drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/types.h>
  15 #include <linux/module.h>
  16 #include <linux/errno.h>
  17 #include <linux/slab.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/rculist.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/in.h>
  22 #include <linux/ip.h>
  23 #include <linux/udp.h>
  24 #include <linux/igmp.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/hash.h>
  29 #include <linux/ethtool.h>
  30 #include <net/arp.h>
  31 #include <net/ndisc.h>
  32 #include <net/ip.h>
  33 #include <net/ip_tunnels.h>
  34 #include <net/icmp.h>
  35 #include <net/udp.h>
  36 #include <net/udp_tunnel.h>
  37 #include <net/rtnetlink.h>
  38 #include <net/route.h>
  39 #include <net/dsfield.h>
  40 #include <net/inet_ecn.h>
  41 #include <net/net_namespace.h>
  42 #include <net/netns/generic.h>
  43 #include <net/vxlan.h>
  44 #include <net/protocol.h>
  45
  46 #if IS_ENABLED(CONFIG_IPV6)
  47 #include <net/ipv6.h>
  48 #include <net/addrconf.h>
  49 #include <net/ip6_tunnel.h>
  50 #include <net/ip6_checksum.h>
  51 #endif
  52 #include <net/dst_metadata.h>
  53
  54 #define VXLAN_VERSION   "0.1"
  55
  56 #define PORT_HASH_BITS  8
  57 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  58 #define FDB_AGE_DEFAULT 300 /* 5 min */
  59 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  60
  61 /* UDP port for VXLAN traffic.
  62  * The IANA assigned port is 4789, but the Linux default is 8472
  63  * for compatibility with early adopters.
  64  */
  65 static unsigned short vxlan_port __read_mostly = 8472;
  66 module_param_named(udp_port, vxlan_port, ushort, 0444);
  67 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  68
  69 static bool log_ecn_error = true;
  70 module_param(log_ecn_error, bool, 0644);
  71 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  72
  73 static int vxlan_net_id;
  74 static struct rtnl_link_ops vxlan_link_ops;
  75
  76 static const u8 all_zeros_mac[ETH_ALEN + 2];
  77
  78 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  79
  80 /* per-network namespace private data for this module */
  81 struct vxlan_net {
  82         struct list_head  vxlan_list;
  83         struct hlist_head sock_list[PORT_HASH_SIZE];
  84         spinlock_t        sock_lock;
  85 };
  86
  87 /* Forwarding table entry */
  88 struct vxlan_fdb {
  89         struct hlist_node hlist;        /* linked list of entries */
  90         struct rcu_head   rcu;
  91         unsigned long     updated;      /* jiffies */
  92         unsigned long     used;
  93         struct list_head  remotes;
  94         u8                eth_addr[ETH_ALEN];
  95         u16               state;        /* see ndm_state */
  96         u8                flags;        /* see ndm_flags */
  97 };
  98
  99 /* salt for hash table */
 100 static u32 vxlan_salt __read_mostly;
 101 static struct workqueue_struct *vxlan_wq;
 102
 103 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
 104 {
 105         return vs->flags & VXLAN_F_COLLECT_METADATA ||
 106                ip_tunnel_collect_metadata();
 107 }
 108
 109 #if IS_ENABLED(CONFIG_IPV6)
 110 static inline
 111 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 112 {
 113         if (a->sa.sa_family != b->sa.sa_family)
 114                 return false;
 115         if (a->sa.sa_family == AF_INET6)
 116                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 117         else
 118                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 119 }
 120
 121 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 122 {
 123         if (ipa->sa.sa_family == AF_INET6)
 124                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 125         else
 126                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 127 }
 128
 129 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 130 {
 131         if (ipa->sa.sa_family == AF_INET6)
 132                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 133         else
 134                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 135 }
 136
 137 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 138 {
 139         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 140                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 141                 ip->sa.sa_family = AF_INET6;
 142                 return 0;
 143         } else if (nla_len(nla) >= sizeof(__be32)) {
 144                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 145                 ip->sa.sa_family = AF_INET;
 146                 return 0;
 147         } else {
 148                 return -EAFNOSUPPORT;
 149         }
 150 }
 151
 152 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 153                               const union vxlan_addr *ip)
 154 {
 155         if (ip->sa.sa_family == AF_INET6)
 156                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 157         else
 158                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 159 }
 160
 161 #else /* !CONFIG_IPV6 */
 162
 163 static inline
 164 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 165 {
 166         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 167 }
 168
 169 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 170 {
 171         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 172 }
 173
 174 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 175 {
 176         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 177 }
 178
 179 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 180 {
 181         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 182                 return -EAFNOSUPPORT;
 183         } else if (nla_len(nla) >= sizeof(__be32)) {
 184                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 185                 ip->sa.sa_family = AF_INET;
 186                 return 0;
 187         } else {
 188                 return -EAFNOSUPPORT;
 189         }
 190 }
 191
 192 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 193                               const union vxlan_addr *ip)
 194 {
 195         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 196 }
 197 #endif
 198
 199 /* Virtual Network hash table head */
 200 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 201 {
 202         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 203 }
 204
 205 /* Socket hash table head */
 206 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 207 {
 208         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 209
 210         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 211 }
 212
 213 /* First remote destination for a forwarding entry.
 214  * Guaranteed to be non-NULL because remotes are never deleted.
 215  */
 216 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 217 {
 218         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 219 }
 220
 221 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 222 {
 223         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 224 }
 225
 226 /* Find VXLAN socket based on network namespace, address family and UDP port
 227  * and enabled unshareable flags.
 228  */
 229 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 230                                           __be16 port, u32 flags)
 231 {
 232         struct vxlan_sock *vs;
 233
 234         flags &= VXLAN_F_RCV_FLAGS;
 235
 236         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 237                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 238                     vxlan_get_sk_family(vs) == family &&
 239                     vs->flags == flags)
 240                         return vs;
 241         }
 242         return NULL;
 243 }
 244
 245 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
 246 {
 247         struct vxlan_dev *vxlan;
 248
 249         /* For flow based devices, map all packets to VNI 0 */
 250         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 251                 vni = 0;
 252
 253         hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
 254                 if (vxlan->default_dst.remote_vni == vni)
 255                         return vxlan;
 256         }
 257
 258         return NULL;
 259 }
 260
 261 /* Look up VNI in a per net namespace table */
 262 static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
 263                                         sa_family_t family, __be16 port,
 264                                         u32 flags)
 265 {
 266         struct vxlan_sock *vs;
 267
 268         vs = vxlan_find_sock(net, family, port, flags);
 269         if (!vs)
 270                 return NULL;
 271
 272         return vxlan_vs_find_vni(vs, vni);
 273 }
 274
 275 /* Fill in neighbour message in skbuff. */
 276 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 277                           const struct vxlan_fdb *fdb,
 278                           u32 portid, u32 seq, int type, unsigned int flags,
 279                           const struct vxlan_rdst *rdst)
 280 {
 281         unsigned long now = jiffies;
 282         struct nda_cacheinfo ci;
 283         struct nlmsghdr *nlh;
 284         struct ndmsg *ndm;
 285         bool send_ip, send_eth;
 286
 287         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 288         if (nlh == NULL)
 289                 return -EMSGSIZE;
 290
 291         ndm = nlmsg_data(nlh);
 292         memset(ndm, 0, sizeof(*ndm));
 293
 294         send_eth = send_ip = true;
 295
 296         if (type == RTM_GETNEIGH) {
 297                 ndm->ndm_family = AF_INET;
 298                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 299                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 300         } else
 301                 ndm->ndm_family = AF_BRIDGE;
 302         ndm->ndm_state = fdb->state;
 303         ndm->ndm_ifindex = vxlan->dev->ifindex;
 304         ndm->ndm_flags = fdb->flags;
 305         ndm->ndm_type = RTN_UNICAST;
 306
 307         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 308             nla_put_s32(skb, NDA_LINK_NETNSID,
 309                         peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
 310                 goto nla_put_failure;
 311
 312         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 313                 goto nla_put_failure;
 314
 315         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 316                 goto nla_put_failure;
 317
 318         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 319             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 320                 goto nla_put_failure;
 321         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 322             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 323                 goto nla_put_failure;
 324         if (rdst->remote_ifindex &&
 325             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 326                 goto nla_put_failure;
 327
 328         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 329         ci.ndm_confirmed = 0;
 330         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 331         ci.ndm_refcnt    = 0;
 332
 333         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 334                 goto nla_put_failure;
 335
 336         nlmsg_end(skb, nlh);
 337         return 0;
 338
 339 nla_put_failure:
 340         nlmsg_cancel(skb, nlh);
 341         return -EMSGSIZE;
 342 }
 343
 344 static inline size_t vxlan_nlmsg_size(void)
 345 {
 346         return NLMSG_ALIGN(sizeof(struct ndmsg))
 347                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 348                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 349                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 350                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 351                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 352                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 353                 + nla_total_size(sizeof(struct nda_cacheinfo));
 354 }
 355
 356 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 357                              struct vxlan_rdst *rd, int type)
 358 {
 359         struct net *net = dev_net(vxlan->dev);
 360         struct sk_buff *skb;
 361         int err = -ENOBUFS;
 362
 363         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 364         if (skb == NULL)
 365                 goto errout;
 366
 367         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 368         if (err < 0) {
 369                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 370                 WARN_ON(err == -EMSGSIZE);
 371                 kfree_skb(skb);
 372                 goto errout;
 373         }
 374
 375         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 376         return;
 377 errout:
 378         if (err < 0)
 379                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 380 }
 381
 382 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 383 {
 384         struct vxlan_dev *vxlan = netdev_priv(dev);
 385         struct vxlan_fdb f = {
 386                 .state = NUD_STALE,
 387         };
 388         struct vxlan_rdst remote = {
 389                 .remote_ip = *ipa, /* goes to NDA_DST */
 390                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 391         };
 392
 393         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 394 }
 395
 396 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 397 {
 398         struct vxlan_fdb f = {
 399                 .state = NUD_STALE,
 400         };
 401         struct vxlan_rdst remote = { };
 402
 403         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 404
 405         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 406 }
 407
 408 /* Hash Ethernet address */
 409 static u32 eth_hash(const unsigned char *addr)
 410 {
 411         u64 value = get_unaligned((u64 *)addr);
 412
 413         /* only want 6 bytes */
 414 #ifdef __BIG_ENDIAN
 415         value >>= 16;
 416 #else
 417         value <<= 16;
 418 #endif
 419         return hash_64(value, FDB_HASH_BITS);
 420 }
 421
 422 /* Hash chain to use given mac address */
 423 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 424                                                 const u8 *mac)
 425 {
 426         return &vxlan->fdb_head[eth_hash(mac)];
 427 }
 428
 429 /* Look up Ethernet address in forwarding table */
 430 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 431                                         const u8 *mac)
 432 {
 433         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 434         struct vxlan_fdb *f;
 435
 436         hlist_for_each_entry_rcu(f, head, hlist) {
 437                 if (ether_addr_equal(mac, f->eth_addr))
 438                         return f;
 439         }
 440
 441         return NULL;
 442 }
 443
 444 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 445                                         const u8 *mac)
 446 {
 447         struct vxlan_fdb *f;
 448
 449         f = __vxlan_find_mac(vxlan, mac);
 450         if (f)
 451                 f->used = jiffies;
 452
 453         return f;
 454 }
 455
 456 /* caller should hold vxlan->hash_lock */
 457 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 458                                               union vxlan_addr *ip, __be16 port,
 459                                               __be32 vni, __u32 ifindex)
 460 {
 461         struct vxlan_rdst *rd;
 462
 463         list_for_each_entry(rd, &f->remotes, list) {
 464                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 465                     rd->remote_port == port &&
 466                     rd->remote_vni == vni &&
 467                     rd->remote_ifindex == ifindex)
 468                         return rd;
 469         }
 470
 471         return NULL;
 472 }
 473
 474 /* Replace destination of unicast mac */
 475 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 476                              union vxlan_addr *ip, __be16 port, __be32 vni,
 477                              __u32 ifindex)
 478 {
 479         struct vxlan_rdst *rd;
 480
 481         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 482         if (rd)
 483                 return 0;
 484
 485         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 486         if (!rd)
 487                 return 0;
 488
 489         dst_cache_reset(&rd->dst_cache);
 490         rd->remote_ip = *ip;
 491         rd->remote_port = port;
 492         rd->remote_vni = vni;
 493         rd->remote_ifindex = ifindex;
 494         return 1;
 495 }
 496
 497 /* Add/update destinations for multicast */
 498 static int vxlan_fdb_append(struct vxlan_fdb *f,
 499                             union vxlan_addr *ip, __be16 port, __be32 vni,
 500                             __u32 ifindex, struct vxlan_rdst **rdp)
 501 {
 502         struct vxlan_rdst *rd;
 503
 504         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 505         if (rd)
