datapath/linux/compat/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 <linux/if_link.h>
  19 #include <linux/if_vlan.h>
  20
  21 #include <net/addrconf.h>
  22 #include <net/dst_cache.h>
  23 #include <net/dst_metadata.h>
  24 #include <net/net_namespace.h>
  25 #include <net/netns/generic.h>
  26 #include <net/rtnetlink.h>
  27 #include <net/geneve.h>
  28 #include <net/protocol.h>
  29 #include <net/udp_tunnel.h>
  30 #include <net/ip6_route.h>
  31 #if IS_ENABLED(CONFIG_IPV6)
  32 #include <net/ipv6.h>
  33 #include <net/addrconf.h>
  34 #include <net/ip6_tunnel.h>
  35 #include <net/ip6_checksum.h>
  36 #endif
  37
  38
  39 #include "gso.h"
  40 #include "vport-netdev.h"
  41 #include "compat.h"
  42
  43 #ifndef USE_UPSTREAM_TUNNEL
  44
  45 #define GENEVE_NETDEV_VER       "0.6"
  46
  47 #define GENEVE_UDP_PORT         6081
  48
  49 #define GENEVE_N_VID            (1u << 24)
  50 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  51
  52 #define VNI_HASH_BITS           10
  53 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  54
  55 #define GENEVE_VER 0
  56 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  57
  58 /* per-network namespace private data for this module */
  59 struct geneve_net {
  60         struct list_head        geneve_list;
  61         struct list_head        sock_list;
  62 };
  63
  64 static int geneve_net_id;
  65
  66 union geneve_addr {
  67         struct sockaddr_in sin;
  68         struct sockaddr_in6 sin6;
  69         struct sockaddr sa;
  70 };
  71
  72 static union geneve_addr geneve_remote_unspec = { .sa.sa_family = AF_UNSPEC, };
  73
  74 /* Pseudo network device */
  75 struct geneve_dev {
  76         struct hlist_node  hlist;       /* vni hash table */
  77         struct net         *net;        /* netns for packet i/o */
  78         struct net_device  *dev;        /* netdev for geneve tunnel */
  79         struct geneve_sock *sock4;      /* IPv4 socket used for geneve tunnel */
  80 #if IS_ENABLED(CONFIG_IPV6)
  81         struct geneve_sock *sock6;      /* IPv6 socket used for geneve tunnel */
  82 #endif
  83         u8                 vni[3];      /* virtual network ID for tunnel */
  84         u8                 ttl;         /* TTL override */
  85         u8                 tos;         /* TOS override */
  86         union geneve_addr  remote;      /* IP address for link partner */
  87         struct list_head   next;        /* geneve's per namespace list */
  88         __be32             label;       /* IPv6 flowlabel override */
  89         __be16             dst_port;
  90         bool               collect_md;
  91         u32                flags;
  92         struct dst_cache   dst_cache;
  93 };
  94
  95 /* Geneve device flags */
  96 #define GENEVE_F_UDP_ZERO_CSUM_TX       BIT(0)
  97 #define GENEVE_F_UDP_ZERO_CSUM6_TX      BIT(1)
  98 #define GENEVE_F_UDP_ZERO_CSUM6_RX      BIT(2)
  99
 100 struct geneve_sock {
 101         bool                    collect_md;
 102         struct list_head        list;
 103         struct socket           *sock;
 104         struct rcu_head         rcu;
 105         int                     refcnt;
 106         struct hlist_head       vni_list[VNI_HASH_SIZE];
 107         u32                     flags;
 108 #ifdef HAVE_UDP_OFFLOAD
 109         struct udp_offload      udp_offloads;
 110 #endif
 111 };
 112
 113 static inline __u32 geneve_net_vni_hash(u8 vni[3])
 114 {
 115         __u32 vnid;
 116
 117         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
 118         return hash_32(vnid, VNI_HASH_BITS);
 119 }
 120
 121 static __be64 vni_to_tunnel_id(const __u8 *vni)
 122 {
 123 #ifdef __BIG_ENDIAN
 124         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
 125 #else
 126         return (__force __be64)(((__force u64)vni[0] << 40) |
 127                                 ((__force u64)vni[1] << 48) |
 128                                 ((__force u64)vni[2] << 56));
 129 #endif
 130 }
 131
 132 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 133 {
 134         return gs->sock->sk->sk_family;
 135 }
 136
 137 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 138                                         __be32 addr, u8 vni[])
 139 {
 140         struct hlist_head *vni_list_head;
 141         struct geneve_dev *geneve;
 142         __u32 hash;
 143
 144         /* Find the device for this VNI */
 145         hash = geneve_net_vni_hash(vni);
 146         vni_list_head = &gs->vni_list[hash];
 147         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 148                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 149                     addr == geneve->remote.sin.sin_addr.s_addr)
 150                         return geneve;
 151         }
 152         return NULL;
 153 }
 154
 155 #if IS_ENABLED(CONFIG_IPV6)
 156 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 157                                          struct in6_addr addr6, u8 vni[])
 158 {
 159         struct hlist_head *vni_list_head;
 160         struct geneve_dev *geneve;
 161         __u32 hash;
 162
 163         /* Find the device for this VNI */
 164         hash = geneve_net_vni_hash(vni);
 165         vni_list_head = &gs->vni_list[hash];
 166         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 167                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 168                     ipv6_addr_equal(&addr6, &geneve->remote.sin6.sin6_addr))
 169                         return geneve;
 170         }
 171         return NULL;
 172 }
 173 #endif
 174
 175 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 176 {
 177         return (struct genevehdr *)(udp_hdr(skb) + 1);
 178 }
 179
 180 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 181                                             struct sk_buff *skb)
 182 {
 183         u8 *vni;
 184         __be32 addr;
 185         static u8 zero_vni[3];
 186 #if IS_ENABLED(CONFIG_IPV6)
 187         static struct in6_addr zero_addr6;
 188 #endif
 189
 190         if (geneve_get_sk_family(gs) == AF_INET) {
 191                 struct iphdr *iph;
 192
 193                 iph = ip_hdr(skb); /* outer IP header... */
 194
 195                 if (gs->collect_md) {
 196                         vni = zero_vni;
 197                         addr = 0;
 198                 } else {
 199                         vni = geneve_hdr(skb)->vni;
 200                         addr = iph->saddr;
 201                 }
 202
 203                 return geneve_lookup(gs, addr, vni);
 204 #if IS_ENABLED(CONFIG_IPV6)
 205         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 206                 struct ipv6hdr *ip6h;
 207                 struct in6_addr addr6;
 208
 209                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 210
 211                 if (gs->collect_md) {
 212                         vni = zero_vni;
 213                         addr6 = zero_addr6;
 214                 } else {
 215                         vni = geneve_hdr(skb)->vni;
 216                         addr6 = ip6h->saddr;
 217                 }
 218
 219                 return geneve6_lookup(gs, addr6, vni);
 220 #endif
 221         }
 222         return NULL;
 223 }
 224
 225 /* geneve receive/decap routine */
 226 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 227                       struct sk_buff *skb)
 228 {
 229         struct genevehdr *gnvh = geneve_hdr(skb);
 230         struct metadata_dst *tun_dst = NULL;
 231         struct pcpu_sw_netstats *stats;
 232         int err = 0;
 233         void *oiph;
 234         union {
 235                 struct metadata_dst dst;
 236                 char buf[sizeof(struct metadata_dst) + 256];
 237         } buf;
 238
 239         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 240                 __be16 flags;
 241
 242                 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
 243                         (gnvh->oam ? TUNNEL_OAM : 0) |
 244                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 245
 246                 tun_dst = &buf.dst;
 247                 ovs_udp_tun_rx_dst(tun_dst,
 248                                    skb, geneve_get_sk_family(gs), flags,
 249                                    vni_to_tunnel_id(gnvh->vni),
 250                                    gnvh->opt_len * 4);
 251                 if (!tun_dst)
 252                         goto drop;
 253                 /* Update tunnel dst according to Geneve options. */
 254                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 255                                         gnvh->options, gnvh->opt_len * 4);
 256         } else {
 257                 /* Drop packets w/ critical options,
 258                  * since we don't support any...
