drivers/net/geneve.c

   1 /*
   2  * GENEVE: Generic Network Virtualization Encapsulation
   3  *
   4  * Copyright (c) 2015 Red Hat, Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/netdevice.h>
  16 #include <linux/etherdevice.h>
  17 #include <linux/hash.h>
  18 #include <net/dst_metadata.h>
  19 #include <net/gro_cells.h>
  20 #include <net/rtnetlink.h>
  21 #include <net/geneve.h>
  22 #include <net/protocol.h>
  23
  24 #define GENEVE_NETDEV_VER       "0.6"
  25
  26 #define GENEVE_UDP_PORT         6081
  27
  28 #define GENEVE_N_VID            (1u << 24)
  29 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  30
  31 #define VNI_HASH_BITS           10
  32 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  33
  34 static bool log_ecn_error = true;
  35 module_param(log_ecn_error, bool, 0644);
  36 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  37
  38 #define GENEVE_VER 0
  39 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  40
  41 /* per-network namespace private data for this module */
  42 struct geneve_net {
  43         struct list_head        geneve_list;
  44         struct list_head        sock_list;
  45 };
  46
  47 static int geneve_net_id;
  48
  49 union geneve_addr {
  50         struct sockaddr_in sin;
  51         struct sockaddr_in6 sin6;
  52         struct sockaddr sa;
  53 };
  54
  55 static union geneve_addr geneve_remote_unspec = { .sa.sa_family = AF_UNSPEC, };
  56
  57 /* Pseudo network device */
  58 struct geneve_dev {
  59         struct hlist_node  hlist;       /* vni hash table */
  60         struct net         *net;        /* netns for packet i/o */
  61         struct net_device  *dev;        /* netdev for geneve tunnel */
  62         struct geneve_sock *sock4;      /* IPv4 socket used for geneve tunnel */
  63 #if IS_ENABLED(CONFIG_IPV6)
  64         struct geneve_sock *sock6;      /* IPv6 socket used for geneve tunnel */
  65 #endif
  66         u8                 vni[3];      /* virtual network ID for tunnel */
  67         u8                 ttl;         /* TTL override */
  68         u8                 tos;         /* TOS override */
  69         union geneve_addr  remote;      /* IP address for link partner */
  70         struct list_head   next;        /* geneve's per namespace list */
  71         __be32             label;       /* IPv6 flowlabel override */
  72         __be16             dst_port;
  73         bool               collect_md;
  74         struct gro_cells   gro_cells;
  75         u32                flags;
  76         struct dst_cache   dst_cache;
  77 };
  78
  79 /* Geneve device flags */
  80 #define GENEVE_F_UDP_ZERO_CSUM_TX       BIT(0)
  81 #define GENEVE_F_UDP_ZERO_CSUM6_TX      BIT(1)
  82 #define GENEVE_F_UDP_ZERO_CSUM6_RX      BIT(2)
  83
  84 struct geneve_sock {
  85         bool                    collect_md;
  86         struct list_head        list;
  87         struct socket           *sock;
  88         struct rcu_head         rcu;
  89         int                     refcnt;
  90         struct udp_offload      udp_offloads;
  91         struct hlist_head       vni_list[VNI_HASH_SIZE];
  92         u32                     flags;
  93 };
  94
  95 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  96 {
  97         __u32 vnid;
  98
  99         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
 100         return hash_32(vnid, VNI_HASH_BITS);
 101 }
 102
 103 static __be64 vni_to_tunnel_id(const __u8 *vni)
 104 {
 105 #ifdef __BIG_ENDIAN
 106         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
 107 #else
 108         return (__force __be64)(((__force u64)vni[0] << 40) |
 109                                 ((__force u64)vni[1] << 48) |
 110                                 ((__force u64)vni[2] << 56));
 111 #endif
 112 }
 113
 114 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 115 {
 116         return gs->sock->sk->sk_family;
 117 }
 118
 119 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 120                                         __be32 addr, u8 vni[])
 121 {
 122         struct hlist_head *vni_list_head;
 123         struct geneve_dev *geneve;
 124         __u32 hash;
 125
 126         /* Find the device for this VNI */
 127         hash = geneve_net_vni_hash(vni);
 128         vni_list_head = &gs->vni_list[hash];
 129         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 130                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 131                     addr == geneve->remote.sin.sin_addr.s_addr)
 132                         return geneve;
 133         }
 134         return NULL;
 135 }
 136
 137 #if IS_ENABLED(CONFIG_IPV6)
 138 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 139                                          struct in6_addr addr6, u8 vni[])
 140 {
 141         struct hlist_head *vni_list_head;
 142         struct geneve_dev *geneve;
 143         __u32 hash;
 144
 145         /* Find the device for this VNI */
 146         hash = geneve_net_vni_hash(vni);
 147         vni_list_head = &gs->vni_list[hash];
 148         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 149                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 150                     ipv6_addr_equal(&addr6, &geneve->remote.sin6.sin6_addr))
 151                         return geneve;
 152         }
 153         return NULL;
 154 }
 155 #endif
 156
 157 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 158 {
 159         return (struct genevehdr *)(udp_hdr(skb) + 1);
 160 }
 161
 162 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 163                                             struct sk_buff *skb)
 164 {
 165         u8 *vni;
 166         __be32 addr;
 167         static u8 zero_vni[3];
 168 #if IS_ENABLED(CONFIG_IPV6)
 169         static struct in6_addr zero_addr6;
 170 #endif
 171
 172         if (geneve_get_sk_family(gs) == AF_INET) {
 173                 struct iphdr *iph;
 174
 175                 iph = ip_hdr(skb); /* outer IP header... */
 176
 177                 if (gs->collect_md) {
 178                         vni = zero_vni;
 179                         addr = 0;
 180                 } else {
 181                         vni = geneve_hdr(skb)->vni;
 182                         addr = iph->saddr;
 183                 }
 184
 185                 return geneve_lookup(gs, addr, vni);
 186 #if IS_ENABLED(CONFIG_IPV6)
 187         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 188                 struct ipv6hdr *ip6h;
 189                 struct in6_addr addr6;
 190
 191                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 192
 193                 if (gs->collect_md) {
 194                         vni = zero_vni;
 195                         addr6 = zero_addr6;
 196                 } else {
 197                         vni = geneve_hdr(skb)->vni;
 198                         addr6 = ip6h->saddr;
 199                 }
 200
 201                 return geneve6_lookup(gs, addr6, vni);
 202 #endif
 203         }
 204         return NULL;
 205 }
 206
 207 /* geneve receive/decap routine */
 208 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 209                       struct sk_buff *skb)
 210 {
 211         struct genevehdr *gnvh = geneve_hdr(skb);
 212         struct metadata_dst *tun_dst = NULL;
 213         struct pcpu_sw_netstats *stats;
 214         int err = 0;
 215         void *oiph;
 216
 217         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 218                 __be16 flags;
 219
 220                 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
 221                         (gnvh->oam ? TUNNEL_OAM : 0) |
 222                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 223
 224                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 225                                          vni_to_tunnel_id(gnvh->vni),
 226                                          gnvh->opt_len * 4);
 227                 if (!tun_dst)
 228                         goto drop;
 229                 /* Update tunnel dst according to Geneve options. */
 230                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 231                                         gnvh->options, gnvh->opt_len * 4);
 232         } else {
 233                 /* Drop packets w/ critical options,
 234                  * since we don't support any...