 506                 return 0;
 507
 508         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 509         if (rd == NULL)
 510                 return -ENOBUFS;
 511
 512         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 513                 kfree(rd);
 514                 return -ENOBUFS;
 515         }
 516
 517         rd->remote_ip = *ip;
 518         rd->remote_port = port;
 519         rd->remote_vni = vni;
 520         rd->remote_ifindex = ifindex;
 521
 522         list_add_tail_rcu(&rd->list, &f->remotes);
 523
 524         *rdp = rd;
 525         return 1;
 526 }
 527
 528 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 529                                           unsigned int off,
 530                                           struct vxlanhdr *vh, size_t hdrlen,
 531                                           __be32 vni_field,
 532                                           struct gro_remcsum *grc,
 533                                           bool nopartial)
 534 {
 535         size_t start, offset;
 536
 537         if (skb->remcsum_offload)
 538                 return vh;
 539
 540         if (!NAPI_GRO_CB(skb)->csum_valid)
 541                 return NULL;
 542
 543         start = vxlan_rco_start(vni_field);
 544         offset = start + vxlan_rco_offset(vni_field);
 545
 546         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 547                                      start, offset, grc, nopartial);
 548
 549         skb->remcsum_offload = 1;
 550
 551         return vh;
 552 }
 553
 554 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
 555                                           struct sk_buff *skb,
 556                                           struct udp_offload *uoff)
 557 {
 558         struct sk_buff *p, **pp = NULL;
 559         struct vxlanhdr *vh, *vh2;
 560         unsigned int hlen, off_vx;
 561         int flush = 1;
 562         struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
 563                                              udp_offloads);
 564         __be32 flags;
 565         struct gro_remcsum grc;
 566
 567         skb_gro_remcsum_init(&grc);
 568
 569         off_vx = skb_gro_offset(skb);
 570         hlen = off_vx + sizeof(*vh);
 571         vh   = skb_gro_header_fast(skb, off_vx);
 572         if (skb_gro_header_hard(skb, hlen)) {
 573                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 574                 if (unlikely(!vh))
 575                         goto out;
 576         }
 577
 578         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 579
 580         flags = vh->vx_flags;
 581
 582         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 583                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 584                                        vh->vx_vni, &grc,
 585                                        !!(vs->flags &
 586                                           VXLAN_F_REMCSUM_NOPARTIAL));
 587
 588                 if (!vh)
 589                         goto out;
 590         }
 591
 592         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 593
 594         for (p = *head; p; p = p->next) {
 595                 if (!NAPI_GRO_CB(p)->same_flow)
 596                         continue;
 597
 598                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 599                 if (vh->vx_flags != vh2->vx_flags ||
 600                     vh->vx_vni != vh2->vx_vni) {
 601                         NAPI_GRO_CB(p)->same_flow = 0;
 602                         continue;
 603                 }
 604         }
 605
 606         pp = eth_gro_receive(head, skb);
 607         flush = 0;
 608
 609 out:
 610         skb_gro_remcsum_cleanup(skb, &grc);
 611         NAPI_GRO_CB(skb)->flush |= flush;
 612
 613         return pp;
 614 }
 615
 616 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
 617                               struct udp_offload *uoff)
 618 {
 619         udp_tunnel_gro_complete(skb, nhoff);
 620
 621         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 622 }
 623
 624 /* Notify netdevs that UDP port started listening */
 625 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 626 {
 627         struct net_device *dev;
 628         struct sock *sk = vs->sock->sk;
 629         struct net *net = sock_net(sk);
 630         sa_family_t sa_family = vxlan_get_sk_family(vs);
 631         __be16 port = inet_sk(sk)->inet_sport;
 632         int err;
 633
 634         if (sa_family == AF_INET) {
 635                 err = udp_add_offload(net, &vs->udp_offloads);
 636                 if (err)
 637                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 638         }
 639
 640         rcu_read_lock();
 641         for_each_netdev_rcu(net, dev) {
 642                 if (dev->netdev_ops->ndo_add_vxlan_port)
 643                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 644                                                             port);
 645         }
 646         rcu_read_unlock();
 647 }
 648
 649 /* Notify netdevs that UDP port is no more listening */
 650 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 651 {
 652         struct net_device *dev;
 653         struct sock *sk = vs->sock->sk;
 654         struct net *net = sock_net(sk);
 655         sa_family_t sa_family = vxlan_get_sk_family(vs);
 656         __be16 port = inet_sk(sk)->inet_sport;
 657
 658         rcu_read_lock();
 659         for_each_netdev_rcu(net, dev) {
 660                 if (dev->netdev_ops->ndo_del_vxlan_port)
 661                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 662                                                             port);
 663         }
 664         rcu_read_unlock();
 665
 666         if (sa_family == AF_INET)
 667                 udp_del_offload(&vs->udp_offloads);
 668 }
 669
 670 /* Add new entry to forwarding table -- assumes lock held */
 671 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 672                             const u8 *mac, union vxlan_addr *ip,
 673                             __u16 state, __u16 flags,
 674                             __be16 port, __be32 vni, __u32 ifindex,
 675                             __u8 ndm_flags)
 676 {
 677         struct vxlan_rdst *rd = NULL;
 678         struct vxlan_fdb *f;
 679         int notify = 0;
 680
 681         f = __vxlan_find_mac(vxlan, mac);
 682         if (f) {
 683                 if (flags & NLM_F_EXCL) {
 684                         netdev_dbg(vxlan->dev,
 685                                    "lost race to create %pM\n", mac);
 686                         return -EEXIST;
 687                 }
 688                 if (f->state != state) {
 689                         f->state = state;
 690                         f->updated = jiffies;
 691                         notify = 1;
 692                 }
 693                 if (f->flags != ndm_flags) {
 694                         f->flags = ndm_flags;
 695                         f->updated = jiffies;
 696                         notify = 1;
 697                 }
 698                 if ((flags & NLM_F_REPLACE)) {
 699                         /* Only change unicasts */
 700                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 701                              is_zero_ether_addr(f->eth_addr))) {
 702                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 703                                                            ifindex);
 704                         } else
 705                                 return -EOPNOTSUPP;
 706                 }
 707                 if ((flags & NLM_F_APPEND) &&
 708                     (is_multicast_ether_addr(f->eth_addr) ||
 709                      is_zero_ether_addr(f->eth_addr))) {
 710                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 711                                                   &rd);
 712
 713                         if (rc < 0)
 714                                 return rc;
 715                         notify |= rc;
 716                 }
 717         } else {
 718                 if (!(flags & NLM_F_CREATE))
 719                         return -ENOENT;
 720
 721                 if (vxlan->cfg.addrmax &&
 722                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 723                         return -ENOSPC;
 724
 725                 /* Disallow replace to add a multicast entry */
 726                 if ((flags & NLM_F_REPLACE) &&
 727                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 728                         return -EOPNOTSUPP;
 729
 730                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 731                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 732                 if (!f)
 733                         return -ENOMEM;
 734
 735                 notify = 1;
 736                 f->state = state;
 737                 f->flags = ndm_flags;
 738                 f->updated = f->used = jiffies;
 739                 INIT_LIST_HEAD(&f->remotes);
 740                 memcpy(f->eth_addr, mac, ETH_ALEN);
 741
 742                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 743
 744                 ++vxlan->addrcnt;
 745                 hlist_add_head_rcu(&f->hlist,
 746                                    vxlan_fdb_head(vxlan, mac));
 747         }
 748
 749         if (notify) {
 750                 if (rd == NULL)
 751                         rd = first_remote_rtnl(f);
 752                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 753         }
 754
 755         return 0;
 756 }
 757
 758 static void vxlan_fdb_free(struct rcu_head *head)
 759 {
 760         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 761         struct vxlan_rdst *rd, *nd;
 762
 763         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 764                 dst_cache_destroy(&rd->dst_cache);
 765                 kfree(rd);
 766         }
 767         kfree(f);
 768 }
 769
 770 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 771 {
 772         netdev_dbg(vxlan->dev,
 773                     "delete %pM\n", f->eth_addr);
 774
 775         --vxlan->addrcnt;
 776         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 777
 778         hlist_del_rcu(&f->hlist);
 779         call_rcu(&f->rcu, vxlan_fdb_free);
 780 }
 781
 782 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 783                            union vxlan_addr *ip, __be16 *port, __be32 *vni,
 784                            u32 *ifindex)
 785 {
 786         struct net *net = dev_net(vxlan->dev);
 787         int err;
 788
 789         if (tb[NDA_DST]) {
 790                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 791                 if (err)
 792                         return err;
 793         } else {
 794                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 795                 if (remote->sa.sa_family == AF_INET) {
 796                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 797                         ip->sa.sa_family = AF_INET;
 798 #if IS_ENABLED(CONFIG_IPV6)
 799                 } else {
 800                         ip->sin6.sin6_addr = in6addr_any;
 801                         ip->sa.sa_family = AF_INET6;
 802 #endif
 803                 }
 804         }
 805
 806         if (tb[NDA_PORT]) {
 807                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 808                         return -EINVAL;
 809                 *port = nla_get_be16(tb[NDA_PORT]);
 810         } else {
 811                 *port = vxlan->cfg.dst_port;
 812         }
 813
 814         if (tb[NDA_VNI]) {
 815                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 816                         return -EINVAL;
 817                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 818         } else {
 819                 *vni = vxlan->default_dst.remote_vni;
 820         }
 821
 822         if (tb[NDA_IFINDEX]) {
 823                 struct net_device *tdev;
 824
 825                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 826                         return -EINVAL;
 827                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 828                 tdev = __dev_get_by_index(net, *ifindex);
 829                 if (!tdev)
 830                         return -EADDRNOTAVAIL;
 831         } else {
 832                 *ifindex = 0;
 833         }
 834
 835         return 0;
 836 }
 837
 838 /* Add static entry (via netlink) */
 839 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 840                          struct net_device *dev,
 841                          const unsigned char *addr, u16 vid, u16 flags)
 842 {
 843         struct vxlan_dev *vxlan = netdev_priv(dev);
 844         /* struct net *net = dev_net(vxlan->dev); */
 845         union vxlan_addr ip;
 846         __be16 port;
 847         __be32 vni;
 848         u32 ifindex;
 849         int err;
 850
 851         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 852                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 853                         ndm->ndm_state);
 854                 return -EINVAL;
 855         }
 856
 857         if (tb[NDA_DST] == NULL)
 858                 return -EINVAL;
 859
 860         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 861         if (err)
 862                 return err;
 863
 864         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 865                 return -EAFNOSUPPORT;
 866
 867         spin_lock_bh(&vxlan->hash_lock);
 868         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 869                                port, vni, ifindex, ndm->ndm_flags);
 870         spin_unlock_bh(&vxlan->hash_lock);
 871
 872         return err;
 873 }
 874
 875 /* Delete entry (via netlink) */
 876 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 877                             struct net_device *dev,
 878                             const unsigned char *addr, u16 vid)
 879 {
 880         struct vxlan_dev *vxlan = netdev_priv(dev);
 881         struct vxlan_fdb *f;
 882         struct vxlan_rdst *rd = NULL;
 883         union vxlan_addr ip;
 884         __be16 port;
 885         __be32 vni;
 886         u32 ifindex;
 887         int err;
 888
 889         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 890         if (err)
 891                 return err;
 892
 893         err = -ENOENT;
 894
 895         spin_lock_bh(&vxlan->hash_lock);
 896         f = vxlan_find_mac(vxlan, addr);
 897         if (!f)
 898                 goto out;
 899
 900         if (!vxlan_addr_any(&ip)) {
 901                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 902                 if (!rd)
 903                         goto out;
 904         }
 905
 906         err = 0;
 907
 908         /* remove a destination if it's not the only one on the list,
 909          * otherwise destroy the fdb entry
 910          */
 911         if (rd && !list_is_singular(&f->remotes)) {
 912                 list_del_rcu(&rd->list);
 913                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 914                 kfree_rcu(rd, rcu);
 915                 goto out;
 916         }
 917
 918         vxlan_fdb_destroy(vxlan, f);
 919
 920 out:
 921         spin_unlock_bh(&vxlan->hash_lock);
 922
 923         return err;
 924 }
 925
 926 /* Dump forwarding table */
 927 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 928                           struct net_device *dev,
 929                           struct net_device *filter_dev, int idx)
 930 {
 931         struct vxlan_dev *vxlan = netdev_priv(dev);
 932         unsigned int h;
 933
 934         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 935                 struct vxlan_fdb *f;
 936                 int err;
 937
 938                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 939                         struct vxlan_rdst *rd;
 940
 941                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 942                                 if (idx < cb->args[0])
 943                                         goto skip;
 944
 945                                 err = vxlan_fdb_info(skb, vxlan, f,
 946                                                      NETLINK_CB(cb->skb).portid,
 947                                                      cb->nlh->nlmsg_seq,
 948                                                      RTM_NEWNEIGH,
 949                                                      NLM_F_MULTI, rd);
 950                                 if (err < 0) {
 951                                         cb->args[1] = err;
 952                                         goto out;
 953                                 }
 954 skip:
 955                                 ++idx;
 956                         }
 957                 }
 958         }
 959 out:
 960         return idx;
 961 }
 962
 963 /* Watch incoming packets to learn mapping between Ethernet address
 964  * and Tunnel endpoint.
