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 | (gnvh->oam ? TUNNEL_OAM : 0) |
 225                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 226
 227                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 228                                          vni_to_tunnel_id(gnvh->vni),
 229                                          gnvh->opt_len * 4);
 230                 if (!tun_dst) {
 231                         geneve->dev->stats.rx_dropped++;
 232                         goto drop;
 233                 }
 234                 /* Update tunnel dst according to Geneve options. */
 235                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 236                                         gnvh->options, gnvh->opt_len * 4,
 237                                         TUNNEL_GENEVE_OPT);
 238         } else {
 239                 /* Drop packets w/ critical options,
 240                  * since we don't support any...
 241                  */
 242                 if (gnvh->critical) {
 243                         geneve->dev->stats.rx_frame_errors++;
 244                         geneve->dev->stats.rx_errors++;
 245                         goto drop;
 246                 }
 247         }
 248
 249         skb_reset_mac_header(skb);
 250         skb->protocol = eth_type_trans(skb, geneve->dev);
 251         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 252
 253         if (tun_dst)
 254                 skb_dst_set(skb, &tun_dst->dst);
 255
 256         /* Ignore packet loops (and multicast echo) */
 257         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
 258                 geneve->dev->stats.rx_errors++;
 259                 goto drop;
 260         }
 261
 262         oiph = skb_network_header(skb);
 263         skb_reset_network_header(skb);
 264
 265         if (geneve_get_sk_family(gs) == AF_INET)
 266                 err = IP_ECN_decapsulate(oiph, skb);
 267 #if IS_ENABLED(CONFIG_IPV6)
 268         else
 269                 err = IP6_ECN_decapsulate(oiph, skb);
 270 #endif
 271
 272         if (unlikely(err)) {
 273                 if (log_ecn_error) {
 274                         if (geneve_get_sk_family(gs) == AF_INET)
 275                                 net_info_ratelimited("non-ECT from %pI4 "
 276                                                      "with TOS=%#x\n",
 277                                                      &((struct iphdr *)oiph)->saddr,
 278                                                      ((struct iphdr *)oiph)->tos);
 279 #if IS_ENABLED(CONFIG_IPV6)
 280                         else
 281                                 net_info_ratelimited("non-ECT from %pI6\n",
 282                                                      &((struct ipv6hdr *)oiph)->saddr);
 283 #endif
 284                 }
 285                 if (err > 1) {
 286                         ++geneve->dev->stats.rx_frame_errors;
 287                         ++geneve->dev->stats.rx_errors;
 288                         goto drop;
 289                 }
 290         }
 291
 292         len = skb->len;
 293         err = gro_cells_receive(&geneve->gro_cells, skb);
 294         if (likely(err == NET_RX_SUCCESS)) {
 295                 stats = this_cpu_ptr(geneve->dev->tstats);
 296                 u64_stats_update_begin(&stats->syncp);
 297                 stats->rx_packets++;
 298                 stats->rx_bytes += len;
 299                 u64_stats_update_end(&stats->syncp);
 300         }
 301         return;
 302 drop:
 303         /* Consume bad packet */
 304         kfree_skb(skb);
 305 }
 306
 307 /* Setup stats when device is created */
 308 static int geneve_init(struct net_device *dev)
 309 {
 310         struct geneve_dev *geneve = netdev_priv(dev);
 311         int err;
 312
 313         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 314         if (!dev->tstats)
 315                 return -ENOMEM;
 316
 317         err = gro_cells_init(&geneve->gro_cells, dev);
 318         if (err) {
 319                 free_percpu(dev->tstats);
 320                 return err;
 321         }
 322
 323         err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL);
 324         if (err) {
 325                 free_percpu(dev->tstats);
 326                 gro_cells_destroy(&geneve->gro_cells);
 327                 return err;
 328         }
 329         return 0;
 330 }
 331
 332 static void geneve_uninit(struct net_device *dev)
 333 {
 334         struct geneve_dev *geneve = netdev_priv(dev);
 335
 336         dst_cache_destroy(&geneve->info.dst_cache);
 337         gro_cells_destroy(&geneve->gro_cells);
 338         free_percpu(dev->tstats);
 339 }
 340
 341 /* Callback from net/ipv4/udp.c to receive packets */
 342 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 343 {
 344         struct genevehdr *geneveh;
 345         struct geneve_dev *geneve;
 346         struct geneve_sock *gs;
 347         int opts_len;
 348
 349         /* Need UDP and Geneve header to be present */
 350         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 351                 goto drop;
 352
 353         /* Return packets with reserved bits set */
 354         geneveh = geneve_hdr(skb);
 355         if (unlikely(geneveh->ver != GENEVE_VER))
 356                 goto drop;
 357
 358         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 359                 goto drop;
 360
 361         gs = rcu_dereference_sk_user_data(sk);
 362         if (!gs)
 363                 goto drop;
 364
 365         geneve = geneve_lookup_skb(gs, skb);
 366         if (!geneve)
 367                 goto drop;
 368
 369         opts_len = geneveh->opt_len * 4;
 370         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 371                                  htons(ETH_P_TEB),
 372                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 373                 geneve->dev->stats.rx_dropped++;
 374                 goto drop;
 375         }
 376
 377         geneve_rx(geneve, gs, skb);
 378         return 0;
 379
 380 drop:
 381         /* Consume bad packet */
 382         kfree_skb(skb);
 383         return 0;
 384 }
 385
 386 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
 387 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
 388 {
 389         struct genevehdr *geneveh;
 390         struct geneve_sock *gs;
 391         u8 zero_vni[3] = { 0 };
 392         u8 *vni = zero_vni;
 393
 394         if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
 395                 return -EINVAL;
 396
 397         geneveh = geneve_hdr(skb);
 398         if (geneveh->ver != GENEVE_VER)
 399                 return -EINVAL;
 400
 401         if (geneveh->proto_type != htons(ETH_P_TEB))
 402                 return -EINVAL;
 403
 404         gs = rcu_dereference_sk_user_data(sk);
 405         if (!gs)
 406                 return -ENOENT;
 407
 408         if (geneve_get_sk_family(gs) == AF_INET) {
 409                 struct iphdr *iph = ip_hdr(skb);
 410                 __be32 addr4 = 0;
 411
 412                 if (!gs->collect_md) {
 413                         vni = geneve_hdr(skb)->vni;
 414                         addr4 = iph->daddr;
 415                 }
 416
 417                 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
 418         }
 419
 420 #if IS_ENABLED(CONFIG_IPV6)
 421         if (geneve_get_sk_family(gs) == AF_INET6) {
 422                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 423                 struct in6_addr addr6;
 424
 425                 memset(&addr6, 0, sizeof(struct in6_addr));
 426
 427                 if (!gs->collect_md) {
 428                         vni = geneve_hdr(skb)->vni;
 429                         addr6 = ip6h->daddr;
 430                 }
 431
 432                 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
 433         }
 434 #endif
 435
 436         return -EPFNOSUPPORT;
 437 }
 438
 439 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 440                                          __be16 port, bool ipv6_rx_csum)
 441 {
 442         struct socket *sock;
 443         struct udp_port_cfg udp_conf;
 444         int err;
 445
 446         memset(&udp_conf, 0, sizeof(udp_conf));
 447
 448         if (ipv6) {
 449                 udp_conf.family = AF_INET6;
