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 #if IS_ENABLED(CONFIG_IPV6)
  46 #include <net/ipv6.h>
  47 #include <net/addrconf.h>
  48 #include <net/ip6_tunnel.h>
  49 #include <net/ip6_checksum.h>
  50 #endif
  51
  52 #define VXLAN_VERSION   "0.1"
  53
  54 #define PORT_HASH_BITS  8
  55 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  56 #define VNI_HASH_BITS   10
  57 #define VNI_HASH_SIZE   (1<<VNI_HASH_BITS)
  58 #define FDB_HASH_BITS   8
  59 #define FDB_HASH_SIZE   (1<<FDB_HASH_BITS)
  60 #define FDB_AGE_DEFAULT 300 /* 5 min */
  61 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  62
  63 #define VXLAN_N_VID     (1u << 24)
  64 #define VXLAN_VID_MASK  (VXLAN_N_VID - 1)
  65 #define VXLAN_HLEN (sizeof(struct udphdr) + sizeof(struct vxlanhdr))
  66
  67 #define VXLAN_FLAGS 0x08000000  /* struct vxlanhdr.vx_flags required value. */
  68
  69 /* VXLAN protocol header */
  70 struct vxlanhdr {
  71         __be32 vx_flags;
  72         __be32 vx_vni;
  73 };
  74
  75 /* UDP port for VXLAN traffic.
  76  * The IANA assigned port is 4789, but the Linux default is 8472
  77  * for compatibility with early adopters.
  78  */
  79 static unsigned short vxlan_port __read_mostly = 8472;
  80 module_param_named(udp_port, vxlan_port, ushort, 0444);
  81 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  82
  83 static bool log_ecn_error = true;
  84 module_param(log_ecn_error, bool, 0644);
  85 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  86
  87 static int vxlan_net_id;
  88
  89 static const u8 all_zeros_mac[ETH_ALEN];
  90
  91 /* per-network namespace private data for this module */
  92 struct vxlan_net {
  93         struct list_head  vxlan_list;
  94         struct hlist_head sock_list[PORT_HASH_SIZE];
  95         spinlock_t        sock_lock;
  96 };
  97
  98 union vxlan_addr {
  99         struct sockaddr_in sin;
 100         struct sockaddr_in6 sin6;
 101         struct sockaddr sa;
 102 };
 103
 104 struct vxlan_rdst {
 105         union vxlan_addr         remote_ip;
 106         __be16                   remote_port;
 107         u32                      remote_vni;
 108         u32                      remote_ifindex;
 109         struct list_head         list;
 110         struct rcu_head          rcu;
 111 };
 112
 113 /* Forwarding table entry */
 114 struct vxlan_fdb {
 115         struct hlist_node hlist;        /* linked list of entries */
 116         struct rcu_head   rcu;
 117         unsigned long     updated;      /* jiffies */
 118         unsigned long     used;
 119         struct list_head  remotes;
 120         u16               state;        /* see ndm_state */
 121         u8                flags;        /* see ndm_flags */
 122         u8                eth_addr[ETH_ALEN];
 123 };
 124
 125 /* Pseudo network device */
 126 struct vxlan_dev {
 127         struct hlist_node hlist;        /* vni hash table */
 128         struct list_head  next;         /* vxlan's per namespace list */
 129         struct vxlan_sock *vn_sock;     /* listening socket */
 130         struct net_device *dev;
 131         struct net        *net;         /* netns for packet i/o */
 132         struct vxlan_rdst default_dst;  /* default destination */
 133         union vxlan_addr  saddr;        /* source address */
 134         __be16            dst_port;
 135         __u16             port_min;     /* source port range */
 136         __u16             port_max;
 137         __u8              tos;          /* TOS override */
 138         __u8              ttl;
 139         u32               flags;        /* VXLAN_F_* in vxlan.h */
 140
 141         struct work_struct sock_work;
 142         struct work_struct igmp_join;
 143         struct work_struct igmp_leave;
 144
 145         unsigned long     age_interval;
 146         struct timer_list age_timer;
 147         spinlock_t        hash_lock;
 148         unsigned int      addrcnt;
 149         unsigned int      addrmax;
 150
 151         struct hlist_head fdb_head[FDB_HASH_SIZE];
 152 };
 153
 154 /* salt for hash table */
 155 static u32 vxlan_salt __read_mostly;
 156 static struct workqueue_struct *vxlan_wq;
 157
 158 static void vxlan_sock_work(struct work_struct *work);
 159
 160 #if IS_ENABLED(CONFIG_IPV6)
 161 static inline
 162 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 163 {
 164        if (a->sa.sa_family != b->sa.sa_family)
 165                return false;
 166        if (a->sa.sa_family == AF_INET6)
 167                return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 168        else
 169                return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 170 }
 171
 172 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 173 {
 174        if (ipa->sa.sa_family == AF_INET6)
 175                return ipv6_addr_any(&ipa->sin6.sin6_addr);
 176        else
 177                return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 178 }
 179
 180 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 181 {
 182        if (ipa->sa.sa_family == AF_INET6)
 183                return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 184        else
 185                return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 186 }
 187
 188 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 189 {
 190        if (nla_len(nla) >= sizeof(struct in6_addr)) {
 191                nla_memcpy(&ip->sin6.sin6_addr, nla, sizeof(struct in6_addr));
 192                ip->sa.sa_family = AF_INET6;
 193                return 0;
 194        } else if (nla_len(nla) >= sizeof(__be32)) {
 195                ip->sin.sin_addr.s_addr = nla_get_be32(nla);
 196                ip->sa.sa_family = AF_INET;
 197                return 0;
 198        } else {
 199                return -EAFNOSUPPORT;
 200        }
 201 }
 202
 203 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 204                              const union vxlan_addr *ip)
 205 {
 206        if (ip->sa.sa_family == AF_INET6)
 207                return nla_put(skb, attr, sizeof(struct in6_addr), &ip->sin6.sin6_addr);
 208        else
 209                return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
 210 }
 211
 212 #else /* !CONFIG_IPV6 */
 213
 214 static inline
 215 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 216 {
 217        return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 218 }
 219
 220 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 221 {
 222        return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 223 }
 224
 225 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 226 {
 227        return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 228 }
 229
 230 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 231 {
 232        if (nla_len(nla) >= sizeof(struct in6_addr)) {
 233                return -EAFNOSUPPORT;
 234        } else if (nla_len(nla) >= sizeof(__be32)) {
 235                ip->sin.sin_addr.s_addr = nla_get_be32(nla);
 236                ip->sa.sa_family = AF_INET;
 237                return 0;
 238        } else {
 239                return -EAFNOSUPPORT;
 240        }
 241 }
 242
 243 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 244                              const union vxlan_addr *ip)
 245 {
 246        return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
 247 }
 248 #endif
 249
 250 /* Virtual Network hash table head */
 251 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
 252 {
 253         return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
 254 }
 255
 256 /* Socket hash table head */
 257 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 258 {
 259         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 260
 261         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 262 }
 263
 264 /* First remote destination for a forwarding entry.
 265  * Guaranteed to be non-NULL because remotes are never deleted.
 266  */
 267 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 268 {
 269         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 270 }
 271
 272 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 273 {
 274         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 275 }
 276
 277 /* Find VXLAN socket based on network namespace and UDP port */
 278 static struct vxlan_sock *vxlan_find_sock(struct net *net, __be16 port)
 279 {
 280         struct vxlan_sock *vs;
 281
 282         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 283                 if (inet_sk(vs->sock->sk)->inet_sport == port)
 284                         return vs;
 285         }
 286         return NULL;
 287 }
 288
 289 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
 290 {
 291         struct vxlan_dev *vxlan;
 292
 293         hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
 294                 if (vxlan->default_dst.remote_vni == id)
 295                         return vxlan;
 296         }
 297
 298         return NULL;
 299 }
 300
 301 /* Look up VNI in a per net namespace table */
 302 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id, __be16 port)
 303 {
 304         struct vxlan_sock *vs;
 305
 306         vs = vxlan_find_sock(net, port);
 307         if (!vs)
 308                 return NULL;
 309
 310         return vxlan_vs_find_vni(vs, id);
 311 }
 312
 313 /* Fill in neighbour message in skbuff. */
 314 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 315                           const struct vxlan_fdb *fdb,
 316                           u32 portid, u32 seq, int type, unsigned int flags,
 317                           const struct vxlan_rdst *rdst)
 318 {
 319         unsigned long now = jiffies;
 320         struct nda_cacheinfo ci;
 321         struct nlmsghdr *nlh;
 322         struct ndmsg *ndm;
 323         bool send_ip, send_eth;
 324
 325         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 326         if (nlh == NULL)
 327                 return -EMSGSIZE;
 328
 329         ndm = nlmsg_data(nlh);
 330         memset(ndm, 0, sizeof(*ndm));
 331
 332         send_eth = send_ip = true;
 333
 334         if (type == RTM_GETNEIGH) {
 335                 ndm->ndm_family = AF_INET;
 336                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 337                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 338         } else
 339                 ndm->ndm_family = AF_BRIDGE;
 340         ndm->ndm_state = fdb->state;
 341         ndm->ndm_ifindex = vxlan->dev->ifindex;
 342         ndm->ndm_flags = fdb->flags;
 343         ndm->ndm_type = RTN_UNICAST;
 344
 345         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 346                 goto nla_put_failure;
 347
 348         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 349                 goto nla_put_failure;
 350
 351         if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
 352             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 353                 goto nla_put_failure;
 354         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 355             nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
 356                 goto nla_put_failure;
 357         if (rdst->remote_ifindex &&
 358             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 359                 goto nla_put_failure;
 360
 361         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 362         ci.ndm_confirmed = 0;
 363         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 364         ci.ndm_refcnt    = 0;
 365
 366         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 367                 goto nla_put_failure;
 368
 369         return nlmsg_end(skb, nlh);
 370
 371 nla_put_failure:
 372         nlmsg_cancel(skb, nlh);
 373         return -EMSGSIZE;
 374 }
 375
 376 static inline size_t vxlan_nlmsg_size(void)
 377 {
 378         return NLMSG_ALIGN(sizeof(struct ndmsg))
 379                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 380                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 381                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 382                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 383                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 384                 + nla_total_size(sizeof(struct nda_cacheinfo));
 385 }
 386
 387 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 388                              struct vxlan_rdst *rd, int type)
