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 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 drop;
 339
 340         /* Return packets with reserved bits set */
 341         geneveh = geneve_hdr(skb);
 342         if (unlikely(geneveh->ver != GENEVE_VER))
 343                 goto drop;
 344
 345         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 346                 goto drop;
 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
 371 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 372                                          __be16 port, u32 flags)
 373 {
 374         struct socket *sock;
 375         struct udp_port_cfg udp_conf;
 376         int err;
 377
 378         memset(&udp_conf, 0, sizeof(udp_conf));
 379
 380         if (ipv6) {
 381                 udp_conf.family = AF_INET6;
 382                 udp_conf.ipv6_v6only = 1;
 383                 udp_conf.use_udp6_rx_checksums =
 384                     !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX);
 385         } else {
 386                 udp_conf.family = AF_INET;
 387                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 388         }
 389
 390         udp_conf.local_udp_port = port;
 391
 392         /* Open UDP socket */
 393         err = udp_sock_create(net, &udp_conf, &sock);
 394         if (err < 0)
 395                 return ERR_PTR(err);
 396
 397         return sock;
 398 }
 399
 400 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
 401 {
 402         struct net_device *dev;
 403         struct sock *sk = gs->sock->sk;
 404         struct net *net = sock_net(sk);
 405         sa_family_t sa_family = geneve_get_sk_family(gs);
 406         __be16 port = inet_sk(sk)->inet_sport;
 407
 408         rcu_read_lock();
 409         for_each_netdev_rcu(net, dev) {
 410                 if (dev->netdev_ops->ndo_add_geneve_port)
 411                         dev->netdev_ops->ndo_add_geneve_port(dev, sa_family,
 412                                                              port);
 413         }
 414         rcu_read_unlock();
 415 }
 416
 417 static int geneve_hlen(struct genevehdr *gh)
 418 {
 419         return sizeof(*gh) + gh->opt_len * 4;
 420 }
 421
 422 static struct sk_buff **geneve_gro_receive(struct sock *sk,
 423                                            struct sk_buff **head,
 424                                            struct sk_buff *skb)
 425 {
 426         struct sk_buff *p, **pp = NULL;
 427         struct genevehdr *gh, *gh2;
 428         unsigned int hlen, gh_len, off_gnv;
 429         const struct packet_offload *ptype;
 430         __be16 type;
 431         int flush = 1;
 432
 433         off_gnv = skb_gro_offset(skb);
 434         hlen = off_gnv + sizeof(*gh);
 435         gh = skb_gro_header_fast(skb, off_gnv);
 436         if (skb_gro_header_hard(skb, hlen)) {
 437                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 438                 if (unlikely(!gh))
 439                         goto out;
 440         }
 441
 442         if (gh->ver != GENEVE_VER || gh->oam)
 443                 goto out;
 444         gh_len = geneve_hlen(gh);
 445
 446         hlen = off_gnv + gh_len;
 447         if (skb_gro_header_hard(skb, hlen)) {
 448                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 449                 if (unlikely(!gh))
 450                         goto out;
 451         }
 452
 453         for (p = *head; p; p = p->next) {
 454                 if (!NAPI_GRO_CB(p)->same_flow)
 455                         continue;
 456
 457                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 458                 if (gh->opt_len != gh2->opt_len ||
 459                     memcmp(gh, gh2, gh_len)) {
 460                         NAPI_GRO_CB(p)->same_flow = 0;
 461                         continue;
 462                 }
 463         }
 464
 465         type = gh->proto_type;
 466
 467         rcu_read_lock();
 468         ptype = gro_find_receive_by_type(type);
 469         if (!ptype)
 470                 goto out_unlock;
 471
 472         skb_gro_pull(skb, gh_len);
 473         skb_gro_postpull_rcsum(skb, gh, gh_len);
 474         pp = ptype->callbacks.gro_receive(head, skb);
 475         flush = 0;
 476
 477 out_unlock:
 478         rcu_read_unlock();
 479 out:
 480         NAPI_GRO_CB(skb)->flush |= flush;
 481
 482         return pp;
 483 }
 484
 485 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 486                                int nhoff)
 487 {
 488         struct genevehdr *gh;
 489         struct packet_offload *ptype;
 490         __be16 type;
 491         int gh_len;
 492         int err = -ENOSYS;
 493
 494         gh = (struct genevehdr *)(skb->data + nhoff);
 495         gh_len = geneve_hlen(gh);
 496         type = gh->proto_type;
 497
 498         rcu_read_lock();
 499         ptype = gro_find_complete_by_type(type);
 500         if (ptype)
 501                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 502
 503         rcu_read_unlock();
 504
 505         skb_set_inner_mac_header(skb, nhoff + gh_len);
 506
 507         return err;
 508 }
 509
 510 /* Create new listen socket if needed */
 511 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 512                                                 bool ipv6, u32 flags)
 513 {
 514         struct geneve_net *gn = net_generic(net, geneve_net_id);
 515         struct geneve_sock *gs;
 516         struct socket *sock;
 517         struct udp_tunnel_sock_cfg tunnel_cfg;
 518         int h;
 519
 520         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 521         if (!gs)
 522                 return ERR_PTR(-ENOMEM);
 523
 524         sock = geneve_create_sock(net, ipv6, port, flags);
 525         if (IS_ERR(sock)) {
 526                 kfree(gs);
 527                 return ERR_CAST(sock);
 528         }
 529
 530         gs->sock = sock;
 531         gs->refcnt = 1;
 532         for (h = 0; h < VNI_HASH_SIZE; ++h)
 533                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 534
