drivers/net/geneve.c

   1 /*
   2  * GENEVE: Generic Network Virtualization Encapsulation
   3  *
   4  * Copyright (c) 2015 Red Hat, Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/netdevice.h>
  16 #include <linux/etherdevice.h>
  17 #include <linux/hash.h>
  18 #include <net/dst_metadata.h>
  19 #include <net/gro_cells.h>
  20 #include <net/rtnetlink.h>
  21 #include <net/geneve.h>
  22 #include <net/protocol.h>
  23
  24 #define GENEVE_NETDEV_VER       "0.6"
  25
  26 #define GENEVE_UDP_PORT         6081
  27
  28 #define GENEVE_N_VID            (1u << 24)
  29 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  30
  31 #define VNI_HASH_BITS           10
  32 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  33
  34 static bool log_ecn_error = true;
  35 module_param(log_ecn_error, bool, 0644);
  36 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  37
  38 #define GENEVE_VER 0
  39 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  40
  41 /* per-network namespace private data for this module */
  42 struct geneve_net {
  43         struct list_head        geneve_list;
  44         struct list_head        sock_list;
  45 };
  46
  47 static int geneve_net_id;
  48
  49 union geneve_addr {
  50         struct sockaddr_in sin;
  51         struct sockaddr_in6 sin6;
  52         struct sockaddr sa;
  53 };
  54
  55 static union geneve_addr geneve_remote_unspec = { .sa.sa_family = AF_UNSPEC, };
  56
  57 /* Pseudo network device */
  58 struct geneve_dev {
  59         struct hlist_node  hlist;       /* vni hash table */
  60         struct net         *net;        /* netns for packet i/o */
  61         struct net_device  *dev;        /* netdev for geneve tunnel */
  62         struct geneve_sock *sock4;      /* IPv4 socket used for geneve tunnel */
  63 #if IS_ENABLED(CONFIG_IPV6)
  64         struct geneve_sock *sock6;      /* IPv6 socket used for geneve tunnel */
  65 #endif
  66         u8                 vni[3];      /* virtual network ID for tunnel */
  67         u8                 ttl;         /* TTL override */
  68         u8                 tos;         /* TOS override */
  69         union geneve_addr  remote;      /* IP address for link partner */
  70         struct list_head   next;        /* geneve's per namespace list */
  71         __be16             dst_port;
  72         bool               collect_md;
  73         struct gro_cells   gro_cells;
  74         u32                flags;
  75 };
  76
  77 /* Geneve device flags */
  78 #define GENEVE_F_UDP_CSUM               BIT(0)
  79 #define GENEVE_F_UDP_ZERO_CSUM6_TX      BIT(1)
  80 #define GENEVE_F_UDP_ZERO_CSUM6_RX      BIT(2)
  81
  82 struct geneve_sock {
  83         bool                    collect_md;
  84         struct list_head        list;
  85         struct socket           *sock;
  86         struct rcu_head         rcu;
  87         int                     refcnt;
  88         struct udp_offload      udp_offloads;
  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 struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 113                                         __be32 addr, u8 vni[])
 114 {
 115         struct hlist_head *vni_list_head;
 116         struct geneve_dev *geneve;
 117         __u32 hash;
 118
 119         /* Find the device for this VNI */
 120         hash = geneve_net_vni_hash(vni);
 121         vni_list_head = &gs->vni_list[hash];
 122         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 123                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 124                     addr == geneve->remote.sin.sin_addr.s_addr)
 125                         return geneve;
 126         }
 127         return NULL;
 128 }
 129
 130 #if IS_ENABLED(CONFIG_IPV6)
 131 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 132                                          struct in6_addr addr6, u8 vni[])
 133 {
 134         struct hlist_head *vni_list_head;
 135         struct geneve_dev *geneve;
 136         __u32 hash;
 137
 138         /* Find the device for this VNI */
 139         hash = geneve_net_vni_hash(vni);
 140         vni_list_head = &gs->vni_list[hash];
 141         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 142                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 143                     ipv6_addr_equal(&addr6, &geneve->remote.sin6.sin6_addr))
 144                         return geneve;
 145         }
 146         return NULL;
 147 }
 148 #endif
 149
 150 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 151 {
 152         return (struct genevehdr *)(udp_hdr(skb) + 1);
 153 }
 154
 155 /* geneve receive/decap routine */
 156 static void geneve_rx(struct geneve_sock *gs, struct sk_buff *skb)
 157 {
 158         struct genevehdr *gnvh = geneve_hdr(skb);
 159         struct metadata_dst *tun_dst = NULL;
 160         struct geneve_dev *geneve = NULL;
 161         struct pcpu_sw_netstats *stats;
 162         struct iphdr *iph = NULL;
 163         __be32 addr;
 164         static u8 zero_vni[3];
 165         u8 *vni;
 166         int err = 0;
 167         sa_family_t sa_family;
 168 #if IS_ENABLED(CONFIG_IPV6)
 169         struct ipv6hdr *ip6h = NULL;
 170         struct in6_addr addr6;
 171         static struct in6_addr zero_addr6;
 172 #endif
 173
 174         sa_family = gs->sock->sk->sk_family;
 175
 176         if (sa_family == AF_INET) {
 177                 iph = ip_hdr(skb); /* outer IP header... */
 178
 179                 if (gs->collect_md) {
 180                         vni = zero_vni;
 181                         addr = 0;
 182                 } else {
 183                         vni = gnvh->vni;
 184
 185                         addr = iph->saddr;
 186                 }
 187
 188                 geneve = geneve_lookup(gs, addr, vni);
 189 #if IS_ENABLED(CONFIG_IPV6)
 190         } else if (sa_family == AF_INET6) {
 191                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 192
 193                 if (gs->collect_md) {
 194                         vni = zero_vni;
 195                         addr6 = zero_addr6;
 196                 } else {
 197                         vni = gnvh->vni;
 198
 199                         addr6 = ip6h->saddr;
 200                 }
 201
 202                 geneve = geneve6_lookup(gs, addr6, vni);
 203 #endif
 204         }
 205         if (!geneve)
 206                 goto drop;
 207
 208         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 209                 __be16 flags;
 210
 211                 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
 212                         (gnvh->oam ? TUNNEL_OAM : 0) |
 213                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 214
 215                 tun_dst = udp_tun_rx_dst(skb, sa_family, flags,
 216                                          vni_to_tunnel_id(gnvh->vni),
 217                                          gnvh->opt_len * 4);
 218                 if (!tun_dst)
 219                         goto drop;
 220                 /* Update tunnel dst according to Geneve options. */
 221                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 222                                         gnvh->options, gnvh->opt_len * 4);
 223         } else {
 224                 /* Drop packets w/ critical options,
 225                  * since we don't support any...
