drivers/net/geneve.c

   1 /*
   2  * GENEVE: Generic Network Virtualization Encapsulation
   3  *
   4  * Copyright (c) 2015 Red Hat, Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/etherdevice.h>
  16 #include <linux/hash.h>
  17 #include <net/dst_metadata.h>
  18 #include <net/gro_cells.h>
  19 #include <net/rtnetlink.h>
  20 #include <net/geneve.h>
  21 #include <net/protocol.h>
  22
  23 #define GENEVE_NETDEV_VER       "0.6"
  24
  25 #define GENEVE_UDP_PORT         6081
  26
  27 #define GENEVE_N_VID            (1u << 24)
  28 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  29
  30 #define VNI_HASH_BITS           10
  31 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  32
  33 static bool log_ecn_error = true;
  34 module_param(log_ecn_error, bool, 0644);
  35 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  36
  37 #define GENEVE_VER 0
  38 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  39 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
  40 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
  41
  42 /* per-network namespace private data for this module */
  43 struct geneve_net {
  44         struct list_head        geneve_list;
  45         struct list_head        sock_list;
  46 };
  47
  48 static unsigned int geneve_net_id;
  49
  50 struct geneve_dev_node {
  51         struct hlist_node hlist;
  52         struct geneve_dev *geneve;
  53 };
  54
  55 /* Pseudo network device */
  56 struct geneve_dev {
  57         struct geneve_dev_node hlist4;  /* vni hash table for IPv4 socket */
  58 #if IS_ENABLED(CONFIG_IPV6)
  59         struct geneve_dev_node hlist6;  /* vni hash table for IPv6 socket */
  60 #endif
  61         struct net         *net;        /* netns for packet i/o */
  62         struct net_device  *dev;        /* netdev for geneve tunnel */
  63         struct ip_tunnel_info info;
  64         struct geneve_sock __rcu *sock4;        /* IPv4 socket used for geneve tunnel */
  65 #if IS_ENABLED(CONFIG_IPV6)
  66         struct geneve_sock __rcu *sock6;        /* IPv6 socket used for geneve tunnel */
  67 #endif
  68         struct list_head   next;        /* geneve's per namespace list */
  69         struct gro_cells   gro_cells;
  70         bool               collect_md;
  71         bool               use_udp6_rx_checksums;
  72         bool               ttl_inherit;
  73         enum ifla_geneve_df df;
  74 };
  75
  76 struct geneve_sock {
  77         bool                    collect_md;
  78         struct list_head        list;
  79         struct socket           *sock;
  80         struct rcu_head         rcu;
  81         int                     refcnt;
  82         struct hlist_head       vni_list[VNI_HASH_SIZE];
  83 };
  84
  85 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  86 {
  87         __u32 vnid;
  88
  89         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  90         return hash_32(vnid, VNI_HASH_BITS);
  91 }
  92
  93 static __be64 vni_to_tunnel_id(const __u8 *vni)
  94 {
  95 #ifdef __BIG_ENDIAN
  96         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
  97 #else
  98         return (__force __be64)(((__force u64)vni[0] << 40) |
  99                                 ((__force u64)vni[1] << 48) |
 100                                 ((__force u64)vni[2] << 56));
 101 #endif
 102 }
 103
 104 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 105 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 106 {
 107 #ifdef __BIG_ENDIAN
 108         vni[0] = (__force __u8)(tun_id >> 16);
 109         vni[1] = (__force __u8)(tun_id >> 8);
 110         vni[2] = (__force __u8)tun_id;
 111 #else
 112         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 113         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 114         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 115 #endif
 116 }
 117
 118 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
 119 {
 120         return !memcmp(vni, &tun_id[5], 3);
 121 }
 122
 123 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 124 {
 125         return gs->sock->sk->sk_family;
 126 }
 127
 128 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 129                                         __be32 addr, u8 vni[])
 130 {
 131         struct hlist_head *vni_list_head;
 132         struct geneve_dev_node *node;
 133         __u32 hash;
 134
 135         /* Find the device for this VNI */
 136         hash = geneve_net_vni_hash(vni);
 137         vni_list_head = &gs->vni_list[hash];
 138         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 139                 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
 140                     addr == node->geneve->info.key.u.ipv4.dst)
 141                         return node->geneve;
 142         }
 143         return NULL;
 144 }
 145
 146 #if IS_ENABLED(CONFIG_IPV6)
 147 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 148                                          struct in6_addr addr6, u8 vni[])
 149 {
 150         struct hlist_head *vni_list_head;
 151         struct geneve_dev_node *node;
 152         __u32 hash;
 153
 154         /* Find the device for this VNI */
 155         hash = geneve_net_vni_hash(vni);
 156         vni_list_head = &gs->vni_list[hash];
 157         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 158                 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
 159                     ipv6_addr_equal(&addr6, &node->geneve->info.key.u.ipv6.dst))
 160                         return node->geneve;
 161         }
 162         return NULL;
 163 }
 164 #endif
 165
 166 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 167 {
 168         return (struct genevehdr *)(udp_hdr(skb) + 1);
 169 }
 170
 171 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 172                                             struct sk_buff *skb)
 173 {
 174         static u8 zero_vni[3];
 175         u8 *vni;
 176
 177         if (geneve_get_sk_family(gs) == AF_INET) {
 178                 struct iphdr *iph;
 179                 __be32 addr;
 180
 181                 iph = ip_hdr(skb); /* outer IP header... */
 182
 183                 if (gs->collect_md) {
 184                         vni = zero_vni;
 185                         addr = 0;
 186                 } else {
 187                         vni = geneve_hdr(skb)->vni;
 188                         addr = iph->saddr;
 189                 }
 190
 191                 return geneve_lookup(gs, addr, vni);
 192 #if IS_ENABLED(CONFIG_IPV6)
 193         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 194                 static struct in6_addr zero_addr6;
 195                 struct ipv6hdr *ip6h;
 196                 struct in6_addr addr6;
 197
 198                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 199
 200                 if (gs->collect_md) {
 201                         vni = zero_vni;
 202                         addr6 = zero_addr6;
 203                 } else {
 204                         vni = geneve_hdr(skb)->vni;
 205                         addr6 = ip6h->saddr;
 206                 }
 207
 208                 return geneve6_lookup(gs, addr6, vni);
 209 #endif
 210         }
 211         return NULL;
 212 }
 213
 214 /* geneve receive/decap routine */
 215 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 216                       struct sk_buff *skb)
 217 {
 218         struct genevehdr *gnvh = geneve_hdr(skb);
 219         struct metadata_dst *tun_dst = NULL;
 220         struct pcpu_sw_netstats *stats;
 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 | TUNNEL_GENEVE_OPT |
 229                         (gnvh->oam ? TUNNEL_OAM : 0) |
 230                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 231
 232                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 233                                          vni_to_tunnel_id(gnvh->vni),
 234                                          gnvh->opt_len * 4);
 235                 if (!tun_dst) {
 236                         geneve->dev->stats.rx_dropped++;
 237                         goto drop;
 238                 }
 239                 /* Update tunnel dst according to Geneve options. */
 240                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 241                                         gnvh->options, gnvh->opt_len * 4,
 242                                         TUNNEL_GENEVE_OPT);
 243         } else {
 244                 /* Drop packets w/ critical options,
 245                  * since we don't support any...
