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         udp_tunnel_gro_complete(skb, nhoff);
 508
 509         gh = (struct genevehdr *)(skb->data + nhoff);
 510         gh_len = geneve_hlen(gh);
 511         type = gh->proto_type;
 512
 513         rcu_read_lock();
 514         ptype = gro_find_complete_by_type(type);
 515         if (ptype)
 516                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 517
 518         rcu_read_unlock();
 519         return err;
 520 }
 521
 522 /* Create new listen socket if needed */
 523 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 524                                                 bool ipv6, u32 flags)
 525 {
 526         struct geneve_net *gn = net_generic(net, geneve_net_id);
 527         struct geneve_sock *gs;
 528         struct socket *sock;
 529         struct udp_tunnel_sock_cfg tunnel_cfg;
 530         int h;
 531
 532         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 533         if (!gs)
 534                 return ERR_PTR(-ENOMEM);
 535
 536         sock = geneve_create_sock(net, ipv6, port, flags);
 537         if (IS_ERR(sock)) {
 538                 kfree(gs);
 539                 return ERR_CAST(sock);
 540         }
 541
 542         gs->sock = sock;
 543         gs->refcnt = 1;
 544         for (h = 0; h < VNI_HASH_SIZE; ++h)
 545                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 546
 547         /* Initialize the geneve udp offloads structure */
 548         gs->udp_offloads.port = port;
 549         gs->udp_offloads.callbacks.gro_receive  = geneve_gro_receive;
 550         gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
 551         geneve_notify_add_rx_port(gs);
 552
 553         /* Mark socket as an encapsulation socket */
 554         tunnel_cfg.sk_user_data = gs;
 555         tunnel_cfg.encap_type = 1;
 556         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 557         tunnel_cfg.encap_destroy = NULL;
 558         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 559         list_add(&gs->list, &gn->sock_list);
 560         return gs;
 561 }
 562
 563 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 564 {
 565         struct net_device *dev;
 566         struct sock *sk = gs->sock->sk;
 567         struct net *net = sock_net(sk);
 568         sa_family_t sa_family = geneve_get_sk_family(gs);
 569         __be16 port = inet_sk(sk)->inet_sport;
 570
 571         rcu_read_lock();
 572         for_each_netdev_rcu(net, dev) {
 573                 if (dev->netdev_ops->ndo_del_geneve_port)
 574                         dev->netdev_ops->ndo_del_geneve_port(dev, sa_family,
 575                                                              port);
 576         }
 577
 578         rcu_read_unlock();
 579
 580         if (sa_family == AF_INET)
 581                 udp_del_offload(&gs->udp_offloads);
 582 }
 583
 584 static void __geneve_sock_release(struct geneve_sock *gs)
 585 {
 586         if (!gs || --gs->refcnt)
 587                 return;
 588
 589         list_del(&gs->list);
 590         geneve_notify_del_rx_port(gs);
 591         udp_tunnel_sock_release(gs->sock);
 592         kfree_rcu(gs, rcu);
 593 }
 594
 595 static void geneve_sock_release(struct geneve_dev *geneve)
 596 {
 597         __geneve_sock_release(geneve->sock4);
 598 #if IS_ENABLED(CONFIG_IPV6)
 599         __geneve_sock_release(geneve->sock6);
 600 #endif
 601 }
 602
 603 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 604                                             sa_family_t family,
 605                                             __be16 dst_port)
 606 {
 607         struct geneve_sock *gs;
 608
 609         list_for_each_entry(gs, &gn->sock_list, list) {
 610                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 611                     geneve_get_sk_family(gs) == family) {
 612                         return gs;
 613                 }
 614         }
 615         return NULL;
 616 }
 617
 618 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 619 {
 620         struct net *net = geneve->net;
 621         struct geneve_net *gn = net_generic(net, geneve_net_id);
 622         struct geneve_sock *gs;
 623         __u32 hash;
 624
 625         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
 626         if (gs) {
 627                 gs->refcnt++;
 628                 goto out;
 629         }
 630
 631         gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
 632         if (IS_ERR(gs))
 633                 return PTR_ERR(gs);
 634
 635 out:
 636         gs->collect_md = geneve->collect_md;
 637         gs->flags = geneve->flags;
 638 #if IS_ENABLED(CONFIG_IPV6)
 639         if (ipv6)
 640                 geneve->sock6 = gs;
 641         else
 642 #endif
 643                 geneve->sock4 = gs;
 644
 645         hash = geneve_net_vni_hash(geneve->vni);
 646         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 647         return 0;
 648 }
 649
 650 static int geneve_open(struct net_device *dev)
 651 {
 652         struct geneve_dev *geneve = netdev_priv(dev);
 653         bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
 654         bool metadata = geneve->collect_md;
 655         int ret = 0;
 656
 657         geneve->sock4 = NULL;
 658 #if IS_ENABLED(CONFIG_IPV6)
 659         geneve->sock6 = NULL;
 660         if (ipv6 || metadata)
 661                 ret = geneve_sock_add(geneve, true);
 662 #endif
 663         if (!ret && (!ipv6 || metadata))
 664                 ret = geneve_sock_add(geneve, false);
 665         if (ret < 0)
