drivers/net/geneve.c

   1 /*
   2  * GENEVE: Generic Network Virtualization Encapsulation
   3  *
   4  * Copyright (c) 2015 Red Hat, Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/etherdevice.h>
  16 #include <linux/hash.h>
  17 #include <net/dst_metadata.h>
  18 #include <net/gro_cells.h>
  19 #include <net/rtnetlink.h>
  20 #include <net/geneve.h>
  21 #include <net/protocol.h>
  22
  23 #define GENEVE_NETDEV_VER       "0.6"
  24
  25 #define GENEVE_UDP_PORT         6081
  26
  27 #define GENEVE_N_VID            (1u << 24)
  28 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  29
  30 #define VNI_HASH_BITS           10
  31 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  32
  33 static bool log_ecn_error = true;
  34 module_param(log_ecn_error, bool, 0644);
  35 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  36
  37 #define GENEVE_VER 0
  38 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  39 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
  40 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
  41
  42 /* per-network namespace private data for this module */
  43 struct geneve_net {
  44         struct list_head        geneve_list;
  45         struct list_head        sock_list;
  46 };
  47
  48 static unsigned int geneve_net_id;
  49
  50 struct geneve_dev_node {
  51         struct hlist_node hlist;
  52         struct geneve_dev *geneve;
  53 };
  54
  55 /* Pseudo network device */
  56 struct geneve_dev {
  57         struct geneve_dev_node hlist4;  /* vni hash table for IPv4 socket */
  58 #if IS_ENABLED(CONFIG_IPV6)
  59         struct geneve_dev_node hlist6;  /* vni hash table for IPv6 socket */
  60 #endif
  61         struct net         *net;        /* netns for packet i/o */
  62         struct net_device  *dev;        /* netdev for geneve tunnel */
  63         struct ip_tunnel_info info;
  64         struct geneve_sock __rcu *sock4;        /* IPv4 socket used for geneve tunnel */
  65 #if IS_ENABLED(CONFIG_IPV6)
  66         struct geneve_sock __rcu *sock6;        /* IPv6 socket used for geneve tunnel */
  67 #endif
  68         struct list_head   next;        /* geneve's per namespace list */
  69         struct gro_cells   gro_cells;
  70         bool               collect_md;
  71         bool               use_udp6_rx_checksums;
  72 };
  73
  74 struct geneve_sock {
  75         bool                    collect_md;
  76         struct list_head        list;
  77         struct socket           *sock;
  78         struct rcu_head         rcu;
  79         int                     refcnt;
  80         struct hlist_head       vni_list[VNI_HASH_SIZE];
  81 };
  82
  83 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  84 {
  85         __u32 vnid;
  86
  87         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  88         return hash_32(vnid, VNI_HASH_BITS);
  89 }
  90
  91 static __be64 vni_to_tunnel_id(const __u8 *vni)
  92 {
  93 #ifdef __BIG_ENDIAN
  94         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
  95 #else
  96         return (__force __be64)(((__force u64)vni[0] << 40) |
  97                                 ((__force u64)vni[1] << 48) |
  98                                 ((__force u64)vni[2] << 56));
  99 #endif
 100 }
 101
 102 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 103 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 104 {
 105 #ifdef __BIG_ENDIAN
 106         vni[0] = (__force __u8)(tun_id >> 16);
 107         vni[1] = (__force __u8)(tun_id >> 8);
 108         vni[2] = (__force __u8)tun_id;
 109 #else
 110         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 111         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 112         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 113 #endif
 114 }
 115
 116 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
 117 {
 118         return !memcmp(vni, &tun_id[5], 3);
 119 }
 120
 121 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 122 {
 123         return gs->sock->sk->sk_family;
 124 }
 125
 126 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 127                                         __be32 addr, u8 vni[])
 128 {
 129         struct hlist_head *vni_list_head;
 130         struct geneve_dev_node *node;
 131         __u32 hash;
 132
 133         /* Find the device for this VNI */
 134         hash = geneve_net_vni_hash(vni);
 135         vni_list_head = &gs->vni_list[hash];
 136         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 137                 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
 138                     addr == node->geneve->info.key.u.ipv4.dst)
 139                         return node->geneve;
 140         }
 141         return NULL;
 142 }
 143
 144 #if IS_ENABLED(CONFIG_IPV6)
 145 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 146                                          struct in6_addr addr6, u8 vni[])
 147 {
 148         struct hlist_head *vni_list_head;
 149         struct geneve_dev_node *node;
 150         __u32 hash;
 151
 152         /* Find the device for this VNI */
 153         hash = geneve_net_vni_hash(vni);
 154         vni_list_head = &gs->vni_list[hash];
 155         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 156                 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
 157                     ipv6_addr_equal(&addr6, &node->geneve->info.key.u.ipv6.dst))
 158                         return node->geneve;
 159         }
 160         return NULL;
 161 }
 162 #endif
 163
 164 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 165 {
 166         return (struct genevehdr *)(udp_hdr(skb) + 1);
 167 }
 168
 169 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 170                                             struct sk_buff *skb)
 171 {
 172         static u8 zero_vni[3];
 173         u8 *vni;
 174
 175         if (geneve_get_sk_family(gs) == AF_INET) {
 176                 struct iphdr *iph;
 177                 __be32 addr;
 178
 179                 iph = ip_hdr(skb); /* outer IP header... */
 180
 181                 if (gs->collect_md) {
 182                         vni = zero_vni;
 183                         addr = 0;
 184                 } else {
 185                         vni = geneve_hdr(skb)->vni;
 186                         addr = iph->saddr;
 187                 }
 188
 189                 return geneve_lookup(gs, addr, vni);
 190 #if IS_ENABLED(CONFIG_IPV6)
 191         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 192                 static struct in6_addr zero_addr6;
 193                 struct ipv6hdr *ip6h;
 194                 struct in6_addr addr6;
 195
 196                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 197
 198                 if (gs->collect_md) {
 199                         vni = zero_vni;
 200                         addr6 = zero_addr6;
 201                 } else {
 202                         vni = geneve_hdr(skb)->vni;
 203                         addr6 = ip6h->saddr;
 204                 }
 205
 206                 return geneve6_lookup(gs, addr6, vni);
 207 #endif
 208         }
 209         return NULL;
 210 }
 211
 212 /* geneve receive/decap routine */
 213 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 214                       struct sk_buff *skb)
 215 {
 216         struct genevehdr *gnvh = geneve_hdr(skb);
 217         struct metadata_dst *tun_dst = NULL;
 218         struct pcpu_sw_netstats *stats;
 219         unsigned int len;
 220         int err = 0;
 221         void *oiph;
 222
 223         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 224                 __be16 flags;
 225
 226                 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
 227                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 228
 229                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 230                                          vni_to_tunnel_id(gnvh->vni),
 231                                          gnvh->opt_len * 4);
 232                 if (!tun_dst) {
 233                         geneve->dev->stats.rx_dropped++;
 234                         goto drop;
 235                 }
 236                 /* Update tunnel dst according to Geneve options. */
 237                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 238                                         gnvh->options, gnvh->opt_len * 4,
 239                                         TUNNEL_GENEVE_OPT);
 240         } else {
 241                 /* Drop packets w/ critical options,
 242                  * since we don't support any...
