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 error;
 339
 340         /* Return packets with reserved bits set */
 341         geneveh = geneve_hdr(skb);
 342         if (unlikely(geneveh->ver != GENEVE_VER))
 343                 goto error;
 344
 345         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 346                 goto error;
 347
 348         gs = rcu_dereference_sk_user_data(sk);
 349         if (!gs)
 350                 goto drop;
 351
 352         geneve = geneve_lookup_skb(gs, skb);
 353         if (!geneve)
 354                 goto drop;
 355
 356         opts_len = geneveh->opt_len * 4;
 357         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 358                                  htons(ETH_P_TEB),
 359                                  !net_eq(geneve->net, dev_net(geneve->dev))))
 360                 goto drop;
 361
 362         geneve_rx(geneve, gs, skb);
 363         return 0;
 364
 365 drop:
 366         /* Consume bad packet */
 367         kfree_skb(skb);
 368         return 0;
 369
 370 error:
 371         /* Let the UDP layer deal with the skb */
 372         return 1;
 373 }
 374
 375 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 376                                          __be16 port, u32 flags)
 377 {
 378         struct socket *sock;
 379         struct udp_port_cfg udp_conf;
 380         int err;
 381
 382         memset(&udp_conf, 0, sizeof(udp_conf));
 383
 384         if (ipv6) {
 385                 udp_conf.family = AF_INET6;
 386                 udp_conf.ipv6_v6only = 1;
 387                 udp_conf.use_udp6_rx_checksums =
 388                     !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX);
 389         } else {
 390                 udp_conf.family = AF_INET;
 391                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 392         }
 393
 394         udp_conf.local_udp_port = port;
 395
 396         /* Open UDP socket */
 397         err = udp_sock_create(net, &udp_conf, &sock);
 398         if (err < 0)
 399                 return ERR_PTR(err);
 400
 401         return sock;
 402 }
 403
 404 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
 405 {
 406         struct net_device *dev;
 407         struct sock *sk = gs->sock->sk;
 408         struct net *net = sock_net(sk);
 409         sa_family_t sa_family = geneve_get_sk_family(gs);
 410         __be16 port = inet_sk(sk)->inet_sport;
 411
 412         rcu_read_lock();
 413         for_each_netdev_rcu(net, dev) {
 414                 if (dev->netdev_ops->ndo_add_geneve_port)
 415                         dev->netdev_ops->ndo_add_geneve_port(dev, sa_family,
 416                                                              port);
 417         }
 418         rcu_read_unlock();
 419 }
 420
 421 static int geneve_hlen(struct genevehdr *gh)
 422 {
 423         return sizeof(*gh) + gh->opt_len * 4;
 424 }
 425
 426 static struct sk_buff **geneve_gro_receive(struct sock *sk,
 427                                            struct sk_buff **head,
 428                                            struct sk_buff *skb)
 429 {
 430         struct sk_buff *p, **pp = NULL;
 431         struct genevehdr *gh, *gh2;
 432         unsigned int hlen, gh_len, off_gnv;
 433         const struct packet_offload *ptype;
 434         __be16 type;
 435         int flush = 1;
 436
 437         off_gnv = skb_gro_offset(skb);
 438         hlen = off_gnv + sizeof(*gh);
 439         gh = skb_gro_header_fast(skb, off_gnv);
 440         if (skb_gro_header_hard(skb, hlen)) {
 441                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 442                 if (unlikely(!gh))
 443                         goto out;
 444         }
 445
 446         if (gh->ver != GENEVE_VER || gh->oam)
 447                 goto out;
 448         gh_len = geneve_hlen(gh);
 449
 450         hlen = off_gnv + gh_len;
 451         if (skb_gro_header_hard(skb, hlen)) {
 452                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 453                 if (unlikely(!gh))
 454                         goto out;
 455         }
 456
 457         for (p = *head; p; p = p->next) {
 458                 if (!NAPI_GRO_CB(p)->same_flow)
 459                         continue;
 460
 461                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 462                 if (gh->opt_len != gh2->opt_len ||
 463                     memcmp(gh, gh2, gh_len)) {
 464                         NAPI_GRO_CB(p)->same_flow = 0;
 465                         continue;
 466                 }
 467         }
 468
 469         type = gh->proto_type;
 470
 471         rcu_read_lock();
 472         ptype = gro_find_receive_by_type(type);
 473         if (!ptype)
 474                 goto out_unlock;
 475
 476         skb_gro_pull(skb, gh_len);
 477         skb_gro_postpull_rcsum(skb, gh, gh_len);
 478         pp = ptype->callbacks.gro_receive(head, skb);
 479         flush = 0;
 480
 481 out_unlock:
 482         rcu_read_unlock();
 483 out:
 484         NAPI_GRO_CB(skb)->flush |= flush;
 485
 486         return pp;
 487 }
 488
 489 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 490                                int nhoff)
 491 {
 492         struct genevehdr *gh;
 493         struct packet_offload *ptype;
 494         __be16 type;
 495         int gh_len;
 496         int err = -ENOSYS;
 497
 498         udp_tunnel_gro_complete(skb, nhoff);
 499
 500         gh = (struct genevehdr *)(skb->data + nhoff);
 501         gh_len = geneve_hlen(gh);