 450                 udp_conf.ipv6_v6only = 1;
 451                 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
 452         } else {
 453                 udp_conf.family = AF_INET;
 454                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 455         }
 456
 457         udp_conf.local_udp_port = port;
 458
 459         /* Open UDP socket */
 460         err = udp_sock_create(net, &udp_conf, &sock);
 461         if (err < 0)
 462                 return ERR_PTR(err);
 463
 464         return sock;
 465 }
 466
 467 static int geneve_hlen(struct genevehdr *gh)
 468 {
 469         return sizeof(*gh) + gh->opt_len * 4;
 470 }
 471
 472 static struct sk_buff *geneve_gro_receive(struct sock *sk,
 473                                           struct list_head *head,
 474                                           struct sk_buff *skb)
 475 {
 476         struct sk_buff *pp = NULL;
 477         struct sk_buff *p;
 478         struct genevehdr *gh, *gh2;
 479         unsigned int hlen, gh_len, off_gnv;
 480         const struct packet_offload *ptype;
 481         __be16 type;
 482         int flush = 1;
 483
 484         off_gnv = skb_gro_offset(skb);
 485         hlen = off_gnv + sizeof(*gh);
 486         gh = skb_gro_header_fast(skb, off_gnv);
 487         if (skb_gro_header_hard(skb, hlen)) {
 488                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 489                 if (unlikely(!gh))
 490                         goto out;
 491         }
 492
 493         if (gh->ver != GENEVE_VER || gh->oam)
 494                 goto out;
 495         gh_len = geneve_hlen(gh);
 496
 497         hlen = off_gnv + gh_len;
 498         if (skb_gro_header_hard(skb, hlen)) {
 499                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 500                 if (unlikely(!gh))
 501                         goto out;
 502         }
 503
 504         list_for_each_entry(p, head, list) {
 505                 if (!NAPI_GRO_CB(p)->same_flow)
 506                         continue;
 507
 508                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 509                 if (gh->opt_len != gh2->opt_len ||
 510                     memcmp(gh, gh2, gh_len)) {
 511                         NAPI_GRO_CB(p)->same_flow = 0;
 512                         continue;
 513                 }
 514         }
 515
 516         type = gh->proto_type;
 517
 518         rcu_read_lock();
 519         ptype = gro_find_receive_by_type(type);
 520         if (!ptype)
 521                 goto out_unlock;
 522
 523         skb_gro_pull(skb, gh_len);
 524         skb_gro_postpull_rcsum(skb, gh, gh_len);
 525         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 526         flush = 0;
 527
 528 out_unlock:
 529         rcu_read_unlock();
 530 out:
 531         skb_gro_flush_final(skb, pp, flush);
 532
 533         return pp;
 534 }
 535
 536 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 537                                int nhoff)
 538 {
 539         struct genevehdr *gh;
 540         struct packet_offload *ptype;
 541         __be16 type;
 542         int gh_len;
 543         int err = -ENOSYS;
 544
 545         gh = (struct genevehdr *)(skb->data + nhoff);
 546         gh_len = geneve_hlen(gh);
 547         type = gh->proto_type;
 548
 549         rcu_read_lock();
 550         ptype = gro_find_complete_by_type(type);
 551         if (ptype)
 552                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 553
 554         rcu_read_unlock();
 555
 556         skb_set_inner_mac_header(skb, nhoff + gh_len);
 557
 558         return err;
 559 }
 560
 561 /* Create new listen socket if needed */
 562 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 563                                                 bool ipv6, bool ipv6_rx_csum)
 564 {
 565         struct geneve_net *gn = net_generic(net, geneve_net_id);
 566         struct geneve_sock *gs;
 567         struct socket *sock;
 568         struct udp_tunnel_sock_cfg tunnel_cfg;
 569         int h;
 570
 571         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 572         if (!gs)
 573                 return ERR_PTR(-ENOMEM);
 574
 575         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 576         if (IS_ERR(sock)) {
 577                 kfree(gs);
 578                 return ERR_CAST(sock);
 579         }
 580
 581         gs->sock = sock;
 582         gs->refcnt = 1;
 583         for (h = 0; h < VNI_HASH_SIZE; ++h)
 584                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 585
 586         /* Initialize the geneve udp offloads structure */
 587         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 588
 589         /* Mark socket as an encapsulation socket */
 590         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 591         tunnel_cfg.sk_user_data = gs;
 592         tunnel_cfg.encap_type = 1;
 593         tunnel_cfg.gro_receive = geneve_gro_receive;
 594         tunnel_cfg.gro_complete = geneve_gro_complete;
 595         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 596         tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
 597         tunnel_cfg.encap_destroy = NULL;
 598         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 599         list_add(&gs->list, &gn->sock_list);
 600         return gs;
 601 }
 602
 603 static void __geneve_sock_release(struct geneve_sock *gs)
 604 {
 605         if (!gs || --gs->refcnt)
 606                 return;
 607
 608         list_del(&gs->list);
 609         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 610         udp_tunnel_sock_release(gs->sock);
 611         kfree_rcu(gs, rcu);
 612 }
 613
 614 static void geneve_sock_release(struct geneve_dev *geneve)
 615 {
 616         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 617 #if IS_ENABLED(CONFIG_IPV6)
 618         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 619
 620         rcu_assign_pointer(geneve->sock6, NULL);
 621 #endif
 622
 623         rcu_assign_pointer(geneve->sock4, NULL);
 624         synchronize_net();
 625
 626         __geneve_sock_release(gs4);
 627 #if IS_ENABLED(CONFIG_IPV6)
 628         __geneve_sock_release(gs6);
 629 #endif
 630 }
 631
 632 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 633                                             sa_family_t family,
 634                                             __be16 dst_port)
 635 {
 636         struct geneve_sock *gs;
 637
 638         list_for_each_entry(gs, &gn->sock_list, list) {
 639                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 640                     geneve_get_sk_family(gs) == family) {
 641                         return gs;
 642                 }
 643         }
 644         return NULL;
 645 }
 646
 647 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 648 {
 649         struct net *net = geneve->net;
 650         struct geneve_net *gn = net_generic(net, geneve_net_id);
 651         struct geneve_dev_node *node;
 652         struct geneve_sock *gs;
 653         __u8 vni[3];
 654         __u32 hash;
 655
 656         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst);
 657         if (gs) {
 658                 gs->refcnt++;
 659                 goto out;
 660         }
 661
 662         gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6,
 663                                   geneve->use_udp6_rx_checksums);
 664         if (IS_ERR(gs))
 665                 return PTR_ERR(gs);
 666
 667 out:
 668         gs->collect_md = geneve->collect_md;
 669 #if IS_ENABLED(CONFIG_IPV6)
 670         if (ipv6) {
 671                 rcu_assign_pointer(geneve->sock6, gs);
 672                 node = &geneve->hlist6;
 673         } else
 674 #endif
 675         {
 676                 rcu_assign_pointer(geneve->sock4, gs);
 677                 node = &geneve->hlist4;
 678         }
 679         node->geneve = geneve;
 680
 681         tunnel_id_to_vni(geneve->info.key.tun_id, vni);
 682         hash = geneve_net_vni_hash(vni);
 683         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 684         return 0;
 685 }
 686
 687 static int geneve_open(struct net_device *dev)
 688 {
 689         struct geneve_dev *geneve = netdev_priv(dev);
 690         bool metadata = geneve->collect_md;
 691         bool ipv4, ipv6;
