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/etherdevice.h>
  16 #include <linux/hash.h>
  17 #include <net/dst_metadata.h>
  18 #include <net/gro_cells.h>
  19 #include <net/rtnetlink.h>
  20 #include <net/geneve.h>
  21 #include <net/protocol.h>
  22
  23 #define GENEVE_NETDEV_VER       "0.6"
  24
  25 #define GENEVE_UDP_PORT         6081
  26
  27 #define GENEVE_N_VID            (1u << 24)
  28 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  29
  30 #define VNI_HASH_BITS           10
  31 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  32
  33 static bool log_ecn_error = true;
  34 module_param(log_ecn_error, bool, 0644);
  35 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  36
  37 #define GENEVE_VER 0
  38 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  39
  40 /* per-network namespace private data for this module */
  41 struct geneve_net {
  42         struct list_head        geneve_list;
  43         struct list_head        sock_list;
  44 };
  45
  46 static int geneve_net_id;
  47
  48 union geneve_addr {
  49         struct sockaddr_in sin;
  50         struct sockaddr_in6 sin6;
  51         struct sockaddr sa;
  52 };
  53
  54 static union geneve_addr geneve_remote_unspec = { .sa.sa_family = AF_UNSPEC, };
  55
  56 /* Pseudo network device */
  57 struct geneve_dev {
  58         struct hlist_node  hlist;       /* vni hash table */
  59         struct net         *net;        /* netns for packet i/o */
  60         struct net_device  *dev;        /* netdev for geneve tunnel */
  61         struct geneve_sock __rcu *sock4;        /* IPv4 socket used for geneve tunnel */
  62 #if IS_ENABLED(CONFIG_IPV6)
  63         struct geneve_sock __rcu *sock6;        /* IPv6 socket used for geneve tunnel */
  64 #endif
  65         u8                 vni[3];      /* virtual network ID for tunnel */
  66         u8                 ttl;         /* TTL override */
  67         u8                 tos;         /* TOS override */
  68         union geneve_addr  remote;      /* IP address for link partner */
  69         struct list_head   next;        /* geneve's per namespace list */
  70         __be32             label;       /* IPv6 flowlabel override */
  71         __be16             dst_port;
  72         bool               collect_md;
  73         struct gro_cells   gro_cells;
  74         u32                flags;
  75         struct dst_cache   dst_cache;
  76 };
  77
  78 /* Geneve device flags */
  79 #define GENEVE_F_UDP_ZERO_CSUM_TX       BIT(0)
  80 #define GENEVE_F_UDP_ZERO_CSUM6_TX      BIT(1)
  81 #define GENEVE_F_UDP_ZERO_CSUM6_RX      BIT(2)
  82
  83 struct geneve_sock {
  84         bool                    collect_md;
  85         struct list_head        list;
  86         struct socket           *sock;
  87         struct rcu_head         rcu;
  88         int                     refcnt;
  89         struct hlist_head       vni_list[VNI_HASH_SIZE];
  90         u32                     flags;
  91 };
  92
  93 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  94 {
  95         __u32 vnid;
  96
  97         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  98         return hash_32(vnid, VNI_HASH_BITS);
  99 }
 100
 101 static __be64 vni_to_tunnel_id(const __u8 *vni)
 102 {
 103 #ifdef __BIG_ENDIAN
 104         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
 105 #else
 106         return (__force __be64)(((__force u64)vni[0] << 40) |
 107                                 ((__force u64)vni[1] << 48) |
 108                                 ((__force u64)vni[2] << 56));
 109 #endif
 110 }
 111
 112 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 113 {
 114         return gs->sock->sk->sk_family;
 115 }
 116
 117 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 118                                         __be32 addr, u8 vni[])
 119 {
 120         struct hlist_head *vni_list_head;
 121         struct geneve_dev *geneve;
 122         __u32 hash;
 123
 124         /* Find the device for this VNI */
 125         hash = geneve_net_vni_hash(vni);
 126         vni_list_head = &gs->vni_list[hash];
 127         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 128                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 129                     addr == geneve->remote.sin.sin_addr.s_addr)
 130                         return geneve;
 131         }
 132         return NULL;
 133 }
 134
 135 #if IS_ENABLED(CONFIG_IPV6)
 136 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 137                                          struct in6_addr addr6, u8 vni[])
 138 {
 139         struct hlist_head *vni_list_head;
 140         struct geneve_dev *geneve;
 141         __u32 hash;
 142
 143         /* Find the device for this VNI */
 144         hash = geneve_net_vni_hash(vni);
 145         vni_list_head = &gs->vni_list[hash];
 146         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 147                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 148                     ipv6_addr_equal(&addr6, &geneve->remote.sin6.sin6_addr))
 149                         return geneve;
 150         }
 151         return NULL;
 152 }
 153 #endif
 154
 155 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 156 {
 157         return (struct genevehdr *)(udp_hdr(skb) + 1);
 158 }
 159
 160 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 161                                             struct sk_buff *skb)
 162 {
 163         u8 *vni;
 164         __be32 addr;
 165         static u8 zero_vni[3];
 166 #if IS_ENABLED(CONFIG_IPV6)
 167         static struct in6_addr zero_addr6;
 168 #endif
 169
 170         if (geneve_get_sk_family(gs) == AF_INET) {
 171                 struct iphdr *iph;
 172
 173                 iph = ip_hdr(skb); /* outer IP header... */
 174
 175                 if (gs->collect_md) {
 176                         vni = zero_vni;
 177                         addr = 0;
 178                 } else {
 179                         vni = geneve_hdr(skb)->vni;
 180                         addr = iph->saddr;
 181                 }
 182
 183                 return geneve_lookup(gs, addr, vni);
 184 #if IS_ENABLED(CONFIG_IPV6)
 185         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 186                 struct ipv6hdr *ip6h;
 187                 struct in6_addr addr6;
 188
 189                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 190
 191                 if (gs->collect_md) {
 192                         vni = zero_vni;
 193                         addr6 = zero_addr6;
 194                 } else {
 195                         vni = geneve_hdr(skb)->vni;
 196                         addr6 = ip6h->saddr;
 197                 }
 198
 199                 return geneve6_lookup(gs, addr6, vni);
 200 #endif
 201         }
 202         return NULL;
 203 }
 204
 205 /* geneve receive/decap routine */
 206 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 207                       struct sk_buff *skb)
 208 {
 209         struct genevehdr *gnvh = geneve_hdr(skb);
 210         struct metadata_dst *tun_dst = NULL;
 211         struct pcpu_sw_netstats *stats;
 212         unsigned int len;
 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                         geneve->dev->stats.rx_dropped++;
 228                         goto drop;
 229                 }
 230                 /* Update tunnel dst according to Geneve options. */
 231                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 232                                         gnvh->options, gnvh->opt_len * 4);
 233         } else {
 234                 /* Drop packets w/ critical options,
 235                  * since we don't support any...
