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/netdevice.h>
  16 #include <linux/etherdevice.h>
  17 #include <linux/hash.h>
  18 #include <net/dst_metadata.h>
  19 #include <net/gro_cells.h>
  20 #include <net/rtnetlink.h>
  21 #include <net/geneve.h>
  22 #include <net/protocol.h>
  23
  24 #define GENEVE_NETDEV_VER       "0.6"
  25
  26 #define GENEVE_UDP_PORT         6081
  27
  28 #define GENEVE_N_VID            (1u << 24)
  29 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  30
  31 #define VNI_HASH_BITS           10
  32 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  33
  34 static bool log_ecn_error = true;
  35 module_param(log_ecn_error, bool, 0644);
  36 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  37
  38 #define GENEVE_VER 0
  39 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  40
  41 /* per-network namespace private data for this module */
  42 struct geneve_net {
  43         struct list_head        geneve_list;
  44         struct list_head        sock_list;
  45 };
  46
  47 static int geneve_net_id;
  48
  49 union geneve_addr {
  50         struct sockaddr_in sin;
  51         struct sockaddr_in6 sin6;
  52         struct sockaddr sa;
  53 };
  54
  55 static union geneve_addr geneve_remote_unspec = { .sa.sa_family = AF_UNSPEC, };
  56
  57 /* Pseudo network device */
  58 struct geneve_dev {
  59         struct hlist_node  hlist;       /* vni hash table */
  60         struct net         *net;        /* netns for packet i/o */
  61         struct net_device  *dev;        /* netdev for geneve tunnel */
  62         struct geneve_sock *sock4;      /* IPv4 socket used for geneve tunnel */
  63 #if IS_ENABLED(CONFIG_IPV6)
  64         struct geneve_sock *sock6;      /* IPv6 socket used for geneve tunnel */
  65 #endif
  66         u8                 vni[3];      /* virtual network ID for tunnel */
  67         u8                 ttl;         /* TTL override */
  68         u8                 tos;         /* TOS override */
  69         union geneve_addr  remote;      /* IP address for link partner */
  70         struct list_head   next;        /* geneve's per namespace list */
  71         __be32             label;       /* IPv6 flowlabel override */
  72         __be16             dst_port;
  73         bool               collect_md;
  74         struct gro_cells   gro_cells;
  75         u32                flags;
  76         struct dst_cache   dst_cache;
  77 };
  78
  79 /* Geneve device flags */
  80 #define GENEVE_F_UDP_ZERO_CSUM_TX       BIT(0)
  81 #define GENEVE_F_UDP_ZERO_CSUM6_TX      BIT(1)
  82 #define GENEVE_F_UDP_ZERO_CSUM6_RX      BIT(2)
  83
  84 struct geneve_sock {
  85         bool                    collect_md;
  86         struct list_head        list;
  87         struct socket           *sock;
  88         struct rcu_head         rcu;
  89         int                     refcnt;
  90         struct udp_offload      udp_offloads;
  91         struct hlist_head       vni_list[VNI_HASH_SIZE];
  92         u32                     flags;
  93 };
  94
  95 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  96 {
  97         __u32 vnid;
  98
  99         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
 100         return hash_32(vnid, VNI_HASH_BITS);
 101 }
 102
 103 static __be64 vni_to_tunnel_id(const __u8 *vni)
 104 {
 105 #ifdef __BIG_ENDIAN
 106         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
 107 #else
 108         return (__force __be64)(((__force u64)vni[0] << 40) |
 109                                 ((__force u64)vni[1] << 48) |
 110                                 ((__force u64)vni[2] << 56));
 111 #endif
 112 }
 113
 114 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 115 {
 116         return gs->sock->sk->sk_family;
 117 }
 118
 119 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 120                                         __be32 addr, u8 vni[])
 121 {
 122         struct hlist_head *vni_list_head;
 123         struct geneve_dev *geneve;
 124         __u32 hash;
 125
 126         /* Find the device for this VNI */
 127         hash = geneve_net_vni_hash(vni);
 128         vni_list_head = &gs->vni_list[hash];
 129         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 130                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 131                     addr == geneve->remote.sin.sin_addr.s_addr)
 132                         return geneve;
 133         }
 134         return NULL;
 135 }
 136
 137 #if IS_ENABLED(CONFIG_IPV6)
 138 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 139                                          struct in6_addr addr6, u8 vni[])
 140 {
 141         struct hlist_head *vni_list_head;
 142         struct geneve_dev *geneve;
 143         __u32 hash;
 144
 145         /* Find the device for this VNI */
 146         hash = geneve_net_vni_hash(vni);
 147         vni_list_head = &gs->vni_list[hash];
 148         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 149                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 150                     ipv6_addr_equal(&addr6, &geneve->remote.sin6.sin6_addr))
 151                         return geneve;
 152         }
 153         return NULL;
 154 }
 155 #endif
 156
 157 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 158 {
 159         return (struct genevehdr *)(udp_hdr(skb) + 1);
 160 }
 161
 162 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 163                                             struct sk_buff *skb)
 164 {
 165         u8 *vni;
 166         __be32 addr;
 167         static u8 zero_vni[3];
 168 #if IS_ENABLED(CONFIG_IPV6)
 169         static struct in6_addr zero_addr6;
 170 #endif
 171
 172         if (geneve_get_sk_family(gs) == AF_INET) {
 173                 struct iphdr *iph;
 174
 175                 iph = ip_hdr(skb); /* outer IP header... */
 176
 177                 if (gs->collect_md) {
 178                         vni = zero_vni;
 179                         addr = 0;
 180                 } else {
 181                         vni = geneve_hdr(skb)->vni;
 182                         addr = iph->saddr;
 183                 }
 184
 185                 return geneve_lookup(gs, addr, vni);
 186 #if IS_ENABLED(CONFIG_IPV6)
 187         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 188                 struct ipv6hdr *ip6h;
 189                 struct in6_addr addr6;
 190
 191                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 192
 193                 if (gs->collect_md) {
 194                         vni = zero_vni;
 195                         addr6 = zero_addr6;
 196                 } else {
 197                         vni = geneve_hdr(skb)->vni;
 198                         addr6 = ip6h->saddr;
 199                 }
 200
 201                 return geneve6_lookup(gs, addr6, vni);
 202 #endif
 203         }
 204         return NULL;
 205 }
 206
 207 /* geneve receive/decap routine */
 208 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 209                       struct sk_buff *skb)
 210 {
 211         struct genevehdr *gnvh = geneve_hdr(skb);
 212         struct metadata_dst *tun_dst = NULL;
 213         struct pcpu_sw_netstats *stats;
 214         int err = 0;
 215         void *oiph;
 216
 217         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 218                 __be16 flags;
 219
 220                 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
 221                         (gnvh->oam ? TUNNEL_OAM : 0) |
 222                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 223
 224                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 225                                          vni_to_tunnel_id(gnvh->vni),
 226                                          gnvh->opt_len * 4);
 227                 if (!tun_dst)
 228                         goto drop;
 229                 /* Update tunnel dst according to Geneve options. */
 230                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 231                                         gnvh->options, gnvh->opt_len * 4);
 232         } else {
 233                 /* Drop packets w/ critical options,
 234                  * since we don't support any...