 535         /* Initialize the geneve udp offloads structure */
 536         geneve_notify_add_rx_port(gs);
 537
 538         /* Mark socket as an encapsulation socket */
 539         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 540         tunnel_cfg.sk_user_data = gs;
 541         tunnel_cfg.encap_type = 1;
 542         tunnel_cfg.gro_receive = geneve_gro_receive;
 543         tunnel_cfg.gro_complete = geneve_gro_complete;
 544         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 545         tunnel_cfg.encap_destroy = NULL;
 546         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 547         list_add(&gs->list, &gn->sock_list);
 548         return gs;
 549 }
 550
 551 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 552 {
 553         struct net_device *dev;
 554         struct sock *sk = gs->sock->sk;
 555         struct net *net = sock_net(sk);
 556         sa_family_t sa_family = geneve_get_sk_family(gs);
 557         __be16 port = inet_sk(sk)->inet_sport;
 558
 559         rcu_read_lock();
 560         for_each_netdev_rcu(net, dev) {
 561                 if (dev->netdev_ops->ndo_del_geneve_port)
 562                         dev->netdev_ops->ndo_del_geneve_port(dev, sa_family,
 563                                                              port);
 564         }
 565
 566         rcu_read_unlock();
 567 }
 568
 569 static void __geneve_sock_release(struct geneve_sock *gs)
 570 {
 571         if (!gs || --gs->refcnt)
 572                 return;
 573
 574         list_del(&gs->list);
 575         geneve_notify_del_rx_port(gs);
 576         udp_tunnel_sock_release(gs->sock);
 577         kfree_rcu(gs, rcu);
 578 }
 579
 580 static void geneve_sock_release(struct geneve_dev *geneve)
 581 {
 582         __geneve_sock_release(geneve->sock4);
 583 #if IS_ENABLED(CONFIG_IPV6)
 584         __geneve_sock_release(geneve->sock6);
 585 #endif
 586 }
 587
 588 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 589                                             sa_family_t family,
 590                                             __be16 dst_port)
 591 {
 592         struct geneve_sock *gs;
 593
 594         list_for_each_entry(gs, &gn->sock_list, list) {
 595                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 596                     geneve_get_sk_family(gs) == family) {
 597                         return gs;
 598                 }
 599         }
 600         return NULL;
 601 }
 602
 603 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 604 {
 605         struct net *net = geneve->net;
 606         struct geneve_net *gn = net_generic(net, geneve_net_id);
 607         struct geneve_sock *gs;
 608         __u32 hash;
 609
 610         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
 611         if (gs) {
 612                 gs->refcnt++;
 613                 goto out;
 614         }
 615
 616         gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
 617         if (IS_ERR(gs))
 618                 return PTR_ERR(gs);
 619
 620 out:
 621         gs->collect_md = geneve->collect_md;
 622         gs->flags = geneve->flags;
 623 #if IS_ENABLED(CONFIG_IPV6)
 624         if (ipv6)
 625                 geneve->sock6 = gs;
 626         else
 627 #endif
 628                 geneve->sock4 = gs;
 629
 630         hash = geneve_net_vni_hash(geneve->vni);
 631         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 632         return 0;
 633 }
 634
 635 static int geneve_open(struct net_device *dev)
 636 {
 637         struct geneve_dev *geneve = netdev_priv(dev);
 638         bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
 639         bool metadata = geneve->collect_md;
 640         int ret = 0;
 641
 642         geneve->sock4 = NULL;
 643 #if IS_ENABLED(CONFIG_IPV6)
 644         geneve->sock6 = NULL;
 645         if (ipv6 || metadata)
 646                 ret = geneve_sock_add(geneve, true);
 647 #endif
 648         if (!ret && (!ipv6 || metadata))
 649                 ret = geneve_sock_add(geneve, false);
 650         if (ret < 0)
 651                 geneve_sock_release(geneve);
 652
 653         return ret;
 654 }
 655
 656 static int geneve_stop(struct net_device *dev)
 657 {
 658         struct geneve_dev *geneve = netdev_priv(dev);
 659
 660         if (!hlist_unhashed(&geneve->hlist))
 661                 hlist_del_rcu(&geneve->hlist);
 662         geneve_sock_release(geneve);
 663         return 0;
 664 }
 665
 666 static void geneve_build_header(struct genevehdr *geneveh,
 667                                 __be16 tun_flags, u8 vni[3],
 668                                 u8 options_len, u8 *options)
 669 {
 670         geneveh->ver = GENEVE_VER;
 671         geneveh->opt_len = options_len / 4;
 672         geneveh->oam = !!(tun_flags & TUNNEL_OAM);
 673         geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 674         geneveh->rsvd1 = 0;
 675         memcpy(geneveh->vni, vni, 3);
 676         geneveh->proto_type = htons(ETH_P_TEB);
 677         geneveh->rsvd2 = 0;
 678
 679         memcpy(geneveh->options, options, options_len);
 680 }
 681
 682 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 683                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 684                             u32 flags, bool xnet)
 685 {
 686         struct genevehdr *gnvh;
 687         int min_headroom;
 688         int err;
 689         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
 690
 691         skb_scrub_packet(skb, xnet);
 692
 693         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 694                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 695         err = skb_cow_head(skb, min_headroom);
 696         if (unlikely(err))
 697                 goto free_rt;
 698
 699         err = udp_tunnel_handle_offloads(skb, udp_sum);
 700         if (err)
 701                 goto free_rt;
 702
 703         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 704         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 705
