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         gh = (struct genevehdr *)(skb->data + nhoff);
 499         gh_len = geneve_hlen(gh);
 500         type = gh->proto_type;
 501
 502         rcu_read_lock();
 503         ptype = gro_find_complete_by_type(type);
 504         if (ptype)
 505                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 506
 507         rcu_read_unlock();
 508
 509         skb_set_inner_mac_header(skb, nhoff + gh_len);
 510
 511         return err;
 512 }
 513
 514 /* Create new listen socket if needed */
 515 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 516                                                 bool ipv6, u32 flags)
 517 {
 518         struct geneve_net *gn = net_generic(net, geneve_net_id);
 519         struct geneve_sock *gs;
 520         struct socket *sock;
 521         struct udp_tunnel_sock_cfg tunnel_cfg;
 522         int h;
 523
 524         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 525         if (!gs)
 526                 return ERR_PTR(-ENOMEM);
 527
 528         sock = geneve_create_sock(net, ipv6, port, flags);
 529         if (IS_ERR(sock)) {
 530                 kfree(gs);
 531                 return ERR_CAST(sock);
 532         }
 533
 534         gs->sock = sock;
 535         gs->refcnt = 1;
 536         for (h = 0; h < VNI_HASH_SIZE; ++h)
 537                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 538
 539         /* Initialize the geneve udp offloads structure */
 540         geneve_notify_add_rx_port(gs);
 541
 542         /* Mark socket as an encapsulation socket */
 543         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 544         tunnel_cfg.sk_user_data = gs;
 545         tunnel_cfg.encap_type = 1;
 546         tunnel_cfg.gro_receive = geneve_gro_receive;
 547         tunnel_cfg.gro_complete = geneve_gro_complete;
 548         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 549         tunnel_cfg.encap_destroy = NULL;
 550         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 551         list_add(&gs->list, &gn->sock_list);
 552         return gs;
 553 }
 554
 555 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 556 {
 557         struct net_device *dev;
 558         struct sock *sk = gs->sock->sk;
 559         struct net *net = sock_net(sk);
 560         sa_family_t sa_family = geneve_get_sk_family(gs);
 561         __be16 port = inet_sk(sk)->inet_sport;
 562
 563         rcu_read_lock();
 564         for_each_netdev_rcu(net, dev) {
 565                 if (dev->netdev_ops->ndo_del_geneve_port)
 566                         dev->netdev_ops->ndo_del_geneve_port(dev, sa_family,
 567                                                              port);
 568         }
 569
 570         rcu_read_unlock();
 571 }
 572
 573 static void __geneve_sock_release(struct geneve_sock *gs)
 574 {
 575         if (!gs || --gs->refcnt)
 576                 return;
 577
 578         list_del(&gs->list);
 579         geneve_notify_del_rx_port(gs);
 580         udp_tunnel_sock_release(gs->sock);
 581         kfree_rcu(gs, rcu);
 582 }
 583
 584 static void geneve_sock_release(struct geneve_dev *geneve)
 585 {
 586         __geneve_sock_release(geneve->sock4);
 587 #if IS_ENABLED(CONFIG_IPV6)
 588         __geneve_sock_release(geneve->sock6);
 589 #endif
 590 }
 591
 592 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 593                                             sa_family_t family,
 594                                             __be16 dst_port)
 595 {
 596         struct geneve_sock *gs;
 597
 598         list_for_each_entry(gs, &gn->sock_list, list) {
 599                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 600                     geneve_get_sk_family(gs) == family) {
 601                         return gs;
 602                 }
 603         }
 604         return NULL;
 605 }
 606
 607 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 608 {
 609         struct net *net = geneve->net;
 610         struct geneve_net *gn = net_generic(net, geneve_net_id);
 611         struct geneve_sock *gs;
 612         __u32 hash;
 613
 614         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
 615         if (gs) {
 616                 gs->refcnt++;
 617                 goto out;
 618         }
 619
 620         gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
 621         if (IS_ERR(gs))
 622                 return PTR_ERR(gs);
 623
 624 out:
 625         gs->collect_md = geneve->collect_md;
 626         gs->flags = geneve->flags;
 627 #if IS_ENABLED(CONFIG_IPV6)
 628         if (ipv6)
 629                 geneve->sock6 = gs;
 630         else
 631 #endif
 632                 geneve->sock4 = gs;
 633
 634         hash = geneve_net_vni_hash(geneve->vni);
 635         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 636         return 0;
 637 }
 638
 639 static int geneve_open(struct net_device *dev)
 640 {
 641         struct geneve_dev *geneve = netdev_priv(dev);
 642         bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
 643         bool metadata = geneve->collect_md;
 644         int ret = 0;
 645
 646         geneve->sock4 = NULL;
 647 #if IS_ENABLED(CONFIG_IPV6)
 648         geneve->sock6 = NULL;
 649         if (ipv6 || metadata)
 650                 ret = geneve_sock_add(geneve, true);
 651 #endif
 652         if (!ret && (!ipv6 || metadata))
 653                 ret = geneve_sock_add(geneve, false);
 654         if (ret < 0)
 655                 geneve_sock_release(geneve);
