drivers/net/geneve.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * GENEVE: Generic Network Virtualization Encapsulation
   4  *
   5  * Copyright (c) 2015 Red Hat, Inc.
   6  */
   7
   8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   9
  10 #include <linux/ethtool.h>
  11 #include <linux/kernel.h>
  12 #include <linux/module.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/hash.h>
  15 #include <net/ipv6_stubs.h>
  16 #include <net/dst_metadata.h>
  17 #include <net/gro_cells.h>
  18 #include <net/rtnetlink.h>
  19 #include <net/geneve.h>
  20 #include <net/protocol.h>
  21
  22 #define GENEVE_NETDEV_VER       "0.6"
  23
  24 #define GENEVE_N_VID            (1u << 24)
  25 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  26
  27 #define VNI_HASH_BITS           10
  28 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  29
  30 static bool log_ecn_error = true;
  31 module_param(log_ecn_error, bool, 0644);
  32 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  33
  34 #define GENEVE_VER 0
  35 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  36 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
  37 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
  38
  39 /* per-network namespace private data for this module */
  40 struct geneve_net {
  41         struct list_head        geneve_list;
  42         struct list_head        sock_list;
  43 };
  44
  45 static unsigned int geneve_net_id;
  46
  47 struct geneve_dev_node {
  48         struct hlist_node hlist;
  49         struct geneve_dev *geneve;
  50 };
  51
  52 struct geneve_config {
  53         struct ip_tunnel_info   info;
  54         bool                    collect_md;
  55         bool                    use_udp6_rx_checksums;
  56         bool                    ttl_inherit;
  57         enum ifla_geneve_df     df;
  58 };
  59
  60 /* Pseudo network device */
  61 struct geneve_dev {
  62         struct geneve_dev_node hlist4;  /* vni hash table for IPv4 socket */
  63 #if IS_ENABLED(CONFIG_IPV6)
  64         struct geneve_dev_node hlist6;  /* vni hash table for IPv6 socket */
  65 #endif
  66         struct net         *net;        /* netns for packet i/o */
  67         struct net_device  *dev;        /* netdev for geneve tunnel */
  68         struct geneve_sock __rcu *sock4;        /* IPv4 socket used for geneve tunnel */
  69 #if IS_ENABLED(CONFIG_IPV6)
  70         struct geneve_sock __rcu *sock6;        /* IPv6 socket used for geneve tunnel */
  71 #endif
  72         struct list_head   next;        /* geneve's per namespace list */
  73         struct gro_cells   gro_cells;
  74         struct geneve_config cfg;
  75 };
  76
  77 struct geneve_sock {
  78         bool                    collect_md;
  79         struct list_head        list;
  80         struct socket           *sock;
  81         struct rcu_head         rcu;
  82         int                     refcnt;
  83         struct hlist_head       vni_list[VNI_HASH_SIZE];
  84 };
  85
  86 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  87 {
  88         __u32 vnid;
  89
  90         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  91         return hash_32(vnid, VNI_HASH_BITS);
  92 }
  93
  94 static __be64 vni_to_tunnel_id(const __u8 *vni)
  95 {
  96 #ifdef __BIG_ENDIAN
  97         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
  98 #else
  99         return (__force __be64)(((__force u64)vni[0] << 40) |
 100                                 ((__force u64)vni[1] << 48) |
 101                                 ((__force u64)vni[2] << 56));
 102 #endif
 103 }
 104
 105 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 106 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 107 {
 108 #ifdef __BIG_ENDIAN
 109         vni[0] = (__force __u8)(tun_id >> 16);
 110         vni[1] = (__force __u8)(tun_id >> 8);
 111         vni[2] = (__force __u8)tun_id;
 112 #else
 113         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 114         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 115         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 116 #endif
 117 }
 118
 119 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
 120 {
 121         return !memcmp(vni, &tun_id[5], 3);
 122 }
 123
 124 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 125 {
 126         return gs->sock->sk->sk_family;
 127 }
 128
 129 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 130                                         __be32 addr, u8 vni[])
 131 {
 132         struct hlist_head *vni_list_head;
 133         struct geneve_dev_node *node;
 134         __u32 hash;
 135
 136         /* Find the device for this VNI */
 137         hash = geneve_net_vni_hash(vni);
 138         vni_list_head = &gs->vni_list[hash];
 139         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 140                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 141                     addr == node->geneve->cfg.info.key.u.ipv4.dst)
 142                         return node->geneve;
 143         }
 144         return NULL;
 145 }
 146
 147 #if IS_ENABLED(CONFIG_IPV6)
 148 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 149                                          struct in6_addr addr6, u8 vni[])
 150 {
 151         struct hlist_head *vni_list_head;
 152         struct geneve_dev_node *node;
 153         __u32 hash;
 154
 155         /* Find the device for this VNI */
 156         hash = geneve_net_vni_hash(vni);
 157         vni_list_head = &gs->vni_list[hash];
 158         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 159                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 160                     ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
 161                         return node->geneve;
 162         }
 163         return NULL;
 164 }
 165 #endif
 166
 167 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 168 {
 169         return (struct genevehdr *)(udp_hdr(skb) + 1);
 170 }
 171
 172 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 173                                             struct sk_buff *skb)
 174 {
 175         static u8 zero_vni[3];
 176         u8 *vni;
 177
 178         if (geneve_get_sk_family(gs) == AF_INET) {
 179                 struct iphdr *iph;
 180                 __be32 addr;
 181
 182                 iph = ip_hdr(skb); /* outer IP header... */
 183
 184                 if (gs->collect_md) {
 185                         vni = zero_vni;
 186                         addr = 0;
 187                 } else {
 188                         vni = geneve_hdr(skb)->vni;
 189                         addr = iph->saddr;
 190                 }
 191
 192                 return geneve_lookup(gs, addr, vni);
 193 #if IS_ENABLED(CONFIG_IPV6)
 194         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 195                 static struct in6_addr zero_addr6;
 196                 struct ipv6hdr *ip6h;
 197                 struct in6_addr addr6;
 198
 199                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 200
 201                 if (gs->collect_md) {
 202                         vni = zero_vni;
 203                         addr6 = zero_addr6;
 204                 } else {
 205                         vni = geneve_hdr(skb)->vni;
 206                         addr6 = ip6h->saddr;
 207                 }
 208
 209                 return geneve6_lookup(gs, addr6, vni);
 210 #endif
 211         }
 212         return NULL;
 213 }
 214
 215 /* geneve receive/decap routine */
 216 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 217                       struct sk_buff *skb)
 218 {
 219         struct genevehdr *gnvh = geneve_hdr(skb);
 220         struct metadata_dst *tun_dst = NULL;
 221         unsigned int len;
 222         int err = 0;
 223         void *oiph;
 224
 225         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 226                 __be16 flags;
 227
 228                 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
 229                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 230
 231                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 232                                          vni_to_tunnel_id(gnvh->vni),
 233                                          gnvh->opt_len * 4);
 234                 if (!tun_dst) {
 235                         geneve->dev->stats.rx_dropped++;
 236                         goto drop;
 237                 }
 238                 /* Update tunnel dst according to Geneve options. */
 239                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 240                                         gnvh->options, gnvh->opt_len * 4,
 241                                         TUNNEL_GENEVE_OPT);
 242         } else {
 243                 /* Drop packets w/ critical options,
 244                  * since we don't support any...
