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, plen;
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         plen = sizeof(struct vxlanhdr) + offset + sizeof(u16);
1155
1156         if (!pskb_may_pull(skb, plen))
1157                 return false;
1158
1159         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1160                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1161 out:
1162         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1163         unparsed->vx_vni &= VXLAN_VNI_MASK;
1164         return true;
1165 }
1166
1167 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1168                                 struct sk_buff *skb, u32 vxflags,
1169                                 struct vxlan_metadata *md)
1170 {
1171         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1172         struct metadata_dst *tun_dst;
1173
1174         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1175                 goto out;
1176
1177         md->gbp = ntohs(gbp->policy_id);
1178
1179         tun_dst = (struct metadata_dst *)skb_dst(skb);
1180         if (tun_dst) {
1181                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1182                 tun_dst->u.tun_info.options_len = sizeof(*md);
1183         }
1184         if (gbp->dont_learn)
1185                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1186
1187         if (gbp->policy_applied)
1188                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1189
1190         /* In flow-based mode, GBP is carried in dst_metadata */
1191         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1192                 skb->mark = md->gbp;
1193 out:
1194         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1195 }
1196
1197 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1198                           struct vxlan_sock *vs,
1199                           struct sk_buff *skb)
1200 {
1201         union vxlan_addr saddr;
1202
1203         skb_reset_mac_header(skb);
1204         skb->protocol = eth_type_trans(skb, vxlan->dev);
1205         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1206
1207         /* Ignore packet loops (and multicast echo) */
1208         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1209                 return false;
1210
1211         /* Get address from the outer IP header */
1212         if (vxlan_get_sk_family(vs) == AF_INET) {
1213                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1214                 saddr.sa.sa_family = AF_INET;
1215 #if IS_ENABLED(CONFIG_IPV6)
1216         } else {
1217                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1218                 saddr.sa.sa_family = AF_INET6;
1219 #endif
1220         }
1221
1222         if ((vxlan->flags & VXLAN_F_LEARN) &&
1223             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1224                 return false;
1225
1226         return true;
1227 }
1228
1229 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1230                                   struct sk_buff *skb)
1231 {
1232         int err = 0;
1233
1234         if (vxlan_get_sk_family(vs) == AF_INET)
1235                 err = IP_ECN_decapsulate(oiph, skb);
1236 #if IS_ENABLED(CONFIG_IPV6)
1237         else
1238                 err = IP6_ECN_decapsulate(oiph, skb);
1239 #endif
1240
1241         if (unlikely(err) && log_ecn_error) {
1242                 if (vxlan_get_sk_family(vs) == AF_INET)
1243                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1244                                              &((struct iphdr *)oiph)->saddr,
1245                                              ((struct iphdr *)oiph)->tos);
1246                 else
1247                         net_info_ratelimited("non-ECT from %pI6\n",
1248                                              &((struct ipv6hdr *)oiph)->saddr);
1249         }
1250         return err <= 1;
1251 }
1252
1253 /* Callback from net/ipv4/udp.c to receive packets */
1254 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1255 {
1256         struct pcpu_sw_netstats *stats;
1257         struct vxlan_dev *vxlan;
1258         struct vxlan_sock *vs;
1259         struct vxlanhdr unparsed;
1260         struct vxlan_metadata _md;
1261         struct vxlan_metadata *md = &_md;
1262         void *oiph;
1263
1264         /* Need Vxlan and inner Ethernet header to be present */
1265         if (!pskb_may_pull(skb, VXLAN_HLEN))
1266                 return 1;
1267
1268         unparsed = *vxlan_hdr(skb);
1269         /* VNI flag always required to be set */
1270         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1271                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1272                            ntohl(vxlan_hdr(skb)->vx_flags),
1273                            ntohl(vxlan_hdr(skb)->vx_vni));
1274                 /* Return non vxlan pkt */
1275                 return 1;
1276         }
1277         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1278         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1279
1280         vs = rcu_dereference_sk_user_data(sk);
1281         if (!vs)
1282                 goto drop;
1283
1284         vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1285         if (!vxlan)
1286                 goto drop;
1287
1288         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB),
1289                                  !net_eq(vxlan->net, dev_net(vxlan->dev))))
1290                 goto drop;
1291
1292         if (vxlan_collect_metadata(vs)) {
1293                 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1294                 struct metadata_dst *tun_dst;
1295
1296                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1297                                          vxlan_vni_to_tun_id(vni), sizeof(*md));
1298
1299                 if (!tun_dst)
1300                         goto drop;
1301
1302                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1303
1304                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1305         } else {
1306                 memset(md, 0, sizeof(*md));
1307         }
1308
1309         /* For backwards compatibility, only allow reserved fields to be
1310          * used by VXLAN extensions if explicitly requested.
1311          */
1312         if (vs->flags & VXLAN_F_REMCSUM_RX)
1313                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1314                         goto drop;
1315         if (vs->flags & VXLAN_F_GBP)
1316                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1317
1318         if (unparsed.vx_flags || unparsed.vx_vni) {
1319                 /* If there are any unprocessed flags remaining treat
1320                  * this as a malformed packet. This behavior diverges from
1321                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1322                  * in reserved fields are to be ignored. The approach here
1323                  * maintains compatibility with previous stack code, and also
1324                  * is more robust and provides a little more security in
1325                  * adding extensions to VXLAN.
1326                  */
1327                 goto drop;
1328         }
1329
1330         if (!vxlan_set_mac(vxlan, vs, skb))
1331                 goto drop;
1332
1333         oiph = skb_network_header(skb);
1334         skb_reset_network_header(skb);
1335
1336         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1337                 ++vxlan->dev->stats.rx_frame_errors;
1338                 ++vxlan->dev->stats.rx_errors;
1339                 goto drop;
1340         }
1341
1342         stats = this_cpu_ptr(vxlan->dev->tstats);
1343         u64_stats_update_begin(&stats->syncp);
1344         stats->rx_packets++;
1345         stats->rx_bytes += skb->len;
1346         u64_stats_update_end(&stats->syncp);
1347
1348         gro_cells_receive(&vxlan->gro_cells, skb);
1349         return 0;
1350
1351 drop:
1352         /* Consume bad packet */
1353         kfree_skb(skb);
1354         return 0;
1355 }
1356
1357 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1358 {
1359         struct vxlan_dev *vxlan = netdev_priv(dev);
1360         struct arphdr *parp;
1361         u8 *arpptr, *sha;
1362         __be32 sip, tip;
1363         struct neighbour *n;
1364
1365         if (dev->flags & IFF_NOARP)
1366                 goto out;
1367
1368         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1369                 dev->stats.tx_dropped++;
1370                 goto out;
1371         }
1372         parp = arp_hdr(skb);
1373
1374         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1375              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1376             parp->ar_pro != htons(ETH_P_IP) ||
1377             parp->ar_op != htons(ARPOP_REQUEST) ||
1378             parp->ar_hln != dev->addr_len ||
1379             parp->ar_pln != 4)
1380                 goto out;
1381         arpptr = (u8 *)parp + sizeof(struct arphdr);
1382         sha = arpptr;
1383         arpptr += dev->addr_len;        /* sha */
1384         memcpy(&sip, arpptr, sizeof(sip));
1385         arpptr += sizeof(sip);
1386         arpptr += dev->addr_len;        /* tha */
1387         memcpy(&tip, arpptr, sizeof(tip));
1388
1389         if (ipv4_is_loopback(tip) ||
1390             ipv4_is_multicast(tip))
1391                 goto out;
1392
1393         n = neigh_lookup(&arp_tbl, &tip, dev);
1394
1395         if (n) {
1396                 struct vxlan_fdb *f;
1397                 struct sk_buff  *reply;
1398
1399                 if (!(n->nud_state & NUD_CONNECTED)) {
1400                         neigh_release(n);
1401                         goto out;
1402                 }
1403
1404                 f = vxlan_find_mac(vxlan, n->ha);
1405                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1406                         /* bridge-local neighbor */
