr_info("IPv6 is disabled via sysctl\n");
3000                 break;
3001
3002         case -EEXIST:
3003                 pr_info("duplicate VNI %u\n", conf.vni);
3004                 break;
3005         }
3006
3007         return err;
3008 }
3009
3010 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3011 {
3012         struct vxlan_dev *vxlan = netdev_priv(dev);
3013         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3014
3015         spin_lock(&vn->sock_lock);
3016         if (!hlist_unhashed(&vxlan->hlist))
3017                 hlist_del_rcu(&vxlan->hlist);
3018         spin_unlock(&vn->sock_lock);
3019
3020         gro_cells_destroy(&vxlan->gro_cells);
3021         list_del(&vxlan->next);
3022         unregister_netdevice_queue(dev, head);
3023 }
3024
3025 static size_t vxlan_get_size(const struct net_device *dev)
3026 {
3027
3028         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3029                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3030                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3031                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3032                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3033                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3034                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3035                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3036                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3037                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3038                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3039                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3040                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3041                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3042                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3043                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3044                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3045                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3046                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3047                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3048                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3049                 0;
3050 }
3051
3052 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3053 {
3054         const struct vxlan_dev *vxlan = netdev_priv(dev);
3055         const struct vxlan_rdst *dst = &vxlan->default_dst;
3056         struct ifla_vxlan_port_range ports = {
3057                 .low =  htons(vxlan->cfg.port_min),
3058                 .high = htons(vxlan->cfg.port_max),
3059         };
3060
3061         if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
3062                 goto nla_put_failure;
3063
3064         if (!vxlan_addr_any(&dst->remote_ip)) {
3065                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3066                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3067                                             dst->remote_ip.sin.sin_addr.s_addr))
3068                                 goto nla_put_failure;
3069 #if IS_ENABLED(CONFIG_IPV6)
3070                 } else {
3071                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3072                                              &dst->remote_ip.sin6.sin6_addr))
3073                                 goto nla_put_failure;
3074 #endif
3075                 }
3076         }
3077
3078         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3079                 goto nla_put_failure;
3080
3081         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3082                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3083                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3084                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3085                                 goto nla_put_failure;
3086 #if IS_ENABLED(CONFIG_IPV6)
3087                 } else {
3088                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3089                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3090                                 goto nla_put_failure;
3091 #endif
3092                 }
3093         }
3094
3095         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3096             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3097             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3098                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3099             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3100                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3101             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3102             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3103                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3104             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3105                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3106             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3107                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3108             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3109             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3110             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3111             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3112                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
3113             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3114                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3115             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3116                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3117             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3118                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3119             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3120                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3121                 goto nla_put_failure;
3122
3123         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3124                 goto nla_put_failure;
3125
3126         if (vxlan->flags & VXLAN_F_GBP &&
3127             nla_put_flag(skb, IFLA_VXLAN_GBP))
3128                 goto nla_put_failure;
3129
3130         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3131             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3132                 goto nla_put_failure;
3133
3134         return 0;
3135
3136 nla_put_failure:
3137         return -EMSGSIZE;
3138 }
3139
3140 static struct net *vxlan_get_link_net(const struct net_device *dev)
3141 {
3142         struct vxlan_dev *vxlan = netdev_priv(dev);
3143
3144         return vxlan->net;
3145 }
3146
3147 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3148         .kind           = "vxlan",
3149         .maxtype        = IFLA_VXLAN_MAX,
3150         .policy         = vxlan_policy,
3151         .priv_size      = sizeof(struct vxlan_dev),
3152         .setup          = vxlan_setup,
3153         .validate       = vxlan_validate,
3154         .newlink        = vxlan_newlink,
3155         .dellink        = vxlan_dellink,
3156         .get_size       = vxlan_get_size,
3157         .fill_info      = vxlan_fill_info,
3158         .get_link_net   = vxlan_get_link_net,
3159 };
3160
3161 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3162                                              struct net_device *dev)
3163 {
3164         struct vxlan_dev *vxlan, *next;
3165         LIST_HEAD(list_kill);
3166
3167         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3168                 struct vxlan_rdst *dst = &vxlan->default_dst;
3169
3170                 /* In case we created vxlan device with carrier
3171                  * and we loose the carrier due to module unload
3172                  * we also need to remove vxlan device. In other
3173                  * cases, it's not necessary and remote_ifindex
3174                  * is 0 here, so no matches.