 656
 657         return ret;
 658 }
 659
 660 static int geneve_stop(struct net_device *dev)
 661 {
 662         struct geneve_dev *geneve = netdev_priv(dev);
 663
 664         if (!hlist_unhashed(&geneve->hlist))
 665                 hlist_del_rcu(&geneve->hlist);
 666         geneve_sock_release(geneve);
 667         return 0;
 668 }
 669
 670 static void geneve_build_header(struct genevehdr *geneveh,
 671                                 __be16 tun_flags, u8 vni[3],
 672                                 u8 options_len, u8 *options)
 673 {
 674         geneveh->ver = GENEVE_VER;
 675         geneveh->opt_len = options_len / 4;
 676         geneveh->oam = !!(tun_flags & TUNNEL_OAM);
 677         geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 678         geneveh->rsvd1 = 0;
 679         memcpy(geneveh->vni, vni, 3);
 680         geneveh->proto_type = htons(ETH_P_TEB);
 681         geneveh->rsvd2 = 0;
 682
 683         memcpy(geneveh->options, options, options_len);
 684 }
 685
 686 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 687                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 688                             u32 flags, bool xnet)
 689 {
 690         struct genevehdr *gnvh;
 691         int min_headroom;
 692         int err;
 693         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
 694
 695         skb_scrub_packet(skb, xnet);
 696
 697         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 698                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 699         err = skb_cow_head(skb, min_headroom);
 700         if (unlikely(err))
 701                 goto free_rt;
 702
 703         err = udp_tunnel_handle_offloads(skb, udp_sum);
 704         if (err)
 705                 goto free_rt;
 706
 707         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 708         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 709
 710         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 711         return 0;
 712
 713 free_rt:
 714         ip_rt_put(rt);
 715         return err;
 716 }
 717
 718 #if IS_ENABLED(CONFIG_IPV6)
 719 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 720                              __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 721                              u32 flags, bool xnet)
 722 {
 723         struct genevehdr *gnvh;
 724         int min_headroom;
 725         int err;
 726         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
 727
 728         skb_scrub_packet(skb, xnet);
 729
 730         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
 731                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
 732         err = skb_cow_head(skb, min_headroom);
 733         if (unlikely(err))
 734                 goto free_dst;
 735
 736         err = udp_tunnel_handle_offloads(skb, udp_sum);
 737         if (err)
 738                 goto free_dst;
 739
 740         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 741         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 742
 743         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 744         return 0;
 745
 746 free_dst:
 747         dst_release(dst);
 748         return err;
 749 }
 750 #endif
 751
 752 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 753                                        struct net_device *dev,
 754                                        struct flowi4 *fl4,
 755                                        struct ip_tunnel_info *info)
 756 {
 757         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 758         struct geneve_dev *geneve = netdev_priv(dev);
 759         struct dst_cache *dst_cache;
 760         struct rtable *rt = NULL;
 761         __u8 tos;
 762
 763         memset(fl4, 0, sizeof(*fl4));
 764         fl4->flowi4_mark = skb->mark;
 765         fl4->flowi4_proto = IPPROTO_UDP;
 766
 767         if (info) {
 768                 fl4->daddr = info->key.u.ipv4.dst;
 769                 fl4->saddr = info->key.u.ipv4.src;
 770                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 771                 dst_cache = &info->dst_cache;
 772         } else {
 773                 tos = geneve->tos;
 774                 if (tos == 1) {
 775                         const struct iphdr *iip = ip_hdr(skb);
 776
 777                         tos = ip_tunnel_get_dsfield(iip, skb);
 778                         use_cache = false;
 779                 }
 780
 781                 fl4->flowi4_tos = RT_TOS(tos);
 782                 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
 783                 dst_cache = &geneve->dst_cache;
 784         }
 785
 786         if (use_cache) {
 787                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 788                 if (rt)
 789                         return rt;
 790         }
 791
 792         rt = ip_route_output_key(geneve->net, fl4);
 793         if (IS_ERR(rt)) {
 794                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 795                 return ERR_PTR(-ENETUNREACH);
 796         }
 797         if (rt->dst.dev == dev) { /* is this necessary? */
 798                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 799                 ip_rt_put(rt);
 800                 return ERR_PTR(-ELOOP);
 801         }
 802         if (use_cache)
 803                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 804         return rt;
 805 }
 806
 807 #if IS_ENABLED(CONFIG_IPV6)
