kernel/drivers/net/geneve.c

   1 /*
   2  * GENEVE: Generic Network Virtualization Encapsulation
   3  *
   4  * Copyright (c) 2015 Red Hat, Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/netdevice.h>
  16 #include <linux/etherdevice.h>
  17 #include <linux/hash.h>
  18 #include <net/dst_metadata.h>
  19 #include <net/gro_cells.h>
  20 #include <net/rtnetlink.h>
  21 #include <net/geneve.h>
  22 #include <net/protocol.h>
  23
  24 #define GENEVE_NETDEV_VER       "0.6"
  25
  26 #define GENEVE_UDP_PORT         6081
  27
  28 #define GENEVE_N_VID            (1u << 24)
  29 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  30
  31 #define VNI_HASH_BITS           10
  32 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  33
  34 static bool log_ecn_error = true;
  35 module_param(log_ecn_error, bool, 0644);
  36 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  37
  38 #define GENEVE_VER 0
  39 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  40
  41 /* per-network namespace private data for this module */
  42 struct geneve_net {
  43         struct list_head        geneve_list;
  44         struct list_head        sock_list;
  45 };
  46
  47 static int geneve_net_id;
  48
  49 union geneve_addr {
  50         struct sockaddr_in sin;
  51         struct sockaddr_in6 sin6;
  52         struct sockaddr sa;
  53 };
  54
  55 static union geneve_addr geneve_remote_unspec = { .sa.sa_family = AF_UNSPEC, };
  56
  57 /* Pseudo network device */
  58 struct geneve_dev {
  59         struct hlist_node  hlist;       /* vni hash table */
  60         struct net         *net;        /* netns for packet i/o */
  61         struct net_device  *dev;        /* netdev for geneve tunnel */
  62         struct geneve_sock *sock4;      /* IPv4 socket used for geneve tunnel */
  63 #if IS_ENABLED(CONFIG_IPV6)
  64         struct geneve_sock *sock6;      /* IPv6 socket used for geneve tunnel */
  65 #endif
  66         u8                 vni[3];      /* virtual network ID for tunnel */
  67         u8                 ttl;         /* TTL override */
  68         u8                 tos;         /* TOS override */
  69         union geneve_addr  remote;      /* IP address for link partner */
  70         struct list_head   next;        /* geneve's per namespace list */
  71         __be16             dst_port;
  72         bool               collect_md;
  73         struct gro_cells   gro_cells;
  74 };
  75
  76 struct geneve_sock {
  77         bool                    collect_md;
  78         struct list_head        list;
  79         struct socket           *sock;
  80         struct rcu_head         rcu;
  81         int                     refcnt;
  82         struct udp_offload      udp_offloads;
  83         struct hlist_head       vni_list[VNI_HASH_SIZE];
  84 };
  85
  86 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  87 {
  88         __u32 vnid;
  89
  90         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  91         return hash_32(vnid, VNI_HASH_BITS);
  92 }
  93
  94 static __be64 vni_to_tunnel_id(const __u8 *vni)
  95 {
  96 #ifdef __BIG_ENDIAN
  97         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
  98 #else
  99         return (__force __be64)(((__force u64)vni[0] << 40) |
 100                                 ((__force u64)vni[1] << 48) |
 101                                 ((__force u64)vni[2] << 56));
 102 #endif
 103 }
 104
 105 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 106                                         __be32 addr, u8 vni[])
 107 {
 108         struct hlist_head *vni_list_head;
 109         struct geneve_dev *geneve;
 110         __u32 hash;
 111
 112         /* Find the device for this VNI */
 113         hash = geneve_net_vni_hash(vni);
 114         vni_list_head = &gs->vni_list[hash];
 115         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 116                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 117                     addr == geneve->remote.sin.sin_addr.s_addr)
 118                         return geneve;
 119         }
 120         return NULL;
 121 }
 122
 123 #if IS_ENABLED(CONFIG_IPV6)
 124 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 125                                          struct in6_addr addr6, u8 vni[])
 126 {
 127         struct hlist_head *vni_list_head;
 128         struct geneve_dev *geneve;
 129         __u32 hash;
 130
 131         /* Find the device for this VNI */
 132         hash = geneve_net_vni_hash(vni);
 133         vni_list_head = &gs->vni_list[hash];
 134         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 135                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 136                     ipv6_addr_equal(&addr6, &geneve->remote.sin6.sin6_addr))
 137                         return geneve;
 138         }
 139         return NULL;
 140 }
 141 #endif
 142
 143 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 144 {
 145         return (struct genevehdr *)(udp_hdr(skb) + 1);
 146 }
 147
 148 /* geneve receive/decap routine */
 149 static void geneve_rx(struct geneve_sock *gs, struct sk_buff *skb)
 150 {
 151         struct genevehdr *gnvh = geneve_hdr(skb);
 152         struct metadata_dst *tun_dst = NULL;
 153         struct geneve_dev *geneve = NULL;
 154         struct pcpu_sw_netstats *stats;
 155         struct iphdr *iph = NULL;
 156         __be32 addr;
 157         static u8 zero_vni[3];
 158         u8 *vni;
 159         int err = 0;
 160         sa_family_t sa_family;
 161 #if IS_ENABLED(CONFIG_IPV6)
 162         struct ipv6hdr *ip6h = NULL;
 163         struct in6_addr addr6;
 164         static struct in6_addr zero_addr6;
 165 #endif
 166
 167         sa_family = gs->sock->sk->sk_family;
 168
 169         if (sa_family == AF_INET) {
 170                 iph = ip_hdr(skb); /* outer IP header... */
 171
 172                 if (gs->collect_md) {
 173                         vni = zero_vni;
 174                         addr = 0;
 175                 } else {
 176                         vni = gnvh->vni;
 177
 178                         addr = iph->saddr;
 179                 }
 180
 181                 geneve = geneve_lookup(gs, addr, vni);
 182 #if IS_ENABLED(CONFIG_IPV6)
 183         } else if (sa_family == AF_INET6) {
 184                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 185
 186                 if (gs->collect_md) {
 187                         vni = zero_vni;
 188                         addr6 = zero_addr6;
 189                 } else {
 190                         vni = gnvh->vni;
 191
 192                         addr6 = ip6h->saddr;
 193                 }
 194
 195                 geneve = geneve6_lookup(gs, addr6, vni);
 196 #endif
 197         }
 198         if (!geneve)
 199                 goto drop;
 200
 201         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 202                 __be16 flags;
 203
 204                 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
 205                         (gnvh->oam ? TUNNEL_OAM : 0) |
 206                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 207
 208                 tun_dst = udp_tun_rx_dst(skb, sa_family, flags,
 209                                          vni_to_tunnel_id(gnvh->vni),
 210                                          gnvh->opt_len * 4);
 211                 if (!tun_dst)
 212                         goto drop;
 213                 /* Update tunnel dst according to Geneve options. */
 214                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 215                                         gnvh->options, gnvh->opt_len * 4);
 216         } else {
 217                 /* Drop packets w/ critical options,
 218                  * since we don't support any...
