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         int err = 0;
 213         void *oiph;
 214
 215         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 216                 __be16 flags;
 217
 218                 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
 219                         (gnvh->oam ? TUNNEL_OAM : 0) |
 220                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 221
 222                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 223                                          vni_to_tunnel_id(gnvh->vni),
 224                                          gnvh->opt_len * 4);
 225                 if (!tun_dst)
 226                         goto drop;
 227                 /* Update tunnel dst according to Geneve options. */
 228                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 229                                         gnvh->options, gnvh->opt_len * 4);
 230         } else {
 231                 /* Drop packets w/ critical options,
 232                  * since we don't support any...
 233                  */
 234                 if (gnvh->critical)
 235                         goto drop;
 236         }
 237
 238         skb_reset_mac_header(skb);
 239         skb->protocol = eth_type_trans(skb, geneve->dev);
 240         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 241
 242         if (tun_dst)
 243                 skb_dst_set(skb, &tun_dst->dst);
 244
 245         /* Ignore packet loops (and multicast echo) */
 246         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
 247                 goto drop;
 248
 249         oiph = skb_network_header(skb);
 250         skb_reset_network_header(skb);
 251
 252         if (geneve_get_sk_family(gs) == AF_INET)
 253                 err = IP_ECN_decapsulate(oiph, skb);
 254 #if IS_ENABLED(CONFIG_IPV6)
 255         else
 256                 err = IP6_ECN_decapsulate(oiph, skb);
 257 #endif
 258
 259         if (unlikely(err)) {
 260                 if (log_ecn_error) {
 261                         if (geneve_get_sk_family(gs) == AF_INET)
 262                                 net_info_ratelimited("non-ECT from %pI4 "
 263                                                      "with TOS=%#x\n",
 264                                                      &((struct iphdr *)oiph)->saddr,
 265                                                      ((struct iphdr *)oiph)->tos);
 266 #if IS_ENABLED(CONFIG_IPV6)
 267                         else
 268                                 net_info_ratelimited("non-ECT from %pI6\n",
 269                                                      &((struct ipv6hdr *)oiph)->saddr);
 270 #endif
 271                 }
 272                 if (err > 1) {
 273                         ++geneve->dev->stats.rx_frame_errors;
 274                         ++geneve->dev->stats.rx_errors;
 275                         goto drop;
 276                 }
 277         }
 278
 279         stats = this_cpu_ptr(geneve->dev->tstats);
 280         u64_stats_update_begin(&stats->syncp);
 281         stats->rx_packets++;
 282         stats->rx_bytes += skb->len;
 283         u64_stats_update_end(&stats->syncp);
 284
 285         gro_cells_receive(&geneve->gro_cells, skb);
 286         return;
 287 drop:
 288         /* Consume bad packet */
 289         kfree_skb(skb);
 290 }
 291
 292 /* Setup stats when device is created */
 293 static int geneve_init(struct net_device *dev)
 294 {
 295         struct geneve_dev *geneve = netdev_priv(dev);
 296         int err;
 297
 298         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 299         if (!dev->tstats)
 300                 return -ENOMEM;
 301
 302         err = gro_cells_init(&geneve->gro_cells, dev);
 303         if (err) {
 304                 free_percpu(dev->tstats);
 305                 return err;
 306         }
 307
 308         err = dst_cache_init(&geneve->dst_cache, GFP_KERNEL);
 309         if (err) {
 310                 free_percpu(dev->tstats);
 311                 gro_cells_destroy(&geneve->gro_cells);
 312                 return err;
 313         }
 314
 315         return 0;
 316 }
 317
 318 static void geneve_uninit(struct net_device *dev)
 319 {
 320         struct geneve_dev *geneve = netdev_priv(dev);
 321
 322         dst_cache_destroy(&geneve->dst_cache);
 323         gro_cells_destroy(&geneve->gro_cells);
 324         free_percpu(dev->tstats);
 325 }
 326
 327 /* Callback from net/ipv4/udp.c to receive packets */
 328 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 329 {
 330         struct genevehdr *geneveh;
 331         struct geneve_dev *geneve;
 332         struct geneve_sock *gs;
 333         int opts_len;
 334
 335         /* Need Geneve and inner Ethernet header to be present */
 336         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 337                 goto drop;
 338
 339         /* Return packets with reserved bits set */
 340         geneveh = geneve_hdr(skb);
 341         if (unlikely(geneveh->ver != GENEVE_VER))
 342                 goto drop;
 343
 344         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 345                 goto drop;
 346
 347         gs = rcu_dereference_sk_user_data(sk);
 348         if (!gs)
 349                 goto drop;
 350
 351         geneve = geneve_lookup_skb(gs, skb);
 352         if (!geneve)
 353                 goto drop;
 354
 355         opts_len = geneveh->opt_len * 4;
 356         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 357                                  htons(ETH_P_TEB),
 358                                  !net_eq(geneve->net, dev_net(geneve->dev))))
 359                 goto drop;
 360
 361         geneve_rx(geneve, gs, skb);
 362         return 0;
 363
 364 drop:
 365         /* Consume bad packet */
 366         kfree_skb(skb);
 367         return 0;
 368 }
 369
 370 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 371                                          __be16 port, u32 flags)
 372 {
 373         struct socket *sock;
 374         struct udp_port_cfg udp_conf;
 375         int err;
 376
 377         memset(&udp_conf, 0, sizeof(udp_conf));
 378
 379         if (ipv6) {
 380                 udp_conf.family = AF_INET6;
 381                 udp_conf.ipv6_v6only = 1;
