drivers/net/geneve.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * GENEVE: Generic Network Virtualization Encapsulation
   4  *
   5  * Copyright (c) 2015 Red Hat, Inc.
   6  */
   7
   8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   9
  10 #include <linux/kernel.h>
  11 #include <linux/module.h>
  12 #include <linux/etherdevice.h>
  13 #include <linux/hash.h>
  14 #include <net/ipv6_stubs.h>
  15 #include <net/dst_metadata.h>
  16 #include <net/gro_cells.h>
  17 #include <net/rtnetlink.h>
  18 #include <net/geneve.h>
  19 #include <net/protocol.h>
  20
  21 #define GENEVE_NETDEV_VER       "0.6"
  22
  23 #define GENEVE_N_VID            (1u << 24)
  24 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  25
  26 #define VNI_HASH_BITS           10
  27 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  28
  29 static bool log_ecn_error = true;
  30 module_param(log_ecn_error, bool, 0644);
  31 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  32
  33 #define GENEVE_VER 0
  34 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  35 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
  36 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
  37
  38 /* per-network namespace private data for this module */
  39 struct geneve_net {
  40         struct list_head        geneve_list;
  41         struct list_head        sock_list;
  42 };
  43
  44 static unsigned int geneve_net_id;
  45
  46 struct geneve_dev_node {
  47         struct hlist_node hlist;
  48         struct geneve_dev *geneve;
  49 };
  50
  51 /* Pseudo network device */
  52 struct geneve_dev {
  53         struct geneve_dev_node hlist4;  /* vni hash table for IPv4 socket */
  54 #if IS_ENABLED(CONFIG_IPV6)
  55         struct geneve_dev_node hlist6;  /* vni hash table for IPv6 socket */
  56 #endif
  57         struct net         *net;        /* netns for packet i/o */
  58         struct net_device  *dev;        /* netdev for geneve tunnel */
  59         struct ip_tunnel_info info;
  60         struct geneve_sock __rcu *sock4;        /* IPv4 socket used for geneve tunnel */
  61 #if IS_ENABLED(CONFIG_IPV6)
  62         struct geneve_sock __rcu *sock6;        /* IPv6 socket used for geneve tunnel */
  63 #endif
  64         struct list_head   next;        /* geneve's per namespace list */
  65         struct gro_cells   gro_cells;
  66         bool               collect_md;
  67         bool               use_udp6_rx_checksums;
  68         bool               ttl_inherit;
  69         enum ifla_geneve_df df;
  70 };
  71
  72 struct geneve_sock {
  73         bool                    collect_md;
  74         struct list_head        list;
  75         struct socket           *sock;
  76         struct rcu_head         rcu;
  77         int                     refcnt;
  78         struct hlist_head       vni_list[VNI_HASH_SIZE];
  79 };
  80
  81 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  82 {
  83         __u32 vnid;
  84
  85         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  86         return hash_32(vnid, VNI_HASH_BITS);
  87 }
  88
  89 static __be64 vni_to_tunnel_id(const __u8 *vni)
  90 {
  91 #ifdef __BIG_ENDIAN
  92         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
  93 #else
  94         return (__force __be64)(((__force u64)vni[0] << 40) |
  95                                 ((__force u64)vni[1] << 48) |
  96                                 ((__force u64)vni[2] << 56));
  97 #endif
  98 }
  99
 100 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 101 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 102 {
 103 #ifdef __BIG_ENDIAN
 104         vni[0] = (__force __u8)(tun_id >> 16);
 105         vni[1] = (__force __u8)(tun_id >> 8);
 106         vni[2] = (__force __u8)tun_id;
 107 #else
 108         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 109         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 110         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 111 #endif
 112 }
 113
 114 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
 115 {
 116         return !memcmp(vni, &tun_id[5], 3);
 117 }
 118
 119 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 120 {
 121         return gs->sock->sk->sk_family;
 122 }
 123
 124 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 125                                         __be32 addr, u8 vni[])
 126 {
 127         struct hlist_head *vni_list_head;
 128         struct geneve_dev_node *node;
 129         __u32 hash;
 130
 131         /* Find the device for this VNI */
 132         hash = geneve_net_vni_hash(vni);
 133         vni_list_head = &gs->vni_list[hash];
 134         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 135                 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
 136                     addr == node->geneve->info.key.u.ipv4.dst)
 137                         return node->geneve;
 138         }
 139         return NULL;
 140 }
 141
 142 #if IS_ENABLED(CONFIG_IPV6)
 143 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 144                                          struct in6_addr addr6, u8 vni[])
 145 {
 146         struct hlist_head *vni_list_head;
 147         struct geneve_dev_node *node;
 148         __u32 hash;
 149
 150         /* Find the device for this VNI */
 151         hash = geneve_net_vni_hash(vni);
 152         vni_list_head = &gs->vni_list[hash];
 153         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 154                 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
 155                     ipv6_addr_equal(&addr6, &node->geneve->info.key.u.ipv6.dst))
 156                         return node->geneve;
 157         }
 158         return NULL;
 159 }
 160 #endif
 161
 162 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 163 {
 164         return (struct genevehdr *)(udp_hdr(skb) + 1);
 165 }
 166
 167 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 168                                             struct sk_buff *skb)
 169 {
 170         static u8 zero_vni[3];
 171         u8 *vni;
 172
 173         if (geneve_get_sk_family(gs) == AF_INET) {
 174                 struct iphdr *iph;
 175                 __be32 addr;
 176
 177                 iph = ip_hdr(skb); /* outer IP header... */
 178
 179                 if (gs->collect_md) {
 180                         vni = zero_vni;
 181                         addr = 0;
 182                 } else {
 183                         vni = geneve_hdr(skb)->vni;
 184                         addr = iph->saddr;
 185                 }
 186
 187                 return geneve_lookup(gs, addr, vni);
 188 #if IS_ENABLED(CONFIG_IPV6)
 189         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 190                 static struct in6_addr zero_addr6;
 191                 struct ipv6hdr *ip6h;
 192                 struct in6_addr addr6;
 193
 194                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 195
 196                 if (gs->collect_md) {
 197                         vni = zero_vni;
 198                         addr6 = zero_addr6;
 199                 } else {
 200                         vni = geneve_hdr(skb)->vni;
 201                         addr6 = ip6h->saddr;
 202                 }
 203
 204                 return geneve6_lookup(gs, addr6, vni);
 205 #endif
 206         }
 207         return NULL;
 208 }
 209
 210 /* geneve receive/decap routine */
 211 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 212                       struct sk_buff *skb)
 213 {
 214         struct genevehdr *gnvh = geneve_hdr(skb);
 215         struct metadata_dst *tun_dst = NULL;
 216         struct pcpu_sw_netstats *stats;
 217         unsigned int len;
 218         int err = 0;
 219         void *oiph;
 220
 221         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 222                 __be16 flags;
 223
 224                 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
 225                         (gnvh->oam ? TUNNEL_OAM : 0) |
 226                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 227
 228                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 229                                          vni_to_tunnel_id(gnvh->vni),
 230                                          gnvh->opt_len * 4);
 231                 if (!tun_dst) {
 232                         geneve->dev->stats.rx_dropped++;
 233                         goto drop;
 234                 }
 235                 /* Update tunnel dst according to Geneve options. */
 236                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 237                                         gnvh->options, gnvh->opt_len * 4,
 238                                         TUNNEL_GENEVE_OPT);
 239         } else {
 240                 /* Drop packets w/ critical options,
 241                  * since we don't support any...