 236                  */
 237                 if (gnvh->critical) {
 238                         geneve->dev->stats.rx_frame_errors++;
 239                         geneve->dev->stats.rx_errors++;
 240                         goto drop;
 241                 }
 242         }
 243
 244         skb_reset_mac_header(skb);
 245         skb->protocol = eth_type_trans(skb, geneve->dev);
 246         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 247
 248         if (tun_dst)
 249                 skb_dst_set(skb, &tun_dst->dst);
 250
 251         /* Ignore packet loops (and multicast echo) */
 252         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
 253                 geneve->dev->stats.rx_errors++;
 254                 goto drop;
 255         }
 256
 257         oiph = skb_network_header(skb);
 258         skb_reset_network_header(skb);
 259
 260         if (geneve_get_sk_family(gs) == AF_INET)
 261                 err = IP_ECN_decapsulate(oiph, skb);
 262 #if IS_ENABLED(CONFIG_IPV6)
 263         else
 264                 err = IP6_ECN_decapsulate(oiph, skb);
 265 #endif
 266
 267         if (unlikely(err)) {
 268                 if (log_ecn_error) {
 269                         if (geneve_get_sk_family(gs) == AF_INET)
 270                                 net_info_ratelimited("non-ECT from %pI4 "
 271                                                      "with TOS=%#x\n",
 272                                                      &((struct iphdr *)oiph)->saddr,
 273                                                      ((struct iphdr *)oiph)->tos);
 274 #if IS_ENABLED(CONFIG_IPV6)
 275                         else
 276                                 net_info_ratelimited("non-ECT from %pI6\n",
 277                                                      &((struct ipv6hdr *)oiph)->saddr);
 278 #endif
 279                 }
 280                 if (err > 1) {
 281                         ++geneve->dev->stats.rx_frame_errors;
 282                         ++geneve->dev->stats.rx_errors;
 283                         goto drop;
 284                 }
 285         }
 286
 287         len = skb->len;
 288         err = gro_cells_receive(&geneve->gro_cells, skb);
 289         if (likely(err == NET_RX_SUCCESS)) {
 290                 stats = this_cpu_ptr(geneve->dev->tstats);
 291                 u64_stats_update_begin(&stats->syncp);
 292                 stats->rx_packets++;
 293                 stats->rx_bytes += len;
 294                 u64_stats_update_end(&stats->syncp);
 295         }
 296         return;
 297 drop:
 298         /* Consume bad packet */
 299         kfree_skb(skb);
 300 }
 301
 302 /* Setup stats when device is created */
 303 static int geneve_init(struct net_device *dev)
 304 {
 305         struct geneve_dev *geneve = netdev_priv(dev);
 306         int err;
 307
 308         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 309         if (!dev->tstats)
 310                 return -ENOMEM;
 311
 312         err = gro_cells_init(&geneve->gro_cells, dev);
 313         if (err) {
 314                 free_percpu(dev->tstats);
 315                 return err;
 316         }
 317
 318         err = dst_cache_init(&geneve->dst_cache, GFP_KERNEL);
 319         if (err) {
 320                 free_percpu(dev->tstats);
 321                 gro_cells_destroy(&geneve->gro_cells);
 322                 return err;
 323         }
 324
 325         return 0;
 326 }
 327
 328 static void geneve_uninit(struct net_device *dev)
 329 {
 330         struct geneve_dev *geneve = netdev_priv(dev);
 331
 332         dst_cache_destroy(&geneve->dst_cache);
 333         gro_cells_destroy(&geneve->gro_cells);
 334         free_percpu(dev->tstats);
 335 }
 336
 337 /* Callback from net/ipv4/udp.c to receive packets */
 338 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 339 {
 340         struct genevehdr *geneveh;
 341         struct geneve_dev *geneve;
 342         struct geneve_sock *gs;
 343         int opts_len;
 344
 345         /* Need UDP and Geneve header to be present */
 346         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 347                 goto drop;
 348
 349         /* Return packets with reserved bits set */
 350         geneveh = geneve_hdr(skb);
 351         if (unlikely(geneveh->ver != GENEVE_VER))
 352                 goto drop;
 353
 354         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 355                 goto drop;
 356
 357         gs = rcu_dereference_sk_user_data(sk);
 358         if (!gs)
 359                 goto drop;
 360
 361         geneve = geneve_lookup_skb(gs, skb);
 362         if (!geneve)
 363                 goto drop;
 364
 365         opts_len = geneveh->opt_len * 4;
 366         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 367                                  htons(ETH_P_TEB),
 368                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 369                 geneve->dev->stats.rx_dropped++;
 370                 goto drop;
 371         }
 372
 373         geneve_rx(geneve, gs, skb);
 374         return 0;
 375
 376 drop:
 377         /* Consume bad packet */
 378         kfree_skb(skb);
 379         return 0;
 380 }
 381
 382 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 383                                          __be16 port, u32 flags)
 384 {
 385         struct socket *sock;
 386         struct udp_port_cfg udp_conf;
 387         int err;
 388
 389         memset(&udp_conf, 0, sizeof(udp_conf));
 390
 391         if (ipv6) {
 392                 udp_conf.family = AF_INET6;
 393                 udp_conf.ipv6_v6only = 1;
 394                 udp_conf.use_udp6_rx_checksums =
 395                     !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX);
 396         } else {
 397                 udp_conf.family = AF_INET;
 398                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 399         }
 400
 401         udp_conf.local_udp_port = port;
 402
 403         /* Open UDP socket */
 404         err = udp_sock_create(net, &udp_conf, &sock);
 405         if (err < 0)
 406                 return ERR_PTR(err);
 407
 408         return sock;
 409 }
 410
 411 static int geneve_hlen(struct genevehdr *gh)
 412 {
 413         return sizeof(*gh) + gh->opt_len * 4;
 414 }
 415
 416 static struct sk_buff **geneve_gro_receive(struct sock *sk,
 417                                            struct sk_buff **head,
 418                                            struct sk_buff *skb)
 419 {
 420         struct sk_buff *p, **pp = NULL;
 421         struct genevehdr *gh, *gh2;
 422         unsigned int hlen, gh_len, off_gnv;
 423         const struct packet_offload *ptype;
 424         __be16 type;
 425         int flush = 1;
 426
 427         off_gnv = skb_gro_offset(skb);
 428         hlen = off_gnv + sizeof(*gh);
 429         gh = skb_gro_header_fast(skb, off_gnv);
 430         if (skb_gro_header_hard(skb, hlen)) {
 431                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 432                 if (unlikely(!gh))
 433                         goto out;
 434         }
 435
 436         if (gh->ver != GENEVE_VER || gh->oam)
 437                 goto out;
 438         gh_len = geneve_hlen(gh);
 439
 440         hlen = off_gnv + gh_len;
 441         if (skb_gro_header_hard(skb, hlen)) {
 442                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 443                 if (unlikely(!gh))
 444                         goto out;
 445         }
 446
 447         for (p = *head; p; p = p->next) {
 448                 if (!NAPI_GRO_CB(p)->same_flow)
 449                         continue;
 450
 451                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 452                 if (gh->opt_len != gh2->opt_len ||