 706         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 707         return 0;
 708
 709 free_rt:
 710         ip_rt_put(rt);
 711         return err;
 712 }
 713
 714 #if IS_ENABLED(CONFIG_IPV6)
 715 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 716                              __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 717                              u32 flags, bool xnet)
 718 {
 719         struct genevehdr *gnvh;
 720         int min_headroom;
 721         int err;
 722         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
 723
 724         skb_scrub_packet(skb, xnet);
 725
 726         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
 727                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
 728         err = skb_cow_head(skb, min_headroom);
 729         if (unlikely(err))
 730                 goto free_dst;
 731
 732         err = udp_tunnel_handle_offloads(skb, udp_sum);
 733         if (err)
 734                 goto free_dst;
 735
 736         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 737         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 738
 739         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 740         return 0;
 741
 742 free_dst:
 743         dst_release(dst);
 744         return err;
 745 }
 746 #endif
 747
 748 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 749                                        struct net_device *dev,
 750                                        struct flowi4 *fl4,
 751                                        struct ip_tunnel_info *info)
 752 {
 753         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 754         struct geneve_dev *geneve = netdev_priv(dev);
 755         struct dst_cache *dst_cache;
 756         struct rtable *rt = NULL;
 757         __u8 tos;
 758
 759         memset(fl4, 0, sizeof(*fl4));
 760         fl4->flowi4_mark = skb->mark;
 761         fl4->flowi4_proto = IPPROTO_UDP;
 762
 763         if (info) {
 764                 fl4->daddr = info->key.u.ipv4.dst;
 765                 fl4->saddr = info->key.u.ipv4.src;
 766                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 767                 dst_cache = &info->dst_cache;
 768         } else {
 769                 tos = geneve->tos;
 770                 if (tos == 1) {
 771                         const struct iphdr *iip = ip_hdr(skb);
 772
 773                         tos = ip_tunnel_get_dsfield(iip, skb);
 774                         use_cache = false;
 775                 }
 776
 777                 fl4->flowi4_tos = RT_TOS(tos);
 778                 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
 779                 dst_cache = &geneve->dst_cache;
 780         }
 781
 782         if (use_cache) {
 783                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 784                 if (rt)
 785                         return rt;
 786         }
 787
 788         rt = ip_route_output_key(geneve->net, fl4);
 789         if (IS_ERR(rt)) {
 790                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 791                 return ERR_PTR(-ENETUNREACH);
 792         }
 793         if (rt->dst.dev == dev) { /* is this necessary? */
 794                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 795                 ip_rt_put(rt);
 796                 return ERR_PTR(-ELOOP);
 797         }
 798         if (use_cache)
 799                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 800         return rt;
 801 }
 802
 803 #if IS_ENABLED(CONFIG_IPV6)
 804 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 805                                            struct net_device *dev,
 806                                            struct flowi6 *fl6,
 807                                            struct ip_tunnel_info *info)
 808 {
 809         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 810         struct geneve_dev *geneve = netdev_priv(dev);
 811         struct geneve_sock *gs6 = geneve->sock6;
 812         struct dst_entry *dst = NULL;
 813         struct dst_cache *dst_cache;
 814         __u8 prio;
 815
 816         memset(fl6, 0, sizeof(*fl6));
 817         fl6->flowi6_mark = skb->mark;
 818         fl6->flowi6_proto = IPPROTO_UDP;
 819
 820         if (info) {
 821                 fl6->daddr = info->key.u.ipv6.dst;
 822                 fl6->saddr = info->key.u.ipv6.src;
 823                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(info->key.tos),
 824                                                    info->key.label);
 825                 dst_cache = &info->dst_cache;
 826         } else {
 827                 prio = geneve->tos;
 828                 if (prio == 1) {
 829                         const struct iphdr *iip = ip_hdr(skb);
 830
 831                         prio = ip_tunnel_get_dsfield(iip, skb);
 832                         use_cache = false;
 833                 }
 834
 835                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 836                                                    geneve->label);
 837                 fl6->daddr = geneve->remote.sin6.sin6_addr;
 838                 dst_cache = &geneve->dst_cache;
 839         }
 840
 841         if (use_cache) {
 842                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 843                 if (dst)
 844                         return dst;
 845         }
 846
 847         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 848                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 849                 return ERR_PTR(-ENETUNREACH);
 850         }
 851         if (dst->dev == dev) { /* is this necessary? */
 852                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 853                 dst_release(dst);
 854                 return ERR_PTR(-ELOOP);
 855         }
 856
 857         if (use_cache)
 858                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 859         return dst;
 860 }
 861 #endif
 862