 808 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 809                                            struct net_device *dev,
 810                                            struct flowi6 *fl6,
 811                                            struct ip_tunnel_info *info)
 812 {
 813         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 814         struct geneve_dev *geneve = netdev_priv(dev);
 815         struct geneve_sock *gs6 = geneve->sock6;
 816         struct dst_entry *dst = NULL;
 817         struct dst_cache *dst_cache;
 818         __u8 prio;
 819
 820         memset(fl6, 0, sizeof(*fl6));
 821         fl6->flowi6_mark = skb->mark;
 822         fl6->flowi6_proto = IPPROTO_UDP;
 823
 824         if (info) {
 825                 fl6->daddr = info->key.u.ipv6.dst;
 826                 fl6->saddr = info->key.u.ipv6.src;
 827                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(info->key.tos),
 828                                                    info->key.label);
 829                 dst_cache = &info->dst_cache;
 830         } else {
 831                 prio = geneve->tos;
 832                 if (prio == 1) {
 833                         const struct iphdr *iip = ip_hdr(skb);
 834
 835                         prio = ip_tunnel_get_dsfield(iip, skb);
 836                         use_cache = false;
 837                 }
 838
 839                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 840                                                    geneve->label);
 841                 fl6->daddr = geneve->remote.sin6.sin6_addr;
 842                 dst_cache = &geneve->dst_cache;
 843         }
 844
 845         if (use_cache) {
 846                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 847                 if (dst)
 848                         return dst;
 849         }
 850
 851         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 852                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 853                 return ERR_PTR(-ENETUNREACH);
 854         }
 855         if (dst->dev == dev) { /* is this necessary? */
 856                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 857                 dst_release(dst);
 858                 return ERR_PTR(-ELOOP);
 859         }
 860
 861         if (use_cache)
 862                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 863         return dst;
 864 }
 865 #endif
 866
 867 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 868 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 869 {
 870 #ifdef __BIG_ENDIAN
 871         vni[0] = (__force __u8)(tun_id >> 16);
 872         vni[1] = (__force __u8)(tun_id >> 8);
 873         vni[2] = (__force __u8)tun_id;
 874 #else
 875         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 876         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 877         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 878 #endif
 879 }
 880
 881 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 882                                    struct ip_tunnel_info *info)
 883 {
 884         struct geneve_dev *geneve = netdev_priv(dev);
 885         struct geneve_sock *gs4 = geneve->sock4;
 886         struct rtable *rt = NULL;
 887         const struct iphdr *iip; /* interior IP header */
 888         int err = -EINVAL;
 889         struct flowi4 fl4;
 890         __u8 tos, ttl;
 891         __be16 sport;
 892         __be16 df;
 893         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 894         u32 flags = geneve->flags;
 895
 896         if (geneve->collect_md) {
 897                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 898                         netdev_dbg(dev, "no tunnel metadata\n");
 899                         goto tx_error;
 900                 }
 901                 if (info && ip_tunnel_info_af(info) != AF_INET)
 902                         goto tx_error;
 903         }
 904
 905         rt = geneve_get_v4_rt(skb, dev, &fl4, info);
 906         if (IS_ERR(rt)) {
 907                 err = PTR_ERR(rt);
 908                 goto tx_error;
 909         }
 910
 911         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 912         skb_reset_mac_header(skb);
 913
 914         iip = ip_hdr(skb);
 915
 916         if (info) {
 917                 const struct ip_tunnel_key *key = &info->key;
 918                 u8 *opts = NULL;
 919                 u8 vni[3];
 920
 921                 tunnel_id_to_vni(key->tun_id, vni);
 922                 if (info->options_len)
 923                         opts = ip_tunnel_info_opts(info);
 924
 925                 if (key->tun_flags & TUNNEL_CSUM)
 926                         flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
 927                 else
 928                         flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
 929
 930                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 931                                        info->options_len, opts, flags, xnet);
 932                 if (unlikely(err))
 933                         goto tx_error;
 934
 935                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
 936                 ttl = key->ttl;
 937                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 938         } else {
 939                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 940                                        0, NULL, flags, xnet);
 941                 if (unlikely(err))