 246                  */
 247                 if (gnvh->critical) {
 248                         geneve->dev->stats.rx_frame_errors++;
 249                         geneve->dev->stats.rx_errors++;
 250                         goto drop;
 251                 }
 252         }
 253
 254         skb_reset_mac_header(skb);
 255         skb->protocol = eth_type_trans(skb, geneve->dev);
 256         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 257
 258         if (tun_dst)
 259                 skb_dst_set(skb, &tun_dst->dst);
 260
 261         /* Ignore packet loops (and multicast echo) */
 262         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
 263                 geneve->dev->stats.rx_errors++;
 264                 goto drop;
 265         }
 266
 267         oiph = skb_network_header(skb);
 268         skb_reset_network_header(skb);
 269
 270         if (geneve_get_sk_family(gs) == AF_INET)
 271                 err = IP_ECN_decapsulate(oiph, skb);
 272 #if IS_ENABLED(CONFIG_IPV6)
 273         else
 274                 err = IP6_ECN_decapsulate(oiph, skb);
 275 #endif
 276
 277         if (unlikely(err)) {
 278                 if (log_ecn_error) {
 279                         if (geneve_get_sk_family(gs) == AF_INET)
 280                                 net_info_ratelimited("non-ECT from %pI4 "
 281                                                      "with TOS=%#x\n",
 282                                                      &((struct iphdr *)oiph)->saddr,
 283                                                      ((struct iphdr *)oiph)->tos);
 284 #if IS_ENABLED(CONFIG_IPV6)
 285                         else
 286                                 net_info_ratelimited("non-ECT from %pI6\n",
 287                                                      &((struct ipv6hdr *)oiph)->saddr);
 288 #endif
 289                 }
 290                 if (err > 1) {
 291                         ++geneve->dev->stats.rx_frame_errors;
 292                         ++geneve->dev->stats.rx_errors;
 293                         goto drop;
 294                 }
 295         }
 296
 297         len = skb->len;
 298         err = gro_cells_receive(&geneve->gro_cells, skb);
 299         if (likely(err == NET_RX_SUCCESS)) {
 300                 stats = this_cpu_ptr(geneve->dev->tstats);
 301                 u64_stats_update_begin(&stats->syncp);
 302                 stats->rx_packets++;
 303                 stats->rx_bytes += len;
 304                 u64_stats_update_end(&stats->syncp);
 305         }
 306         return;
 307 drop:
 308         /* Consume bad packet */
 309         kfree_skb(skb);
 310 }
 311
 312 /* Setup stats when device is created */
 313 static int geneve_init(struct net_device *dev)
 314 {
 315         struct geneve_dev *geneve = netdev_priv(dev);
 316         int err;
 317
 318         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 319         if (!dev->tstats)
 320                 return -ENOMEM;
 321
 322         err = gro_cells_init(&geneve->gro_cells, dev);
 323         if (err) {
 324                 free_percpu(dev->tstats);
 325                 return err;
 326         }
 327
 328         err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL);
 329         if (err) {
 330                 free_percpu(dev->tstats);
 331                 gro_cells_destroy(&geneve->gro_cells);
 332                 return err;
 333         }
 334         return 0;
 335 }
 336
 337 static void geneve_uninit(struct net_device *dev)
 338 {
 339         struct geneve_dev *geneve = netdev_priv(dev);
 340
 341         dst_cache_destroy(&geneve->info.dst_cache);
 342         gro_cells_destroy(&geneve->gro_cells);
 343         free_percpu(dev->tstats);
 344 }
 345
 346 /* Callback from net/ipv4/udp.c to receive packets */
 347 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 348 {
 349         struct genevehdr *geneveh;
 350         struct geneve_dev *geneve;
 351         struct geneve_sock *gs;
 352         int opts_len;
 353
 354         /* Need UDP and Geneve header to be present */
 355         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 356                 goto drop;
 357
 358         /* Return packets with reserved bits set */
 359         geneveh = geneve_hdr(skb);
 360         if (unlikely(geneveh->ver != GENEVE_VER))
 361                 goto drop;
 362
 363         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 364                 goto drop;
 365
 366         gs = rcu_dereference_sk_user_data(sk);
 367         if (!gs)
 368                 goto drop;
 369
 370         geneve = geneve_lookup_skb(gs, skb);
 371         if (!geneve)
 372                 goto drop;
 373
 374         opts_len = geneveh->opt_len * 4;
 375         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 376                                  htons(ETH_P_TEB),
 377                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 378                 geneve->dev->stats.rx_dropped++;
 379                 goto drop;
 380         }
 381
 382         geneve_rx(geneve, gs, skb);
 383         return 0;
 384
 385 drop:
 386         /* Consume bad packet */
 387         kfree_skb(skb);
 388         return 0;
 389 }
 390
 391 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
 392 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
 393 {
 394         struct genevehdr *geneveh;
 395         struct geneve_sock *gs;
 396         u8 zero_vni[3] = { 0 };
 397         u8 *vni = zero_vni;
 398
 399         if (skb->len < GENEVE_BASE_HLEN)
 400                 return -EINVAL;
 401
 402         geneveh = geneve_hdr(skb);
 403         if (geneveh->ver != GENEVE_VER)
 404                 return -EINVAL;
 405
 406         if (geneveh->proto_type != htons(ETH_P_TEB))
 407                 return -EINVAL;
 408
 409         gs = rcu_dereference_sk_user_data(sk);
 410         if (!gs)
 411                 return -ENOENT;
 412
 413         if (geneve_get_sk_family(gs) == AF_INET) {
 414                 struct iphdr *iph = ip_hdr(skb);
 415                 __be32 addr4 = 0;
 416
 417                 if (!gs->collect_md) {
 418                         vni = geneve_hdr(skb)->vni;
 419                         addr4 = iph->daddr;
 420                 }
 421
 422                 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
 423         }
 424
 425 #if IS_ENABLED(CONFIG_IPV6)
 426         if (geneve_get_sk_family(gs) == AF_INET6) {
 427                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 428                 struct in6_addr addr6;
 429
 430                 memset(&addr6, 0, sizeof(struct in6_addr));
 431
 432                 if (!gs->collect_md) {
 433                         vni = geneve_hdr(skb)->vni;
 434                         addr6 = ip6h->daddr;
 435                 }
 436
 437                 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
 438         }
 439 #endif
 440
 441         return -EPFNOSUPPORT;
 442 }
 443
 444 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 445                                          __be16 port, bool ipv6_rx_csum)
 446 {
 447         struct socket *sock;
 448         struct udp_port_cfg udp_conf;
 449         int err;
 450
 451         memset(&udp_conf, 0, sizeof(udp_conf));
 452
 453         if (ipv6) {
 454                 udp_conf.family = AF_INET6;
 455                 udp_conf.ipv6_v6only = 1;
 456                 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
 457         } else {
 458                 udp_conf.family = AF_INET;
 459                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 460         }
 461
 462         udp_conf.local_udp_port = port;
 463