 502         type = gh->proto_type;
 503
 504         rcu_read_lock();
 505         ptype = gro_find_complete_by_type(type);
 506         if (ptype)
 507                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 508
 509         rcu_read_unlock();
 510         return err;
 511 }
 512
 513 /* Create new listen socket if needed */
 514 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 515                                                 bool ipv6, u32 flags)
 516 {
 517         struct geneve_net *gn = net_generic(net, geneve_net_id);
 518         struct geneve_sock *gs;
 519         struct socket *sock;
 520         struct udp_tunnel_sock_cfg tunnel_cfg;
 521         int h;
 522
 523         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 524         if (!gs)
 525                 return ERR_PTR(-ENOMEM);
 526
 527         sock = geneve_create_sock(net, ipv6, port, flags);
 528         if (IS_ERR(sock)) {
 529                 kfree(gs);
 530                 return ERR_CAST(sock);
 531         }
 532
 533         gs->sock = sock;
 534         gs->refcnt = 1;
 535         for (h = 0; h < VNI_HASH_SIZE; ++h)
 536                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 537
 538         /* Initialize the geneve udp offloads structure */
 539         geneve_notify_add_rx_port(gs);
 540
 541         /* Mark socket as an encapsulation socket */
 542         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 543         tunnel_cfg.sk_user_data = gs;
 544         tunnel_cfg.encap_type = 1;
 545         tunnel_cfg.gro_receive = geneve_gro_receive;
 546         tunnel_cfg.gro_complete = geneve_gro_complete;
 547         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 548         tunnel_cfg.encap_destroy = NULL;
 549         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 550         list_add(&gs->list, &gn->sock_list);
 551         return gs;
 552 }
 553
 554 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 555 {
 556         struct net_device *dev;
 557         struct sock *sk = gs->sock->sk;
 558         struct net *net = sock_net(sk);
 559         sa_family_t sa_family = geneve_get_sk_family(gs);
 560         __be16 port = inet_sk(sk)->inet_sport;
 561
 562         rcu_read_lock();
 563         for_each_netdev_rcu(net, dev) {
 564                 if (dev->netdev_ops->ndo_del_geneve_port)
 565                         dev->netdev_ops->ndo_del_geneve_port(dev, sa_family,
 566                                                              port);
 567         }
 568
 569         rcu_read_unlock();
 570 }
 571
 572 static void __geneve_sock_release(struct geneve_sock *gs)
 573 {
 574         if (!gs || --gs->refcnt)
 575                 return;
 576
 577         list_del(&gs->list);
 578         geneve_notify_del_rx_port(gs);
 579         udp_tunnel_sock_release(gs->sock);
 580         kfree_rcu(gs, rcu);
 581 }
 582
 583 static void geneve_sock_release(struct geneve_dev *geneve)
 584 {
 585         __geneve_sock_release(geneve->sock4);
 586 #if IS_ENABLED(CONFIG_IPV6)
 587         __geneve_sock_release(geneve->sock6);
 588 #endif
 589 }
 590
 591 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 592                                             sa_family_t family,
 593                                             __be16 dst_port)
 594 {
 595         struct geneve_sock *gs;
 596
 597         list_for_each_entry(gs, &gn->sock_list, list) {
 598                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 599                     geneve_get_sk_family(gs) == family) {
 600                         return gs;
 601                 }
 602         }
 603         return NULL;
 604 }
 605
 606 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 607 {
 608         struct net *net = geneve->net;
 609         struct geneve_net *gn = net_generic(net, geneve_net_id);
 610         struct geneve_sock *gs;
 611         __u32 hash;
 612
 613         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
 614         if (gs) {
 615                 gs->refcnt++;
 616                 goto out;
 617         }
 618
 619         gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
 620         if (IS_ERR(gs))
 621                 return PTR_ERR(gs);
 622
 623 out:
 624         gs->collect_md = geneve->collect_md;
 625         gs->flags = geneve->flags;
 626 #if IS_ENABLED(CONFIG_IPV6)
 627         if (ipv6)
 628                 geneve->sock6 = gs;
 629         else
 630 #endif
 631                 geneve->sock4 = gs;
 632
 633         hash = geneve_net_vni_hash(geneve->vni);
 634         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 635         return 0;
 636 }
 637
 638 static int geneve_open(struct net_device *dev)
 639 {
 640         struct geneve_dev *geneve = netdev_priv(dev);
 641         bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
 642         bool metadata = geneve->collect_md;
 643         int ret = 0;
 644
 645         geneve->sock4 = NULL;
 646 #if IS_ENABLED(CONFIG_IPV6)
 647         geneve->sock6 = NULL;
 648         if (ipv6 || metadata)
 649                 ret = geneve_sock_add(geneve, true);
 650 #endif
 651         if (!ret && (!ipv6 || metadata))
 652                 ret = geneve_sock_add(geneve, false);
 653         if (ret < 0)
 654                 geneve_sock_release(geneve);
 655
 656         return ret;
 657 }
 658
 659 static int geneve_stop(struct net_device *dev)
 660 {
 661         struct geneve_dev *geneve = netdev_priv(dev);
 662
 663         if (!hlist_unhashed(&geneve->hlist))