 245                  */
 246                 if (gnvh->critical) {
 247                         geneve->dev->stats.rx_frame_errors++;
 248                         geneve->dev->stats.rx_errors++;
 249                         goto drop;
 250                 }
 251         }
 252
 253         skb_reset_mac_header(skb);
 254         skb->protocol = eth_type_trans(skb, geneve->dev);
 255         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 256
 257         if (tun_dst)
 258                 skb_dst_set(skb, &tun_dst->dst);
 259
 260         /* Ignore packet loops (and multicast echo) */
 261         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
 262                 geneve->dev->stats.rx_errors++;
 263                 goto drop;
 264         }
 265
 266         oiph = skb_network_header(skb);
 267         skb_reset_network_header(skb);
 268
 269         if (geneve_get_sk_family(gs) == AF_INET)
 270                 err = IP_ECN_decapsulate(oiph, skb);
 271 #if IS_ENABLED(CONFIG_IPV6)
 272         else
 273                 err = IP6_ECN_decapsulate(oiph, skb);
 274 #endif
 275
 276         if (unlikely(err)) {
 277                 if (log_ecn_error) {
 278                         if (geneve_get_sk_family(gs) == AF_INET)
 279                                 net_info_ratelimited("non-ECT from %pI4 "
 280                                                      "with TOS=%#x\n",
 281                                                      &((struct iphdr *)oiph)->saddr,
 282                                                      ((struct iphdr *)oiph)->tos);
 283 #if IS_ENABLED(CONFIG_IPV6)
 284                         else
 285                                 net_info_ratelimited("non-ECT from %pI6\n",
 286                                                      &((struct ipv6hdr *)oiph)->saddr);
 287 #endif
 288                 }
 289                 if (err > 1) {
 290                         ++geneve->dev->stats.rx_frame_errors;
 291                         ++geneve->dev->stats.rx_errors;
 292                         goto drop;
 293                 }
 294         }
 295
 296         len = skb->len;
 297         err = gro_cells_receive(&geneve->gro_cells, skb);
 298         if (likely(err == NET_RX_SUCCESS))
 299                 dev_sw_netstats_rx_add(geneve->dev, len);
 300
 301         return;
 302 drop:
 303         /* Consume bad packet */
 304         kfree_skb(skb);
 305 }
 306
 307 /* Setup stats when device is created */
 308 static int geneve_init(struct net_device *dev)
 309 {
 310         struct geneve_dev *geneve = netdev_priv(dev);
 311         int err;
 312
 313         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 314         if (!dev->tstats)
 315                 return -ENOMEM;
 316
 317         err = gro_cells_init(&geneve->gro_cells, dev);
 318         if (err) {
 319                 free_percpu(dev->tstats);
 320                 return err;
 321         }
 322
 323         err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
 324         if (err) {
 325                 free_percpu(dev->tstats);
 326                 gro_cells_destroy(&geneve->gro_cells);
 327                 return err;
 328         }
 329         return 0;
 330 }
 331
 332 static void geneve_uninit(struct net_device *dev)
 333 {
 334         struct geneve_dev *geneve = netdev_priv(dev);
 335
 336         dst_cache_destroy(&geneve->cfg.info.dst_cache);
 337         gro_cells_destroy(&geneve->gro_cells);
 338         free_percpu(dev->tstats);
 339 }
 340
 341 /* Callback from net/ipv4/udp.c to receive packets */
 342 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 343 {
 344         struct genevehdr *geneveh;
 345         struct geneve_dev *geneve;
 346         struct geneve_sock *gs;
 347         int opts_len;
 348
 349         /* Need UDP and Geneve header to be present */
 350         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 351                 goto drop;
 352
 353         /* Return packets with reserved bits set */
 354         geneveh = geneve_hdr(skb);
 355         if (unlikely(geneveh->ver != GENEVE_VER))
 356                 goto drop;
 357
 358         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 359                 goto drop;
 360
 361         gs = rcu_dereference_sk_user_data(sk);
 362         if (!gs)
 363                 goto drop;
 364
 365         geneve = geneve_lookup_skb(gs, skb);
 366         if (!geneve)
 367                 goto drop;
 368
 369         opts_len = geneveh->opt_len * 4;
 370         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 371                                  htons(ETH_P_TEB),
 372                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 373                 geneve->dev->stats.rx_dropped++;
 374                 goto drop;
 375         }
 376
 377         geneve_rx(geneve, gs, skb);
 378         return 0;
 379
 380 drop:
 381         /* Consume bad packet */
 382         kfree_skb(skb);
 383         return 0;
 384 }
 385
 386 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
 387 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
 388 {
 389         struct genevehdr *geneveh;
 390         struct geneve_sock *gs;
 391         u8 zero_vni[3] = { 0 };
 392         u8 *vni = zero_vni;
 393
 394         if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
 395                 return -EINVAL;
 396
 397         geneveh = geneve_hdr(skb);
 398         if (geneveh->ver != GENEVE_VER)
 399                 return -EINVAL;
 400
 401         if (geneveh->proto_type != htons(ETH_P_TEB))
 402                 return -EINVAL;
 403
 404         gs = rcu_dereference_sk_user_data(sk);
 405         if (!gs)
 406                 return -ENOENT;
 407
 408         if (geneve_get_sk_family(gs) == AF_INET) {
 409                 struct iphdr *iph = ip_hdr(skb);
 410                 __be32 addr4 = 0;
 411
 412                 if (!gs->collect_md) {
 413                         vni = geneve_hdr(skb)->vni;
 414                         addr4 = iph->daddr;
 415                 }
 416
 417                 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
 418         }
 419
 420 #if IS_ENABLED(CONFIG_IPV6)
 421         if (geneve_get_sk_family(gs) == AF_INET6) {
 422                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 423                 struct in6_addr addr6;
 424
 425                 memset(&addr6, 0, sizeof(struct in6_addr));
 426
 427                 if (!gs->collect_md) {
 428                         vni = geneve_hdr(skb)->vni;
 429                         addr6 = ip6h->daddr;
 430                 }
 431
 432                 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
 433         }
 434 #endif
 435
 436         return -EPFNOSUPPORT;
 437 }
 438
 439 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 440                                          __be16 port, bool ipv6_rx_csum)
 441 {
 442         struct socket *sock;
 443         struct udp_port_cfg udp_conf;
 444         int err;
 445
 446         memset(&udp_conf, 0, sizeof(udp_conf));
 447
 448         if (ipv6) {
 449                 udp_conf.family = AF_INET6;
 450                 udp_conf.ipv6_v6only = 1;
 451                 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
 452         } else {
 453                 udp_conf.family = AF_INET;
 454                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 455         }
 456
 457         udp_conf.local_udp_port = port;
 458
 459         /* Open UDP socket */
 460         err = udp_sock_create(net, &udp_conf, &sock);
 461         if (err < 0)
 462                 return ERR_PTR(err);
 463
 464         udp_allow_gso(sock->sk);
 465         return sock;
 466 }
 467
 468 static int geneve_hlen(struct genevehdr *gh)
 469 {
 470         return sizeof(*gh) + gh->opt_len * 4;
 471 }
 472
 473 static struct sk_buff *geneve_gro_receive(struct sock *sk,
 474                                           struct list_head *head,
 475                                           struct sk_buff *skb)
 476 {
 477         struct sk_buff *pp = NULL;
 478         struct sk_buff *p;
 479         struct genevehdr *gh, *gh2;
 480         unsigned int hlen, gh_len, off_gnv;
 481         const struct packet_offload *ptype;
 482         __be16 type;
 483         int flush = 1;
 484
 485         off_gnv = skb_gro_offset(skb);
 486         hlen = off_gnv + sizeof(*gh);
 487         gh = skb_gro_header_fast(skb, off_gnv);
 488         if (skb_gro_header_hard(skb, hlen)) {
 489                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 490                 if (unlikely(!gh))
 491                         goto out;
 492         }
 493
 494         if (gh->ver != GENEVE_VER || gh->oam)
 495                 goto out;
 496         gh_len = geneve_hlen(gh);
 497
 498         hlen = off_gnv + gh_len;
 499         if (skb_gro_header_hard(skb, hlen)) {
 500                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 501                 if (unlikely(!gh))
 502                         goto out;
 503         }
 504