 453                     memcmp(gh, gh2, gh_len)) {
 454                         NAPI_GRO_CB(p)->same_flow = 0;
 455                         continue;
 456                 }
 457         }
 458
 459         type = gh->proto_type;
 460
 461         rcu_read_lock();
 462         ptype = gro_find_receive_by_type(type);
 463         if (!ptype)
 464                 goto out_unlock;
 465
 466         skb_gro_pull(skb, gh_len);
 467         skb_gro_postpull_rcsum(skb, gh, gh_len);
 468         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 469         flush = 0;
 470
 471 out_unlock:
 472         rcu_read_unlock();
 473 out:
 474         NAPI_GRO_CB(skb)->flush |= flush;
 475
 476         return pp;
 477 }
 478
 479 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 480                                int nhoff)
 481 {
 482         struct genevehdr *gh;
 483         struct packet_offload *ptype;
 484         __be16 type;
 485         int gh_len;
 486         int err = -ENOSYS;
 487
 488         gh = (struct genevehdr *)(skb->data + nhoff);
 489         gh_len = geneve_hlen(gh);
 490         type = gh->proto_type;
 491
 492         rcu_read_lock();
 493         ptype = gro_find_complete_by_type(type);
 494         if (ptype)
 495                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 496
 497         rcu_read_unlock();
 498
 499         skb_set_inner_mac_header(skb, nhoff + gh_len);
 500
 501         return err;
 502 }
 503
 504 /* Create new listen socket if needed */
 505 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 506                                                 bool ipv6, u32 flags)
 507 {
 508         struct geneve_net *gn = net_generic(net, geneve_net_id);
 509         struct geneve_sock *gs;
 510         struct socket *sock;
 511         struct udp_tunnel_sock_cfg tunnel_cfg;
 512         int h;
 513
 514         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 515         if (!gs)
 516                 return ERR_PTR(-ENOMEM);
 517
 518         sock = geneve_create_sock(net, ipv6, port, flags);
 519         if (IS_ERR(sock)) {
 520                 kfree(gs);
 521                 return ERR_CAST(sock);
 522         }
 523
 524         gs->sock = sock;
 525         gs->refcnt = 1;
 526         for (h = 0; h < VNI_HASH_SIZE; ++h)
 527                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 528
 529         /* Initialize the geneve udp offloads structure */
 530         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 531
 532         /* Mark socket as an encapsulation socket */
 533         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 534         tunnel_cfg.sk_user_data = gs;
 535         tunnel_cfg.encap_type = 1;
 536         tunnel_cfg.gro_receive = geneve_gro_receive;
 537         tunnel_cfg.gro_complete = geneve_gro_complete;
 538         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 539         tunnel_cfg.encap_destroy = NULL;
 540         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 541         list_add(&gs->list, &gn->sock_list);
 542         return gs;
 543 }
 544
 545 static void __geneve_sock_release(struct geneve_sock *gs)
 546 {
 547         if (!gs || --gs->refcnt)
 548                 return;
 549
 550         list_del(&gs->list);
 551         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 552         udp_tunnel_sock_release(gs->sock);
 553         kfree_rcu(gs, rcu);
 554 }
 555
 556 static void geneve_sock_release(struct geneve_dev *geneve)
 557 {
 558         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 559 #if IS_ENABLED(CONFIG_IPV6)
 560         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 561
 562         rcu_assign_pointer(geneve->sock6, NULL);
 563 #endif
 564
 565         rcu_assign_pointer(geneve->sock4, NULL);
 566         synchronize_net();
 567
 568         __geneve_sock_release(gs4);
 569 #if IS_ENABLED(CONFIG_IPV6)
 570         __geneve_sock_release(gs6);
 571 #endif
 572 }
 573
 574 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 575                                             sa_family_t family,
 576                                             __be16 dst_port)
 577 {
 578         struct geneve_sock *gs;
 579
 580         list_for_each_entry(gs, &gn->sock_list, list) {
 581                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 582                     geneve_get_sk_family(gs) == family) {
 583                         return gs;
 584                 }
 585         }
 586         return NULL;
 587 }
 588
 589 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 590 {
 591         struct net *net = geneve->net;
 592         struct geneve_net *gn = net_generic(net, geneve_net_id);
 593         struct geneve_sock *gs;
 594         __u32 hash;
 595
 596         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
 597         if (gs) {
 598                 gs->refcnt++;
 599                 goto out;
 600         }
 601
 602         gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
 603         if (IS_ERR(gs))
 604                 return PTR_ERR(gs);
 605
 606 out:
 607         gs->collect_md = geneve->collect_md;
 608         gs->flags = geneve->flags;
 609 #if IS_ENABLED(CONFIG_IPV6)
 610         if (ipv6)
 611                 rcu_assign_pointer(geneve->sock6, gs);
 612         else
 613 #endif
 614                 rcu_assign_pointer(geneve->sock4, gs);
 615
 616         hash = geneve_net_vni_hash(geneve->vni);
 617         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 618         return 0;
 619 }
 620
 621 static int geneve_open(struct net_device *dev)
 622 {
 623         struct geneve_dev *geneve = netdev_priv(dev);
 624         bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
 625         bool metadata = geneve->collect_md;
 626         int ret = 0;
 627
 628 #if IS_ENABLED(CONFIG_IPV6)
 629         if (ipv6 || metadata)
 630                 ret = geneve_sock_add(geneve, true);
 631 #endif
 632         if (!ret && (!ipv6 || metadata))
 633                 ret = geneve_sock_add(geneve, false);
 634         if (ret < 0)
 635                 geneve_sock_release(geneve);
 636
 637         return ret;
 638 }
 639
 640 static int geneve_stop(struct net_device *dev)
 641 {
 642         struct geneve_dev *geneve = netdev_priv(dev);
 643
 644         if (!hlist_unhashed(&geneve->hlist))
 645                 hlist_del_rcu(&geneve->hlist);
 646         geneve_sock_release(geneve);
 647         return 0;
 648 }
 649
 650 static void geneve_build_header(struct genevehdr *geneveh,
 651                                 __be16 tun_flags, u8 vni[3],
 652                                 u8 options_len, u8 *options)
 653 {
 654         geneveh->ver = GENEVE_VER;
 655         geneveh->opt_len = options_len / 4;
 656         geneveh->oam = !!(tun_flags & TUNNEL_OAM);
 657         geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 658         geneveh->rsvd1 = 0;
 659         memcpy(geneveh->vni, vni, 3);
 660         geneveh->proto_type = htons(ETH_P_TEB);
 661         geneveh->rsvd2 = 0;
 662
 663         memcpy(geneveh->options, options, options_len);
 664 }
 665
 666 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 667                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 668                             u32 flags, bool xnet)
 669 {
 670         struct genevehdr *gnvh;
 671         int min_headroom;
 672         int err;
 673         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
 674
 675         skb_scrub_packet(skb, xnet);
 676
 677         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 678                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 679         err = skb_cow_head(skb, min_headroom);
 680         if (unlikely(err))
