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         return sock;
 465 }
 466
 467 static int geneve_hlen(struct genevehdr *gh)
 468 {
 469         return sizeof(*gh) + gh->opt_len * 4;
 470 }
 471
 472 static struct sk_buff *geneve_gro_receive(struct sock *sk,
 473                                           struct list_head *head,
 474                                           struct sk_buff *skb)
 475 {
 476         struct sk_buff *pp = NULL;
 477         struct sk_buff *p;
 478         struct genevehdr *gh, *gh2;
 479         unsigned int hlen, gh_len, off_gnv;
 480         const struct packet_offload *ptype;
 481         __be16 type;
 482         int flush = 1;
 483
 484         off_gnv = skb_gro_offset(skb);
 485         hlen = off_gnv + sizeof(*gh);
 486         gh = skb_gro_header_fast(skb, off_gnv);
 487         if (skb_gro_header_hard(skb, hlen)) {
 488                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 489                 if (unlikely(!gh))
 490                         goto out;
 491         }
 492
 493         if (gh->ver != GENEVE_VER || gh->oam)
 494                 goto out;
 495         gh_len = geneve_hlen(gh);
 496
 497         hlen = off_gnv + gh_len;
 498         if (skb_gro_header_hard(skb, hlen)) {
 499                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 500                 if (unlikely(!gh))
 501                         goto out;
 502         }
 503
 504         list_for_each_entry(p, head, list) {
 505                 if (!NAPI_GRO_CB(p)->same_flow)
 506                         continue;
 507
 508                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 509                 if (gh->opt_len != gh2->opt_len ||
 510                     memcmp(gh, gh2, gh_len)) {
 511                         NAPI_GRO_CB(p)->same_flow = 0;
 512                         continue;
 513                 }
 514         }
 515
 516         type = gh->proto_type;
 517
 518         rcu_read_lock();
 519         ptype = gro_find_receive_by_type(type);
 520         if (!ptype)
 521                 goto out_unlock;
 522
 523         skb_gro_pull(skb, gh_len);
 524         skb_gro_postpull_rcsum(skb, gh, gh_len);
 525         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 526         flush = 0;
 527
 528 out_unlock:
 529         rcu_read_unlock();
 530 out:
 531         skb_gro_flush_final(skb, pp, flush);
 532
 533         return pp;
 534 }
 535
 536 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 537                                int nhoff)
 538 {
 539         struct genevehdr *gh;
 540         struct packet_offload *ptype;
 541         __be16 type;
 542         int gh_len;
 543         int err = -ENOSYS;
 544
 545         gh = (struct genevehdr *)(skb->data + nhoff);
 546         gh_len = geneve_hlen(gh);
 547         type = gh->proto_type;
 548
 549         rcu_read_lock();
 550         ptype = gro_find_complete_by_type(type);
 551         if (ptype)
 552                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 553
 554         rcu_read_unlock();
 555
 556         skb_set_inner_mac_header(skb, nhoff + gh_len);
 557
 558         return err;
 559 }
 560
 561 /* Create new listen socket if needed */
 562 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 563                                                 bool ipv6, bool ipv6_rx_csum)
 564 {
 565         struct geneve_net *gn = net_generic(net, geneve_net_id);
 566         struct geneve_sock *gs;
 567         struct socket *sock;
 568         struct udp_tunnel_sock_cfg tunnel_cfg;
 569         int h;
 570
 571         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 572         if (!gs)
 573                 return ERR_PTR(-ENOMEM);
 574
 575         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 576         if (IS_ERR(sock)) {
 577                 kfree(gs);
 578                 return ERR_CAST(sock);
 579         }
 580
 581         gs->sock = sock;
 582         gs->refcnt = 1;
 583         for (h = 0; h < VNI_HASH_SIZE; ++h)
 584                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 585
 586         /* Initialize the geneve udp offloads structure */
 587         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 588
 589         /* Mark socket as an encapsulation socket */
 590         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 591         tunnel_cfg.sk_user_data = gs;
 592         tunnel_cfg.encap_type = 1;
 593         tunnel_cfg.gro_receive = geneve_gro_receive;
 594         tunnel_cfg.gro_complete = geneve_gro_complete;
 595         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 596         tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
 597         tunnel_cfg.encap_destroy = NULL;
 598         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 599         list_add(&gs->list, &gn->sock_list);
 600         return gs;
 601 }
 602
 603 static void __geneve_sock_release(struct geneve_sock *gs)
 604 {
 605         if (!gs || --gs->refcnt)
 606                 return;
 607
 608         list_del(&gs->list);
 609         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 610         udp_tunnel_sock_release(gs->sock);
 611         kfree_rcu(gs, rcu);
 612 }
 613
 614 static void geneve_sock_release(struct geneve_dev *geneve)
 615 {
 616         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 617 #if IS_ENABLED(CONFIG_IPV6)
 618         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 619
 620         rcu_assign_pointer(geneve->sock6, NULL);
 621 #endif
 622
 623         rcu_assign_pointer(geneve->sock4, NULL);
 624         synchronize_net();
 625
 626         __geneve_sock_release(gs4);
 627 #if IS_ENABLED(CONFIG_IPV6)
 628         __geneve_sock_release(gs6);
 629 #endif
 630 }
 631
 632 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 633                                             sa_family_t family,
 634                                             __be16 dst_port)
 635 {
 636         struct geneve_sock *gs;
 637
 638         list_for_each_entry(gs, &gn->sock_list, list) {
 639                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 640                     geneve_get_sk_family(gs) == family) {
 641                         return gs;
 642                 }
 643         }
 644         return NULL;
 645 }
 646
 647 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 648 {
 649         struct net *net = geneve->net;
 650         struct geneve_net *gn = net_generic(net, geneve_net_id);
 651         struct geneve_dev_node *node;
 652         struct geneve_sock *gs;
 653         __u8 vni[3];
 654         __u32 hash;
 655
 656         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
 657         if (gs) {
 658                 gs->refcnt++;
 659                 goto out;
 660         }
 661
 662         gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
 663                                   geneve->cfg.use_udp6_rx_checksums);
 664         if (IS_ERR(gs))
 665                 return PTR_ERR(gs);
 666
 667 out:
 668         gs->collect_md = geneve->cfg.collect_md;
 669 #if IS_ENABLED(CONFIG_IPV6)
 670         if (ipv6) {