 389 {
 390         struct net *net = dev_net(vxlan->dev);
 391         struct sk_buff *skb;
 392         int err = -ENOBUFS;
 393
 394         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 395         if (skb == NULL)
 396                 goto errout;
 397
 398         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 399         if (err < 0) {
 400                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 401                 WARN_ON(err == -EMSGSIZE);
 402                 kfree_skb(skb);
 403                 goto errout;
 404         }
 405
 406         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 407         return;
 408 errout:
 409         if (err < 0)
 410                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 411 }
 412
 413 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 414 {
 415         struct vxlan_dev *vxlan = netdev_priv(dev);
 416         struct vxlan_fdb f = {
 417                 .state = NUD_STALE,
 418         };
 419         struct vxlan_rdst remote = {
 420                 .remote_ip = *ipa, /* goes to NDA_DST */
 421                 .remote_vni = VXLAN_N_VID,
 422         };
 423
 424         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 425 }
 426
 427 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 428 {
 429         struct vxlan_fdb f = {
 430                 .state = NUD_STALE,
 431         };
 432         struct vxlan_rdst remote = { };
 433
 434         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 435
 436         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 437 }
 438
 439 /* Hash Ethernet address */
 440 static u32 eth_hash(const unsigned char *addr)
 441 {
 442         u64 value = get_unaligned((u64 *)addr);
 443
 444         /* only want 6 bytes */
 445 #ifdef __BIG_ENDIAN
 446         value >>= 16;
 447 #else
 448         value <<= 16;
 449 #endif
 450         return hash_64(value, FDB_HASH_BITS);
 451 }
 452
 453 /* Hash chain to use given mac address */
 454 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 455                                                 const u8 *mac)
 456 {
 457         return &vxlan->fdb_head[eth_hash(mac)];
 458 }
 459
 460 /* Look up Ethernet address in forwarding table */
 461 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 462                                         const u8 *mac)
 463 {
 464         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 465         struct vxlan_fdb *f;
 466
 467         hlist_for_each_entry_rcu(f, head, hlist) {
 468                 if (ether_addr_equal(mac, f->eth_addr))
 469                         return f;
 470         }
 471
 472         return NULL;
 473 }
 474
 475 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 476                                         const u8 *mac)
 477 {
 478         struct vxlan_fdb *f;
 479
 480         f = __vxlan_find_mac(vxlan, mac);
 481         if (f)
 482                 f->used = jiffies;
 483
 484         return f;
 485 }
 486
 487 /* caller should hold vxlan->hash_lock */
 488 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 489                                               union vxlan_addr *ip, __be16 port,
 490                                               __u32 vni, __u32 ifindex)
 491 {
 492         struct vxlan_rdst *rd;
 493
 494         list_for_each_entry(rd, &f->remotes, list) {
 495                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 496                     rd->remote_port == port &&
 497                     rd->remote_vni == vni &&
 498                     rd->remote_ifindex == ifindex)
 499                         return rd;
 500         }
 501
 502         return NULL;
 503 }
 504
 505 /* Replace destination of unicast mac */
 506 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 507                              union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
 508 {
 509         struct vxlan_rdst *rd;
 510
 511         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 512         if (rd)
 513                 return 0;
 514
 515         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 516         if (!rd)
 517                 return 0;
 518         rd->remote_ip = *ip;
 519         rd->remote_port = port;
 520         rd->remote_vni = vni;
 521         rd->remote_ifindex = ifindex;
 522         return 1;
 523 }
 524
 525 /* Add/update destinations for multicast */
 526 static int vxlan_fdb_append(struct vxlan_fdb *f,
 527                             union vxlan_addr *ip, __be16 port, __u32 vni,
 528                             __u32 ifindex, struct vxlan_rdst **rdp)
 529 {
 530         struct vxlan_rdst *rd;
 531
 532         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 533         if (rd)
 534                 return 0;
 535
 536         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 537         if (rd == NULL)
 538                 return -ENOBUFS;
 539         rd->remote_ip = *ip;
 540         rd->remote_port = port;
 541         rd->remote_vni = vni;
 542         rd->remote_ifindex = ifindex;
 543
 544         list_add_tail_rcu(&rd->list, &f->remotes);
 545
 546         *rdp = rd;
 547         return 1;
 548 }
 549
 550 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head, struct sk_buff *skb)
 551 {
 552         struct sk_buff *p, **pp = NULL;
 553         struct vxlanhdr *vh, *vh2;
 554         struct ethhdr *eh, *eh2;
 555         unsigned int hlen, off_vx, off_eth;
 556         const struct packet_offload *ptype;
 557         __be16 type;
 558         int flush = 1;
 559
 560         off_vx = skb_gro_offset(skb);
 561         hlen = off_vx + sizeof(*vh);
 562         vh   = skb_gro_header_fast(skb, off_vx);
 563         if (skb_gro_header_hard(skb, hlen)) {
 564                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 565                 if (unlikely(!vh))
 566                         goto out;
 567         }
 568         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 569         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 570
 571         off_eth = skb_gro_offset(skb);
 572         hlen = off_eth + sizeof(*eh);
 573         eh   = skb_gro_header_fast(skb, off_eth);
 574         if (skb_gro_header_hard(skb, hlen)) {
 575                 eh = skb_gro_header_slow(skb, hlen, off_eth);
 576                 if (unlikely(!eh))
 577                         goto out;
 578         }
 579
 580         flush = 0;
 581
 582         for (p = *head; p; p = p->next) {
 583                 if (!NAPI_GRO_CB(p)->same_flow)
 584                         continue;
 585
 586                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 587                 eh2 = (struct ethhdr   *)(p->data + off_eth);
 588                 if (vh->vx_vni != vh2->vx_vni || compare_ether_header(eh, eh2)) {
 589                         NAPI_GRO_CB(p)->same_flow = 0;
 590                         continue;
 591                 }
 592         }
 593
 594         type = eh->h_proto;
 595
 596         rcu_read_lock();
 597         ptype = gro_find_receive_by_type(type);
 598         if (ptype == NULL) {
 599                 flush = 1;
 600                 goto out_unlock;
 601         }
 602
 603         skb_gro_pull(skb, sizeof(*eh)); /* pull inner eth header */
 604         skb_gro_postpull_rcsum(skb, eh, sizeof(*eh));
 605         pp = ptype->callbacks.gro_receive(head, skb);
 606
 607 out_unlock:
 608         rcu_read_unlock();
 609 out:
 610         NAPI_GRO_CB(skb)->flush |= flush;
 611
 612         return pp;
 613 }
 614
 615 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff)
 616 {
 617         struct ethhdr *eh;
 618         struct packet_offload *ptype;
 619         __be16 type;
 620         int vxlan_len  = sizeof(struct vxlanhdr) + sizeof(struct ethhdr);
 621         int err = -ENOSYS;
 622
 623         eh = (struct ethhdr *)(skb->data + nhoff + sizeof(struct vxlanhdr));
 624         type = eh->h_proto;
 625
 626         rcu_read_lock();
 627         ptype = gro_find_complete_by_type(type);
 628         if (ptype != NULL)
 629                 err = ptype->callbacks.gro_complete(skb, nhoff + vxlan_len);
 630
 631         rcu_read_unlock();
 632         return err;
 633 }
 634
 635 /* Notify netdevs that UDP port started listening */
 636 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 637 {
 638         struct net_device *dev;
 639         struct sock *sk = vs->sock->sk;
 640         struct net *net = sock_net(sk);
 641         sa_family_t sa_family = sk->sk_family;
 642         __be16 port = inet_sk(sk)->inet_sport;
 643         int err;
 644
 645         if (sa_family == AF_INET) {
 646                 err = udp_add_offload(&vs->udp_offloads);
 647                 if (err)
 648                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 649         }
 650
 651         rcu_read_lock();
 652         for_each_netdev_rcu(net, dev) {
 653                 if (dev->netdev_ops->ndo_add_vxlan_port)
 654                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 655                                                             port);
 656         }
 657         rcu_read_unlock();
 658 }
 659
 660 /* Notify netdevs that UDP port is no more listening */
 661 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 662 {
 663         struct net_device *dev;
 664         struct sock *sk = vs->sock->sk;
 665         struct net *net = sock_net(sk);
 666         sa_family_t sa_family = sk->sk_family;
 667         __be16 port = inet_sk(sk)->inet_sport;
 668
 669         rcu_read_lock();
 670         for_each_netdev_rcu(net, dev) {
 671                 if (dev->netdev_ops->ndo_del_vxlan_port)
 672                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 673                                                             port);
 674         }
 675         rcu_read_unlock();
 676
 677         if (sa_family == AF_INET)
 678                 udp_del_offload(&vs->udp_offloads);
 679 }
 680
 681 /* Add new entry to forwarding table -- assumes lock held */
 682 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 683                             const u8 *mac, union vxlan_addr *ip,
 684                             __u16 state, __u16 flags,
 685                             __be16 port, __u32 vni, __u32 ifindex,
 686                             __u8 ndm_flags)
 687 {
 688         struct vxlan_rdst *rd = NULL;
 689         struct vxlan_fdb *f;
 690         int notify = 0;
 691
 692         f = __vxlan_find_mac(vxlan, mac);
 693         if (f) {
 694                 if (flags & NLM_F_EXCL) {
 695                         netdev_dbg(vxlan->dev,
 696                                    "lost race to create %pM\n", mac);
 697                         return -EEXIST;
 698                 }
 699                 if (f->state != state) {
 700                         f->state = state;
 701                         f->updated = jiffies;
 702                         notify = 1;
 703                 }
 704                 if (f->flags != ndm_flags) {
 705                         f->flags = ndm_flags;
 706                         f->updated = jiffies;
 707                         notify = 1;
 708                 }
 709                 if ((flags & NLM_F_REPLACE)) {
 710                         /* Only change unicasts */
 711                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 712                              is_zero_ether_addr(f->eth_addr))) {
 713                                 int rc = vxlan_fdb_replace(f, ip, port, vni,
 714                                                            ifindex);
 715
 716                                 if (rc < 0)
 717                                         return rc;
 718                                 notify |= rc;
 719                         } else
 720                                 return -EOPNOTSUPP;
 721                 }
 722                 if ((flags & NLM_F_APPEND) &&
 723                     (is_multicast_ether_addr(f->eth_addr) ||
 724                      is_zero_ether_addr(f->eth_addr))) {
 725                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 726                                                   &rd);
 727
 728                         if (rc < 0)
 729                                 return rc;
 730                         notify |= rc;
 731                 }
 732         } else {
 733                 if (!(flags & NLM_F_CREATE))
 734                         return -ENOENT;
