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         __be16             dst_port;
  72         bool               collect_md;
  73         struct gro_cells   gro_cells;
  74         u32                flags;
  75         struct dst_cache   dst_cache;
  76 };
  77
  78 /* Geneve device flags */
  79 #define GENEVE_F_UDP_ZERO_CSUM_TX       BIT(0)
  80 #define GENEVE_F_UDP_ZERO_CSUM6_TX      BIT(1)
  81 #define GENEVE_F_UDP_ZERO_CSUM6_RX      BIT(2)
  82
  83 struct geneve_sock {
  84         bool                    collect_md;
  85         struct list_head        list;
  86         struct socket           *sock;
  87         struct rcu_head         rcu;
  88         int                     refcnt;
  89         struct udp_offload      udp_offloads;
  90         struct hlist_head       vni_list[VNI_HASH_SIZE];
  91         u32                     flags;
  92 };
  93
  94 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  95 {
  96         __u32 vnid;
  97
  98         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  99         return hash_32(vnid, VNI_HASH_BITS);
 100 }
 101
 102 static __be64 vni_to_tunnel_id(const __u8 *vni)
 103 {
 104 #ifdef __BIG_ENDIAN
 105         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
 106 #else
 107         return (__force __be64)(((__force u64)vni[0] << 40) |
 108                                 ((__force u64)vni[1] << 48) |
 109                                 ((__force u64)vni[2] << 56));
 110 #endif
 111 }
 112
 113 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 114 {
 115         return gs->sock->sk->sk_family;
 116 }
 117
 118 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 119                                         __be32 addr, u8 vni[])
 120 {
 121         struct hlist_head *vni_list_head;
 122         struct geneve_dev *geneve;
 123         __u32 hash;
 124
 125         /* Find the device for this VNI */
 126         hash = geneve_net_vni_hash(vni);
 127         vni_list_head = &gs->vni_list[hash];
 128         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 129                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 130                     addr == geneve->remote.sin.sin_addr.s_addr)
 131                         return geneve;
 132         }
 133         return NULL;
 134 }
 135
 136 #if IS_ENABLED(CONFIG_IPV6)
 137 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 138                                          struct in6_addr addr6, u8 vni[])
 139 {
 140         struct hlist_head *vni_list_head;
 141         struct geneve_dev *geneve;
 142         __u32 hash;
 143
 144         /* Find the device for this VNI */
 145         hash = geneve_net_vni_hash(vni);
 146         vni_list_head = &gs->vni_list[hash];
 147         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 148                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 149                     ipv6_addr_equal(&addr6, &geneve->remote.sin6.sin6_addr))
 150                         return geneve;
 151         }
 152         return NULL;
 153 }
 154 #endif
 155
 156 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 157 {
 158         return (struct genevehdr *)(udp_hdr(skb) + 1);
 159 }
 160
 161 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 162                                             struct sk_buff *skb)
 163 {
 164         u8 *vni;
 165         __be32 addr;
 166         static u8 zero_vni[3];
 167 #if IS_ENABLED(CONFIG_IPV6)
 168         static struct in6_addr zero_addr6;
 169 #endif
 170
 171         if (geneve_get_sk_family(gs) == AF_INET) {
 172                 struct iphdr *iph;
 173
 174                 iph = ip_hdr(skb); /* outer IP header... */
 175
 176                 if (gs->collect_md) {
 177                         vni = zero_vni;
 178                         addr = 0;
 179                 } else {
 180                         vni = geneve_hdr(skb)->vni;
 181                         addr = iph->saddr;
 182                 }
 183
 184                 return geneve_lookup(gs, addr, vni);
 185 #if IS_ENABLED(CONFIG_IPV6)
 186         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 187                 struct ipv6hdr *ip6h;
 188                 struct in6_addr addr6;
 189
 190                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 191
 192                 if (gs->collect_md) {
 193                         vni = zero_vni;
 194                         addr6 = zero_addr6;
 195                 } else {
 196                         vni = geneve_hdr(skb)->vni;
 197                         addr6 = ip6h->saddr;
 198                 }
 199
 200                 return geneve6_lookup(gs, addr6, vni);
 201 #endif
 202         }
 203         return NULL;
 204 }
 205
 206 /* geneve receive/decap routine */
 207 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 208                       struct sk_buff *skb)
 209 {
 210         struct genevehdr *gnvh = geneve_hdr(skb);
 211         struct metadata_dst *tun_dst = NULL;
 212         struct pcpu_sw_netstats *stats;
 213         int err = 0;
 214         void *oiph;
 215
 216         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 217                 __be16 flags;
 218
 219                 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
 220                         (gnvh->oam ? TUNNEL_OAM : 0) |
 221                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 222
 223                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 224                                          vni_to_tunnel_id(gnvh->vni),
 225                                          gnvh->opt_len * 4);
 226                 if (!tun_dst)
 227                         goto drop;
 228                 /* Update tunnel dst according to Geneve options. */
 229                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 230                                         gnvh->options, gnvh->opt_len * 4);
 231         } else {
 232                 /* Drop packets w/ critical options,
 233                  * since we don't support any...