 242                  */
 243                 if (gnvh->critical) {
 244                         geneve->dev->stats.rx_frame_errors++;
 245                         geneve->dev->stats.rx_errors++;
 246                         goto drop;
 247                 }
 248         }
 249
 250         skb_reset_mac_header(skb);
 251         skb->protocol = eth_type_trans(skb, geneve->dev);
 252         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 253
 254         if (tun_dst)
 255                 skb_dst_set(skb, &tun_dst->dst);
 256
 257         /* Ignore packet loops (and multicast echo) */
 258         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
 259                 geneve->dev->stats.rx_errors++;
 260                 goto drop;
 261         }
 262
 263         oiph = skb_network_header(skb);
 264         skb_reset_network_header(skb);
 265
 266         if (geneve_get_sk_family(gs) == AF_INET)
 267                 err = IP_ECN_decapsulate(oiph, skb);
 268 #if IS_ENABLED(CONFIG_IPV6)
 269         else
 270                 err = IP6_ECN_decapsulate(oiph, skb);
 271 #endif
 272
 273         if (unlikely(err)) {
 274                 if (log_ecn_error) {
 275                         if (geneve_get_sk_family(gs) == AF_INET)
 276                                 net_info_ratelimited("non-ECT from %pI4 "
 277                                                      "with TOS=%#x\n",
 278                                                      &((struct iphdr *)oiph)->saddr,
 279                                                      ((struct iphdr *)oiph)->tos);
 280 #if IS_ENABLED(CONFIG_IPV6)
 281                         else
 282                                 net_info_ratelimited("non-ECT from %pI6\n",
 283                                                      &((struct ipv6hdr *)oiph)->saddr);
 284 #endif
 285                 }
 286                 if (err > 1) {
 287                         ++geneve->dev->stats.rx_frame_errors;
 288                         ++geneve->dev->stats.rx_errors;
 289                         goto drop;
 290                 }
 291         }
 292
 293         len = skb->len;
 294         err = gro_cells_receive(&geneve->gro_cells, skb);
 295         if (likely(err == NET_RX_SUCCESS)) {
 296                 stats = this_cpu_ptr(geneve->dev->tstats);
 297                 u64_stats_update_begin(&stats->syncp);
 298                 stats->rx_packets++;
 299                 stats->rx_bytes += len;
 300                 u64_stats_update_end(&stats->syncp);
 301         }
 302         return;
 303 drop:
 304         /* Consume bad packet */
 305         kfree_skb(skb);
 306 }
 307
 308 /* Setup stats when device is created */
 309 static int geneve_init(struct net_device *dev)
 310 {
 311         struct geneve_dev *geneve = netdev_priv(dev);
 312         int err;
 313
 314         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 315         if (!dev->tstats)
 316                 return -ENOMEM;
 317
 318         err = gro_cells_init(&geneve->gro_cells, dev);
 319         if (err) {
 320                 free_percpu(dev->tstats);
 321                 return err;
 322         }
 323
 324         err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL);
 325         if (err) {
 326                 free_percpu(dev->tstats);
 327                 gro_cells_destroy(&geneve->gro_cells);
 328                 return err;
 329         }
 330         return 0;
 331 }
 332
 333 static void geneve_uninit(struct net_device *dev)
 334 {
 335         struct geneve_dev *geneve = netdev_priv(dev);
 336
 337         dst_cache_destroy(&geneve->info.dst_cache);
 338         gro_cells_destroy(&geneve->gro_cells);
 339         free_percpu(dev->tstats);
 340 }
 341
 342 /* Callback from net/ipv4/udp.c to receive packets */
 343 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 344 {
 345         struct genevehdr *geneveh;
 346         struct geneve_dev *geneve;
 347         struct geneve_sock *gs;
 348         int opts_len;
 349
 350         /* Need UDP and Geneve header to be present */
 351         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 352                 goto drop;
 353
 354         /* Return packets with reserved bits set */
 355         geneveh = geneve_hdr(skb);
 356         if (unlikely(geneveh->ver != GENEVE_VER))
 357                 goto drop;
 358
 359         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 360                 goto drop;
 361
 362         gs = rcu_dereference_sk_user_data(sk);
 363         if (!gs)
 364                 goto drop;
 365
 366         geneve = geneve_lookup_skb(gs, skb);
 367         if (!geneve)
 368                 goto drop;
 369
 370         opts_len = geneveh->opt_len * 4;
 371         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 372                                  htons(ETH_P_TEB),
 373                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 374                 geneve->dev->stats.rx_dropped++;
 375                 goto drop;
 376         }
 377
 378         geneve_rx(geneve, gs, skb);
 379         return 0;
 380
 381 drop:
 382         /* Consume bad packet */
 383         kfree_skb(skb);
 384         return 0;
 385 }
 386
 387 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
 388 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
 389 {
 390         struct genevehdr *geneveh;
 391         struct geneve_sock *gs;
 392         u8 zero_vni[3] = { 0 };
 393         u8 *vni = zero_vni;
 394
 395         if (skb->len < GENEVE_BASE_HLEN)
 396                 return -EINVAL;
 397
 398         geneveh = geneve_hdr(skb);
 399         if (geneveh->ver != GENEVE_VER)
 400                 return -EINVAL;
 401
 402         if (geneveh->proto_type != htons(ETH_P_TEB))
 403                 return -EINVAL;
 404
 405         gs = rcu_dereference_sk_user_data(sk);
 406         if (!gs)
 407                 return -ENOENT;
 408
 409         if (geneve_get_sk_family(gs) == AF_INET) {
 410                 struct iphdr *iph = ip_hdr(skb);
 411                 __be32 addr4 = 0;
 412
 413                 if (!gs->collect_md) {
 414                         vni = geneve_hdr(skb)->vni;
 415                         addr4 = iph->daddr;
 416                 }
 417
 418                 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
 419         }
 420
 421 #if IS_ENABLED(CONFIG_IPV6)
 422         if (geneve_get_sk_family(gs) == AF_INET6) {
 423                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 424                 struct in6_addr addr6;
 425
 426                 memset(&addr6, 0, sizeof(struct in6_addr));
 427
 428                 if (!gs->collect_md) {
 429                         vni = geneve_hdr(skb)->vni;
 430                         addr6 = ip6h->daddr;
 431                 }
 432
 433                 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
 434         }
 435 #endif
 436
 437         return -EPFNOSUPPORT;
 438 }
 439
 440 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 441                                          __be16 port, bool ipv6_rx_csum)
 442 {
 443         struct socket *sock;
 444         struct udp_port_cfg udp_conf;
 445         int err;
 446
 447         memset(&udp_conf, 0, sizeof(udp_conf));
 448
 449         if (ipv6) {
 450                 udp_conf.family = AF_INET6;
 451                 udp_conf.ipv6_v6only = 1;
 452                 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
 453         } else {
 454                 udp_conf.family = AF_INET;
 455                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 456         }
 457
 458         udp_conf.local_udp_port = port;
 459
 460         /* Open UDP socket */
 461         err = udp_sock_create(net, &udp_conf, &sock);
 462         if (err < 0)
 463                 return ERR_PTR(err);
 464
 465         return sock;
 466 }
 467
 468 static int geneve_hlen(struct genevehdr *gh)
 469 {
 470         return sizeof(*gh) + gh->opt_len * 4;
 471 }
 472
 473 static struct sk_buff *geneve_gro_receive(struct sock *sk,
 474                                           struct list_head *head,
 475                                           struct sk_buff *skb)
 476 {
 477         struct sk_buff *pp = NULL;
 478         struct sk_buff *p;
 479         struct genevehdr *gh, *gh2;
 480         unsigned int hlen, gh_len, off_gnv;