 235                  */
 236                 if (gnvh->critical)
 237                         goto drop;
 238         }
 239
 240         skb_reset_mac_header(skb);
 241         skb->protocol = eth_type_trans(skb, geneve->dev);
 242         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 243
 244         if (tun_dst)
 245                 skb_dst_set(skb, &tun_dst->dst);
 246
 247         /* Ignore packet loops (and multicast echo) */
 248         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
 249                 goto drop;
 250
 251         oiph = skb_network_header(skb);
 252         skb_reset_network_header(skb);
 253
 254         if (geneve_get_sk_family(gs) == AF_INET)
 255                 err = IP_ECN_decapsulate(oiph, skb);
 256 #if IS_ENABLED(CONFIG_IPV6)
 257         else
 258                 err = IP6_ECN_decapsulate(oiph, skb);
 259 #endif
 260
 261         if (unlikely(err)) {
 262                 if (log_ecn_error) {
 263                         if (geneve_get_sk_family(gs) == AF_INET)
 264                                 net_info_ratelimited("non-ECT from %pI4 "
 265                                                      "with TOS=%#x\n",
 266                                                      &((struct iphdr *)oiph)->saddr,
 267                                                      ((struct iphdr *)oiph)->tos);
 268 #if IS_ENABLED(CONFIG_IPV6)
 269                         else
 270                                 net_info_ratelimited("non-ECT from %pI6\n",
 271                                                      &((struct ipv6hdr *)oiph)->saddr);
 272 #endif
 273                 }
 274                 if (err > 1) {
 275                         ++geneve->dev->stats.rx_frame_errors;
 276                         ++geneve->dev->stats.rx_errors;
 277                         goto drop;
 278                 }
 279         }
 280
 281         stats = this_cpu_ptr(geneve->dev->tstats);
 282         u64_stats_update_begin(&stats->syncp);
 283         stats->rx_packets++;
 284         stats->rx_bytes += skb->len;
 285         u64_stats_update_end(&stats->syncp);
 286
 287         gro_cells_receive(&geneve->gro_cells, skb);
 288         return;
 289 drop:
 290         /* Consume bad packet */
 291         kfree_skb(skb);
 292 }
 293
 294 /* Setup stats when device is created */
 295 static int geneve_init(struct net_device *dev)
 296 {
 297         struct geneve_dev *geneve = netdev_priv(dev);
 298         int err;
 299
 300         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 301         if (!dev->tstats)
 302                 return -ENOMEM;
 303
 304         err = gro_cells_init(&geneve->gro_cells, dev);
 305         if (err) {
 306                 free_percpu(dev->tstats);
 307                 return err;
 308         }
 309
 310         err = dst_cache_init(&geneve->dst_cache, GFP_KERNEL);
 311         if (err) {
 312                 free_percpu(dev->tstats);
 313                 gro_cells_destroy(&geneve->gro_cells);
 314                 return err;
 315         }
 316
 317         return 0;
 318 }
 319
 320 static void geneve_uninit(struct net_device *dev)
 321 {
 322         struct geneve_dev *geneve = netdev_priv(dev);
 323
 324         dst_cache_destroy(&geneve->dst_cache);
 325         gro_cells_destroy(&geneve->gro_cells);
 326         free_percpu(dev->tstats);
 327 }
 328
 329 /* Callback from net/ipv4/udp.c to receive packets */
 330 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 331 {
 332         struct genevehdr *geneveh;
 333         struct geneve_dev *geneve;
 334         struct geneve_sock *gs;
 335         int opts_len;
 336
 337         /* Need Geneve and inner Ethernet header to be present */
 338         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 339                 goto error;
 340
 341         /* Return packets with reserved bits set */
 342         geneveh = geneve_hdr(skb);
 343         if (unlikely(geneveh->ver != GENEVE_VER))
 344                 goto error;
 345
 346         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 347                 goto error;
 348
 349         gs = rcu_dereference_sk_user_data(sk);
 350         if (!gs)
 351                 goto drop;
 352
 353         geneve = geneve_lookup_skb(gs, skb);
 354         if (!geneve)
 355                 goto drop;
 356
 357         opts_len = geneveh->opt_len * 4;
 358         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 359                                  htons(ETH_P_TEB),
 360                                  !net_eq(geneve->net, dev_net(geneve->dev))))
 361                 goto drop;
 362
 363         geneve_rx(geneve, gs, skb);
 364         return 0;
 365
 366 drop:
 367         /* Consume bad packet */
 368         kfree_skb(skb);
 369         return 0;
 370
 371 error:
 372         /* Let the UDP layer deal with the skb */
 373         return 1;
 374 }
 375
 376 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 377                                          __be16 port, u32 flags)
 378 {
 379         struct socket *sock;
 380         struct udp_port_cfg udp_conf;
 381         int err;
 382
 383         memset(&udp_conf, 0, sizeof(udp_conf));
 384
 385         if (ipv6) {
 386                 udp_conf.family = AF_INET6;
 387                 udp_conf.ipv6_v6only = 1;
 388                 udp_conf.use_udp6_rx_checksums =
 389                     !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX);
 390         } else {
 391                 udp_conf.family = AF_INET;
 392                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 393         }
 394
 395         udp_conf.local_udp_port = port;
 396
 397         /* Open UDP socket */
 398         err = udp_sock_create(net, &udp_conf, &sock);
 399         if (err < 0)
 400                 return ERR_PTR(err);
 401
 402         return sock;
 403 }
 404
 405 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
 406 {
 407         struct net_device *dev;
 408         struct sock *sk = gs->sock->sk;
 409         struct net *net = sock_net(sk);
 410         sa_family_t sa_family = geneve_get_sk_family(gs);
 411         __be16 port = inet_sk(sk)->inet_sport;
 412         int err;
 413
 414         if (sa_family == AF_INET) {
 415                 err = udp_add_offload(sock_net(sk), &gs->udp_offloads);
 416                 if (err)
 417                         pr_warn("geneve: udp_add_offload failed with status %d\n",
 418                                 err);
 419         }
 420
 421         rcu_read_lock();
 422         for_each_netdev_rcu(net, dev) {
 423                 if (dev->netdev_ops->ndo_add_geneve_port)
 424                         dev->netdev_ops->ndo_add_geneve_port(dev, sa_family,
 425                                                              port);
 426         }
 427         rcu_read_unlock();
 428 }
 429
 430 static int geneve_hlen(struct genevehdr *gh)
 431 {
 432         return sizeof(*gh) + gh->opt_len * 4;
 433 }
 434
 435 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