 219                  */
 220                 if (gnvh->critical)
 221                         goto drop;
 222         }
 223
 224         skb_reset_mac_header(skb);
 225         skb_scrub_packet(skb, !net_eq(geneve->net, dev_net(geneve->dev)));
 226         skb->protocol = eth_type_trans(skb, geneve->dev);
 227         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 228
 229         if (tun_dst)
 230                 skb_dst_set(skb, &tun_dst->dst);
 231
 232         /* Ignore packet loops (and multicast echo) */
 233         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
 234                 goto drop;
 235
 236         skb_reset_network_header(skb);
 237
 238         if (iph)
 239                 err = IP_ECN_decapsulate(iph, skb);
 240 #if IS_ENABLED(CONFIG_IPV6)
 241         if (ip6h)
 242                 err = IP6_ECN_decapsulate(ip6h, skb);
 243 #endif
 244
 245         if (unlikely(err)) {
 246                 if (log_ecn_error) {
 247                         if (iph)
 248                                 net_info_ratelimited("non-ECT from %pI4 "
 249                                                      "with TOS=%#x\n",
 250                                                      &iph->saddr, iph->tos);
 251 #if IS_ENABLED(CONFIG_IPV6)
 252                         if (ip6h)
 253                                 net_info_ratelimited("non-ECT from %pI6\n",
 254                                                      &ip6h->saddr);
 255 #endif
 256                 }
 257                 if (err > 1) {
 258                         ++geneve->dev->stats.rx_frame_errors;
 259                         ++geneve->dev->stats.rx_errors;
 260                         goto drop;
 261                 }
 262         }
 263
 264         stats = this_cpu_ptr(geneve->dev->tstats);
 265         u64_stats_update_begin(&stats->syncp);
 266         stats->rx_packets++;
 267         stats->rx_bytes += skb->len;
 268         u64_stats_update_end(&stats->syncp);
 269
 270         gro_cells_receive(&geneve->gro_cells, skb);
 271         return;
 272 drop:
 273         /* Consume bad packet */
 274         kfree_skb(skb);
 275 }
 276
 277 /* Setup stats when device is created */
 278 static int geneve_init(struct net_device *dev)
 279 {
 280         struct geneve_dev *geneve = netdev_priv(dev);
 281         int err;
 282
 283         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 284         if (!dev->tstats)
 285                 return -ENOMEM;
 286
 287         err = gro_cells_init(&geneve->gro_cells, dev);
 288         if (err) {
 289                 free_percpu(dev->tstats);
 290                 return err;
 291         }
 292
 293         return 0;
 294 }
 295
 296 static void geneve_uninit(struct net_device *dev)
 297 {
 298         struct geneve_dev *geneve = netdev_priv(dev);
 299
 300         gro_cells_destroy(&geneve->gro_cells);
 301         free_percpu(dev->tstats);
 302 }
 303
 304 /* Callback from net/ipv4/udp.c to receive packets */
 305 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 306 {
 307         struct genevehdr *geneveh;
 308         struct geneve_sock *gs;
 309         int opts_len;
 310
 311         /* Need Geneve and inner Ethernet header to be present */
 312         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 313                 goto drop;
 314
 315         /* Return packets with reserved bits set */
 316         geneveh = geneve_hdr(skb);
 317         if (unlikely(geneveh->ver != GENEVE_VER))
 318                 goto drop;
 319
 320         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 321                 goto drop;
 322
 323         opts_len = geneveh->opt_len * 4;
 324         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 325                                  htons(ETH_P_TEB)))
 326                 goto drop;
 327
 328         gs = rcu_dereference_sk_user_data(sk);
 329         if (!gs)
 330                 goto drop;
 331
 332         geneve_rx(gs, skb);
 333         return 0;
 334
 335 drop:
 336         /* Consume bad packet */
 337         kfree_skb(skb);
 338         return 0;
 339 }
 340
 341 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 342                                          __be16 port)
 343 {
 344         struct socket *sock;
 345         struct udp_port_cfg udp_conf;
 346         int err;
 347
 348         memset(&udp_conf, 0, sizeof(udp_conf));
 349
 350         if (ipv6) {
 351                 udp_conf.family = AF_INET6;
 352                 udp_conf.ipv6_v6only = 1;
 353         } else {
 354                 udp_conf.family = AF_INET;
 355                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 356         }
 357
 358         udp_conf.local_udp_port = port;
 359
 360         /* Open UDP socket */
 361         err = udp_sock_create(net, &udp_conf, &sock);
 362         if (err < 0)
 363                 return ERR_PTR(err);
 364
 365         return sock;
 366 }
 367
 368 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
 369 {
 370         struct sock *sk = gs->sock->sk;
 371         sa_family_t sa_family = sk->sk_family;
 372         int err;
 373
 374         if (sa_family == AF_INET) {
 375                 err = udp_add_offload(&gs->udp_offloads);
 376                 if (err)
 377                         pr_warn("geneve: udp_add_offload failed with status %d\n",
 378                                 err);
 379         }
 380 }
 381
 382 static int geneve_hlen(struct genevehdr *gh)
 383 {
 384         return sizeof(*gh) + gh->opt_len * 4;
 385 }
 386
 387 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
 388                                            struct sk_buff *skb,
 389                                            struct udp_offload *uoff)
 390 {
 391         struct sk_buff *p, **pp = NULL;
 392         struct genevehdr *gh, *gh2;
 393         unsigned int hlen, gh_len, off_gnv;
 394         const struct packet_offload *ptype;
 395         __be16 type;
 396         int flush = 1;
 397
 398         off_gnv = skb_gro_offset(skb);
 399         hlen = off_gnv + sizeof(*gh);
 400         gh = skb_gro_header_fast(skb, off_gnv);
 401         if (skb_gro_header_hard(skb, hlen)) {
 402                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 403                 if (unlikely(!gh))
 404                         goto out;
 405         }
 406
 407         if (gh->ver != GENEVE_VER || gh->oam)