 735
 736                 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
 737                         return -ENOSPC;
 738
 739                 /* Disallow replace to add a multicast entry */
 740                 if ((flags & NLM_F_REPLACE) &&
 741                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 742                         return -EOPNOTSUPP;
 743
 744                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 745                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 746                 if (!f)
 747                         return -ENOMEM;
 748
 749                 notify = 1;
 750                 f->state = state;
 751                 f->flags = ndm_flags;
 752                 f->updated = f->used = jiffies;
 753                 INIT_LIST_HEAD(&f->remotes);
 754                 memcpy(f->eth_addr, mac, ETH_ALEN);
 755
 756                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 757
 758                 ++vxlan->addrcnt;
 759                 hlist_add_head_rcu(&f->hlist,
 760                                    vxlan_fdb_head(vxlan, mac));
 761         }
 762
 763         if (notify) {
 764                 if (rd == NULL)
 765                         rd = first_remote_rtnl(f);
 766                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 767         }
 768
 769         return 0;
 770 }
 771
 772 static void vxlan_fdb_free(struct rcu_head *head)
 773 {
 774         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 775         struct vxlan_rdst *rd, *nd;
 776
 777         list_for_each_entry_safe(rd, nd, &f->remotes, list)
 778                 kfree(rd);
 779         kfree(f);
 780 }
 781
 782 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 783 {
 784         netdev_dbg(vxlan->dev,
 785                     "delete %pM\n", f->eth_addr);
 786
 787         --vxlan->addrcnt;
 788         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 789
 790         hlist_del_rcu(&f->hlist);
 791         call_rcu(&f->rcu, vxlan_fdb_free);
 792 }
 793
 794 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 795                            union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
 796 {
 797         struct net *net = dev_net(vxlan->dev);
 798         int err;
 799
 800         if (tb[NDA_DST]) {
 801                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 802                 if (err)
 803                         return err;
 804         } else {
 805                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 806                 if (remote->sa.sa_family == AF_INET) {
 807                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 808                         ip->sa.sa_family = AF_INET;
 809 #if IS_ENABLED(CONFIG_IPV6)
 810                 } else {
 811                         ip->sin6.sin6_addr = in6addr_any;
 812                         ip->sa.sa_family = AF_INET6;
 813 #endif
 814                 }
 815         }
 816
 817         if (tb[NDA_PORT]) {
 818                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 819                         return -EINVAL;
 820                 *port = nla_get_be16(tb[NDA_PORT]);
 821         } else {
 822                 *port = vxlan->dst_port;
 823         }
 824
 825         if (tb[NDA_VNI]) {
 826                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 827                         return -EINVAL;
 828                 *vni = nla_get_u32(tb[NDA_VNI]);
 829         } else {
 830                 *vni = vxlan->default_dst.remote_vni;
 831         }
 832
 833         if (tb[NDA_IFINDEX]) {
 834                 struct net_device *tdev;
 835
 836                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 837                         return -EINVAL;
 838                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 839                 tdev = __dev_get_by_index(net, *ifindex);
 840                 if (!tdev)
 841                         return -EADDRNOTAVAIL;
 842         } else {
 843                 *ifindex = 0;
 844         }
 845
 846         return 0;
 847 }
 848
 849 /* Add static entry (via netlink) */
 850 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 851                          struct net_device *dev,
 852                          const unsigned char *addr, u16 flags)
 853 {
 854         struct vxlan_dev *vxlan = netdev_priv(dev);
 855         /* struct net *net = dev_net(vxlan->dev); */
 856         union vxlan_addr ip;
 857         __be16 port;
 858         u32 vni, ifindex;
 859         int err;
 860
 861         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 862                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 863                         ndm->ndm_state);
 864                 return -EINVAL;
 865         }
 866
 867         if (tb[NDA_DST] == NULL)
 868                 return -EINVAL;
 869
 870         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 871         if (err)
 872                 return err;
 873
 874         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 875                 return -EAFNOSUPPORT;
 876
 877         spin_lock_bh(&vxlan->hash_lock);
 878         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 879                                port, vni, ifindex, ndm->ndm_flags);
 880         spin_unlock_bh(&vxlan->hash_lock);
 881
 882         return err;
 883 }
 884
 885 /* Delete entry (via netlink) */
 886 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 887                             struct net_device *dev,
 888                             const unsigned char *addr)
 889 {
 890         struct vxlan_dev *vxlan = netdev_priv(dev);
 891         struct vxlan_fdb *f;
 892         struct vxlan_rdst *rd = NULL;
 893         union vxlan_addr ip;
 894         __be16 port;
 895         u32 vni, ifindex;
 896         int err;
 897
 898         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 899         if (err)
 900                 return err;
 901
 902         err = -ENOENT;
 903
 904         spin_lock_bh(&vxlan->hash_lock);
 905         f = vxlan_find_mac(vxlan, addr);
 906         if (!f)
 907                 goto out;
 908
 909         if (!vxlan_addr_any(&ip)) {
 910                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 911                 if (!rd)
 912                         goto out;
 913         }
 914
 915         err = 0;
 916
 917         /* remove a destination if it's not the only one on the list,
 918          * otherwise destroy the fdb entry
 919          */
 920         if (rd && !list_is_singular(&f->remotes)) {
 921                 list_del_rcu(&rd->list);
 922                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 923                 kfree_rcu(rd, rcu);
 924                 goto out;
 925         }
 926
 927         vxlan_fdb_destroy(vxlan, f);
 928
 929 out:
 930         spin_unlock_bh(&vxlan->hash_lock);
 931
 932         return err;
 933 }
 934
 935 /* Dump forwarding table */
 936 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 937                           struct net_device *dev,
 938                           struct net_device *filter_dev, int idx)
 939 {
 940         struct vxlan_dev *vxlan = netdev_priv(dev);
 941         unsigned int h;
 942
 943         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 944                 struct vxlan_fdb *f;
 945                 int err;
 946
 947                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 948                         struct vxlan_rdst *rd;
 949
 950                         if (idx < cb->args[0])
 951                                 goto skip;
 952
 953                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 954                                 err = vxlan_fdb_info(skb, vxlan, f,
 955                                                      NETLINK_CB(cb->skb).portid,
 956                                                      cb->nlh->nlmsg_seq,
 957                                                      RTM_NEWNEIGH,
 958                                                      NLM_F_MULTI, rd);
 959                                 if (err < 0)
 960                                         goto out;
 961                         }
 962 skip:
 963                         ++idx;
 964                 }
 965         }
 966 out:
 967         return idx;
 968 }
 969
 970 /* Watch incoming packets to learn mapping between Ethernet address
 971  * and Tunnel endpoint.
 972  * Return true if packet is bogus and should be droppped.
 973  */
 974 static bool vxlan_snoop(struct net_device *dev,
 975                         union vxlan_addr *src_ip, const u8 *src_mac)
 976 {
 977         struct vxlan_dev *vxlan = netdev_priv(dev);
 978         struct vxlan_fdb *f;
 979
 980         f = vxlan_find_mac(vxlan, src_mac);
 981         if (likely(f)) {
 982                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 983
 984                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 985                         return false;
 986
 987                 /* Don't migrate static entries, drop packets */
 988                 if (f->state & NUD_NOARP)
 989                         return true;
 990
 991                 if (net_ratelimit())
 992                         netdev_info(dev,
 993                                     "%pM migrated from %pIS to %pIS\n",
 994                                     src_mac, &rdst->remote_ip, &src_ip);
 995
 996                 rdst->remote_ip = *src_ip;
 997                 f->updated = jiffies;
 998                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 999         } else {
1000                 /* learned new entry */
1001                 spin_lock(&vxlan->hash_lock);
1002
1003                 /* close off race between vxlan_flush and incoming packets */
1004                 if (netif_running(dev))
1005                         vxlan_fdb_create(vxlan, src_mac, src_ip,
1006                                          NUD_REACHABLE,
1007                                          NLM_F_EXCL|NLM_F_CREATE,
1008                                          vxlan->dst_port,
1009                                          vxlan->default_dst.remote_vni,
1010                                          0, NTF_SELF);
1011                 spin_unlock(&vxlan->hash_lock);
1012         }
1013
1014         return false;
1015 }
1016
1017 /* See if multicast group is already in use by other ID */
1018 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1019 {
1020         struct vxlan_dev *vxlan;
1021
1022         /* The vxlan_sock is only used by dev, leaving group has
1023          * no effect on other vxlan devices.
1024          */
1025         if (atomic_read(&dev->vn_sock->refcnt) == 1)
1026                 return false;
1027
1028         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1029                 if (!netif_running(vxlan->dev) || vxlan == dev)
1030                         continue;
1031
1032                 if (vxlan->vn_sock != dev->vn_sock)
1033                         continue;
1034
1035                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1036                                       &dev->default_dst.remote_ip))
1037                         continue;
1038
1039                 if (vxlan->default_dst.remote_ifindex !=
1040                     dev->default_dst.remote_ifindex)
1041                         continue;
1042
1043                 return true;
1044         }
1045
1046         return false;
1047 }
1048
1049 static void vxlan_sock_hold(struct vxlan_sock *vs)
1050 {
1051         atomic_inc(&vs->refcnt);
1052 }
1053
1054 void vxlan_sock_release(struct vxlan_sock *vs)
1055 {
1056         struct sock *sk = vs->sock->sk;
1057         struct net *net = sock_net(sk);
1058         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1059
1060         if (!atomic_dec_and_test(&vs->refcnt))
1061                 return;
1062
1063         spin_lock(&vn->sock_lock);
1064         hlist_del_rcu(&vs->hlist);
1065         rcu_assign_sk_user_data(vs->sock->sk, NULL);
1066         vxlan_notify_del_rx_port(vs);
1067         spin_unlock(&vn->sock_lock);
1068
1069         queue_work(vxlan_wq, &vs->del_work);
1070 }
1071 EXPORT_SYMBOL_GPL(vxlan_sock_release);
1072
1073 /* Callback to update multicast group membership when first VNI on
1074  * multicast asddress is brought up
1075  * Done as workqueue because ip_mc_join_group acquires RTNL.