 666                 geneve_sock_release(geneve);
 667
 668         return ret;
 669 }
 670
 671 static int geneve_stop(struct net_device *dev)
 672 {
 673         struct geneve_dev *geneve = netdev_priv(dev);
 674
 675         if (!hlist_unhashed(&geneve->hlist))
 676                 hlist_del_rcu(&geneve->hlist);
 677         geneve_sock_release(geneve);
 678         return 0;
 679 }
 680
 681 static void geneve_build_header(struct genevehdr *geneveh,
 682                                 __be16 tun_flags, u8 vni[3],
 683                                 u8 options_len, u8 *options)
 684 {
 685         geneveh->ver = GENEVE_VER;
 686         geneveh->opt_len = options_len / 4;
 687         geneveh->oam = !!(tun_flags & TUNNEL_OAM);
 688         geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 689         geneveh->rsvd1 = 0;
 690         memcpy(geneveh->vni, vni, 3);
 691         geneveh->proto_type = htons(ETH_P_TEB);
 692         geneveh->rsvd2 = 0;
 693
 694         memcpy(geneveh->options, options, options_len);
 695 }
 696
 697 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 698                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 699                             u32 flags, bool xnet)
 700 {
 701         struct genevehdr *gnvh;
 702         int min_headroom;
 703         int err;
 704         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
 705
 706         skb_scrub_packet(skb, xnet);
 707
 708         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 709                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 710         err = skb_cow_head(skb, min_headroom);
 711         if (unlikely(err)) {
 712                 kfree_skb(skb);
 713                 goto free_rt;
 714         }
 715
 716         skb = udp_tunnel_handle_offloads(skb, udp_sum);
 717         if (IS_ERR(skb)) {
 718                 err = PTR_ERR(skb);
 719                 goto free_rt;
 720         }
 721
 722         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 723         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 724
 725         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 726         return 0;
 727
 728 free_rt:
 729         ip_rt_put(rt);
 730         return err;
 731 }
 732
 733 #if IS_ENABLED(CONFIG_IPV6)
 734 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 735                              __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 736                              u32 flags, bool xnet)
 737 {
 738         struct genevehdr *gnvh;
 739         int min_headroom;
 740         int err;
 741         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
 742
 743         skb_scrub_packet(skb, xnet);
 744
 745         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
 746                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
 747         err = skb_cow_head(skb, min_headroom);
 748         if (unlikely(err)) {
 749                 kfree_skb(skb);
 750                 goto free_dst;
 751         }
 752
 753         skb = udp_tunnel_handle_offloads(skb, udp_sum);
 754         if (IS_ERR(skb)) {
 755                 err = PTR_ERR(skb);
 756                 goto free_dst;
 757         }
 758
 759         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 760         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 761
 762         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 763         return 0;
 764
 765 free_dst:
 766         dst_release(dst);
 767         return err;
 768 }
 769 #endif
 770
 771 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 772                                        struct net_device *dev,
 773                                        struct flowi4 *fl4,
 774                                        struct ip_tunnel_info *info)
 775 {
 776         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 777         struct geneve_dev *geneve = netdev_priv(dev);
 778         struct dst_cache *dst_cache;
 779         struct rtable *rt = NULL;
 780         __u8 tos;
 781
 782         memset(fl4, 0, sizeof(*fl4));
 783         fl4->flowi4_mark = skb->mark;
 784         fl4->flowi4_proto = IPPROTO_UDP;
 785
 786         if (info) {
 787                 fl4->daddr = info->key.u.ipv4.dst;
 788                 fl4->saddr = info->key.u.ipv4.src;
 789                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 790                 dst_cache = &info->dst_cache;
 791         } else {
 792                 tos = geneve->tos;
 793                 if (tos == 1) {
 794                         const struct iphdr *iip = ip_hdr(skb);
 795
 796                         tos = ip_tunnel_get_dsfield(iip, skb);
 797                         use_cache = false;
 798                 }
 799
 800                 fl4->flowi4_tos = RT_TOS(tos);
 801                 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
 802                 dst_cache = &geneve->dst_cache;
 803         }
 804
 805         if (use_cache) {
 806                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 807                 if (rt)
 808                         return rt;
 809         }
 810
 811         rt = ip_route_output_key(geneve->net, fl4);
 812         if (IS_ERR(rt)) {
 813                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 814                 return ERR_PTR(-ENETUNREACH);
 815         }
 816         if (rt->dst.dev == dev) { /* is this necessary? */
 817                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 818                 ip_rt_put(rt);
 819                 return ERR_PTR(-ELOOP);
 820         }