 408                 goto out;
 409         gh_len = geneve_hlen(gh);
 410
 411         hlen = off_gnv + gh_len;
 412         if (skb_gro_header_hard(skb, hlen)) {
 413                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 414                 if (unlikely(!gh))
 415                         goto out;
 416         }
 417
 418         flush = 0;
 419
 420         for (p = *head; p; p = p->next) {
 421                 if (!NAPI_GRO_CB(p)->same_flow)
 422                         continue;
 423
 424                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 425                 if (gh->opt_len != gh2->opt_len ||
 426                     memcmp(gh, gh2, gh_len)) {
 427                         NAPI_GRO_CB(p)->same_flow = 0;
 428                         continue;
 429                 }
 430         }
 431
 432         type = gh->proto_type;
 433
 434         rcu_read_lock();
 435         ptype = gro_find_receive_by_type(type);
 436         if (!ptype) {
 437                 flush = 1;
 438                 goto out_unlock;
 439         }
 440
 441         skb_gro_pull(skb, gh_len);
 442         skb_gro_postpull_rcsum(skb, gh, gh_len);
 443         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 444
 445 out_unlock:
 446         rcu_read_unlock();
 447 out:
 448         NAPI_GRO_CB(skb)->flush |= flush;
 449
 450         return pp;
 451 }
 452
 453 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
 454                                struct udp_offload *uoff)
 455 {
 456         struct genevehdr *gh;
 457         struct packet_offload *ptype;
 458         __be16 type;
 459         int gh_len;
 460         int err = -ENOSYS;
 461
 462         udp_tunnel_gro_complete(skb, nhoff);
 463
 464         gh = (struct genevehdr *)(skb->data + nhoff);
 465         gh_len = geneve_hlen(gh);
 466         type = gh->proto_type;
 467
 468         rcu_read_lock();
 469         ptype = gro_find_complete_by_type(type);
 470         if (ptype)
 471                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 472
 473         rcu_read_unlock();
 474         return err;
 475 }
 476
 477 /* Create new listen socket if needed */
 478 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 479                                                 bool ipv6)
 480 {
 481         struct geneve_net *gn = net_generic(net, geneve_net_id);
 482         struct geneve_sock *gs;
 483         struct socket *sock;
 484         struct udp_tunnel_sock_cfg tunnel_cfg;
 485         int h;
 486
 487         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 488         if (!gs)
 489                 return ERR_PTR(-ENOMEM);
 490
 491         sock = geneve_create_sock(net, ipv6, port);
 492         if (IS_ERR(sock)) {
 493                 kfree(gs);
 494                 return ERR_CAST(sock);
 495         }
 496
 497         gs->sock = sock;
 498         gs->refcnt = 1;
 499         for (h = 0; h < VNI_HASH_SIZE; ++h)
 500                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 501
 502         /* Initialize the geneve udp offloads structure */
 503         gs->udp_offloads.port = port;
 504         gs->udp_offloads.callbacks.gro_receive  = geneve_gro_receive;
 505         gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
 506         geneve_notify_add_rx_port(gs);
 507
 508         /* Mark socket as an encapsulation socket */
 509         tunnel_cfg.sk_user_data = gs;
 510         tunnel_cfg.encap_type = 1;
 511         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 512         tunnel_cfg.encap_destroy = NULL;
 513         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 514         list_add(&gs->list, &gn->sock_list);
 515         return gs;
 516 }
 517
 518 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 519 {
 520         struct sock *sk = gs->sock->sk;
 521         sa_family_t sa_family = sk->sk_family;
 522
 523         if (sa_family == AF_INET)
 524                 udp_del_offload(&gs->udp_offloads);
 525 }
 526
 527 static void __geneve_sock_release(struct geneve_sock *gs)
 528 {
 529         if (!gs || --gs->refcnt)
 530                 return;
 531
 532         list_del(&gs->list);
 533         geneve_notify_del_rx_port(gs);
 534         udp_tunnel_sock_release(gs->sock);
 535         kfree_rcu(gs, rcu);
 536 }
 537
 538 static void geneve_sock_release(struct geneve_dev *geneve)
 539 {
 540         __geneve_sock_release(geneve->sock4);
 541 #if IS_ENABLED(CONFIG_IPV6)
 542         __geneve_sock_release(geneve->sock6);
 543 #endif
 544 }
 545
 546 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 547                                             sa_family_t family,
 548                                             __be16 dst_port)
 549 {
 550         struct geneve_sock *gs;
 551
 552         list_for_each_entry(gs, &gn->sock_list, list) {
 553                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 554                     inet_sk(gs->sock->sk)->sk.sk_family == family) {
 555                         return gs;
 556                 }
 557         }
 558         return NULL;
 559 }
 560
 561 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 562 {
 563         struct net *net = geneve->net;
 564         struct geneve_net *gn = net_generic(net, geneve_net_id);
 565         struct geneve_sock *gs;
 566         __u32 hash;
 567
 568         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
 569         if (gs) {
 570                 gs->refcnt++;
 571                 goto out;
 572         }
 573
 574         gs = geneve_socket_create(net, geneve->dst_port, ipv6);
 575         if (IS_ERR(gs))
 576                 return PTR_ERR(gs);
 577
 578 out:
 579         gs->collect_md = geneve->collect_md;
 580 #if IS_ENABLED(CONFIG_IPV6)
 581         if (ipv6)
 582                 geneve->sock6 = gs;
 583         else
 584 #endif
 585                 geneve->sock4 = gs;
 586
 587         hash = geneve_net_vni_hash(geneve->vni);
 588         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 589         return 0;
 590 }
 591
 592 static int geneve_open(struct net_device *dev)
 593 {
 594         struct geneve_dev *geneve = netdev_priv(dev);
 595         bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
 596         bool metadata = geneve->collect_md;
 597         int ret = 0;
 598
 599         geneve->sock4 = NULL;
