drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/types.h>
  15 #include <linux/module.h>
  16 #include <linux/errno.h>
  17 #include <linux/slab.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/rculist.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/in.h>
  22 #include <linux/ip.h>
  23 #include <linux/udp.h>
  24 #include <linux/igmp.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/hash.h>
  29 #include <linux/ethtool.h>
  30 #include <net/arp.h>
  31 #include <net/ndisc.h>
  32 #include <net/ip.h>
  33 #include <net/ip_tunnels.h>
  34 #include <net/icmp.h>
  35 #include <net/udp.h>
  36 #include <net/udp_tunnel.h>
  37 #include <net/rtnetlink.h>
  38 #include <net/route.h>
  39 #include <net/dsfield.h>
  40 #include <net/inet_ecn.h>
  41 #include <net/net_namespace.h>
  42 #include <net/netns/generic.h>
  43 #include <net/vxlan.h>
  44 #include <net/protocol.h>
  45
  46 #if IS_ENABLED(CONFIG_IPV6)
  47 #include <net/ipv6.h>
  48 #include <net/addrconf.h>
  49 #include <net/ip6_tunnel.h>
  50 #include <net/ip6_checksum.h>
  51 #endif
  52 #include <net/dst_metadata.h>
  53
  54 #define VXLAN_VERSION   "0.1"
  55
  56 #define PORT_HASH_BITS  8
  57 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  58 #define FDB_AGE_DEFAULT 300 /* 5 min */
  59 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  60
  61 /* UDP port for VXLAN traffic.
  62  * The IANA assigned port is 4789, but the Linux default is 8472
  63  * for compatibility with early adopters.
  64  */
  65 static unsigned short vxlan_port __read_mostly = 8472;
  66 module_param_named(udp_port, vxlan_port, ushort, 0444);
  67 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  68
  69 static bool log_ecn_error = true;
  70 module_param(log_ecn_error, bool, 0644);
  71 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  72
  73 static int vxlan_net_id;
  74 static struct rtnl_link_ops vxlan_link_ops;
  75
  76 static const u8 all_zeros_mac[ETH_ALEN + 2];
  77
  78 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  79
  80 /* per-network namespace private data for this module */
  81 struct vxlan_net {
  82         struct list_head  vxlan_list;
  83         struct hlist_head sock_list[PORT_HASH_SIZE];
  84         spinlock_t        sock_lock;
  85 };
  86
  87 /* Forwarding table entry */
  88 struct vxlan_fdb {
  89         struct hlist_node hlist;        /* linked list of entries */
  90         struct rcu_head   rcu;
  91         unsigned long     updated;      /* jiffies */
  92         unsigned long     used;
  93         struct list_head  remotes;
  94         u8                eth_addr[ETH_ALEN];
  95         u16               state;        /* see ndm_state */
  96         u8                flags;        /* see ndm_flags */
  97 };
  98
  99 /* salt for hash table */
 100 static u32 vxlan_salt __read_mostly;
 101 static struct workqueue_struct *vxlan_wq;
 102
 103 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
 104 {
 105         return vs->flags & VXLAN_F_COLLECT_METADATA ||
 106                ip_tunnel_collect_metadata();
 107 }
 108
 109 #if IS_ENABLED(CONFIG_IPV6)
 110 static inline
 111 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 112 {
 113         if (a->sa.sa_family != b->sa.sa_family)
 114                 return false;
 115         if (a->sa.sa_family == AF_INET6)
 116                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 117         else
 118                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 119 }
 120
 121 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 122 {
 123         if (ipa->sa.sa_family == AF_INET6)
 124                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 125         else
 126                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 127 }
 128
 129 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 130 {
 131         if (ipa->sa.sa_family == AF_INET6)
 132                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 133         else
 134                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 135 }
 136
 137 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 138 {
 139         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 140                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 141                 ip->sa.sa_family = AF_INET6;
 142                 return 0;
 143         } else if (nla_len(nla) >= sizeof(__be32)) {
 144                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 145                 ip->sa.sa_family = AF_INET;
 146                 return 0;
 147         } else {
 148                 return -EAFNOSUPPORT;
 149         }
 150 }
 151
 152 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 153                               const union vxlan_addr *ip)
 154 {
 155         if (ip->sa.sa_family == AF_INET6)
 156                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 157         else
 158                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 159 }
 160
 161 #else /* !CONFIG_IPV6 */
 162
 163 static inline
 164 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 165 {
 166         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 167 }
 168
 169 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 170 {
 171         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 172 }
 173
 174 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 175 {
 176         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 177 }
 178
 179 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 180 {
 181         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 182                 return -EAFNOSUPPORT;
 183         } else if (nla_len(nla) >= sizeof(__be32)) {
 184                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 185                 ip->sa.sa_family = AF_INET;
 186                 return 0;
 187         } else {
 188                 return -EAFNOSUPPORT;
 189         }
 190 }
 191
 192 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 193                               const union vxlan_addr *ip)
 194 {
 195         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 196 }
 197 #endif
 198
 199 /* Virtual Network hash table head */
 200 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 201 {
 202         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 203 }
 204
 205 /* Socket hash table head */
 206 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 207 {
 208         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 209
 210         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 211 }
 212
 213 /* First remote destination for a forwarding entry.
 214  * Guaranteed to be non-NULL because remotes are never deleted.
 215  */
 216 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 217 {
 218         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 219 }
 220
 221 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 222 {
 223         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 224 }
 225
 226 /* Find VXLAN socket based on network namespace, address family and UDP port
 227  * and enabled unshareable flags.
 228  */
 229 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 230                                           __be16 port, u32 flags)
 231 {
 232         struct vxlan_sock *vs;
 233
 234         flags &= VXLAN_F_RCV_FLAGS;
 235
 236         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 237                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 238                     vxlan_get_sk_family(vs) == family &&
 239                     vs->flags == flags)
 240                         return vs;
 241         }
 242         return NULL;
 243 }
 244
 245 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
 246 {
 247         struct vxlan_dev *vxlan;
 248
 249         hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
 250                 if (vxlan->default_dst.remote_vni == vni)
 251                         return vxlan;
 252         }
 253
 254         return NULL;
 255 }
 256
 257 /* Look up VNI in a per net namespace table */
 258 static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
 259                                         sa_family_t family, __be16 port,
 260                                         u32 flags)
 261 {
 262         struct vxlan_sock *vs;
 263
 264         vs = vxlan_find_sock(net, family, port, flags);
 265         if (!vs)
 266                 return NULL;
 267
 268         return vxlan_vs_find_vni(vs, vni);
 269 }
 270
 271 /* Fill in neighbour message in skbuff. */
 272 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 273                           const struct vxlan_fdb *fdb,
 274                           u32 portid, u32 seq, int type, unsigned int flags,
 275                           const struct vxlan_rdst *rdst)
 276 {
 277         unsigned long now = jiffies;
 278         struct nda_cacheinfo ci;
 279         struct nlmsghdr *nlh;
 280         struct ndmsg *ndm;
 281         bool send_ip, send_eth;
 282
 283         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 284         if (nlh == NULL)
 285                 return -EMSGSIZE;
 286
 287         ndm = nlmsg_data(nlh);
 288         memset(ndm, 0, sizeof(*ndm));
 289
 290         send_eth = send_ip = true;
 291
 292         if (type == RTM_GETNEIGH) {
 293                 ndm->ndm_family = AF_INET;
 294                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 295                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 296         } else
 297                 ndm->ndm_family = AF_BRIDGE;
 298         ndm->ndm_state = fdb->state;
 299         ndm->ndm_ifindex = vxlan->dev->ifindex;
 300         ndm->ndm_flags = fdb->flags;
 301         ndm->ndm_type = RTN_UNICAST;
 302
 303         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 304             nla_put_s32(skb, NDA_LINK_NETNSID,
 305                         peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
 306                 goto nla_put_failure;
 307
 308         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 309                 goto nla_put_failure;
 310
 311         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 312                 goto nla_put_failure;
 313
 314         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 315             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 316                 goto nla_put_failure;
 317         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 318             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 319                 goto nla_put_failure;
 320         if (rdst->remote_ifindex &&
 321             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 322                 goto nla_put_failure;
 323
 324         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 325         ci.ndm_confirmed = 0;
 326         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 327         ci.ndm_refcnt    = 0;
 328
 329         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 330                 goto nla_put_failure;
 331
 332         nlmsg_end(skb, nlh);
 333         return 0;
 334
 335 nla_put_failure:
 336         nlmsg_cancel(skb, nlh);
 337         return -EMSGSIZE;
 338 }
 339
 340 static inline size_t vxlan_nlmsg_size(void)
 341 {
 342         return NLMSG_ALIGN(sizeof(struct ndmsg))
 343                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 344                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 345                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 346                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 347                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 348                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 349                 + nla_total_size(sizeof(struct nda_cacheinfo));
 350 }
 351
 352 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 353                              struct vxlan_rdst *rd, int type)
 354 {
 355         struct net *net = dev_net(vxlan->dev);
 356         struct sk_buff *skb;
 357         int err = -ENOBUFS;
 358
 359         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 360         if (skb == NULL)
 361                 goto errout;
 362
 363         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 364         if (err < 0) {
 365                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 366                 WARN_ON(err == -EMSGSIZE);
 367                 kfree_skb(skb);
 368                 goto errout;
 369         }
 370
 371         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 372         return;
 373 errout:
 374         if (err < 0)
 375                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 376 }
 377
 378 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 379 {
 380         struct vxlan_dev *vxlan = netdev_priv(dev);
 381         struct vxlan_fdb f = {
 382                 .state = NUD_STALE,
 383         };
 384         struct vxlan_rdst remote = {
 385                 .remote_ip = *ipa, /* goes to NDA_DST */
 386                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 387         };
 388
 389         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 390 }
 391
 392 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 393 {
 394         struct vxlan_fdb f = {
 395                 .state = NUD_STALE,
 396         };
 397         struct vxlan_rdst remote = { };
 398
 399         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 400
 401         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 402 }
 403
 404 /* Hash Ethernet address */
 405 static u32 eth_hash(const unsigned char *addr)
 406 {
 407         u64 value = get_unaligned((u64 *)addr);
 408
 409         /* only want 6 bytes */
 410 #ifdef __BIG_ENDIAN
 411         value >>= 16;
 412 #else
 413         value <<= 16;
 414 #endif
 415         return hash_64(value, FDB_HASH_BITS);
 416 }
 417
 418 /* Hash chain to use given mac address */
 419 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 420                                                 const u8 *mac)
 421 {
 422         return &vxlan->fdb_head[eth_hash(mac)];
 423 }
 424
 425 /* Look up Ethernet address in forwarding table */
 426 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 427                                         const u8 *mac)
 428 {
 429         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 430         struct vxlan_fdb *f;
 431
 432         hlist_for_each_entry_rcu(f, head, hlist) {
 433                 if (ether_addr_equal(mac, f->eth_addr))
 434                         return f;
 435         }
 436
 437         return NULL;
 438 }
 439
 440 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 441                                         const u8 *mac)
 442 {
 443         struct vxlan_fdb *f;
 444
 445         f = __vxlan_find_mac(vxlan, mac);
 446         if (f)
 447                 f->used = jiffies;
 448
 449         return f;
 450 }
 451
 452 /* caller should hold vxlan->hash_lock */
 453 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 454                                               union vxlan_addr *ip, __be16 port,
 455                                               __be32 vni, __u32 ifindex)
 456 {
 457         struct vxlan_rdst *rd;
 458
 459         list_for_each_entry(rd, &f->remotes, list) {
 460                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 461                     rd->remote_port == port &&
 462                     rd->remote_vni == vni &&
 463                     rd->remote_ifindex == ifindex)
 464                         return rd;
 465         }
 466
 467         return NULL;
 468 }
 469
 470 /* Replace destination of unicast mac */
 471 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 472                              union vxlan_addr *ip, __be16 port, __be32 vni,
 473                              __u32 ifindex)
 474 {
 475         struct vxlan_rdst *rd;
 476
 477         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 478         if (rd)
 479                 return 0;
 480
 481         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 482         if (!rd)
 483                 return 0;
 484
 485         dst_cache_reset(&rd->dst_cache);
 486         rd->remote_ip = *ip;
 487         rd->remote_port = port;
 488         rd->remote_vni = vni;
 489         rd->remote_ifindex = ifindex;
 490         return 1;
 491 }
 492
 493 /* Add/update destinations for multicast */
 494 static int vxlan_fdb_append(struct vxlan_fdb *f,
 495                             union vxlan_addr *ip, __be16 port, __be32 vni,
 496                             __u32 ifindex, struct vxlan_rdst **rdp)
 497 {
 498         struct vxlan_rdst *rd;
 499
 500         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 501         if (rd)