 464         /* Open UDP socket */
 465         err = udp_sock_create(net, &udp_conf, &sock);
 466         if (err < 0)
 467                 return ERR_PTR(err);
 468
 469         return sock;
 470 }
 471
 472 static int geneve_hlen(struct genevehdr *gh)
 473 {
 474         return sizeof(*gh) + gh->opt_len * 4;
 475 }
 476
 477 static struct sk_buff *geneve_gro_receive(struct sock *sk,
 478                                           struct list_head *head,
 479                                           struct sk_buff *skb)
 480 {
 481         struct sk_buff *pp = NULL;
 482         struct sk_buff *p;
 483         struct genevehdr *gh, *gh2;
 484         unsigned int hlen, gh_len, off_gnv;
 485         const struct packet_offload *ptype;
 486         __be16 type;
 487         int flush = 1;
 488
 489         off_gnv = skb_gro_offset(skb);
 490         hlen = off_gnv + sizeof(*gh);
 491         gh = skb_gro_header_fast(skb, off_gnv);
 492         if (skb_gro_header_hard(skb, hlen)) {
 493                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 494                 if (unlikely(!gh))
 495                         goto out;
 496         }
 497
 498         if (gh->ver != GENEVE_VER || gh->oam)
 499                 goto out;
 500         gh_len = geneve_hlen(gh);
 501
 502         hlen = off_gnv + gh_len;
 503         if (skb_gro_header_hard(skb, hlen)) {
 504                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 505                 if (unlikely(!gh))
 506                         goto out;
 507         }
 508
 509         list_for_each_entry(p, head, list) {
 510                 if (!NAPI_GRO_CB(p)->same_flow)
 511                         continue;
 512
 513                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 514                 if (gh->opt_len != gh2->opt_len ||
 515                     memcmp(gh, gh2, gh_len)) {
 516                         NAPI_GRO_CB(p)->same_flow = 0;
 517                         continue;
 518                 }
 519         }
 520
 521         type = gh->proto_type;
 522
 523         rcu_read_lock();
 524         ptype = gro_find_receive_by_type(type);
 525         if (!ptype)
 526                 goto out_unlock;
 527
 528         skb_gro_pull(skb, gh_len);
 529         skb_gro_postpull_rcsum(skb, gh, gh_len);
 530         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 531         flush = 0;
 532
 533 out_unlock:
 534         rcu_read_unlock();
 535 out:
 536         skb_gro_flush_final(skb, pp, flush);
 537
 538         return pp;
 539 }
 540
 541 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 542                                int nhoff)
 543 {
 544         struct genevehdr *gh;
 545         struct packet_offload *ptype;
 546         __be16 type;
 547         int gh_len;
 548         int err = -ENOSYS;
 549
 550         gh = (struct genevehdr *)(skb->data + nhoff);
 551         gh_len = geneve_hlen(gh);
 552         type = gh->proto_type;
 553
 554         rcu_read_lock();
 555         ptype = gro_find_complete_by_type(type);
 556         if (ptype)
 557                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 558
 559         rcu_read_unlock();
 560
 561         skb_set_inner_mac_header(skb, nhoff + gh_len);
 562
 563         return err;
 564 }
 565
 566 /* Create new listen socket if needed */
 567 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 568                                                 bool ipv6, bool ipv6_rx_csum)
 569 {
 570         struct geneve_net *gn = net_generic(net, geneve_net_id);
 571         struct geneve_sock *gs;
 572         struct socket *sock;
 573         struct udp_tunnel_sock_cfg tunnel_cfg;
 574         int h;
 575
 576         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 577         if (!gs)
 578                 return ERR_PTR(-ENOMEM);
 579
 580         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 581         if (IS_ERR(sock)) {
 582                 kfree(gs);
 583                 return ERR_CAST(sock);
 584         }
 585
 586         gs->sock = sock;
 587         gs->refcnt = 1;
 588         for (h = 0; h < VNI_HASH_SIZE; ++h)
 589                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 590
 591         /* Initialize the geneve udp offloads structure */
 592         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 593
 594         /* Mark socket as an encapsulation socket */
 595         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 596         tunnel_cfg.sk_user_data = gs;
 597         tunnel_cfg.encap_type = 1;
 598         tunnel_cfg.gro_receive = geneve_gro_receive;
 599         tunnel_cfg.gro_complete = geneve_gro_complete;
 600         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 601         tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
 602         tunnel_cfg.encap_destroy = NULL;
 603         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 604         list_add(&gs->list, &gn->sock_list);
 605         return gs;
 606 }
 607
 608 static void __geneve_sock_release(struct geneve_sock *gs)
 609 {
 610         if (!gs || --gs->refcnt)
 611                 return;
 612
 613         list_del(&gs->list);
 614         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 615         udp_tunnel_sock_release(gs->sock);
 616         kfree_rcu(gs, rcu);
 617 }
 618
 619 static void geneve_sock_release(struct geneve_dev *geneve)
 620 {
 621         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 622 #if IS_ENABLED(CONFIG_IPV6)
 623         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 624
 625         rcu_assign_pointer(geneve->sock6, NULL);
 626 #endif
 627
 628         rcu_assign_pointer(geneve->sock4, NULL);
 629         synchronize_net();
 630
 631         __geneve_sock_release(gs4);
 632 #if IS_ENABLED(CONFIG_IPV6)
 633         __geneve_sock_release(gs6);
 634 #endif
 635 }
 636
 637 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 638                                             sa_family_t family,
 639                                             __be16 dst_port)
 640 {
 641         struct geneve_sock *gs;
 642
 643         list_for_each_entry(gs, &gn->sock_list, list) {
 644                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 645                     geneve_get_sk_family(gs) == family) {
 646                         return gs;
 647                 }
 648         }
 649         return NULL;
 650 }
 651
 652 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 653 {
 654         struct net *net = geneve->net;
 655         struct geneve_net *gn = net_generic(net, geneve_net_id);
 656         struct geneve_dev_node *node;
 657         struct geneve_sock *gs;
 658         __u8 vni[3];
 659         __u32 hash;
 660
 661         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst);
 662         if (gs) {
 663                 gs->refcnt++;
 664                 goto out;
 665         }
 666
 667         gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6,
 668                                   geneve->use_udp6_rx_checksums);
 669         if (IS_ERR(gs))
 670                 return PTR_ERR(gs);
 671
 672 out:
 673         gs->collect_md = geneve->collect_md;
 674 #if IS_ENABLED(CONFIG_IPV6)
 675         if (ipv6) {
 676                 rcu_assign_pointer(geneve->sock6, gs);
 677                 node = &geneve->hlist6;
 678         } else
 679 #endif
 680         {
 681                 rcu_assign_pointer(geneve->sock4, gs);
 682                 node = &geneve->hlist4;
 683         }
 684         node->geneve = geneve;
 685
 686         tunnel_id_to_vni(geneve->info.key.tun_id, vni);
 687         hash = geneve_net_vni_hash(vni);