 692         int ret = 0;
 693
 694         ipv6 = geneve->info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 695         ipv4 = !ipv6 || metadata;
 696 #if IS_ENABLED(CONFIG_IPV6)
 697         if (ipv6) {
 698                 ret = geneve_sock_add(geneve, true);
 699                 if (ret < 0 && ret != -EAFNOSUPPORT)
 700                         ipv4 = false;
 701         }
 702 #endif
 703         if (ipv4)
 704                 ret = geneve_sock_add(geneve, false);
 705         if (ret < 0)
 706                 geneve_sock_release(geneve);
 707
 708         return ret;
 709 }
 710
 711 static int geneve_stop(struct net_device *dev)
 712 {
 713         struct geneve_dev *geneve = netdev_priv(dev);
 714
 715         hlist_del_init_rcu(&geneve->hlist4.hlist);
 716 #if IS_ENABLED(CONFIG_IPV6)
 717         hlist_del_init_rcu(&geneve->hlist6.hlist);
 718 #endif
 719         geneve_sock_release(geneve);
 720         return 0;
 721 }
 722
 723 static void geneve_build_header(struct genevehdr *geneveh,
 724                                 const struct ip_tunnel_info *info)
 725 {
 726         geneveh->ver = GENEVE_VER;
 727         geneveh->opt_len = info->options_len / 4;
 728         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 729         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 730         geneveh->rsvd1 = 0;
 731         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 732         geneveh->proto_type = htons(ETH_P_TEB);
 733         geneveh->rsvd2 = 0;
 734
 735         if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
 736                 ip_tunnel_info_opts_get(geneveh->options, info);
 737 }
 738
 739 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 740                             const struct ip_tunnel_info *info,
 741                             bool xnet, int ip_hdr_len)
 742 {
 743         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 744         struct genevehdr *gnvh;
 745         int min_headroom;
 746         int err;
 747
 748         skb_reset_mac_header(skb);
 749         skb_scrub_packet(skb, xnet);
 750
 751         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 752                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 753         err = skb_cow_head(skb, min_headroom);
 754         if (unlikely(err))
 755                 goto free_dst;
 756
 757         err = udp_tunnel_handle_offloads(skb, udp_sum);
 758         if (err)
 759                 goto free_dst;
 760
 761         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 762         geneve_build_header(gnvh, info);
 763         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 764         return 0;
 765
 766 free_dst:
 767         dst_release(dst);
 768         return err;
 769 }
 770
 771 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 772                                        struct net_device *dev,
 773                                        struct geneve_sock *gs4,
 774                                        struct flowi4 *fl4,
 775                                        const struct ip_tunnel_info *info,
 776                                        __be16 dport, __be16 sport)
 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         fl4->fl4_dport = dport;
 793         fl4->fl4_sport = sport;
 794
 795         tos = info->key.tos;
 796         if ((tos == 1) && !geneve->collect_md) {
 797                 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 798                 use_cache = false;
 799         }
 800         fl4->flowi4_tos = RT_TOS(tos);
 801
 802         dst_cache = (struct dst_cache *)&info->dst_cache;
 803         if (use_cache) {
 804                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 805                 if (rt)
 806                         return rt;
 807         }
 808         rt = ip_route_output_key(geneve->net, fl4);
 809         if (IS_ERR(rt)) {
 810                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 811                 return ERR_PTR(-ENETUNREACH);
 812         }
 813         if (rt->dst.dev == dev) { /* is this necessary? */
 814                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 815                 ip_rt_put(rt);
 816                 return ERR_PTR(-ELOOP);
 817         }
 818         if (use_cache)
 819                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 820         return rt;
 821 }
 822
 823 #if IS_ENABLED(CONFIG_IPV6)
 824 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 825                                            struct net_device *dev,
 826                                            struct geneve_sock *gs6,
 827                                            struct flowi6 *fl6,
 828                                            const struct ip_tunnel_info *info,
 829                                            __be16 dport, __be16 sport)
 830 {
 831         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 832         struct geneve_dev *geneve = netdev_priv(dev);
 833         struct dst_entry *dst = NULL;
 834         struct dst_cache *dst_cache;
 835         __u8 prio;
 836
 837         if (!gs6)
 838                 return ERR_PTR(-EIO);
 839
 840         memset(fl6, 0, sizeof(*fl6));
 841         fl6->flowi6_mark = skb->mark;
 842         fl6->flowi6_proto = IPPROTO_UDP;
 843         fl6->daddr = info->key.u.ipv6.dst;
 844         fl6->saddr = info->key.u.ipv6.src;
 845         fl6->fl6_dport = dport;
 846         fl6->fl6_sport = sport;
 847
 848         prio = info->key.tos;
 849         if ((prio == 1) && !geneve->collect_md) {
 850                 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 851                 use_cache = false;
 852         }
 853
 854         fl6->flowlabel = ip6_make_flowinfo(prio, info->key.label);
 855         dst_cache = (struct dst_cache *)&info->dst_cache;
 856         if (use_cache) {
 857                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 858                 if (dst)
 859                         return dst;
 860         }
 861         dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
 862                                               NULL);
 863         if (IS_ERR(dst)) {
 864                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 865                 return ERR_PTR(-ENETUNREACH);
 866         }
 867         if (dst->dev == dev) { /* is this necessary? */
 868                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 869                 dst_release(dst);
 870                 return ERR_PTR(-ELOOP);
 871         }
 872
 873         if (use_cache)
 874                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 875         return dst;
 876 }
 877 #endif
 878
 879 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 880                            struct geneve_dev *geneve,
 881                            const struct ip_tunnel_info *info)
 882 {
 883         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 884         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 885         const struct ip_tunnel_key *key = &info->key;
 886         struct rtable *rt;
 887         struct flowi4 fl4;
 888         __u8 tos, ttl;
 889         __be16 df = 0;
 890         __be16 sport;
 891         int err;
 892
 893         if (!pskb_inet_may_pull(skb))
 894                 return -EINVAL;
 895
 896         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 897         rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
 898                               geneve->info.key.tp_dst, sport);
 899         if (IS_ERR(rt))
 900                 return PTR_ERR(rt);
 901
 902         skb_tunnel_check_pmtu(skb, &rt->dst,
 903                               GENEVE_IPV4_HLEN + info->options_len);
 904
 905         if (geneve->collect_md) {
 906                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 907                 ttl = key->ttl;
 908
 909                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 910         } else {
 911                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 912                 if (geneve->ttl_inherit)
 913                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 914                 else
 915                         ttl = key->ttl;