 481         const struct packet_offload *ptype;
 482         __be16 type;
 483         int flush = 1;
 484
 485         off_gnv = skb_gro_offset(skb);
 486         hlen = off_gnv + sizeof(*gh);
 487         gh = skb_gro_header_fast(skb, off_gnv);
 488         if (skb_gro_header_hard(skb, hlen)) {
 489                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 490                 if (unlikely(!gh))
 491                         goto out;
 492         }
 493
 494         if (gh->ver != GENEVE_VER || gh->oam)
 495                 goto out;
 496         gh_len = geneve_hlen(gh);
 497
 498         hlen = off_gnv + gh_len;
 499         if (skb_gro_header_hard(skb, hlen)) {
 500                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 501                 if (unlikely(!gh))
 502                         goto out;
 503         }
 504
 505         list_for_each_entry(p, head, list) {
 506                 if (!NAPI_GRO_CB(p)->same_flow)
 507                         continue;
 508
 509                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 510                 if (gh->opt_len != gh2->opt_len ||
 511                     memcmp(gh, gh2, gh_len)) {
 512                         NAPI_GRO_CB(p)->same_flow = 0;
 513                         continue;
 514                 }
 515         }
 516
 517         type = gh->proto_type;
 518
 519         rcu_read_lock();
 520         ptype = gro_find_receive_by_type(type);
 521         if (!ptype)
 522                 goto out_unlock;
 523
 524         skb_gro_pull(skb, gh_len);
 525         skb_gro_postpull_rcsum(skb, gh, gh_len);
 526         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 527         flush = 0;
 528
 529 out_unlock:
 530         rcu_read_unlock();
 531 out:
 532         skb_gro_flush_final(skb, pp, flush);
 533
 534         return pp;
 535 }
 536
 537 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 538                                int nhoff)
 539 {
 540         struct genevehdr *gh;
 541         struct packet_offload *ptype;
 542         __be16 type;
 543         int gh_len;
 544         int err = -ENOSYS;
 545
 546         gh = (struct genevehdr *)(skb->data + nhoff);
 547         gh_len = geneve_hlen(gh);
 548         type = gh->proto_type;
 549
 550         rcu_read_lock();
 551         ptype = gro_find_complete_by_type(type);
 552         if (ptype)
 553                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 554
 555         rcu_read_unlock();
 556
 557         skb_set_inner_mac_header(skb, nhoff + gh_len);
 558
 559         return err;
 560 }
 561
 562 /* Create new listen socket if needed */
 563 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 564                                                 bool ipv6, bool ipv6_rx_csum)
 565 {
 566         struct geneve_net *gn = net_generic(net, geneve_net_id);
 567         struct geneve_sock *gs;
 568         struct socket *sock;
 569         struct udp_tunnel_sock_cfg tunnel_cfg;
 570         int h;
 571
 572         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 573         if (!gs)
 574                 return ERR_PTR(-ENOMEM);
 575
 576         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 577         if (IS_ERR(sock)) {
 578                 kfree(gs);
 579                 return ERR_CAST(sock);
 580         }
 581
 582         gs->sock = sock;
 583         gs->refcnt = 1;
 584         for (h = 0; h < VNI_HASH_SIZE; ++h)
 585                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 586
 587         /* Initialize the geneve udp offloads structure */
 588         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 589
 590         /* Mark socket as an encapsulation socket */
 591         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 592         tunnel_cfg.sk_user_data = gs;
 593         tunnel_cfg.encap_type = 1;
 594         tunnel_cfg.gro_receive = geneve_gro_receive;
 595         tunnel_cfg.gro_complete = geneve_gro_complete;
 596         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 597         tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
 598         tunnel_cfg.encap_destroy = NULL;
 599         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 600         list_add(&gs->list, &gn->sock_list);
 601         return gs;
 602 }
 603
 604 static void __geneve_sock_release(struct geneve_sock *gs)
 605 {
 606         if (!gs || --gs->refcnt)
 607                 return;
 608
 609         list_del(&gs->list);
 610         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 611         udp_tunnel_sock_release(gs->sock);
 612         kfree_rcu(gs, rcu);
 613 }
 614
 615 static void geneve_sock_release(struct geneve_dev *geneve)
 616 {
 617         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 618 #if IS_ENABLED(CONFIG_IPV6)
 619         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 620
 621         rcu_assign_pointer(geneve->sock6, NULL);
 622 #endif
 623
 624         rcu_assign_pointer(geneve->sock4, NULL);
 625         synchronize_net();
 626
 627         __geneve_sock_release(gs4);
 628 #if IS_ENABLED(CONFIG_IPV6)
 629         __geneve_sock_release(gs6);
 630 #endif
 631 }
 632
 633 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 634                                             sa_family_t family,
 635                                             __be16 dst_port)
 636 {
 637         struct geneve_sock *gs;
 638
 639         list_for_each_entry(gs, &gn->sock_list, list) {
 640                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 641                     geneve_get_sk_family(gs) == family) {
 642                         return gs;
 643                 }
 644         }
 645         return NULL;
 646 }
 647
 648 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 649 {
 650         struct net *net = geneve->net;
 651         struct geneve_net *gn = net_generic(net, geneve_net_id);
 652         struct geneve_dev_node *node;
 653         struct geneve_sock *gs;
 654         __u8 vni[3];
 655         __u32 hash;
 656
 657         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst);
 658         if (gs) {
 659                 gs->refcnt++;
 660                 goto out;
 661         }
 662
 663         gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6,
 664                                   geneve->use_udp6_rx_checksums);
 665         if (IS_ERR(gs))
 666                 return PTR_ERR(gs);
 667
 668 out:
 669         gs->collect_md = geneve->collect_md;
 670 #if IS_ENABLED(CONFIG_IPV6)
 671         if (ipv6) {
 672                 rcu_assign_pointer(geneve->sock6, gs);
 673                 node = &geneve->hlist6;
 674         } else
 675 #endif
 676         {
 677                 rcu_assign_pointer(geneve->sock4, gs);
 678                 node = &geneve->hlist4;
 679         }
 680         node->geneve = geneve;
 681
 682         tunnel_id_to_vni(geneve->info.key.tun_id, vni);
 683         hash = geneve_net_vni_hash(vni);
 684         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 685         return 0;
 686 }
 687
 688 static int geneve_open(struct net_device *dev)
 689 {
 690         struct geneve_dev *geneve = netdev_priv(dev);
 691         bool metadata = geneve->collect_md;
 692         bool ipv4, ipv6;
 693         int ret = 0;
 694
 695         ipv6 = geneve->info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 696         ipv4 = !ipv6 || metadata;
 697 #if IS_ENABLED(CONFIG_IPV6)
 698         if (ipv6) {
 699                 ret = geneve_sock_add(geneve, true);
 700                 if (ret < 0 && ret != -EAFNOSUPPORT)
 701                         ipv4 = false;
 702         }
 703 #endif
 704         if (ipv4)
 705                 ret = geneve_sock_add(geneve, false);
 706         if (ret < 0)
 707                 geneve_sock_release(geneve);
 708
 709         return ret;
 710 }
 711
 712 static int geneve_stop(struct net_device *dev)
 713 {
 714         struct geneve_dev *geneve = netdev_priv(dev);
 715
 716         hlist_del_init_rcu(&geneve->hlist4.hlist);
 717 #if IS_ENABLED(CONFIG_IPV6)
 718         hlist_del_init_rcu(&geneve->hlist6.hlist);
 719 #endif
 720         geneve_sock_release(geneve);
 721         return 0;
 722 }
 723
 724 static void geneve_build_header(struct genevehdr *geneveh,
 725                                 const struct ip_tunnel_info *info)
 726 {