 600 #if IS_ENABLED(CONFIG_IPV6)
 601         geneve->sock6 = NULL;
 602         if (ipv6 || metadata)
 603                 ret = geneve_sock_add(geneve, true);
 604 #endif
 605         if (!ret && (!ipv6 || metadata))
 606                 ret = geneve_sock_add(geneve, false);
 607         if (ret < 0)
 608                 geneve_sock_release(geneve);
 609
 610         return ret;
 611 }
 612
 613 static int geneve_stop(struct net_device *dev)
 614 {
 615         struct geneve_dev *geneve = netdev_priv(dev);
 616
 617         if (!hlist_unhashed(&geneve->hlist))
 618                 hlist_del_rcu(&geneve->hlist);
 619         geneve_sock_release(geneve);
 620         return 0;
 621 }
 622
 623 static void geneve_build_header(struct genevehdr *geneveh,
 624                                 __be16 tun_flags, u8 vni[3],
 625                                 u8 options_len, u8 *options)
 626 {
 627         geneveh->ver = GENEVE_VER;
 628         geneveh->opt_len = options_len / 4;
 629         geneveh->oam = !!(tun_flags & TUNNEL_OAM);
 630         geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 631         geneveh->rsvd1 = 0;
 632         memcpy(geneveh->vni, vni, 3);
 633         geneveh->proto_type = htons(ETH_P_TEB);
 634         geneveh->rsvd2 = 0;
 635
 636         memcpy(geneveh->options, options, options_len);
 637 }
 638
 639 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 640                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 641                             bool csum, bool xnet)
 642 {
 643         struct genevehdr *gnvh;
 644         int min_headroom;
 645         int err;
 646
 647         skb_scrub_packet(skb, xnet);
 648
 649         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 650                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 651         err = skb_cow_head(skb, min_headroom);
 652         if (unlikely(err)) {
 653                 kfree_skb(skb);
 654                 goto free_rt;
 655         }
 656
 657         skb = udp_tunnel_handle_offloads(skb, csum);
 658         if (IS_ERR(skb)) {
 659                 err = PTR_ERR(skb);
 660                 goto free_rt;
 661         }
 662
 663         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 664         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 665
 666         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 667         return 0;
 668
 669 free_rt:
 670         ip_rt_put(rt);
 671         return err;
 672 }
 673
 674 #if IS_ENABLED(CONFIG_IPV6)
 675 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 676                              __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 677                              bool csum, bool xnet)
 678 {
 679         struct genevehdr *gnvh;
 680         int min_headroom;
 681         int err;
 682
 683         skb_scrub_packet(skb, xnet);
 684
 685         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
 686                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
 687         err = skb_cow_head(skb, min_headroom);
 688         if (unlikely(err)) {
 689                 kfree_skb(skb);
 690                 goto free_dst;
 691         }
 692
 693         skb = udp_tunnel_handle_offloads(skb, csum);
 694         if (IS_ERR(skb)) {
 695                 err = PTR_ERR(skb);
 696                 goto free_dst;
 697         }
 698
 699         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 700         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 701
 702         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 703         return 0;
 704
 705 free_dst:
 706         dst_release(dst);
 707         return err;
 708 }
 709 #endif
 710
 711 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 712                                        struct net_device *dev,
 713                                        struct flowi4 *fl4,
 714                                        struct ip_tunnel_info *info)
 715 {
 716         struct geneve_dev *geneve = netdev_priv(dev);
 717         struct rtable *rt = NULL;
 718         __u8 tos;
 719
 720         memset(fl4, 0, sizeof(*fl4));
 721         fl4->flowi4_mark = skb->mark;
 722         fl4->flowi4_proto = IPPROTO_UDP;
 723
 724         if (info) {
 725                 fl4->daddr = info->key.u.ipv4.dst;
 726                 fl4->saddr = info->key.u.ipv4.src;
 727                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 728         } else {
 729                 tos = geneve->tos;
 730                 if (tos == 1) {
 731                         const struct iphdr *iip = ip_hdr(skb);
 732
 733                         tos = ip_tunnel_get_dsfield(iip, skb);
 734                 }
 735
 736                 fl4->flowi4_tos = RT_TOS(tos);
 737                 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
 738         }
 739
 740         rt = ip_route_output_key(geneve->net, fl4);
 741         if (IS_ERR(rt)) {
 742                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 743                 return ERR_PTR(-ENETUNREACH);
 744         }
 745         if (rt->dst.dev == dev) { /* is this necessary? */
 746                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 747                 ip_rt_put(rt);
 748                 return ERR_PTR(-ELOOP);
 749         }
 750         return rt;
 751 }
 752
 753 #if IS_ENABLED(CONFIG_IPV6)
 754 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 755                                            struct net_device *dev,
 756                                            struct flowi6 *fl6,
 757                                            struct ip_tunnel_info *info)
 758 {
 759         struct geneve_dev *geneve = netdev_priv(dev);
 760         struct geneve_sock *gs6 = geneve->sock6;
 761         struct dst_entry *dst = NULL;
 762         __u8 prio;
 763
 764         memset(fl6, 0, sizeof(*fl6));
 765         fl6->flowi6_mark = skb->mark;
 766         fl6->flowi6_proto = IPPROTO_UDP;
 767
 768         if (info) {
 769                 fl6->daddr = info->key.u.ipv6.dst;
 770                 fl6->saddr = info->key.u.ipv6.src;
 771                 fl6->flowi6_tos = RT_TOS(info->key.tos);
 772         } else {
 773                 prio = geneve->tos;
 774                 if (prio == 1) {
 775                         const struct iphdr *iip = ip_hdr(skb);
 776
