drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/types.h>
  15 #include <linux/module.h>
  16 #include <linux/errno.h>
  17 #include <linux/slab.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/rculist.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/in.h>
  22 #include <linux/ip.h>
  23 #include <linux/udp.h>
  24 #include <linux/igmp.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/hash.h>
  29 #include <linux/ethtool.h>
  30 #include <net/arp.h>
  31 #include <net/ndisc.h>
  32 #include <net/ip.h>
  33 #include <net/ip_tunnels.h>
  34 #include <net/icmp.h>
  35 #include <net/udp.h>
  36 #include <net/udp_tunnel.h>
  37 #include <net/rtnetlink.h>
  38 #include <net/route.h>
  39 #include <net/dsfield.h>
  40 #include <net/inet_ecn.h>
  41 #include <net/net_namespace.h>
  42 #include <net/netns/generic.h>
  43 #include <net/vxlan.h>
  44 #include <net/protocol.h>
  45 #include <net/udp_tunnel.h>
  46 #if IS_ENABLED(CONFIG_IPV6)
  47 #include <net/ipv6.h>
  48 #include <net/addrconf.h>
  49 #include <net/ip6_tunnel.h>
  50 #include <net/ip6_checksum.h>
  51 #endif
  52 #include <net/dst_metadata.h>
  53
  54 #define VXLAN_VERSION   "0.1"
  55
  56 #define PORT_HASH_BITS  8
  57 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  58 #define FDB_AGE_DEFAULT 300 /* 5 min */
  59 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  60
  61 /* UDP port for VXLAN traffic.
  62  * The IANA assigned port is 4789, but the Linux default is 8472
  63  * for compatibility with early adopters.
  64  */
  65 static unsigned short vxlan_port __read_mostly = 8472;
  66 module_param_named(udp_port, vxlan_port, ushort, 0444);
  67 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  68
  69 static bool log_ecn_error = true;
  70 module_param(log_ecn_error, bool, 0644);
  71 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  72
  73 static int vxlan_net_id;
  74 static struct rtnl_link_ops vxlan_link_ops;
  75
  76 static const u8 all_zeros_mac[ETH_ALEN];
  77
  78 static struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
  79                                          bool no_share, u32 flags);
  80
  81 /* per-network namespace private data for this module */
  82 struct vxlan_net {
  83         struct list_head  vxlan_list;
  84         struct hlist_head sock_list[PORT_HASH_SIZE];
  85         spinlock_t        sock_lock;
  86 };
  87
  88 /* Forwarding table entry */
  89 struct vxlan_fdb {
  90         struct hlist_node hlist;        /* linked list of entries */
  91         struct rcu_head   rcu;
  92         unsigned long     updated;      /* jiffies */
  93         unsigned long     used;
  94         struct list_head  remotes;
  95         u8                eth_addr[ETH_ALEN];
  96         u16               state;        /* see ndm_state */
  97         u8                flags;        /* see ndm_flags */
  98 };
  99
 100 /* salt for hash table */
 101 static u32 vxlan_salt __read_mostly;
 102 static struct workqueue_struct *vxlan_wq;
 103
 104 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
 105 {
 106         return vs->flags & VXLAN_F_COLLECT_METADATA ||
 107                ip_tunnel_collect_metadata();
 108 }
 109
 110 #if IS_ENABLED(CONFIG_IPV6)
 111 static inline
 112 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 113 {
 114         if (a->sa.sa_family != b->sa.sa_family)
 115                 return false;
 116         if (a->sa.sa_family == AF_INET6)
 117                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 118         else
 119                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 120 }
 121
 122 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 123 {
 124         if (ipa->sa.sa_family == AF_INET6)
 125                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 126         else
 127                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 128 }
 129
 130 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 131 {
 132         if (ipa->sa.sa_family == AF_INET6)
 133                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 134         else
 135                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 136 }
 137
 138 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 139 {
 140         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 141                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 142                 ip->sa.sa_family = AF_INET6;
 143                 return 0;
 144         } else if (nla_len(nla) >= sizeof(__be32)) {
 145                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 146                 ip->sa.sa_family = AF_INET;
 147                 return 0;
 148         } else {
 149                 return -EAFNOSUPPORT;
 150         }
 151 }
 152
 153 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 154                               const union vxlan_addr *ip)
 155 {
 156         if (ip->sa.sa_family == AF_INET6)
 157                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 158         else
 159                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 160 }
 161
 162 #else /* !CONFIG_IPV6 */
 163
 164 static inline
 165 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 166 {
 167         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 168 }
 169
 170 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 171 {
 172         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 173 }
 174
 175 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 176 {
 177         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 178 }
 179
 180 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 181 {
 182         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 183                 return -EAFNOSUPPORT;
 184         } else if (nla_len(nla) >= sizeof(__be32)) {
 185                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 186                 ip->sa.sa_family = AF_INET;
 187                 return 0;
 188         } else {
 189                 return -EAFNOSUPPORT;
 190         }
 191 }
 192
 193 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 194                               const union vxlan_addr *ip)
 195 {
 196         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 197 }
 198 #endif
 199
 200 /* Virtual Network hash table head */
 201 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
 202 {
 203         return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
 204 }
 205
 206 /* Socket hash table head */
 207 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 208 {
 209         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 210
 211         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 212 }
 213
 214 /* First remote destination for a forwarding entry.
 215  * Guaranteed to be non-NULL because remotes are never deleted.
 216  */
 217 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 218 {
 219         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 220 }
 221
 222 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 223 {
 224         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 225 }
 226
 227 /* Find VXLAN socket based on network namespace, address family and UDP port
 228  * and enabled unshareable flags.
 229  */
 230 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 231                                           __be16 port, u32 flags)
 232 {
 233         struct vxlan_sock *vs;
 234
 235         flags &= VXLAN_F_RCV_FLAGS;
 236
 237         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 238                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 239                     vxlan_get_sk_family(vs) == family &&
 240                     vs->flags == flags)
 241                         return vs;
 242         }
 243         return NULL;
 244 }
 245
 246 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
 247 {
 248         struct vxlan_dev *vxlan;
 249
 250         hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
 251                 if (vxlan->default_dst.remote_vni == id)
 252                         return vxlan;
 253         }
 254
 255         return NULL;
 256 }
 257
 258 /* Look up VNI in a per net namespace table */
 259 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id,
 260                                         sa_family_t family, __be16 port,
 261                                         u32 flags)
 262 {
 263         struct vxlan_sock *vs;
 264
 265         vs = vxlan_find_sock(net, family, port, flags);
 266         if (!vs)
 267                 return NULL;
 268
 269         return vxlan_vs_find_vni(vs, id);
 270 }
 271
 272 /* Fill in neighbour message in skbuff. */
 273 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 274                           const struct vxlan_fdb *fdb,
 275                           u32 portid, u32 seq, int type, unsigned int flags,
 276                           const struct vxlan_rdst *rdst)
 277 {
 278         unsigned long now = jiffies;
 279         struct nda_cacheinfo ci;
 280         struct nlmsghdr *nlh;
 281         struct ndmsg *ndm;
 282         bool send_ip, send_eth;
 283
 284         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 285         if (nlh == NULL)
 286                 return -EMSGSIZE;
 287
 288         ndm = nlmsg_data(nlh);
 289         memset(ndm, 0, sizeof(*ndm));
 290
 291         send_eth = send_ip = true;
 292
 293         if (type == RTM_GETNEIGH) {
 294                 ndm->ndm_family = AF_INET;
 295                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 296                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 297         } else
 298                 ndm->ndm_family = AF_BRIDGE;
 299         ndm->ndm_state = fdb->state;
 300         ndm->ndm_ifindex = vxlan->dev->ifindex;
 301         ndm->ndm_flags = fdb->flags;
 302         ndm->ndm_type = RTN_UNICAST;
 303
 304         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 305             nla_put_s32(skb, NDA_LINK_NETNSID,
 306                         peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
 307                 goto nla_put_failure;
 308
 309         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 310                 goto nla_put_failure;
 311
 312         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 313                 goto nla_put_failure;
 314
 315         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 316             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 317                 goto nla_put_failure;
 318         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 319             nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
 320                 goto nla_put_failure;
 321         if (rdst->remote_ifindex &&
 322             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 323                 goto nla_put_failure;
 324
 325         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 326         ci.ndm_confirmed = 0;
 327         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 328         ci.ndm_refcnt    = 0;
 329
 330         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 331                 goto nla_put_failure;
 332
 333         nlmsg_end(skb, nlh);
 334         return 0;
 335
 336 nla_put_failure:
 337         nlmsg_cancel(skb, nlh);
 338         return -EMSGSIZE;
 339 }
 340
 341 static inline size_t vxlan_nlmsg_size(void)
 342 {
 343         return NLMSG_ALIGN(sizeof(struct ndmsg))
 344                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 345                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 346                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 347                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 348                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 349                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 350                 + nla_total_size(sizeof(struct nda_cacheinfo));
 351 }
 352
 353 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 354                              struct vxlan_rdst *rd, int type)
 355 {
 356         struct net *net = dev_net(vxlan->dev);
 357         struct sk_buff *skb;
 358         int err = -ENOBUFS;
 359
 360         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 361         if (skb == NULL)
 362                 goto errout;
 363
 364         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 365         if (err < 0) {
 366                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 367                 WARN_ON(err == -EMSGSIZE);
 368                 kfree_skb(skb);
 369                 goto errout;
 370         }
 371
 372         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 373         return;
 374 errout:
 375         if (err < 0)
 376                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 377 }
 378
 379 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 380 {
 381         struct vxlan_dev *vxlan = netdev_priv(dev);
 382         struct vxlan_fdb f = {
 383                 .state = NUD_STALE,
 384         };
 385         struct vxlan_rdst remote = {
 386                 .remote_ip = *ipa, /* goes to NDA_DST */
 387                 .remote_vni = VXLAN_N_VID,
 388         };
 389
 390         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 391 }
 392
 393 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 394 {
 395         struct vxlan_fdb f = {
 396                 .state = NUD_STALE,
 397         };
 398         struct vxlan_rdst remote = { };
 399
 400         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 401
 402         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 403 }
 404
 405 /* Hash Ethernet address */
 406 static u32 eth_hash(const unsigned char *addr)
 407 {
 408         u64 value = get_unaligned((u64 *)addr);
 409
 410         /* only want 6 bytes */
 411 #ifdef __BIG_ENDIAN
 412         value >>= 16;
 413 #else
 414         value <<= 16;
 415 #endif
 416         return hash_64(value, FDB_HASH_BITS);
 417 }
 418
 419 /* Hash chain to use given mac address */
 420 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 421                                                 const u8 *mac)
 422 {
 423         return &vxlan->fdb_head[eth_hash(mac)];
 424 }
 425
 426 /* Look up Ethernet address in forwarding table */
 427 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 428                                         const u8 *mac)
 429 {
 430         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 431         struct vxlan_fdb *f;
 432
 433         hlist_for_each_entry_rcu(f, head, hlist) {
 434                 if (ether_addr_equal(mac, f->eth_addr))
 435                         return f;
 436         }
 437
 438         return NULL;
 439 }
 440
 441 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 442                                         const u8 *mac)
 443 {
 444         struct vxlan_fdb *f;
 445
 446         f = __vxlan_find_mac(vxlan, mac);
 447         if (f)
 448                 f->used = jiffies;
 449
 450         return f;
 451 }
 452
 453 /* caller should hold vxlan->hash_lock */
 454 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 455                                               union vxlan_addr *ip, __be16 port,
 456                                               __u32 vni, __u32 ifindex)
 457 {
 458         struct vxlan_rdst *rd;
 459
 460         list_for_each_entry(rd, &f->remotes, list) {
 461                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 462                     rd->remote_port == port &&
 463                     rd->remote_vni == vni &&
 464                     rd->remote_ifindex == ifindex)
 465                         return rd;
 466         }
 467
 468         return NULL;
 469 }
 470
 471 /* Replace destination of unicast mac */
 472 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 473                              union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
 474 {
 475         struct vxlan_rdst *rd;
 476
 477         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 478         if (rd)
 479                 return 0;
 480
 481         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 482         if (!rd)
 483                 return 0;
 484         rd->remote_ip = *ip;
 485         rd->remote_port = port;
 486         rd->remote_vni = vni;
 487         rd->remote_ifindex = ifindex;
 488         return 1;
 489 }
 490
 491 /* Add/update destinations for multicast */
 492 static int vxlan_fdb_append(struct vxlan_fdb *f,
 493                             union vxlan_addr *ip, __be16 port, __u32 vni,
 494                             __u32 ifindex, struct vxlan_rdst **rdp)
 495 {
 496         struct vxlan_rdst *rd;
 497
 498         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 499         if (rd)
 500                 return 0;
 501
 502         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 503         if (rd == NULL)
 504                 return -ENOBUFS;
 505         rd->remote_ip = *ip;
 506         rd->remote_port = port;
 507         rd->remote_vni = vni;
 508         rd->remote_ifindex = ifindex;
 509
 510         list_add_tail_rcu(&rd->list, &f->remotes);
 511
 512         *rdp = rd;
 513         return 1;
 514 }
 515
 516 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 517                                           unsigned int off,
 518                                           struct vxlanhdr *vh, size_t hdrlen,
 519                                           u32 data, struct gro_remcsum *grc,
 520                                           bool nopartial)
 521 {
 522         size_t start, offset, plen;
 523
 524         if (skb->remcsum_offload)
 525                 return NULL;
 526
 527         if (!NAPI_GRO_CB(skb)->csum_valid)
 528                 return NULL;
 529
 530         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
 531         offset = start + ((data & VXLAN_RCO_UDP) ?