 681                 goto free_rt;
 682
 683         err = udp_tunnel_handle_offloads(skb, udp_sum);
 684         if (err)
 685                 goto free_rt;
 686
 687         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 688         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 689
 690         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 691         return 0;
 692
 693 free_rt:
 694         ip_rt_put(rt);
 695         return err;
 696 }
 697
 698 #if IS_ENABLED(CONFIG_IPV6)
 699 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 700                              __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 701                              u32 flags, bool xnet)
 702 {
 703         struct genevehdr *gnvh;
 704         int min_headroom;
 705         int err;
 706         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
 707
 708         skb_scrub_packet(skb, xnet);
 709
 710         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
 711                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
 712         err = skb_cow_head(skb, min_headroom);
 713         if (unlikely(err))
 714                 goto free_dst;
 715
 716         err = udp_tunnel_handle_offloads(skb, udp_sum);
 717         if (err)
 718                 goto free_dst;
 719
 720         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 721         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 722
 723         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 724         return 0;
 725
 726 free_dst:
 727         dst_release(dst);
 728         return err;
 729 }
 730 #endif
 731
 732 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 733                                        struct net_device *dev,
 734                                        struct flowi4 *fl4,
 735                                        struct ip_tunnel_info *info,
 736                                        __be16 dport, __be16 sport)
 737 {
 738         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 739         struct geneve_dev *geneve = netdev_priv(dev);
 740         struct dst_cache *dst_cache;
 741         struct rtable *rt = NULL;
 742         __u8 tos;
 743
 744         if (!rcu_dereference(geneve->sock4))
 745                 return ERR_PTR(-EIO);
 746
 747         memset(fl4, 0, sizeof(*fl4));
 748         fl4->flowi4_mark = skb->mark;
 749         fl4->flowi4_proto = IPPROTO_UDP;
 750         fl4->fl4_dport = dport;
 751         fl4->fl4_sport = sport;
 752
 753         if (info) {
 754                 fl4->daddr = info->key.u.ipv4.dst;
 755                 fl4->saddr = info->key.u.ipv4.src;
 756                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 757                 dst_cache = &info->dst_cache;
 758         } else {
 759                 tos = geneve->tos;
 760                 if (tos == 1) {
 761                         const struct iphdr *iip = ip_hdr(skb);
 762
 763                         tos = ip_tunnel_get_dsfield(iip, skb);
 764                         use_cache = false;
 765                 }
 766
 767                 fl4->flowi4_tos = RT_TOS(tos);
 768                 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
 769                 dst_cache = &geneve->dst_cache;
 770         }
 771
 772         if (use_cache) {
 773                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 774                 if (rt)
 775                         return rt;
 776         }
 777
 778         rt = ip_route_output_key(geneve->net, fl4);
 779         if (IS_ERR(rt)) {
 780                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 781                 return ERR_PTR(-ENETUNREACH);
 782         }
 783         if (rt->dst.dev == dev) { /* is this necessary? */
 784                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 785                 ip_rt_put(rt);
 786                 return ERR_PTR(-ELOOP);
 787         }
 788         if (use_cache)
 789                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 790         return rt;
 791 }
 792
 793 #if IS_ENABLED(CONFIG_IPV6)
 794 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 795                                            struct net_device *dev,
 796                                            struct flowi6 *fl6,
 797                                            struct ip_tunnel_info *info,
 798                                            __be16 dport, __be16 sport)
 799 {
 800         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 801         struct geneve_dev *geneve = netdev_priv(dev);
 802         struct dst_entry *dst = NULL;
 803         struct dst_cache *dst_cache;
 804         struct geneve_sock *gs6;
 805         __u8 prio;
 806
 807         gs6 = rcu_dereference(geneve->sock6);
 808         if (!gs6)
 809                 return ERR_PTR(-EIO);
 810
 811         memset(fl6, 0, sizeof(*fl6));
 812         fl6->flowi6_mark = skb->mark;
 813         fl6->flowi6_proto = IPPROTO_UDP;
 814         fl6->fl6_dport = dport;
 815         fl6->fl6_sport = sport;
 816
 817         if (info) {
 818                 fl6->daddr = info->key.u.ipv6.dst;
 819                 fl6->saddr = info->key.u.ipv6.src;
 820                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(info->key.tos),
 821                                                    info->key.label);
 822                 dst_cache = &info->dst_cache;
 823         } else {
 824                 prio = geneve->tos;
 825                 if (prio == 1) {
 826                         const struct iphdr *iip = ip_hdr(skb);
 827
 828                         prio = ip_tunnel_get_dsfield(iip, skb);
 829                         use_cache = false;
 830                 }
 831
 832                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 833                                                    geneve->label);
 834                 fl6->daddr = geneve->remote.sin6.sin6_addr;
 835                 dst_cache = &geneve->dst_cache;
 836         }
 837
 838         if (use_cache) {
 839                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 840                 if (dst)
 841                         return dst;
 842         }
 843
 844         dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
 845                                               NULL);
 846         if (IS_ERR(dst)) {
 847                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 848                 return ERR_PTR(-ENETUNREACH);
 849         }
 850         if (dst->dev == dev) { /* is this necessary? */
 851                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 852                 dst_release(dst);
 853                 return ERR_PTR(-ELOOP);
 854         }
 855
 856         if (use_cache)
 857                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 858         return dst;
 859 }
 860 #endif
 861
 862 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 863 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 864 {
 865 #ifdef __BIG_ENDIAN
 866         vni[0] = (__force __u8)(tun_id >> 16);
 867         vni[1] = (__force __u8)(tun_id >> 8);
 868         vni[2] = (__force __u8)tun_id;
 869 #else
 870         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 871         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 872         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 873 #endif
 874 }
 875
 876 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 877                                    struct ip_tunnel_info *info)
 878 {
 879         struct geneve_dev *geneve = netdev_priv(dev);
 880         struct geneve_sock *gs4;
 881         struct rtable *rt = NULL;
 882         int err = -EINVAL;
 883         struct flowi4 fl4;