 234                  */
 235                 if (gnvh->critical)
 236                         goto drop;
 237         }
 238
 239         skb_reset_mac_header(skb);
 240         skb->protocol = eth_type_trans(skb, geneve->dev);
 241         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 242
 243         if (tun_dst)
 244                 skb_dst_set(skb, &tun_dst->dst);
 245
 246         /* Ignore packet loops (and multicast echo) */
 247         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
 248                 goto drop;
 249
 250         oiph = skb_network_header(skb);
 251         skb_reset_network_header(skb);
 252
 253         if (geneve_get_sk_family(gs) == AF_INET)
 254                 err = IP_ECN_decapsulate(oiph, skb);
 255 #if IS_ENABLED(CONFIG_IPV6)
 256         else
 257                 err = IP6_ECN_decapsulate(oiph, skb);
 258 #endif
 259
 260         if (unlikely(err)) {
 261                 if (log_ecn_error) {
 262                         if (geneve_get_sk_family(gs) == AF_INET)
 263                                 net_info_ratelimited("non-ECT from %pI4 "
 264                                                      "with TOS=%#x\n",
 265                                                      &((struct iphdr *)oiph)->saddr,
 266                                                      ((struct iphdr *)oiph)->tos);
 267 #if IS_ENABLED(CONFIG_IPV6)
 268                         else
 269                                 net_info_ratelimited("non-ECT from %pI6\n",
 270                                                      &((struct ipv6hdr *)oiph)->saddr);
 271 #endif
 272                 }
 273                 if (err > 1) {
 274                         ++geneve->dev->stats.rx_frame_errors;
 275                         ++geneve->dev->stats.rx_errors;
 276                         goto drop;
 277                 }
 278         }
 279
 280         stats = this_cpu_ptr(geneve->dev->tstats);
 281         u64_stats_update_begin(&stats->syncp);
 282         stats->rx_packets++;
 283         stats->rx_bytes += skb->len;
 284         u64_stats_update_end(&stats->syncp);
 285
 286         gro_cells_receive(&geneve->gro_cells, skb);
 287         return;
 288 drop:
 289         /* Consume bad packet */
 290         kfree_skb(skb);
 291 }
 292
 293 /* Setup stats when device is created */
 294 static int geneve_init(struct net_device *dev)
 295 {
 296         struct geneve_dev *geneve = netdev_priv(dev);
 297         int err;
 298
 299         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 300         if (!dev->tstats)
 301                 return -ENOMEM;
 302
 303         err = gro_cells_init(&geneve->gro_cells, dev);
 304         if (err) {
 305                 free_percpu(dev->tstats);
 306                 return err;
 307         }
 308
 309         err = dst_cache_init(&geneve->dst_cache, GFP_KERNEL);
 310         if (err) {
 311                 free_percpu(dev->tstats);
 312                 gro_cells_destroy(&geneve->gro_cells);
 313                 return err;
 314         }
 315
 316         return 0;
 317 }
 318
 319 static void geneve_uninit(struct net_device *dev)
 320 {
 321         struct geneve_dev *geneve = netdev_priv(dev);
 322
 323         dst_cache_destroy(&geneve->dst_cache);
 324         gro_cells_destroy(&geneve->gro_cells);
 325         free_percpu(dev->tstats);
 326 }
 327
 328 /* Callback from net/ipv4/udp.c to receive packets */
 329 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 330 {
 331         struct genevehdr *geneveh;
 332         struct geneve_dev *geneve;
 333         struct geneve_sock *gs;
 334         int opts_len;
 335
 336         /* Need Geneve and inner Ethernet header to be present */
 337         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 338                 goto error;
 339
 340         /* Return packets with reserved bits set */
 341         geneveh = geneve_hdr(skb);
 342         if (unlikely(geneveh->ver != GENEVE_VER))
 343                 goto error;
 344
 345         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 346                 goto error;
 347
 348         gs = rcu_dereference_sk_user_data(sk);
 349         if (!gs)
 350                 goto drop;
 351
 352         geneve = geneve_lookup_skb(gs, skb);
 353         if (!geneve)
 354                 goto drop;
 355
 356         opts_len = geneveh->opt_len * 4;
 357         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 358                                  htons(ETH_P_TEB),
 359                                  !net_eq(geneve->net, dev_net(geneve->dev))))
 360                 goto drop;
 361
 362         geneve_rx(geneve, gs, skb);
 363         return 0;
 364
 365 drop:
 366         /* Consume bad packet */
 367         kfree_skb(skb);
 368         return 0;
 369
 370 error:
 371         /* Let the UDP layer deal with the skb */
 372         return 1;
 373 }
 374
 375 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 376                                          __be16 port, u32 flags)
 377 {
 378         struct socket *sock;
 379         struct udp_port_cfg udp_conf;
 380         int err;
 381
 382         memset(&udp_conf, 0, sizeof(udp_conf));
 383
 384         if (ipv6) {
 385                 udp_conf.family = AF_INET6;
 386                 udp_conf.ipv6_v6only = 1;
 387                 udp_conf.use_udp6_rx_checksums =
 388                     !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX);
 389         } else {
 390                 udp_conf.family = AF_INET;
 391                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 392         }
 393
 394         udp_conf.local_udp_port = port;
 395
 396         /* Open UDP socket */
 397         err = udp_sock_create(net, &udp_conf, &sock);
 398         if (err < 0)
 399                 return ERR_PTR(err);
 400
 401         return sock;
 402 }
 403
 404 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
 405 {
 406         struct net_device *dev;
 407         struct sock *sk = gs->sock->sk;
 408         struct net *net = sock_net(sk);
 409         sa_family_t sa_family = geneve_get_sk_family(gs);
 410         __be16 port = inet_sk(sk)->inet_sport;
 411         int err;
 412
 413         if (sa_family == AF_INET) {
 414                 err = udp_add_offload(sock_net(sk), &gs->udp_offloads);
 415                 if (err)
 416                         pr_warn("geneve: udp_add_offload failed with status %d\n",
 417                                 err);
 418         }
 419
 420         rcu_read_lock();
 421         for_each_netdev_rcu(net, dev) {
 422                 if (dev->netdev_ops->ndo_add_geneve_port)