 532                           offsetof(struct udphdr, check) :
 533                           offsetof(struct tcphdr, check));
 534
 535         plen = hdrlen + offset + sizeof(u16);
 536
 537         /* Pull checksum that will be written */
 538         if (skb_gro_header_hard(skb, off + plen)) {
 539                 vh = skb_gro_header_slow(skb, off + plen, off);
 540                 if (!vh)
 541                         return NULL;
 542         }
 543
 544         skb_gro_remcsum_process(skb, (void *)vh + hdrlen,
 545                                 start, offset, grc, nopartial);
 546
 547         skb->remcsum_offload = 1;
 548
 549         return vh;
 550 }
 551
 552 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
 553                                           struct sk_buff *skb,
 554                                           struct udp_offload *uoff)
 555 {
 556         struct sk_buff *p, **pp = NULL;
 557         struct vxlanhdr *vh, *vh2;
 558         unsigned int hlen, off_vx;
 559         int flush = 1;
 560         struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
 561                                              udp_offloads);
 562         u32 flags;
 563         struct gro_remcsum grc;
 564
 565         skb_gro_remcsum_init(&grc);
 566
 567         off_vx = skb_gro_offset(skb);
 568         hlen = off_vx + sizeof(*vh);
 569         vh   = skb_gro_header_fast(skb, off_vx);
 570         if (skb_gro_header_hard(skb, hlen)) {
 571                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 572                 if (unlikely(!vh))
 573                         goto out;
 574         }
 575
 576         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 577         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 578
 579         flags = ntohl(vh->vx_flags);
 580
 581         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 582                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 583                                        ntohl(vh->vx_vni), &grc,
 584                                        !!(vs->flags &
 585                                           VXLAN_F_REMCSUM_NOPARTIAL));
 586
 587                 if (!vh)
 588                         goto out;
 589         }
 590
 591         flush = 0;
 592
 593         for (p = *head; p; p = p->next) {
 594                 if (!NAPI_GRO_CB(p)->same_flow)
 595                         continue;
 596
 597                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 598                 if (vh->vx_flags != vh2->vx_flags ||
 599                     vh->vx_vni != vh2->vx_vni) {
 600                         NAPI_GRO_CB(p)->same_flow = 0;
 601                         continue;
 602                 }
 603         }
 604
 605         pp = eth_gro_receive(head, skb);
 606
 607 out:
 608         skb_gro_remcsum_cleanup(skb, &grc);
 609         NAPI_GRO_CB(skb)->flush |= flush;
 610
 611         return pp;
 612 }
 613
 614 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
 615                               struct udp_offload *uoff)
 616 {
 617         udp_tunnel_gro_complete(skb, nhoff);
 618
 619         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 620 }
 621
 622 /* Notify netdevs that UDP port started listening */
 623 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 624 {
 625         struct net_device *dev;
 626         struct sock *sk = vs->sock->sk;
 627         struct net *net = sock_net(sk);
 628         sa_family_t sa_family = vxlan_get_sk_family(vs);
 629         __be16 port = inet_sk(sk)->inet_sport;
 630         int err;
 631
 632         if (sa_family == AF_INET) {
 633                 err = udp_add_offload(&vs->udp_offloads);
 634                 if (err)
 635                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 636         }
 637
 638         rcu_read_lock();
 639         for_each_netdev_rcu(net, dev) {
 640                 if (dev->netdev_ops->ndo_add_vxlan_port)
 641                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 642                                                             port);
 643         }
 644         rcu_read_unlock();
 645 }
 646
 647 /* Notify netdevs that UDP port is no more listening */
 648 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 649 {
 650         struct net_device *dev;
 651         struct sock *sk = vs->sock->sk;
 652         struct net *net = sock_net(sk);
 653         sa_family_t sa_family = vxlan_get_sk_family(vs);
 654         __be16 port = inet_sk(sk)->inet_sport;
 655
 656         rcu_read_lock();
 657         for_each_netdev_rcu(net, dev) {
 658                 if (dev->netdev_ops->ndo_del_vxlan_port)
 659                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 660                                                             port);
 661         }
 662         rcu_read_unlock();
 663
 664         if (sa_family == AF_INET)
 665                 udp_del_offload(&vs->udp_offloads);
 666 }
 667
 668 /* Add new entry to forwarding table -- assumes lock held */
 669 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 670                             const u8 *mac, union vxlan_addr *ip,
 671                             __u16 state, __u16 flags,
 672                             __be16 port, __u32 vni, __u32 ifindex,
 673                             __u8 ndm_flags)
 674 {
 675         struct vxlan_rdst *rd = NULL;
 676         struct vxlan_fdb *f;
 677         int notify = 0;
 678
 679         f = __vxlan_find_mac(vxlan, mac);
 680         if (f) {
 681                 if (flags & NLM_F_EXCL) {
 682                         netdev_dbg(vxlan->dev,
 683                                    "lost race to create %pM\n", mac);
 684                         return -EEXIST;
 685                 }
 686                 if (f->state != state) {
 687                         f->state = state;
 688                         f->updated = jiffies;
 689                         notify = 1;
 690                 }
 691                 if (f->flags != ndm_flags) {
 692                         f->flags = ndm_flags;
 693                         f->updated = jiffies;
 694                         notify = 1;
 695                 }
 696                 if ((flags & NLM_F_REPLACE)) {
 697                         /* Only change unicasts */
 698                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 699                              is_zero_ether_addr(f->eth_addr))) {
 700                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 701                                                            ifindex);
 702                         } else
 703                                 return -EOPNOTSUPP;
 704                 }
 705                 if ((flags & NLM_F_APPEND) &&
 706                     (is_multicast_ether_addr(f->eth_addr) ||
 707                      is_zero_ether_addr(f->eth_addr))) {
 708                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 709                                                   &rd);
 710
 711                         if (rc < 0)
 712                                 return rc;
 713                         notify |= rc;
 714                 }
 715         } else {
 716                 if (!(flags & NLM_F_CREATE))
 717                         return -ENOENT;
 718
 719                 if (vxlan->cfg.addrmax &&
 720                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 721                         return -ENOSPC;
 722
 723                 /* Disallow replace to add a multicast entry */
 724                 if ((flags & NLM_F_REPLACE) &&
 725                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 726                         return -EOPNOTSUPP;
 727
 728                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 729                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 730                 if (!f)
 731                         return -ENOMEM;
 732
 733                 notify = 1;
 734                 f->state = state;
 735                 f->flags = ndm_flags;
 736                 f->updated = f->used = jiffies;
 737                 INIT_LIST_HEAD(&f->remotes);
 738                 memcpy(f->eth_addr, mac, ETH_ALEN);
 739
 740                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 741
 742                 ++vxlan->addrcnt;
 743                 hlist_add_head_rcu(&f->hlist,
 744                                    vxlan_fdb_head(vxlan, mac));
 745         }
 746
 747         if (notify) {
 748                 if (rd == NULL)
 749                         rd = first_remote_rtnl(f);
 750                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 751         }
 752
 753         return 0;
 754 }
 755
 756 static void vxlan_fdb_free(struct rcu_head *head)
 757 {
 758         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 759         struct vxlan_rdst *rd, *nd;
 760
 761         list_for_each_entry_safe(rd, nd, &f->remotes, list)
 762                 kfree(rd);
 763         kfree(f);
 764 }
 765
 766 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 767 {
 768         netdev_dbg(vxlan->dev,
 769                     "delete %pM\n", f->eth_addr);
 770
 771         --vxlan->addrcnt;
 772         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 773
 774         hlist_del_rcu(&f->hlist);
 775         call_rcu(&f->rcu, vxlan_fdb_free);
 776 }
 777
 778 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 779                            union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
 780 {
 781         struct net *net = dev_net(vxlan->dev);
 782         int err;
 783
 784         if (tb[NDA_DST]) {
 785                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 786                 if (err)
 787                         return err;
 788         } else {
 789                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 790                 if (remote->sa.sa_family == AF_INET) {
 791                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 792                         ip->sa.sa_family = AF_INET;
 793 #if IS_ENABLED(CONFIG_IPV6)
 794                 } else {
 795                         ip->sin6.sin6_addr = in6addr_any;
 796                         ip->sa.sa_family = AF_INET6;
 797 #endif
 798                 }
 799         }
 800
 801         if (tb[NDA_PORT]) {
 802                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 803                         return -EINVAL;
 804                 *port = nla_get_be16(tb[NDA_PORT]);
 805         } else {
 806                 *port = vxlan->cfg.dst_port;
 807         }
 808
 809         if (tb[NDA_VNI]) {
 810                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 811                         return -EINVAL;
 812                 *vni = nla_get_u32(tb[NDA_VNI]);
 813         } else {
 814                 *vni = vxlan->default_dst.remote_vni;
 815         }
 816
 817         if (tb[NDA_IFINDEX]) {
 818                 struct net_device *tdev;
 819
 820                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 821                         return -EINVAL;
 822                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 823                 tdev = __dev_get_by_index(net, *ifindex);
 824                 if (!tdev)
 825                         return -EADDRNOTAVAIL;
 826         } else {
 827                 *ifindex = 0;
 828         }
 829
 830         return 0;
 831 }
 832
 833 /* Add static entry (via netlink) */
 834 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 835                          struct net_device *dev,
 836                          const unsigned char *addr, u16 vid, u16 flags)
 837 {
 838         struct vxlan_dev *vxlan = netdev_priv(dev);
 839         /* struct net *net = dev_net(vxlan->dev); */
 840         union vxlan_addr ip;
 841         __be16 port;
 842         u32 vni, ifindex;
 843         int err;
 844
 845         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 846                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 847                         ndm->ndm_state);
 848                 return -EINVAL;
 849         }
 850
 851         if (tb[NDA_DST] == NULL)
 852                 return -EINVAL;
 853
 854         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 855         if (err)
 856                 return err;
 857
 858         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 859                 return -EAFNOSUPPORT;
 860
 861         spin_lock_bh(&vxlan->hash_lock);
 862         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 863                                port, vni, ifindex, ndm->ndm_flags);
 864         spin_unlock_bh(&vxlan->hash_lock);
 865
 866         return err;
 867 }
 868
 869 /* Delete entry (via netlink) */
 870 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 871                             struct net_device *dev,
 872                             const unsigned char *addr, u16 vid)
 873 {
 874         struct vxlan_dev *vxlan = netdev_priv(dev);
 875         struct vxlan_fdb *f;
 876         struct vxlan_rdst *rd = NULL;
 877         union vxlan_addr ip;
 878         __be16 port;
 879         u32 vni, ifindex;
 880         int err;
 881
 882         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 883         if (err)
 884                 return err;
 885
 886         err = -ENOENT;
 887
 888         spin_lock_bh(&vxlan->hash_lock);
 889         f = vxlan_find_mac(vxlan, addr);
 890         if (!f)
 891                 goto out;
 892
 893         if (!vxlan_addr_any(&ip)) {
 894                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 895                 if (!rd)
 896                         goto out;
 897         }
 898
 899         err = 0;
 900
 901         /* remove a destination if it's not the only one on the list,
 902          * otherwise destroy the fdb entry
 903          */
 904         if (rd && !list_is_singular(&f->remotes)) {
 905                 list_del_rcu(&rd->list);
 906                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 907                 kfree_rcu(rd, rcu);
 908                 goto out;
 909         }
 910
 911         vxlan_fdb_destroy(vxlan, f);
 912
 913 out:
 914         spin_unlock_bh(&vxlan->hash_lock);
 915
 916         return err;
 917 }
 918
 919 /* Dump forwarding table */
 920 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 921                           struct net_device *dev,
 922                           struct net_device *filter_dev, int idx)
 923 {
 924         struct vxlan_dev *vxlan = netdev_priv(dev);
 925         unsigned int h;
 926
 927         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 928                 struct vxlan_fdb *f;
 929                 int err;
 930
 931                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 932                         struct vxlan_rdst *rd;
 933
 934                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 935                                 if (idx < cb->args[0])
 936                                         goto skip;
 937
 938                                 err = vxlan_fdb_info(skb, vxlan, f,
 939                                                      NETLINK_CB(cb->skb).portid,
 940                                                      cb->nlh->nlmsg_seq,
 941                                                      RTM_NEWNEIGH,
 942                                                      NLM_F_MULTI, rd);
 943                                 if (err < 0)
 944                                         goto out;
 945 skip:
 946                                 ++idx;
 947                         }
 948                 }
 949         }
 950 out:
 951         return idx;
 952 }
 953
 954 /* Watch incoming packets to learn mapping between Ethernet address
 955  * and Tunnel endpoint.