1407                         neigh_release(n);
1408                         goto out;
1409                 }
1410
1411                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1412                                 n->ha, sha);
1413
1414                 neigh_release(n);
1415
1416                 if (reply == NULL)
1417                         goto out;
1418
1419                 skb_reset_mac_header(reply);
1420                 __skb_pull(reply, skb_network_offset(reply));
1421                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1422                 reply->pkt_type = PACKET_HOST;
1423
1424                 if (netif_rx_ni(reply) == NET_RX_DROP)
1425                         dev->stats.rx_dropped++;
1426         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1427                 union vxlan_addr ipa = {
1428                         .sin.sin_addr.s_addr = tip,
1429                         .sin.sin_family = AF_INET,
1430                 };
1431
1432                 vxlan_ip_miss(dev, &ipa);
1433         }
1434 out:
1435         consume_skb(skb);
1436         return NETDEV_TX_OK;
1437 }
1438
1439 #if IS_ENABLED(CONFIG_IPV6)
1440 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1441         struct neighbour *n, bool isrouter)
1442 {
1443         struct net_device *dev = request->dev;
1444         struct sk_buff *reply;
1445         struct nd_msg *ns, *na;
1446         struct ipv6hdr *pip6;
1447         u8 *daddr;
1448         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1449         int ns_olen;
1450         int i, len;
1451
1452         if (dev == NULL)
1453                 return NULL;
1454
1455         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1456                 sizeof(*na) + na_olen + dev->needed_tailroom;
1457         reply = alloc_skb(len, GFP_ATOMIC);
1458         if (reply == NULL)
1459                 return NULL;
1460
1461         reply->protocol = htons(ETH_P_IPV6);
1462         reply->dev = dev;
1463         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1464         skb_push(reply, sizeof(struct ethhdr));
1465         skb_reset_mac_header(reply);
1466
1467         ns = (struct nd_msg *)skb_transport_header(request);
1468
1469         daddr = eth_hdr(request)->h_source;
1470         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1471         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1472                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1473                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1474                         break;
1475                 }
1476         }
1477
1478         /* Ethernet header */
1479         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1480         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1481         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1482         reply->protocol = htons(ETH_P_IPV6);
1483
1484         skb_pull(reply, sizeof(struct ethhdr));
1485         skb_reset_network_header(reply);
1486         skb_put(reply, sizeof(struct ipv6hdr));
1487
1488         /* IPv6 header */
1489
1490         pip6 = ipv6_hdr(reply);
1491         memset(pip6, 0, sizeof(struct ipv6hdr));
1492         pip6->version = 6;
1493         pip6->priority = ipv6_hdr(request)->priority;
1494         pip6->nexthdr = IPPROTO_ICMPV6;
1495         pip6->hop_limit = 255;
1496         pip6->daddr = ipv6_hdr(request)->saddr;
1497         pip6->saddr = *(struct in6_addr *)n->primary_key;
1498
1499         skb_pull(reply, sizeof(struct ipv6hdr));
1500         skb_reset_transport_header(reply);
1501
1502         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1503
1504         /* Neighbor Advertisement */
1505         memset(na, 0, sizeof(*na)+na_olen);
1506         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1507         na->icmph.icmp6_router = isrouter;
1508         na->icmph.icmp6_override = 1;
1509         na->icmph.icmp6_solicited = 1;
1510         na->target = ns->target;
1511         ether_addr_copy(&na->opt[2], n->ha);
1512         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1513         na->opt[1] = na_olen >> 3;
1514
1515         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1516                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1517                 csum_partial(na, sizeof(*na)+na_olen, 0));
1518
1519         pip6->payload_len = htons(sizeof(*na)+na_olen);
1520
1521         skb_push(reply, sizeof(struct ipv6hdr));
1522
1523         reply->ip_summed = CHECKSUM_UNNECESSARY;
1524
1525         return reply;
1526 }
1527
1528 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1529 {
1530         struct vxlan_dev *vxlan = netdev_priv(dev);
1531         struct nd_msg *msg;
1532         const struct ipv6hdr *iphdr;
1533         const struct in6_addr *saddr, *daddr;
1534         struct neighbour *n;
1535         struct inet6_dev *in6_dev;
1536
1537         in6_dev = __in6_dev_get(dev);
1538         if (!in6_dev)
1539                 goto out;
1540
1541         iphdr = ipv6_hdr(skb);
1542         saddr = &iphdr->saddr;
1543         daddr = &iphdr->daddr;
1544
1545         msg = (struct nd_msg *)skb_transport_header(skb);
1546         if (msg->icmph.icmp6_code != 0 ||
1547             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1548                 goto out;
1549
1550         if (ipv6_addr_loopback(daddr) ||
1551             ipv6_addr_is_multicast(&msg->target))
1552                 goto out;
1553
1554         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1555
1556         if (n) {
1557                 struct vxlan_fdb *f;
1558                 struct sk_buff *reply;
1559
1560                 if (!(n->nud_state & NUD_CONNECTED)) {
1561                         neigh_release(n);
1562                         goto out;
1563                 }
1564
1565                 f = vxlan_find_mac(vxlan, n->ha);
1566                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1567                         /* bridge-local neighbor */
1568                         neigh_release(n);
1569                         goto out;
1570                 }
1571
1572                 reply = vxlan_na_create(skb, n,
1573                                         !!(f ? f->flags & NTF_ROUTER : 0));
1574
1575                 neigh_release(n);
1576
1577                 if (reply == NULL)
1578                         goto out;
1579
1580                 if (netif_rx_ni(reply) == NET_RX_DROP)
1581                         dev->stats.rx_dropped++;
1582
1583         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1584                 union vxlan_addr ipa = {
1585                         .sin6.sin6_addr = msg->target,
1586                         .sin6.sin6_family = AF_INET6,
1587                 };
1588
1589                 vxlan_ip_miss(dev, &ipa);
1590         }
1591
1592 out:
1593         consume_skb(skb);
1594         return NETDEV_TX_OK;
1595 }
1596 #endif
1597
1598 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1599 {
1600         struct vxlan_dev *vxlan = netdev_priv(dev);
1601         struct neighbour *n;
1602
1603         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1604                 return false;
1605
1606         n = NULL;
1607         switch (ntohs(eth_hdr(skb)->h_proto)) {
1608         case ETH_P_IP:
1609         {
1610                 struct iphdr *pip;
1611
1612                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1613                         return false;
1614                 pip = ip_hdr(skb);
1615                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1616                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1617                         union vxlan_addr ipa = {
1618                                 .sin.sin_addr.s_addr = pip->daddr,
1619                                 .sin.sin_family = AF_INET,
1620                         };
1621
1622                         vxlan_ip_miss(dev, &ipa);
1623                         return false;
1624                 }
1625
1626                 break;
1627         }
1628 #if IS_ENABLED(CONFIG_IPV6)
1629         case ETH_P_IPV6:
1630         {
1631                 struct ipv6hdr *pip6;
1632
1633                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1634                         return false;
1635                 pip6 = ipv6_hdr(skb);
1636                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1637                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1638                         union vxlan_addr ipa = {
1639                                 .sin6.sin6_addr = pip6->daddr,
1640                                 .sin6.sin6_family = AF_INET6,
1641                         };
1642
1643                         vxlan_ip_miss(dev, &ipa);
1644                         return false;
1645                 }
1646
1647                 break;
1648         }
1649 #endif
1650         default:
1651                 return false;
1652         }
1653
1654         if (n) {
1655                 bool diff;
1656
1657                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1658                 if (diff) {
1659                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1660                                 dev->addr_len);
1661                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1662                 }
1663                 neigh_release(n);
1664                 return diff;
1665         }
1666
1667         return false;
1668 }
1669
1670 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1671                                 struct vxlan_metadata *md)
1672 {
1673         struct vxlanhdr_gbp *gbp;
1674
1675         if (!md->gbp)
1676                 return;
1677
1678         gbp = (struct vxlanhdr_gbp *)vxh;
1679         vxh->vx_flags |= VXLAN_HF_GBP;
1680
1681         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1682                 gbp->dont_learn = 1;
1683
