drivers/net/vxlan.c

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