 956  * Return true if packet is bogus and should be dropped.
 957  */
 958 static bool vxlan_snoop(struct net_device *dev,
 959                         union vxlan_addr *src_ip, const u8 *src_mac)
 960 {
 961         struct vxlan_dev *vxlan = netdev_priv(dev);
 962         struct vxlan_fdb *f;
 963
 964         f = vxlan_find_mac(vxlan, src_mac);
 965         if (likely(f)) {
 966                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 967
 968                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 969                         return false;
 970
 971                 /* Don't migrate static entries, drop packets */
 972                 if (f->state & NUD_NOARP)
 973                         return true;
 974
 975                 if (net_ratelimit())
 976                         netdev_info(dev,
 977                                     "%pM migrated from %pIS to %pIS\n",
 978                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 979
 980                 rdst->remote_ip = *src_ip;
 981                 f->updated = jiffies;
 982                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 983         } else {
 984                 /* learned new entry */
 985                 spin_lock(&vxlan->hash_lock);
 986
 987                 /* close off race between vxlan_flush and incoming packets */
 988                 if (netif_running(dev))
 989                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 990                                          NUD_REACHABLE,
 991                                          NLM_F_EXCL|NLM_F_CREATE,
 992                                          vxlan->cfg.dst_port,
 993                                          vxlan->default_dst.remote_vni,
 994                                          0, NTF_SELF);
 995                 spin_unlock(&vxlan->hash_lock);
 996         }
 997
 998         return false;
 999 }
1000
1001 /* See if multicast group is already in use by other ID */
1002 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1003 {
1004         struct vxlan_dev *vxlan;
1005
1006         /* The vxlan_sock is only used by dev, leaving group has
1007          * no effect on other vxlan devices.
1008          */
1009         if (atomic_read(&dev->vn_sock->refcnt) == 1)
1010                 return false;
1011
1012         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1013                 if (!netif_running(vxlan->dev) || vxlan == dev)
1014                         continue;
1015
1016                 if (vxlan->vn_sock != dev->vn_sock)
1017                         continue;
1018
1019                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1020                                       &dev->default_dst.remote_ip))
1021                         continue;
1022
1023                 if (vxlan->default_dst.remote_ifindex !=
1024                     dev->default_dst.remote_ifindex)
1025                         continue;
1026
1027                 return true;
1028         }
1029
1030         return false;
1031 }
1032
1033 static void vxlan_sock_release(struct vxlan_sock *vs)
1034 {
1035         struct sock *sk = vs->sock->sk;
1036         struct net *net = sock_net(sk);
1037         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1038
1039         if (!atomic_dec_and_test(&vs->refcnt))
1040                 return;
1041
1042         spin_lock(&vn->sock_lock);
1043         hlist_del_rcu(&vs->hlist);
1044         vxlan_notify_del_rx_port(vs);
1045         spin_unlock(&vn->sock_lock);
1046
1047         queue_work(vxlan_wq, &vs->del_work);
1048 }
1049
1050 /* Update multicast group membership when first VNI on
1051  * multicast address is brought up
1052  */
1053 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1054 {
1055         struct vxlan_sock *vs = vxlan->vn_sock;
1056         struct sock *sk = vs->sock->sk;
1057         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1058         int ifindex = vxlan->default_dst.remote_ifindex;
1059         int ret = -EINVAL;
1060
1061         lock_sock(sk);
1062         if (ip->sa.sa_family == AF_INET) {
1063                 struct ip_mreqn mreq = {
1064                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1065                         .imr_ifindex            = ifindex,
1066                 };
1067
1068                 ret = ip_mc_join_group(sk, &mreq);
1069 #if IS_ENABLED(CONFIG_IPV6)
1070         } else {
1071                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1072                                                    &ip->sin6.sin6_addr);
1073 #endif
1074         }
1075         release_sock(sk);
1076
1077         return ret;
1078 }
1079
1080 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1081 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1082 {
1083         struct vxlan_sock *vs = vxlan->vn_sock;
1084         struct sock *sk = vs->sock->sk;
1085         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1086         int ifindex = vxlan->default_dst.remote_ifindex;
1087         int ret = -EINVAL;
1088
1089         lock_sock(sk);
1090         if (ip->sa.sa_family == AF_INET) {
1091                 struct ip_mreqn mreq = {
1092                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1093                         .imr_ifindex            = ifindex,
1094                 };
1095
1096                 ret = ip_mc_leave_group(sk, &mreq);
1097 #if IS_ENABLED(CONFIG_IPV6)
1098         } else {
1099                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1100                                                    &ip->sin6.sin6_addr);
1101 #endif
1102         }
1103         release_sock(sk);
1104
1105         return ret;
1106 }
1107
1108 static struct vxlanhdr *vxlan_remcsum(struct sk_buff *skb, struct vxlanhdr *vh,
1109                                       size_t hdrlen, u32 data, bool nopartial)
1110 {
1111         size_t start, offset, plen;
1112
1113         start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
1114         offset = start + ((data & VXLAN_RCO_UDP) ?
1115                           offsetof(struct udphdr, check) :
1116                           offsetof(struct tcphdr, check));
1117
1118         plen = hdrlen + offset + sizeof(u16);
1119
1120         if (!pskb_may_pull(skb, plen))
1121                 return NULL;
1122
1123         vh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1124
1125         skb_remcsum_process(skb, (void *)vh + hdrlen, start, offset,
1126                             nopartial);
1127
1128         return vh;
1129 }
1130
1131 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1132                       struct vxlan_metadata *md, u32 vni,
1133                       struct metadata_dst *tun_dst)
1134 {
1135         struct iphdr *oip = NULL;
1136         struct ipv6hdr *oip6 = NULL;
1137         struct vxlan_dev *vxlan;
1138         struct pcpu_sw_netstats *stats;
1139         union vxlan_addr saddr;
1140         int err = 0;
1141         union vxlan_addr *remote_ip;
1142
1143         /* For flow based devices, map all packets to VNI 0 */
1144         if (vs->flags & VXLAN_F_COLLECT_METADATA)
1145                 vni = 0;
1146
1147         /* Is this VNI defined? */
1148         vxlan = vxlan_vs_find_vni(vs, vni);
1149         if (!vxlan)
1150                 goto drop;
1151
1152         remote_ip = &vxlan->default_dst.remote_ip;
1153         skb_reset_mac_header(skb);
1154         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1155         skb->protocol = eth_type_trans(skb, vxlan->dev);
1156         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1157
1158         /* Ignore packet loops (and multicast echo) */
1159         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1160                 goto drop;
1161
1162         /* Re-examine inner Ethernet packet */
1163         if (remote_ip->sa.sa_family == AF_INET) {
1164                 oip = ip_hdr(skb);
1165                 saddr.sin.sin_addr.s_addr = oip->saddr;
1166                 saddr.sa.sa_family = AF_INET;
1167 #if IS_ENABLED(CONFIG_IPV6)
1168         } else {
1169                 oip6 = ipv6_hdr(skb);
1170                 saddr.sin6.sin6_addr = oip6->saddr;
1171                 saddr.sa.sa_family = AF_INET6;
1172 #endif
1173         }
1174
1175         if (tun_dst) {
1176                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1177                 tun_dst = NULL;
1178         }
1179
1180         if ((vxlan->flags & VXLAN_F_LEARN) &&
1181             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1182                 goto drop;
1183
1184         skb_reset_network_header(skb);
1185         /* In flow-based mode, GBP is carried in dst_metadata */
1186         if (!(vs->flags & VXLAN_F_COLLECT_METADATA))
1187                 skb->mark = md->gbp;
1188
1189         if (oip6)
1190                 err = IP6_ECN_decapsulate(oip6, skb);
1191         if (oip)
1192                 err = IP_ECN_decapsulate(oip, skb);
1193
1194         if (unlikely(err)) {
1195                 if (log_ecn_error) {
1196                         if (oip6)
1197                                 net_info_ratelimited("non-ECT from %pI6\n",
1198                                                      &oip6->saddr);
1199                         if (oip)
1200                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1201                                                      &oip->saddr, oip->tos);
1202                 }
1203                 if (err > 1) {
1204                         ++vxlan->dev->stats.rx_frame_errors;
1205                         ++vxlan->dev->stats.rx_errors;
1206                         goto drop;
1207                 }
1208         }
1209
1210         stats = this_cpu_ptr(vxlan->dev->tstats);
1211         u64_stats_update_begin(&stats->syncp);
1212         stats->rx_packets++;
1213         stats->rx_bytes += skb->len;
1214         u64_stats_update_end(&stats->syncp);
1215
1216         netif_rx(skb);
1217
1218         return;
1219 drop:
1220         if (tun_dst)
1221                 dst_release((struct dst_entry *)tun_dst);
1222
1223         /* Consume bad packet */
1224         kfree_skb(skb);
1225 }
1226
1227 /* Callback from net/ipv4/udp.c to receive packets */
1228 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1229 {
1230         struct metadata_dst *tun_dst = NULL;
1231         struct ip_tunnel_info *info;
1232         struct vxlan_sock *vs;
1233         struct vxlanhdr *vxh;
1234         u32 flags, vni;
1235         struct vxlan_metadata _md;
1236         struct vxlan_metadata *md = &_md;
1237
1238         /* Need Vxlan and inner Ethernet header to be present */
1239         if (!pskb_may_pull(skb, VXLAN_HLEN))
1240                 goto error;
1241
1242         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1243         flags = ntohl(vxh->vx_flags);
1244         vni = ntohl(vxh->vx_vni);
1245
1246         if (flags & VXLAN_HF_VNI) {
1247                 flags &= ~VXLAN_HF_VNI;
1248         } else {
1249                 /* VNI flag always required to be set */
1250                 goto bad_flags;
1251         }
1252
1253         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1254                 goto drop;
1255         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1256
1257         vs = rcu_dereference_sk_user_data(sk);
1258         if (!vs)
1259                 goto drop;
1260
1261         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1262                 vxh = vxlan_remcsum(skb, vxh, sizeof(struct vxlanhdr), vni,
1263                                     !!(vs->flags & VXLAN_F_REMCSUM_NOPARTIAL));
1264                 if (!vxh)
1265                         goto drop;
1266
1267                 flags &= ~VXLAN_HF_RCO;
1268                 vni &= VXLAN_VNI_MASK;
1269         }
1270
1271         if (vxlan_collect_metadata(vs)) {
1272                 const struct iphdr *iph = ip_hdr(skb);
1273
1274                 tun_dst = metadata_dst_alloc(sizeof(*md), GFP_ATOMIC);
1275                 if (!tun_dst)
1276                         goto drop;
1277
1278                 info = &tun_dst->u.tun_info;
1279                 info->key.u.ipv4.src = iph->saddr;
1280                 info->key.u.ipv4.dst = iph->daddr;
1281                 info->key.tos = iph->tos;
1282                 info->key.ttl = iph->ttl;
1283                 info->key.tp_src = udp_hdr(skb)->source;
1284                 info->key.tp_dst = udp_hdr(skb)->dest;
1285
1286                 info->mode = IP_TUNNEL_INFO_RX;
1287                 info->key.tun_flags = TUNNEL_KEY;
1288                 info->key.tun_id = cpu_to_be64(vni >> 8);
1289                 if (udp_hdr(skb)->check != 0)
1290                         info->key.tun_flags |= TUNNEL_CSUM;
1291
1292                 md = ip_tunnel_info_opts(info, sizeof(*md));
1293         } else {
1294                 memset(md, 0, sizeof(*md));
1295         }
1296
1297         /* For backwards compatibility, only allow reserved fields to be
1298          * used by VXLAN extensions if explicitly requested.