 505         list_for_each_entry(p, head, list) {
 506                 if (!NAPI_GRO_CB(p)->same_flow)
 507                         continue;
 508
 509                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 510                 if (gh->opt_len != gh2->opt_len ||
 511                     memcmp(gh, gh2, gh_len)) {
 512                         NAPI_GRO_CB(p)->same_flow = 0;
 513                         continue;
 514                 }
 515         }
 516
 517         type = gh->proto_type;
 518
 519         rcu_read_lock();
 520         ptype = gro_find_receive_by_type(type);
 521         if (!ptype)
 522                 goto out_unlock;
 523
 524         skb_gro_pull(skb, gh_len);
 525         skb_gro_postpull_rcsum(skb, gh, gh_len);
 526         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 527         flush = 0;
 528
 529 out_unlock:
 530         rcu_read_unlock();
 531 out:
 532         skb_gro_flush_final(skb, pp, flush);
 533
 534         return pp;
 535 }
 536
 537 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 538                                int nhoff)
 539 {
 540         struct genevehdr *gh;
 541         struct packet_offload *ptype;
 542         __be16 type;
 543         int gh_len;
 544         int err = -ENOSYS;
 545
 546         gh = (struct genevehdr *)(skb->data + nhoff);
 547         gh_len = geneve_hlen(gh);
 548         type = gh->proto_type;
 549
 550         rcu_read_lock();
 551         ptype = gro_find_complete_by_type(type);
 552         if (ptype)
 553                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 554
 555         rcu_read_unlock();
 556
 557         skb_set_inner_mac_header(skb, nhoff + gh_len);
 558
 559         return err;
 560 }
 561
 562 /* Create new listen socket if needed */
 563 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 564                                                 bool ipv6, bool ipv6_rx_csum)
 565 {
 566         struct geneve_net *gn = net_generic(net, geneve_net_id);
 567         struct geneve_sock *gs;
 568         struct socket *sock;
 569         struct udp_tunnel_sock_cfg tunnel_cfg;
 570         int h;
 571
 572         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 573         if (!gs)
 574                 return ERR_PTR(-ENOMEM);
 575
 576         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 577         if (IS_ERR(sock)) {
 578                 kfree(gs);
 579                 return ERR_CAST(sock);
 580         }
 581
 582         gs->sock = sock;
 583         gs->refcnt = 1;
 584         for (h = 0; h < VNI_HASH_SIZE; ++h)
 585                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 586
 587         /* Initialize the geneve udp offloads structure */
 588         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 589
 590         /* Mark socket as an encapsulation socket */
 591         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 592         tunnel_cfg.sk_user_data = gs;
 593         tunnel_cfg.encap_type = 1;
 594         tunnel_cfg.gro_receive = geneve_gro_receive;
 595         tunnel_cfg.gro_complete = geneve_gro_complete;
 596         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 597         tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
 598         tunnel_cfg.encap_destroy = NULL;
 599         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 600         list_add(&gs->list, &gn->sock_list);
 601         return gs;
 602 }
 603
 604 static void __geneve_sock_release(struct geneve_sock *gs)
 605 {
 606         if (!gs || --gs->refcnt)
 607                 return;
 608
 609         list_del(&gs->list);
 610         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 611         udp_tunnel_sock_release(gs->sock);
 612         kfree_rcu(gs, rcu);
 613 }
 614
 615 static void geneve_sock_release(struct geneve_dev *geneve)
 616 {
 617         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 618 #if IS_ENABLED(CONFIG_IPV6)
 619         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 620
 621         rcu_assign_pointer(geneve->sock6, NULL);
 622 #endif
 623
 624         rcu_assign_pointer(geneve->sock4, NULL);
 625         synchronize_net();
 626
 627         __geneve_sock_release(gs4);
 628 #if IS_ENABLED(CONFIG_IPV6)
 629         __geneve_sock_release(gs6);
 630 #endif
 631 }
 632
 633 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 634                                             sa_family_t family,
 635                                             __be16 dst_port)
 636 {
 637         struct geneve_sock *gs;
 638
 639         list_for_each_entry(gs, &gn->sock_list, list) {
 640                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 641                     geneve_get_sk_family(gs) == family) {
 642                         return gs;
 643                 }
 644         }
 645         return NULL;
 646 }
 647
 648 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 649 {
 650         struct net *net = geneve->net;
 651         struct geneve_net *gn = net_generic(net, geneve_net_id);
 652         struct geneve_dev_node *node;
 653         struct geneve_sock *gs;
 654         __u8 vni[3];
 655         __u32 hash;
 656
 657         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
 658         if (gs) {
 659                 gs->refcnt++;
 660                 goto out;
 661         }
 662
 663         gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
 664                                   geneve->cfg.use_udp6_rx_checksums);
 665         if (IS_ERR(gs))
 666                 return PTR_ERR(gs);
 667
 668 out:
 669         gs->collect_md = geneve->cfg.collect_md;
 670 #if IS_ENABLED(CONFIG_IPV6)
 671         if (ipv6) {
 672                 rcu_assign_pointer(geneve->sock6, gs);
 673                 node = &geneve->hlist6;
 674         } else
 675 #endif
 676         {
 677                 rcu_assign_pointer(geneve->sock4, gs);
 678                 node = &geneve->hlist4;
 679         }
 680         node->geneve = geneve;
 681
 682         tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
 683         hash = geneve_net_vni_hash(vni);
 684         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 685         return 0;
 686 }
 687
 688 static int geneve_open(struct net_device *dev)
 689 {
 690         struct geneve_dev *geneve = netdev_priv(dev);
 691         bool metadata = geneve->cfg.collect_md;
 692         bool ipv4, ipv6;
 693         int ret = 0;
 694
 695         ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 696         ipv4 = !ipv6 || metadata;
 697 #if IS_ENABLED(CONFIG_IPV6)
 698         if (ipv6) {
 699                 ret = geneve_sock_add(geneve, true);
 700                 if (ret < 0 && ret != -EAFNOSUPPORT)
 701                         ipv4 = false;
 702         }
 703 #endif
 704         if (ipv4)
 705                 ret = geneve_sock_add(geneve, false);
 706         if (ret < 0)
 707                 geneve_sock_release(geneve);
 708
 709         return ret;
 710 }
 711
 712 static int geneve_stop(struct net_device *dev)
 713 {
 714         struct geneve_dev *geneve = netdev_priv(dev);
 715
 716         hlist_del_init_rcu(&geneve->hlist4.hlist);
 717 #if IS_ENABLED(CONFIG_IPV6)
 718         hlist_del_init_rcu(&geneve->hlist6.hlist);
 719 #endif
 720         geneve_sock_release(geneve);
 721         return 0;
 722 }
 723
 724 static void geneve_build_header(struct genevehdr *geneveh,
 725                                 const struct ip_tunnel_info *info)
 726 {
 727         geneveh->ver = GENEVE_VER;
 728         geneveh->opt_len = info->options_len / 4;
 729         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 730         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 731         geneveh->rsvd1 = 0;
 732         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 733         geneveh->proto_type = htons(ETH_P_TEB);
 734         geneveh->rsvd2 = 0;
 735
 736         if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
 737                 ip_tunnel_info_opts_get(geneveh->options, info);
 738 }
 739
 740 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 741                             const struct ip_tunnel_info *info,
 742                             bool xnet, int ip_hdr_len)
 743 {
 744         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 745         struct genevehdr *gnvh;
 746         int min_headroom;
 747         int err;
 748
 749         skb_reset_mac_header(skb);
 750         skb_scrub_packet(skb, xnet);
 751
 752         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 753                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 754         err = skb_cow_head(skb, min_headroom);
 755         if (unlikely(err))
 756                 goto free_dst;
 757
 758         err = udp_tunnel_handle_offloads(skb, udp_sum);
 759         if (err)
 760                 goto free_dst;
 761
 762         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 763         geneve_build_header(gnvh, info);
 764         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 765         return 0;