 727         geneveh->ver = GENEVE_VER;
 728         geneveh->opt_len = info->options_len / 4;
 729         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 730         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 731         geneveh->rsvd1 = 0;
 732         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 733         geneveh->proto_type = htons(ETH_P_TEB);
 734         geneveh->rsvd2 = 0;
 735
 736         if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
 737                 ip_tunnel_info_opts_get(geneveh->options, info);
 738 }
 739
 740 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 741                             const struct ip_tunnel_info *info,
 742                             bool xnet, int ip_hdr_len)
 743 {
 744         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 745         struct genevehdr *gnvh;
 746         int min_headroom;
 747         int err;
 748
 749         skb_reset_mac_header(skb);
 750         skb_scrub_packet(skb, xnet);
 751
 752         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 753                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 754         err = skb_cow_head(skb, min_headroom);
 755         if (unlikely(err))
 756                 goto free_dst;
 757
 758         err = udp_tunnel_handle_offloads(skb, udp_sum);
 759         if (err)
 760                 goto free_dst;
 761
 762         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 763         geneve_build_header(gnvh, info);
 764         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 765         return 0;
 766
 767 free_dst:
 768         dst_release(dst);
 769         return err;
 770 }
 771
 772 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 773                                        struct net_device *dev,
 774                                        struct geneve_sock *gs4,
 775                                        struct flowi4 *fl4,
 776                                        const struct ip_tunnel_info *info)
 777 {
 778         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 779         struct geneve_dev *geneve = netdev_priv(dev);
 780         struct dst_cache *dst_cache;
 781         struct rtable *rt = NULL;
 782         __u8 tos;
 783
 784         if (!gs4)
 785                 return ERR_PTR(-EIO);
 786
 787         memset(fl4, 0, sizeof(*fl4));
 788         fl4->flowi4_mark = skb->mark;
 789         fl4->flowi4_proto = IPPROTO_UDP;
 790         fl4->daddr = info->key.u.ipv4.dst;
 791         fl4->saddr = info->key.u.ipv4.src;
 792
 793         tos = info->key.tos;
 794         if ((tos == 1) && !geneve->collect_md) {
 795                 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 796                 use_cache = false;
 797         }
 798         fl4->flowi4_tos = RT_TOS(tos);
 799
 800         dst_cache = (struct dst_cache *)&info->dst_cache;
 801         if (use_cache) {
 802                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 803                 if (rt)
 804                         return rt;
 805         }
 806         rt = ip_route_output_key(geneve->net, fl4);
 807         if (IS_ERR(rt)) {
 808                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 809                 return ERR_PTR(-ENETUNREACH);
 810         }
 811         if (rt->dst.dev == dev) { /* is this necessary? */
 812                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 813                 ip_rt_put(rt);
 814                 return ERR_PTR(-ELOOP);
 815         }
 816         if (use_cache)
 817                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 818         return rt;
 819 }
 820
 821 #if IS_ENABLED(CONFIG_IPV6)
 822 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 823                                            struct net_device *dev,
 824                                            struct geneve_sock *gs6,
 825                                            struct flowi6 *fl6,
 826                                            const struct ip_tunnel_info *info)
 827 {
 828         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 829         struct geneve_dev *geneve = netdev_priv(dev);
 830         struct dst_entry *dst = NULL;
 831         struct dst_cache *dst_cache;
 832         __u8 prio;
 833
 834         if (!gs6)
 835                 return ERR_PTR(-EIO);
 836
 837         memset(fl6, 0, sizeof(*fl6));
 838         fl6->flowi6_mark = skb->mark;
 839         fl6->flowi6_proto = IPPROTO_UDP;
 840         fl6->daddr = info->key.u.ipv6.dst;
 841         fl6->saddr = info->key.u.ipv6.src;
 842         prio = info->key.tos;
 843         if ((prio == 1) && !geneve->collect_md) {
 844                 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 845                 use_cache = false;
 846         }
 847
 848         fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 849                                            info->key.label);
 850         dst_cache = (struct dst_cache *)&info->dst_cache;
 851         if (use_cache) {
 852                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 853                 if (dst)
 854                         return dst;
 855         }
 856         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 857                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 858                 return ERR_PTR(-ENETUNREACH);
 859         }
 860         if (dst->dev == dev) { /* is this necessary? */
 861                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 862                 dst_release(dst);
 863                 return ERR_PTR(-ELOOP);
 864         }
 865
 866         if (use_cache)
 867                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 868         return dst;
 869 }
 870 #endif
 871
 872 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 873                            struct geneve_dev *geneve,
 874                            const struct ip_tunnel_info *info)
 875 {
 876         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 877         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 878         const struct ip_tunnel_key *key = &info->key;
 879         struct rtable *rt;
 880         struct flowi4 fl4;
 881         __u8 tos, ttl;
 882         __be16 df = 0;
 883         __be16 sport;
 884         int err;
 885
 886         rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
 887         if (IS_ERR(rt))
 888                 return PTR_ERR(rt);
 889
 890         skb_tunnel_check_pmtu(skb, &rt->dst,
 891                               GENEVE_IPV4_HLEN + info->options_len);
 892
 893         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 894         if (geneve->collect_md) {
 895                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 896                 ttl = key->ttl;
 897
 898                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 899         } else {
 900                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 901                 if (geneve->ttl_inherit)
 902                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 903                 else
 904                         ttl = key->ttl;
 905                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 906
 907                 if (geneve->df == GENEVE_DF_SET) {
 908                         df = htons(IP_DF);
 909                 } else if (geneve->df == GENEVE_DF_INHERIT) {
 910                         struct ethhdr *eth = eth_hdr(skb);
 911
 912                         if (ntohs(eth->h_proto) == ETH_P_IPV6) {
 913                                 df = htons(IP_DF);
 914                         } else if (ntohs(eth->h_proto) == ETH_P_IP) {
 915                                 struct iphdr *iph = ip_hdr(skb);
 916
 917                                 if (iph->frag_off & htons(IP_DF))
 918                                         df = htons(IP_DF);
 919                         }
 920                 }
 921         }
 922
 923         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
 924         if (unlikely(err))
 925                 return err;
 926
 927         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 928                             tos, ttl, df, sport, geneve->info.key.tp_dst,
 929                             !net_eq(geneve->net, dev_net(geneve->dev)),