 226                  */
 227                 if (gnvh->critical)
 228                         goto drop;
 229         }
 230
 231         skb_reset_mac_header(skb);
 232         skb_scrub_packet(skb, !net_eq(geneve->net, dev_net(geneve->dev)));
 233         skb->protocol = eth_type_trans(skb, geneve->dev);
 234         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 235
 236         if (tun_dst)
 237                 skb_dst_set(skb, &tun_dst->dst);
 238
 239         /* Ignore packet loops (and multicast echo) */
 240         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
 241                 goto drop;
 242
 243         skb_reset_network_header(skb);
 244
 245         if (iph)
 246                 err = IP_ECN_decapsulate(iph, skb);
 247 #if IS_ENABLED(CONFIG_IPV6)
 248         if (ip6h)
 249                 err = IP6_ECN_decapsulate(ip6h, skb);
 250 #endif
 251
 252         if (unlikely(err)) {
 253                 if (log_ecn_error) {
 254                         if (iph)
 255                                 net_info_ratelimited("non-ECT from %pI4 "
 256                                                      "with TOS=%#x\n",
 257                                                      &iph->saddr, iph->tos);
 258 #if IS_ENABLED(CONFIG_IPV6)
 259                         if (ip6h)
 260                                 net_info_ratelimited("non-ECT from %pI6\n",
 261                                                      &ip6h->saddr);
 262 #endif
 263                 }
 264                 if (err > 1) {
 265                         ++geneve->dev->stats.rx_frame_errors;
 266                         ++geneve->dev->stats.rx_errors;
 267                         goto drop;
 268                 }
 269         }
 270
 271         stats = this_cpu_ptr(geneve->dev->tstats);
 272         u64_stats_update_begin(&stats->syncp);
 273         stats->rx_packets++;
 274         stats->rx_bytes += skb->len;
 275         u64_stats_update_end(&stats->syncp);
 276
 277         gro_cells_receive(&geneve->gro_cells, skb);
 278         return;
 279 drop:
 280         /* Consume bad packet */
 281         kfree_skb(skb);
 282 }
 283
 284 /* Setup stats when device is created */
 285 static int geneve_init(struct net_device *dev)
 286 {
 287         struct geneve_dev *geneve = netdev_priv(dev);
 288         int err;
 289
 290         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 291         if (!dev->tstats)
 292                 return -ENOMEM;
 293
 294         err = gro_cells_init(&geneve->gro_cells, dev);
 295         if (err) {
 296                 free_percpu(dev->tstats);
 297                 return err;
 298         }
 299
 300         return 0;
 301 }
 302
 303 static void geneve_uninit(struct net_device *dev)
 304 {
 305         struct geneve_dev *geneve = netdev_priv(dev);
 306
 307         gro_cells_destroy(&geneve->gro_cells);
 308         free_percpu(dev->tstats);
 309 }
 310
 311 /* Callback from net/ipv4/udp.c to receive packets */
 312 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 313 {
 314         struct genevehdr *geneveh;
 315         struct geneve_sock *gs;
 316         int opts_len;
 317
 318         /* Need Geneve and inner Ethernet header to be present */
 319         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 320                 goto error;
 321
 322         /* Return packets with reserved bits set */
 323         geneveh = geneve_hdr(skb);
 324         if (unlikely(geneveh->ver != GENEVE_VER))
 325                 goto error;
 326
 327         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 328                 goto error;
 329
 330         opts_len = geneveh->opt_len * 4;
 331         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 332                                  htons(ETH_P_TEB)))
 333                 goto drop;
 334
 335         gs = rcu_dereference_sk_user_data(sk);
 336         if (!gs)
 337                 goto drop;
 338
 339         geneve_rx(gs, skb);
 340         return 0;
 341
 342 drop:
 343         /* Consume bad packet */
 344         kfree_skb(skb);
 345         return 0;
 346
 347 error:
 348         /* Let the UDP layer deal with the skb */
 349         return 1;
 350 }
 351
 352 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 353                                          __be16 port, u32 flags)
 354 {
 355         struct socket *sock;
 356         struct udp_port_cfg udp_conf;
 357         int err;
 358
 359         memset(&udp_conf, 0, sizeof(udp_conf));
 360
 361         if (ipv6) {
 362                 udp_conf.family = AF_INET6;
 363                 udp_conf.ipv6_v6only = 1;
 364                 udp_conf.use_udp6_rx_checksums =
 365                     !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX);
 366         } else {
 367                 udp_conf.family = AF_INET;
 368                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 369         }
 370
 371         udp_conf.local_udp_port = port;
 372
 373         /* Open UDP socket */
 374         err = udp_sock_create(net, &udp_conf, &sock);
 375         if (err < 0)
 376                 return ERR_PTR(err);
 377
 378         return sock;
 379 }
 380
 381 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
 382 {
 383         struct net_device *dev;
 384         struct sock *sk = gs->sock->sk;
 385         struct net *net = sock_net(sk);
 386         sa_family_t sa_family = sk->sk_family;
 387         __be16 port = inet_sk(sk)->inet_sport;
 388         int err;
 389
 390         if (sa_family == AF_INET) {
 391                 err = udp_add_offload(sock_net(sk), &gs->udp_offloads);
 392                 if (err)
 393                         pr_warn("geneve: udp_add_offload failed with status %d\n",
 394                                 err);
 395         }
 396
 397         rcu_read_lock();
 398         for_each_netdev_rcu(net, dev) {
 399                 if (dev->netdev_ops->ndo_add_geneve_port)
 400                         dev->netdev_ops->ndo_add_geneve_port(dev, sa_family,
 401                                                              port);
 402         }
 403         rcu_read_unlock();
 404 }
 405
 406 static int geneve_hlen(struct genevehdr *gh)
 407 {
 408         return sizeof(*gh) + gh->opt_len * 4;
 409 }
 410
 411 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
 412                                            struct sk_buff *skb,
 413                                            struct udp_offload *uoff)
 414 {
 415         struct sk_buff *p, **pp = NULL;
 416         struct genevehdr *gh, *gh2;
 417         unsigned int hlen, gh_len, off_gnv;
