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 *sock4;      /* IPv4 socket used for geneve tunnel */
  62 #if IS_ENABLED(CONFIG_IPV6)
  63         struct geneve_sock *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 = 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         __geneve_sock_release(geneve->sock4);
 547 #if IS_ENABLED(CONFIG_IPV6)
 548         __geneve_sock_release(geneve->sock6);
 549 #endif
 550 }
 551
 552 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 553                                             sa_family_t family,
 554                                             __be16 dst_port)
 555 {
 556         struct geneve_sock *gs;
 557
 558         list_for_each_entry(gs, &gn->sock_list, list) {
 559                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 560                     geneve_get_sk_family(gs) == family) {
 561                         return gs;
 562                 }
 563         }
 564         return NULL;
 565 }
 566
 567 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 568 {
 569         struct net *net = geneve->net;
 570         struct geneve_net *gn = net_generic(net, geneve_net_id);
 571         struct geneve_sock *gs;
 572         __u32 hash;
 573
 574         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
 575         if (gs) {
 576                 gs->refcnt++;
 577                 goto out;
 578         }
 579
 580         gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
 581         if (IS_ERR(gs))
 582                 return PTR_ERR(gs);
 583
 584 out:
 585         gs->collect_md = geneve->collect_md;
 586         gs->flags = geneve->flags;
 587 #if IS_ENABLED(CONFIG_IPV6)
 588         if (ipv6)
 589                 geneve->sock6 = gs;
 590         else
 591 #endif
 592                 geneve->sock4 = gs;
 593
 594         hash = geneve_net_vni_hash(geneve->vni);
 595         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 596         return 0;
 597 }
 598
 599 static int geneve_open(struct net_device *dev)
 600 {
 601         struct geneve_dev *geneve = netdev_priv(dev);
 602         bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
 603         bool metadata = geneve->collect_md;
 604         int ret = 0;
 605
 606         geneve->sock4 = NULL;
 607 #if IS_ENABLED(CONFIG_IPV6)
 608         geneve->sock6 = NULL;
 609         if (ipv6 || metadata)
 610                 ret = geneve_sock_add(geneve, true);
 611 #endif
 612         if (!ret && (!ipv6 || metadata))
 613                 ret = geneve_sock_add(geneve, false);
 614         if (ret < 0)
 615                 geneve_sock_release(geneve);
 616
 617         return ret;
 618 }
 619
 620 static int geneve_stop(struct net_device *dev)
 621 {
 622         struct geneve_dev *geneve = netdev_priv(dev);
 623
 624         if (!hlist_unhashed(&geneve->hlist))
 625                 hlist_del_rcu(&geneve->hlist);
 626         geneve_sock_release(geneve);
 627         return 0;
 628 }
 629
 630 static void geneve_build_header(struct genevehdr *geneveh,
 631                                 __be16 tun_flags, u8 vni[3],
 632                                 u8 options_len, u8 *options)
 633 {
 634         geneveh->ver = GENEVE_VER;
 635         geneveh->opt_len = options_len / 4;
 636         geneveh->oam = !!(tun_flags & TUNNEL_OAM);
 637         geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 638         geneveh->rsvd1 = 0;
 639         memcpy(geneveh->vni, vni, 3);
 640         geneveh->proto_type = htons(ETH_P_TEB);
 641         geneveh->rsvd2 = 0;
 642
 643         memcpy(geneveh->options, options, options_len);
 644 }
 645
 646 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 647                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 648                             u32 flags, bool xnet)
 649 {
 650         struct genevehdr *gnvh;
 651         int min_headroom;
 652         int err;
 653         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
 654
 655         skb_scrub_packet(skb, xnet);
 656
 657         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 658                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 659         err = skb_cow_head(skb, min_headroom);
 660         if (unlikely(err))
 661                 goto free_rt;
 662
 663         err = udp_tunnel_handle_offloads(skb, udp_sum);
 664         if (err)
 665                 goto free_rt;
 666
 667         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 668         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 669
 670         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 671         return 0;
 672
 673 free_rt:
 674         ip_rt_put(rt);
 675         return err;
 676 }
 677
 678 #if IS_ENABLED(CONFIG_IPV6)