 502                 return 0;
 503
 504         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 505         if (rd == NULL)
 506                 return -ENOBUFS;
 507
 508         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 509                 kfree(rd);
 510                 return -ENOBUFS;
 511         }
 512
 513         rd->remote_ip = *ip;
 514         rd->remote_port = port;
 515         rd->remote_vni = vni;
 516         rd->remote_ifindex = ifindex;
 517
 518         list_add_tail_rcu(&rd->list, &f->remotes);
 519
 520         *rdp = rd;
 521         return 1;
 522 }
 523
 524 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 525                                           unsigned int off,
 526                                           struct vxlanhdr *vh, size_t hdrlen,
 527                                           __be32 vni_field,
 528                                           struct gro_remcsum *grc,
 529                                           bool nopartial)
 530 {
 531         size_t start, offset;
 532
 533         if (skb->remcsum_offload)
 534                 return vh;
 535
 536         if (!NAPI_GRO_CB(skb)->csum_valid)
 537                 return NULL;
 538
 539         start = vxlan_rco_start(vni_field);
 540         offset = start + vxlan_rco_offset(vni_field);
 541
 542         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 543                                      start, offset, grc, nopartial);
 544
 545         skb->remcsum_offload = 1;
 546
 547         return vh;
 548 }
 549
 550 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
 551                                           struct sk_buff *skb,
 552                                           struct udp_offload *uoff)
 553 {
 554         struct sk_buff *p, **pp = NULL;
 555         struct vxlanhdr *vh, *vh2;
 556         unsigned int hlen, off_vx;
 557         int flush = 1;
 558         struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
 559                                              udp_offloads);
 560         __be32 flags;
 561         struct gro_remcsum grc;
 562
 563         skb_gro_remcsum_init(&grc);
 564
 565         off_vx = skb_gro_offset(skb);
 566         hlen = off_vx + sizeof(*vh);
 567         vh   = skb_gro_header_fast(skb, off_vx);
 568         if (skb_gro_header_hard(skb, hlen)) {
 569                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 570                 if (unlikely(!vh))
 571                         goto out;
 572         }
 573
 574         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 575
 576         flags = vh->vx_flags;
 577
 578         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 579                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 580                                        vh->vx_vni, &grc,
 581                                        !!(vs->flags &
 582                                           VXLAN_F_REMCSUM_NOPARTIAL));
 583
 584                 if (!vh)
 585                         goto out;
 586         }
 587
 588         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 589
 590         flush = 0;
 591
 592         for (p = *head; p; p = p->next) {
 593                 if (!NAPI_GRO_CB(p)->same_flow)
 594                         continue;
 595
 596                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 597                 if (vh->vx_flags != vh2->vx_flags ||
 598                     vh->vx_vni != vh2->vx_vni) {
 599                         NAPI_GRO_CB(p)->same_flow = 0;
 600                         continue;
 601                 }
 602         }
 603
 604         pp = eth_gro_receive(head, skb);
 605
 606 out:
 607         skb_gro_remcsum_cleanup(skb, &grc);
 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                               struct udp_offload *uoff)
 615 {
 616         udp_tunnel_gro_complete(skb, nhoff);
 617
 618         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 619 }
 620
 621 /* Notify netdevs that UDP port started listening */
 622 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
 623 {
 624         struct net_device *dev;
 625         struct sock *sk = vs->sock->sk;
 626         struct net *net = sock_net(sk);
 627         sa_family_t sa_family = vxlan_get_sk_family(vs);
 628         __be16 port = inet_sk(sk)->inet_sport;
 629         int err;
 630
 631         if (sa_family == AF_INET) {
 632                 err = udp_add_offload(net, &vs->udp_offloads);
 633                 if (err)
 634                         pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
 635         }
 636
 637         rcu_read_lock();
 638         for_each_netdev_rcu(net, dev) {
 639                 if (dev->netdev_ops->ndo_add_vxlan_port)
 640                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
 641                                                             port);
 642         }
 643         rcu_read_unlock();
 644 }
 645
 646 /* Notify netdevs that UDP port is no more listening */
 647 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
 648 {
 649         struct net_device *dev;
 650         struct sock *sk = vs->sock->sk;
 651         struct net *net = sock_net(sk);
 652         sa_family_t sa_family = vxlan_get_sk_family(vs);
 653         __be16 port = inet_sk(sk)->inet_sport;
 654
 655         rcu_read_lock();
 656         for_each_netdev_rcu(net, dev) {
 657                 if (dev->netdev_ops->ndo_del_vxlan_port)
 658                         dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
 659                                                             port);
 660         }
 661         rcu_read_unlock();
 662
 663         if (sa_family == AF_INET)
 664                 udp_del_offload(&vs->udp_offloads);
 665 }
 666
 667 /* Add new entry to forwarding table -- assumes lock held */
 668 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 669                             const u8 *mac, union vxlan_addr *ip,
 670                             __u16 state, __u16 flags,
 671                             __be16 port, __be32 vni, __u32 ifindex,
 672                             __u8 ndm_flags)
 673 {
 674         struct vxlan_rdst *rd = NULL;
 675         struct vxlan_fdb *f;
 676         int notify = 0;
 677
 678         f = __vxlan_find_mac(vxlan, mac);
 679         if (f) {
 680                 if (flags & NLM_F_EXCL) {
 681                         netdev_dbg(vxlan->dev,
 682                                    "lost race to create %pM\n", mac);
 683                         return -EEXIST;
 684                 }
 685                 if (f->state != state) {
 686                         f->state = state;
 687                         f->updated = jiffies;
 688                         notify = 1;
 689                 }
 690                 if (f->flags != ndm_flags) {
 691                         f->flags = ndm_flags;
 692                         f->updated = jiffies;
 693                         notify = 1;
 694                 }
 695                 if ((flags & NLM_F_REPLACE)) {
 696                         /* Only change unicasts */
 697                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 698                              is_zero_ether_addr(f->eth_addr))) {
 699                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 700                                                            ifindex);
 701                         } else
 702                                 return -EOPNOTSUPP;
 703                 }
 704                 if ((flags & NLM_F_APPEND) &&
 705                     (is_multicast_ether_addr(f->eth_addr) ||
 706                      is_zero_ether_addr(f->eth_addr))) {
 707                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 708                                                   &rd);
 709
 710                         if (rc < 0)
 711                                 return rc;
 712                         notify |= rc;
 713                 }
 714         } else {
 715                 if (!(flags & NLM_F_CREATE))
 716                         return -ENOENT;
 717
 718                 if (vxlan->cfg.addrmax &&
 719                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 720                         return -ENOSPC;
 721
 722                 /* Disallow replace to add a multicast entry */
 723                 if ((flags & NLM_F_REPLACE) &&
 724                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 725                         return -EOPNOTSUPP;
 726
 727                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 728                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 729                 if (!f)
 730                         return -ENOMEM;
 731
 732                 notify = 1;
 733                 f->state = state;
 734                 f->flags = ndm_flags;
 735                 f->updated = f->used = jiffies;
 736                 INIT_LIST_HEAD(&f->remotes);
 737                 memcpy(f->eth_addr, mac, ETH_ALEN);
 738
 739                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 740
 741                 ++vxlan->addrcnt;
 742                 hlist_add_head_rcu(&f->hlist,
 743                                    vxlan_fdb_head(vxlan, mac));
 744         }
 745
 746         if (notify) {
 747                 if (rd == NULL)
 748                         rd = first_remote_rtnl(f);
 749                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 750         }
 751
 752         return 0;
 753 }
 754
 755 static void vxlan_fdb_free(struct rcu_head *head)
 756 {
 757         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 758         struct vxlan_rdst *rd, *nd;
 759
 760         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 761                 dst_cache_destroy(&rd->dst_cache);
 762                 kfree(rd);
 763         }
 764         kfree(f);
 765 }
 766
 767 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 768 {
 769         netdev_dbg(vxlan->dev,
 770                     "delete %pM\n", f->eth_addr);
 771
 772         --vxlan->addrcnt;
 773         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 774
 775         hlist_del_rcu(&f->hlist);
 776         call_rcu(&f->rcu, vxlan_fdb_free);
 777 }
 778
 779 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 780                            union vxlan_addr *ip, __be16 *port, __be32 *vni,
 781                            u32 *ifindex)
 782 {
 783         struct net *net = dev_net(vxlan->dev);
 784         int err;
 785
 786         if (tb[NDA_DST]) {
 787                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 788                 if (err)
 789                         return err;
 790         } else {
 791                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 792                 if (remote->sa.sa_family == AF_INET) {
 793                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 794                         ip->sa.sa_family = AF_INET;
 795 #if IS_ENABLED(CONFIG_IPV6)
 796                 } else {
 797                         ip->sin6.sin6_addr = in6addr_any;
 798                         ip->sa.sa_family = AF_INET6;
 799 #endif
 800                 }
 801         }
 802
 803         if (tb[NDA_PORT]) {
 804                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 805                         return -EINVAL;
 806                 *port = nla_get_be16(tb[NDA_PORT]);
 807         } else {
 808                 *port = vxlan->cfg.dst_port;
 809         }
 810
 811         if (tb[NDA_VNI]) {
 812                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 813                         return -EINVAL;
 814                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 815         } else {
 816                 *vni = vxlan->default_dst.remote_vni;
 817         }
 818
 819         if (tb[NDA_IFINDEX]) {
 820                 struct net_device *tdev;
 821
 822                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 823                         return -EINVAL;
 824                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 825                 tdev = __dev_get_by_index(net, *ifindex);
 826                 if (!tdev)
 827                         return -EADDRNOTAVAIL;
 828         } else {
 829                 *ifindex = 0;
 830         }
 831
 832         return 0;
 833 }
 834
 835 /* Add static entry (via netlink) */
 836 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 837                          struct net_device *dev,
 838                          const unsigned char *addr, u16 vid, u16 flags)
 839 {
 840         struct vxlan_dev *vxlan = netdev_priv(dev);
 841         /* struct net *net = dev_net(vxlan->dev); */
 842         union vxlan_addr ip;
 843         __be16 port;
 844         __be32 vni;
 845         u32 ifindex;
 846         int err;
 847
 848         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 849                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 850                         ndm->ndm_state);
 851                 return -EINVAL;
 852         }
 853
 854         if (tb[NDA_DST] == NULL)
 855                 return -EINVAL;
 856
 857         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 858         if (err)
 859                 return err;
 860
 861         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 862                 return -EAFNOSUPPORT;
 863
 864         spin_lock_bh(&vxlan->hash_lock);
 865         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 866                                port, vni, ifindex, ndm->ndm_flags);
 867         spin_unlock_bh(&vxlan->hash_lock);
 868
 869         return err;
 870 }
 871
 872 /* Delete entry (via netlink) */
 873 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 874                             struct net_device *dev,
 875                             const unsigned char *addr, u16 vid)
 876 {
 877         struct vxlan_dev *vxlan = netdev_priv(dev);
 878         struct vxlan_fdb *f;
 879         struct vxlan_rdst *rd = NULL;
 880         union vxlan_addr ip;
 881         __be16 port;
 882         __be32 vni;
 883         u32 ifindex;
 884         int err;
 885
 886         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 887         if (err)
 888                 return err;
 889
 890         err = -ENOENT;
 891
 892         spin_lock_bh(&vxlan->hash_lock);
 893         f = vxlan_find_mac(vxlan, addr);
 894         if (!f)
 895                 goto out;
 896
 897         if (!vxlan_addr_any(&ip)) {
 898                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 899                 if (!rd)
 900                         goto out;
 901         }
 902
 903         err = 0;
 904
 905         /* remove a destination if it's not the only one on the list,
 906          * otherwise destroy the fdb entry
 907          */
 908         if (rd && !list_is_singular(&f->remotes)) {
 909                 list_del_rcu(&rd->list);
 910                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 911                 kfree_rcu(rd, rcu);
 912                 goto out;
 913         }
 914
 915         vxlan_fdb_destroy(vxlan, f);
 916
 917 out:
 918         spin_unlock_bh(&vxlan->hash_lock);
 919
 920         return err;
 921 }
 922
 923 /* Dump forwarding table */
 924 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 925                           struct net_device *dev,
 926                           struct net_device *filter_dev, int idx)
 927 {
 928         struct vxlan_dev *vxlan = netdev_priv(dev);
 929         unsigned int h;
 930
 931         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 932                 struct vxlan_fdb *f;
 933                 int err;
 934
 935                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 936                         struct vxlan_rdst *rd;
 937
 938                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 939                                 if (idx < cb->args[0])
 940                                         goto skip;
 941
 942                                 err = vxlan_fdb_info(skb, vxlan, f,
 943                                                      NETLINK_CB(cb->skb).portid,
 944                                                      cb->nlh->nlmsg_seq,
 945                                                      RTM_NEWNEIGH,
 946                                                      NLM_F_MULTI, rd);
 947                                 if (err < 0)
 948                                         goto out;
 949 skip:
 950                                 ++idx;
 951                         }
 952                 }
 953         }
 954 out:
 955         return idx;
 956 }
 957
 958 /* Watch incoming packets to learn mapping between Ethernet address
 959  * and Tunnel endpoint.
