drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/types.h>
  15 #include <linux/module.h>
  16 #include <linux/errno.h>
  17 #include <linux/slab.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/rculist.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/in.h>
  22 #include <linux/ip.h>
  23 #include <linux/udp.h>
  24 #include <linux/igmp.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/hash.h>
  29 #include <linux/ethtool.h>
  30 #include <net/arp.h>
  31 #include <net/ndisc.h>
  32 #include <net/ip.h>
  33 #include <net/ip_tunnels.h>
  34 #include <net/icmp.h>
  35 #include <net/udp.h>
  36 #include <net/udp_tunnel.h>
  37 #include <net/rtnetlink.h>
  38 #include <net/route.h>
  39 #include <net/dsfield.h>
  40 #include <net/inet_ecn.h>
  41 #include <net/net_namespace.h>
  42 #include <net/netns/generic.h>
  43 #include <net/vxlan.h>
  44 #include <net/protocol.h>
  45 #include <net/udp_tunnel.h>
  46 #if IS_ENABLED(CONFIG_IPV6)
  47 #include <net/ipv6.h>
  48 #include <net/addrconf.h>
  49 #include <net/ip6_tunnel.h>
  50 #include <net/ip6_checksum.h>
  51 #endif
  52
  53 #define VXLAN_VERSION   "0.1"
  54
  55 #define PORT_HASH_BITS  8
  56 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  57 #define VNI_HASH_BITS   10
  58 #define VNI_HASH_SIZE   (1<<VNI_HASH_BITS)
  59 #define FDB_HASH_BITS   8
  60 #define FDB_HASH_SIZE   (1<<FDB_HASH_BITS)
  61 #define FDB_AGE_DEFAULT 300 /* 5 min */
  62 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  63
  64 #define VXLAN_N_VID     (1u << 24)
  65 #define VXLAN_VID_MASK  (VXLAN_N_VID - 1)
  66 #define VXLAN_HLEN (sizeof(struct udphdr) + sizeof(struct vxlanhdr))
  67
  68 #define VXLAN_FLAGS 0x08000000  /* struct vxlanhdr.vx_flags required value. */
  69
  70 /* UDP port for VXLAN traffic.
  71  * The IANA assigned port is 4789, but the Linux default is 8472
  72  * for compatibility with early adopters.
  73  */
  74 static unsigned short vxlan_port __read_mostly = 8472;
  75 module_param_named(udp_port, vxlan_port, ushort, 0444);
  76 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  77
  78 static bool log_ecn_error = true;
  79 module_param(log_ecn_error, bool, 0644);
  80 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  81
  82 static int vxlan_net_id;
  83
  84 static const u8 all_zeros_mac[ETH_ALEN];
  85
  86 /* per-network namespace private data for this module */
  87 struct vxlan_net {
  88         struct list_head  vxlan_list;
  89         struct hlist_head sock_list[PORT_HASH_SIZE];
  90         spinlock_t        sock_lock;
  91 };
  92
  93 union vxlan_addr {
  94         struct sockaddr_in sin;
  95         struct sockaddr_in6 sin6;
  96         struct sockaddr sa;
  97 };
  98
  99 struct vxlan_rdst {
 100         union vxlan_addr         remote_ip;
 101         __be16                   remote_port;
 102         u32                      remote_vni;
 103         u32                      remote_ifindex;
 104         struct list_head         list;
 105         struct rcu_head          rcu;
 106 };
 107
 108 /* Forwarding table entry */
 109 struct vxlan_fdb {
 110         struct hlist_node hlist;        /* linked list of entries */
 111         struct rcu_head   rcu;
 112         unsigned long     updated;      /* jiffies */
 113         unsigned long     used;
 114         struct list_head  remotes;
 115         u16               state;        /* see ndm_state */
 116         u8                flags;        /* see ndm_flags */
 117         u8                eth_addr[ETH_ALEN];
 118 };
 119
 120 /* Pseudo network device */
 121 struct vxlan_dev {
 122         struct hlist_node hlist;        /* vni hash table */
 123         struct list_head  next;         /* vxlan's per namespace list */
 124         struct vxlan_sock *vn_sock;     /* listening socket */
 125         struct net_device *dev;
 126         struct net        *net;         /* netns for packet i/o */
 127         struct vxlan_rdst default_dst;  /* default destination */
 128         union vxlan_addr  saddr;        /* source address */
 129         __be16            dst_port;
 130         __u16             port_min;     /* source port range */
 131         __u16             port_max;
 132         __u8              tos;          /* TOS override */
 133         __u8              ttl;
 134         u32               flags;        /* VXLAN_F_* in vxlan.h */
 135
 136         struct work_struct sock_work;
 137         struct work_struct igmp_join;
 138         struct work_struct igmp_leave;
 139
 140         unsigned long     age_interval;
 141         struct timer_list age_timer;
 142         spinlock_t        hash_lock;
 143         unsigned int      addrcnt;
 144         unsigned int      addrmax;
 145
 146         struct hlist_head fdb_head[FDB_HASH_SIZE];
 147 };
 148
 149 /* salt for hash table */
 150 static u32 vxlan_salt __read_mostly;
 151 static struct workqueue_struct *vxlan_wq;
 152
 153 static void vxlan_sock_work(struct work_struct *work);
 154
 155 #if IS_ENABLED(CONFIG_IPV6)
 156 static inline
 157 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 158 {
 159        if (a->sa.sa_family != b->sa.sa_family)
 160                return false;
 161        if (a->sa.sa_family == AF_INET6)
 162                return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 163        else
 164                return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 165 }
 166
 167 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 168 {
 169        if (ipa->sa.sa_family == AF_INET6)
 170                return ipv6_addr_any(&ipa->sin6.sin6_addr);
 171        else
 172                return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 173 }
 174
 175 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 176 {
 177        if (ipa->sa.sa_family == AF_INET6)
 178                return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 179        else
 180                return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 181 }
 182
 183 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 184 {
 185        if (nla_len(nla) >= sizeof(struct in6_addr)) {
 186                nla_memcpy(&ip->sin6.sin6_addr, nla, sizeof(struct in6_addr));
 187                ip->sa.sa_family = AF_INET6;
 188                return 0;
 189        } else if (nla_len(nla) >= sizeof(__be32)) {
 190                ip->sin.sin_addr.s_addr = nla_get_be32(nla);
 191                ip->sa.sa_family = AF_INET;
 192                return 0;
 193        } else {
 194                return -EAFNOSUPPORT;
 195        }
 196 }
 197
 198 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 199                              const union vxlan_addr *ip)
 200 {
 201        if (ip->sa.sa_family == AF_INET6)
 202                return nla_put(skb, attr, sizeof(struct in6_addr), &ip->sin6.sin6_addr);
 203        else
 204                return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
 205 }
 206
 207 #else /* !CONFIG_IPV6 */
 208
 209 static inline
 210 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 211 {
 212        return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 213 }
 214
 215 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 216 {
 217        return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 218 }
 219
 220 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 221 {
 222        return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 223 }
 224
 225 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 226 {
 227        if (nla_len(nla) >= sizeof(struct in6_addr)) {
 228                return -EAFNOSUPPORT;
 229        } else if (nla_len(nla) >= sizeof(__be32)) {
 230                ip->sin.sin_addr.s_addr = nla_get_be32(nla);
 231                ip->sa.sa_family = AF_INET;
 232                return 0;
 233        } else {
 234                return -EAFNOSUPPORT;
 235        }
 236 }
 237
 238 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 239                              const union vxlan_addr *ip)
 240 {
 241        return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
 242 }
 243 #endif
 244
 245 /* Virtual Network hash table head */
 246 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
 247 {
 248         return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
 249 }
 250
 251 /* Socket hash table head */
 252 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 253 {
 254         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 255
 256         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 257 }
 258
 259 /* First remote destination for a forwarding entry.
 260  * Guaranteed to be non-NULL because remotes are never deleted.