 664                 hlist_del_rcu(&geneve->hlist);
 665         geneve_sock_release(geneve);
 666         return 0;
 667 }
 668
 669 static void geneve_build_header(struct genevehdr *geneveh,
 670                                 __be16 tun_flags, u8 vni[3],
 671                                 u8 options_len, u8 *options)
 672 {
 673         geneveh->ver = GENEVE_VER;
 674         geneveh->opt_len = options_len / 4;
 675         geneveh->oam = !!(tun_flags & TUNNEL_OAM);
 676         geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 677         geneveh->rsvd1 = 0;
 678         memcpy(geneveh->vni, vni, 3);
 679         geneveh->proto_type = htons(ETH_P_TEB);
 680         geneveh->rsvd2 = 0;
 681
 682         memcpy(geneveh->options, options, options_len);
 683 }
 684
 685 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 686                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 687                             u32 flags, bool xnet)
 688 {
 689         struct genevehdr *gnvh;
 690         int min_headroom;
 691         int err;
 692         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
 693
 694         skb_scrub_packet(skb, xnet);
 695
 696         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 697                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 698         err = skb_cow_head(skb, min_headroom);
 699         if (unlikely(err)) {
 700                 kfree_skb(skb);
 701                 goto free_rt;
 702         }
 703
 704         skb = udp_tunnel_handle_offloads(skb, udp_sum);
 705         if (IS_ERR(skb)) {
 706                 err = PTR_ERR(skb);
 707                 goto free_rt;
 708         }
 709
 710         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 711         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 712
 713         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 714         return 0;
 715
 716 free_rt:
 717         ip_rt_put(rt);
 718         return err;
 719 }
 720
 721 #if IS_ENABLED(CONFIG_IPV6)
 722 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 723                              __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 724                              u32 flags, bool xnet)
 725 {
 726         struct genevehdr *gnvh;
 727         int min_headroom;
 728         int err;
 729         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
 730
 731         skb_scrub_packet(skb, xnet);
 732
 733         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
 734                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
 735         err = skb_cow_head(skb, min_headroom);
 736         if (unlikely(err)) {
 737                 kfree_skb(skb);
 738                 goto free_dst;
 739         }
 740
 741         skb = udp_tunnel_handle_offloads(skb, udp_sum);
 742         if (IS_ERR(skb)) {
 743                 err = PTR_ERR(skb);
 744                 goto free_dst;
 745         }
 746
 747         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 748         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 749
 750         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 751         return 0;
 752
 753 free_dst:
 754         dst_release(dst);
 755         return err;
 756 }
 757 #endif
 758
 759 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 760                                        struct net_device *dev,
 761                                        struct flowi4 *fl4,
 762                                        struct ip_tunnel_info *info)
 763 {
 764         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 765         struct geneve_dev *geneve = netdev_priv(dev);
 766         struct dst_cache *dst_cache;
 767         struct rtable *rt = NULL;
 768         __u8 tos;
 769
 770         memset(fl4, 0, sizeof(*fl4));
 771         fl4->flowi4_mark = skb->mark;
 772         fl4->flowi4_proto = IPPROTO_UDP;
 773
 774         if (info) {
 775                 fl4->daddr = info->key.u.ipv4.dst;
 776                 fl4->saddr = info->key.u.ipv4.src;
 777                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 778                 dst_cache = &info->dst_cache;
 779         } else {
 780                 tos = geneve->tos;
 781                 if (tos == 1) {
 782                         const struct iphdr *iip = ip_hdr(skb);
 783
 784                         tos = ip_tunnel_get_dsfield(iip, skb);
 785                         use_cache = false;
 786                 }
 787
 788                 fl4->flowi4_tos = RT_TOS(tos);
 789                 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
 790                 dst_cache = &geneve->dst_cache;
 791         }
 792
 793         if (use_cache) {
 794                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 795                 if (rt)
 796                         return rt;
 797         }
 798
 799         rt = ip_route_output_key(geneve->net, fl4);
 800         if (IS_ERR(rt)) {
 801                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 802                 return ERR_PTR(-ENETUNREACH);
 803         }
 804         if (rt->dst.dev == dev) { /* is this necessary? */
 805                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 806                 ip_rt_put(rt);
 807                 return ERR_PTR(-ELOOP);
 808         }
 809         if (use_cache)
 810                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 811         return rt;
 812 }
 813
 814 #if IS_ENABLED(CONFIG_IPV6)