 382                 udp_conf.use_udp6_rx_checksums =
 383                     !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX);
 384         } else {
 385                 udp_conf.family = AF_INET;
 386                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 387         }
 388
 389         udp_conf.local_udp_port = port;
 390
 391         /* Open UDP socket */
 392         err = udp_sock_create(net, &udp_conf, &sock);
 393         if (err < 0)
 394                 return ERR_PTR(err);
 395
 396         return sock;
 397 }
 398
 399 static int geneve_hlen(struct genevehdr *gh)
 400 {
 401         return sizeof(*gh) + gh->opt_len * 4;
 402 }
 403
 404 static struct sk_buff **geneve_gro_receive(struct sock *sk,
 405                                            struct sk_buff **head,
 406                                            struct sk_buff *skb)
 407 {
 408         struct sk_buff *p, **pp = NULL;
 409         struct genevehdr *gh, *gh2;
 410         unsigned int hlen, gh_len, off_gnv;
 411         const struct packet_offload *ptype;
 412         __be16 type;
 413         int flush = 1;
 414
 415         off_gnv = skb_gro_offset(skb);
 416         hlen = off_gnv + sizeof(*gh);
 417         gh = skb_gro_header_fast(skb, off_gnv);
 418         if (skb_gro_header_hard(skb, hlen)) {
 419                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 420                 if (unlikely(!gh))
 421                         goto out;
 422         }
 423
 424         if (gh->ver != GENEVE_VER || gh->oam)
 425                 goto out;
 426         gh_len = geneve_hlen(gh);
 427
 428         hlen = off_gnv + gh_len;
 429         if (skb_gro_header_hard(skb, hlen)) {
 430                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 431                 if (unlikely(!gh))
 432                         goto out;
 433         }
 434
 435         for (p = *head; p; p = p->next) {
 436                 if (!NAPI_GRO_CB(p)->same_flow)
 437                         continue;
 438
 439                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 440                 if (gh->opt_len != gh2->opt_len ||
 441                     memcmp(gh, gh2, gh_len)) {
 442                         NAPI_GRO_CB(p)->same_flow = 0;
 443                         continue;
 444                 }
 445         }
 446
 447         type = gh->proto_type;
 448
 449         rcu_read_lock();
 450         ptype = gro_find_receive_by_type(type);
 451         if (!ptype)
 452                 goto out_unlock;
 453
 454         skb_gro_pull(skb, gh_len);
 455         skb_gro_postpull_rcsum(skb, gh, gh_len);
 456         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 457         flush = 0;
 458
 459 out_unlock:
 460         rcu_read_unlock();
 461 out:
 462         NAPI_GRO_CB(skb)->flush |= flush;
 463
 464         return pp;
 465 }
 466
 467 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 468                                int nhoff)
 469 {
 470         struct genevehdr *gh;
 471         struct packet_offload *ptype;
 472         __be16 type;
 473         int gh_len;
 474         int err = -ENOSYS;
 475
 476         gh = (struct genevehdr *)(skb->data + nhoff);
 477         gh_len = geneve_hlen(gh);
 478         type = gh->proto_type;
 479
 480         rcu_read_lock();
 481         ptype = gro_find_complete_by_type(type);
 482         if (ptype)
 483                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 484
 485         rcu_read_unlock();
 486
 487         skb_set_inner_mac_header(skb, nhoff + gh_len);
 488
 489         return err;
 490 }
 491
 492 /* Create new listen socket if needed */
 493 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 494                                                 bool ipv6, u32 flags)
 495 {
 496         struct geneve_net *gn = net_generic(net, geneve_net_id);
 497         struct geneve_sock *gs;
 498         struct socket *sock;
 499         struct udp_tunnel_sock_cfg tunnel_cfg;
 500         int h;
 501
 502         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 503         if (!gs)
 504                 return ERR_PTR(-ENOMEM);
 505
 506         sock = geneve_create_sock(net, ipv6, port, flags);
 507         if (IS_ERR(sock)) {
 508                 kfree(gs);
 509                 return ERR_CAST(sock);
 510         }
 511
 512         gs->sock = sock;
 513         gs->refcnt = 1;
 514         for (h = 0; h < VNI_HASH_SIZE; ++h)
 515                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 516
 517         /* Initialize the geneve udp offloads structure */
 518         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 519
 520         /* Mark socket as an encapsulation socket */
 521         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 522         tunnel_cfg.sk_user_data = gs;
 523         tunnel_cfg.encap_type = 1;
 524         tunnel_cfg.gro_receive = geneve_gro_receive;
 525         tunnel_cfg.gro_complete = geneve_gro_complete;
 526         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 527         tunnel_cfg.encap_destroy = NULL;
 528         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 529         list_add(&gs->list, &gn->sock_list);
 530         return gs;
 531 }
 532
 533 static void __geneve_sock_release(struct geneve_sock *gs)
 534 {
 535         if (!gs || --gs->refcnt)
 536                 return;
 537
 538         list_del(&gs->list);
 539         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 540         udp_tunnel_sock_release(gs->sock);
 541         kfree_rcu(gs, rcu);
 542 }
 543
 544 static void geneve_sock_release(struct geneve_dev *geneve)
 545 {
 546         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 547 #if IS_ENABLED(CONFIG_IPV6)
 548         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 549
 550         rcu_assign_pointer(geneve->sock6, NULL);
 551 #endif
 552
 553         rcu_assign_pointer(geneve->sock4, NULL);
 554         synchronize_net();
 555
 556         __geneve_sock_release(gs4);
 557 #if IS_ENABLED(CONFIG_IPV6)
 558         __geneve_sock_release(gs6);
 559 #endif
 560 }
 561
 562 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 563                                             sa_family_t family,
 564                                             __be16 dst_port)
 565 {