 418         const struct packet_offload *ptype;
 419         __be16 type;
 420         int flush = 1;
 421
 422         off_gnv = skb_gro_offset(skb);
 423         hlen = off_gnv + sizeof(*gh);
 424         gh = skb_gro_header_fast(skb, off_gnv);
 425         if (skb_gro_header_hard(skb, hlen)) {
 426                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 427                 if (unlikely(!gh))
 428                         goto out;
 429         }
 430
 431         if (gh->ver != GENEVE_VER || gh->oam)
 432                 goto out;
 433         gh_len = geneve_hlen(gh);
 434
 435         hlen = off_gnv + gh_len;
 436         if (skb_gro_header_hard(skb, hlen)) {
 437                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 438                 if (unlikely(!gh))
 439                         goto out;
 440         }
 441
 442         flush = 0;
 443
 444         for (p = *head; p; p = p->next) {
 445                 if (!NAPI_GRO_CB(p)->same_flow)
 446                         continue;
 447
 448                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 449                 if (gh->opt_len != gh2->opt_len ||
 450                     memcmp(gh, gh2, gh_len)) {
 451                         NAPI_GRO_CB(p)->same_flow = 0;
 452                         continue;
 453                 }
 454         }
 455
 456         type = gh->proto_type;
 457
 458         rcu_read_lock();
 459         ptype = gro_find_receive_by_type(type);
 460         if (!ptype) {
 461                 flush = 1;
 462                 goto out_unlock;
 463         }
 464
 465         skb_gro_pull(skb, gh_len);
 466         skb_gro_postpull_rcsum(skb, gh, gh_len);
 467         pp = ptype->callbacks.gro_receive(head, skb);
 468
 469 out_unlock:
 470         rcu_read_unlock();
 471 out:
 472         NAPI_GRO_CB(skb)->flush |= flush;
 473
 474         return pp;
 475 }
 476
 477 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
 478                                struct udp_offload *uoff)
 479 {
 480         struct genevehdr *gh;
 481         struct packet_offload *ptype;
 482         __be16 type;
 483         int gh_len;
 484         int err = -ENOSYS;
 485
 486         udp_tunnel_gro_complete(skb, nhoff);
 487
 488         gh = (struct genevehdr *)(skb->data + nhoff);
 489         gh_len = geneve_hlen(gh);
 490         type = gh->proto_type;
 491
 492         rcu_read_lock();
 493         ptype = gro_find_complete_by_type(type);
 494         if (ptype)
 495                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 496
 497         rcu_read_unlock();
 498         return err;
 499 }
 500
 501 /* Create new listen socket if needed */
 502 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 503                                                 bool ipv6, u32 flags)
 504 {
 505         struct geneve_net *gn = net_generic(net, geneve_net_id);
 506         struct geneve_sock *gs;
 507         struct socket *sock;
 508         struct udp_tunnel_sock_cfg tunnel_cfg;
 509         int h;
 510
 511         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 512         if (!gs)
 513                 return ERR_PTR(-ENOMEM);
 514
 515         sock = geneve_create_sock(net, ipv6, port, flags);
 516         if (IS_ERR(sock)) {
 517                 kfree(gs);
 518                 return ERR_CAST(sock);
 519         }
 520
 521         gs->sock = sock;
 522         gs->refcnt = 1;
 523         for (h = 0; h < VNI_HASH_SIZE; ++h)
 524                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 525
 526         /* Initialize the geneve udp offloads structure */
 527         gs->udp_offloads.port = port;
 528         gs->udp_offloads.callbacks.gro_receive  = geneve_gro_receive;
 529         gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
 530         geneve_notify_add_rx_port(gs);
 531
 532         /* Mark socket as an encapsulation socket */
 533         tunnel_cfg.sk_user_data = gs;
 534         tunnel_cfg.encap_type = 1;
 535         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 536         tunnel_cfg.encap_destroy = NULL;
 537         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 538         list_add(&gs->list, &gn->sock_list);
 539         return gs;
 540 }
 541
 542 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 543 {
 544         struct net_device *dev;
 545         struct sock *sk = gs->sock->sk;
 546         struct net *net = sock_net(sk);
 547         sa_family_t sa_family = sk->sk_family;
 548         __be16 port = inet_sk(sk)->inet_sport;
 549
 550         rcu_read_lock();
 551         for_each_netdev_rcu(net, dev) {
 552                 if (dev->netdev_ops->ndo_del_geneve_port)
 553                         dev->netdev_ops->ndo_del_geneve_port(dev, sa_family,
 554                                                              port);
 555         }
 556
 557         rcu_read_unlock();
 558
 559         if (sa_family == AF_INET)
 560                 udp_del_offload(&gs->udp_offloads);
 561 }
 562
 563 static void __geneve_sock_release(struct geneve_sock *gs)
 564 {
 565         if (!gs || --gs->refcnt)
 566                 return;
 567
 568         list_del(&gs->list);
 569         geneve_notify_del_rx_port(gs);
 570         udp_tunnel_sock_release(gs->sock);
 571         kfree_rcu(gs, rcu);
 572 }
 573
 574 static void geneve_sock_release(struct geneve_dev *geneve)
 575 {
 576         __geneve_sock_release(geneve->sock4);
 577 #if IS_ENABLED(CONFIG_IPV6)
 578         __geneve_sock_release(geneve->sock6);
 579 #endif
 580 }
 581
 582 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 583                                             sa_family_t family,
 584                                             __be16 dst_port)
 585 {
 586         struct geneve_sock *gs;
 587
 588         list_for_each_entry(gs, &gn->sock_list, list) {
 589                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 590                     inet_sk(gs->sock->sk)->sk.sk_family == family) {
 591                         return gs;
 592                 }
 593         }
 594         return NULL;
 595 }
 596
 597 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 598 {
 599         struct net *net = geneve->net;
 600         struct geneve_net *gn = net_generic(net, geneve_net_id);
 601         struct geneve_sock *gs;
 602         __u32 hash;
 603
 604         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
 605         if (gs) {
 606                 gs->refcnt++;
 607                 goto out;
 608         }
 609