 821         if (use_cache)
 822                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 823         return rt;
 824 }
 825
 826 #if IS_ENABLED(CONFIG_IPV6)
 827 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 828                                            struct net_device *dev,
 829                                            struct flowi6 *fl6,
 830                                            struct ip_tunnel_info *info)
 831 {
 832         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 833         struct geneve_dev *geneve = netdev_priv(dev);
 834         struct geneve_sock *gs6 = geneve->sock6;
 835         struct dst_entry *dst = NULL;
 836         struct dst_cache *dst_cache;
 837         __u8 prio;
 838
 839         memset(fl6, 0, sizeof(*fl6));
 840         fl6->flowi6_mark = skb->mark;
 841         fl6->flowi6_proto = IPPROTO_UDP;
 842
 843         if (info) {
 844                 fl6->daddr = info->key.u.ipv6.dst;
 845                 fl6->saddr = info->key.u.ipv6.src;
 846                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(info->key.tos),
 847                                                    info->key.label);
 848                 dst_cache = &info->dst_cache;
 849         } else {
 850                 prio = geneve->tos;
 851                 if (prio == 1) {
 852                         const struct iphdr *iip = ip_hdr(skb);
 853
 854                         prio = ip_tunnel_get_dsfield(iip, skb);
 855                         use_cache = false;
 856                 }
 857
 858                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 859                                                    geneve->label);
 860                 fl6->daddr = geneve->remote.sin6.sin6_addr;
 861                 dst_cache = &geneve->dst_cache;
 862         }
 863
 864         if (use_cache) {
 865                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 866                 if (dst)
 867                         return dst;
 868         }
 869
 870         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 871                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 872                 return ERR_PTR(-ENETUNREACH);
 873         }
 874         if (dst->dev == dev) { /* is this necessary? */
 875                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 876                 dst_release(dst);
 877                 return ERR_PTR(-ELOOP);
 878         }
 879
 880         if (use_cache)
 881                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 882         return dst;
 883 }
 884 #endif
 885
 886 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 887 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 888 {
 889 #ifdef __BIG_ENDIAN
 890         vni[0] = (__force __u8)(tun_id >> 16);
 891         vni[1] = (__force __u8)(tun_id >> 8);
 892         vni[2] = (__force __u8)tun_id;
 893 #else
 894         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 895         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 896         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 897 #endif
 898 }
 899
 900 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 901                                    struct ip_tunnel_info *info)
 902 {
 903         struct geneve_dev *geneve = netdev_priv(dev);
 904         struct geneve_sock *gs4 = geneve->sock4;
 905         struct rtable *rt = NULL;
 906         const struct iphdr *iip; /* interior IP header */
 907         int err = -EINVAL;
 908         struct flowi4 fl4;
 909         __u8 tos, ttl;
 910         __be16 sport;
 911         __be16 df;
 912         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 913         u32 flags = geneve->flags;
 914
 915         if (geneve->collect_md) {
 916                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 917                         netdev_dbg(dev, "no tunnel metadata\n");
 918                         goto tx_error;
 919                 }
 920                 if (info && ip_tunnel_info_af(info) != AF_INET)
 921                         goto tx_error;
 922         }
 923
 924         rt = geneve_get_v4_rt(skb, dev, &fl4, info);
 925         if (IS_ERR(rt)) {
 926                 err = PTR_ERR(rt);
 927                 goto tx_error;
 928         }
 929
 930         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 931         skb_reset_mac_header(skb);
 932
 933         iip = ip_hdr(skb);
 934
 935         if (info) {
 936                 const struct ip_tunnel_key *key = &info->key;
 937                 u8 *opts = NULL;
 938                 u8 vni[3];
 939
 940                 tunnel_id_to_vni(key->tun_id, vni);
 941                 if (info->options_len)
 942                         opts = ip_tunnel_info_opts(info);
 943
 944                 if (key->tun_flags & TUNNEL_CSUM)
 945                         flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
 946                 else
 947                         flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
 948
 949                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 950                                        info->options_len, opts, flags, xnet);
 951                 if (unlikely(err))
 952                         goto err;
 953
 954                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
 955                 ttl = key->ttl;
 956                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 957         } else {
 958                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 959                                        0, NULL, flags, xnet);