1684         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1685                 gbp->policy_applied = 1;
1686
1687         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1688 }
1689
1690 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1691                            int iphdr_len, __be32 vni,
1692                            struct vxlan_metadata *md, u32 vxflags,
1693                            bool udp_sum)
1694 {
1695         struct vxlanhdr *vxh;
1696         int min_headroom;
1697         int err;
1698         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1699
1700         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1701             skb->ip_summed == CHECKSUM_PARTIAL) {
1702                 int csum_start = skb_checksum_start_offset(skb);
1703
1704                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1705                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1706                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1707                      skb->csum_offset == offsetof(struct tcphdr, check)))
1708                         type |= SKB_GSO_TUNNEL_REMCSUM;
1709         }
1710
1711         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1712                         + VXLAN_HLEN + iphdr_len
1713                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1714
1715         /* Need space for new headers (invalidates iph ptr) */
1716         err = skb_cow_head(skb, min_headroom);
1717         if (unlikely(err)) {
1718                 kfree_skb(skb);
1719                 return err;
1720         }
1721
1722         skb = vlan_hwaccel_push_inside(skb);
1723         if (WARN_ON(!skb))
1724                 return -ENOMEM;
1725
1726         skb = iptunnel_handle_offloads(skb, type);
1727         if (IS_ERR(skb))
1728                 return PTR_ERR(skb);
1729
1730         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1731         vxh->vx_flags = VXLAN_HF_VNI;
1732         vxh->vx_vni = vxlan_vni_field(vni);
1733
1734         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1735                 unsigned int start;
1736
1737                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1738                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1739                 vxh->vx_flags |= VXLAN_HF_RCO;
1740
1741                 if (!skb_is_gso(skb)) {
1742                         skb->ip_summed = CHECKSUM_NONE;
1743                         skb->encapsulation = 0;
1744                 }
1745         }
1746
1747         if (vxflags & VXLAN_F_GBP)
1748                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1749
1750         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1751         return 0;
1752 }
1753
1754 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1755                                       struct sk_buff *skb, int oif, u8 tos,
1756                                       __be32 daddr, __be32 *saddr,
1757                                       struct dst_cache *dst_cache,
1758                                       const struct ip_tunnel_info *info)
1759 {
1760         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1761         struct rtable *rt = NULL;
1762         struct flowi4 fl4;
1763
1764         if (tos && !info)
1765                 use_cache = false;
1766         if (use_cache) {
1767                 rt = dst_cache_get_ip4(dst_cache, saddr);
1768                 if (rt)
1769                         return rt;
1770         }
1771
1772         memset(&fl4, 0, sizeof(fl4));
1773         fl4.flowi4_oif = oif;
1774         fl4.flowi4_tos = RT_TOS(tos);
1775         fl4.flowi4_mark = skb->mark;
1776         fl4.flowi4_proto = IPPROTO_UDP;
1777         fl4.daddr = daddr;
1778         fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
1779
1780         rt = ip_route_output_key(vxlan->net, &fl4);
1781         if (!IS_ERR(rt)) {
1782                 *saddr = fl4.saddr;
1783                 if (use_cache)
1784                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1785         }
1786         return rt;
1787 }
1788
1789 #if IS_ENABLED(CONFIG_IPV6)
1790 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1791                                           struct sk_buff *skb, int oif, u8 tos,
1792                                           __be32 label,
1793                                           const struct in6_addr *daddr,
1794                                           struct in6_addr *saddr,
1795                                           struct dst_cache *dst_cache,
1796                                           const struct ip_tunnel_info *info)
1797 {
1798         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1799         struct dst_entry *ndst;
1800         struct flowi6 fl6;
1801         int err;
1802
1803         if (tos && !info)
1804                 use_cache = false;
1805         if (use_cache) {
1806                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1807                 if (ndst)
1808                         return ndst;
1809         }
1810
1811         memset(&fl6, 0, sizeof(fl6));
1812         fl6.flowi6_oif = oif;
1813         fl6.flowi6_tos = RT_TOS(tos);
1814         fl6.daddr = *daddr;
1815         fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
1816         fl6.flowlabel = label;
1817         fl6.flowi6_mark = skb->mark;
1818         fl6.flowi6_proto = IPPROTO_UDP;
1819
1820         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1821                                          vxlan->vn6_sock->sock->sk,
1822                                          &ndst, &fl6);
1823         if (err < 0)
1824                 return ERR_PTR(err);
1825
1826         *saddr = fl6.saddr;
1827         if (use_cache)
1828                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1829         return ndst;
1830 }
1831 #endif
1832
1833 /* Bypass encapsulation if the destination is local */
1834 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1835                                struct vxlan_dev *dst_vxlan)
1836 {
1837         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1838         union vxlan_addr loopback;
1839         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1840         struct net_device *dev = skb->dev;
1841         int len = skb->len;
1842
1843         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1844         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1845         skb->pkt_type = PACKET_HOST;
1846         skb->encapsulation = 0;
1847         skb->dev = dst_vxlan->dev;
1848         __skb_pull(skb, skb_network_offset(skb));
1849
1850         if (remote_ip->sa.sa_family == AF_INET) {
1851                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1852                 loopback.sa.sa_family =  AF_INET;
1853 #if IS_ENABLED(CONFIG_IPV6)
1854         } else {
1855                 loopback.sin6.sin6_addr = in6addr_loopback;
1856                 loopback.sa.sa_family =  AF_INET6;
1857 #endif
1858         }
1859
1860         if (dst_vxlan->flags & VXLAN_F_LEARN)
1861                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1862
1863         u64_stats_update_begin(&tx_stats->syncp);
1864         tx_stats->tx_packets++;
1865         tx_stats->tx_bytes += len;
1866         u64_stats_update_end(&tx_stats->syncp);
1867
1868         if (netif_rx(skb) == NET_RX_SUCCESS) {
1869                 u64_stats_update_begin(&rx_stats->syncp);
1870                 rx_stats->rx_packets++;
1871                 rx_stats->rx_bytes += len;
1872                 u64_stats_update_end(&rx_stats->syncp);
1873         } else {
1874                 dev->stats.rx_dropped++;
1875         }
1876 }
1877
1878 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1879                            struct vxlan_rdst *rdst, bool did_rsc)
1880 {
1881         struct dst_cache *dst_cache;
1882         struct ip_tunnel_info *info;
1883         struct vxlan_dev *vxlan = netdev_priv(dev);
1884         struct sock *sk;
1885         struct rtable *rt = NULL;
1886         const struct iphdr *old_iph;
1887         union vxlan_addr *dst;
1888         union vxlan_addr remote_ip;
1889         struct vxlan_metadata _md;
1890         struct vxlan_metadata *md = &_md;
1891         __be16 src_port = 0, dst_port;
1892         __be32 vni, label;
1893         __be16 df = 0;
1894         __u8 tos, ttl;
1895         int err;
1896         u32 flags = vxlan->flags;
1897         bool udp_sum = false;
1898         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1899
1900         info = skb_tunnel_info(skb);
1901
1902         if (rdst) {
1903                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1904                 vni = rdst->remote_vni;
1905                 dst = &rdst->remote_ip;
1906                 dst_cache = &rdst->dst_cache;
1907         } else {
1908                 if (!info) {
1909                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1910                                   dev->name);
1911                         goto drop;
1912                 }
1913                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1914                 vni = vxlan_tun_id_to_vni(info->key.tun_id);
1915                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1916                 if (remote_ip.sa.sa_family == AF_INET)
1917                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1918                 else
1919                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1920                 dst = &remote_ip;
1921                 dst_cache = &info->dst_cache;
1922         }
1923
1924         if (vxlan_addr_any(dst)) {
1925                 if (did_rsc) {
1926                         /* short-circuited back to local bridge */
1927                         vxlan_encap_bypass(skb, vxlan, vxlan);
1928                         return;
1929                 }
1930                 goto drop;
1931         }
1932