 261  */
 262 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 263 {
 264         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 265 }
 266
 267 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 268 {
 269         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 270 }
 271
 272 /* Find VXLAN socket based on network namespace, address family and UDP port */
 273 static struct vxlan_sock *vxlan_find_sock(struct net *net,
 274                                           sa_family_t family, __be16 port)
 275 {
 276         struct vxlan_sock *vs;
 277
 278         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 279                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 280                     inet_sk(vs->sock->sk)->sk.sk_family == family)
 281                         return vs;
 282         }
 283         return NULL;
 284 }
 285
 286 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
 287 {
 288         struct vxlan_dev *vxlan;
 289
 290         hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
 291                 if (vxlan->default_dst.remote_vni == id)
 292                         return vxlan;
 293         }
 294
 295         return NULL;
 296 }
 297
 298 /* Look up VNI in a per net namespace table */
 299 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id,
 300                                         sa_family_t family, __be16 port)
 301 {
 302         struct vxlan_sock *vs;
 303
 304         vs = vxlan_find_sock(net, family, port);
 305         if (!vs)
 306                 return NULL;
 307
 308         return vxlan_vs_find_vni(vs, id);
 309 }
 310
 311 /* Fill in neighbour message in skbuff. */
 312 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 313                           const struct vxlan_fdb *fdb,
 314                           u32 portid, u32 seq, int type, unsigned int flags,
 315                           const struct vxlan_rdst *rdst)
 316 {
 317         unsigned long now = jiffies;
 318         struct nda_cacheinfo ci;
 319         struct nlmsghdr *nlh;
 320         struct ndmsg *ndm;
 321         bool send_ip, send_eth;
 322
 323         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 324         if (nlh == NULL)
 325                 return -EMSGSIZE;
 326
 327         ndm = nlmsg_data(nlh);
 328         memset(ndm, 0, sizeof(*ndm));
 329
 330         send_eth = send_ip = true;
 331
 332         if (type == RTM_GETNEIGH) {
 333                 ndm->ndm_family = AF_INET;
 334                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 335                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 336         } else
 337                 ndm->ndm_family = AF_BRIDGE;
 338         ndm->ndm_state = fdb->state;
 339         ndm->ndm_ifindex = vxlan->dev->ifindex;
 340         ndm->ndm_flags = fdb->flags;
 341         ndm->ndm_type = RTN_UNICAST;
 342
 343         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 344                 goto nla_put_failure;
 345
 346         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 347                 goto nla_put_failure;
 348
 349         if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
 350             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 351                 goto nla_put_failure;
 352         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 353             nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
 354                 goto nla_put_failure;
 355         if (rdst->remote_ifindex &&
 356             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 357                 goto nla_put_failure;
 358
 359         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 360         ci.ndm_confirmed = 0;
 361         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 362         ci.ndm_refcnt    = 0;
 363
 364         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 365                 goto nla_put_failure;
 366
 367         return nlmsg_end(skb, nlh);
 368
 369 nla_put_failure:
 370         nlmsg_cancel(skb, nlh);
 371         return -EMSGSIZE;
 372 }
 373
 374 static inline size_t vxlan_nlmsg_size(void)
 375 {
 376         return NLMSG_ALIGN(sizeof(struct ndmsg))
 377                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 378                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 379                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 380                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 381                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 382                 + nla_total_size(sizeof(struct nda_cacheinfo));
 383 }
 384
 385 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 386                              struct vxlan_rdst *rd, int type)
 387 {
 388         struct net *net = dev_net(vxlan->dev);
 389         struct sk_buff *skb;
 390         int err = -ENOBUFS;
 391
 392         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 393         if (skb == NULL)
 394                 goto errout;
 395
 396         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 397         if (err < 0) {
 398                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 399                 WARN_ON(err == -EMSGSIZE);
 400                 kfree_skb(skb);
 401                 goto errout;
 402         }
 403
 404         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 405         return;
 406 errout:
 407         if (err < 0)
 408                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 409 }
 410
 411 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 412 {
 413         struct vxlan_dev *vxlan = netdev_priv(dev);
 414         struct vxlan_fdb f = {
 415                 .state = NUD_STALE,
 416         };
 417         struct vxlan_rdst remote = {
 418                 .remote_ip = *ipa, /* goes to NDA_DST */
 419                 .remote_vni = VXLAN_N_VID,
 420         };
 421
 422         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 423 }
 424
 425 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 426 {
 427         struct vxlan_fdb f = {
 428                 .state = NUD_STALE,
 429         };
 430         struct vxlan_rdst remote = { };
 431
 432         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 433
 434         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 435 }
 436
 437 /* Hash Ethernet address */
 438 static u32 eth_hash(const unsigned char *addr)
 439 {
 440         u64 value = get_unaligned((u64 *)addr);
 441
 442         /* only want 6 bytes */
 443 #ifdef __BIG_ENDIAN
 444         value >>= 16;
 445 #else
 446         value <<= 16;
 447 #endif
 448         return hash_64(value, FDB_HASH_BITS);
 449 }
 450
 451 /* Hash chain to use given mac address */
 452 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 453                                                 const u8 *mac)
 454 {
 455         return &vxlan->fdb_head[eth_hash(mac)];
 456 }
 457
 458 /* Look up Ethernet address in forwarding table */
 459 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 460                                         const u8 *mac)
 461 {
 462         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 463         struct vxlan_fdb *f;
 464
 465         hlist_for_each_entry_rcu(f, head, hlist) {
 466                 if (ether_addr_equal(mac, f->eth_addr))
 467                         return f;
 468         }
 469
 470         return NULL;
 471 }
 472
 473 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 474                                         const u8 *mac)
 475 {
 476         struct vxlan_fdb *f;
 477
 478         f = __vxlan_find_mac(vxlan, mac);
 479         if (f)
 480                 f->used = jiffies;
 481
 482         return f;
 483 }
 484
 485 /* caller should hold vxlan->hash_lock */
 486 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 487                                               union vxlan_addr *ip, __be16 port,
 488                                               __u32 vni, __u32 ifindex)
 489 {
 490         struct vxlan_rdst *rd;
 491
 492         list_for_each_entry(rd, &f->remotes, list) {
 493                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 494                     rd->remote_port == port &&
 495                     rd->remote_vni == vni &&
 496                     rd->remote_ifindex == ifindex)
 497                         return rd;
 498         }
 499
 500         return NULL;
 501 }
 502
 503 /* Replace destination of unicast mac */
 504 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 505                              union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
 506 {
 507         struct vxlan_rdst *rd;
 508
 509         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 510         if (rd)
 511                 return 0;
 512
 513         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 514         if (!rd)
 515                 return 0;
 516         rd->remote_ip = *ip;
 517         rd->remote_port = port;
 518         rd->remote_vni = vni;
 519         rd->remote_ifindex = ifindex;
 520         return 1;
 521 }
 522
 523 /* Add/update destinations for multicast */
 524 static int vxlan_fdb_append(struct vxlan_fdb *f,
 525                             union vxlan_addr *ip, __be16 port, __u32 vni,
 526                             __u32 ifindex, struct vxlan_rdst **rdp)
 527 {
 528         struct vxlan_rdst *rd;
 529
 530         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 531         if (rd)
 532                 return 0;
 533
 534         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 535         if (rd == NULL)
 536                 return -ENOBUFS;
 537         rd->remote_ip = *ip;
 538         rd->remote_port = port;
 539         rd->remote_vni = vni;
 540         rd->remote_ifindex = ifindex;
 541
 542         list_add_tail_rcu(&rd->list, &f->remotes);
 543
 544         *rdp = rd;
 545         return 1;
 546 }
 547
 548 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head, struct sk_buff *skb)
 549 {
 550         struct sk_buff *p, **pp = NULL;
 551         struct vxlanhdr *vh, *vh2;
 552         struct ethhdr *eh, *eh2;
 553         unsigned int hlen, off_vx, off_eth;
 554         const struct packet_offload *ptype;
 555         __be16 type;
 556         int flush = 1;
 557
 558         off_vx = skb_gro_offset(skb);
 559         hlen = off_vx + sizeof(*vh);
 560         vh   = skb_gro_header_fast(skb, off_vx);
 561         if (skb_gro_header_hard(skb, hlen)) {
 562                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 563                 if (unlikely(!vh))
 564                         goto out;
 565         }
 566         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 567         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 568
 569         off_eth = skb_gro_offset(skb);
 570         hlen = off_eth + sizeof(*eh);
 571         eh   = skb_gro_header_fast(skb, off_eth);
 572         if (skb_gro_header_hard(skb, hlen)) {
 573                 eh = skb_gro_header_slow(skb, hlen, off_eth);
 574                 if (unlikely(!eh))
 575                         goto out;
 576         }
 577
 578         flush = 0;
 579
 580         for (p = *head; p; p = p->next) {
 581                 if (!NAPI_GRO_CB(p)->same_flow)
 582                         continue;
 583
 584                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 585                 eh2 = (struct ethhdr   *)(p->data + off_eth);
 586                 if (vh->vx_vni != vh2->vx_vni || compare_ether_header(eh, eh2)) {
 587                         NAPI_GRO_CB(p)->same_flow = 0;
 588                         continue;
 589                 }
 590         }
 591
 592         type = eh->h_proto;
 593
 594         rcu_read_lock();
 595         ptype = gro_find_receive_by_type(type);
 596         if (ptype == NULL) {
 597                 flush = 1;
 598                 goto out_unlock;
 599         }
 600
 601         skb_gro_pull(skb, sizeof(*eh)); /* pull inner eth header */
 602         skb_gro_postpull_rcsum(skb, eh, sizeof(*eh));
 603         pp = ptype->callbacks.gro_receive(head, skb);
 604
 605 out_unlock:
 606         rcu_read_unlock();
 607 out:
 608         NAPI_GRO_CB(skb)->flush |= flush;
 609
 610         return pp;
 611 }
 612
 613 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff)
 614 {
 615         struct ethhdr *eh;
 616         struct packet_offload *ptype;
 617         __be16 type;
 618         int vxlan_len  = sizeof(struct vxlanhdr) + sizeof(struct ethhdr);
 619         int err = -ENOSYS;
 620
 621         udp_tunnel_gro_complete(skb, nhoff);
 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         vxlan_notify_del_rx_port(vs);
1066         spin_unlock(&vn->sock_lock);
1067
1068         queue_work(vxlan_wq, &vs->del_work);
1069 }
1070 EXPORT_SYMBOL_GPL(vxlan_sock_release);
1071
1072 /* Callback to update multicast group membership when first VNI on
1073  * multicast asddress is brought up
1074  * Done as workqueue because ip_mc_join_group acquires RTNL.