 815 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 816                                            struct net_device *dev,
 817                                            struct flowi6 *fl6,
 818                                            struct ip_tunnel_info *info)
 819 {
 820         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 821         struct geneve_dev *geneve = netdev_priv(dev);
 822         struct geneve_sock *gs6 = geneve->sock6;
 823         struct dst_entry *dst = NULL;
 824         struct dst_cache *dst_cache;
 825         __u8 prio;
 826
 827         memset(fl6, 0, sizeof(*fl6));
 828         fl6->flowi6_mark = skb->mark;
 829         fl6->flowi6_proto = IPPROTO_UDP;
 830
 831         if (info) {
 832                 fl6->daddr = info->key.u.ipv6.dst;
 833                 fl6->saddr = info->key.u.ipv6.src;
 834                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(info->key.tos),
 835                                                    info->key.label);
 836                 dst_cache = &info->dst_cache;
 837         } else {
 838                 prio = geneve->tos;
 839                 if (prio == 1) {
 840                         const struct iphdr *iip = ip_hdr(skb);
 841
 842                         prio = ip_tunnel_get_dsfield(iip, skb);
 843                         use_cache = false;
 844                 }
 845
 846                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 847                                                    geneve->label);
 848                 fl6->daddr = geneve->remote.sin6.sin6_addr;
 849                 dst_cache = &geneve->dst_cache;
 850         }
 851
 852         if (use_cache) {
 853                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 854                 if (dst)
 855                         return dst;
 856         }
 857
 858         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 859                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 860                 return ERR_PTR(-ENETUNREACH);
 861         }
 862         if (dst->dev == dev) { /* is this necessary? */
 863                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 864                 dst_release(dst);
 865                 return ERR_PTR(-ELOOP);
 866         }
 867
 868         if (use_cache)
 869                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 870         return dst;
 871 }
 872 #endif
 873
 874 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 875 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 876 {
 877 #ifdef __BIG_ENDIAN
 878         vni[0] = (__force __u8)(tun_id >> 16);
 879         vni[1] = (__force __u8)(tun_id >> 8);
 880         vni[2] = (__force __u8)tun_id;
 881 #else
 882         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 883         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 884         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 885 #endif
 886 }
 887
 888 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 889                                    struct ip_tunnel_info *info)
 890 {
 891         struct geneve_dev *geneve = netdev_priv(dev);
 892         struct geneve_sock *gs4 = geneve->sock4;
 893         struct rtable *rt = NULL;
 894         const struct iphdr *iip; /* interior IP header */
 895         int err = -EINVAL;
 896         struct flowi4 fl4;
 897         __u8 tos, ttl;
 898         __be16 sport;
 899         __be16 df;
 900         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 901         u32 flags = geneve->flags;
 902
 903         if (geneve->collect_md) {
 904                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 905                         netdev_dbg(dev, "no tunnel metadata\n");
 906                         goto tx_error;
 907                 }
 908                 if (info && ip_tunnel_info_af(info) != AF_INET)
 909                         goto tx_error;
 910         }
 911
 912         rt = geneve_get_v4_rt(skb, dev, &fl4, info);
 913         if (IS_ERR(rt)) {
 914                 err = PTR_ERR(rt);
 915                 goto tx_error;
 916         }
 917
 918         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 919         skb_reset_mac_header(skb);
 920
 921         iip = ip_hdr(skb);
 922
 923         if (info) {
 924                 const struct ip_tunnel_key *key = &info->key;
 925                 u8 *opts = NULL;
 926                 u8 vni[3];
 927
 928                 tunnel_id_to_vni(key->tun_id, vni);
 929                 if (info->options_len)
 930                         opts = ip_tunnel_info_opts(info);
 931
 932                 if (key->tun_flags & TUNNEL_CSUM)
 933                         flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
 934                 else
 935                         flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
 936
 937                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 938                                        info->options_len, opts, flags, xnet);
 939                 if (unlikely(err))
 940                         goto err;
 941
 942                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
 943                 ttl = key->ttl;
 944                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 945         } else {
 946                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 947                                        0, NULL, flags, xnet);
 948                 if (unlikely(err))