 965  * Return true if packet is bogus and should be dropped.
 966  */
 967 static bool vxlan_snoop(struct net_device *dev,
 968                         union vxlan_addr *src_ip, const u8 *src_mac)
 969 {
 970         struct vxlan_dev *vxlan = netdev_priv(dev);
 971         struct vxlan_fdb *f;
 972
 973         f = vxlan_find_mac(vxlan, src_mac);
 974         if (likely(f)) {
 975                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 976
 977                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 978                         return false;
 979
 980                 /* Don't migrate static entries, drop packets */
 981                 if (f->state & NUD_NOARP)
 982                         return true;
 983
 984                 if (net_ratelimit())
 985                         netdev_info(dev,
 986                                     "%pM migrated from %pIS to %pIS\n",
 987                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 988
 989                 rdst->remote_ip = *src_ip;
 990                 f->updated = jiffies;
 991                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 992         } else {
 993                 /* learned new entry */
 994                 spin_lock(&vxlan->hash_lock);
 995
 996                 /* close off race between vxlan_flush and incoming packets */
 997                 if (netif_running(dev))
 998                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 999                                          NUD_REACHABLE,
1000                                          NLM_F_EXCL|NLM_F_CREATE,
1001                                          vxlan->cfg.dst_port,
1002                                          vxlan->default_dst.remote_vni,
1003                                          0, NTF_SELF);
1004                 spin_unlock(&vxlan->hash_lock);
1005         }
1006
1007         return false;
1008 }
1009
1010 /* See if multicast group is already in use by other ID */
1011 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1012 {
1013         struct vxlan_dev *vxlan;
1014         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1015
1016         /* The vxlan_sock is only used by dev, leaving group has
1017          * no effect on other vxlan devices.
1018          */
1019         if (family == AF_INET && dev->vn4_sock &&
1020             atomic_read(&dev->vn4_sock->refcnt) == 1)
1021                 return false;
1022 #if IS_ENABLED(CONFIG_IPV6)
1023         if (family == AF_INET6 && dev->vn6_sock &&
1024             atomic_read(&dev->vn6_sock->refcnt) == 1)
1025                 return false;
1026 #endif
1027
1028         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1029                 if (!netif_running(vxlan->dev) || vxlan == dev)
1030                         continue;
1031
1032                 if (family == AF_INET && vxlan->vn4_sock != dev->vn4_sock)
1033                         continue;
1034 #if IS_ENABLED(CONFIG_IPV6)
1035                 if (family == AF_INET6 && vxlan->vn6_sock != dev->vn6_sock)
1036                         continue;
1037 #endif
1038
1039                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1040                                       &dev->default_dst.remote_ip))
1041                         continue;
1042
1043                 if (vxlan->default_dst.remote_ifindex !=
1044                     dev->default_dst.remote_ifindex)
1045                         continue;
1046
1047                 return true;
1048         }
1049
1050         return false;
1051 }
1052
1053 static void __vxlan_sock_release(struct vxlan_sock *vs)
1054 {
1055         struct vxlan_net *vn;
1056
1057         if (!vs)
1058                 return;
1059         if (!atomic_dec_and_test(&vs->refcnt))
1060                 return;
1061
1062         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1063         spin_lock(&vn->sock_lock);
1064         hlist_del_rcu(&vs->hlist);
1065         vxlan_notify_del_rx_port(vs);
1066         spin_unlock(&vn->sock_lock);
1067
1068         queue_work(vxlan_wq, &vs->del_work);
1069 }
1070
1071 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1072 {
1073         __vxlan_sock_release(vxlan->vn4_sock);
1074 #if IS_ENABLED(CONFIG_IPV6)
1075         __vxlan_sock_release(vxlan->vn6_sock);
1076 #endif
1077 }
1078
1079 /* Update multicast group membership when first VNI on
1080  * multicast address is brought up
1081  */
1082 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1083 {
1084         struct 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         if (ip->sa.sa_family == AF_INET) {
1090                 struct ip_mreqn mreq = {
1091                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1092                         .imr_ifindex            = ifindex,
1093                 };
1094
1095                 sk = vxlan->vn4_sock->sock->sk;
1096                 lock_sock(sk);
1097                 ret = ip_mc_join_group(sk, &mreq);
1098                 release_sock(sk);
1099 #if IS_ENABLED(CONFIG_IPV6)
1100         } else {
1101                 sk = vxlan->vn6_sock->sock->sk;
1102                 lock_sock(sk);
1103                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1104                                                    &ip->sin6.sin6_addr);
1105                 release_sock(sk);
1106 #endif
1107         }
1108
1109         return ret;
1110 }
1111
1112 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1113 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1114 {
1115         struct sock *sk;
1116         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1117         int ifindex = vxlan->default_dst.remote_ifindex;
1118         int ret = -EINVAL;
1119
1120         if (ip->sa.sa_family == AF_INET) {
1121                 struct ip_mreqn mreq = {
1122                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1123                         .imr_ifindex            = ifindex,
1124                 };
1125
1126                 sk = vxlan->vn4_sock->sock->sk;
1127                 lock_sock(sk);
1128                 ret = ip_mc_leave_group(sk, &mreq);
1129                 release_sock(sk);
1130 #if IS_ENABLED(CONFIG_IPV6)
1131         } else {
1132                 sk = vxlan->vn6_sock->sock->sk;
1133                 lock_sock(sk);
1134                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1135                                                    &ip->sin6.sin6_addr);
1136                 release_sock(sk);
1137 #endif
1138         }
1139
1140         return ret;
1141 }
1142
1143 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1144                           struct sk_buff *skb, u32 vxflags)
1145 {
1146         size_t start, offset;
1147
1148         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1149                 goto out;
1150
1151         start = vxlan_rco_start(unparsed->vx_vni);
1152         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1153
1154         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1155                 return false;
1156
1157         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1158                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1159 out:
1160         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1161         unparsed->vx_vni &= VXLAN_VNI_MASK;
1162         return true;
1163 }
1164
1165 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1166                                 struct sk_buff *skb, u32 vxflags,
1167                                 struct vxlan_metadata *md)
1168 {
1169         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1170         struct metadata_dst *tun_dst;
1171
1172         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1173                 goto out;
1174
1175         md->gbp = ntohs(gbp->policy_id);
1176
1177         tun_dst = (struct metadata_dst *)skb_dst(skb);
1178         if (tun_dst) {
1179                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1180                 tun_dst->u.tun_info.options_len = sizeof(*md);
1181         }
1182         if (gbp->dont_learn)
1183                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1184
1185         if (gbp->policy_applied)
1186                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1187
1188         /* In flow-based mode, GBP is carried in dst_metadata */
1189         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1190                 skb->mark = md->gbp;
1191 out:
1192         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1193 }
1194
1195 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1196                                 __be32 *protocol,
1197                                 struct sk_buff *skb, u32 vxflags)
1198 {
1199         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1200
1201         /* Need to have Next Protocol set for interfaces in GPE mode. */
1202         if (!gpe->np_applied)
1203                 return false;
1204         /* "The initial version is 0. If a receiver does not support the
1205          * version indicated it MUST drop the packet.
1206          */
1207         if (gpe->version != 0)
1208                 return false;
1209         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1210          * processing MUST occur." However, we don't implement OAM
1211          * processing, thus drop the packet.
1212          */
1213         if (gpe->oam_flag)
1214                 return false;
1215
1216         switch (gpe->next_protocol) {
1217         case VXLAN_GPE_NP_IPV4:
1218                 *protocol = htons(ETH_P_IP);
1219                 break;
1220         case VXLAN_GPE_NP_IPV6:
1221                 *protocol = htons(ETH_P_IPV6);
1222                 break;
1223         case VXLAN_GPE_NP_ETHERNET:
1224                 *protocol = htons(ETH_P_TEB);
1225                 break;
1226         default:
1227                 return false;
1228         }
1229
1230         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1231         return true;
1232 }
1233
1234 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1235                           struct vxlan_sock *vs,
1236                           struct sk_buff *skb)
1237 {
1238         union vxlan_addr saddr;
1239
1240         skb_reset_mac_header(skb);
1241         skb->protocol = eth_type_trans(skb, vxlan->dev);
1242         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1243
1244         /* Ignore packet loops (and multicast echo) */
1245         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1246                 return false;
1247
1248         /* Get address from the outer IP header */
1249         if (vxlan_get_sk_family(vs) == AF_INET) {
1250                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1251                 saddr.sa.sa_family = AF_INET;
1252 #if IS_ENABLED(CONFIG_IPV6)
1253         } else {
1254                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1255                 saddr.sa.sa_family = AF_INET6;
1256 #endif
1257         }
1258
1259         if ((vxlan->flags & VXLAN_F_LEARN) &&
1260             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1261                 return false;
1262
1263         return true;
1264 }
1265
1266 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1267                                   struct sk_buff *skb)
1268 {
1269         int err = 0;
1270
1271         if (vxlan_get_sk_family(vs) == AF_INET)
1272                 err = IP_ECN_decapsulate(oiph, skb);
1273 #if IS_ENABLED(CONFIG_IPV6)
1274         else
1275                 err = IP6_ECN_decapsulate(oiph, skb);
1276 #endif
1277
1278         if (unlikely(err) && log_ecn_error) {
1279                 if (vxlan_get_sk_family(vs) == AF_INET)
1280                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1281                                              &((struct iphdr *)oiph)->saddr,
1282                                              ((struct iphdr *)oiph)->tos);
1283                 else
1284                         net_info_ratelimited("non-ECT from %pI6\n",
1285                                              &((struct ipv6hdr *)oiph)->saddr);
1286         }
1287         return err <= 1;
1288 }
1289
1290 /* Callback from net/ipv4/udp.c to receive packets */
1291 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1292 {
1293         struct pcpu_sw_netstats *stats;
1294         struct vxlan_dev *vxlan;
1295         struct vxlan_sock *vs;
1296         struct vxlanhdr unparsed;
1297         struct vxlan_metadata _md;
1298         struct vxlan_metadata *md = &_md;
1299         __be32 protocol = htons(ETH_P_TEB);
1300         bool raw_proto = false;
1301         void *oiph;
1302
1303         /* Need UDP and VXLAN header to be present */
1304         if (!pskb_may_pull(skb, VXLAN_HLEN))
1305                 return 1;
1306
1307         unparsed = *vxlan_hdr(skb);
1308         /* VNI flag always required to be set */
1309         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1310                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1311                            ntohl(vxlan_hdr(skb)->vx_flags),
1312                            ntohl(vxlan_hdr(skb)->vx_vni));
1313                 /* Return non vxlan pkt */
1314                 return 1;
1315         }
1316         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1317         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1318
1319         vs = rcu_dereference_sk_user_data(sk);
1320         if (!vs)
1321                 goto drop;
1322
1323         vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1324         if (!vxlan)
1325                 goto drop;
1326
1327         /* For backwards compatibility, only allow reserved fields to be
1328          * used by VXLAN extensions if explicitly requested.