 688         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 689         return 0;
 690 }
 691
 692 static int geneve_open(struct net_device *dev)
 693 {
 694         struct geneve_dev *geneve = netdev_priv(dev);
 695         bool ipv6 = !!(geneve->info.mode & IP_TUNNEL_INFO_IPV6);
 696         bool metadata = geneve->collect_md;
 697         int ret = 0;
 698
 699 #if IS_ENABLED(CONFIG_IPV6)
 700         if (ipv6 || metadata)
 701                 ret = geneve_sock_add(geneve, true);
 702 #endif
 703         if (!ret && (!ipv6 || metadata))
 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 {
 777         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 778         struct geneve_dev *geneve = netdev_priv(dev);
 779         struct dst_cache *dst_cache;
 780         struct rtable *rt = NULL;
 781         __u8 tos;
 782
 783         if (!gs4)
 784                 return ERR_PTR(-EIO);
 785
 786         memset(fl4, 0, sizeof(*fl4));
 787         fl4->flowi4_mark = skb->mark;
 788         fl4->flowi4_proto = IPPROTO_UDP;
 789         fl4->daddr = info->key.u.ipv4.dst;
 790         fl4->saddr = info->key.u.ipv4.src;
 791
 792         tos = info->key.tos;
 793         if ((tos == 1) && !geneve->collect_md) {
 794                 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 795                 use_cache = false;
 796         }
 797         fl4->flowi4_tos = RT_TOS(tos);
 798
 799         dst_cache = (struct dst_cache *)&info->dst_cache;
 800         if (use_cache) {
 801                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 802                 if (rt)
 803                         return rt;
 804         }
 805         rt = ip_route_output_key(geneve->net, fl4);
 806         if (IS_ERR(rt)) {
 807                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 808                 return ERR_PTR(-ENETUNREACH);
 809         }
 810         if (rt->dst.dev == dev) { /* is this necessary? */
 811                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 812                 ip_rt_put(rt);
 813                 return ERR_PTR(-ELOOP);
 814         }
 815         if (use_cache)
 816                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 817         return rt;
 818 }
 819
 820 #if IS_ENABLED(CONFIG_IPV6)
 821 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 822                                            struct net_device *dev,
 823                                            struct geneve_sock *gs6,
 824                                            struct flowi6 *fl6,
 825                                            const struct ip_tunnel_info *info)
 826 {
 827         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 828         struct geneve_dev *geneve = netdev_priv(dev);
 829         struct dst_entry *dst = NULL;
 830         struct dst_cache *dst_cache;
 831         __u8 prio;
 832
 833         if (!gs6)
 834                 return ERR_PTR(-EIO);
 835
 836         memset(fl6, 0, sizeof(*fl6));
 837         fl6->flowi6_mark = skb->mark;
 838         fl6->flowi6_proto = IPPROTO_UDP;
 839         fl6->daddr = info->key.u.ipv6.dst;
 840         fl6->saddr = info->key.u.ipv6.src;
 841         prio = info->key.tos;
 842         if ((prio == 1) && !geneve->collect_md) {
 843                 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 844                 use_cache = false;
 845         }
 846
 847         fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 848                                            info->key.label);
 849         dst_cache = (struct dst_cache *)&info->dst_cache;
 850         if (use_cache) {
 851                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 852                 if (dst)
 853                         return dst;
 854         }
 855         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 856                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 857                 return ERR_PTR(-ENETUNREACH);
 858         }
 859         if (dst->dev == dev) { /* is this necessary? */
 860                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 861                 dst_release(dst);
 862                 return ERR_PTR(-ELOOP);
 863         }
 864
 865         if (use_cache)
 866                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 867         return dst;
 868 }
 869 #endif
 870
 871 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 872                            struct geneve_dev *geneve,
 873                            const struct ip_tunnel_info *info)
 874 {
 875         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 876         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 877         const struct ip_tunnel_key *key = &info->key;
 878         struct rtable *rt;
 879         struct flowi4 fl4;
 880         __u8 tos, ttl;
 881         __be16 df = 0;
 882         __be16 sport;
 883         int err;
 884
 885         rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
 886         if (IS_ERR(rt))
 887                 return PTR_ERR(rt);
 888
 889         skb_tunnel_check_pmtu(skb, &rt->dst,
 890                               GENEVE_IPV4_HLEN + info->options_len);
 891
 892         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 893         if (geneve->collect_md) {
 894                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 895                 ttl = key->ttl;
 896
 897                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 898         } else {
 899                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 900                 if (geneve->ttl_inherit)
 901                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 902                 else
 903                         ttl = key->ttl;
 904                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 905
 906                 if (geneve->df == GENEVE_DF_SET) {
 907                         df = htons(IP_DF);
 908                 } else if (geneve->df == GENEVE_DF_INHERIT) {
 909                         struct ethhdr *eth = eth_hdr(skb);
 910
 911                         if (ntohs(eth->h_proto) == ETH_P_IPV6) {
 912                                 df = htons(IP_DF);
 913                         } else if (ntohs(eth->h_proto) == ETH_P_IP) {
 914                                 struct iphdr *iph = ip_hdr(skb);
 915
 916                                 if (iph->frag_off & htons(IP_DF))
 917                                         df = htons(IP_DF);
 918                         }
 919                 }
 920         }
 921
 922         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
 923         if (unlikely(err))
 924                 return err;
 925
 926         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 927                             tos, ttl, df, sport, geneve->info.key.tp_dst,
 928                             !net_eq(geneve->net, dev_net(geneve->dev)),
 929                             !(info->key.tun_flags & TUNNEL_CSUM));
 930         return 0;
 931 }
 932
 933 #if IS_ENABLED(CONFIG_IPV6)
 934 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 935                             struct geneve_dev *geneve,
 936                             const struct ip_tunnel_info *info)
 937 {
 938         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 939         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 940         const struct ip_tunnel_key *key = &info->key;
 941         struct dst_entry *dst = NULL;
 942         struct flowi6 fl6;
 943         __u8 prio, ttl;
 944         __be16 sport;
 945         int err;
 946
 947         dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