1299          */
1300         if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1301                 struct vxlanhdr_gbp *gbp;
1302
1303                 gbp = (struct vxlanhdr_gbp *)vxh;
1304                 md->gbp = ntohs(gbp->policy_id);
1305
1306                 if (tun_dst)
1307                         info->key.tun_flags |= TUNNEL_VXLAN_OPT;
1308
1309                 if (gbp->dont_learn)
1310                         md->gbp |= VXLAN_GBP_DONT_LEARN;
1311
1312                 if (gbp->policy_applied)
1313                         md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1314
1315                 flags &= ~VXLAN_GBP_USED_BITS;
1316         }
1317
1318         if (flags || vni & ~VXLAN_VNI_MASK) {
1319                 /* If there are any unprocessed flags remaining treat
1320                  * this as a malformed packet. This behavior diverges from
1321                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1322                  * in reserved fields are to be ignored. The approach here
1323                  * maintains compatibility with previous stack code, and also
1324                  * is more robust and provides a little more security in
1325                  * adding extensions to VXLAN.
1326                  */
1327
1328                 goto bad_flags;
1329         }
1330
1331         vxlan_rcv(vs, skb, md, vni >> 8, tun_dst);
1332         return 0;
1333
1334 drop:
1335         /* Consume bad packet */
1336         kfree_skb(skb);
1337         return 0;
1338
1339 bad_flags:
1340         netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1341                    ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1342
1343 error:
1344         if (tun_dst)
1345                 dst_release((struct dst_entry *)tun_dst);
1346
1347         /* Return non vxlan pkt */
1348         return 1;
1349 }
1350
1351 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1352 {
1353         struct vxlan_dev *vxlan = netdev_priv(dev);
1354         struct arphdr *parp;
1355         u8 *arpptr, *sha;
1356         __be32 sip, tip;
1357         struct neighbour *n;
1358
1359         if (dev->flags & IFF_NOARP)
1360                 goto out;
1361
1362         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1363                 dev->stats.tx_dropped++;
1364                 goto out;
1365         }
1366         parp = arp_hdr(skb);
1367
1368         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1369              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1370             parp->ar_pro != htons(ETH_P_IP) ||
1371             parp->ar_op != htons(ARPOP_REQUEST) ||
1372             parp->ar_hln != dev->addr_len ||
1373             parp->ar_pln != 4)
1374                 goto out;
1375         arpptr = (u8 *)parp + sizeof(struct arphdr);
1376         sha = arpptr;
1377         arpptr += dev->addr_len;        /* sha */
1378         memcpy(&sip, arpptr, sizeof(sip));
1379         arpptr += sizeof(sip);
1380         arpptr += dev->addr_len;        /* tha */
1381         memcpy(&tip, arpptr, sizeof(tip));
1382
1383         if (ipv4_is_loopback(tip) ||
1384             ipv4_is_multicast(tip))
1385                 goto out;
1386
1387         n = neigh_lookup(&arp_tbl, &tip, dev);
1388
1389         if (n) {
1390                 struct vxlan_fdb *f;
1391                 struct sk_buff  *reply;
1392
1393                 if (!(n->nud_state & NUD_CONNECTED)) {
1394                         neigh_release(n);
1395                         goto out;
1396                 }
1397
1398                 f = vxlan_find_mac(vxlan, n->ha);
1399                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1400                         /* bridge-local neighbor */
1401                         neigh_release(n);
1402                         goto out;
1403                 }
1404
1405                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1406                                 n->ha, sha);
1407
1408                 neigh_release(n);
1409
1410                 if (reply == NULL)
1411                         goto out;
1412
1413                 skb_reset_mac_header(reply);
1414                 __skb_pull(reply, skb_network_offset(reply));
1415                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1416                 reply->pkt_type = PACKET_HOST;
1417
1418                 if (netif_rx_ni(reply) == NET_RX_DROP)
1419                         dev->stats.rx_dropped++;
1420         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1421                 union vxlan_addr ipa = {
1422                         .sin.sin_addr.s_addr = tip,
1423                         .sin.sin_family = AF_INET,
1424                 };
1425
1426                 vxlan_ip_miss(dev, &ipa);
1427         }
1428 out:
1429         consume_skb(skb);
1430         return NETDEV_TX_OK;
1431 }
1432
1433 #if IS_ENABLED(CONFIG_IPV6)
1434 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1435         struct neighbour *n, bool isrouter)
1436 {
1437         struct net_device *dev = request->dev;
1438         struct sk_buff *reply;
1439         struct nd_msg *ns, *na;
1440         struct ipv6hdr *pip6;
1441         u8 *daddr;
1442         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1443         int ns_olen;
1444         int i, len;
1445
1446         if (dev == NULL)
1447                 return NULL;
1448
1449         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1450                 sizeof(*na) + na_olen + dev->needed_tailroom;
1451         reply = alloc_skb(len, GFP_ATOMIC);
1452         if (reply == NULL)
1453                 return NULL;
1454
1455         reply->protocol = htons(ETH_P_IPV6);
1456         reply->dev = dev;
1457         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1458         skb_push(reply, sizeof(struct ethhdr));
1459         skb_set_mac_header(reply, 0);
1460
1461         ns = (struct nd_msg *)skb_transport_header(request);
1462
1463         daddr = eth_hdr(request)->h_source;
1464         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1465         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1466                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1467                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1468                         break;
1469                 }
1470         }
1471
1472         /* Ethernet header */
1473         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1474         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1475         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1476         reply->protocol = htons(ETH_P_IPV6);
1477
1478         skb_pull(reply, sizeof(struct ethhdr));
1479         skb_set_network_header(reply, 0);
1480         skb_put(reply, sizeof(struct ipv6hdr));
1481
1482         /* IPv6 header */
1483
1484         pip6 = ipv6_hdr(reply);
1485         memset(pip6, 0, sizeof(struct ipv6hdr));
1486         pip6->version = 6;
1487         pip6->priority = ipv6_hdr(request)->priority;
1488         pip6->nexthdr = IPPROTO_ICMPV6;
1489         pip6->hop_limit = 255;
1490         pip6->daddr = ipv6_hdr(request)->saddr;
1491         pip6->saddr = *(struct in6_addr *)n->primary_key;
1492
1493         skb_pull(reply, sizeof(struct ipv6hdr));
1494         skb_set_transport_header(reply, 0);
1495
1496         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1497
1498         /* Neighbor Advertisement */
1499         memset(na, 0, sizeof(*na)+na_olen);
1500         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1501         na->icmph.icmp6_router = isrouter;
1502         na->icmph.icmp6_override = 1;
1503         na->icmph.icmp6_solicited = 1;
1504         na->target = ns->target;
1505         ether_addr_copy(&na->opt[2], n->ha);
1506         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1507         na->opt[1] = na_olen >> 3;
1508
1509         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1510                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1511                 csum_partial(na, sizeof(*na)+na_olen, 0));
1512
1513         pip6->payload_len = htons(sizeof(*na)+na_olen);
1514
1515         skb_push(reply, sizeof(struct ipv6hdr));
1516
1517         reply->ip_summed = CHECKSUM_UNNECESSARY;
1518
1519         return reply;
1520 }
1521
1522 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1523 {
1524         struct vxlan_dev *vxlan = netdev_priv(dev);
1525         struct nd_msg *msg;
1526         const struct ipv6hdr *iphdr;
1527         const struct in6_addr *saddr, *daddr;
1528         struct neighbour *n;
1529         struct inet6_dev *in6_dev;
1530
1531         in6_dev = __in6_dev_get(dev);
1532         if (!in6_dev)
1533                 goto out;
1534
1535         iphdr = ipv6_hdr(skb);
1536         saddr = &iphdr->saddr;
1537         daddr = &iphdr->daddr;
1538
1539         msg = (struct nd_msg *)skb_transport_header(skb);
1540         if (msg->icmph.icmp6_code != 0 ||
1541             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1542                 goto out;
1543
1544         if (ipv6_addr_loopback(daddr) ||
1545             ipv6_addr_is_multicast(&msg->target))
1546                 goto out;
1547
1548         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1549
1550         if (n) {
1551                 struct vxlan_fdb *f;
1552                 struct sk_buff *reply;
1553
1554                 if (!(n->nud_state & NUD_CONNECTED)) {
1555                         neigh_release(n);
1556                         goto out;
1557                 }
1558
1559                 f = vxlan_find_mac(vxlan, n->ha);
1560                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1561                         /* bridge-local neighbor */
1562                         neigh_release(n);
1563                         goto out;
1564                 }
1565
1566                 reply = vxlan_na_create(skb, n,
1567                                         !!(f ? f->flags & NTF_ROUTER : 0));
1568
1569                 neigh_release(n);
1570
1571                 if (reply == NULL)
1572                         goto out;
1573
1574                 if (netif_rx_ni(reply) == NET_RX_DROP)
1575                         dev->stats.rx_dropped++;
1576
1577         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1578                 union vxlan_addr ipa = {
1579                         .sin6.sin6_addr = msg->target,
1580                         .sin6.sin6_family = AF_INET6,
1581                 };
1582
1583                 vxlan_ip_miss(dev, &ipa);
1584         }
1585
1586 out:
1587         consume_skb(skb);
1588         return NETDEV_TX_OK;
1589 }
1590 #endif
1591
1592 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1593 {
1594         struct vxlan_dev *vxlan = netdev_priv(dev);
1595         struct neighbour *n;
1596
1597         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1598                 return false;
1599
1600         n = NULL;
1601         switch (ntohs(eth_hdr(skb)->h_proto)) {
1602         case ETH_P_IP:
1603         {
1604                 struct iphdr *pip;
1605
1606                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1607                         return false;
1608                 pip = ip_hdr(skb);
1609                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1610                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1611                         union vxlan_addr ipa = {
1612                                 .sin.sin_addr.s_addr = pip->daddr,
1613                                 .sin.sin_family = AF_INET,
1614                         };
1615
1616                         vxlan_ip_miss(dev, &ipa);
1617                         return false;
1618                 }
1619
1620                 break;
1621         }
1622 #if IS_ENABLED(CONFIG_IPV6)
1623         case ETH_P_IPV6:
1624         {
1625                 struct ipv6hdr *pip6;
1626
1627                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1628                         return false;
1629                 pip6 = ipv6_hdr(skb);
1630                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1631                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1632                         union vxlan_addr ipa = {
1633                                 .sin6.sin6_addr = pip6->daddr,
1634                                 .sin6.sin6_family = AF_INET6,
1635                         };
1636
1637                         vxlan_ip_miss(dev, &ipa);
1638                         return false;
1639                 }
1640
1641                 break;
1642         }
1643 #endif
1644         default:
1645                 return false;
1646         }
1647
1648         if (n) {
1649                 bool diff;
1650
1651                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1652                 if (diff) {
1653                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1654                                 dev->addr_len);
1655                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1656                 }
1657                 neigh_release(n);
1658                 return diff;
1659         }
1660
1661         return false;
1662 }
1663
1664 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1665                                 struct vxlan_metadata *md)
1666 {
1667         struct vxlanhdr_gbp *gbp;
1668
1669         if (!md->gbp)
1670                 return;
1671
1672         gbp = (struct vxlanhdr_gbp *)vxh;
1673         vxh->vx_flags |= htonl(VXLAN_HF_GBP);
1674
1675         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1676                 gbp->dont_learn = 1;
1677
1678         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1679                 gbp->policy_applied = 1;
1680
1681         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1682 }
1683
1684 #if IS_ENABLED(CONFIG_IPV6)
1685 static int vxlan6_xmit_skb(struct dst_entry *dst, struct sock *sk,
1686                            struct sk_buff *skb,