 566         struct geneve_sock *gs;
 567
 568         list_for_each_entry(gs, &gn->sock_list, list) {
 569                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 570                     geneve_get_sk_family(gs) == family) {
 571                         return gs;
 572                 }
 573         }
 574         return NULL;
 575 }
 576
 577 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 578 {
 579         struct net *net = geneve->net;
 580         struct geneve_net *gn = net_generic(net, geneve_net_id);
 581         struct geneve_sock *gs;
 582         __u32 hash;
 583
 584         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
 585         if (gs) {
 586                 gs->refcnt++;
 587                 goto out;
 588         }
 589
 590         gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
 591         if (IS_ERR(gs))
 592                 return PTR_ERR(gs);
 593
 594 out:
 595         gs->collect_md = geneve->collect_md;
 596         gs->flags = geneve->flags;
 597 #if IS_ENABLED(CONFIG_IPV6)
 598         if (ipv6)
 599                 rcu_assign_pointer(geneve->sock6, gs);
 600         else
 601 #endif
 602                 rcu_assign_pointer(geneve->sock4, gs);
 603
 604         hash = geneve_net_vni_hash(geneve->vni);
 605         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 606         return 0;
 607 }
 608
 609 static int geneve_open(struct net_device *dev)
 610 {
 611         struct geneve_dev *geneve = netdev_priv(dev);
 612         bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
 613         bool metadata = geneve->collect_md;
 614         int ret = 0;
 615
 616 #if IS_ENABLED(CONFIG_IPV6)
 617         if (ipv6 || metadata)
 618                 ret = geneve_sock_add(geneve, true);
 619 #endif
 620         if (!ret && (!ipv6 || metadata))
 621                 ret = geneve_sock_add(geneve, false);
 622         if (ret < 0)
 623                 geneve_sock_release(geneve);
 624
 625         return ret;
 626 }
 627
 628 static int geneve_stop(struct net_device *dev)
 629 {
 630         struct geneve_dev *geneve = netdev_priv(dev);
 631
 632         if (!hlist_unhashed(&geneve->hlist))
 633                 hlist_del_rcu(&geneve->hlist);
 634         geneve_sock_release(geneve);
 635         return 0;
 636 }
 637
 638 static void geneve_build_header(struct genevehdr *geneveh,
 639                                 __be16 tun_flags, u8 vni[3],
 640                                 u8 options_len, u8 *options)
 641 {
 642         geneveh->ver = GENEVE_VER;
 643         geneveh->opt_len = options_len / 4;
 644         geneveh->oam = !!(tun_flags & TUNNEL_OAM);
 645         geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 646         geneveh->rsvd1 = 0;
 647         memcpy(geneveh->vni, vni, 3);
 648         geneveh->proto_type = htons(ETH_P_TEB);
 649         geneveh->rsvd2 = 0;
 650
 651         memcpy(geneveh->options, options, options_len);
 652 }
 653
 654 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 655                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 656                             u32 flags, bool xnet)
 657 {
 658         struct genevehdr *gnvh;
 659         int min_headroom;
 660         int err;
 661         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
 662
 663         skb_scrub_packet(skb, xnet);
 664
 665         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 666                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 667         err = skb_cow_head(skb, min_headroom);
 668         if (unlikely(err))
 669                 goto free_rt;
 670
 671         err = udp_tunnel_handle_offloads(skb, udp_sum);
 672         if (err)
 673                 goto free_rt;
 674
 675         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 676         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 677
 678         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 679         return 0;
 680
 681 free_rt:
 682         ip_rt_put(rt);
 683         return err;
 684 }
 685
 686 #if IS_ENABLED(CONFIG_IPV6)
 687 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 688                              __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 689                              u32 flags, bool xnet)
 690 {
 691         struct genevehdr *gnvh;
 692         int min_headroom;
 693         int err;
 694         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
 695
 696         skb_scrub_packet(skb, xnet);
 697
 698         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
 699                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
 700         err = skb_cow_head(skb, min_headroom);
 701         if (unlikely(err))
 702                 goto free_dst;
 703
 704         err = udp_tunnel_handle_offloads(skb, udp_sum);
 705         if (err)
 706                 goto free_dst;
 707
 708         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 709         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 710
 711         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 712         return 0;
 713
 714 free_dst:
 715         dst_release(dst);
 716         return err;
 717 }
 718 #endif
 719
 720 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 721                                        struct net_device *dev,
 722                                        struct flowi4 *fl4,
 723                                        struct ip_tunnel_info *info)
 724 {
 725         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 726         struct geneve_dev *geneve = netdev_priv(dev);
 727         struct dst_cache *dst_cache;
 728         struct rtable *rt = NULL;
 729         __u8 tos;
 730
 731         if (!rcu_dereference(geneve->sock4))
 732                 return ERR_PTR(-EIO);
 733
 734         memset(fl4, 0, sizeof(*fl4));
 735         fl4->flowi4_mark = skb->mark;
 736         fl4->flowi4_proto = IPPROTO_UDP;
 737
 738         if (info) {
 739                 fl4->daddr = info->key.u.ipv4.dst;
 740                 fl4->saddr = info->key.u.ipv4.src;
 741                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 742                 dst_cache = &info->dst_cache;
 743         } else {
 744                 tos = geneve->tos;