 960                 if (unlikely(err))
 961                         goto err;
 962
 963                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 964                 ttl = geneve->ttl;
 965                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 966                         ttl = 1;
 967                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 968                 df = 0;
 969         }
 970         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 971                             tos, ttl, df, sport, geneve->dst_port,
 972                             !net_eq(geneve->net, dev_net(geneve->dev)),
 973                             !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
 974
 975         return NETDEV_TX_OK;
 976
 977 tx_error:
 978         dev_kfree_skb(skb);
 979 err:
 980         if (err == -ELOOP)
 981                 dev->stats.collisions++;
 982         else if (err == -ENETUNREACH)
 983                 dev->stats.tx_carrier_errors++;
 984         else
 985                 dev->stats.tx_errors++;
 986         return NETDEV_TX_OK;
 987 }
 988
 989 #if IS_ENABLED(CONFIG_IPV6)
 990 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 991                                     struct ip_tunnel_info *info)
 992 {
 993         struct geneve_dev *geneve = netdev_priv(dev);
 994         struct geneve_sock *gs6 = geneve->sock6;
 995         struct dst_entry *dst = NULL;
 996         const struct iphdr *iip; /* interior IP header */
 997         int err = -EINVAL;
 998         struct flowi6 fl6;
 999         __u8 prio, ttl;
1000         __be16 sport;
1001         __be32 label;
1002         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
1003         u32 flags = geneve->flags;
1004
1005         if (geneve->collect_md) {
1006                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1007                         netdev_dbg(dev, "no tunnel metadata\n");
1008                         goto tx_error;
1009                 }
1010         }
1011
1012         dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1013         if (IS_ERR(dst)) {
1014                 err = PTR_ERR(dst);
1015                 goto tx_error;
1016         }
1017
1018         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1019         skb_reset_mac_header(skb);
1020
1021         iip = ip_hdr(skb);
1022
1023         if (info) {
1024                 const struct ip_tunnel_key *key = &info->key;
1025                 u8 *opts = NULL;
1026                 u8 vni[3];
1027
1028                 tunnel_id_to_vni(key->tun_id, vni);
1029                 if (info->options_len)
1030                         opts = ip_tunnel_info_opts(info);
1031
1032                 if (key->tun_flags & TUNNEL_CSUM)
1033                         flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
1034                 else
1035                         flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1036
1037                 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
1038                                         info->options_len, opts,
1039                                         flags, xnet);
1040                 if (unlikely(err))
1041                         goto err;
1042
1043                 prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
1044                 ttl = key->ttl;
1045                 label = info->key.label;
1046         } else {
1047                 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
1048                                         0, NULL, flags, xnet);
1049                 if (unlikely(err))
1050                         goto err;
1051
1052                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1053                                            iip, skb);
1054                 ttl = geneve->ttl;
1055                 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1056                         ttl = 1;
1057                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1058                 label = geneve->label;
1059         }
1060
1061         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1062                              &fl6.saddr, &fl6.daddr, prio, ttl, label,
1063                              sport, geneve->dst_port,
1064                              !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1065         return NETDEV_TX_OK;
1066
1067 tx_error:
1068         dev_kfree_skb(skb);
1069 err:
1070         if (err == -ELOOP)
1071                 dev->stats.collisions++;
1072         else if (err == -ENETUNREACH)
1073                 dev->stats.tx_carrier_errors++;
1074         else
1075                 dev->stats.tx_errors++;
1076         return NETDEV_TX_OK;
1077 }
1078 #endif
1079
1080 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1081 {
1082         struct geneve_dev *geneve = netdev_priv(dev);
1083         struct ip_tunnel_info *info = NULL;
1084
1085         if (geneve->collect_md)
1086                 info = skb_tunnel_info(skb);
1087
1088 #if IS_ENABLED(CONFIG_IPV6)
1089         if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1090             (!info && geneve->remote.sa.sa_family == AF_INET6))
1091                 return geneve6_xmit_skb(skb, dev, info);
1092 #endif
1093         return geneve_xmit_skb(skb, dev, info);
1094 }
1095
1096 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1097 {
1098         /* The max_mtu calculation does not take account of GENEVE
1099          * options, to avoid excluding potentially valid
1100          * configurations.