 948         if (IS_ERR(dst))
 949                 return PTR_ERR(dst);
 950
 951         skb_tunnel_check_pmtu(skb, dst, GENEVE_IPV6_HLEN + info->options_len);
 952
 953         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 954         if (geneve->collect_md) {
 955                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 956                 ttl = key->ttl;
 957         } else {
 958                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
 959                                            ip_hdr(skb), skb);
 960                 if (geneve->ttl_inherit)
 961                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 962                 else
 963                         ttl = key->ttl;
 964                 ttl = ttl ? : ip6_dst_hoplimit(dst);
 965         }
 966         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
 967         if (unlikely(err))
 968                 return err;
 969
 970         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
 971                              &fl6.saddr, &fl6.daddr, prio, ttl,
 972                              info->key.label, sport, geneve->info.key.tp_dst,
 973                              !(info->key.tun_flags & TUNNEL_CSUM));
 974         return 0;
 975 }
 976 #endif
 977
 978 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
 979 {
 980         struct geneve_dev *geneve = netdev_priv(dev);
 981         struct ip_tunnel_info *info = NULL;
 982         int err;
 983
 984         if (geneve->collect_md) {
 985                 info = skb_tunnel_info(skb);
 986                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 987                         err = -EINVAL;
 988                         netdev_dbg(dev, "no tunnel metadata\n");
 989                         goto tx_error;
 990                 }
 991         } else {
 992                 info = &geneve->info;
 993         }
 994
 995         rcu_read_lock();
 996 #if IS_ENABLED(CONFIG_IPV6)
 997         if (info->mode & IP_TUNNEL_INFO_IPV6)
 998                 err = geneve6_xmit_skb(skb, dev, geneve, info);
 999         else
1000 #endif
1001                 err = geneve_xmit_skb(skb, dev, geneve, info);
1002         rcu_read_unlock();
1003
1004         if (likely(!err))
1005                 return NETDEV_TX_OK;
1006 tx_error:
1007         dev_kfree_skb(skb);
1008
1009         if (err == -ELOOP)
1010                 dev->stats.collisions++;
1011         else if (err == -ENETUNREACH)
1012                 dev->stats.tx_carrier_errors++;
1013
1014         dev->stats.tx_errors++;
1015         return NETDEV_TX_OK;
1016 }
1017
1018 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1019 {
1020         if (new_mtu > dev->max_mtu)
1021                 new_mtu = dev->max_mtu;
1022         else if (new_mtu < dev->min_mtu)
1023                 new_mtu = dev->min_mtu;
1024
1025         dev->mtu = new_mtu;
1026         return 0;
1027 }
1028
1029 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1030 {
1031         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1032         struct geneve_dev *geneve = netdev_priv(dev);
1033
1034         if (ip_tunnel_info_af(info) == AF_INET) {
1035                 struct rtable *rt;
1036                 struct flowi4 fl4;
1037                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1038
1039                 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
1040                 if (IS_ERR(rt))
1041                         return PTR_ERR(rt);
1042
1043                 ip_rt_put(rt);
1044                 info->key.u.ipv4.src = fl4.saddr;
1045 #if IS_ENABLED(CONFIG_IPV6)
1046         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1047                 struct dst_entry *dst;
1048                 struct flowi6 fl6;
1049                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1050
1051                 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
1052                 if (IS_ERR(dst))
1053                         return PTR_ERR(dst);
1054
1055                 dst_release(dst);
1056                 info->key.u.ipv6.src = fl6.saddr;
1057 #endif
1058         } else {
1059                 return -EINVAL;
1060         }
1061
1062         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1063                                              1, USHRT_MAX, true);
1064         info->key.tp_dst = geneve->info.key.tp_dst;
1065         return 0;
1066 }
1067
1068 static const struct net_device_ops geneve_netdev_ops = {
1069         .ndo_init               = geneve_init,
1070         .ndo_uninit             = geneve_uninit,
1071         .ndo_open               = geneve_open,
1072         .ndo_stop               = geneve_stop,
1073         .ndo_start_xmit         = geneve_xmit,
1074         .ndo_get_stats64        = ip_tunnel_get_stats64,
1075         .ndo_change_mtu         = geneve_change_mtu,
1076         .ndo_validate_addr      = eth_validate_addr,
1077         .ndo_set_mac_address    = eth_mac_addr,
1078         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1079 };
1080
1081 static void geneve_get_drvinfo(struct net_device *dev,
1082                                struct ethtool_drvinfo *drvinfo)
1083 {
1084         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1085         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1086 }
1087
1088 static const struct ethtool_ops geneve_ethtool_ops = {
1089         .get_drvinfo    = geneve_get_drvinfo,
1090         .get_link       = ethtool_op_get_link,
1091 };
1092
1093 /* Info for udev, that this is a virtual tunnel endpoint */
1094 static struct device_type geneve_type = {
1095         .name = "geneve",
1096 };
1097
1098 /* Calls the ndo_udp_tunnel_add of the caller in order to
1099  * supply the listening GENEVE udp ports. Callers are expected
1100  * to implement the ndo_udp_tunnel_add.
1101  */
1102 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1103 {
1104         struct net *net = dev_net(dev);
1105         struct geneve_net *gn = net_generic(net, geneve_net_id);
1106         struct geneve_sock *gs;
1107
1108         rcu_read_lock();
1109         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1110                 if (push) {
1111                         udp_tunnel_push_rx_port(dev, gs->sock,
1112                                                 UDP_TUNNEL_TYPE_GENEVE);
1113                 } else {
1114                         udp_tunnel_drop_rx_port(dev, gs->sock,
1115                                                 UDP_TUNNEL_TYPE_GENEVE);
1116                 }
1117         }
1118         rcu_read_unlock();
1119 }
1120
1121 /* Initialize the device structure. */
1122 static void geneve_setup(struct net_device *dev)
1123 {
1124         ether_setup(dev);
1125
1126         dev->netdev_ops = &geneve_netdev_ops;
1127         dev->ethtool_ops = &geneve_ethtool_ops;
1128         dev->needs_free_netdev = true;
1129
1130         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1131
1132         dev->features    |= NETIF_F_LLTX;
1133         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1134         dev->features    |= NETIF_F_RXCSUM;
1135         dev->features    |= NETIF_F_GSO_SOFTWARE;
1136
1137         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1138         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1139
1140         /* MTU range: 68 - (something less than 65535) */
1141         dev->min_mtu = ETH_MIN_MTU;
1142         /* The max_mtu calculation does not take account of GENEVE
1143          * options, to avoid excluding potentially valid
1144          * configurations. This will be further reduced by IPvX hdr size.