 930                             !(info->key.tun_flags & TUNNEL_CSUM));
 931         return 0;
 932 }
 933
 934 #if IS_ENABLED(CONFIG_IPV6)
 935 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 936                             struct geneve_dev *geneve,
 937                             const struct ip_tunnel_info *info)
 938 {
 939         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 940         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 941         const struct ip_tunnel_key *key = &info->key;
 942         struct dst_entry *dst = NULL;
 943         struct flowi6 fl6;
 944         __u8 prio, ttl;
 945         __be16 sport;
 946         int err;
 947
 948         dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
 949         if (IS_ERR(dst))
 950                 return PTR_ERR(dst);
 951
 952         skb_tunnel_check_pmtu(skb, dst, GENEVE_IPV6_HLEN + info->options_len);
 953
 954         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 955         if (geneve->collect_md) {
 956                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 957                 ttl = key->ttl;
 958         } else {
 959                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
 960                                            ip_hdr(skb), skb);
 961                 if (geneve->ttl_inherit)
 962                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 963                 else
 964                         ttl = key->ttl;
 965                 ttl = ttl ? : ip6_dst_hoplimit(dst);
 966         }
 967         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
 968         if (unlikely(err))
 969                 return err;
 970
 971         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
 972                              &fl6.saddr, &fl6.daddr, prio, ttl,
 973                              info->key.label, sport, geneve->info.key.tp_dst,
 974                              !(info->key.tun_flags & TUNNEL_CSUM));
 975         return 0;
 976 }
 977 #endif
 978
 979 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
 980 {
 981         struct geneve_dev *geneve = netdev_priv(dev);
 982         struct ip_tunnel_info *info = NULL;
 983         int err;
 984
 985         if (geneve->collect_md) {
 986                 info = skb_tunnel_info(skb);
 987                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 988                         err = -EINVAL;
 989                         netdev_dbg(dev, "no tunnel metadata\n");
 990                         goto tx_error;
 991                 }
 992         } else {
 993                 info = &geneve->info;
 994         }
 995
 996         rcu_read_lock();
 997 #if IS_ENABLED(CONFIG_IPV6)
 998         if (info->mode & IP_TUNNEL_INFO_IPV6)
 999                 err = geneve6_xmit_skb(skb, dev, geneve, info);
1000         else
1001 #endif
1002                 err = geneve_xmit_skb(skb, dev, geneve, info);
1003         rcu_read_unlock();
1004
1005         if (likely(!err))
1006                 return NETDEV_TX_OK;
1007 tx_error:
1008         dev_kfree_skb(skb);
1009
1010         if (err == -ELOOP)
1011                 dev->stats.collisions++;
1012         else if (err == -ENETUNREACH)
1013                 dev->stats.tx_carrier_errors++;
1014
1015         dev->stats.tx_errors++;
1016         return NETDEV_TX_OK;
1017 }
1018
1019 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1020 {
1021         if (new_mtu > dev->max_mtu)
1022                 new_mtu = dev->max_mtu;
1023         else if (new_mtu < dev->min_mtu)
1024                 new_mtu = dev->min_mtu;
1025
1026         dev->mtu = new_mtu;
1027         return 0;
1028 }
1029
1030 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1031 {
1032         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1033         struct geneve_dev *geneve = netdev_priv(dev);
1034
1035         if (ip_tunnel_info_af(info) == AF_INET) {
1036                 struct rtable *rt;
1037                 struct flowi4 fl4;
1038                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1039
1040                 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
1041                 if (IS_ERR(rt))
1042                         return PTR_ERR(rt);
1043
1044                 ip_rt_put(rt);
1045                 info->key.u.ipv4.src = fl4.saddr;
1046 #if IS_ENABLED(CONFIG_IPV6)
1047         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1048                 struct dst_entry *dst;
1049                 struct flowi6 fl6;
1050                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1051
1052                 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
1053                 if (IS_ERR(dst))
1054                         return PTR_ERR(dst);
1055
1056                 dst_release(dst);
1057                 info->key.u.ipv6.src = fl6.saddr;
1058 #endif
1059         } else {
1060                 return -EINVAL;
1061         }
1062
1063         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1064                                              1, USHRT_MAX, true);
1065         info->key.tp_dst = geneve->info.key.tp_dst;
1066         return 0;
1067 }
1068
1069 static const struct net_device_ops geneve_netdev_ops = {
1070         .ndo_init               = geneve_init,
1071         .ndo_uninit             = geneve_uninit,
1072         .ndo_open               = geneve_open,
1073         .ndo_stop               = geneve_stop,
1074         .ndo_start_xmit         = geneve_xmit,
1075         .ndo_get_stats64        = ip_tunnel_get_stats64,
1076         .ndo_change_mtu         = geneve_change_mtu,
1077         .ndo_validate_addr      = eth_validate_addr,
1078         .ndo_set_mac_address    = eth_mac_addr,
1079         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1080 };
1081
1082 static void geneve_get_drvinfo(struct net_device *dev,
1083                                struct ethtool_drvinfo *drvinfo)
1084 {
1085         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1086         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1087 }
1088
1089 static const struct ethtool_ops geneve_ethtool_ops = {
1090         .get_drvinfo    = geneve_get_drvinfo,
1091         .get_link       = ethtool_op_get_link,
1092 };
1093
1094 /* Info for udev, that this is a virtual tunnel endpoint */
1095 static struct device_type geneve_type = {
1096         .name = "geneve",
1097 };
1098
1099 /* Calls the ndo_udp_tunnel_add of the caller in order to
1100  * supply the listening GENEVE udp ports. Callers are expected
1101  * to implement the ndo_udp_tunnel_add.
1102  */
1103 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1104 {
1105         struct net *net = dev_net(dev);
1106         struct geneve_net *gn = net_generic(net, geneve_net_id);
1107         struct geneve_sock *gs;
1108
1109         rcu_read_lock();
1110         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1111                 if (push) {
1112                         udp_tunnel_push_rx_port(dev, gs->sock,
1113                                                 UDP_TUNNEL_TYPE_GENEVE);
1114                 } else {
1115                         udp_tunnel_drop_rx_port(dev, gs->sock,
1116                                                 UDP_TUNNEL_TYPE_GENEVE);
1117                 }
1118         }
1119         rcu_read_unlock();
1120 }
1121
1122 /* Initialize the device structure. */
1123 static void geneve_setup(struct net_device *dev)
1124 {
1125         ether_setup(dev);
1126
1127         dev->netdev_ops = &geneve_netdev_ops;
1128         dev->ethtool_ops = &geneve_ethtool_ops;
1129         dev->needs_free_netdev = true;
1130
1131         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1132
1133         dev->features    |= NETIF_F_LLTX;
1134         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1135         dev->features    |= NETIF_F_RXCSUM;
1136         dev->features    |= NETIF_F_GSO_SOFTWARE;
1137
1138         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1139         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1140
1141         /* MTU range: 68 - (something less than 65535) */
1142         dev->min_mtu = ETH_MIN_MTU;
1143         /* The max_mtu calculation does not take account of GENEVE
1144          * options, to avoid excluding potentially valid
1145          * configurations. This will be further reduced by IPvX hdr size.