 679 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 680                              __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 681                              u32 flags, bool xnet)
 682 {
 683         struct genevehdr *gnvh;
 684         int min_headroom;
 685         int err;
 686         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
 687
 688         skb_scrub_packet(skb, xnet);
 689
 690         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
 691                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
 692         err = skb_cow_head(skb, min_headroom);
 693         if (unlikely(err))
 694                 goto free_dst;
 695
 696         err = udp_tunnel_handle_offloads(skb, udp_sum);
 697         if (err)
 698                 goto free_dst;
 699
 700         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 701         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 702
 703         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 704         return 0;
 705
 706 free_dst:
 707         dst_release(dst);
 708         return err;
 709 }
 710 #endif
 711
 712 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 713                                        struct net_device *dev,
 714                                        struct flowi4 *fl4,
 715                                        struct ip_tunnel_info *info)
 716 {
 717         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 718         struct geneve_dev *geneve = netdev_priv(dev);
 719         struct dst_cache *dst_cache;
 720         struct rtable *rt = NULL;
 721         __u8 tos;
 722
 723         memset(fl4, 0, sizeof(*fl4));
 724         fl4->flowi4_mark = skb->mark;
 725         fl4->flowi4_proto = IPPROTO_UDP;
 726
 727         if (info) {
 728                 fl4->daddr = info->key.u.ipv4.dst;
 729                 fl4->saddr = info->key.u.ipv4.src;
 730                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 731                 dst_cache = &info->dst_cache;
 732         } else {
 733                 tos = geneve->tos;
 734                 if (tos == 1) {
 735                         const struct iphdr *iip = ip_hdr(skb);
 736
 737                         tos = ip_tunnel_get_dsfield(iip, skb);
 738                         use_cache = false;
 739                 }
 740
 741                 fl4->flowi4_tos = RT_TOS(tos);
 742                 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
 743                 dst_cache = &geneve->dst_cache;
 744         }
 745
 746         if (use_cache) {
 747                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 748                 if (rt)
 749                         return rt;
 750         }
 751
 752         rt = ip_route_output_key(geneve->net, fl4);
 753         if (IS_ERR(rt)) {
 754                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 755                 return ERR_PTR(-ENETUNREACH);
 756         }
 757         if (rt->dst.dev == dev) { /* is this necessary? */
 758                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 759                 ip_rt_put(rt);
 760                 return ERR_PTR(-ELOOP);
 761         }
 762         if (use_cache)
 763                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 764         return rt;
 765 }
 766
 767 #if IS_ENABLED(CONFIG_IPV6)
 768 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 769                                            struct net_device *dev,
 770                                            struct flowi6 *fl6,
 771                                            struct ip_tunnel_info *info)
 772 {
 773         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 774         struct geneve_dev *geneve = netdev_priv(dev);
 775         struct geneve_sock *gs6 = geneve->sock6;
 776         struct dst_entry *dst = NULL;
 777         struct dst_cache *dst_cache;
 778         __u8 prio;
 779
 780         memset(fl6, 0, sizeof(*fl6));
 781         fl6->flowi6_mark = skb->mark;
 782         fl6->flowi6_proto = IPPROTO_UDP;
 783
 784         if (info) {
 785                 fl6->daddr = info->key.u.ipv6.dst;
 786                 fl6->saddr = info->key.u.ipv6.src;
 787                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(info->key.tos),
 788                                                    info->key.label);
 789                 dst_cache = &info->dst_cache;
 790         } else {
 791                 prio = geneve->tos;
 792                 if (prio == 1) {
 793                         const struct iphdr *iip = ip_hdr(skb);
 794
 795                         prio = ip_tunnel_get_dsfield(iip, skb);
 796                         use_cache = false;
 797                 }
 798