 942                         goto tx_error;
 943
 944                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 945                 ttl = geneve->ttl;
 946                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 947                         ttl = 1;
 948                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 949                 df = 0;
 950         }
 951         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 952                             tos, ttl, df, sport, geneve->dst_port,
 953                             !net_eq(geneve->net, dev_net(geneve->dev)),
 954                             !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
 955
 956         return NETDEV_TX_OK;
 957
 958 tx_error:
 959         dev_kfree_skb(skb);
 960
 961         if (err == -ELOOP)
 962                 dev->stats.collisions++;
 963         else if (err == -ENETUNREACH)
 964                 dev->stats.tx_carrier_errors++;
 965         else
 966                 dev->stats.tx_errors++;
 967         return NETDEV_TX_OK;
 968 }
 969
 970 #if IS_ENABLED(CONFIG_IPV6)
 971 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 972                                     struct ip_tunnel_info *info)
 973 {
 974         struct geneve_dev *geneve = netdev_priv(dev);
 975         struct geneve_sock *gs6 = geneve->sock6;
 976         struct dst_entry *dst = NULL;
 977         const struct iphdr *iip; /* interior IP header */
 978         int err = -EINVAL;
 979         struct flowi6 fl6;
 980         __u8 prio, ttl;
 981         __be16 sport;
 982         __be32 label;
 983         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 984         u32 flags = geneve->flags;
 985
 986         if (geneve->collect_md) {
 987                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 988                         netdev_dbg(dev, "no tunnel metadata\n");
 989                         goto tx_error;
 990                 }
 991         }
 992
 993         dst = geneve_get_v6_dst(skb, dev, &fl6, info);
 994         if (IS_ERR(dst)) {
 995                 err = PTR_ERR(dst);
 996                 goto tx_error;
 997         }
 998
 999         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1000         skb_reset_mac_header(skb);
1001
1002         iip = ip_hdr(skb);
1003
1004         if (info) {
1005                 const struct ip_tunnel_key *key = &info->key;
1006                 u8 *opts = NULL;
1007                 u8 vni[3];
1008
1009                 tunnel_id_to_vni(key->tun_id, vni);
1010                 if (info->options_len)
1011                         opts = ip_tunnel_info_opts(info);
1012
1013                 if (key->tun_flags & TUNNEL_CSUM)
1014                         flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
1015                 else
1016                         flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1017
1018                 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
1019                                         info->options_len, opts,
1020                                         flags, xnet);
1021                 if (unlikely(err))
1022                         goto tx_error;
1023
1024                 prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
1025                 ttl = key->ttl;
1026                 label = info->key.label;
1027         } else {
1028                 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
1029                                         0, NULL, flags, xnet);
1030                 if (unlikely(err))
1031                         goto tx_error;
1032
1033                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1034                                            iip, skb);
1035                 ttl = geneve->ttl;
1036                 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1037                         ttl = 1;
1038                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1039                 label = geneve->label;
1040         }
1041
1042         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1043                              &fl6.saddr, &fl6.daddr, prio, ttl, label,
1044                              sport, geneve->dst_port,
1045                              !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1046         return NETDEV_TX_OK;
1047
1048 tx_error:
1049         dev_kfree_skb(skb);
1050
1051         if (err == -ELOOP)
1052                 dev->stats.collisions++;
1053         else if (err == -ENETUNREACH)
1054                 dev->stats.tx_carrier_errors++;
1055         else
1056                 dev->stats.tx_errors++;
1057         return NETDEV_TX_OK;
1058 }
1059 #endif
1060
1061 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1062 {
1063         struct geneve_dev *geneve = netdev_priv(dev);
1064         struct ip_tunnel_info *info = NULL;
1065
1066         if (geneve->collect_md)
1067                 info = skb_tunnel_info(skb);
1068
1069 #if IS_ENABLED(CONFIG_IPV6)
1070         if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1071             (!info && geneve->remote.sa.sa_family == AF_INET6))
1072                 return geneve6_xmit_skb(skb, dev, info);
1073 #endif
1074         return geneve_xmit_skb(skb, dev, info);
1075 }
1076
1077 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1078 {
1079         /* The max_mtu calculation does not take account of GENEVE
1080          * options, to avoid excluding potentially valid
1081          * configurations.