 960  * Return true if packet is bogus and should be dropped.
 961  */
 962 static bool vxlan_snoop(struct net_device *dev,
 963                         union vxlan_addr *src_ip, const u8 *src_mac)
 964 {
 965         struct vxlan_dev *vxlan = netdev_priv(dev);
 966         struct vxlan_fdb *f;
 967
 968         f = vxlan_find_mac(vxlan, src_mac);
 969         if (likely(f)) {
 970                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 971
 972                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 973                         return false;
 974
 975                 /* Don't migrate static entries, drop packets */
 976                 if (f->state & NUD_NOARP)
 977                         return true;
 978
 979                 if (net_ratelimit())
 980                         netdev_info(dev,
 981                                     "%pM migrated from %pIS to %pIS\n",
 982                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 983
 984                 rdst->remote_ip = *src_ip;
 985                 f->updated = jiffies;
 986                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 987         } else {
 988                 /* learned new entry */
 989                 spin_lock(&vxlan->hash_lock);
 990
 991                 /* close off race between vxlan_flush and incoming packets */
 992                 if (netif_running(dev))
 993                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 994                                          NUD_REACHABLE,
 995                                          NLM_F_EXCL|NLM_F_CREATE,
 996                                          vxlan->cfg.dst_port,
 997                                          vxlan->default_dst.remote_vni,
 998                                          0, NTF_SELF);
 999                 spin_unlock(&vxlan->hash_lock);
1000         }
1001
1002         return false;
1003 }
1004
1005 /* See if multicast group is already in use by other ID */
1006 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1007 {
1008         struct vxlan_dev *vxlan;
1009         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1010
1011         /* The vxlan_sock is only used by dev, leaving group has
1012          * no effect on other vxlan devices.
1013          */
1014         if (family == AF_INET && dev->vn4_sock &&
1015             atomic_read(&dev->vn4_sock->refcnt) == 1)
1016                 return false;
1017 #if IS_ENABLED(CONFIG_IPV6)
1018         if (family == AF_INET6 && dev->vn6_sock &&
1019             atomic_read(&dev->vn6_sock->refcnt) == 1)
1020                 return false;
1021 #endif
1022
1023         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1024                 if (!netif_running(vxlan->dev) || vxlan == dev)
1025                         continue;
1026
1027                 if (family == AF_INET && vxlan->vn4_sock != dev->vn4_sock)
1028                         continue;
1029 #if IS_ENABLED(CONFIG_IPV6)
1030                 if (family == AF_INET6 && vxlan->vn6_sock != dev->vn6_sock)
1031                         continue;
1032 #endif
1033
1034                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1035                                       &dev->default_dst.remote_ip))
1036                         continue;
1037
1038                 if (vxlan->default_dst.remote_ifindex !=
1039                     dev->default_dst.remote_ifindex)
1040                         continue;
1041
1042                 return true;
1043         }
1044
1045         return false;
1046 }
1047
1048 static void __vxlan_sock_release(struct vxlan_sock *vs)
1049 {
1050         struct vxlan_net *vn;
1051
1052         if (!vs)
1053                 return;
1054         if (!atomic_dec_and_test(&vs->refcnt))
1055                 return;
1056
1057         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1058         spin_lock(&vn->sock_lock);
1059         hlist_del_rcu(&vs->hlist);
1060         vxlan_notify_del_rx_port(vs);
1061         spin_unlock(&vn->sock_lock);
1062
1063         queue_work(vxlan_wq, &vs->del_work);
1064 }
1065
1066 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1067 {
1068         __vxlan_sock_release(vxlan->vn4_sock);
1069 #if IS_ENABLED(CONFIG_IPV6)
1070         __vxlan_sock_release(vxlan->vn6_sock);
1071 #endif
1072 }
1073
1074 /* Update multicast group membership when first VNI on
1075  * multicast address is brought up
1076  */
1077 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1078 {
1079         struct sock *sk;
1080         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1081         int ifindex = vxlan->default_dst.remote_ifindex;
1082         int ret = -EINVAL;
1083
1084         if (ip->sa.sa_family == AF_INET) {
1085                 struct ip_mreqn mreq = {
1086                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1087                         .imr_ifindex            = ifindex,
1088                 };
1089
1090                 sk = vxlan->vn4_sock->sock->sk;
1091                 lock_sock(sk);
1092                 ret = ip_mc_join_group(sk, &mreq);
1093                 release_sock(sk);
1094 #if IS_ENABLED(CONFIG_IPV6)
1095         } else {
1096                 sk = vxlan->vn6_sock->sock->sk;
1097                 lock_sock(sk);
1098                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1099                                                    &ip->sin6.sin6_addr);
1100                 release_sock(sk);
1101 #endif
1102         }
1103
1104         return ret;
1105 }
1106
1107 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1108 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1109 {
1110         struct sock *sk;
1111         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1112         int ifindex = vxlan->default_dst.remote_ifindex;
1113         int ret = -EINVAL;
1114
1115         if (ip->sa.sa_family == AF_INET) {
1116                 struct ip_mreqn mreq = {
1117                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1118                         .imr_ifindex            = ifindex,
1119                 };
1120
1121                 sk = vxlan->vn4_sock->sock->sk;
1122                 lock_sock(sk);
1123                 ret = ip_mc_leave_group(sk, &mreq);
1124                 release_sock(sk);
1125 #if IS_ENABLED(CONFIG_IPV6)
1126         } else {
1127                 sk = vxlan->vn6_sock->sock->sk;
1128                 lock_sock(sk);
1129                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1130                                                    &ip->sin6.sin6_addr);
1131                 release_sock(sk);
1132 #endif
1133         }
1134
1135         return ret;
1136 }
1137
1138 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1139                           struct sk_buff *skb, u32 vxflags)
1140 {
1141         size_t start, offset, plen;
1142
1143         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1144                 goto out;
1145
1146         start = vxlan_rco_start(unparsed->vx_vni);
1147         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1148
1149         plen = sizeof(struct vxlanhdr) + offset + sizeof(u16);
1150
1151         if (!pskb_may_pull(skb, plen))
1152                 return false;
1153
1154         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1155                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1156 out:
1157         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1158         unparsed->vx_vni &= VXLAN_VNI_MASK;
1159         return true;
1160 }
1161
1162 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1163                                 struct vxlan_metadata *md,
1164                                 struct metadata_dst *tun_dst)
1165 {
1166         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1167
1168         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1169                 goto out;
1170
1171         md->gbp = ntohs(gbp->policy_id);
1172
1173         if (tun_dst)
1174                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1175
1176         if (gbp->dont_learn)
1177                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1178
1179         if (gbp->policy_applied)
1180                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1181
1182 out:
1183         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1184 }
1185
1186 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1187                       struct vxlan_metadata *md, __be32 vni,
1188                       struct metadata_dst *tun_dst)
1189 {
1190         struct iphdr *oip = NULL;
1191         struct ipv6hdr *oip6 = NULL;
1192         struct vxlan_dev *vxlan;
1193         struct pcpu_sw_netstats *stats;
1194         union vxlan_addr saddr;
1195         int err = 0;
1196
1197         /* For flow based devices, map all packets to VNI 0 */
1198         if (vs->flags & VXLAN_F_COLLECT_METADATA)
1199                 vni = 0;
1200
1201         /* Is this VNI defined? */
1202         vxlan = vxlan_vs_find_vni(vs, vni);
1203         if (!vxlan)
1204                 goto drop;
1205
1206         skb_reset_mac_header(skb);
1207         skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1208         skb->protocol = eth_type_trans(skb, vxlan->dev);
1209         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1210
1211         /* Ignore packet loops (and multicast echo) */
1212         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1213                 goto drop;
1214
1215         /* Get data from the outer IP header */
1216         if (vxlan_get_sk_family(vs) == AF_INET) {
1217                 oip = ip_hdr(skb);
1218                 saddr.sin.sin_addr.s_addr = oip->saddr;
1219                 saddr.sa.sa_family = AF_INET;
1220 #if IS_ENABLED(CONFIG_IPV6)
1221         } else {
1222                 oip6 = ipv6_hdr(skb);
1223                 saddr.sin6.sin6_addr = oip6->saddr;
1224                 saddr.sa.sa_family = AF_INET6;
1225 #endif
1226         }
1227
1228         if (tun_dst) {
1229                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1230                 tun_dst = NULL;
1231         }
1232
1233         if ((vxlan->flags & VXLAN_F_LEARN) &&
1234             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1235                 goto drop;
1236
1237         skb_reset_network_header(skb);
1238         /* In flow-based mode, GBP is carried in dst_metadata */
1239         if (!(vs->flags & VXLAN_F_COLLECT_METADATA))
1240                 skb->mark = md->gbp;
1241
1242         if (oip6)
1243                 err = IP6_ECN_decapsulate(oip6, skb);
1244         if (oip)
1245                 err = IP_ECN_decapsulate(oip, skb);
1246
1247         if (unlikely(err)) {
1248                 if (log_ecn_error) {
1249                         if (oip6)
1250                                 net_info_ratelimited("non-ECT from %pI6\n",
1251                                                      &oip6->saddr);
1252                         if (oip)
1253                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1254                                                      &oip->saddr, oip->tos);
1255                 }
1256                 if (err > 1) {
1257                         ++vxlan->dev->stats.rx_frame_errors;
1258                         ++vxlan->dev->stats.rx_errors;
1259                         goto drop;
1260                 }
1261         }
1262
1263         stats = this_cpu_ptr(vxlan->dev->tstats);
1264         u64_stats_update_begin(&stats->syncp);
1265         stats->rx_packets++;
1266         stats->rx_bytes += skb->len;
1267         u64_stats_update_end(&stats->syncp);
1268
1269         gro_cells_receive(&vxlan->gro_cells, skb);
1270
1271         return;
1272 drop:
1273         if (tun_dst)
1274                 dst_release((struct dst_entry *)tun_dst);
1275
1276         /* Consume bad packet */
1277         kfree_skb(skb);
1278 }
1279
1280 /* Callback from net/ipv4/udp.c to receive packets */
1281 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1282 {
1283         struct metadata_dst *tun_dst = NULL;
1284         struct vxlan_sock *vs;
1285         struct vxlanhdr unparsed;
1286         struct vxlan_metadata _md;
1287         struct vxlan_metadata *md = &_md;
1288
1289         /* Need Vxlan and inner Ethernet header to be present */
1290         if (!pskb_may_pull(skb, VXLAN_HLEN))
1291                 goto error;
1292
1293         unparsed = *vxlan_hdr(skb);
1294         if (unparsed.vx_flags & VXLAN_HF_VNI) {
1295                 unparsed.vx_flags &= ~VXLAN_HF_VNI;
1296                 unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1297         } else {
1298                 /* VNI flag always required to be set */
1299                 goto bad_flags;
1300         }
1301
1302         if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1303                 goto drop;
1304
1305         vs = rcu_dereference_sk_user_data(sk);
1306         if (!vs)
1307                 goto drop;
1308
1309         if (vxlan_collect_metadata(vs)) {
1310                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1311                                          vxlan_vni(vxlan_hdr(skb)->vx_vni),
1312                                          sizeof(*md));
1313
1314                 if (!tun_dst)
1315                         goto drop;
1316
1317                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1318         } else {
1319                 memset(md, 0, sizeof(*md));
1320         }
1321
1322         /* For backwards compatibility, only allow reserved fields to be
1323          * used by VXLAN extensions if explicitly requested.