 884         __u8 tos, ttl;
 885         __be16 sport;
 886         __be16 df;
 887         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 888         u32 flags = geneve->flags;
 889
 890         gs4 = rcu_dereference(geneve->sock4);
 891         if (!gs4)
 892                 goto tx_error;
 893
 894         if (geneve->collect_md) {
 895                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 896                         netdev_dbg(dev, "no tunnel metadata\n");
 897                         goto tx_error;
 898                 }
 899                 if (info && ip_tunnel_info_af(info) != AF_INET)
 900                         goto tx_error;
 901         }
 902
 903         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 904         rt = geneve_get_v4_rt(skb, dev, &fl4, info,
 905                               geneve->dst_port, sport);
 906         if (IS_ERR(rt)) {
 907                 err = PTR_ERR(rt);
 908                 goto tx_error;
 909         }
 910
 911         skb_reset_mac_header(skb);
 912
 913         if (info) {
 914                 const struct ip_tunnel_key *key = &info->key;
 915                 u8 *opts = NULL;
 916                 u8 vni[3];
 917
 918                 tunnel_id_to_vni(key->tun_id, vni);
 919                 if (info->options_len)
 920                         opts = ip_tunnel_info_opts(info);
 921
 922                 if (key->tun_flags & TUNNEL_CSUM)
 923                         flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
 924                 else
 925                         flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
 926
 927                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 928                                        info->options_len, opts, flags, xnet);
 929                 if (unlikely(err))
 930                         goto tx_error;
 931
 932                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 933                 ttl = key->ttl;
 934                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 935         } else {
 936                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 937                                        0, NULL, flags, xnet);
 938                 if (unlikely(err))
 939                         goto tx_error;
 940
 941                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 942                 ttl = geneve->ttl;
 943                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 944                         ttl = 1;
 945                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 946                 df = 0;
 947         }
 948         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 949                             tos, ttl, df, sport, geneve->dst_port,
 950                             !net_eq(geneve->net, dev_net(geneve->dev)),
 951                             !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
 952
 953         return NETDEV_TX_OK;
 954
 955 tx_error:
 956         dev_kfree_skb(skb);
 957
 958         if (err == -ELOOP)
 959                 dev->stats.collisions++;
 960         else if (err == -ENETUNREACH)
 961                 dev->stats.tx_carrier_errors++;
 962
 963         dev->stats.tx_errors++;
 964         return NETDEV_TX_OK;
 965 }
 966
 967 #if IS_ENABLED(CONFIG_IPV6)
 968 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 969                                     struct ip_tunnel_info *info)
 970 {
 971         struct geneve_dev *geneve = netdev_priv(dev);
 972         struct dst_entry *dst = NULL;
 973         struct geneve_sock *gs6;
 974         int err = -EINVAL;
 975         struct flowi6 fl6;
 976         __u8 prio, ttl;
 977         __be16 sport;
 978         __be32 label;
 979         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 980         u32 flags = geneve->flags;
 981
 982         gs6 = rcu_dereference(geneve->sock6);
 983         if (!gs6)
 984                 goto tx_error;
 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         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 994         dst = geneve_get_v6_dst(skb, dev, &fl6, info,
 995                                 geneve->dst_port, sport);
 996         if (IS_ERR(dst)) {
 997                 err = PTR_ERR(dst);
 998                 goto tx_error;
 999         }
1000
1001         skb_reset_mac_header(skb);
1002
1003         if (info) {
1004                 const struct ip_tunnel_key *key = &info->key;
1005                 u8 *opts = NULL;
1006                 u8 vni[3];
1007
1008                 tunnel_id_to_vni(key->tun_id, vni);
1009                 if (info->options_len)
1010                         opts = ip_tunnel_info_opts(info);
1011
1012                 if (key->tun_flags & TUNNEL_CSUM)
1013                         flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
1014                 else
1015                         flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1016
1017                 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
1018                                         info->options_len, opts,
1019                                         flags, xnet);
1020                 if (unlikely(err))
1021                         goto tx_error;
1022
1023                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1024                 ttl = key->ttl;
1025                 label = info->key.label;
1026         } else {
1027                 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
1028                                         0, NULL, flags, xnet);
1029                 if (unlikely(err))
1030                         goto tx_error;
1031
1032                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1033                                            ip_hdr(skb), skb);
1034                 ttl = geneve->ttl;
1035                 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1036                         ttl = 1;
1037                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1038                 label = geneve->label;
1039         }
1040
1041         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1042                              &fl6.saddr, &fl6.daddr, prio, ttl, label,
1043                              sport, geneve->dst_port,
1044                              !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1045         return NETDEV_TX_OK;
1046
1047 tx_error:
1048         dev_kfree_skb(skb);
1049
1050         if (err == -ELOOP)
1051                 dev->stats.collisions++;
1052         else if (err == -ENETUNREACH)
1053                 dev->stats.tx_carrier_errors++;
1054
1055         dev->stats.tx_errors++;
1056         return NETDEV_TX_OK;
1057 }
1058 #endif
1059
1060 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1061 {
1062         struct geneve_dev *geneve = netdev_priv(dev);
1063         struct ip_tunnel_info *info = NULL;
1064         int err;
1065
1066         if (geneve->collect_md)
1067                 info = skb_tunnel_info(skb);
1068
1069         rcu_read_lock();
1070 #if IS_ENABLED(CONFIG_IPV6)
1071         if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1072             (!info && geneve->remote.sa.sa_family == AF_INET6))
1073                 err = geneve6_xmit_skb(skb, dev, info);
1074         else
1075 #endif
1076                 err = geneve_xmit_skb(skb, dev, info);
1077         rcu_read_unlock();
1078
1079         return err;
1080 }
1081
1082 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1083 {
1084         struct geneve_dev *geneve = netdev_priv(dev);
1085         /* The max_mtu calculation does not take account of GENEVE
1086          * options, to avoid excluding potentially valid
1087          * configurations.