1101          */
1102         int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - sizeof(struct iphdr)
1103                 - dev->hard_header_len;
1104
1105         if (new_mtu < 68)
1106                 return -EINVAL;
1107
1108         if (new_mtu > max_mtu) {
1109                 if (strict)
1110                         return -EINVAL;
1111
1112                 new_mtu = max_mtu;
1113         }
1114
1115         dev->mtu = new_mtu;
1116         return 0;
1117 }
1118
1119 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1120 {
1121         return __geneve_change_mtu(dev, new_mtu, true);
1122 }
1123
1124 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1125 {
1126         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1127         struct geneve_dev *geneve = netdev_priv(dev);
1128         struct rtable *rt;
1129         struct flowi4 fl4;
1130 #if IS_ENABLED(CONFIG_IPV6)
1131         struct dst_entry *dst;
1132         struct flowi6 fl6;
1133 #endif
1134
1135         if (ip_tunnel_info_af(info) == AF_INET) {
1136                 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1137                 if (IS_ERR(rt))
1138                         return PTR_ERR(rt);
1139
1140                 ip_rt_put(rt);
1141                 info->key.u.ipv4.src = fl4.saddr;
1142 #if IS_ENABLED(CONFIG_IPV6)
1143         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1144                 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1145                 if (IS_ERR(dst))
1146                         return PTR_ERR(dst);
1147
1148                 dst_release(dst);
1149                 info->key.u.ipv6.src = fl6.saddr;
1150 #endif
1151         } else {
1152                 return -EINVAL;
1153         }
1154
1155         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1156                                              1, USHRT_MAX, true);
1157         info->key.tp_dst = geneve->dst_port;
1158         return 0;
1159 }
1160
1161 static const struct net_device_ops geneve_netdev_ops = {
1162         .ndo_init               = geneve_init,
1163         .ndo_uninit             = geneve_uninit,
1164         .ndo_open               = geneve_open,
1165         .ndo_stop               = geneve_stop,
1166         .ndo_start_xmit         = geneve_xmit,
1167         .ndo_get_stats64        = ip_tunnel_get_stats64,
1168         .ndo_change_mtu         = geneve_change_mtu,
1169         .ndo_validate_addr      = eth_validate_addr,
1170         .ndo_set_mac_address    = eth_mac_addr,
1171         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1172 };
1173
1174 static void geneve_get_drvinfo(struct net_device *dev,
1175                                struct ethtool_drvinfo *drvinfo)
1176 {
1177         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1178         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1179 }
1180
1181 static const struct ethtool_ops geneve_ethtool_ops = {
1182         .get_drvinfo    = geneve_get_drvinfo,
1183         .get_link       = ethtool_op_get_link,
1184 };
1185
1186 /* Info for udev, that this is a virtual tunnel endpoint */
1187 static struct device_type geneve_type = {
1188         .name = "geneve",
1189 };
1190
1191 /* Calls the ndo_add_geneve_port of the caller in order to
1192  * supply the listening GENEVE udp ports. Callers are expected
1193  * to implement the ndo_add_geneve_port.