3175                  */
3176                 if (dst->remote_ifindex == dev->ifindex)
3177                         vxlan_dellink(vxlan->dev, &list_kill);
3178         }
3179
3180         unregister_netdevice_many(&list_kill);
3181 }
3182
3183 static int vxlan_lowerdev_event(struct notifier_block *unused,
3184                                 unsigned long event, void *ptr)
3185 {
3186         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3187         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3188
3189         if (event == NETDEV_UNREGISTER)
3190                 vxlan_handle_lowerdev_unregister(vn, dev);
3191
3192         return NOTIFY_DONE;
3193 }
3194
3195 static struct notifier_block vxlan_notifier_block __read_mostly = {
3196         .notifier_call = vxlan_lowerdev_event,
3197 };
3198
3199 static __net_init int vxlan_init_net(struct net *net)
3200 {
3201         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3202         unsigned int h;
3203
3204         INIT_LIST_HEAD(&vn->vxlan_list);
3205         spin_lock_init(&vn->sock_lock);
3206
3207         for (h = 0; h < PORT_HASH_SIZE; ++h)
3208                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3209
3210         return 0;
3211 }
3212
3213 static void __net_exit vxlan_exit_net(struct net *net)
3214 {
3215         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3216         struct vxlan_dev *vxlan, *next;
3217         struct net_device *dev, *aux;
3218         LIST_HEAD(list);
3219
3220         rtnl_lock();
3221         for_each_netdev_safe(net, dev, aux)
3222                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3223                         unregister_netdevice_queue(dev, &list);
3224
3225         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3226                 /* If vxlan->dev is in the same netns, it has already been added
3227                  * to the list by the previous loop.
3228                  */
3229                 if (!net_eq(dev_net(vxlan->dev), net)) {
3230                         gro_cells_destroy(&vxlan->gro_cells);
3231                         unregister_netdevice_queue(vxlan->dev, &list);
3232                 }
3233         }
3234
3235         unregister_netdevice_many(&list);
3236         rtnl_unlock();
3237 }
3238
3239 static struct pernet_operations vxlan_net_ops = {
3240         .init = vxlan_init_net,
3241         .exit = vxlan_exit_net,
3242         .id   = &vxlan_net_id,
3243         .size = sizeof(struct vxlan_net),
3244 };
3245
3246 static int __init vxlan_init_module(void)
3247 {
3248         int rc;
3249
3250         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3251         if (!vxlan_wq)
3252                 return -ENOMEM;
3253
3254         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3255
3256         rc = register_pernet_subsys(&vxlan_net_ops);
3257         if (rc)
3258                 goto out1;
3259
3260         rc = register_netdevice_notifier(&vxlan_notifier_block);
3261         if (rc)
3262                 goto out2;
3263
3264         rc = rtnl_link_register(&vxlan_link_ops);
3265         if (rc)
3266                 goto out3;
3267
3268         return 0;
3269 out3:
3270         unregister_netdevice_notifier(&vxlan_notifier_block);
3271 out2:
3272         unregister_pernet_subsys(&vxlan_net_ops);
3273 out1:
3274         destroy_workqueue(vxlan_wq);
3275         return rc;
3276 }
3277 late_initcall(vxlan_init_module);
3278
3279 static void __exit vxlan_cleanup_module(void)
3280 {
3281         rtnl_link_unregister(&vxlan_link_ops);
3282         unregister_netdevice_notifier(&vxlan_notifier_block);
3283         destroy_workqueue(vxlan_wq);
3284         unregister_pernet_subsys(&vxlan_net_ops);
3285         /* rcu_barrier() is called by netns */
3286 }
3287 module_exit(vxlan_cleanup_module);
3288
3289 MODULE_LICENSE("GPL");
3290 MODULE_VERSION(VXLAN_VERSION);
3291 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3292 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3293 MODULE_ALIAS_RTNL_LINK("vxlan");