 235                  */
 236                 if (gnvh->critical)
 237                         goto drop;
 238         }
 239
 240         skb_reset_mac_header(skb);
 241         skb->protocol = eth_type_trans(skb, geneve->dev);
 242         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 243
 244         if (tun_dst)
 245                 skb_dst_set(skb, &tun_dst->dst);
 246
 247         /* Ignore packet loops (and multicast echo) */
 248         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
 249                 goto drop;
 250
 251         oiph = skb_network_header(skb);
 252         skb_reset_network_header(skb);
 253
 254         if (geneve_get_sk_family(gs) == AF_INET)
 255                 err = IP_ECN_decapsulate(oiph, skb);
 256 #if IS_ENABLED(CONFIG_IPV6)
 257         else
 258                 err = IP6_ECN_decapsulate(oiph, skb);
 259 #endif
 260
 261         if (unlikely(err)) {
 262                 if (log_ecn_error) {
 263                         if (geneve_get_sk_family(gs) == AF_INET)
 264                                 net_info_ratelimited("non-ECT from %pI4 "
 265                                                      "with TOS=%#x\n",
 266                                                      &((struct iphdr *)oiph)->saddr,
 267                                                      ((struct iphdr *)oiph)->tos);
 268 #if IS_ENABLED(CONFIG_IPV6)
 269                         else
 270                                 net_info_ratelimited("non-ECT from %pI6\n",
 271                                                      &((struct ipv6hdr *)oiph)->saddr);
 272 #endif
 273                 }
 274                 if (err > 1) {
 275                         ++geneve->dev->stats.rx_frame_errors;
 276                         ++geneve->dev->stats.rx_errors;
 277                         goto drop;
 278                 }
 279         }
 280
 281         stats = this_cpu_ptr(geneve->dev->tstats);
 282         u64_stats_update_begin(&stats->syncp);
 283         stats->rx_packets++;
 284         stats->rx_bytes += skb->len;
 285         u64_stats_update_end(&stats->syncp);
 286
 287         gro_cells_receive(&geneve->gro_cells, skb);
 288         return;
 289 drop:
 290         /* Consume bad packet */
 291         kfree_skb(skb);
 292 }
 293
 294 /* Setup stats when device is created */
 295 static int geneve_init(struct net_device *dev)
 296 {
 297         struct geneve_dev *geneve = netdev_priv(dev);
 298         int err;
 299
 300         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 301         if (!dev->tstats)
 302                 return -ENOMEM;
 303
 304         err = gro_cells_init(&geneve->gro_cells, dev);
 305         if (err) {
 306                 free_percpu(dev->tstats);
 307                 return err;
 308         }
 309
 310         err = dst_cache_init(&geneve->dst_cache, GFP_KERNEL);
 311         if (err) {
 312                 free_percpu(dev->tstats);
 313                 gro_cells_destroy(&geneve->gro_cells);
 314                 return err;
 315         }
 316
 317         return 0;
 318 }
 319
 320 static void geneve_uninit(struct net_device *dev)
 321 {
 322         struct geneve_dev *geneve = netdev_priv(dev);
 323
 324         dst_cache_destroy(&geneve->dst_cache);
 325         gro_cells_destroy(&geneve->gro_cells);
 326         free_percpu(dev->tstats);
 327 }
 328
 329 /* Callback from net/ipv4/udp.c to receive packets */
 330 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 331 {
 332         struct genevehdr *geneveh;
 333         struct geneve_dev *geneve;
 334         struct geneve_sock *gs;
 335         int opts_len;
 336
 337         /* Need Geneve and inner Ethernet header to be present */
 338         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 339                 goto error;
 340
 341         /* Return packets with reserved bits set */
 342         geneveh = geneve_hdr(skb);
 343         if (unlikely(geneveh->ver != GENEVE_VER))
 344                 goto error;
 345
 346         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 347                 goto error;
 348
 349         gs = rcu_dereference_sk_user_data(sk);
 350         if (!gs)
 351                 goto drop;
 352
 353         geneve = geneve_lookup_skb(gs, skb);
 354         if (!geneve)
 355                 goto drop;
 356
 357         opts_len = geneveh->opt_len * 4;
 358         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 359                                  htons(ETH_P_TEB),
 360                                  !net_eq(geneve->net, dev_net(geneve->dev))))
 361                 goto drop;
 362
 363         geneve_rx(geneve, gs, skb);
 364         return 0;
 365
 366 drop:
 367         /* Consume bad packet */
 368         kfree_skb(skb);
 369         return 0;
 370
 371 error:
 372         /* Let the UDP layer deal with the skb */
 373         return 1;
 374 }
 375
 376 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 377                                          __be16 port, u32 flags)
 378 {
 379         struct socket *sock;
 380         struct udp_port_cfg udp_conf;
 381         int err;
 382
 383         memset(&udp_conf, 0, sizeof(udp_conf));
 384
 385         if (ipv6) {
 386                 udp_conf.family = AF_INET6;
 387                 udp_conf.ipv6_v6only = 1;
 388                 udp_conf.use_udp6_rx_checksums =
 389                     !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX);
 390         } else {
 391                 udp_conf.family = AF_INET;
 392                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 393         }
 394
 395         udp_conf.local_udp_port = port;
 396
 397         /* Open UDP socket */
 398         err = udp_sock_create(net, &udp_conf, &sock);
 399         if (err < 0)
 400                 return ERR_PTR(err);
 401
 402         return sock;
 403 }
 404
 405 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
 406 {
 407         struct net_device *dev;
 408         struct sock *sk = gs->sock->sk;
 409         struct net *net = sock_net(sk);
 410         sa_family_t sa_family = geneve_get_sk_family(gs);
 411         __be16 port = inet_sk(sk)->inet_sport;
 412         int err;
 413
 414         if (sa_family == AF_INET) {
 415                 err = udp_add_offload(sock_net(sk), &gs->udp_offloads);
 416                 if (err)
 417                         pr_warn("geneve: udp_add_offload failed with status %d\n",
 418                                 err);
 419         }
 420
 421         rcu_read_lock();
 422         for_each_netdev_rcu(net, dev) {
 423                 if (dev->netdev_ops->ndo_add_geneve_port)
 424                         dev->netdev_ops->ndo_add_geneve_port(dev, sa_family,