 259                  */
 260                 if (gnvh->critical)
 261                         goto drop;
 262         }
 263
 264         skb_reset_mac_header(skb);
 265         skb->protocol = eth_type_trans(skb, geneve->dev);
 266         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 267
 268         if (tun_dst)
 269                 ovs_skb_dst_set(skb, &tun_dst->dst);
 270
 271         /* Ignore packet loops (and multicast echo) */
 272         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
 273                 goto drop;
 274
 275         oiph = skb_network_header(skb);
 276         skb_reset_network_header(skb);
 277
 278         if (geneve_get_sk_family(gs) == AF_INET)
 279                 err = IP_ECN_decapsulate(oiph, skb);
 280 #if IS_ENABLED(CONFIG_IPV6)
 281         else
 282                 err = IP6_ECN_decapsulate(oiph, skb);
 283 #endif
 284         if (unlikely(err > 1)) {
 285                 ++geneve->dev->stats.rx_frame_errors;
 286                 ++geneve->dev->stats.rx_errors;
 287                 goto drop;
 288         }
 289
 290         stats = this_cpu_ptr(geneve->dev->tstats);
 291         u64_stats_update_begin(&stats->syncp);
 292         stats->rx_packets++;
 293         stats->rx_bytes += skb->len;
 294         u64_stats_update_end(&stats->syncp);
 295
 296         netdev_port_receive(skb, skb_tunnel_info(skb));
 297         return;
 298 drop:
 299         /* Consume bad packet */
 300         kfree_skb(skb);
 301 }
 302
 303 /* Setup stats when device is created */
 304 static int geneve_init(struct net_device *dev)
 305 {
 306         struct geneve_dev *geneve = netdev_priv(dev);
 307         int err;
 308
 309         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 310         if (!dev->tstats)
 311                 return -ENOMEM;
 312
 313         err = dst_cache_init(&geneve->dst_cache, GFP_KERNEL);
 314         if (err) {
 315                 free_percpu(dev->tstats);
 316                 return err;
 317         }
 318
 319         return 0;
 320 }
 321
 322 static void geneve_uninit(struct net_device *dev)
 323 {
 324         struct geneve_dev *geneve = netdev_priv(dev);
 325
 326         dst_cache_destroy(&geneve->dst_cache);
 327         free_percpu(dev->tstats);
 328 }
 329
 330 /* Callback from net/ipv4/udp.c to receive packets */
 331 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 332 {
 333         struct genevehdr *geneveh;
 334         struct geneve_dev *geneve;
 335         struct geneve_sock *gs;
 336         int opts_len;
 337
 338         /* Need Geneve and inner Ethernet header to be present */
 339         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 340                 goto drop;
 341
 342         /* Return packets with reserved bits set */
 343         geneveh = geneve_hdr(skb);
 344         if (unlikely(geneveh->ver != GENEVE_VER))
 345                 goto drop;
 346
 347         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 348                 goto drop;
 349
 350         gs = rcu_dereference_sk_user_data(sk);
 351         if (!gs)
 352                 goto drop;
 353
 354 #if IS_ENABLED(CONFIG_IPV6)
 355 #ifdef OVS_CHECK_UDP_TUNNEL_ZERO_CSUM
 356         if (geneve_get_sk_family(gs) == AF_INET6 &&
 357             !udp_hdr(skb)->check &&
 358             !(gs->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)) {
 359                 udp6_csum_zero_error(skb);
 360                 goto drop;
 361         }
 362 #endif
 363 #endif
 364         geneve = geneve_lookup_skb(gs, skb);
 365         if (!geneve)
 366                 goto drop;
 367
 368         opts_len = geneveh->opt_len * 4;
 369         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 370                                  htons(ETH_P_TEB),
 371                                  !net_eq(geneve->net, dev_net(geneve->dev))))
 372                 goto drop;
 373
 374         geneve_rx(geneve, gs, skb);
 375         return 0;
 376
 377 drop:
 378         /* Consume bad packet */
 379         kfree_skb(skb);
 380         return 0;
 381 }
 382
 383 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 384                                          __be16 port, u32 flags)
 385 {
 386         struct socket *sock;
 387         struct udp_port_cfg udp_conf;
 388         int err;
 389
 390         memset(&udp_conf, 0, sizeof(udp_conf));
 391
 392         if (ipv6) {
 393                 udp_conf.family = AF_INET6;
 394                 udp_conf.ipv6_v6only = 1;
 395                 udp_conf.use_udp6_rx_checksums =
 396                     !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX);
 397         } else {
 398                 udp_conf.family = AF_INET;
 399                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 400         }
 401
 402         udp_conf.local_udp_port = port;
 403
 404         /* Open UDP socket */
 405         err = udp_sock_create(net, &udp_conf, &sock);
 406         if (err < 0)
 407                 return ERR_PTR(err);
 408
 409         return sock;
 410 }
 411
 412 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
 413 {
 414 #ifdef HAVE_NDO_ADD_GENEVE_PORT
 415         struct net_device *dev;
 416         struct sock *sk = gs->sock->sk;
 417         struct net *net = sock_net(sk);
 418         sa_family_t sa_family = geneve_get_sk_family(gs);
 419         __be16 port = inet_sk(sk)->inet_sport;
 420
 421         rcu_read_lock();
 422         for_each_netdev_rcu(net, dev) {
 423                 if (dev->netdev_ops->ndo_add_geneve_port)
 424                         dev->netdev_ops->ndo_add_geneve_port(dev, sa_family,
 425                                                              port);
 426         }
 427         rcu_read_unlock();
 428 #else
 429 #ifdef HAVE_UDP_OFFLOAD
 430         struct sock *sk = gs->sock->sk;
 431         sa_family_t sa_family = sk->sk_family;
 432         int err;
 433
 434         if (sa_family == AF_INET) {
 435                 err = udp_add_offload(&gs->udp_offloads);
 436                 if (err)
 437                         pr_warn("geneve: udp_add_offload failed with status %d\n",
 438                                 err);
 439         }
 440 #endif
 441 #endif
 442
 443 }
 444
 445 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 446 {
 447 #ifdef HAVE_NDO_ADD_GENEVE_PORT
 448         struct net_device *dev;
 449         struct sock *sk = gs->sock->sk;
 450         struct net *net = sock_net(sk);
 451         sa_family_t sa_family = geneve_get_sk_family(gs);
 452         __be16 port = inet_sk(sk)->inet_sport;
 453
 454         rcu_read_lock();
 455         for_each_netdev_rcu(net, dev) {
 456                 if (dev->netdev_ops->ndo_del_geneve_port)
 457                         dev->netdev_ops->ndo_del_geneve_port(dev, sa_family,
 458                                                              port);
 459         }
 460
 461         rcu_read_unlock();
 462 #else
 463 #ifdef HAVE_UDP_OFFLOAD
 464         struct sock *sk = gs->sock->sk;
 465         sa_family_t sa_family = sk->sk_family;
 466
 467         if (sa_family == AF_INET)
 468                 udp_del_offload(&gs->udp_offloads);
 469 #endif
 470
 471 #endif
 472 }
 473
 474 #if defined(HAVE_UDP_OFFLOAD) || \