1933         old_iph = ip_hdr(skb);
1934
1935         ttl = vxlan->cfg.ttl;
1936         if (!ttl && vxlan_addr_multicast(dst))
1937                 ttl = 1;
1938
1939         tos = vxlan->cfg.tos;
1940         if (tos == 1)
1941                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1942
1943         label = vxlan->cfg.label;
1944         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
1945                                      vxlan->cfg.port_max, true);
1946
1947         if (info) {
1948                 ttl = info->key.ttl;
1949                 tos = info->key.tos;
1950                 label = info->key.label;
1951                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
1952
1953                 if (info->options_len)
1954                         md = ip_tunnel_info_opts(info);
1955         } else {
1956                 md->gbp = skb->mark;
1957         }
1958
1959         if (dst->sa.sa_family == AF_INET) {
1960                 __be32 saddr;
1961
1962                 if (!vxlan->vn4_sock)
1963                         goto drop;
1964                 sk = vxlan->vn4_sock->sock->sk;
1965
1966                 rt = vxlan_get_route(vxlan, skb,
1967                                      rdst ? rdst->remote_ifindex : 0, tos,
1968                                      dst->sin.sin_addr.s_addr, &saddr,
1969                                      dst_cache, info);
1970                 if (IS_ERR(rt)) {
1971                         netdev_dbg(dev, "no route to %pI4\n",
1972                                    &dst->sin.sin_addr.s_addr);
1973                         dev->stats.tx_carrier_errors++;
1974                         goto tx_error;
1975                 }
1976
1977                 if (rt->dst.dev == dev) {
1978                         netdev_dbg(dev, "circular route to %pI4\n",
1979                                    &dst->sin.sin_addr.s_addr);
1980                         dev->stats.collisions++;
1981                         goto rt_tx_error;
1982                 }
1983
1984                 /* Bypass encapsulation if the destination is local */
1985                 if (rt->rt_flags & RTCF_LOCAL &&
1986                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1987                         struct vxlan_dev *dst_vxlan;
1988
1989                         ip_rt_put(rt);
1990                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1991                                                    dst->sa.sa_family, dst_port,
1992                                                    vxlan->flags);
1993                         if (!dst_vxlan)
1994                                 goto tx_error;
1995                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1996                         return;
1997                 }
1998
1999                 if (!info)
2000                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2001                 else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2002                         df = htons(IP_DF);
2003
2004                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2005                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2006                 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2007                                       vni, md, flags, udp_sum);
2008                 if (err < 0)
2009                         goto xmit_tx_error;
2010
2011                 udp_tunnel_xmit_skb(rt, sk, skb, saddr,
2012                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2013                                     src_port, dst_port, xnet, !udp_sum);
2014 #if IS_ENABLED(CONFIG_IPV6)
2015         } else {
2016                 struct dst_entry *ndst;
2017                 struct in6_addr saddr;
2018                 u32 rt6i_flags;
2019
2020                 if (!vxlan->vn6_sock)
2021                         goto drop;
2022                 sk = vxlan->vn6_sock->sock->sk;
2023
2024                 ndst = vxlan6_get_route(vxlan, skb,
2025                                         rdst ? rdst->remote_ifindex : 0, tos,
2026                                         label, &dst->sin6.sin6_addr, &saddr,
2027                                         dst_cache, info);
2028                 if (IS_ERR(ndst)) {
2029                         netdev_dbg(dev, "no route to %pI6\n",
2030                                    &dst->sin6.sin6_addr);
2031                         dev->stats.tx_carrier_errors++;
2032                         goto tx_error;
2033                 }
2034
2035                 if (ndst->dev == dev) {
2036                         netdev_dbg(dev, "circular route to %pI6\n",
2037                                    &dst->sin6.sin6_addr);
2038                         dst_release(ndst);
2039                         dev->stats.collisions++;
2040                         goto tx_error;
2041                 }
2042
2043                 /* Bypass encapsulation if the destination is local */
2044                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2045                 if (rt6i_flags & RTF_LOCAL &&
2046                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2047                         struct vxlan_dev *dst_vxlan;
2048
2049                         dst_release(ndst);
2050                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2051                                                    dst->sa.sa_family, dst_port,
2052                                                    vxlan->flags);
2053                         if (!dst_vxlan)
2054                                 goto tx_error;
2055                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2056                         return;
2057                 }
2058
2059                 if (!info)
2060                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2061
2062                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2063                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2064                 skb_scrub_packet(skb, xnet);
2065                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2066                                       vni, md, flags, udp_sum);
2067                 if (err < 0) {
2068                         dst_release(ndst);
2069                         return;
2070                 }
2071                 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2072                                      &saddr, &dst->sin6.sin6_addr, tos, ttl,
2073                                      label, src_port, dst_port, !udp_sum);
2074 #endif
2075         }
2076
2077         return;
2078
2079 drop:
2080         dev->stats.tx_dropped++;
2081         goto tx_free;
2082
2083 xmit_tx_error:
2084         /* skb is already freed. */
2085         skb = NULL;
2086 rt_tx_error:
2087         ip_rt_put(rt);
2088 tx_error:
2089         dev->stats.tx_errors++;
2090 tx_free:
2091         dev_kfree_skb(skb);
2092 }
2093
2094 /* Transmit local packets over Vxlan
2095  *
2096  * Outer IP header inherits ECN and DF from inner header.
2097  * Outer UDP destination is the VXLAN assigned port.
2098  *           source port is based on hash of flow
2099  */
2100 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2101 {
2102         struct vxlan_dev *vxlan = netdev_priv(dev);
2103         const struct ip_tunnel_info *info;
2104         struct ethhdr *eth;
2105         bool did_rsc = false;
2106         struct vxlan_rdst *rdst, *fdst = NULL;
2107         struct vxlan_fdb *f;
2108
2109         info = skb_tunnel_info(skb);
2110
2111         skb_reset_mac_header(skb);
2112         eth = eth_hdr(skb);
2113
2114         if ((vxlan->flags & VXLAN_F_PROXY)) {
2115                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2116                         return arp_reduce(dev, skb);
2117 #if IS_ENABLED(CONFIG_IPV6)
2118                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2119                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2120                                        + sizeof(struct nd_msg)) &&
2121                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2122                                 struct nd_msg *msg;
2123
2124                                 msg = (struct nd_msg *)skb_transport_header(skb);
2125                                 if (msg->icmph.icmp6_code == 0 &&
2126                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2127                                         return neigh_reduce(dev, skb);
2128                 }
2129                 eth = eth_hdr(skb);
2130 #endif
2131         }
2132
2133         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2134                 if (info && info->mode & IP_TUNNEL_INFO_TX)
2135                         vxlan_xmit_one(skb, dev, NULL, false);
2136                 else
2137                         kfree_skb(skb);
2138                 return NETDEV_TX_OK;
2139         }
2140
2141         f = vxlan_find_mac(vxlan, eth->h_dest);
2142         did_rsc = false;
2143
2144         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2145             (ntohs(eth->h_proto) == ETH_P_IP ||
2146              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2147                 did_rsc = route_shortcircuit(dev, skb);
2148                 if (did_rsc)
2149                         f = vxlan_find_mac(vxlan, eth->h_dest);
2150         }
2151
2152         if (f == NULL) {
2153                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2154                 if (f == NULL) {
2155                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2156                             !is_multicast_ether_addr(eth->h_dest))
2157                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2158
2159                         dev->stats.tx_dropped++;
2160                         kfree_skb(skb);
2161                         return NETDEV_TX_OK;
2162                 }
2163         }
2164