 425                                                              port);
 426         }
 427         rcu_read_unlock();
 428 }
 429
 430 static int geneve_hlen(struct genevehdr *gh)
 431 {
 432         return sizeof(*gh) + gh->opt_len * 4;
 433 }
 434
 435 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
 436                                            struct sk_buff *skb,
 437                                            struct udp_offload *uoff)
 438 {
 439         struct sk_buff *p, **pp = NULL;
 440         struct genevehdr *gh, *gh2;
 441         unsigned int hlen, gh_len, off_gnv;
 442         const struct packet_offload *ptype;
 443         __be16 type;
 444         int flush = 1;
 445
 446         off_gnv = skb_gro_offset(skb);
 447         hlen = off_gnv + sizeof(*gh);
 448         gh = skb_gro_header_fast(skb, off_gnv);
 449         if (skb_gro_header_hard(skb, hlen)) {
 450                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 451                 if (unlikely(!gh))
 452                         goto out;
 453         }
 454
 455         if (gh->ver != GENEVE_VER || gh->oam)
 456                 goto out;
 457         gh_len = geneve_hlen(gh);
 458
 459         hlen = off_gnv + gh_len;
 460         if (skb_gro_header_hard(skb, hlen)) {
 461                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 462                 if (unlikely(!gh))
 463                         goto out;
 464         }
 465
 466         flush = 0;
 467
 468         for (p = *head; p; p = p->next) {
 469                 if (!NAPI_GRO_CB(p)->same_flow)
 470                         continue;
 471
 472                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 473                 if (gh->opt_len != gh2->opt_len ||
 474                     memcmp(gh, gh2, gh_len)) {
 475                         NAPI_GRO_CB(p)->same_flow = 0;
 476                         continue;
 477                 }
 478         }
 479
 480         type = gh->proto_type;
 481
 482         rcu_read_lock();
 483         ptype = gro_find_receive_by_type(type);
 484         if (!ptype) {
 485                 flush = 1;
 486                 goto out_unlock;
 487         }
 488
 489         skb_gro_pull(skb, gh_len);
 490         skb_gro_postpull_rcsum(skb, gh, gh_len);
 491         pp = ptype->callbacks.gro_receive(head, skb);
 492
 493 out_unlock:
 494         rcu_read_unlock();
 495 out:
 496         NAPI_GRO_CB(skb)->flush |= flush;
 497
 498         return pp;
 499 }
 500
 501 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
 502                                struct udp_offload *uoff)
 503 {
 504         struct genevehdr *gh;
 505         struct packet_offload *ptype;
 506         __be16 type;
 507         int gh_len;
 508         int err = -ENOSYS;
 509
 510         udp_tunnel_gro_complete(skb, nhoff);
 511
 512         gh = (struct genevehdr *)(skb->data + nhoff);
 513         gh_len = geneve_hlen(gh);
 514         type = gh->proto_type;
 515
 516         rcu_read_lock();
 517         ptype = gro_find_complete_by_type(type);
 518         if (ptype)
 519                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 520
 521         rcu_read_unlock();
 522         return err;
 523 }
 524
 525 /* Create new listen socket if needed */
 526 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 527                                                 bool ipv6, u32 flags)
 528 {
 529         struct geneve_net *gn = net_generic(net, geneve_net_id);
 530         struct geneve_sock *gs;
 531         struct socket *sock;
 532         struct udp_tunnel_sock_cfg tunnel_cfg;
 533         int h;
 534
 535         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 536         if (!gs)
 537                 return ERR_PTR(-ENOMEM);
 538
 539         sock = geneve_create_sock(net, ipv6, port, flags);
 540         if (IS_ERR(sock)) {
 541                 kfree(gs);
 542                 return ERR_CAST(sock);
 543         }
 544
 545         gs->sock = sock;
 546         gs->refcnt = 1;
 547         for (h = 0; h < VNI_HASH_SIZE; ++h)
 548                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 549
 550         /* Initialize the geneve udp offloads structure */
 551         gs->udp_offloads.port = port;
 552         gs->udp_offloads.callbacks.gro_receive  = geneve_gro_receive;
 553         gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
 554         geneve_notify_add_rx_port(gs);
 555
 556         /* Mark socket as an encapsulation socket */
 557         tunnel_cfg.sk_user_data = gs;
 558         tunnel_cfg.encap_type = 1;
 559         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 560         tunnel_cfg.encap_destroy = NULL;
 561         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 562         list_add(&gs->list, &gn->sock_list);
 563         return gs;
 564 }
 565
 566 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 567 {
 568         struct net_device *dev;
 569         struct sock *sk = gs->sock->sk;
 570         struct net *net = sock_net(sk);
 571         sa_family_t sa_family = geneve_get_sk_family(gs);
 572         __be16 port = inet_sk(sk)->inet_sport;
 573
 574         rcu_read_lock();
 575         for_each_netdev_rcu(net, dev) {
 576                 if (dev->netdev_ops->ndo_del_geneve_port)
 577                         dev->netdev_ops->ndo_del_geneve_port(dev, sa_family,
 578                                                              port);
 579         }
 580
 581         rcu_read_unlock();
 582
 583         if (sa_family == AF_INET)
 584                 udp_del_offload(&gs->udp_offloads);
 585 }
 586
 587 static void __geneve_sock_release(struct geneve_sock *gs)
 588 {
 589         if (!gs || --gs->refcnt)
 590                 return;
 591
 592         list_del(&gs->list);
 593         geneve_notify_del_rx_port(gs);
 594         udp_tunnel_sock_release(gs->sock);
 595         kfree_rcu(gs, rcu);
 596 }
 597
 598 static void geneve_sock_release(struct geneve_dev *geneve)
 599 {
 600         __geneve_sock_release(geneve->sock4);
 601 #if IS_ENABLED(CONFIG_IPV6)
 602         __geneve_sock_release(geneve->sock6);
 603 #endif
 604 }
 605
 606 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 607                                             sa_family_t family,
 608                                             __be16 dst_port)
 609 {
 610         struct geneve_sock *gs;
 611
 612         list_for_each_entry(gs, &gn->sock_list, list) {
 613                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 614                     geneve_get_sk_family(gs) == family) {
 615                         return gs;
 616                 }
 617         }
 618         return NULL;
 619 }
 620
 621 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 622 {