 610         gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
 611         if (IS_ERR(gs))
 612                 return PTR_ERR(gs);
 613
 614 out:
 615         gs->collect_md = geneve->collect_md;
 616         gs->flags = geneve->flags;
 617 #if IS_ENABLED(CONFIG_IPV6)
 618         if (ipv6)
 619                 geneve->sock6 = gs;
 620         else
 621 #endif
 622                 geneve->sock4 = gs;
 623
 624         hash = geneve_net_vni_hash(geneve->vni);
 625         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 626         return 0;
 627 }
 628
 629 static int geneve_open(struct net_device *dev)
 630 {
 631         struct geneve_dev *geneve = netdev_priv(dev);
 632         bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
 633         bool metadata = geneve->collect_md;
 634         int ret = 0;
 635
 636         geneve->sock4 = NULL;
 637 #if IS_ENABLED(CONFIG_IPV6)
 638         geneve->sock6 = NULL;
 639         if (ipv6 || metadata)
 640                 ret = geneve_sock_add(geneve, true);
 641 #endif
 642         if (!ret && (!ipv6 || metadata))
 643                 ret = geneve_sock_add(geneve, false);
 644         if (ret < 0)
 645                 geneve_sock_release(geneve);
 646
 647         return ret;
 648 }
 649
 650 static int geneve_stop(struct net_device *dev)
 651 {
 652         struct geneve_dev *geneve = netdev_priv(dev);
 653
 654         if (!hlist_unhashed(&geneve->hlist))
 655                 hlist_del_rcu(&geneve->hlist);
 656         geneve_sock_release(geneve);
 657         return 0;
 658 }
 659
 660 static void geneve_build_header(struct genevehdr *geneveh,
 661                                 __be16 tun_flags, u8 vni[3],
 662                                 u8 options_len, u8 *options)
 663 {
 664         geneveh->ver = GENEVE_VER;
 665         geneveh->opt_len = options_len / 4;
 666         geneveh->oam = !!(tun_flags & TUNNEL_OAM);
 667         geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 668         geneveh->rsvd1 = 0;
 669         memcpy(geneveh->vni, vni, 3);
 670         geneveh->proto_type = htons(ETH_P_TEB);
 671         geneveh->rsvd2 = 0;
 672
 673         memcpy(geneveh->options, options, options_len);
 674 }
 675
 676 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 677                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 678                             u32 flags, bool xnet)
 679 {
 680         struct genevehdr *gnvh;
 681         int min_headroom;
 682         int err;
 683         bool udp_sum = !!(flags & GENEVE_F_UDP_CSUM);
 684
 685         skb_scrub_packet(skb, xnet);
 686
 687         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 688                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 689         err = skb_cow_head(skb, min_headroom);
 690         if (unlikely(err)) {
 691                 kfree_skb(skb);
 692                 goto free_rt;
 693         }
 694
 695         skb = udp_tunnel_handle_offloads(skb, udp_sum);
 696         if (IS_ERR(skb)) {
 697                 err = PTR_ERR(skb);
 698                 goto free_rt;
 699         }
 700
 701         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 702         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 703
 704         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 705         return 0;
 706
 707 free_rt:
 708         ip_rt_put(rt);
 709         return err;
 710 }
 711
 712 #if IS_ENABLED(CONFIG_IPV6)
 713 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 714                              __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 715                              u32 flags, bool xnet)
 716 {
 717         struct genevehdr *gnvh;
 718         int min_headroom;
 719         int err;
 720         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
 721
 722         skb_scrub_packet(skb, xnet);
 723
 724         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
 725                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
 726         err = skb_cow_head(skb, min_headroom);
 727         if (unlikely(err)) {
 728                 kfree_skb(skb);
 729                 goto free_dst;
 730         }
 731
 732         skb = udp_tunnel_handle_offloads(skb, udp_sum);
 733         if (IS_ERR(skb)) {
 734                 err = PTR_ERR(skb);
 735                 goto free_dst;
 736         }
 737
 738         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 739         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 740
 741         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 742         return 0;
 743
 744 free_dst:
 745         dst_release(dst);
 746         return err;
 747 }
 748 #endif
 749
 750 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 751                                        struct net_device *dev,
 752                                        struct flowi4 *fl4,
 753                                        struct ip_tunnel_info *info)
 754 {
 755         struct geneve_dev *geneve = netdev_priv(dev);
 756         struct rtable *rt = NULL;
 757         __u8 tos;
 758
 759         memset(fl4, 0, sizeof(*fl4));
 760         fl4->flowi4_mark = skb->mark;
 761         fl4->flowi4_proto = IPPROTO_UDP;
 762
 763         if (info) {
 764                 fl4->daddr = info->key.u.ipv4.dst;
 765                 fl4->saddr = info->key.u.ipv4.src;
 766                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 767         } else {
 768                 tos = geneve->tos;
 769                 if (tos == 1) {
 770                         const struct iphdr *iip = ip_hdr(skb);
 771
 772                         tos = ip_tunnel_get_dsfield(iip, skb);
 773                 }
 774
 775                 fl4->flowi4_tos = RT_TOS(tos);
 776                 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
 777         }
 778
 779         rt = ip_route_output_key(geneve->net, fl4);
 780         if (IS_ERR(rt)) {
 781                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 782                 return ERR_PTR(-ENETUNREACH);
 783         }
 784         if (rt->dst.dev == dev) { /* is this necessary? */
 785                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 786                 ip_rt_put(rt);
 787                 return ERR_PTR(-ELOOP);
 788         }
 789         return rt;
 790 }
 791
 792 #if IS_ENABLED(CONFIG_IPV6)
 793 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 794                                            struct net_device *dev,
 795                                            struct flowi6 *fl6,