 916                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 917
 918                 if (geneve->df == GENEVE_DF_SET) {
 919                         df = htons(IP_DF);
 920                 } else if (geneve->df == GENEVE_DF_INHERIT) {
 921                         struct ethhdr *eth = eth_hdr(skb);
 922
 923                         if (ntohs(eth->h_proto) == ETH_P_IPV6) {
 924                                 df = htons(IP_DF);
 925                         } else if (ntohs(eth->h_proto) == ETH_P_IP) {
 926                                 struct iphdr *iph = ip_hdr(skb);
 927
 928                                 if (iph->frag_off & htons(IP_DF))
 929                                         df = htons(IP_DF);
 930                         }
 931                 }
 932         }
 933
 934         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
 935         if (unlikely(err))
 936                 return err;
 937
 938         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 939                             tos, ttl, df, sport, geneve->info.key.tp_dst,
 940                             !net_eq(geneve->net, dev_net(geneve->dev)),
 941                             !(info->key.tun_flags & TUNNEL_CSUM));
 942         return 0;
 943 }
 944
 945 #if IS_ENABLED(CONFIG_IPV6)
 946 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 947                             struct geneve_dev *geneve,
 948                             const struct ip_tunnel_info *info)
 949 {
 950         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 951         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 952         const struct ip_tunnel_key *key = &info->key;
 953         struct dst_entry *dst = NULL;
 954         struct flowi6 fl6;
 955         __u8 prio, ttl;
 956         __be16 sport;
 957         int err;
 958
 959         if (!pskb_inet_may_pull(skb))
 960                 return -EINVAL;
 961
 962         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 963         dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
 964                                 geneve->info.key.tp_dst, sport);
 965         if (IS_ERR(dst))
 966                 return PTR_ERR(dst);
 967
 968         skb_tunnel_check_pmtu(skb, dst, GENEVE_IPV6_HLEN + info->options_len);
 969
 970         if (geneve->collect_md) {
 971                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 972                 ttl = key->ttl;
 973         } else {
 974                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
 975                                            ip_hdr(skb), skb);
 976                 if (geneve->ttl_inherit)
 977                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 978                 else
 979                         ttl = key->ttl;
 980                 ttl = ttl ? : ip6_dst_hoplimit(dst);
 981         }
 982         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
 983         if (unlikely(err))
 984                 return err;
 985
 986         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
 987                              &fl6.saddr, &fl6.daddr, prio, ttl,
 988                              info->key.label, sport, geneve->info.key.tp_dst,
 989                              !(info->key.tun_flags & TUNNEL_CSUM));
 990         return 0;
 991 }
 992 #endif
 993
 994 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
 995 {
 996         struct geneve_dev *geneve = netdev_priv(dev);
 997         struct ip_tunnel_info *info = NULL;
 998         int err;
 999
1000         if (geneve->collect_md) {
1001                 info = skb_tunnel_info(skb);
1002                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1003                         netdev_dbg(dev, "no tunnel metadata\n");
1004                         dev_kfree_skb(skb);
1005                         dev->stats.tx_dropped++;
1006                         return NETDEV_TX_OK;
1007                 }
1008         } else {
1009                 info = &geneve->info;
1010         }
1011
1012         rcu_read_lock();
1013 #if IS_ENABLED(CONFIG_IPV6)
1014         if (info->mode & IP_TUNNEL_INFO_IPV6)
1015                 err = geneve6_xmit_skb(skb, dev, geneve, info);
1016         else
1017 #endif
1018                 err = geneve_xmit_skb(skb, dev, geneve, info);
1019         rcu_read_unlock();
1020
1021         if (likely(!err))
1022                 return NETDEV_TX_OK;
1023
1024         dev_kfree_skb(skb);
1025
1026         if (err == -ELOOP)
1027                 dev->stats.collisions++;
1028         else if (err == -ENETUNREACH)
1029                 dev->stats.tx_carrier_errors++;
1030
1031         dev->stats.tx_errors++;
1032         return NETDEV_TX_OK;
1033 }
1034
1035 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1036 {
1037         if (new_mtu > dev->max_mtu)
1038                 new_mtu = dev->max_mtu;
1039         else if (new_mtu < dev->min_mtu)
1040                 new_mtu = dev->min_mtu;
1041
1042         dev->mtu = new_mtu;
1043         return 0;
1044 }
1045
1046 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1047 {
1048         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1049         struct geneve_dev *geneve = netdev_priv(dev);
1050         __be16 sport;
1051
1052         if (ip_tunnel_info_af(info) == AF_INET) {
1053                 struct rtable *rt;
1054                 struct flowi4 fl4;
1055
1056                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1057                 sport = udp_flow_src_port(geneve->net, skb,
1058                                           1, USHRT_MAX, true);
1059
1060                 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1061                                       geneve->info.key.tp_dst, sport);
1062                 if (IS_ERR(rt))
1063                         return PTR_ERR(rt);
1064
1065                 ip_rt_put(rt);
1066                 info->key.u.ipv4.src = fl4.saddr;
1067 #if IS_ENABLED(CONFIG_IPV6)
1068         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1069                 struct dst_entry *dst;
1070                 struct flowi6 fl6;
1071
1072                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1073                 sport = udp_flow_src_port(geneve->net, skb,
1074                                           1, USHRT_MAX, true);
1075
1076                 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1077                                         geneve->info.key.tp_dst, sport);
1078                 if (IS_ERR(dst))
1079                         return PTR_ERR(dst);
1080
1081                 dst_release(dst);
1082                 info->key.u.ipv6.src = fl6.saddr;
1083 #endif
1084         } else {
1085                 return -EINVAL;
1086         }
1087
1088         info->key.tp_src = sport;
1089         info->key.tp_dst = geneve->info.key.tp_dst;
1090         return 0;
1091 }
1092
1093 static const struct net_device_ops geneve_netdev_ops = {
1094         .ndo_init               = geneve_init,
1095         .ndo_uninit             = geneve_uninit,
1096         .ndo_open               = geneve_open,
1097         .ndo_stop               = geneve_stop,
1098         .ndo_start_xmit         = geneve_xmit,
1099         .ndo_get_stats64        = ip_tunnel_get_stats64,
1100         .ndo_change_mtu         = geneve_change_mtu,
1101         .ndo_validate_addr      = eth_validate_addr,
1102         .ndo_set_mac_address    = eth_mac_addr,
1103         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1104 };
1105
1106 static void geneve_get_drvinfo(struct net_device *dev,
1107                                struct ethtool_drvinfo *drvinfo)
1108 {
1109         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1110         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1111 }
1112
1113 static const struct ethtool_ops geneve_ethtool_ops = {
1114         .get_drvinfo    = geneve_get_drvinfo,
1115         .get_link       = ethtool_op_get_link,
1116 };
1117
1118 /* Info for udev, that this is a virtual tunnel endpoint */
1119 static struct device_type geneve_type = {
1120         .name = "geneve",
1121 };
1122
1123 /* Calls the ndo_udp_tunnel_add of the caller in order to
1124  * supply the listening GENEVE udp ports. Callers are expected
1125  * to implement the ndo_udp_tunnel_add.