1076  */
1077 static void vxlan_igmp_join(struct work_struct *work)
1078 {
1079         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_join);
1080         struct vxlan_sock *vs = vxlan->vn_sock;
1081         struct sock *sk = vs->sock->sk;
1082         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1083         int ifindex = vxlan->default_dst.remote_ifindex;
1084
1085         lock_sock(sk);
1086         if (ip->sa.sa_family == AF_INET) {
1087                 struct ip_mreqn mreq = {
1088                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1089                         .imr_ifindex            = ifindex,
1090                 };
1091
1092                 ip_mc_join_group(sk, &mreq);
1093 #if IS_ENABLED(CONFIG_IPV6)
1094         } else {
1095                 ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1096                                              &ip->sin6.sin6_addr);
1097 #endif
1098         }
1099         release_sock(sk);
1100
1101         vxlan_sock_release(vs);
1102         dev_put(vxlan->dev);
1103 }
1104
1105 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1106 static void vxlan_igmp_leave(struct work_struct *work)
1107 {
1108         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_leave);
1109         struct vxlan_sock *vs = vxlan->vn_sock;
1110         struct sock *sk = vs->sock->sk;
1111         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1112         int ifindex = vxlan->default_dst.remote_ifindex;
1113
1114         lock_sock(sk);
1115         if (ip->sa.sa_family == AF_INET) {
1116                 struct ip_mreqn mreq = {
1117                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1118                         .imr_ifindex            = ifindex,
1119                 };
1120
1121                 ip_mc_leave_group(sk, &mreq);
1122 #if IS_ENABLED(CONFIG_IPV6)
1123         } else {
1124                 ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1125                                              &ip->sin6.sin6_addr);
1126 #endif
1127         }
1128
1129         release_sock(sk);
1130
1131         vxlan_sock_release(vs);
1132         dev_put(vxlan->dev);
1133 }
1134
1135 /* Callback from net/ipv4/udp.c to receive packets */
1136 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1137 {
1138         struct vxlan_sock *vs;
1139         struct vxlanhdr *vxh;
1140
1141         /* Need Vxlan and inner Ethernet header to be present */
1142         if (!pskb_may_pull(skb, VXLAN_HLEN))
1143                 goto error;
1144
1145         /* Return packets with reserved bits set */
1146         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1147         if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
1148             (vxh->vx_vni & htonl(0xff))) {
1149                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1150                            ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1151                 goto error;
1152         }
1153
1154         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1155                 goto drop;
1156
1157         vs = rcu_dereference_sk_user_data(sk);
1158         if (!vs)
1159                 goto drop;
1160
1161         skb_pop_rcv_encapsulation(skb);
1162
1163         vs->rcv(vs, skb, vxh->vx_vni);
1164         return 0;
1165
1166 drop:
1167         /* Consume bad packet */
1168         kfree_skb(skb);
1169         return 0;
1170
1171 error:
1172         /* Return non vxlan pkt */
1173         return 1;
1174 }
1175
1176 static void vxlan_rcv(struct vxlan_sock *vs,
1177                       struct sk_buff *skb, __be32 vx_vni)
1178 {
1179         struct iphdr *oip = NULL;
1180         struct ipv6hdr *oip6 = NULL;
1181         struct vxlan_dev *vxlan;
1182         struct pcpu_sw_netstats *stats;
1183         union vxlan_addr saddr;
1184         __u32 vni;
1185         int err = 0;
1186         union vxlan_addr *remote_ip;
1187
1188         vni = ntohl(vx_vni) >> 8;
1189         /* Is this VNI defined? */
1190         vxlan = vxlan_vs_find_vni(vs, vni);
1191         if (!vxlan)
1192                 goto drop;
1193
1194         remote_ip = &vxlan->default_dst.remote_ip;
1195         skb_reset_mac_header(skb);
1196         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1197         skb->protocol = eth_type_trans(skb, vxlan->dev);
1198         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1199
1200         /* Ignore packet loops (and multicast echo) */
1201         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1202                 goto drop;
1203
1204         /* Re-examine inner Ethernet packet */
1205         if (remote_ip->sa.sa_family == AF_INET) {
1206                 oip = ip_hdr(skb);
1207                 saddr.sin.sin_addr.s_addr = oip->saddr;
1208                 saddr.sa.sa_family = AF_INET;
1209 #if IS_ENABLED(CONFIG_IPV6)
1210         } else {
1211                 oip6 = ipv6_hdr(skb);
1212                 saddr.sin6.sin6_addr = oip6->saddr;
1213                 saddr.sa.sa_family = AF_INET6;
1214 #endif
1215         }
1216
1217         if ((vxlan->flags & VXLAN_F_LEARN) &&
1218             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1219                 goto drop;
1220
1221         skb_reset_network_header(skb);
1222
1223         if (oip6)
1224                 err = IP6_ECN_decapsulate(oip6, skb);
1225         if (oip)
1226                 err = IP_ECN_decapsulate(oip, skb);
1227
1228         if (unlikely(err)) {
1229                 if (log_ecn_error) {
1230                         if (oip6)
1231                                 net_info_ratelimited("non-ECT from %pI6\n",
1232                                                      &oip6->saddr);
1233                         if (oip)
1234                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1235                                                      &oip->saddr, oip->tos);
1236                 }
1237                 if (err > 1) {
1238                         ++vxlan->dev->stats.rx_frame_errors;
1239                         ++vxlan->dev->stats.rx_errors;
1240                         goto drop;
1241                 }
1242         }
1243
1244         stats = this_cpu_ptr(vxlan->dev->tstats);
1245         u64_stats_update_begin(&stats->syncp);
1246         stats->rx_packets++;
1247         stats->rx_bytes += skb->len;
1248         u64_stats_update_end(&stats->syncp);
1249
1250         netif_rx(skb);
1251
1252         return;
1253 drop:
1254         /* Consume bad packet */
1255         kfree_skb(skb);
1256 }
1257
1258 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1259 {
1260         struct vxlan_dev *vxlan = netdev_priv(dev);
1261         struct arphdr *parp;
1262         u8 *arpptr, *sha;
1263         __be32 sip, tip;
1264         struct neighbour *n;
1265
1266         if (dev->flags & IFF_NOARP)
1267                 goto out;
1268
1269         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1270                 dev->stats.tx_dropped++;
1271                 goto out;
1272         }
1273         parp = arp_hdr(skb);
1274
1275         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1276              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1277             parp->ar_pro != htons(ETH_P_IP) ||
1278             parp->ar_op != htons(ARPOP_REQUEST) ||
1279             parp->ar_hln != dev->addr_len ||
1280             parp->ar_pln != 4)
1281                 goto out;
1282         arpptr = (u8 *)parp + sizeof(struct arphdr);
1283         sha = arpptr;
1284         arpptr += dev->addr_len;        /* sha */
1285         memcpy(&sip, arpptr, sizeof(sip));
1286         arpptr += sizeof(sip);
1287         arpptr += dev->addr_len;        /* tha */
1288         memcpy(&tip, arpptr, sizeof(tip));
1289
1290         if (ipv4_is_loopback(tip) ||
1291             ipv4_is_multicast(tip))
1292                 goto out;
1293
1294         n = neigh_lookup(&arp_tbl, &tip, dev);
1295
1296         if (n) {
1297                 struct vxlan_fdb *f;
1298                 struct sk_buff  *reply;
1299
1300                 if (!(n->nud_state & NUD_CONNECTED)) {
1301                         neigh_release(n);
1302                         goto out;
1303                 }
1304
1305                 f = vxlan_find_mac(vxlan, n->ha);
1306                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1307                         /* bridge-local neighbor */
1308                         neigh_release(n);
1309                         goto out;
1310                 }
1311
1312                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1313                                 n->ha, sha);
1314
1315                 neigh_release(n);
1316
1317                 if (reply == NULL)
1318                         goto out;
1319
1320                 skb_reset_mac_header(reply);
1321                 __skb_pull(reply, skb_network_offset(reply));
1322                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1323                 reply->pkt_type = PACKET_HOST;
1324
1325                 if (netif_rx_ni(reply) == NET_RX_DROP)
1326                         dev->stats.rx_dropped++;
1327         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1328                 union vxlan_addr ipa = {
1329                         .sin.sin_addr.s_addr = tip,
1330                         .sa.sa_family = AF_INET,
1331                 };
1332
1333                 vxlan_ip_miss(dev, &ipa);
1334         }
1335 out:
1336         consume_skb(skb);
1337         return NETDEV_TX_OK;
1338 }
1339
1340 #if IS_ENABLED(CONFIG_IPV6)
1341
1342 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1343         struct neighbour *n, bool isrouter)
1344 {
1345         struct net_device *dev = request->dev;
1346         struct sk_buff *reply;
1347         struct nd_msg *ns, *na;
1348         struct ipv6hdr *pip6;
1349         u8 *daddr;
1350         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1351         int ns_olen;
1352         int i, len;
1353
1354         if (dev == NULL)
1355                 return NULL;
1356
1357         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1358                 sizeof(*na) + na_olen + dev->needed_tailroom;
1359         reply = alloc_skb(len, GFP_ATOMIC);
1360         if (reply == NULL)
1361                 return NULL;
1362
1363         reply->protocol = htons(ETH_P_IPV6);
1364         reply->dev = dev;
1365         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1366         skb_push(reply, sizeof(struct ethhdr));
1367         skb_set_mac_header(reply, 0);
1368
1369         ns = (struct nd_msg *)skb_transport_header(request);
1370
1371         daddr = eth_hdr(request)->h_source;
1372         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1373         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1374                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1375                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1376                         break;
1377                 }
1378         }
1379
1380         /* Ethernet header */
1381         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1382         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1383         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1384         reply->protocol = htons(ETH_P_IPV6);
1385
1386         skb_pull(reply, sizeof(struct ethhdr));
1387         skb_set_network_header(reply, 0);
1388         skb_put(reply, sizeof(struct ipv6hdr));
1389
1390         /* IPv6 header */
1391
1392         pip6 = ipv6_hdr(reply);
1393         memset(pip6, 0, sizeof(struct ipv6hdr));
1394         pip6->version = 6;
1395         pip6->priority = ipv6_hdr(request)->priority;
1396         pip6->nexthdr = IPPROTO_ICMPV6;
1397         pip6->hop_limit = 255;
1398         pip6->daddr = ipv6_hdr(request)->saddr;
1399         pip6->saddr = *(struct in6_addr *)n->primary_key;
1400
1401         skb_pull(reply, sizeof(struct ipv6hdr));
1402         skb_set_transport_header(reply, 0);
1403
1404         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1405
1406         /* Neighbor Advertisement */
1407         memset(na, 0, sizeof(*na)+na_olen);
1408         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1409         na->icmph.icmp6_router = isrouter;
1410         na->icmph.icmp6_override = 1;
1411         na->icmph.icmp6_solicited = 1;
1412         na->target = ns->target;
1413         ether_addr_copy(&na->opt[2], n->ha);
1414         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1415         na->opt[1] = na_olen >> 3;
1416
1417         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1418                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1419                 csum_partial(na, sizeof(*na)+na_olen, 0));
1420
1421         pip6->payload_len = htons(sizeof(*na)+na_olen);
1422
1423         skb_push(reply, sizeof(struct ipv6hdr));
1424
1425         reply->ip_summed = CHECKSUM_UNNECESSARY;
1426
1427         return reply;
1428 }
1429
1430 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1431 {
1432         struct vxlan_dev *vxlan = netdev_priv(dev);
1433         struct nd_msg *msg;
1434         const struct ipv6hdr *iphdr;
1435         const struct in6_addr *saddr, *daddr;
1436         struct neighbour *n;
1437         struct inet6_dev *in6_dev;
1438
1439         in6_dev = __in6_dev_get(dev);
1440         if (!in6_dev)
1441                 goto out;
1442
1443         if (!pskb_may_pull(skb, skb->len))
1444                 goto out;
1445
1446         iphdr = ipv6_hdr(skb);
1447         saddr = &iphdr->saddr;
1448         daddr = &iphdr->daddr;
1449
1450         msg = (struct nd_msg *)skb_transport_header(skb);
1451         if (msg->icmph.icmp6_code != 0 ||
1452             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1453                 goto out;
1454
1455         if (ipv6_addr_loopback(daddr) ||
1456             ipv6_addr_is_multicast(&msg->target))
1457                 goto out;
1458
1459         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1460
1461         if (n) {
1462                 struct vxlan_fdb *f;
1463                 struct sk_buff *reply;
1464
1465                 if (!(n->nud_state & NUD_CONNECTED)) {
1466                         neigh_release(n);
1467                         goto out;
1468                 }
1469
1470                 f = vxlan_find_mac(vxlan, n->ha);
1471                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1472                         /* bridge-local neighbor */
1473                         neigh_release(n);
1474                         goto out;
1475                 }
1476
1477                 reply = vxlan_na_create(skb, n,
1478                                         !!(f ? f->flags & NTF_ROUTER : 0));
1479
1480                 neigh_release(n);
1481
1482                 if (reply == NULL)
1483                         goto out;
1484
1485                 if (netif_rx_ni(reply) == NET_RX_DROP)
1486                         dev->stats.rx_dropped++;
1487
1488         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1489                 union vxlan_addr ipa = {
1490                         .sin6.sin6_addr = msg->target,
1491                         .sa.sa_family = AF_INET6,
1492                 };
1493
1494                 vxlan_ip_miss(dev, &ipa);
1495         }
1496
1497 out:
1498         consume_skb(skb);
1499         return NETDEV_TX_OK;
1500 }
1501 #endif
1502
1503 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1504 {
1505         struct vxlan_dev *vxlan = netdev_priv(dev);
1506         struct neighbour *n;
1507
1508         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1509                 return false;
1510
1511         n = NULL;