 623         struct net *net = geneve->net;
 624         struct geneve_net *gn = net_generic(net, geneve_net_id);
 625         struct geneve_sock *gs;
 626         __u32 hash;
 627
 628         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
 629         if (gs) {
 630                 gs->refcnt++;
 631                 goto out;
 632         }
 633
 634         gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
 635         if (IS_ERR(gs))
 636                 return PTR_ERR(gs);
 637
 638 out:
 639         gs->collect_md = geneve->collect_md;
 640         gs->flags = geneve->flags;
 641 #if IS_ENABLED(CONFIG_IPV6)
 642         if (ipv6)
 643                 geneve->sock6 = gs;
 644         else
 645 #endif
 646                 geneve->sock4 = gs;
 647
 648         hash = geneve_net_vni_hash(geneve->vni);
 649         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 650         return 0;
 651 }
 652
 653 static int geneve_open(struct net_device *dev)
 654 {
 655         struct geneve_dev *geneve = netdev_priv(dev);
 656         bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
 657         bool metadata = geneve->collect_md;
 658         int ret = 0;
 659
 660         geneve->sock4 = NULL;
 661 #if IS_ENABLED(CONFIG_IPV6)
 662         geneve->sock6 = NULL;
 663         if (ipv6 || metadata)
 664                 ret = geneve_sock_add(geneve, true);
 665 #endif
 666         if (!ret && (!ipv6 || metadata))
 667                 ret = geneve_sock_add(geneve, false);
 668         if (ret < 0)
 669                 geneve_sock_release(geneve);
 670
 671         return ret;
 672 }
 673
 674 static int geneve_stop(struct net_device *dev)
 675 {
 676         struct geneve_dev *geneve = netdev_priv(dev);
 677
 678         if (!hlist_unhashed(&geneve->hlist))
 679                 hlist_del_rcu(&geneve->hlist);
 680         geneve_sock_release(geneve);
 681         return 0;
 682 }
 683
 684 static void geneve_build_header(struct genevehdr *geneveh,
 685                                 __be16 tun_flags, u8 vni[3],
 686                                 u8 options_len, u8 *options)
 687 {
 688         geneveh->ver = GENEVE_VER;
 689         geneveh->opt_len = options_len / 4;
 690         geneveh->oam = !!(tun_flags & TUNNEL_OAM);
 691         geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 692         geneveh->rsvd1 = 0;
 693         memcpy(geneveh->vni, vni, 3);
 694         geneveh->proto_type = htons(ETH_P_TEB);
 695         geneveh->rsvd2 = 0;
 696
 697         memcpy(geneveh->options, options, options_len);
 698 }
 699
 700 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 701                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 702                             u32 flags, bool xnet)
 703 {
 704         struct genevehdr *gnvh;
 705         int min_headroom;
 706         int err;
 707         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
 708
 709         skb_scrub_packet(skb, xnet);
 710
 711         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 712                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 713         err = skb_cow_head(skb, min_headroom);
 714         if (unlikely(err)) {
 715                 kfree_skb(skb);
 716                 goto free_rt;
 717         }
 718
 719         skb = udp_tunnel_handle_offloads(skb, udp_sum);
 720         if (IS_ERR(skb)) {
 721                 err = PTR_ERR(skb);
 722                 goto free_rt;
 723         }
 724
 725         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 726         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 727
 728         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 729         return 0;
 730
 731 free_rt:
 732         ip_rt_put(rt);
 733         return err;
 734 }
 735
 736 #if IS_ENABLED(CONFIG_IPV6)
 737 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 738                              __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 739                              u32 flags, bool xnet)
 740 {
 741         struct genevehdr *gnvh;
 742         int min_headroom;
 743         int err;
 744         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
 745
 746         skb_scrub_packet(skb, xnet);
 747
 748         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
 749                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
 750         err = skb_cow_head(skb, min_headroom);
 751         if (unlikely(err)) {
 752                 kfree_skb(skb);
 753                 goto free_dst;
 754         }
 755
 756         skb = udp_tunnel_handle_offloads(skb, udp_sum);
 757         if (IS_ERR(skb)) {
 758                 err = PTR_ERR(skb);
 759                 goto free_dst;
 760         }
 761
 762         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 763         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 764
 765         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 766         return 0;
 767
 768 free_dst:
 769         dst_release(dst);
 770         return err;
 771 }
 772 #endif
 773
 774 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 775                                        struct net_device *dev,
 776                                        struct flowi4 *fl4,
 777                                        struct ip_tunnel_info *info)
 778 {
 779         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 780         struct geneve_dev *geneve = netdev_priv(dev);
 781         struct dst_cache *dst_cache;
 782         struct rtable *rt = NULL;
 783         __u8 tos;
 784
 785         memset(fl4, 0, sizeof(*fl4));
 786         fl4->flowi4_mark = skb->mark;
 787         fl4->flowi4_proto = IPPROTO_UDP;
 788
 789         if (info) {
 790                 fl4->daddr = info->key.u.ipv4.dst;
 791                 fl4->saddr = info->key.u.ipv4.src;
 792                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 793                 dst_cache = &info->dst_cache;
 794         } else {
 795                 tos = geneve->tos;
 796                 if (tos == 1) {
 797                         const struct iphdr *iip = ip_hdr(skb);
 798
 799                         tos = ip_tunnel_get_dsfield(iip, skb);
 800                         use_cache = false;
 801                 }
 802
 803                 fl4->flowi4_tos = RT_TOS(tos);
 804                 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
 805                 dst_cache = &geneve->dst_cache;
 806         }
 807
 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         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 836         struct geneve_dev *geneve = netdev_priv(dev);
 837         struct geneve_sock *gs6 = geneve->sock6;