 863 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 864 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 865 {
 866 #ifdef __BIG_ENDIAN
 867         vni[0] = (__force __u8)(tun_id >> 16);
 868         vni[1] = (__force __u8)(tun_id >> 8);
 869         vni[2] = (__force __u8)tun_id;
 870 #else
 871         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 872         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 873         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 874 #endif
 875 }
 876
 877 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 878                                    struct ip_tunnel_info *info)
 879 {
 880         struct geneve_dev *geneve = netdev_priv(dev);
 881         struct geneve_sock *gs4 = geneve->sock4;
 882         struct rtable *rt = NULL;
 883         const struct iphdr *iip; /* interior IP header */
 884         int err = -EINVAL;
 885         struct flowi4 fl4;
 886         __u8 tos, ttl;
 887         __be16 sport;
 888         __be16 df;
 889         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 890         u32 flags = geneve->flags;
 891
 892         if (geneve->collect_md) {
 893                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 894                         netdev_dbg(dev, "no tunnel metadata\n");
 895                         goto tx_error;
 896                 }
 897                 if (info && ip_tunnel_info_af(info) != AF_INET)
 898                         goto tx_error;
 899         }
 900
 901         rt = geneve_get_v4_rt(skb, dev, &fl4, info);
 902         if (IS_ERR(rt)) {
 903                 err = PTR_ERR(rt);
 904                 goto tx_error;
 905         }
 906
 907         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 908         skb_reset_mac_header(skb);
 909
 910         iip = ip_hdr(skb);
 911
 912         if (info) {
 913                 const struct ip_tunnel_key *key = &info->key;
 914                 u8 *opts = NULL;
 915                 u8 vni[3];
 916
 917                 tunnel_id_to_vni(key->tun_id, vni);
 918                 if (info->options_len)
 919                         opts = ip_tunnel_info_opts(info);
 920
 921                 if (key->tun_flags & TUNNEL_CSUM)
 922                         flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
 923                 else
 924                         flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
 925
 926                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 927                                        info->options_len, opts, flags, xnet);
 928                 if (unlikely(err))
 929                         goto tx_error;
 930
 931                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
 932                 ttl = key->ttl;
 933                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 934         } else {
 935                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 936                                        0, NULL, flags, xnet);
 937                 if (unlikely(err))
 938                         goto tx_error;
 939
 940                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 941                 ttl = geneve->ttl;
 942                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 943                         ttl = 1;
 944                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 945                 df = 0;
 946         }
 947         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 948                             tos, ttl, df, sport, geneve->dst_port,
 949                             !net_eq(geneve->net, dev_net(geneve->dev)),
 950                             !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
 951
 952         return NETDEV_TX_OK;
 953
 954 tx_error:
 955         dev_kfree_skb(skb);
 956
 957         if (err == -ELOOP)
 958                 dev->stats.collisions++;
 959         else if (err == -ENETUNREACH)
 960                 dev->stats.tx_carrier_errors++;
 961
 962         dev->stats.tx_errors++;
 963         return NETDEV_TX_OK;
 964 }
 965
 966 #if IS_ENABLED(CONFIG_IPV6)
 967 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 968                                     struct ip_tunnel_info *info)
 969 {
 970         struct geneve_dev *geneve = netdev_priv(dev);
 971         struct geneve_sock *gs6 = geneve->sock6;
 972         struct dst_entry *dst = NULL;
 973         const struct iphdr *iip; /* interior IP header */
 974         int err = -EINVAL;
 975         struct flowi6 fl6;
 976         __u8 prio, ttl;
 977         __be16 sport;
 978         __be32 label;
 979         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 980         u32 flags = geneve->flags;
 981
 982         if (geneve->collect_md) {
 983                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 984                         netdev_dbg(dev, "no tunnel metadata\n");
 985                         goto tx_error;
 986                 }
 987         }
 988
 989         dst = geneve_get_v6_dst(skb, dev, &fl6, info);
 990         if (IS_ERR(dst)) {
 991                 err = PTR_ERR(dst);
 992                 goto tx_error;
 993         }
 994
 995         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 996         skb_reset_mac_header(skb);
 997
 998         iip = ip_hdr(skb);
 999
1000         if (info) {
1001                 const struct ip_tunnel_key *key = &info->key;
1002                 u8 *opts = NULL;
1003                 u8 vni[3];
1004
1005                 tunnel_id_to_vni(key->tun_id, vni);
1006                 if (info->options_len)
1007                         opts = ip_tunnel_info_opts(info);
1008
1009                 if (key->tun_flags & TUNNEL_CSUM)
1010                         flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
1011                 else
1012                         flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1013
1014                 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
1015                                         info->options_len, opts,