1075  */
1076 static void vxlan_igmp_join(struct work_struct *work)
1077 {
1078         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_join);
1079         struct vxlan_sock *vs = vxlan->vn_sock;
1080         struct sock *sk = vs->sock->sk;
1081         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1082         int ifindex = vxlan->default_dst.remote_ifindex;
1083
1084         lock_sock(sk);
1085         if (ip->sa.sa_family == AF_INET) {
1086                 struct ip_mreqn mreq = {
1087                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1088                         .imr_ifindex            = ifindex,
1089                 };
1090
1091                 ip_mc_join_group(sk, &mreq);
1092 #if IS_ENABLED(CONFIG_IPV6)
1093         } else {
1094                 ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1095                                              &ip->sin6.sin6_addr);
1096 #endif
1097         }
1098         release_sock(sk);
1099
1100         vxlan_sock_release(vs);
1101         dev_put(vxlan->dev);
1102 }
1103
1104 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1105 static void vxlan_igmp_leave(struct work_struct *work)
1106 {
1107         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_leave);
1108         struct vxlan_sock *vs = vxlan->vn_sock;
1109         struct sock *sk = vs->sock->sk;
1110         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1111         int ifindex = vxlan->default_dst.remote_ifindex;
1112
1113         lock_sock(sk);
1114         if (ip->sa.sa_family == AF_INET) {
1115                 struct ip_mreqn mreq = {
1116                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1117                         .imr_ifindex            = ifindex,
1118                 };
1119
1120                 ip_mc_leave_group(sk, &mreq);
1121 #if IS_ENABLED(CONFIG_IPV6)
1122         } else {
1123                 ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1124                                              &ip->sin6.sin6_addr);
1125 #endif
1126         }
1127
1128         release_sock(sk);
1129
1130         vxlan_sock_release(vs);
1131         dev_put(vxlan->dev);
1132 }
1133
1134 /* Callback from net/ipv4/udp.c to receive packets */
1135 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1136 {
1137         struct vxlan_sock *vs;
1138         struct vxlanhdr *vxh;
1139
1140         /* Need Vxlan and inner Ethernet header to be present */
1141         if (!pskb_may_pull(skb, VXLAN_HLEN))
1142                 goto error;
1143
1144         /* Return packets with reserved bits set */
1145         vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1146         if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
1147             (vxh->vx_vni & htonl(0xff))) {
1148                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1149                            ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1150                 goto error;
1151         }
1152
1153         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1154                 goto drop;
1155
1156         vs = rcu_dereference_sk_user_data(sk);
1157         if (!vs)
1158                 goto drop;
1159
1160         vs->rcv(vs, skb, vxh->vx_vni);
1161         return 0;
1162
1163 drop:
1164         /* Consume bad packet */
1165         kfree_skb(skb);
1166         return 0;
1167
1168 error:
1169         /* Return non vxlan pkt */
1170         return 1;
1171 }
1172
1173 static void vxlan_rcv(struct vxlan_sock *vs,
1174                       struct sk_buff *skb, __be32 vx_vni)
1175 {
1176         struct iphdr *oip = NULL;
1177         struct ipv6hdr *oip6 = NULL;
1178         struct vxlan_dev *vxlan;
1179         struct pcpu_sw_netstats *stats;
1180         union vxlan_addr saddr;
1181         __u32 vni;
1182         int err = 0;
1183         union vxlan_addr *remote_ip;
1184
1185         vni = ntohl(vx_vni) >> 8;
1186         /* Is this VNI defined? */
1187         vxlan = vxlan_vs_find_vni(vs, vni);
1188         if (!vxlan)
1189                 goto drop;
1190
1191         remote_ip = &vxlan->default_dst.remote_ip;
1192         skb_reset_mac_header(skb);
1193         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1194         skb->protocol = eth_type_trans(skb, vxlan->dev);
1195         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1196
1197         /* Ignore packet loops (and multicast echo) */
1198         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1199                 goto drop;
1200
1201         /* Re-examine inner Ethernet packet */
1202         if (remote_ip->sa.sa_family == AF_INET) {
1203                 oip = ip_hdr(skb);
1204                 saddr.sin.sin_addr.s_addr = oip->saddr;
1205                 saddr.sa.sa_family = AF_INET;
1206 #if IS_ENABLED(CONFIG_IPV6)
1207         } else {
1208                 oip6 = ipv6_hdr(skb);
1209                 saddr.sin6.sin6_addr = oip6->saddr;
1210                 saddr.sa.sa_family = AF_INET6;
1211 #endif
1212         }
1213
1214         if ((vxlan->flags & VXLAN_F_LEARN) &&
1215             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1216                 goto drop;
1217
1218         skb_reset_network_header(skb);
1219
1220         if (oip6)
1221                 err = IP6_ECN_decapsulate(oip6, skb);
1222         if (oip)
1223                 err = IP_ECN_decapsulate(oip, skb);
1224
1225         if (unlikely(err)) {
1226                 if (log_ecn_error) {
1227                         if (oip6)
1228                                 net_info_ratelimited("non-ECT from %pI6\n",
1229                                                      &oip6->saddr);
1230                         if (oip)
1231                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1232                                                      &oip->saddr, oip->tos);
1233                 }
1234                 if (err > 1) {
1235                         ++vxlan->dev->stats.rx_frame_errors;
1236                         ++vxlan->dev->stats.rx_errors;
1237                         goto drop;
1238                 }
1239         }
1240
1241         stats = this_cpu_ptr(vxlan->dev->tstats);
1242         u64_stats_update_begin(&stats->syncp);
1243         stats->rx_packets++;
1244         stats->rx_bytes += skb->len;
1245         u64_stats_update_end(&stats->syncp);
1246
1247         netif_rx(skb);
1248
1249         return;
1250 drop:
1251         /* Consume bad packet */
1252         kfree_skb(skb);
1253 }
1254
1255 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1256 {
1257         struct vxlan_dev *vxlan = netdev_priv(dev);
1258         struct arphdr *parp;
1259         u8 *arpptr, *sha;
1260         __be32 sip, tip;
1261         struct neighbour *n;
1262
1263         if (dev->flags & IFF_NOARP)
1264                 goto out;
1265
1266         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1267                 dev->stats.tx_dropped++;
1268                 goto out;
1269         }
1270         parp = arp_hdr(skb);
1271
1272         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1273              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1274             parp->ar_pro != htons(ETH_P_IP) ||
1275             parp->ar_op != htons(ARPOP_REQUEST) ||
1276             parp->ar_hln != dev->addr_len ||
1277             parp->ar_pln != 4)
1278                 goto out;
1279         arpptr = (u8 *)parp + sizeof(struct arphdr);
1280         sha = arpptr;
1281         arpptr += dev->addr_len;        /* sha */
1282         memcpy(&sip, arpptr, sizeof(sip));
1283         arpptr += sizeof(sip);
1284         arpptr += dev->addr_len;        /* tha */
1285         memcpy(&tip, arpptr, sizeof(tip));
1286
1287         if (ipv4_is_loopback(tip) ||
1288             ipv4_is_multicast(tip))
1289                 goto out;
1290
1291         n = neigh_lookup(&arp_tbl, &tip, dev);
1292
1293         if (n) {
1294                 struct vxlan_fdb *f;
1295                 struct sk_buff  *reply;
1296
1297                 if (!(n->nud_state & NUD_CONNECTED)) {
1298                         neigh_release(n);
1299                         goto out;
1300                 }
1301
1302                 f = vxlan_find_mac(vxlan, n->ha);
1303                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1304                         /* bridge-local neighbor */
1305                         neigh_release(n);
1306                         goto out;
1307                 }
1308
1309                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1310                                 n->ha, sha);
1311
1312                 neigh_release(n);
1313
1314                 if (reply == NULL)
1315                         goto out;
1316
1317                 skb_reset_mac_header(reply);
1318                 __skb_pull(reply, skb_network_offset(reply));
1319                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1320                 reply->pkt_type = PACKET_HOST;
1321
1322                 if (netif_rx_ni(reply) == NET_RX_DROP)
1323                         dev->stats.rx_dropped++;
1324         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1325                 union vxlan_addr ipa = {
1326                         .sin.sin_addr.s_addr = tip,
1327                         .sin.sin_family = AF_INET,
1328                 };
1329
1330                 vxlan_ip_miss(dev, &ipa);
1331         }
1332 out:
1333         consume_skb(skb);
1334         return NETDEV_TX_OK;
1335 }
1336
1337 #if IS_ENABLED(CONFIG_IPV6)
1338 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1339         struct neighbour *n, bool isrouter)
1340 {
1341         struct net_device *dev = request->dev;
1342         struct sk_buff *reply;
1343         struct nd_msg *ns, *na;
1344         struct ipv6hdr *pip6;
1345         u8 *daddr;
1346         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1347         int ns_olen;
1348         int i, len;
1349
1350         if (dev == NULL)
1351                 return NULL;
1352
1353         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1354                 sizeof(*na) + na_olen + dev->needed_tailroom;
1355         reply = alloc_skb(len, GFP_ATOMIC);
1356         if (reply == NULL)
1357                 return NULL;
1358
1359         reply->protocol = htons(ETH_P_IPV6);
1360         reply->dev = dev;
1361         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1362         skb_push(reply, sizeof(struct ethhdr));
1363         skb_set_mac_header(reply, 0);
1364
1365         ns = (struct nd_msg *)skb_transport_header(request);
1366
1367         daddr = eth_hdr(request)->h_source;
1368         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1369         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1370                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1371                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1372                         break;
1373                 }
1374         }
1375
1376         /* Ethernet header */
1377         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1378         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1379         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1380         reply->protocol = htons(ETH_P_IPV6);
1381
1382         skb_pull(reply, sizeof(struct ethhdr));
1383         skb_set_network_header(reply, 0);
1384         skb_put(reply, sizeof(struct ipv6hdr));
1385
1386         /* IPv6 header */
1387
1388         pip6 = ipv6_hdr(reply);
1389         memset(pip6, 0, sizeof(struct ipv6hdr));