1145          */
1146         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1147
1148         netif_keep_dst(dev);
1149         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1150         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1151         eth_hw_addr_random(dev);
1152 }
1153
1154 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1155         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1156         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1157         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1158         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1159         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1160         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1161         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1162         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1163         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1164         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1165         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1166         [IFLA_GENEVE_TTL_INHERIT]       = { .type = NLA_U8 },
1167         [IFLA_GENEVE_DF]                = { .type = NLA_U8 },
1168 };
1169
1170 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1171                            struct netlink_ext_ack *extack)
1172 {
1173         if (tb[IFLA_ADDRESS]) {
1174                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1175                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1176                                             "Provided link layer address is not Ethernet");
1177                         return -EINVAL;
1178                 }
1179
1180                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1181                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1182                                             "Provided Ethernet address is not unicast");
1183                         return -EADDRNOTAVAIL;
1184                 }
1185         }
1186
1187         if (!data) {
1188                 NL_SET_ERR_MSG(extack,
1189                                "Not enough attributes provided to perform the operation");
1190                 return -EINVAL;
1191         }
1192
1193         if (data[IFLA_GENEVE_ID]) {
1194                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1195
1196                 if (vni >= GENEVE_N_VID) {
1197                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1198                                             "Geneve ID must be lower than 16777216");
1199                         return -ERANGE;
1200                 }
1201         }
1202
1203         if (data[IFLA_GENEVE_DF]) {
1204                 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1205
1206                 if (df < 0 || df > GENEVE_DF_MAX) {
1207                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_GENEVE_DF],
1208                                             "Invalid DF attribute");
1209                         return -EINVAL;
1210                 }
1211         }
1212
1213         return 0;
1214 }
1215
1216 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1217                                           const struct ip_tunnel_info *info,
1218                                           bool *tun_on_same_port,
1219                                           bool *tun_collect_md)
1220 {
1221         struct geneve_dev *geneve, *t = NULL;
1222
1223         *tun_on_same_port = false;
1224         *tun_collect_md = false;
1225         list_for_each_entry(geneve, &gn->geneve_list, next) {
1226                 if (info->key.tp_dst == geneve->info.key.tp_dst) {
1227                         *tun_collect_md = geneve->collect_md;
1228                         *tun_on_same_port = true;
1229                 }
1230                 if (info->key.tun_id == geneve->info.key.tun_id &&
1231                     info->key.tp_dst == geneve->info.key.tp_dst &&
1232                     !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1233                         t = geneve;
1234         }
1235         return t;
1236 }
1237
1238 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1239 {
1240         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1241                  info->key.ttl || info->key.label || info->key.tp_src ||
1242                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1243 }
1244
1245 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1246                                   struct ip_tunnel_info *b)
1247 {
1248         if (ip_tunnel_info_af(a) == AF_INET)
1249                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1250         else
1251                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1252 }
1253
1254 static int geneve_configure(struct net *net, struct net_device *dev,
1255                             struct netlink_ext_ack *extack,
1256                             const struct ip_tunnel_info *info,
1257                             bool metadata, bool ipv6_rx_csum,
1258                             bool ttl_inherit, enum ifla_geneve_df df)
1259 {
1260         struct geneve_net *gn = net_generic(net, geneve_net_id);
1261         struct geneve_dev *t, *geneve = netdev_priv(dev);
1262         bool tun_collect_md, tun_on_same_port;
1263         int err, encap_len;
1264
1265         if (metadata && !is_tnl_info_zero(info)) {
1266                 NL_SET_ERR_MSG(extack,
1267                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1268                 return -EINVAL;
1269         }
1270
1271         geneve->net = net;
1272         geneve->dev = dev;
1273
1274         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1275         if (t)
1276                 return -EBUSY;
1277
1278         /* make enough headroom for basic scenario */
1279         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1280         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1281                 encap_len += sizeof(struct iphdr);
1282                 dev->max_mtu -= sizeof(struct iphdr);
1283         } else {
1284                 encap_len += sizeof(struct ipv6hdr);
1285                 dev->max_mtu -= sizeof(struct ipv6hdr);
1286         }
1287         dev->needed_headroom = encap_len + ETH_HLEN;
1288
1289         if (metadata) {
1290                 if (tun_on_same_port) {
1291                         NL_SET_ERR_MSG(extack,
1292                                        "There can be only one externally controlled device on a destination port");
1293                         return -EPERM;
1294                 }
1295         } else {
1296                 if (tun_collect_md) {
1297                         NL_SET_ERR_MSG(extack,
1298                                        "There already exists an externally controlled device on this destination port");
1299                         return -EPERM;
1300                 }
1301         }
1302
1303         dst_cache_reset(&geneve->info.dst_cache);
1304         geneve->info = *info;
1305         geneve->collect_md = metadata;
1306         geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1307         geneve->ttl_inherit = ttl_inherit;
1308         geneve->df = df;
1309
1310         err = register_netdevice(dev);
1311         if (err)
1312                 return err;
1313
1314         list_add(&geneve->next, &gn->geneve_list);
1315         return 0;
1316 }
1317
1318 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1319 {
1320         memset(info, 0, sizeof(*info));
1321         info->key.tp_dst = htons(dst_port);
1322 }
1323
1324 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1325                           struct netlink_ext_ack *extack,
1326                           struct ip_tunnel_info *info, bool *metadata,
1327                           bool *use_udp6_rx_checksums, bool *ttl_inherit,
1328                           enum ifla_geneve_df *df, bool changelink)
1329 {
1330         int attrtype;
1331
1332         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1333                 NL_SET_ERR_MSG(extack,
1334                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1335                 return -EINVAL;
1336         }
1337
1338         if (data[IFLA_GENEVE_REMOTE]) {
1339                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1340                         attrtype = IFLA_GENEVE_REMOTE;
1341                         goto change_notsup;
1342                 }
1343
1344                 info->key.u.ipv4.dst =
1345                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1346
1347                 if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) {
1348                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1349                                             "Remote IPv4 address cannot be Multicast");
1350                         return -EINVAL;
1351                 }
1352         }
1353
1354         if (data[IFLA_GENEVE_REMOTE6]) {
1355 #if IS_ENABLED(CONFIG_IPV6)
1356                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1357                         attrtype = IFLA_GENEVE_REMOTE6;
1358                         goto change_notsup;
1359                 }
1360
1361                 info->mode = IP_TUNNEL_INFO_IPV6;
1362                 info->key.u.ipv6.dst =
1363                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1364
1365                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1366                     IPV6_ADDR_LINKLOCAL) {
1367                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1368                                             "Remote IPv6 address cannot be link-local");
1369                         return -EINVAL;
1370                 }
1371                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1372                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1373                                             "Remote IPv6 address cannot be Multicast");