1088          */
1089         int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1090
1091         if (geneve->remote.sa.sa_family == AF_INET6)
1092                 max_mtu -= sizeof(struct ipv6hdr);
1093         else
1094                 max_mtu -= sizeof(struct iphdr);
1095
1096         if (new_mtu < 68)
1097                 return -EINVAL;
1098
1099         if (new_mtu > max_mtu) {
1100                 if (strict)
1101                         return -EINVAL;
1102
1103                 new_mtu = max_mtu;
1104         }
1105
1106         dev->mtu = new_mtu;
1107         return 0;
1108 }
1109
1110 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1111 {
1112         return __geneve_change_mtu(dev, new_mtu, true);
1113 }
1114
1115 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1116 {
1117         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1118         struct geneve_dev *geneve = netdev_priv(dev);
1119         struct rtable *rt;
1120         struct flowi4 fl4;
1121 #if IS_ENABLED(CONFIG_IPV6)
1122         struct dst_entry *dst;
1123         struct flowi6 fl6;
1124 #endif
1125         __be16 sport;
1126
1127         if (ip_tunnel_info_af(info) == AF_INET) {
1128                 sport = udp_flow_src_port(geneve->net, skb,
1129                                           1, USHRT_MAX, true);
1130
1131                 rt = geneve_get_v4_rt(skb, dev, &fl4, info,
1132                                       geneve->dst_port, sport);
1133                 if (IS_ERR(rt))
1134                         return PTR_ERR(rt);
1135
1136                 ip_rt_put(rt);
1137                 info->key.u.ipv4.src = fl4.saddr;
1138 #if IS_ENABLED(CONFIG_IPV6)
1139         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1140                 sport = udp_flow_src_port(geneve->net, skb,
1141                                           1, USHRT_MAX, true);
1142
1143                 dst = geneve_get_v6_dst(skb, dev, &fl6, info,
1144                                         geneve->dst_port, sport);
1145                 if (IS_ERR(dst))
1146                         return PTR_ERR(dst);
1147
1148                 dst_release(dst);
1149                 info->key.u.ipv6.src = fl6.saddr;
1150 #endif
1151         } else {
1152                 return -EINVAL;
1153         }
1154
1155         info->key.tp_src = sport;
1156         info->key.tp_dst = geneve->dst_port;
1157         return 0;
1158 }
1159
1160 static const struct net_device_ops geneve_netdev_ops = {
1161         .ndo_init               = geneve_init,
1162         .ndo_uninit             = geneve_uninit,
1163         .ndo_open               = geneve_open,
1164         .ndo_stop               = geneve_stop,
1165         .ndo_start_xmit         = geneve_xmit,
1166         .ndo_get_stats64        = ip_tunnel_get_stats64,
1167         .ndo_change_mtu         = geneve_change_mtu,
1168         .ndo_validate_addr      = eth_validate_addr,
1169         .ndo_set_mac_address    = eth_mac_addr,
1170         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1171 };
1172
1173 static void geneve_get_drvinfo(struct net_device *dev,
1174                                struct ethtool_drvinfo *drvinfo)
1175 {
1176         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1177         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1178 }
1179
1180 static const struct ethtool_ops geneve_ethtool_ops = {
1181         .get_drvinfo    = geneve_get_drvinfo,
1182         .get_link       = ethtool_op_get_link,
1183 };
1184
1185 /* Info for udev, that this is a virtual tunnel endpoint */
1186 static struct device_type geneve_type = {
1187         .name = "geneve",
1188 };
1189
1190 /* Calls the ndo_udp_tunnel_add of the caller in order to
1191  * supply the listening GENEVE udp ports. Callers are expected
1192  * to implement the ndo_udp_tunnel_add.