1687                            struct net_device *dev, struct in6_addr *saddr,
1688                            struct in6_addr *daddr, __u8 prio, __u8 ttl,
1689                            __be16 src_port, __be16 dst_port, __be32 vni,
1690                            struct vxlan_metadata *md, bool xnet, u32 vxflags)
1691 {
1692         struct vxlanhdr *vxh;
1693         int min_headroom;
1694         int err;
1695         bool udp_sum = !(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX);
1696         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1697         u16 hdrlen = sizeof(struct vxlanhdr);
1698
1699         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1700             skb->ip_summed == CHECKSUM_PARTIAL) {
1701                 int csum_start = skb_checksum_start_offset(skb);
1702
1703                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1704                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1705                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1706                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1707                         udp_sum = false;
1708                         type |= SKB_GSO_TUNNEL_REMCSUM;
1709                 }
1710         }
1711
1712         skb_scrub_packet(skb, xnet);
1713
1714         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1715                         + VXLAN_HLEN + sizeof(struct ipv6hdr)
1716                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1717
1718         /* Need space for new headers (invalidates iph ptr) */
1719         err = skb_cow_head(skb, min_headroom);
1720         if (unlikely(err)) {
1721                 kfree_skb(skb);
1722                 goto err;
1723         }
1724
1725         skb = vlan_hwaccel_push_inside(skb);
1726         if (WARN_ON(!skb)) {
1727                 err = -ENOMEM;
1728                 goto err;
1729         }
1730
1731         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1732         if (IS_ERR(skb)) {
1733                 err = -EINVAL;
1734                 goto err;
1735         }
1736
1737         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1738         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1739         vxh->vx_vni = vni;
1740
1741         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1742                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1743                            VXLAN_RCO_SHIFT;
1744
1745                 if (skb->csum_offset == offsetof(struct udphdr, check))
1746                         data |= VXLAN_RCO_UDP;
1747
1748                 vxh->vx_vni |= htonl(data);
1749                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1750
1751                 if (!skb_is_gso(skb)) {
1752                         skb->ip_summed = CHECKSUM_NONE;
1753                         skb->encapsulation = 0;
1754                 }
1755         }
1756
1757         if (vxflags & VXLAN_F_GBP)
1758                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1759
1760         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1761
1762         udp_tunnel6_xmit_skb(dst, sk, skb, dev, saddr, daddr, prio,
1763                              ttl, src_port, dst_port,
1764                              !!(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX));
1765         return 0;
1766 err:
1767         dst_release(dst);
1768         return err;
1769 }
1770 #endif
1771
1772 static int vxlan_xmit_skb(struct rtable *rt, struct sock *sk, struct sk_buff *skb,
1773                           __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1774                           __be16 src_port, __be16 dst_port, __be32 vni,
1775                           struct vxlan_metadata *md, bool xnet, u32 vxflags)
1776 {
1777         struct vxlanhdr *vxh;
1778         int min_headroom;
1779         int err;
1780         bool udp_sum = !!(vxflags & VXLAN_F_UDP_CSUM);
1781         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1782         u16 hdrlen = sizeof(struct vxlanhdr);
1783
1784         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1785             skb->ip_summed == CHECKSUM_PARTIAL) {
1786                 int csum_start = skb_checksum_start_offset(skb);
1787
1788                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1789                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1790                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1791                      skb->csum_offset == offsetof(struct tcphdr, check))) {
1792                         udp_sum = false;
1793                         type |= SKB_GSO_TUNNEL_REMCSUM;
1794                 }
1795         }
1796
1797         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1798                         + VXLAN_HLEN + sizeof(struct iphdr)
1799                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1800
1801         /* Need space for new headers (invalidates iph ptr) */
1802         err = skb_cow_head(skb, min_headroom);
1803         if (unlikely(err)) {
1804                 kfree_skb(skb);
1805                 return err;
1806         }
1807
1808         skb = vlan_hwaccel_push_inside(skb);
1809         if (WARN_ON(!skb))
1810                 return -ENOMEM;
1811
1812         skb = iptunnel_handle_offloads(skb, udp_sum, type);
1813         if (IS_ERR(skb))
1814                 return PTR_ERR(skb);
1815
1816         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1817         vxh->vx_flags = htonl(VXLAN_HF_VNI);
1818         vxh->vx_vni = vni;
1819
1820         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1821                 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1822                            VXLAN_RCO_SHIFT;
1823
1824                 if (skb->csum_offset == offsetof(struct udphdr, check))
1825                         data |= VXLAN_RCO_UDP;
1826
1827                 vxh->vx_vni |= htonl(data);
1828                 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1829
1830                 if (!skb_is_gso(skb)) {
1831                         skb->ip_summed = CHECKSUM_NONE;
1832                         skb->encapsulation = 0;
1833                 }
1834         }
1835
1836         if (vxflags & VXLAN_F_GBP)
1837                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1838
1839         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1840
1841         return udp_tunnel_xmit_skb(rt, sk, skb, src, dst, tos,
1842                                    ttl, df, src_port, dst_port, xnet,
1843                                    !(vxflags & VXLAN_F_UDP_CSUM));
1844 }
1845
1846 /* Bypass encapsulation if the destination is local */
1847 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1848                                struct vxlan_dev *dst_vxlan)
1849 {
1850         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1851         union vxlan_addr loopback;
1852         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1853         struct net_device *dev = skb->dev;
1854         int len = skb->len;
1855
1856         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1857         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1858         skb->pkt_type = PACKET_HOST;
1859         skb->encapsulation = 0;
1860         skb->dev = dst_vxlan->dev;
1861         __skb_pull(skb, skb_network_offset(skb));
1862
1863         if (remote_ip->sa.sa_family == AF_INET) {
1864                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1865                 loopback.sa.sa_family =  AF_INET;
1866 #if IS_ENABLED(CONFIG_IPV6)
1867         } else {
1868                 loopback.sin6.sin6_addr = in6addr_loopback;
1869                 loopback.sa.sa_family =  AF_INET6;
1870 #endif
1871         }
1872
1873         if (dst_vxlan->flags & VXLAN_F_LEARN)
1874                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1875
1876         u64_stats_update_begin(&tx_stats->syncp);
1877         tx_stats->tx_packets++;
1878         tx_stats->tx_bytes += len;
1879         u64_stats_update_end(&tx_stats->syncp);
1880
1881         if (netif_rx(skb) == NET_RX_SUCCESS) {
1882                 u64_stats_update_begin(&rx_stats->syncp);
1883                 rx_stats->rx_packets++;
1884                 rx_stats->rx_bytes += len;
1885                 u64_stats_update_end(&rx_stats->syncp);
1886         } else {
1887                 dev->stats.rx_dropped++;
1888         }
1889 }
1890
1891 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1892                            struct vxlan_rdst *rdst, bool did_rsc)
1893 {
1894         struct ip_tunnel_info *info;
1895         struct vxlan_dev *vxlan = netdev_priv(dev);
1896         struct sock *sk = vxlan->vn_sock->sock->sk;
1897         struct rtable *rt = NULL;
1898         const struct iphdr *old_iph;
1899         struct flowi4 fl4;
1900         union vxlan_addr *dst;
1901         union vxlan_addr remote_ip;
1902         struct vxlan_metadata _md;
1903         struct vxlan_metadata *md = &_md;
1904         __be16 src_port = 0, dst_port;
1905         u32 vni;
1906         __be16 df = 0;
1907         __u8 tos, ttl;
1908         int err;
1909         u32 flags = vxlan->flags;
1910
1911         /* FIXME: Support IPv6 */
1912         info = skb_tunnel_info(skb);
1913
1914         if (rdst) {
1915                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1916                 vni = rdst->remote_vni;
1917                 dst = &rdst->remote_ip;
1918         } else {
1919                 if (!info) {
1920                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1921                                   dev->name);
1922                         goto drop;
1923                 }
1924
1925                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1926                 vni = be64_to_cpu(info->key.tun_id);
1927                 remote_ip.sin.sin_family = AF_INET;
1928                 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1929                 dst = &remote_ip;
1930         }
1931
1932         if (vxlan_addr_any(dst)) {
1933                 if (did_rsc) {
1934                         /* short-circuited back to local bridge */
1935                         vxlan_encap_bypass(skb, vxlan, vxlan);
1936                         return;
1937                 }
1938                 goto drop;
1939         }
1940
1941         old_iph = ip_hdr(skb);
1942
1943         ttl = vxlan->cfg.ttl;
1944         if (!ttl && vxlan_addr_multicast(dst))
1945                 ttl = 1;
1946
1947         tos = vxlan->cfg.tos;
1948         if (tos == 1)
1949                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1950
1951         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
1952                                      vxlan->cfg.port_max, true);
1953
1954         if (dst->sa.sa_family == AF_INET) {
1955                 if (info) {
1956                         if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
1957                                 df = htons(IP_DF);
1958                         if (info->key.tun_flags & TUNNEL_CSUM)
1959                                 flags |= VXLAN_F_UDP_CSUM;
1960                         else
1961                                 flags &= ~VXLAN_F_UDP_CSUM;
1962
1963                         ttl = info->key.ttl;
1964                         tos = info->key.tos;
1965
1966                         if (info->options_len)
1967                                 md = ip_tunnel_info_opts(info, sizeof(*md));
1968                 } else {
1969                         md->gbp = skb->mark;
1970                 }
1971
1972                 memset(&fl4, 0, sizeof(fl4));
1973                 fl4.flowi4_oif = rdst ? rdst->remote_ifindex : 0;
1974                 fl4.flowi4_tos = RT_TOS(tos);
1975                 fl4.flowi4_mark = skb->mark;
1976                 fl4.flowi4_proto = IPPROTO_UDP;
1977                 fl4.daddr = dst->sin.sin_addr.s_addr;
1978                 fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
1979
1980                 rt = ip_route_output_key(vxlan->net, &fl4);
1981                 if (IS_ERR(rt)) {
1982                         netdev_dbg(dev, "no route to %pI4\n",
1983                                    &dst->sin.sin_addr.s_addr);
1984                         dev->stats.tx_carrier_errors++;
1985                         goto tx_error;
1986                 }
1987
1988                 if (rt->dst.dev == dev) {
1989                         netdev_dbg(dev, "circular route to %pI4\n",
1990                                    &dst->sin.sin_addr.s_addr);
1991                         dev->stats.collisions++;
1992                         goto rt_tx_error;
1993                 }
1994
1995                 /* Bypass encapsulation if the destination is local */
1996                 if (rt->rt_flags & RTCF_LOCAL &&
1997                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1998                         struct vxlan_dev *dst_vxlan;
1999
2000                         ip_rt_put(rt);
2001                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2002                                                    dst->sa.sa_family, dst_port,
2003                                                    vxlan->flags);
2004                         if (!dst_vxlan)
2005                                 goto tx_error;
2006                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2007                         return;
2008                 }
2009
2010                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2011                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2012                 err = vxlan_xmit_skb(rt, sk, skb, fl4.saddr,
2013                                      dst->sin.sin_addr.s_addr, tos, ttl, df,
2014                                      src_port, dst_port, htonl(vni << 8), md,
2015                                      !net_eq(vxlan->net, dev_net(vxlan->dev)),
2016                                      flags);