 799                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 800                                                    geneve->label);
 801                 fl6->daddr = geneve->remote.sin6.sin6_addr;
 802                 dst_cache = &geneve->dst_cache;
 803         }
 804
 805         if (use_cache) {
 806                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 807                 if (dst)
 808                         return dst;
 809         }
 810
 811         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 812                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 813                 return ERR_PTR(-ENETUNREACH);
 814         }
 815         if (dst->dev == dev) { /* is this necessary? */
 816                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 817                 dst_release(dst);
 818                 return ERR_PTR(-ELOOP);
 819         }
 820
 821         if (use_cache)
 822                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 823         return dst;
 824 }
 825 #endif
 826
 827 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 828 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 829 {
 830 #ifdef __BIG_ENDIAN
 831         vni[0] = (__force __u8)(tun_id >> 16);
 832         vni[1] = (__force __u8)(tun_id >> 8);
 833         vni[2] = (__force __u8)tun_id;
 834 #else
 835         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 836         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 837         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 838 #endif
 839 }
 840
 841 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 842                                    struct ip_tunnel_info *info)
 843 {
 844         struct geneve_dev *geneve = netdev_priv(dev);
 845         struct geneve_sock *gs4 = geneve->sock4;
 846         struct rtable *rt = NULL;
 847         const struct iphdr *iip; /* interior IP header */
 848         int err = -EINVAL;
 849         struct flowi4 fl4;
 850         __u8 tos, ttl;
 851         __be16 sport;
 852         __be16 df;
 853         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 854         u32 flags = geneve->flags;
 855
 856         if (geneve->collect_md) {
 857                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 858                         netdev_dbg(dev, "no tunnel metadata\n");
 859                         goto tx_error;
 860                 }
 861                 if (info && ip_tunnel_info_af(info) != AF_INET)
 862                         goto tx_error;
 863         }
 864
 865         rt = geneve_get_v4_rt(skb, dev, &fl4, info);
 866         if (IS_ERR(rt)) {
 867                 err = PTR_ERR(rt);
 868                 goto tx_error;
 869         }
 870
 871         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 872         skb_reset_mac_header(skb);
 873
 874         iip = ip_hdr(skb);
 875
 876         if (info) {
 877                 const struct ip_tunnel_key *key = &info->key;
 878                 u8 *opts = NULL;
 879                 u8 vni[3];
 880
 881                 tunnel_id_to_vni(key->tun_id, vni);
 882                 if (info->options_len)
 883                         opts = ip_tunnel_info_opts(info);
 884
 885                 if (key->tun_flags & TUNNEL_CSUM)
 886                         flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
 887                 else
 888                         flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
 889
 890                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 891                                        info->options_len, opts, flags, xnet);
 892                 if (unlikely(err))
 893                         goto tx_error;
 894
 895                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
 896                 ttl = key->ttl;
 897                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 898         } else {
 899                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 900                                        0, NULL, flags, xnet);
 901                 if (unlikely(err))
 902                         goto tx_error;
 903
 904                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 905                 ttl = geneve->ttl;
 906                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 907                         ttl = 1;
 908                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 909                 df = 0;
 910         }
 911         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 912                             tos, ttl, df, sport, geneve->dst_port,
 913                             !net_eq(geneve->net, dev_net(geneve->dev)),
 914                             !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