1126  */
1127 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1128 {
1129         struct net *net = dev_net(dev);
1130         struct geneve_net *gn = net_generic(net, geneve_net_id);
1131         struct geneve_sock *gs;
1132
1133         rcu_read_lock();
1134         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1135                 if (push) {
1136                         udp_tunnel_push_rx_port(dev, gs->sock,
1137                                                 UDP_TUNNEL_TYPE_GENEVE);
1138                 } else {
1139                         udp_tunnel_drop_rx_port(dev, gs->sock,
1140                                                 UDP_TUNNEL_TYPE_GENEVE);
1141                 }
1142         }
1143         rcu_read_unlock();
1144 }
1145
1146 /* Initialize the device structure. */
1147 static void geneve_setup(struct net_device *dev)
1148 {
1149         ether_setup(dev);
1150
1151         dev->netdev_ops = &geneve_netdev_ops;
1152         dev->ethtool_ops = &geneve_ethtool_ops;
1153         dev->needs_free_netdev = true;
1154
1155         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1156
1157         dev->features    |= NETIF_F_LLTX;
1158         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1159         dev->features    |= NETIF_F_RXCSUM;
1160         dev->features    |= NETIF_F_GSO_SOFTWARE;
1161
1162         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1163         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1164
1165         /* MTU range: 68 - (something less than 65535) */
1166         dev->min_mtu = ETH_MIN_MTU;
1167         /* The max_mtu calculation does not take account of GENEVE
1168          * options, to avoid excluding potentially valid
1169          * configurations. This will be further reduced by IPvX hdr size.
1170          */
1171         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1172
1173         netif_keep_dst(dev);
1174         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1175         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1176         eth_hw_addr_random(dev);
1177 }
1178
1179 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1180         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1181         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1182         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1183         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1184         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1185         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1186         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1187         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1188         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1189         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1190         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1191         [IFLA_GENEVE_TTL_INHERIT]       = { .type = NLA_U8 },
1192         [IFLA_GENEVE_DF]                = { .type = NLA_U8 },
1193 };
1194
1195 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1196                            struct netlink_ext_ack *extack)
1197 {
1198         if (tb[IFLA_ADDRESS]) {
1199                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1200                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1201                                             "Provided link layer address is not Ethernet");
1202                         return -EINVAL;
1203                 }
1204
1205                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1206                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1207                                             "Provided Ethernet address is not unicast");
1208                         return -EADDRNOTAVAIL;
1209                 }
1210         }
1211
1212         if (!data) {
1213                 NL_SET_ERR_MSG(extack,
1214                                "Not enough attributes provided to perform the operation");
1215                 return -EINVAL;
1216         }
1217
1218         if (data[IFLA_GENEVE_ID]) {
1219                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1220
1221                 if (vni >= GENEVE_N_VID) {
1222                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1223                                             "Geneve ID must be lower than 16777216");
1224                         return -ERANGE;
1225                 }
1226         }
1227
1228         if (data[IFLA_GENEVE_DF]) {
1229                 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1230
1231                 if (df < 0 || df > GENEVE_DF_MAX) {
1232                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1233                                             "Invalid DF attribute");
1234                         return -EINVAL;
1235                 }
1236         }
1237
1238         return 0;
1239 }
1240
1241 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1242                                           const struct ip_tunnel_info *info,
1243                                           bool *tun_on_same_port,
1244                                           bool *tun_collect_md)
1245 {
1246         struct geneve_dev *geneve, *t = NULL;
1247
1248         *tun_on_same_port = false;
1249         *tun_collect_md = false;
1250         list_for_each_entry(geneve, &gn->geneve_list, next) {
1251                 if (info->key.tp_dst == geneve->info.key.tp_dst) {
1252                         *tun_collect_md = geneve->collect_md;
1253                         *tun_on_same_port = true;
1254                 }
1255                 if (info->key.tun_id == geneve->info.key.tun_id &&
1256                     info->key.tp_dst == geneve->info.key.tp_dst &&
1257                     !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1258                         t = geneve;
1259         }
1260         return t;
1261 }
1262
1263 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1264 {
1265         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1266                  info->key.ttl || info->key.label || info->key.tp_src ||
1267                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1268 }
1269
1270 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1271                                   struct ip_tunnel_info *b)
1272 {
1273         if (ip_tunnel_info_af(a) == AF_INET)
1274                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1275         else
1276                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1277 }
1278
1279 static int geneve_configure(struct net *net, struct net_device *dev,
1280                             struct netlink_ext_ack *extack,
1281                             const struct ip_tunnel_info *info,
1282                             bool metadata, bool ipv6_rx_csum,
1283                             bool ttl_inherit, enum ifla_geneve_df df)
1284 {
1285         struct geneve_net *gn = net_generic(net, geneve_net_id);
1286         struct geneve_dev *t, *geneve = netdev_priv(dev);
1287         bool tun_collect_md, tun_on_same_port;
1288         int err, encap_len;
1289
1290         if (metadata && !is_tnl_info_zero(info)) {
1291                 NL_SET_ERR_MSG(extack,
1292                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1293                 return -EINVAL;
1294         }
1295
1296         geneve->net = net;
1297         geneve->dev = dev;
1298
1299         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1300         if (t)
1301                 return -EBUSY;
1302
1303         /* make enough headroom for basic scenario */
1304         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1305         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1306                 encap_len += sizeof(struct iphdr);
1307                 dev->max_mtu -= sizeof(struct iphdr);
1308         } else {
1309                 encap_len += sizeof(struct ipv6hdr);
1310                 dev->max_mtu -= sizeof(struct ipv6hdr);
1311         }
1312         dev->needed_headroom = encap_len + ETH_HLEN;
1313
1314         if (metadata) {
1315                 if (tun_on_same_port) {
1316                         NL_SET_ERR_MSG(extack,