1329          */
1330         if (vs->flags & VXLAN_F_GPE) {
1331                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1332                         goto drop;
1333                 raw_proto = true;
1334         }
1335
1336         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1337                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1338                         goto drop;
1339
1340         if (vxlan_collect_metadata(vs)) {
1341                 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1342                 struct metadata_dst *tun_dst;
1343
1344                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1345                                          vxlan_vni_to_tun_id(vni), sizeof(*md));
1346
1347                 if (!tun_dst)
1348                         goto drop;
1349
1350                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1351
1352                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1353         } else {
1354                 memset(md, 0, sizeof(*md));
1355         }
1356
1357         if (vs->flags & VXLAN_F_REMCSUM_RX)
1358                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1359                         goto drop;
1360         if (vs->flags & VXLAN_F_GBP)
1361                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1362         /* Note that GBP and GPE can never be active together. This is
1363          * ensured in vxlan_dev_configure.
1364          */
1365
1366         if (unparsed.vx_flags || unparsed.vx_vni) {
1367                 /* If there are any unprocessed flags remaining treat
1368                  * this as a malformed packet. This behavior diverges from
1369                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1370                  * in reserved fields are to be ignored. The approach here
1371                  * maintains compatibility with previous stack code, and also
1372                  * is more robust and provides a little more security in
1373                  * adding extensions to VXLAN.
1374                  */
1375                 goto drop;
1376         }
1377
1378         if (!raw_proto) {
1379                 if (!vxlan_set_mac(vxlan, vs, skb))
1380                         goto drop;
1381         } else {
1382                 skb->dev = vxlan->dev;
1383                 skb->pkt_type = PACKET_HOST;
1384         }
1385
1386         oiph = skb_network_header(skb);
1387         skb_reset_network_header(skb);
1388
1389         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1390                 ++vxlan->dev->stats.rx_frame_errors;
1391                 ++vxlan->dev->stats.rx_errors;
1392                 goto drop;
1393         }
1394
1395         stats = this_cpu_ptr(vxlan->dev->tstats);
1396         u64_stats_update_begin(&stats->syncp);
1397         stats->rx_packets++;
1398         stats->rx_bytes += skb->len;
1399         u64_stats_update_end(&stats->syncp);
1400
1401         gro_cells_receive(&vxlan->gro_cells, skb);
1402         return 0;
1403
1404 drop:
1405         /* Consume bad packet */
1406         kfree_skb(skb);
1407         return 0;
1408 }
1409
1410 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1411 {
1412         struct vxlan_dev *vxlan = netdev_priv(dev);
1413         struct arphdr *parp;
1414         u8 *arpptr, *sha;
1415         __be32 sip, tip;
1416         struct neighbour *n;
1417
1418         if (dev->flags & IFF_NOARP)
1419                 goto out;
1420
1421         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1422                 dev->stats.tx_dropped++;
1423                 goto out;
1424         }
1425         parp = arp_hdr(skb);
1426
1427         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1428              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1429             parp->ar_pro != htons(ETH_P_IP) ||
1430             parp->ar_op != htons(ARPOP_REQUEST) ||
1431             parp->ar_hln != dev->addr_len ||
1432             parp->ar_pln != 4)
1433                 goto out;
1434         arpptr = (u8 *)parp + sizeof(struct arphdr);
1435         sha = arpptr;
1436         arpptr += dev->addr_len;        /* sha */
1437         memcpy(&sip, arpptr, sizeof(sip));
1438         arpptr += sizeof(sip);
1439         arpptr += dev->addr_len;        /* tha */
1440         memcpy(&tip, arpptr, sizeof(tip));
1441
1442         if (ipv4_is_loopback(tip) ||
1443             ipv4_is_multicast(tip))
1444                 goto out;
1445
1446         n = neigh_lookup(&arp_tbl, &tip, dev);
1447
1448         if (n) {
1449                 struct vxlan_fdb *f;
1450                 struct sk_buff  *reply;
1451
1452                 if (!(n->nud_state & NUD_CONNECTED)) {
1453                         neigh_release(n);
1454                         goto out;
1455                 }
1456
1457                 f = vxlan_find_mac(vxlan, n->ha);
1458                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1459                         /* bridge-local neighbor */
1460                         neigh_release(n);
1461                         goto out;
1462                 }
1463
1464                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1465                                 n->ha, sha);
1466
1467                 neigh_release(n);
1468
1469                 if (reply == NULL)
1470                         goto out;
1471
1472                 skb_reset_mac_header(reply);
1473                 __skb_pull(reply, skb_network_offset(reply));
1474                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1475                 reply->pkt_type = PACKET_HOST;
1476
1477                 if (netif_rx_ni(reply) == NET_RX_DROP)
1478                         dev->stats.rx_dropped++;
1479         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1480                 union vxlan_addr ipa = {
1481                         .sin.sin_addr.s_addr = tip,
1482                         .sin.sin_family = AF_INET,
1483                 };
1484
1485                 vxlan_ip_miss(dev, &ipa);
1486         }
1487 out:
1488         consume_skb(skb);
1489         return NETDEV_TX_OK;
1490 }
1491
1492 #if IS_ENABLED(CONFIG_IPV6)
1493 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1494         struct neighbour *n, bool isrouter)
1495 {
1496         struct net_device *dev = request->dev;
1497         struct sk_buff *reply;
1498         struct nd_msg *ns, *na;
1499         struct ipv6hdr *pip6;
1500         u8 *daddr;
1501         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1502         int ns_olen;
1503         int i, len;
1504
1505         if (dev == NULL)
1506                 return NULL;
1507
1508         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1509                 sizeof(*na) + na_olen + dev->needed_tailroom;
1510         reply = alloc_skb(len, GFP_ATOMIC);
1511         if (reply == NULL)
1512                 return NULL;
1513
1514         reply->protocol = htons(ETH_P_IPV6);
1515         reply->dev = dev;
1516         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1517         skb_push(reply, sizeof(struct ethhdr));
1518         skb_reset_mac_header(reply);
1519
1520         ns = (struct nd_msg *)skb_transport_header(request);
1521
1522         daddr = eth_hdr(request)->h_source;
1523         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1524         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1525                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1526                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1527                         break;
1528                 }
1529         }
1530
1531         /* Ethernet header */
1532         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1533         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1534         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1535         reply->protocol = htons(ETH_P_IPV6);
1536
1537         skb_pull(reply, sizeof(struct ethhdr));
1538         skb_reset_network_header(reply);
1539         skb_put(reply, sizeof(struct ipv6hdr));
1540
1541         /* IPv6 header */
1542
1543         pip6 = ipv6_hdr(reply);
1544         memset(pip6, 0, sizeof(struct ipv6hdr));
1545         pip6->version = 6;
1546         pip6->priority = ipv6_hdr(request)->priority;
1547         pip6->nexthdr = IPPROTO_ICMPV6;
1548         pip6->hop_limit = 255;
1549         pip6->daddr = ipv6_hdr(request)->saddr;
1550         pip6->saddr = *(struct in6_addr *)n->primary_key;
1551
1552         skb_pull(reply, sizeof(struct ipv6hdr));
1553         skb_reset_transport_header(reply);
1554
1555         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1556
1557         /* Neighbor Advertisement */
1558         memset(na, 0, sizeof(*na)+na_olen);
1559         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1560         na->icmph.icmp6_router = isrouter;
1561         na->icmph.icmp6_override = 1;
1562         na->icmph.icmp6_solicited = 1;
1563         na->target = ns->target;
1564         ether_addr_copy(&na->opt[2], n->ha);
1565         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1566         na->opt[1] = na_olen >> 3;
1567
1568         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1569                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1570                 csum_partial(na, sizeof(*na)+na_olen, 0));
1571
1572         pip6->payload_len = htons(sizeof(*na)+na_olen);
1573
1574         skb_push(reply, sizeof(struct ipv6hdr));
1575
1576         reply->ip_summed = CHECKSUM_UNNECESSARY;
1577
1578         return reply;
1579 }
1580
1581 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1582 {
1583         struct vxlan_dev *vxlan = netdev_priv(dev);
1584         struct nd_msg *msg;
1585         const struct ipv6hdr *iphdr;
1586         const struct in6_addr *saddr, *daddr;
1587         struct neighbour *n;
1588         struct inet6_dev *in6_dev;
1589
1590         in6_dev = __in6_dev_get(dev);
1591         if (!in6_dev)
1592                 goto out;
1593
1594         iphdr = ipv6_hdr(skb);
1595         saddr = &iphdr->saddr;
1596         daddr = &iphdr->daddr;
1597
1598         msg = (struct nd_msg *)skb_transport_header(skb);
1599         if (msg->icmph.icmp6_code != 0 ||
1600             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1601                 goto out;
1602
1603         if (ipv6_addr_loopback(daddr) ||
1604             ipv6_addr_is_multicast(&msg->target))
1605                 goto out;
1606
1607         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1608
1609         if (n) {
1610                 struct vxlan_fdb *f;
1611                 struct sk_buff *reply;
1612
1613                 if (!(n->nud_state & NUD_CONNECTED)) {
1614                         neigh_release(n);
1615                         goto out;
1616                 }
1617
1618                 f = vxlan_find_mac(vxlan, n->ha);
1619                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1620                         /* bridge-local neighbor */
1621                         neigh_release(n);
1622                         goto out;
1623                 }
1624
1625                 reply = vxlan_na_create(skb, n,
1626                                         !!(f ? f->flags & NTF_ROUTER : 0));
1627
1628                 neigh_release(n);
1629
1630                 if (reply == NULL)
1631                         goto out;
1632
1633                 if (netif_rx_ni(reply) == NET_RX_DROP)
1634                         dev->stats.rx_dropped++;
1635
1636         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1637                 union vxlan_addr ipa = {
1638                         .sin6.sin6_addr = msg->target,
1639                         .sin6.sin6_family = AF_INET6,
1640                 };
1641
1642                 vxlan_ip_miss(dev, &ipa);
1643         }
1644
1645 out:
1646         consume_skb(skb);
1647         return NETDEV_TX_OK;
1648 }
1649 #endif
1650
1651 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1652 {
1653         struct vxlan_dev *vxlan = netdev_priv(dev);
1654         struct neighbour *n;
1655
1656         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1657                 return false;
1658
1659         n = NULL;
1660         switch (ntohs(eth_hdr(skb)->h_proto)) {
1661         case ETH_P_IP:
1662         {
1663                 struct iphdr *pip;
1664
1665                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1666                         return false;
1667                 pip = ip_hdr(skb);
1668                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1669                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1670                         union vxlan_addr ipa = {
1671                                 .sin.sin_addr.s_addr = pip->daddr,
1672                                 .sin.sin_family = AF_INET,
1673                         };
1674
1675                         vxlan_ip_miss(dev, &ipa);
1676                         return false;
1677                 }
1678
1679                 break;
1680         }
1681 #if IS_ENABLED(CONFIG_IPV6)
1682         case ETH_P_IPV6:
1683         {
1684                 struct ipv6hdr *pip6;
1685
1686                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1687                         return false;
1688                 pip6 = ipv6_hdr(skb);
1689                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1690                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1691                         union vxlan_addr ipa = {
1692                                 .sin6.sin6_addr = pip6->daddr,
1693                                 .sin6.sin6_family = AF_INET6,
1694                         };
1695
1696                         vxlan_ip_miss(dev, &ipa);
1697                         return false;
1698                 }
1699
1700                 break;
1701         }
1702 #endif
1703         default:
1704                 return false;
1705         }
1706
1707         if (n) {
1708                 bool diff;
1709
1710                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1711                 if (diff) {
1712                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1713                                 dev->addr_len);
1714                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1715                 }
1716                 neigh_release(n);
1717                 return diff;
1718         }
1719
1720         return false;
1721 }
1722
1723 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1724                                 struct vxlan_metadata *md)
1725 {
1726         struct vxlanhdr_gbp *gbp;
1727
1728         if (!md->gbp)
1729                 return;
1730
1731         gbp = (struct vxlanhdr_gbp *)vxh;
1732         vxh->vx_flags |= VXLAN_HF_GBP;
1733
1734         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1735                 gbp->dont_learn = 1;
1736
1737         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1738                 gbp->policy_applied = 1;
1739
1740         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1741 }
1742
1743 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1744                                __be16 protocol)
1745 {
1746         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1747
1748         gpe->np_applied = 1;
1749
1750         switch (protocol) {
1751         case htons(ETH_P_IP):
1752                 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1753                 return 0;
1754         case htons(ETH_P_IPV6):
1755                 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1756                 return 0;
1757         case htons(ETH_P_TEB):
1758                 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1759                 return 0;
1760         }
1761         return -EPFNOSUPPORT;
1762 }
1763
1764 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1765                            int iphdr_len, __be32 vni,
1766                            struct vxlan_metadata *md, u32 vxflags,
1767                            bool udp_sum)
1768 {
1769         struct vxlanhdr *vxh;
1770         int min_headroom;
1771         int err;
1772         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1773         __be16 inner_protocol = htons(ETH_P_TEB);
1774
1775         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1776             skb->ip_summed == CHECKSUM_PARTIAL) {
1777                 int csum_start = skb_checksum_start_offset(skb);