 243                  */
 244                 if (gnvh->critical) {
 245                         geneve->dev->stats.rx_frame_errors++;
 246                         geneve->dev->stats.rx_errors++;
 247                         goto drop;
 248                 }
 249         }
 250
 251         skb_reset_mac_header(skb);
 252         skb->protocol = eth_type_trans(skb, geneve->dev);
 253         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 254
 255         if (tun_dst)
 256                 skb_dst_set(skb, &tun_dst->dst);
 257
 258         /* Ignore packet loops (and multicast echo) */
 259         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
 260                 geneve->dev->stats.rx_errors++;
 261                 goto drop;
 262         }
 263
 264         oiph = skb_network_header(skb);
 265         skb_reset_network_header(skb);
 266
 267         if (geneve_get_sk_family(gs) == AF_INET)
 268                 err = IP_ECN_decapsulate(oiph, skb);
 269 #if IS_ENABLED(CONFIG_IPV6)
 270         else
 271                 err = IP6_ECN_decapsulate(oiph, skb);
 272 #endif
 273
 274         if (unlikely(err)) {
 275                 if (log_ecn_error) {
 276                         if (geneve_get_sk_family(gs) == AF_INET)
 277                                 net_info_ratelimited("non-ECT from %pI4 "
 278                                                      "with TOS=%#x\n",
 279                                                      &((struct iphdr *)oiph)->saddr,
 280                                                      ((struct iphdr *)oiph)->tos);
 281 #if IS_ENABLED(CONFIG_IPV6)
 282                         else
 283                                 net_info_ratelimited("non-ECT from %pI6\n",
 284                                                      &((struct ipv6hdr *)oiph)->saddr);
 285 #endif
 286                 }
 287                 if (err > 1) {
 288                         ++geneve->dev->stats.rx_frame_errors;
 289                         ++geneve->dev->stats.rx_errors;
 290                         goto drop;
 291                 }
 292         }
 293
 294         len = skb->len;
 295         err = gro_cells_receive(&geneve->gro_cells, skb);
 296         if (likely(err == NET_RX_SUCCESS)) {
 297                 stats = this_cpu_ptr(geneve->dev->tstats);
 298                 u64_stats_update_begin(&stats->syncp);
 299                 stats->rx_packets++;
 300                 stats->rx_bytes += len;
 301                 u64_stats_update_end(&stats->syncp);
 302         }
 303         return;
 304 drop:
 305         /* Consume bad packet */
 306         kfree_skb(skb);
 307 }
 308
 309 /* Setup stats when device is created */
 310 static int geneve_init(struct net_device *dev)
 311 {
 312         struct geneve_dev *geneve = netdev_priv(dev);
 313         int err;
 314
 315         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 316         if (!dev->tstats)
 317                 return -ENOMEM;
 318
 319         err = gro_cells_init(&geneve->gro_cells, dev);
 320         if (err) {
 321                 free_percpu(dev->tstats);
 322                 return err;
 323         }
 324
 325         err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL);
 326         if (err) {
 327                 free_percpu(dev->tstats);
 328                 gro_cells_destroy(&geneve->gro_cells);
 329                 return err;
 330         }
 331         return 0;
 332 }
 333
 334 static void geneve_uninit(struct net_device *dev)
 335 {
 336         struct geneve_dev *geneve = netdev_priv(dev);
 337
 338         dst_cache_destroy(&geneve->info.dst_cache);
 339         gro_cells_destroy(&geneve->gro_cells);
 340         free_percpu(dev->tstats);
 341 }
 342
 343 /* Callback from net/ipv4/udp.c to receive packets */
 344 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 345 {
 346         struct genevehdr *geneveh;
 347         struct geneve_dev *geneve;
 348         struct geneve_sock *gs;
 349         int opts_len;
 350
 351         /* Need UDP and Geneve header to be present */
 352         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 353                 goto drop;
 354
 355         /* Return packets with reserved bits set */
 356         geneveh = geneve_hdr(skb);
 357         if (unlikely(geneveh->ver != GENEVE_VER))
 358                 goto drop;
 359
 360         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 361                 goto drop;
 362
 363         gs = rcu_dereference_sk_user_data(sk);
 364         if (!gs)
 365                 goto drop;
 366
 367         geneve = geneve_lookup_skb(gs, skb);
 368         if (!geneve)
 369                 goto drop;
 370
 371         opts_len = geneveh->opt_len * 4;
 372         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 373                                  htons(ETH_P_TEB),
 374                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 375                 geneve->dev->stats.rx_dropped++;
 376                 goto drop;
 377         }
 378
 379         geneve_rx(geneve, gs, skb);
 380         return 0;
 381
 382 drop:
 383         /* Consume bad packet */
 384         kfree_skb(skb);
 385         return 0;
 386 }
 387
 388 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 389                                          __be16 port, bool ipv6_rx_csum)
 390 {
 391         struct socket *sock;
 392         struct udp_port_cfg udp_conf;
 393         int err;
 394
 395         memset(&udp_conf, 0, sizeof(udp_conf));
 396
 397         if (ipv6) {
 398                 udp_conf.family = AF_INET6;
 399                 udp_conf.ipv6_v6only = 1;
 400                 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
 401         } else {
 402                 udp_conf.family = AF_INET;
 403                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 404         }
 405
 406         udp_conf.local_udp_port = port;
 407
 408         /* Open UDP socket */
 409         err = udp_sock_create(net, &udp_conf, &sock);
 410         if (err < 0)
 411                 return ERR_PTR(err);
 412
 413         return sock;
 414 }
 415
 416 static int geneve_hlen(struct genevehdr *gh)
 417 {
 418         return sizeof(*gh) + gh->opt_len * 4;
 419 }
 420
 421 static struct sk_buff *geneve_gro_receive(struct sock *sk,
 422                                           struct list_head *head,
 423                                           struct sk_buff *skb)
 424 {
 425         struct sk_buff *pp = NULL;
 426         struct sk_buff *p;
 427         struct genevehdr *gh, *gh2;
 428         unsigned int hlen, gh_len, off_gnv;
 429         const struct packet_offload *ptype;
 430         __be16 type;
 431         int flush = 1;
 432
 433         off_gnv = skb_gro_offset(skb);
 434         hlen = off_gnv + sizeof(*gh);
 435         gh = skb_gro_header_fast(skb, off_gnv);
 436         if (skb_gro_header_hard(skb, hlen)) {
 437                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 438                 if (unlikely(!gh))
 439                         goto out;
 440         }
 441
 442         if (gh->ver != GENEVE_VER || gh->oam)
 443                 goto out;
 444         gh_len = geneve_hlen(gh);
 445
 446         hlen = off_gnv + gh_len;
 447         if (skb_gro_header_hard(skb, hlen)) {
 448                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 449                 if (unlikely(!gh))
 450                         goto out;
 451         }
 452
 453         list_for_each_entry(p, head, list) {
 454                 if (!NAPI_GRO_CB(p)->same_flow)
 455                         continue;
 456
 457                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 458                 if (gh->opt_len != gh2->opt_len ||
 459                     memcmp(gh, gh2, gh_len)) {
 460                         NAPI_GRO_CB(p)->same_flow = 0;
 461                         continue;
 462                 }