 766
 767 free_dst:
 768         dst_release(dst);
 769         return err;
 770 }
 771
 772 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 773                                        struct net_device *dev,
 774                                        struct geneve_sock *gs4,
 775                                        struct flowi4 *fl4,
 776                                        const struct ip_tunnel_info *info,
 777                                        __be16 dport, __be16 sport)
 778 {
 779         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 780         struct geneve_dev *geneve = netdev_priv(dev);
 781         struct dst_cache *dst_cache;
 782         struct rtable *rt = NULL;
 783         __u8 tos;
 784
 785         if (!gs4)
 786                 return ERR_PTR(-EIO);
 787
 788         memset(fl4, 0, sizeof(*fl4));
 789         fl4->flowi4_mark = skb->mark;
 790         fl4->flowi4_proto = IPPROTO_UDP;
 791         fl4->daddr = info->key.u.ipv4.dst;
 792         fl4->saddr = info->key.u.ipv4.src;
 793         fl4->fl4_dport = dport;
 794         fl4->fl4_sport = sport;
 795
 796         tos = info->key.tos;
 797         if ((tos == 1) && !geneve->cfg.collect_md) {
 798                 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 799                 use_cache = false;
 800         }
 801         fl4->flowi4_tos = RT_TOS(tos);
 802
 803         dst_cache = (struct dst_cache *)&info->dst_cache;
 804         if (use_cache) {
 805                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 806                 if (rt)
 807                         return rt;
 808         }
 809         rt = ip_route_output_key(geneve->net, fl4);
 810         if (IS_ERR(rt)) {
 811                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 812                 return ERR_PTR(-ENETUNREACH);
 813         }
 814         if (rt->dst.dev == dev) { /* is this necessary? */
 815                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 816                 ip_rt_put(rt);
 817                 return ERR_PTR(-ELOOP);
 818         }
 819         if (use_cache)
 820                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 821         return rt;
 822 }
 823
 824 #if IS_ENABLED(CONFIG_IPV6)
 825 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 826                                            struct net_device *dev,
 827                                            struct geneve_sock *gs6,
 828                                            struct flowi6 *fl6,
 829                                            const struct ip_tunnel_info *info,
 830                                            __be16 dport, __be16 sport)
 831 {
 832         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 833         struct geneve_dev *geneve = netdev_priv(dev);
 834         struct dst_entry *dst = NULL;
 835         struct dst_cache *dst_cache;
 836         __u8 prio;
 837
 838         if (!gs6)
 839                 return ERR_PTR(-EIO);
 840
 841         memset(fl6, 0, sizeof(*fl6));
 842         fl6->flowi6_mark = skb->mark;
 843         fl6->flowi6_proto = IPPROTO_UDP;
 844         fl6->daddr = info->key.u.ipv6.dst;
 845         fl6->saddr = info->key.u.ipv6.src;
 846         fl6->fl6_dport = dport;
 847         fl6->fl6_sport = sport;
 848
 849         prio = info->key.tos;
 850         if ((prio == 1) && !geneve->cfg.collect_md) {
 851                 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 852                 use_cache = false;
 853         }
 854
 855         fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 856                                            info->key.label);
 857         dst_cache = (struct dst_cache *)&info->dst_cache;
 858         if (use_cache) {
 859                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 860                 if (dst)
 861                         return dst;
 862         }
 863         dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
 864                                               NULL);
 865         if (IS_ERR(dst)) {
 866                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 867                 return ERR_PTR(-ENETUNREACH);
 868         }
 869         if (dst->dev == dev) { /* is this necessary? */
 870                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 871                 dst_release(dst);
 872                 return ERR_PTR(-ELOOP);
 873         }
 874
 875         if (use_cache)
 876                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 877         return dst;
 878 }
 879 #endif
 880
 881 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 882                            struct geneve_dev *geneve,
 883                            const struct ip_tunnel_info *info)
 884 {
 885         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 886         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 887         const struct ip_tunnel_key *key = &info->key;
 888         struct rtable *rt;
 889         struct flowi4 fl4;
 890         __u8 tos, ttl;
 891         __be16 df = 0;
 892         __be16 sport;
 893         int err;
 894
 895         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 896         rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
 897                               geneve->cfg.info.key.tp_dst, sport);
 898         if (IS_ERR(rt))
 899                 return PTR_ERR(rt);
 900
 901         err = skb_tunnel_check_pmtu(skb, &rt->dst,
 902                                     GENEVE_IPV4_HLEN + info->options_len,
 903                                     netif_is_any_bridge_port(dev));
 904         if (err < 0) {
 905                 dst_release(&rt->dst);
 906                 return err;
 907         } else if (err) {
 908                 struct ip_tunnel_info *info;
 909
 910                 info = skb_tunnel_info(skb);
 911                 if (info) {
 912                         info->key.u.ipv4.dst = fl4.saddr;
 913                         info->key.u.ipv4.src = fl4.daddr;
 914                 }
 915
 916                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 917                         dst_release(&rt->dst);
 918                         return -EINVAL;
 919                 }
 920
 921                 skb->protocol = eth_type_trans(skb, geneve->dev);
 922                 netif_rx(skb);
 923                 dst_release(&rt->dst);
 924                 return -EMSGSIZE;
 925         }
 926
 927         if (geneve->cfg.collect_md) {
 928                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 929                 ttl = key->ttl;
 930
 931                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 932         } else {
 933                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 934                 if (geneve->cfg.ttl_inherit)
 935                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 936                 else
 937                         ttl = key->ttl;
 938                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 939
 940                 if (geneve->cfg.df == GENEVE_DF_SET) {
 941                         df = htons(IP_DF);
 942                 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
 943                         struct ethhdr *eth = eth_hdr(skb);
 944
 945                         if (ntohs(eth->h_proto) == ETH_P_IPV6) {
 946                                 df = htons(IP_DF);
 947                         } else if (ntohs(eth->h_proto) == ETH_P_IP) {
 948                                 struct iphdr *iph = ip_hdr(skb);
 949
 950                                 if (iph->frag_off & htons(IP_DF))
 951                                         df = htons(IP_DF);
 952                         }
 953                 }
 954         }
 955
 956         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
 957         if (unlikely(err))
 958                 return err;
 959
 960         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 961                             tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
 962                             !net_eq(geneve->net, dev_net(geneve->dev)),
 963                             !(info->key.tun_flags & TUNNEL_CSUM));
 964         return 0;
 965 }
 966
 967 #if IS_ENABLED(CONFIG_IPV6)
 968 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 969                             struct geneve_dev *geneve,
 970                             const struct ip_tunnel_info *info)
 971 {
 972         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 973         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 974         const struct ip_tunnel_key *key = &info->key;
 975         struct dst_entry *dst = NULL;
 976         struct flowi6 fl6;
 977         __u8 prio, ttl;
 978         __be16 sport;
 979         int err;
 980
 981         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 982         dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
 983                                 geneve->cfg.info.key.tp_dst, sport);