 671                 rcu_assign_pointer(geneve->sock6, gs);
 672                 node = &geneve->hlist6;
 673         } else
 674 #endif
 675         {
 676                 rcu_assign_pointer(geneve->sock4, gs);
 677                 node = &geneve->hlist4;
 678         }
 679         node->geneve = geneve;
 680
 681         tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
 682         hash = geneve_net_vni_hash(vni);
 683         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 684         return 0;
 685 }
 686
 687 static int geneve_open(struct net_device *dev)
 688 {
 689         struct geneve_dev *geneve = netdev_priv(dev);
 690         bool metadata = geneve->cfg.collect_md;
 691         bool ipv4, ipv6;
 692         int ret = 0;
 693
 694         ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 695         ipv4 = !ipv6 || metadata;
 696 #if IS_ENABLED(CONFIG_IPV6)
 697         if (ipv6) {
 698                 ret = geneve_sock_add(geneve, true);
 699                 if (ret < 0 && ret != -EAFNOSUPPORT)
 700                         ipv4 = false;
 701         }
 702 #endif
 703         if (ipv4)
 704                 ret = geneve_sock_add(geneve, false);
 705         if (ret < 0)
 706                 geneve_sock_release(geneve);
 707
 708         return ret;
 709 }
 710
 711 static int geneve_stop(struct net_device *dev)
 712 {
 713         struct geneve_dev *geneve = netdev_priv(dev);
 714
 715         hlist_del_init_rcu(&geneve->hlist4.hlist);
 716 #if IS_ENABLED(CONFIG_IPV6)
 717         hlist_del_init_rcu(&geneve->hlist6.hlist);
 718 #endif
 719         geneve_sock_release(geneve);
 720         return 0;
 721 }
 722
 723 static void geneve_build_header(struct genevehdr *geneveh,
 724                                 const struct ip_tunnel_info *info)
 725 {
 726         geneveh->ver = GENEVE_VER;
 727         geneveh->opt_len = info->options_len / 4;
 728         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 729         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 730         geneveh->rsvd1 = 0;
 731         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 732         geneveh->proto_type = htons(ETH_P_TEB);
 733         geneveh->rsvd2 = 0;
 734
 735         if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
 736                 ip_tunnel_info_opts_get(geneveh->options, info);
 737 }
 738
 739 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 740                             const struct ip_tunnel_info *info,
 741                             bool xnet, int ip_hdr_len)
 742 {
 743         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 744         struct genevehdr *gnvh;
 745         int min_headroom;
 746         int err;
 747
 748         skb_reset_mac_header(skb);
 749         skb_scrub_packet(skb, xnet);
 750
 751         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 752                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 753         err = skb_cow_head(skb, min_headroom);
 754         if (unlikely(err))
 755                 goto free_dst;
 756
 757         err = udp_tunnel_handle_offloads(skb, udp_sum);
 758         if (err)
 759                 goto free_dst;
 760
 761         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 762         geneve_build_header(gnvh, info);
 763         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 764         return 0;
 765
 766 free_dst:
 767         dst_release(dst);
 768         return err;
 769 }
 770
 771 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 772                                        struct net_device *dev,
 773                                        struct geneve_sock *gs4,
 774                                        struct flowi4 *fl4,
 775                                        const struct ip_tunnel_info *info,
 776                                        __be16 dport, __be16 sport)
 777 {
 778         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 779         struct geneve_dev *geneve = netdev_priv(dev);
 780         struct dst_cache *dst_cache;
 781         struct rtable *rt = NULL;
 782         __u8 tos;
 783
 784         if (!gs4)
 785                 return ERR_PTR(-EIO);
 786
 787         memset(fl4, 0, sizeof(*fl4));
 788         fl4->flowi4_mark = skb->mark;
 789         fl4->flowi4_proto = IPPROTO_UDP;
 790         fl4->daddr = info->key.u.ipv4.dst;
 791         fl4->saddr = info->key.u.ipv4.src;
 792         fl4->fl4_dport = dport;
 793         fl4->fl4_sport = sport;
 794
 795         tos = info->key.tos;
 796         if ((tos == 1) && !geneve->cfg.collect_md) {
 797                 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 798                 use_cache = false;
 799         }
 800         fl4->flowi4_tos = RT_TOS(tos);
 801
 802         dst_cache = (struct dst_cache *)&info->dst_cache;
 803         if (use_cache) {
 804                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 805                 if (rt)
 806                         return rt;
 807         }
 808         rt = ip_route_output_key(geneve->net, fl4);
 809         if (IS_ERR(rt)) {
 810                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 811                 return ERR_PTR(-ENETUNREACH);
 812         }
 813         if (rt->dst.dev == dev) { /* is this necessary? */
 814                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 815                 ip_rt_put(rt);
 816                 return ERR_PTR(-ELOOP);
 817         }
 818         if (use_cache)
 819                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 820         return rt;
 821 }
 822
 823 #if IS_ENABLED(CONFIG_IPV6)
 824 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 825                                            struct net_device *dev,
 826                                            struct geneve_sock *gs6,
 827                                            struct flowi6 *fl6,
 828                                            const struct ip_tunnel_info *info,
 829                                            __be16 dport, __be16 sport)
 830 {
 831         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 832         struct geneve_dev *geneve = netdev_priv(dev);
 833         struct dst_entry *dst = NULL;
 834         struct dst_cache *dst_cache;
 835         __u8 prio;
 836
 837         if (!gs6)
 838                 return ERR_PTR(-EIO);
 839
 840         memset(fl6, 0, sizeof(*fl6));
 841         fl6->flowi6_mark = skb->mark;
 842         fl6->flowi6_proto = IPPROTO_UDP;
 843         fl6->daddr = info->key.u.ipv6.dst;
 844         fl6->saddr = info->key.u.ipv6.src;
 845         fl6->fl6_dport = dport;
 846         fl6->fl6_sport = sport;
 847
 848         prio = info->key.tos;
 849         if ((prio == 1) && !geneve->cfg.collect_md) {
 850                 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 851                 use_cache = false;
 852         }
 853
 854         fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 855                                            info->key.label);
 856         dst_cache = (struct dst_cache *)&info->dst_cache;
 857         if (use_cache) {