 475     defined(HAVE_UDP_TUNNEL_SOCK_CFG_GRO_RECEIVE)
 476
 477 static int geneve_hlen(struct genevehdr *gh)
 478 {
 479         return sizeof(*gh) + gh->opt_len * 4;
 480 }
 481
 482 #ifndef HAVE_UDP_OFFLOAD_ARG_UOFF
 483 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
 484                                            struct sk_buff *skb)
 485 #else
 486 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
 487                                            struct sk_buff *skb,
 488                                            struct udp_offload *uoff)
 489 #endif
 490 {
 491         struct sk_buff *p, **pp = NULL;
 492         struct genevehdr *gh, *gh2;
 493         unsigned int hlen, gh_len, off_gnv;
 494         const struct packet_offload *ptype;
 495         __be16 type;
 496         int flush = 1;
 497
 498         off_gnv = skb_gro_offset(skb);
 499         hlen = off_gnv + sizeof(*gh);
 500         gh = skb_gro_header_fast(skb, off_gnv);
 501         if (skb_gro_header_hard(skb, hlen)) {
 502                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 503                 if (unlikely(!gh))
 504                         goto out;
 505         }
 506
 507         if (gh->ver != GENEVE_VER || gh->oam)
 508                 goto out;
 509         gh_len = geneve_hlen(gh);
 510
 511         hlen = off_gnv + gh_len;
 512         if (skb_gro_header_hard(skb, hlen)) {
 513                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 514                 if (unlikely(!gh))
 515                         goto out;
 516         }
 517
 518         for (p = *head; p; p = p->next) {
 519                 if (!NAPI_GRO_CB(p)->same_flow)
 520                         continue;
 521
 522                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 523                 if (gh->opt_len != gh2->opt_len ||
 524                     memcmp(gh, gh2, gh_len)) {
 525                         NAPI_GRO_CB(p)->same_flow = 0;
 526                         continue;
 527                 }
 528         }
 529
 530         type = gh->proto_type;
 531
 532         rcu_read_lock();
 533         ptype = gro_find_receive_by_type(type);
 534         if (!ptype)
 535                 goto out_unlock;
 536
 537         skb_gro_pull(skb, gh_len);
 538         skb_gro_postpull_rcsum(skb, gh, gh_len);
 539         pp = ptype->callbacks.gro_receive(head, skb);
 540         flush = 0;
 541
 542 out_unlock:
 543         rcu_read_unlock();
 544 out:
 545         NAPI_GRO_CB(skb)->flush |= flush;
 546
 547         return pp;
 548 }
 549
 550 #ifndef HAVE_UDP_OFFLOAD_ARG_UOFF
 551 static int geneve_gro_complete(struct sk_buff *skb, int nhoff)
 552 #else
 553 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
 554                                struct udp_offload *uoff)
 555 #endif
 556 {
 557         struct genevehdr *gh;
 558         struct packet_offload *ptype;
 559         __be16 type;
 560         int gh_len;
 561         int err = -ENOSYS;
 562
 563         udp_tunnel_gro_complete(skb, nhoff);
 564
 565         gh = (struct genevehdr *)(skb->data + nhoff);
 566         gh_len = geneve_hlen(gh);
 567         type = gh->proto_type;
 568
 569         rcu_read_lock();
 570         ptype = gro_find_complete_by_type(type);
 571         if (ptype)
 572                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 573
 574         rcu_read_unlock();
 575
 576         skb_set_inner_mac_header(skb, nhoff + gh_len);
 577         return err;
 578 }
 579 #endif
 580
 581 /* Create new listen socket if needed */
 582 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 583                                                 bool ipv6, u32 flags)
 584 {
 585         struct geneve_net *gn = net_generic(net, geneve_net_id);
 586         struct geneve_sock *gs;
 587         struct socket *sock;
 588         struct udp_tunnel_sock_cfg tunnel_cfg;
 589         int h;
 590
 591         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 592         if (!gs)
 593                 return ERR_PTR(-ENOMEM);
 594
 595         sock = geneve_create_sock(net, ipv6, port, flags);
 596         if (IS_ERR(sock)) {
 597                 kfree(gs);
 598                 return ERR_CAST(sock);
 599         }
 600
 601         gs->sock = sock;
 602         gs->refcnt = 1;
 603         for (h = 0; h < VNI_HASH_SIZE; ++h)
 604                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 605
 606         /* Initialize the geneve udp offloads structure */
 607 #ifdef HAVE_UDP_OFFLOAD
 608         gs->udp_offloads.port = port;
 609         gs->udp_offloads.callbacks.gro_receive  = geneve_gro_receive;
 610         gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
 611 #endif
 612
 613         geneve_notify_add_rx_port(gs);
 614         /* Mark socket as an encapsulation socket */
 615         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 616         tunnel_cfg.sk_user_data = gs;
 617         tunnel_cfg.encap_type = 1;
 618 #ifdef HAVE_UDP_TUNNEL_SOCK_CFG_GRO_RECEIVE
 619         tunnel_cfg.gro_receive = geneve_gro_receive;
 620         tunnel_cfg.gro_complete = geneve_gro_complete;
 621 #endif
 622         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 623         tunnel_cfg.encap_destroy = NULL;
 624         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 625         list_add(&gs->list, &gn->sock_list);
 626         return gs;
 627 }
 628
 629 static void __geneve_sock_release(struct geneve_sock *gs)
 630 {
 631         if (!gs || --gs->refcnt)
 632                 return;
 633
 634         list_del(&gs->list);
 635         geneve_notify_del_rx_port(gs);
 636         udp_tunnel_sock_release(gs->sock);
 637         kfree_rcu(gs, rcu);
 638 }
 639
 640 static void geneve_sock_release(struct geneve_dev *geneve)
 641 {
 642         __geneve_sock_release(geneve->sock4);
 643 #if IS_ENABLED(CONFIG_IPV6)
 644         __geneve_sock_release(geneve->sock6);
 645 #endif
 646 }
 647
 648 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 649                                             sa_family_t family,
 650                                             __be16 dst_port)
 651 {
 652         struct geneve_sock *gs;
 653
 654         list_for_each_entry(gs, &gn->sock_list, list) {
 655                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 656                     geneve_get_sk_family(gs) == family) {
 657                         return gs;
 658                 }
 659         }
 660         return NULL;
 661 }
 662
 663 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 664 {
 665         struct net *net = geneve->net;
 666         struct geneve_net *gn = net_generic(net, geneve_net_id);
 667         struct geneve_sock *gs;
 668         __u32 hash;
 669
 670         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
 671         if (gs) {
 672                 gs->refcnt++;
 673                 goto out;
 674         }
 675
 676         gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
 677         if (IS_ERR(gs))
 678                 return PTR_ERR(gs);
 679
 680 out:
 681         gs->collect_md = geneve->collect_md;
 682         gs->flags = geneve->flags;
 683 #if IS_ENABLED(CONFIG_IPV6)