1512         switch (ntohs(eth_hdr(skb)->h_proto)) {
1513         case ETH_P_IP:
1514         {
1515                 struct iphdr *pip;
1516
1517                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1518                         return false;
1519                 pip = ip_hdr(skb);
1520                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1521                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1522                         union vxlan_addr ipa = {
1523                                 .sin.sin_addr.s_addr = pip->daddr,
1524                                 .sa.sa_family = AF_INET,
1525                         };
1526
1527                         vxlan_ip_miss(dev, &ipa);
1528                         return false;
1529                 }
1530
1531                 break;
1532         }
1533 #if IS_ENABLED(CONFIG_IPV6)
1534         case ETH_P_IPV6:
1535         {
1536                 struct ipv6hdr *pip6;
1537
1538                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1539                         return false;
1540                 pip6 = ipv6_hdr(skb);
1541                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1542                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1543                         union vxlan_addr ipa = {
1544                                 .sin6.sin6_addr = pip6->daddr,
1545                                 .sa.sa_family = AF_INET6,
1546                         };
1547
1548                         vxlan_ip_miss(dev, &ipa);
1549                         return false;
1550                 }
1551
1552                 break;
1553         }
1554 #endif
1555         default:
1556                 return false;
1557         }
1558
1559         if (n) {
1560                 bool diff;
1561
1562                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1563                 if (diff) {
1564                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1565                                 dev->addr_len);
1566                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1567                 }
1568                 neigh_release(n);
1569                 return diff;
1570         }
1571
1572         return false;
1573 }
1574
1575 static inline struct sk_buff *vxlan_handle_offloads(struct sk_buff *skb,
1576                                                     bool udp_csum)
1577 {
1578         int type = udp_csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1579         return iptunnel_handle_offloads(skb, udp_csum, type);
1580 }
1581
1582 #if IS_ENABLED(CONFIG_IPV6)
1583 static int vxlan6_xmit_skb(struct vxlan_sock *vs,
1584                            struct dst_entry *dst, struct sk_buff *skb,
1585                            struct net_device *dev, struct in6_addr *saddr,
1586                            struct in6_addr *daddr, __u8 prio, __u8 ttl,
1587                            __be16 src_port, __be16 dst_port, __be32 vni,
1588                            bool xnet)
1589 {
1590         struct ipv6hdr *ip6h;
1591         struct vxlanhdr *vxh;
1592         struct udphdr *uh;
1593         int min_headroom;
1594         int err;
1595
1596         skb = vxlan_handle_offloads(skb, !udp_get_no_check6_tx(vs->sock->sk));
1597         if (IS_ERR(skb))
1598                 return -EINVAL;
1599
1600         skb_scrub_packet(skb, xnet);
1601
1602         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1603                         + VXLAN_HLEN + sizeof(struct ipv6hdr)
1604                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1605
1606         /* Need space for new headers (invalidates iph ptr) */
1607         err = skb_cow_head(skb, min_headroom);
1608         if (unlikely(err))
1609                 return err;
1610
1611         if (vlan_tx_tag_present(skb)) {
1612                 if (WARN_ON(!__vlan_put_tag(skb,
1613                                             skb->vlan_proto,
1614                                             vlan_tx_tag_get(skb))))
1615                         return -ENOMEM;
1616
1617                 skb->vlan_tci = 0;
1618         }
1619
1620         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1621         vxh->vx_flags = htonl(VXLAN_FLAGS);
1622         vxh->vx_vni = vni;
1623
1624         __skb_push(skb, sizeof(*uh));
1625         skb_reset_transport_header(skb);
1626         uh = udp_hdr(skb);
1627
1628         uh->dest = dst_port;
1629         uh->source = src_port;
1630
1631         uh->len = htons(skb->len);
1632
1633         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1634         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1635                               IPSKB_REROUTED);
1636         skb_dst_set(skb, dst);
1637
1638         udp6_set_csum(udp_get_no_check6_tx(vs->sock->sk), skb,
1639                       saddr, daddr, skb->len);
1640
1641         __skb_push(skb, sizeof(*ip6h));
1642         skb_reset_network_header(skb);
1643         ip6h              = ipv6_hdr(skb);
1644         ip6h->version     = 6;
1645         ip6h->priority    = prio;
1646         ip6h->flow_lbl[0] = 0;
1647         ip6h->flow_lbl[1] = 0;
1648         ip6h->flow_lbl[2] = 0;
1649         ip6h->payload_len = htons(skb->len);
1650         ip6h->nexthdr     = IPPROTO_UDP;
1651         ip6h->hop_limit   = ttl;
1652         ip6h->daddr       = *daddr;
1653         ip6h->saddr       = *saddr;
1654
1655         ip6tunnel_xmit(skb, dev);
1656         return 0;
1657 }
1658 #endif
1659
1660 int vxlan_xmit_skb(struct vxlan_sock *vs,
1661                    struct rtable *rt, struct sk_buff *skb,
1662                    __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1663                    __be16 src_port, __be16 dst_port, __be32 vni, bool xnet)
1664 {
1665         struct vxlanhdr *vxh;
1666         struct udphdr *uh;
1667         int min_headroom;
1668         int err;
1669
1670         skb = vxlan_handle_offloads(skb, !vs->sock->sk->sk_no_check_tx);
1671         if (IS_ERR(skb))
1672                 return -EINVAL;
1673
1674         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1675                         + VXLAN_HLEN + sizeof(struct iphdr)
1676                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1677
1678         /* Need space for new headers (invalidates iph ptr) */
1679         err = skb_cow_head(skb, min_headroom);
1680         if (unlikely(err))
1681                 return err;
1682
1683         if (vlan_tx_tag_present(skb)) {
1684                 if (WARN_ON(!__vlan_put_tag(skb,
1685                                             skb->vlan_proto,
1686                                             vlan_tx_tag_get(skb))))
1687                         return -ENOMEM;
1688
1689                 skb->vlan_tci = 0;
1690         }
1691
1692         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1693         vxh->vx_flags = htonl(VXLAN_FLAGS);
1694         vxh->vx_vni = vni;
1695
1696         __skb_push(skb, sizeof(*uh));
1697         skb_reset_transport_header(skb);
1698         uh = udp_hdr(skb);
1699
1700         uh->dest = dst_port;
1701         uh->source = src_port;
1702
1703         uh->len = htons(skb->len);
1704
1705         udp_set_csum(vs->sock->sk->sk_no_check_tx, skb,
1706                      src, dst, skb->len);
1707
1708         return iptunnel_xmit(vs->sock->sk, rt, skb, src, dst, IPPROTO_UDP,
1709                              tos, ttl, df, xnet);
1710 }
1711 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1712
1713 /* Bypass encapsulation if the destination is local */
1714 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1715                                struct vxlan_dev *dst_vxlan)
1716 {
1717         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1718         union vxlan_addr loopback;
1719         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1720
1721         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1722         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1723         skb->pkt_type = PACKET_HOST;
1724         skb->encapsulation = 0;
1725         skb->dev = dst_vxlan->dev;
1726         __skb_pull(skb, skb_network_offset(skb));
1727
1728         if (remote_ip->sa.sa_family == AF_INET) {
1729                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1730                 loopback.sa.sa_family =  AF_INET;
1731 #if IS_ENABLED(CONFIG_IPV6)
1732         } else {
1733                 loopback.sin6.sin6_addr = in6addr_loopback;
1734                 loopback.sa.sa_family =  AF_INET6;
1735 #endif
1736         }
1737
1738         if (dst_vxlan->flags & VXLAN_F_LEARN)
1739                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1740
1741         u64_stats_update_begin(&tx_stats->syncp);
1742         tx_stats->tx_packets++;
1743         tx_stats->tx_bytes += skb->len;
1744         u64_stats_update_end(&tx_stats->syncp);
1745
1746         if (netif_rx(skb) == NET_RX_SUCCESS) {
1747                 u64_stats_update_begin(&rx_stats->syncp);
1748                 rx_stats->rx_packets++;
1749                 rx_stats->rx_bytes += skb->len;
1750                 u64_stats_update_end(&rx_stats->syncp);
1751         } else {
1752                 skb->dev->stats.rx_dropped++;
1753         }
1754 }
1755
1756 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1757                            struct vxlan_rdst *rdst, bool did_rsc)
1758 {
1759         struct vxlan_dev *vxlan = netdev_priv(dev);
1760         struct rtable *rt = NULL;
1761         const struct iphdr *old_iph;
1762         struct flowi4 fl4;
1763         union vxlan_addr *dst;
1764         __be16 src_port = 0, dst_port;
1765         u32 vni;
1766         __be16 df = 0;
1767         __u8 tos, ttl;
1768         int err;
1769
1770         dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1771         vni = rdst->remote_vni;
1772         dst = &rdst->remote_ip;
1773
1774         if (vxlan_addr_any(dst)) {
1775                 if (did_rsc) {
1776                         /* short-circuited back to local bridge */
1777                         vxlan_encap_bypass(skb, vxlan, vxlan);
1778                         return;
1779                 }
1780                 goto drop;
1781         }
1782
1783         old_iph = ip_hdr(skb);
1784
1785         ttl = vxlan->ttl;
1786         if (!ttl && vxlan_addr_multicast(dst))
1787                 ttl = 1;
1788
1789         tos = vxlan->tos;
1790         if (tos == 1)
1791                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1792
1793         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->port_min,
1794                                      vxlan->port_max, true);
1795
1796         if (dst->sa.sa_family == AF_INET) {
1797                 memset(&fl4, 0, sizeof(fl4));
1798                 fl4.flowi4_oif = rdst->remote_ifindex;
1799                 fl4.flowi4_tos = RT_TOS(tos);
1800                 fl4.daddr = dst->sin.sin_addr.s_addr;
1801                 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1802
1803                 rt = ip_route_output_key(vxlan->net, &fl4);
1804                 if (IS_ERR(rt)) {
1805                         netdev_dbg(dev, "no route to %pI4\n",
1806                                    &dst->sin.sin_addr.s_addr);
1807                         dev->stats.tx_carrier_errors++;
1808                         goto tx_error;
1809                 }
1810
1811                 if (rt->dst.dev == dev) {
1812                         netdev_dbg(dev, "circular route to %pI4\n",
1813                                    &dst->sin.sin_addr.s_addr);
1814                         dev->stats.collisions++;
1815                         goto rt_tx_error;
1816                 }
1817
1818                 /* Bypass encapsulation if the destination is local */
1819                 if (rt->rt_flags & RTCF_LOCAL &&
1820                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1821                         struct vxlan_dev *dst_vxlan;
1822
1823                         ip_rt_put(rt);
1824                         dst_vxlan = vxlan_find_vni(vxlan->net, vni, dst_port);
1825                         if (!dst_vxlan)
1826                                 goto tx_error;
1827                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1828                         return;
1829                 }
1830
1831                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1832                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1833
1834                 err = vxlan_xmit_skb(vxlan->vn_sock, rt, skb,
1835                                      fl4.saddr, dst->sin.sin_addr.s_addr,
1836                                      tos, ttl, df, src_port, dst_port,
1837                                      htonl(vni << 8),
1838                                      !net_eq(vxlan->net, dev_net(vxlan->dev)));
1839
1840                 if (err < 0)
1841                         goto rt_tx_error;
1842                 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1843 #if IS_ENABLED(CONFIG_IPV6)
1844         } else {
1845                 struct sock *sk = vxlan->vn_sock->sock->sk;
1846                 struct dst_entry *ndst;
1847                 struct flowi6 fl6;
1848                 u32 flags;
1849
1850                 memset(&fl6, 0, sizeof(fl6));
1851                 fl6.flowi6_oif = rdst->remote_ifindex;
1852                 fl6.daddr = dst->sin6.sin6_addr;
1853                 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
1854                 fl6.flowi6_proto = IPPROTO_UDP;
1855
1856                 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
1857                         netdev_dbg(dev, "no route to %pI6\n",
1858                                    &dst->sin6.sin6_addr);
1859                         dev->stats.tx_carrier_errors++;
1860                         goto tx_error;
1861                 }
1862
1863                 if (ndst->dev == dev) {
1864                         netdev_dbg(dev, "circular route to %pI6\n",
1865                                    &dst->sin6.sin6_addr);
1866                         dst_release(ndst);
1867                         dev->stats.collisions++;
1868                         goto tx_error;
1869                 }
1870
1871                 /* Bypass encapsulation if the destination is local */
1872                 flags = ((struct rt6_info *)ndst)->rt6i_flags;
1873                 if (flags & RTF_LOCAL &&
1874                     !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1875                         struct vxlan_dev *dst_vxlan;
1876
1877                         dst_release(ndst);
1878                         dst_vxlan = vxlan_find_vni(vxlan->net, vni, dst_port);
1879                         if (!dst_vxlan)
1880                                 goto tx_error;
1881                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1882                         return;
1883                 }
1884
1885                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
1886
1887                 err = vxlan6_xmit_skb(vxlan->vn_sock, ndst, skb,
1888                                       dev, &fl6.saddr, &fl6.daddr, 0, ttl,
1889                                       src_port, dst_port, htonl(vni << 8),
1890                                       !net_eq(vxlan->net, dev_net(vxlan->dev)));
1891 #endif
1892         }
1893
1894         return;
1895
1896 drop:
1897         dev->stats.tx_dropped++;
1898         goto tx_free;
1899
1900 rt_tx_error:
1901         ip_rt_put(rt);
1902 tx_error:
1903         dev->stats.tx_errors++;
1904 tx_free:
1905         dev_kfree_skb(skb);
1906 }
1907
1908 /* Transmit local packets over Vxlan
1909  *
1910  * Outer IP header inherits ECN and DF from inner header.