 463         }
 464
 465         type = gh->proto_type;
 466
 467         rcu_read_lock();
 468         ptype = gro_find_receive_by_type(type);
 469         if (!ptype)
 470                 goto out_unlock;
 471
 472         skb_gro_pull(skb, gh_len);
 473         skb_gro_postpull_rcsum(skb, gh, gh_len);
 474         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 475         flush = 0;
 476
 477 out_unlock:
 478         rcu_read_unlock();
 479 out:
 480         skb_gro_flush_final(skb, pp, flush);
 481
 482         return pp;
 483 }
 484
 485 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 486                                int nhoff)
 487 {
 488         struct genevehdr *gh;
 489         struct packet_offload *ptype;
 490         __be16 type;
 491         int gh_len;
 492         int err = -ENOSYS;
 493
 494         gh = (struct genevehdr *)(skb->data + nhoff);
 495         gh_len = geneve_hlen(gh);
 496         type = gh->proto_type;
 497
 498         rcu_read_lock();
 499         ptype = gro_find_complete_by_type(type);
 500         if (ptype)
 501                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 502
 503         rcu_read_unlock();
 504
 505         skb_set_inner_mac_header(skb, nhoff + gh_len);
 506
 507         return err;
 508 }
 509
 510 /* Create new listen socket if needed */
 511 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 512                                                 bool ipv6, bool ipv6_rx_csum)
 513 {
 514         struct geneve_net *gn = net_generic(net, geneve_net_id);
 515         struct geneve_sock *gs;
 516         struct socket *sock;
 517         struct udp_tunnel_sock_cfg tunnel_cfg;
 518         int h;
 519
 520         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 521         if (!gs)
 522                 return ERR_PTR(-ENOMEM);
 523
 524         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 525         if (IS_ERR(sock)) {
 526                 kfree(gs);
 527                 return ERR_CAST(sock);
 528         }
 529
 530         gs->sock = sock;
 531         gs->refcnt = 1;
 532         for (h = 0; h < VNI_HASH_SIZE; ++h)
 533                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 534
 535         /* Initialize the geneve udp offloads structure */
 536         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 537
 538         /* Mark socket as an encapsulation socket */
 539         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 540         tunnel_cfg.sk_user_data = gs;
 541         tunnel_cfg.encap_type = 1;
 542         tunnel_cfg.gro_receive = geneve_gro_receive;
 543         tunnel_cfg.gro_complete = geneve_gro_complete;
 544         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 545         tunnel_cfg.encap_destroy = NULL;
 546         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 547         list_add(&gs->list, &gn->sock_list);
 548         return gs;
 549 }
 550
 551 static void __geneve_sock_release(struct geneve_sock *gs)
 552 {
 553         if (!gs || --gs->refcnt)
 554                 return;
 555
 556         list_del(&gs->list);
 557         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 558         udp_tunnel_sock_release(gs->sock);
 559         kfree_rcu(gs, rcu);
 560 }
 561
 562 static void geneve_sock_release(struct geneve_dev *geneve)
 563 {
 564         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 565 #if IS_ENABLED(CONFIG_IPV6)
 566         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 567
 568         rcu_assign_pointer(geneve->sock6, NULL);
 569 #endif
 570
 571         rcu_assign_pointer(geneve->sock4, NULL);
 572         synchronize_net();
 573
 574         __geneve_sock_release(gs4);
 575 #if IS_ENABLED(CONFIG_IPV6)
 576         __geneve_sock_release(gs6);
 577 #endif
 578 }
 579
 580 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 581                                             sa_family_t family,
 582                                             __be16 dst_port)
 583 {
 584         struct geneve_sock *gs;
 585
 586         list_for_each_entry(gs, &gn->sock_list, list) {
 587                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 588                     geneve_get_sk_family(gs) == family) {
 589                         return gs;
 590                 }
 591         }
 592         return NULL;
 593 }
 594
 595 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 596 {
 597         struct net *net = geneve->net;
 598         struct geneve_net *gn = net_generic(net, geneve_net_id);
 599         struct geneve_dev_node *node;
 600         struct geneve_sock *gs;
 601         __u8 vni[3];
 602         __u32 hash;
 603
 604         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst);
 605         if (gs) {
 606                 gs->refcnt++;
 607                 goto out;
 608         }
 609
 610         gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6,
 611                                   geneve->use_udp6_rx_checksums);
 612         if (IS_ERR(gs))
 613                 return PTR_ERR(gs);
 614
 615 out:
 616         gs->collect_md = geneve->collect_md;
 617 #if IS_ENABLED(CONFIG_IPV6)
 618         if (ipv6) {
 619                 rcu_assign_pointer(geneve->sock6, gs);
 620                 node = &geneve->hlist6;
 621         } else
 622 #endif
 623         {
 624                 rcu_assign_pointer(geneve->sock4, gs);
 625                 node = &geneve->hlist4;
 626         }
 627         node->geneve = geneve;
 628
 629         tunnel_id_to_vni(geneve->info.key.tun_id, vni);
 630         hash = geneve_net_vni_hash(vni);
 631         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 632         return 0;
 633 }
 634
 635 static int geneve_open(struct net_device *dev)
 636 {
 637         struct geneve_dev *geneve = netdev_priv(dev);
 638         bool metadata = geneve->collect_md;
 639         bool ipv4, ipv6;
 640         int ret = 0;
 641
 642         ipv6 = geneve->info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 643         ipv4 = !ipv6 || metadata;
 644 #if IS_ENABLED(CONFIG_IPV6)
 645         if (ipv6) {
 646                 ret = geneve_sock_add(geneve, true);
 647                 if (ret < 0 && ret != -EAFNOSUPPORT)
 648                         ipv4 = false;
 649         }
 650 #endif
 651         if (ipv4)
 652                 ret = geneve_sock_add(geneve, false);
 653         if (ret < 0)
 654                 geneve_sock_release(geneve);
 655
 656         return ret;
 657 }
 658
 659 static int geneve_stop(struct net_device *dev)
 660 {
 661         struct geneve_dev *geneve = netdev_priv(dev);
 662
 663         hlist_del_init_rcu(&geneve->hlist4.hlist);
 664 #if IS_ENABLED(CONFIG_IPV6)
 665         hlist_del_init_rcu(&geneve->hlist6.hlist);
 666 #endif
 667         geneve_sock_release(geneve);
 668         return 0;
 669 }
 670
 671 static void geneve_build_header(struct genevehdr *geneveh,
 672                                 const struct ip_tunnel_info *info)
 673 {
 674         geneveh->ver = GENEVE_VER;
 675         geneveh->opt_len = info->options_len / 4;
 676         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 677         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 678         geneveh->rsvd1 = 0;
 679         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 680         geneveh->proto_type = htons(ETH_P_TEB);
 681         geneveh->rsvd2 = 0;
 682
 683         if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
 684                 ip_tunnel_info_opts_get(geneveh->options, info);
 685 }
 686
 687 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 688                             const struct ip_tunnel_info *info,
 689                             bool xnet, int ip_hdr_len)
 690 {
 691         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 692         struct genevehdr *gnvh;
 693         int min_headroom;