2017                 if (err < 0) {
2018                         /* skb is already freed. */
2019                         skb = NULL;
2020                         goto rt_tx_error;
2021                 }
2022
2023                 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
2024 #if IS_ENABLED(CONFIG_IPV6)
2025         } else {
2026                 struct dst_entry *ndst;
2027                 struct flowi6 fl6;
2028                 u32 rt6i_flags;
2029
2030                 memset(&fl6, 0, sizeof(fl6));
2031                 fl6.flowi6_oif = rdst ? rdst->remote_ifindex : 0;
2032                 fl6.daddr = dst->sin6.sin6_addr;
2033                 fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
2034                 fl6.flowi6_mark = skb->mark;
2035                 fl6.flowi6_proto = IPPROTO_UDP;
2036
2037                 if (ipv6_stub->ipv6_dst_lookup(vxlan->net, sk, &ndst, &fl6)) {
2038                         netdev_dbg(dev, "no route to %pI6\n",
2039                                    &dst->sin6.sin6_addr);
2040                         dev->stats.tx_carrier_errors++;
2041                         goto tx_error;
2042                 }
2043
2044                 if (ndst->dev == dev) {
2045                         netdev_dbg(dev, "circular route to %pI6\n",
2046                                    &dst->sin6.sin6_addr);
2047                         dst_release(ndst);
2048                         dev->stats.collisions++;
2049                         goto tx_error;
2050                 }
2051
2052                 /* Bypass encapsulation if the destination is local */
2053                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2054                 if (rt6i_flags & RTF_LOCAL &&
2055                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2056                         struct vxlan_dev *dst_vxlan;
2057
2058                         dst_release(ndst);
2059                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2060                                                    dst->sa.sa_family, dst_port,
2061                                                    vxlan->flags);
2062                         if (!dst_vxlan)
2063                                 goto tx_error;
2064                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2065                         return;
2066                 }
2067
2068                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2069                 md->gbp = skb->mark;
2070
2071                 err = vxlan6_xmit_skb(ndst, sk, skb, dev, &fl6.saddr, &fl6.daddr,
2072                                       0, ttl, src_port, dst_port, htonl(vni << 8), md,
2073                                       !net_eq(vxlan->net, dev_net(vxlan->dev)),
2074                                       vxlan->flags);
2075 #endif
2076         }
2077
2078         return;
2079
2080 drop:
2081         dev->stats.tx_dropped++;
2082         goto tx_free;
2083
2084 rt_tx_error:
2085         ip_rt_put(rt);
2086 tx_error:
2087         dev->stats.tx_errors++;
2088 tx_free:
2089         dev_kfree_skb(skb);
2090 }
2091
2092 /* Transmit local packets over Vxlan
2093  *
2094  * Outer IP header inherits ECN and DF from inner header.
2095  * Outer UDP destination is the VXLAN assigned port.
2096  *           source port is based on hash of flow
2097  */
2098 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2099 {
2100         struct vxlan_dev *vxlan = netdev_priv(dev);
2101         const struct ip_tunnel_info *info;
2102         struct ethhdr *eth;
2103         bool did_rsc = false;
2104         struct vxlan_rdst *rdst, *fdst = NULL;
2105         struct vxlan_fdb *f;
2106
2107         /* FIXME: Support IPv6 */
2108         info = skb_tunnel_info(skb);
2109
2110         skb_reset_mac_header(skb);
2111         eth = eth_hdr(skb);
2112
2113         if ((vxlan->flags & VXLAN_F_PROXY)) {
2114                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2115                         return arp_reduce(dev, skb);
2116 #if IS_ENABLED(CONFIG_IPV6)
2117                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2118                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2119                                        + sizeof(struct nd_msg)) &&
2120                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2121                                 struct nd_msg *msg;
2122
2123                                 msg = (struct nd_msg *)skb_transport_header(skb);
2124                                 if (msg->icmph.icmp6_code == 0 &&
2125                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2126                                         return neigh_reduce(dev, skb);
2127                 }
2128                 eth = eth_hdr(skb);
2129 #endif
2130         }
2131
2132         if (vxlan->flags & VXLAN_F_COLLECT_METADATA &&
2133             info && info->mode == IP_TUNNEL_INFO_TX) {
2134                 vxlan_xmit_one(skb, dev, NULL, false);
2135                 return NETDEV_TX_OK;
2136         }
2137
2138         f = vxlan_find_mac(vxlan, eth->h_dest);
2139         did_rsc = false;
2140
2141         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2142             (ntohs(eth->h_proto) == ETH_P_IP ||
2143              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2144                 did_rsc = route_shortcircuit(dev, skb);
2145                 if (did_rsc)
2146                         f = vxlan_find_mac(vxlan, eth->h_dest);
2147         }
2148
2149         if (f == NULL) {
2150                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2151                 if (f == NULL) {
2152                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2153                             !is_multicast_ether_addr(eth->h_dest))
2154                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2155
2156                         dev->stats.tx_dropped++;
2157                         kfree_skb(skb);
2158                         return NETDEV_TX_OK;
2159                 }
2160         }
2161
2162         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2163                 struct sk_buff *skb1;
2164
2165                 if (!fdst) {
2166                         fdst = rdst;
2167                         continue;
2168                 }
2169                 skb1 = skb_clone(skb, GFP_ATOMIC);
2170                 if (skb1)
2171                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2172         }
2173
2174         if (fdst)
2175                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2176         else
2177                 kfree_skb(skb);
2178         return NETDEV_TX_OK;
2179 }
2180
2181 /* Walk the forwarding table and purge stale entries */
2182 static void vxlan_cleanup(unsigned long arg)
2183 {
2184         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2185         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2186         unsigned int h;
2187
2188         if (!netif_running(vxlan->dev))
2189                 return;
2190
2191         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2192                 struct hlist_node *p, *n;
2193
2194                 spin_lock_bh(&vxlan->hash_lock);
2195                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2196                         struct vxlan_fdb *f
2197                                 = container_of(p, struct vxlan_fdb, hlist);
2198                         unsigned long timeout;
2199
2200                         if (f->state & NUD_PERMANENT)
2201                                 continue;
2202
2203                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2204                         if (time_before_eq(timeout, jiffies)) {
2205                                 netdev_dbg(vxlan->dev,
2206                                            "garbage collect %pM\n",
2207                                            f->eth_addr);
2208                                 f->state = NUD_STALE;
2209                                 vxlan_fdb_destroy(vxlan, f);
2210                         } else if (time_before(timeout, next_timer))
2211                                 next_timer = timeout;
2212                 }
2213                 spin_unlock_bh(&vxlan->hash_lock);
2214         }
2215
2216         mod_timer(&vxlan->age_timer, next_timer);
2217 }
2218
2219 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2220 {
2221         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2222         __u32 vni = vxlan->default_dst.remote_vni;
2223
2224         vxlan->vn_sock = vs;
2225         spin_lock(&vn->sock_lock);
2226         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2227         spin_unlock(&vn->sock_lock);
2228 }
2229
2230 /* Setup stats when device is created */
2231 static int vxlan_init(struct net_device *dev)
2232 {
2233         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2234         if (!dev->tstats)
2235                 return -ENOMEM;
2236
2237         return 0;
2238 }
2239
2240 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2241 {
2242         struct vxlan_fdb *f;
2243
2244         spin_lock_bh(&vxlan->hash_lock);
2245         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2246         if (f)
2247                 vxlan_fdb_destroy(vxlan, f);
2248         spin_unlock_bh(&vxlan->hash_lock);
2249 }
2250
2251 static void vxlan_uninit(struct net_device *dev)
2252 {
2253         struct vxlan_dev *vxlan = netdev_priv(dev);
2254
2255         vxlan_fdb_delete_default(vxlan);
2256
2257         free_percpu(dev->tstats);
2258 }
2259
2260 /* Start ageing timer and join group when device is brought up */
2261 static int vxlan_open(struct net_device *dev)
2262 {
2263         struct vxlan_dev *vxlan = netdev_priv(dev);
2264         struct vxlan_sock *vs;
2265         int ret = 0;
2266
2267         vs = vxlan_sock_add(vxlan->net, vxlan->cfg.dst_port,
2268                             vxlan->cfg.no_share, vxlan->flags);
2269         if (IS_ERR(vs))
2270                 return PTR_ERR(vs);
2271
2272         vxlan_vs_add_dev(vs, vxlan);
2273
2274         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2275                 ret = vxlan_igmp_join(vxlan);
2276                 if (ret) {
2277                         vxlan_sock_release(vs);
2278                         return ret;
2279                 }
2280         }
2281
2282         if (vxlan->cfg.age_interval)
2283                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2284
2285         return ret;
2286 }
2287
2288 /* Purge the forwarding table */
2289 static void vxlan_flush(struct vxlan_dev *vxlan)
2290 {
2291         unsigned int h;
2292
2293         spin_lock_bh(&vxlan->hash_lock);
2294         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2295                 struct hlist_node *p, *n;
2296                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2297                         struct vxlan_fdb *f
2298                                 = container_of(p, struct vxlan_fdb, hlist);
2299                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2300                         if (!is_zero_ether_addr(f->eth_addr))
2301                                 vxlan_fdb_destroy(vxlan, f);
2302                 }
2303         }
2304         spin_unlock_bh(&vxlan->hash_lock);
2305 }
2306
2307 /* Cleanup timer and forwarding table on shutdown */
2308 static int vxlan_stop(struct net_device *dev)
2309 {
2310         struct vxlan_dev *vxlan = netdev_priv(dev);
2311         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2312         struct vxlan_sock *vs = vxlan->vn_sock;
2313         int ret = 0;
2314
2315         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2316             !vxlan_group_used(vn, vxlan))
2317                 ret = vxlan_igmp_leave(vxlan);
2318
2319         del_timer_sync(&vxlan->age_timer);
2320
2321         vxlan_flush(vxlan);
2322         vxlan_sock_release(vs);
2323
2324         return ret;
2325 }
2326
2327 /* Stub, nothing needs to be done. */
2328 static void vxlan_set_multicast_list(struct net_device *dev)
2329 {
2330 }
2331
2332 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2333 {
2334         struct vxlan_dev *vxlan = netdev_priv(dev);
2335         struct vxlan_rdst *dst = &vxlan->default_dst;
2336         struct net_device *lowerdev;
2337         int max_mtu;
2338
2339         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2340         if (lowerdev == NULL)
2341                 return eth_change_mtu(dev, new_mtu);
2342
2343         if (dst->remote_ip.sa.sa_family == AF_INET6)
2344                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2345         else
2346                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2347
2348         if (new_mtu < 68 || new_mtu > max_mtu)
2349                 return -EINVAL;
2350
2351         dev->mtu = new_mtu;
2352         return 0;
2353 }
2354
2355 static const struct net_device_ops vxlan_netdev_ops = {
2356         .ndo_init               = vxlan_init,
2357         .ndo_uninit             = vxlan_uninit,
2358         .ndo_open               = vxlan_open,
2359         .ndo_stop               = vxlan_stop,
2360         .ndo_start_xmit         = vxlan_xmit,
2361         .ndo_get_stats64        = ip_tunnel_get_stats64,
2362         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2363         .ndo_change_mtu         = vxlan_change_mtu,
2364         .ndo_validate_addr      = eth_validate_addr,
2365         .ndo_set_mac_address    = eth_mac_addr,
2366         .ndo_fdb_add            = vxlan_fdb_add,
2367         .ndo_fdb_del            = vxlan_fdb_delete,
2368         .ndo_fdb_dump           = vxlan_fdb_dump,
2369 };
2370
2371 /* Info for udev, that this is a virtual tunnel endpoint */
2372 static struct device_type vxlan_type = {
2373         .name = "vxlan",
2374 };
2375
2376 /* Calls the ndo_add_vxlan_port of the caller in order to
2377  * supply the listening VXLAN udp ports. Callers are expected
2378  * to implement the ndo_add_vxlan_port.