 436                                            struct sk_buff *skb,
 437                                            struct udp_offload *uoff)
 438 {
 439         struct sk_buff *p, **pp = NULL;
 440         struct genevehdr *gh, *gh2;
 441         unsigned int hlen, gh_len, off_gnv;
 442         const struct packet_offload *ptype;
 443         __be16 type;
 444         int flush = 1;
 445
 446         off_gnv = skb_gro_offset(skb);
 447         hlen = off_gnv + sizeof(*gh);
 448         gh = skb_gro_header_fast(skb, off_gnv);
 449         if (skb_gro_header_hard(skb, hlen)) {
 450                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 451                 if (unlikely(!gh))
 452                         goto out;
 453         }
 454
 455         if (gh->ver != GENEVE_VER || gh->oam)
 456                 goto out;
 457         gh_len = geneve_hlen(gh);
 458
 459         hlen = off_gnv + gh_len;
 460         if (skb_gro_header_hard(skb, hlen)) {
 461                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 462                 if (unlikely(!gh))
 463                         goto out;
 464         }
 465
 466         for (p = *head; p; p = p->next) {
 467                 if (!NAPI_GRO_CB(p)->same_flow)
 468                         continue;
 469
 470                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 471                 if (gh->opt_len != gh2->opt_len ||
 472                     memcmp(gh, gh2, gh_len)) {
 473                         NAPI_GRO_CB(p)->same_flow = 0;
 474                         continue;
 475                 }
 476         }
 477
 478         type = gh->proto_type;
 479
 480         rcu_read_lock();
 481         ptype = gro_find_receive_by_type(type);
 482         if (!ptype)
 483                 goto out_unlock;
 484
 485         skb_gro_pull(skb, gh_len);
 486         skb_gro_postpull_rcsum(skb, gh, gh_len);
 487         pp = ptype->callbacks.gro_receive(head, skb);
 488         flush = 0;
 489
 490 out_unlock:
 491         rcu_read_unlock();
 492 out:
 493         NAPI_GRO_CB(skb)->flush |= flush;
 494
 495         return pp;
 496 }
 497
 498 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
 499                                struct udp_offload *uoff)
 500 {
 501         struct genevehdr *gh;
 502         struct packet_offload *ptype;
 503         __be16 type;
 504         int gh_len;
 505         int err = -ENOSYS;
 506
 507         gh = (struct genevehdr *)(skb->data + nhoff);
 508         gh_len = geneve_hlen(gh);
 509         type = gh->proto_type;
 510
 511         rcu_read_lock();
 512         ptype = gro_find_complete_by_type(type);
 513         if (ptype)
 514                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 515
 516         rcu_read_unlock();
 517
 518         skb_set_inner_mac_header(skb, nhoff + gh_len);
 519
 520         return err;
 521 }
 522
 523 /* Create new listen socket if needed */
 524 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 525                                                 bool ipv6, u32 flags)
 526 {
 527         struct geneve_net *gn = net_generic(net, geneve_net_id);
 528         struct geneve_sock *gs;
 529         struct socket *sock;
 530         struct udp_tunnel_sock_cfg tunnel_cfg;
 531         int h;
 532
 533         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 534         if (!gs)
 535                 return ERR_PTR(-ENOMEM);
 536
 537         sock = geneve_create_sock(net, ipv6, port, flags);
 538         if (IS_ERR(sock)) {
 539                 kfree(gs);
 540                 return ERR_CAST(sock);
 541         }
 542
 543         gs->sock = sock;
 544         gs->refcnt = 1;
 545         for (h = 0; h < VNI_HASH_SIZE; ++h)
 546                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 547
 548         /* Initialize the geneve udp offloads structure */
 549         gs->udp_offloads.port = port;
 550         gs->udp_offloads.callbacks.gro_receive  = geneve_gro_receive;
 551         gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
 552         geneve_notify_add_rx_port(gs);
 553
 554         /* Mark socket as an encapsulation socket */
 555         tunnel_cfg.sk_user_data = gs;
 556         tunnel_cfg.encap_type = 1;
 557         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 558         tunnel_cfg.encap_destroy = NULL;
 559         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 560         list_add(&gs->list, &gn->sock_list);
 561         return gs;
 562 }
 563
 564 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 565 {
 566         struct net_device *dev;
 567         struct sock *sk = gs->sock->sk;
 568         struct net *net = sock_net(sk);
 569         sa_family_t sa_family = geneve_get_sk_family(gs);
 570         __be16 port = inet_sk(sk)->inet_sport;
 571
 572         rcu_read_lock();
 573         for_each_netdev_rcu(net, dev) {
 574                 if (dev->netdev_ops->ndo_del_geneve_port)
 575                         dev->netdev_ops->ndo_del_geneve_port(dev, sa_family,
 576                                                              port);
 577         }
 578
 579         rcu_read_unlock();
 580
 581         if (sa_family == AF_INET)
 582                 udp_del_offload(&gs->udp_offloads);
 583 }
 584
 585 static void __geneve_sock_release(struct geneve_sock *gs)
 586 {
 587         if (!gs || --gs->refcnt)
 588                 return;
 589
 590         list_del(&gs->list);
 591         geneve_notify_del_rx_port(gs);
 592         udp_tunnel_sock_release(gs->sock);
 593         kfree_rcu(gs, rcu);
 594 }
 595
 596 static void geneve_sock_release(struct geneve_dev *geneve)
 597 {
 598         __geneve_sock_release(geneve->sock4);
 599 #if IS_ENABLED(CONFIG_IPV6)
 600         __geneve_sock_release(geneve->sock6);
 601 #endif
 602 }
 603
 604 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 605                                             sa_family_t family,
 606                                             __be16 dst_port)
 607 {
 608         struct geneve_sock *gs;
 609
 610         list_for_each_entry(gs, &gn->sock_list, list) {
 611                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 612                     geneve_get_sk_family(gs) == family) {
 613                         return gs;
 614                 }
 615         }
 616         return NULL;
 617 }
 618
 619 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 620 {
 621         struct net *net = geneve->net;
 622         struct geneve_net *gn = net_generic(net, geneve_net_id);
 623         struct geneve_sock *gs;
 624         __u32 hash;
 625
 626         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
 627         if (gs) {
 628                 gs->refcnt++;
 629                 goto out;
 630         }
 631
 632         gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
 633         if (IS_ERR(gs))
 634                 return PTR_ERR(gs);
 635
 636 out:
 637         gs->collect_md = geneve->collect_md;