1317                                        "There can be only one externally controlled device on a destination port");
1318                         return -EPERM;
1319                 }
1320         } else {
1321                 if (tun_collect_md) {
1322                         NL_SET_ERR_MSG(extack,
1323                                        "There already exists an externally controlled device on this destination port");
1324                         return -EPERM;
1325                 }
1326         }
1327
1328         dst_cache_reset(&geneve->info.dst_cache);
1329         geneve->info = *info;
1330         geneve->collect_md = metadata;
1331         geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1332         geneve->ttl_inherit = ttl_inherit;
1333         geneve->df = df;
1334
1335         err = register_netdevice(dev);
1336         if (err)
1337                 return err;
1338
1339         list_add(&geneve->next, &gn->geneve_list);
1340         return 0;
1341 }
1342
1343 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1344 {
1345         memset(info, 0, sizeof(*info));
1346         info->key.tp_dst = htons(dst_port);
1347 }
1348
1349 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1350                           struct netlink_ext_ack *extack,
1351                           struct ip_tunnel_info *info, bool *metadata,
1352                           bool *use_udp6_rx_checksums, bool *ttl_inherit,
1353                           enum ifla_geneve_df *df, bool changelink)
1354 {
1355         int attrtype;
1356
1357         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1358                 NL_SET_ERR_MSG(extack,
1359                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1360                 return -EINVAL;
1361         }
1362
1363         if (data[IFLA_GENEVE_REMOTE]) {
1364                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1365                         attrtype = IFLA_GENEVE_REMOTE;
1366                         goto change_notsup;
1367                 }
1368
1369                 info->key.u.ipv4.dst =
1370                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1371
1372                 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1373                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1374                                             "Remote IPv4 address cannot be Multicast");
1375                         return -EINVAL;
1376                 }
1377         }
1378
1379         if (data[IFLA_GENEVE_REMOTE6]) {
1380 #if IS_ENABLED(CONFIG_IPV6)
1381                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1382                         attrtype = IFLA_GENEVE_REMOTE6;
1383                         goto change_notsup;
1384                 }
1385
1386                 info->mode = IP_TUNNEL_INFO_IPV6;
1387                 info->key.u.ipv6.dst =
1388                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1389
1390                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1391                     IPV6_ADDR_LINKLOCAL) {
1392                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1393                                             "Remote IPv6 address cannot be link-local");
1394                         return -EINVAL;
1395                 }
1396                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1397                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1398                                             "Remote IPv6 address cannot be Multicast");
1399                         return -EINVAL;
1400                 }
1401                 info->key.tun_flags |= TUNNEL_CSUM;
1402                 *use_udp6_rx_checksums = true;
1403 #else
1404                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1405                                     "IPv6 support not enabled in the kernel");
1406                 return -EPFNOSUPPORT;
1407 #endif
1408         }
1409
1410         if (data[IFLA_GENEVE_ID]) {
1411                 __u32 vni;
1412                 __u8 tvni[3];
1413                 __be64 tunid;
1414
1415                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1416                 tvni[0] = (vni & 0x00ff0000) >> 16;
1417                 tvni[1] = (vni & 0x0000ff00) >> 8;
1418                 tvni[2] =  vni & 0x000000ff;
1419
1420                 tunid = vni_to_tunnel_id(tvni);
1421                 if (changelink && (tunid != info->key.tun_id)) {
1422                         attrtype = IFLA_GENEVE_ID;
1423                         goto change_notsup;
1424                 }
1425                 info->key.tun_id = tunid;
1426         }
1427
1428         if (data[IFLA_GENEVE_TTL_INHERIT]) {
1429                 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1430                         *ttl_inherit = true;
1431                 else
1432                         *ttl_inherit = false;
1433         } else if (data[IFLA_GENEVE_TTL]) {
1434                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1435                 *ttl_inherit = false;
1436         }
1437
1438         if (data[IFLA_GENEVE_TOS])
1439                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1440
1441         if (data[IFLA_GENEVE_DF])
1442                 *df = nla_get_u8(data[IFLA_GENEVE_DF]);
1443
1444         if (data[IFLA_GENEVE_LABEL]) {
1445                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1446                                   IPV6_FLOWLABEL_MASK;
1447                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1448                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1449                                             "Label attribute only applies for IPv6 Geneve devices");
1450                         return -EINVAL;
1451                 }
1452         }
1453
1454         if (data[IFLA_GENEVE_PORT]) {
1455                 if (changelink) {
1456                         attrtype = IFLA_GENEVE_PORT;
1457                         goto change_notsup;
1458                 }
1459                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1460         }
1461
1462         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1463                 if (changelink) {
1464                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1465                         goto change_notsup;
1466                 }
1467                 *metadata = true;
1468         }
1469
1470         if (data[IFLA_GENEVE_UDP_CSUM]) {
1471                 if (changelink) {
1472                         attrtype = IFLA_GENEVE_UDP_CSUM;
1473                         goto change_notsup;
1474                 }
1475                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1476                         info->key.tun_flags |= TUNNEL_CSUM;
1477         }
1478
1479         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1480 #if IS_ENABLED(CONFIG_IPV6)
1481                 if (changelink) {
1482                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1483                         goto change_notsup;
1484                 }
1485                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1486                         info->key.tun_flags &= ~TUNNEL_CSUM;
1487 #else
1488                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1489                                     "IPv6 support not enabled in the kernel");
1490                 return -EPFNOSUPPORT;
1491 #endif
1492         }
1493
1494         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1495 #if IS_ENABLED(CONFIG_IPV6)
1496                 if (changelink) {
1497                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1498                         goto change_notsup;
1499                 }
1500                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1501                         *use_udp6_rx_checksums = false;
1502 #else
1503                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1504                                     "IPv6 support not enabled in the kernel");
1505                 return -EPFNOSUPPORT;
1506 #endif
1507         }
1508
1509         return 0;
1510 change_notsup:
1511         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1512                             "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1513         return -EOPNOTSUPP;
1514 }
1515
1516 static void geneve_link_config(struct net_device *dev,
1517                                struct ip_tunnel_info *info, struct nlattr *tb[])
1518 {
1519         struct geneve_dev *geneve = netdev_priv(dev);
1520         int ldev_mtu = 0;
1521
1522         if (tb[IFLA_MTU]) {
1523                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1524                 return;
1525         }
1526
1527         switch (ip_tunnel_info_af(info)) {
1528         case AF_INET: {
1529                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1530                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1531
1532                 if (!IS_ERR(rt) && rt->dst.dev) {
1533                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1534                         ip_rt_put(rt);
1535                 }
1536                 break;
1537         }
1538 #if IS_ENABLED(CONFIG_IPV6)
1539         case AF_INET6: {
1540                 struct rt6_info *rt;
1541
1542                 if (!__in6_dev_get(dev))
1543                         break;
1544
1545                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1546                                 NULL, 0);
1547
1548                 if (rt && rt->dst.dev)
1549                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1550                 ip6_rt_put(rt);
1551                 break;
1552         }
1553 #endif
1554         }
1555
1556         if (ldev_mtu <= 0)
1557                 return;
1558
1559         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1560 }
1561
1562 static int geneve_newlink(struct net *net, struct net_device *dev,
1563                           struct nlattr *tb[], struct nlattr *data[],
1564                           struct netlink_ext_ack *extack)
1565 {
1566         enum ifla_geneve_df df = GENEVE_DF_UNSET;
1567         bool use_udp6_rx_checksums = false;
1568         struct ip_tunnel_info info;
1569         bool ttl_inherit = false;
1570         bool metadata = false;
1571         int err;
1572
1573         init_tnl_info(&info, GENEVE_UDP_PORT);
1574         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1575                              &use_udp6_rx_checksums, &ttl_inherit, &df, false);
1576         if (err)
1577                 return err;
1578
1579         err = geneve_configure(net, dev, extack, &info, metadata,
1580                                use_udp6_rx_checksums, ttl_inherit, df);
1581         if (err)
1582                 return err;
1583
1584         geneve_link_config(dev, &info, tb);
1585
1586         return 0;
1587 }
1588
1589 /* Quiesces the geneve device data path for both TX and RX.
1590  *
1591  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1592  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1593  * to complete for the existing set of in-flight packets to be transmitted,
1594  * then we would have quiesced the transmit data path. All the future packets
1595  * will get dropped until we unquiesce the data path.
1596  *
1597  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1598  * if we set that to NULL under RCU and wait for synchronize_net() to
1599  * complete, then we would have quiesced the receive data path.
1600  */
1601 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1602                            struct geneve_sock **gs6)
1603 {
1604         *gs4 = rtnl_dereference(geneve->sock4);
1605         rcu_assign_pointer(geneve->sock4, NULL);
1606         if (*gs4)
1607                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1608 #if IS_ENABLED(CONFIG_IPV6)
1609         *gs6 = rtnl_dereference(geneve->sock6);
1610         rcu_assign_pointer(geneve->sock6, NULL);
1611         if (*gs6)
1612                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1613 #else
1614         *gs6 = NULL;
1615 #endif
1616         synchronize_net();
1617 }
1618
1619 /* Resumes the geneve device data path for both TX and RX. */
1620 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1621                              struct geneve_sock __maybe_unused *gs6)
1622 {
1623         rcu_assign_pointer(geneve->sock4, gs4);
1624         if (gs4)
1625                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1626 #if IS_ENABLED(CONFIG_IPV6)
1627         rcu_assign_pointer(geneve->sock6, gs6);
1628         if (gs6)
1629                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1630 #endif
1631         synchronize_net();
1632 }
1633
1634 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1635                              struct nlattr *data[],
1636                              struct netlink_ext_ack *extack)
1637 {
1638         struct geneve_dev *geneve = netdev_priv(dev);
1639         enum ifla_geneve_df df = geneve->df;
1640         struct geneve_sock *gs4, *gs6;
1641         struct ip_tunnel_info info;
1642         bool metadata;
1643         bool use_udp6_rx_checksums;
1644         bool ttl_inherit;
1645         int err;
1646
1647         /* If the geneve device is configured for metadata (or externally
1648          * controlled, for example, OVS), then nothing can be changed.
1649          */
1650         if (geneve->collect_md)
1651                 return -EOPNOTSUPP;
1652
1653         /* Start with the existing info. */
1654         memcpy(&info, &geneve->info, sizeof(info));
1655         metadata = geneve->collect_md;
1656         use_udp6_rx_checksums = geneve->use_udp6_rx_checksums;
1657         ttl_inherit = geneve->ttl_inherit;
1658         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1659                              &use_udp6_rx_checksums, &ttl_inherit, &df, true);
1660         if (err)
1661                 return err;
1662
1663         if (!geneve_dst_addr_equal(&geneve->info, &info)) {
1664                 dst_cache_reset(&info.dst_cache);
1665                 geneve_link_config(dev, &info, tb);
1666         }
1667
1668         geneve_quiesce(geneve, &gs4, &gs6);
1669         geneve->info = info;
1670         geneve->collect_md = metadata;
1671         geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
1672         geneve->ttl_inherit = ttl_inherit;
1673         geneve->df = df;
1674         geneve_unquiesce(geneve, gs4, gs6);
1675
1676         return 0;
1677 }
1678
1679 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1680 {
1681         struct geneve_dev *geneve = netdev_priv(dev);
1682
1683         list_del(&geneve->next);
1684         unregister_netdevice_queue(dev, head);
1685 }
1686
1687 static size_t geneve_get_size(const struct net_device *dev)
1688 {
1689         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1690                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1691                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1692                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1693                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_DF */
1694                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1695                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1696                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1697                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1698                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1699                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1700                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1701                 0;
1702 }
1703
1704 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1705 {
1706         struct geneve_dev *geneve = netdev_priv(dev);
1707         struct ip_tunnel_info *info = &geneve->info;
1708         bool ttl_inherit = geneve->ttl_inherit;
1709         bool metadata = geneve->collect_md;
1710         __u8 tmp_vni[3];
1711         __u32 vni;
1712
1713         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1714         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1715         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1716                 goto nla_put_failure;
1717