 949                         goto err;
 950
 951                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 952                 ttl = geneve->ttl;
 953                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 954                         ttl = 1;
 955                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 956                 df = 0;
 957         }
 958         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 959                             tos, ttl, df, sport, geneve->dst_port,
 960                             !net_eq(geneve->net, dev_net(geneve->dev)),
 961                             !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
 962
 963         return NETDEV_TX_OK;
 964
 965 tx_error:
 966         dev_kfree_skb(skb);
 967 err:
 968         if (err == -ELOOP)
 969                 dev->stats.collisions++;
 970         else if (err == -ENETUNREACH)
 971                 dev->stats.tx_carrier_errors++;
 972         else
 973                 dev->stats.tx_errors++;
 974         return NETDEV_TX_OK;
 975 }
 976
 977 #if IS_ENABLED(CONFIG_IPV6)
 978 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 979                                     struct ip_tunnel_info *info)
 980 {
 981         struct geneve_dev *geneve = netdev_priv(dev);
 982         struct geneve_sock *gs6 = geneve->sock6;
 983         struct dst_entry *dst = NULL;
 984         const struct iphdr *iip; /* interior IP header */
 985         int err = -EINVAL;
 986         struct flowi6 fl6;
 987         __u8 prio, ttl;
 988         __be16 sport;
 989         __be32 label;
 990         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 991         u32 flags = geneve->flags;
 992
 993         if (geneve->collect_md) {
 994                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 995                         netdev_dbg(dev, "no tunnel metadata\n");
 996                         goto tx_error;
 997                 }
 998         }
 999
1000         dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1001         if (IS_ERR(dst)) {
1002                 err = PTR_ERR(dst);
1003                 goto tx_error;
1004         }
1005
1006         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1007         skb_reset_mac_header(skb);
1008
1009         iip = ip_hdr(skb);
1010
1011         if (info) {
1012                 const struct ip_tunnel_key *key = &info->key;
1013                 u8 *opts = NULL;
1014                 u8 vni[3];
1015
1016                 tunnel_id_to_vni(key->tun_id, vni);
1017                 if (info->options_len)
1018                         opts = ip_tunnel_info_opts(info);
1019
1020                 if (key->tun_flags & TUNNEL_CSUM)
1021                         flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
1022                 else
1023                         flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1024
1025                 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
1026                                         info->options_len, opts,
1027                                         flags, xnet);
1028                 if (unlikely(err))
1029                         goto err;
1030
1031                 prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
1032                 ttl = key->ttl;
1033                 label = info->key.label;
1034         } else {
1035                 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
1036                                         0, NULL, flags, xnet);
1037                 if (unlikely(err))
1038                         goto err;
1039
1040                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1041                                            iip, skb);
1042                 ttl = geneve->ttl;
1043                 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1044                         ttl = 1;
1045                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1046                 label = geneve->label;
1047         }
1048
1049         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1050                              &fl6.saddr, &fl6.daddr, prio, ttl, label,
1051                              sport, geneve->dst_port,
1052                              !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1053         return NETDEV_TX_OK;
1054
1055 tx_error:
1056         dev_kfree_skb(skb);
1057 err:
1058         if (err == -ELOOP)
1059                 dev->stats.collisions++;
1060         else if (err == -ENETUNREACH)
1061                 dev->stats.tx_carrier_errors++;
1062         else
1063                 dev->stats.tx_errors++;
1064         return NETDEV_TX_OK;
1065 }
1066 #endif
1067
1068 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1069 {
1070         struct geneve_dev *geneve = netdev_priv(dev);
1071         struct ip_tunnel_info *info = NULL;
1072
1073         if (geneve->collect_md)
1074                 info = skb_tunnel_info(skb);
1075
1076 #if IS_ENABLED(CONFIG_IPV6)
1077         if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1078             (!info && geneve->remote.sa.sa_family == AF_INET6))
1079                 return geneve6_xmit_skb(skb, dev, info);
1080 #endif
1081         return geneve_xmit_skb(skb, dev, info);
1082 }
1083
1084 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1085 {
1086         /* The max_mtu calculation does not take account of GENEVE
1087          * options, to avoid excluding potentially valid
1088          * configurations.