1911  * Outer UDP destination is the VXLAN assigned port.
1912  *           source port is based on hash of flow
1913  */
1914 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1915 {
1916         struct vxlan_dev *vxlan = netdev_priv(dev);
1917         struct ethhdr *eth;
1918         bool did_rsc = false;
1919         struct vxlan_rdst *rdst, *fdst = NULL;
1920         struct vxlan_fdb *f;
1921
1922         skb_reset_mac_header(skb);
1923         eth = eth_hdr(skb);
1924
1925         if ((vxlan->flags & VXLAN_F_PROXY)) {
1926                 if (ntohs(eth->h_proto) == ETH_P_ARP)
1927                         return arp_reduce(dev, skb);
1928 #if IS_ENABLED(CONFIG_IPV6)
1929                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
1930                          skb->len >= sizeof(struct ipv6hdr) + sizeof(struct nd_msg) &&
1931                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
1932                                 struct nd_msg *msg;
1933
1934                                 msg = (struct nd_msg *)skb_transport_header(skb);
1935                                 if (msg->icmph.icmp6_code == 0 &&
1936                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
1937                                         return neigh_reduce(dev, skb);
1938                 }
1939 #endif
1940         }
1941
1942         f = vxlan_find_mac(vxlan, eth->h_dest);
1943         did_rsc = false;
1944
1945         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
1946             (ntohs(eth->h_proto) == ETH_P_IP ||
1947              ntohs(eth->h_proto) == ETH_P_IPV6)) {
1948                 did_rsc = route_shortcircuit(dev, skb);
1949                 if (did_rsc)
1950                         f = vxlan_find_mac(vxlan, eth->h_dest);
1951         }
1952
1953         if (f == NULL) {
1954                 f = vxlan_find_mac(vxlan, all_zeros_mac);
1955                 if (f == NULL) {
1956                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
1957                             !is_multicast_ether_addr(eth->h_dest))
1958                                 vxlan_fdb_miss(vxlan, eth->h_dest);
1959
1960                         dev->stats.tx_dropped++;
1961                         kfree_skb(skb);
1962                         return NETDEV_TX_OK;
1963                 }
1964         }
1965
1966         list_for_each_entry_rcu(rdst, &f->remotes, list) {
1967                 struct sk_buff *skb1;
1968
1969                 if (!fdst) {
1970                         fdst = rdst;
1971                         continue;
1972                 }
1973                 skb1 = skb_clone(skb, GFP_ATOMIC);
1974                 if (skb1)
1975                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1976         }
1977
1978         if (fdst)
1979                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
1980         else
1981                 kfree_skb(skb);
1982         return NETDEV_TX_OK;
1983 }
1984
1985 /* Walk the forwarding table and purge stale entries */
1986 static void vxlan_cleanup(unsigned long arg)
1987 {
1988         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1989         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1990         unsigned int h;
1991
1992         if (!netif_running(vxlan->dev))
1993                 return;
1994
1995         spin_lock_bh(&vxlan->hash_lock);
1996         for (h = 0; h < FDB_HASH_SIZE; ++h) {
1997                 struct hlist_node *p, *n;
1998                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1999                         struct vxlan_fdb *f
2000                                 = container_of(p, struct vxlan_fdb, hlist);
2001                         unsigned long timeout;
2002
2003                         if (f->state & NUD_PERMANENT)
2004                                 continue;
2005
2006                         timeout = f->used + vxlan->age_interval * HZ;
2007                         if (time_before_eq(timeout, jiffies)) {
2008                                 netdev_dbg(vxlan->dev,
2009                                            "garbage collect %pM\n",
2010                                            f->eth_addr);
2011                                 f->state = NUD_STALE;
2012                                 vxlan_fdb_destroy(vxlan, f);
2013                         } else if (time_before(timeout, next_timer))
2014                                 next_timer = timeout;
2015                 }
2016         }
2017         spin_unlock_bh(&vxlan->hash_lock);
2018
2019         mod_timer(&vxlan->age_timer, next_timer);
2020 }
2021
2022 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2023 {
2024         __u32 vni = vxlan->default_dst.remote_vni;
2025
2026         vxlan->vn_sock = vs;
2027         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2028 }
2029
2030 /* Setup stats when device is created */
2031 static int vxlan_init(struct net_device *dev)
2032 {
2033         struct vxlan_dev *vxlan = netdev_priv(dev);
2034         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2035         struct vxlan_sock *vs;
2036
2037         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2038         if (!dev->tstats)
2039                 return -ENOMEM;
2040
2041         spin_lock(&vn->sock_lock);
2042         vs = vxlan_find_sock(vxlan->net, vxlan->dst_port);
2043         if (vs) {
2044                 /* If we have a socket with same port already, reuse it */
2045                 atomic_inc(&vs->refcnt);
2046                 vxlan_vs_add_dev(vs, vxlan);
2047         } else {
2048                 /* otherwise make new socket outside of RTNL */
2049                 dev_hold(dev);
2050                 queue_work(vxlan_wq, &vxlan->sock_work);
2051         }
2052         spin_unlock(&vn->sock_lock);
2053
2054         return 0;
2055 }
2056
2057 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2058 {
2059         struct vxlan_fdb *f;
2060
2061         spin_lock_bh(&vxlan->hash_lock);
2062         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2063         if (f)
2064                 vxlan_fdb_destroy(vxlan, f);
2065         spin_unlock_bh(&vxlan->hash_lock);
2066 }
2067
2068 static void vxlan_uninit(struct net_device *dev)
2069 {
2070         struct vxlan_dev *vxlan = netdev_priv(dev);
2071         struct vxlan_sock *vs = vxlan->vn_sock;
2072
2073         vxlan_fdb_delete_default(vxlan);
2074
2075         if (vs)
2076                 vxlan_sock_release(vs);
2077         free_percpu(dev->tstats);
2078 }
2079
2080 /* Start ageing timer and join group when device is brought up */
2081 static int vxlan_open(struct net_device *dev)
2082 {
2083         struct vxlan_dev *vxlan = netdev_priv(dev);
2084         struct vxlan_sock *vs = vxlan->vn_sock;
2085
2086         /* socket hasn't been created */
2087         if (!vs)
2088                 return -ENOTCONN;
2089
2090         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2091                 vxlan_sock_hold(vs);
2092                 dev_hold(dev);
2093                 queue_work(vxlan_wq, &vxlan->igmp_join);
2094         }
2095
2096         if (vxlan->age_interval)
2097                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2098
2099         return 0;
2100 }
2101
2102 /* Purge the forwarding table */
2103 static void vxlan_flush(struct vxlan_dev *vxlan)
2104 {
2105         unsigned int h;
2106
2107         spin_lock_bh(&vxlan->hash_lock);
2108         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2109                 struct hlist_node *p, *n;
2110                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2111                         struct vxlan_fdb *f
2112                                 = container_of(p, struct vxlan_fdb, hlist);
2113                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2114                         if (!is_zero_ether_addr(f->eth_addr))
2115                                 vxlan_fdb_destroy(vxlan, f);
2116                 }
2117         }
2118         spin_unlock_bh(&vxlan->hash_lock);
2119 }
2120
2121 /* Cleanup timer and forwarding table on shutdown */
2122 static int vxlan_stop(struct net_device *dev)
2123 {
2124         struct vxlan_dev *vxlan = netdev_priv(dev);
2125         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2126         struct vxlan_sock *vs = vxlan->vn_sock;
2127
2128         if (vs && vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2129             !vxlan_group_used(vn, vxlan)) {
2130                 vxlan_sock_hold(vs);
2131                 dev_hold(dev);
2132                 queue_work(vxlan_wq, &vxlan->igmp_leave);
2133         }
2134
2135         del_timer_sync(&vxlan->age_timer);
2136
2137         vxlan_flush(vxlan);
2138
2139         return 0;
2140 }
2141
2142 /* Stub, nothing needs to be done. */
2143 static void vxlan_set_multicast_list(struct net_device *dev)
2144 {
2145 }
2146
2147 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2148 {
2149         struct vxlan_dev *vxlan = netdev_priv(dev);
2150         struct vxlan_rdst *dst = &vxlan->default_dst;
2151         struct net_device *lowerdev;
2152         int max_mtu;
2153
2154         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2155         if (lowerdev == NULL)
2156                 return eth_change_mtu(dev, new_mtu);
2157
2158         if (dst->remote_ip.sa.sa_family == AF_INET6)
2159                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2160         else
2161                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2162
2163         if (new_mtu < 68 || new_mtu > max_mtu)
2164                 return -EINVAL;
2165
2166         dev->mtu = new_mtu;
2167         return 0;
2168 }
2169
2170 static const struct net_device_ops vxlan_netdev_ops = {
2171         .ndo_init               = vxlan_init,
2172         .ndo_uninit             = vxlan_uninit,
2173         .ndo_open               = vxlan_open,
2174         .ndo_stop               = vxlan_stop,
2175         .ndo_start_xmit         = vxlan_xmit,
2176         .ndo_get_stats64        = ip_tunnel_get_stats64,
2177         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2178         .ndo_change_mtu         = vxlan_change_mtu,
2179         .ndo_validate_addr      = eth_validate_addr,
2180         .ndo_set_mac_address    = eth_mac_addr,
2181         .ndo_fdb_add            = vxlan_fdb_add,
2182         .ndo_fdb_del            = vxlan_fdb_delete,
2183         .ndo_fdb_dump           = vxlan_fdb_dump,
2184 };
2185
2186 /* Info for udev, that this is a virtual tunnel endpoint */
2187 static struct device_type vxlan_type = {
2188         .name = "vxlan",
2189 };
2190
2191 /* Calls the ndo_add_vxlan_port of the caller in order to
2192  * supply the listening VXLAN udp ports. Callers are expected
2193  * to implement the ndo_add_vxlan_port.