 694         int err;
 695
 696         skb_reset_mac_header(skb);
 697         skb_scrub_packet(skb, xnet);
 698
 699         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 700                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 701         err = skb_cow_head(skb, min_headroom);
 702         if (unlikely(err))
 703                 goto free_dst;
 704
 705         err = udp_tunnel_handle_offloads(skb, udp_sum);
 706         if (err)
 707                 goto free_dst;
 708
 709         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 710         geneve_build_header(gnvh, info);
 711         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 712         return 0;
 713
 714 free_dst:
 715         dst_release(dst);
 716         return err;
 717 }
 718
 719 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 720                                        struct net_device *dev,
 721                                        struct geneve_sock *gs4,
 722                                        struct flowi4 *fl4,
 723                                        const struct ip_tunnel_info *info,
 724                                        __be16 dport, __be16 sport)
 725 {
 726         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 727         struct geneve_dev *geneve = netdev_priv(dev);
 728         struct dst_cache *dst_cache;
 729         struct rtable *rt = NULL;
 730         __u8 tos;
 731
 732         if (!gs4)
 733                 return ERR_PTR(-EIO);
 734
 735         memset(fl4, 0, sizeof(*fl4));
 736         fl4->flowi4_mark = skb->mark;
 737         fl4->flowi4_proto = IPPROTO_UDP;
 738         fl4->daddr = info->key.u.ipv4.dst;
 739         fl4->saddr = info->key.u.ipv4.src;
 740         fl4->fl4_dport = dport;
 741         fl4->fl4_sport = sport;
 742
 743         tos = info->key.tos;
 744         if ((tos == 1) && !geneve->collect_md) {
 745                 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 746                 use_cache = false;
 747         }
 748         fl4->flowi4_tos = RT_TOS(tos);
 749
 750         dst_cache = (struct dst_cache *)&info->dst_cache;
 751         if (use_cache) {
 752                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 753                 if (rt)
 754                         return rt;
 755         }
 756         rt = ip_route_output_key(geneve->net, fl4);
 757         if (IS_ERR(rt)) {
 758                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 759                 return ERR_PTR(-ENETUNREACH);
 760         }
 761         if (rt->dst.dev == dev) { /* is this necessary? */
 762                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 763                 ip_rt_put(rt);
 764                 return ERR_PTR(-ELOOP);
 765         }
 766         if (use_cache)
 767                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 768         return rt;
 769 }
 770
 771 #if IS_ENABLED(CONFIG_IPV6)
 772 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 773                                            struct net_device *dev,
 774                                            struct geneve_sock *gs6,
 775                                            struct flowi6 *fl6,
 776                                            const struct ip_tunnel_info *info,
 777                                            __be16 dport, __be16 sport)
 778 {
 779         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 780         struct geneve_dev *geneve = netdev_priv(dev);
 781         struct dst_entry *dst = NULL;
 782         struct dst_cache *dst_cache;
 783         __u8 prio;
 784
 785         if (!gs6)
 786                 return ERR_PTR(-EIO);
 787
 788         memset(fl6, 0, sizeof(*fl6));
 789         fl6->flowi6_mark = skb->mark;
 790         fl6->flowi6_proto = IPPROTO_UDP;
 791         fl6->daddr = info->key.u.ipv6.dst;
 792         fl6->saddr = info->key.u.ipv6.src;
 793         fl6->fl6_dport = dport;
 794         fl6->fl6_sport = sport;
 795
 796         prio = info->key.tos;
 797         if ((prio == 1) && !geneve->collect_md) {
 798                 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 799                 use_cache = false;
 800         }
 801
 802         fl6->flowlabel = ip6_make_flowinfo(prio, info->key.label);
 803         dst_cache = (struct dst_cache *)&info->dst_cache;
 804         if (use_cache) {
 805                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 806                 if (dst)
 807                         return dst;
 808         }
 809         dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
 810                                               NULL);
 811         if (IS_ERR(dst)) {
 812                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 813                 return ERR_PTR(-ENETUNREACH);
 814         }
 815         if (dst->dev == dev) { /* is this necessary? */
 816                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 817                 dst_release(dst);
 818                 return ERR_PTR(-ELOOP);
 819         }
 820
 821         if (use_cache)
 822                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 823         return dst;
 824 }
 825 #endif
 826
 827 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 828                            struct geneve_dev *geneve,
 829                            const struct ip_tunnel_info *info)
 830 {
 831         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 832         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 833         const struct ip_tunnel_key *key = &info->key;
 834         struct rtable *rt;
 835         struct flowi4 fl4;
 836         __u8 tos, ttl;
 837         __be16 sport;
 838         __be16 df;
 839         int err;
 840
 841         if (!pskb_inet_may_pull(skb))
 842                 return -EINVAL;
 843
 844         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 845         rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
 846                               geneve->info.key.tp_dst, sport);
 847         if (IS_ERR(rt))
 848                 return PTR_ERR(rt);
 849
 850         skb_tunnel_check_pmtu(skb, &rt->dst,
 851                               GENEVE_IPV4_HLEN + info->options_len);
 852
 853         if (geneve->collect_md) {
 854                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 855                 ttl = key->ttl;
 856         } else {
 857                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 858                 ttl = key->ttl ? : ip4_dst_hoplimit(&rt->dst);
 859         }
 860         df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 861
 862         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
 863         if (unlikely(err))
 864                 return err;
 865
 866         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 867                             tos, ttl, df, sport, geneve->info.key.tp_dst,
 868                             !net_eq(geneve->net, dev_net(geneve->dev)),
 869                             !(info->key.tun_flags & TUNNEL_CSUM));
 870         return 0;
 871 }
 872
 873 #if IS_ENABLED(CONFIG_IPV6)
 874 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 875                             struct geneve_dev *geneve,
 876                             const struct ip_tunnel_info *info)
 877 {
 878         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 879         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 880         const struct ip_tunnel_key *key = &info->key;
 881         struct dst_entry *dst = NULL;
 882         struct flowi6 fl6;
 883         __u8 prio, ttl;
 884         __be16 sport;
 885         int err;
 886
 887         if (!pskb_inet_may_pull(skb))
 888                 return -EINVAL;
 889
 890         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 891         dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