 423                         dev->netdev_ops->ndo_add_geneve_port(dev, sa_family,
 424                                                              port);
 425         }
 426         rcu_read_unlock();
 427 }
 428
 429 static int geneve_hlen(struct genevehdr *gh)
 430 {
 431         return sizeof(*gh) + gh->opt_len * 4;
 432 }
 433
 434 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
 435                                            struct sk_buff *skb,
 436                                            struct udp_offload *uoff)
 437 {
 438         struct sk_buff *p, **pp = NULL;
 439         struct genevehdr *gh, *gh2;
 440         unsigned int hlen, gh_len, off_gnv;
 441         const struct packet_offload *ptype;
 442         __be16 type;
 443         int flush = 1;
 444
 445         off_gnv = skb_gro_offset(skb);
 446         hlen = off_gnv + sizeof(*gh);
 447         gh = skb_gro_header_fast(skb, off_gnv);
 448         if (skb_gro_header_hard(skb, hlen)) {
 449                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 450                 if (unlikely(!gh))
 451                         goto out;
 452         }
 453
 454         if (gh->ver != GENEVE_VER || gh->oam)
 455                 goto out;
 456         gh_len = geneve_hlen(gh);
 457
 458         hlen = off_gnv + gh_len;
 459         if (skb_gro_header_hard(skb, hlen)) {
 460                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 461                 if (unlikely(!gh))
 462                         goto out;
 463         }
 464
 465         flush = 0;
 466
 467         for (p = *head; p; p = p->next) {
 468                 if (!NAPI_GRO_CB(p)->same_flow)
 469                         continue;
 470
 471                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 472                 if (gh->opt_len != gh2->opt_len ||
 473                     memcmp(gh, gh2, gh_len)) {
 474                         NAPI_GRO_CB(p)->same_flow = 0;
 475                         continue;
 476                 }
 477         }
 478
 479         type = gh->proto_type;
 480
 481         rcu_read_lock();
 482         ptype = gro_find_receive_by_type(type);
 483         if (!ptype) {
 484                 flush = 1;
 485                 goto out_unlock;
 486         }
 487
 488         skb_gro_pull(skb, gh_len);
 489         skb_gro_postpull_rcsum(skb, gh, gh_len);
 490         pp = ptype->callbacks.gro_receive(head, skb);
 491
 492 out_unlock:
 493         rcu_read_unlock();
 494 out:
 495         NAPI_GRO_CB(skb)->flush |= flush;
 496
 497         return pp;
 498 }
 499
 500 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
 501                                struct udp_offload *uoff)
 502 {
 503         struct genevehdr *gh;
 504         struct packet_offload *ptype;
 505         __be16 type;
 506         int gh_len;
 507         int err = -ENOSYS;
 508
 509         udp_tunnel_gro_complete(skb, nhoff);
 510
 511         gh = (struct genevehdr *)(skb->data + nhoff);
 512         gh_len = geneve_hlen(gh);
 513         type = gh->proto_type;
 514
 515         rcu_read_lock();
 516         ptype = gro_find_complete_by_type(type);
 517         if (ptype)
 518                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 519
 520         rcu_read_unlock();
 521         return err;
 522 }
 523
 524 /* Create new listen socket if needed */
 525 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 526                                                 bool ipv6, u32 flags)
 527 {
 528         struct geneve_net *gn = net_generic(net, geneve_net_id);
 529         struct geneve_sock *gs;
 530         struct socket *sock;
 531         struct udp_tunnel_sock_cfg tunnel_cfg;
 532         int h;
 533
 534         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 535         if (!gs)
 536                 return ERR_PTR(-ENOMEM);
 537
 538         sock = geneve_create_sock(net, ipv6, port, flags);
 539         if (IS_ERR(sock)) {
 540                 kfree(gs);
 541                 return ERR_CAST(sock);
 542         }
 543
 544         gs->sock = sock;
 545         gs->refcnt = 1;
 546         for (h = 0; h < VNI_HASH_SIZE; ++h)
 547                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 548
 549         /* Initialize the geneve udp offloads structure */
 550         gs->udp_offloads.port = port;
 551         gs->udp_offloads.callbacks.gro_receive  = geneve_gro_receive;
 552         gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
 553         geneve_notify_add_rx_port(gs);
 554
 555         /* Mark socket as an encapsulation socket */
 556         tunnel_cfg.sk_user_data = gs;
 557         tunnel_cfg.encap_type = 1;
 558         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 559         tunnel_cfg.encap_destroy = NULL;
 560         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 561         list_add(&gs->list, &gn->sock_list);
 562         return gs;
 563 }
 564
 565 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 566 {
 567         struct net_device *dev;
 568         struct sock *sk = gs->sock->sk;
 569         struct net *net = sock_net(sk);
 570         sa_family_t sa_family = geneve_get_sk_family(gs);
 571         __be16 port = inet_sk(sk)->inet_sport;
 572
 573         rcu_read_lock();
 574         for_each_netdev_rcu(net, dev) {
 575                 if (dev->netdev_ops->ndo_del_geneve_port)
 576                         dev->netdev_ops->ndo_del_geneve_port(dev, sa_family,
 577                                                              port);
 578         }
 579
 580         rcu_read_unlock();
 581
 582         if (sa_family == AF_INET)
 583                 udp_del_offload(&gs->udp_offloads);
 584 }
 585
 586 static void __geneve_sock_release(struct geneve_sock *gs)
 587 {
 588         if (!gs || --gs->refcnt)
 589                 return;
 590
 591         list_del(&gs->list);
 592         geneve_notify_del_rx_port(gs);
 593         udp_tunnel_sock_release(gs->sock);
 594         kfree_rcu(gs, rcu);
 595 }
 596
 597 static void geneve_sock_release(struct geneve_dev *geneve)
 598 {
 599         __geneve_sock_release(geneve->sock4);
 600 #if IS_ENABLED(CONFIG_IPV6)
 601         __geneve_sock_release(geneve->sock6);
 602 #endif
 603 }
 604
 605 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 606                                             sa_family_t family,
 607                                             __be16 dst_port)
 608 {
 609         struct geneve_sock *gs;
 610
 611         list_for_each_entry(gs, &gn->sock_list, list) {