 838         struct dst_entry *dst = NULL;
 839         struct dst_cache *dst_cache;
 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                 fl6->flowlabel = info->key.label;
 851                 dst_cache = &info->dst_cache;
 852         } else {
 853                 prio = geneve->tos;
 854                 if (prio == 1) {
 855                         const struct iphdr *iip = ip_hdr(skb);
 856
 857                         prio = ip_tunnel_get_dsfield(iip, skb);
 858                         use_cache = false;
 859                 }
 860
 861                 fl6->flowi6_tos = RT_TOS(prio);
 862                 fl6->flowlabel = geneve->label;
 863                 fl6->daddr = geneve->remote.sin6.sin6_addr;
 864                 dst_cache = &geneve->dst_cache;
 865         }
 866
 867         if (use_cache) {
 868                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 869                 if (dst)
 870                         return dst;
 871         }
 872
 873         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 874                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 875                 return ERR_PTR(-ENETUNREACH);
 876         }
 877         if (dst->dev == dev) { /* is this necessary? */
 878                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 879                 dst_release(dst);
 880                 return ERR_PTR(-ELOOP);
 881         }
 882
 883         if (use_cache)
 884                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 885         return dst;
 886 }
 887 #endif
 888
 889 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 890 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 891 {
 892 #ifdef __BIG_ENDIAN
 893         vni[0] = (__force __u8)(tun_id >> 16);
 894         vni[1] = (__force __u8)(tun_id >> 8);
 895         vni[2] = (__force __u8)tun_id;
 896 #else
 897         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 898         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 899         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 900 #endif
 901 }
 902
 903 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 904                                    struct ip_tunnel_info *info)
 905 {
 906         struct geneve_dev *geneve = netdev_priv(dev);
 907         struct geneve_sock *gs4 = geneve->sock4;
 908         struct rtable *rt = NULL;
 909         const struct iphdr *iip; /* interior IP header */
 910         int err = -EINVAL;
 911         struct flowi4 fl4;
 912         __u8 tos, ttl;
 913         __be16 sport;
 914         __be16 df;
 915         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 916         u32 flags = geneve->flags;
 917
 918         if (geneve->collect_md) {
 919                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 920                         netdev_dbg(dev, "no tunnel metadata\n");
 921                         goto tx_error;
 922                 }
 923                 if (info && ip_tunnel_info_af(info) != AF_INET)
 924                         goto tx_error;
 925         }
 926
 927         rt = geneve_get_v4_rt(skb, dev, &fl4, info);
 928         if (IS_ERR(rt)) {
 929                 err = PTR_ERR(rt);
 930                 goto tx_error;
 931         }
 932
 933         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 934         skb_reset_mac_header(skb);
 935
 936         iip = ip_hdr(skb);
 937
 938         if (info) {
 939                 const struct ip_tunnel_key *key = &info->key;
 940                 u8 *opts = NULL;
 941                 u8 vni[3];
 942
 943                 tunnel_id_to_vni(key->tun_id, vni);
 944                 if (info->options_len)
 945                         opts = ip_tunnel_info_opts(info);
 946
 947                 if (key->tun_flags & TUNNEL_CSUM)
 948                         flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
 949                 else
 950                         flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
 951
 952                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 953                                        info->options_len, opts, flags, xnet);
 954                 if (unlikely(err))
 955                         goto err;
 956
 957                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
 958                 ttl = key->ttl;
 959                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 960         } else {
 961                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 962                                        0, NULL, flags, xnet);
 963                 if (unlikely(err))
 964                         goto err;
 965
 966                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 967                 ttl = geneve->ttl;
 968                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 969                         ttl = 1;
 970                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 971                 df = 0;
 972         }
 973         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 974                             tos, ttl, df, sport, geneve->dst_port,
 975                             !net_eq(geneve->net, dev_net(geneve->dev)),
 976                             !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
 977
 978         return NETDEV_TX_OK;
 979
 980 tx_error:
 981         dev_kfree_skb(skb);
 982 err:
 983         if (err == -ELOOP)
 984                 dev->stats.collisions++;
 985         else if (err == -ENETUNREACH)
 986                 dev->stats.tx_carrier_errors++;
 987         else
 988                 dev->stats.tx_errors++;
 989         return NETDEV_TX_OK;
 990 }
 991
 992 #if IS_ENABLED(CONFIG_IPV6)
 993 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 994                                     struct ip_tunnel_info *info)
 995 {
 996         struct geneve_dev *geneve = netdev_priv(dev);
 997         struct geneve_sock *gs6 = geneve->sock6;
 998         struct dst_entry *dst = NULL;
 999         const struct iphdr *iip; /* interior IP header */
1000         int err = -EINVAL;
1001         struct flowi6 fl6;
1002         __u8 prio, ttl;
1003         __be16 sport;
1004         __be32 label;
1005         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
1006         u32 flags = geneve->flags;
1007
1008         if (geneve->collect_md) {
1009                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1010                         netdev_dbg(dev, "no tunnel metadata\n");
1011                         goto tx_error;
1012                 }
1013         }
1014
1015         dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1016         if (IS_ERR(dst)) {
1017                 err = PTR_ERR(dst);