 777                         prio = ip_tunnel_get_dsfield(iip, skb);
 778                 }
 779
 780                 fl6->flowi6_tos = RT_TOS(prio);
 781                 fl6->daddr = geneve->remote.sin6.sin6_addr;
 782         }
 783
 784         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 785                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 786                 return ERR_PTR(-ENETUNREACH);
 787         }
 788         if (dst->dev == dev) { /* is this necessary? */
 789                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 790                 dst_release(dst);
 791                 return ERR_PTR(-ELOOP);
 792         }
 793
 794         return dst;
 795 }
 796 #endif
 797
 798 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 799 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 800 {
 801 #ifdef __BIG_ENDIAN
 802         vni[0] = (__force __u8)(tun_id >> 16);
 803         vni[1] = (__force __u8)(tun_id >> 8);
 804         vni[2] = (__force __u8)tun_id;
 805 #else
 806         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 807         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 808         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 809 #endif
 810 }
 811
 812 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 813                                    struct ip_tunnel_info *info)
 814 {
 815         struct geneve_dev *geneve = netdev_priv(dev);
 816         struct geneve_sock *gs4 = geneve->sock4;
 817         struct rtable *rt = NULL;
 818         int err = -EINVAL;
 819         struct flowi4 fl4;
 820         __u8 tos, ttl;
 821         __be16 sport;
 822         bool udp_csum;
 823         __be16 df;
 824         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 825
 826         if (geneve->collect_md) {
 827                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 828                         netdev_dbg(dev, "no tunnel metadata\n");
 829                         goto tx_error;
 830                 }
 831                 if (info && ip_tunnel_info_af(info) != AF_INET)
 832                         goto tx_error;
 833         }
 834
 835         rt = geneve_get_v4_rt(skb, dev, &fl4, info);
 836         if (IS_ERR(rt)) {
 837                 err = PTR_ERR(rt);
 838                 goto tx_error;
 839         }
 840
 841         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 842         skb_reset_mac_header(skb);
 843
 844         if (info) {
 845                 const struct ip_tunnel_key *key = &info->key;
 846                 u8 *opts = NULL;
 847                 u8 vni[3];
 848
 849                 tunnel_id_to_vni(key->tun_id, vni);
 850                 if (key->tun_flags & TUNNEL_GENEVE_OPT)
 851                         opts = ip_tunnel_info_opts(info);
 852
 853                 udp_csum = !!(key->tun_flags & TUNNEL_CSUM);
 854                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 855                                        info->options_len, opts, udp_csum, xnet);
 856                 if (unlikely(err))
 857                         goto err;
 858
 859                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 860                 ttl = key->ttl;
 861                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 862         } else {
 863                 udp_csum = false;
 864                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 865                                        0, NULL, udp_csum, xnet);
 866                 if (unlikely(err))
 867                         goto err;
 868
 869                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 870                 ttl = geneve->ttl;
 871                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 872                         ttl = 1;
 873                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 874                 df = 0;
 875         }
 876         err = udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 877                                   tos, ttl, df, sport, geneve->dst_port,
 878                                   !net_eq(geneve->net, dev_net(geneve->dev)),
 879                                   !udp_csum);
 880
 881         iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
 882         return NETDEV_TX_OK;
 883
 884 tx_error:
 885         dev_kfree_skb(skb);
 886 err:
 887         if (err == -ELOOP)
 888                 dev->stats.collisions++;
 889         else if (err == -ENETUNREACH)
 890                 dev->stats.tx_carrier_errors++;
 891         else
 892                 dev->stats.tx_errors++;
 893         return NETDEV_TX_OK;
 894 }
 895
 896 #if IS_ENABLED(CONFIG_IPV6)
 897 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 898                                     struct ip_tunnel_info *info)
 899 {
 900         struct geneve_dev *geneve = netdev_priv(dev);
 901         struct geneve_sock *gs6 = geneve->sock6;
 902         struct dst_entry *dst = NULL;
 903         int err = -EINVAL;
 904         struct flowi6 fl6;
 905         __u8 prio, ttl;
 906         __be16 sport;
 907         bool udp_csum;
 908         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 909
 910         if (geneve->collect_md) {
 911                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 912                         netdev_dbg(dev, "no tunnel metadata\n");
 913                         goto tx_error;
 914                 }
 915         }
 916
 917         dst = geneve_get_v6_dst(skb, dev, &fl6, info);
 918         if (IS_ERR(dst)) {
 919                 err = PTR_ERR(dst);
 920                 goto tx_error;
 921         }
 922
 923         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 924         skb_reset_mac_header(skb);
 925
 926         if (info) {
 927                 const struct ip_tunnel_key *key = &info->key;
 928                 u8 *opts = NULL;
 929                 u8 vni[3];
 930
 931                 tunnel_id_to_vni(key->tun_id, vni);
 932                 if (key->tun_flags & TUNNEL_GENEVE_OPT)
 933                         opts = ip_tunnel_info_opts(info);
 934
 935                 udp_csum = !!(key->tun_flags & TUNNEL_CSUM);
 936                 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
 937                                         info->options_len, opts,
 938                                         udp_csum, xnet);