 638         gs->flags = geneve->flags;
 639 #if IS_ENABLED(CONFIG_IPV6)
 640         if (ipv6)
 641                 geneve->sock6 = gs;
 642         else
 643 #endif
 644                 geneve->sock4 = gs;
 645
 646         hash = geneve_net_vni_hash(geneve->vni);
 647         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 648         return 0;
 649 }
 650
 651 static int geneve_open(struct net_device *dev)
 652 {
 653         struct geneve_dev *geneve = netdev_priv(dev);
 654         bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
 655         bool metadata = geneve->collect_md;
 656         int ret = 0;
 657
 658         geneve->sock4 = NULL;
 659 #if IS_ENABLED(CONFIG_IPV6)
 660         geneve->sock6 = NULL;
 661         if (ipv6 || metadata)
 662                 ret = geneve_sock_add(geneve, true);
 663 #endif
 664         if (!ret && (!ipv6 || metadata))
 665                 ret = geneve_sock_add(geneve, false);
 666         if (ret < 0)
 667                 geneve_sock_release(geneve);
 668
 669         return ret;
 670 }
 671
 672 static int geneve_stop(struct net_device *dev)
 673 {
 674         struct geneve_dev *geneve = netdev_priv(dev);
 675
 676         if (!hlist_unhashed(&geneve->hlist))
 677                 hlist_del_rcu(&geneve->hlist);
 678         geneve_sock_release(geneve);
 679         return 0;
 680 }
 681
 682 static void geneve_build_header(struct genevehdr *geneveh,
 683                                 __be16 tun_flags, u8 vni[3],
 684                                 u8 options_len, u8 *options)
 685 {
 686         geneveh->ver = GENEVE_VER;
 687         geneveh->opt_len = options_len / 4;
 688         geneveh->oam = !!(tun_flags & TUNNEL_OAM);
 689         geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 690         geneveh->rsvd1 = 0;
 691         memcpy(geneveh->vni, vni, 3);
 692         geneveh->proto_type = htons(ETH_P_TEB);
 693         geneveh->rsvd2 = 0;
 694
 695         memcpy(geneveh->options, options, options_len);
 696 }
 697
 698 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 699                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 700                             u32 flags, bool xnet)
 701 {
 702         struct genevehdr *gnvh;
 703         int min_headroom;
 704         int err;
 705         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
 706
 707         skb_scrub_packet(skb, xnet);
 708
 709         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 710                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 711         err = skb_cow_head(skb, min_headroom);
 712         if (unlikely(err)) {
 713                 kfree_skb(skb);
 714                 goto free_rt;
 715         }
 716
 717         skb = udp_tunnel_handle_offloads(skb, udp_sum);
 718         if (IS_ERR(skb)) {
 719                 err = PTR_ERR(skb);
 720                 goto free_rt;
 721         }
 722
 723         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 724         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 725
 726         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 727         return 0;
 728
 729 free_rt:
 730         ip_rt_put(rt);
 731         return err;
 732 }
 733
 734 #if IS_ENABLED(CONFIG_IPV6)
 735 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 736                              __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 737                              u32 flags, bool xnet)
 738 {
 739         struct genevehdr *gnvh;
 740         int min_headroom;
 741         int err;
 742         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
 743
 744         skb_scrub_packet(skb, xnet);
 745
 746         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
 747                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
 748         err = skb_cow_head(skb, min_headroom);
 749         if (unlikely(err)) {
 750                 kfree_skb(skb);
 751                 goto free_dst;
 752         }
 753
 754         skb = udp_tunnel_handle_offloads(skb, udp_sum);
 755         if (IS_ERR(skb)) {
 756                 err = PTR_ERR(skb);
 757                 goto free_dst;
 758         }
 759
 760         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 761         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 762
 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 #endif
 771
 772 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 773                                        struct net_device *dev,
 774                                        struct flowi4 *fl4,
 775                                        struct ip_tunnel_info *info)
 776 {
 777         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 778         struct geneve_dev *geneve = netdev_priv(dev);
 779         struct dst_cache *dst_cache;
 780         struct rtable *rt = NULL;
 781         __u8 tos;
 782
 783         memset(fl4, 0, sizeof(*fl4));
 784         fl4->flowi4_mark = skb->mark;
 785         fl4->flowi4_proto = IPPROTO_UDP;
 786
 787         if (info) {
 788                 fl4->daddr = info->key.u.ipv4.dst;
 789                 fl4->saddr = info->key.u.ipv4.src;
 790                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 791                 dst_cache = &info->dst_cache;
 792         } else {
 793                 tos = geneve->tos;
 794                 if (tos == 1) {
 795                         const struct iphdr *iip = ip_hdr(skb);
 796
 797                         tos = ip_tunnel_get_dsfield(iip, skb);
 798                         use_cache = false;
 799                 }
 800
 801                 fl4->flowi4_tos = RT_TOS(tos);
 802                 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
 803                 dst_cache = &geneve->dst_cache;
 804         }
 805
 806         if (use_cache) {
 807                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 808                 if (rt)
 809                         return rt;
 810         }
 811
 812         rt = ip_route_output_key(geneve->net, fl4);
 813         if (IS_ERR(rt)) {
 814                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 815                 return ERR_PTR(-ENETUNREACH);
 816         }
 817         if (rt->dst.dev == dev) { /* is this necessary? */
 818                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 819                 ip_rt_put(rt);
 820                 return ERR_PTR(-ELOOP);
 821         }
 822         if (use_cache)
 823                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 824         return rt;
 825 }
 826
 827 #if IS_ENABLED(CONFIG_IPV6)
 828 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 829                                            struct net_device *dev,
 830                                            struct flowi6 *fl6,
 831                                            struct ip_tunnel_info *info)
 832 {
 833         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 834         struct geneve_dev *geneve = netdev_priv(dev);