1390         pip6->version = 6;
1391         pip6->priority = ipv6_hdr(request)->priority;
1392         pip6->nexthdr = IPPROTO_ICMPV6;
1393         pip6->hop_limit = 255;
1394         pip6->daddr = ipv6_hdr(request)->saddr;
1395         pip6->saddr = *(struct in6_addr *)n->primary_key;
1396
1397         skb_pull(reply, sizeof(struct ipv6hdr));
1398         skb_set_transport_header(reply, 0);
1399
1400         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1401
1402         /* Neighbor Advertisement */
1403         memset(na, 0, sizeof(*na)+na_olen);
1404         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1405         na->icmph.icmp6_router = isrouter;
1406         na->icmph.icmp6_override = 1;
1407         na->icmph.icmp6_solicited = 1;
1408         na->target = ns->target;
1409         ether_addr_copy(&na->opt[2], n->ha);
1410         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1411         na->opt[1] = na_olen >> 3;
1412
1413         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1414                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1415                 csum_partial(na, sizeof(*na)+na_olen, 0));
1416
1417         pip6->payload_len = htons(sizeof(*na)+na_olen);
1418
1419         skb_push(reply, sizeof(struct ipv6hdr));
1420
1421         reply->ip_summed = CHECKSUM_UNNECESSARY;
1422
1423         return reply;
1424 }
1425
1426 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1427 {
1428         struct vxlan_dev *vxlan = netdev_priv(dev);
1429         struct nd_msg *msg;
1430         const struct ipv6hdr *iphdr;
1431         const struct in6_addr *saddr, *daddr;
1432         struct neighbour *n;
1433         struct inet6_dev *in6_dev;
1434
1435         in6_dev = __in6_dev_get(dev);
1436         if (!in6_dev)
1437                 goto out;
1438
1439         iphdr = ipv6_hdr(skb);
1440         saddr = &iphdr->saddr;
1441         daddr = &iphdr->daddr;
1442
1443         msg = (struct nd_msg *)skb_transport_header(skb);
1444         if (msg->icmph.icmp6_code != 0 ||
1445             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1446                 goto out;
1447
1448         if (ipv6_addr_loopback(daddr) ||
1449             ipv6_addr_is_multicast(&msg->target))
1450                 goto out;
1451
1452         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1453
1454         if (n) {
1455                 struct vxlan_fdb *f;
1456                 struct sk_buff *reply;
1457
1458                 if (!(n->nud_state & NUD_CONNECTED)) {
1459                         neigh_release(n);
1460                         goto out;
1461                 }
1462
1463                 f = vxlan_find_mac(vxlan, n->ha);
1464                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1465                         /* bridge-local neighbor */
1466                         neigh_release(n);
1467                         goto out;
1468                 }
1469
1470                 reply = vxlan_na_create(skb, n,
1471                                         !!(f ? f->flags & NTF_ROUTER : 0));
1472
1473                 neigh_release(n);
1474
1475                 if (reply == NULL)
1476                         goto out;
1477
1478                 if (netif_rx_ni(reply) == NET_RX_DROP)
1479                         dev->stats.rx_dropped++;
1480
1481         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1482                 union vxlan_addr ipa = {
1483                         .sin6.sin6_addr = msg->target,
1484                         .sin6.sin6_family = AF_INET6,
1485                 };
1486
1487                 vxlan_ip_miss(dev, &ipa);
1488         }
1489
1490 out:
1491         consume_skb(skb);
1492         return NETDEV_TX_OK;
1493 }
1494 #endif
1495
1496 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1497 {
1498         struct vxlan_dev *vxlan = netdev_priv(dev);
1499         struct neighbour *n;
1500
1501         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1502                 return false;
1503
1504         n = NULL;
1505         switch (ntohs(eth_hdr(skb)->h_proto)) {
1506         case ETH_P_IP:
1507         {
1508                 struct iphdr *pip;
1509
1510                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1511                         return false;
1512                 pip = ip_hdr(skb);
1513                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1514                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1515                         union vxlan_addr ipa = {
1516                                 .sin.sin_addr.s_addr = pip->daddr,
1517                                 .sin.sin_family = AF_INET,
1518                         };
1519
1520                         vxlan_ip_miss(dev, &ipa);
1521                         return false;
1522                 }
1523
1524                 break;
1525         }
1526 #if IS_ENABLED(CONFIG_IPV6)
1527         case ETH_P_IPV6:
1528         {
1529                 struct ipv6hdr *pip6;
1530
1531                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1532                         return false;
1533                 pip6 = ipv6_hdr(skb);
1534                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1535                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1536                         union vxlan_addr ipa = {
1537                                 .sin6.sin6_addr = pip6->daddr,
1538                                 .sin6.sin6_family = AF_INET6,
1539                         };
1540
1541                         vxlan_ip_miss(dev, &ipa);
1542                         return false;
1543                 }
1544
1545                 break;
1546         }
1547 #endif
1548         default:
1549                 return false;
1550         }
1551
1552         if (n) {
1553                 bool diff;
1554
1555                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1556                 if (diff) {
1557                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1558                                 dev->addr_len);
1559                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1560                 }
1561                 neigh_release(n);
1562                 return diff;
1563         }
1564
1565         return false;
1566 }
1567
1568 #if IS_ENABLED(CONFIG_IPV6)
1569 static int vxlan6_xmit_skb(struct vxlan_sock *vs,
1570                            struct dst_entry *dst, struct sk_buff *skb,
1571                            struct net_device *dev, struct in6_addr *saddr,
1572                            struct in6_addr *daddr, __u8 prio, __u8 ttl,
1573                            __be16 src_port, __be16 dst_port, __be32 vni,
1574                            bool xnet)
1575 {
1576         struct vxlanhdr *vxh;
1577         int min_headroom;
1578         int err;
1579         bool udp_sum = !udp_get_no_check6_tx(vs->sock->sk);
1580
1581         skb = udp_tunnel_handle_offloads(skb, udp_sum);
1582         if (IS_ERR(skb))
1583                 return -EINVAL;
1584
1585         skb_scrub_packet(skb, xnet);
1586
1587         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1588                         + VXLAN_HLEN + sizeof(struct ipv6hdr)
1589                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1590
1591         /* Need space for new headers (invalidates iph ptr) */
1592         err = skb_cow_head(skb, min_headroom);
1593         if (unlikely(err))
1594                 return err;
1595
1596         if (vlan_tx_tag_present(skb)) {
1597                 if (WARN_ON(!__vlan_put_tag(skb,
1598                                             skb->vlan_proto,
1599                                             vlan_tx_tag_get(skb))))
1600                         return -ENOMEM;
1601
1602                 skb->vlan_tci = 0;
1603         }
1604
1605         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1606         vxh->vx_flags = htonl(VXLAN_FLAGS);
1607         vxh->vx_vni = vni;
1608
1609         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1610
1611         udp_tunnel6_xmit_skb(vs->sock, dst, skb, dev, saddr, daddr, prio,
1612                              ttl, src_port, dst_port);
1613         return 0;
1614 }
1615 #endif
1616
1617 int vxlan_xmit_skb(struct vxlan_sock *vs,
1618                    struct rtable *rt, struct sk_buff *skb,
1619                    __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1620                    __be16 src_port, __be16 dst_port, __be32 vni, bool xnet)
1621 {
1622         struct vxlanhdr *vxh;
1623         int min_headroom;
1624         int err;
1625         bool udp_sum = !vs->sock->sk->sk_no_check_tx;
1626
1627         skb = udp_tunnel_handle_offloads(skb, udp_sum);
1628         if (IS_ERR(skb))
1629                 return -EINVAL;
1630
1631         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1632                         + VXLAN_HLEN + sizeof(struct iphdr)
1633                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1634
1635         /* Need space for new headers (invalidates iph ptr) */
1636         err = skb_cow_head(skb, min_headroom);
1637         if (unlikely(err))
1638                 return err;
1639
1640         if (vlan_tx_tag_present(skb)) {
1641                 if (WARN_ON(!__vlan_put_tag(skb,
1642                                             skb->vlan_proto,
1643                                             vlan_tx_tag_get(skb))))
1644                         return -ENOMEM;
1645
1646                 skb->vlan_tci = 0;
1647         }
1648
1649         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1650         vxh->vx_flags = htonl(VXLAN_FLAGS);
1651         vxh->vx_vni = vni;
1652
1653         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1654
1655         return udp_tunnel_xmit_skb(vs->sock, rt, skb, src, dst, tos,
1656                                    ttl, df, src_port, dst_port, xnet);
1657 }
1658 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1659
1660 /* Bypass encapsulation if the destination is local */
1661 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1662                                struct vxlan_dev *dst_vxlan)
1663 {
1664         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1665         union vxlan_addr loopback;
1666         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1667         struct net_device *dev = skb->dev;
1668         int len = skb->len;
1669
1670         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1671         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1672         skb->pkt_type = PACKET_HOST;
1673         skb->encapsulation = 0;
1674         skb->dev = dst_vxlan->dev;
1675         __skb_pull(skb, skb_network_offset(skb));
1676
1677         if (remote_ip->sa.sa_family == AF_INET) {
1678                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1679                 loopback.sa.sa_family =  AF_INET;
1680 #if IS_ENABLED(CONFIG_IPV6)
1681         } else {
1682                 loopback.sin6.sin6_addr = in6addr_loopback;
1683                 loopback.sa.sa_family =  AF_INET6;
1684 #endif
1685         }
1686
1687         if (dst_vxlan->flags & VXLAN_F_LEARN)
1688                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1689
1690         u64_stats_update_begin(&tx_stats->syncp);
1691         tx_stats->tx_packets++;
1692         tx_stats->tx_bytes += len;
1693         u64_stats_update_end(&tx_stats->syncp);
1694
1695         if (netif_rx(skb) == NET_RX_SUCCESS) {
1696                 u64_stats_update_begin(&rx_stats->syncp);
1697                 rx_stats->rx_packets++;