 892                                 geneve->info.key.tp_dst, sport);
 893         if (IS_ERR(dst))
 894                 return PTR_ERR(dst);
 895
 896         skb_tunnel_check_pmtu(skb, dst, GENEVE_IPV6_HLEN + info->options_len);
 897
 898         if (geneve->collect_md) {
 899                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 900                 ttl = key->ttl;
 901         } else {
 902                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
 903                                            ip_hdr(skb), skb);
 904                 ttl = key->ttl ? : ip6_dst_hoplimit(dst);
 905         }
 906         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
 907         if (unlikely(err))
 908                 return err;
 909
 910         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
 911                              &fl6.saddr, &fl6.daddr, prio, ttl,
 912                              info->key.label, sport, geneve->info.key.tp_dst,
 913                              !(info->key.tun_flags & TUNNEL_CSUM));
 914         return 0;
 915 }
 916 #endif
 917
 918 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
 919 {
 920         struct geneve_dev *geneve = netdev_priv(dev);
 921         struct ip_tunnel_info *info = NULL;
 922         int err;
 923
 924         if (geneve->collect_md) {
 925                 info = skb_tunnel_info(skb);
 926                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 927                         netdev_dbg(dev, "no tunnel metadata\n");
 928                         dev_kfree_skb(skb);
 929                         dev->stats.tx_dropped++;
 930                         return NETDEV_TX_OK;
 931                 }
 932         } else {
 933                 info = &geneve->info;
 934         }
 935
 936         rcu_read_lock();
 937 #if IS_ENABLED(CONFIG_IPV6)
 938         if (info->mode & IP_TUNNEL_INFO_IPV6)
 939                 err = geneve6_xmit_skb(skb, dev, geneve, info);
 940         else
 941 #endif
 942                 err = geneve_xmit_skb(skb, dev, geneve, info);
 943         rcu_read_unlock();
 944
 945         if (likely(!err))
 946                 return NETDEV_TX_OK;
 947
 948         dev_kfree_skb(skb);
 949
 950         if (err == -ELOOP)
 951                 dev->stats.collisions++;
 952         else if (err == -ENETUNREACH)
 953                 dev->stats.tx_carrier_errors++;
 954
 955         dev->stats.tx_errors++;
 956         return NETDEV_TX_OK;
 957 }
 958
 959 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
 960 {
 961         if (new_mtu > dev->max_mtu)
 962                 new_mtu = dev->max_mtu;
 963         else if (new_mtu < dev->min_mtu)
 964                 new_mtu = dev->min_mtu;
 965
 966         dev->mtu = new_mtu;
 967         return 0;
 968 }
 969
 970 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
 971 {
 972         struct ip_tunnel_info *info = skb_tunnel_info(skb);
 973         struct geneve_dev *geneve = netdev_priv(dev);
 974         __be16 sport;
 975
 976         if (ip_tunnel_info_af(info) == AF_INET) {
 977                 struct rtable *rt;
 978                 struct flowi4 fl4;
 979
 980                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 981                 sport = udp_flow_src_port(geneve->net, skb,
 982                                           1, USHRT_MAX, true);
 983
 984                 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
 985                                       geneve->info.key.tp_dst, sport);
 986                 if (IS_ERR(rt))
 987                         return PTR_ERR(rt);
 988
 989                 ip_rt_put(rt);
 990                 info->key.u.ipv4.src = fl4.saddr;
 991 #if IS_ENABLED(CONFIG_IPV6)
 992         } else if (ip_tunnel_info_af(info) == AF_INET6) {
 993                 struct dst_entry *dst;
 994                 struct flowi6 fl6;
 995
 996                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 997                 sport = udp_flow_src_port(geneve->net, skb,
 998                                           1, USHRT_MAX, true);
 999
1000                 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1001                                         geneve->info.key.tp_dst, sport);
1002                 if (IS_ERR(dst))
1003                         return PTR_ERR(dst);
1004
1005                 dst_release(dst);
1006                 info->key.u.ipv6.src = fl6.saddr;
1007 #endif
1008         } else {
1009                 return -EINVAL;
1010         }
1011
1012         info->key.tp_src = sport;
1013         info->key.tp_dst = geneve->info.key.tp_dst;
1014         return 0;
1015 }
1016
1017 static const struct net_device_ops geneve_netdev_ops = {
1018         .ndo_init               = geneve_init,
1019         .ndo_uninit             = geneve_uninit,
1020         .ndo_open               = geneve_open,
1021         .ndo_stop               = geneve_stop,
1022         .ndo_start_xmit         = geneve_xmit,
1023         .ndo_get_stats64        = ip_tunnel_get_stats64,
1024         .ndo_change_mtu         = geneve_change_mtu,
1025         .ndo_validate_addr      = eth_validate_addr,
1026         .ndo_set_mac_address    = eth_mac_addr,
1027         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1028 };
1029
1030 static void geneve_get_drvinfo(struct net_device *dev,
1031                                struct ethtool_drvinfo *drvinfo)
1032 {
1033         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1034         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1035 }
1036
1037 static const struct ethtool_ops geneve_ethtool_ops = {
1038         .get_drvinfo    = geneve_get_drvinfo,
1039         .get_link       = ethtool_op_get_link,
1040 };
1041
1042 /* Info for udev, that this is a virtual tunnel endpoint */
1043 static struct device_type geneve_type = {
1044         .name = "geneve",
1045 };
1046
1047 /* Calls the ndo_udp_tunnel_add of the caller in order to
1048  * supply the listening GENEVE udp ports. Callers are expected
1049  * to implement the ndo_udp_tunnel_add.
1050  */
1051 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1052 {
1053         struct net *net = dev_net(dev);
1054         struct geneve_net *gn = net_generic(net, geneve_net_id);
1055         struct geneve_sock *gs;
1056
1057         rcu_read_lock();
1058         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1059                 if (push) {
1060                         udp_tunnel_push_rx_port(dev, gs->sock,
1061                                                 UDP_TUNNEL_TYPE_GENEVE);
1062                 } else {
1063                         udp_tunnel_drop_rx_port(dev, gs->sock,
1064                                                 UDP_TUNNEL_TYPE_GENEVE);
1065                 }
1066         }
1067         rcu_read_unlock();
1068 }
1069
1070 /* Initialize the device structure. */
1071 static void geneve_setup(struct net_device *dev)
1072 {
1073         ether_setup(dev);
1074
1075         dev->netdev_ops = &geneve_netdev_ops;
1076         dev->ethtool_ops = &geneve_ethtool_ops;
1077         dev->needs_free_netdev = true;
1078
1079         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1080
1081         dev->features    |= NETIF_F_LLTX;
1082         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1083         dev->features    |= NETIF_F_RXCSUM;
1084         dev->features    |= NETIF_F_GSO_SOFTWARE;
1085
1086         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1087         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1088
1089         /* MTU range: 68 - (something less than 65535) */
1090         dev->min_mtu = ETH_MIN_MTU;
1091         /* The max_mtu calculation does not take account of GENEVE
1092          * options, to avoid excluding potentially valid
1093          * configurations. This will be further reduced by IPvX hdr size.