 796                                            struct ip_tunnel_info *info)
 797 {
 798         struct geneve_dev *geneve = netdev_priv(dev);
 799         struct geneve_sock *gs6 = geneve->sock6;
 800         struct dst_entry *dst = NULL;
 801         __u8 prio;
 802
 803         memset(fl6, 0, sizeof(*fl6));
 804         fl6->flowi6_mark = skb->mark;
 805         fl6->flowi6_proto = IPPROTO_UDP;
 806
 807         if (info) {
 808                 fl6->daddr = info->key.u.ipv6.dst;
 809                 fl6->saddr = info->key.u.ipv6.src;
 810                 fl6->flowi6_tos = RT_TOS(info->key.tos);
 811         } else {
 812                 prio = geneve->tos;
 813                 if (prio == 1) {
 814                         const struct iphdr *iip = ip_hdr(skb);
 815
 816                         prio = ip_tunnel_get_dsfield(iip, skb);
 817                 }
 818
 819                 fl6->flowi6_tos = RT_TOS(prio);
 820                 fl6->daddr = geneve->remote.sin6.sin6_addr;
 821         }
 822
 823         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 824                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 825                 return ERR_PTR(-ENETUNREACH);
 826         }
 827         if (dst->dev == dev) { /* is this necessary? */
 828                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 829                 dst_release(dst);
 830                 return ERR_PTR(-ELOOP);
 831         }
 832
 833         return dst;
 834 }
 835 #endif
 836
 837 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 838 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 839 {
 840 #ifdef __BIG_ENDIAN
 841         vni[0] = (__force __u8)(tun_id >> 16);
 842         vni[1] = (__force __u8)(tun_id >> 8);
 843         vni[2] = (__force __u8)tun_id;
 844 #else
 845         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 846         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 847         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 848 #endif
 849 }
 850
 851 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 852                                    struct ip_tunnel_info *info)
 853 {
 854         struct geneve_dev *geneve = netdev_priv(dev);
 855         struct geneve_sock *gs4 = geneve->sock4;
 856         struct rtable *rt = NULL;
 857         const struct iphdr *iip; /* interior IP header */
 858         int err = -EINVAL;
 859         struct flowi4 fl4;
 860         __u8 tos, ttl;
 861         __be16 sport;
 862         __be16 df;
 863         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 864         u32 flags = geneve->flags;
 865
 866         if (geneve->collect_md) {
 867                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 868                         netdev_dbg(dev, "no tunnel metadata\n");
 869                         goto tx_error;
 870                 }
 871                 if (info && ip_tunnel_info_af(info) != AF_INET)
 872                         goto tx_error;
 873         }
 874
 875         rt = geneve_get_v4_rt(skb, dev, &fl4, info);
 876         if (IS_ERR(rt)) {
 877                 err = PTR_ERR(rt);
 878                 goto tx_error;
 879         }
 880
 881         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 882         skb_reset_mac_header(skb);
 883
 884         iip = ip_hdr(skb);
 885
 886         if (info) {
 887                 const struct ip_tunnel_key *key = &info->key;
 888                 u8 *opts = NULL;
 889                 u8 vni[3];
 890
 891                 tunnel_id_to_vni(key->tun_id, vni);
 892                 if (key->tun_flags & TUNNEL_GENEVE_OPT)
 893                         opts = ip_tunnel_info_opts(info);
 894
 895                 if (key->tun_flags & TUNNEL_CSUM)
 896                         flags |= GENEVE_F_UDP_CSUM;
 897                 else
 898                         flags &= ~GENEVE_F_UDP_CSUM;
 899
 900                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 901                                        info->options_len, opts, flags, xnet);
 902                 if (unlikely(err))
 903                         goto err;
 904
 905                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
 906                 ttl = key->ttl;
 907                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 908         } else {
 909                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 910                                        0, NULL, flags, xnet);
 911                 if (unlikely(err))
 912                         goto err;
 913
 914                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 915                 ttl = geneve->ttl;
 916                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 917                         ttl = 1;
 918                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 919                 df = 0;
 920         }
 921         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 922                             tos, ttl, df, sport, geneve->dst_port,
 923                             !net_eq(geneve->net, dev_net(geneve->dev)),
 924                             !(flags & GENEVE_F_UDP_CSUM));
 925
 926         return NETDEV_TX_OK;
 927
 928 tx_error:
 929         dev_kfree_skb(skb);
 930 err:
 931         if (err == -ELOOP)
 932                 dev->stats.collisions++;
 933         else if (err == -ENETUNREACH)
 934                 dev->stats.tx_carrier_errors++;
 935         else
 936                 dev->stats.tx_errors++;
 937         return NETDEV_TX_OK;
 938 }
 939
 940 #if IS_ENABLED(CONFIG_IPV6)
 941 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 942                                     struct ip_tunnel_info *info)
 943 {
 944         struct geneve_dev *geneve = netdev_priv(dev);
 945         struct geneve_sock *gs6 = geneve->sock6;
 946         struct dst_entry *dst = NULL;
 947         const struct iphdr *iip; /* interior IP header */
 948         int err = -EINVAL;
 949         struct flowi6 fl6;
 950         __u8 prio, ttl;
 951         __be16 sport;
 952         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 953         u32 flags = geneve->flags;
 954
 955         if (geneve->collect_md) {
 956                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 957                         netdev_dbg(dev, "no tunnel metadata\n");
 958                         goto tx_error;
 959                 }
 960         }
 961
 962         dst = geneve_get_v6_dst(skb, dev, &fl6, info);
 963         if (IS_ERR(dst)) {
 964                 err = PTR_ERR(dst);
 965                 goto tx_error;
 966         }