 612                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 613                     geneve_get_sk_family(gs) == family) {
 614                         return gs;
 615                 }
 616         }
 617         return NULL;
 618 }
 619
 620 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 621 {
 622         struct net *net = geneve->net;
 623         struct geneve_net *gn = net_generic(net, geneve_net_id);
 624         struct geneve_sock *gs;
 625         __u32 hash;
 626
 627         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
 628         if (gs) {
 629                 gs->refcnt++;
 630                 goto out;
 631         }
 632
 633         gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
 634         if (IS_ERR(gs))
 635                 return PTR_ERR(gs);
 636
 637 out:
 638         gs->collect_md = geneve->collect_md;
 639         gs->flags = geneve->flags;
 640 #if IS_ENABLED(CONFIG_IPV6)
 641         if (ipv6)
 642                 geneve->sock6 = gs;
 643         else
 644 #endif
 645                 geneve->sock4 = gs;
 646
 647         hash = geneve_net_vni_hash(geneve->vni);
 648         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 649         return 0;
 650 }
 651
 652 static int geneve_open(struct net_device *dev)
 653 {
 654         struct geneve_dev *geneve = netdev_priv(dev);
 655         bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
 656         bool metadata = geneve->collect_md;
 657         int ret = 0;
 658
 659         geneve->sock4 = NULL;
 660 #if IS_ENABLED(CONFIG_IPV6)
 661         geneve->sock6 = NULL;
 662         if (ipv6 || metadata)
 663                 ret = geneve_sock_add(geneve, true);
 664 #endif
 665         if (!ret && (!ipv6 || metadata))
 666                 ret = geneve_sock_add(geneve, false);
 667         if (ret < 0)
 668                 geneve_sock_release(geneve);
 669
 670         return ret;
 671 }
 672
 673 static int geneve_stop(struct net_device *dev)
 674 {
 675         struct geneve_dev *geneve = netdev_priv(dev);
 676
 677         if (!hlist_unhashed(&geneve->hlist))
 678                 hlist_del_rcu(&geneve->hlist);
 679         geneve_sock_release(geneve);
 680         return 0;
 681 }
 682
 683 static void geneve_build_header(struct genevehdr *geneveh,
 684                                 __be16 tun_flags, u8 vni[3],
 685                                 u8 options_len, u8 *options)
 686 {
 687         geneveh->ver = GENEVE_VER;
 688         geneveh->opt_len = options_len / 4;
 689         geneveh->oam = !!(tun_flags & TUNNEL_OAM);
 690         geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 691         geneveh->rsvd1 = 0;
 692         memcpy(geneveh->vni, vni, 3);
 693         geneveh->proto_type = htons(ETH_P_TEB);
 694         geneveh->rsvd2 = 0;
 695
 696         memcpy(geneveh->options, options, options_len);
 697 }
 698
 699 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 700                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 701                             u32 flags, bool xnet)
 702 {
 703         struct genevehdr *gnvh;
 704         int min_headroom;
 705         int err;
 706         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
 707
 708         skb_scrub_packet(skb, xnet);
 709
 710         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 711                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 712         err = skb_cow_head(skb, min_headroom);
 713         if (unlikely(err)) {
 714                 kfree_skb(skb);
 715                 goto free_rt;
 716         }
 717
 718         skb = udp_tunnel_handle_offloads(skb, udp_sum);
 719         if (IS_ERR(skb)) {
 720                 err = PTR_ERR(skb);
 721                 goto free_rt;
 722         }
 723
 724         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 725         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 726
 727         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 728         return 0;
 729
 730 free_rt:
 731         ip_rt_put(rt);
 732         return err;
 733 }
 734
 735 #if IS_ENABLED(CONFIG_IPV6)
 736 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 737                              __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 738                              u32 flags, bool xnet)
 739 {
 740         struct genevehdr *gnvh;
 741         int min_headroom;
 742         int err;
 743         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
 744
 745         skb_scrub_packet(skb, xnet);
 746
 747         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
 748                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
 749         err = skb_cow_head(skb, min_headroom);
 750         if (unlikely(err)) {
 751                 kfree_skb(skb);
 752                 goto free_dst;
 753         }
 754
 755         skb = udp_tunnel_handle_offloads(skb, udp_sum);
 756         if (IS_ERR(skb)) {
 757                 err = PTR_ERR(skb);
 758                 goto free_dst;
 759         }
 760
 761         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 762         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 763
 764         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 765         return 0;
 766
 767 free_dst:
 768         dst_release(dst);
 769         return err;
 770 }
 771 #endif
 772
 773 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 774                                        struct net_device *dev,
 775                                        struct flowi4 *fl4,
 776                                        struct ip_tunnel_info *info)
 777 {
 778         struct geneve_dev *geneve = netdev_priv(dev);
 779         struct dst_cache *dst_cache;
 780         struct rtable *rt = NULL;
 781         bool use_cache = true;
 782         __u8 tos;
 783
 784         memset(fl4, 0, sizeof(*fl4));
 785         fl4->flowi4_mark = skb->mark;
 786         fl4->flowi4_proto = IPPROTO_UDP;
 787
 788         if (info) {
 789                 fl4->daddr = info->key.u.ipv4.dst;
 790                 fl4->saddr = info->key.u.ipv4.src;
 791                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 792                 dst_cache = &info->dst_cache;
 793         } else {
 794                 tos = geneve->tos;
 795                 if (tos == 1) {
 796                         const struct iphdr *iip = ip_hdr(skb);
 797
 798                         tos = ip_tunnel_get_dsfield(iip, skb);
 799                         use_cache = false;
 800                 }
 801
 802                 fl4->flowi4_tos = RT_TOS(tos);
 803                 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