 745                 if (tos == 1) {
 746                         const struct iphdr *iip = ip_hdr(skb);
 747
 748                         tos = ip_tunnel_get_dsfield(iip, skb);
 749                         use_cache = false;
 750                 }
 751
 752                 fl4->flowi4_tos = RT_TOS(tos);
 753                 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
 754                 dst_cache = &geneve->dst_cache;
 755         }
 756
 757         if (use_cache) {
 758                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 759                 if (rt)
 760                         return rt;
 761         }
 762
 763         rt = ip_route_output_key(geneve->net, fl4);
 764         if (IS_ERR(rt)) {
 765                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 766                 return ERR_PTR(-ENETUNREACH);
 767         }
 768         if (rt->dst.dev == dev) { /* is this necessary? */
 769                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 770                 ip_rt_put(rt);
 771                 return ERR_PTR(-ELOOP);
 772         }
 773         if (use_cache)
 774                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 775         return rt;
 776 }
 777
 778 #if IS_ENABLED(CONFIG_IPV6)
 779 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 780                                            struct net_device *dev,
 781                                            struct flowi6 *fl6,
 782                                            struct ip_tunnel_info *info)
 783 {
 784         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 785         struct geneve_dev *geneve = netdev_priv(dev);
 786         struct dst_entry *dst = NULL;
 787         struct dst_cache *dst_cache;
 788         struct geneve_sock *gs6;
 789         __u8 prio;
 790
 791         gs6 = rcu_dereference(geneve->sock6);
 792         if (!gs6)
 793                 return ERR_PTR(-EIO);
 794
 795         memset(fl6, 0, sizeof(*fl6));
 796         fl6->flowi6_mark = skb->mark;
 797         fl6->flowi6_proto = IPPROTO_UDP;
 798
 799         if (info) {
 800                 fl6->daddr = info->key.u.ipv6.dst;
 801                 fl6->saddr = info->key.u.ipv6.src;
 802                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(info->key.tos),
 803                                                    info->key.label);
 804                 dst_cache = &info->dst_cache;
 805         } else {
 806                 prio = geneve->tos;
 807                 if (prio == 1) {
 808                         const struct iphdr *iip = ip_hdr(skb);
 809
 810                         prio = ip_tunnel_get_dsfield(iip, skb);
 811                         use_cache = false;
 812                 }
 813
 814                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 815                                                    geneve->label);
 816                 fl6->daddr = geneve->remote.sin6.sin6_addr;
 817                 dst_cache = &geneve->dst_cache;
 818         }
 819
 820         if (use_cache) {
 821                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 822                 if (dst)
 823                         return dst;
 824         }
 825
 826         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 827                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 828                 return ERR_PTR(-ENETUNREACH);
 829         }
 830         if (dst->dev == dev) { /* is this necessary? */
 831                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 832                 dst_release(dst);
 833                 return ERR_PTR(-ELOOP);
 834         }
 835
 836         if (use_cache)
 837                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 838         return dst;
 839 }
 840 #endif
 841
 842 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 843 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 844 {
 845 #ifdef __BIG_ENDIAN
 846         vni[0] = (__force __u8)(tun_id >> 16);
 847         vni[1] = (__force __u8)(tun_id >> 8);
 848         vni[2] = (__force __u8)tun_id;
 849 #else
 850         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 851         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 852         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 853 #endif
 854 }
 855
 856 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 857                                    struct ip_tunnel_info *info)
 858 {
 859         struct geneve_dev *geneve = netdev_priv(dev);
 860         struct geneve_sock *gs4;
 861         struct rtable *rt = NULL;
 862         const struct iphdr *iip; /* interior IP header */
 863         int err = -EINVAL;
 864         struct flowi4 fl4;
 865         __u8 tos, ttl;
 866         __be16 sport;
 867         __be16 df;
 868         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 869         u32 flags = geneve->flags;
 870
 871         gs4 = rcu_dereference(geneve->sock4);
 872         if (!gs4)
 873                 goto tx_error;
 874
 875         if (geneve->collect_md) {
 876                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 877                         netdev_dbg(dev, "no tunnel metadata\n");
 878                         goto tx_error;
 879                 }
 880                 if (info && ip_tunnel_info_af(info) != AF_INET)
 881                         goto tx_error;
 882         }
 883
 884         rt = geneve_get_v4_rt(skb, dev, &fl4, info);
 885         if (IS_ERR(rt)) {
 886                 err = PTR_ERR(rt);
 887                 goto tx_error;
 888         }
 889
 890         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 891         skb_reset_mac_header(skb);
 892
 893         iip = ip_hdr(skb);
 894
 895         if (info) {
 896                 const struct ip_tunnel_key *key = &info->key;
 897                 u8 *opts = NULL;
 898                 u8 vni[3];
 899
 900                 tunnel_id_to_vni(key->tun_id, vni);
 901                 if (info->options_len)
 902                         opts = ip_tunnel_info_opts(info);
 903
 904                 if (key->tun_flags & TUNNEL_CSUM)
 905                         flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
 906                 else
 907                         flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