 835         struct geneve_sock *gs6 = geneve->sock6;
 836         struct dst_entry *dst = NULL;
 837         struct dst_cache *dst_cache;
 838         __u8 prio;
 839
 840         memset(fl6, 0, sizeof(*fl6));
 841         fl6->flowi6_mark = skb->mark;
 842         fl6->flowi6_proto = IPPROTO_UDP;
 843
 844         if (info) {
 845                 fl6->daddr = info->key.u.ipv6.dst;
 846                 fl6->saddr = info->key.u.ipv6.src;
 847                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(info->key.tos),
 848                                                    info->key.label);
 849                 dst_cache = &info->dst_cache;
 850         } else {
 851                 prio = geneve->tos;
 852                 if (prio == 1) {
 853                         const struct iphdr *iip = ip_hdr(skb);
 854
 855                         prio = ip_tunnel_get_dsfield(iip, skb);
 856                         use_cache = false;
 857                 }
 858
 859                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 860                                                    geneve->label);
 861                 fl6->daddr = geneve->remote.sin6.sin6_addr;
 862                 dst_cache = &geneve->dst_cache;
 863         }
 864
 865         if (use_cache) {
 866                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 867                 if (dst)
 868                         return dst;
 869         }
 870
 871         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 872                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 873                 return ERR_PTR(-ENETUNREACH);
 874         }
 875         if (dst->dev == dev) { /* is this necessary? */
 876                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 877                 dst_release(dst);
 878                 return ERR_PTR(-ELOOP);
 879         }
 880
 881         if (use_cache)
 882                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 883         return dst;
 884 }
 885 #endif
 886
 887 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 888 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 889 {
 890 #ifdef __BIG_ENDIAN
 891         vni[0] = (__force __u8)(tun_id >> 16);
 892         vni[1] = (__force __u8)(tun_id >> 8);
 893         vni[2] = (__force __u8)tun_id;
 894 #else
 895         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 896         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 897         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 898 #endif
 899 }
 900
 901 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 902                                    struct ip_tunnel_info *info)
 903 {
 904         struct geneve_dev *geneve = netdev_priv(dev);
 905         struct geneve_sock *gs4 = geneve->sock4;
 906         struct rtable *rt = NULL;
 907         const struct iphdr *iip; /* interior IP header */
 908         int err = -EINVAL;
 909         struct flowi4 fl4;
 910         __u8 tos, ttl;
 911         __be16 sport;
 912         __be16 df;
 913         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 914         u32 flags = geneve->flags;
 915
 916         if (geneve->collect_md) {
 917                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 918                         netdev_dbg(dev, "no tunnel metadata\n");
 919                         goto tx_error;
 920                 }
 921                 if (info && ip_tunnel_info_af(info) != AF_INET)
 922                         goto tx_error;
 923         }
 924
 925         rt = geneve_get_v4_rt(skb, dev, &fl4, info);
 926         if (IS_ERR(rt)) {
 927                 err = PTR_ERR(rt);
 928                 goto tx_error;
 929         }
 930
 931         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 932         skb_reset_mac_header(skb);
 933
 934         iip = ip_hdr(skb);
 935
 936         if (info) {
 937                 const struct ip_tunnel_key *key = &info->key;
 938                 u8 *opts = NULL;
 939                 u8 vni[3];
 940
 941                 tunnel_id_to_vni(key->tun_id, vni);
 942                 if (info->options_len)
 943                         opts = ip_tunnel_info_opts(info);
 944
 945                 if (key->tun_flags & TUNNEL_CSUM)
 946                         flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
 947                 else
 948                         flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
 949
 950                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 951                                        info->options_len, opts, flags, xnet);
 952                 if (unlikely(err))
 953                         goto err;
 954
 955                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
 956                 ttl = key->ttl;
 957                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 958         } else {
 959                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 960                                        0, NULL, flags, xnet);
 961                 if (unlikely(err))
 962                         goto err;
 963
 964                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 965                 ttl = geneve->ttl;
 966                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 967                         ttl = 1;
 968                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 969                 df = 0;
 970         }
 971         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 972                             tos, ttl, df, sport, geneve->dst_port,
 973                             !net_eq(geneve->net, dev_net(geneve->dev)),
 974                             !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
 975
 976         return NETDEV_TX_OK;
 977
 978 tx_error:
 979         dev_kfree_skb(skb);
 980 err:
 981         if (err == -ELOOP)
 982                 dev->stats.collisions++;
 983         else if (err == -ENETUNREACH)
 984                 dev->stats.tx_carrier_errors++;
 985         else
 986                 dev->stats.tx_errors++;
 987         return NETDEV_TX_OK;
 988 }
 989
 990 #if IS_ENABLED(CONFIG_IPV6)
 991 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 992                                     struct ip_tunnel_info *info)
 993 {
 994         struct geneve_dev *geneve = netdev_priv(dev);
 995         struct geneve_sock *gs6 = geneve->sock6;
 996         struct dst_entry *dst = NULL;
 997         const struct iphdr *iip; /* interior IP header */
 998         int err = -EINVAL;
 999         struct flowi6 fl6;
1000         __u8 prio, ttl;
1001         __be16 sport;
1002         __be32 label;
1003         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
1004         u32 flags = geneve->flags;
1005
1006         if (geneve->collect_md) {
1007                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1008                         netdev_dbg(dev, "no tunnel metadata\n");