 858                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 859                 if (dst)
 860                         return dst;
 861         }
 862         dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
 863                                               NULL);
 864         if (IS_ERR(dst)) {
 865                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 866                 return ERR_PTR(-ENETUNREACH);
 867         }
 868         if (dst->dev == dev) { /* is this necessary? */
 869                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 870                 dst_release(dst);
 871                 return ERR_PTR(-ELOOP);
 872         }
 873
 874         if (use_cache)
 875                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 876         return dst;
 877 }
 878 #endif
 879
 880 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 881                            struct geneve_dev *geneve,
 882                            const struct ip_tunnel_info *info)
 883 {
 884         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 885         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 886         const struct ip_tunnel_key *key = &info->key;
 887         struct rtable *rt;
 888         struct flowi4 fl4;
 889         __u8 tos, ttl;
 890         __be16 df = 0;
 891         __be16 sport;
 892         int err;
 893
 894         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 895         rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
 896                               geneve->cfg.info.key.tp_dst, sport);
 897         if (IS_ERR(rt))
 898                 return PTR_ERR(rt);
 899
 900         err = skb_tunnel_check_pmtu(skb, &rt->dst,
 901                                     GENEVE_IPV4_HLEN + info->options_len,
 902                                     netif_is_any_bridge_port(dev));
 903         if (err < 0) {
 904                 dst_release(&rt->dst);
 905                 return err;
 906         } else if (err) {
 907                 struct ip_tunnel_info *info;
 908
 909                 info = skb_tunnel_info(skb);
 910                 if (info) {
 911                         info->key.u.ipv4.dst = fl4.saddr;
 912                         info->key.u.ipv4.src = fl4.daddr;
 913                 }
 914
 915                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 916                         dst_release(&rt->dst);
 917                         return -EINVAL;
 918                 }
 919
 920                 skb->protocol = eth_type_trans(skb, geneve->dev);
 921                 netif_rx(skb);
 922                 dst_release(&rt->dst);
 923                 return -EMSGSIZE;
 924         }
 925
 926         if (geneve->cfg.collect_md) {
 927                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 928                 ttl = key->ttl;
 929
 930                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 931         } else {
 932                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 933                 if (geneve->cfg.ttl_inherit)
 934                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 935                 else
 936                         ttl = key->ttl;
 937                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 938
 939                 if (geneve->cfg.df == GENEVE_DF_SET) {
 940                         df = htons(IP_DF);
 941                 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
 942                         struct ethhdr *eth = eth_hdr(skb);
 943
 944                         if (ntohs(eth->h_proto) == ETH_P_IPV6) {
 945                                 df = htons(IP_DF);
 946                         } else if (ntohs(eth->h_proto) == ETH_P_IP) {
 947                                 struct iphdr *iph = ip_hdr(skb);
 948
 949                                 if (iph->frag_off & htons(IP_DF))
 950                                         df = htons(IP_DF);
 951                         }
 952                 }
 953         }
 954
 955         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
 956         if (unlikely(err))
 957                 return err;
 958
 959         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 960                             tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
 961                             !net_eq(geneve->net, dev_net(geneve->dev)),
 962                             !(info->key.tun_flags & TUNNEL_CSUM));
 963         return 0;
 964 }
 965
 966 #if IS_ENABLED(CONFIG_IPV6)
 967 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 968                             struct geneve_dev *geneve,
 969                             const struct ip_tunnel_info *info)
 970 {
 971         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 972         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 973         const struct ip_tunnel_key *key = &info->key;
 974         struct dst_entry *dst = NULL;
 975         struct flowi6 fl6;
 976         __u8 prio, ttl;
 977         __be16 sport;
 978         int err;
 979
 980         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 981         dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
 982                                 geneve->cfg.info.key.tp_dst, sport);
 983         if (IS_ERR(dst))
 984                 return PTR_ERR(dst);
 985
 986         err = skb_tunnel_check_pmtu(skb, dst,
 987                                     GENEVE_IPV6_HLEN + info->options_len,
 988                                     netif_is_any_bridge_port(dev));
 989         if (err < 0) {
 990                 dst_release(dst);
 991                 return err;
 992         } else if (err) {
 993                 struct ip_tunnel_info *info = skb_tunnel_info(skb);
 994
 995                 if (info) {
 996                         info->key.u.ipv6.dst = fl6.saddr;
 997                         info->key.u.ipv6.src = fl6.daddr;
 998                 }
 999
1000                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1001                         dst_release(dst);
1002                         return -EINVAL;
1003                 }
1004
1005                 skb->protocol = eth_type_trans(skb, geneve->dev);
1006                 netif_rx(skb);
1007                 dst_release(dst);
1008                 return -EMSGSIZE;
1009         }
1010
1011         if (geneve->cfg.collect_md) {
1012                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1013                 ttl = key->ttl;
1014         } else {
1015                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1016                                            ip_hdr(skb), skb);
1017                 if (geneve->cfg.ttl_inherit)
1018                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1019                 else
1020                         ttl = key->ttl;
1021                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1022         }
1023         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
1024         if (unlikely(err))
1025                 return err;
1026