2379  */
2380 void vxlan_get_rx_port(struct net_device *dev)
2381 {
2382         struct vxlan_sock *vs;
2383         struct net *net = dev_net(dev);
2384         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2385         sa_family_t sa_family;
2386         __be16 port;
2387         unsigned int i;
2388
2389         spin_lock(&vn->sock_lock);
2390         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2391                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2392                         port = inet_sk(vs->sock->sk)->inet_sport;
2393                         sa_family = vxlan_get_sk_family(vs);
2394                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2395                                                             port);
2396                 }
2397         }
2398         spin_unlock(&vn->sock_lock);
2399 }
2400 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2401
2402 /* Initialize the device structure. */
2403 static void vxlan_setup(struct net_device *dev)
2404 {
2405         struct vxlan_dev *vxlan = netdev_priv(dev);
2406         unsigned int h;
2407
2408         eth_hw_addr_random(dev);
2409         ether_setup(dev);
2410         if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2411                 dev->needed_headroom = ETH_HLEN + VXLAN6_HEADROOM;
2412         else
2413                 dev->needed_headroom = ETH_HLEN + VXLAN_HEADROOM;
2414
2415         dev->netdev_ops = &vxlan_netdev_ops;
2416         dev->destructor = free_netdev;
2417         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2418
2419         dev->features   |= NETIF_F_LLTX;
2420         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2421         dev->features   |= NETIF_F_RXCSUM;
2422         dev->features   |= NETIF_F_GSO_SOFTWARE;
2423
2424         dev->vlan_features = dev->features;
2425         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2426         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2427         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2428         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2429         netif_keep_dst(dev);
2430         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
2431
2432         INIT_LIST_HEAD(&vxlan->next);
2433         spin_lock_init(&vxlan->hash_lock);
2434
2435         init_timer_deferrable(&vxlan->age_timer);
2436         vxlan->age_timer.function = vxlan_cleanup;
2437         vxlan->age_timer.data = (unsigned long) vxlan;
2438
2439         vxlan->cfg.dst_port = htons(vxlan_port);
2440
2441         vxlan->dev = dev;
2442
2443         for (h = 0; h < FDB_HASH_SIZE; ++h)
2444                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2445 }
2446
2447 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2448         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2449         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2450         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2451         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2452         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2453         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2454         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2455         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2456         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2457         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2458         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2459         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2460         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2461         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2462         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2463         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2464         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2465         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2466         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2467         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2468         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2469         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2470         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2471         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2472         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2473 };
2474
2475 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2476 {
2477         if (tb[IFLA_ADDRESS]) {
2478                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2479                         pr_debug("invalid link address (not ethernet)\n");
2480                         return -EINVAL;
2481                 }
2482
2483                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2484                         pr_debug("invalid all zero ethernet address\n");
2485                         return -EADDRNOTAVAIL;
2486                 }
2487         }
2488
2489         if (!data)
2490                 return -EINVAL;
2491
2492         if (data[IFLA_VXLAN_ID]) {
2493                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2494                 if (id >= VXLAN_VID_MASK)
2495                         return -ERANGE;
2496         }
2497
2498         if (data[IFLA_VXLAN_PORT_RANGE]) {
2499                 const struct ifla_vxlan_port_range *p
2500                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2501
2502                 if (ntohs(p->high) < ntohs(p->low)) {
2503                         pr_debug("port range %u .. %u not valid\n",
2504                                  ntohs(p->low), ntohs(p->high));
2505                         return -EINVAL;
2506                 }
2507         }
2508
2509         return 0;
2510 }
2511
2512 static void vxlan_get_drvinfo(struct net_device *netdev,
2513                               struct ethtool_drvinfo *drvinfo)
2514 {
2515         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2516         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2517 }
2518
2519 static const struct ethtool_ops vxlan_ethtool_ops = {
2520         .get_drvinfo    = vxlan_get_drvinfo,
2521         .get_link       = ethtool_op_get_link,
2522 };
2523
2524 static void vxlan_del_work(struct work_struct *work)
2525 {
2526         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2527         udp_tunnel_sock_release(vs->sock);
2528         kfree_rcu(vs, rcu);
2529 }
2530
2531 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2532                                         __be16 port, u32 flags)
2533 {
2534         struct socket *sock;
2535         struct udp_port_cfg udp_conf;
2536         int err;
2537
2538         memset(&udp_conf, 0, sizeof(udp_conf));
2539
2540         if (ipv6) {
2541                 udp_conf.family = AF_INET6;
2542                 udp_conf.use_udp6_rx_checksums =
2543                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2544         } else {
2545                 udp_conf.family = AF_INET;
2546         }
2547
2548         udp_conf.local_udp_port = port;
2549
2550         /* Open UDP socket */
2551         err = udp_sock_create(net, &udp_conf, &sock);
2552         if (err < 0)
2553                 return ERR_PTR(err);
2554
2555         return sock;
2556 }
2557
2558 /* Create new listen socket if needed */
2559 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2560                                               u32 flags)
2561 {
2562         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2563         struct vxlan_sock *vs;
2564         struct socket *sock;
2565         unsigned int h;
2566         bool ipv6 = !!(flags & VXLAN_F_IPV6);
2567         struct udp_tunnel_sock_cfg tunnel_cfg;
2568
2569         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2570         if (!vs)
2571                 return ERR_PTR(-ENOMEM);
2572
2573         for (h = 0; h < VNI_HASH_SIZE; ++h)
2574                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2575
2576         INIT_WORK(&vs->del_work, vxlan_del_work);
2577
2578         sock = vxlan_create_sock(net, ipv6, port, flags);
2579         if (IS_ERR(sock)) {
2580                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2581                         PTR_ERR(sock));
2582                 kfree(vs);
2583                 return ERR_CAST(sock);
2584         }
2585
2586         vs->sock = sock;
2587         atomic_set(&vs->refcnt, 1);
2588         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2589
2590         /* Initialize the vxlan udp offloads structure */
2591         vs->udp_offloads.port = port;
2592         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2593         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2594
2595         spin_lock(&vn->sock_lock);
2596         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2597         vxlan_notify_add_rx_port(vs);
2598         spin_unlock(&vn->sock_lock);
2599
2600         /* Mark socket as an encapsulation socket. */
2601         tunnel_cfg.sk_user_data = vs;
2602         tunnel_cfg.encap_type = 1;
2603         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2604         tunnel_cfg.encap_destroy = NULL;
2605
2606         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2607
2608         return vs;
2609 }
2610
2611 static struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2612                                          bool no_share, u32 flags)
2613 {
2614         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2615         struct vxlan_sock *vs;
2616         bool ipv6 = flags & VXLAN_F_IPV6;
2617
2618         if (!no_share) {
2619                 spin_lock(&vn->sock_lock);
2620                 vs = vxlan_find_sock(net, ipv6 ? AF_INET6 : AF_INET, port,
2621                                      flags);
2622                 if (vs) {
2623                         if (!atomic_add_unless(&vs->refcnt, 1, 0))
2624                                 vs = ERR_PTR(-EBUSY);
2625                         spin_unlock(&vn->sock_lock);
2626                         return vs;
2627                 }
2628                 spin_unlock(&vn->sock_lock);
2629         }
2630
2631         return vxlan_socket_create(net, port, flags);
2632 }
2633
2634 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2635                                struct vxlan_config *conf)
2636 {
2637         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2638         struct vxlan_dev *vxlan = netdev_priv(dev);
2639         struct vxlan_rdst *dst = &vxlan->default_dst;
2640         int err;
2641         bool use_ipv6 = false;
2642         __be16 default_port = vxlan->cfg.dst_port;
2643
2644         vxlan->net = src_net;
2645
2646         dst->remote_vni = conf->vni;
2647
2648         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2649
2650         /* Unless IPv6 is explicitly requested, assume IPv4 */
2651         if (!dst->remote_ip.sa.sa_family)
2652                 dst->remote_ip.sa.sa_family = AF_INET;
2653
2654         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2655             vxlan->cfg.saddr.sa.sa_family == AF_INET6)
2656                 use_ipv6 = true;
2657
2658         if (conf->remote_ifindex) {
2659                 struct net_device *lowerdev
2660                          = __dev_get_by_index(src_net, conf->remote_ifindex);
2661
2662                 dst->remote_ifindex = conf->remote_ifindex;
2663
2664                 if (!lowerdev) {
2665                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2666                         return -ENODEV;
2667                 }
2668
2669 #if IS_ENABLED(CONFIG_IPV6)
2670                 if (use_ipv6) {
2671                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2672                         if (idev && idev->cnf.disable_ipv6) {
2673                                 pr_info("IPv6 is disabled via sysctl\n");
2674                                 return -EPERM;
2675                         }
2676                         vxlan->flags |= VXLAN_F_IPV6;
2677                 }
2678 #endif
2679
2680                 if (!conf->mtu)
2681                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2682
2683                 dev->needed_headroom = lowerdev->hard_header_len +
2684                                        (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2685         } else if (use_ipv6)
2686                 vxlan->flags |= VXLAN_F_IPV6;
2687
2688         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2689         if (!vxlan->cfg.dst_port)
2690                 vxlan->cfg.dst_port = default_port;
2691         vxlan->flags |= conf->flags;
2692
2693         if (!vxlan->cfg.age_interval)
2694                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2695
2696         if (vxlan_find_vni(src_net, conf->vni, use_ipv6 ? AF_INET6 : AF_INET,
2697                            vxlan->cfg.dst_port, vxlan->flags))
2698                 return -EEXIST;
2699
2700         dev->ethtool_ops = &vxlan_ethtool_ops;
2701
2702         /* create an fdb entry for a valid default destination */
2703         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2704                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2705                                        &vxlan->default_dst.remote_ip,
2706                                        NUD_REACHABLE|NUD_PERMANENT,
2707                                        NLM_F_EXCL|NLM_F_CREATE,
2708                                        vxlan->cfg.dst_port,
2709                                        vxlan->default_dst.remote_vni,
2710                                        vxlan->default_dst.remote_ifindex,
2711                                        NTF_SELF);
2712                 if (err)
2713                         return err;
2714         }
2715
2716         err = register_netdevice(dev);
2717         if (err) {
2718                 vxlan_fdb_delete_default(vxlan);
2719                 return err;
2720         }
2721
2722         list_add(&vxlan->next, &vn->vxlan_list);
2723
2724         return 0;
2725 }
2726
2727 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2728                                     u8 name_assign_type, struct vxlan_config *conf)
2729 {
2730         struct nlattr *tb[IFLA_MAX+1];
2731         struct net_device *dev;
2732         int err;
2733
2734         memset(&tb, 0, sizeof(tb));
2735
2736         dev = rtnl_create_link(net, name, name_assign_type,
2737                                &vxlan_link_ops, tb);
2738         if (IS_ERR(dev))
2739                 return dev;
2740
2741         err = vxlan_dev_configure(net, dev, conf);
2742         if (err < 0) {
2743                 free_netdev(dev);
2744                 return ERR_PTR(err);
2745         }
2746
2747         return dev;
2748 }
2749 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2750
2751 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2752                          struct nlattr *tb[], struct nlattr *data[])
2753 {
2754         struct vxlan_config conf;
2755         int err;
2756
2757         if (!data[IFLA_VXLAN_ID])
2758                 return -EINVAL;
2759
2760         memset(&conf, 0, sizeof(conf));
2761         conf.vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2762
2763         if (data[IFLA_VXLAN_GROUP]) {
2764                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2765         } else if (data[IFLA_VXLAN_GROUP6]) {
2766                 if (!IS_ENABLED(CONFIG_IPV6))
2767                         return -EPFNOSUPPORT;
2768
2769                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2770                 conf.remote_ip.sa.sa_family = AF_INET6;
2771         }
2772
2773         if (data[IFLA_VXLAN_LOCAL]) {