1324          */
1325         if (vs->flags & VXLAN_F_REMCSUM_RX)
1326                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1327                         goto drop;
1328         if (vs->flags & VXLAN_F_GBP)
1329                 vxlan_parse_gbp_hdr(&unparsed, md, tun_dst);
1330
1331         if (unparsed.vx_flags || unparsed.vx_vni) {
1332                 /* If there are any unprocessed flags remaining treat
1333                  * this as a malformed packet. This behavior diverges from
1334                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1335                  * in reserved fields are to be ignored. The approach here
1336                  * maintains compatibility with previous stack code, and also
1337                  * is more robust and provides a little more security in
1338                  * adding extensions to VXLAN.
1339                  */
1340
1341                 goto bad_flags;
1342         }
1343
1344         vxlan_rcv(vs, skb, md, vxlan_vni(vxlan_hdr(skb)->vx_vni), tun_dst);
1345         return 0;
1346
1347 drop:
1348         /* Consume bad packet */
1349         kfree_skb(skb);
1350         return 0;
1351
1352 bad_flags:
1353         netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1354                    ntohl(vxlan_hdr(skb)->vx_flags),
1355                    ntohl(vxlan_hdr(skb)->vx_vni));
1356
1357 error:
1358         if (tun_dst)
1359                 dst_release((struct dst_entry *)tun_dst);
1360
1361         /* Return non vxlan pkt */
1362         return 1;
1363 }
1364
1365 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1366 {
1367         struct vxlan_dev *vxlan = netdev_priv(dev);
1368         struct arphdr *parp;
1369         u8 *arpptr, *sha;
1370         __be32 sip, tip;
1371         struct neighbour *n;
1372
1373         if (dev->flags & IFF_NOARP)
1374                 goto out;
1375
1376         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1377                 dev->stats.tx_dropped++;
1378                 goto out;
1379         }
1380         parp = arp_hdr(skb);
1381
1382         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1383              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1384             parp->ar_pro != htons(ETH_P_IP) ||
1385             parp->ar_op != htons(ARPOP_REQUEST) ||
1386             parp->ar_hln != dev->addr_len ||
1387             parp->ar_pln != 4)
1388                 goto out;
1389         arpptr = (u8 *)parp + sizeof(struct arphdr);
1390         sha = arpptr;
1391         arpptr += dev->addr_len;        /* sha */
1392         memcpy(&sip, arpptr, sizeof(sip));
1393         arpptr += sizeof(sip);
1394         arpptr += dev->addr_len;        /* tha */
1395         memcpy(&tip, arpptr, sizeof(tip));
1396
1397         if (ipv4_is_loopback(tip) ||
1398             ipv4_is_multicast(tip))
1399                 goto out;
1400
1401         n = neigh_lookup(&arp_tbl, &tip, dev);
1402
1403         if (n) {
1404                 struct vxlan_fdb *f;
1405                 struct sk_buff  *reply;
1406
1407                 if (!(n->nud_state & NUD_CONNECTED)) {
1408                         neigh_release(n);
1409                         goto out;
1410                 }
1411
1412                 f = vxlan_find_mac(vxlan, n->ha);
1413                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1414                         /* bridge-local neighbor */
1415                         neigh_release(n);
1416                         goto out;
1417                 }
1418
1419                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1420                                 n->ha, sha);
1421
1422                 neigh_release(n);
1423
1424                 if (reply == NULL)
1425                         goto out;
1426
1427                 skb_reset_mac_header(reply);
1428                 __skb_pull(reply, skb_network_offset(reply));
1429                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1430                 reply->pkt_type = PACKET_HOST;
1431
1432                 if (netif_rx_ni(reply) == NET_RX_DROP)
1433                         dev->stats.rx_dropped++;
1434         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1435                 union vxlan_addr ipa = {
1436                         .sin.sin_addr.s_addr = tip,
1437                         .sin.sin_family = AF_INET,
1438                 };
1439
1440                 vxlan_ip_miss(dev, &ipa);
1441         }
1442 out:
1443         consume_skb(skb);
1444         return NETDEV_TX_OK;
1445 }
1446
1447 #if IS_ENABLED(CONFIG_IPV6)
1448 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1449         struct neighbour *n, bool isrouter)
1450 {
1451         struct net_device *dev = request->dev;
1452         struct sk_buff *reply;
1453         struct nd_msg *ns, *na;
1454         struct ipv6hdr *pip6;
1455         u8 *daddr;
1456         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1457         int ns_olen;
1458         int i, len;
1459
1460         if (dev == NULL)
1461                 return NULL;
1462
1463         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1464                 sizeof(*na) + na_olen + dev->needed_tailroom;
1465         reply = alloc_skb(len, GFP_ATOMIC);
1466         if (reply == NULL)
1467                 return NULL;
1468
1469         reply->protocol = htons(ETH_P_IPV6);
1470         reply->dev = dev;
1471         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1472         skb_push(reply, sizeof(struct ethhdr));
1473         skb_set_mac_header(reply, 0);
1474
1475         ns = (struct nd_msg *)skb_transport_header(request);
1476
1477         daddr = eth_hdr(request)->h_source;
1478         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1479         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1480                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1481                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1482                         break;
1483                 }
1484         }
1485
1486         /* Ethernet header */
1487         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1488         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1489         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1490         reply->protocol = htons(ETH_P_IPV6);
1491
1492         skb_pull(reply, sizeof(struct ethhdr));
1493         skb_set_network_header(reply, 0);
1494         skb_put(reply, sizeof(struct ipv6hdr));
1495
1496         /* IPv6 header */
1497
1498         pip6 = ipv6_hdr(reply);
1499         memset(pip6, 0, sizeof(struct ipv6hdr));
1500         pip6->version = 6;
1501         pip6->priority = ipv6_hdr(request)->priority;
1502         pip6->nexthdr = IPPROTO_ICMPV6;
1503         pip6->hop_limit = 255;
1504         pip6->daddr = ipv6_hdr(request)->saddr;
1505         pip6->saddr = *(struct in6_addr *)n->primary_key;
1506
1507         skb_pull(reply, sizeof(struct ipv6hdr));
1508         skb_set_transport_header(reply, 0);
1509
1510         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1511
1512         /* Neighbor Advertisement */
1513         memset(na, 0, sizeof(*na)+na_olen);
1514         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1515         na->icmph.icmp6_router = isrouter;
1516         na->icmph.icmp6_override = 1;
1517         na->icmph.icmp6_solicited = 1;
1518         na->target = ns->target;
1519         ether_addr_copy(&na->opt[2], n->ha);
1520         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1521         na->opt[1] = na_olen >> 3;
1522
1523         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1524                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1525                 csum_partial(na, sizeof(*na)+na_olen, 0));
1526
1527         pip6->payload_len = htons(sizeof(*na)+na_olen);
1528
1529         skb_push(reply, sizeof(struct ipv6hdr));
1530
1531         reply->ip_summed = CHECKSUM_UNNECESSARY;
1532
1533         return reply;
1534 }
1535
1536 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1537 {
1538         struct vxlan_dev *vxlan = netdev_priv(dev);
1539         struct nd_msg *msg;
1540         const struct ipv6hdr *iphdr;
1541         const struct in6_addr *saddr, *daddr;
1542         struct neighbour *n;
1543         struct inet6_dev *in6_dev;
1544
1545         in6_dev = __in6_dev_get(dev);
1546         if (!in6_dev)
1547                 goto out;
1548
1549         iphdr = ipv6_hdr(skb);
1550         saddr = &iphdr->saddr;
1551         daddr = &iphdr->daddr;
1552
1553         msg = (struct nd_msg *)skb_transport_header(skb);
1554         if (msg->icmph.icmp6_code != 0 ||
1555             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1556                 goto out;
1557
1558         if (ipv6_addr_loopback(daddr) ||
1559             ipv6_addr_is_multicast(&msg->target))
1560                 goto out;
1561
1562         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1563
1564         if (n) {
1565                 struct vxlan_fdb *f;
1566                 struct sk_buff *reply;
1567
1568                 if (!(n->nud_state & NUD_CONNECTED)) {
1569                         neigh_release(n);
1570                         goto out;
1571                 }
1572
1573                 f = vxlan_find_mac(vxlan, n->ha);
1574                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1575                         /* bridge-local neighbor */
1576                         neigh_release(n);
1577                         goto out;
1578                 }
1579
1580                 reply = vxlan_na_create(skb, n,
1581                                         !!(f ? f->flags & NTF_ROUTER : 0));
1582
1583                 neigh_release(n);
1584
1585                 if (reply == NULL)
1586                         goto out;
1587
1588                 if (netif_rx_ni(reply) == NET_RX_DROP)
1589                         dev->stats.rx_dropped++;
1590
1591         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1592                 union vxlan_addr ipa = {
1593                         .sin6.sin6_addr = msg->target,
1594                         .sin6.sin6_family = AF_INET6,
1595                 };
1596
1597                 vxlan_ip_miss(dev, &ipa);
1598         }
1599
1600 out:
1601         consume_skb(skb);
1602         return NETDEV_TX_OK;
1603 }
1604 #endif
1605
1606 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1607 {
1608         struct vxlan_dev *vxlan = netdev_priv(dev);
1609         struct neighbour *n;
1610
1611         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1612                 return false;
1613
1614         n = NULL;
1615         switch (ntohs(eth_hdr(skb)->h_proto)) {
1616         case ETH_P_IP:
1617         {
1618                 struct iphdr *pip;
1619
1620                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1621                         return false;
1622                 pip = ip_hdr(skb);
1623                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1624                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1625                         union vxlan_addr ipa = {
1626                                 .sin.sin_addr.s_addr = pip->daddr,
1627                                 .sin.sin_family = AF_INET,
1628                         };
1629
1630                         vxlan_ip_miss(dev, &ipa);
1631                         return false;
1632                 }
1633
1634                 break;
1635         }
1636 #if IS_ENABLED(CONFIG_IPV6)
1637         case ETH_P_IPV6:
1638         {
1639                 struct ipv6hdr *pip6;
1640
1641                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1642                         return false;
1643                 pip6 = ipv6_hdr(skb);
1644                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1645                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1646                         union vxlan_addr ipa = {
1647                                 .sin6.sin6_addr = pip6->daddr,
1648                                 .sin6.sin6_family = AF_INET6,
1649                         };
1650
1651                         vxlan_ip_miss(dev, &ipa);
1652                         return false;
1653                 }
1654
1655                 break;
1656         }
1657 #endif
1658         default:
1659                 return false;
1660         }
1661
1662         if (n) {
1663                 bool diff;
1664
1665                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1666                 if (diff) {
1667                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1668                                 dev->addr_len);
1669                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1670                 }
1671                 neigh_release(n);
1672                 return diff;
1673         }
1674
1675         return false;
1676 }
1677
1678 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1679                                 struct vxlan_metadata *md)
1680 {
1681         struct vxlanhdr_gbp *gbp;
1682
1683         if (!md->gbp)
1684                 return;
1685
1686         gbp = (struct vxlanhdr_gbp *)vxh;
1687         vxh->vx_flags |= VXLAN_HF_GBP;
1688
1689         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1690                 gbp->dont_learn = 1;
1691
1692         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1693                 gbp->policy_applied = 1;
1694
1695         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1696 }
1697
1698 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1699                            int iphdr_len, __be32 vni,
1700                            struct vxlan_metadata *md, u32 vxflags,
1701                            bool udp_sum)
1702 {
1703         struct vxlanhdr *vxh;
1704         int min_headroom;
1705         int err;
1706         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1707
1708         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1709             skb->ip_summed == CHECKSUM_PARTIAL) {
1710                 int csum_start = skb_checksum_start_offset(skb);
1711
1712                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1713                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1714                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1715                      skb->csum_offset == offsetof(struct tcphdr, check)))
1716                         type |= SKB_GSO_TUNNEL_REMCSUM;
1717         }
1718
1719         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1720                         + VXLAN_HLEN + iphdr_len
1721                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1722
1723         /* Need space for new headers (invalidates iph ptr) */
1724         err = skb_cow_head(skb, min_headroom);
1725         if (unlikely(err)) {
1726                 kfree_skb(skb);
1727                 return err;
1728         }
1729
1730         skb = vlan_hwaccel_push_inside(skb);