1082          */
1083         int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - sizeof(struct iphdr)
1084                 - dev->hard_header_len;
1085
1086         if (new_mtu < 68)
1087                 return -EINVAL;
1088
1089         if (new_mtu > max_mtu) {
1090                 if (strict)
1091                         return -EINVAL;
1092
1093                 new_mtu = max_mtu;
1094         }
1095
1096         dev->mtu = new_mtu;
1097         return 0;
1098 }
1099
1100 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1101 {
1102         return __geneve_change_mtu(dev, new_mtu, true);
1103 }
1104
1105 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1106 {
1107         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1108         struct geneve_dev *geneve = netdev_priv(dev);
1109         struct rtable *rt;
1110         struct flowi4 fl4;
1111 #if IS_ENABLED(CONFIG_IPV6)
1112         struct dst_entry *dst;
1113         struct flowi6 fl6;
1114 #endif
1115
1116         if (ip_tunnel_info_af(info) == AF_INET) {
1117                 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1118                 if (IS_ERR(rt))
1119                         return PTR_ERR(rt);
1120
1121                 ip_rt_put(rt);
1122                 info->key.u.ipv4.src = fl4.saddr;
1123 #if IS_ENABLED(CONFIG_IPV6)
1124         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1125                 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1126                 if (IS_ERR(dst))
1127                         return PTR_ERR(dst);
1128
1129                 dst_release(dst);
1130                 info->key.u.ipv6.src = fl6.saddr;
1131 #endif
1132         } else {
1133                 return -EINVAL;
1134         }
1135
1136         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1137                                              1, USHRT_MAX, true);
1138         info->key.tp_dst = geneve->dst_port;
1139         return 0;
1140 }
1141
1142 static const struct net_device_ops geneve_netdev_ops = {
1143         .ndo_init               = geneve_init,
1144         .ndo_uninit             = geneve_uninit,
1145         .ndo_open               = geneve_open,
1146         .ndo_stop               = geneve_stop,
1147         .ndo_start_xmit         = geneve_xmit,
1148         .ndo_get_stats64        = ip_tunnel_get_stats64,
1149         .ndo_change_mtu         = geneve_change_mtu,
1150         .ndo_validate_addr      = eth_validate_addr,
1151         .ndo_set_mac_address    = eth_mac_addr,
1152         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1153 };
1154
1155 static void geneve_get_drvinfo(struct net_device *dev,
1156                                struct ethtool_drvinfo *drvinfo)
1157 {
1158         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1159         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1160 }
1161
1162 static const struct ethtool_ops geneve_ethtool_ops = {
1163         .get_drvinfo    = geneve_get_drvinfo,
1164         .get_link       = ethtool_op_get_link,
1165 };
1166
1167 /* Info for udev, that this is a virtual tunnel endpoint */
1168 static struct device_type geneve_type = {
1169         .name = "geneve",
1170 };
1171
1172 /* Calls the ndo_add_geneve_port of the caller in order to
1173  * supply the listening GENEVE udp ports. Callers are expected
1174  * to implement the ndo_add_geneve_port.