1194  */
1195 void geneve_get_rx_port(struct net_device *dev)
1196 {
1197         struct net *net = dev_net(dev);
1198         struct geneve_net *gn = net_generic(net, geneve_net_id);
1199         struct geneve_sock *gs;
1200         sa_family_t sa_family;
1201         struct sock *sk;
1202         __be16 port;
1203
1204         rcu_read_lock();
1205         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1206                 sk = gs->sock->sk;
1207                 sa_family = sk->sk_family;
1208                 port = inet_sk(sk)->inet_sport;
1209                 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family, port);
1210         }
1211         rcu_read_unlock();
1212 }
1213 EXPORT_SYMBOL_GPL(geneve_get_rx_port);
1214
1215 /* Initialize the device structure. */
1216 static void geneve_setup(struct net_device *dev)
1217 {
1218         ether_setup(dev);
1219
1220         dev->netdev_ops = &geneve_netdev_ops;
1221         dev->ethtool_ops = &geneve_ethtool_ops;
1222         dev->destructor = free_netdev;
1223
1224         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1225
1226         dev->features    |= NETIF_F_LLTX;
1227         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1228         dev->features    |= NETIF_F_RXCSUM;
1229         dev->features    |= NETIF_F_GSO_SOFTWARE;
1230
1231         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1232         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1233
1234         netif_keep_dst(dev);
1235         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1236         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1237         eth_hw_addr_random(dev);
1238 }
1239
1240 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1241         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1242         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1243         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1244         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1245         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1246         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1247         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1248         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1249         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1250         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1251         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1252 };
1253
1254 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1255 {
1256         if (tb[IFLA_ADDRESS]) {
1257                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1258                         return -EINVAL;
1259
1260                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1261                         return -EADDRNOTAVAIL;
1262         }
1263
1264         if (!data)
1265                 return -EINVAL;
1266
1267         if (data[IFLA_GENEVE_ID]) {
1268                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1269
1270                 if (vni >= GENEVE_VID_MASK)
1271                         return -ERANGE;
1272         }
1273
1274         return 0;
1275 }
1276
1277 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1278                                           __be16 dst_port,
1279                                           union geneve_addr *remote,
1280                                           u8 vni[],
1281                                           bool *tun_on_same_port,
1282                                           bool *tun_collect_md)
1283 {
1284         struct geneve_dev *geneve, *t;
1285
1286         *tun_on_same_port = false;
1287         *tun_collect_md = false;
1288         t = NULL;
1289         list_for_each_entry(geneve, &gn->geneve_list, next) {
1290                 if (geneve->dst_port == dst_port) {
1291                         *tun_collect_md = geneve->collect_md;
1292                         *tun_on_same_port = true;
1293                 }
1294                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1295                     !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1296                     dst_port == geneve->dst_port)
1297                         t = geneve;
1298         }
1299         return t;
1300 }
1301
1302 static int geneve_configure(struct net *net, struct net_device *dev,
1303                             union geneve_addr *remote,
1304                             __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1305                             __be16 dst_port, bool metadata, u32 flags)
1306 {
1307         struct geneve_net *gn = net_generic(net, geneve_net_id);
1308         struct geneve_dev *t, *geneve = netdev_priv(dev);
1309         bool tun_collect_md, tun_on_same_port;
1310         int err, encap_len;
1311
1312         if (!remote)
1313                 return -EINVAL;
1314         if (metadata &&
1315             (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1316                 return -EINVAL;
1317
1318         geneve->net = net;
1319         geneve->dev = dev;
1320
1321         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1322         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1323         geneve->vni[2] =  vni & 0x000000ff;
1324
1325         if ((remote->sa.sa_family == AF_INET &&
1326              IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1327             (remote->sa.sa_family == AF_INET6 &&
1328              ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1329                 return -EINVAL;
1330         if (label && remote->sa.sa_family != AF_INET6)
1331                 return -EINVAL;
1332
1333         geneve->remote = *remote;
1334
1335         geneve->ttl = ttl;
1336         geneve->tos = tos;