 684         if (ipv6)
 685                 geneve->sock6 = gs;
 686         else
 687 #endif
 688                 geneve->sock4 = gs;
 689
 690         hash = geneve_net_vni_hash(geneve->vni);
 691         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 692         return 0;
 693 }
 694
 695 static int geneve_open(struct net_device *dev)
 696 {
 697         struct geneve_dev *geneve = netdev_priv(dev);
 698         bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
 699         bool metadata = geneve->collect_md;
 700         int ret = 0;
 701
 702         geneve->sock4 = NULL;
 703 #if IS_ENABLED(CONFIG_IPV6)
 704         geneve->sock6 = NULL;
 705         if (ipv6 || metadata)
 706                 ret = geneve_sock_add(geneve, true);
 707 #endif
 708
 709         if (!ret && (!ipv6 || metadata))
 710                 ret = geneve_sock_add(geneve, false);
 711         if (ret < 0)
 712                 geneve_sock_release(geneve);
 713
 714         return ret;
 715 }
 716
 717 static int geneve_stop(struct net_device *dev)
 718 {
 719         struct geneve_dev *geneve = netdev_priv(dev);
 720
 721         if (!hlist_unhashed(&geneve->hlist))
 722                 hlist_del_rcu(&geneve->hlist);
 723         geneve_sock_release(geneve);
 724         return 0;
 725 }
 726
 727 static void geneve_build_header(struct genevehdr *geneveh,
 728                                 __be16 tun_flags, u8 vni[3],
 729                                 u8 options_len, u8 *options)
 730 {
 731         geneveh->ver = GENEVE_VER;
 732         geneveh->opt_len = options_len / 4;
 733         geneveh->oam = !!(tun_flags & TUNNEL_OAM);
 734         geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 735         geneveh->rsvd1 = 0;
 736         memcpy(geneveh->vni, vni, 3);
 737         geneveh->proto_type = htons(ETH_P_TEB);
 738         geneveh->rsvd2 = 0;
 739
 740         memcpy(geneveh->options, options, options_len);
 741 }
 742
 743 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 744                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 745                             u32 flags, bool xnet)
 746 {
 747         struct genevehdr *gnvh;
 748         int min_headroom;
 749         int err;
 750         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
 751
 752         skb_scrub_packet(skb, xnet);
 753
 754         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 755                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 756         err = skb_cow_head(skb, min_headroom);
 757         if (unlikely(err))
 758                 goto free_rt;
 759
 760         err = udp_tunnel_handle_offloads(skb, udp_sum);
 761         if (err)
 762                 goto free_rt;
 763
 764         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 765         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 766
 767         ovs_skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 768         return 0;
 769
 770 free_rt:
 771         ip_rt_put(rt);
 772         return err;
 773 }
 774
 775 #if IS_ENABLED(CONFIG_IPV6)
 776 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 777                              __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 778                              u32 flags, bool xnet)
 779 {
 780         struct genevehdr *gnvh;
 781         int min_headroom;
 782         int err;
 783         bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
 784
 785         skb_scrub_packet(skb, xnet);
 786
 787         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
 788                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
 789         err = skb_cow_head(skb, min_headroom);
 790         if (unlikely(err))
 791                 goto free_dst;
 792
 793         err = udp_tunnel_handle_offloads(skb, udp_sum);
 794         if (err)
 795                 goto free_dst;
 796
 797         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 798         geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
 799
 800         ovs_skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 801         return 0;
 802
 803 free_dst:
 804         dst_release(dst);
 805         return err;
 806 }
 807 #endif
 808
 809 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 810                                        struct net_device *dev,
 811                                        struct flowi4 *fl4,
 812                                        struct ip_tunnel_info *info)
 813 {
 814         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 815         struct geneve_dev *geneve = netdev_priv(dev);
 816         struct dst_cache *dst_cache;
 817         struct rtable *rt = NULL;
 818         __u8 tos;
 819
 820         memset(fl4, 0, sizeof(*fl4));
 821         fl4->flowi4_mark = skb->mark;
 822         fl4->flowi4_proto = IPPROTO_UDP;
 823
 824         if (info) {
 825                 fl4->daddr = info->key.u.ipv4.dst;
 826                 fl4->saddr = info->key.u.ipv4.src;
 827                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 828                 dst_cache = &info->dst_cache;
 829         } else {
 830                 tos = geneve->tos;
 831                 if (tos == 1) {
 832                         const struct iphdr *iip = ip_hdr(skb);
 833
 834                         tos = ip_tunnel_get_dsfield(iip, skb);
 835                         use_cache = false;
 836                 }
 837
 838                 fl4->flowi4_tos = RT_TOS(tos);
 839                 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
 840                 dst_cache = &geneve->dst_cache;
 841         }
 842
 843         if (use_cache) {
 844                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 845                 if (rt)
 846                         return rt;
 847         }
 848
 849         rt = ip_route_output_key(geneve->net, fl4);
 850         if (IS_ERR(rt)) {
 851                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 852                 return ERR_PTR(-ENETUNREACH);
 853         }
 854         if (rt->dst.dev == dev) { /* is this necessary? */
 855                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 856                 ip_rt_put(rt);
 857                 return ERR_PTR(-ELOOP);
 858         }
 859         if (use_cache)
 860                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 861         return rt;
 862 }
 863
 864 #if IS_ENABLED(CONFIG_IPV6)
 865 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 866                                            struct net_device *dev,
 867                                            struct flowi6 *fl6,
 868                                            struct ip_tunnel_info *info)
 869 {
 870         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 871         struct geneve_dev *geneve = netdev_priv(dev);
 872         struct geneve_sock *gs6 = geneve->sock6;
 873         struct dst_entry *dst = NULL;
 874         struct dst_cache *dst_cache;
 875         __u8 prio;
 876
 877         memset(fl6, 0, sizeof(*fl6));
 878         fl6->flowi6_mark = skb->mark;
 879         fl6->flowi6_proto = IPPROTO_UDP;
 880
 881         if (info) {