 908
 909                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 910                                        info->options_len, opts, flags, xnet);
 911                 if (unlikely(err))
 912                         goto tx_error;
 913
 914                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
 915                 ttl = key->ttl;
 916                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 917         } else {
 918                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 919                                        0, NULL, flags, xnet);
 920                 if (unlikely(err))
 921                         goto tx_error;
 922
 923                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 924                 ttl = geneve->ttl;
 925                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 926                         ttl = 1;
 927                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 928                 df = 0;
 929         }
 930         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 931                             tos, ttl, df, sport, geneve->dst_port,
 932                             !net_eq(geneve->net, dev_net(geneve->dev)),
 933                             !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
 934
 935         return NETDEV_TX_OK;
 936
 937 tx_error:
 938         dev_kfree_skb(skb);
 939
 940         if (err == -ELOOP)
 941                 dev->stats.collisions++;
 942         else if (err == -ENETUNREACH)
 943                 dev->stats.tx_carrier_errors++;
 944
 945         dev->stats.tx_errors++;
 946         return NETDEV_TX_OK;
 947 }
 948
 949 #if IS_ENABLED(CONFIG_IPV6)
 950 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 951                                     struct ip_tunnel_info *info)
 952 {
 953         struct geneve_dev *geneve = netdev_priv(dev);
 954         struct dst_entry *dst = NULL;
 955         const struct iphdr *iip; /* interior IP header */
 956         struct geneve_sock *gs6;
 957         int err = -EINVAL;
 958         struct flowi6 fl6;
 959         __u8 prio, ttl;
 960         __be16 sport;
 961         __be32 label;
 962         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 963         u32 flags = geneve->flags;
 964
 965         gs6 = rcu_dereference(geneve->sock6);
 966         if (!gs6)
 967                 goto tx_error;
 968
 969         if (geneve->collect_md) {
 970                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 971                         netdev_dbg(dev, "no tunnel metadata\n");
 972                         goto tx_error;
 973                 }
 974         }
 975
 976         dst = geneve_get_v6_dst(skb, dev, &fl6, info);
 977         if (IS_ERR(dst)) {
 978                 err = PTR_ERR(dst);
 979                 goto tx_error;
 980         }
 981
 982         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 983         skb_reset_mac_header(skb);
 984
 985         iip = ip_hdr(skb);
 986
 987         if (info) {
 988                 const struct ip_tunnel_key *key = &info->key;
 989                 u8 *opts = NULL;
 990                 u8 vni[3];
 991
 992                 tunnel_id_to_vni(key->tun_id, vni);
 993                 if (info->options_len)
 994                         opts = ip_tunnel_info_opts(info);
 995
 996                 if (key->tun_flags & TUNNEL_CSUM)
 997                         flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
 998                 else
 999                         flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1000
1001                 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
1002                                         info->options_len, opts,
1003                                         flags, xnet);
1004                 if (unlikely(err))
1005                         goto tx_error;
1006
1007                 prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
1008                 ttl = key->ttl;
1009                 label = info->key.label;
1010         } else {
1011                 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
1012                                         0, NULL, flags, xnet);
1013                 if (unlikely(err))
1014                         goto tx_error;
1015
1016                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1017                                            iip, skb);
1018                 ttl = geneve->ttl;
1019                 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1020                         ttl = 1;
1021                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1022                 label = geneve->label;
1023         }
1024
1025         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1026                              &fl6.saddr, &fl6.daddr, prio, ttl, label,
1027                              sport, geneve->dst_port,
1028                              !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1029         return NETDEV_TX_OK;
1030
1031 tx_error:
1032         dev_kfree_skb(skb);
1033
1034         if (err == -ELOOP)
1035                 dev->stats.collisions++;
1036         else if (err == -ENETUNREACH)
1037                 dev->stats.tx_carrier_errors++;
1038
1039         dev->stats.tx_errors++;
1040         return NETDEV_TX_OK;
1041 }
1042 #endif
1043
1044 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1045 {
1046         struct geneve_dev *geneve = netdev_priv(dev);
1047         struct ip_tunnel_info *info = NULL;
1048
1049         if (geneve->collect_md)
1050                 info = skb_tunnel_info(skb);
1051
1052 #if IS_ENABLED(CONFIG_IPV6)
1053         if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1054             (!info && geneve->remote.sa.sa_family == AF_INET6))
1055                 return geneve6_xmit_skb(skb, dev, info);
1056 #endif
1057         return geneve_xmit_skb(skb, dev, info);
1058 }
1059
1060 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1061 {
1062         struct geneve_dev *geneve = netdev_priv(dev);
1063         /* The max_mtu calculation does not take account of GENEVE
1064          * options, to avoid excluding potentially valid
1065          * configurations.