2165         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2166                 struct sk_buff *skb1;
2167
2168                 if (!fdst) {
2169                         fdst = rdst;
2170                         continue;
2171                 }
2172                 skb1 = skb_clone(skb, GFP_ATOMIC);
2173                 if (skb1)
2174                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2175         }
2176
2177         if (fdst)
2178                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2179         else
2180                 kfree_skb(skb);
2181         return NETDEV_TX_OK;
2182 }
2183
2184 /* Walk the forwarding table and purge stale entries */
2185 static void vxlan_cleanup(unsigned long arg)
2186 {
2187         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2188         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2189         unsigned int h;
2190
2191         if (!netif_running(vxlan->dev))
2192                 return;
2193
2194         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2195                 struct hlist_node *p, *n;
2196
2197                 spin_lock_bh(&vxlan->hash_lock);
2198                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2199                         struct vxlan_fdb *f
2200                                 = container_of(p, struct vxlan_fdb, hlist);
2201                         unsigned long timeout;
2202
2203                         if (f->state & NUD_PERMANENT)
2204                                 continue;
2205
2206                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2207                         if (time_before_eq(timeout, jiffies)) {
2208                                 netdev_dbg(vxlan->dev,
2209                                            "garbage collect %pM\n",
2210                                            f->eth_addr);
2211                                 f->state = NUD_STALE;
2212                                 vxlan_fdb_destroy(vxlan, f);
2213                         } else if (time_before(timeout, next_timer))
2214                                 next_timer = timeout;
2215                 }
2216                 spin_unlock_bh(&vxlan->hash_lock);
2217         }
2218
2219         mod_timer(&vxlan->age_timer, next_timer);
2220 }
2221
2222 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2223 {
2224         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2225         __be32 vni = vxlan->default_dst.remote_vni;
2226
2227         spin_lock(&vn->sock_lock);
2228         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2229         spin_unlock(&vn->sock_lock);
2230 }
2231
2232 /* Setup stats when device is created */
2233 static int vxlan_init(struct net_device *dev)
2234 {
2235         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2236         if (!dev->tstats)
2237                 return -ENOMEM;
2238
2239         return 0;
2240 }
2241
2242 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2243 {
2244         struct vxlan_fdb *f;
2245
2246         spin_lock_bh(&vxlan->hash_lock);
2247         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2248         if (f)
2249                 vxlan_fdb_destroy(vxlan, f);
2250         spin_unlock_bh(&vxlan->hash_lock);
2251 }
2252
2253 static void vxlan_uninit(struct net_device *dev)
2254 {
2255         struct vxlan_dev *vxlan = netdev_priv(dev);
2256
2257         vxlan_fdb_delete_default(vxlan);
2258
2259         free_percpu(dev->tstats);
2260 }
2261
2262 /* Start ageing timer and join group when device is brought up */
2263 static int vxlan_open(struct net_device *dev)
2264 {
2265         struct vxlan_dev *vxlan = netdev_priv(dev);
2266         int ret;
2267
2268         ret = vxlan_sock_add(vxlan);
2269         if (ret < 0)
2270                 return ret;
2271
2272         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2273                 ret = vxlan_igmp_join(vxlan);
2274                 if (ret == -EADDRINUSE)
2275                         ret = 0;
2276                 if (ret) {
2277                         vxlan_sock_release(vxlan);
2278                         return ret;
2279                 }
2280         }
2281
2282         if (vxlan->cfg.age_interval)
2283                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2284
2285         return ret;
2286 }
2287
2288 /* Purge the forwarding table */
2289 static void vxlan_flush(struct vxlan_dev *vxlan)
2290 {
2291         unsigned int h;
2292
2293         spin_lock_bh(&vxlan->hash_lock);
2294         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2295                 struct hlist_node *p, *n;
2296                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2297                         struct vxlan_fdb *f
2298                                 = container_of(p, struct vxlan_fdb, hlist);
2299                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2300                         if (!is_zero_ether_addr(f->eth_addr))
2301                                 vxlan_fdb_destroy(vxlan, f);
2302                 }
2303         }
2304         spin_unlock_bh(&vxlan->hash_lock);
2305 }
2306
2307 /* Cleanup timer and forwarding table on shutdown */
2308 static int vxlan_stop(struct net_device *dev)
2309 {
2310         struct vxlan_dev *vxlan = netdev_priv(dev);
2311         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2312         int ret = 0;
2313
2314         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2315             !vxlan_group_used(vn, vxlan))
2316                 ret = vxlan_igmp_leave(vxlan);
2317
2318         del_timer_sync(&vxlan->age_timer);
2319
2320         vxlan_flush(vxlan);
2321         vxlan_sock_release(vxlan);
2322
2323         return ret;
2324 }
2325
2326 /* Stub, nothing needs to be done. */
2327 static void vxlan_set_multicast_list(struct net_device *dev)
2328 {
2329 }
2330
2331 static int __vxlan_change_mtu(struct net_device *dev,
2332                               struct net_device *lowerdev,
2333                               struct vxlan_rdst *dst, int new_mtu, bool strict)
2334 {
2335         int max_mtu = IP_MAX_MTU;
2336
2337         if (lowerdev)
2338                 max_mtu = lowerdev->mtu;
2339
2340         if (dst->remote_ip.sa.sa_family == AF_INET6)
2341                 max_mtu -= VXLAN6_HEADROOM;
2342         else
2343                 max_mtu -= VXLAN_HEADROOM;
2344
2345         if (new_mtu < 68)
2346                 return -EINVAL;
2347
2348         if (new_mtu > max_mtu) {
2349                 if (strict)
2350                         return -EINVAL;
2351
2352                 new_mtu = max_mtu;
2353         }
2354
2355         dev->mtu = new_mtu;
2356         return 0;
2357 }
2358
2359 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2360 {
2361         struct vxlan_dev *vxlan = netdev_priv(dev);
2362         struct vxlan_rdst *dst = &vxlan->default_dst;
2363         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2364                                                          dst->remote_ifindex);
2365         return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2366 }
2367
2368 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2369 {
2370         struct vxlan_dev *vxlan = netdev_priv(dev);
2371         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2372         __be16 sport, dport;
2373
2374         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2375                                   vxlan->cfg.port_max, true);
2376         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2377
2378         if (ip_tunnel_info_af(info) == AF_INET) {
2379                 struct rtable *rt;
2380
2381                 if (!vxlan->vn4_sock)
2382                         return -EINVAL;
2383                 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2384                                      info->key.u.ipv4.dst,
2385                                      &info->key.u.ipv4.src, NULL, info);
2386                 if (IS_ERR(rt))
2387                         return PTR_ERR(rt);
2388                 ip_rt_put(rt);
2389         } else {
2390 #if IS_ENABLED(CONFIG_IPV6)
2391                 struct dst_entry *ndst;
2392
2393                 if (!vxlan->vn6_sock)
2394                         return -EINVAL;
2395                 ndst = vxlan6_get_route(vxlan, skb, 0, info->key.tos,
2396                                         info->key.label, &info->key.u.ipv6.dst,
2397                                         &info->key.u.ipv6.src, NULL, info);
2398                 if (IS_ERR(ndst))
2399                         return PTR_ERR(ndst);
2400                 dst_release(ndst);
2401 #else /* !CONFIG_IPV6 */
2402                 return -EPFNOSUPPORT;
2403 #endif
2404         }
2405         info->key.tp_src = sport;
2406         info->key.tp_dst = dport;
2407         return 0;
2408 }
2409
2410 static const struct net_device_ops vxlan_netdev_ops = {
2411         .ndo_init               = vxlan_init,
2412         .ndo_uninit             = vxlan_uninit,
2413         .ndo_open               = vxlan_open,
2414         .ndo_stop               = vxlan_stop,
2415         .ndo_start_xmit         = vxlan_xmit,
2416         .ndo_get_stats64        = ip_tunnel_get_stats64,
2417         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2418         .ndo_change_mtu         = vxlan_change_mtu,
2419         .ndo_validate_addr      = eth_validate_addr,
2420         .ndo_set_mac_address    = eth_mac_addr,
2421         .ndo_fdb_add            = vxlan_fdb_add,
2422         .ndo_fdb_del            = vxlan_fdb_delete,
2423         .ndo_fdb_dump           = vxlan_fdb_dump,
2424         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2425 };
2426
2427 /* Info for udev, that this is a virtual tunnel endpoint */
2428 static struct device_type vxlan_type = {
2429         .name = "vxlan",
2430 };
2431
2432 /* Calls the ndo_add_vxlan_port of the caller in order to
2433  * supply the listening VXLAN udp ports. Callers are expected
2434  * to implement the ndo_add_vxlan_port.