 882                 fl6->daddr = info->key.u.ipv6.dst;
 883                 fl6->saddr = info->key.u.ipv6.src;
 884                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(info->key.tos),
 885                                                    info->key.label);
 886                 dst_cache = &info->dst_cache;
 887         } else {
 888                 prio = geneve->tos;
 889                 if (prio == 1) {
 890                         const struct iphdr *iip = ip_hdr(skb);
 891
 892                         prio = ip_tunnel_get_dsfield(iip, skb);
 893                         use_cache = false;
 894                 }
 895
 896                 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 897                                                    geneve->label);
 898                 fl6->daddr = geneve->remote.sin6.sin6_addr;
 899                 dst_cache = &geneve->dst_cache;
 900         }
 901
 902         if (use_cache) {
 903                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 904                 if (dst)
 905                         return dst;
 906         }
 907
 908 #ifdef HAVE_IPV6_DST_LOOKUP_NET
 909         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 910 #else
 911 #ifdef HAVE_IPV6_STUB
 912         if (ipv6_stub->ipv6_dst_lookup(gs6->sock->sk, &dst, fl6)) {
 913 #else
 914         if (ip6_dst_lookup(gs6->sock->sk, &dst, fl6)) {
 915 #endif
 916 #endif
 917                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 918                 return ERR_PTR(-ENETUNREACH);
 919         }
 920         if (dst->dev == dev) { /* is this necessary? */
 921                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 922                 dst_release(dst);
 923                 return ERR_PTR(-ELOOP);
 924         }
 925
 926         if (use_cache)
 927                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 928         return dst;
 929 }
 930 #endif
 931
 932 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 933 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 934 {
 935 #ifdef __BIG_ENDIAN
 936         vni[0] = (__force __u8)(tun_id >> 16);
 937         vni[1] = (__force __u8)(tun_id >> 8);
 938         vni[2] = (__force __u8)tun_id;
 939 #else
 940         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 941         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 942         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 943 #endif
 944 }
 945
 946 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 947                                    struct ip_tunnel_info *info)
 948 {
 949         struct geneve_dev *geneve = netdev_priv(dev);
 950         struct geneve_sock *gs4 = geneve->sock4;
 951         struct rtable *rt = NULL;
 952         const struct iphdr *iip; /* interior IP header */
 953         int err = -EINVAL;
 954         struct flowi4 fl4;
 955         __u8 tos, ttl;
 956         __be16 sport;
 957         __be16 df;
 958         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 959         u32 flags = geneve->flags;
 960
 961         if (geneve->collect_md) {
 962                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 963                         netdev_dbg(dev, "no tunnel metadata\n");
 964                         goto tx_error;
 965                 }
 966                 if (info && ip_tunnel_info_af(info) != AF_INET)
 967                         goto tx_error;
 968         }
 969
 970         rt = geneve_get_v4_rt(skb, dev, &fl4, info);
 971         if (IS_ERR(rt)) {
 972                 err = PTR_ERR(rt);
 973                 goto tx_error;
 974         }
 975
 976         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 977         skb_reset_mac_header(skb);
 978
 979         iip = ip_hdr(skb);
 980
 981         if (info) {
 982                 const struct ip_tunnel_key *key = &info->key;
 983                 u8 *opts = NULL;
 984                 u8 vni[3];
 985
 986                 tunnel_id_to_vni(key->tun_id, vni);
 987                 if (info->options_len)
 988                         opts = ip_tunnel_info_opts(info);
 989
 990                 if (key->tun_flags & TUNNEL_CSUM)
 991                         flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
 992                 else
 993                         flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
 994
 995                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 996                                        info->options_len, opts, flags, xnet);
 997                 if (unlikely(err))
 998                         goto tx_error;
 999
1000                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
1001                 ttl = key->ttl;
1002                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
1003         } else {
1004                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
1005                                        0, NULL, flags, xnet);
1006                 if (unlikely(err))
1007                         goto tx_error;
1008
1009                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
1010                 ttl = geneve->ttl;
1011                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
1012                         ttl = 1;
1013                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1014                 df = 0;
1015         }
1016         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
1017                             tos, ttl, df, sport, geneve->dst_port,
1018                             !net_eq(geneve->net, dev_net(geneve->dev)),
1019                             !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
1020
1021         return NETDEV_TX_OK;
1022
1023 tx_error:
1024         dev_kfree_skb(skb);
1025
1026         if (err == -ELOOP)
1027                 dev->stats.collisions++;
1028         else if (err == -ENETUNREACH)
1029                 dev->stats.tx_carrier_errors++;
1030
1031         dev->stats.tx_errors++;
1032         return NETDEV_TX_OK;
1033 }
1034
1035 #if IS_ENABLED(CONFIG_IPV6)
1036 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
1037                                     struct ip_tunnel_info *info)
1038 {
1039         struct geneve_dev *geneve = netdev_priv(dev);
1040         struct geneve_sock *gs6 = geneve->sock6;
1041         struct dst_entry *dst = NULL;
1042         const struct iphdr *iip; /* interior IP header */
1043         int err = -EINVAL;
1044         struct flowi6 fl6;
1045         __u8 prio, ttl;
1046         __be16 sport;
1047         __be32 label;
1048         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
1049         u32 flags = geneve->flags;
1050
1051         if (geneve->collect_md) {
1052                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1053                         netdev_dbg(dev, "no tunnel metadata\n");
1054                         goto tx_error;
1055                 }
1056         }
1057
1058         dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1059         if (IS_ERR(dst)) {
1060                 err = PTR_ERR(dst);
1061                 goto tx_error;
1062         }
1063