1018                 goto tx_error;
1019         }
1020
1021         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1022         skb_reset_mac_header(skb);
1023
1024         iip = ip_hdr(skb);
1025
1026         if (info) {
1027                 const struct ip_tunnel_key *key = &info->key;
1028                 u8 *opts = NULL;
1029                 u8 vni[3];
1030
1031                 tunnel_id_to_vni(key->tun_id, vni);
1032                 if (info->options_len)
1033                         opts = ip_tunnel_info_opts(info);
1034
1035                 if (key->tun_flags & TUNNEL_CSUM)
1036                         flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
1037                 else
1038                         flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1039
1040                 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
1041                                         info->options_len, opts,
1042                                         flags, xnet);
1043                 if (unlikely(err))
1044                         goto err;
1045
1046                 prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
1047                 ttl = key->ttl;
1048                 label = info->key.label;
1049         } else {
1050                 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
1051                                         0, NULL, flags, xnet);
1052                 if (unlikely(err))
1053                         goto err;
1054
1055                 prio = ip_tunnel_ecn_encap(fl6.flowi6_tos, iip, skb);
1056                 ttl = geneve->ttl;
1057                 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1058                         ttl = 1;
1059                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1060                 label = geneve->label;
1061         }
1062
1063         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1064                              &fl6.saddr, &fl6.daddr, prio, ttl, label,
1065                              sport, geneve->dst_port,
1066                              !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1067         return NETDEV_TX_OK;
1068
1069 tx_error:
1070         dev_kfree_skb(skb);
1071 err:
1072         if (err == -ELOOP)
1073                 dev->stats.collisions++;
1074         else if (err == -ENETUNREACH)
1075                 dev->stats.tx_carrier_errors++;
1076         else
1077                 dev->stats.tx_errors++;
1078         return NETDEV_TX_OK;
1079 }
1080 #endif
1081
1082 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1083 {
1084         struct geneve_dev *geneve = netdev_priv(dev);
1085         struct ip_tunnel_info *info = NULL;
1086
1087         if (geneve->collect_md)
1088                 info = skb_tunnel_info(skb);
1089
1090 #if IS_ENABLED(CONFIG_IPV6)
1091         if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1092             (!info && geneve->remote.sa.sa_family == AF_INET6))
1093                 return geneve6_xmit_skb(skb, dev, info);
1094 #endif
1095         return geneve_xmit_skb(skb, dev, info);
1096 }
1097
1098 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1099 {
1100         /* The max_mtu calculation does not take account of GENEVE
1101          * options, to avoid excluding potentially valid
1102          * configurations.
1103          */
1104         int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - sizeof(struct iphdr)
1105                 - dev->hard_header_len;
1106
1107         if (new_mtu < 68)
1108                 return -EINVAL;
1109
1110         if (new_mtu > max_mtu) {
1111                 if (strict)
1112                         return -EINVAL;
1113
1114                 new_mtu = max_mtu;
1115         }
1116
1117         dev->mtu = new_mtu;
1118         return 0;
1119 }
1120
1121 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1122 {
1123         return __geneve_change_mtu(dev, new_mtu, true);
1124 }
1125
1126 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1127 {
1128         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1129         struct geneve_dev *geneve = netdev_priv(dev);
1130         struct rtable *rt;
1131         struct flowi4 fl4;
1132 #if IS_ENABLED(CONFIG_IPV6)
1133         struct dst_entry *dst;
1134         struct flowi6 fl6;
1135 #endif
1136
1137         if (ip_tunnel_info_af(info) == AF_INET) {
1138                 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1139                 if (IS_ERR(rt))
1140                         return PTR_ERR(rt);
1141
1142                 ip_rt_put(rt);
1143                 info->key.u.ipv4.src = fl4.saddr;
1144 #if IS_ENABLED(CONFIG_IPV6)
1145         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1146                 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1147                 if (IS_ERR(dst))
1148                         return PTR_ERR(dst);
1149
1150                 dst_release(dst);
1151                 info->key.u.ipv6.src = fl6.saddr;
1152 #endif
1153         } else {
1154                 return -EINVAL;
1155         }
1156
1157         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1158                                              1, USHRT_MAX, true);
1159         info->key.tp_dst = geneve->dst_port;
1160         return 0;
1161 }
1162
1163 static const struct net_device_ops geneve_netdev_ops = {
1164         .ndo_init               = geneve_init,
1165         .ndo_uninit             = geneve_uninit,
1166         .ndo_open               = geneve_open,
1167         .ndo_stop               = geneve_stop,
1168         .ndo_start_xmit         = geneve_xmit,
1169         .ndo_get_stats64        = ip_tunnel_get_stats64,
1170         .ndo_change_mtu         = geneve_change_mtu,
1171         .ndo_validate_addr      = eth_validate_addr,
1172         .ndo_set_mac_address    = eth_mac_addr,
1173         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1174 };
1175
1176 static void geneve_get_drvinfo(struct net_device *dev,
1177                                struct ethtool_drvinfo *drvinfo)
1178 {
1179         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1180         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1181 }
1182
1183 static const struct ethtool_ops geneve_ethtool_ops = {
1184         .get_drvinfo    = geneve_get_drvinfo,
1185         .get_link       = ethtool_op_get_link,
1186 };
1187
1188 /* Info for udev, that this is a virtual tunnel endpoint */
1189 static struct device_type geneve_type = {
1190         .name = "geneve",
1191 };
1192
1193 /* Calls the ndo_add_geneve_port of the caller in order to
1194  * supply the listening GENEVE udp ports. Callers are expected
1195  * to implement the ndo_add_geneve_port.