2435  */
2436 void vxlan_get_rx_port(struct net_device *dev)
2437 {
2438         struct vxlan_sock *vs;
2439         struct net *net = dev_net(dev);
2440         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2441         sa_family_t sa_family;
2442         __be16 port;
2443         unsigned int i;
2444
2445         spin_lock(&vn->sock_lock);
2446         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2447                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2448                         port = inet_sk(vs->sock->sk)->inet_sport;
2449                         sa_family = vxlan_get_sk_family(vs);
2450                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2451                                                             port);
2452                 }
2453         }
2454         spin_unlock(&vn->sock_lock);
2455 }
2456 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2457
2458 /* Initialize the device structure. */
2459 static void vxlan_setup(struct net_device *dev)
2460 {
2461         struct vxlan_dev *vxlan = netdev_priv(dev);
2462         unsigned int h;
2463
2464         eth_hw_addr_random(dev);
2465         ether_setup(dev);
2466
2467         dev->netdev_ops = &vxlan_netdev_ops;
2468         dev->destructor = free_netdev;
2469         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2470
2471         dev->features   |= NETIF_F_LLTX;
2472         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2473         dev->features   |= NETIF_F_RXCSUM;
2474         dev->features   |= NETIF_F_GSO_SOFTWARE;
2475
2476         dev->vlan_features = dev->features;
2477         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2478         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2479         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2480         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2481         netif_keep_dst(dev);
2482         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2483         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
2484
2485         INIT_LIST_HEAD(&vxlan->next);
2486         spin_lock_init(&vxlan->hash_lock);
2487
2488         init_timer_deferrable(&vxlan->age_timer);
2489         vxlan->age_timer.function = vxlan_cleanup;
2490         vxlan->age_timer.data = (unsigned long) vxlan;
2491
2492         vxlan->cfg.dst_port = htons(vxlan_port);
2493
2494         vxlan->dev = dev;
2495
2496         gro_cells_init(&vxlan->gro_cells, dev);
2497
2498         for (h = 0; h < FDB_HASH_SIZE; ++h)
2499                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2500 }
2501
2502 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2503         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2504         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2505         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2506         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2507         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2508         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2509         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2510         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2511         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2512         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2513         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2514         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2515         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2516         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2517         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2518         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2519         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2520         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2521         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2522         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2523         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2524         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2525         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2526         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2527         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2528         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2529 };
2530
2531 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2532 {
2533         if (tb[IFLA_ADDRESS]) {
2534                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2535                         pr_debug("invalid link address (not ethernet)\n");
2536                         return -EINVAL;
2537                 }
2538
2539                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2540                         pr_debug("invalid all zero ethernet address\n");
2541                         return -EADDRNOTAVAIL;
2542                 }
2543         }
2544
2545         if (!data)
2546                 return -EINVAL;
2547
2548         if (data[IFLA_VXLAN_ID]) {
2549                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2550                 if (id >= VXLAN_VID_MASK)
2551                         return -ERANGE;
2552         }
2553
2554         if (data[IFLA_VXLAN_PORT_RANGE]) {
2555                 const struct ifla_vxlan_port_range *p
2556                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2557
2558                 if (ntohs(p->high) < ntohs(p->low)) {
2559                         pr_debug("port range %u .. %u not valid\n",
2560                                  ntohs(p->low), ntohs(p->high));
2561                         return -EINVAL;
2562                 }
2563         }
2564
2565         return 0;
2566 }
2567
2568 static void vxlan_get_drvinfo(struct net_device *netdev,
2569                               struct ethtool_drvinfo *drvinfo)
2570 {
2571         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2572         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2573 }
2574
2575 static const struct ethtool_ops vxlan_ethtool_ops = {
2576         .get_drvinfo    = vxlan_get_drvinfo,
2577         .get_link       = ethtool_op_get_link,
2578 };
2579
2580 static void vxlan_del_work(struct work_struct *work)
2581 {
2582         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2583         udp_tunnel_sock_release(vs->sock);
2584         kfree_rcu(vs, rcu);
2585 }
2586
2587 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2588                                         __be16 port, u32 flags)
2589 {
2590         struct socket *sock;
2591         struct udp_port_cfg udp_conf;
2592         int err;
2593
2594         memset(&udp_conf, 0, sizeof(udp_conf));
2595
2596         if (ipv6) {
2597                 udp_conf.family = AF_INET6;
2598                 udp_conf.use_udp6_rx_checksums =
2599                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2600                 udp_conf.ipv6_v6only = 1;
2601         } else {
2602                 udp_conf.family = AF_INET;
2603         }
2604
2605         udp_conf.local_udp_port = port;
2606
2607         /* Open UDP socket */
2608         err = udp_sock_create(net, &udp_conf, &sock);
2609         if (err < 0)
2610                 return ERR_PTR(err);
2611
2612         return sock;
2613 }
2614
2615 /* Create new listen socket if needed */
2616 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2617                                               __be16 port, u32 flags)
2618 {
2619         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2620         struct vxlan_sock *vs;
2621         struct socket *sock;
2622         unsigned int h;
2623         struct udp_tunnel_sock_cfg tunnel_cfg;
2624
2625         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2626         if (!vs)
2627                 return ERR_PTR(-ENOMEM);
2628
2629         for (h = 0; h < VNI_HASH_SIZE; ++h)
2630                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2631
2632         INIT_WORK(&vs->del_work, vxlan_del_work);
2633
2634         sock = vxlan_create_sock(net, ipv6, port, flags);
2635         if (IS_ERR(sock)) {
2636                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2637                         PTR_ERR(sock));
2638                 kfree(vs);
2639                 return ERR_CAST(sock);
2640         }
2641
2642         vs->sock = sock;
2643         atomic_set(&vs->refcnt, 1);
2644         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2645
2646         /* Initialize the vxlan udp offloads structure */
2647         vs->udp_offloads.port = port;
2648         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2649         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2650
2651         spin_lock(&vn->sock_lock);
2652         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2653         vxlan_notify_add_rx_port(vs);
2654         spin_unlock(&vn->sock_lock);
2655
2656         /* Mark socket as an encapsulation socket. */
2657         tunnel_cfg.sk_user_data = vs;
2658         tunnel_cfg.encap_type = 1;
2659         tunnel_cfg.encap_rcv = vxlan_rcv;
2660         tunnel_cfg.encap_destroy = NULL;
2661
2662         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2663
2664         return vs;
2665 }
2666
2667 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2668 {
2669         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2670         struct vxlan_sock *vs = NULL;
2671
2672         if (!vxlan->cfg.no_share) {
2673                 spin_lock(&vn->sock_lock);
2674                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2675                                      vxlan->cfg.dst_port, vxlan->flags);
2676                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2677                         spin_unlock(&vn->sock_lock);
2678                         return -EBUSY;
2679                 }
2680                 spin_unlock(&vn->sock_lock);
2681         }
2682         if (!vs)
2683                 vs = vxlan_socket_create(vxlan->net, ipv6,
2684                                          vxlan->cfg.dst_port, vxlan->flags);
2685         if (IS_ERR(vs))
2686                 return PTR_ERR(vs);
2687 #if IS_ENABLED(CONFIG_IPV6)
2688         if (ipv6)
2689                 vxlan->vn6_sock = vs;
2690         else
2691 #endif
2692                 vxlan->vn4_sock = vs;
2693         vxlan_vs_add_dev(vs, vxlan);
2694         return 0;
2695 }
2696
2697 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2698 {
2699         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2700         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2701         int ret = 0;
2702
2703         vxlan->vn4_sock = NULL;
2704 #if IS_ENABLED(CONFIG_IPV6)
2705         vxlan->vn6_sock = NULL;
2706         if (ipv6 || metadata)
2707                 ret = __vxlan_sock_add(vxlan, true);
2708 #endif
2709         if (!ret && (!ipv6 || metadata))
2710                 ret = __vxlan_sock_add(vxlan, false);
2711         if (ret < 0)
2712                 vxlan_sock_release(vxlan);
2713         return ret;
2714 }
2715
2716 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2717                                struct vxlan_config *conf)
2718 {