1778
1779                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1780                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1781                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1782                      skb->csum_offset == offsetof(struct tcphdr, check)))
1783                         type |= SKB_GSO_TUNNEL_REMCSUM;
1784         }
1785
1786         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1787                         + VXLAN_HLEN + iphdr_len
1788                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1789
1790         /* Need space for new headers (invalidates iph ptr) */
1791         err = skb_cow_head(skb, min_headroom);
1792         if (unlikely(err))
1793                 goto out_free;
1794
1795         skb = vlan_hwaccel_push_inside(skb);
1796         if (WARN_ON(!skb))
1797                 return -ENOMEM;
1798
1799         skb = iptunnel_handle_offloads(skb, type);
1800         if (IS_ERR(skb))
1801                 return PTR_ERR(skb);
1802
1803         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1804         vxh->vx_flags = VXLAN_HF_VNI;
1805         vxh->vx_vni = vxlan_vni_field(vni);
1806
1807         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1808                 unsigned int start;
1809
1810                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1811                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1812                 vxh->vx_flags |= VXLAN_HF_RCO;
1813
1814                 if (!skb_is_gso(skb)) {
1815                         skb->ip_summed = CHECKSUM_NONE;
1816                         skb->encapsulation = 0;
1817                 }
1818         }
1819
1820         if (vxflags & VXLAN_F_GBP)
1821                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1822         if (vxflags & VXLAN_F_GPE) {
1823                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1824                 if (err < 0)
1825                         goto out_free;
1826                 inner_protocol = skb->protocol;
1827         }
1828
1829         skb_set_inner_protocol(skb, inner_protocol);
1830         return 0;
1831
1832 out_free:
1833         kfree_skb(skb);
1834         return err;
1835 }
1836
1837 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1838                                       struct sk_buff *skb, int oif, u8 tos,
1839                                       __be32 daddr, __be32 *saddr,
1840                                       struct dst_cache *dst_cache,
1841                                       const struct ip_tunnel_info *info)
1842 {
1843         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1844         struct rtable *rt = NULL;
1845         struct flowi4 fl4;
1846
1847         if (tos && !info)
1848                 use_cache = false;
1849         if (use_cache) {
1850                 rt = dst_cache_get_ip4(dst_cache, saddr);
1851                 if (rt)
1852                         return rt;
1853         }
1854
1855         memset(&fl4, 0, sizeof(fl4));
1856         fl4.flowi4_oif = oif;
1857         fl4.flowi4_tos = RT_TOS(tos);
1858         fl4.flowi4_mark = skb->mark;
1859         fl4.flowi4_proto = IPPROTO_UDP;
1860         fl4.daddr = daddr;
1861         fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
1862
1863         rt = ip_route_output_key(vxlan->net, &fl4);
1864         if (!IS_ERR(rt)) {
1865                 *saddr = fl4.saddr;
1866                 if (use_cache)
1867                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1868         }
1869         return rt;
1870 }
1871
1872 #if IS_ENABLED(CONFIG_IPV6)
1873 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1874                                           struct sk_buff *skb, int oif, u8 tos,
1875                                           __be32 label,
1876                                           const struct in6_addr *daddr,
1877                                           struct in6_addr *saddr,
1878                                           struct dst_cache *dst_cache,
1879                                           const struct ip_tunnel_info *info)
1880 {
1881         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1882         struct dst_entry *ndst;
1883         struct flowi6 fl6;
1884         int err;
1885
1886         if (tos && !info)
1887                 use_cache = false;
1888         if (use_cache) {
1889                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1890                 if (ndst)
1891                         return ndst;
1892         }
1893
1894         memset(&fl6, 0, sizeof(fl6));
1895         fl6.flowi6_oif = oif;
1896         fl6.daddr = *daddr;
1897         fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
1898         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1899         fl6.flowi6_mark = skb->mark;
1900         fl6.flowi6_proto = IPPROTO_UDP;
1901
1902         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1903                                          vxlan->vn6_sock->sock->sk,
1904                                          &ndst, &fl6);
1905         if (err < 0)
1906                 return ERR_PTR(err);
1907
1908         *saddr = fl6.saddr;
1909         if (use_cache)
1910                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1911         return ndst;
1912 }
1913 #endif
1914
1915 /* Bypass encapsulation if the destination is local */
1916 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1917                                struct vxlan_dev *dst_vxlan)
1918 {
1919         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1920         union vxlan_addr loopback;
1921         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1922         struct net_device *dev = skb->dev;
1923         int len = skb->len;
1924
1925         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1926         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1927         skb->pkt_type = PACKET_HOST;
1928         skb->encapsulation = 0;
1929         skb->dev = dst_vxlan->dev;
1930         __skb_pull(skb, skb_network_offset(skb));
1931
1932         if (remote_ip->sa.sa_family == AF_INET) {
1933                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1934                 loopback.sa.sa_family =  AF_INET;
1935 #if IS_ENABLED(CONFIG_IPV6)
1936         } else {
1937                 loopback.sin6.sin6_addr = in6addr_loopback;
1938                 loopback.sa.sa_family =  AF_INET6;
1939 #endif
1940         }
1941
1942         if (dst_vxlan->flags & VXLAN_F_LEARN)
1943                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1944
1945         u64_stats_update_begin(&tx_stats->syncp);
1946         tx_stats->tx_packets++;
1947         tx_stats->tx_bytes += len;
1948         u64_stats_update_end(&tx_stats->syncp);
1949
1950         if (netif_rx(skb) == NET_RX_SUCCESS) {
1951                 u64_stats_update_begin(&rx_stats->syncp);
1952                 rx_stats->rx_packets++;
1953                 rx_stats->rx_bytes += len;
1954                 u64_stats_update_end(&rx_stats->syncp);
1955         } else {
1956                 dev->stats.rx_dropped++;
1957         }
1958 }
1959
1960 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1961                            struct vxlan_rdst *rdst, bool did_rsc)
1962 {
1963         struct dst_cache *dst_cache;
1964         struct ip_tunnel_info *info;
1965         struct vxlan_dev *vxlan = netdev_priv(dev);
1966         struct sock *sk;
1967         struct rtable *rt = NULL;
1968         const struct iphdr *old_iph;
1969         union vxlan_addr *dst;
1970         union vxlan_addr remote_ip;
1971         struct vxlan_metadata _md;
1972         struct vxlan_metadata *md = &_md;
1973         __be16 src_port = 0, dst_port;
1974         __be32 vni, label;
1975         __be16 df = 0;
1976         __u8 tos, ttl;
1977         int err;
1978         u32 flags = vxlan->flags;
1979         bool udp_sum = false;
1980         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1981
1982         info = skb_tunnel_info(skb);
1983
1984         if (rdst) {
1985                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1986                 vni = rdst->remote_vni;
1987                 dst = &rdst->remote_ip;
1988                 dst_cache = &rdst->dst_cache;
1989         } else {
1990                 if (!info) {
1991                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1992                                   dev->name);
1993                         goto drop;
1994                 }
1995                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1996                 vni = vxlan_tun_id_to_vni(info->key.tun_id);
1997                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1998                 if (remote_ip.sa.sa_family == AF_INET)
1999                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2000                 else
2001                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2002                 dst = &remote_ip;
2003                 dst_cache = &info->dst_cache;
2004         }
2005
2006         if (vxlan_addr_any(dst)) {
2007                 if (did_rsc) {
2008                         /* short-circuited back to local bridge */
2009                         vxlan_encap_bypass(skb, vxlan, vxlan);
2010                         return;
2011                 }
2012                 goto drop;
2013         }
2014
2015         old_iph = ip_hdr(skb);
2016
2017         ttl = vxlan->cfg.ttl;
2018         if (!ttl && vxlan_addr_multicast(dst))
2019                 ttl = 1;
2020
2021         tos = vxlan->cfg.tos;
2022         if (tos == 1)
2023                 tos = ip_tunnel_get_dsfield(old_iph, skb);
2024
2025         label = vxlan->cfg.label;
2026         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2027                                      vxlan->cfg.port_max, true);
2028
2029         if (info) {
2030                 ttl = info->key.ttl;
2031                 tos = info->key.tos;
2032                 label = info->key.label;
2033                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2034
2035                 if (info->options_len)
2036                         md = ip_tunnel_info_opts(info);
2037         } else {
2038                 md->gbp = skb->mark;
2039         }
2040
2041         if (dst->sa.sa_family == AF_INET) {
2042                 __be32 saddr;
2043
2044                 if (!vxlan->vn4_sock)
2045                         goto drop;
2046                 sk = vxlan->vn4_sock->sock->sk;
2047
2048                 rt = vxlan_get_route(vxlan, skb,
2049                                      rdst ? rdst->remote_ifindex : 0, tos,
2050                                      dst->sin.sin_addr.s_addr, &saddr,
2051                                      dst_cache, info);
2052                 if (IS_ERR(rt)) {
2053                         netdev_dbg(dev, "no route to %pI4\n",
2054                                    &dst->sin.sin_addr.s_addr);
2055                         dev->stats.tx_carrier_errors++;
2056                         goto tx_error;
2057                 }
2058
2059                 if (rt->dst.dev == dev) {
2060                         netdev_dbg(dev, "circular route to %pI4\n",
2061                                    &dst->sin.sin_addr.s_addr);
2062                         dev->stats.collisions++;
2063                         goto rt_tx_error;
2064                 }
2065
2066                 /* Bypass encapsulation if the destination is local */
2067                 if (rt->rt_flags & RTCF_LOCAL &&
2068                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2069                         struct vxlan_dev *dst_vxlan;
2070
2071                         ip_rt_put(rt);
2072                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2073                                                    dst->sa.sa_family, dst_port,
2074                                                    vxlan->flags);
2075                         if (!dst_vxlan)
2076                                 goto tx_error;
2077                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2078                         return;
2079                 }
2080
2081                 if (!info)
2082                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2083                 else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2084                         df = htons(IP_DF);
2085
2086                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2087                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2088                 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2089                                       vni, md, flags, udp_sum);
2090                 if (err < 0)
2091                         goto xmit_tx_error;
2092
2093                 udp_tunnel_xmit_skb(rt, sk, skb, saddr,
2094                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2095                                     src_port, dst_port, xnet, !udp_sum);
2096 #if IS_ENABLED(CONFIG_IPV6)
2097         } else {
2098                 struct dst_entry *ndst;
2099                 struct in6_addr saddr;
2100                 u32 rt6i_flags;
2101
2102                 if (!vxlan->vn6_sock)
2103                         goto drop;
2104                 sk = vxlan->vn6_sock->sock->sk;
2105
2106                 ndst = vxlan6_get_route(vxlan, skb,
2107                                         rdst ? rdst->remote_ifindex : 0, tos,
2108                                         label, &dst->sin6.sin6_addr, &saddr,
2109                                         dst_cache, info);
2110                 if (IS_ERR(ndst)) {
2111                         netdev_dbg(dev, "no route to %pI6\n",
2112                                    &dst->sin6.sin6_addr);
2113                         dev->stats.tx_carrier_errors++;
2114                         goto tx_error;
2115                 }
2116
2117                 if (ndst->dev == dev) {
2118                         netdev_dbg(dev, "circular route to %pI6\n",
2119                                    &dst->sin6.sin6_addr);
2120                         dst_release(ndst);
2121                         dev->stats.collisions++;
2122                         goto tx_error;
2123                 }
2124
2125                 /* Bypass encapsulation if the destination is local */
2126                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2127                 if (rt6i_flags & RTF_LOCAL &&
2128                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2129                         struct vxlan_dev *dst_vxlan;
2130
2131                         dst_release(ndst);
2132                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2133                                                    dst->sa.sa_family, dst_port,
2134                                                    vxlan->flags);
2135                         if (!dst_vxlan)
2136                                 goto tx_error;
2137                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2138                         return;
2139                 }
2140
2141                 if (!info)
2142                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2143
2144                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2145                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2146                 skb_scrub_packet(skb, xnet);
2147                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2148                                       vni, md, flags, udp_sum);
2149                 if (err < 0) {
2150                         dst_release(ndst);
2151                         return;
2152                 }
2153                 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2154                                      &saddr, &dst->sin6.sin6_addr, tos, ttl,
2155                                      label, src_port, dst_port, !udp_sum);
2156 #endif
2157         }
2158
2159         return;
2160
2161 drop:
2162         dev->stats.tx_dropped++;
2163         goto tx_free;
2164
2165 xmit_tx_error:
2166         /* skb is already freed. */
2167         skb = NULL;
2168 rt_tx_error:
2169         ip_rt_put(rt);
2170 tx_error:
2171         dev->stats.tx_errors++;
2172 tx_free:
2173         dev_kfree_skb(skb);
2174 }
2175
2176 /* Transmit local packets over Vxlan
2177  *
2178  * Outer IP header inherits ECN and DF from inner header.