1337         geneve->label = label;
1338         geneve->dst_port = dst_port;
1339         geneve->collect_md = metadata;
1340         geneve->flags = flags;
1341
1342         t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1343                             &tun_on_same_port, &tun_collect_md);
1344         if (t)
1345                 return -EBUSY;
1346
1347         /* make enough headroom for basic scenario */
1348         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1349         if (remote->sa.sa_family == AF_INET)
1350                 encap_len += sizeof(struct iphdr);
1351         else
1352                 encap_len += sizeof(struct ipv6hdr);
1353         dev->needed_headroom = encap_len + ETH_HLEN;
1354
1355         if (metadata) {
1356                 if (tun_on_same_port)
1357                         return -EPERM;
1358         } else {
1359                 if (tun_collect_md)
1360                         return -EPERM;
1361         }
1362
1363         dst_cache_reset(&geneve->dst_cache);
1364
1365         err = register_netdevice(dev);
1366         if (err)
1367                 return err;
1368
1369         list_add(&geneve->next, &gn->geneve_list);
1370         return 0;
1371 }
1372
1373 static int geneve_newlink(struct net *net, struct net_device *dev,
1374                           struct nlattr *tb[], struct nlattr *data[])
1375 {
1376         __be16 dst_port = htons(GENEVE_UDP_PORT);
1377         __u8 ttl = 0, tos = 0;
1378         bool metadata = false;
1379         union geneve_addr remote = geneve_remote_unspec;
1380         __be32 label = 0;
1381         __u32 vni = 0;
1382         u32 flags = 0;
1383
1384         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1385                 return -EINVAL;
1386
1387         if (data[IFLA_GENEVE_REMOTE]) {
1388                 remote.sa.sa_family = AF_INET;
1389                 remote.sin.sin_addr.s_addr =
1390                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1391         }
1392
1393         if (data[IFLA_GENEVE_REMOTE6]) {
1394                 if (!IS_ENABLED(CONFIG_IPV6))
1395                         return -EPFNOSUPPORT;
1396
1397                 remote.sa.sa_family = AF_INET6;
1398                 remote.sin6.sin6_addr =
1399                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1400
1401                 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1402                     IPV6_ADDR_LINKLOCAL) {
1403                         netdev_dbg(dev, "link-local remote is unsupported\n");
1404                         return -EINVAL;
1405                 }
1406         }
1407
1408         if (data[IFLA_GENEVE_ID])
1409                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1410
1411         if (data[IFLA_GENEVE_TTL])
1412                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1413
1414         if (data[IFLA_GENEVE_TOS])
1415                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1416
1417         if (data[IFLA_GENEVE_LABEL])
1418                 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1419                         IPV6_FLOWLABEL_MASK;
1420
1421         if (data[IFLA_GENEVE_PORT])
1422                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1423
1424         if (data[IFLA_GENEVE_COLLECT_METADATA])
1425                 metadata = true;
1426
1427         if (data[IFLA_GENEVE_UDP_CSUM] &&
1428             !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1429                 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1430
1431         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1432             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1433                 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1434
1435         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1436             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1437                 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1438
1439         return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1440                                 dst_port, metadata, flags);
1441 }
1442
1443 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1444 {
1445         struct geneve_dev *geneve = netdev_priv(dev);
1446
1447         list_del(&geneve->next);
1448         unregister_netdevice_queue(dev, head);
1449 }
1450
1451 static size_t geneve_get_size(const struct net_device *dev)
1452 {
1453         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1454                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1455                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1456                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1457                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1458                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1459                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1460                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1461                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1462                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1463                 0;
1464 }
1465
1466 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1467 {
1468         struct geneve_dev *geneve = netdev_priv(dev);
1469         __u32 vni;
1470
1471         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1472         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1473                 goto nla_put_failure;
1474
1475         if (geneve->remote.sa.sa_family == AF_INET) {
1476                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1477                                     geneve->remote.sin.sin_addr.s_addr))
1478                         goto nla_put_failure;