1374                         return -EINVAL;
1375                 }
1376                 info->key.tun_flags |= TUNNEL_CSUM;
1377                 *use_udp6_rx_checksums = true;
1378 #else
1379                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1380                                     "IPv6 support not enabled in the kernel");
1381                 return -EPFNOSUPPORT;
1382 #endif
1383         }
1384
1385         if (data[IFLA_GENEVE_ID]) {
1386                 __u32 vni;
1387                 __u8 tvni[3];
1388                 __be64 tunid;
1389
1390                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1391                 tvni[0] = (vni & 0x00ff0000) >> 16;
1392                 tvni[1] = (vni & 0x0000ff00) >> 8;
1393                 tvni[2] =  vni & 0x000000ff;
1394
1395                 tunid = vni_to_tunnel_id(tvni);
1396                 if (changelink && (tunid != info->key.tun_id)) {
1397                         attrtype = IFLA_GENEVE_ID;
1398                         goto change_notsup;
1399                 }
1400                 info->key.tun_id = tunid;
1401         }
1402
1403         if (data[IFLA_GENEVE_TTL_INHERIT]) {
1404                 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1405                         *ttl_inherit = true;
1406                 else
1407                         *ttl_inherit = false;
1408         } else if (data[IFLA_GENEVE_TTL]) {
1409                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1410                 *ttl_inherit = false;
1411         }
1412
1413         if (data[IFLA_GENEVE_TOS])
1414                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1415
1416         if (data[IFLA_GENEVE_DF])
1417                 *df = nla_get_u8(data[IFLA_GENEVE_DF]);
1418
1419         if (data[IFLA_GENEVE_LABEL]) {
1420                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1421                                   IPV6_FLOWLABEL_MASK;
1422                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1423                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1424                                             "Label attribute only applies for IPv6 Geneve devices");
1425                         return -EINVAL;
1426                 }
1427         }
1428
1429         if (data[IFLA_GENEVE_PORT]) {
1430                 if (changelink) {
1431                         attrtype = IFLA_GENEVE_PORT;
1432                         goto change_notsup;
1433                 }
1434                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1435         }
1436
1437         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1438                 if (changelink) {
1439                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1440                         goto change_notsup;
1441                 }
1442                 *metadata = true;
1443         }
1444
1445         if (data[IFLA_GENEVE_UDP_CSUM]) {
1446                 if (changelink) {
1447                         attrtype = IFLA_GENEVE_UDP_CSUM;
1448                         goto change_notsup;
1449                 }
1450                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1451                         info->key.tun_flags |= TUNNEL_CSUM;
1452         }
1453
1454         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1455 #if IS_ENABLED(CONFIG_IPV6)
1456                 if (changelink) {
1457                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1458                         goto change_notsup;
1459                 }
1460                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1461                         info->key.tun_flags &= ~TUNNEL_CSUM;
1462 #else
1463                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1464                                     "IPv6 support not enabled in the kernel");
1465                 return -EPFNOSUPPORT;
1466 #endif
1467         }
1468
1469         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1470 #if IS_ENABLED(CONFIG_IPV6)
1471                 if (changelink) {
1472                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1473                         goto change_notsup;
1474                 }
1475                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1476                         *use_udp6_rx_checksums = false;
1477 #else
1478                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1479                                     "IPv6 support not enabled in the kernel");
1480                 return -EPFNOSUPPORT;
1481 #endif
1482         }
1483
1484         return 0;
1485 change_notsup:
1486         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1487                             "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1488         return -EOPNOTSUPP;
1489 }
1490
1491 static void geneve_link_config(struct net_device *dev,
1492                                struct ip_tunnel_info *info, struct nlattr *tb[])
1493 {
1494         struct geneve_dev *geneve = netdev_priv(dev);
1495         int ldev_mtu = 0;
1496
1497         if (tb[IFLA_MTU]) {
1498                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1499                 return;
1500         }
1501
1502         switch (ip_tunnel_info_af(info)) {
1503         case AF_INET: {
1504                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1505                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1506
1507                 if (!IS_ERR(rt) && rt->dst.dev) {
1508                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1509                         ip_rt_put(rt);
1510                 }
1511                 break;
1512         }
1513 #if IS_ENABLED(CONFIG_IPV6)
1514         case AF_INET6: {
1515                 struct rt6_info *rt;
1516
1517                 if (!__in6_dev_get(dev))
1518                         break;
1519
1520                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1521                                 NULL, 0);
1522
1523                 if (rt && rt->dst.dev)
1524                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1525                 ip6_rt_put(rt);
1526                 break;
1527         }
1528 #endif
1529         }
1530
1531         if (ldev_mtu <= 0)
1532                 return;
1533
1534         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1535 }
1536
1537 static int geneve_newlink(struct net *net, struct net_device *dev,
1538                           struct nlattr *tb[], struct nlattr *data[],
1539                           struct netlink_ext_ack *extack)
1540 {
1541         enum ifla_geneve_df df = GENEVE_DF_UNSET;
1542         bool use_udp6_rx_checksums = false;
1543         struct ip_tunnel_info info;
1544         bool ttl_inherit = false;
1545         bool metadata = false;
1546         int err;
1547
1548         init_tnl_info(&info, GENEVE_UDP_PORT);
1549         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1550                              &use_udp6_rx_checksums, &ttl_inherit, &df, false);
1551         if (err)
1552                 return err;
1553
1554         err = geneve_configure(net, dev, extack, &info, metadata,
1555                                use_udp6_rx_checksums, ttl_inherit, df);
1556         if (err)
1557                 return err;
1558
1559         geneve_link_config(dev, &info, tb);
1560
1561         return 0;
1562 }
1563
1564 /* Quiesces the geneve device data path for both TX and RX.
1565  *
1566  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1567  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1568  * to complete for the existing set of in-flight packets to be transmitted,
1569  * then we would have quiesced the transmit data path. All the future packets
1570  * will get dropped until we unquiesce the data path.
1571  *
1572  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1573  * if we set that to NULL under RCU and wait for synchronize_net() to
1574  * complete, then we would have quiesced the receive data path.
1575  */
1576 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1577                            struct geneve_sock **gs6)
1578 {
1579         *gs4 = rtnl_dereference(geneve->sock4);
1580         rcu_assign_pointer(geneve->sock4, NULL);
1581         if (*gs4)
1582                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1583 #if IS_ENABLED(CONFIG_IPV6)
1584         *gs6 = rtnl_dereference(geneve->sock6);
1585         rcu_assign_pointer(geneve->sock6, NULL);
1586         if (*gs6)
1587                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1588 #else
1589         *gs6 = NULL;
1590 #endif
1591         synchronize_net();
1592 }
1593
1594 /* Resumes the geneve device data path for both TX and RX. */
1595 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1596                              struct geneve_sock __maybe_unused *gs6)
1597 {
1598         rcu_assign_pointer(geneve->sock4, gs4);
1599         if (gs4)
1600                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1601 #if IS_ENABLED(CONFIG_IPV6)
1602         rcu_assign_pointer(geneve->sock6, gs6);
1603         if (gs6)
1604                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1605 #endif
1606         synchronize_net();
1607 }
1608
1609 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1610                              struct nlattr *data[],
1611                              struct netlink_ext_ack *extack)
1612 {
1613         struct geneve_dev *geneve = netdev_priv(dev);
1614         struct geneve_sock *gs4, *gs6;
1615         struct ip_tunnel_info info;
1616         bool metadata;
1617         bool use_udp6_rx_checksums;
1618         enum ifla_geneve_df df;
1619         bool ttl_inherit;
1620         int err;
1621
1622         /* If the geneve device is configured for metadata (or externally
1623          * controlled, for example, OVS), then nothing can be changed.