1016                                         flags, xnet);
1017                 if (unlikely(err))
1018                         goto tx_error;
1019
1020                 prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
1021                 ttl = key->ttl;
1022                 label = info->key.label;
1023         } else {
1024                 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
1025                                         0, NULL, flags, xnet);
1026                 if (unlikely(err))
1027                         goto tx_error;
1028
1029                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1030                                            iip, skb);
1031                 ttl = geneve->ttl;
1032                 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1033                         ttl = 1;
1034                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1035                 label = geneve->label;
1036         }
1037
1038         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1039                              &fl6.saddr, &fl6.daddr, prio, ttl, label,
1040                              sport, geneve->dst_port,
1041                              !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1042         return NETDEV_TX_OK;
1043
1044 tx_error:
1045         dev_kfree_skb(skb);
1046
1047         if (err == -ELOOP)
1048                 dev->stats.collisions++;
1049         else if (err == -ENETUNREACH)
1050                 dev->stats.tx_carrier_errors++;
1051
1052         dev->stats.tx_errors++;
1053         return NETDEV_TX_OK;
1054 }
1055 #endif
1056
1057 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1058 {
1059         struct geneve_dev *geneve = netdev_priv(dev);
1060         struct ip_tunnel_info *info = NULL;
1061
1062         if (geneve->collect_md)
1063                 info = skb_tunnel_info(skb);
1064
1065 #if IS_ENABLED(CONFIG_IPV6)
1066         if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1067             (!info && geneve->remote.sa.sa_family == AF_INET6))
1068                 return geneve6_xmit_skb(skb, dev, info);
1069 #endif
1070         return geneve_xmit_skb(skb, dev, info);
1071 }
1072
1073 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1074 {
1075         struct geneve_dev *geneve = netdev_priv(dev);
1076         /* The max_mtu calculation does not take account of GENEVE
1077          * options, to avoid excluding potentially valid
1078          * configurations.
1079          */
1080         int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1081
1082         if (geneve->remote.sa.sa_family == AF_INET6)
1083                 max_mtu -= sizeof(struct ipv6hdr);
1084         else
1085                 max_mtu -= sizeof(struct iphdr);
1086
1087         if (new_mtu < 68)
1088                 return -EINVAL;
1089
1090         if (new_mtu > max_mtu) {
1091                 if (strict)
1092                         return -EINVAL;
1093
1094                 new_mtu = max_mtu;
1095         }
1096
1097         dev->mtu = new_mtu;
1098         return 0;
1099 }
1100
1101 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1102 {
1103         return __geneve_change_mtu(dev, new_mtu, true);
1104 }
1105
1106 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1107 {
1108         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1109         struct geneve_dev *geneve = netdev_priv(dev);
1110         struct rtable *rt;
1111         struct flowi4 fl4;
1112 #if IS_ENABLED(CONFIG_IPV6)
1113         struct dst_entry *dst;
1114         struct flowi6 fl6;
1115 #endif
1116
1117         if (ip_tunnel_info_af(info) == AF_INET) {
1118                 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1119                 if (IS_ERR(rt))
1120                         return PTR_ERR(rt);
1121
1122                 ip_rt_put(rt);
1123                 info->key.u.ipv4.src = fl4.saddr;
1124 #if IS_ENABLED(CONFIG_IPV6)
1125         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1126                 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1127                 if (IS_ERR(dst))
1128                         return PTR_ERR(dst);
1129
1130                 dst_release(dst);
1131                 info->key.u.ipv6.src = fl6.saddr;
1132 #endif
1133         } else {
1134                 return -EINVAL;
1135         }
1136
1137         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1138                                              1, USHRT_MAX, true);
1139         info->key.tp_dst = geneve->dst_port;
1140         return 0;
1141 }
1142
1143 static const struct net_device_ops geneve_netdev_ops = {
1144         .ndo_init               = geneve_init,
1145         .ndo_uninit             = geneve_uninit,
1146         .ndo_open               = geneve_open,
1147         .ndo_stop               = geneve_stop,
1148         .ndo_start_xmit         = geneve_xmit,
1149         .ndo_get_stats64        = ip_tunnel_get_stats64,
1150         .ndo_change_mtu         = geneve_change_mtu,
1151         .ndo_validate_addr      = eth_validate_addr,
1152         .ndo_set_mac_address    = eth_mac_addr,
1153         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1154 };
1155
1156 static void geneve_get_drvinfo(struct net_device *dev,
1157                                struct ethtool_drvinfo *drvinfo)
1158 {
1159         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1160         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1161 }
1162
1163 static const struct ethtool_ops geneve_ethtool_ops = {
1164         .get_drvinfo    = geneve_get_drvinfo,
1165         .get_link       = ethtool_op_get_link,
1166 };
1167
1168 /* Info for udev, that this is a virtual tunnel endpoint */
1169 static struct device_type geneve_type = {
1170         .name = "geneve",
1171 };
1172
1173 /* Calls the ndo_add_geneve_port of the caller in order to
1174  * supply the listening GENEVE udp ports. Callers are expected
1175  * to implement the ndo_add_geneve_port.