1027         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1028                              &fl6.saddr, &fl6.daddr, prio, ttl,
1029                              info->key.label, sport, geneve->cfg.info.key.tp_dst,
1030                              !(info->key.tun_flags & TUNNEL_CSUM));
1031         return 0;
1032 }
1033 #endif
1034
1035 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1036 {
1037         struct geneve_dev *geneve = netdev_priv(dev);
1038         struct ip_tunnel_info *info = NULL;
1039         int err;
1040
1041         if (geneve->cfg.collect_md) {
1042                 info = skb_tunnel_info(skb);
1043                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1044                         netdev_dbg(dev, "no tunnel metadata\n");
1045                         dev_kfree_skb(skb);
1046                         dev->stats.tx_dropped++;
1047                         return NETDEV_TX_OK;
1048                 }
1049         } else {
1050                 info = &geneve->cfg.info;
1051         }
1052
1053         rcu_read_lock();
1054 #if IS_ENABLED(CONFIG_IPV6)
1055         if (info->mode & IP_TUNNEL_INFO_IPV6)
1056                 err = geneve6_xmit_skb(skb, dev, geneve, info);
1057         else
1058 #endif
1059                 err = geneve_xmit_skb(skb, dev, geneve, info);
1060         rcu_read_unlock();
1061
1062         if (likely(!err))
1063                 return NETDEV_TX_OK;
1064
1065         if (err != -EMSGSIZE)
1066                 dev_kfree_skb(skb);
1067
1068         if (err == -ELOOP)
1069                 dev->stats.collisions++;
1070         else if (err == -ENETUNREACH)
1071                 dev->stats.tx_carrier_errors++;
1072
1073         dev->stats.tx_errors++;
1074         return NETDEV_TX_OK;
1075 }
1076
1077 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1078 {
1079         if (new_mtu > dev->max_mtu)
1080                 new_mtu = dev->max_mtu;
1081         else if (new_mtu < dev->min_mtu)
1082                 new_mtu = dev->min_mtu;
1083
1084         dev->mtu = new_mtu;
1085         return 0;
1086 }
1087
1088 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1089 {
1090         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1091         struct geneve_dev *geneve = netdev_priv(dev);
1092         __be16 sport;
1093
1094         if (ip_tunnel_info_af(info) == AF_INET) {
1095                 struct rtable *rt;
1096                 struct flowi4 fl4;
1097
1098                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1099                 sport = udp_flow_src_port(geneve->net, skb,
1100                                           1, USHRT_MAX, true);
1101
1102                 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1103                                       geneve->cfg.info.key.tp_dst, sport);
1104                 if (IS_ERR(rt))
1105                         return PTR_ERR(rt);
1106
1107                 ip_rt_put(rt);
1108                 info->key.u.ipv4.src = fl4.saddr;
1109 #if IS_ENABLED(CONFIG_IPV6)
1110         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1111                 struct dst_entry *dst;
1112                 struct flowi6 fl6;
1113
1114                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1115                 sport = udp_flow_src_port(geneve->net, skb,
1116                                           1, USHRT_MAX, true);
1117
1118                 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1119                                         geneve->cfg.info.key.tp_dst, sport);
1120                 if (IS_ERR(dst))
1121                         return PTR_ERR(dst);
1122
1123                 dst_release(dst);
1124                 info->key.u.ipv6.src = fl6.saddr;
1125 #endif
1126         } else {
1127                 return -EINVAL;
1128         }
1129
1130         info->key.tp_src = sport;
1131         info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1132         return 0;
1133 }
1134
1135 static const struct net_device_ops geneve_netdev_ops = {
1136         .ndo_init               = geneve_init,
1137         .ndo_uninit             = geneve_uninit,
1138         .ndo_open               = geneve_open,
1139         .ndo_stop               = geneve_stop,
1140         .ndo_start_xmit         = geneve_xmit,
1141         .ndo_get_stats64        = dev_get_tstats64,
1142         .ndo_change_mtu         = geneve_change_mtu,
1143         .ndo_validate_addr      = eth_validate_addr,
1144         .ndo_set_mac_address    = eth_mac_addr,
1145         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1146 };
1147
1148 static void geneve_get_drvinfo(struct net_device *dev,
1149                                struct ethtool_drvinfo *drvinfo)
1150 {
1151         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1152         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1153 }
1154
1155 static const struct ethtool_ops geneve_ethtool_ops = {
1156         .get_drvinfo    = geneve_get_drvinfo,
1157         .get_link       = ethtool_op_get_link,
1158 };
1159
1160 /* Info for udev, that this is a virtual tunnel endpoint */
1161 static struct device_type geneve_type = {
1162         .name = "geneve",
1163 };
1164
1165 /* Calls the ndo_udp_tunnel_add of the caller in order to
1166  * supply the listening GENEVE udp ports. Callers are expected
1167  * to implement the ndo_udp_tunnel_add.
1168  */
1169 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1170 {
1171         struct net *net = dev_net(dev);
1172         struct geneve_net *gn = net_generic(net, geneve_net_id);
1173         struct geneve_sock *gs;
1174
1175         rcu_read_lock();
1176         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1177                 if (push) {
1178                         udp_tunnel_push_rx_port(dev, gs->sock,
1179                                                 UDP_TUNNEL_TYPE_GENEVE);
1180                 } else {
1181                         udp_tunnel_drop_rx_port(dev, gs->sock,
1182                                                 UDP_TUNNEL_TYPE_GENEVE);
1183                 }
1184         }
1185         rcu_read_unlock();
1186 }
1187
1188 /* Initialize the device structure. */
1189 static void geneve_setup(struct net_device *dev)
1190 {
1191         ether_setup(dev);
1192
1193         dev->netdev_ops = &geneve_netdev_ops;
1194         dev->ethtool_ops = &geneve_ethtool_ops;
1195         dev->needs_free_netdev = true;
1196
1197         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1198
1199         dev->features    |= NETIF_F_LLTX;
1200         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1201         dev->features    |= NETIF_F_RXCSUM;
1202         dev->features    |= NETIF_F_GSO_SOFTWARE;
1203
1204         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1205         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1206
1207         /* MTU range: 68 - (something less than 65535) */
1208         dev->min_mtu = ETH_MIN_MTU;
1209         /* The max_mtu calculation does not take account of GENEVE
1210          * options, to avoid excluding potentially valid
1211          * configurations. This will be further reduced by IPvX hdr size.