1009                         goto tx_error;
1010                 }
1011         }
1012
1013         dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1014         if (IS_ERR(dst)) {
1015                 err = PTR_ERR(dst);
1016                 goto tx_error;
1017         }
1018
1019         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1020         skb_reset_mac_header(skb);
1021
1022         iip = ip_hdr(skb);
1023
1024         if (info) {
1025                 const struct ip_tunnel_key *key = &info->key;
1026                 u8 *opts = NULL;
1027                 u8 vni[3];
1028
1029                 tunnel_id_to_vni(key->tun_id, vni);
1030                 if (info->options_len)
1031                         opts = ip_tunnel_info_opts(info);
1032
1033                 if (key->tun_flags & TUNNEL_CSUM)
1034                         flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
1035                 else
1036                         flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1037
1038                 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
1039                                         info->options_len, opts,
1040                                         flags, xnet);
1041                 if (unlikely(err))
1042                         goto err;
1043
1044                 prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
1045                 ttl = key->ttl;
1046                 label = info->key.label;
1047         } else {
1048                 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
1049                                         0, NULL, flags, xnet);
1050                 if (unlikely(err))
1051                         goto err;
1052
1053                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1054                                            iip, skb);
1055                 ttl = geneve->ttl;
1056                 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1057                         ttl = 1;
1058                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1059                 label = geneve->label;
1060         }
1061
1062         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1063                              &fl6.saddr, &fl6.daddr, prio, ttl, label,
1064                              sport, geneve->dst_port,
1065                              !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1066         return NETDEV_TX_OK;
1067
1068 tx_error:
1069         dev_kfree_skb(skb);
1070 err:
1071         if (err == -ELOOP)
1072                 dev->stats.collisions++;
1073         else if (err == -ENETUNREACH)
1074                 dev->stats.tx_carrier_errors++;
1075         else
1076                 dev->stats.tx_errors++;
1077         return NETDEV_TX_OK;
1078 }
1079 #endif
1080
1081 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1082 {
1083         struct geneve_dev *geneve = netdev_priv(dev);
1084         struct ip_tunnel_info *info = NULL;
1085
1086         if (geneve->collect_md)
1087                 info = skb_tunnel_info(skb);
1088
1089 #if IS_ENABLED(CONFIG_IPV6)
1090         if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1091             (!info && geneve->remote.sa.sa_family == AF_INET6))
1092                 return geneve6_xmit_skb(skb, dev, info);
1093 #endif
1094         return geneve_xmit_skb(skb, dev, info);
1095 }
1096
1097 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1098 {
1099         /* The max_mtu calculation does not take account of GENEVE
1100          * options, to avoid excluding potentially valid
1101          * configurations.
1102          */
1103         int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - sizeof(struct iphdr)
1104                 - dev->hard_header_len;
1105
1106         if (new_mtu < 68)
1107                 return -EINVAL;
1108
1109         if (new_mtu > max_mtu) {
1110                 if (strict)
1111                         return -EINVAL;
1112
1113                 new_mtu = max_mtu;
1114         }
1115
1116         dev->mtu = new_mtu;
1117         return 0;
1118 }
1119
1120 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1121 {
1122         return __geneve_change_mtu(dev, new_mtu, true);
1123 }
1124
1125 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1126 {
1127         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1128         struct geneve_dev *geneve = netdev_priv(dev);
1129         struct rtable *rt;
1130         struct flowi4 fl4;
1131 #if IS_ENABLED(CONFIG_IPV6)
1132         struct dst_entry *dst;
1133         struct flowi6 fl6;
1134 #endif
1135
1136         if (ip_tunnel_info_af(info) == AF_INET) {
1137                 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1138                 if (IS_ERR(rt))
1139                         return PTR_ERR(rt);
1140
1141                 ip_rt_put(rt);
1142                 info->key.u.ipv4.src = fl4.saddr;
1143 #if IS_ENABLED(CONFIG_IPV6)
1144         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1145                 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1146                 if (IS_ERR(dst))
1147                         return PTR_ERR(dst);
1148
1149                 dst_release(dst);
1150                 info->key.u.ipv6.src = fl6.saddr;
1151 #endif
1152         } else {
1153                 return -EINVAL;
1154         }
1155
1156         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1157                                              1, USHRT_MAX, true);
1158         info->key.tp_dst = geneve->dst_port;
1159         return 0;
1160 }
1161
1162 static const struct net_device_ops geneve_netdev_ops = {
1163         .ndo_init               = geneve_init,
1164         .ndo_uninit             = geneve_uninit,
1165         .ndo_open               = geneve_open,
1166         .ndo_stop               = geneve_stop,
1167         .ndo_start_xmit         = geneve_xmit,
1168         .ndo_get_stats64        = ip_tunnel_get_stats64,
1169         .ndo_change_mtu         = geneve_change_mtu,
1170         .ndo_validate_addr      = eth_validate_addr,
1171         .ndo_set_mac_address    = eth_mac_addr,
1172         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1173 };
1174
1175 static void geneve_get_drvinfo(struct net_device *dev,
1176                                struct ethtool_drvinfo *drvinfo)
1177 {
1178         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1179         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1180 }
1181
1182 static const struct ethtool_ops geneve_ethtool_ops = {
1183         .get_drvinfo    = geneve_get_drvinfo,
1184         .get_link       = ethtool_op_get_link,
1185 };
1186
1187 /* Info for udev, that this is a virtual tunnel endpoint */
1188 static struct device_type geneve_type = {
1189         .name = "geneve",
1190 };
1191
1192 /* Calls the ndo_add_geneve_port of the caller in order to
1193  * supply the listening GENEVE udp ports. Callers are expected
1194  * to implement the ndo_add_geneve_port.