1731         if (WARN_ON(!skb))
1732                 return -ENOMEM;
1733
1734         skb = iptunnel_handle_offloads(skb, type);
1735         if (IS_ERR(skb))
1736                 return PTR_ERR(skb);
1737
1738         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1739         vxh->vx_flags = VXLAN_HF_VNI;
1740         vxh->vx_vni = vxlan_vni_field(vni);
1741
1742         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1743                 unsigned int start;
1744
1745                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1746                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1747                 vxh->vx_flags |= VXLAN_HF_RCO;
1748
1749                 if (!skb_is_gso(skb)) {
1750                         skb->ip_summed = CHECKSUM_NONE;
1751                         skb->encapsulation = 0;
1752                 }
1753         }
1754
1755         if (vxflags & VXLAN_F_GBP)
1756                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1757
1758         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1759         return 0;
1760 }
1761
1762 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1763                                       struct sk_buff *skb, int oif, u8 tos,
1764                                       __be32 daddr, __be32 *saddr,
1765                                       struct dst_cache *dst_cache,
1766                                       struct ip_tunnel_info *info)
1767 {
1768         struct rtable *rt = NULL;
1769         bool use_cache = false;
1770         struct flowi4 fl4;
1771
1772         /* when the ip_tunnel_info is availble, the tos used for lookup is
1773          * packet independent, so we can use the cache
1774          */
1775         if (!skb->mark && (!tos || info)) {
1776                 use_cache = true;
1777                 rt = dst_cache_get_ip4(dst_cache, saddr);
1778                 if (rt)
1779                         return rt;
1780         }
1781
1782         memset(&fl4, 0, sizeof(fl4));
1783         fl4.flowi4_oif = oif;
1784         fl4.flowi4_tos = RT_TOS(tos);
1785         fl4.flowi4_mark = skb->mark;
1786         fl4.flowi4_proto = IPPROTO_UDP;
1787         fl4.daddr = daddr;
1788         fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
1789
1790         rt = ip_route_output_key(vxlan->net, &fl4);
1791         if (!IS_ERR(rt)) {
1792                 *saddr = fl4.saddr;
1793                 if (use_cache)
1794                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1795         }
1796         return rt;
1797 }
1798
1799 #if IS_ENABLED(CONFIG_IPV6)
1800 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1801                                           struct sk_buff *skb, int oif,
1802                                           const struct in6_addr *daddr,
1803                                           struct in6_addr *saddr,
1804                                           struct dst_cache *dst_cache)
1805 {
1806         struct dst_entry *ndst;
1807         struct flowi6 fl6;
1808         int err;
1809
1810         if (!skb->mark) {
1811                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1812                 if (ndst)
1813                         return ndst;
1814         }
1815
1816         memset(&fl6, 0, sizeof(fl6));
1817         fl6.flowi6_oif = oif;
1818         fl6.daddr = *daddr;
1819         fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
1820         fl6.flowi6_mark = skb->mark;
1821         fl6.flowi6_proto = IPPROTO_UDP;
1822
1823         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1824                                          vxlan->vn6_sock->sock->sk,
1825                                          &ndst, &fl6);
1826         if (err < 0)
1827                 return ERR_PTR(err);
1828
1829         *saddr = fl6.saddr;
1830         if (!skb->mark)
1831                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1832         return ndst;
1833 }
1834 #endif
1835
1836 /* Bypass encapsulation if the destination is local */
1837 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1838                                struct vxlan_dev *dst_vxlan)
1839 {
1840         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1841         union vxlan_addr loopback;
1842         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1843         struct net_device *dev = skb->dev;
1844         int len = skb->len;
1845
1846         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1847         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1848         skb->pkt_type = PACKET_HOST;
1849         skb->encapsulation = 0;
1850         skb->dev = dst_vxlan->dev;
1851         __skb_pull(skb, skb_network_offset(skb));
1852
1853         if (remote_ip->sa.sa_family == AF_INET) {
1854                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1855                 loopback.sa.sa_family =  AF_INET;
1856 #if IS_ENABLED(CONFIG_IPV6)
1857         } else {
1858                 loopback.sin6.sin6_addr = in6addr_loopback;
1859                 loopback.sa.sa_family =  AF_INET6;
1860 #endif
1861         }
1862
1863         if (dst_vxlan->flags & VXLAN_F_LEARN)
1864                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1865
1866         u64_stats_update_begin(&tx_stats->syncp);
1867         tx_stats->tx_packets++;
1868         tx_stats->tx_bytes += len;
1869         u64_stats_update_end(&tx_stats->syncp);
1870
1871         if (netif_rx(skb) == NET_RX_SUCCESS) {
1872                 u64_stats_update_begin(&rx_stats->syncp);
1873                 rx_stats->rx_packets++;
1874                 rx_stats->rx_bytes += len;
1875                 u64_stats_update_end(&rx_stats->syncp);
1876         } else {
1877                 dev->stats.rx_dropped++;
1878         }
1879 }
1880
1881 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1882                            struct vxlan_rdst *rdst, bool did_rsc)
1883 {
1884         struct dst_cache *dst_cache;
1885         struct ip_tunnel_info *info;
1886         struct vxlan_dev *vxlan = netdev_priv(dev);
1887         struct sock *sk;
1888         struct rtable *rt = NULL;
1889         const struct iphdr *old_iph;
1890         union vxlan_addr *dst;
1891         union vxlan_addr remote_ip;
1892         struct vxlan_metadata _md;
1893         struct vxlan_metadata *md = &_md;
1894         __be16 src_port = 0, dst_port;
1895         __be32 vni;
1896         __be16 df = 0;
1897         __u8 tos, ttl;
1898         int err;
1899         u32 flags = vxlan->flags;
1900         bool udp_sum = false;
1901         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1902
1903         info = skb_tunnel_info(skb);
1904
1905         if (rdst) {
1906                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1907                 vni = rdst->remote_vni;
1908                 dst = &rdst->remote_ip;
1909                 dst_cache = &rdst->dst_cache;
1910         } else {
1911                 if (!info) {
1912                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1913                                   dev->name);
1914                         goto drop;
1915                 }
1916                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1917                 vni = vxlan_tun_id_to_vni(info->key.tun_id);
1918                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1919                 if (remote_ip.sa.sa_family == AF_INET)
1920                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1921                 else
1922                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1923                 dst = &remote_ip;
1924                 dst_cache = &info->dst_cache;
1925         }
1926
1927         if (vxlan_addr_any(dst)) {
1928                 if (did_rsc) {
1929                         /* short-circuited back to local bridge */
1930                         vxlan_encap_bypass(skb, vxlan, vxlan);
1931                         return;
1932                 }
1933                 goto drop;
1934         }
1935
1936         old_iph = ip_hdr(skb);
1937
1938         ttl = vxlan->cfg.ttl;
1939         if (!ttl && vxlan_addr_multicast(dst))
1940                 ttl = 1;
1941
1942         tos = vxlan->cfg.tos;
1943         if (tos == 1)
1944                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1945
1946         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
1947                                      vxlan->cfg.port_max, true);
1948
1949         if (info) {
1950                 ttl = info->key.ttl;
1951                 tos = info->key.tos;
1952                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
1953
1954                 if (info->options_len)
1955                         md = ip_tunnel_info_opts(info);
1956         } else {
1957                 md->gbp = skb->mark;
1958         }
1959
1960         if (dst->sa.sa_family == AF_INET) {
1961                 __be32 saddr;
1962
1963                 if (!vxlan->vn4_sock)
1964                         goto drop;
1965                 sk = vxlan->vn4_sock->sock->sk;
1966
1967                 if (info) {
1968                         if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
1969                                 df = htons(IP_DF);
1970                 } else {
1971                         udp_sum = !!(flags & VXLAN_F_UDP_CSUM);
1972                 }
1973
1974                 rt = vxlan_get_route(vxlan, skb,
1975                                      rdst ? rdst->remote_ifindex : 0, tos,
1976                                      dst->sin.sin_addr.s_addr, &saddr,
1977                                      dst_cache, info);
1978                 if (IS_ERR(rt)) {
1979                         netdev_dbg(dev, "no route to %pI4\n",
1980                                    &dst->sin.sin_addr.s_addr);
1981                         dev->stats.tx_carrier_errors++;
1982                         goto tx_error;
1983                 }
1984
1985                 if (rt->dst.dev == dev) {
1986                         netdev_dbg(dev, "circular route to %pI4\n",
1987                                    &dst->sin.sin_addr.s_addr);
1988                         dev->stats.collisions++;
1989                         goto rt_tx_error;
1990                 }
1991
1992                 /* Bypass encapsulation if the destination is local */
1993                 if (rt->rt_flags & RTCF_LOCAL &&
1994                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1995                         struct vxlan_dev *dst_vxlan;
1996
1997                         ip_rt_put(rt);
1998                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1999                                                    dst->sa.sa_family, dst_port,
2000                                                    vxlan->flags);
2001                         if (!dst_vxlan)
2002                                 goto tx_error;
2003                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2004                         return;
2005                 }
2006
2007                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2008                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2009                 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2010                                       vni, md, flags, udp_sum);
2011                 if (err < 0)
2012                         goto xmit_tx_error;
2013
2014                 udp_tunnel_xmit_skb(rt, sk, skb, saddr,
2015                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2016                                     src_port, dst_port, xnet, !udp_sum);
2017 #if IS_ENABLED(CONFIG_IPV6)
2018         } else {
2019                 struct dst_entry *ndst;
2020                 struct in6_addr saddr;
2021                 u32 rt6i_flags;
2022
2023                 if (!vxlan->vn6_sock)
2024                         goto drop;
2025                 sk = vxlan->vn6_sock->sock->sk;
2026
2027                 ndst = vxlan6_get_route(vxlan, skb,
2028                                         rdst ? rdst->remote_ifindex : 0,
2029                                         &dst->sin6.sin6_addr, &saddr,
2030                                         dst_cache);
2031                 if (IS_ERR(ndst)) {
2032                         netdev_dbg(dev, "no route to %pI6\n",
2033                                    &dst->sin6.sin6_addr);
2034                         dev->stats.tx_carrier_errors++;
2035                         goto tx_error;
2036                 }
2037
2038                 if (ndst->dev == dev) {
2039                         netdev_dbg(dev, "circular route to %pI6\n",
2040                                    &dst->sin6.sin6_addr);
2041                         dst_release(ndst);
2042                         dev->stats.collisions++;
2043                         goto tx_error;
2044                 }
2045
2046                 /* Bypass encapsulation if the destination is local */
2047                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2048                 if (rt6i_flags & RTF_LOCAL &&
2049                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2050                         struct vxlan_dev *dst_vxlan;
2051
2052                         dst_release(ndst);
2053                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2054                                                    dst->sa.sa_family, dst_port,
2055                                                    vxlan->flags);
2056                         if (!dst_vxlan)
2057                                 goto tx_error;
2058                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2059                         return;
2060                 }
2061
2062                 if (!info)
2063                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2064
2065                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2066                 skb_scrub_packet(skb, xnet);
2067                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2068                                       vni, md, flags, udp_sum);
2069                 if (err < 0) {
2070                         dst_release(ndst);
2071                         return;
2072                 }
2073                 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2074                                      &saddr, &dst->sin6.sin6_addr,
2075                                      0, ttl, src_port, dst_port, !udp_sum);
2076 #endif
2077         }
2078
2079         return;
2080
2081 drop:
2082         dev->stats.tx_dropped++;
2083         goto tx_free;
2084
2085 xmit_tx_error:
2086         /* skb is already freed. */
2087         skb = NULL;
2088 rt_tx_error:
2089         ip_rt_put(rt);
2090 tx_error:
2091         dev->stats.tx_errors++;
2092 tx_free:
2093         dev_kfree_skb(skb);
2094 }
2095
2096 /* Transmit local packets over Vxlan
2097  *
2098  * Outer IP header inherits ECN and DF from inner header.