1064         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1065         skb_reset_mac_header(skb);
1066
1067         iip = ip_hdr(skb);
1068
1069         if (info) {
1070                 const struct ip_tunnel_key *key = &info->key;
1071                 u8 *opts = NULL;
1072                 u8 vni[3];
1073
1074                 tunnel_id_to_vni(key->tun_id, vni);
1075                 if (info->options_len)
1076                         opts = ip_tunnel_info_opts(info);
1077
1078                 if (key->tun_flags & TUNNEL_CSUM)
1079                         flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
1080                 else
1081                         flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1082
1083                 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
1084                                         info->options_len, opts,
1085                                         flags, xnet);
1086                 if (unlikely(err))
1087                         goto tx_error;
1088
1089                 prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
1090                 ttl = key->ttl;
1091                 label = info->key.label;
1092         } else {
1093                 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
1094                                         0, NULL, flags, xnet);
1095                 if (unlikely(err))
1096                         goto tx_error;
1097
1098                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1099                                            iip, skb);
1100                 ttl = geneve->ttl;
1101                 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1102                         ttl = 1;
1103                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1104                 label = geneve->label;
1105         }
1106         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1107                              &fl6.saddr, &fl6.daddr, prio, ttl, label,
1108                              sport, geneve->dst_port,
1109                              !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1110         return NETDEV_TX_OK;
1111
1112 tx_error:
1113         dev_kfree_skb(skb);
1114
1115         if (err == -ELOOP)
1116                 dev->stats.collisions++;
1117         else if (err == -ENETUNREACH)
1118                 dev->stats.tx_carrier_errors++;
1119
1120         dev->stats.tx_errors++;
1121         return NETDEV_TX_OK;
1122 }
1123 #endif
1124
1125 netdev_tx_t rpl_geneve_xmit(struct sk_buff *skb)
1126 {
1127         struct net_device *dev = skb->dev;
1128         struct geneve_dev *geneve = netdev_priv(dev);
1129         struct ip_tunnel_info *info = NULL;
1130
1131         if (geneve->collect_md)
1132                 info = skb_tunnel_info(skb);
1133
1134 #if IS_ENABLED(CONFIG_IPV6)
1135         if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1136             (!info && geneve->remote.sa.sa_family == AF_INET6))
1137                 return geneve6_xmit_skb(skb, dev, info);
1138 #endif
1139         return geneve_xmit_skb(skb, dev, info);
1140 }
1141 EXPORT_SYMBOL_GPL(rpl_geneve_xmit);
1142
1143 static netdev_tx_t geneve_dev_xmit(struct sk_buff *skb, struct net_device *dev)
1144 {
1145         /* Drop All packets coming from networking stack. OVS-CB is
1146          * not initialized for these packets.
1147          */
1148
1149         dev_kfree_skb(skb);
1150         dev->stats.tx_dropped++;
1151         return NETDEV_TX_OK;
1152 }
1153
1154 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1155 {
1156         /* The max_mtu calculation does not take account of GENEVE
1157          * options, to avoid excluding potentially valid
1158          * configurations.
1159          */
1160         int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - sizeof(struct iphdr)
1161                 - dev->hard_header_len;
1162
1163         if (new_mtu < 68)
1164                 return -EINVAL;
1165
1166         if (new_mtu > max_mtu) {
1167                 if (strict)
1168                         return -EINVAL;
1169
1170                 new_mtu = max_mtu;
1171         }
1172
1173         dev->mtu = new_mtu;
1174         return 0;
1175 }
1176
1177 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1178 {
1179         return __geneve_change_mtu(dev, new_mtu, true);
1180 }
1181
1182 int ovs_geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1183 {
1184         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1185         struct geneve_dev *geneve = netdev_priv(dev);
1186         struct rtable *rt;
1187         struct flowi4 fl4;
1188 #if IS_ENABLED(CONFIG_IPV6)
1189         struct dst_entry *dst;
1190         struct flowi6 fl6;
1191 #endif
1192
1193         if (ip_tunnel_info_af(info) == AF_INET) {
1194                 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1195                 if (IS_ERR(rt))
1196                         return PTR_ERR(rt);
1197
1198                 ip_rt_put(rt);
1199                 info->key.u.ipv4.src = fl4.saddr;
1200 #if IS_ENABLED(CONFIG_IPV6)
1201         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1202                 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1203                 if (IS_ERR(dst))
1204                         return PTR_ERR(dst);
1205
1206                 dst_release(dst);
1207                 info->key.u.ipv6.src = fl6.saddr;
1208 #endif
1209         } else {
1210                 return -EINVAL;
1211         }
1212
1213         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1214                                              1, USHRT_MAX, true);
1215         info->key.tp_dst = geneve->dst_port;
1216         return 0;
1217 }
1218 EXPORT_SYMBOL_GPL(ovs_geneve_fill_metadata_dst);
1219
1220 static const struct net_device_ops geneve_netdev_ops = {
1221         .ndo_init               = geneve_init,
1222         .ndo_uninit             = geneve_uninit,
1223         .ndo_open               = geneve_open,
1224         .ndo_stop               = geneve_stop,
1225         .ndo_start_xmit         = geneve_dev_xmit,
1226         .ndo_get_stats64        = ip_tunnel_get_stats64,
1227         .ndo_change_mtu         = geneve_change_mtu,
1228         .ndo_validate_addr      = eth_validate_addr,
1229         .ndo_set_mac_address    = eth_mac_addr,
1230 #ifdef HAVE_NDO_FILL_METADATA_DST
1231         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1232 #endif
1233 };
1234
1235 static void geneve_get_drvinfo(struct net_device *dev,
1236                                struct ethtool_drvinfo *drvinfo)
1237 {
1238         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1239         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1240 }
1241
1242 static const struct ethtool_ops geneve_ethtool_ops = {
1243         .get_drvinfo    = geneve_get_drvinfo,
1244         .get_link       = ethtool_op_get_link,
1245 };
1246
1247 /* Info for udev, that this is a virtual tunnel endpoint */
1248 static struct device_type geneve_type = {
1249         .name = "geneve",
1250 };
1251
1252 /* Calls the ndo_add_geneve_port of the caller in order to
1253  * supply the listening GENEVE udp ports. Callers are expected
1254  * to implement the ndo_add_geneve_port.