1066          */
1067         int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1068
1069         if (geneve->remote.sa.sa_family == AF_INET6)
1070                 max_mtu -= sizeof(struct ipv6hdr);
1071         else
1072                 max_mtu -= sizeof(struct iphdr);
1073
1074         if (new_mtu < 68)
1075                 return -EINVAL;
1076
1077         if (new_mtu > max_mtu) {
1078                 if (strict)
1079                         return -EINVAL;
1080
1081                 new_mtu = max_mtu;
1082         }
1083
1084         dev->mtu = new_mtu;
1085         return 0;
1086 }
1087
1088 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1089 {
1090         return __geneve_change_mtu(dev, new_mtu, true);
1091 }
1092
1093 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1094 {
1095         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1096         struct geneve_dev *geneve = netdev_priv(dev);
1097         struct rtable *rt;
1098         struct flowi4 fl4;
1099 #if IS_ENABLED(CONFIG_IPV6)
1100         struct dst_entry *dst;
1101         struct flowi6 fl6;
1102 #endif
1103
1104         if (ip_tunnel_info_af(info) == AF_INET) {
1105                 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1106                 if (IS_ERR(rt))
1107                         return PTR_ERR(rt);
1108
1109                 ip_rt_put(rt);
1110                 info->key.u.ipv4.src = fl4.saddr;
1111 #if IS_ENABLED(CONFIG_IPV6)
1112         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1113                 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1114                 if (IS_ERR(dst))
1115                         return PTR_ERR(dst);
1116
1117                 dst_release(dst);
1118                 info->key.u.ipv6.src = fl6.saddr;
1119 #endif
1120         } else {
1121                 return -EINVAL;
1122         }
1123
1124         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1125                                              1, USHRT_MAX, true);
1126         info->key.tp_dst = geneve->dst_port;
1127         return 0;
1128 }
1129
1130 static const struct net_device_ops geneve_netdev_ops = {
1131         .ndo_init               = geneve_init,
1132         .ndo_uninit             = geneve_uninit,
1133         .ndo_open               = geneve_open,
1134         .ndo_stop               = geneve_stop,
1135         .ndo_start_xmit         = geneve_xmit,
1136         .ndo_get_stats64        = ip_tunnel_get_stats64,
1137         .ndo_change_mtu         = geneve_change_mtu,
1138         .ndo_validate_addr      = eth_validate_addr,
1139         .ndo_set_mac_address    = eth_mac_addr,
1140         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1141 };
1142
1143 static void geneve_get_drvinfo(struct net_device *dev,
1144                                struct ethtool_drvinfo *drvinfo)
1145 {
1146         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1147         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1148 }
1149
1150 static const struct ethtool_ops geneve_ethtool_ops = {
1151         .get_drvinfo    = geneve_get_drvinfo,
1152         .get_link       = ethtool_op_get_link,
1153 };
1154
1155 /* Info for udev, that this is a virtual tunnel endpoint */
1156 static struct device_type geneve_type = {
1157         .name = "geneve",
1158 };
1159
1160 /* Calls the ndo_udp_tunnel_add of the caller in order to
1161  * supply the listening GENEVE udp ports. Callers are expected
1162  * to implement the ndo_udp_tunnel_add.
1163  */
1164 static void geneve_push_rx_ports(struct net_device *dev)
1165 {
1166         struct net *net = dev_net(dev);
1167         struct geneve_net *gn = net_generic(net, geneve_net_id);
1168         struct geneve_sock *gs;
1169
1170         rcu_read_lock();
1171         list_for_each_entry_rcu(gs, &gn->sock_list, list)
1172                 udp_tunnel_push_rx_port(dev, gs->sock,
1173                                         UDP_TUNNEL_TYPE_GENEVE);
1174         rcu_read_unlock();
1175 }
1176
1177 /* Initialize the device structure. */
1178 static void geneve_setup(struct net_device *dev)
1179 {
1180         ether_setup(dev);
1181
1182         dev->netdev_ops = &geneve_netdev_ops;
1183         dev->ethtool_ops = &geneve_ethtool_ops;
1184         dev->destructor = free_netdev;
1185
1186         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1187
1188         dev->features    |= NETIF_F_LLTX;
1189         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1190         dev->features    |= NETIF_F_RXCSUM;
1191         dev->features    |= NETIF_F_GSO_SOFTWARE;
1192
1193         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1194         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1195
1196         netif_keep_dst(dev);
1197         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1198         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1199         eth_hw_addr_random(dev);
1200 }
1201
1202 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1203         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1204         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1205         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1206         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1207         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1208         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1209         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1210         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1211         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1212         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1213         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1214 };
1215
1216 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1217 {
1218         if (tb[IFLA_ADDRESS]) {
1219                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1220                         return -EINVAL;
1221
1222                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1223                         return -EADDRNOTAVAIL;
1224         }
1225
1226         if (!data)
1227                 return -EINVAL;
1228
1229         if (data[IFLA_GENEVE_ID]) {
1230                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1231
1232                 if (vni >= GENEVE_VID_MASK)
1233                         return -ERANGE;
1234         }
1235
1236         return 0;
1237 }
1238
1239 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1240                                           __be16 dst_port,
1241                                           union geneve_addr *remote,
1242                                           u8 vni[],
1243                                           bool *tun_on_same_port,
1244                                           bool *tun_collect_md)
1245 {
1246         struct geneve_dev *geneve, *t;
1247
1248         *tun_on_same_port = false;
1249         *tun_collect_md = false;
1250         t = NULL;
1251         list_for_each_entry(geneve, &gn->geneve_list, next) {
1252                 if (geneve->dst_port == dst_port) {
1253                         *tun_collect_md = geneve->collect_md;
1254                         *tun_on_same_port = true;