 915
 916         return NETDEV_TX_OK;
 917
 918 tx_error:
 919         dev_kfree_skb(skb);
 920
 921         if (err == -ELOOP)
 922                 dev->stats.collisions++;
 923         else if (err == -ENETUNREACH)
 924                 dev->stats.tx_carrier_errors++;
 925
 926         dev->stats.tx_errors++;
 927         return NETDEV_TX_OK;
 928 }
 929
 930 #if IS_ENABLED(CONFIG_IPV6)
 931 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 932                                     struct ip_tunnel_info *info)
 933 {
 934         struct geneve_dev *geneve = netdev_priv(dev);
 935         struct geneve_sock *gs6 = geneve->sock6;
 936         struct dst_entry *dst = NULL;
 937         const struct iphdr *iip; /* interior IP header */
 938         int err = -EINVAL;
 939         struct flowi6 fl6;
 940         __u8 prio, ttl;
 941         __be16 sport;
 942         __be32 label;
 943         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 944         u32 flags = geneve->flags;
 945
 946         if (geneve->collect_md) {
 947                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 948                         netdev_dbg(dev, "no tunnel metadata\n");
 949                         goto tx_error;
 950                 }
 951         }
 952
 953         dst = geneve_get_v6_dst(skb, dev, &fl6, info);
 954         if (IS_ERR(dst)) {
 955                 err = PTR_ERR(dst);
 956                 goto tx_error;
 957         }
 958
 959         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 960         skb_reset_mac_header(skb);
 961
 962         iip = ip_hdr(skb);
 963
 964         if (info) {
 965                 const struct ip_tunnel_key *key = &info->key;
 966                 u8 *opts = NULL;
 967                 u8 vni[3];
 968
 969                 tunnel_id_to_vni(key->tun_id, vni);
 970                 if (info->options_len)
 971                         opts = ip_tunnel_info_opts(info);
 972
 973                 if (key->tun_flags & TUNNEL_CSUM)
 974                         flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
 975                 else
 976                         flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
 977
 978                 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
 979                                         info->options_len, opts,
 980                                         flags, xnet);
 981                 if (unlikely(err))
 982                         goto tx_error;
 983
 984                 prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
 985                 ttl = key->ttl;
 986                 label = info->key.label;
 987         } else {
 988                 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
 989                                         0, NULL, flags, xnet);
 990                 if (unlikely(err))
 991                         goto tx_error;
 992
 993                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
 994                                            iip, skb);
 995                 ttl = geneve->ttl;
 996                 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
 997                         ttl = 1;
 998                 ttl = ttl ? : ip6_dst_hoplimit(dst);
 999                 label = geneve->label;
1000         }
1001
1002         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1003                              &fl6.saddr, &fl6.daddr, prio, ttl, label,
1004                              sport, geneve->dst_port,
1005                              !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1006         return NETDEV_TX_OK;
1007
1008 tx_error:
1009         dev_kfree_skb(skb);
1010
1011         if (err == -ELOOP)
1012                 dev->stats.collisions++;
1013         else if (err == -ENETUNREACH)
1014                 dev->stats.tx_carrier_errors++;
1015
1016         dev->stats.tx_errors++;
1017         return NETDEV_TX_OK;
1018 }
1019 #endif
1020
1021 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1022 {
1023         struct geneve_dev *geneve = netdev_priv(dev);
1024         struct ip_tunnel_info *info = NULL;
1025
1026         if (geneve->collect_md)
1027                 info = skb_tunnel_info(skb);
1028
1029 #if IS_ENABLED(CONFIG_IPV6)
1030         if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1031             (!info && geneve->remote.sa.sa_family == AF_INET6))
1032                 return geneve6_xmit_skb(skb, dev, info);
1033 #endif
1034         return geneve_xmit_skb(skb, dev, info);
1035 }
1036
1037 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1038 {
1039         /* Only possible if called internally, ndo_change_mtu path's new_mtu
1040          * is guaranteed to be between dev->min_mtu and dev->max_mtu.