1089          */
1090         int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - sizeof(struct iphdr)
1091                 - dev->hard_header_len;
1092
1093         if (new_mtu < 68)
1094                 return -EINVAL;
1095
1096         if (new_mtu > max_mtu) {
1097                 if (strict)
1098                         return -EINVAL;
1099
1100                 new_mtu = max_mtu;
1101         }
1102
1103         dev->mtu = new_mtu;
1104         return 0;
1105 }
1106
1107 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1108 {
1109         return __geneve_change_mtu(dev, new_mtu, true);
1110 }
1111
1112 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1113 {
1114         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1115         struct geneve_dev *geneve = netdev_priv(dev);
1116         struct rtable *rt;
1117         struct flowi4 fl4;
1118 #if IS_ENABLED(CONFIG_IPV6)
1119         struct dst_entry *dst;
1120         struct flowi6 fl6;
1121 #endif
1122
1123         if (ip_tunnel_info_af(info) == AF_INET) {
1124                 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1125                 if (IS_ERR(rt))
1126                         return PTR_ERR(rt);
1127
1128                 ip_rt_put(rt);
1129                 info->key.u.ipv4.src = fl4.saddr;
1130 #if IS_ENABLED(CONFIG_IPV6)
1131         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1132                 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1133                 if (IS_ERR(dst))
1134                         return PTR_ERR(dst);
1135
1136                 dst_release(dst);
1137                 info->key.u.ipv6.src = fl6.saddr;
1138 #endif
1139         } else {
1140                 return -EINVAL;
1141         }
1142
1143         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1144                                              1, USHRT_MAX, true);
1145         info->key.tp_dst = geneve->dst_port;
1146         return 0;
1147 }
1148
1149 static const struct net_device_ops geneve_netdev_ops = {
1150         .ndo_init               = geneve_init,
1151         .ndo_uninit             = geneve_uninit,
1152         .ndo_open               = geneve_open,
1153         .ndo_stop               = geneve_stop,
1154         .ndo_start_xmit         = geneve_xmit,
1155         .ndo_get_stats64        = ip_tunnel_get_stats64,
1156         .ndo_change_mtu         = geneve_change_mtu,
1157         .ndo_validate_addr      = eth_validate_addr,
1158         .ndo_set_mac_address    = eth_mac_addr,
1159         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1160 };
1161
1162 static void geneve_get_drvinfo(struct net_device *dev,
1163                                struct ethtool_drvinfo *drvinfo)
1164 {
1165         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1166         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1167 }
1168
1169 static const struct ethtool_ops geneve_ethtool_ops = {
1170         .get_drvinfo    = geneve_get_drvinfo,
1171         .get_link       = ethtool_op_get_link,
1172 };
1173
1174 /* Info for udev, that this is a virtual tunnel endpoint */
1175 static struct device_type geneve_type = {
1176         .name = "geneve",
1177 };
1178
1179 /* Calls the ndo_add_geneve_port of the caller in order to
1180  * supply the listening GENEVE udp ports. Callers are expected
1181  * to implement the ndo_add_geneve_port.
1182  */
1183 void geneve_get_rx_port(struct net_device *dev)
1184 {
1185         struct net *net = dev_net(dev);
1186         struct geneve_net *gn = net_generic(net, geneve_net_id);
1187         struct geneve_sock *gs;
1188         sa_family_t sa_family;
1189         struct sock *sk;
1190         __be16 port;
1191
1192         rcu_read_lock();
1193         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1194                 sk = gs->sock->sk;
1195                 sa_family = sk->sk_family;
1196                 port = inet_sk(sk)->inet_sport;
1197                 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family, port);
1198         }
1199         rcu_read_unlock();
1200 }
1201 EXPORT_SYMBOL_GPL(geneve_get_rx_port);
1202
1203 /* Initialize the device structure. */
1204 static void geneve_setup(struct net_device *dev)
1205 {
1206         ether_setup(dev);
1207
1208         dev->netdev_ops = &geneve_netdev_ops;
1209         dev->ethtool_ops = &geneve_ethtool_ops;
1210         dev->destructor = free_netdev;
1211
1212         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1213
1214         dev->features    |= NETIF_F_LLTX;
1215         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1216         dev->features    |= NETIF_F_RXCSUM;
1217         dev->features    |= NETIF_F_GSO_SOFTWARE;
1218
1219         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1220         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1221
1222         netif_keep_dst(dev);
1223         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1224         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1225         eth_hw_addr_random(dev);
1226 }
1227
1228 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1229         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1230         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1231         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1232         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1233         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1234         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1235         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1236         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1237         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1238         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1239         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1240 };
1241
1242 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1243 {
1244         if (tb[IFLA_ADDRESS]) {
1245                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1246                         return -EINVAL;
1247
1248                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1249                         return -EADDRNOTAVAIL;
1250         }
1251
1252         if (!data)
1253                 return -EINVAL;
1254
1255         if (data[IFLA_GENEVE_ID]) {
1256                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1257
1258                 if (vni >= GENEVE_VID_MASK)
1259                         return -ERANGE;
1260         }
1261
1262         return 0;
1263 }
1264
1265 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1266                                           __be16 dst_port,
1267                                           union geneve_addr *remote,
1268                                           u8 vni[],
1269                                           bool *tun_on_same_port,