 984         if (IS_ERR(dst))
 985                 return PTR_ERR(dst);
 986
 987         err = skb_tunnel_check_pmtu(skb, dst,
 988                                     GENEVE_IPV6_HLEN + info->options_len,
 989                                     netif_is_any_bridge_port(dev));
 990         if (err < 0) {
 991                 dst_release(dst);
 992                 return err;
 993         } else if (err) {
 994                 struct ip_tunnel_info *info = skb_tunnel_info(skb);
 995
 996                 if (info) {
 997                         info->key.u.ipv6.dst = fl6.saddr;
 998                         info->key.u.ipv6.src = fl6.daddr;
 999                 }
1000
1001                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1002                         dst_release(dst);
1003                         return -EINVAL;
1004                 }
1005
1006                 skb->protocol = eth_type_trans(skb, geneve->dev);
1007                 netif_rx(skb);
1008                 dst_release(dst);
1009                 return -EMSGSIZE;
1010         }
1011
1012         if (geneve->cfg.collect_md) {
1013                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1014                 ttl = key->ttl;
1015         } else {
1016                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1017                                            ip_hdr(skb), skb);
1018                 if (geneve->cfg.ttl_inherit)
1019                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1020                 else
1021                         ttl = key->ttl;
1022                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1023         }
1024         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
1025         if (unlikely(err))
1026                 return err;
1027
1028         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1029                              &fl6.saddr, &fl6.daddr, prio, ttl,
1030                              info->key.label, sport, geneve->cfg.info.key.tp_dst,
1031                              !(info->key.tun_flags & TUNNEL_CSUM));
1032         return 0;
1033 }
1034 #endif
1035
1036 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1037 {
1038         struct geneve_dev *geneve = netdev_priv(dev);
1039         struct ip_tunnel_info *info = NULL;
1040         int err;
1041
1042         if (geneve->cfg.collect_md) {
1043                 info = skb_tunnel_info(skb);
1044                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1045                         netdev_dbg(dev, "no tunnel metadata\n");
1046                         dev_kfree_skb(skb);
1047                         dev->stats.tx_dropped++;
1048                         return NETDEV_TX_OK;
1049                 }
1050         } else {
1051                 info = &geneve->cfg.info;
1052         }
1053
1054         rcu_read_lock();
1055 #if IS_ENABLED(CONFIG_IPV6)
1056         if (info->mode & IP_TUNNEL_INFO_IPV6)
1057                 err = geneve6_xmit_skb(skb, dev, geneve, info);
1058         else
1059 #endif
1060                 err = geneve_xmit_skb(skb, dev, geneve, info);
1061         rcu_read_unlock();
1062
1063         if (likely(!err))
1064                 return NETDEV_TX_OK;
1065
1066         if (err != -EMSGSIZE)
1067                 dev_kfree_skb(skb);
1068
1069         if (err == -ELOOP)
1070                 dev->stats.collisions++;
1071         else if (err == -ENETUNREACH)
1072                 dev->stats.tx_carrier_errors++;
1073
1074         dev->stats.tx_errors++;
1075         return NETDEV_TX_OK;
1076 }
1077
1078 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1079 {
1080         if (new_mtu > dev->max_mtu)
1081                 new_mtu = dev->max_mtu;
1082         else if (new_mtu < dev->min_mtu)
1083                 new_mtu = dev->min_mtu;
1084
1085         dev->mtu = new_mtu;
1086         return 0;
1087 }
1088
1089 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1090 {
1091         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1092         struct geneve_dev *geneve = netdev_priv(dev);
1093         __be16 sport;
1094
1095         if (ip_tunnel_info_af(info) == AF_INET) {
1096                 struct rtable *rt;
1097                 struct flowi4 fl4;
1098
1099                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1100                 sport = udp_flow_src_port(geneve->net, skb,
1101                                           1, USHRT_MAX, true);
1102
1103                 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1104                                       geneve->cfg.info.key.tp_dst, sport);
1105                 if (IS_ERR(rt))
1106                         return PTR_ERR(rt);
1107
1108                 ip_rt_put(rt);
1109                 info->key.u.ipv4.src = fl4.saddr;
1110 #if IS_ENABLED(CONFIG_IPV6)
1111         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1112                 struct dst_entry *dst;
1113                 struct flowi6 fl6;
1114
1115                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1116                 sport = udp_flow_src_port(geneve->net, skb,
1117                                           1, USHRT_MAX, true);
1118
1119                 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1120                                         geneve->cfg.info.key.tp_dst, sport);
1121                 if (IS_ERR(dst))
1122                         return PTR_ERR(dst);
1123
1124                 dst_release(dst);
1125                 info->key.u.ipv6.src = fl6.saddr;
1126 #endif
1127         } else {
1128                 return -EINVAL;
1129         }
1130
1131         info->key.tp_src = sport;
1132         info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1133         return 0;
1134 }
1135
1136 static const struct net_device_ops geneve_netdev_ops = {
1137         .ndo_init               = geneve_init,
1138         .ndo_uninit             = geneve_uninit,
1139         .ndo_open               = geneve_open,
1140         .ndo_stop               = geneve_stop,
1141         .ndo_start_xmit         = geneve_xmit,
1142         .ndo_get_stats64        = dev_get_tstats64,
1143         .ndo_change_mtu         = geneve_change_mtu,
1144         .ndo_validate_addr      = eth_validate_addr,
1145         .ndo_set_mac_address    = eth_mac_addr,
1146         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1147 };
1148
1149 static void geneve_get_drvinfo(struct net_device *dev,
1150                                struct ethtool_drvinfo *drvinfo)
1151 {
1152         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1153         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1154 }
1155
1156 static const struct ethtool_ops geneve_ethtool_ops = {
1157         .get_drvinfo    = geneve_get_drvinfo,
1158         .get_link       = ethtool_op_get_link,
1159 };
1160
1161 /* Info for udev, that this is a virtual tunnel endpoint */
1162 static struct device_type geneve_type = {
1163         .name = "geneve",
1164 };
1165
1166 /* Calls the ndo_udp_tunnel_add of the caller in order to
1167  * supply the listening GENEVE udp ports. Callers are expected
1168  * to implement the ndo_udp_tunnel_add.
1169  */
1170 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1171 {
1172         struct net *net = dev_net(dev);
1173         struct geneve_net *gn = net_generic(net, geneve_net_id);
1174         struct geneve_sock *gs;
1175
1176         rcu_read_lock();
1177         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1178                 if (push) {
1179                         udp_tunnel_push_rx_port(dev, gs->sock,
1180                                                 UDP_TUNNEL_TYPE_GENEVE);
1181                 } else {
1182                         udp_tunnel_drop_rx_port(dev, gs->sock,
1183                                                 UDP_TUNNEL_TYPE_GENEVE);
1184                 }
1185         }
1186         rcu_read_unlock();
1187 }
1188
1189 /* Initialize the device structure. */
1190 static void geneve_setup(struct net_device *dev)
1191 {
1192         ether_setup(dev);
1193
1194         dev->netdev_ops = &geneve_netdev_ops;
1195         dev->ethtool_ops = &geneve_ethtool_ops;
1196         dev->needs_free_netdev = true;
1197
1198         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1199
1200         dev->features    |= NETIF_F_LLTX;
1201         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1202         dev->features    |= NETIF_F_RXCSUM;
1203         dev->features    |= NETIF_F_GSO_SOFTWARE;
1204
1205         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1206         dev->hw_features |= NETIF_F_RXCSUM;
1207         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1208
1209         /* MTU range: 68 - (something less than 65535) */
1210         dev->min_mtu = ETH_MIN_MTU;
1211         /* The max_mtu calculation does not take account of GENEVE
1212          * options, to avoid excluding potentially valid
1213          * configurations. This will be further reduced by IPvX hdr size.