2179  * Outer UDP destination is the VXLAN assigned port.
2180  *           source port is based on hash of flow
2181  */
2182 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2183 {
2184         struct vxlan_dev *vxlan = netdev_priv(dev);
2185         const struct ip_tunnel_info *info;
2186         struct ethhdr *eth;
2187         bool did_rsc = false;
2188         struct vxlan_rdst *rdst, *fdst = NULL;
2189         struct vxlan_fdb *f;
2190
2191         info = skb_tunnel_info(skb);
2192
2193         skb_reset_mac_header(skb);
2194
2195         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2196                 if (info && info->mode & IP_TUNNEL_INFO_TX)
2197                         vxlan_xmit_one(skb, dev, NULL, false);
2198                 else
2199                         kfree_skb(skb);
2200                 return NETDEV_TX_OK;
2201         }
2202
2203         if (vxlan->flags & VXLAN_F_PROXY) {
2204                 eth = eth_hdr(skb);
2205                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2206                         return arp_reduce(dev, skb);
2207 #if IS_ENABLED(CONFIG_IPV6)
2208                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2209                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2210                                        + sizeof(struct nd_msg)) &&
2211                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2212                                 struct nd_msg *msg;
2213
2214                                 msg = (struct nd_msg *)skb_transport_header(skb);
2215                                 if (msg->icmph.icmp6_code == 0 &&
2216                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2217                                         return neigh_reduce(dev, skb);
2218                 }
2219 #endif
2220         }
2221
2222         eth = eth_hdr(skb);
2223         f = vxlan_find_mac(vxlan, eth->h_dest);
2224         did_rsc = false;
2225
2226         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2227             (ntohs(eth->h_proto) == ETH_P_IP ||
2228              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2229                 did_rsc = route_shortcircuit(dev, skb);
2230                 if (did_rsc)
2231                         f = vxlan_find_mac(vxlan, eth->h_dest);
2232         }
2233
2234         if (f == NULL) {
2235                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2236                 if (f == NULL) {
2237                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2238                             !is_multicast_ether_addr(eth->h_dest))
2239                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2240
2241                         dev->stats.tx_dropped++;
2242                         kfree_skb(skb);
2243                         return NETDEV_TX_OK;
2244                 }
2245         }
2246
2247         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2248                 struct sk_buff *skb1;
2249
2250                 if (!fdst) {
2251                         fdst = rdst;
2252                         continue;
2253                 }
2254                 skb1 = skb_clone(skb, GFP_ATOMIC);
2255                 if (skb1)
2256                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2257         }
2258
2259         if (fdst)
2260                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2261         else
2262                 kfree_skb(skb);
2263         return NETDEV_TX_OK;
2264 }
2265
2266 /* Walk the forwarding table and purge stale entries */
2267 static void vxlan_cleanup(unsigned long arg)
2268 {
2269         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2270         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2271         unsigned int h;
2272
2273         if (!netif_running(vxlan->dev))
2274                 return;
2275
2276         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2277                 struct hlist_node *p, *n;
2278
2279                 spin_lock_bh(&vxlan->hash_lock);
2280                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2281                         struct vxlan_fdb *f
2282                                 = container_of(p, struct vxlan_fdb, hlist);
2283                         unsigned long timeout;
2284
2285                         if (f->state & NUD_PERMANENT)
2286                                 continue;
2287
2288                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2289                         if (time_before_eq(timeout, jiffies)) {
2290                                 netdev_dbg(vxlan->dev,
2291                                            "garbage collect %pM\n",
2292                                            f->eth_addr);
2293                                 f->state = NUD_STALE;
2294                                 vxlan_fdb_destroy(vxlan, f);
2295                         } else if (time_before(timeout, next_timer))
2296                                 next_timer = timeout;
2297                 }
2298                 spin_unlock_bh(&vxlan->hash_lock);
2299         }
2300
2301         mod_timer(&vxlan->age_timer, next_timer);
2302 }
2303
2304 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2305 {
2306         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2307         __be32 vni = vxlan->default_dst.remote_vni;
2308
2309         spin_lock(&vn->sock_lock);
2310         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2311         spin_unlock(&vn->sock_lock);
2312 }
2313
2314 /* Setup stats when device is created */
2315 static int vxlan_init(struct net_device *dev)
2316 {
2317         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2318         if (!dev->tstats)
2319                 return -ENOMEM;
2320
2321         return 0;
2322 }
2323
2324 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2325 {
2326         struct vxlan_fdb *f;
2327
2328         spin_lock_bh(&vxlan->hash_lock);
2329         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2330         if (f)
2331                 vxlan_fdb_destroy(vxlan, f);
2332         spin_unlock_bh(&vxlan->hash_lock);
2333 }
2334
2335 static void vxlan_uninit(struct net_device *dev)
2336 {
2337         struct vxlan_dev *vxlan = netdev_priv(dev);
2338
2339         vxlan_fdb_delete_default(vxlan);
2340
2341         free_percpu(dev->tstats);
2342 }
2343
2344 /* Start ageing timer and join group when device is brought up */
2345 static int vxlan_open(struct net_device *dev)
2346 {
2347         struct vxlan_dev *vxlan = netdev_priv(dev);
2348         int ret;
2349
2350         ret = vxlan_sock_add(vxlan);
2351         if (ret < 0)
2352                 return ret;
2353
2354         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2355                 ret = vxlan_igmp_join(vxlan);
2356                 if (ret == -EADDRINUSE)
2357                         ret = 0;
2358                 if (ret) {
2359                         vxlan_sock_release(vxlan);
2360                         return ret;
2361                 }
2362         }
2363
2364         if (vxlan->cfg.age_interval)
2365                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2366
2367         return ret;
2368 }
2369
2370 /* Purge the forwarding table */
2371 static void vxlan_flush(struct vxlan_dev *vxlan)
2372 {
2373         unsigned int h;
2374
2375         spin_lock_bh(&vxlan->hash_lock);
2376         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2377                 struct hlist_node *p, *n;
2378                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2379                         struct vxlan_fdb *f
2380                                 = container_of(p, struct vxlan_fdb, hlist);
2381                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2382                         if (!is_zero_ether_addr(f->eth_addr))
2383                                 vxlan_fdb_destroy(vxlan, f);
2384                 }
2385         }
2386         spin_unlock_bh(&vxlan->hash_lock);
2387 }
2388
2389 /* Cleanup timer and forwarding table on shutdown */
2390 static int vxlan_stop(struct net_device *dev)
2391 {
2392         struct vxlan_dev *vxlan = netdev_priv(dev);
2393         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2394         int ret = 0;
2395
2396         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2397             !vxlan_group_used(vn, vxlan))
2398                 ret = vxlan_igmp_leave(vxlan);
2399
2400         del_timer_sync(&vxlan->age_timer);
2401
2402         vxlan_flush(vxlan);
2403         vxlan_sock_release(vxlan);
2404
2405         return ret;
2406 }
2407
2408 /* Stub, nothing needs to be done. */
2409 static void vxlan_set_multicast_list(struct net_device *dev)
2410 {
2411 }
2412
2413 static int __vxlan_change_mtu(struct net_device *dev,
2414                               struct net_device *lowerdev,
2415                               struct vxlan_rdst *dst, int new_mtu, bool strict)
2416 {
2417         int max_mtu = IP_MAX_MTU;
2418
2419         if (lowerdev)
2420                 max_mtu = lowerdev->mtu;
2421
2422         if (dst->remote_ip.sa.sa_family == AF_INET6)
2423                 max_mtu -= VXLAN6_HEADROOM;
2424         else
2425                 max_mtu -= VXLAN_HEADROOM;
2426
2427         if (new_mtu < 68)
2428                 return -EINVAL;
2429
2430         if (new_mtu > max_mtu) {
2431                 if (strict)
2432                         return -EINVAL;
2433
2434                 new_mtu = max_mtu;
2435         }
2436
2437         dev->mtu = new_mtu;
2438         return 0;
2439 }
2440
2441 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2442 {
2443         struct vxlan_dev *vxlan = netdev_priv(dev);
2444         struct vxlan_rdst *dst = &vxlan->default_dst;
2445         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2446                                                          dst->remote_ifindex);
2447         return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2448 }
2449
2450 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2451 {
2452         struct vxlan_dev *vxlan = netdev_priv(dev);
2453         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2454         __be16 sport, dport;
2455
2456         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2457                                   vxlan->cfg.port_max, true);
2458         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2459
2460         if (ip_tunnel_info_af(info) == AF_INET) {
2461                 struct rtable *rt;
2462
2463                 if (!vxlan->vn4_sock)
2464                         return -EINVAL;
2465                 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2466                                      info->key.u.ipv4.dst,
2467                                      &info->key.u.ipv4.src, NULL, info);
2468                 if (IS_ERR(rt))
2469                         return PTR_ERR(rt);
2470                 ip_rt_put(rt);
2471         } else {
2472 #if IS_ENABLED(CONFIG_IPV6)
2473                 struct dst_entry *ndst;
2474
2475                 if (!vxlan->vn6_sock)
2476                         return -EINVAL;
2477                 ndst = vxlan6_get_route(vxlan, skb, 0, info->key.tos,
2478                                         info->key.label, &info->key.u.ipv6.dst,
2479                                         &info->key.u.ipv6.src, NULL, info);
2480                 if (IS_ERR(ndst))
2481                         return PTR_ERR(ndst);
2482                 dst_release(ndst);
2483 #else /* !CONFIG_IPV6 */
2484                 return -EPFNOSUPPORT;
2485 #endif
2486         }
2487         info->key.tp_src = sport;
2488         info->key.tp_dst = dport;
2489         return 0;
2490 }
2491
2492 static const struct net_device_ops vxlan_netdev_ether_ops = {
2493         .ndo_init               = vxlan_init,
2494         .ndo_uninit             = vxlan_uninit,
2495         .ndo_open               = vxlan_open,
2496         .ndo_stop               = vxlan_stop,
2497         .ndo_start_xmit         = vxlan_xmit,
2498         .ndo_get_stats64        = ip_tunnel_get_stats64,
2499         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2500         .ndo_change_mtu         = vxlan_change_mtu,
2501         .ndo_validate_addr      = eth_validate_addr,
2502         .ndo_set_mac_address    = eth_mac_addr,
2503         .ndo_fdb_add            = vxlan_fdb_add,
2504         .ndo_fdb_del            = vxlan_fdb_delete,
2505         .ndo_fdb_dump           = vxlan_fdb_dump,
2506         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2507 };
2508
2509 static const struct net_device_ops vxlan_netdev_raw_ops = {
2510         .ndo_init               = vxlan_init,
2511         .ndo_uninit             = vxlan_uninit,
2512         .ndo_open               = vxlan_open,
2513         .ndo_stop               = vxlan_stop,
2514         .ndo_start_xmit         = vxlan_xmit,
2515         .ndo_get_stats64        = ip_tunnel_get_stats64,
2516         .ndo_change_mtu         = vxlan_change_mtu,
2517         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2518 };
2519
2520 /* Info for udev, that this is a virtual tunnel endpoint */
2521 static struct device_type vxlan_type = {
2522         .name = "vxlan",
2523 };
2524
2525 /* Calls the ndo_add_vxlan_port of the caller in order to
2526  * supply the listening VXLAN udp ports. Callers are expected
2527  * to implement the ndo_add_vxlan_port.