1698                 rx_stats->rx_bytes += len;
1699                 u64_stats_update_end(&rx_stats->syncp);
1700         } else {
1701                 dev->stats.rx_dropped++;
1702         }
1703 }
1704
1705 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1706                            struct vxlan_rdst *rdst, bool did_rsc)
1707 {
1708         struct vxlan_dev *vxlan = netdev_priv(dev);
1709         struct rtable *rt = NULL;
1710         const struct iphdr *old_iph;
1711         struct flowi4 fl4;
1712         union vxlan_addr *dst;
1713         __be16 src_port = 0, dst_port;
1714         u32 vni;
1715         __be16 df = 0;
1716         __u8 tos, ttl;
1717         int err;
1718
1719         dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1720         vni = rdst->remote_vni;
1721         dst = &rdst->remote_ip;
1722
1723         if (vxlan_addr_any(dst)) {
1724                 if (did_rsc) {
1725                         /* short-circuited back to local bridge */
1726                         vxlan_encap_bypass(skb, vxlan, vxlan);
1727                         return;
1728                 }
1729                 goto drop;
1730         }
1731
1732         old_iph = ip_hdr(skb);
1733
1734         ttl = vxlan->ttl;
1735         if (!ttl && vxlan_addr_multicast(dst))
1736                 ttl = 1;
1737
1738         tos = vxlan->tos;
1739         if (tos == 1)
1740                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1741
1742         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->port_min,
1743                                      vxlan->port_max, true);
1744
1745         if (dst->sa.sa_family == AF_INET) {
1746                 memset(&fl4, 0, sizeof(fl4));
1747                 fl4.flowi4_oif = rdst->remote_ifindex;
1748                 fl4.flowi4_tos = RT_TOS(tos);
1749                 fl4.daddr = dst->sin.sin_addr.s_addr;
1750                 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1751
1752                 rt = ip_route_output_key(vxlan->net, &fl4);
1753                 if (IS_ERR(rt)) {
1754                         netdev_dbg(dev, "no route to %pI4\n",
1755                                    &dst->sin.sin_addr.s_addr);
1756                         dev->stats.tx_carrier_errors++;
1757                         goto tx_error;
1758                 }
1759
1760                 if (rt->dst.dev == dev) {
1761                         netdev_dbg(dev, "circular route to %pI4\n",
1762                                    &dst->sin.sin_addr.s_addr);
1763                         dev->stats.collisions++;
1764                         goto rt_tx_error;
1765                 }
1766
1767                 /* Bypass encapsulation if the destination is local */
1768                 if (rt->rt_flags & RTCF_LOCAL &&
1769                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1770                         struct vxlan_dev *dst_vxlan;
1771
1772                         ip_rt_put(rt);
1773                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1774                                                    dst->sa.sa_family, dst_port);
1775                         if (!dst_vxlan)
1776                                 goto tx_error;
1777                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1778                         return;
1779                 }
1780
1781                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1782                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1783
1784                 err = vxlan_xmit_skb(vxlan->vn_sock, rt, skb,
1785                                      fl4.saddr, dst->sin.sin_addr.s_addr,
1786                                      tos, ttl, df, src_port, dst_port,
1787                                      htonl(vni << 8),
1788                                      !net_eq(vxlan->net, dev_net(vxlan->dev)));
1789
1790                 if (err < 0)
1791                         goto rt_tx_error;
1792                 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1793 #if IS_ENABLED(CONFIG_IPV6)
1794         } else {
1795                 struct sock *sk = vxlan->vn_sock->sock->sk;
1796                 struct dst_entry *ndst;
1797                 struct flowi6 fl6;
1798                 u32 flags;
1799
1800                 memset(&fl6, 0, sizeof(fl6));
1801                 fl6.flowi6_oif = rdst->remote_ifindex;
1802                 fl6.daddr = dst->sin6.sin6_addr;
1803                 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
1804                 fl6.flowi6_proto = IPPROTO_UDP;
1805
1806                 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
1807                         netdev_dbg(dev, "no route to %pI6\n",
1808                                    &dst->sin6.sin6_addr);
1809                         dev->stats.tx_carrier_errors++;
1810                         goto tx_error;
1811                 }
1812
1813                 if (ndst->dev == dev) {
1814                         netdev_dbg(dev, "circular route to %pI6\n",
1815                                    &dst->sin6.sin6_addr);
1816                         dst_release(ndst);
1817                         dev->stats.collisions++;
1818                         goto tx_error;
1819                 }
1820
1821                 /* Bypass encapsulation if the destination is local */
1822                 flags = ((struct rt6_info *)ndst)->rt6i_flags;
1823                 if (flags & RTF_LOCAL &&
1824                     !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1825                         struct vxlan_dev *dst_vxlan;
1826
1827                         dst_release(ndst);
1828                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1829                                                    dst->sa.sa_family, dst_port);
1830                         if (!dst_vxlan)
1831                                 goto tx_error;
1832                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1833                         return;
1834                 }
1835
1836                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
1837
1838                 err = vxlan6_xmit_skb(vxlan->vn_sock, ndst, skb,
1839                                       dev, &fl6.saddr, &fl6.daddr, 0, ttl,
1840                                       src_port, dst_port, htonl(vni << 8),
1841                                       !net_eq(vxlan->net, dev_net(vxlan->dev)));
1842 #endif
1843         }
1844
1845         return;
1846
1847 drop:
1848         dev->stats.tx_dropped++;
1849         goto tx_free;
1850
1851 rt_tx_error:
1852         ip_rt_put(rt);
1853 tx_error:
1854         dev->stats.tx_errors++;
1855 tx_free:
1856         dev_kfree_skb(skb);
1857 }
1858
1859 /* Transmit local packets over Vxlan
1860  *
1861  * Outer IP header inherits ECN and DF from inner header.
1862  * Outer UDP destination is the VXLAN assigned port.
1863  *           source port is based on hash of flow
1864  */
1865 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1866 {
1867         struct vxlan_dev *vxlan = netdev_priv(dev);
1868         struct ethhdr *eth;
1869         bool did_rsc = false;
1870         struct vxlan_rdst *rdst, *fdst = NULL;
1871         struct vxlan_fdb *f;
1872
1873         skb_reset_mac_header(skb);
1874         eth = eth_hdr(skb);
1875
1876         if ((vxlan->flags & VXLAN_F_PROXY)) {
1877                 if (ntohs(eth->h_proto) == ETH_P_ARP)
1878                         return arp_reduce(dev, skb);
1879 #if IS_ENABLED(CONFIG_IPV6)
1880                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
1881                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
1882                                        + sizeof(struct nd_msg)) &&
1883                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
1884                                 struct nd_msg *msg;
1885
1886                                 msg = (struct nd_msg *)skb_transport_header(skb);
1887                                 if (msg->icmph.icmp6_code == 0 &&
1888                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
1889                                         return neigh_reduce(dev, skb);
1890                 }
1891                 eth = eth_hdr(skb);
1892 #endif
1893         }
1894
1895         f = vxlan_find_mac(vxlan, eth->h_dest);
1896         did_rsc = false;
1897
1898         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
1899             (ntohs(eth->h_proto) == ETH_P_IP ||
1900              ntohs(eth->h_proto) == ETH_P_IPV6)) {
1901                 did_rsc = route_shortcircuit(dev, skb);
1902                 if (did_rsc)
1903                         f = vxlan_find_mac(vxlan, eth->h_dest);
1904         }
1905
1906         if (f == NULL) {
1907                 f = vxlan_find_mac(vxlan, all_zeros_mac);
1908                 if (f == NULL) {
1909                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
1910                             !is_multicast_ether_addr(eth->h_dest))
1911                                 vxlan_fdb_miss(vxlan, eth->h_dest);
1912
1913                         dev->stats.tx_dropped++;
1914                         kfree_skb(skb);
1915                         return NETDEV_TX_OK;
1916                 }
1917         }
1918
1919         list_for_each_entry_rcu(rdst, &f->remotes, list) {
1920                 struct sk_buff *skb1;
1921
1922                 if (!fdst) {
1923                         fdst = rdst;
1924                         continue;
1925                 }
1926                 skb1 = skb_clone(skb, GFP_ATOMIC);
1927                 if (skb1)
1928                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1929         }
1930
1931         if (fdst)
1932                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
1933         else
1934                 kfree_skb(skb);
1935         return NETDEV_TX_OK;
1936 }
1937
1938 /* Walk the forwarding table and purge stale entries */
1939 static void vxlan_cleanup(unsigned long arg)
1940 {
1941         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1942         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1943         unsigned int h;
1944
1945         if (!netif_running(vxlan->dev))
1946                 return;
1947
1948         spin_lock_bh(&vxlan->hash_lock);
1949         for (h = 0; h < FDB_HASH_SIZE; ++h) {
1950                 struct hlist_node *p, *n;
1951                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1952                         struct vxlan_fdb *f
1953                                 = container_of(p, struct vxlan_fdb, hlist);
1954                         unsigned long timeout;
1955
1956                         if (f->state & NUD_PERMANENT)
1957                                 continue;
1958
1959                         timeout = f->used + vxlan->age_interval * HZ;
1960                         if (time_before_eq(timeout, jiffies)) {
1961                                 netdev_dbg(vxlan->dev,
1962                                            "garbage collect %pM\n",
1963                                            f->eth_addr);
1964                                 f->state = NUD_STALE;
1965                                 vxlan_fdb_destroy(vxlan, f);
1966                         } else if (time_before(timeout, next_timer))
1967                                 next_timer = timeout;
1968                 }
1969         }
1970         spin_unlock_bh(&vxlan->hash_lock);
1971
1972         mod_timer(&vxlan->age_timer, next_timer);
1973 }
1974
1975 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