1624          */
1625         if (geneve->collect_md)
1626                 return -EOPNOTSUPP;
1627
1628         /* Start with the existing info. */
1629         memcpy(&info, &geneve->info, sizeof(info));
1630         metadata = geneve->collect_md;
1631         use_udp6_rx_checksums = geneve->use_udp6_rx_checksums;
1632         ttl_inherit = geneve->ttl_inherit;
1633         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1634                              &use_udp6_rx_checksums, &ttl_inherit, &df, true);
1635         if (err)
1636                 return err;
1637
1638         if (!geneve_dst_addr_equal(&geneve->info, &info)) {
1639                 dst_cache_reset(&info.dst_cache);
1640                 geneve_link_config(dev, &info, tb);
1641         }
1642
1643         geneve_quiesce(geneve, &gs4, &gs6);
1644         geneve->info = info;
1645         geneve->collect_md = metadata;
1646         geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
1647         geneve->ttl_inherit = ttl_inherit;
1648         geneve_unquiesce(geneve, gs4, gs6);
1649
1650         return 0;
1651 }
1652
1653 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1654 {
1655         struct geneve_dev *geneve = netdev_priv(dev);
1656
1657         list_del(&geneve->next);
1658         unregister_netdevice_queue(dev, head);
1659 }
1660
1661 static size_t geneve_get_size(const struct net_device *dev)
1662 {
1663         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1664                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1665                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1666                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1667                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_DF */
1668                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1669                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1670                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1671                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1672                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1673                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1674                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1675                 0;
1676 }
1677
1678 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1679 {
1680         struct geneve_dev *geneve = netdev_priv(dev);
1681         struct ip_tunnel_info *info = &geneve->info;
1682         bool ttl_inherit = geneve->ttl_inherit;
1683         bool metadata = geneve->collect_md;
1684         __u8 tmp_vni[3];
1685         __u32 vni;
1686
1687         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1688         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1689         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1690                 goto nla_put_failure;
1691
1692         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1693                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1694                                     info->key.u.ipv4.dst))
1695                         goto nla_put_failure;
1696                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1697                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1698                         goto nla_put_failure;
1699
1700 #if IS_ENABLED(CONFIG_IPV6)
1701         } else if (!metadata) {
1702                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1703                                      &info->key.u.ipv6.dst))
1704                         goto nla_put_failure;
1705                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1706                                !(info->key.tun_flags & TUNNEL_CSUM)))
1707                         goto nla_put_failure;
1708 #endif
1709         }
1710
1711         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1712             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1713             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1714                 goto nla_put_failure;
1715
1716         if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->df))
1717                 goto nla_put_failure;
1718
1719         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1720                 goto nla_put_failure;
1721
1722         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1723                 goto nla_put_failure;
1724
1725 #if IS_ENABLED(CONFIG_IPV6)
1726         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1727                        !geneve->use_udp6_rx_checksums))
1728                 goto nla_put_failure;
1729 #endif
1730
1731         if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1732                 goto nla_put_failure;
1733
1734         return 0;
1735
1736 nla_put_failure:
1737         return -EMSGSIZE;
1738 }
1739
1740 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1741         .kind           = "geneve",
1742         .maxtype        = IFLA_GENEVE_MAX,
1743         .policy         = geneve_policy,
1744         .priv_size      = sizeof(struct geneve_dev),
1745         .setup          = geneve_setup,
1746         .validate       = geneve_validate,
1747         .newlink        = geneve_newlink,
1748         .changelink     = geneve_changelink,
1749         .dellink        = geneve_dellink,
1750         .get_size       = geneve_get_size,
1751         .fill_info      = geneve_fill_info,
1752 };
1753
1754 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1755                                         u8 name_assign_type, u16 dst_port)
1756 {
1757         struct nlattr *tb[IFLA_MAX + 1];
1758         struct ip_tunnel_info info;
1759         struct net_device *dev;
1760         LIST_HEAD(list_kill);
1761         int err;
1762
1763         memset(tb, 0, sizeof(tb));
1764         dev = rtnl_create_link(net, name, name_assign_type,
1765                                &geneve_link_ops, tb, NULL);
1766         if (IS_ERR(dev))
1767                 return dev;
1768
1769         init_tnl_info(&info, dst_port);
1770         err = geneve_configure(net, dev, NULL, &info,
1771                                true, true, false, GENEVE_DF_UNSET);
1772         if (err) {
1773                 free_netdev(dev);
1774                 return ERR_PTR(err);
1775         }
1776
1777         /* openvswitch users expect packet sizes to be unrestricted,
1778          * so set the largest MTU we can.
1779          */
1780         err = geneve_change_mtu(dev, IP_MAX_MTU);
1781         if (err)
1782                 goto err;
1783
1784         err = rtnl_configure_link(dev, NULL);
1785         if (err < 0)
1786                 goto err;
1787
1788         return dev;
1789 err:
1790         geneve_dellink(dev, &list_kill);
1791         unregister_netdevice_many(&list_kill);
1792         return ERR_PTR(err);
1793 }
1794 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1795
1796 static int geneve_netdevice_event(struct notifier_block *unused,
1797                                   unsigned long event, void *ptr)
1798 {
1799         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1800
1801         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1802             event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1803                 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1804         } else if (event == NETDEV_UNREGISTER) {
1805                 geneve_offload_rx_ports(dev, false);
1806         } else if (event == NETDEV_REGISTER) {
1807                 geneve_offload_rx_ports(dev, true);
1808         }
1809
1810         return NOTIFY_DONE;
1811 }
1812
1813 static struct notifier_block geneve_notifier_block __read_mostly = {
1814         .notifier_call = geneve_netdevice_event,
1815 };
1816
1817 static __net_init int geneve_init_net(struct net *net)
1818 {
1819         struct geneve_net *gn = net_generic(net, geneve_net_id);
1820
1821         INIT_LIST_HEAD(&gn->geneve_list);
1822         INIT_LIST_HEAD(&gn->sock_list);
1823         return 0;
1824 }
1825
1826 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1827 {
1828         struct geneve_net *gn = net_generic(net, geneve_net_id);
1829         struct geneve_dev *geneve, *next;
1830         struct net_device *dev, *aux;
1831
1832         /* gather any geneve devices that were moved into this ns */
1833         for_each_netdev_safe(net, dev, aux)
1834                 if (dev->rtnl_link_ops == &geneve_link_ops)
1835                         unregister_netdevice_queue(dev, head);
1836
1837         /* now gather any other geneve devices that were created in this ns */
1838         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1839                 /* If geneve->dev is in the same netns, it was already added
1840                  * to the list by the previous loop.
1841                  */
1842                 if (!net_eq(dev_net(geneve->dev), net))
1843                         unregister_netdevice_queue(geneve->dev, head);
1844         }
1845
1846         WARN_ON_ONCE(!list_empty(&gn->sock_list));
1847 }
1848
1849 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1850 {
1851         struct net *net;
1852         LIST_HEAD(list);
1853
1854         rtnl_lock();
1855         list_for_each_entry(net, net_list, exit_list)
1856                 geneve_destroy_tunnels(net, &list);
1857
1858         /* unregister the devices gathered above */
1859         unregister_netdevice_many(&list);
1860         rtnl_unlock();
1861 }
1862
1863 static struct pernet_operations geneve_net_ops = {
1864         .init = geneve_init_net,
1865         .exit_batch = geneve_exit_batch_net,
1866         .id   = &geneve_net_id,
1867         .size = sizeof(struct geneve_net),
1868 };
1869
1870 static int __init geneve_init_module(void)
1871 {
1872         int rc;
1873
1874         rc = register_pernet_subsys(&geneve_net_ops);
1875         if (rc)
1876                 goto out1;
1877
1878         rc = register_netdevice_notifier(&geneve_notifier_block);
1879         if (rc)
1880                 goto out2;
1881
1882         rc = rtnl_link_register(&geneve_link_ops);
1883         if (rc)
1884                 goto out3;
1885
1886         return 0;
1887 out3:
1888         unregister_netdevice_notifier(&geneve_notifier_block);
1889 out2:
1890         unregister_pernet_subsys(&geneve_net_ops);
1891 out1:
1892         return rc;
1893 }
1894 late_initcall(geneve_init_module);
1895
1896 static void __exit geneve_cleanup_module(void)
1897 {
1898         rtnl_link_unregister(&geneve_link_ops);
1899         unregister_netdevice_notifier(&geneve_notifier_block);
1900         unregister_pernet_subsys(&geneve_net_ops);
1901 }
1902 module_exit(geneve_cleanup_module);
1903
1904 MODULE_LICENSE("GPL");
1905 MODULE_VERSION(GENEVE_NETDEV_VER);
1906 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1907 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1908 MODULE_ALIAS_RTNL_LINK("geneve");