1196  */
1197 void geneve_get_rx_port(struct net_device *dev)
1198 {
1199         struct net *net = dev_net(dev);
1200         struct geneve_net *gn = net_generic(net, geneve_net_id);
1201         struct geneve_sock *gs;
1202         sa_family_t sa_family;
1203         struct sock *sk;
1204         __be16 port;
1205
1206         rcu_read_lock();
1207         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1208                 sk = gs->sock->sk;
1209                 sa_family = sk->sk_family;
1210                 port = inet_sk(sk)->inet_sport;
1211                 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family, port);
1212         }
1213         rcu_read_unlock();
1214 }
1215 EXPORT_SYMBOL_GPL(geneve_get_rx_port);
1216
1217 /* Initialize the device structure. */
1218 static void geneve_setup(struct net_device *dev)
1219 {
1220         ether_setup(dev);
1221
1222         dev->netdev_ops = &geneve_netdev_ops;
1223         dev->ethtool_ops = &geneve_ethtool_ops;
1224         dev->destructor = free_netdev;
1225
1226         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1227
1228         dev->features    |= NETIF_F_LLTX;
1229         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1230         dev->features    |= NETIF_F_RXCSUM;
1231         dev->features    |= NETIF_F_GSO_SOFTWARE;
1232
1233         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1234         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1235
1236         netif_keep_dst(dev);
1237         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1238         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1239         eth_hw_addr_random(dev);
1240 }
1241
1242 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1243         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1244         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1245         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1246         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1247         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1248         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1249         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1250         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1251         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1252         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1253         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1254 };
1255
1256 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1257 {
1258         if (tb[IFLA_ADDRESS]) {
1259                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1260                         return -EINVAL;
1261
1262                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1263                         return -EADDRNOTAVAIL;
1264         }
1265
1266         if (!data)
1267                 return -EINVAL;
1268
1269         if (data[IFLA_GENEVE_ID]) {
1270                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1271
1272                 if (vni >= GENEVE_VID_MASK)
1273                         return -ERANGE;
1274         }
1275
1276         return 0;
1277 }
1278
1279 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1280                                           __be16 dst_port,
1281                                           union geneve_addr *remote,
1282                                           u8 vni[],
1283                                           bool *tun_on_same_port,
1284                                           bool *tun_collect_md)
1285 {
1286         struct geneve_dev *geneve, *t;
1287
1288         *tun_on_same_port = false;
1289         *tun_collect_md = false;
1290         t = NULL;
1291         list_for_each_entry(geneve, &gn->geneve_list, next) {
1292                 if (geneve->dst_port == dst_port) {
1293                         *tun_collect_md = geneve->collect_md;
1294                         *tun_on_same_port = true;
1295                 }
1296                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1297                     !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1298                     dst_port == geneve->dst_port)
1299                         t = geneve;
1300         }
1301         return t;
1302 }
1303
1304 static int geneve_configure(struct net *net, struct net_device *dev,
1305                             union geneve_addr *remote,
1306                             __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1307                             __be16 dst_port, bool metadata, u32 flags)
1308 {
1309         struct geneve_net *gn = net_generic(net, geneve_net_id);
1310         struct geneve_dev *t, *geneve = netdev_priv(dev);
1311         bool tun_collect_md, tun_on_same_port;
1312         int err, encap_len;
1313
1314         if (!remote)
1315                 return -EINVAL;
1316         if (metadata &&
1317             (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1318                 return -EINVAL;
1319
1320         geneve->net = net;
1321         geneve->dev = dev;
1322
1323         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1324         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1325         geneve->vni[2] =  vni & 0x000000ff;
1326
1327         if ((remote->sa.sa_family == AF_INET &&
1328              IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1329             (remote->sa.sa_family == AF_INET6 &&
1330              ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1331                 return -EINVAL;
1332         if (label && remote->sa.sa_family != AF_INET6)
1333                 return -EINVAL;
1334
1335         geneve->remote = *remote;
1336
1337         geneve->ttl = ttl;
1338         geneve->tos = tos;
1339         geneve->label = label;
1340         geneve->dst_port = dst_port;
1341         geneve->collect_md = metadata;
1342         geneve->flags = flags;
1343
1344         t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1345                             &tun_on_same_port, &tun_collect_md);
1346         if (t)
1347                 return -EBUSY;
1348
1349         /* make enough headroom for basic scenario */
1350         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1351         if (remote->sa.sa_family == AF_INET)
1352                 encap_len += sizeof(struct iphdr);
1353         else
1354                 encap_len += sizeof(struct ipv6hdr);
1355         dev->needed_headroom = encap_len + ETH_HLEN;
1356
1357         if (metadata) {
1358                 if (tun_on_same_port)
1359                         return -EPERM;
1360         } else {
1361                 if (tun_collect_md)
1362                         return -EPERM;
1363         }
1364
1365         dst_cache_reset(&geneve->dst_cache);
1366
1367         err = register_netdevice(dev);
1368         if (err)
1369                 return err;
1370
1371         list_add(&geneve->next, &gn->geneve_list);
1372         return 0;
1373 }
1374
1375 static int geneve_newlink(struct net *net, struct net_device *dev,
1376                           struct nlattr *tb[], struct nlattr *data[])
1377 {
1378         __be16 dst_port = htons(GENEVE_UDP_PORT);
1379         __u8 ttl = 0, tos = 0;
1380         bool metadata = false;
1381         union geneve_addr remote = geneve_remote_unspec;
1382         __be32 label = 0;
1383         __u32 vni = 0;
1384         u32 flags = 0;
1385
1386         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1387                 return -EINVAL;
1388
1389         if (data[IFLA_GENEVE_REMOTE]) {
1390                 remote.sa.sa_family = AF_INET;
1391                 remote.sin.sin_addr.s_addr =
1392                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1393         }
1394
1395         if (data[IFLA_GENEVE_REMOTE6]) {
1396                 if (!IS_ENABLED(CONFIG_IPV6))
1397                         return -EPFNOSUPPORT;
1398
1399                 remote.sa.sa_family = AF_INET6;
1400                 remote.sin6.sin6_addr =
1401                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1402
1403                 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1404                     IPV6_ADDR_LINKLOCAL) {
1405                         netdev_dbg(dev, "link-local remote is unsupported\n");
1406                         return -EINVAL;
1407                 }
1408         }
1409
1410         if (data[IFLA_GENEVE_ID])
1411                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1412
1413         if (data[IFLA_GENEVE_TTL])
1414                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1415