2099  * Outer UDP destination is the VXLAN assigned port.
2100  *           source port is based on hash of flow
2101  */
2102 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2103 {
2104         struct vxlan_dev *vxlan = netdev_priv(dev);
2105         const struct ip_tunnel_info *info;
2106         struct ethhdr *eth;
2107         bool did_rsc = false;
2108         struct vxlan_rdst *rdst, *fdst = NULL;
2109         struct vxlan_fdb *f;
2110
2111         info = skb_tunnel_info(skb);
2112
2113         skb_reset_mac_header(skb);
2114         eth = eth_hdr(skb);
2115
2116         if ((vxlan->flags & VXLAN_F_PROXY)) {
2117                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2118                         return arp_reduce(dev, skb);
2119 #if IS_ENABLED(CONFIG_IPV6)
2120                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2121                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2122                                        + sizeof(struct nd_msg)) &&
2123                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2124                                 struct nd_msg *msg;
2125
2126                                 msg = (struct nd_msg *)skb_transport_header(skb);
2127                                 if (msg->icmph.icmp6_code == 0 &&
2128                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2129                                         return neigh_reduce(dev, skb);
2130                 }
2131                 eth = eth_hdr(skb);
2132 #endif
2133         }
2134
2135         if (vxlan->flags & VXLAN_F_COLLECT_METADATA &&
2136             info && info->mode & IP_TUNNEL_INFO_TX) {
2137                 vxlan_xmit_one(skb, dev, NULL, false);
2138                 return NETDEV_TX_OK;
2139         }
2140
2141         f = vxlan_find_mac(vxlan, eth->h_dest);
2142         did_rsc = false;
2143
2144         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2145             (ntohs(eth->h_proto) == ETH_P_IP ||
2146              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2147                 did_rsc = route_shortcircuit(dev, skb);
2148                 if (did_rsc)
2149                         f = vxlan_find_mac(vxlan, eth->h_dest);
2150         }
2151
2152         if (f == NULL) {
2153                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2154                 if (f == NULL) {
2155                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2156                             !is_multicast_ether_addr(eth->h_dest))
2157                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2158
2159                         dev->stats.tx_dropped++;
2160                         kfree_skb(skb);
2161                         return NETDEV_TX_OK;
2162                 }
2163         }
2164
2165         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2166                 struct sk_buff *skb1;
2167
2168                 if (!fdst) {
2169                         fdst = rdst;
2170                         continue;
2171                 }
2172                 skb1 = skb_clone(skb, GFP_ATOMIC);
2173                 if (skb1)
2174                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2175         }
2176
2177         if (fdst)
2178                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2179         else
2180                 kfree_skb(skb);
2181         return NETDEV_TX_OK;
2182 }
2183
2184 /* Walk the forwarding table and purge stale entries */
2185 static void vxlan_cleanup(unsigned long arg)
2186 {
2187         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2188         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2189         unsigned int h;
2190
2191         if (!netif_running(vxlan->dev))
2192                 return;
2193
2194         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2195                 struct hlist_node *p, *n;
2196
2197                 spin_lock_bh(&vxlan->hash_lock);
2198                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2199                         struct vxlan_fdb *f
2200                                 = container_of(p, struct vxlan_fdb, hlist);
2201                         unsigned long timeout;
2202
2203                         if (f->state & NUD_PERMANENT)
2204                                 continue;
2205
2206                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2207                         if (time_before_eq(timeout, jiffies)) {
2208                                 netdev_dbg(vxlan->dev,
2209                                            "garbage collect %pM\n",
2210                                            f->eth_addr);
2211                                 f->state = NUD_STALE;
2212                                 vxlan_fdb_destroy(vxlan, f);
2213                         } else if (time_before(timeout, next_timer))
2214                                 next_timer = timeout;
2215                 }
2216                 spin_unlock_bh(&vxlan->hash_lock);
2217         }
2218
2219         mod_timer(&vxlan->age_timer, next_timer);
2220 }
2221
2222 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2223 {
2224         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2225         __be32 vni = vxlan->default_dst.remote_vni;
2226
2227         spin_lock(&vn->sock_lock);
2228         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2229         spin_unlock(&vn->sock_lock);
2230 }
2231
2232 /* Setup stats when device is created */
2233 static int vxlan_init(struct net_device *dev)
2234 {
2235         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2236         if (!dev->tstats)
2237                 return -ENOMEM;
2238
2239         return 0;
2240 }
2241
2242 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2243 {
2244         struct vxlan_fdb *f;
2245
2246         spin_lock_bh(&vxlan->hash_lock);
2247         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2248         if (f)
2249                 vxlan_fdb_destroy(vxlan, f);
2250         spin_unlock_bh(&vxlan->hash_lock);
2251 }
2252
2253 static void vxlan_uninit(struct net_device *dev)
2254 {
2255         struct vxlan_dev *vxlan = netdev_priv(dev);
2256
2257         vxlan_fdb_delete_default(vxlan);
2258
2259         free_percpu(dev->tstats);
2260 }
2261
2262 /* Start ageing timer and join group when device is brought up */
2263 static int vxlan_open(struct net_device *dev)
2264 {
2265         struct vxlan_dev *vxlan = netdev_priv(dev);
2266         int ret;
2267
2268         ret = vxlan_sock_add(vxlan);
2269         if (ret < 0)
2270                 return ret;
2271
2272         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2273                 ret = vxlan_igmp_join(vxlan);
2274                 if (ret == -EADDRINUSE)
2275                         ret = 0;
2276                 if (ret) {
2277                         vxlan_sock_release(vxlan);
2278                         return ret;
2279                 }
2280         }
2281
2282         if (vxlan->cfg.age_interval)
2283                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2284
2285         return ret;
2286 }
2287
2288 /* Purge the forwarding table */
2289 static void vxlan_flush(struct vxlan_dev *vxlan)
2290 {
2291         unsigned int h;
2292
2293         spin_lock_bh(&vxlan->hash_lock);
2294         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2295                 struct hlist_node *p, *n;
2296                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2297                         struct vxlan_fdb *f
2298                                 = container_of(p, struct vxlan_fdb, hlist);
2299                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2300                         if (!is_zero_ether_addr(f->eth_addr))
2301                                 vxlan_fdb_destroy(vxlan, f);
2302                 }
2303         }
2304         spin_unlock_bh(&vxlan->hash_lock);
2305 }
2306
2307 /* Cleanup timer and forwarding table on shutdown */
2308 static int vxlan_stop(struct net_device *dev)
2309 {
2310         struct vxlan_dev *vxlan = netdev_priv(dev);
2311         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2312         int ret = 0;
2313
2314         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2315             !vxlan_group_used(vn, vxlan))
2316                 ret = vxlan_igmp_leave(vxlan);
2317
2318         del_timer_sync(&vxlan->age_timer);
2319
2320         vxlan_flush(vxlan);
2321         vxlan_sock_release(vxlan);
2322
2323         return ret;
2324 }
2325
2326 /* Stub, nothing needs to be done. */
2327 static void vxlan_set_multicast_list(struct net_device *dev)
2328 {
2329 }
2330
2331 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2332 {
2333         struct vxlan_dev *vxlan = netdev_priv(dev);
2334         struct vxlan_rdst *dst = &vxlan->default_dst;
2335         struct net_device *lowerdev;
2336         int max_mtu;
2337
2338         lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2339         if (lowerdev == NULL)
2340                 return eth_change_mtu(dev, new_mtu);
2341
2342         if (dst->remote_ip.sa.sa_family == AF_INET6)
2343                 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2344         else
2345                 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2346
2347         if (new_mtu < 68 || new_mtu > max_mtu)
2348                 return -EINVAL;
2349
2350         dev->mtu = new_mtu;
2351         return 0;
2352 }
2353
2354 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2355 {
2356         struct vxlan_dev *vxlan = netdev_priv(dev);
2357         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2358         __be16 sport, dport;
2359
2360         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2361                                   vxlan->cfg.port_max, true);
2362         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2363
2364         if (ip_tunnel_info_af(info) == AF_INET) {
2365                 struct rtable *rt;
2366
2367                 if (!vxlan->vn4_sock)
2368                         return -EINVAL;
2369                 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2370                                      info->key.u.ipv4.dst,
2371                                      &info->key.u.ipv4.src, NULL, info);
2372                 if (IS_ERR(rt))
2373                         return PTR_ERR(rt);
2374                 ip_rt_put(rt);
2375         } else {
2376 #if IS_ENABLED(CONFIG_IPV6)
2377                 struct dst_entry *ndst;
2378
2379                 if (!vxlan->vn6_sock)
2380                         return -EINVAL;
2381                 ndst = vxlan6_get_route(vxlan, skb, 0,
2382                                         &info->key.u.ipv6.dst,
2383                                         &info->key.u.ipv6.src, NULL);
2384                 if (IS_ERR(ndst))
2385                         return PTR_ERR(ndst);
2386                 dst_release(ndst);
2387 #else /* !CONFIG_IPV6 */
2388                 return -EPFNOSUPPORT;
2389 #endif
2390         }
2391         info->key.tp_src = sport;
2392         info->key.tp_dst = dport;
2393         return 0;
2394 }
2395
2396 static const struct net_device_ops vxlan_netdev_ops = {
2397         .ndo_init               = vxlan_init,
2398         .ndo_uninit             = vxlan_uninit,
2399         .ndo_open               = vxlan_open,
2400         .ndo_stop               = vxlan_stop,
2401         .ndo_start_xmit         = vxlan_xmit,
2402         .ndo_get_stats64        = ip_tunnel_get_stats64,
2403         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2404         .ndo_change_mtu         = vxlan_change_mtu,
2405         .ndo_validate_addr      = eth_validate_addr,
2406         .ndo_set_mac_address    = eth_mac_addr,
2407         .ndo_fdb_add            = vxlan_fdb_add,
2408         .ndo_fdb_del            = vxlan_fdb_delete,
2409         .ndo_fdb_dump           = vxlan_fdb_dump,
2410         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2411 };
2412
2413 /* Info for udev, that this is a virtual tunnel endpoint */
2414 static struct device_type vxlan_type = {
2415         .name = "vxlan",
2416 };
2417
2418 /* Calls the ndo_add_vxlan_port of the caller in order to
2419  * supply the listening VXLAN udp ports. Callers are expected
2420  * to implement the ndo_add_vxlan_port.