1212          */
1213         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1214
1215         netif_keep_dst(dev);
1216         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1217         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1218         eth_hw_addr_random(dev);
1219 }
1220
1221 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1222         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1223         [IFLA_GENEVE_REMOTE]            = { .len = sizeof_field(struct iphdr, daddr) },
1224         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1225         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1226         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1227         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1228         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1229         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1230         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1231         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1232         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1233         [IFLA_GENEVE_TTL_INHERIT]       = { .type = NLA_U8 },
1234         [IFLA_GENEVE_DF]                = { .type = NLA_U8 },
1235 };
1236
1237 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1238                            struct netlink_ext_ack *extack)
1239 {
1240         if (tb[IFLA_ADDRESS]) {
1241                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1242                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1243                                             "Provided link layer address is not Ethernet");
1244                         return -EINVAL;
1245                 }
1246
1247                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1248                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1249                                             "Provided Ethernet address is not unicast");
1250                         return -EADDRNOTAVAIL;
1251                 }
1252         }
1253
1254         if (!data) {
1255                 NL_SET_ERR_MSG(extack,
1256                                "Not enough attributes provided to perform the operation");
1257                 return -EINVAL;
1258         }
1259
1260         if (data[IFLA_GENEVE_ID]) {
1261                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1262
1263                 if (vni >= GENEVE_N_VID) {
1264                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1265                                             "Geneve ID must be lower than 16777216");
1266                         return -ERANGE;
1267                 }
1268         }
1269
1270         if (data[IFLA_GENEVE_DF]) {
1271                 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1272
1273                 if (df < 0 || df > GENEVE_DF_MAX) {
1274                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1275                                             "Invalid DF attribute");
1276                         return -EINVAL;
1277                 }
1278         }
1279
1280         return 0;
1281 }
1282
1283 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1284                                           const struct ip_tunnel_info *info,
1285                                           bool *tun_on_same_port,
1286                                           bool *tun_collect_md)
1287 {
1288         struct geneve_dev *geneve, *t = NULL;
1289
1290         *tun_on_same_port = false;
1291         *tun_collect_md = false;
1292         list_for_each_entry(geneve, &gn->geneve_list, next) {
1293                 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1294                         *tun_collect_md = geneve->cfg.collect_md;
1295                         *tun_on_same_port = true;
1296                 }
1297                 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1298                     info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1299                     !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1300                         t = geneve;
1301         }
1302         return t;
1303 }
1304
1305 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1306 {
1307         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1308                  info->key.ttl || info->key.label || info->key.tp_src ||
1309                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1310 }
1311
1312 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1313                                   struct ip_tunnel_info *b)
1314 {
1315         if (ip_tunnel_info_af(a) == AF_INET)
1316                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1317         else
1318                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1319 }
1320
1321 static int geneve_configure(struct net *net, struct net_device *dev,
1322                             struct netlink_ext_ack *extack,
1323                             const struct geneve_config *cfg)
1324 {
1325         struct geneve_net *gn = net_generic(net, geneve_net_id);
1326         struct geneve_dev *t, *geneve = netdev_priv(dev);
1327         const struct ip_tunnel_info *info = &cfg->info;
1328         bool tun_collect_md, tun_on_same_port;
1329         int err, encap_len;
1330
1331         if (cfg->collect_md && !is_tnl_info_zero(info)) {
1332                 NL_SET_ERR_MSG(extack,
1333                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1334                 return -EINVAL;
1335         }
1336
1337         geneve->net = net;
1338         geneve->dev = dev;
1339
1340         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1341         if (t)
1342                 return -EBUSY;
1343
1344         /* make enough headroom for basic scenario */
1345         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1346         if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1347                 encap_len += sizeof(struct iphdr);
1348                 dev->max_mtu -= sizeof(struct iphdr);
1349         } else {
1350                 encap_len += sizeof(struct ipv6hdr);
1351                 dev->max_mtu -= sizeof(struct ipv6hdr);
1352         }
1353         dev->needed_headroom = encap_len + ETH_HLEN;
1354
1355         if (cfg->collect_md) {
1356                 if (tun_on_same_port) {
1357                         NL_SET_ERR_MSG(extack,
1358                                        "There can be only one externally controlled device on a destination port");
1359                         return -EPERM;
1360                 }
1361         } else {
1362                 if (tun_collect_md) {
1363                         NL_SET_ERR_MSG(extack,
1364                                        "There already exists an externally controlled device on this destination port");
1365                         return -EPERM;
1366                 }
1367         }
1368
1369         dst_cache_reset(&geneve->cfg.info.dst_cache);
1370         memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1371
1372         err = register_netdevice(dev);
1373         if (err)
1374                 return err;
1375
1376         list_add(&geneve->next, &gn->geneve_list);
1377         return 0;
1378 }
1379
1380 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1381 {
1382         memset(info, 0, sizeof(*info));
1383         info->key.tp_dst = htons(dst_port);
1384 }
1385
1386 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1387                           struct netlink_ext_ack *extack,
1388                           struct geneve_config *cfg, bool changelink)
1389 {
1390         struct ip_tunnel_info *info = &cfg->info;
1391         int attrtype;
1392
1393         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1394                 NL_SET_ERR_MSG(extack,
1395                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1396                 return -EINVAL;
1397         }
1398
1399         if (data[IFLA_GENEVE_REMOTE]) {
1400                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1401                         attrtype = IFLA_GENEVE_REMOTE;
1402                         goto change_notsup;
1403                 }
1404
1405                 info->key.u.ipv4.dst =
1406                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1407
1408                 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1409                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1410                                             "Remote IPv4 address cannot be Multicast");