1176  */
1177 static void geneve_push_rx_ports(struct net_device *dev)
1178 {
1179         struct net *net = dev_net(dev);
1180         struct geneve_net *gn = net_generic(net, geneve_net_id);
1181         struct geneve_sock *gs;
1182         sa_family_t sa_family;
1183         struct sock *sk;
1184         __be16 port;
1185
1186         if (!dev->netdev_ops->ndo_add_geneve_port)
1187                 return;
1188
1189         rcu_read_lock();
1190         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1191                 sk = gs->sock->sk;
1192                 sa_family = sk->sk_family;
1193                 port = inet_sk(sk)->inet_sport;
1194                 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family, port);
1195         }
1196         rcu_read_unlock();
1197 }
1198
1199 /* Initialize the device structure. */
1200 static void geneve_setup(struct net_device *dev)
1201 {
1202         ether_setup(dev);
1203
1204         dev->netdev_ops = &geneve_netdev_ops;
1205         dev->ethtool_ops = &geneve_ethtool_ops;
1206         dev->destructor = free_netdev;
1207
1208         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1209
1210         dev->features    |= NETIF_F_LLTX;
1211         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1212         dev->features    |= NETIF_F_RXCSUM;
1213         dev->features    |= NETIF_F_GSO_SOFTWARE;
1214
1215         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1216         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1217
1218         netif_keep_dst(dev);
1219         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1220         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1221         eth_hw_addr_random(dev);
1222 }
1223
1224 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1225         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1226         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1227         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1228         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1229         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1230         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1231         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1232         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1233         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1234         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1235         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1236 };
1237
1238 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1239 {
1240         if (tb[IFLA_ADDRESS]) {
1241                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1242                         return -EINVAL;
1243
1244                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1245                         return -EADDRNOTAVAIL;
1246         }
1247
1248         if (!data)
1249                 return -EINVAL;
1250
1251         if (data[IFLA_GENEVE_ID]) {
1252                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1253
1254                 if (vni >= GENEVE_VID_MASK)
1255                         return -ERANGE;
1256         }
1257
1258         return 0;
1259 }
1260
1261 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1262                                           __be16 dst_port,
1263                                           union geneve_addr *remote,
1264                                           u8 vni[],
1265                                           bool *tun_on_same_port,
1266                                           bool *tun_collect_md)
1267 {
1268         struct geneve_dev *geneve, *t;
1269
1270         *tun_on_same_port = false;
1271         *tun_collect_md = false;
1272         t = NULL;
1273         list_for_each_entry(geneve, &gn->geneve_list, next) {
1274                 if (geneve->dst_port == dst_port) {
1275                         *tun_collect_md = geneve->collect_md;
1276                         *tun_on_same_port = true;
1277                 }
1278                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1279                     !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1280                     dst_port == geneve->dst_port)
1281                         t = geneve;
1282         }
1283         return t;
1284 }
1285
1286 static int geneve_configure(struct net *net, struct net_device *dev,
1287                             union geneve_addr *remote,
1288                             __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1289                             __be16 dst_port, bool metadata, u32 flags)
1290 {
1291         struct geneve_net *gn = net_generic(net, geneve_net_id);
1292         struct geneve_dev *t, *geneve = netdev_priv(dev);
1293         bool tun_collect_md, tun_on_same_port;
1294         int err, encap_len;
1295
1296         if (!remote)
1297                 return -EINVAL;
1298         if (metadata &&
1299             (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1300                 return -EINVAL;
1301
1302         geneve->net = net;
1303         geneve->dev = dev;
1304
1305         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1306         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1307         geneve->vni[2] =  vni & 0x000000ff;
1308
1309         if ((remote->sa.sa_family == AF_INET &&
1310              IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1311             (remote->sa.sa_family == AF_INET6 &&
1312              ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1313                 return -EINVAL;
1314         if (label && remote->sa.sa_family != AF_INET6)
1315                 return -EINVAL;
1316
1317         geneve->remote = *remote;
1318
1319         geneve->ttl = ttl;
1320         geneve->tos = tos;
1321         geneve->label = label;
1322         geneve->dst_port = dst_port;
1323         geneve->collect_md = metadata;
1324         geneve->flags = flags;
1325
1326         t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1327                             &tun_on_same_port, &tun_collect_md);
1328         if (t)
1329                 return -EBUSY;
1330
1331         /* make enough headroom for basic scenario */
1332         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1333         if (remote->sa.sa_family == AF_INET)
1334                 encap_len += sizeof(struct iphdr);
1335         else
1336                 encap_len += sizeof(struct ipv6hdr);
1337         dev->needed_headroom = encap_len + ETH_HLEN;
1338
1339         if (metadata) {
1340                 if (tun_on_same_port)
1341                         return -EPERM;
1342         } else {
1343                 if (tun_collect_md)
1344                         return -EPERM;
1345         }
1346
1347         dst_cache_reset(&geneve->dst_cache);
1348
1349         err = register_netdevice(dev);
1350         if (err)
1351                 return err;
1352
1353         list_add(&geneve->next, &gn->geneve_list);
1354         return 0;
1355 }
1356
1357 static int geneve_newlink(struct net *net, struct net_device *dev,