1146          */
1147         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1148
1149         netif_keep_dst(dev);
1150         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1151         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1152         eth_hw_addr_random(dev);
1153 }
1154
1155 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1156         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1157         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1158         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1159         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1160         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1161         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1162         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1163         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1164         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1165         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1166         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1167         [IFLA_GENEVE_TTL_INHERIT]       = { .type = NLA_U8 },
1168         [IFLA_GENEVE_DF]                = { .type = NLA_U8 },
1169 };
1170
1171 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1172                            struct netlink_ext_ack *extack)
1173 {
1174         if (tb[IFLA_ADDRESS]) {
1175                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1176                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1177                                             "Provided link layer address is not Ethernet");
1178                         return -EINVAL;
1179                 }
1180
1181                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1182                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1183                                             "Provided Ethernet address is not unicast");
1184                         return -EADDRNOTAVAIL;
1185                 }
1186         }
1187
1188         if (!data) {
1189                 NL_SET_ERR_MSG(extack,
1190                                "Not enough attributes provided to perform the operation");
1191                 return -EINVAL;
1192         }
1193
1194         if (data[IFLA_GENEVE_ID]) {
1195                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1196
1197                 if (vni >= GENEVE_N_VID) {
1198                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1199                                             "Geneve ID must be lower than 16777216");
1200                         return -ERANGE;
1201                 }
1202         }
1203
1204         if (data[IFLA_GENEVE_DF]) {
1205                 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1206
1207                 if (df < 0 || df > GENEVE_DF_MAX) {
1208                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_GENEVE_DF],
1209                                             "Invalid DF attribute");
1210                         return -EINVAL;
1211                 }
1212         }
1213
1214         return 0;
1215 }
1216
1217 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1218                                           const struct ip_tunnel_info *info,
1219                                           bool *tun_on_same_port,
1220                                           bool *tun_collect_md)
1221 {
1222         struct geneve_dev *geneve, *t = NULL;
1223
1224         *tun_on_same_port = false;
1225         *tun_collect_md = false;
1226         list_for_each_entry(geneve, &gn->geneve_list, next) {
1227                 if (info->key.tp_dst == geneve->info.key.tp_dst) {
1228                         *tun_collect_md = geneve->collect_md;
1229                         *tun_on_same_port = true;
1230                 }
1231                 if (info->key.tun_id == geneve->info.key.tun_id &&
1232                     info->key.tp_dst == geneve->info.key.tp_dst &&
1233                     !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1234                         t = geneve;
1235         }
1236         return t;
1237 }
1238
1239 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1240 {
1241         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1242                  info->key.ttl || info->key.label || info->key.tp_src ||
1243                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1244 }
1245
1246 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1247                                   struct ip_tunnel_info *b)
1248 {
1249         if (ip_tunnel_info_af(a) == AF_INET)
1250                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1251         else
1252                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1253 }
1254
1255 static int geneve_configure(struct net *net, struct net_device *dev,
1256                             struct netlink_ext_ack *extack,
1257                             const struct ip_tunnel_info *info,
1258                             bool metadata, bool ipv6_rx_csum,
1259                             bool ttl_inherit, enum ifla_geneve_df df)
1260 {
1261         struct geneve_net *gn = net_generic(net, geneve_net_id);
1262         struct geneve_dev *t, *geneve = netdev_priv(dev);
1263         bool tun_collect_md, tun_on_same_port;
1264         int err, encap_len;
1265
1266         if (metadata && !is_tnl_info_zero(info)) {
1267                 NL_SET_ERR_MSG(extack,
1268                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1269                 return -EINVAL;
1270         }
1271
1272         geneve->net = net;
1273         geneve->dev = dev;
1274
1275         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1276         if (t)
1277                 return -EBUSY;
1278
1279         /* make enough headroom for basic scenario */
1280         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1281         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1282                 encap_len += sizeof(struct iphdr);
1283                 dev->max_mtu -= sizeof(struct iphdr);
1284         } else {
1285                 encap_len += sizeof(struct ipv6hdr);
1286                 dev->max_mtu -= sizeof(struct ipv6hdr);
1287         }
1288         dev->needed_headroom = encap_len + ETH_HLEN;
1289
1290         if (metadata) {
1291                 if (tun_on_same_port) {
1292                         NL_SET_ERR_MSG(extack,
1293                                        "There can be only one externally controlled device on a destination port");
1294                         return -EPERM;
1295                 }
1296         } else {
1297                 if (tun_collect_md) {
1298                         NL_SET_ERR_MSG(extack,
1299                                        "There already exists an externally controlled device on this destination port");
1300                         return -EPERM;
1301                 }
1302         }
1303
1304         dst_cache_reset(&geneve->info.dst_cache);
1305         geneve->info = *info;
1306         geneve->collect_md = metadata;
1307         geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1308         geneve->ttl_inherit = ttl_inherit;
1309         geneve->df = df;
1310
1311         err = register_netdevice(dev);
1312         if (err)
1313                 return err;
1314
1315         list_add(&geneve->next, &gn->geneve_list);
1316         return 0;
1317 }
1318
1319 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1320 {
1321         memset(info, 0, sizeof(*info));
1322         info->key.tp_dst = htons(dst_port);
1323 }
1324
1325 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1326                           struct netlink_ext_ack *extack,
1327                           struct ip_tunnel_info *info, bool *metadata,
1328                           bool *use_udp6_rx_checksums, bool *ttl_inherit,
1329                           enum ifla_geneve_df *df, bool changelink)
1330 {
1331         int attrtype;
1332
1333         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1334                 NL_SET_ERR_MSG(extack,
1335                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1336                 return -EINVAL;
1337         }
1338
1339         if (data[IFLA_GENEVE_REMOTE]) {
1340                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1341                         attrtype = IFLA_GENEVE_REMOTE;
1342                         goto change_notsup;
1343                 }
1344
1345                 info->key.u.ipv4.dst =
1346                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1347
1348                 if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) {
1349                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1350                                             "Remote IPv4 address cannot be Multicast");
1351                         return -EINVAL;
1352                 }
1353         }
1354
1355         if (data[IFLA_GENEVE_REMOTE6]) {
1356 #if IS_ENABLED(CONFIG_IPV6)
1357                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1358                         attrtype = IFLA_GENEVE_REMOTE6;
1359                         goto change_notsup;
1360                 }
1361
1362                 info->mode = IP_TUNNEL_INFO_IPV6;
1363                 info->key.u.ipv6.dst =
1364                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1365
1366                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1367                     IPV6_ADDR_LINKLOCAL) {
1368                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1369                                             "Remote IPv6 address cannot be link-local");
1370                         return -EINVAL;
1371                 }
1372                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1373                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1374                                             "Remote IPv6 address cannot be Multicast");