1214          */
1215         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1216
1217         netif_keep_dst(dev);
1218         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1219         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1220         eth_hw_addr_random(dev);
1221 }
1222
1223 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1224         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1225         [IFLA_GENEVE_REMOTE]            = { .len = sizeof_field(struct iphdr, daddr) },
1226         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1227         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1228         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1229         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1230         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1231         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1232         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1233         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1234         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1235         [IFLA_GENEVE_TTL_INHERIT]       = { .type = NLA_U8 },
1236         [IFLA_GENEVE_DF]                = { .type = NLA_U8 },
1237 };
1238
1239 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1240                            struct netlink_ext_ack *extack)
1241 {
1242         if (tb[IFLA_ADDRESS]) {
1243                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1244                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1245                                             "Provided link layer address is not Ethernet");
1246                         return -EINVAL;
1247                 }
1248
1249                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1250                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1251                                             "Provided Ethernet address is not unicast");
1252                         return -EADDRNOTAVAIL;
1253                 }
1254         }
1255
1256         if (!data) {
1257                 NL_SET_ERR_MSG(extack,
1258                                "Not enough attributes provided to perform the operation");
1259                 return -EINVAL;
1260         }
1261
1262         if (data[IFLA_GENEVE_ID]) {
1263                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1264
1265                 if (vni >= GENEVE_N_VID) {
1266                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1267                                             "Geneve ID must be lower than 16777216");
1268                         return -ERANGE;
1269                 }
1270         }
1271
1272         if (data[IFLA_GENEVE_DF]) {
1273                 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1274
1275                 if (df < 0 || df > GENEVE_DF_MAX) {
1276                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1277                                             "Invalid DF attribute");
1278                         return -EINVAL;
1279                 }
1280         }
1281
1282         return 0;
1283 }
1284
1285 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1286                                           const struct ip_tunnel_info *info,
1287                                           bool *tun_on_same_port,
1288                                           bool *tun_collect_md)
1289 {
1290         struct geneve_dev *geneve, *t = NULL;
1291
1292         *tun_on_same_port = false;
1293         *tun_collect_md = false;
1294         list_for_each_entry(geneve, &gn->geneve_list, next) {
1295                 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1296                         *tun_collect_md = geneve->cfg.collect_md;
1297                         *tun_on_same_port = true;
1298                 }
1299                 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1300                     info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1301                     !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1302                         t = geneve;
1303         }
1304         return t;
1305 }
1306
1307 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1308 {
1309         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1310                  info->key.ttl || info->key.label || info->key.tp_src ||
1311                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1312 }
1313
1314 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1315                                   struct ip_tunnel_info *b)
1316 {
1317         if (ip_tunnel_info_af(a) == AF_INET)
1318                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1319         else
1320                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1321 }
1322
1323 static int geneve_configure(struct net *net, struct net_device *dev,
1324                             struct netlink_ext_ack *extack,
1325                             const struct geneve_config *cfg)
1326 {
1327         struct geneve_net *gn = net_generic(net, geneve_net_id);
1328         struct geneve_dev *t, *geneve = netdev_priv(dev);
1329         const struct ip_tunnel_info *info = &cfg->info;
1330         bool tun_collect_md, tun_on_same_port;
1331         int err, encap_len;
1332
1333         if (cfg->collect_md && !is_tnl_info_zero(info)) {
1334                 NL_SET_ERR_MSG(extack,
1335                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1336                 return -EINVAL;
1337         }
1338
1339         geneve->net = net;
1340         geneve->dev = dev;
1341
1342         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1343         if (t)
1344                 return -EBUSY;
1345
1346         /* make enough headroom for basic scenario */
1347         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1348         if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1349                 encap_len += sizeof(struct iphdr);
1350                 dev->max_mtu -= sizeof(struct iphdr);
1351         } else {
1352                 encap_len += sizeof(struct ipv6hdr);
1353                 dev->max_mtu -= sizeof(struct ipv6hdr);
1354         }
1355         dev->needed_headroom = encap_len + ETH_HLEN;
1356
1357         if (cfg->collect_md) {
1358                 if (tun_on_same_port) {
1359                         NL_SET_ERR_MSG(extack,
1360                                        "There can be only one externally controlled device on a destination port");
1361                         return -EPERM;
1362                 }
1363         } else {
1364                 if (tun_collect_md) {
1365                         NL_SET_ERR_MSG(extack,
1366                                        "There already exists an externally controlled device on this destination port");
1367                         return -EPERM;
1368                 }
1369         }
1370
1371         dst_cache_reset(&geneve->cfg.info.dst_cache);
1372         memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1373
1374         err = register_netdevice(dev);
1375         if (err)
1376                 return err;
1377
1378         list_add(&geneve->next, &gn->geneve_list);
1379         return 0;
1380 }
1381
1382 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1383 {
1384         memset(info, 0, sizeof(*info));
1385         info->key.tp_dst = htons(dst_port);
1386 }
1387
1388 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1389                           struct netlink_ext_ack *extack,
1390                           struct geneve_config *cfg, bool changelink)
1391 {
1392         struct ip_tunnel_info *info = &cfg->info;
1393         int attrtype;
1394
1395         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1396                 NL_SET_ERR_MSG(extack,
1397                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1398                 return -EINVAL;
1399         }
1400
1401         if (data[IFLA_GENEVE_REMOTE]) {
1402                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1403                         attrtype = IFLA_GENEVE_REMOTE;
1404                         goto change_notsup;
1405                 }
1406
1407                 info->key.u.ipv4.dst =
1408                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1409
1410                 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1411                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1412                                             "Remote IPv4 address cannot be Multicast");
1413                         return -EINVAL;
1414                 }
1415         }
1416
1417         if (data[IFLA_GENEVE_REMOTE6]) {
1418 #if IS_ENABLED(CONFIG_IPV6)
1419                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1420                         attrtype = IFLA_GENEVE_REMOTE6;
1421                         goto change_notsup;
1422                 }
1423
1424                 info->mode = IP_TUNNEL_INFO_IPV6;
1425                 info->key.u.ipv6.dst =
1426                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1427
1428                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1429                     IPV6_ADDR_LINKLOCAL) {
1430                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1431                                             "Remote IPv6 address cannot be link-local");
1432                         return -EINVAL;
1433                 }
1434                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1435                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1436                                             "Remote IPv6 address cannot be Multicast");