 939                 if (unlikely(err))
 940                         goto err;
 941
 942                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 943                 ttl = key->ttl;
 944         } else {
 945                 udp_csum = false;
 946                 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
 947                                         0, NULL, udp_csum, xnet);
 948                 if (unlikely(err))
 949                         goto err;
 950
 951                 prio = ip_tunnel_ecn_encap(fl6.flowi6_tos, ip_hdr(skb), skb);
 952                 ttl = geneve->ttl;
 953                 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
 954                         ttl = 1;
 955                 ttl = ttl ? : ip6_dst_hoplimit(dst);
 956         }
 957         err = udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
 958                                    &fl6.saddr, &fl6.daddr, prio, ttl,
 959                                    sport, geneve->dst_port, !udp_csum);
 960         return NETDEV_TX_OK;
 961
 962 tx_error:
 963         dev_kfree_skb(skb);
 964 err:
 965         if (err == -ELOOP)
 966                 dev->stats.collisions++;
 967         else if (err == -ENETUNREACH)
 968                 dev->stats.tx_carrier_errors++;
 969         else
 970                 dev->stats.tx_errors++;
 971         return NETDEV_TX_OK;
 972 }
 973 #endif
 974
 975 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
 976 {
 977         struct geneve_dev *geneve = netdev_priv(dev);
 978         struct ip_tunnel_info *info = NULL;
 979
 980         if (geneve->collect_md)
 981                 info = skb_tunnel_info(skb);
 982
 983 #if IS_ENABLED(CONFIG_IPV6)
 984         if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
 985             (!info && geneve->remote.sa.sa_family == AF_INET6))
 986                 return geneve6_xmit_skb(skb, dev, info);
 987 #endif
 988         return geneve_xmit_skb(skb, dev, info);
 989 }
 990
 991 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
 992 {
 993         /* GENEVE overhead is not fixed, so we can't enforce a more
 994          * precise max MTU.
 995          */
 996         if (new_mtu < 68 || new_mtu > IP_MAX_MTU)
 997                 return -EINVAL;
 998         dev->mtu = new_mtu;
 999         return 0;
1000 }
1001
1002 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1003 {
1004         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1005         struct geneve_dev *geneve = netdev_priv(dev);
1006         struct rtable *rt;
1007         struct flowi4 fl4;
1008 #if IS_ENABLED(CONFIG_IPV6)
1009         struct dst_entry *dst;
1010         struct flowi6 fl6;
1011 #endif
1012
1013         if (ip_tunnel_info_af(info) == AF_INET) {
1014                 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1015                 if (IS_ERR(rt))
1016                         return PTR_ERR(rt);
1017
1018                 ip_rt_put(rt);
1019                 info->key.u.ipv4.src = fl4.saddr;
1020 #if IS_ENABLED(CONFIG_IPV6)
1021         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1022                 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1023                 if (IS_ERR(dst))
1024                         return PTR_ERR(dst);
1025
1026                 dst_release(dst);
1027                 info->key.u.ipv6.src = fl6.saddr;
1028 #endif
1029         } else {
1030                 return -EINVAL;
1031         }
1032
1033         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1034                                              1, USHRT_MAX, true);
1035         info->key.tp_dst = geneve->dst_port;
1036         return 0;
1037 }
1038
1039 static const struct net_device_ops geneve_netdev_ops = {
1040         .ndo_init               = geneve_init,
1041         .ndo_uninit             = geneve_uninit,
1042         .ndo_open               = geneve_open,
1043         .ndo_stop               = geneve_stop,
1044         .ndo_start_xmit         = geneve_xmit,
1045         .ndo_get_stats64        = ip_tunnel_get_stats64,
1046         .ndo_change_mtu         = geneve_change_mtu,
1047         .ndo_validate_addr      = eth_validate_addr,
1048         .ndo_set_mac_address    = eth_mac_addr,
1049         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1050 };
1051
1052 static void geneve_get_drvinfo(struct net_device *dev,
1053                                struct ethtool_drvinfo *drvinfo)
1054 {
1055         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1056         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1057 }
1058
1059 static const struct ethtool_ops geneve_ethtool_ops = {
1060         .get_drvinfo    = geneve_get_drvinfo,
1061         .get_link       = ethtool_op_get_link,
1062 };
1063
1064 /* Info for udev, that this is a virtual tunnel endpoint */
1065 static struct device_type geneve_type = {
1066         .name = "geneve",
1067 };
1068
1069 /* Initialize the device structure. */
1070 static void geneve_setup(struct net_device *dev)
1071 {
1072         ether_setup(dev);
1073
1074         dev->netdev_ops = &geneve_netdev_ops;
1075         dev->ethtool_ops = &geneve_ethtool_ops;
1076         dev->destructor = free_netdev;
1077
1078         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1079
1080         dev->features    |= NETIF_F_LLTX;
1081         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1082         dev->features    |= NETIF_F_RXCSUM;
1083         dev->features    |= NETIF_F_GSO_SOFTWARE;
1084
1085         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1086         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1087
1088         netif_keep_dst(dev);
1089         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1090         eth_hw_addr_random(dev);
1091 }
1092
1093 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1094         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1095         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1096         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1097         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1098         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1099         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1100         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1101 };
1102