1416         if (data[IFLA_GENEVE_TOS])
1417                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1418
1419         if (data[IFLA_GENEVE_LABEL])
1420                 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1421                         IPV6_FLOWLABEL_MASK;
1422
1423         if (data[IFLA_GENEVE_PORT])
1424                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1425
1426         if (data[IFLA_GENEVE_COLLECT_METADATA])
1427                 metadata = true;
1428
1429         if (data[IFLA_GENEVE_UDP_CSUM] &&
1430             !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1431                 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1432
1433         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1434             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1435                 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1436
1437         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1438             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1439                 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1440
1441         return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1442                                 dst_port, metadata, flags);
1443 }
1444
1445 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1446 {
1447         struct geneve_dev *geneve = netdev_priv(dev);
1448
1449         list_del(&geneve->next);
1450         unregister_netdevice_queue(dev, head);
1451 }
1452
1453 static size_t geneve_get_size(const struct net_device *dev)
1454 {
1455         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1456                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1457                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1458                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1459                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1460                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1461                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1462                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1463                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1464                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1465                 0;
1466 }
1467
1468 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1469 {
1470         struct geneve_dev *geneve = netdev_priv(dev);
1471         __u32 vni;
1472
1473         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1474         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1475                 goto nla_put_failure;
1476
1477         if (geneve->remote.sa.sa_family == AF_INET) {
1478                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1479                                     geneve->remote.sin.sin_addr.s_addr))
1480                         goto nla_put_failure;
1481 #if IS_ENABLED(CONFIG_IPV6)
1482         } else {
1483                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1484                                      &geneve->remote.sin6.sin6_addr))
1485                         goto nla_put_failure;
1486 #endif
1487         }
1488
1489         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1490             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1491             nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1492                 goto nla_put_failure;
1493
1494         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1495                 goto nla_put_failure;
1496
1497         if (geneve->collect_md) {
1498                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1499                         goto nla_put_failure;
1500         }
1501
1502         if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1503                        !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1504             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1505                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1506             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1507                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1508                 goto nla_put_failure;
1509
1510         return 0;
1511
1512 nla_put_failure:
1513         return -EMSGSIZE;
1514 }
1515
1516 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1517         .kind           = "geneve",
1518         .maxtype        = IFLA_GENEVE_MAX,
1519         .policy         = geneve_policy,
1520         .priv_size      = sizeof(struct geneve_dev),
1521         .setup          = geneve_setup,
1522         .validate       = geneve_validate,
1523         .newlink        = geneve_newlink,
1524         .dellink        = geneve_dellink,
1525         .get_size       = geneve_get_size,
1526         .fill_info      = geneve_fill_info,
1527 };
1528
1529 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1530                                         u8 name_assign_type, u16 dst_port)
1531 {
1532         struct nlattr *tb[IFLA_MAX + 1];
1533         struct net_device *dev;
1534         int err;
1535
1536         memset(tb, 0, sizeof(tb));
1537         dev = rtnl_create_link(net, name, name_assign_type,
1538                                &geneve_link_ops, tb);
1539         if (IS_ERR(dev))
1540                 return dev;
1541
1542         err = geneve_configure(net, dev, &geneve_remote_unspec,
1543                                0, 0, 0, 0, htons(dst_port), true,
1544                                GENEVE_F_UDP_ZERO_CSUM6_RX);
1545         if (err)
1546                 goto err;
1547
1548         /* openvswitch users expect packet sizes to be unrestricted,
1549          * so set the largest MTU we can.
1550          */
1551         err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1552         if (err)
1553                 goto err;
1554
1555         return dev;
1556
1557  err:
1558         free_netdev(dev);
1559         return ERR_PTR(err);
1560 }
1561 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1562
1563 static __net_init int geneve_init_net(struct net *net)
1564 {
1565         struct geneve_net *gn = net_generic(net, geneve_net_id);
1566
1567         INIT_LIST_HEAD(&gn->geneve_list);
1568         INIT_LIST_HEAD(&gn->sock_list);
1569         return 0;
1570 }
1571
1572 static void __net_exit geneve_exit_net(struct net *net)
1573 {
1574         struct geneve_net *gn = net_generic(net, geneve_net_id);
1575         struct geneve_dev *geneve, *next;
1576         struct net_device *dev, *aux;
1577         LIST_HEAD(list);
1578
1579         rtnl_lock();
1580
1581         /* gather any geneve devices that were moved into this ns */
1582         for_each_netdev_safe(net, dev, aux)
1583                 if (dev->rtnl_link_ops == &geneve_link_ops)
1584                         unregister_netdevice_queue(dev, &list);
1585
1586         /* now gather any other geneve devices that were created in this ns */
1587         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1588                 /* If geneve->dev is in the same netns, it was already added
1589                  * to the list by the previous loop.
1590                  */
1591                 if (!net_eq(dev_net(geneve->dev), net))
1592                         unregister_netdevice_queue(geneve->dev, &list);
1593         }
1594
1595         /* unregister the devices gathered above */
1596         unregister_netdevice_many(&list);
1597         rtnl_unlock();
1598 }
1599
1600 static struct pernet_operations geneve_net_ops = {
1601         .init = geneve_init_net,
1602         .exit = geneve_exit_net,
1603         .id   = &geneve_net_id,
1604         .size = sizeof(struct geneve_net),
1605 };
1606
1607 static int __init geneve_init_module(void)
1608 {
1609         int rc;
1610
1611         rc = register_pernet_subsys(&geneve_net_ops);
1612         if (rc)
1613                 goto out1;
1614
1615         rc = rtnl_link_register(&geneve_link_ops);
1616         if (rc)
1617                 goto out2;
1618
1619         return 0;
1620 out2:
1621         unregister_pernet_subsys(&geneve_net_ops);
1622 out1:
1623         return rc;
1624 }
1625 late_initcall(geneve_init_module);
1626
1627 static void __exit geneve_cleanup_module(void)
1628 {
1629         rtnl_link_unregister(&geneve_link_ops);
1630         unregister_pernet_subsys(&geneve_net_ops);
1631 }
1632 module_exit(geneve_cleanup_module);
1633
1634 MODULE_LICENSE("GPL");
1635 MODULE_VERSION(GENEVE_NETDEV_VER);
1636 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1637 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1638 MODULE_ALIAS_RTNL_LINK("geneve");