2719         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2720         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2721         struct vxlan_rdst *dst = &vxlan->default_dst;
2722         unsigned short needed_headroom = ETH_HLEN;
2723         int err;
2724         bool use_ipv6 = false;
2725         __be16 default_port = vxlan->cfg.dst_port;
2726         struct net_device *lowerdev = NULL;
2727
2728         vxlan->net = src_net;
2729
2730         dst->remote_vni = conf->vni;
2731
2732         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2733
2734         /* Unless IPv6 is explicitly requested, assume IPv4 */
2735         if (!dst->remote_ip.sa.sa_family)
2736                 dst->remote_ip.sa.sa_family = AF_INET;
2737
2738         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2739             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2740                 if (!IS_ENABLED(CONFIG_IPV6))
2741                         return -EPFNOSUPPORT;
2742                 use_ipv6 = true;
2743                 vxlan->flags |= VXLAN_F_IPV6;
2744         }
2745
2746         if (conf->label && !use_ipv6) {
2747                 pr_info("label only supported in use with IPv6\n");
2748                 return -EINVAL;
2749         }
2750
2751         if (conf->remote_ifindex) {
2752                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2753                 dst->remote_ifindex = conf->remote_ifindex;
2754
2755                 if (!lowerdev) {
2756                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2757                         return -ENODEV;
2758                 }
2759
2760 #if IS_ENABLED(CONFIG_IPV6)
2761                 if (use_ipv6) {
2762                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2763                         if (idev && idev->cnf.disable_ipv6) {
2764                                 pr_info("IPv6 is disabled via sysctl\n");
2765                                 return -EPERM;
2766                         }
2767                 }
2768 #endif
2769
2770                 if (!conf->mtu)
2771                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2772
2773                 needed_headroom = lowerdev->hard_header_len;
2774         }
2775
2776         if (conf->mtu) {
2777                 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2778                 if (err)
2779                         return err;
2780         }
2781
2782         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2783                 needed_headroom += VXLAN6_HEADROOM;
2784         else
2785                 needed_headroom += VXLAN_HEADROOM;
2786         dev->needed_headroom = needed_headroom;
2787
2788         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2789         if (!vxlan->cfg.dst_port)
2790                 vxlan->cfg.dst_port = default_port;
2791         vxlan->flags |= conf->flags;
2792
2793         if (!vxlan->cfg.age_interval)
2794                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2795
2796         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2797                 if (tmp->cfg.vni == conf->vni &&
2798                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2799                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2800                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2801                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2802                     (vxlan->flags & VXLAN_F_RCV_FLAGS))
2803                 return -EEXIST;
2804         }
2805
2806         dev->ethtool_ops = &vxlan_ethtool_ops;
2807
2808         /* create an fdb entry for a valid default destination */
2809         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2810                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2811                                        &vxlan->default_dst.remote_ip,
2812                                        NUD_REACHABLE|NUD_PERMANENT,
2813                                        NLM_F_EXCL|NLM_F_CREATE,
2814                                        vxlan->cfg.dst_port,
2815                                        vxlan->default_dst.remote_vni,
2816                                        vxlan->default_dst.remote_ifindex,
2817                                        NTF_SELF);
2818                 if (err)
2819                         return err;
2820         }
2821
2822         err = register_netdevice(dev);
2823         if (err) {
2824                 vxlan_fdb_delete_default(vxlan);
2825                 return err;
2826         }
2827
2828         list_add(&vxlan->next, &vn->vxlan_list);
2829
2830         return 0;
2831 }
2832
2833 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2834                                     u8 name_assign_type, struct vxlan_config *conf)
2835 {
2836         struct nlattr *tb[IFLA_MAX+1];
2837         struct net_device *dev;
2838         int err;
2839
2840         memset(&tb, 0, sizeof(tb));
2841
2842         dev = rtnl_create_link(net, name, name_assign_type,
2843                                &vxlan_link_ops, tb);
2844         if (IS_ERR(dev))
2845                 return dev;
2846
2847         err = vxlan_dev_configure(net, dev, conf);
2848         if (err < 0) {
2849                 free_netdev(dev);
2850                 return ERR_PTR(err);
2851         }
2852
2853         return dev;
2854 }
2855 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2856
2857 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2858                          struct nlattr *tb[], struct nlattr *data[])
2859 {
2860         struct vxlan_config conf;
2861         int err;
2862
2863         memset(&conf, 0, sizeof(conf));
2864
2865         if (data[IFLA_VXLAN_ID])
2866                 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2867
2868         if (data[IFLA_VXLAN_GROUP]) {
2869                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2870         } else if (data[IFLA_VXLAN_GROUP6]) {
2871                 if (!IS_ENABLED(CONFIG_IPV6))
2872                         return -EPFNOSUPPORT;
2873
2874                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2875                 conf.remote_ip.sa.sa_family = AF_INET6;
2876         }
2877
2878         if (data[IFLA_VXLAN_LOCAL]) {
2879                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2880                 conf.saddr.sa.sa_family = AF_INET;
2881         } else if (data[IFLA_VXLAN_LOCAL6]) {
2882                 if (!IS_ENABLED(CONFIG_IPV6))
2883                         return -EPFNOSUPPORT;
2884
2885                 /* TODO: respect scope id */
2886                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2887                 conf.saddr.sa.sa_family = AF_INET6;
2888         }
2889
2890         if (data[IFLA_VXLAN_LINK])
2891                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2892
2893         if (data[IFLA_VXLAN_TOS])
2894                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2895
2896         if (data[IFLA_VXLAN_TTL])
2897                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2898
2899         if (data[IFLA_VXLAN_LABEL])
2900                 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
2901                              IPV6_FLOWLABEL_MASK;
2902
2903         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2904                 conf.flags |= VXLAN_F_LEARN;
2905
2906         if (data[IFLA_VXLAN_AGEING])
2907                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2908
2909         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2910                 conf.flags |= VXLAN_F_PROXY;
2911
2912         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2913                 conf.flags |= VXLAN_F_RSC;
2914
2915         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2916                 conf.flags |= VXLAN_F_L2MISS;
2917
2918         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2919                 conf.flags |= VXLAN_F_L3MISS;
2920
2921         if (data[IFLA_VXLAN_LIMIT])
2922                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2923
2924         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2925             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2926                 conf.flags |= VXLAN_F_COLLECT_METADATA;
2927
2928         if (data[IFLA_VXLAN_PORT_RANGE]) {
2929                 const struct ifla_vxlan_port_range *p
2930                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2931                 conf.port_min = ntohs(p->low);
2932                 conf.port_max = ntohs(p->high);
2933         }
2934
2935         if (data[IFLA_VXLAN_PORT])
2936                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2937
2938         if (data[IFLA_VXLAN_UDP_CSUM] &&
2939             !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2940                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
2941
2942         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2943             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2944                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2945
2946         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2947             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2948                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2949
2950         if (data[IFLA_VXLAN_REMCSUM_TX] &&
2951             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2952                 conf.flags |= VXLAN_F_REMCSUM_TX;
2953
2954         if (data[IFLA_VXLAN_REMCSUM_RX] &&
2955             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2956                 conf.flags |= VXLAN_F_REMCSUM_RX;
2957
2958         if (data[IFLA_VXLAN_GBP])
2959                 conf.flags |= VXLAN_F_GBP;
2960
2961         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2962                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2963
2964         err = vxlan_dev_configure(src_net, dev, &conf);
2965         switch (err) {
2966         case -ENODEV:
2967                 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
2968                 break;
2969
2970         case -EPERM:
2971                 pr_info("IPv6 is disabled via sysctl\n");
2972                 break;
2973
2974         case -EEXIST:
2975                 pr_info("duplicate VNI %u\n", be32_to_cpu(conf.vni));
2976                 break;
2977         }
2978
2979         return err;
2980 }
2981
2982 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2983 {
2984         struct vxlan_dev *vxlan = netdev_priv(dev);
2985         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2986
2987         spin_lock(&vn->sock_lock);
2988         if (!hlist_unhashed(&vxlan->hlist))
2989                 hlist_del_rcu(&vxlan->hlist);
2990         spin_unlock(&vn->sock_lock);
2991
2992         gro_cells_destroy(&vxlan->gro_cells);
2993         list_del(&vxlan->next);
2994         unregister_netdevice_queue(dev, head);
2995 }
2996
2997 static size_t vxlan_get_size(const struct net_device *dev)
2998 {
2999