1103 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1104 {
1105         if (tb[IFLA_ADDRESS]) {
1106                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1107                         return -EINVAL;
1108
1109                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1110                         return -EADDRNOTAVAIL;
1111         }
1112
1113         if (!data)
1114                 return -EINVAL;
1115
1116         if (data[IFLA_GENEVE_ID]) {
1117                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1118
1119                 if (vni >= GENEVE_VID_MASK)
1120                         return -ERANGE;
1121         }
1122
1123         return 0;
1124 }
1125
1126 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1127                                           __be16 dst_port,
1128                                           union geneve_addr *remote,
1129                                           u8 vni[],
1130                                           bool *tun_on_same_port,
1131                                           bool *tun_collect_md)
1132 {
1133         struct geneve_dev *geneve, *t;
1134
1135         *tun_on_same_port = false;
1136         *tun_collect_md = false;
1137         t = NULL;
1138         list_for_each_entry(geneve, &gn->geneve_list, next) {
1139                 if (geneve->dst_port == dst_port) {
1140                         *tun_collect_md = geneve->collect_md;
1141                         *tun_on_same_port = true;
1142                 }
1143                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1144                     !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1145                     dst_port == geneve->dst_port)
1146                         t = geneve;
1147         }
1148         return t;
1149 }
1150
1151 static int geneve_configure(struct net *net, struct net_device *dev,
1152                             union geneve_addr *remote,
1153                             __u32 vni, __u8 ttl, __u8 tos, __be16 dst_port,
1154                             bool metadata)
1155 {
1156         struct geneve_net *gn = net_generic(net, geneve_net_id);
1157         struct geneve_dev *t, *geneve = netdev_priv(dev);
1158         bool tun_collect_md, tun_on_same_port;
1159         int err, encap_len;
1160
1161         if (!remote)
1162                 return -EINVAL;
1163         if (metadata &&
1164             (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl))
1165                 return -EINVAL;
1166
1167         geneve->net = net;
1168         geneve->dev = dev;
1169
1170         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1171         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1172         geneve->vni[2] =  vni & 0x000000ff;
1173
1174         if ((remote->sa.sa_family == AF_INET &&
1175              IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1176             (remote->sa.sa_family == AF_INET6 &&
1177              ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1178                 return -EINVAL;
1179         geneve->remote = *remote;
1180
1181         geneve->ttl = ttl;
1182         geneve->tos = tos;
1183         geneve->dst_port = dst_port;
1184         geneve->collect_md = metadata;
1185
1186         t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1187                             &tun_on_same_port, &tun_collect_md);
1188         if (t)
1189                 return -EBUSY;
1190
1191         /* make enough headroom for basic scenario */
1192         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1193         if (remote->sa.sa_family == AF_INET)
1194                 encap_len += sizeof(struct iphdr);
1195         else
1196                 encap_len += sizeof(struct ipv6hdr);
1197         dev->needed_headroom = encap_len + ETH_HLEN;
1198
1199         if (metadata) {
1200                 if (tun_on_same_port)
1201                         return -EPERM;
1202         } else {
1203                 if (tun_collect_md)
1204                         return -EPERM;
1205         }
1206
1207         err = register_netdevice(dev);
1208         if (err)
1209                 return err;
1210
1211         list_add(&geneve->next, &gn->geneve_list);
1212         return 0;
1213 }
1214
1215 static int geneve_newlink(struct net *net, struct net_device *dev,
1216                           struct nlattr *tb[], struct nlattr *data[])
1217 {
1218         __be16 dst_port = htons(GENEVE_UDP_PORT);
1219         __u8 ttl = 0, tos = 0;
1220         bool metadata = false;
1221         union geneve_addr remote = geneve_remote_unspec;
1222         __u32 vni = 0;
1223
1224         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1225                 return -EINVAL;
1226
1227         if (data[IFLA_GENEVE_REMOTE]) {
1228                 remote.sa.sa_family = AF_INET;
1229                 remote.sin.sin_addr.s_addr =
1230                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1231         }
1232
1233         if (data[IFLA_GENEVE_REMOTE6]) {
1234                 if (!IS_ENABLED(CONFIG_IPV6))
1235                         return -EPFNOSUPPORT;
1236
1237                 remote.sa.sa_family = AF_INET6;
1238                 remote.sin6.sin6_addr =
1239                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1240
1241                 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1242                     IPV6_ADDR_LINKLOCAL) {
1243                         netdev_dbg(dev, "link-local remote is unsupported\n");
1244                         return -EINVAL;
1245                 }
1246         }
1247
1248         if (data[IFLA_GENEVE_ID])
1249                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1250
1251         if (data[IFLA_GENEVE_TTL])
1252                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1253
1254         if (data[IFLA_GENEVE_TOS])
1255                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1256
1257         if (data[IFLA_GENEVE_PORT])
1258                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1259
1260         if (data[IFLA_GENEVE_COLLECT_METADATA])
1261                 metadata = true;
1262
1263         return geneve_configure(net, dev, &remote, vni, ttl, tos, dst_port,
1264                                 metadata);
1265 }
1266
1267 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1268 {
1269         struct geneve_dev *geneve = netdev_priv(dev);