1358                           struct nlattr *tb[], struct nlattr *data[])
1359 {
1360         __be16 dst_port = htons(GENEVE_UDP_PORT);
1361         __u8 ttl = 0, tos = 0;
1362         bool metadata = false;
1363         union geneve_addr remote = geneve_remote_unspec;
1364         __be32 label = 0;
1365         __u32 vni = 0;
1366         u32 flags = 0;
1367
1368         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1369                 return -EINVAL;
1370
1371         if (data[IFLA_GENEVE_REMOTE]) {
1372                 remote.sa.sa_family = AF_INET;
1373                 remote.sin.sin_addr.s_addr =
1374                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1375         }
1376
1377         if (data[IFLA_GENEVE_REMOTE6]) {
1378                 if (!IS_ENABLED(CONFIG_IPV6))
1379                         return -EPFNOSUPPORT;
1380
1381                 remote.sa.sa_family = AF_INET6;
1382                 remote.sin6.sin6_addr =
1383                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1384
1385                 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1386                     IPV6_ADDR_LINKLOCAL) {
1387                         netdev_dbg(dev, "link-local remote is unsupported\n");
1388                         return -EINVAL;
1389                 }
1390         }
1391
1392         if (data[IFLA_GENEVE_ID])
1393                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1394
1395         if (data[IFLA_GENEVE_TTL])
1396                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1397
1398         if (data[IFLA_GENEVE_TOS])
1399                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1400
1401         if (data[IFLA_GENEVE_LABEL])
1402                 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1403                         IPV6_FLOWLABEL_MASK;
1404
1405         if (data[IFLA_GENEVE_PORT])
1406                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1407
1408         if (data[IFLA_GENEVE_COLLECT_METADATA])
1409                 metadata = true;
1410
1411         if (data[IFLA_GENEVE_UDP_CSUM] &&
1412             !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1413                 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1414
1415         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1416             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1417                 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1418
1419         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1420             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1421                 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1422
1423         return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1424                                 dst_port, metadata, flags);
1425 }
1426
1427 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1428 {
1429         struct geneve_dev *geneve = netdev_priv(dev);
1430
1431         list_del(&geneve->next);
1432         unregister_netdevice_queue(dev, head);
1433 }
1434
1435 static size_t geneve_get_size(const struct net_device *dev)
1436 {
1437         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1438                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1439                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1440                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1441                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1442                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1443                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1444                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1445                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1446                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1447                 0;
1448 }
1449
1450 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1451 {
1452         struct geneve_dev *geneve = netdev_priv(dev);
1453         __u32 vni;
1454
1455         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1456         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1457                 goto nla_put_failure;
1458
1459         if (geneve->remote.sa.sa_family == AF_INET) {
1460                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1461                                     geneve->remote.sin.sin_addr.s_addr))
1462                         goto nla_put_failure;
1463 #if IS_ENABLED(CONFIG_IPV6)
1464         } else {
1465                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1466                                      &geneve->remote.sin6.sin6_addr))
1467                         goto nla_put_failure;
1468 #endif
1469         }
1470
1471         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1472             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1473             nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1474                 goto nla_put_failure;
1475
1476         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1477                 goto nla_put_failure;
1478
1479         if (geneve->collect_md) {
1480                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1481                         goto nla_put_failure;
1482         }
1483
1484         if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1485                        !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1486             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1487                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1488             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1489                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1490                 goto nla_put_failure;
1491
1492         return 0;
1493
1494 nla_put_failure:
1495         return -EMSGSIZE;
1496 }
1497
1498 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1499         .kind           = "geneve",
1500         .maxtype        = IFLA_GENEVE_MAX,
1501         .policy         = geneve_policy,
1502         .priv_size      = sizeof(struct geneve_dev),
1503         .setup          = geneve_setup,
1504         .validate       = geneve_validate,
1505         .newlink        = geneve_newlink,
1506         .dellink        = geneve_dellink,
1507         .get_size       = geneve_get_size,
1508         .fill_info      = geneve_fill_info,
1509 };
1510
1511 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1512                                         u8 name_assign_type, u16 dst_port)
1513 {
1514         struct nlattr *tb[IFLA_MAX + 1];
1515         struct net_device *dev;
1516         LIST_HEAD(list_kill);
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, 0, htons(dst_port), true,
1527                                GENEVE_F_UDP_ZERO_CSUM6_RX);
1528         if (err) {
1529                 free_netdev(dev);
1530                 return ERR_PTR(err);
1531         }
1532
1533         /* openvswitch users expect packet sizes to be unrestricted,
1534          * so set the largest MTU we can.