 804                 dst_cache = &geneve->dst_cache;
 805         }
 806
 807         use_cache = use_cache && !skb->mark;
 808         if (use_cache) {
 809                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 810                 if (rt)
 811                         return rt;
 812         }
 813
 814         rt = ip_route_output_key(geneve->net, fl4);
 815         if (IS_ERR(rt)) {
 816                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 817                 return ERR_PTR(-ENETUNREACH);
 818         }
 819         if (rt->dst.dev == dev) { /* is this necessary? */
 820                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 821                 ip_rt_put(rt);
 822                 return ERR_PTR(-ELOOP);
 823         }
 824         if (use_cache)
 825                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 826         return rt;
 827 }
 828
 829 #if IS_ENABLED(CONFIG_IPV6)
 830 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 831                                            struct net_device *dev,
 832                                            struct flowi6 *fl6,
 833                                            struct ip_tunnel_info *info)
 834 {
 835         struct geneve_dev *geneve = netdev_priv(dev);
 836         struct geneve_sock *gs6 = geneve->sock6;
 837         struct dst_entry *dst = NULL;
 838         struct dst_cache *dst_cache;
 839         bool use_cache = true;
 840         __u8 prio;
 841
 842         memset(fl6, 0, sizeof(*fl6));
 843         fl6->flowi6_mark = skb->mark;
 844         fl6->flowi6_proto = IPPROTO_UDP;
 845
 846         if (info) {
 847                 fl6->daddr = info->key.u.ipv6.dst;
 848                 fl6->saddr = info->key.u.ipv6.src;
 849                 fl6->flowi6_tos = RT_TOS(info->key.tos);
 850                 dst_cache = &info->dst_cache;
 851         } else {
 852                 prio = geneve->tos;
 853                 if (prio == 1) {
 854                         const struct iphdr *iip = ip_hdr(skb);
 855
 856                         prio = ip_tunnel_get_dsfield(iip, skb);
 857                         use_cache = false;
 858                 }
 859
 860                 fl6->flowi6_tos = RT_TOS(prio);
 861                 fl6->daddr = geneve->remote.sin6.sin6_addr;
 862                 dst_cache = &geneve->dst_cache;
 863         }
 864
 865         use_cache = use_cache && !skb->mark;
 866         if (use_cache) {
 867                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 868                 if (dst)
 869                         return dst;
 870         }
 871
 872         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 873                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 874                 return ERR_PTR(-ENETUNREACH);
 875         }
 876         if (dst->dev == dev) { /* is this necessary? */
 877                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 878                 dst_release(dst);
 879                 return ERR_PTR(-ELOOP);
 880         }
 881
 882         if (use_cache)
 883                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 884         return dst;
 885 }
 886 #endif
 887
 888 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 889 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 890 {
 891 #ifdef __BIG_ENDIAN
 892         vni[0] = (__force __u8)(tun_id >> 16);
 893         vni[1] = (__force __u8)(tun_id >> 8);
 894         vni[2] = (__force __u8)tun_id;
 895 #else
 896         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 897         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 898         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 899 #endif
 900 }
 901
 902 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 903                                    struct ip_tunnel_info *info)
 904 {
 905         struct geneve_dev *geneve = netdev_priv(dev);
 906         struct geneve_sock *gs4 = geneve->sock4;
 907         struct rtable *rt = NULL;
 908         const struct iphdr *iip; /* interior IP header */
 909         int err = -EINVAL;
 910         struct flowi4 fl4;
 911         __u8 tos, ttl;
 912         __be16 sport;
 913         __be16 df;
 914         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 915         u32 flags = geneve->flags;
 916
 917         if (geneve->collect_md) {
 918                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 919                         netdev_dbg(dev, "no tunnel metadata\n");
 920                         goto tx_error;
 921                 }
 922                 if (info && ip_tunnel_info_af(info) != AF_INET)
 923                         goto tx_error;
 924         }
 925
 926         rt = geneve_get_v4_rt(skb, dev, &fl4, info);
 927         if (IS_ERR(rt)) {
 928                 err = PTR_ERR(rt);
 929                 goto tx_error;
 930         }
 931
 932         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 933         skb_reset_mac_header(skb);
 934
 935         iip = ip_hdr(skb);
 936
 937         if (info) {
 938                 const struct ip_tunnel_key *key = &info->key;
 939                 u8 *opts = NULL;
 940                 u8 vni[3];
 941
 942                 tunnel_id_to_vni(key->tun_id, vni);
 943                 if (key->tun_flags & TUNNEL_GENEVE_OPT)
 944                         opts = ip_tunnel_info_opts(info);
 945
 946                 if (key->tun_flags & TUNNEL_CSUM)
 947                         flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
 948                 else
 949                         flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
 950
 951                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 952                                        info->options_len, opts, flags, xnet);
 953                 if (unlikely(err))
 954                         goto err;
 955
 956                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
 957                 ttl = key->ttl;
 958                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 959         } else {
 960                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 961                                        0, NULL, flags, xnet);
 962                 if (unlikely(err))
 963                         goto err;
 964
 965                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 966                 ttl = geneve->ttl;
 967                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 968                         ttl = 1;