1437                         return -EINVAL;
1438                 }
1439                 info->key.tun_flags |= TUNNEL_CSUM;
1440                 cfg->use_udp6_rx_checksums = true;
1441 #else
1442                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1443                                     "IPv6 support not enabled in the kernel");
1444                 return -EPFNOSUPPORT;
1445 #endif
1446         }
1447
1448         if (data[IFLA_GENEVE_ID]) {
1449                 __u32 vni;
1450                 __u8 tvni[3];
1451                 __be64 tunid;
1452
1453                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1454                 tvni[0] = (vni & 0x00ff0000) >> 16;
1455                 tvni[1] = (vni & 0x0000ff00) >> 8;
1456                 tvni[2] =  vni & 0x000000ff;
1457
1458                 tunid = vni_to_tunnel_id(tvni);
1459                 if (changelink && (tunid != info->key.tun_id)) {
1460                         attrtype = IFLA_GENEVE_ID;
1461                         goto change_notsup;
1462                 }
1463                 info->key.tun_id = tunid;
1464         }
1465
1466         if (data[IFLA_GENEVE_TTL_INHERIT]) {
1467                 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1468                         cfg->ttl_inherit = true;
1469                 else
1470                         cfg->ttl_inherit = false;
1471         } else if (data[IFLA_GENEVE_TTL]) {
1472                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1473                 cfg->ttl_inherit = false;
1474         }
1475
1476         if (data[IFLA_GENEVE_TOS])
1477                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1478
1479         if (data[IFLA_GENEVE_DF])
1480                 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1481
1482         if (data[IFLA_GENEVE_LABEL]) {
1483                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1484                                   IPV6_FLOWLABEL_MASK;
1485                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1486                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1487                                             "Label attribute only applies for IPv6 Geneve devices");
1488                         return -EINVAL;
1489                 }
1490         }
1491
1492         if (data[IFLA_GENEVE_PORT]) {
1493                 if (changelink) {
1494                         attrtype = IFLA_GENEVE_PORT;
1495                         goto change_notsup;
1496                 }
1497                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1498         }
1499
1500         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1501                 if (changelink) {
1502                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1503                         goto change_notsup;
1504                 }
1505                 cfg->collect_md = true;
1506         }
1507
1508         if (data[IFLA_GENEVE_UDP_CSUM]) {
1509                 if (changelink) {
1510                         attrtype = IFLA_GENEVE_UDP_CSUM;
1511                         goto change_notsup;
1512                 }
1513                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1514                         info->key.tun_flags |= TUNNEL_CSUM;
1515         }
1516
1517         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1518 #if IS_ENABLED(CONFIG_IPV6)
1519                 if (changelink) {
1520                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1521                         goto change_notsup;
1522                 }
1523                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1524                         info->key.tun_flags &= ~TUNNEL_CSUM;
1525 #else
1526                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1527                                     "IPv6 support not enabled in the kernel");
1528                 return -EPFNOSUPPORT;
1529 #endif
1530         }
1531
1532         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1533 #if IS_ENABLED(CONFIG_IPV6)
1534                 if (changelink) {
1535                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1536                         goto change_notsup;
1537                 }
1538                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1539                         cfg->use_udp6_rx_checksums = false;
1540 #else
1541                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1542                                     "IPv6 support not enabled in the kernel");
1543                 return -EPFNOSUPPORT;
1544 #endif
1545         }
1546
1547         return 0;
1548 change_notsup:
1549         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1550                             "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1551         return -EOPNOTSUPP;
1552 }
1553
1554 static void geneve_link_config(struct net_device *dev,
1555                                struct ip_tunnel_info *info, struct nlattr *tb[])
1556 {
1557         struct geneve_dev *geneve = netdev_priv(dev);
1558         int ldev_mtu = 0;
1559
1560         if (tb[IFLA_MTU]) {
1561                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1562                 return;
1563         }
1564
1565         switch (ip_tunnel_info_af(info)) {
1566         case AF_INET: {
1567                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1568                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1569
1570                 if (!IS_ERR(rt) && rt->dst.dev) {
1571                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1572                         ip_rt_put(rt);
1573                 }
1574                 break;
1575         }
1576 #if IS_ENABLED(CONFIG_IPV6)
1577         case AF_INET6: {
1578                 struct rt6_info *rt;
1579
1580                 if (!__in6_dev_get(dev))
1581                         break;
1582
1583                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1584                                 NULL, 0);
1585
1586                 if (rt && rt->dst.dev)
1587                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1588                 ip6_rt_put(rt);
1589                 break;
1590         }
1591 #endif
1592         }
1593
1594         if (ldev_mtu <= 0)
1595                 return;
1596
1597         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1598 }
1599
1600 static int geneve_newlink(struct net *net, struct net_device *dev,
1601                           struct nlattr *tb[], struct nlattr *data[],
1602                           struct netlink_ext_ack *extack)
1603 {
1604         struct geneve_config cfg = {
1605                 .df = GENEVE_DF_UNSET,
1606                 .use_udp6_rx_checksums = false,
1607                 .ttl_inherit = false,
1608                 .collect_md = false,
1609         };
1610         int err;
1611
1612         init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1613         err = geneve_nl2info(tb, data, extack, &cfg, false);
1614         if (err)
1615                 return err;
1616
1617         err = geneve_configure(net, dev, extack, &cfg);
1618         if (err)
1619                 return err;
1620
1621         geneve_link_config(dev, &cfg.info, tb);
1622
1623         return 0;
1624 }
1625
1626 /* Quiesces the geneve device data path for both TX and RX.
1627  *
1628  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1629  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1630  * to complete for the existing set of in-flight packets to be transmitted,
1631  * then we would have quiesced the transmit data path. All the future packets
1632  * will get dropped until we unquiesce the data path.
1633  *
1634  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1635  * if we set that to NULL under RCU and wait for synchronize_net() to
1636  * complete, then we would have quiesced the receive data path.
1637  */
1638 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1639                            struct geneve_sock **gs6)
1640 {
1641         *gs4 = rtnl_dereference(geneve->sock4);
1642         rcu_assign_pointer(geneve->sock4, NULL);
1643         if (*gs4)
1644                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1645 #if IS_ENABLED(CONFIG_IPV6)
1646         *gs6 = rtnl_dereference(geneve->sock6);
1647         rcu_assign_pointer(geneve->sock6, NULL);
1648         if (*gs6)
1649                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1650 #else
1651         *gs6 = NULL;
1652 #endif
1653         synchronize_net();
1654 }
1655
1656 /* Resumes the geneve device data path for both TX and RX. */
1657 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1658                              struct geneve_sock __maybe_unused *gs6)
1659 {
1660         rcu_assign_pointer(geneve->sock4, gs4);
1661         if (gs4)
1662                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1663 #if IS_ENABLED(CONFIG_IPV6)
1664         rcu_assign_pointer(geneve->sock6, gs6);
1665         if (gs6)
1666                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1667 #endif
1668         synchronize_net();
1669 }
1670
1671 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1672                              struct nlattr *data[],
1673                              struct netlink_ext_ack *extack)
1674 {
1675         struct geneve_dev *geneve = netdev_priv(dev);
1676         struct geneve_sock *gs4, *gs6;
1677         struct geneve_config cfg;
1678         int err;
1679
1680         /* If the geneve device is configured for metadata (or externally
1681          * controlled, for example, OVS), then nothing can be changed.