3000         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3001                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3002                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3003                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3004                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3005                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3006                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3007                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3008                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3009                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3010                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3011                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3012                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3013                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3014                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3015                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3016                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3017                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3018                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3019                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3020                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3021                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3022                 0;
3023 }
3024
3025 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3026 {
3027         const struct vxlan_dev *vxlan = netdev_priv(dev);
3028         const struct vxlan_rdst *dst = &vxlan->default_dst;
3029         struct ifla_vxlan_port_range ports = {
3030                 .low =  htons(vxlan->cfg.port_min),
3031                 .high = htons(vxlan->cfg.port_max),
3032         };
3033
3034         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3035                 goto nla_put_failure;
3036
3037         if (!vxlan_addr_any(&dst->remote_ip)) {
3038                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3039                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3040                                             dst->remote_ip.sin.sin_addr.s_addr))
3041                                 goto nla_put_failure;
3042 #if IS_ENABLED(CONFIG_IPV6)
3043                 } else {
3044                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3045                                              &dst->remote_ip.sin6.sin6_addr))
3046                                 goto nla_put_failure;
3047 #endif
3048                 }
3049         }
3050
3051         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3052                 goto nla_put_failure;
3053
3054         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3055                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3056                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3057                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3058                                 goto nla_put_failure;
3059 #if IS_ENABLED(CONFIG_IPV6)
3060                 } else {
3061                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3062                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3063                                 goto nla_put_failure;
3064 #endif
3065                 }
3066         }
3067
3068         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3069             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3070             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3071             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3072                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3073             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3074                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3075             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3076             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3077                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3078             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3079                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3080             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3081                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3082             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3083             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3084             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3085             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3086                         !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3087             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3088                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3089             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3090                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3091             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3092                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3093             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3094                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3095                 goto nla_put_failure;
3096
3097         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3098                 goto nla_put_failure;
3099
3100         if (vxlan->flags & VXLAN_F_GBP &&
3101             nla_put_flag(skb, IFLA_VXLAN_GBP))
3102                 goto nla_put_failure;
3103
3104         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3105             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3106                 goto nla_put_failure;
3107
3108         return 0;
3109
3110 nla_put_failure:
3111         return -EMSGSIZE;
3112 }
3113
3114 static struct net *vxlan_get_link_net(const struct net_device *dev)
3115 {
3116         struct vxlan_dev *vxlan = netdev_priv(dev);
3117
3118         return vxlan->net;
3119 }
3120
3121 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3122         .kind           = "vxlan",
3123         .maxtype        = IFLA_VXLAN_MAX,
3124         .policy         = vxlan_policy,
3125         .priv_size      = sizeof(struct vxlan_dev),
3126         .setup          = vxlan_setup,
3127         .validate       = vxlan_validate,
3128         .newlink        = vxlan_newlink,
3129         .dellink        = vxlan_dellink,
3130         .get_size       = vxlan_get_size,
3131         .fill_info      = vxlan_fill_info,
3132         .get_link_net   = vxlan_get_link_net,
3133 };
3134
3135 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3136                                              struct net_device *dev)
3137 {
3138         struct vxlan_dev *vxlan, *next;
3139         LIST_HEAD(list_kill);
3140
3141         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3142                 struct vxlan_rdst *dst = &vxlan->default_dst;
3143
3144                 /* In case we created vxlan device with carrier
3145                  * and we loose the carrier due to module unload
3146                  * we also need to remove vxlan device. In other
3147                  * cases, it's not necessary and remote_ifindex
3148                  * is 0 here, so no matches.
3149                  */
3150                 if (dst->remote_ifindex == dev->ifindex)
3151                         vxlan_dellink(vxlan->dev, &list_kill);
3152         }
3153
3154         unregister_netdevice_many(&list_kill);
3155 }
3156
3157 static int vxlan_lowerdev_event(struct notifier_block *unused,
3158                                 unsigned long event, void *ptr)
3159 {
3160         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3161         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3162
3163         if (event == NETDEV_UNREGISTER)
3164                 vxlan_handle_lowerdev_unregister(vn, dev);
3165
3166         return NOTIFY_DONE;
3167 }
3168
3169 static struct notifier_block vxlan_notifier_block __read_mostly = {
3170         .notifier_call = vxlan_lowerdev_event,
3171 };
3172
3173 static __net_init int vxlan_init_net(struct net *net)
3174 {
3175         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3176         unsigned int h;
3177
3178         INIT_LIST_HEAD(&vn->vxlan_list);
3179         spin_lock_init(&vn->sock_lock);
3180
3181         for (h = 0; h < PORT_HASH_SIZE; ++h)
3182                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3183
3184         return 0;
3185 }
3186
3187 static void __net_exit vxlan_exit_net(struct net *net)
3188 {
3189         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3190         struct vxlan_dev *vxlan, *next;
3191         struct net_device *dev, *aux;
3192         LIST_HEAD(list);
3193
3194         rtnl_lock();
3195         for_each_netdev_safe(net, dev, aux)
3196                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3197                         unregister_netdevice_queue(dev, &list);
3198
3199         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3200                 /* If vxlan->dev is in the same netns, it has already been added
3201                  * to the list by the previous loop.
3202                  */
3203                 if (!net_eq(dev_net(vxlan->dev), net)) {
3204                         gro_cells_destroy(&vxlan->gro_cells);
3205                         unregister_netdevice_queue(vxlan->dev, &list);
3206                 }
3207         }
3208
3209         unregister_netdevice_many(&list);
3210         rtnl_unlock();
3211 }
3212
3213 static struct pernet_operations vxlan_net_ops = {
3214         .init = vxlan_init_net,
3215         .exit = vxlan_exit_net,
3216         .id   = &vxlan_net_id,
3217         .size = sizeof(struct vxlan_net),
3218 };
3219
3220 static int __init vxlan_init_module(void)
3221 {
3222         int rc;
3223
3224         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3225         if (!vxlan_wq)
3226                 return -ENOMEM;
3227
3228         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3229
3230         rc = register_pernet_subsys(&vxlan_net_ops);
3231         if (rc)
3232                 goto out1;
3233
3234         rc = register_netdevice_notifier(&vxlan_notifier_block);
3235         if (rc)
3236                 goto out2;
3237
3238         rc = rtnl_link_register(&vxlan_link_ops);
3239         if (rc)
3240                 goto out3;
3241
3242         return 0;
3243 out3:
3244         unregister_netdevice_notifier(&vxlan_notifier_block);
3245 out2:
3246         unregister_pernet_subsys(&vxlan_net_ops);
3247 out1:
3248         destroy_workqueue(vxlan_wq);
3249         return rc;
3250 }
3251 late_initcall(vxlan_init_module);
3252
3253 static void __exit vxlan_cleanup_module(void)
3254 {
3255         rtnl_link_unregister(&vxlan_link_ops);
3256         unregister_netdevice_notifier(&vxlan_notifier_block);
3257         destroy_workqueue(vxlan_wq);
3258         unregister_pernet_subsys(&vxlan_net_ops);
3259         /* rcu_barrier() is called by netns */
3260 }
3261 module_exit(vxlan_cleanup_module);
3262
3263 MODULE_LICENSE("GPL");
3264 MODULE_VERSION(VXLAN_VERSION);
3265 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3266 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3267 MODULE_ALIAS_RTNL_LINK("vxlan");