1094          */
1095         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1096
1097         netif_keep_dst(dev);
1098         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1099         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1100         eth_hw_addr_random(dev);
1101 }
1102
1103 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1104         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1105         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1106         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1107         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1108         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1109         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1110         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1111         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1112         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1113         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1114         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1115 };
1116
1117 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1118                            struct netlink_ext_ack *extack)
1119 {
1120         if (tb[IFLA_ADDRESS]) {
1121                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1122                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1123                                             "Provided link layer address is not Ethernet");
1124                         return -EINVAL;
1125                 }
1126
1127                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1128                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1129                                             "Provided Ethernet address is not unicast");
1130                         return -EADDRNOTAVAIL;
1131                 }
1132         }
1133
1134         if (!data) {
1135                 NL_SET_ERR_MSG(extack,
1136                                "Not enough attributes provided to perform the operation");
1137                 return -EINVAL;
1138         }
1139
1140         if (data[IFLA_GENEVE_ID]) {
1141                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1142
1143                 if (vni >= GENEVE_N_VID) {
1144                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1145                                             "Geneve ID must be lower than 16777216");
1146                         return -ERANGE;
1147                 }
1148         }
1149
1150         return 0;
1151 }
1152
1153 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1154                                           const struct ip_tunnel_info *info,
1155                                           bool *tun_on_same_port,
1156                                           bool *tun_collect_md)
1157 {
1158         struct geneve_dev *geneve, *t = NULL;
1159
1160         *tun_on_same_port = false;
1161         *tun_collect_md = false;
1162         list_for_each_entry(geneve, &gn->geneve_list, next) {
1163                 if (info->key.tp_dst == geneve->info.key.tp_dst) {
1164                         *tun_collect_md = geneve->collect_md;
1165                         *tun_on_same_port = true;
1166                 }
1167                 if (info->key.tun_id == geneve->info.key.tun_id &&
1168                     info->key.tp_dst == geneve->info.key.tp_dst &&
1169                     !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1170                         t = geneve;
1171         }
1172         return t;
1173 }
1174
1175 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1176 {
1177         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1178                  info->key.ttl || info->key.label || info->key.tp_src ||
1179                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1180 }
1181
1182 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1183                                   struct ip_tunnel_info *b)
1184 {
1185         if (ip_tunnel_info_af(a) == AF_INET)
1186                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1187         else
1188                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1189 }
1190
1191 static int geneve_configure(struct net *net, struct net_device *dev,
1192                             struct netlink_ext_ack *extack,
1193                             const struct ip_tunnel_info *info,
1194                             bool metadata, bool ipv6_rx_csum)
1195 {
1196         struct geneve_net *gn = net_generic(net, geneve_net_id);
1197         struct geneve_dev *t, *geneve = netdev_priv(dev);
1198         bool tun_collect_md, tun_on_same_port;
1199         int err, encap_len;
1200
1201         if (metadata && !is_tnl_info_zero(info)) {
1202                 NL_SET_ERR_MSG(extack,
1203                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1204                 return -EINVAL;
1205         }
1206
1207         geneve->net = net;
1208         geneve->dev = dev;
1209
1210         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1211         if (t)
1212                 return -EBUSY;
1213
1214         /* make enough headroom for basic scenario */
1215         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1216         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1217                 encap_len += sizeof(struct iphdr);
1218                 dev->max_mtu -= sizeof(struct iphdr);
1219         } else {
1220                 encap_len += sizeof(struct ipv6hdr);
1221                 dev->max_mtu -= sizeof(struct ipv6hdr);
1222         }
1223         dev->needed_headroom = encap_len + ETH_HLEN;
1224
1225         if (metadata) {
1226                 if (tun_on_same_port) {
1227                         NL_SET_ERR_MSG(extack,
1228                                        "There can be only one externally controlled device on a destination port");
1229                         return -EPERM;
1230                 }
1231         } else {
1232                 if (tun_collect_md) {
1233                         NL_SET_ERR_MSG(extack,
1234                                        "There already exists an externally controlled device on this destination port");
1235                         return -EPERM;
1236                 }
1237         }
1238
1239         dst_cache_reset(&geneve->info.dst_cache);
1240         geneve->info = *info;
1241         geneve->collect_md = metadata;
1242         geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1243
1244         err = register_netdevice(dev);
1245         if (err)
1246                 return err;
1247
1248         list_add(&geneve->next, &gn->geneve_list);
1249         return 0;
1250 }
1251
1252 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1253 {
1254         memset(info, 0, sizeof(*info));
1255         info->key.tp_dst = htons(dst_port);
1256 }
1257
1258 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1259                           struct netlink_ext_ack *extack,
1260                           struct ip_tunnel_info *info, bool *metadata,
1261                           bool *use_udp6_rx_checksums, bool changelink)
1262 {
1263         int attrtype;
1264
1265         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1266                 NL_SET_ERR_MSG(extack,
1267                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1268                 return -EINVAL;
1269         }
1270
1271         if (data[IFLA_GENEVE_REMOTE]) {
1272                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1273                         attrtype = IFLA_GENEVE_REMOTE;
1274                         goto change_notsup;
1275                 }
1276
1277                 info->key.u.ipv4.dst =
1278                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1279
1280                 if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) {
1281                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1282                                             "Remote IPv4 address cannot be Multicast");
1283                         return -EINVAL;
1284                 }
1285         }
1286
1287         if (data[IFLA_GENEVE_REMOTE6]) {
1288 #if IS_ENABLED(CONFIG_IPV6)
1289                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1290                         attrtype = IFLA_GENEVE_REMOTE6;
1291                         goto change_notsup;
1292                 }
1293
1294                 info->mode = IP_TUNNEL_INFO_IPV6;
1295                 info->key.u.ipv6.dst =
1296                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1297
1298                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1299                     IPV6_ADDR_LINKLOCAL) {
1300                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1301                                             "Remote IPv6 address cannot be link-local");
1302                         return -EINVAL;
1303                 }
1304                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1305                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1306                                             "Remote IPv6 address cannot be Multicast");
1307                         return -EINVAL;