1411                         return -EINVAL;
1412                 }
1413         }
1414
1415         if (data[IFLA_GENEVE_REMOTE6]) {
1416 #if IS_ENABLED(CONFIG_IPV6)
1417                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1418                         attrtype = IFLA_GENEVE_REMOTE6;
1419                         goto change_notsup;
1420                 }
1421
1422                 info->mode = IP_TUNNEL_INFO_IPV6;
1423                 info->key.u.ipv6.dst =
1424                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1425
1426                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1427                     IPV6_ADDR_LINKLOCAL) {
1428                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1429                                             "Remote IPv6 address cannot be link-local");
1430                         return -EINVAL;
1431                 }
1432                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1433                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1434                                             "Remote IPv6 address cannot be Multicast");
1435                         return -EINVAL;
1436                 }
1437                 info->key.tun_flags |= TUNNEL_CSUM;
1438                 cfg->use_udp6_rx_checksums = true;
1439 #else
1440                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1441                                     "IPv6 support not enabled in the kernel");
1442                 return -EPFNOSUPPORT;
1443 #endif
1444         }
1445
1446         if (data[IFLA_GENEVE_ID]) {
1447                 __u32 vni;
1448                 __u8 tvni[3];
1449                 __be64 tunid;
1450
1451                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1452                 tvni[0] = (vni & 0x00ff0000) >> 16;
1453                 tvni[1] = (vni & 0x0000ff00) >> 8;
1454                 tvni[2] =  vni & 0x000000ff;
1455
1456                 tunid = vni_to_tunnel_id(tvni);
1457                 if (changelink && (tunid != info->key.tun_id)) {
1458                         attrtype = IFLA_GENEVE_ID;
1459                         goto change_notsup;
1460                 }
1461                 info->key.tun_id = tunid;
1462         }
1463
1464         if (data[IFLA_GENEVE_TTL_INHERIT]) {
1465                 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1466                         cfg->ttl_inherit = true;
1467                 else
1468                         cfg->ttl_inherit = false;
1469         } else if (data[IFLA_GENEVE_TTL]) {
1470                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1471                 cfg->ttl_inherit = false;
1472         }
1473
1474         if (data[IFLA_GENEVE_TOS])
1475                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1476
1477         if (data[IFLA_GENEVE_DF])
1478                 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1479
1480         if (data[IFLA_GENEVE_LABEL]) {
1481                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1482                                   IPV6_FLOWLABEL_MASK;
1483                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1484                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1485                                             "Label attribute only applies for IPv6 Geneve devices");
1486                         return -EINVAL;
1487                 }
1488         }
1489
1490         if (data[IFLA_GENEVE_PORT]) {
1491                 if (changelink) {
1492                         attrtype = IFLA_GENEVE_PORT;
1493                         goto change_notsup;
1494                 }
1495                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1496         }
1497
1498         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1499                 if (changelink) {
1500                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1501                         goto change_notsup;
1502                 }
1503                 cfg->collect_md = true;
1504         }
1505
1506         if (data[IFLA_GENEVE_UDP_CSUM]) {
1507                 if (changelink) {
1508                         attrtype = IFLA_GENEVE_UDP_CSUM;
1509                         goto change_notsup;
1510                 }
1511                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1512                         info->key.tun_flags |= TUNNEL_CSUM;
1513         }
1514
1515         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1516 #if IS_ENABLED(CONFIG_IPV6)
1517                 if (changelink) {
1518                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1519                         goto change_notsup;
1520                 }
1521                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1522                         info->key.tun_flags &= ~TUNNEL_CSUM;
1523 #else
1524                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1525                                     "IPv6 support not enabled in the kernel");
1526                 return -EPFNOSUPPORT;
1527 #endif
1528         }
1529
1530         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1531 #if IS_ENABLED(CONFIG_IPV6)
1532                 if (changelink) {
1533                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1534                         goto change_notsup;
1535                 }
1536                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1537                         cfg->use_udp6_rx_checksums = false;
1538 #else
1539                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1540                                     "IPv6 support not enabled in the kernel");
1541                 return -EPFNOSUPPORT;
1542 #endif
1543         }
1544
1545         return 0;
1546 change_notsup:
1547         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1548                             "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1549         return -EOPNOTSUPP;
1550 }
1551
1552 static void geneve_link_config(struct net_device *dev,
1553                                struct ip_tunnel_info *info, struct nlattr *tb[])
1554 {
1555         struct geneve_dev *geneve = netdev_priv(dev);
1556         int ldev_mtu = 0;
1557
1558         if (tb[IFLA_MTU]) {
1559                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1560                 return;
1561         }
1562
1563         switch (ip_tunnel_info_af(info)) {
1564         case AF_INET: {
1565                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1566                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1567
1568                 if (!IS_ERR(rt) && rt->dst.dev) {
1569                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1570                         ip_rt_put(rt);
1571                 }
1572                 break;
1573         }
1574 #if IS_ENABLED(CONFIG_IPV6)
1575         case AF_INET6: {
1576                 struct rt6_info *rt;
1577
1578                 if (!__in6_dev_get(dev))
1579                         break;
1580
1581                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1582                                 NULL, 0);
1583
1584                 if (rt && rt->dst.dev)
1585                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1586                 ip6_rt_put(rt);
1587                 break;
1588         }
1589 #endif
1590         }
1591
1592         if (ldev_mtu <= 0)
1593                 return;
1594
1595         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1596 }
1597
1598 static int geneve_newlink(struct net *net, struct net_device *dev,
1599                           struct nlattr *tb[], struct nlattr *data[],
1600                           struct netlink_ext_ack *extack)
1601 {
1602         struct geneve_config cfg = {
1603                 .df = GENEVE_DF_UNSET,
1604                 .use_udp6_rx_checksums = false,
1605                 .ttl_inherit = false,
1606                 .collect_md = false,
1607         };
1608         int err;
1609
1610         init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1611         err = geneve_nl2info(tb, data, extack, &cfg, false);
1612         if (err)
1613                 return err;
1614
1615         err = geneve_configure(net, dev, extack, &cfg);
1616         if (err)
1617                 return err;
1618
1619         geneve_link_config(dev, &cfg.info, tb);
1620
1621         return 0;
1622 }
1623
1624 /* Quiesces the geneve device data path for both TX and RX.