1375                         return -EINVAL;
1376                 }
1377                 info->key.tun_flags |= TUNNEL_CSUM;
1378                 *use_udp6_rx_checksums = true;
1379 #else
1380                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1381                                     "IPv6 support not enabled in the kernel");
1382                 return -EPFNOSUPPORT;
1383 #endif
1384         }
1385
1386         if (data[IFLA_GENEVE_ID]) {
1387                 __u32 vni;
1388                 __u8 tvni[3];
1389                 __be64 tunid;
1390
1391                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1392                 tvni[0] = (vni & 0x00ff0000) >> 16;
1393                 tvni[1] = (vni & 0x0000ff00) >> 8;
1394                 tvni[2] =  vni & 0x000000ff;
1395
1396                 tunid = vni_to_tunnel_id(tvni);
1397                 if (changelink && (tunid != info->key.tun_id)) {
1398                         attrtype = IFLA_GENEVE_ID;
1399                         goto change_notsup;
1400                 }
1401                 info->key.tun_id = tunid;
1402         }
1403
1404         if (data[IFLA_GENEVE_TTL_INHERIT]) {
1405                 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1406                         *ttl_inherit = true;
1407                 else
1408                         *ttl_inherit = false;
1409         } else if (data[IFLA_GENEVE_TTL]) {
1410                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1411                 *ttl_inherit = false;
1412         }
1413
1414         if (data[IFLA_GENEVE_TOS])
1415                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1416
1417         if (data[IFLA_GENEVE_DF])
1418                 *df = nla_get_u8(data[IFLA_GENEVE_DF]);
1419
1420         if (data[IFLA_GENEVE_LABEL]) {
1421                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1422                                   IPV6_FLOWLABEL_MASK;
1423                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1424                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1425                                             "Label attribute only applies for IPv6 Geneve devices");
1426                         return -EINVAL;
1427                 }
1428         }
1429
1430         if (data[IFLA_GENEVE_PORT]) {
1431                 if (changelink) {
1432                         attrtype = IFLA_GENEVE_PORT;
1433                         goto change_notsup;
1434                 }
1435                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1436         }
1437
1438         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1439                 if (changelink) {
1440                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1441                         goto change_notsup;
1442                 }
1443                 *metadata = true;
1444         }
1445
1446         if (data[IFLA_GENEVE_UDP_CSUM]) {
1447                 if (changelink) {
1448                         attrtype = IFLA_GENEVE_UDP_CSUM;
1449                         goto change_notsup;
1450                 }
1451                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1452                         info->key.tun_flags |= TUNNEL_CSUM;
1453         }
1454
1455         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1456 #if IS_ENABLED(CONFIG_IPV6)
1457                 if (changelink) {
1458                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1459                         goto change_notsup;
1460                 }
1461                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1462                         info->key.tun_flags &= ~TUNNEL_CSUM;
1463 #else
1464                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1465                                     "IPv6 support not enabled in the kernel");
1466                 return -EPFNOSUPPORT;
1467 #endif
1468         }
1469
1470         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1471 #if IS_ENABLED(CONFIG_IPV6)
1472                 if (changelink) {
1473                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1474                         goto change_notsup;
1475                 }
1476                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1477                         *use_udp6_rx_checksums = false;
1478 #else
1479                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1480                                     "IPv6 support not enabled in the kernel");
1481                 return -EPFNOSUPPORT;
1482 #endif
1483         }
1484
1485         return 0;
1486 change_notsup:
1487         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1488                             "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1489         return -EOPNOTSUPP;
1490 }
1491
1492 static void geneve_link_config(struct net_device *dev,
1493                                struct ip_tunnel_info *info, struct nlattr *tb[])
1494 {
1495         struct geneve_dev *geneve = netdev_priv(dev);
1496         int ldev_mtu = 0;
1497
1498         if (tb[IFLA_MTU]) {
1499                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1500                 return;
1501         }
1502
1503         switch (ip_tunnel_info_af(info)) {
1504         case AF_INET: {
1505                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1506                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1507
1508                 if (!IS_ERR(rt) && rt->dst.dev) {
1509                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1510                         ip_rt_put(rt);
1511                 }
1512                 break;
1513         }
1514 #if IS_ENABLED(CONFIG_IPV6)
1515         case AF_INET6: {
1516                 struct rt6_info *rt;
1517
1518                 if (!__in6_dev_get(dev))
1519                         break;
1520
1521                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1522                                 NULL, 0);
1523
1524                 if (rt && rt->dst.dev)
1525                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1526                 ip6_rt_put(rt);
1527                 break;
1528         }
1529 #endif
1530         }
1531
1532         if (ldev_mtu <= 0)
1533                 return;
1534
1535         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1536 }
1537
1538 static int geneve_newlink(struct net *net, struct net_device *dev,
1539                           struct nlattr *tb[], struct nlattr *data[],
1540                           struct netlink_ext_ack *extack)
1541 {
1542         enum ifla_geneve_df df = GENEVE_DF_UNSET;
1543         bool use_udp6_rx_checksums = false;
1544         struct ip_tunnel_info info;
1545         bool ttl_inherit = false;
1546         bool metadata = false;
1547         int err;
1548
1549         init_tnl_info(&info, GENEVE_UDP_PORT);
1550         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1551                              &use_udp6_rx_checksums, &ttl_inherit, &df, false);
1552         if (err)
1553                 return err;
1554
1555         err = geneve_configure(net, dev, extack, &info, metadata,
1556                                use_udp6_rx_checksums, ttl_inherit, df);
1557         if (err)
1558                 return err;
1559
1560         geneve_link_config(dev, &info, tb);
1561
1562         return 0;
1563 }
1564
1565 /* Quiesces the geneve device data path for both TX and RX.
1566  *
1567  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1568  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1569  * to complete for the existing set of in-flight packets to be transmitted,
1570  * then we would have quiesced the transmit data path. All the future packets
1571  * will get dropped until we unquiesce the data path.
1572  *
1573  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1574  * if we set that to NULL under RCU and wait for synchronize_net() to
1575  * complete, then we would have quiesced the receive data path.
1576  */
1577 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1578                            struct geneve_sock **gs6)
1579 {
1580         *gs4 = rtnl_dereference(geneve->sock4);
1581         rcu_assign_pointer(geneve->sock4, NULL);
1582         if (*gs4)
1583                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1584 #if IS_ENABLED(CONFIG_IPV6)
1585         *gs6 = rtnl_dereference(geneve->sock6);
1586         rcu_assign_pointer(geneve->sock6, NULL);
1587         if (*gs6)
1588                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1589 #else
1590         *gs6 = NULL;
1591 #endif
1592         synchronize_net();
1593 }
1594
1595 /* Resumes the geneve device data path for both TX and RX. */
1596 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1597                              struct geneve_sock __maybe_unused *gs6)
1598 {
1599         rcu_assign_pointer(geneve->sock4, gs4);
1600         if (gs4)
1601                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1602 #if IS_ENABLED(CONFIG_IPV6)
1603         rcu_assign_pointer(geneve->sock6, gs6);
1604         if (gs6)
1605                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1606 #endif
1607         synchronize_net();
1608 }
1609
1610 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1611                              struct nlattr *data[],
1612                              struct netlink_ext_ack *extack)
1613 {
1614         struct geneve_dev *geneve = netdev_priv(dev);
1615         struct geneve_sock *gs4, *gs6;
1616         struct ip_tunnel_info info;
1617         bool metadata;
1618         bool use_udp6_rx_checksums;
1619         enum ifla_geneve_df df;
1620         bool ttl_inherit;
1621         int err;
1622
1623         /* If the geneve device is configured for metadata (or externally
1624          * controlled, for example, OVS), then nothing can be changed.