1175  */
1176 static void geneve_push_rx_ports(struct net_device *dev)
1177 {
1178         struct net *net = dev_net(dev);
1179         struct geneve_net *gn = net_generic(net, geneve_net_id);
1180         struct geneve_sock *gs;
1181         sa_family_t sa_family;
1182         struct sock *sk;
1183         __be16 port;
1184
1185         if (!dev->netdev_ops->ndo_add_geneve_port)
1186                 return;
1187
1188         rcu_read_lock();
1189         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1190                 sk = gs->sock->sk;
1191                 sa_family = sk->sk_family;
1192                 port = inet_sk(sk)->inet_sport;
1193                 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family, port);
1194         }
1195         rcu_read_unlock();
1196 }
1197
1198 /* Initialize the device structure. */
1199 static void geneve_setup(struct net_device *dev)
1200 {
1201         ether_setup(dev);
1202
1203         dev->netdev_ops = &geneve_netdev_ops;
1204         dev->ethtool_ops = &geneve_ethtool_ops;
1205         dev->destructor = free_netdev;
1206
1207         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1208
1209         dev->features    |= NETIF_F_LLTX;
1210         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1211         dev->features    |= NETIF_F_RXCSUM;
1212         dev->features    |= NETIF_F_GSO_SOFTWARE;
1213
1214         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1215         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1216
1217         netif_keep_dst(dev);
1218         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1219         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1220         eth_hw_addr_random(dev);
1221 }
1222
1223 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1224         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1225         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1226         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1227         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1228         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1229         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1230         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1231         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1232         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1233         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1234         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1235 };
1236
1237 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1238 {
1239         if (tb[IFLA_ADDRESS]) {
1240                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1241                         return -EINVAL;
1242
1243                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1244                         return -EADDRNOTAVAIL;
1245         }
1246
1247         if (!data)
1248                 return -EINVAL;
1249
1250         if (data[IFLA_GENEVE_ID]) {
1251                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1252
1253                 if (vni >= GENEVE_VID_MASK)
1254                         return -ERANGE;
1255         }
1256
1257         return 0;
1258 }
1259
1260 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1261                                           __be16 dst_port,
1262                                           union geneve_addr *remote,
1263                                           u8 vni[],
1264                                           bool *tun_on_same_port,
1265                                           bool *tun_collect_md)
1266 {
1267         struct geneve_dev *geneve, *t;
1268
1269         *tun_on_same_port = false;
1270         *tun_collect_md = false;
1271         t = NULL;
1272         list_for_each_entry(geneve, &gn->geneve_list, next) {
1273                 if (geneve->dst_port == dst_port) {
1274                         *tun_collect_md = geneve->collect_md;
1275                         *tun_on_same_port = true;
1276                 }
1277                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1278                     !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1279                     dst_port == geneve->dst_port)
1280                         t = geneve;
1281         }
1282         return t;
1283 }
1284
1285 static int geneve_configure(struct net *net, struct net_device *dev,
1286                             union geneve_addr *remote,
1287                             __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1288                             __be16 dst_port, bool metadata, u32 flags)
1289 {
1290         struct geneve_net *gn = net_generic(net, geneve_net_id);
1291         struct geneve_dev *t, *geneve = netdev_priv(dev);
1292         bool tun_collect_md, tun_on_same_port;
1293         int err, encap_len;
1294
1295         if (!remote)
1296                 return -EINVAL;
1297         if (metadata &&
1298             (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1299                 return -EINVAL;
1300
1301         geneve->net = net;
1302         geneve->dev = dev;
1303
1304         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1305         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1306         geneve->vni[2] =  vni & 0x000000ff;
1307
1308         if ((remote->sa.sa_family == AF_INET &&
1309              IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1310             (remote->sa.sa_family == AF_INET6 &&
1311              ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1312                 return -EINVAL;
1313         if (label && remote->sa.sa_family != AF_INET6)
1314                 return -EINVAL;
1315
1316         geneve->remote = *remote;
1317
1318         geneve->ttl = ttl;
1319         geneve->tos = tos;
1320         geneve->label = label;
1321         geneve->dst_port = dst_port;
1322         geneve->collect_md = metadata;
1323         geneve->flags = flags;
1324
1325         t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1326                             &tun_on_same_port, &tun_collect_md);
1327         if (t)
1328                 return -EBUSY;
1329
1330         /* make enough headroom for basic scenario */
1331         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1332         if (remote->sa.sa_family == AF_INET)
1333                 encap_len += sizeof(struct iphdr);
1334         else
1335                 encap_len += sizeof(struct ipv6hdr);
1336         dev->needed_headroom = encap_len + ETH_HLEN;
1337
1338         if (metadata) {
1339                 if (tun_on_same_port)
1340                         return -EPERM;
1341         } else {
1342                 if (tun_collect_md)
1343                         return -EPERM;
1344         }
1345
1346         dst_cache_reset(&geneve->dst_cache);
1347
1348         err = register_netdevice(dev);
1349         if (err)
1350                 return err;
1351
1352         list_add(&geneve->next, &gn->geneve_list);
1353         return 0;
1354 }
1355
1356 static int geneve_newlink(struct net *net, struct net_device *dev,
1357                           struct nlattr *tb[], struct nlattr *data[])
1358 {