1976 {
1977         __u32 vni = vxlan->default_dst.remote_vni;
1978
1979         vxlan->vn_sock = vs;
1980         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
1981 }
1982
1983 /* Setup stats when device is created */
1984 static int vxlan_init(struct net_device *dev)
1985 {
1986         struct vxlan_dev *vxlan = netdev_priv(dev);
1987         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
1988         struct vxlan_sock *vs;
1989         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
1990
1991         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1992         if (!dev->tstats)
1993                 return -ENOMEM;
1994
1995         spin_lock(&vn->sock_lock);
1996         vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
1997                              vxlan->dst_port);
1998         if (vs) {
1999                 /* If we have a socket with same port already, reuse it */
2000                 atomic_inc(&vs->refcnt);
2001                 vxlan_vs_add_dev(vs, vxlan);
2002         } else {
2003                 /* otherwise make new socket outside of RTNL */
2004                 dev_hold(dev);
2005                 queue_work(vxlan_wq, &vxlan->sock_work);
2006         }
2007         spin_unlock(&vn->sock_lock);
2008
2009         return 0;
2010 }
2011
2012 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2013 {
2014         struct vxlan_fdb *f;
2015
2016         spin_lock_bh(&vxlan->hash_lock);
2017         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2018         if (f)
2019                 vxlan_fdb_destroy(vxlan, f);
2020         spin_unlock_bh(&vxlan->hash_lock);
2021 }
2022
2023 static void vxlan_uninit(struct net_device *dev)
2024 {
2025         struct vxlan_dev *vxlan = netdev_priv(dev);
2026         struct vxlan_sock *vs = vxlan->vn_sock;
2027
2028         vxlan_fdb_delete_default(vxlan);
2029
2030         if (vs)
2031                 vxlan_sock_release(vs);
2032         free_percpu(dev->tstats);
2033 }
2034
2035 /* Start ageing timer and join group when device is brought up */
2036 static int vxlan_open(struct net_device *dev)
2037 {
2038         struct vxlan_dev *vxlan = netdev_priv(dev);
2039         struct vxlan_sock *vs = vxlan->vn_sock;
2040
2041         /* socket hasn't been created */
2042         if (!vs)
2043                 return -ENOTCONN;
2044
2045         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2046                 vxlan_sock_hold(vs);
2047                 dev_hold(dev);
2048                 queue_work(vxlan_wq, &vxlan->igmp_join);
2049         }
2050
2051         if (vxlan->age_interval)
2052                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2053
2054         return 0;
2055 }
2056
2057 /* Purge the forwarding table */
2058 static void vxlan_flush(struct vxlan_dev *vxlan)
2059 {
2060         unsigned int h;
2061
2062         spin_lock_bh(&vxlan->hash_lock);
2063         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2064                 struct hlist_node *p, *n;
2065                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2066                         struct vxlan_fdb *f
2067                                 = container_of(p, struct vxlan_fdb, hlist);
2068                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2069                         if (!is_zero_ether_addr(f->eth_addr))
2070                                 vxlan_fdb_destroy(vxlan, f);
2071                 }
2072         }
2073         spin_unlock_bh(&vxlan->hash_lock);
2074 }
2075
2076 /* Cleanup timer and forwarding table on shutdown */
2077 static int vxlan_stop(struct net_device *dev)
2078 {
2079         struct vxlan_dev *vxlan = netdev_priv(dev);
2080         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2081         struct vxlan_sock *vs = vxlan->vn_sock;
2082
2083         if (vs && vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2084             !vxlan_group_used(vn, vxlan)) {
2085                 vxlan_sock_hold(vs);
2086                 dev_hold(dev);
2087                 queue_work(vxlan_wq, &vxlan->igmp_leave);
2088         }
2089
2090         del_timer_sync(&vxlan->age_timer);
2091
2092         vxlan_flush(vxlan);
2093
2094         return 0;
2095 }
2096
2097 /* Stub, nothing needs to be done. */
2098 static void vxlan_set_multicast_list(struct net_device *dev)
2099 {
2100 }
2101
2102 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2103 {
2104         struct vxlan_dev *vxlan = netdev_priv(dev);
2105         struct vxlan_rdst *dst = &vxlan->default_dst;
2106         struct net_device *lowerdev;
2107         int max_mtu;
2108
2109         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2110         if (lowerdev == NULL)
2111                 return eth_change_mtu(dev, new_mtu);
2112
2113         if (dst->remote_ip.sa.sa_family == AF_INET6)
2114                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2115         else
2116                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2117
2118         if (new_mtu < 68 || new_mtu > max_mtu)
2119                 return -EINVAL;
2120
2121         dev->mtu = new_mtu;
2122         return 0;
2123 }
2124
2125 static const struct net_device_ops vxlan_netdev_ops = {
2126         .ndo_init               = vxlan_init,
2127         .ndo_uninit             = vxlan_uninit,
2128         .ndo_open               = vxlan_open,
2129         .ndo_stop               = vxlan_stop,
2130         .ndo_start_xmit         = vxlan_xmit,
2131         .ndo_get_stats64        = ip_tunnel_get_stats64,
2132         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2133         .ndo_change_mtu         = vxlan_change_mtu,
2134         .ndo_validate_addr      = eth_validate_addr,
2135         .ndo_set_mac_address    = eth_mac_addr,
2136         .ndo_fdb_add            = vxlan_fdb_add,
2137         .ndo_fdb_del            = vxlan_fdb_delete,
2138         .ndo_fdb_dump           = vxlan_fdb_dump,
2139 };
2140
2141 /* Info for udev, that this is a virtual tunnel endpoint */
2142 static struct device_type vxlan_type = {
2143         .name = "vxlan",
2144 };
2145
2146 /* Calls the ndo_add_vxlan_port of the caller in order to
2147  * supply the listening VXLAN udp ports. Callers are expected
2148  * to implement the ndo_add_vxlan_port.
2149  */
2150 void vxlan_get_rx_port(struct net_device *dev)
2151 {
2152         struct vxlan_sock *vs;
2153         struct net *net = dev_net(dev);
2154         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2155         sa_family_t sa_family;
2156         __be16 port;
2157         unsigned int i;
2158
2159         spin_lock(&vn->sock_lock);
2160         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2161                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2162                         port = inet_sk(vs->sock->sk)->inet_sport;
2163                         sa_family = vs->sock->sk->sk_family;
2164                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2165                                                             port);
2166                 }
2167         }
2168         spin_unlock(&vn->sock_lock);
2169 }
2170 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2171
2172 /* Initialize the device structure. */
2173 static void vxlan_setup(struct net_device *dev)
2174 {
2175         struct vxlan_dev *vxlan = netdev_priv(dev);
2176         unsigned int h;
2177
2178         eth_hw_addr_random(dev);
2179         ether_setup(dev);
2180         if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2181                 dev->needed_headroom = ETH_HLEN + VXLAN6_HEADROOM;
2182         else
2183                 dev->needed_headroom = ETH_HLEN + VXLAN_HEADROOM;
2184
2185         dev->netdev_ops = &vxlan_netdev_ops;
2186         dev->destructor = free_netdev;
2187         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2188
2189         dev->tx_queue_len = 0;
2190         dev->features   |= NETIF_F_LLTX;
2191         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2192         dev->features   |= NETIF_F_RXCSUM;
2193         dev->features   |= NETIF_F_GSO_SOFTWARE;
2194
2195         dev->vlan_features = dev->features;
2196         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2197         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2198         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2199         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2200         netif_keep_dst(dev);
2201         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2202
2203         INIT_LIST_HEAD(&vxlan->next);
2204         spin_lock_init(&vxlan->hash_lock);
2205         INIT_WORK(&vxlan->igmp_join, vxlan_igmp_join);
2206         INIT_WORK(&vxlan->igmp_leave, vxlan_igmp_leave);
2207         INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
2208
2209         init_timer_deferrable(&vxlan->age_timer);
2210         vxlan->age_timer.function = vxlan_cleanup;
2211         vxlan->age_timer.data = (unsigned long) vxlan;
2212
2213         vxlan->dst_port = htons(vxlan_port);
2214
2215         vxlan->dev = dev;
2216
2217         for (h = 0; h < FDB_HASH_SIZE; ++h)
2218                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2219 }
2220
2221 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2222         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2223         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2224         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2225         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2226         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2227         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2228         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2229         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2230         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2231         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2232         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2233         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2234         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2235         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2236         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2237         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2238         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2239 };
2240
2241 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2242 {
2243         if (tb[IFLA_ADDRESS]) {
2244                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2245                         pr_debug("invalid link address (not ethernet)\n");
2246                         return -EINVAL;
2247                 }
2248
2249                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2250                         pr_debug("invalid all zero ethernet address\n");
2251                         return -EADDRNOTAVAIL;
2252                 }
2253         }
2254
2255         if (!data)
2256                 return -EINVAL;
2257
2258         if (data[IFLA_VXLAN_ID]) {
2259                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2260                 if (id >= VXLAN_VID_MASK)
2261                         return -ERANGE;
2262         }
2263
2264         if (data[IFLA_VXLAN_PORT_RANGE]) {
2265                 const struct ifla_vxlan_port_range *p
2266                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2267
2268                 if (ntohs(p->high) < ntohs(p->low)) {
2269                         pr_debug("port range %u .. %u not valid\n",