1270
1271         list_del(&geneve->next);
1272         unregister_netdevice_queue(dev, head);
1273 }
1274
1275 static size_t geneve_get_size(const struct net_device *dev)
1276 {
1277         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1278                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1279                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1280                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1281                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1282                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1283                 0;
1284 }
1285
1286 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1287 {
1288         struct geneve_dev *geneve = netdev_priv(dev);
1289         __u32 vni;
1290
1291         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1292         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1293                 goto nla_put_failure;
1294
1295         if (geneve->remote.sa.sa_family == AF_INET) {
1296                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1297                                     geneve->remote.sin.sin_addr.s_addr))
1298                         goto nla_put_failure;
1299 #if IS_ENABLED(CONFIG_IPV6)
1300         } else {
1301                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1302                                      &geneve->remote.sin6.sin6_addr))
1303                         goto nla_put_failure;
1304 #endif
1305         }
1306
1307         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1308             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos))
1309                 goto nla_put_failure;
1310
1311         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1312                 goto nla_put_failure;
1313
1314         if (geneve->collect_md) {
1315                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1316                         goto nla_put_failure;
1317         }
1318
1319         return 0;
1320
1321 nla_put_failure:
1322         return -EMSGSIZE;
1323 }
1324
1325 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1326         .kind           = "geneve",
1327         .maxtype        = IFLA_GENEVE_MAX,
1328         .policy         = geneve_policy,
1329         .priv_size      = sizeof(struct geneve_dev),
1330         .setup          = geneve_setup,
1331         .validate       = geneve_validate,
1332         .newlink        = geneve_newlink,
1333         .dellink        = geneve_dellink,
1334         .get_size       = geneve_get_size,
1335         .fill_info      = geneve_fill_info,
1336 };
1337
1338 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1339                                         u8 name_assign_type, u16 dst_port)
1340 {
1341         struct nlattr *tb[IFLA_MAX + 1];
1342         struct net_device *dev;
1343         int err;
1344
1345         memset(tb, 0, sizeof(tb));
1346         dev = rtnl_create_link(net, name, name_assign_type,
1347                                &geneve_link_ops, tb);
1348         if (IS_ERR(dev))
1349                 return dev;
1350
1351         err = geneve_configure(net, dev, &geneve_remote_unspec,
1352                                0, 0, 0, htons(dst_port), true);
1353         if (err)
1354                 goto err;
1355
1356         /* openvswitch users expect packet sizes to be unrestricted,
1357          * so set the largest MTU we can.
1358          */
1359         err = geneve_change_mtu(dev, IP_MAX_MTU);
1360         if (err)
1361                 goto err;
1362
1363         return dev;
1364
1365  err:
1366         free_netdev(dev);
1367         return ERR_PTR(err);
1368 }
1369 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1370
1371 static __net_init int geneve_init_net(struct net *net)
1372 {
1373         struct geneve_net *gn = net_generic(net, geneve_net_id);
1374
1375         INIT_LIST_HEAD(&gn->geneve_list);
1376         INIT_LIST_HEAD(&gn->sock_list);
1377         return 0;
1378 }
1379
1380 static void __net_exit geneve_exit_net(struct net *net)
1381 {
1382         struct geneve_net *gn = net_generic(net, geneve_net_id);
1383         struct geneve_dev *geneve, *next;
1384         struct net_device *dev, *aux;
1385         LIST_HEAD(list);
1386
1387         rtnl_lock();
1388
1389         /* gather any geneve devices that were moved into this ns */
1390         for_each_netdev_safe(net, dev, aux)
1391                 if (dev->rtnl_link_ops == &geneve_link_ops)
1392                         unregister_netdevice_queue(dev, &list);
1393
1394         /* now gather any other geneve devices that were created in this ns */
1395         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1396                 /* If geneve->dev is in the same netns, it was already added
1397                  * to the list by the previous loop.
1398                  */
1399                 if (!net_eq(dev_net(geneve->dev), net))
1400                         unregister_netdevice_queue(geneve->dev, &list);
1401         }
1402
1403         /* unregister the devices gathered above */
1404         unregister_netdevice_many(&list);
1405         rtnl_unlock();
1406 }
1407
1408 static struct pernet_operations geneve_net_ops = {
1409         .init = geneve_init_net,
1410         .exit = geneve_exit_net,
1411         .id   = &geneve_net_id,
1412         .size = sizeof(struct geneve_net),
1413 };
1414
1415 static int __init geneve_init_module(void)
1416 {
1417         int rc;
1418
1419         rc = register_pernet_subsys(&geneve_net_ops);
1420         if (rc)
1421                 goto out1;
1422
1423         rc = rtnl_link_register(&geneve_link_ops);
1424         if (rc)
1425                 goto out2;
1426
1427         return 0;
1428 out2:
1429         unregister_pernet_subsys(&geneve_net_ops);
1430 out1:
1431         return rc;
1432 }
1433 late_initcall(geneve_init_module);
1434
1435 static void __exit geneve_cleanup_module(void)
1436 {
1437         rtnl_link_unregister(&geneve_link_ops);
1438         unregister_pernet_subsys(&geneve_net_ops);
1439 }
1440 module_exit(geneve_cleanup_module);
1441
1442 MODULE_LICENSE("GPL");
1443 MODULE_VERSION(GENEVE_NETDEV_VER);
1444 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1445 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1446 MODULE_ALIAS_RTNL_LINK("geneve");