1359         __be16 dst_port = htons(GENEVE_UDP_PORT);
1360         __u8 ttl = 0, tos = 0;
1361         bool metadata = false;
1362         union geneve_addr remote = geneve_remote_unspec;
1363         __be32 label = 0;
1364         __u32 vni = 0;
1365         u32 flags = 0;
1366
1367         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1368                 return -EINVAL;
1369
1370         if (data[IFLA_GENEVE_REMOTE]) {
1371                 remote.sa.sa_family = AF_INET;
1372                 remote.sin.sin_addr.s_addr =
1373                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1374         }
1375
1376         if (data[IFLA_GENEVE_REMOTE6]) {
1377                 if (!IS_ENABLED(CONFIG_IPV6))
1378                         return -EPFNOSUPPORT;
1379
1380                 remote.sa.sa_family = AF_INET6;
1381                 remote.sin6.sin6_addr =
1382                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1383
1384                 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1385                     IPV6_ADDR_LINKLOCAL) {
1386                         netdev_dbg(dev, "link-local remote is unsupported\n");
1387                         return -EINVAL;
1388                 }
1389         }
1390
1391         if (data[IFLA_GENEVE_ID])
1392                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1393
1394         if (data[IFLA_GENEVE_TTL])
1395                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1396
1397         if (data[IFLA_GENEVE_TOS])
1398                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1399
1400         if (data[IFLA_GENEVE_LABEL])
1401                 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1402                         IPV6_FLOWLABEL_MASK;
1403
1404         if (data[IFLA_GENEVE_PORT])
1405                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1406
1407         if (data[IFLA_GENEVE_COLLECT_METADATA])
1408                 metadata = true;
1409
1410         if (data[IFLA_GENEVE_UDP_CSUM] &&
1411             !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1412                 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1413
1414         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1415             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1416                 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1417
1418         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1419             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1420                 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1421
1422         return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1423                                 dst_port, metadata, flags);
1424 }
1425
1426 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1427 {
1428         struct geneve_dev *geneve = netdev_priv(dev);
1429
1430         list_del(&geneve->next);
1431         unregister_netdevice_queue(dev, head);
1432 }
1433
1434 static size_t geneve_get_size(const struct net_device *dev)
1435 {
1436         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1437                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1438                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1439                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1440                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1441                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1442                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1443                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1444                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1445                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1446                 0;
1447 }
1448
1449 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1450 {
1451         struct geneve_dev *geneve = netdev_priv(dev);
1452         __u32 vni;
1453
1454         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1455         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1456                 goto nla_put_failure;
1457
1458         if (geneve->remote.sa.sa_family == AF_INET) {
1459                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1460                                     geneve->remote.sin.sin_addr.s_addr))
1461                         goto nla_put_failure;
1462 #if IS_ENABLED(CONFIG_IPV6)
1463         } else {
1464                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1465                                      &geneve->remote.sin6.sin6_addr))
1466                         goto nla_put_failure;
1467 #endif
1468         }
1469
1470         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1471             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1472             nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1473                 goto nla_put_failure;
1474
1475         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1476                 goto nla_put_failure;
1477
1478         if (geneve->collect_md) {
1479                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1480                         goto nla_put_failure;
1481         }
1482
1483         if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1484                        !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1485             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1486                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1487             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1488                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1489                 goto nla_put_failure;
1490
1491         return 0;
1492
1493 nla_put_failure:
1494         return -EMSGSIZE;
1495 }
1496
1497 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1498         .kind           = "geneve",
1499         .maxtype        = IFLA_GENEVE_MAX,
1500         .policy         = geneve_policy,
1501         .priv_size      = sizeof(struct geneve_dev),
1502         .setup          = geneve_setup,
1503         .validate       = geneve_validate,
1504         .newlink        = geneve_newlink,
1505         .dellink        = geneve_dellink,
1506         .get_size       = geneve_get_size,
1507         .fill_info      = geneve_fill_info,
1508 };
1509
1510 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1511                                         u8 name_assign_type, u16 dst_port)
1512 {
1513         struct nlattr *tb[IFLA_MAX + 1];
1514         struct net_device *dev;
1515         int err;
1516
1517         memset(tb, 0, sizeof(tb));
1518         dev = rtnl_create_link(net, name, name_assign_type,
1519                                &geneve_link_ops, tb);
1520         if (IS_ERR(dev))
1521                 return dev;
1522
1523         err = geneve_configure(net, dev, &geneve_remote_unspec,
1524                                0, 0, 0, 0, htons(dst_port), true,
1525                                GENEVE_F_UDP_ZERO_CSUM6_RX);
1526         if (err)
1527                 goto err;
1528
1529         /* openvswitch users expect packet sizes to be unrestricted,
1530          * so set the largest MTU we can.