2270                                  ntohs(p->low), ntohs(p->high));
2271                         return -EINVAL;
2272                 }
2273         }
2274
2275         return 0;
2276 }
2277
2278 static void vxlan_get_drvinfo(struct net_device *netdev,
2279                               struct ethtool_drvinfo *drvinfo)
2280 {
2281         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2282         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2283 }
2284
2285 static const struct ethtool_ops vxlan_ethtool_ops = {
2286         .get_drvinfo    = vxlan_get_drvinfo,
2287         .get_link       = ethtool_op_get_link,
2288 };
2289
2290 static void vxlan_del_work(struct work_struct *work)
2291 {
2292         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2293         udp_tunnel_sock_release(vs->sock);
2294         kfree_rcu(vs, rcu);
2295 }
2296
2297 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2298                                         __be16 port, u32 flags)
2299 {
2300         struct socket *sock;
2301         struct udp_port_cfg udp_conf;
2302         int err;
2303
2304         memset(&udp_conf, 0, sizeof(udp_conf));
2305
2306         if (ipv6) {
2307                 udp_conf.family = AF_INET6;
2308                 udp_conf.use_udp6_tx_checksums =
2309                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2310                 udp_conf.use_udp6_rx_checksums =
2311                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2312         } else {
2313                 udp_conf.family = AF_INET;
2314                 udp_conf.local_ip.s_addr = INADDR_ANY;
2315                 udp_conf.use_udp_checksums =
2316                     !!(flags & VXLAN_F_UDP_CSUM);
2317         }
2318
2319         udp_conf.local_udp_port = port;
2320
2321         /* Open UDP socket */
2322         err = udp_sock_create(net, &udp_conf, &sock);
2323         if (err < 0)
2324                 return ERR_PTR(err);
2325
2326         return sock;
2327 }
2328
2329 /* Create new listen socket if needed */
2330 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2331                                               vxlan_rcv_t *rcv, void *data,
2332                                               u32 flags)
2333 {
2334         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2335         struct vxlan_sock *vs;
2336         struct socket *sock;
2337         unsigned int h;
2338         bool ipv6 = !!(flags & VXLAN_F_IPV6);
2339         struct udp_tunnel_sock_cfg tunnel_cfg;
2340
2341         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2342         if (!vs)
2343                 return ERR_PTR(-ENOMEM);
2344
2345         for (h = 0; h < VNI_HASH_SIZE; ++h)
2346                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2347
2348         INIT_WORK(&vs->del_work, vxlan_del_work);
2349
2350         sock = vxlan_create_sock(net, ipv6, port, flags);
2351         if (IS_ERR(sock)) {
2352                 kfree(vs);
2353                 return ERR_CAST(sock);
2354         }
2355
2356         vs->sock = sock;
2357         atomic_set(&vs->refcnt, 1);
2358         vs->rcv = rcv;
2359         vs->data = data;
2360
2361         /* Initialize the vxlan udp offloads structure */
2362         vs->udp_offloads.port = port;
2363         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2364         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2365
2366         spin_lock(&vn->sock_lock);
2367         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2368         vxlan_notify_add_rx_port(vs);
2369         spin_unlock(&vn->sock_lock);
2370
2371         /* Mark socket as an encapsulation socket. */
2372         tunnel_cfg.sk_user_data = vs;
2373         tunnel_cfg.encap_type = 1;
2374         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2375         tunnel_cfg.encap_destroy = NULL;
2376
2377         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2378
2379         return vs;
2380 }
2381
2382 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2383                                   vxlan_rcv_t *rcv, void *data,
2384                                   bool no_share, u32 flags)
2385 {
2386         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2387         struct vxlan_sock *vs;
2388         bool ipv6 = flags & VXLAN_F_IPV6;
2389
2390         vs = vxlan_socket_create(net, port, rcv, data, flags);
2391         if (!IS_ERR(vs))
2392                 return vs;
2393
2394         if (no_share)   /* Return error if sharing is not allowed. */
2395                 return vs;
2396
2397         spin_lock(&vn->sock_lock);
2398         vs = vxlan_find_sock(net, ipv6 ? AF_INET6 : AF_INET, port);
2399         if (vs) {
2400                 if (vs->rcv == rcv)
2401                         atomic_inc(&vs->refcnt);
2402                 else
2403                         vs = ERR_PTR(-EBUSY);
2404         }
2405         spin_unlock(&vn->sock_lock);
2406
2407         if (!vs)
2408                 vs = ERR_PTR(-EINVAL);
2409
2410         return vs;
2411 }
2412 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2413
2414 /* Scheduled at device creation to bind to a socket */
2415 static void vxlan_sock_work(struct work_struct *work)
2416 {
2417         struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, sock_work);
2418         struct net *net = vxlan->net;
2419         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2420         __be16 port = vxlan->dst_port;
2421         struct vxlan_sock *nvs;
2422
2423         nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags);
2424         spin_lock(&vn->sock_lock);
2425         if (!IS_ERR(nvs))
2426                 vxlan_vs_add_dev(nvs, vxlan);
2427         spin_unlock(&vn->sock_lock);
2428
2429         dev_put(vxlan->dev);
2430 }
2431
2432 static int vxlan_newlink(struct net *net, struct net_device *dev,
2433                          struct nlattr *tb[], struct nlattr *data[])
2434 {
2435         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2436         struct vxlan_dev *vxlan = netdev_priv(dev);
2437         struct vxlan_rdst *dst = &vxlan->default_dst;
2438         __u32 vni;
2439         int err;
2440         bool use_ipv6 = false;
2441
2442         if (!data[IFLA_VXLAN_ID])
2443                 return -EINVAL;
2444
2445         vxlan->net = dev_net(dev);
2446
2447         vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2448         dst->remote_vni = vni;
2449
2450         /* Unless IPv6 is explicitly requested, assume IPv4 */
2451         dst->remote_ip.sa.sa_family = AF_INET;
2452         if (data[IFLA_VXLAN_GROUP]) {
2453                 dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
2454         } else if (data[IFLA_VXLAN_GROUP6]) {
2455                 if (!IS_ENABLED(CONFIG_IPV6))
2456                         return -EPFNOSUPPORT;
2457
2458                 nla_memcpy(&dst->remote_ip.sin6.sin6_addr, data[IFLA_VXLAN_GROUP6],
2459                            sizeof(struct in6_addr));
2460                 dst->remote_ip.sa.sa_family = AF_INET6;
2461                 use_ipv6 = true;
2462         }
2463
2464         if (data[IFLA_VXLAN_LOCAL]) {
2465                 vxlan->saddr.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
2466                 vxlan->saddr.sa.sa_family = AF_INET;
2467         } else if (data[IFLA_VXLAN_LOCAL6]) {
2468                 if (!IS_ENABLED(CONFIG_IPV6))
2469                         return -EPFNOSUPPORT;
2470
2471                 /* TODO: respect scope id */
2472                 nla_memcpy(&vxlan->saddr.sin6.sin6_addr, data[IFLA_VXLAN_LOCAL6],
2473                            sizeof(struct in6_addr));
2474                 vxlan->saddr.sa.sa_family = AF_INET6;
2475                 use_ipv6 = true;
2476         }
2477
2478         if (data[IFLA_VXLAN_LINK] &&
2479             (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2480                 struct net_device *lowerdev
2481                          = __dev_get_by_index(net, dst->remote_ifindex);
2482
2483                 if (!lowerdev) {
2484                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2485                         return -ENODEV;
2486                 }
2487
2488 #if IS_ENABLED(CONFIG_IPV6)
2489                 if (use_ipv6) {
2490                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2491                         if (idev && idev->cnf.disable_ipv6) {
2492                                 pr_info("IPv6 is disabled via sysctl\n");
2493                                 return -EPERM;
2494                         }
2495                         vxlan->flags |= VXLAN_F_IPV6;
2496                 }
2497 #endif
2498
2499                 if (!tb[IFLA_MTU])
2500                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2501
2502                 dev->needed_headroom = lowerdev->hard_header_len +
2503                                        (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2504         } else if (use_ipv6)
2505                 vxlan->flags |= VXLAN_F_IPV6;
2506
2507         if (data[IFLA_VXLAN_TOS])
2508                 vxlan->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2509
2510         if (data[IFLA_VXLAN_TTL])
2511                 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2512
2513         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2514                 vxlan->flags |= VXLAN_F_LEARN;
2515
2516         if (data[IFLA_VXLAN_AGEING])
2517                 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2518         else
2519                 vxlan->age_interval = FDB_AGE_DEFAULT;
2520
2521         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2522                 vxlan->flags |= VXLAN_F_PROXY;
2523
2524         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2525                 vxlan->flags |= VXLAN_F_RSC;
2526
2527         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2528                 vxlan->flags |= VXLAN_F_L2MISS;
2529
2530         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2531                 vxlan->flags |= VXLAN_F_L3MISS;
2532
2533         if (data[IFLA_VXLAN_LIMIT])
2534                 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2535
2536         if (data[IFLA_VXLAN_PORT_RANGE]) {
2537                 const struct ifla_vxlan_port_range *p
2538                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2539                 vxlan->port_min = ntohs(p->low);
2540                 vxlan->port_max = ntohs(p->high);
2541         }
2542
2543         if (data[IFLA_VXLAN_PORT])
2544                 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2545
2546         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2547                 vxlan->flags |= VXLAN_F_UDP_CSUM;
2548
2549         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2550             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2551                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2552
2553         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2554             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2555                 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2556
2557         if (vxlan_find_vni(net, vni, use_ipv6 ? AF_INET6 : AF_INET,
2558                            vxlan->dst_port)) {
2559                 pr_info("duplicate VNI %u\n", vni);
2560                 return -EEXIST;
2561         }
2562
2563         dev->ethtool_ops = &vxlan_ethtool_ops;
2564
2565         /* create an fdb entry for a valid default destination */