2421  */
2422 void vxlan_get_rx_port(struct net_device *dev)
2423 {
2424         struct vxlan_sock *vs;
2425         struct net *net = dev_net(dev);
2426         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2427         sa_family_t sa_family;
2428         __be16 port;
2429         unsigned int i;
2430
2431         spin_lock(&vn->sock_lock);
2432         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2433                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2434                         port = inet_sk(vs->sock->sk)->inet_sport;
2435                         sa_family = vxlan_get_sk_family(vs);
2436                         dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2437                                                             port);
2438                 }
2439         }
2440         spin_unlock(&vn->sock_lock);
2441 }
2442 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2443
2444 /* Initialize the device structure. */
2445 static void vxlan_setup(struct net_device *dev)
2446 {
2447         struct vxlan_dev *vxlan = netdev_priv(dev);
2448         unsigned int h;
2449
2450         eth_hw_addr_random(dev);
2451         ether_setup(dev);
2452
2453         dev->netdev_ops = &vxlan_netdev_ops;
2454         dev->destructor = free_netdev;
2455         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2456
2457         dev->features   |= NETIF_F_LLTX;
2458         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2459         dev->features   |= NETIF_F_RXCSUM;
2460         dev->features   |= NETIF_F_GSO_SOFTWARE;
2461
2462         dev->vlan_features = dev->features;
2463         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2464         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2465         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2466         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2467         netif_keep_dst(dev);
2468         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
2469
2470         INIT_LIST_HEAD(&vxlan->next);
2471         spin_lock_init(&vxlan->hash_lock);
2472
2473         init_timer_deferrable(&vxlan->age_timer);
2474         vxlan->age_timer.function = vxlan_cleanup;
2475         vxlan->age_timer.data = (unsigned long) vxlan;
2476
2477         vxlan->cfg.dst_port = htons(vxlan_port);
2478
2479         vxlan->dev = dev;
2480
2481         gro_cells_init(&vxlan->gro_cells, dev);
2482
2483         for (h = 0; h < FDB_HASH_SIZE; ++h)
2484                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2485 }
2486
2487 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2488         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2489         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2490         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2491         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2492         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2493         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2494         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2495         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2496         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2497         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2498         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2499         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2500         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2501         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2502         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2503         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2504         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2505         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2506         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2507         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2508         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2509         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2510         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2511         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2512         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2513 };
2514
2515 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2516 {
2517         if (tb[IFLA_ADDRESS]) {
2518                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2519                         pr_debug("invalid link address (not ethernet)\n");
2520                         return -EINVAL;
2521                 }
2522
2523                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2524                         pr_debug("invalid all zero ethernet address\n");
2525                         return -EADDRNOTAVAIL;
2526                 }
2527         }
2528
2529         if (!data)
2530                 return -EINVAL;
2531
2532         if (data[IFLA_VXLAN_ID]) {
2533                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2534                 if (id >= VXLAN_VID_MASK)
2535                         return -ERANGE;
2536         }
2537
2538         if (data[IFLA_VXLAN_PORT_RANGE]) {
2539                 const struct ifla_vxlan_port_range *p
2540                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2541
2542                 if (ntohs(p->high) < ntohs(p->low)) {
2543                         pr_debug("port range %u .. %u not valid\n",
2544                                  ntohs(p->low), ntohs(p->high));
2545                         return -EINVAL;
2546                 }
2547         }
2548
2549         return 0;
2550 }
2551
2552 static void vxlan_get_drvinfo(struct net_device *netdev,
2553                               struct ethtool_drvinfo *drvinfo)
2554 {
2555         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2556         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2557 }
2558
2559 static const struct ethtool_ops vxlan_ethtool_ops = {
2560         .get_drvinfo    = vxlan_get_drvinfo,
2561         .get_link       = ethtool_op_get_link,
2562 };
2563
2564 static void vxlan_del_work(struct work_struct *work)
2565 {
2566         struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2567         udp_tunnel_sock_release(vs->sock);
2568         kfree_rcu(vs, rcu);
2569 }
2570
2571 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2572                                         __be16 port, u32 flags)
2573 {
2574         struct socket *sock;
2575         struct udp_port_cfg udp_conf;
2576         int err;
2577
2578         memset(&udp_conf, 0, sizeof(udp_conf));
2579
2580         if (ipv6) {
2581                 udp_conf.family = AF_INET6;
2582                 udp_conf.use_udp6_rx_checksums =
2583                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2584                 udp_conf.ipv6_v6only = 1;
2585         } else {
2586                 udp_conf.family = AF_INET;
2587         }
2588
2589         udp_conf.local_udp_port = port;
2590
2591         /* Open UDP socket */
2592         err = udp_sock_create(net, &udp_conf, &sock);
2593         if (err < 0)
2594                 return ERR_PTR(err);
2595
2596         return sock;
2597 }
2598
2599 /* Create new listen socket if needed */
2600 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2601                                               __be16 port, u32 flags)
2602 {
2603         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2604         struct vxlan_sock *vs;
2605         struct socket *sock;
2606         unsigned int h;
2607         struct udp_tunnel_sock_cfg tunnel_cfg;
2608
2609         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2610         if (!vs)
2611                 return ERR_PTR(-ENOMEM);
2612
2613         for (h = 0; h < VNI_HASH_SIZE; ++h)
2614                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2615
2616         INIT_WORK(&vs->del_work, vxlan_del_work);
2617
2618         sock = vxlan_create_sock(net, ipv6, port, flags);
2619         if (IS_ERR(sock)) {
2620                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2621                         PTR_ERR(sock));
2622                 kfree(vs);
2623                 return ERR_CAST(sock);
2624         }
2625
2626         vs->sock = sock;
2627         atomic_set(&vs->refcnt, 1);
2628         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2629
2630         /* Initialize the vxlan udp offloads structure */
2631         vs->udp_offloads.port = port;
2632         vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
2633         vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2634
2635         spin_lock(&vn->sock_lock);
2636         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2637         vxlan_notify_add_rx_port(vs);
2638         spin_unlock(&vn->sock_lock);
2639
2640         /* Mark socket as an encapsulation socket. */
2641         tunnel_cfg.sk_user_data = vs;
2642         tunnel_cfg.encap_type = 1;
2643         tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2644         tunnel_cfg.encap_destroy = NULL;
2645
2646         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2647
2648         return vs;
2649 }
2650
2651 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2652 {
2653         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2654         struct vxlan_sock *vs = NULL;
2655
2656         if (!vxlan->cfg.no_share) {
2657                 spin_lock(&vn->sock_lock);
2658                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2659                                      vxlan->cfg.dst_port, vxlan->flags);
2660                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2661                         spin_unlock(&vn->sock_lock);
2662                         return -EBUSY;
2663                 }
2664                 spin_unlock(&vn->sock_lock);
2665         }
2666         if (!vs)
2667                 vs = vxlan_socket_create(vxlan->net, ipv6,
2668                                          vxlan->cfg.dst_port, vxlan->flags);
2669         if (IS_ERR(vs))
2670                 return PTR_ERR(vs);
2671 #if IS_ENABLED(CONFIG_IPV6)
2672         if (ipv6)
2673                 vxlan->vn6_sock = vs;
2674         else
2675 #endif
2676                 vxlan->vn4_sock = vs;
2677         vxlan_vs_add_dev(vs, vxlan);
2678         return 0;
2679 }
2680
2681 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2682 {
2683         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2684         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2685         int ret = 0;
2686
2687         vxlan->vn4_sock = NULL;
2688 #if IS_ENABLED(CONFIG_IPV6)
2689         vxlan->vn6_sock = NULL;
2690         if (ipv6 || metadata)
2691                 ret = __vxlan_sock_add(vxlan, true);
2692 #endif
2693         if (!ret && (!ipv6 || metadata))
2694                 ret = __vxlan_sock_add(vxlan, false);
2695         if (ret < 0)
2696                 vxlan_sock_release(vxlan);
2697         return ret;
2698 }
2699
2700 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2701                                struct vxlan_config *conf)
2702 {
2703         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2704         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2705         struct vxlan_rdst *dst = &vxlan->default_dst;
2706         unsigned short needed_headroom = ETH_HLEN;
2707         int err;
2708         bool use_ipv6 = false;
2709         __be16 default_port = vxlan->cfg.dst_port;
2710
2711         vxlan->net = src_net;
2712
2713         dst->remote_vni = conf->vni;
2714
2715         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2716
2717         /* Unless IPv6 is explicitly requested, assume IPv4 */
2718         if (!dst->remote_ip.sa.sa_family)
2719                 dst->remote_ip.sa.sa_family = AF_INET;
2720
2721         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2722             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2723                 if (!IS_ENABLED(CONFIG_IPV6))
2724                         return -EPFNOSUPPORT;
2725                 use_ipv6 = true;
2726                 vxlan->flags |= VXLAN_F_IPV6;
2727         }
2728
2729         if (conf->remote_ifindex) {
2730                 struct net_device *lowerdev
2731                          = __dev_get_by_index(src_net, conf->remote_ifindex);
2732
2733                 dst->remote_ifindex = conf->remote_ifindex;
2734
2735                 if (!lowerdev) {
2736                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2737                         return -ENODEV;
2738                 }
2739
2740 #if IS_ENABLED(CONFIG_IPV6)
2741                 if (use_ipv6) {
2742                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2743                         if (idev && idev->cnf.disable_ipv6) {
2744                                 pr_info("IPv6 is disabled via sysctl\n");
2745                                 return -EPERM;
2746                         }
2747                 }
2748 #endif
2749
2750                 if (!conf->mtu)
2751                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2752
2753                 needed_headroom = lowerdev->hard_header_len;
2754         }
2755
2756         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2757                 needed_headroom += VXLAN6_HEADROOM;
2758         else
2759                 needed_headroom += VXLAN_HEADROOM;
2760         dev->needed_headroom = needed_headroom;
2761
2762         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2763         if (!vxlan->cfg.dst_port)
2764                 vxlan->cfg.dst_port = default_port;
2765         vxlan->flags |= conf->flags;
2766
2767         if (!vxlan->cfg.age_interval)
2768                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2769
2770         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2771                 if (tmp->cfg.vni == conf->vni &&
2772                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2773                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2774                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2775                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2776                     (vxlan->flags & VXLAN_F_RCV_FLAGS))
2777                 return -EEXIST;
2778         }
2779
2780         dev->ethtool_ops = &vxlan_ethtool_ops;
2781
2782         /* create an fdb entry for a valid default destination */
2783         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2784                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2785                                        &vxlan->default_dst.remote_ip,
2786                                        NUD_REACHABLE|NUD_PERMANENT,
2787                                        NLM_F_EXCL|NLM_F_CREATE,
2788                                        vxlan->cfg.dst_port,
2789                                        vxlan->default_dst.remote_vni,
2790                                        vxlan->default_dst.remote_ifindex,
2791                                        NTF_SELF);
2792                 if (err)
2793                         return err;
2794         }
2795
2796         err = register_netdevice(dev);
2797         if (err) {
2798                 vxlan_fdb_delete_default(vxlan);
2799                 return err;
2800         }
2801
2802         list_add(&vxlan->next, &vn->vxlan_list);
2803
2804         return 0;
2805 }
2806
2807 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2808                                     u8 name_assign_type, struct vxlan_config *conf)
2809 {
2810         struct nlattr *tb[IFLA_MAX+1];
2811         struct net_device *dev;
2812         int err;
2813
2814         memset(&tb, 0, sizeof(tb));
2815
2816         dev = rtnl_create_link(net, name, name_assign_type,
2817                                &vxlan_link_ops, tb);
2818         if (IS_ERR(dev))
2819                 return dev;
2820
2821         err = vxlan_dev_configure(net, dev, conf);
2822         if (err < 0) {
2823                 free_netdev(dev);
2824                 return ERR_PTR(err);
2825         }
2826
2827         return dev;
2828 }
2829 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2830
2831 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2832                          struct nlattr *tb[], struct nlattr *data[])
2833 {
2834         struct vxlan_config conf;
2835         int err;
2836
2837         memset(&conf, 0, sizeof(conf));
2838
2839         if (data[IFLA_VXLAN_ID])