1625          */
1626         if (geneve->collect_md)
1627                 return -EOPNOTSUPP;
1628
1629         /* Start with the existing info. */
1630         memcpy(&info, &geneve->info, sizeof(info));
1631         metadata = geneve->collect_md;
1632         use_udp6_rx_checksums = geneve->use_udp6_rx_checksums;
1633         ttl_inherit = geneve->ttl_inherit;
1634         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1635                              &use_udp6_rx_checksums, &ttl_inherit, &df, true);
1636         if (err)
1637                 return err;
1638
1639         if (!geneve_dst_addr_equal(&geneve->info, &info)) {
1640                 dst_cache_reset(&info.dst_cache);
1641                 geneve_link_config(dev, &info, tb);
1642         }
1643
1644         geneve_quiesce(geneve, &gs4, &gs6);
1645         geneve->info = info;
1646         geneve->collect_md = metadata;
1647         geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
1648         geneve->ttl_inherit = ttl_inherit;
1649         geneve_unquiesce(geneve, gs4, gs6);
1650
1651         return 0;
1652 }
1653
1654 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1655 {
1656         struct geneve_dev *geneve = netdev_priv(dev);
1657
1658         list_del(&geneve->next);
1659         unregister_netdevice_queue(dev, head);
1660 }
1661
1662 static size_t geneve_get_size(const struct net_device *dev)
1663 {
1664         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1665                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1666                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1667                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1668                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_DF */
1669                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1670                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1671                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1672                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1673                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1674                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1675                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1676                 0;
1677 }
1678
1679 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1680 {
1681         struct geneve_dev *geneve = netdev_priv(dev);
1682         struct ip_tunnel_info *info = &geneve->info;
1683         bool ttl_inherit = geneve->ttl_inherit;
1684         bool metadata = geneve->collect_md;
1685         __u8 tmp_vni[3];
1686         __u32 vni;
1687
1688         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1689         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1690         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1691                 goto nla_put_failure;
1692
1693         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1694                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1695                                     info->key.u.ipv4.dst))
1696                         goto nla_put_failure;
1697                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1698                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1699                         goto nla_put_failure;
1700
1701 #if IS_ENABLED(CONFIG_IPV6)
1702         } else if (!metadata) {
1703                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1704                                      &info->key.u.ipv6.dst))
1705                         goto nla_put_failure;
1706                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1707                                !(info->key.tun_flags & TUNNEL_CSUM)))
1708                         goto nla_put_failure;
1709 #endif
1710         }
1711
1712         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1713             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1714             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1715                 goto nla_put_failure;
1716
1717         if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->df))
1718                 goto nla_put_failure;
1719
1720         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1721                 goto nla_put_failure;
1722
1723         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1724                 goto nla_put_failure;
1725
1726 #if IS_ENABLED(CONFIG_IPV6)
1727         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1728                        !geneve->use_udp6_rx_checksums))
1729                 goto nla_put_failure;
1730 #endif
1731
1732         if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1733                 goto nla_put_failure;
1734
1735         return 0;
1736
1737 nla_put_failure:
1738         return -EMSGSIZE;
1739 }
1740
1741 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1742         .kind           = "geneve",
1743         .maxtype        = IFLA_GENEVE_MAX,
1744         .policy         = geneve_policy,
1745         .priv_size      = sizeof(struct geneve_dev),
1746         .setup          = geneve_setup,
1747         .validate       = geneve_validate,
1748         .newlink        = geneve_newlink,
1749         .changelink     = geneve_changelink,
1750         .dellink        = geneve_dellink,
1751         .get_size       = geneve_get_size,
1752         .fill_info      = geneve_fill_info,
1753 };
1754
1755 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1756                                         u8 name_assign_type, u16 dst_port)
1757 {
1758         struct nlattr *tb[IFLA_MAX + 1];
1759         struct ip_tunnel_info info;
1760         struct net_device *dev;
1761         LIST_HEAD(list_kill);
1762         int err;
1763
1764         memset(tb, 0, sizeof(tb));
1765         dev = rtnl_create_link(net, name, name_assign_type,
1766                                &geneve_link_ops, tb, NULL);
1767         if (IS_ERR(dev))
1768                 return dev;
1769
1770         init_tnl_info(&info, dst_port);
1771         err = geneve_configure(net, dev, NULL, &info,
1772                                true, true, false, GENEVE_DF_UNSET);
1773         if (err) {
1774                 free_netdev(dev);
1775                 return ERR_PTR(err);
1776         }
1777
1778         /* openvswitch users expect packet sizes to be unrestricted,
1779          * so set the largest MTU we can.
1780          */
1781         err = geneve_change_mtu(dev, IP_MAX_MTU);
1782         if (err)
1783                 goto err;
1784
1785         err = rtnl_configure_link(dev, NULL);
1786         if (err < 0)
1787                 goto err;
1788
1789         return dev;
1790 err:
1791         geneve_dellink(dev, &list_kill);
1792         unregister_netdevice_many(&list_kill);
1793         return ERR_PTR(err);
1794 }
1795 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1796
1797 static int geneve_netdevice_event(struct notifier_block *unused,
1798                                   unsigned long event, void *ptr)
1799 {
1800         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1801
1802         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1803             event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1804                 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1805         } else if (event == NETDEV_UNREGISTER) {
1806                 geneve_offload_rx_ports(dev, false);
1807         } else if (event == NETDEV_REGISTER) {
1808                 geneve_offload_rx_ports(dev, true);
1809         }
1810
1811         return NOTIFY_DONE;
1812 }
1813
1814 static struct notifier_block geneve_notifier_block __read_mostly = {
1815         .notifier_call = geneve_netdevice_event,
1816 };
1817
1818 static __net_init int geneve_init_net(struct net *net)
1819 {
1820         struct geneve_net *gn = net_generic(net, geneve_net_id);
1821
1822         INIT_LIST_HEAD(&gn->geneve_list);
1823         INIT_LIST_HEAD(&gn->sock_list);
1824         return 0;
1825 }
1826
1827 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1828 {
1829         struct geneve_net *gn = net_generic(net, geneve_net_id);
1830         struct geneve_dev *geneve, *next;
1831         struct net_device *dev, *aux;
1832
1833         /* gather any geneve devices that were moved into this ns */
1834         for_each_netdev_safe(net, dev, aux)
1835                 if (dev->rtnl_link_ops == &geneve_link_ops)
1836                         unregister_netdevice_queue(dev, head);
1837
1838         /* now gather any other geneve devices that were created in this ns */
1839         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1840                 /* If geneve->dev is in the same netns, it was already added
1841                  * to the list by the previous loop.
1842                  */
1843                 if (!net_eq(dev_net(geneve->dev), net))
1844                         unregister_netdevice_queue(geneve->dev, head);
1845         }
1846
1847         WARN_ON_ONCE(!list_empty(&gn->sock_list));
1848 }
1849
1850 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1851 {
1852         struct net *net;
1853         LIST_HEAD(list);
1854
1855         rtnl_lock();
1856         list_for_each_entry(net, net_list, exit_list)
1857                 geneve_destroy_tunnels(net, &list);
1858
1859         /* unregister the devices gathered above */
1860         unregister_netdevice_many(&list);
1861         rtnl_unlock();
1862 }
1863
1864 static struct pernet_operations geneve_net_ops = {
1865         .init = geneve_init_net,
1866         .exit_batch = geneve_exit_batch_net,
1867         .id   = &geneve_net_id,
1868         .size = sizeof(struct geneve_net),
1869 };
1870
1871 static int __init geneve_init_module(void)
1872 {
1873         int rc;
1874
1875         rc = register_pernet_subsys(&geneve_net_ops);
1876         if (rc)
1877                 goto out1;
1878
1879         rc = register_netdevice_notifier(&geneve_notifier_block);
1880         if (rc)
1881                 goto out2;
1882
1883         rc = rtnl_link_register(&geneve_link_ops);
1884         if (rc)
1885                 goto out3;
1886
1887         return 0;
1888 out3:
1889         unregister_netdevice_notifier(&geneve_notifier_block);
1890 out2:
1891         unregister_pernet_subsys(&geneve_net_ops);
1892 out1:
1893         return rc;
1894 }
1895 late_initcall(geneve_init_module);
1896
1897 static void __exit geneve_cleanup_module(void)
1898 {
1899         rtnl_link_unregister(&geneve_link_ops);
1900         unregister_netdevice_notifier(&geneve_notifier_block);
1901         unregister_pernet_subsys(&geneve_net_ops);
1902 }
1903 module_exit(geneve_cleanup_module);
1904
1905 MODULE_LICENSE("GPL");
1906 MODULE_VERSION(GENEVE_NETDEV_VER);
1907 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1908 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1909 MODULE_ALIAS_RTNL_LINK("geneve");