2194  */
2195 void vxlan_get_rx_port(struct net_device *dev)
2196 {
2197         struct vxlan_sock *vs;
2198         struct net *net = dev_net(dev);
2199         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2200         sa_family_t sa_family;
2201         __be16 port;
2202         unsigned int i;
2203
2204         spin_lock(&vn->sock_lock);
2205         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2206                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2207                         port = inet_sk(vs->sock->sk)->inet_sport;
2208                         sa_family = vs->sock->sk->sk_family;
2209                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2210                                                             port);
2211                 }
2212         }
2213         spin_unlock(&vn->sock_lock);
2214 }
2215 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2216
2217 /* Initialize the device structure. */
2218 static void vxlan_setup(struct net_device *dev)
2219 {
2220         struct vxlan_dev *vxlan = netdev_priv(dev);
2221         unsigned int h;
2222
2223         eth_hw_addr_random(dev);
2224         ether_setup(dev);
2225         if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2226                 dev->needed_headroom = ETH_HLEN + VXLAN6_HEADROOM;
2227         else
2228                 dev->needed_headroom = ETH_HLEN + VXLAN_HEADROOM;
2229
2230         dev->netdev_ops = &vxlan_netdev_ops;
2231         dev->destructor = free_netdev;
2232         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2233
2234         dev->tx_queue_len = 0;
2235         dev->features   |= NETIF_F_LLTX;
2236         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2237         dev->features   |= NETIF_F_RXCSUM;
2238         dev->features   |= NETIF_F_GSO_SOFTWARE;
2239
2240         dev->vlan_features = dev->features;
2241         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2242         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2243         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2244         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2245         dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
2246         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2247
2248         INIT_LIST_HEAD(&vxlan->next);
2249         spin_lock_init(&vxlan->hash_lock);
2250         INIT_WORK(&vxlan->igmp_join, vxlan_igmp_join);
2251         INIT_WORK(&vxlan->igmp_leave, vxlan_igmp_leave);
2252         INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
2253
2254         init_timer_deferrable(&vxlan->age_timer);
2255         vxlan->age_timer.function = vxlan_cleanup;
2256         vxlan->age_timer.data = (unsigned long) vxlan;
2257
2258         vxlan->dst_port = htons(vxlan_port);
2259
2260         vxlan->dev = dev;
2261
2262         for (h = 0; h < FDB_HASH_SIZE; ++h)
2263                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2264 }
2265
2266 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2267         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2268         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2269         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2270         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2271         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2272         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2273         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2274         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2275         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2276         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2277         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2278         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2279         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2280         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2281         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2282         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2283         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2284 };
2285
2286 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2287 {
2288         if (tb[IFLA_ADDRESS]) {
2289                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2290                         pr_debug("invalid link address (not ethernet)\n");
2291                         return -EINVAL;
2292                 }
2293
2294                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2295                         pr_debug("invalid all zero ethernet address\n");
2296                         return -EADDRNOTAVAIL;
2297                 }
2298         }
2299
2300         if (!data)
2301                 return -EINVAL;
2302
2303         if (data[IFLA_VXLAN_ID]) {
2304                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2305                 if (id >= VXLAN_VID_MASK)
2306                         return -ERANGE;
2307         }
2308
2309         if (data[IFLA_VXLAN_PORT_RANGE]) {
2310                 const struct ifla_vxlan_port_range *p
2311                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2312
2313                 if (ntohs(p->high) < ntohs(p->low)) {
2314                         pr_debug("port range %u .. %u not valid\n",
2315                                  ntohs(p->low), ntohs(p->high));
2316                         return -EINVAL;
2317                 }
2318         }
2319
2320         return 0;
2321 }
2322
2323 static void vxlan_get_drvinfo(struct net_device *netdev,
2324                               struct ethtool_drvinfo *drvinfo)
2325 {
2326         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2327         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2328 }
2329
2330 static const struct ethtool_ops vxlan_ethtool_ops = {
2331         .get_drvinfo    = vxlan_get_drvinfo,
2332         .get_link       = ethtool_op_get_link,
2333 };
2334
2335 static void vxlan_del_work(struct work_struct *work)
2336 {
2337         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2338
2339         sk_release_kernel(vs->sock->sk);
2340         kfree_rcu(vs, rcu);
2341 }
2342
2343 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2344                                         __be16 port, u32 flags)
2345 {
2346         struct socket *sock;
2347         struct udp_port_cfg udp_conf;
2348         int err;
2349
2350         memset(&udp_conf, 0, sizeof(udp_conf));
2351
2352         if (ipv6) {
2353                 udp_conf.family = AF_INET6;
2354                 udp_conf.use_udp6_tx_checksums =
2355                     !!(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2356                 udp_conf.use_udp6_rx_checksums =
2357                     !!(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2358         } else {
2359                 udp_conf.family = AF_INET;
2360                 udp_conf.local_ip.s_addr = INADDR_ANY;
2361                 udp_conf.use_udp_checksums =
2362                     !!(flags & VXLAN_F_UDP_CSUM);
2363         }
2364
2365         udp_conf.local_udp_port = port;
2366
2367         /* Open UDP socket */
2368         err = udp_sock_create(net, &udp_conf, &sock);
2369         if (err < 0)
2370                 return ERR_PTR(err);
2371
2372         /* Disable multicast loopback */
2373         inet_sk(sock->sk)->mc_loop = 0;
2374
2375         return sock;
2376 }
2377
2378 /* Create new listen socket if needed */
2379 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2380                                               vxlan_rcv_t *rcv, void *data,
2381                                               u32 flags)
2382 {
2383         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2384         struct vxlan_sock *vs;
2385         struct socket *sock;
2386         struct sock *sk;
2387         unsigned int h;
2388         bool ipv6 = !!(flags & VXLAN_F_IPV6);
2389
2390         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2391         if (!vs)
2392                 return ERR_PTR(-ENOMEM);
2393
2394         for (h = 0; h < VNI_HASH_SIZE; ++h)
2395                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2396
2397         INIT_WORK(&vs->del_work, vxlan_del_work);
2398
2399         sock = vxlan_create_sock(net, ipv6, port, flags);
2400         if (IS_ERR(sock)) {
2401                 kfree(vs);
2402                 return ERR_CAST(sock);
2403         }
2404
2405         vs->sock = sock;
2406         sk = sock->sk;
2407         atomic_set(&vs->refcnt, 1);
2408         vs->rcv = rcv;
2409         vs->data = data;
2410         rcu_assign_sk_user_data(vs->sock->sk, vs);
2411
2412         /* Initialize the vxlan udp offloads structure */
2413         vs->udp_offloads.port = port;
2414         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2415         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2416
2417         spin_lock(&vn->sock_lock);
2418         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2419         vxlan_notify_add_rx_port(vs);
2420         spin_unlock(&vn->sock_lock);
2421
2422         /* Mark socket as an encapsulation socket. */
2423         udp_sk(sk)->encap_type = 1;
2424         udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
2425 #if IS_ENABLED(CONFIG_IPV6)
2426         if (ipv6)
2427                 ipv6_stub->udpv6_encap_enable();
2428         else
2429 #endif
2430                 udp_encap_enable();
2431
2432         return vs;
2433 }
2434
2435 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2436                                   vxlan_rcv_t *rcv, void *data,
2437                                   bool no_share, u32 flags)
2438 {
2439         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2440         struct vxlan_sock *vs;
2441
2442         vs = vxlan_socket_create(net, port, rcv, data, flags);
2443         if (!IS_ERR(vs))
2444                 return vs;
2445
2446         if (no_share)   /* Return error if sharing is not allowed. */
2447                 return vs;
2448
2449         spin_lock(&vn->sock_lock);
2450         vs = vxlan_find_sock(net, port);
2451         if (vs) {
2452                 if (vs->rcv == rcv)
2453                         atomic_inc(&vs->refcnt);
2454                 else
2455                         vs = ERR_PTR(-EBUSY);
2456         }
2457         spin_unlock(&vn->sock_lock);
2458
2459         if (!vs)
2460                 vs = ERR_PTR(-EINVAL);
2461
2462         return vs;
2463 }
2464 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2465
2466 /* Scheduled at device creation to bind to a socket */
2467 static void vxlan_sock_work(struct work_struct *work)
2468 {
2469         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, sock_work);
2470         struct net *net = vxlan->net;
2471         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2472         __be16 port = vxlan->dst_port;
2473         struct vxlan_sock *nvs;
2474
2475         nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags);
2476         spin_lock(&vn->sock_lock);
2477         if (!IS_ERR(nvs))
2478                 vxlan_vs_add_dev(nvs, vxlan);
2479         spin_unlock(&vn->sock_lock);
2480
2481         dev_put(vxlan->dev);
2482 }
2483
2484 static int vxlan_newlink(struct net *net, struct net_device *dev,
2485                          struct nlattr *tb[], struct nlattr *data[])
2486 {
2487         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2488         struct vxlan_dev *vxlan = netdev_priv(dev);
2489         struct vxlan_rdst *dst = &vxlan->default_dst;
2490         __u32 vni;
2491         int err;
2492         bool use_ipv6 = false;
2493
2494         if (!data[IFLA_VXLAN_ID])
2495                 return -EINVAL;
2496
2497         vxlan->net = dev_net(dev);
2498
2499         vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2500         dst->remote_vni = vni;
2501
2502         /* Unless IPv6 is explicitly requested, assume IPv4 */
2503         dst->remote_ip.sa.sa_family = AF_INET;
2504         if (data[IFLA_VXLAN_GROUP]) {
2505                 dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
2506         } else if (data[IFLA_VXLAN_GROUP6]) {
2507                 if (!IS_ENABLED(CONFIG_IPV6))
2508                         return -EPFNOSUPPORT;
2509
2510                 nla_memcpy(&dst->remote_ip.sin6.sin6_addr, data[IFLA_VXLAN_GROUP6],
2511                            sizeof(struct in6_addr));
2512                 dst->remote_ip.sa.sa_family = AF_INET6;
2513                 use_ipv6 = true;
2514         }
2515
2516         if (data[IFLA_VXLAN_LOCAL]) {
2517                 vxlan->saddr.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
2518                 vxlan->saddr.sa.sa_family = AF_INET;
2519         } else if (data[IFLA_VXLAN_LOCAL6]) {
2520                 if (!IS_ENABLED(CONFIG_IPV6))
2521                         return -EPFNOSUPPORT;
2522
2523                 /* TODO: respect scope id */
2524                 nla_memcpy(&vxlan->saddr.sin6.sin6_addr, data[IFLA_VXLAN_LOCAL6],
2525                            sizeof(struct in6_addr));
2526                 vxlan->saddr.sa.sa_family = AF_INET6;
2527                 use_ipv6 = true;
2528         }
2529
2530         if (data[IFLA_VXLAN_LINK] &&
2531             (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2532                 struct net_device *lowerdev
2533                          = __dev_get_by_index(net, dst->remote_ifindex);
2534
2535                 if (!lowerdev) {
2536                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2537                         return -ENODEV;
2538                 }
2539
2540 #if IS_ENABLED(CONFIG_IPV6)
2541                 if (use_ipv6) {
2542                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2543                         if (idev && idev->cnf.disable_ipv6) {
2544                                 pr_info("IPv6 is disabled via sysctl\n");
2545                                 return -EPERM;
2546                         }
2547                         vxlan->flags |= VXLAN_F_IPV6;
2548                 }
2549 #endif
2550
2551                 if (!tb[IFLA_MTU])
2552                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2553
2554                 dev->needed_headroom = lowerdev->hard_header_len +