1041          */
1042         if (new_mtu > dev->max_mtu)
1043                 new_mtu = dev->max_mtu;
1044
1045         dev->mtu = new_mtu;
1046         return 0;
1047 }
1048
1049 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1050 {
1051         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1052         struct geneve_dev *geneve = netdev_priv(dev);
1053         struct rtable *rt;
1054         struct flowi4 fl4;
1055 #if IS_ENABLED(CONFIG_IPV6)
1056         struct dst_entry *dst;
1057         struct flowi6 fl6;
1058 #endif
1059
1060         if (ip_tunnel_info_af(info) == AF_INET) {
1061                 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1062                 if (IS_ERR(rt))
1063                         return PTR_ERR(rt);
1064
1065                 ip_rt_put(rt);
1066                 info->key.u.ipv4.src = fl4.saddr;
1067 #if IS_ENABLED(CONFIG_IPV6)
1068         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1069                 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1070                 if (IS_ERR(dst))
1071                         return PTR_ERR(dst);
1072
1073                 dst_release(dst);
1074                 info->key.u.ipv6.src = fl6.saddr;
1075 #endif
1076         } else {
1077                 return -EINVAL;
1078         }
1079
1080         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1081                                              1, USHRT_MAX, true);
1082         info->key.tp_dst = geneve->dst_port;
1083         return 0;
1084 }
1085
1086 static const struct net_device_ops geneve_netdev_ops = {
1087         .ndo_init               = geneve_init,
1088         .ndo_uninit             = geneve_uninit,
1089         .ndo_open               = geneve_open,
1090         .ndo_stop               = geneve_stop,
1091         .ndo_start_xmit         = geneve_xmit,
1092         .ndo_get_stats64        = ip_tunnel_get_stats64,
1093         .ndo_change_mtu         = geneve_change_mtu,
1094         .ndo_validate_addr      = eth_validate_addr,
1095         .ndo_set_mac_address    = eth_mac_addr,
1096         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1097 };
1098
1099 static void geneve_get_drvinfo(struct net_device *dev,
1100                                struct ethtool_drvinfo *drvinfo)
1101 {
1102         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1103         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1104 }
1105
1106 static const struct ethtool_ops geneve_ethtool_ops = {
1107         .get_drvinfo    = geneve_get_drvinfo,
1108         .get_link       = ethtool_op_get_link,
1109 };
1110
1111 /* Info for udev, that this is a virtual tunnel endpoint */
1112 static struct device_type geneve_type = {
1113         .name = "geneve",
1114 };
1115
1116 /* Calls the ndo_udp_tunnel_add of the caller in order to
1117  * supply the listening GENEVE udp ports. Callers are expected
1118  * to implement the ndo_udp_tunnel_add.
1119  */
1120 static void geneve_push_rx_ports(struct net_device *dev)
1121 {
1122         struct net *net = dev_net(dev);
1123         struct geneve_net *gn = net_generic(net, geneve_net_id);
1124         struct geneve_sock *gs;
1125
1126         rcu_read_lock();
1127         list_for_each_entry_rcu(gs, &gn->sock_list, list)
1128                 udp_tunnel_push_rx_port(dev, gs->sock,
1129                                         UDP_TUNNEL_TYPE_GENEVE);
1130         rcu_read_unlock();
1131 }
1132
1133 /* Initialize the device structure. */
1134 static void geneve_setup(struct net_device *dev)
1135 {
1136         ether_setup(dev);
1137
1138         dev->netdev_ops = &geneve_netdev_ops;
1139         dev->ethtool_ops = &geneve_ethtool_ops;
1140         dev->destructor = free_netdev;
1141
1142         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1143
1144         dev->features    |= NETIF_F_LLTX;
1145         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1146         dev->features    |= NETIF_F_RXCSUM;
1147         dev->features    |= NETIF_F_GSO_SOFTWARE;
1148
1149         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1150         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1151
1152         /* MTU range: 68 - (something less than 65535) */
1153         dev->min_mtu = ETH_MIN_MTU;
1154         /* The max_mtu calculation does not take account of GENEVE
1155          * options, to avoid excluding potentially valid
1156          * configurations. This will be further reduced by IPvX hdr size.