1682          */
1683         if (geneve->cfg.collect_md)
1684                 return -EOPNOTSUPP;
1685
1686         /* Start with the existing info. */
1687         memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1688         err = geneve_nl2info(tb, data, extack, &cfg, true);
1689         if (err)
1690                 return err;
1691
1692         if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1693                 dst_cache_reset(&cfg.info.dst_cache);
1694                 geneve_link_config(dev, &cfg.info, tb);
1695         }
1696
1697         geneve_quiesce(geneve, &gs4, &gs6);
1698         memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1699         geneve_unquiesce(geneve, gs4, gs6);
1700
1701         return 0;
1702 }
1703
1704 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1705 {
1706         struct geneve_dev *geneve = netdev_priv(dev);
1707
1708         list_del(&geneve->next);
1709         unregister_netdevice_queue(dev, head);
1710 }
1711
1712 static size_t geneve_get_size(const struct net_device *dev)
1713 {
1714         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1715                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1716                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1717                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1718                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_DF */
1719                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1720                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1721                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1722                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1723                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1724                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1725                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1726                 0;
1727 }
1728
1729 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1730 {
1731         struct geneve_dev *geneve = netdev_priv(dev);
1732         struct ip_tunnel_info *info = &geneve->cfg.info;
1733         bool ttl_inherit = geneve->cfg.ttl_inherit;
1734         bool metadata = geneve->cfg.collect_md;
1735         __u8 tmp_vni[3];
1736         __u32 vni;
1737
1738         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1739         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1740         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1741                 goto nla_put_failure;
1742
1743         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1744                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1745                                     info->key.u.ipv4.dst))
1746                         goto nla_put_failure;
1747                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1748                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1749                         goto nla_put_failure;
1750
1751 #if IS_ENABLED(CONFIG_IPV6)
1752         } else if (!metadata) {
1753                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1754                                      &info->key.u.ipv6.dst))
1755                         goto nla_put_failure;
1756                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1757                                !(info->key.tun_flags & TUNNEL_CSUM)))
1758                         goto nla_put_failure;
1759 #endif
1760         }
1761
1762         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1763             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1764             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1765                 goto nla_put_failure;
1766
1767         if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1768                 goto nla_put_failure;
1769
1770         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1771                 goto nla_put_failure;
1772
1773         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1774                 goto nla_put_failure;
1775
1776 #if IS_ENABLED(CONFIG_IPV6)
1777         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1778                        !geneve->cfg.use_udp6_rx_checksums))
1779                 goto nla_put_failure;
1780 #endif
1781
1782         if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1783                 goto nla_put_failure;
1784
1785         return 0;
1786
1787 nla_put_failure:
1788         return -EMSGSIZE;
1789 }
1790
1791 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1792         .kind           = "geneve",
1793         .maxtype        = IFLA_GENEVE_MAX,
1794         .policy         = geneve_policy,
1795         .priv_size      = sizeof(struct geneve_dev),
1796         .setup          = geneve_setup,
1797         .validate       = geneve_validate,
1798         .newlink        = geneve_newlink,
1799         .changelink     = geneve_changelink,
1800         .dellink        = geneve_dellink,
1801         .get_size       = geneve_get_size,
1802         .fill_info      = geneve_fill_info,
1803 };
1804
1805 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1806                                         u8 name_assign_type, u16 dst_port)
1807 {
1808         struct nlattr *tb[IFLA_MAX + 1];
1809         struct net_device *dev;
1810         LIST_HEAD(list_kill);
1811         int err;
1812         struct geneve_config cfg = {
1813                 .df = GENEVE_DF_UNSET,
1814                 .use_udp6_rx_checksums = true,
1815                 .ttl_inherit = false,
1816                 .collect_md = true,
1817         };
1818
1819         memset(tb, 0, sizeof(tb));
1820         dev = rtnl_create_link(net, name, name_assign_type,
1821                                &geneve_link_ops, tb, NULL);
1822         if (IS_ERR(dev))
1823                 return dev;
1824
1825         init_tnl_info(&cfg.info, dst_port);
1826         err = geneve_configure(net, dev, NULL, &cfg);
1827         if (err) {
1828                 free_netdev(dev);
1829                 return ERR_PTR(err);
1830         }
1831
1832         /* openvswitch users expect packet sizes to be unrestricted,
1833          * so set the largest MTU we can.
1834          */
1835         err = geneve_change_mtu(dev, IP_MAX_MTU);
1836         if (err)
1837                 goto err;
1838
1839         err = rtnl_configure_link(dev, NULL);
1840         if (err < 0)
1841                 goto err;
1842
1843         return dev;
1844 err:
1845         geneve_dellink(dev, &list_kill);
1846         unregister_netdevice_many(&list_kill);
1847         return ERR_PTR(err);
1848 }
1849 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1850
1851 static int geneve_netdevice_event(struct notifier_block *unused,
1852                                   unsigned long event, void *ptr)
1853 {
1854         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1855
1856         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1857                 geneve_offload_rx_ports(dev, true);
1858         else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1859                 geneve_offload_rx_ports(dev, false);
1860
1861         return NOTIFY_DONE;
1862 }
1863
1864 static struct notifier_block geneve_notifier_block __read_mostly = {
1865         .notifier_call = geneve_netdevice_event,
1866 };
1867
1868 static __net_init int geneve_init_net(struct net *net)
1869 {
1870         struct geneve_net *gn = net_generic(net, geneve_net_id);
1871
1872         INIT_LIST_HEAD(&gn->geneve_list);
1873         INIT_LIST_HEAD(&gn->sock_list);
1874         return 0;
1875 }
1876
1877 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1878 {
1879         struct geneve_net *gn = net_generic(net, geneve_net_id);
1880         struct geneve_dev *geneve, *next;
1881         struct net_device *dev, *aux;
1882
1883         /* gather any geneve devices that were moved into this ns */
1884         for_each_netdev_safe(net, dev, aux)
1885                 if (dev->rtnl_link_ops == &geneve_link_ops)
1886                         unregister_netdevice_queue(dev, head);
1887
1888         /* now gather any other geneve devices that were created in this ns */
1889         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1890                 /* If geneve->dev is in the same netns, it was already added
1891                  * to the list by the previous loop.
1892                  */
1893                 if (!net_eq(dev_net(geneve->dev), net))
1894                         unregister_netdevice_queue(geneve->dev, head);
1895         }
1896 }
1897
1898 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1899 {
1900         struct net *net;
1901         LIST_HEAD(list);
1902
1903         rtnl_lock();
1904         list_for_each_entry(net, net_list, exit_list)
1905                 geneve_destroy_tunnels(net, &list);
1906
1907         /* unregister the devices gathered above */
1908         unregister_netdevice_many(&list);
1909         rtnl_unlock();
1910
1911         list_for_each_entry(net, net_list, exit_list) {
1912                 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1913
1914                 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1915         }
1916 }
1917
1918 static struct pernet_operations geneve_net_ops = {
1919         .init = geneve_init_net,
1920         .exit_batch = geneve_exit_batch_net,
1921         .id   = &geneve_net_id,
1922         .size = sizeof(struct geneve_net),
1923 };
1924
1925 static int __init geneve_init_module(void)
1926 {
1927         int rc;
1928
1929         rc = register_pernet_subsys(&geneve_net_ops);
1930         if (rc)
1931                 goto out1;
1932
1933         rc = register_netdevice_notifier(&geneve_notifier_block);
1934         if (rc)
1935                 goto out2;
1936
1937         rc = rtnl_link_register(&geneve_link_ops);
1938         if (rc)
1939                 goto out3;
1940
1941         return 0;
1942 out3:
1943         unregister_netdevice_notifier(&geneve_notifier_block);
1944 out2:
1945         unregister_pernet_subsys(&geneve_net_ops);
1946 out1:
1947         return rc;
1948 }
1949 late_initcall(geneve_init_module);
1950
1951 static void __exit geneve_cleanup_module(void)
1952 {
1953         rtnl_link_unregister(&geneve_link_ops);
1954         unregister_netdevice_notifier(&geneve_notifier_block);
1955         unregister_pernet_subsys(&geneve_net_ops);
1956 }
1957 module_exit(geneve_cleanup_module);
1958
1959 MODULE_LICENSE("GPL");
1960 MODULE_VERSION(GENEVE_NETDEV_VER);
1961 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1962 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1963 MODULE_ALIAS_RTNL_LINK("geneve");