2774                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2775                 conf.saddr.sa.sa_family = AF_INET;
2776         } else if (data[IFLA_VXLAN_LOCAL6]) {
2777                 if (!IS_ENABLED(CONFIG_IPV6))
2778                         return -EPFNOSUPPORT;
2779
2780                 /* TODO: respect scope id */
2781                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2782                 conf.saddr.sa.sa_family = AF_INET6;
2783         }
2784
2785         if (data[IFLA_VXLAN_LINK])
2786                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2787
2788         if (data[IFLA_VXLAN_TOS])
2789                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2790
2791         if (data[IFLA_VXLAN_TTL])
2792                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2793
2794         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2795                 conf.flags |= VXLAN_F_LEARN;
2796
2797         if (data[IFLA_VXLAN_AGEING])
2798                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2799
2800         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2801                 conf.flags |= VXLAN_F_PROXY;
2802
2803         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2804                 conf.flags |= VXLAN_F_RSC;
2805
2806         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2807                 conf.flags |= VXLAN_F_L2MISS;
2808
2809         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2810                 conf.flags |= VXLAN_F_L3MISS;
2811
2812         if (data[IFLA_VXLAN_LIMIT])
2813                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2814
2815         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2816             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2817                 conf.flags |= VXLAN_F_COLLECT_METADATA;
2818
2819         if (data[IFLA_VXLAN_PORT_RANGE]) {
2820                 const struct ifla_vxlan_port_range *p
2821                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2822                 conf.port_min = ntohs(p->low);
2823                 conf.port_max = ntohs(p->high);
2824         }
2825
2826         if (data[IFLA_VXLAN_PORT])
2827                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2828
2829         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2830                 conf.flags |= VXLAN_F_UDP_CSUM;
2831
2832         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2833             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2834                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2835
2836         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2837             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2838                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2839
2840         if (data[IFLA_VXLAN_REMCSUM_TX] &&
2841             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2842                 conf.flags |= VXLAN_F_REMCSUM_TX;
2843
2844         if (data[IFLA_VXLAN_REMCSUM_RX] &&
2845             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2846                 conf.flags |= VXLAN_F_REMCSUM_RX;
2847
2848         if (data[IFLA_VXLAN_GBP])
2849                 conf.flags |= VXLAN_F_GBP;
2850
2851         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2852                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2853
2854         err = vxlan_dev_configure(src_net, dev, &conf);
2855         switch (err) {
2856         case -ENODEV:
2857                 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
2858                 break;
2859
2860         case -EPERM:
2861                 pr_info("IPv6 is disabled via sysctl\n");
2862                 break;
2863
2864         case -EEXIST:
2865                 pr_info("duplicate VNI %u\n", conf.vni);
2866                 break;
2867         }
2868
2869         return err;
2870 }
2871
2872 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2873 {
2874         struct vxlan_dev *vxlan = netdev_priv(dev);
2875         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2876
2877         spin_lock(&vn->sock_lock);
2878         if (!hlist_unhashed(&vxlan->hlist))
2879                 hlist_del_rcu(&vxlan->hlist);
2880         spin_unlock(&vn->sock_lock);
2881
2882         list_del(&vxlan->next);
2883         unregister_netdevice_queue(dev, head);
2884 }
2885
2886 static size_t vxlan_get_size(const struct net_device *dev)
2887 {
2888
2889         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2890                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2891                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2892                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2893                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2894                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2895                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2896                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2897                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2898                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2899                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2900                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
2901                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2902                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2903                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2904                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2905                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2906                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2907                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2908                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
2909                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
2910                 0;
2911 }
2912
2913 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2914 {
2915         const struct vxlan_dev *vxlan = netdev_priv(dev);
2916         const struct vxlan_rdst *dst = &vxlan->default_dst;
2917         struct ifla_vxlan_port_range ports = {
2918                 .low =  htons(vxlan->cfg.port_min),
2919                 .high = htons(vxlan->cfg.port_max),
2920         };
2921
2922         if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2923                 goto nla_put_failure;
2924
2925         if (!vxlan_addr_any(&dst->remote_ip)) {
2926                 if (dst->remote_ip.sa.sa_family == AF_INET) {
2927                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
2928                                             dst->remote_ip.sin.sin_addr.s_addr))
2929                                 goto nla_put_failure;
2930 #if IS_ENABLED(CONFIG_IPV6)
2931                 } else {
2932                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
2933                                              &dst->remote_ip.sin6.sin6_addr))
2934                                 goto nla_put_failure;
2935 #endif
2936                 }
2937         }
2938
2939         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2940                 goto nla_put_failure;
2941
2942         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
2943                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
2944                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
2945                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
2946                                 goto nla_put_failure;
2947 #if IS_ENABLED(CONFIG_IPV6)
2948                 } else {
2949                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
2950                                              &vxlan->cfg.saddr.sin6.sin6_addr))
2951                                 goto nla_put_failure;
2952 #endif
2953                 }
2954         }
2955
2956         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
2957             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
2958             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2959                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
2960             nla_put_u8(skb, IFLA_VXLAN_PROXY,
2961                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
2962             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2963             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2964                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2965             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2966                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2967             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
2968                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
2969             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
2970             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
2971             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
2972             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2973                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2974             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2975                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2976             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2977                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
2978             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
2979                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
2980             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
2981                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
2982                 goto nla_put_failure;
2983
2984         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2985                 goto nla_put_failure;
2986
2987         if (vxlan->flags & VXLAN_F_GBP &&
2988             nla_put_flag(skb, IFLA_VXLAN_GBP))
2989                 goto nla_put_failure;
2990
2991         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
2992             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
2993                 goto nla_put_failure;
2994
2995         return 0;
2996
2997 nla_put_failure:
2998         return -EMSGSIZE;
2999 }
3000
3001 static struct net *vxlan_get_link_net(const struct net_device *dev)
3002 {
3003         struct vxlan_dev *vxlan = netdev_priv(dev);
3004
3005         return vxlan->net;
3006 }
3007
3008 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3009         .kind           = "vxlan",
3010         .maxtype        = IFLA_VXLAN_MAX,
3011         .policy         = vxlan_policy,
3012         .priv_size      = sizeof(struct vxlan_dev),
3013         .setup          = vxlan_setup,
3014         .validate       = vxlan_validate,
3015         .newlink        = vxlan_newlink,
3016         .dellink        = vxlan_dellink,
3017         .get_size       = vxlan_get_size,
3018         .fill_info      = vxlan_fill_info,
3019         .get_link_net   = vxlan_get_link_net,
3020 };
3021
3022 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3023                                              struct net_device *dev)
3024 {
3025         struct vxlan_dev *vxlan, *next;
3026         LIST_HEAD(list_kill);
3027
3028         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3029                 struct vxlan_rdst *dst = &vxlan->default_dst;
3030
3031                 /* In case we created vxlan device with carrier
3032                  * and we loose the carrier due to module unload
3033                  * we also need to remove vxlan device. In other
3034                  * cases, it's not necessary and remote_ifindex
3035                  * is 0 here, so no matches.
3036                  */
3037                 if (dst->remote_ifindex == dev->ifindex)
3038                         vxlan_dellink(vxlan->dev, &list_kill);
3039         }
3040
3041         unregister_netdevice_many(&list_kill);
3042 }
3043
3044 static int vxlan_lowerdev_event(struct notifier_block *unused,
3045                                 unsigned long event, void *ptr)
3046 {
3047         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3048         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3049
3050         if (event == NETDEV_UNREGISTER)
3051                 vxlan_handle_lowerdev_unregister(vn, dev);
3052
3053         return NOTIFY_DONE;
3054 }
3055
3056 static struct notifier_block vxlan_notifier_block __read_mostly = {
3057         .notifier_call = vxlan_lowerdev_event,
3058 };
3059
3060 static __net_init int vxlan_init_net(struct net *net)
3061 {
3062         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3063         unsigned int h;
3064
3065         INIT_LIST_HEAD(&vn->vxlan_list);
3066         spin_lock_init(&vn->sock_lock);
3067
3068         for (h = 0; h < PORT_HASH_SIZE; ++h)
3069                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3070
3071         return 0;
3072 }
3073
3074 static void __net_exit vxlan_exit_net(struct net *net)
3075 {
3076         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3077         struct vxlan_dev *vxlan, *next;
3078         struct net_device *dev, *aux;
3079         LIST_HEAD(list);
3080
3081         rtnl_lock();
3082         for_each_netdev_safe(net, dev, aux)
3083                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3084                         unregister_netdevice_queue(dev, &list);
3085
3086         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3087                 /* If vxlan->dev is in the same netns, it has already been added
3088                  * to the list by the previous loop.
3089                  */
3090                 if (!net_eq(dev_net(vxlan->dev), net))
3091                         unregister_netdevice_queue(vxlan->dev, &list);
3092         }
3093
3094         unregister_netdevice_many(&list);
3095         rtnl_unlock();
3096 }
3097
3098 static struct pernet_operations vxlan_net_ops = {
3099         .init = vxlan_init_net,
3100         .exit = vxlan_exit_net,
3101         .id   = &vxlan_net_id,
3102         .size = sizeof(struct vxlan_net),
3103 };
3104
3105 static int __init vxlan_init_module(void)
3106 {
3107         int rc;
3108
3109         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3110         if (!vxlan_wq)
3111                 return -ENOMEM;
3112
3113         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3114
3115         rc = register_pernet_subsys(&vxlan_net_ops);
3116         if (rc)
3117                 goto out1;
3118
3119         rc = register_netdevice_notifier(&vxlan_notifier_block);
3120         if (rc)
3121                 goto out2;
3122
3123         rc = rtnl_link_register(&vxlan_link_ops);
3124         if (rc)
3125                 goto out3;
3126
3127         return 0;
3128 out3:
3129         unregister_netdevice_notifier(&vxlan_notifier_block);
3130 out2:
3131         unregister_pernet_subsys(&vxlan_net_ops);
3132 out1:
3133         destroy_workqueue(vxlan_wq);
3134         return rc;
3135 }
3136 late_initcall(vxlan_init_module);
3137
3138 static void __exit vxlan_cleanup_module(void)
3139 {
3140         rtnl_link_unregister(&vxlan_link_ops);
3141         unregister_netdevice_notifier(&vxlan_notifier_block);
3142         destroy_workqueue(vxlan_wq);
3143         unregister_pernet_subsys(&vxlan_net_ops);
3144         /* rcu_barrier() is called by netns */
3145 }
3146 module_exit(vxlan_cleanup_module);
3147
3148 MODULE_LICENSE("GPL");
3149 MODULE_VERSION(VXLAN_VERSION);
3150 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3151 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3152 MODULE_ALIAS_RTNL_LINK("vxlan");