 967
 968         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 969         skb_reset_mac_header(skb);
 970
 971         iip = ip_hdr(skb);
 972
 973         if (info) {
 974                 const struct ip_tunnel_key *key = &info->key;
 975                 u8 *opts = NULL;
 976                 u8 vni[3];
 977
 978                 tunnel_id_to_vni(key->tun_id, vni);
 979                 if (key->tun_flags & TUNNEL_GENEVE_OPT)
 980                         opts = ip_tunnel_info_opts(info);
 981
 982                 if (key->tun_flags & TUNNEL_CSUM)
 983                         flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
 984                 else
 985                         flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
 986
 987                 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
 988                                         info->options_len, opts,
 989                                         flags, xnet);
 990                 if (unlikely(err))
 991                         goto err;
 992
 993                 prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
 994                 ttl = key->ttl;
 995         } else {
 996                 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
 997                                         0, NULL, flags, xnet);
 998                 if (unlikely(err))
 999                         goto err;
1000
1001                 prio = ip_tunnel_ecn_encap(fl6.flowi6_tos, iip, skb);
1002                 ttl = geneve->ttl;
1003                 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1004                         ttl = 1;
1005                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1006         }
1007         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1008                              &fl6.saddr, &fl6.daddr, prio, ttl,
1009                              sport, geneve->dst_port,
1010                              !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1011         return NETDEV_TX_OK;
1012
1013 tx_error:
1014         dev_kfree_skb(skb);
1015 err:
1016         if (err == -ELOOP)
1017                 dev->stats.collisions++;
1018         else if (err == -ENETUNREACH)
1019                 dev->stats.tx_carrier_errors++;
1020         else
1021                 dev->stats.tx_errors++;
1022         return NETDEV_TX_OK;
1023 }
1024 #endif
1025
1026 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1027 {
1028         struct geneve_dev *geneve = netdev_priv(dev);
1029         struct ip_tunnel_info *info = NULL;
1030
1031         if (geneve->collect_md)
1032                 info = skb_tunnel_info(skb);
1033
1034 #if IS_ENABLED(CONFIG_IPV6)
1035         if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1036             (!info && geneve->remote.sa.sa_family == AF_INET6))
1037                 return geneve6_xmit_skb(skb, dev, info);
1038 #endif
1039         return geneve_xmit_skb(skb, dev, info);
1040 }
1041
1042 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1043 {
1044         /* The max_mtu calculation does not take account of GENEVE
1045          * options, to avoid excluding potentially valid
1046          * configurations.
1047          */
1048         int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - sizeof(struct iphdr)
1049                 - dev->hard_header_len;
1050
1051         if (new_mtu < 68)
1052                 return -EINVAL;
1053
1054         if (new_mtu > max_mtu) {
1055                 if (strict)
1056                         return -EINVAL;
1057
1058                 new_mtu = max_mtu;
1059         }
1060
1061         dev->mtu = new_mtu;
1062         return 0;
1063 }
1064
1065 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1066 {
1067         return __geneve_change_mtu(dev, new_mtu, true);
1068 }
1069
1070 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1071 {
1072         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1073         struct geneve_dev *geneve = netdev_priv(dev);
1074         struct rtable *rt;
1075         struct flowi4 fl4;
1076 #if IS_ENABLED(CONFIG_IPV6)
1077         struct dst_entry *dst;
1078         struct flowi6 fl6;
1079 #endif
1080
1081         if (ip_tunnel_info_af(info) == AF_INET) {
1082                 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1083                 if (IS_ERR(rt))
1084                         return PTR_ERR(rt);
1085
1086                 ip_rt_put(rt);
1087                 info->key.u.ipv4.src = fl4.saddr;
1088 #if IS_ENABLED(CONFIG_IPV6)
1089         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1090                 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1091                 if (IS_ERR(dst))
1092                         return PTR_ERR(dst);
1093
1094                 dst_release(dst);
1095                 info->key.u.ipv6.src = fl6.saddr;
1096 #endif
1097         } else {
1098                 return -EINVAL;
1099         }
1100
1101         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1102                                              1, USHRT_MAX, true);
1103         info->key.tp_dst = geneve->dst_port;
1104         return 0;
1105 }
1106
1107 static const struct net_device_ops geneve_netdev_ops = {
1108         .ndo_init               = geneve_init,
1109         .ndo_uninit             = geneve_uninit,
1110         .ndo_open               = geneve_open,
1111         .ndo_stop               = geneve_stop,
1112         .ndo_start_xmit         = geneve_xmit,
1113         .ndo_get_stats64        = ip_tunnel_get_stats64,
1114         .ndo_change_mtu         = geneve_change_mtu,
1115         .ndo_validate_addr      = eth_validate_addr,
1116         .ndo_set_mac_address    = eth_mac_addr,
1117         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1118 };
1119
1120 static void geneve_get_drvinfo(struct net_device *dev,
1121                                struct ethtool_drvinfo *drvinfo)
1122 {
1123         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1124         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1125 }
1126
1127 static const struct ethtool_ops geneve_ethtool_ops = {
1128         .get_drvinfo    = geneve_get_drvinfo,
1129         .get_link       = ethtool_op_get_link,
1130 };
1131
1132 /* Info for udev, that this is a virtual tunnel endpoint */
1133 static struct device_type geneve_type = {
1134         .name = "geneve",
1135 };
1136
1137 /* Calls the ndo_add_geneve_port of the caller in order to
1138  * supply the listening GENEVE udp ports. Callers are expected
1139  * to implement the ndo_add_geneve_port.