1535          */
1536         err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1537         if (err)
1538                 goto err;
1539
1540         err = rtnl_configure_link(dev, NULL);
1541         if (err < 0)
1542                 goto err;
1543
1544         return dev;
1545
1546  err:
1547         geneve_dellink(dev, &list_kill);
1548         unregister_netdevice_many(&list_kill);
1549         return ERR_PTR(err);
1550 }
1551 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1552
1553 static int geneve_netdevice_event(struct notifier_block *unused,
1554                                   unsigned long event, void *ptr)
1555 {
1556         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1557
1558         if (event == NETDEV_OFFLOAD_PUSH_GENEVE)
1559                 geneve_push_rx_ports(dev);
1560
1561         return NOTIFY_DONE;
1562 }
1563
1564 static struct notifier_block geneve_notifier_block __read_mostly = {
1565         .notifier_call = geneve_netdevice_event,
1566 };
1567
1568 static __net_init int geneve_init_net(struct net *net)
1569 {
1570         struct geneve_net *gn = net_generic(net, geneve_net_id);
1571
1572         INIT_LIST_HEAD(&gn->geneve_list);
1573         INIT_LIST_HEAD(&gn->sock_list);
1574         return 0;
1575 }
1576
1577 static void __net_exit geneve_exit_net(struct net *net)
1578 {
1579         struct geneve_net *gn = net_generic(net, geneve_net_id);
1580         struct geneve_dev *geneve, *next;
1581         struct net_device *dev, *aux;
1582         LIST_HEAD(list);
1583
1584         rtnl_lock();
1585
1586         /* gather any geneve devices that were moved into this ns */
1587         for_each_netdev_safe(net, dev, aux)
1588                 if (dev->rtnl_link_ops == &geneve_link_ops)
1589                         unregister_netdevice_queue(dev, &list);
1590
1591         /* now gather any other geneve devices that were created in this ns */
1592         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1593                 /* If geneve->dev is in the same netns, it was already added
1594                  * to the list by the previous loop.
1595                  */
1596                 if (!net_eq(dev_net(geneve->dev), net))
1597                         unregister_netdevice_queue(geneve->dev, &list);
1598         }
1599
1600         /* unregister the devices gathered above */
1601         unregister_netdevice_many(&list);
1602         rtnl_unlock();
1603 }
1604
1605 static struct pernet_operations geneve_net_ops = {
1606         .init = geneve_init_net,
1607         .exit = geneve_exit_net,
1608         .id   = &geneve_net_id,
1609         .size = sizeof(struct geneve_net),
1610 };
1611
1612 static int __init geneve_init_module(void)
1613 {
1614         int rc;
1615
1616         rc = register_pernet_subsys(&geneve_net_ops);
1617         if (rc)
1618                 goto out1;
1619
1620         rc = register_netdevice_notifier(&geneve_notifier_block);
1621         if (rc)
1622                 goto out2;
1623
1624         rc = rtnl_link_register(&geneve_link_ops);
1625         if (rc)
1626                 goto out3;
1627
1628         return 0;
1629
1630 out3:
1631         unregister_netdevice_notifier(&geneve_notifier_block);
1632 out2:
1633         unregister_pernet_subsys(&geneve_net_ops);
1634 out1:
1635         return rc;
1636 }
1637 late_initcall(geneve_init_module);
1638
1639 static void __exit geneve_cleanup_module(void)
1640 {
1641         rtnl_link_unregister(&geneve_link_ops);
1642         unregister_netdevice_notifier(&geneve_notifier_block);
1643         unregister_pernet_subsys(&geneve_net_ops);
1644 }
1645 module_exit(geneve_cleanup_module);
1646
1647 MODULE_LICENSE("GPL");
1648 MODULE_VERSION(GENEVE_NETDEV_VER);
1649 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1650 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1651 MODULE_ALIAS_RTNL_LINK("geneve");