1718         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1719                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1720                                     info->key.u.ipv4.dst))
1721                         goto nla_put_failure;
1722                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1723                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1724                         goto nla_put_failure;
1725
1726 #if IS_ENABLED(CONFIG_IPV6)
1727         } else if (!metadata) {
1728                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1729                                      &info->key.u.ipv6.dst))
1730                         goto nla_put_failure;
1731                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1732                                !(info->key.tun_flags & TUNNEL_CSUM)))
1733                         goto nla_put_failure;
1734 #endif
1735         }
1736
1737         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1738             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1739             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1740                 goto nla_put_failure;
1741
1742         if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->df))
1743                 goto nla_put_failure;
1744
1745         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1746                 goto nla_put_failure;
1747
1748         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1749                 goto nla_put_failure;
1750
1751 #if IS_ENABLED(CONFIG_IPV6)
1752         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1753                        !geneve->use_udp6_rx_checksums))
1754                 goto nla_put_failure;
1755 #endif
1756
1757         if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1758                 goto nla_put_failure;
1759
1760         return 0;
1761
1762 nla_put_failure:
1763         return -EMSGSIZE;
1764 }
1765
1766 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1767         .kind           = "geneve",
1768         .maxtype        = IFLA_GENEVE_MAX,
1769         .policy         = geneve_policy,
1770         .priv_size      = sizeof(struct geneve_dev),
1771         .setup          = geneve_setup,
1772         .validate       = geneve_validate,
1773         .newlink        = geneve_newlink,
1774         .changelink     = geneve_changelink,
1775         .dellink        = geneve_dellink,
1776         .get_size       = geneve_get_size,
1777         .fill_info      = geneve_fill_info,
1778 };
1779
1780 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1781                                         u8 name_assign_type, u16 dst_port)
1782 {
1783         struct nlattr *tb[IFLA_MAX + 1];
1784         struct ip_tunnel_info info;
1785         struct net_device *dev;
1786         LIST_HEAD(list_kill);
1787         int err;
1788
1789         memset(tb, 0, sizeof(tb));
1790         dev = rtnl_create_link(net, name, name_assign_type,
1791                                &geneve_link_ops, tb, NULL);
1792         if (IS_ERR(dev))
1793                 return dev;
1794
1795         init_tnl_info(&info, dst_port);
1796         err = geneve_configure(net, dev, NULL, &info,
1797                                true, true, false, GENEVE_DF_UNSET);
1798         if (err) {
1799                 free_netdev(dev);
1800                 return ERR_PTR(err);
1801         }
1802
1803         /* openvswitch users expect packet sizes to be unrestricted,
1804          * so set the largest MTU we can.
1805          */
1806         err = geneve_change_mtu(dev, IP_MAX_MTU);
1807         if (err)
1808                 goto err;
1809
1810         err = rtnl_configure_link(dev, NULL);
1811         if (err < 0)
1812                 goto err;
1813
1814         return dev;
1815 err:
1816         geneve_dellink(dev, &list_kill);
1817         unregister_netdevice_many(&list_kill);
1818         return ERR_PTR(err);
1819 }
1820 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1821
1822 static int geneve_netdevice_event(struct notifier_block *unused,
1823                                   unsigned long event, void *ptr)
1824 {
1825         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1826
1827         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1828             event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1829                 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1830         } else if (event == NETDEV_UNREGISTER) {
1831                 geneve_offload_rx_ports(dev, false);
1832         } else if (event == NETDEV_REGISTER) {
1833                 geneve_offload_rx_ports(dev, true);
1834         }
1835
1836         return NOTIFY_DONE;
1837 }
1838
1839 static struct notifier_block geneve_notifier_block __read_mostly = {
1840         .notifier_call = geneve_netdevice_event,
1841 };
1842
1843 static __net_init int geneve_init_net(struct net *net)
1844 {
1845         struct geneve_net *gn = net_generic(net, geneve_net_id);
1846
1847         INIT_LIST_HEAD(&gn->geneve_list);
1848         INIT_LIST_HEAD(&gn->sock_list);
1849         return 0;
1850 }
1851
1852 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1853 {
1854         struct geneve_net *gn = net_generic(net, geneve_net_id);
1855         struct geneve_dev *geneve, *next;
1856         struct net_device *dev, *aux;
1857
1858         /* gather any geneve devices that were moved into this ns */
1859         for_each_netdev_safe(net, dev, aux)
1860                 if (dev->rtnl_link_ops == &geneve_link_ops)
1861                         unregister_netdevice_queue(dev, head);
1862
1863         /* now gather any other geneve devices that were created in this ns */
1864         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1865                 /* If geneve->dev is in the same netns, it was already added
1866                  * to the list by the previous loop.
1867                  */
1868                 if (!net_eq(dev_net(geneve->dev), net))
1869                         unregister_netdevice_queue(geneve->dev, head);
1870         }
1871 }
1872
1873 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1874 {
1875         struct net *net;
1876         LIST_HEAD(list);
1877
1878         rtnl_lock();
1879         list_for_each_entry(net, net_list, exit_list)
1880                 geneve_destroy_tunnels(net, &list);
1881
1882         /* unregister the devices gathered above */
1883         unregister_netdevice_many(&list);
1884         rtnl_unlock();
1885
1886         list_for_each_entry(net, net_list, exit_list) {
1887                 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1888
1889                 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1890         }
1891 }
1892
1893 static struct pernet_operations geneve_net_ops = {
1894         .init = geneve_init_net,
1895         .exit_batch = geneve_exit_batch_net,
1896         .id   = &geneve_net_id,
1897         .size = sizeof(struct geneve_net),
1898 };
1899
1900 static int __init geneve_init_module(void)
1901 {
1902         int rc;
1903
1904         rc = register_pernet_subsys(&geneve_net_ops);
1905         if (rc)
1906                 goto out1;
1907
1908         rc = register_netdevice_notifier(&geneve_notifier_block);
1909         if (rc)
1910                 goto out2;
1911
1912         rc = rtnl_link_register(&geneve_link_ops);
1913         if (rc)
1914                 goto out3;
1915
1916         return 0;
1917 out3:
1918         unregister_netdevice_notifier(&geneve_notifier_block);
1919 out2:
1920         unregister_pernet_subsys(&geneve_net_ops);
1921 out1:
1922         return rc;
1923 }
1924 late_initcall(geneve_init_module);
1925
1926 static void __exit geneve_cleanup_module(void)
1927 {
1928         rtnl_link_unregister(&geneve_link_ops);
1929         unregister_netdevice_notifier(&geneve_notifier_block);
1930         unregister_pernet_subsys(&geneve_net_ops);
1931 }
1932 module_exit(geneve_cleanup_module);
1933
1934 MODULE_LICENSE("GPL");
1935 MODULE_VERSION(GENEVE_NETDEV_VER);
1936 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1937 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1938 MODULE_ALIAS_RTNL_LINK("geneve");