 969                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 970                 df = 0;
 971         }
 972         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 973                             tos, ttl, df, sport, geneve->dst_port,
 974                             !net_eq(geneve->net, dev_net(geneve->dev)),
 975                             !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
 976
 977         return NETDEV_TX_OK;
 978
 979 tx_error:
 980         dev_kfree_skb(skb);
 981 err:
 982         if (err == -ELOOP)
 983                 dev->stats.collisions++;
 984         else if (err == -ENETUNREACH)
 985                 dev->stats.tx_carrier_errors++;
 986         else
 987                 dev->stats.tx_errors++;
 988         return NETDEV_TX_OK;
 989 }
 990
 991 #if IS_ENABLED(CONFIG_IPV6)
 992 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 993                                     struct ip_tunnel_info *info)
 994 {
 995         struct geneve_dev *geneve = netdev_priv(dev);
 996         struct geneve_sock *gs6 = geneve->sock6;
 997         struct dst_entry *dst = NULL;
 998         const struct iphdr *iip; /* interior IP header */
 999         int err = -EINVAL;
1000         struct flowi6 fl6;
1001         __u8 prio, ttl;
1002         __be16 sport;
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 (key->tun_flags & TUNNEL_GENEVE_OPT)
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         } 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(fl6.flowi6_tos, iip, skb);
1053                 ttl = geneve->ttl;
1054                 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1055                         ttl = 1;
1056                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1057         }
1058         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1059                              &fl6.saddr, &fl6.daddr, prio, ttl,
1060                              sport, geneve->dst_port,
1061                              !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1062         return NETDEV_TX_OK;
1063
1064 tx_error:
1065         dev_kfree_skb(skb);
1066 err:
1067         if (err == -ELOOP)
1068                 dev->stats.collisions++;
1069         else if (err == -ENETUNREACH)
1070                 dev->stats.tx_carrier_errors++;
1071         else
1072                 dev->stats.tx_errors++;
1073         return NETDEV_TX_OK;
1074 }
1075 #endif
1076
1077 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1078 {
1079         struct geneve_dev *geneve = netdev_priv(dev);
1080         struct ip_tunnel_info *info = NULL;
1081
1082         if (geneve->collect_md)
1083                 info = skb_tunnel_info(skb);
1084
1085 #if IS_ENABLED(CONFIG_IPV6)
1086         if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1087             (!info && geneve->remote.sa.sa_family == AF_INET6))
1088                 return geneve6_xmit_skb(skb, dev, info);
1089 #endif
1090         return geneve_xmit_skb(skb, dev, info);
1091 }
1092
1093 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1094 {
1095         /* The max_mtu calculation does not take account of GENEVE
1096          * options, to avoid excluding potentially valid
1097          * configurations.
1098          */
1099         int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - sizeof(struct iphdr)
1100                 - dev->hard_header_len;
1101
1102         if (new_mtu < 68)
1103                 return -EINVAL;
1104
1105         if (new_mtu > max_mtu) {
1106                 if (strict)
1107                         return -EINVAL;
1108
1109                 new_mtu = max_mtu;
1110         }
1111
1112         dev->mtu = new_mtu;
1113         return 0;
1114 }
1115
1116 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1117 {
1118         return __geneve_change_mtu(dev, new_mtu, true);
1119 }
1120
1121 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1122 {
1123         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1124         struct geneve_dev *geneve = netdev_priv(dev);
1125         struct rtable *rt;
1126         struct flowi4 fl4;
1127 #if IS_ENABLED(CONFIG_IPV6)
1128         struct dst_entry *dst;
1129         struct flowi6 fl6;
1130 #endif
1131
1132         if (ip_tunnel_info_af(info) == AF_INET) {
1133                 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1134                 if (IS_ERR(rt))
1135                         return PTR_ERR(rt);
1136
1137                 ip_rt_put(rt);
1138                 info->key.u.ipv4.src = fl4.saddr;
1139 #if IS_ENABLED(CONFIG_IPV6)
1140         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1141                 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1142                 if (IS_ERR(dst))
1143                         return PTR_ERR(dst);
1144
1145                 dst_release(dst);
1146                 info->key.u.ipv6.src = fl6.saddr;
1147 #endif
1148         } else {
1149                 return -EINVAL;
1150         }
1151
1152         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1153                                              1, USHRT_MAX, true);
1154         info->key.tp_dst = geneve->dst_port;
1155         return 0;
1156 }
1157
1158 static const struct net_device_ops geneve_netdev_ops = {
1159         .ndo_init               = geneve_init,
1160         .ndo_uninit             = geneve_uninit,
1161         .ndo_open               = geneve_open,
1162         .ndo_stop               = geneve_stop,
1163         .ndo_start_xmit         = geneve_xmit,
1164         .ndo_get_stats64        = ip_tunnel_get_stats64,
1165         .ndo_change_mtu         = geneve_change_mtu,
1166         .ndo_validate_addr      = eth_validate_addr,
1167         .ndo_set_mac_address    = eth_mac_addr,
1168         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1169 };
1170
1171 static void geneve_get_drvinfo(struct net_device *dev,
1172                                struct ethtool_drvinfo *drvinfo)
1173 {
1174         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1175         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1176 }
1177
1178 static const struct ethtool_ops geneve_ethtool_ops = {
1179         .get_drvinfo    = geneve_get_drvinfo,
1180         .get_link       = ethtool_op_get_link,
1181 };
1182
1183 /* Info for udev, that this is a virtual tunnel endpoint */
1184 static struct device_type geneve_type = {
1185         .name = "geneve",
1186 };
1187
1188 /* Calls the ndo_add_geneve_port of the caller in order to
1189  * supply the listening GENEVE udp ports. Callers are expected
1190  * to implement the ndo_add_geneve_port.