1255  */
1256 static void geneve_push_rx_ports(struct net_device *dev)
1257 {
1258 #ifdef HAVE_NDO_ADD_GENEVE_PORT
1259         struct net *net = dev_net(dev);
1260         struct geneve_net *gn = net_generic(net, geneve_net_id);
1261         struct geneve_sock *gs;
1262         sa_family_t sa_family;
1263         struct sock *sk;
1264         __be16 port;
1265
1266         if (!dev->netdev_ops->ndo_add_geneve_port)
1267                 return;
1268
1269         rcu_read_lock();
1270         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1271                 sk = gs->sock->sk;
1272                 sa_family = sk->sk_family;
1273                 port = inet_sk(sk)->inet_sport;
1274                 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family, port);
1275         }
1276         rcu_read_unlock();
1277 #endif
1278 }
1279
1280 /* Initialize the device structure. */
1281 static void geneve_setup(struct net_device *dev)
1282 {
1283         ether_setup(dev);
1284
1285         dev->netdev_ops = &geneve_netdev_ops;
1286         dev->ethtool_ops = &geneve_ethtool_ops;
1287         dev->destructor = free_netdev;
1288
1289         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1290
1291         dev->features    |= NETIF_F_LLTX;
1292         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1293         dev->features    |= NETIF_F_RXCSUM;
1294         dev->features    |= NETIF_F_GSO_SOFTWARE;
1295
1296         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1297         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1298
1299 #if 0
1300         netif_keep_dst(dev);
1301 #endif
1302         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1303         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1304         eth_hw_addr_random(dev);
1305 }
1306
1307 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1308         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1309         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1310         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1311         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1312         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1313         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1314         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1315         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1316         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1317         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1318         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1319 };
1320
1321 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1322 {
1323         if (tb[IFLA_ADDRESS]) {
1324                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1325                         return -EINVAL;
1326
1327                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1328                         return -EADDRNOTAVAIL;
1329         }
1330
1331         if (!data)
1332                 return -EINVAL;
1333
1334         if (data[IFLA_GENEVE_ID]) {
1335                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1336
1337                 if (vni >= GENEVE_VID_MASK)
1338                         return -ERANGE;
1339         }
1340
1341         return 0;
1342 }
1343
1344 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1345                                           __be16 dst_port,
1346                                           union geneve_addr *remote,
1347                                           u8 vni[],
1348                                           bool *tun_on_same_port,
1349                                           bool *tun_collect_md)
1350 {
1351         struct geneve_dev *geneve, *t;
1352
1353         *tun_on_same_port = false;
1354         *tun_collect_md = false;
1355         t = NULL;
1356         list_for_each_entry(geneve, &gn->geneve_list, next) {
1357                 if (geneve->dst_port == dst_port) {
1358                         *tun_collect_md = geneve->collect_md;
1359                         *tun_on_same_port = true;
1360                 }
1361                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1362                     !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1363                     dst_port == geneve->dst_port)
1364                         t = geneve;
1365         }
1366         return t;
1367 }
1368
1369 static int geneve_configure(struct net *net, struct net_device *dev,
1370                             union geneve_addr *remote,
1371                             __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1372                             __be16 dst_port, bool metadata, u32 flags)
1373 {
1374         struct geneve_net *gn = net_generic(net, geneve_net_id);
1375         struct geneve_dev *t, *geneve = netdev_priv(dev);
1376         bool tun_collect_md, tun_on_same_port;
1377         int err, encap_len;
1378
1379         if (!remote)
1380                 return -EINVAL;
1381         if (metadata &&
1382             (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1383                 return -EINVAL;
1384
1385         geneve->net = net;
1386         geneve->dev = dev;
1387
1388         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1389         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1390         geneve->vni[2] =  vni & 0x000000ff;
1391
1392         if ((remote->sa.sa_family == AF_INET &&
1393              IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1394             (remote->sa.sa_family == AF_INET6 &&
1395              ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1396                 return -EINVAL;
1397         if (label && remote->sa.sa_family != AF_INET6)
1398                 return -EINVAL;
1399
1400         geneve->remote = *remote;
1401
1402         geneve->ttl = ttl;
1403         geneve->tos = tos;
1404         geneve->label = label;
1405         geneve->dst_port = dst_port;
1406         geneve->collect_md = metadata;
1407         geneve->flags = flags;
1408
1409         t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1410                             &tun_on_same_port, &tun_collect_md);
1411         if (t)
1412                 return -EBUSY;
1413
1414         /* make enough headroom for basic scenario */
1415         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1416         if (remote->sa.sa_family == AF_INET)
1417                 encap_len += sizeof(struct iphdr);
1418         else
1419                 encap_len += sizeof(struct ipv6hdr);
1420         dev->needed_headroom = encap_len + ETH_HLEN;
1421
1422         if (metadata) {
1423                 if (tun_on_same_port)
1424                         return -EPERM;
1425         } else {
1426                 if (tun_collect_md)
1427                         return -EPERM;
1428         }
1429
1430         dst_cache_reset(&geneve->dst_cache);
1431
1432         err = register_netdevice(dev);
1433         if (err)
1434                 return err;
1435
1436         list_add(&geneve->next, &gn->geneve_list);
1437         return 0;
1438 }
1439
1440 static int geneve_newlink(struct net *net, struct net_device *dev,
1441                           struct nlattr *tb[], struct nlattr *data[])
1442 {
1443         __be16 dst_port = htons(GENEVE_UDP_PORT);
1444         __u8 ttl = 0, tos = 0;
1445         bool metadata = false;
1446         union geneve_addr remote = geneve_remote_unspec;
1447         __be32 label = 0;
1448         __u32 vni = 0;
1449         u32 flags = 0;
1450
1451         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1452                 return -EINVAL;
1453
1454         if (data[IFLA_GENEVE_REMOTE]) {
1455                 remote.sa.sa_family = AF_INET;
1456                 remote.sin.sin_addr.s_addr =
1457                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1458         }
1459
1460         if (data[IFLA_GENEVE_REMOTE6]) {
1461                 if (!IS_ENABLED(CONFIG_IPV6))
1462                         return -EPFNOSUPPORT;
1463
1464                 remote.sa.sa_family = AF_INET6;
1465                 remote.sin6.sin6_addr =
1466                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1467
1468                 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1469                     IPV6_ADDR_LINKLOCAL) {
1470                         netdev_dbg(dev, "link-local remote is unsupported\n");
1471                         return -EINVAL;
1472                 }
1473         }
1474
1475         if (data[IFLA_GENEVE_ID])
1476                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1477
1478         if (data[IFLA_GENEVE_TTL])
1479                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1480
1481         if (data[IFLA_GENEVE_TOS])
1482                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1483
1484         if (data[IFLA_GENEVE_LABEL])
1485                 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1486                         IPV6_FLOWLABEL_MASK;
1487
1488         if (data[IFLA_GENEVE_PORT])
1489                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1490