1157          */
1158         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1159
1160         netif_keep_dst(dev);
1161         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1162         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1163         eth_hw_addr_random(dev);
1164 }
1165
1166 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1167         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1168         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1169         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1170         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1171         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1172         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1173         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1174         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1175         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1176         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1177         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1178 };
1179
1180 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1181 {
1182         if (tb[IFLA_ADDRESS]) {
1183                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1184                         return -EINVAL;
1185
1186                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1187                         return -EADDRNOTAVAIL;
1188         }
1189
1190         if (!data)
1191                 return -EINVAL;
1192
1193         if (data[IFLA_GENEVE_ID]) {
1194                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1195
1196                 if (vni >= GENEVE_VID_MASK)
1197                         return -ERANGE;
1198         }
1199
1200         return 0;
1201 }
1202
1203 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1204                                           __be16 dst_port,
1205                                           union geneve_addr *remote,
1206                                           u8 vni[],
1207                                           bool *tun_on_same_port,
1208                                           bool *tun_collect_md)
1209 {
1210         struct geneve_dev *geneve, *t;
1211
1212         *tun_on_same_port = false;
1213         *tun_collect_md = false;
1214         t = NULL;
1215         list_for_each_entry(geneve, &gn->geneve_list, next) {
1216                 if (geneve->dst_port == dst_port) {
1217                         *tun_collect_md = geneve->collect_md;
1218                         *tun_on_same_port = true;
1219                 }
1220                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1221                     !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1222                     dst_port == geneve->dst_port)
1223                         t = geneve;
1224         }
1225         return t;
1226 }
1227
1228 static int geneve_configure(struct net *net, struct net_device *dev,
1229                             union geneve_addr *remote,
1230                             __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1231                             __be16 dst_port, bool metadata, u32 flags)
1232 {
1233         struct geneve_net *gn = net_generic(net, geneve_net_id);
1234         struct geneve_dev *t, *geneve = netdev_priv(dev);
1235         bool tun_collect_md, tun_on_same_port;
1236         int err, encap_len;
1237
1238         if (!remote)
1239                 return -EINVAL;
1240         if (metadata &&
1241             (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1242                 return -EINVAL;
1243
1244         geneve->net = net;
1245         geneve->dev = dev;
1246
1247         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1248         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1249         geneve->vni[2] =  vni & 0x000000ff;
1250
1251         if ((remote->sa.sa_family == AF_INET &&
1252              IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1253             (remote->sa.sa_family == AF_INET6 &&
1254              ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1255                 return -EINVAL;
1256         if (label && remote->sa.sa_family != AF_INET6)
1257                 return -EINVAL;
1258
1259         geneve->remote = *remote;
1260
1261         geneve->ttl = ttl;
1262         geneve->tos = tos;
1263         geneve->label = label;
1264         geneve->dst_port = dst_port;
1265         geneve->collect_md = metadata;
1266         geneve->flags = flags;
1267
1268         t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1269                             &tun_on_same_port, &tun_collect_md);
1270         if (t)
1271                 return -EBUSY;
1272
1273         /* make enough headroom for basic scenario */
1274         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1275         if (remote->sa.sa_family == AF_INET) {
1276                 encap_len += sizeof(struct iphdr);
1277                 dev->max_mtu -= sizeof(struct iphdr);
1278         } else {
1279                 encap_len += sizeof(struct ipv6hdr);
1280                 dev->max_mtu -= sizeof(struct ipv6hdr);
1281         }
1282         dev->needed_headroom = encap_len + ETH_HLEN;
1283
1284         if (metadata) {
1285                 if (tun_on_same_port)
1286                         return -EPERM;
1287         } else {
1288                 if (tun_collect_md)
1289                         return -EPERM;
1290         }
1291
1292         dst_cache_reset(&geneve->dst_cache);
1293
1294         err = register_netdevice(dev);
1295         if (err)
1296                 return err;
1297
1298         list_add(&geneve->next, &gn->geneve_list);
1299         return 0;
1300 }
1301
1302 static int geneve_newlink(struct net *net, struct net_device *dev,
1303                           struct nlattr *tb[], struct nlattr *data[])
1304 {
1305         __be16 dst_port = htons(GENEVE_UDP_PORT);
1306         __u8 ttl = 0, tos = 0;
1307         bool metadata = false;
1308         union geneve_addr remote = geneve_remote_unspec;
1309         __be32 label = 0;
1310         __u32 vni = 0;
1311         u32 flags = 0;
1312
1313         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1314                 return -EINVAL;
1315
1316         if (data[IFLA_GENEVE_REMOTE]) {
1317                 remote.sa.sa_family = AF_INET;
1318                 remote.sin.sin_addr.s_addr =
1319                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1320         }
1321
1322         if (data[IFLA_GENEVE_REMOTE6]) {
1323                 if (!IS_ENABLED(CONFIG_IPV6))