2566         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2567                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2568                                        &vxlan->default_dst.remote_ip,
2569                                        NUD_REACHABLE|NUD_PERMANENT,
2570                                        NLM_F_EXCL|NLM_F_CREATE,
2571                                        vxlan->dst_port,
2572                                        vxlan->default_dst.remote_vni,
2573                                        vxlan->default_dst.remote_ifindex,
2574                                        NTF_SELF);
2575                 if (err)
2576                         return err;
2577         }
2578
2579         err = register_netdevice(dev);
2580         if (err) {
2581                 vxlan_fdb_delete_default(vxlan);
2582                 return err;
2583         }
2584
2585         list_add(&vxlan->next, &vn->vxlan_list);
2586
2587         return 0;
2588 }
2589
2590 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2591 {
2592         struct vxlan_dev *vxlan = netdev_priv(dev);
2593         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2594
2595         spin_lock(&vn->sock_lock);
2596         if (!hlist_unhashed(&vxlan->hlist))
2597                 hlist_del_rcu(&vxlan->hlist);
2598         spin_unlock(&vn->sock_lock);
2599
2600         list_del(&vxlan->next);
2601         unregister_netdevice_queue(dev, head);
2602 }
2603
2604 static size_t vxlan_get_size(const struct net_device *dev)
2605 {
2606
2607         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2608                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2609                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2610                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2611                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2612                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2613                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2614                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2615                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2616                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2617                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2618                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2619                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2620                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2621                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2622                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2623                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2624                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2625                 0;
2626 }
2627
2628 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2629 {
2630         const struct vxlan_dev *vxlan = netdev_priv(dev);
2631         const struct vxlan_rdst *dst = &vxlan->default_dst;
2632         struct ifla_vxlan_port_range ports = {
2633                 .low =  htons(vxlan->port_min),
2634                 .high = htons(vxlan->port_max),
2635         };
2636
2637         if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2638                 goto nla_put_failure;
2639
2640         if (!vxlan_addr_any(&dst->remote_ip)) {
2641                 if (dst->remote_ip.sa.sa_family == AF_INET) {
2642                         if (nla_put_be32(skb, IFLA_VXLAN_GROUP,
2643                                          dst->remote_ip.sin.sin_addr.s_addr))
2644                                 goto nla_put_failure;
2645 #if IS_ENABLED(CONFIG_IPV6)
2646                 } else {
2647                         if (nla_put(skb, IFLA_VXLAN_GROUP6, sizeof(struct in6_addr),
2648                                     &dst->remote_ip.sin6.sin6_addr))
2649                                 goto nla_put_failure;
2650 #endif
2651                 }
2652         }
2653
2654         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2655                 goto nla_put_failure;
2656
2657         if (!vxlan_addr_any(&vxlan->saddr)) {
2658                 if (vxlan->saddr.sa.sa_family == AF_INET) {
2659                         if (nla_put_be32(skb, IFLA_VXLAN_LOCAL,
2660                                          vxlan->saddr.sin.sin_addr.s_addr))
2661                                 goto nla_put_failure;
2662 #if IS_ENABLED(CONFIG_IPV6)
2663                 } else {
2664                         if (nla_put(skb, IFLA_VXLAN_LOCAL6, sizeof(struct in6_addr),
2665                                     &vxlan->saddr.sin6.sin6_addr))
2666                                 goto nla_put_failure;
2667 #endif
2668                 }
2669         }
2670
2671         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2672             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2673             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2674                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
2675             nla_put_u8(skb, IFLA_VXLAN_PROXY,
2676                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
2677             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2678             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2679                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2680             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2681                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2682             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2683             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2684             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port) ||
2685             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2686                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2687             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2688                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2689             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2690                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)))
2691                 goto nla_put_failure;
2692
2693         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2694                 goto nla_put_failure;
2695
2696         return 0;
2697
2698 nla_put_failure:
2699         return -EMSGSIZE;
2700 }
2701
2702 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2703         .kind           = "vxlan",
2704         .maxtype        = IFLA_VXLAN_MAX,
2705         .policy         = vxlan_policy,
2706         .priv_size      = sizeof(struct vxlan_dev),
2707         .setup          = vxlan_setup,
2708         .validate       = vxlan_validate,
2709         .newlink        = vxlan_newlink,
2710         .dellink        = vxlan_dellink,
2711         .get_size       = vxlan_get_size,
2712         .fill_info      = vxlan_fill_info,
2713 };
2714
2715 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
2716                                              struct net_device *dev)
2717 {
2718         struct vxlan_dev *vxlan, *next;
2719         LIST_HEAD(list_kill);
2720
2721         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2722                 struct vxlan_rdst *dst = &vxlan->default_dst;
2723
2724                 /* In case we created vxlan device with carrier
2725                  * and we loose the carrier due to module unload
2726                  * we also need to remove vxlan device. In other
2727                  * cases, it's not necessary and remote_ifindex
2728                  * is 0 here, so no matches.
2729                  */
2730                 if (dst->remote_ifindex == dev->ifindex)
2731                         vxlan_dellink(vxlan->dev, &list_kill);
2732         }
2733
2734         unregister_netdevice_many(&list_kill);
2735 }
2736
2737 static int vxlan_lowerdev_event(struct notifier_block *unused,
2738                                 unsigned long event, void *ptr)
2739 {
2740         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2741         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2742
2743         if (event == NETDEV_UNREGISTER)
2744                 vxlan_handle_lowerdev_unregister(vn, dev);
2745
2746         return NOTIFY_DONE;
2747 }
2748
2749 static struct notifier_block vxlan_notifier_block __read_mostly = {
2750         .notifier_call = vxlan_lowerdev_event,
2751 };
2752
2753 static __net_init int vxlan_init_net(struct net *net)
2754 {
2755         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2756         unsigned int h;
2757
2758         INIT_LIST_HEAD(&vn->vxlan_list);
2759         spin_lock_init(&vn->sock_lock);
2760
2761         for (h = 0; h < PORT_HASH_SIZE; ++h)
2762                 INIT_HLIST_HEAD(&vn->sock_list[h]);
2763
2764         return 0;
2765 }
2766
2767 static void __net_exit vxlan_exit_net(struct net *net)
2768 {
2769         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2770         struct vxlan_dev *vxlan, *next;
2771         struct net_device *dev, *aux;
2772         LIST_HEAD(list);
2773
2774         rtnl_lock();
2775         for_each_netdev_safe(net, dev, aux)
2776                 if (dev->rtnl_link_ops == &vxlan_link_ops)
2777                         unregister_netdevice_queue(dev, &list);
2778
2779         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2780                 /* If vxlan->dev is in the same netns, it has already been added
2781                  * to the list by the previous loop.
2782                  */
2783                 if (!net_eq(dev_net(vxlan->dev), net))
2784                         unregister_netdevice_queue(dev, &list);
2785         }
2786
2787         unregister_netdevice_many(&list);
2788         rtnl_unlock();
2789 }
2790
2791 static struct pernet_operations vxlan_net_ops = {
2792         .init = vxlan_init_net,
2793         .exit = vxlan_exit_net,
2794         .id   = &vxlan_net_id,
2795         .size = sizeof(struct vxlan_net),
2796 };
2797
2798 static int __init vxlan_init_module(void)
2799 {
2800         int rc;
2801
2802         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
2803         if (!vxlan_wq)
2804                 return -ENOMEM;
2805
2806         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
2807
2808         rc = register_pernet_subsys(&vxlan_net_ops);
2809         if (rc)
2810                 goto out1;
2811
2812         rc = register_netdevice_notifier(&vxlan_notifier_block);
2813         if (rc)
2814                 goto out2;
2815
2816         rc = rtnl_link_register(&vxlan_link_ops);
2817         if (rc)
2818                 goto out3;
2819
2820         return 0;
2821 out3:
2822         unregister_netdevice_notifier(&vxlan_notifier_block);
2823 out2:
2824         unregister_pernet_subsys(&vxlan_net_ops);
2825 out1:
2826         destroy_workqueue(vxlan_wq);
2827         return rc;
2828 }
2829 late_initcall(vxlan_init_module);
2830
2831 static void __exit vxlan_cleanup_module(void)
2832 {
2833         rtnl_link_unregister(&vxlan_link_ops);
2834         unregister_netdevice_notifier(&vxlan_notifier_block);
2835         destroy_workqueue(vxlan_wq);
2836         unregister_pernet_subsys(&vxlan_net_ops);
2837         /* rcu_barrier() is called by netns */
2838 }
2839 module_exit(vxlan_cleanup_module);
2840
2841 MODULE_LICENSE("GPL");
2842 MODULE_VERSION(VXLAN_VERSION);
2843 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
2844 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
2845 MODULE_ALIAS_RTNL_LINK("vxlan");