1308                 }
1309                 info->key.tun_flags |= TUNNEL_CSUM;
1310                 *use_udp6_rx_checksums = true;
1311 #else
1312                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1313                                     "IPv6 support not enabled in the kernel");
1314                 return -EPFNOSUPPORT;
1315 #endif
1316         }
1317
1318         if (data[IFLA_GENEVE_ID]) {
1319                 __u32 vni;
1320                 __u8 tvni[3];
1321                 __be64 tunid;
1322
1323                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1324                 tvni[0] = (vni & 0x00ff0000) >> 16;
1325                 tvni[1] = (vni & 0x0000ff00) >> 8;
1326                 tvni[2] =  vni & 0x000000ff;
1327
1328                 tunid = vni_to_tunnel_id(tvni);
1329                 if (changelink && (tunid != info->key.tun_id)) {
1330                         attrtype = IFLA_GENEVE_ID;
1331                         goto change_notsup;
1332                 }
1333                 info->key.tun_id = tunid;
1334         }
1335
1336         if (data[IFLA_GENEVE_TTL])
1337                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1338
1339         if (data[IFLA_GENEVE_TOS])
1340                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1341
1342         if (data[IFLA_GENEVE_LABEL]) {
1343                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1344                                   IPV6_FLOWLABEL_MASK;
1345                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1346                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1347                                             "Label attribute only applies for IPv6 Geneve devices");
1348                         return -EINVAL;
1349                 }
1350         }
1351
1352         if (data[IFLA_GENEVE_PORT]) {
1353                 if (changelink) {
1354                         attrtype = IFLA_GENEVE_PORT;
1355                         goto change_notsup;
1356                 }
1357                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1358         }
1359
1360         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1361                 if (changelink) {
1362                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1363                         goto change_notsup;
1364                 }
1365                 *metadata = true;
1366         }
1367
1368         if (data[IFLA_GENEVE_UDP_CSUM]) {
1369                 if (changelink) {
1370                         attrtype = IFLA_GENEVE_UDP_CSUM;
1371                         goto change_notsup;
1372                 }
1373                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1374                         info->key.tun_flags |= TUNNEL_CSUM;
1375         }
1376
1377         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1378 #if IS_ENABLED(CONFIG_IPV6)
1379                 if (changelink) {
1380                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1381                         goto change_notsup;
1382                 }
1383                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1384                         info->key.tun_flags &= ~TUNNEL_CSUM;
1385 #else
1386                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1387                                     "IPv6 support not enabled in the kernel");
1388                 return -EPFNOSUPPORT;
1389 #endif
1390         }
1391
1392         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1393 #if IS_ENABLED(CONFIG_IPV6)
1394                 if (changelink) {
1395                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1396                         goto change_notsup;
1397                 }
1398                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1399                         *use_udp6_rx_checksums = false;
1400 #else
1401                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1402                                     "IPv6 support not enabled in the kernel");
1403                 return -EPFNOSUPPORT;
1404 #endif
1405         }
1406
1407         return 0;
1408 change_notsup:
1409         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1410                             "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1411         return -EOPNOTSUPP;
1412 }
1413
1414 static void geneve_link_config(struct net_device *dev,
1415                                struct ip_tunnel_info *info, struct nlattr *tb[])
1416 {
1417         struct geneve_dev *geneve = netdev_priv(dev);
1418         int ldev_mtu = 0;
1419
1420         if (tb[IFLA_MTU]) {
1421                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1422                 return;
1423         }
1424
1425         switch (ip_tunnel_info_af(info)) {
1426         case AF_INET: {
1427                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1428                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1429
1430                 if (!IS_ERR(rt) && rt->dst.dev) {
1431                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1432                         ip_rt_put(rt);
1433                 }
1434                 break;
1435         }
1436 #if IS_ENABLED(CONFIG_IPV6)
1437         case AF_INET6: {
1438                 struct rt6_info *rt;
1439
1440                 if (!__in6_dev_get(dev))
1441                         break;
1442
1443                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1444                                 NULL, 0);
1445
1446                 if (rt && rt->dst.dev)
1447                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1448                 ip6_rt_put(rt);
1449                 break;
1450         }
1451 #endif
1452         }
1453
1454         if (ldev_mtu <= 0)
1455                 return;
1456
1457         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1458 }
1459
1460 static int geneve_newlink(struct net *net, struct net_device *dev,
1461                           struct nlattr *tb[], struct nlattr *data[],
1462                           struct netlink_ext_ack *extack)
1463 {
1464         bool use_udp6_rx_checksums = false;
1465         struct ip_tunnel_info info;
1466         bool metadata = false;
1467         int err;
1468
1469         init_tnl_info(&info, GENEVE_UDP_PORT);
1470         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1471                              &use_udp6_rx_checksums, false);
1472         if (err)
1473                 return err;
1474
1475         err = geneve_configure(net, dev, extack, &info, metadata,
1476                                use_udp6_rx_checksums);
1477         if (err)
1478                 return err;
1479
1480         geneve_link_config(dev, &info, tb);
1481
1482         return 0;
1483 }
1484
1485 /* Quiesces the geneve device data path for both TX and RX.
1486  *
1487  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1488  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1489  * to complete for the existing set of in-flight packets to be transmitted,
1490  * then we would have quiesced the transmit data path. All the future packets
1491  * will get dropped until we unquiesce the data path.
1492  *
1493  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1494  * if we set that to NULL under RCU and wait for synchronize_net() to
1495  * complete, then we would have quiesced the receive data path.
1496  */
1497 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1498                            struct geneve_sock **gs6)
1499 {
1500         *gs4 = rtnl_dereference(geneve->sock4);
1501         rcu_assign_pointer(geneve->sock4, NULL);
1502         if (*gs4)
1503                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1504 #if IS_ENABLED(CONFIG_IPV6)
1505         *gs6 = rtnl_dereference(geneve->sock6);
1506         rcu_assign_pointer(geneve->sock6, NULL);
1507         if (*gs6)
1508                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1509 #else
1510         *gs6 = NULL;
1511 #endif
1512         synchronize_net();
1513 }
1514
1515 /* Resumes the geneve device data path for both TX and RX. */
1516 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1517                              struct geneve_sock __maybe_unused *gs6)
1518 {
1519         rcu_assign_pointer(geneve->sock4, gs4);
1520         if (gs4)
1521                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1522 #if IS_ENABLED(CONFIG_IPV6)
1523         rcu_assign_pointer(geneve->sock6, gs6);
1524         if (gs6)
1525                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1526 #endif
1527         synchronize_net();
1528 }
1529
1530 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1531                              struct nlattr *data[],
1532                              struct netlink_ext_ack *extack)
1533 {
1534         struct geneve_dev *geneve = netdev_priv(dev);
1535         struct geneve_sock *gs4, *gs6;
1536         struct ip_tunnel_info info;
1537         bool metadata;
1538         bool use_udp6_rx_checksums;
1539         int err;
1540
1541         /* If the geneve device is configured for metadata (or externally
1542          * controlled, for example, OVS), then nothing can be changed.