2840                 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2841
2842         if (data[IFLA_VXLAN_GROUP]) {
2843                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2844         } else if (data[IFLA_VXLAN_GROUP6]) {
2845                 if (!IS_ENABLED(CONFIG_IPV6))
2846                         return -EPFNOSUPPORT;
2847
2848                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2849                 conf.remote_ip.sa.sa_family = AF_INET6;
2850         }
2851
2852         if (data[IFLA_VXLAN_LOCAL]) {
2853                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2854                 conf.saddr.sa.sa_family = AF_INET;
2855         } else if (data[IFLA_VXLAN_LOCAL6]) {
2856                 if (!IS_ENABLED(CONFIG_IPV6))
2857                         return -EPFNOSUPPORT;
2858
2859                 /* TODO: respect scope id */
2860                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2861                 conf.saddr.sa.sa_family = AF_INET6;
2862         }
2863
2864         if (data[IFLA_VXLAN_LINK])
2865                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2866
2867         if (data[IFLA_VXLAN_TOS])
2868                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2869
2870         if (data[IFLA_VXLAN_TTL])
2871                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2872
2873         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2874                 conf.flags |= VXLAN_F_LEARN;
2875
2876         if (data[IFLA_VXLAN_AGEING])
2877                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2878
2879         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2880                 conf.flags |= VXLAN_F_PROXY;
2881
2882         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2883                 conf.flags |= VXLAN_F_RSC;
2884
2885         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2886                 conf.flags |= VXLAN_F_L2MISS;
2887
2888         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2889                 conf.flags |= VXLAN_F_L3MISS;
2890
2891         if (data[IFLA_VXLAN_LIMIT])
2892                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2893
2894         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2895             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2896                 conf.flags |= VXLAN_F_COLLECT_METADATA;
2897
2898         if (data[IFLA_VXLAN_PORT_RANGE]) {
2899                 const struct ifla_vxlan_port_range *p
2900                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2901                 conf.port_min = ntohs(p->low);
2902                 conf.port_max = ntohs(p->high);
2903         }
2904
2905         if (data[IFLA_VXLAN_PORT])
2906                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2907
2908         if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2909                 conf.flags |= VXLAN_F_UDP_CSUM;
2910
2911         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2912             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2913                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2914
2915         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2916             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2917                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2918
2919         if (data[IFLA_VXLAN_REMCSUM_TX] &&
2920             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2921                 conf.flags |= VXLAN_F_REMCSUM_TX;
2922
2923         if (data[IFLA_VXLAN_REMCSUM_RX] &&
2924             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2925                 conf.flags |= VXLAN_F_REMCSUM_RX;
2926
2927         if (data[IFLA_VXLAN_GBP])
2928                 conf.flags |= VXLAN_F_GBP;
2929
2930         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2931                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2932
2933         err = vxlan_dev_configure(src_net, dev, &conf);
2934         switch (err) {
2935         case -ENODEV:
2936                 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
2937                 break;
2938
2939         case -EPERM:
2940                 pr_info("IPv6 is disabled via sysctl\n");
2941                 break;
2942
2943         case -EEXIST:
2944                 pr_info("duplicate VNI %u\n", be32_to_cpu(conf.vni));
2945                 break;
2946         }
2947
2948         return err;
2949 }
2950
2951 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2952 {
2953         struct vxlan_dev *vxlan = netdev_priv(dev);
2954         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2955
2956         spin_lock(&vn->sock_lock);
2957         if (!hlist_unhashed(&vxlan->hlist))
2958                 hlist_del_rcu(&vxlan->hlist);
2959         spin_unlock(&vn->sock_lock);
2960
2961         gro_cells_destroy(&vxlan->gro_cells);
2962         list_del(&vxlan->next);
2963         unregister_netdevice_queue(dev, head);
2964 }
2965
2966 static size_t vxlan_get_size(const struct net_device *dev)
2967 {
2968
2969         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
2970                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2971                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2972                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2973                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
2974                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
2975                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
2976                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
2977                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
2978                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
2979                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
2980                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
2981                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2982                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2983                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2984                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2985                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2986                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2987                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2988                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
2989                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
2990                 0;
2991 }
2992
2993 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2994 {
2995         const struct vxlan_dev *vxlan = netdev_priv(dev);
2996         const struct vxlan_rdst *dst = &vxlan->default_dst;
2997         struct ifla_vxlan_port_range ports = {
2998                 .low =  htons(vxlan->cfg.port_min),
2999                 .high = htons(vxlan->cfg.port_max),
3000         };
3001
3002         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3003                 goto nla_put_failure;
3004
3005         if (!vxlan_addr_any(&dst->remote_ip)) {
3006                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3007                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3008                                             dst->remote_ip.sin.sin_addr.s_addr))
3009                                 goto nla_put_failure;
3010 #if IS_ENABLED(CONFIG_IPV6)
3011                 } else {
3012                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3013                                              &dst->remote_ip.sin6.sin6_addr))
3014                                 goto nla_put_failure;
3015 #endif
3016                 }
3017         }
3018
3019         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3020                 goto nla_put_failure;
3021
3022         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3023                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3024                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3025                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3026                                 goto nla_put_failure;
3027 #if IS_ENABLED(CONFIG_IPV6)
3028                 } else {
3029                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3030                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3031                                 goto nla_put_failure;
3032 #endif
3033                 }
3034         }
3035
3036         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3037             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3038             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3039                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3040             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3041                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3042             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3043             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3044                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3045             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3046                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3047             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3048                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3049             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3050             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3051             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3052             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3053                         !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
3054             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3055                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3056             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3057                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3058             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3059                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3060             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3061                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3062                 goto nla_put_failure;
3063
3064         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3065                 goto nla_put_failure;
3066
3067         if (vxlan->flags & VXLAN_F_GBP &&
3068             nla_put_flag(skb, IFLA_VXLAN_GBP))
3069                 goto nla_put_failure;
3070
3071         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3072             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3073                 goto nla_put_failure;
3074
3075         return 0;
3076
3077 nla_put_failure:
3078         return -EMSGSIZE;
3079 }
3080
3081 static struct net *vxlan_get_link_net(const struct net_device *dev)
3082 {
3083         struct vxlan_dev *vxlan = netdev_priv(dev);
3084
3085         return vxlan->net;
3086 }
3087
3088 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3089         .kind           = "vxlan",
3090         .maxtype        = IFLA_VXLAN_MAX,
3091         .policy         = vxlan_policy,
3092         .priv_size      = sizeof(struct vxlan_dev),
3093         .setup          = vxlan_setup,
3094         .validate       = vxlan_validate,
3095         .newlink        = vxlan_newlink,
3096         .dellink        = vxlan_dellink,
3097         .get_size       = vxlan_get_size,
3098         .fill_info      = vxlan_fill_info,
3099         .get_link_net   = vxlan_get_link_net,
3100 };
3101
3102 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3103                                              struct net_device *dev)
3104 {
3105         struct vxlan_dev *vxlan, *next;
3106         LIST_HEAD(list_kill);
3107
3108         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3109                 struct vxlan_rdst *dst = &vxlan->default_dst;
3110
3111                 /* In case we created vxlan device with carrier
3112                  * and we loose the carrier due to module unload
3113                  * we also need to remove vxlan device. In other
3114                  * cases, it's not necessary and remote_ifindex
3115                  * is 0 here, so no matches.
3116                  */
3117                 if (dst->remote_ifindex == dev->ifindex)
3118                         vxlan_dellink(vxlan->dev, &list_kill);
3119         }
3120
3121         unregister_netdevice_many(&list_kill);
3122 }
3123
3124 static int vxlan_lowerdev_event(struct notifier_block *unused,
3125                                 unsigned long event, void *ptr)
3126 {
3127         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3128         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3129
3130         if (event == NETDEV_UNREGISTER)
3131                 vxlan_handle_lowerdev_unregister(vn, dev);
3132
3133         return NOTIFY_DONE;
3134 }
3135
3136 static struct notifier_block vxlan_notifier_block __read_mostly = {
3137         .notifier_call = vxlan_lowerdev_event,
3138 };
3139
3140 static __net_init int vxlan_init_net(struct net *net)
3141 {
3142         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3143         unsigned int h;
3144
3145         INIT_LIST_HEAD(&vn->vxlan_list);
3146         spin_lock_init(&vn->sock_lock);
3147
3148         for (h = 0; h < PORT_HASH_SIZE; ++h)
3149                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3150
3151         return 0;
3152 }
3153
3154 static void __net_exit vxlan_exit_net(struct net *net)
3155 {
3156         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3157         struct vxlan_dev *vxlan, *next;
3158         struct net_device *dev, *aux;
3159         LIST_HEAD(list);
3160
3161         rtnl_lock();
3162         for_each_netdev_safe(net, dev, aux)
3163                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3164                         unregister_netdevice_queue(dev, &list);
3165
3166         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3167                 /* If vxlan->dev is in the same netns, it has already been added
3168                  * to the list by the previous loop.
3169                  */
3170                 if (!net_eq(dev_net(vxlan->dev), net)) {
3171                         gro_cells_destroy(&vxlan->gro_cells);
3172                         unregister_netdevice_queue(vxlan->dev, &list);
3173                 }
3174         }
3175
3176         unregister_netdevice_many(&list);
3177         rtnl_unlock();
3178 }
3179
3180 static struct pernet_operations vxlan_net_ops = {
3181         .init = vxlan_init_net,
3182         .exit = vxlan_exit_net,
3183         .id   = &vxlan_net_id,
3184         .size = sizeof(struct vxlan_net),
3185 };
3186
3187 static int __init vxlan_init_module(void)
3188 {
3189         int rc;
3190
3191         vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3192         if (!vxlan_wq)
3193                 return -ENOMEM;
3194
3195         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3196
3197         rc = register_pernet_subsys(&vxlan_net_ops);
3198         if (rc)
3199                 goto out1;
3200
3201         rc = register_netdevice_notifier(&vxlan_notifier_block);
3202         if (rc)
3203                 goto out2;
3204
3205         rc = rtnl_link_register(&vxlan_link_ops);
3206         if (rc)
3207                 goto out3;
3208
3209         return 0;
3210 out3:
3211         unregister_netdevice_notifier(&vxlan_notifier_block);
3212 out2:
3213         unregister_pernet_subsys(&vxlan_net_ops);
3214 out1:
3215         destroy_workqueue(vxlan_wq);
3216         return rc;
3217 }
3218 late_initcall(vxlan_init_module);
3219
3220 static void __exit vxlan_cleanup_module(void)
3221 {
3222         rtnl_link_unregister(&vxlan_link_ops);
3223         unregister_netdevice_notifier(&vxlan_notifier_block);
3224         destroy_workqueue(vxlan_wq);
3225         unregister_pernet_subsys(&vxlan_net_ops);
3226         /* rcu_barrier() is called by netns */
3227 }
3228 module_exit(vxlan_cleanup_module);
3229
3230 MODULE_LICENSE("GPL");
3231 MODULE_VERSION(VXLAN_VERSION);
3232 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3233 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3234 MODULE_ALIAS_RTNL_LINK("vxlan");