2528  */
2529 void vxlan_get_rx_port(struct net_device *dev)
2530 {
2531         struct vxlan_sock *vs;
2532         struct net *net = dev_net(dev);
2533         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2534         sa_family_t sa_family;
2535         __be16 port;
2536         unsigned int i;
2537
2538         spin_lock(&vn->sock_lock);
2539         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2540                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2541                         port = inet_sk(vs->sock->sk)->inet_sport;
2542                         sa_family = vxlan_get_sk_family(vs);
2543                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2544                                                             port);
2545                 }
2546         }
2547         spin_unlock(&vn->sock_lock);
2548 }
2549 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2550
2551 /* Initialize the device structure. */
2552 static void vxlan_setup(struct net_device *dev)
2553 {
2554         struct vxlan_dev *vxlan = netdev_priv(dev);
2555         unsigned int h;
2556
2557         dev->destructor = free_netdev;
2558         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2559
2560         dev->features   |= NETIF_F_LLTX;
2561         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2562         dev->features   |= NETIF_F_RXCSUM;
2563         dev->features   |= NETIF_F_GSO_SOFTWARE;
2564
2565         dev->vlan_features = dev->features;
2566         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2567         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2568         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2569         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2570         netif_keep_dst(dev);
2571         dev->priv_flags |= IFF_NO_QUEUE;
2572
2573         INIT_LIST_HEAD(&vxlan->next);
2574         spin_lock_init(&vxlan->hash_lock);
2575
2576         init_timer_deferrable(&vxlan->age_timer);
2577         vxlan->age_timer.function = vxlan_cleanup;
2578         vxlan->age_timer.data = (unsigned long) vxlan;
2579
2580         vxlan->cfg.dst_port = htons(vxlan_port);
2581
2582         vxlan->dev = dev;
2583
2584         gro_cells_init(&vxlan->gro_cells, dev);
2585
2586         for (h = 0; h < FDB_HASH_SIZE; ++h)
2587                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2588 }
2589
2590 static void vxlan_ether_setup(struct net_device *dev)
2591 {
2592         eth_hw_addr_random(dev);
2593         ether_setup(dev);
2594         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2595         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2596         dev->netdev_ops = &vxlan_netdev_ether_ops;
2597 }
2598
2599 static void vxlan_raw_setup(struct net_device *dev)
2600 {
2601         dev->type = ARPHRD_NONE;
2602         dev->hard_header_len = 0;
2603         dev->addr_len = 0;
2604         dev->mtu = ETH_DATA_LEN;
2605         dev->tx_queue_len = 1000;
2606         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2607         dev->netdev_ops = &vxlan_netdev_raw_ops;
2608 }
2609
2610 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2611         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2612         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2613         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2614         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2615         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2616         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2617         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2618         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2619         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2620         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2621         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2622         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2623         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2624         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2625         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2626         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2627         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2628         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2629         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2630         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2631         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2632         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2633         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2634         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2635         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2636         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2637         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2638 };
2639
2640 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2641 {
2642         if (tb[IFLA_ADDRESS]) {
2643                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2644                         pr_debug("invalid link address (not ethernet)\n");
2645                         return -EINVAL;
2646                 }
2647
2648                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2649                         pr_debug("invalid all zero ethernet address\n");
2650                         return -EADDRNOTAVAIL;
2651                 }
2652         }
2653
2654         if (!data)
2655                 return -EINVAL;
2656
2657         if (data[IFLA_VXLAN_ID]) {
2658                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2659                 if (id >= VXLAN_VID_MASK)
2660                         return -ERANGE;
2661         }
2662
2663         if (data[IFLA_VXLAN_PORT_RANGE]) {
2664                 const struct ifla_vxlan_port_range *p
2665                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2666
2667                 if (ntohs(p->high) < ntohs(p->low)) {
2668                         pr_debug("port range %u .. %u not valid\n",
2669                                  ntohs(p->low), ntohs(p->high));
2670                         return -EINVAL;
2671                 }
2672         }
2673
2674         return 0;
2675 }
2676
2677 static void vxlan_get_drvinfo(struct net_device *netdev,
2678                               struct ethtool_drvinfo *drvinfo)
2679 {
2680         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2681         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2682 }
2683
2684 static const struct ethtool_ops vxlan_ethtool_ops = {
2685         .get_drvinfo    = vxlan_get_drvinfo,
2686         .get_link       = ethtool_op_get_link,
2687 };
2688
2689 static void vxlan_del_work(struct work_struct *work)
2690 {
2691         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2692         udp_tunnel_sock_release(vs->sock);
2693         kfree_rcu(vs, rcu);
2694 }
2695
2696 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2697                                         __be16 port, u32 flags)
2698 {
2699         struct socket *sock;
2700         struct udp_port_cfg udp_conf;
2701         int err;
2702
2703         memset(&udp_conf, 0, sizeof(udp_conf));
2704
2705         if (ipv6) {
2706                 udp_conf.family = AF_INET6;
2707                 udp_conf.use_udp6_rx_checksums =
2708                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2709                 udp_conf.ipv6_v6only = 1;
2710         } else {
2711                 udp_conf.family = AF_INET;
2712         }
2713
2714         udp_conf.local_udp_port = port;
2715
2716         /* Open UDP socket */
2717         err = udp_sock_create(net, &udp_conf, &sock);
2718         if (err < 0)
2719                 return ERR_PTR(err);
2720
2721         return sock;
2722 }
2723
2724 /* Create new listen socket if needed */
2725 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2726                                               __be16 port, u32 flags)
2727 {
2728         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2729         struct vxlan_sock *vs;
2730         struct socket *sock;
2731         unsigned int h;
2732         struct udp_tunnel_sock_cfg tunnel_cfg;
2733
2734         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2735         if (!vs)
2736                 return ERR_PTR(-ENOMEM);
2737
2738         for (h = 0; h < VNI_HASH_SIZE; ++h)
2739                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2740
2741         INIT_WORK(&vs->del_work, vxlan_del_work);
2742
2743         sock = vxlan_create_sock(net, ipv6, port, flags);
2744         if (IS_ERR(sock)) {
2745                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2746                         PTR_ERR(sock));
2747                 kfree(vs);
2748                 return ERR_CAST(sock);
2749         }
2750
2751         vs->sock = sock;
2752         atomic_set(&vs->refcnt, 1);
2753         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2754
2755         /* Initialize the vxlan udp offloads structure */
2756         vs->udp_offloads.port = port;
2757         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2758         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2759
2760         spin_lock(&vn->sock_lock);
2761         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2762         vxlan_notify_add_rx_port(vs);
2763         spin_unlock(&vn->sock_lock);
2764
2765         /* Mark socket as an encapsulation socket. */
2766         tunnel_cfg.sk_user_data = vs;
2767         tunnel_cfg.encap_type = 1;
2768         tunnel_cfg.encap_rcv = vxlan_rcv;
2769         tunnel_cfg.encap_destroy = NULL;
2770
2771         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2772
2773         return vs;
2774 }
2775
2776 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2777 {
2778         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2779         struct vxlan_sock *vs = NULL;
2780
2781         if (!vxlan->cfg.no_share) {
2782                 spin_lock(&vn->sock_lock);
2783                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2784                                      vxlan->cfg.dst_port, vxlan->flags);
2785                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2786                         spin_unlock(&vn->sock_lock);
2787                         return -EBUSY;
2788                 }
2789                 spin_unlock(&vn->sock_lock);
2790         }
2791         if (!vs)
2792                 vs = vxlan_socket_create(vxlan->net, ipv6,
2793                                          vxlan->cfg.dst_port, vxlan->flags);
2794         if (IS_ERR(vs))
2795                 return PTR_ERR(vs);
2796 #if IS_ENABLED(CONFIG_IPV6)
2797         if (ipv6)
2798                 vxlan->vn6_sock = vs;
2799         else
2800 #endif
2801                 vxlan->vn4_sock = vs;
2802         vxlan_vs_add_dev(vs, vxlan);
2803         return 0;
2804 }
2805
2806 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2807 {
2808         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2809         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2810         int ret = 0;
2811
2812         vxlan->vn4_sock = NULL;
2813 #if IS_ENABLED(CONFIG_IPV6)
2814         vxlan->vn6_sock = NULL;
2815         if (ipv6 || metadata)
2816                 ret = __vxlan_sock_add(vxlan, true);
2817 #endif
2818         if (!ret && (!ipv6 || metadata))
2819                 ret = __vxlan_sock_add(vxlan, false);
2820         if (ret < 0)
2821                 vxlan_sock_release(vxlan);
2822         return ret;
2823 }
2824
2825 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2826                                struct vxlan_config *conf)
2827 {
2828         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2829         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2830         struct vxlan_rdst *dst = &vxlan->default_dst;
2831         unsigned short needed_headroom = ETH_HLEN;
2832         int err;
2833         bool use_ipv6 = false;
2834         __be16 default_port = vxlan->cfg.dst_port;
2835         struct net_device *lowerdev = NULL;
2836
2837         if (conf->flags & VXLAN_F_GPE) {
2838                 if (conf->flags & ~VXLAN_F_ALLOWED_GPE)
2839                         return -EINVAL;
2840                 /* For now, allow GPE only together with COLLECT_METADATA.
2841                  * This can be relaxed later; in such case, the other side
2842                  * of the PtP link will have to be provided.