2555                                        (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2556         } else if (use_ipv6)
2557                 vxlan->flags |= VXLAN_F_IPV6;
2558
2559         if (data[IFLA_VXLAN_TOS])
2560                 vxlan->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2561
2562         if (data[IFLA_VXLAN_TTL])
2563                 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2564
2565         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2566                 vxlan->flags |= VXLAN_F_LEARN;
2567
2568         if (data[IFLA_VXLAN_AGEING])
2569                 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2570         else
2571                 vxlan->age_interval = FDB_AGE_DEFAULT;
2572
2573         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2574                 vxlan->flags |= VXLAN_F_PROXY;
2575
2576         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2577                 vxlan->flags |= VXLAN_F_RSC;
2578
2579         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2580                 vxlan->flags |= VXLAN_F_L2MISS;
2581
2582         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2583                 vxlan->flags |= VXLAN_F_L3MISS;
2584
2585         if (data[IFLA_VXLAN_LIMIT])
2586                 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2587
2588         if (data[IFLA_VXLAN_PORT_RANGE]) {
2589                 const struct ifla_vxlan_port_range *p
2590                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2591                 vxlan->port_min = ntohs(p->low);
2592                 vxlan->port_max = ntohs(p->high);
2593         }
2594
2595         if (data[IFLA_VXLAN_PORT])
2596                 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2597
2598         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2599                 vxlan->flags |= VXLAN_F_UDP_CSUM;
2600
2601         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2602             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2603                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2604
2605         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2606             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2607                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2608
2609         if (vxlan_find_vni(net, vni, vxlan->dst_port)) {
2610                 pr_info("duplicate VNI %u\n", vni);
2611                 return -EEXIST;
2612         }
2613
2614         dev->ethtool_ops = &vxlan_ethtool_ops;
2615
2616         /* create an fdb entry for a valid default destination */
2617         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2618                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2619                                        &vxlan->default_dst.remote_ip,
2620                                        NUD_REACHABLE|NUD_PERMANENT,
2621                                        NLM_F_EXCL|NLM_F_CREATE,
2622                                        vxlan->dst_port,
2623                                        vxlan->default_dst.remote_vni,
2624                                        vxlan->default_dst.remote_ifindex,
2625                                        NTF_SELF);
2626                 if (err)
2627                         return err;
2628         }
2629
2630         err = register_netdevice(dev);
2631         if (err) {
2632                 vxlan_fdb_delete_default(vxlan);
2633                 return err;
2634         }
2635
2636         list_add(&vxlan->next, &vn->vxlan_list);
2637
2638         return 0;
2639 }
2640
2641 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2642 {
2643         struct vxlan_dev *vxlan = netdev_priv(dev);
2644         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2645
2646         spin_lock(&vn->sock_lock);
2647         if (!hlist_unhashed(&vxlan->hlist))
2648                 hlist_del_rcu(&vxlan->hlist);
2649         spin_unlock(&vn->sock_lock);
2650
2651         list_del(&vxlan->next);
2652         unregister_netdevice_queue(dev, head);
2653 }
2654
2655 static size_t vxlan_get_size(const struct net_device *dev)
2656 {
2657
2658         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2659                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2660                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2661                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2662                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2663                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2664                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2665                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2666                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2667                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2668                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2669                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2670                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2671                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2672                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2673                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2674                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2675                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2676                 0;
2677 }
2678
2679 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2680 {
2681         const struct vxlan_dev *vxlan = netdev_priv(dev);
2682         const struct vxlan_rdst *dst = &vxlan->default_dst;
2683         struct ifla_vxlan_port_range ports = {
2684                 .low =  htons(vxlan->port_min),
2685                 .high = htons(vxlan->port_max),
2686         };
2687
2688         if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2689                 goto nla_put_failure;
2690
2691         if (!vxlan_addr_any(&dst->remote_ip)) {
2692                 if (dst->remote_ip.sa.sa_family == AF_INET) {
2693                         if (nla_put_be32(skb, IFLA_VXLAN_GROUP,
2694                                          dst->remote_ip.sin.sin_addr.s_addr))
2695                                 goto nla_put_failure;
2696 #if IS_ENABLED(CONFIG_IPV6)
2697                 } else {
2698                         if (nla_put(skb, IFLA_VXLAN_GROUP6, sizeof(struct in6_addr),
2699                                     &dst->remote_ip.sin6.sin6_addr))
2700                                 goto nla_put_failure;
2701 #endif
2702                 }
2703         }
2704
2705         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2706                 goto nla_put_failure;
2707
2708         if (!vxlan_addr_any(&vxlan->saddr)) {
2709                 if (vxlan->saddr.sa.sa_family == AF_INET) {
2710                         if (nla_put_be32(skb, IFLA_VXLAN_LOCAL,
2711                                          vxlan->saddr.sin.sin_addr.s_addr))
2712                                 goto nla_put_failure;
2713 #if IS_ENABLED(CONFIG_IPV6)
2714                 } else {
2715                         if (nla_put(skb, IFLA_VXLAN_LOCAL6, sizeof(struct in6_addr),
2716                                     &vxlan->saddr.sin6.sin6_addr))
2717                                 goto nla_put_failure;
2718 #endif
2719                 }
2720         }
2721
2722         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2723             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2724             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2725                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
2726             nla_put_u8(skb, IFLA_VXLAN_PROXY,
2727                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
2728             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2729             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2730                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2731             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2732                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2733             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2734             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2735             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port) ||
2736             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2737                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2738             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2739                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2740             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2741                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)))
2742                 goto nla_put_failure;
2743
2744         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2745                 goto nla_put_failure;
2746
2747         return 0;
2748
2749 nla_put_failure:
2750         return -EMSGSIZE;
2751 }
2752
2753 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2754         .kind           = "vxlan",
2755         .maxtype        = IFLA_VXLAN_MAX,
2756         .policy         = vxlan_policy,
2757         .priv_size      = sizeof(struct vxlan_dev),
2758         .setup          = vxlan_setup,
2759         .validate       = vxlan_validate,
2760         .newlink        = vxlan_newlink,
2761         .dellink        = vxlan_dellink,
2762         .get_size       = vxlan_get_size,
2763         .fill_info      = vxlan_fill_info,
2764 };
2765
2766 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
2767                                              struct net_device *dev)
2768 {
2769         struct vxlan_dev *vxlan, *next;
2770         LIST_HEAD(list_kill);
2771
2772         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2773                 struct vxlan_rdst *dst = &vxlan->default_dst;
2774
2775                 /* In case we created vxlan device with carrier
2776                  * and we loose the carrier due to module unload
2777                  * we also need to remove vxlan device. In other
2778                  * cases, it's not necessary and remote_ifindex
2779                  * is 0 here, so no matches.
2780                  */
2781                 if (dst->remote_ifindex == dev->ifindex)
2782                         vxlan_dellink(vxlan->dev, &list_kill);
2783         }
2784
2785         unregister_netdevice_many(&list_kill);
2786 }
2787
2788 static int vxlan_lowerdev_event(struct notifier_block *unused,
2789                                 unsigned long event, void *ptr)
2790 {
2791         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2792         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2793
2794         if (event == NETDEV_UNREGISTER)
2795                 vxlan_handle_lowerdev_unregister(vn, dev);
2796
2797         return NOTIFY_DONE;
2798 }
2799
2800 static struct notifier_block vxlan_notifier_block __read_mostly = {
2801         .notifier_call = vxlan_lowerdev_event,
2802 };
2803
2804 static __net_init int vxlan_init_net(struct net *net)
2805 {
2806         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2807         unsigned int h;
2808
2809         INIT_LIST_HEAD(&vn->vxlan_list);
2810         spin_lock_init(&vn->sock_lock);
2811
2812         for (h = 0; h < PORT_HASH_SIZE; ++h)
2813                 INIT_HLIST_HEAD(&vn->sock_list[h]);
2814
2815         return 0;
2816 }
2817
2818 static void __net_exit vxlan_exit_net(struct net *net)
2819 {
2820         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2821         struct vxlan_dev *vxlan, *next;
2822         struct net_device *dev, *aux;
2823         LIST_HEAD(list);
2824
2825         rtnl_lock();
2826         for_each_netdev_safe(net, dev, aux)
2827                 if (dev->rtnl_link_ops == &vxlan_link_ops)
2828                         unregister_netdevice_queue(dev, &list);
2829
2830         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2831                 /* If vxlan->dev is in the same netns, it has already been added
2832                  * to the list by the previous loop.
2833                  */
2834                 if (!net_eq(dev_net(vxlan->dev), net))
2835                         unregister_netdevice_queue(dev, &list);
2836         }
2837
2838         unregister_netdevice_many(&list);
2839         rtnl_unlock();
2840 }
2841
2842 static struct pernet_operations vxlan_net_ops = {
2843         .init = vxlan_init_net,
2844         .exit = vxlan_exit_net,
2845         .id   = &vxlan_net_id,
2846         .size = sizeof(struct vxlan_net),
2847 };
2848
2849 static int __init vxlan_init_module(void)
2850 {
2851         int rc;
2852
2853         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
2854         if (!vxlan_wq)
2855                 return -ENOMEM;
2856
2857         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
2858
2859         rc = register_pernet_subsys(&vxlan_net_ops);
2860         if (rc)
2861                 goto out1;
2862
2863         rc = register_netdevice_notifier(&vxlan_notifier_block);
2864         if (rc)
2865                 goto out2;
2866
2867         rc = rtnl_link_register(&vxlan_link_ops);
2868         if (rc)
2869                 goto out3;
2870
2871         return 0;
2872 out3:
2873         unregister_netdevice_notifier(&vxlan_notifier_block);
2874 out2:
2875         unregister_pernet_subsys(&vxlan_net_ops);
2876 out1:
2877         destroy_workqueue(vxlan_wq);
2878         return rc;
2879 }
2880 late_initcall(vxlan_init_module);
2881
2882 static void __exit vxlan_cleanup_module(void)
2883 {
2884         rtnl_link_unregister(&vxlan_link_ops);
2885         unregister_netdevice_notifier(&vxlan_notifier_block);
2886         destroy_workqueue(vxlan_wq);
2887         unregister_pernet_subsys(&vxlan_net_ops);
2888         /* rcu_barrier() is called by netns */
2889 }
2890 module_exit(vxlan_cleanup_module);
2891
2892 MODULE_LICENSE("GPL");
2893 MODULE_VERSION(VXLAN_VERSION);
2894 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
2895 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
2896 MODULE_ALIAS_RTNL_LINK("vxlan");