1270                                           bool *tun_collect_md)
1271 {
1272         struct geneve_dev *geneve, *t;
1273
1274         *tun_on_same_port = false;
1275         *tun_collect_md = false;
1276         t = NULL;
1277         list_for_each_entry(geneve, &gn->geneve_list, next) {
1278                 if (geneve->dst_port == dst_port) {
1279                         *tun_collect_md = geneve->collect_md;
1280                         *tun_on_same_port = true;
1281                 }
1282                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1283                     !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1284                     dst_port == geneve->dst_port)
1285                         t = geneve;
1286         }
1287         return t;
1288 }
1289
1290 static int geneve_configure(struct net *net, struct net_device *dev,
1291                             union geneve_addr *remote,
1292                             __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1293                             __be16 dst_port, bool metadata, u32 flags)
1294 {
1295         struct geneve_net *gn = net_generic(net, geneve_net_id);
1296         struct geneve_dev *t, *geneve = netdev_priv(dev);
1297         bool tun_collect_md, tun_on_same_port;
1298         int err, encap_len;
1299
1300         if (!remote)
1301                 return -EINVAL;
1302         if (metadata &&
1303             (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1304                 return -EINVAL;
1305
1306         geneve->net = net;
1307         geneve->dev = dev;
1308
1309         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1310         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1311         geneve->vni[2] =  vni & 0x000000ff;
1312
1313         if ((remote->sa.sa_family == AF_INET &&
1314              IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1315             (remote->sa.sa_family == AF_INET6 &&
1316              ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1317                 return -EINVAL;
1318         if (label && remote->sa.sa_family != AF_INET6)
1319                 return -EINVAL;
1320
1321         geneve->remote = *remote;
1322
1323         geneve->ttl = ttl;
1324         geneve->tos = tos;
1325         geneve->label = label;
1326         geneve->dst_port = dst_port;
1327         geneve->collect_md = metadata;
1328         geneve->flags = flags;
1329
1330         t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1331                             &tun_on_same_port, &tun_collect_md);
1332         if (t)
1333                 return -EBUSY;
1334
1335         /* make enough headroom for basic scenario */
1336         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1337         if (remote->sa.sa_family == AF_INET)
1338                 encap_len += sizeof(struct iphdr);
1339         else
1340                 encap_len += sizeof(struct ipv6hdr);
1341         dev->needed_headroom = encap_len + ETH_HLEN;
1342
1343         if (metadata) {
1344                 if (tun_on_same_port)
1345                         return -EPERM;
1346         } else {
1347                 if (tun_collect_md)
1348                         return -EPERM;
1349         }
1350
1351         dst_cache_reset(&geneve->dst_cache);
1352
1353         err = register_netdevice(dev);
1354         if (err)
1355                 return err;
1356
1357         list_add(&geneve->next, &gn->geneve_list);
1358         return 0;
1359 }
1360
1361 static int geneve_newlink(struct net *net, struct net_device *dev,
1362                           struct nlattr *tb[], struct nlattr *data[])
1363 {
1364         __be16 dst_port = htons(GENEVE_UDP_PORT);
1365         __u8 ttl = 0, tos = 0;
1366         bool metadata = false;
1367         union geneve_addr remote = geneve_remote_unspec;
1368         __be32 label = 0;
1369         __u32 vni = 0;
1370         u32 flags = 0;
1371
1372         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1373                 return -EINVAL;
1374
1375         if (data[IFLA_GENEVE_REMOTE]) {
1376                 remote.sa.sa_family = AF_INET;
1377                 remote.sin.sin_addr.s_addr =
1378                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1379         }
1380
1381         if (data[IFLA_GENEVE_REMOTE6]) {
1382                 if (!IS_ENABLED(CONFIG_IPV6))
1383                         return -EPFNOSUPPORT;
1384
1385                 remote.sa.sa_family = AF_INET6;
1386                 remote.sin6.sin6_addr =
1387                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1388
1389                 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1390                     IPV6_ADDR_LINKLOCAL) {
1391                         netdev_dbg(dev, "link-local remote is unsupported\n");
1392                         return -EINVAL;
1393                 }
1394         }
1395
1396         if (data[IFLA_GENEVE_ID])
1397                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1398
1399         if (data[IFLA_GENEVE_TTL])
1400                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1401
1402         if (data[IFLA_GENEVE_TOS])
1403                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1404
1405         if (data[IFLA_GENEVE_LABEL])
1406                 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1407                         IPV6_FLOWLABEL_MASK;
1408
1409         if (data[IFLA_GENEVE_PORT])
1410                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1411
1412         if (data[IFLA_GENEVE_COLLECT_METADATA])
1413                 metadata = true;
1414
1415         if (data[IFLA_GENEVE_UDP_CSUM] &&
1416             !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1417                 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1418
1419         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1420             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1421                 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1422
1423         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1424             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1425                 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1426
1427         return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1428                                 dst_port, metadata, flags);
1429 }
1430
1431 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1432 {
1433         struct geneve_dev *geneve = netdev_priv(dev);
1434
1435         list_del(&geneve->next);
1436         unregister_netdevice_queue(dev, head);
1437 }
1438
1439 static size_t geneve_get_size(const struct net_device *dev)
1440 {