1193  */
1194 static void geneve_push_rx_ports(struct net_device *dev)
1195 {
1196         struct net *net = dev_net(dev);
1197         struct geneve_net *gn = net_generic(net, geneve_net_id);
1198         struct geneve_sock *gs;
1199
1200         rcu_read_lock();
1201         list_for_each_entry_rcu(gs, &gn->sock_list, list)
1202                 udp_tunnel_push_rx_port(dev, gs->sock,
1203                                         UDP_TUNNEL_TYPE_GENEVE);
1204         rcu_read_unlock();
1205 }
1206
1207 /* Initialize the device structure. */
1208 static void geneve_setup(struct net_device *dev)
1209 {
1210         ether_setup(dev);
1211
1212         dev->netdev_ops = &geneve_netdev_ops;
1213         dev->ethtool_ops = &geneve_ethtool_ops;
1214         dev->destructor = free_netdev;
1215
1216         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1217
1218         dev->features    |= NETIF_F_LLTX;
1219         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1220         dev->features    |= NETIF_F_RXCSUM;
1221         dev->features    |= NETIF_F_GSO_SOFTWARE;
1222
1223         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1224         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1225
1226         netif_keep_dst(dev);
1227         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1228         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1229         eth_hw_addr_random(dev);
1230 }
1231
1232 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1233         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1234         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1235         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1236         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1237         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1238         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1239         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1240         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1241         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1242         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1243         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1244 };
1245
1246 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1247 {
1248         if (tb[IFLA_ADDRESS]) {
1249                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1250                         return -EINVAL;
1251
1252                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1253                         return -EADDRNOTAVAIL;
1254         }
1255
1256         if (!data)
1257                 return -EINVAL;
1258
1259         if (data[IFLA_GENEVE_ID]) {
1260                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1261
1262                 if (vni >= GENEVE_VID_MASK)
1263                         return -ERANGE;
1264         }
1265
1266         return 0;
1267 }
1268
1269 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1270                                           __be16 dst_port,
1271                                           union geneve_addr *remote,
1272                                           u8 vni[],
1273                                           bool *tun_on_same_port,
1274                                           bool *tun_collect_md)
1275 {
1276         struct geneve_dev *geneve, *t;
1277
1278         *tun_on_same_port = false;
1279         *tun_collect_md = false;
1280         t = NULL;
1281         list_for_each_entry(geneve, &gn->geneve_list, next) {
1282                 if (geneve->dst_port == dst_port) {
1283                         *tun_collect_md = geneve->collect_md;
1284                         *tun_on_same_port = true;
1285                 }
1286                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1287                     !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1288                     dst_port == geneve->dst_port)
1289                         t = geneve;
1290         }
1291         return t;
1292 }
1293
1294 static int geneve_configure(struct net *net, struct net_device *dev,
1295                             union geneve_addr *remote,
1296                             __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1297                             __be16 dst_port, bool metadata, u32 flags)
1298 {
1299         struct geneve_net *gn = net_generic(net, geneve_net_id);
1300         struct geneve_dev *t, *geneve = netdev_priv(dev);
1301         bool tun_collect_md, tun_on_same_port;
1302         int err, encap_len;
1303
1304         if (!remote)
1305                 return -EINVAL;
1306         if (metadata &&
1307             (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1308                 return -EINVAL;
1309
1310         geneve->net = net;
1311         geneve->dev = dev;
1312
1313         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1314         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1315         geneve->vni[2] =  vni & 0x000000ff;
1316
1317         if ((remote->sa.sa_family == AF_INET &&
1318              IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1319             (remote->sa.sa_family == AF_INET6 &&
1320              ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1321                 return -EINVAL;
1322         if (label && remote->sa.sa_family != AF_INET6)
1323                 return -EINVAL;
1324
1325         geneve->remote = *remote;
1326
1327         geneve->ttl = ttl;
1328         geneve->tos = tos;
1329         geneve->label = label;
1330         geneve->dst_port = dst_port;
1331         geneve->collect_md = metadata;
1332         geneve->flags = flags;
1333
1334         t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1335                             &tun_on_same_port, &tun_collect_md);
1336         if (t)
1337                 return -EBUSY;
1338
1339         /* make enough headroom for basic scenario */
1340         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1341         if (remote->sa.sa_family == AF_INET)
1342                 encap_len += sizeof(struct iphdr);
1343         else
1344                 encap_len += sizeof(struct ipv6hdr);
1345         dev->needed_headroom = encap_len + ETH_HLEN;
1346
1347         if (metadata) {
1348                 if (tun_on_same_port)
1349                         return -EPERM;
1350         } else {
1351                 if (tun_collect_md)
1352                         return -EPERM;
1353         }
1354
1355         dst_cache_reset(&geneve->dst_cache);
1356
1357         err = register_netdevice(dev);
1358         if (err)
1359                 return err;
1360
1361         list_add(&geneve->next, &gn->geneve_list);
1362         return 0;
1363 }
1364
1365 static int geneve_newlink(struct net *net, struct net_device *dev,
1366                           struct nlattr *tb[], struct nlattr *data[])
1367 {
1368         __be16 dst_port = htons(GENEVE_UDP_PORT);
1369         __u8 ttl = 0, tos = 0;
1370         bool metadata = false;
1371         union geneve_addr remote = geneve_remote_unspec;
1372         __be32 label = 0;
1373         __u32 vni = 0;
1374         u32 flags = 0;
1375
1376         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1377                 return -EINVAL;
1378
1379         if (data[IFLA_GENEVE_REMOTE]) {
1380                 remote.sa.sa_family = AF_INET;
1381                 remote.sin.sin_addr.s_addr =
1382                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1383         }
1384
1385         if (data[IFLA_GENEVE_REMOTE6]) {
1386                 if (!IS_ENABLED(CONFIG_IPV6))
1387                         return -EPFNOSUPPORT;
1388
1389                 remote.sa.sa_family = AF_INET6;
1390                 remote.sin6.sin6_addr =
1391                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1392
1393                 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1394                     IPV6_ADDR_LINKLOCAL) {
1395                         netdev_dbg(dev, "link-local remote is unsupported\n");
1396                         return -EINVAL;
1397                 }
1398         }
1399
1400         if (data[IFLA_GENEVE_ID])
1401                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1402
1403         if (data[IFLA_GENEVE_TTL])
1404                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1405
1406         if (data[IFLA_GENEVE_TOS])
1407                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1408
1409         if (data[IFLA_GENEVE_LABEL])
1410                 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1411                         IPV6_FLOWLABEL_MASK;
1412
1413         if (data[IFLA_GENEVE_PORT])
1414                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1415
1416         if (data[IFLA_GENEVE_COLLECT_METADATA])
1417                 metadata = true;
1418
1419         if (data[IFLA_GENEVE_UDP_CSUM] &&
1420             !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1421                 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1422