1531          */
1532         err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1533         if (err)
1534                 goto err;
1535
1536         return dev;
1537
1538  err:
1539         free_netdev(dev);
1540         return ERR_PTR(err);
1541 }
1542 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1543
1544 static int geneve_netdevice_event(struct notifier_block *unused,
1545                                   unsigned long event, void *ptr)
1546 {
1547         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1548
1549         if (event == NETDEV_OFFLOAD_PUSH_GENEVE)
1550                 geneve_push_rx_ports(dev);
1551
1552         return NOTIFY_DONE;
1553 }
1554
1555 static struct notifier_block geneve_notifier_block __read_mostly = {
1556         .notifier_call = geneve_netdevice_event,
1557 };
1558
1559 static __net_init int geneve_init_net(struct net *net)
1560 {
1561         struct geneve_net *gn = net_generic(net, geneve_net_id);
1562
1563         INIT_LIST_HEAD(&gn->geneve_list);
1564         INIT_LIST_HEAD(&gn->sock_list);
1565         return 0;
1566 }
1567
1568 static void __net_exit geneve_exit_net(struct net *net)
1569 {
1570         struct geneve_net *gn = net_generic(net, geneve_net_id);
1571         struct geneve_dev *geneve, *next;
1572         struct net_device *dev, *aux;
1573         LIST_HEAD(list);
1574
1575         rtnl_lock();
1576
1577         /* gather any geneve devices that were moved into this ns */
1578         for_each_netdev_safe(net, dev, aux)
1579                 if (dev->rtnl_link_ops == &geneve_link_ops)
1580                         unregister_netdevice_queue(dev, &list);
1581
1582         /* now gather any other geneve devices that were created in this ns */
1583         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1584                 /* If geneve->dev is in the same netns, it was already added
1585                  * to the list by the previous loop.
1586                  */
1587                 if (!net_eq(dev_net(geneve->dev), net))
1588                         unregister_netdevice_queue(geneve->dev, &list);
1589         }
1590
1591         /* unregister the devices gathered above */
1592         unregister_netdevice_many(&list);
1593         rtnl_unlock();
1594 }
1595
1596 static struct pernet_operations geneve_net_ops = {
1597         .init = geneve_init_net,
1598         .exit = geneve_exit_net,
1599         .id   = &geneve_net_id,
1600         .size = sizeof(struct geneve_net),
1601 };
1602
1603 static int __init geneve_init_module(void)
1604 {
1605         int rc;
1606
1607         rc = register_pernet_subsys(&geneve_net_ops);
1608         if (rc)
1609                 goto out1;
1610
1611         rc = register_netdevice_notifier(&geneve_notifier_block);
1612         if (rc)
1613                 goto out2;
1614
1615         rc = rtnl_link_register(&geneve_link_ops);
1616         if (rc)
1617                 goto out3;
1618
1619         return 0;
1620
1621 out3:
1622         unregister_netdevice_notifier(&geneve_notifier_block);
1623 out2:
1624         unregister_pernet_subsys(&geneve_net_ops);
1625 out1:
1626         return rc;
1627 }
1628 late_initcall(geneve_init_module);
1629
1630 static void __exit geneve_cleanup_module(void)
1631 {
1632         rtnl_link_unregister(&geneve_link_ops);
1633         unregister_netdevice_notifier(&geneve_notifier_block);
1634         unregister_pernet_subsys(&geneve_net_ops);
1635 }
1636 module_exit(geneve_cleanup_module);
1637
1638 MODULE_LICENSE("GPL");
1639 MODULE_VERSION(GENEVE_NETDEV_VER);
1640 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1641 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1642 MODULE_ALIAS_RTNL_LINK("geneve");