1255                 }
1256                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1257                     !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1258                     dst_port == geneve->dst_port)
1259                         t = geneve;
1260         }
1261         return t;
1262 }
1263
1264 static int geneve_configure(struct net *net, struct net_device *dev,
1265                             union geneve_addr *remote,
1266                             __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1267                             __be16 dst_port, bool metadata, u32 flags)
1268 {
1269         struct geneve_net *gn = net_generic(net, geneve_net_id);
1270         struct geneve_dev *t, *geneve = netdev_priv(dev);
1271         bool tun_collect_md, tun_on_same_port;
1272         int err, encap_len;
1273
1274         if (!remote)
1275                 return -EINVAL;
1276         if (metadata &&
1277             (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1278                 return -EINVAL;
1279
1280         geneve->net = net;
1281         geneve->dev = dev;
1282
1283         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1284         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1285         geneve->vni[2] =  vni & 0x000000ff;
1286
1287         if ((remote->sa.sa_family == AF_INET &&
1288              IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1289             (remote->sa.sa_family == AF_INET6 &&
1290              ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1291                 return -EINVAL;
1292         if (label && remote->sa.sa_family != AF_INET6)
1293                 return -EINVAL;
1294
1295         geneve->remote = *remote;
1296
1297         geneve->ttl = ttl;
1298         geneve->tos = tos;
1299         geneve->label = label;
1300         geneve->dst_port = dst_port;
1301         geneve->collect_md = metadata;
1302         geneve->flags = flags;
1303
1304         t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1305                             &tun_on_same_port, &tun_collect_md);
1306         if (t)
1307                 return -EBUSY;
1308
1309         /* make enough headroom for basic scenario */
1310         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1311         if (remote->sa.sa_family == AF_INET)
1312                 encap_len += sizeof(struct iphdr);
1313         else
1314                 encap_len += sizeof(struct ipv6hdr);
1315         dev->needed_headroom = encap_len + ETH_HLEN;
1316
1317         if (metadata) {
1318                 if (tun_on_same_port)
1319                         return -EPERM;
1320         } else {
1321                 if (tun_collect_md)
1322                         return -EPERM;
1323         }
1324
1325         dst_cache_reset(&geneve->dst_cache);
1326
1327         err = register_netdevice(dev);
1328         if (err)
1329                 return err;
1330
1331         list_add(&geneve->next, &gn->geneve_list);
1332         return 0;
1333 }
1334
1335 static int geneve_newlink(struct net *net, struct net_device *dev,
1336                           struct nlattr *tb[], struct nlattr *data[])
1337 {
1338         __be16 dst_port = htons(GENEVE_UDP_PORT);
1339         __u8 ttl = 0, tos = 0;
1340         bool metadata = false;
1341         union geneve_addr remote = geneve_remote_unspec;
1342         __be32 label = 0;
1343         __u32 vni = 0;
1344         u32 flags = 0;
1345
1346         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1347                 return -EINVAL;
1348
1349         if (data[IFLA_GENEVE_REMOTE]) {
1350                 remote.sa.sa_family = AF_INET;
1351                 remote.sin.sin_addr.s_addr =
1352                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1353         }
1354
1355         if (data[IFLA_GENEVE_REMOTE6]) {
1356                 if (!IS_ENABLED(CONFIG_IPV6))
1357                         return -EPFNOSUPPORT;
1358
1359                 remote.sa.sa_family = AF_INET6;
1360                 remote.sin6.sin6_addr =
1361                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1362
1363                 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1364                     IPV6_ADDR_LINKLOCAL) {
1365                         netdev_dbg(dev, "link-local remote is unsupported\n");
1366                         return -EINVAL;
1367                 }
1368         }
1369
1370         if (data[IFLA_GENEVE_ID])
1371                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1372
1373         if (data[IFLA_GENEVE_TTL])
1374                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1375
1376         if (data[IFLA_GENEVE_TOS])
1377                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1378
1379         if (data[IFLA_GENEVE_LABEL])
1380                 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1381                         IPV6_FLOWLABEL_MASK;
1382
1383         if (data[IFLA_GENEVE_PORT])
1384                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1385
1386         if (data[IFLA_GENEVE_COLLECT_METADATA])
1387                 metadata = true;
1388
1389         if (data[IFLA_GENEVE_UDP_CSUM] &&
1390             !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1391                 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1392
1393         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1394             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1395                 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1396
1397         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1398             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1399                 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1400
1401         return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1402                                 dst_port, metadata, flags);
1403 }
1404
1405 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1406 {
1407         struct geneve_dev *geneve = netdev_priv(dev);
1408
1409         list_del(&geneve->next);
1410         unregister_netdevice_queue(dev, head);
1411 }
1412
1413 static size_t geneve_get_size(const struct net_device *dev)
1414 {
1415         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1416                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1417                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1418                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1419                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1420                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1421                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1422                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1423                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1424                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1425                 0;
1426 }
1427
1428 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1429 {
1430         struct geneve_dev *geneve = netdev_priv(dev);
1431         __u32 vni;
1432