1140  */
1141 void geneve_get_rx_port(struct net_device *dev)
1142 {
1143         struct net *net = dev_net(dev);
1144         struct geneve_net *gn = net_generic(net, geneve_net_id);
1145         struct geneve_sock *gs;
1146         sa_family_t sa_family;
1147         struct sock *sk;
1148         __be16 port;
1149
1150         rcu_read_lock();
1151         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1152                 sk = gs->sock->sk;
1153                 sa_family = sk->sk_family;
1154                 port = inet_sk(sk)->inet_sport;
1155                 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family, port);
1156         }
1157         rcu_read_unlock();
1158 }
1159 EXPORT_SYMBOL_GPL(geneve_get_rx_port);
1160
1161 /* Initialize the device structure. */
1162 static void geneve_setup(struct net_device *dev)
1163 {
1164         ether_setup(dev);
1165
1166         dev->netdev_ops = &geneve_netdev_ops;
1167         dev->ethtool_ops = &geneve_ethtool_ops;
1168         dev->destructor = free_netdev;
1169
1170         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1171
1172         dev->features    |= NETIF_F_LLTX;
1173         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1174         dev->features    |= NETIF_F_RXCSUM;
1175         dev->features    |= NETIF_F_GSO_SOFTWARE;
1176
1177         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1178         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1179
1180         netif_keep_dst(dev);
1181         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1182         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1183         eth_hw_addr_random(dev);
1184 }
1185
1186 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1187         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1188         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1189         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1190         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1191         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1192         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1193         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1194         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1195         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1196         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1197 };
1198
1199 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1200 {
1201         if (tb[IFLA_ADDRESS]) {
1202                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1203                         return -EINVAL;
1204
1205                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1206                         return -EADDRNOTAVAIL;
1207         }
1208
1209         if (!data)
1210                 return -EINVAL;
1211
1212         if (data[IFLA_GENEVE_ID]) {
1213                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1214
1215                 if (vni >= GENEVE_VID_MASK)
1216                         return -ERANGE;
1217         }
1218
1219         return 0;
1220 }
1221
1222 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1223                                           __be16 dst_port,
1224                                           union geneve_addr *remote,
1225                                           u8 vni[],
1226                                           bool *tun_on_same_port,
1227                                           bool *tun_collect_md)
1228 {
1229         struct geneve_dev *geneve, *t;
1230
1231         *tun_on_same_port = false;
1232         *tun_collect_md = false;
1233         t = NULL;
1234         list_for_each_entry(geneve, &gn->geneve_list, next) {
1235                 if (geneve->dst_port == dst_port) {
1236                         *tun_collect_md = geneve->collect_md;
1237                         *tun_on_same_port = true;
1238                 }
1239                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1240                     !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1241                     dst_port == geneve->dst_port)
1242                         t = geneve;
1243         }
1244         return t;
1245 }
1246
1247 static int geneve_configure(struct net *net, struct net_device *dev,
1248                             union geneve_addr *remote,
1249                             __u32 vni, __u8 ttl, __u8 tos, __be16 dst_port,
1250                             bool metadata, u32 flags)
1251 {
1252         struct geneve_net *gn = net_generic(net, geneve_net_id);
1253         struct geneve_dev *t, *geneve = netdev_priv(dev);
1254         bool tun_collect_md, tun_on_same_port;
1255         int err, encap_len;
1256
1257         if (!remote)
1258                 return -EINVAL;
1259         if (metadata &&
1260             (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl))
1261                 return -EINVAL;
1262
1263         geneve->net = net;
1264         geneve->dev = dev;
1265
1266         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1267         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1268         geneve->vni[2] =  vni & 0x000000ff;
1269
1270         if ((remote->sa.sa_family == AF_INET &&
1271              IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1272             (remote->sa.sa_family == AF_INET6 &&
1273              ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1274                 return -EINVAL;
1275         geneve->remote = *remote;
1276
1277         geneve->ttl = ttl;
1278         geneve->tos = tos;
1279         geneve->dst_port = dst_port;
1280         geneve->collect_md = metadata;
1281         geneve->flags = flags;
1282
1283         t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1284                             &tun_on_same_port, &tun_collect_md);
1285         if (t)
1286                 return -EBUSY;
1287
1288         /* make enough headroom for basic scenario */
1289         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1290         if (remote->sa.sa_family == AF_INET)
1291                 encap_len += sizeof(struct iphdr);
1292         else
1293                 encap_len += sizeof(struct ipv6hdr);
1294         dev->needed_headroom = encap_len + ETH_HLEN;
1295
1296         if (metadata) {
1297                 if (tun_on_same_port)
1298                         return -EPERM;
1299         } else {
1300                 if (tun_collect_md)
1301                         return -EPERM;
1302         }
1303
1304         err = register_netdevice(dev);
1305         if (err)
1306                 return err;
1307
1308         list_add(&geneve->next, &gn->geneve_list);
1309         return 0;
1310 }
1311
1312 static int geneve_newlink(struct net *net, struct net_device *dev,
1313                           struct nlattr *tb[], struct nlattr *data[])
1314 {
1315         __be16 dst_port = htons(GENEVE_UDP_PORT);
1316         __u8 ttl = 0, tos = 0;
1317         bool metadata = false;
1318         union geneve_addr remote = geneve_remote_unspec;
1319         __u32 vni = 0;
1320         u32 flags = 0;
1321
1322         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1323                 return -EINVAL;
1324
1325         if (data[IFLA_GENEVE_REMOTE]) {
1326                 remote.sa.sa_family = AF_INET;
1327                 remote.sin.sin_addr.s_addr =
1328                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1329         }
1330
1331         if (data[IFLA_GENEVE_REMOTE6]) {
1332                 if (!IS_ENABLED(CONFIG_IPV6))
1333                         return -EPFNOSUPPORT;
1334
1335                 remote.sa.sa_family = AF_INET6;
1336                 remote.sin6.sin6_addr =
1337                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1338
1339                 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1340                     IPV6_ADDR_LINKLOCAL) {
1341                         netdev_dbg(dev, "link-local remote is unsupported\n");
1342                         return -EINVAL;
1343                 }
1344         }
1345
1346         if (data[IFLA_GENEVE_ID])
1347                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1348
1349         if (data[IFLA_GENEVE_TTL])