2843                  */
2844                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA))
2845                         return -EINVAL;
2846
2847                 vxlan_raw_setup(dev);
2848         } else {
2849                 vxlan_ether_setup(dev);
2850         }
2851
2852         vxlan->net = src_net;
2853
2854         dst->remote_vni = conf->vni;
2855
2856         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2857
2858         /* Unless IPv6 is explicitly requested, assume IPv4 */
2859         if (!dst->remote_ip.sa.sa_family)
2860                 dst->remote_ip.sa.sa_family = AF_INET;
2861
2862         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2863             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2864                 if (!IS_ENABLED(CONFIG_IPV6))
2865                         return -EPFNOSUPPORT;
2866                 use_ipv6 = true;
2867                 vxlan->flags |= VXLAN_F_IPV6;
2868         }
2869
2870         if (conf->label && !use_ipv6) {
2871                 pr_info("label only supported in use with IPv6\n");
2872                 return -EINVAL;
2873         }
2874
2875         if (conf->remote_ifindex) {
2876                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2877                 dst->remote_ifindex = conf->remote_ifindex;
2878
2879                 if (!lowerdev) {
2880                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2881                         return -ENODEV;
2882                 }
2883
2884 #if IS_ENABLED(CONFIG_IPV6)
2885                 if (use_ipv6) {
2886                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2887                         if (idev && idev->cnf.disable_ipv6) {
2888                                 pr_info("IPv6 is disabled via sysctl\n");
2889                                 return -EPERM;
2890                         }
2891                 }
2892 #endif
2893
2894                 if (!conf->mtu)
2895                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2896
2897                 needed_headroom = lowerdev->hard_header_len;
2898         }
2899
2900         if (conf->mtu) {
2901                 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2902                 if (err)
2903                         return err;
2904         }
2905
2906         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2907                 needed_headroom += VXLAN6_HEADROOM;
2908         else
2909                 needed_headroom += VXLAN_HEADROOM;
2910         dev->needed_headroom = needed_headroom;
2911
2912         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2913         if (!vxlan->cfg.dst_port) {
2914                 if (conf->flags & VXLAN_F_GPE)
2915                         vxlan->cfg.dst_port = 4790; /* IANA assigned VXLAN-GPE port */
2916                 else
2917                         vxlan->cfg.dst_port = default_port;
2918         }
2919         vxlan->flags |= conf->flags;
2920
2921         if (!vxlan->cfg.age_interval)
2922                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2923
2924         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2925                 if (tmp->cfg.vni == conf->vni &&
2926                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2927                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2928                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2929                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2930                     (vxlan->flags & VXLAN_F_RCV_FLAGS))
2931                 return -EEXIST;
2932         }
2933
2934         dev->ethtool_ops = &vxlan_ethtool_ops;
2935
2936         /* create an fdb entry for a valid default destination */
2937         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2938                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2939                                        &vxlan->default_dst.remote_ip,
2940                                        NUD_REACHABLE|NUD_PERMANENT,
2941                                        NLM_F_EXCL|NLM_F_CREATE,
2942                                        vxlan->cfg.dst_port,
2943                                        vxlan->default_dst.remote_vni,
2944                                        vxlan->default_dst.remote_ifindex,
2945                                        NTF_SELF);
2946                 if (err)
2947                         return err;
2948         }
2949
2950         err = register_netdevice(dev);
2951         if (err) {
2952                 vxlan_fdb_delete_default(vxlan);
2953                 return err;
2954         }
2955
2956         list_add(&vxlan->next, &vn->vxlan_list);
2957
2958         return 0;
2959 }
2960
2961 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2962                                     u8 name_assign_type, struct vxlan_config *conf)
2963 {
2964         struct nlattr *tb[IFLA_MAX+1];
2965         struct net_device *dev;
2966         int err;
2967
2968         memset(&tb, 0, sizeof(tb));
2969
2970         dev = rtnl_create_link(net, name, name_assign_type,
2971                                &vxlan_link_ops, tb);
2972         if (IS_ERR(dev))
2973                 return dev;
2974
2975         err = vxlan_dev_configure(net, dev, conf);
2976         if (err < 0) {
2977                 free_netdev(dev);
2978                 return ERR_PTR(err);
2979         }
2980
2981         return dev;
2982 }
2983 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2984
2985 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2986                          struct nlattr *tb[], struct nlattr *data[])
2987 {
2988         struct vxlan_config conf;
2989         int err;
2990
2991         memset(&conf, 0, sizeof(conf));
2992
2993         if (data[IFLA_VXLAN_ID])
2994                 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2995
2996         if (data[IFLA_VXLAN_GROUP]) {
2997                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2998         } else if (data[IFLA_VXLAN_GROUP6]) {
2999                 if (!IS_ENABLED(CONFIG_IPV6))
3000                         return -EPFNOSUPPORT;
3001
3002                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3003                 conf.remote_ip.sa.sa_family = AF_INET6;
3004         }
3005
3006         if (data[IFLA_VXLAN_LOCAL]) {
3007                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3008                 conf.saddr.sa.sa_family = AF_INET;
3009         } else if (data[IFLA_VXLAN_LOCAL6]) {
3010                 if (!IS_ENABLED(CONFIG_IPV6))
3011                         return -EPFNOSUPPORT;
3012
3013                 /* TODO: respect scope id */
3014                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3015                 conf.saddr.sa.sa_family = AF_INET6;
3016         }
3017
3018         if (data[IFLA_VXLAN_LINK])
3019                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3020
3021         if (data[IFLA_VXLAN_TOS])
3022                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3023
3024         if (data[IFLA_VXLAN_TTL])
3025                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3026
3027         if (data[IFLA_VXLAN_LABEL])
3028                 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3029                              IPV6_FLOWLABEL_MASK;
3030
3031         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3032                 conf.flags |= VXLAN_F_LEARN;
3033
3034         if (data[IFLA_VXLAN_AGEING])
3035                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3036
3037         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
3038                 conf.flags |= VXLAN_F_PROXY;
3039
3040         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
3041                 conf.flags |= VXLAN_F_RSC;
3042
3043         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3044                 conf.flags |= VXLAN_F_L2MISS;
3045
3046         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3047                 conf.flags |= VXLAN_F_L3MISS;
3048
3049         if (data[IFLA_VXLAN_LIMIT])
3050                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3051
3052         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
3053             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3054                 conf.flags |= VXLAN_F_COLLECT_METADATA;
3055
3056         if (data[IFLA_VXLAN_PORT_RANGE]) {
3057                 const struct ifla_vxlan_port_range *p
3058                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3059                 conf.port_min = ntohs(p->low);
3060                 conf.port_max = ntohs(p->high);
3061         }
3062
3063         if (data[IFLA_VXLAN_PORT])
3064                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3065
3066         if (data[IFLA_VXLAN_UDP_CSUM] &&
3067             !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3068                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3069
3070         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
3071             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3072                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3073
3074         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3075             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3076                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3077
3078         if (data[IFLA_VXLAN_REMCSUM_TX] &&
3079             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3080                 conf.flags |= VXLAN_F_REMCSUM_TX;
3081
3082         if (data[IFLA_VXLAN_REMCSUM_RX] &&
3083             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3084                 conf.flags |= VXLAN_F_REMCSUM_RX;
3085
3086         if (data[IFLA_VXLAN_GBP])
3087                 conf.flags |= VXLAN_F_GBP;
3088
3089         if (data[IFLA_VXLAN_GPE])
3090                 conf.flags |= VXLAN_F_GPE;
3091
3092         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3093                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3094
3095         err = vxlan_dev_configure(src_net, dev, &conf);
3096         switch (err) {
3097         case -ENODEV:
3098                 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
3099                 break;
3100
3101         case -EPERM:
3102                 pr_info("IPv6 is disabled via sysctl\n");
3103                 break;
3104
3105         case -EEXIST:
3106                 pr_info("duplicate VNI %u\n", be32_to_cpu(conf.vni));
3107                 break;
3108
3109         case -EINVAL:
3110                 pr_info("unsupported combination of extensions\n");
3111                 break;
3112         }
3113
3114         return err;
3115 }
3116
3117 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3118 {
3119         struct vxlan_dev *vxlan = netdev_priv(dev);
3120         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3121
3122         spin_lock(&vn->sock_lock);
3123         if (!hlist_unhashed(&vxlan->hlist))
3124                 hlist_del_rcu(&vxlan->hlist);
3125         spin_unlock(&vn->sock_lock);
3126
3127         gro_cells_destroy(&vxlan->gro_cells);
3128         list_del(&vxlan->next);
3129         unregister_netdevice_queue(dev, head);
3130 }
3131
3132 static size_t vxlan_get_size(const struct net_device *dev)
3133 {
3134
3135         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3136                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3137                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3138                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3139                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3140                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3141                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3142                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3143                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3144                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3145                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3146                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3147                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3148                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3149                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3150                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3151                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3152                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3153                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3154                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3155                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3156                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3157                 0;
3158 }
3159
3160 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3161 {
3162         const struct vxlan_dev *vxlan = netdev_priv(dev);
3163         const struct vxlan_rdst *dst = &vxlan->default_dst;
3164         struct ifla_vxlan_port_range ports = {
3165                 .low =  htons(vxlan->cfg.port_min),
3166                 .high = htons(vxlan->cfg.port_max),
3167         };
3168
3169         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3170                 goto nla_put_failure;
3171
3172         if (!vxlan_addr_any(&dst->remote_ip)) {
3173                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3174                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3175                                             dst->remote_ip.sin.sin_addr.s_addr))
3176                                 goto nla_put_failure;
3177 #if IS_ENABLED(CONFIG_IPV6)
3178                 } else {
3179                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3180                                              &dst->remote_ip.sin6.sin6_addr))
3181                                 goto nla_put_failure;
3182 #endif
3183                 }
3184         }
3185
3186         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3187                 goto nla_put_failure;
3188
3189         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3190                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3191                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3192                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3193                                 goto nla_put_failure;
3194 #if IS_ENABLED(CONFIG_IPV6)
3195                 } else {
3196                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3197                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3198                                 goto nla_put_failure;
3199 #endif
3200                 }
3201         }
3202
3203         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3204             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3205             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3206             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3207                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3208             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3209                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3210             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3211             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3212                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3213             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3214                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3215             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3216                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3217             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3218             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3219             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3220             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3221                         !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3222             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3223                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3224             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3225                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3226             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3227                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3228             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3229                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3230                 goto nla_put_failure;
3231
3232         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3233                 goto nla_put_failure;
3234
3235         if (vxlan->flags & VXLAN_F_GBP &&
3236             nla_put_flag(skb, IFLA_VXLAN_GBP))
3237                 goto nla_put_failure;
3238
3239         if (vxlan->flags & VXLAN_F_GPE &&
3240             nla_put_flag(skb, IFLA_VXLAN_GPE))
3241                 goto nla_put_failure;
3242
3243         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3244             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3245                 goto nla_put_failure;
3246
3247         return 0;
3248
3249 nla_put_failure:
3250         return -EMSGSIZE;
3251 }
3252
3253 static struct net *vxlan_get_link_net(const struct net_device *dev)
3254 {
3255         struct vxlan_dev *vxlan = netdev_priv(dev);
3256
3257         return vxlan->net;
3258 }
3259
3260 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3261         .kind           = "vxlan",
3262         .maxtype        = IFLA_VXLAN_MAX,
3263         .policy         = vxlan_policy,
3264         .priv_size      = sizeof(struct vxlan_dev),
3265         .setup          = vxlan_setup,
3266         .validate       = vxlan_validate,
3267         .newlink        = vxlan_newlink,
3268         .dellink        = vxlan_dellink,
3269         .get_size       = vxlan_get_size,
3270         .fill_info      = vxlan_fill_info,
3271         .get_link_net   = vxlan_get_link_net,
3272 };
3273
3274 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3275                                              struct net_device *dev)
3276 {
3277         struct vxlan_dev *vxlan, *next;
3278         LIST_HEAD(list_kill);
3279
3280         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3281                 struct vxlan_rdst *dst = &vxlan->default_dst;
3282
3283                 /* In case we created vxlan device with carrier
3284                  * and we loose the carrier due to module unload
3285                  * we also need to remove vxlan device. In other
3286                  * cases, it's not necessary and remote_ifindex
3287                  * is 0 here, so no matches.
3288                  */
3289                 if (dst->remote_ifindex == dev->ifindex)
3290                         vxlan_dellink(vxlan->dev, &list_kill);
3291         }
3292
3293         unregister_netdevice_many(&list_kill);
3294 }
3295
3296 static int vxlan_lowerdev_event(struct notifier_block *unused,
3297                                 unsigned long event, void *ptr)
3298 {
3299         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3300         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3301
3302         if (event == NETDEV_UNREGISTER)
3303                 vxlan_handle_lowerdev_unregister(vn, dev);
3304
3305         return NOTIFY_DONE;
3306 }
3307
3308 static struct notifier_block vxlan_notifier_block __read_mostly = {
3309         .notifier_call = vxlan_lowerdev_event,
3310 };
3311
3312 static __net_init int vxlan_init_net(struct net *net)
3313 {
3314         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3315         unsigned int h;
3316
3317         INIT_LIST_HEAD(&vn->vxlan_list);
3318         spin_lock_init(&vn->sock_lock);
3319
3320         for (h = 0; h < PORT_HASH_SIZE; ++h)
3321                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3322
3323         return 0;
3324 }
3325
3326 static void __net_exit vxlan_exit_net(struct net *net)
3327 {
3328         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3329         struct vxlan_dev *vxlan, *next;
3330         struct net_device *dev, *aux;
3331         LIST_HEAD(list);
3332
3333         rtnl_lock();
3334         for_each_netdev_safe(net, dev, aux)
3335                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3336                         unregister_netdevice_queue(dev, &list);
3337
3338         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3339                 /* If vxlan->dev is in the same netns, it has already been added
3340                  * to the list by the previous loop.
3341                  */
3342                 if (!net_eq(dev_net(vxlan->dev), net)) {
3343                         gro_cells_destroy(&vxlan->gro_cells);
3344                         unregister_netdevice_queue(vxlan->dev, &list);
3345                 }
3346         }
3347
3348         unregister_netdevice_many(&list);
3349         rtnl_unlock();
3350 }
3351
3352 static struct pernet_operations vxlan_net_ops = {
3353         .init = vxlan_init_net,
3354         .exit = vxlan_exit_net,
3355         .id   = &vxlan_net_id,
3356         .size = sizeof(struct vxlan_net),
3357 };
3358
3359 static int __init vxlan_init_module(void)
3360 {
3361         int rc;
3362
3363         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3364         if (!vxlan_wq)
3365                 return -ENOMEM;
3366
3367         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3368
3369         rc = register_pernet_subsys(&vxlan_net_ops);
3370         if (rc)
3371                 goto out1;
3372
3373         rc = register_netdevice_notifier(&vxlan_notifier_block);
3374         if (rc)
3375                 goto out2;
3376
3377         rc = rtnl_link_register(&vxlan_link_ops);
3378         if (rc)
3379                 goto out3;
3380
3381         return 0;
3382 out3:
3383         unregister_netdevice_notifier(&vxlan_notifier_block);
3384 out2:
3385         unregister_pernet_subsys(&vxlan_net_ops);
3386 out1:
3387         destroy_workqueue(vxlan_wq);
3388         return rc;
3389 }
3390 late_initcall(vxlan_init_module);
3391
3392 static void __exit vxlan_cleanup_module(void)
3393 {
3394         rtnl_link_unregister(&vxlan_link_ops);
3395         unregister_netdevice_notifier(&vxlan_notifier_block);
3396         destroy_workqueue(vxlan_wq);
3397         unregister_pernet_subsys(&vxlan_net_ops);
3398         /* rcu_barrier() is called by netns */
3399 }
3400 module_exit(vxlan_cleanup_module);
3401
3402 MODULE_LICENSE("GPL");
3403 MODULE_VERSION(VXLAN_VERSION);
3404 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3405 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3406 MODULE_ALIAS_RTNL_LINK("vxlan");