1625  *
1626  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1627  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1628  * to complete for the existing set of in-flight packets to be transmitted,
1629  * then we would have quiesced the transmit data path. All the future packets
1630  * will get dropped until we unquiesce the data path.
1631  *
1632  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1633  * if we set that to NULL under RCU and wait for synchronize_net() to
1634  * complete, then we would have quiesced the receive data path.
1635  */
1636 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1637                            struct geneve_sock **gs6)
1638 {
1639         *gs4 = rtnl_dereference(geneve->sock4);
1640         rcu_assign_pointer(geneve->sock4, NULL);
1641         if (*gs4)
1642                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1643 #if IS_ENABLED(CONFIG_IPV6)
1644         *gs6 = rtnl_dereference(geneve->sock6);
1645         rcu_assign_pointer(geneve->sock6, NULL);
1646         if (*gs6)
1647                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1648 #else
1649         *gs6 = NULL;
1650 #endif
1651         synchronize_net();
1652 }
1653
1654 /* Resumes the geneve device data path for both TX and RX. */
1655 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1656                              struct geneve_sock __maybe_unused *gs6)
1657 {
1658         rcu_assign_pointer(geneve->sock4, gs4);
1659         if (gs4)
1660                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1661 #if IS_ENABLED(CONFIG_IPV6)
1662         rcu_assign_pointer(geneve->sock6, gs6);
1663         if (gs6)
1664                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1665 #endif
1666         synchronize_net();
1667 }
1668
1669 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1670                              struct nlattr *data[],
1671                              struct netlink_ext_ack *extack)
1672 {
1673         struct geneve_dev *geneve = netdev_priv(dev);
1674         struct geneve_sock *gs4, *gs6;
1675         struct geneve_config cfg;
1676         int err;
1677
1678         /* If the geneve device is configured for metadata (or externally
1679          * controlled, for example, OVS), then nothing can be changed.
1680          */
1681         if (geneve->cfg.collect_md)
1682                 return -EOPNOTSUPP;
1683
1684         /* Start with the existing info. */
1685         memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1686         err = geneve_nl2info(tb, data, extack, &cfg, true);
1687         if (err)
1688                 return err;
1689
1690         if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1691                 dst_cache_reset(&cfg.info.dst_cache);
1692                 geneve_link_config(dev, &cfg.info, tb);
1693         }
1694
1695         geneve_quiesce(geneve, &gs4, &gs6);
1696         memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1697         geneve_unquiesce(geneve, gs4, gs6);
1698
1699         return 0;
1700 }
1701
1702 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1703 {
1704         struct geneve_dev *geneve = netdev_priv(dev);
1705
1706         list_del(&geneve->next);
1707         unregister_netdevice_queue(dev, head);
1708 }
1709
1710 static size_t geneve_get_size(const struct net_device *dev)
1711 {
1712         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1713                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1714                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1715                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1716                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_DF */
1717                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1718                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1719                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1720                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1721                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1722                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1723                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1724                 0;
1725 }
1726
1727 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1728 {
1729         struct geneve_dev *geneve = netdev_priv(dev);
1730         struct ip_tunnel_info *info = &geneve->cfg.info;
1731         bool ttl_inherit = geneve->cfg.ttl_inherit;
1732         bool metadata = geneve->cfg.collect_md;
1733         __u8 tmp_vni[3];
1734         __u32 vni;
1735
1736         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1737         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1738         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1739                 goto nla_put_failure;
1740
1741         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1742                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1743                                     info->key.u.ipv4.dst))
1744                         goto nla_put_failure;
1745                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1746                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1747                         goto nla_put_failure;
1748
1749 #if IS_ENABLED(CONFIG_IPV6)
1750         } else if (!metadata) {
1751                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1752                                      &info->key.u.ipv6.dst))
1753                         goto nla_put_failure;
1754                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1755                                !(info->key.tun_flags & TUNNEL_CSUM)))
1756                         goto nla_put_failure;
1757 #endif
1758         }
1759
1760         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1761             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1762             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1763                 goto nla_put_failure;
1764
1765         if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1766                 goto nla_put_failure;
1767
1768         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1769                 goto nla_put_failure;
1770
1771         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1772                 goto nla_put_failure;
1773
1774 #if IS_ENABLED(CONFIG_IPV6)
1775         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1776                        !geneve->cfg.use_udp6_rx_checksums))
1777                 goto nla_put_failure;
1778 #endif
1779
1780         if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1781                 goto nla_put_failure;
1782
1783         return 0;
1784
1785 nla_put_failure:
1786         return -EMSGSIZE;
1787 }
1788
1789 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1790         .kind           = "geneve",
1791         .maxtype        = IFLA_GENEVE_MAX,
1792         .policy         = geneve_policy,
1793         .priv_size      = sizeof(struct geneve_dev),
1794         .setup          = geneve_setup,
1795         .validate       = geneve_validate,
1796         .newlink        = geneve_newlink,
1797         .changelink     = geneve_changelink,
1798         .dellink        = geneve_dellink,
1799         .get_size       = geneve_get_size,
1800         .fill_info      = geneve_fill_info,
1801 };
1802
1803 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1804                                         u8 name_assign_type, u16 dst_port)
1805 {
1806         struct nlattr *tb[IFLA_MAX + 1];
1807         struct net_device *dev;
1808         LIST_HEAD(list_kill);
1809         int err;
1810         struct geneve_config cfg = {
1811                 .df = GENEVE_DF_UNSET,
1812                 .use_udp6_rx_checksums = true,
1813                 .ttl_inherit = false,
1814                 .collect_md = true,
1815         };
1816
1817         memset(tb, 0, sizeof(tb));
1818         dev = rtnl_create_link(net, name, name_assign_type,
1819                                &geneve_link_ops, tb, NULL);
1820         if (IS_ERR(dev))
1821                 return dev;
1822
1823         init_tnl_info(&cfg.info, dst_port);
1824         err = geneve_configure(net, dev, NULL, &cfg);
1825         if (err) {
1826                 free_netdev(dev);
1827                 return ERR_PTR(err);
1828         }
1829
1830         /* openvswitch users expect packet sizes to be unrestricted,
1831          * so set the largest MTU we can.