1423         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1424             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1425                 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1426
1427         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1428             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1429                 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1430
1431         return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1432                                 dst_port, metadata, flags);
1433 }
1434
1435 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1436 {
1437         struct geneve_dev *geneve = netdev_priv(dev);
1438
1439         list_del(&geneve->next);
1440         unregister_netdevice_queue(dev, head);
1441 }
1442
1443 static size_t geneve_get_size(const struct net_device *dev)
1444 {
1445         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1446                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1447                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1448                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1449                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1450                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1451                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1452                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1453                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1454                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1455                 0;
1456 }
1457
1458 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1459 {
1460         struct geneve_dev *geneve = netdev_priv(dev);
1461         __u32 vni;
1462
1463         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1464         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1465                 goto nla_put_failure;
1466
1467         if (geneve->remote.sa.sa_family == AF_INET) {
1468                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1469                                     geneve->remote.sin.sin_addr.s_addr))
1470                         goto nla_put_failure;
1471 #if IS_ENABLED(CONFIG_IPV6)
1472         } else {
1473                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1474                                      &geneve->remote.sin6.sin6_addr))
1475                         goto nla_put_failure;
1476 #endif
1477         }
1478
1479         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1480             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1481             nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1482                 goto nla_put_failure;
1483
1484         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1485                 goto nla_put_failure;
1486
1487         if (geneve->collect_md) {
1488                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1489                         goto nla_put_failure;
1490         }
1491
1492         if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1493                        !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1494             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1495                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1496             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1497                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1498                 goto nla_put_failure;
1499
1500         return 0;
1501
1502 nla_put_failure:
1503         return -EMSGSIZE;
1504 }
1505
1506 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1507         .kind           = "geneve",
1508         .maxtype        = IFLA_GENEVE_MAX,
1509         .policy         = geneve_policy,
1510         .priv_size      = sizeof(struct geneve_dev),
1511         .setup          = geneve_setup,
1512         .validate       = geneve_validate,
1513         .newlink        = geneve_newlink,
1514         .dellink        = geneve_dellink,
1515         .get_size       = geneve_get_size,
1516         .fill_info      = geneve_fill_info,
1517 };
1518
1519 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1520                                         u8 name_assign_type, u16 dst_port)
1521 {
1522         struct nlattr *tb[IFLA_MAX + 1];
1523         struct net_device *dev;
1524         LIST_HEAD(list_kill);
1525         int err;
1526
1527         memset(tb, 0, sizeof(tb));
1528         dev = rtnl_create_link(net, name, name_assign_type,
1529                                &geneve_link_ops, tb);
1530         if (IS_ERR(dev))
1531                 return dev;
1532
1533         err = geneve_configure(net, dev, &geneve_remote_unspec,
1534                                0, 0, 0, 0, htons(dst_port), true,
1535                                GENEVE_F_UDP_ZERO_CSUM6_RX);
1536         if (err) {
1537                 free_netdev(dev);
1538                 return ERR_PTR(err);
1539         }
1540
1541         /* openvswitch users expect packet sizes to be unrestricted,
1542          * so set the largest MTU we can.
1543          */
1544         err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1545         if (err)
1546                 goto err;
1547
1548         err = rtnl_configure_link(dev, NULL);
1549         if (err < 0)
1550                 goto err;
1551
1552         return dev;
1553
1554  err:
1555         geneve_dellink(dev, &list_kill);
1556         unregister_netdevice_many(&list_kill);
1557         return ERR_PTR(err);
1558 }
1559 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1560
1561 static int geneve_netdevice_event(struct notifier_block *unused,
1562                                   unsigned long event, void *ptr)
1563 {
1564         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1565
1566         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1567                 geneve_push_rx_ports(dev);
1568
1569         return NOTIFY_DONE;
1570 }
1571
1572 static struct notifier_block geneve_notifier_block __read_mostly = {
1573         .notifier_call = geneve_netdevice_event,
1574 };
1575
1576 static __net_init int geneve_init_net(struct net *net)
1577 {
1578         struct geneve_net *gn = net_generic(net, geneve_net_id);
1579
1580         INIT_LIST_HEAD(&gn->geneve_list);
1581         INIT_LIST_HEAD(&gn->sock_list);
1582         return 0;
1583 }
1584
1585 static void __net_exit geneve_exit_net(struct net *net)
1586 {
1587         struct geneve_net *gn = net_generic(net, geneve_net_id);
1588         struct geneve_dev *geneve, *next;
1589         struct net_device *dev, *aux;
1590         LIST_HEAD(list);
1591
1592         rtnl_lock();
1593
1594         /* gather any geneve devices that were moved into this ns */
1595         for_each_netdev_safe(net, dev, aux)
1596                 if (dev->rtnl_link_ops == &geneve_link_ops)
1597                         unregister_netdevice_queue(dev, &list);
1598
1599         /* now gather any other geneve devices that were created in this ns */
1600         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1601                 /* If geneve->dev is in the same netns, it was already added
1602                  * to the list by the previous loop.
1603                  */
1604                 if (!net_eq(dev_net(geneve->dev), net))
1605                         unregister_netdevice_queue(geneve->dev, &list);
1606         }
1607
1608         /* unregister the devices gathered above */
1609         unregister_netdevice_many(&list);
1610         rtnl_unlock();
1611 }
1612
1613 static struct pernet_operations geneve_net_ops = {
1614         .init = geneve_init_net,
1615         .exit = geneve_exit_net,
1616         .id   = &geneve_net_id,
1617         .size = sizeof(struct geneve_net),
1618 };
1619
1620 static int __init geneve_init_module(void)
1621 {
1622         int rc;
1623
1624         rc = register_pernet_subsys(&geneve_net_ops);
1625         if (rc)
1626                 goto out1;
1627
1628         rc = register_netdevice_notifier(&geneve_notifier_block);
1629         if (rc)
1630                 goto out2;
1631
1632         rc = rtnl_link_register(&geneve_link_ops);
1633         if (rc)
1634                 goto out3;
1635
1636         return 0;
1637
1638 out3:
1639         unregister_netdevice_notifier(&geneve_notifier_block);
1640 out2:
1641         unregister_pernet_subsys(&geneve_net_ops);
1642 out1:
1643         return rc;
1644 }
1645 late_initcall(geneve_init_module);
1646
1647 static void __exit geneve_cleanup_module(void)
1648 {
1649         rtnl_link_unregister(&geneve_link_ops);
1650         unregister_netdevice_notifier(&geneve_notifier_block);
1651         unregister_pernet_subsys(&geneve_net_ops);
1652 }
1653 module_exit(geneve_cleanup_module);
1654
1655 MODULE_LICENSE("GPL");
1656 MODULE_VERSION(GENEVE_NETDEV_VER);
1657 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1658 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1659 MODULE_ALIAS_RTNL_LINK("geneve");