1350                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1351
1352         if (data[IFLA_GENEVE_TOS])
1353                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1354
1355         if (data[IFLA_GENEVE_PORT])
1356                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1357
1358         if (data[IFLA_GENEVE_COLLECT_METADATA])
1359                 metadata = true;
1360
1361         if (data[IFLA_GENEVE_UDP_CSUM] &&
1362             nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1363                 flags |= GENEVE_F_UDP_CSUM;
1364
1365         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1366             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1367                 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1368
1369         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1370             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1371                 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1372
1373         return geneve_configure(net, dev, &remote, vni, ttl, tos, dst_port,
1374                                 metadata, flags);
1375 }
1376
1377 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1378 {
1379         struct geneve_dev *geneve = netdev_priv(dev);
1380
1381         list_del(&geneve->next);
1382         unregister_netdevice_queue(dev, head);
1383 }
1384
1385 static size_t geneve_get_size(const struct net_device *dev)
1386 {
1387         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1388                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1389                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1390                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1391                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1392                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1393                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1394                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1395                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1396                 0;
1397 }
1398
1399 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1400 {
1401         struct geneve_dev *geneve = netdev_priv(dev);
1402         __u32 vni;
1403
1404         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1405         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1406                 goto nla_put_failure;
1407
1408         if (geneve->remote.sa.sa_family == AF_INET) {
1409                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1410                                     geneve->remote.sin.sin_addr.s_addr))
1411                         goto nla_put_failure;
1412 #if IS_ENABLED(CONFIG_IPV6)
1413         } else {
1414                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1415                                      &geneve->remote.sin6.sin6_addr))
1416                         goto nla_put_failure;
1417 #endif
1418         }
1419
1420         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1421             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos))
1422                 goto nla_put_failure;
1423
1424         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1425                 goto nla_put_failure;
1426
1427         if (geneve->collect_md) {
1428                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1429                         goto nla_put_failure;
1430         }
1431
1432         if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1433                        !!(geneve->flags & GENEVE_F_UDP_CSUM)) ||
1434             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1435                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1436             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1437                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1438                 goto nla_put_failure;
1439
1440         return 0;
1441
1442 nla_put_failure:
1443         return -EMSGSIZE;
1444 }
1445
1446 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1447         .kind           = "geneve",
1448         .maxtype        = IFLA_GENEVE_MAX,
1449         .policy         = geneve_policy,
1450         .priv_size      = sizeof(struct geneve_dev),
1451         .setup          = geneve_setup,
1452         .validate       = geneve_validate,
1453         .newlink        = geneve_newlink,
1454         .dellink        = geneve_dellink,
1455         .get_size       = geneve_get_size,
1456         .fill_info      = geneve_fill_info,
1457 };
1458
1459 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1460                                         u8 name_assign_type, u16 dst_port)
1461 {
1462         struct nlattr *tb[IFLA_MAX + 1];
1463         struct net_device *dev;
1464         int err;
1465
1466         memset(tb, 0, sizeof(tb));
1467         dev = rtnl_create_link(net, name, name_assign_type,
1468                                &geneve_link_ops, tb);
1469         if (IS_ERR(dev))
1470                 return dev;
1471
1472         err = geneve_configure(net, dev, &geneve_remote_unspec,
1473                                0, 0, 0, htons(dst_port), true,
1474                                GENEVE_F_UDP_ZERO_CSUM6_RX);
1475         if (err)
1476                 goto err;
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, false);
1482         if (err)
1483                 goto err;
1484
1485         return dev;
1486
1487  err:
1488         free_netdev(dev);
1489         return ERR_PTR(err);
1490 }
1491 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1492
1493 static __net_init int geneve_init_net(struct net *net)
1494 {
1495         struct geneve_net *gn = net_generic(net, geneve_net_id);
1496
1497         INIT_LIST_HEAD(&gn->geneve_list);
1498         INIT_LIST_HEAD(&gn->sock_list);
1499         return 0;
1500 }
1501
1502 static void __net_exit geneve_exit_net(struct net *net)
1503 {
1504         struct geneve_net *gn = net_generic(net, geneve_net_id);
1505         struct geneve_dev *geneve, *next;
1506         struct net_device *dev, *aux;
1507         LIST_HEAD(list);
1508
1509         rtnl_lock();
1510
1511         /* gather any geneve devices that were moved into this ns */
1512         for_each_netdev_safe(net, dev, aux)
1513                 if (dev->rtnl_link_ops == &geneve_link_ops)
1514                         unregister_netdevice_queue(dev, &list);
1515
1516         /* now gather any other geneve devices that were created in this ns */
1517         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1518                 /* If geneve->dev is in the same netns, it was already added
1519                  * to the list by the previous loop.
1520                  */
1521                 if (!net_eq(dev_net(geneve->dev), net))
1522                         unregister_netdevice_queue(geneve->dev, &list);
1523         }
1524
1525         /* unregister the devices gathered above */
1526         unregister_netdevice_many(&list);
1527         rtnl_unlock();
1528 }
1529
1530 static struct pernet_operations geneve_net_ops = {
1531         .init = geneve_init_net,
1532         .exit = geneve_exit_net,
1533         .id   = &geneve_net_id,
1534         .size = sizeof(struct geneve_net),
1535 };
1536
1537 static int __init geneve_init_module(void)
1538 {
1539         int rc;
1540
1541         rc = register_pernet_subsys(&geneve_net_ops);
1542         if (rc)
1543                 goto out1;
1544
1545         rc = rtnl_link_register(&geneve_link_ops);
1546         if (rc)
1547                 goto out2;
1548
1549         return 0;
1550 out2:
1551         unregister_pernet_subsys(&geneve_net_ops);
1552 out1:
1553         return rc;
1554 }
1555 late_initcall(geneve_init_module);
1556
1557 static void __exit geneve_cleanup_module(void)
1558 {
1559         rtnl_link_unregister(&geneve_link_ops);
1560         unregister_pernet_subsys(&geneve_net_ops);
1561 }
1562 module_exit(geneve_cleanup_module);
1563
1564 MODULE_LICENSE("GPL");
1565 MODULE_VERSION(GENEVE_NETDEV_VER);
1566 MODULE_AUTHOR("John W. Linville <[email protected]>");
1567 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1568 MODULE_ALIAS_RTNL_LINK("geneve");