1832          */
1833         err = geneve_change_mtu(dev, IP_MAX_MTU);
1834         if (err)
1835                 goto err;
1836
1837         err = rtnl_configure_link(dev, NULL);
1838         if (err < 0)
1839                 goto err;
1840
1841         return dev;
1842 err:
1843         geneve_dellink(dev, &list_kill);
1844         unregister_netdevice_many(&list_kill);
1845         return ERR_PTR(err);
1846 }
1847 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1848
1849 static int geneve_netdevice_event(struct notifier_block *unused,
1850                                   unsigned long event, void *ptr)
1851 {
1852         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1853
1854         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1855             event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1856                 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1857         } else if (event == NETDEV_UNREGISTER) {
1858                 if (!dev->udp_tunnel_nic_info)
1859                         geneve_offload_rx_ports(dev, false);
1860         } else if (event == NETDEV_REGISTER) {
1861                 if (!dev->udp_tunnel_nic_info)
1862                         geneve_offload_rx_ports(dev, true);
1863         }
1864
1865         return NOTIFY_DONE;
1866 }
1867
1868 static struct notifier_block geneve_notifier_block __read_mostly = {
1869         .notifier_call = geneve_netdevice_event,
1870 };
1871
1872 static __net_init int geneve_init_net(struct net *net)
1873 {
1874         struct geneve_net *gn = net_generic(net, geneve_net_id);
1875
1876         INIT_LIST_HEAD(&gn->geneve_list);
1877         INIT_LIST_HEAD(&gn->sock_list);
1878         return 0;
1879 }
1880
1881 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1882 {
1883         struct geneve_net *gn = net_generic(net, geneve_net_id);
1884         struct geneve_dev *geneve, *next;
1885         struct net_device *dev, *aux;
1886
1887         /* gather any geneve devices that were moved into this ns */
1888         for_each_netdev_safe(net, dev, aux)
1889                 if (dev->rtnl_link_ops == &geneve_link_ops)
1890                         unregister_netdevice_queue(dev, head);
1891
1892         /* now gather any other geneve devices that were created in this ns */
1893         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1894                 /* If geneve->dev is in the same netns, it was already added
1895                  * to the list by the previous loop.
1896                  */
1897                 if (!net_eq(dev_net(geneve->dev), net))
1898                         unregister_netdevice_queue(geneve->dev, head);
1899         }
1900 }
1901
1902 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1903 {
1904         struct net *net;
1905         LIST_HEAD(list);
1906
1907         rtnl_lock();
1908         list_for_each_entry(net, net_list, exit_list)
1909                 geneve_destroy_tunnels(net, &list);
1910
1911         /* unregister the devices gathered above */
1912         unregister_netdevice_many(&list);
1913         rtnl_unlock();
1914
1915         list_for_each_entry(net, net_list, exit_list) {
1916                 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1917
1918                 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1919         }
1920 }
1921
1922 static struct pernet_operations geneve_net_ops = {
1923         .init = geneve_init_net,
1924         .exit_batch = geneve_exit_batch_net,
1925         .id   = &geneve_net_id,
1926         .size = sizeof(struct geneve_net),
1927 };
1928
1929 static int __init geneve_init_module(void)
1930 {
1931         int rc;
1932
1933         rc = register_pernet_subsys(&geneve_net_ops);
1934         if (rc)
1935                 goto out1;
1936
1937         rc = register_netdevice_notifier(&geneve_notifier_block);
1938         if (rc)
1939                 goto out2;
1940
1941         rc = rtnl_link_register(&geneve_link_ops);
1942         if (rc)
1943                 goto out3;
1944
1945         return 0;
1946 out3:
1947         unregister_netdevice_notifier(&geneve_notifier_block);
1948 out2:
1949         unregister_pernet_subsys(&geneve_net_ops);
1950 out1:
1951         return rc;
1952 }
1953 late_initcall(geneve_init_module);
1954
1955 static void __exit geneve_cleanup_module(void)
1956 {
1957         rtnl_link_unregister(&geneve_link_ops);
1958         unregister_netdevice_notifier(&geneve_notifier_block);
1959         unregister_pernet_subsys(&geneve_net_ops);
1960 }
1961 module_exit(geneve_cleanup_module);
1962
1963 MODULE_LICENSE("GPL");
1964 MODULE_VERSION(GENEVE_NETDEV_VER);
1965 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1966 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1967 MODULE_ALIAS_RTNL_LINK("geneve");