1441         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1442                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1443                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1444                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1445                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1446                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1447                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1448                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1449                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1450                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1451                 0;
1452 }
1453
1454 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1455 {
1456         struct geneve_dev *geneve = netdev_priv(dev);
1457         __u32 vni;
1458
1459         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1460         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1461                 goto nla_put_failure;
1462
1463         if (geneve->remote.sa.sa_family == AF_INET) {
1464                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1465                                     geneve->remote.sin.sin_addr.s_addr))
1466                         goto nla_put_failure;
1467 #if IS_ENABLED(CONFIG_IPV6)
1468         } else {
1469                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1470                                      &geneve->remote.sin6.sin6_addr))
1471                         goto nla_put_failure;
1472 #endif
1473         }
1474
1475         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1476             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1477             nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1478                 goto nla_put_failure;
1479
1480         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1481                 goto nla_put_failure;
1482
1483         if (geneve->collect_md) {
1484                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1485                         goto nla_put_failure;
1486         }
1487
1488         if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1489                        !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1490             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1491                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1492             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1493                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1494                 goto nla_put_failure;
1495
1496         return 0;
1497
1498 nla_put_failure:
1499         return -EMSGSIZE;
1500 }
1501
1502 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1503         .kind           = "geneve",
1504         .maxtype        = IFLA_GENEVE_MAX,
1505         .policy         = geneve_policy,
1506         .priv_size      = sizeof(struct geneve_dev),
1507         .setup          = geneve_setup,
1508         .validate       = geneve_validate,
1509         .newlink        = geneve_newlink,
1510         .dellink        = geneve_dellink,
1511         .get_size       = geneve_get_size,
1512         .fill_info      = geneve_fill_info,
1513 };
1514
1515 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1516                                         u8 name_assign_type, u16 dst_port)
1517 {
1518         struct nlattr *tb[IFLA_MAX + 1];
1519         struct net_device *dev;
1520         int err;
1521
1522         memset(tb, 0, sizeof(tb));
1523         dev = rtnl_create_link(net, name, name_assign_type,
1524                                &geneve_link_ops, tb);
1525         if (IS_ERR(dev))
1526                 return dev;
1527
1528         err = geneve_configure(net, dev, &geneve_remote_unspec,
1529                                0, 0, 0, 0, htons(dst_port), true,
1530                                GENEVE_F_UDP_ZERO_CSUM6_RX);
1531         if (err)
1532                 goto err;
1533
1534         /* openvswitch users expect packet sizes to be unrestricted,
1535          * so set the largest MTU we can.
1536          */
1537         err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1538         if (err)
1539                 goto err;
1540
1541         return dev;
1542
1543  err:
1544         free_netdev(dev);
1545         return ERR_PTR(err);
1546 }
1547 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1548
1549 static __net_init int geneve_init_net(struct net *net)
1550 {
1551         struct geneve_net *gn = net_generic(net, geneve_net_id);
1552
1553         INIT_LIST_HEAD(&gn->geneve_list);
1554         INIT_LIST_HEAD(&gn->sock_list);
1555         return 0;
1556 }
1557
1558 static void __net_exit geneve_exit_net(struct net *net)
1559 {
1560         struct geneve_net *gn = net_generic(net, geneve_net_id);
1561         struct geneve_dev *geneve, *next;
1562         struct net_device *dev, *aux;
1563         LIST_HEAD(list);
1564
1565         rtnl_lock();
1566
1567         /* gather any geneve devices that were moved into this ns */
1568         for_each_netdev_safe(net, dev, aux)
1569                 if (dev->rtnl_link_ops == &geneve_link_ops)
1570                         unregister_netdevice_queue(dev, &list);
1571
1572         /* now gather any other geneve devices that were created in this ns */
1573         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1574                 /* If geneve->dev is in the same netns, it was already added
1575                  * to the list by the previous loop.
1576                  */
1577                 if (!net_eq(dev_net(geneve->dev), net))
1578                         unregister_netdevice_queue(geneve->dev, &list);
1579         }
1580
1581         /* unregister the devices gathered above */
1582         unregister_netdevice_many(&list);
1583         rtnl_unlock();
1584 }
1585
1586 static struct pernet_operations geneve_net_ops = {
1587         .init = geneve_init_net,
1588         .exit = geneve_exit_net,
1589         .id   = &geneve_net_id,
1590         .size = sizeof(struct geneve_net),
1591 };
1592
1593 static int __init geneve_init_module(void)
1594 {
1595         int rc;
1596
1597         rc = register_pernet_subsys(&geneve_net_ops);
1598         if (rc)
1599                 goto out1;
1600
1601         rc = rtnl_link_register(&geneve_link_ops);
1602         if (rc)
1603                 goto out2;
1604
1605         return 0;
1606 out2:
1607         unregister_pernet_subsys(&geneve_net_ops);
1608 out1:
1609         return rc;
1610 }
1611 late_initcall(geneve_init_module);
1612
1613 static void __exit geneve_cleanup_module(void)
1614 {
1615         rtnl_link_unregister(&geneve_link_ops);
1616         unregister_pernet_subsys(&geneve_net_ops);
1617 }
1618 module_exit(geneve_cleanup_module);
1619
1620 MODULE_LICENSE("GPL");
1621 MODULE_VERSION(GENEVE_NETDEV_VER);
1622 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1623 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1624 MODULE_ALIAS_RTNL_LINK("geneve");