1191  */
1192 void geneve_get_rx_port(struct net_device *dev)
1193 {
1194         struct net *net = dev_net(dev);
1195         struct geneve_net *gn = net_generic(net, geneve_net_id);
1196         struct geneve_sock *gs;
1197         sa_family_t sa_family;
1198         struct sock *sk;
1199         __be16 port;
1200
1201         rcu_read_lock();
1202         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1203                 sk = gs->sock->sk;
1204                 sa_family = sk->sk_family;
1205                 port = inet_sk(sk)->inet_sport;
1206                 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family, port);
1207         }
1208         rcu_read_unlock();
1209 }
1210 EXPORT_SYMBOL_GPL(geneve_get_rx_port);
1211
1212 /* Initialize the device structure. */
1213 static void geneve_setup(struct net_device *dev)
1214 {
1215         ether_setup(dev);
1216
1217         dev->netdev_ops = &geneve_netdev_ops;
1218         dev->ethtool_ops = &geneve_ethtool_ops;
1219         dev->destructor = free_netdev;
1220
1221         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1222
1223         dev->features    |= NETIF_F_LLTX;
1224         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1225         dev->features    |= NETIF_F_RXCSUM;
1226         dev->features    |= NETIF_F_GSO_SOFTWARE;
1227
1228         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1229         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1230
1231         netif_keep_dst(dev);
1232         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1233         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1234         eth_hw_addr_random(dev);
1235 }
1236
1237 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1238         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1239         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1240         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1241         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1242         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1243         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1244         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1245         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1246         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1247         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1248 };
1249
1250 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1251 {
1252         if (tb[IFLA_ADDRESS]) {
1253                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1254                         return -EINVAL;
1255
1256                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1257                         return -EADDRNOTAVAIL;
1258         }
1259
1260         if (!data)
1261                 return -EINVAL;
1262
1263         if (data[IFLA_GENEVE_ID]) {
1264                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1265
1266                 if (vni >= GENEVE_VID_MASK)
1267                         return -ERANGE;
1268         }
1269
1270         return 0;
1271 }
1272
1273 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1274                                           __be16 dst_port,
1275                                           union geneve_addr *remote,
1276                                           u8 vni[],
1277                                           bool *tun_on_same_port,
1278                                           bool *tun_collect_md)
1279 {
1280         struct geneve_dev *geneve, *t;
1281
1282         *tun_on_same_port = false;
1283         *tun_collect_md = false;
1284         t = NULL;
1285         list_for_each_entry(geneve, &gn->geneve_list, next) {
1286                 if (geneve->dst_port == dst_port) {
1287                         *tun_collect_md = geneve->collect_md;
1288                         *tun_on_same_port = true;
1289                 }
1290                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1291                     !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1292                     dst_port == geneve->dst_port)
1293                         t = geneve;
1294         }
1295         return t;
1296 }
1297
1298 static int geneve_configure(struct net *net, struct net_device *dev,
1299                             union geneve_addr *remote,
1300                             __u32 vni, __u8 ttl, __u8 tos, __be16 dst_port,
1301                             bool metadata, u32 flags)
1302 {
1303         struct geneve_net *gn = net_generic(net, geneve_net_id);
1304         struct geneve_dev *t, *geneve = netdev_priv(dev);
1305         bool tun_collect_md, tun_on_same_port;
1306         int err, encap_len;
1307
1308         if (!remote)
1309                 return -EINVAL;
1310         if (metadata &&
1311             (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl))
1312                 return -EINVAL;
1313
1314         geneve->net = net;
1315         geneve->dev = dev;
1316
1317         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1318         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1319         geneve->vni[2] =  vni & 0x000000ff;
1320
1321         if ((remote->sa.sa_family == AF_INET &&
1322              IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1323             (remote->sa.sa_family == AF_INET6 &&
1324              ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1325                 return -EINVAL;
1326         geneve->remote = *remote;
1327
1328         geneve->ttl = ttl;
1329         geneve->tos = tos;
1330         geneve->dst_port = dst_port;
1331         geneve->collect_md = metadata;
1332         geneve->flags = flags;
1333
1334         t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1335                             &tun_on_same_port, &tun_collect_md);
1336         if (t)
1337                 return -EBUSY;
1338
1339         /* make enough headroom for basic scenario */
1340         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1341         if (remote->sa.sa_family == AF_INET)
1342                 encap_len += sizeof(struct iphdr);
1343         else
1344                 encap_len += sizeof(struct ipv6hdr);
1345         dev->needed_headroom = encap_len + ETH_HLEN;
1346
1347         if (metadata) {
1348                 if (tun_on_same_port)
1349                         return -EPERM;
1350         } else {
1351                 if (tun_collect_md)
1352                         return -EPERM;
1353         }
1354
1355         dst_cache_reset(&geneve->dst_cache);
1356
1357         err = register_netdevice(dev);
1358         if (err)
1359                 return err;
1360
1361         list_add(&geneve->next, &gn->geneve_list);
1362         return 0;
1363 }
1364
1365 static int geneve_newlink(struct net *net, struct net_device *dev,
1366                           struct nlattr *tb[], struct nlattr *data[])
1367 {
1368         __be16 dst_port = htons(GENEVE_UDP_PORT);
1369         __u8 ttl = 0, tos = 0;
1370         bool metadata = false;
1371         union geneve_addr remote = geneve_remote_unspec;
1372         __u32 vni = 0;
1373         u32 flags = 0;
1374
1375         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1376                 return -EINVAL;
1377
1378         if (data[IFLA_GENEVE_REMOTE]) {
1379                 remote.sa.sa_family = AF_INET;
1380                 remote.sin.sin_addr.s_addr =
1381                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1382         }
1383
1384         if (data[IFLA_GENEVE_REMOTE6]) {
1385                 if (!IS_ENABLED(CONFIG_IPV6))
1386                         return -EPFNOSUPPORT;
1387
1388                 remote.sa.sa_family = AF_INET6;
1389                 remote.sin6.sin6_addr =
1390                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1391
1392                 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1393                     IPV6_ADDR_LINKLOCAL) {
1394                         netdev_dbg(dev, "link-local remote is unsupported\n");
1395                         return -EINVAL;
1396                 }
1397         }
1398
1399         if (data[IFLA_GENEVE_ID])
1400                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1401
1402         if (data[IFLA_GENEVE_TTL])
1403                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1404