1195  */
1196 void geneve_get_rx_port(struct net_device *dev)
1197 {
1198         struct net *net = dev_net(dev);
1199         struct geneve_net *gn = net_generic(net, geneve_net_id);
1200         struct geneve_sock *gs;
1201         sa_family_t sa_family;
1202         struct sock *sk;
1203         __be16 port;
1204
1205         rcu_read_lock();
1206         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1207                 sk = gs->sock->sk;
1208                 sa_family = sk->sk_family;
1209                 port = inet_sk(sk)->inet_sport;
1210                 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family, port);
1211         }
1212         rcu_read_unlock();
1213 }
1214 EXPORT_SYMBOL_GPL(geneve_get_rx_port);
1215
1216 /* Initialize the device structure. */
1217 static void geneve_setup(struct net_device *dev)
1218 {
1219         ether_setup(dev);
1220
1221         dev->netdev_ops = &geneve_netdev_ops;
1222         dev->ethtool_ops = &geneve_ethtool_ops;
1223         dev->destructor = free_netdev;
1224
1225         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1226
1227         dev->features    |= NETIF_F_LLTX;
1228         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1229         dev->features    |= NETIF_F_RXCSUM;
1230         dev->features    |= NETIF_F_GSO_SOFTWARE;
1231
1232         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1233         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1234
1235         netif_keep_dst(dev);
1236         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1237         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1238         eth_hw_addr_random(dev);
1239 }
1240
1241 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1242         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1243         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1244         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1245         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1246         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1247         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1248         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1249         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1250         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1251         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1252         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1253 };
1254
1255 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1256 {
1257         if (tb[IFLA_ADDRESS]) {
1258                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1259                         return -EINVAL;
1260
1261                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1262                         return -EADDRNOTAVAIL;
1263         }
1264
1265         if (!data)
1266                 return -EINVAL;
1267
1268         if (data[IFLA_GENEVE_ID]) {
1269                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1270
1271                 if (vni >= GENEVE_VID_MASK)
1272                         return -ERANGE;
1273         }
1274
1275         return 0;
1276 }
1277
1278 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1279                                           __be16 dst_port,
1280                                           union geneve_addr *remote,
1281                                           u8 vni[],
1282                                           bool *tun_on_same_port,
1283                                           bool *tun_collect_md)
1284 {
1285         struct geneve_dev *geneve, *t;
1286
1287         *tun_on_same_port = false;
1288         *tun_collect_md = false;
1289         t = NULL;
1290         list_for_each_entry(geneve, &gn->geneve_list, next) {
1291                 if (geneve->dst_port == dst_port) {
1292                         *tun_collect_md = geneve->collect_md;
1293                         *tun_on_same_port = true;
1294                 }
1295                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1296                     !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1297                     dst_port == geneve->dst_port)
1298                         t = geneve;
1299         }
1300         return t;
1301 }
1302
1303 static int geneve_configure(struct net *net, struct net_device *dev,
1304                             union geneve_addr *remote,
1305                             __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1306                             __be16 dst_port, bool metadata, u32 flags)
1307 {
1308         struct geneve_net *gn = net_generic(net, geneve_net_id);
1309         struct geneve_dev *t, *geneve = netdev_priv(dev);
1310         bool tun_collect_md, tun_on_same_port;
1311         int err, encap_len;
1312
1313         if (!remote)
1314                 return -EINVAL;
1315         if (metadata &&
1316             (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1317                 return -EINVAL;
1318
1319         geneve->net = net;
1320         geneve->dev = dev;
1321
1322         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1323         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1324         geneve->vni[2] =  vni & 0x000000ff;
1325
1326         if ((remote->sa.sa_family == AF_INET &&
1327              IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1328             (remote->sa.sa_family == AF_INET6 &&
1329              ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1330                 return -EINVAL;
1331         if (label && remote->sa.sa_family != AF_INET6)
1332                 return -EINVAL;
1333
1334         geneve->remote = *remote;
1335
1336         geneve->ttl = ttl;
1337         geneve->tos = tos;
1338         geneve->label = label;
1339         geneve->dst_port = dst_port;
1340         geneve->collect_md = metadata;
1341         geneve->flags = flags;
1342
1343         t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1344                             &tun_on_same_port, &tun_collect_md);
1345         if (t)
1346                 return -EBUSY;
1347
1348         /* make enough headroom for basic scenario */
1349         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1350         if (remote->sa.sa_family == AF_INET)
1351                 encap_len += sizeof(struct iphdr);
1352         else
1353                 encap_len += sizeof(struct ipv6hdr);
1354         dev->needed_headroom = encap_len + ETH_HLEN;
1355
1356         if (metadata) {
1357                 if (tun_on_same_port)
1358                         return -EPERM;
1359         } else {
1360                 if (tun_collect_md)
1361                         return -EPERM;
1362         }
1363
1364         dst_cache_reset(&geneve->dst_cache);
1365
1366         err = register_netdevice(dev);
1367         if (err)
1368                 return err;
1369
1370         list_add(&geneve->next, &gn->geneve_list);
1371         return 0;
1372 }
1373
1374 static int geneve_newlink(struct net *net, struct net_device *dev,
1375                           struct nlattr *tb[], struct nlattr *data[])
1376 {
1377         __be16 dst_port = htons(GENEVE_UDP_PORT);
1378         __u8 ttl = 0, tos = 0;
1379         bool metadata = false;
1380         union geneve_addr remote = geneve_remote_unspec;
1381         __be32 label = 0;
1382         __u32 vni = 0;
1383         u32 flags = 0;
1384
1385         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1386                 return -EINVAL;
1387
1388         if (data[IFLA_GENEVE_REMOTE]) {
1389                 remote.sa.sa_family = AF_INET;
1390                 remote.sin.sin_addr.s_addr =
1391                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1392         }
1393
1394         if (data[IFLA_GENEVE_REMOTE6]) {
1395                 if (!IS_ENABLED(CONFIG_IPV6))
1396                         return -EPFNOSUPPORT;
1397
1398                 remote.sa.sa_family = AF_INET6;
1399                 remote.sin6.sin6_addr =
1400                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1401
1402                 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1403                     IPV6_ADDR_LINKLOCAL) {
1404                         netdev_dbg(dev, "link-local remote is unsupported\n");
1405                         return -EINVAL;
1406                 }
1407         }
1408
1409         if (data[IFLA_GENEVE_ID])
1410                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1411
1412         if (data[IFLA_GENEVE_TTL])
1413                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1414