1324                         return -EPFNOSUPPORT;
1325
1326                 remote.sa.sa_family = AF_INET6;
1327                 remote.sin6.sin6_addr =
1328                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1329
1330                 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1331                     IPV6_ADDR_LINKLOCAL) {
1332                         netdev_dbg(dev, "link-local remote is unsupported\n");
1333                         return -EINVAL;
1334                 }
1335         }
1336
1337         if (data[IFLA_GENEVE_ID])
1338                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1339
1340         if (data[IFLA_GENEVE_TTL])
1341                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1342
1343         if (data[IFLA_GENEVE_TOS])
1344                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1345
1346         if (data[IFLA_GENEVE_LABEL])
1347                 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1348                         IPV6_FLOWLABEL_MASK;
1349
1350         if (data[IFLA_GENEVE_PORT])
1351                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1352
1353         if (data[IFLA_GENEVE_COLLECT_METADATA])
1354                 metadata = true;
1355
1356         if (data[IFLA_GENEVE_UDP_CSUM] &&
1357             !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1358                 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1359
1360         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1361             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1362                 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1363
1364         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1365             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1366                 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1367
1368         return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1369                                 dst_port, metadata, flags);
1370 }
1371
1372 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1373 {
1374         struct geneve_dev *geneve = netdev_priv(dev);
1375
1376         list_del(&geneve->next);
1377         unregister_netdevice_queue(dev, head);
1378 }
1379
1380 static size_t geneve_get_size(const struct net_device *dev)
1381 {
1382         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1383                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1384                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1385                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1386                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1387                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1388                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1389                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1390                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1391                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1392                 0;
1393 }
1394
1395 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1396 {
1397         struct geneve_dev *geneve = netdev_priv(dev);
1398         __u32 vni;
1399
1400         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1401         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1402                 goto nla_put_failure;
1403
1404         if (geneve->remote.sa.sa_family == AF_INET) {
1405                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1406                                     geneve->remote.sin.sin_addr.s_addr))
1407                         goto nla_put_failure;
1408 #if IS_ENABLED(CONFIG_IPV6)
1409         } else {
1410                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1411                                      &geneve->remote.sin6.sin6_addr))
1412                         goto nla_put_failure;
1413 #endif
1414         }
1415
1416         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1417             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1418             nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1419                 goto nla_put_failure;
1420
1421         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1422                 goto nla_put_failure;
1423
1424         if (geneve->collect_md) {
1425                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1426                         goto nla_put_failure;
1427         }
1428
1429         if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1430                        !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1431             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1432                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1433             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1434                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1435                 goto nla_put_failure;
1436
1437         return 0;
1438
1439 nla_put_failure:
1440         return -EMSGSIZE;
1441 }
1442
1443 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1444         .kind           = "geneve",
1445         .maxtype        = IFLA_GENEVE_MAX,
1446         .policy         = geneve_policy,
1447         .priv_size      = sizeof(struct geneve_dev),
1448         .setup          = geneve_setup,
1449         .validate       = geneve_validate,
1450         .newlink        = geneve_newlink,
1451         .dellink        = geneve_dellink,
1452         .get_size       = geneve_get_size,
1453         .fill_info      = geneve_fill_info,
1454 };
1455
1456 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1457                                         u8 name_assign_type, u16 dst_port)
1458 {
1459         struct nlattr *tb[IFLA_MAX + 1];
1460         struct net_device *dev;
1461         LIST_HEAD(list_kill);
1462         int err;
1463
1464         memset(tb, 0, sizeof(tb));
1465         dev = rtnl_create_link(net, name, name_assign_type,
1466                                &geneve_link_ops, tb);
1467         if (IS_ERR(dev))
1468                 return dev;
1469
1470         err = geneve_configure(net, dev, &geneve_remote_unspec,
1471                                0, 0, 0, 0, htons(dst_port), true,
1472                                GENEVE_F_UDP_ZERO_CSUM6_RX);
1473         if (err) {
1474                 free_netdev(dev);
1475                 return ERR_PTR(err);
1476         }
1477
1478         /* openvswitch users expect packet sizes to be unrestricted,
1479          * so set the largest MTU we can.