1405         if (data[IFLA_GENEVE_TOS])
1406                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1407
1408         if (data[IFLA_GENEVE_PORT])
1409                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1410
1411         if (data[IFLA_GENEVE_COLLECT_METADATA])
1412                 metadata = true;
1413
1414         if (data[IFLA_GENEVE_UDP_CSUM] &&
1415             !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1416                 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1417
1418         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1419             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1420                 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1421
1422         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1423             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1424                 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1425
1426         return geneve_configure(net, dev, &remote, vni, ttl, tos, dst_port,
1427                                 metadata, flags);
1428 }
1429
1430 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1431 {
1432         struct geneve_dev *geneve = netdev_priv(dev);
1433
1434         list_del(&geneve->next);
1435         unregister_netdevice_queue(dev, head);
1436 }
1437
1438 static size_t geneve_get_size(const struct net_device *dev)
1439 {
1440         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1441                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1442                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1443                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1444                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1445                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1446                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1447                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1448                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1449                 0;
1450 }
1451
1452 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1453 {
1454         struct geneve_dev *geneve = netdev_priv(dev);
1455         __u32 vni;
1456
1457         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1458         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1459                 goto nla_put_failure;
1460
1461         if (geneve->remote.sa.sa_family == AF_INET) {
1462                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1463                                     geneve->remote.sin.sin_addr.s_addr))
1464                         goto nla_put_failure;
1465 #if IS_ENABLED(CONFIG_IPV6)
1466         } else {
1467                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1468                                      &geneve->remote.sin6.sin6_addr))
1469                         goto nla_put_failure;
1470 #endif
1471         }
1472
1473         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1474             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos))
1475                 goto nla_put_failure;
1476
1477         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1478                 goto nla_put_failure;
1479
1480         if (geneve->collect_md) {
1481                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1482                         goto nla_put_failure;
1483         }
1484
1485         if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1486                        !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1487             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1488                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1489             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1490                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1491                 goto nla_put_failure;
1492
1493         return 0;
1494
1495 nla_put_failure:
1496         return -EMSGSIZE;
1497 }
1498
1499 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1500         .kind           = "geneve",
1501         .maxtype        = IFLA_GENEVE_MAX,
1502         .policy         = geneve_policy,
1503         .priv_size      = sizeof(struct geneve_dev),
1504         .setup          = geneve_setup,
1505         .validate       = geneve_validate,
1506         .newlink        = geneve_newlink,
1507         .dellink        = geneve_dellink,
1508         .get_size       = geneve_get_size,
1509         .fill_info      = geneve_fill_info,
1510 };
1511
1512 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1513                                         u8 name_assign_type, u16 dst_port)
1514 {
1515         struct nlattr *tb[IFLA_MAX + 1];
1516         struct net_device *dev;
1517         int err;
1518
1519         memset(tb, 0, sizeof(tb));
1520         dev = rtnl_create_link(net, name, name_assign_type,
1521                                &geneve_link_ops, tb);
1522         if (IS_ERR(dev))
1523                 return dev;
1524
1525         err = geneve_configure(net, dev, &geneve_remote_unspec,
1526                                0, 0, 0, htons(dst_port), true,
1527                                GENEVE_F_UDP_ZERO_CSUM6_RX);
1528         if (err)
1529                 goto err;
1530
1531         /* openvswitch users expect packet sizes to be unrestricted,
1532          * so set the largest MTU we can.
1533          */
1534         err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1535         if (err)
1536                 goto err;
1537
1538         return dev;
1539
1540  err:
1541         free_netdev(dev);
1542         return ERR_PTR(err);
1543 }
1544 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1545
1546 static __net_init int geneve_init_net(struct net *net)
1547 {
1548         struct geneve_net *gn = net_generic(net, geneve_net_id);
1549
1550         INIT_LIST_HEAD(&gn->geneve_list);
1551         INIT_LIST_HEAD(&gn->sock_list);
1552         return 0;
1553 }
1554
1555 static void __net_exit geneve_exit_net(struct net *net)
1556 {
1557         struct geneve_net *gn = net_generic(net, geneve_net_id);
1558         struct geneve_dev *geneve, *next;
1559         struct net_device *dev, *aux;
1560         LIST_HEAD(list);
1561
1562         rtnl_lock();
1563
1564         /* gather any geneve devices that were moved into this ns */
1565         for_each_netdev_safe(net, dev, aux)
1566                 if (dev->rtnl_link_ops == &geneve_link_ops)
1567                         unregister_netdevice_queue(dev, &list);
1568
1569         /* now gather any other geneve devices that were created in this ns */
1570         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1571                 /* If geneve->dev is in the same netns, it was already added
1572                  * to the list by the previous loop.
1573                  */
1574                 if (!net_eq(dev_net(geneve->dev), net))
1575                         unregister_netdevice_queue(geneve->dev, &list);
1576         }
1577
1578         /* unregister the devices gathered above */
1579         unregister_netdevice_many(&list);
1580         rtnl_unlock();
1581 }
1582
1583 static struct pernet_operations geneve_net_ops = {
1584         .init = geneve_init_net,
1585         .exit = geneve_exit_net,
1586         .id   = &geneve_net_id,
1587         .size = sizeof(struct geneve_net),
1588 };
1589
1590 static int __init geneve_init_module(void)
1591 {
1592         int rc;
1593
1594         rc = register_pernet_subsys(&geneve_net_ops);
1595         if (rc)
1596                 goto out1;
1597
1598         rc = rtnl_link_register(&geneve_link_ops);
1599         if (rc)
1600                 goto out2;
1601
1602         return 0;
1603 out2:
1604         unregister_pernet_subsys(&geneve_net_ops);
1605 out1:
1606         return rc;
1607 }
1608 late_initcall(geneve_init_module);
1609
1610 static void __exit geneve_cleanup_module(void)
1611 {
1612         rtnl_link_unregister(&geneve_link_ops);
1613         unregister_pernet_subsys(&geneve_net_ops);
1614 }
1615 module_exit(geneve_cleanup_module);
1616
1617 MODULE_LICENSE("GPL");
1618 MODULE_VERSION(GENEVE_NETDEV_VER);
1619 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1620 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1621 MODULE_ALIAS_RTNL_LINK("geneve");