1415         if (data[IFLA_GENEVE_TOS])
1416                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1417
1418         if (data[IFLA_GENEVE_LABEL])
1419                 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1420                         IPV6_FLOWLABEL_MASK;
1421
1422         if (data[IFLA_GENEVE_PORT])
1423                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1424
1425         if (data[IFLA_GENEVE_COLLECT_METADATA])
1426                 metadata = true;
1427
1428         if (data[IFLA_GENEVE_UDP_CSUM] &&
1429             !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1430                 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1431
1432         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1433             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1434                 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1435
1436         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1437             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1438                 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1439
1440         return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1441                                 dst_port, metadata, flags);
1442 }
1443
1444 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1445 {
1446         struct geneve_dev *geneve = netdev_priv(dev);
1447
1448         list_del(&geneve->next);
1449         unregister_netdevice_queue(dev, head);
1450 }
1451
1452 static size_t geneve_get_size(const struct net_device *dev)
1453 {
1454         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1455                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1456                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1457                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1458                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1459                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1460                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1461                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1462                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1463                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1464                 0;
1465 }
1466
1467 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1468 {
1469         struct geneve_dev *geneve = netdev_priv(dev);
1470         __u32 vni;
1471
1472         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1473         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1474                 goto nla_put_failure;
1475
1476         if (geneve->remote.sa.sa_family == AF_INET) {
1477                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1478                                     geneve->remote.sin.sin_addr.s_addr))
1479                         goto nla_put_failure;
1480 #if IS_ENABLED(CONFIG_IPV6)
1481         } else {
1482                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1483                                      &geneve->remote.sin6.sin6_addr))
1484                         goto nla_put_failure;
1485 #endif
1486         }
1487
1488         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1489             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1490             nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1491                 goto nla_put_failure;
1492
1493         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1494                 goto nla_put_failure;
1495
1496         if (geneve->collect_md) {
1497                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1498                         goto nla_put_failure;
1499         }
1500
1501         if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1502                        !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1503             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1504                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1505             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1506                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1507                 goto nla_put_failure;
1508
1509         return 0;
1510
1511 nla_put_failure:
1512         return -EMSGSIZE;
1513 }
1514
1515 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1516         .kind           = "geneve",
1517         .maxtype        = IFLA_GENEVE_MAX,
1518         .policy         = geneve_policy,
1519         .priv_size      = sizeof(struct geneve_dev),
1520         .setup          = geneve_setup,
1521         .validate       = geneve_validate,
1522         .newlink        = geneve_newlink,
1523         .dellink        = geneve_dellink,
1524         .get_size       = geneve_get_size,
1525         .fill_info      = geneve_fill_info,
1526 };
1527
1528 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1529                                         u8 name_assign_type, u16 dst_port)
1530 {
1531         struct nlattr *tb[IFLA_MAX + 1];
1532         struct net_device *dev;
1533         int err;
1534
1535         memset(tb, 0, sizeof(tb));
1536         dev = rtnl_create_link(net, name, name_assign_type,
1537                                &geneve_link_ops, tb);
1538         if (IS_ERR(dev))
1539                 return dev;
1540
1541         err = geneve_configure(net, dev, &geneve_remote_unspec,
1542                                0, 0, 0, 0, htons(dst_port), true,
1543                                GENEVE_F_UDP_ZERO_CSUM6_RX);
1544         if (err)
1545                 goto err;
1546
1547         /* openvswitch users expect packet sizes to be unrestricted,
1548          * so set the largest MTU we can.
1549          */
1550         err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1551         if (err)
1552                 goto err;
1553
1554         return dev;
1555
1556  err:
1557         free_netdev(dev);
1558         return ERR_PTR(err);
1559 }
1560 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1561
1562 static __net_init int geneve_init_net(struct net *net)
1563 {
1564         struct geneve_net *gn = net_generic(net, geneve_net_id);
1565
1566         INIT_LIST_HEAD(&gn->geneve_list);
1567         INIT_LIST_HEAD(&gn->sock_list);
1568         return 0;
1569 }
1570
1571 static void __net_exit geneve_exit_net(struct net *net)
1572 {
1573         struct geneve_net *gn = net_generic(net, geneve_net_id);
1574         struct geneve_dev *geneve, *next;
1575         struct net_device *dev, *aux;
1576         LIST_HEAD(list);
1577
1578         rtnl_lock();
1579
1580         /* gather any geneve devices that were moved into this ns */
1581         for_each_netdev_safe(net, dev, aux)
1582                 if (dev->rtnl_link_ops == &geneve_link_ops)
1583                         unregister_netdevice_queue(dev, &list);
1584
1585         /* now gather any other geneve devices that were created in this ns */
1586         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1587                 /* If geneve->dev is in the same netns, it was already added
1588                  * to the list by the previous loop.
1589                  */
1590                 if (!net_eq(dev_net(geneve->dev), net))
1591                         unregister_netdevice_queue(geneve->dev, &list);
1592         }
1593
1594         /* unregister the devices gathered above */
1595         unregister_netdevice_many(&list);
1596         rtnl_unlock();
1597 }
1598
1599 static struct pernet_operations geneve_net_ops = {
1600         .init = geneve_init_net,
1601         .exit = geneve_exit_net,
1602         .id   = &geneve_net_id,
1603         .size = sizeof(struct geneve_net),
1604 };
1605
1606 static int __init geneve_init_module(void)
1607 {
1608         int rc;
1609
1610         rc = register_pernet_subsys(&geneve_net_ops);
1611         if (rc)
1612                 goto out1;
1613
1614         rc = rtnl_link_register(&geneve_link_ops);
1615         if (rc)
1616                 goto out2;
1617
1618         return 0;
1619 out2:
1620         unregister_pernet_subsys(&geneve_net_ops);
1621 out1:
1622         return rc;
1623 }
1624 late_initcall(geneve_init_module);
1625
1626 static void __exit geneve_cleanup_module(void)
1627 {
1628         rtnl_link_unregister(&geneve_link_ops);
1629         unregister_pernet_subsys(&geneve_net_ops);
1630 }
1631 module_exit(geneve_cleanup_module);
1632
1633 MODULE_LICENSE("GPL");
1634 MODULE_VERSION(GENEVE_NETDEV_VER);
1635 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1636 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1637 MODULE_ALIAS_RTNL_LINK("geneve");