1491         if (data[IFLA_GENEVE_COLLECT_METADATA])
1492                 metadata = true;
1493
1494         if (data[IFLA_GENEVE_UDP_CSUM] &&
1495             !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1496                 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1497
1498         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1499             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1500                 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1501
1502         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1503             nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1504                 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1505
1506         return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1507                                 dst_port, metadata, flags);
1508 }
1509
1510 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1511 {
1512         struct geneve_dev *geneve = netdev_priv(dev);
1513
1514         list_del(&geneve->next);
1515         unregister_netdevice_queue(dev, head);
1516 }
1517
1518 static size_t geneve_get_size(const struct net_device *dev)
1519 {
1520         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1521                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1522                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1523                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1524                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1525                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1526                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1527                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1528                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1529                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1530                 0;
1531 }
1532
1533 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1534 {
1535         struct geneve_dev *geneve = netdev_priv(dev);
1536         __u32 vni;
1537
1538         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1539         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1540                 goto nla_put_failure;
1541
1542         if (geneve->remote.sa.sa_family == AF_INET) {
1543                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1544                                     geneve->remote.sin.sin_addr.s_addr))
1545                         goto nla_put_failure;
1546 #if IS_ENABLED(CONFIG_IPV6)
1547         } else {
1548                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1549                                      &geneve->remote.sin6.sin6_addr))
1550                         goto nla_put_failure;
1551 #endif
1552         }
1553
1554         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1555             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1556             nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1557                 goto nla_put_failure;
1558
1559         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1560                 goto nla_put_failure;
1561
1562         if (geneve->collect_md) {
1563                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1564                         goto nla_put_failure;
1565         }
1566
1567         if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1568                        !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1569             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1570                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1571             nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1572                        !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1573                 goto nla_put_failure;
1574
1575         return 0;
1576
1577 nla_put_failure:
1578         return -EMSGSIZE;
1579 }
1580
1581 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1582         .kind           = "geneve",
1583         .maxtype        = IFLA_GENEVE_MAX,
1584         .policy         = geneve_policy,
1585         .priv_size      = sizeof(struct geneve_dev),
1586         .setup          = geneve_setup,
1587         .validate       = geneve_validate,
1588         .newlink        = geneve_newlink,
1589         .dellink        = geneve_dellink,
1590         .get_size       = geneve_get_size,
1591         .fill_info      = geneve_fill_info,
1592 };
1593
1594 struct net_device *rpl_geneve_dev_create_fb(struct net *net, const char *name,
1595                                             u8 name_assign_type, u16 dst_port)
1596 {
1597         struct nlattr *tb[IFLA_MAX + 1];
1598         struct net_device *dev;
1599         LIST_HEAD(list_kill);
1600         int err;
1601
1602         memset(tb, 0, sizeof(tb));
1603         dev = rtnl_create_link(net, name, name_assign_type,
1604                                &geneve_link_ops, tb);
1605         if (IS_ERR(dev))
1606                 return dev;
1607
1608         err = geneve_configure(net, dev, &geneve_remote_unspec,
1609                                0, 0, 0, 0, htons(dst_port), true,
1610                                GENEVE_F_UDP_ZERO_CSUM6_RX);
1611         if (err) {
1612                 free_netdev(dev);
1613                 return ERR_PTR(err);
1614         }
1615
1616         /* openvswitch users expect packet sizes to be unrestricted,
1617          * so set the largest MTU we can.
1618          */
1619         err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1620         if (err)
1621                 goto err;
1622
1623         err = rtnl_configure_link(dev, NULL);
1624         if (err < 0)
1625                 goto err;
1626
1627         return dev;
1628
1629  err:
1630         geneve_dellink(dev, &list_kill);
1631         unregister_netdevice_many(&list_kill);
1632         return ERR_PTR(err);
1633 }
1634 EXPORT_SYMBOL_GPL(rpl_geneve_dev_create_fb);
1635
1636 static int geneve_netdevice_event(struct notifier_block *unused,
1637                                   unsigned long event, void *ptr)
1638 {
1639         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1640
1641         if (event == NETDEV_OFFLOAD_PUSH_GENEVE)
1642                 geneve_push_rx_ports(dev);
1643
1644         return NOTIFY_DONE;
1645 }
1646
1647 static struct notifier_block geneve_notifier_block __read_mostly = {
1648         .notifier_call = geneve_netdevice_event,
1649 };
1650
1651 static __net_init int geneve_init_net(struct net *net)
1652 {
1653         struct geneve_net *gn = net_generic(net, geneve_net_id);
1654
1655         INIT_LIST_HEAD(&gn->geneve_list);
1656         INIT_LIST_HEAD(&gn->sock_list);
1657         return 0;
1658 }
1659
1660 static void __net_exit geneve_exit_net(struct net *net)
1661 {
1662         struct geneve_net *gn = net_generic(net, geneve_net_id);
1663         struct geneve_dev *geneve, *next;
1664         struct net_device *dev, *aux;
1665         LIST_HEAD(list);
1666
1667         rtnl_lock();
1668
1669         /* gather any geneve devices that were moved into this ns */
1670         for_each_netdev_safe(net, dev, aux)
1671                 if (dev->rtnl_link_ops == &geneve_link_ops)
1672                         unregister_netdevice_queue(dev, &list);
1673
1674         /* now gather any other geneve devices that were created in this ns */
1675         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1676                 /* If geneve->dev is in the same netns, it was already added
1677                  * to the list by the previous loop.
1678                  */
1679                 if (!net_eq(dev_net(geneve->dev), net))
1680                         unregister_netdevice_queue(geneve->dev, &list);
1681         }
1682
1683         /* unregister the devices gathered above */
1684         unregister_netdevice_many(&list);
1685         rtnl_unlock();
1686 }
1687
1688 static struct pernet_operations geneve_net_ops = {
1689         .init = geneve_init_net,
1690         .exit = geneve_exit_net,
1691         .id   = &geneve_net_id,
1692         .size = sizeof(struct geneve_net),
1693 };
1694
1695 int rpl_geneve_init_module(void)
1696 {
1697         int rc;
1698
1699         rc = register_pernet_subsys(&geneve_net_ops);
1700         if (rc)
1701                 goto out1;
1702
1703         rc = register_netdevice_notifier(&geneve_notifier_block);
1704         if (rc)
1705                 goto out2;
1706
1707         rc = rtnl_link_register(&geneve_link_ops);
1708         if (rc)
1709                 goto out3;
1710
1711         pr_info("Geneve tunneling driver\n");
1712         return 0;
1713
1714 out3:
1715         unregister_netdevice_notifier(&geneve_notifier_block);
1716 out2:
1717         unregister_pernet_subsys(&geneve_net_ops);
1718 out1:
1719         return rc;
1720 }
1721
1722 void rpl_geneve_cleanup_module(void)
1723 {
1724         rtnl_link_unregister(&geneve_link_ops);
1725         unregister_netdevice_notifier(&geneve_notifier_block);
1726         unregister_pernet_subsys(&geneve_net_ops);
1727 }
1728
1729 #endif