1480          */
1481         err = geneve_change_mtu(dev, IP_MAX_MTU);
1482         if (err)
1483                 goto err;
1484
1485         err = rtnl_configure_link(dev, NULL);
1486         if (err < 0)
1487                 goto err;
1488
1489         return dev;
1490
1491  err:
1492         geneve_dellink(dev, &list_kill);
1493         unregister_netdevice_many(&list_kill);
1494         return ERR_PTR(err);
1495 }
1496 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1497
1498 static int geneve_netdevice_event(struct notifier_block *unused,
1499                                   unsigned long event, void *ptr)
1500 {
1501         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1502
1503         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1504                 geneve_push_rx_ports(dev);
1505
1506         return NOTIFY_DONE;
1507 }
1508
1509 static struct notifier_block geneve_notifier_block __read_mostly = {
1510         .notifier_call = geneve_netdevice_event,
1511 };
1512
1513 static __net_init int geneve_init_net(struct net *net)
1514 {
1515         struct geneve_net *gn = net_generic(net, geneve_net_id);
1516
1517         INIT_LIST_HEAD(&gn->geneve_list);
1518         INIT_LIST_HEAD(&gn->sock_list);
1519         return 0;
1520 }
1521
1522 static void __net_exit geneve_exit_net(struct net *net)
1523 {
1524         struct geneve_net *gn = net_generic(net, geneve_net_id);
1525         struct geneve_dev *geneve, *next;
1526         struct net_device *dev, *aux;
1527         LIST_HEAD(list);
1528
1529         rtnl_lock();
1530
1531         /* gather any geneve devices that were moved into this ns */
1532         for_each_netdev_safe(net, dev, aux)
1533                 if (dev->rtnl_link_ops == &geneve_link_ops)
1534                         unregister_netdevice_queue(dev, &list);
1535
1536         /* now gather any other geneve devices that were created in this ns */
1537         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1538                 /* If geneve->dev is in the same netns, it was already added
1539                  * to the list by the previous loop.
1540                  */
1541                 if (!net_eq(dev_net(geneve->dev), net))
1542                         unregister_netdevice_queue(geneve->dev, &list);
1543         }
1544
1545         /* unregister the devices gathered above */
1546         unregister_netdevice_many(&list);
1547         rtnl_unlock();
1548 }
1549
1550 static struct pernet_operations geneve_net_ops = {
1551         .init = geneve_init_net,
1552         .exit = geneve_exit_net,
1553         .id   = &geneve_net_id,
1554         .size = sizeof(struct geneve_net),
1555 };
1556
1557 static int __init geneve_init_module(void)
1558 {
1559         int rc;
1560
1561         rc = register_pernet_subsys(&geneve_net_ops);
1562         if (rc)
1563                 goto out1;
1564
1565         rc = register_netdevice_notifier(&geneve_notifier_block);
1566         if (rc)
1567                 goto out2;
1568
1569         rc = rtnl_link_register(&geneve_link_ops);
1570         if (rc)
1571                 goto out3;
1572
1573         return 0;
1574
1575 out3:
1576         unregister_netdevice_notifier(&geneve_notifier_block);
1577 out2:
1578         unregister_pernet_subsys(&geneve_net_ops);
1579 out1:
1580         return rc;
1581 }
1582 late_initcall(geneve_init_module);
1583
1584 static void __exit geneve_cleanup_module(void)
1585 {
1586         rtnl_link_unregister(&geneve_link_ops);
1587         unregister_netdevice_notifier(&geneve_notifier_block);
1588         unregister_pernet_subsys(&geneve_net_ops);
1589 }
1590 module_exit(geneve_cleanup_module);
1591
1592 MODULE_LICENSE("GPL");
1593 MODULE_VERSION(GENEVE_NETDEV_VER);
1594 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1595 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1596 MODULE_ALIAS_RTNL_LINK("geneve");