1479 #if IS_ENABLED(CONFIG_IPV6)
1480         } else {
1481                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1482                                      &geneve->remote.sin6.sin6_addr))
1483                         goto nla_put_failure;
1484 #endif
1485         }
1486
1487         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1488             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1489             nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1490                 goto nla_put_failure;
1491
1492         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1493                 goto nla_put_failure;
1494
1495         if (geneve->collect_md) {
1496                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1497                         goto nla_put_failure;
1498         }
1499
1500         if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1501                        !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1502             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1503                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1504             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1505                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1506                 goto nla_put_failure;
1507
1508         return 0;
1509
1510 nla_put_failure:
1511         return -EMSGSIZE;
1512 }
1513
1514 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1515         .kind           = "geneve",
1516         .maxtype        = IFLA_GENEVE_MAX,
1517         .policy         = geneve_policy,
1518         .priv_size      = sizeof(struct geneve_dev),
1519         .setup          = geneve_setup,
1520         .validate       = geneve_validate,
1521         .newlink        = geneve_newlink,
1522         .dellink        = geneve_dellink,
1523         .get_size       = geneve_get_size,
1524         .fill_info      = geneve_fill_info,
1525 };
1526
1527 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1528                                         u8 name_assign_type, u16 dst_port)
1529 {
1530         struct nlattr *tb[IFLA_MAX + 1];
1531         struct net_device *dev;
1532         int err;
1533
1534         memset(tb, 0, sizeof(tb));
1535         dev = rtnl_create_link(net, name, name_assign_type,
1536                                &geneve_link_ops, tb);
1537         if (IS_ERR(dev))
1538                 return dev;
1539
1540         err = geneve_configure(net, dev, &geneve_remote_unspec,
1541                                0, 0, 0, 0, htons(dst_port), true,
1542                                GENEVE_F_UDP_ZERO_CSUM6_RX);
1543         if (err)
1544                 goto err;
1545
1546         /* openvswitch users expect packet sizes to be unrestricted,
1547          * so set the largest MTU we can.
1548          */
1549         err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1550         if (err)
1551                 goto err;
1552
1553         return dev;
1554
1555  err:
1556         free_netdev(dev);
1557         return ERR_PTR(err);
1558 }
1559 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1560
1561 static __net_init int geneve_init_net(struct net *net)
1562 {
1563         struct geneve_net *gn = net_generic(net, geneve_net_id);
1564
1565         INIT_LIST_HEAD(&gn->geneve_list);
1566         INIT_LIST_HEAD(&gn->sock_list);
1567         return 0;
1568 }
1569
1570 static void __net_exit geneve_exit_net(struct net *net)
1571 {
1572         struct geneve_net *gn = net_generic(net, geneve_net_id);
1573         struct geneve_dev *geneve, *next;
1574         struct net_device *dev, *aux;
1575         LIST_HEAD(list);
1576
1577         rtnl_lock();
1578
1579         /* gather any geneve devices that were moved into this ns */
1580         for_each_netdev_safe(net, dev, aux)
1581                 if (dev->rtnl_link_ops == &geneve_link_ops)
1582                         unregister_netdevice_queue(dev, &list);
1583
1584         /* now gather any other geneve devices that were created in this ns */
1585         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1586                 /* If geneve->dev is in the same netns, it was already added
1587                  * to the list by the previous loop.
1588                  */
1589                 if (!net_eq(dev_net(geneve->dev), net))
1590                         unregister_netdevice_queue(geneve->dev, &list);
1591         }
1592
1593         /* unregister the devices gathered above */
1594         unregister_netdevice_many(&list);
1595         rtnl_unlock();
1596 }
1597
1598 static struct pernet_operations geneve_net_ops = {
1599         .init = geneve_init_net,
1600         .exit = geneve_exit_net,
1601         .id   = &geneve_net_id,
1602         .size = sizeof(struct geneve_net),
1603 };
1604
1605 static int __init geneve_init_module(void)
1606 {
1607         int rc;
1608
1609         rc = register_pernet_subsys(&geneve_net_ops);
1610         if (rc)
1611                 goto out1;
1612
1613         rc = rtnl_link_register(&geneve_link_ops);
1614         if (rc)
1615                 goto out2;
1616
1617         return 0;
1618 out2:
1619         unregister_pernet_subsys(&geneve_net_ops);
1620 out1:
1621         return rc;
1622 }
1623 late_initcall(geneve_init_module);
1624
1625 static void __exit geneve_cleanup_module(void)
1626 {
1627         rtnl_link_unregister(&geneve_link_ops);
1628         unregister_pernet_subsys(&geneve_net_ops);
1629 }
1630 module_exit(geneve_cleanup_module);
1631
1632 MODULE_LICENSE("GPL");
1633 MODULE_VERSION(GENEVE_NETDEV_VER);
1634 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1635 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1636 MODULE_ALIAS_RTNL_LINK("geneve");