1433         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1434         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1435                 goto nla_put_failure;
1436
1437         if (geneve->remote.sa.sa_family == AF_INET) {
1438                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1439                                     geneve->remote.sin.sin_addr.s_addr))
1440                         goto nla_put_failure;
1441 #if IS_ENABLED(CONFIG_IPV6)
1442         } else {
1443                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1444                                      &geneve->remote.sin6.sin6_addr))
1445                         goto nla_put_failure;
1446 #endif
1447         }
1448
1449         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1450             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1451             nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1452                 goto nla_put_failure;
1453
1454         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1455                 goto nla_put_failure;
1456
1457         if (geneve->collect_md) {
1458                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1459                         goto nla_put_failure;
1460         }
1461
1462         if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1463                        !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1464             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1465                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1466             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1467                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1468                 goto nla_put_failure;
1469
1470         return 0;
1471
1472 nla_put_failure:
1473         return -EMSGSIZE;
1474 }
1475
1476 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1477         .kind           = "geneve",
1478         .maxtype        = IFLA_GENEVE_MAX,
1479         .policy         = geneve_policy,
1480         .priv_size      = sizeof(struct geneve_dev),
1481         .setup          = geneve_setup,
1482         .validate       = geneve_validate,
1483         .newlink        = geneve_newlink,
1484         .dellink        = geneve_dellink,
1485         .get_size       = geneve_get_size,
1486         .fill_info      = geneve_fill_info,
1487 };
1488
1489 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1490                                         u8 name_assign_type, u16 dst_port)
1491 {
1492         struct nlattr *tb[IFLA_MAX + 1];
1493         struct net_device *dev;
1494         LIST_HEAD(list_kill);
1495         int err;
1496
1497         memset(tb, 0, sizeof(tb));
1498         dev = rtnl_create_link(net, name, name_assign_type,
1499                                &geneve_link_ops, tb);
1500         if (IS_ERR(dev))
1501                 return dev;
1502
1503         err = geneve_configure(net, dev, &geneve_remote_unspec,
1504                                0, 0, 0, 0, htons(dst_port), true,
1505                                GENEVE_F_UDP_ZERO_CSUM6_RX);
1506         if (err) {
1507                 free_netdev(dev);
1508                 return ERR_PTR(err);
1509         }
1510
1511         /* openvswitch users expect packet sizes to be unrestricted,
1512          * so set the largest MTU we can.
1513          */
1514         err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1515         if (err)
1516                 goto err;
1517
1518         err = rtnl_configure_link(dev, NULL);
1519         if (err < 0)
1520                 goto err;
1521
1522         return dev;
1523
1524  err:
1525         geneve_dellink(dev, &list_kill);
1526         unregister_netdevice_many(&list_kill);
1527         return ERR_PTR(err);
1528 }
1529 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1530
1531 static int geneve_netdevice_event(struct notifier_block *unused,
1532                                   unsigned long event, void *ptr)
1533 {
1534         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1535
1536         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1537                 geneve_push_rx_ports(dev);
1538
1539         return NOTIFY_DONE;
1540 }
1541
1542 static struct notifier_block geneve_notifier_block __read_mostly = {
1543         .notifier_call = geneve_netdevice_event,
1544 };
1545
1546 static __net_init int geneve_init_net(struct net *net)
1547 {
1548         struct geneve_net *gn = net_generic(net, geneve_net_id);
1549
1550         INIT_LIST_HEAD(&gn->geneve_list);
1551         INIT_LIST_HEAD(&gn->sock_list);
1552         return 0;
1553 }
1554
1555 static void __net_exit geneve_exit_net(struct net *net)
1556 {
1557         struct geneve_net *gn = net_generic(net, geneve_net_id);
1558         struct geneve_dev *geneve, *next;
1559         struct net_device *dev, *aux;
1560         LIST_HEAD(list);
1561
1562         rtnl_lock();
1563
1564         /* gather any geneve devices that were moved into this ns */
1565         for_each_netdev_safe(net, dev, aux)
1566                 if (dev->rtnl_link_ops == &geneve_link_ops)
1567                         unregister_netdevice_queue(dev, &list);
1568
1569         /* now gather any other geneve devices that were created in this ns */
1570         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1571                 /* If geneve->dev is in the same netns, it was already added
1572                  * to the list by the previous loop.
1573                  */
1574                 if (!net_eq(dev_net(geneve->dev), net))
1575                         unregister_netdevice_queue(geneve->dev, &list);
1576         }
1577
1578         /* unregister the devices gathered above */
1579         unregister_netdevice_many(&list);
1580         rtnl_unlock();
1581 }
1582
1583 static struct pernet_operations geneve_net_ops = {
1584         .init = geneve_init_net,
1585         .exit = geneve_exit_net,
1586         .id   = &geneve_net_id,
1587         .size = sizeof(struct geneve_net),
1588 };
1589
1590 static int __init geneve_init_module(void)
1591 {
1592         int rc;
1593
1594         rc = register_pernet_subsys(&geneve_net_ops);
1595         if (rc)
1596                 goto out1;
1597
1598         rc = register_netdevice_notifier(&geneve_notifier_block);
1599         if (rc)
1600                 goto out2;
1601
1602         rc = rtnl_link_register(&geneve_link_ops);
1603         if (rc)
1604                 goto out3;
1605
1606         return 0;
1607
1608 out3:
1609         unregister_netdevice_notifier(&geneve_notifier_block);
1610 out2:
1611         unregister_pernet_subsys(&geneve_net_ops);
1612 out1:
1613         return rc;
1614 }
1615 late_initcall(geneve_init_module);
1616
1617 static void __exit geneve_cleanup_module(void)
1618 {
1619         rtnl_link_unregister(&geneve_link_ops);
1620         unregister_netdevice_notifier(&geneve_notifier_block);
1621         unregister_pernet_subsys(&geneve_net_ops);
1622 }
1623 module_exit(geneve_cleanup_module);
1624
1625 MODULE_LICENSE("GPL");
1626 MODULE_VERSION(GENEVE_NETDEV_VER);
1627 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1628 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1629 MODULE_ALIAS_RTNL_LINK("geneve");