1543          */
1544         if (geneve->collect_md)
1545                 return -EOPNOTSUPP;
1546
1547         /* Start with the existing info. */
1548         memcpy(&info, &geneve->info, sizeof(info));
1549         metadata = geneve->collect_md;
1550         use_udp6_rx_checksums = geneve->use_udp6_rx_checksums;
1551         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1552                              &use_udp6_rx_checksums, true);
1553         if (err)
1554                 return err;
1555
1556         if (!geneve_dst_addr_equal(&geneve->info, &info)) {
1557                 dst_cache_reset(&info.dst_cache);
1558                 geneve_link_config(dev, &info, tb);
1559         }
1560
1561         geneve_quiesce(geneve, &gs4, &gs6);
1562         geneve->info = info;
1563         geneve->collect_md = metadata;
1564         geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
1565         geneve_unquiesce(geneve, gs4, gs6);
1566
1567         return 0;
1568 }
1569
1570 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1571 {
1572         struct geneve_dev *geneve = netdev_priv(dev);
1573
1574         list_del(&geneve->next);
1575         unregister_netdevice_queue(dev, head);
1576 }
1577
1578 static size_t geneve_get_size(const struct net_device *dev)
1579 {
1580         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1581                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1582                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1583                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1584                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1585                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1586                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1587                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1588                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1589                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1590                 0;
1591 }
1592
1593 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1594 {
1595         struct geneve_dev *geneve = netdev_priv(dev);
1596         struct ip_tunnel_info *info = &geneve->info;
1597         bool metadata = geneve->collect_md;
1598         __u8 tmp_vni[3];
1599         __u32 vni;
1600
1601         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1602         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1603         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1604                 goto nla_put_failure;
1605
1606         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1607                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1608                                     info->key.u.ipv4.dst))
1609                         goto nla_put_failure;
1610                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1611                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1612                         goto nla_put_failure;
1613
1614 #if IS_ENABLED(CONFIG_IPV6)
1615         } else if (!metadata) {
1616                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1617                                      &info->key.u.ipv6.dst))
1618                         goto nla_put_failure;
1619                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1620                                !(info->key.tun_flags & TUNNEL_CSUM)))
1621                         goto nla_put_failure;
1622 #endif
1623         }
1624
1625         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1626             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1627             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1628                 goto nla_put_failure;
1629
1630         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1631                 goto nla_put_failure;
1632
1633         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1634                 goto nla_put_failure;
1635
1636 #if IS_ENABLED(CONFIG_IPV6)
1637         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1638                        !geneve->use_udp6_rx_checksums))
1639                 goto nla_put_failure;
1640 #endif
1641
1642         return 0;
1643
1644 nla_put_failure:
1645         return -EMSGSIZE;
1646 }
1647
1648 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1649         .kind           = "geneve",
1650         .maxtype        = IFLA_GENEVE_MAX,
1651         .policy         = geneve_policy,
1652         .priv_size      = sizeof(struct geneve_dev),
1653         .setup          = geneve_setup,
1654         .validate       = geneve_validate,
1655         .newlink        = geneve_newlink,
1656         .changelink     = geneve_changelink,
1657         .dellink        = geneve_dellink,
1658         .get_size       = geneve_get_size,
1659         .fill_info      = geneve_fill_info,
1660 };
1661
1662 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1663                                         u8 name_assign_type, u16 dst_port)
1664 {
1665         struct nlattr *tb[IFLA_MAX + 1];
1666         struct ip_tunnel_info info;
1667         struct net_device *dev;
1668         LIST_HEAD(list_kill);
1669         int err;
1670
1671         memset(tb, 0, sizeof(tb));
1672         dev = rtnl_create_link(net, name, name_assign_type,
1673                                &geneve_link_ops, tb);
1674         if (IS_ERR(dev))
1675                 return dev;
1676
1677         init_tnl_info(&info, dst_port);
1678         err = geneve_configure(net, dev, NULL, &info, true, true);
1679         if (err) {
1680                 free_netdev(dev);
1681                 return ERR_PTR(err);
1682         }
1683
1684         /* openvswitch users expect packet sizes to be unrestricted,
1685          * so set the largest MTU we can.
1686          */
1687         err = geneve_change_mtu(dev, IP_MAX_MTU);
1688         if (err)
1689                 goto err;
1690
1691         err = rtnl_configure_link(dev, NULL);
1692         if (err < 0)
1693                 goto err;
1694
1695         return dev;
1696 err:
1697         geneve_dellink(dev, &list_kill);
1698         unregister_netdevice_many(&list_kill);
1699         return ERR_PTR(err);
1700 }
1701 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1702
1703 static int geneve_netdevice_event(struct notifier_block *unused,
1704                                   unsigned long event, void *ptr)
1705 {
1706         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1707
1708         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1709             event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1710                 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1711         } else if (event == NETDEV_UNREGISTER) {
1712                 geneve_offload_rx_ports(dev, false);
1713         } else if (event == NETDEV_REGISTER) {
1714                 geneve_offload_rx_ports(dev, true);
1715         }
1716
1717         return NOTIFY_DONE;
1718 }
1719
1720 static struct notifier_block geneve_notifier_block __read_mostly = {
1721         .notifier_call = geneve_netdevice_event,
1722 };
1723
1724 static __net_init int geneve_init_net(struct net *net)
1725 {
1726         struct geneve_net *gn = net_generic(net, geneve_net_id);
1727
1728         INIT_LIST_HEAD(&gn->geneve_list);
1729         INIT_LIST_HEAD(&gn->sock_list);
1730         return 0;
1731 }
1732
1733 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1734 {
1735         struct geneve_net *gn = net_generic(net, geneve_net_id);
1736         struct geneve_dev *geneve, *next;
1737         struct net_device *dev, *aux;
1738
1739         /* gather any geneve devices that were moved into this ns */
1740         for_each_netdev_safe(net, dev, aux)
1741                 if (dev->rtnl_link_ops == &geneve_link_ops)
1742                         unregister_netdevice_queue(dev, head);
1743
1744         /* now gather any other geneve devices that were created in this ns */
1745         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1746                 /* If geneve->dev is in the same netns, it was already added
1747                  * to the list by the previous loop.
1748                  */
1749                 if (!net_eq(dev_net(geneve->dev), net))
1750                         unregister_netdevice_queue(geneve->dev, head);
1751         }
1752 }
1753
1754 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1755 {
1756         struct net *net;
1757         LIST_HEAD(list);
1758
1759         rtnl_lock();
1760         list_for_each_entry(net, net_list, exit_list)
1761                 geneve_destroy_tunnels(net, &list);
1762
1763         /* unregister the devices gathered above */
1764         unregister_netdevice_many(&list);
1765         rtnl_unlock();
1766
1767         list_for_each_entry(net, net_list, exit_list) {
1768                 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1769
1770                 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1771         }
1772 }
1773
1774 static struct pernet_operations geneve_net_ops = {
1775         .init = geneve_init_net,
1776         .exit_batch = geneve_exit_batch_net,
1777         .id   = &geneve_net_id,
1778         .size = sizeof(struct geneve_net),
1779 };
1780
1781 static int __init geneve_init_module(void)
1782 {
1783         int rc;
1784
1785         rc = register_pernet_subsys(&geneve_net_ops);
1786         if (rc)
1787                 goto out1;
1788
1789         rc = register_netdevice_notifier(&geneve_notifier_block);
1790         if (rc)
1791                 goto out2;
1792
1793         rc = rtnl_link_register(&geneve_link_ops);
1794         if (rc)
1795                 goto out3;
1796
1797         return 0;
1798 out3:
1799         unregister_netdevice_notifier(&geneve_notifier_block);
1800 out2:
1801         unregister_pernet_subsys(&geneve_net_ops);
1802 out1:
1803         return rc;
1804 }
1805 late_initcall(geneve_init_module);
1806
1807 static void __exit geneve_cleanup_module(void)
1808 {
1809         rtnl_link_unregister(&geneve_link_ops);
1810         unregister_netdevice_notifier(&geneve_notifier_block);
1811         unregister_pernet_subsys(&geneve_net_ops);
1812 }
1813 module_exit(geneve_cleanup_module);
1814
1815 MODULE_LICENSE("GPL");
1816 MODULE_VERSION(GENEVE_NETDEV_VER);
1817 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1818 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1819 MODULE_ALIAS_RTNL_LINK("geneve");