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 metadata = geneve->collect_md;
 696         bool ipv4, ipv6;
 697         int ret = 0;
 698
 699         ipv6 = geneve->info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 700         ipv4 = !ipv6 || metadata;
 701 #if IS_ENABLED(CONFIG_IPV6)
 702         if (ipv6) {
 703                 ret = geneve_sock_add(geneve, true);
 704                 if (ret < 0 && ret != -EAFNOSUPPORT)
 705                         ipv4 = false;
 706         }
 707 #endif
 708         if (ipv4)
 709                 ret = geneve_sock_add(geneve, false);
 710         if (ret < 0)
 711                 geneve_sock_release(geneve);
 712
 713         return ret;
 714 }
 715
 716 static int geneve_stop(struct net_device *dev)
 717 {
 718         struct geneve_dev *geneve = netdev_priv(dev);
 719
 720         hlist_del_init_rcu(&geneve->hlist4.hlist);
 721 #if IS_ENABLED(CONFIG_IPV6)
 722         hlist_del_init_rcu(&geneve->hlist6.hlist);
 723 #endif
 724         geneve_sock_release(geneve);
 725         return 0;
 726 }
 727
 728 static void geneve_build_header(struct genevehdr *geneveh,
 729                                 const struct ip_tunnel_info *info)
 730 {
 731         geneveh->ver = GENEVE_VER;
 732         geneveh->opt_len = info->options_len / 4;
 733         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 734         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 735         geneveh->rsvd1 = 0;
 736         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 737         geneveh->proto_type = htons(ETH_P_TEB);
 738         geneveh->rsvd2 = 0;
 739
 740         if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
 741                 ip_tunnel_info_opts_get(geneveh->options, info);
 742 }
 743
 744 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 745                             const struct ip_tunnel_info *info,
 746                             bool xnet, int ip_hdr_len)
 747 {
 748         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 749         struct genevehdr *gnvh;
 750         int min_headroom;
 751         int err;
 752
 753         skb_reset_mac_header(skb);
 754         skb_scrub_packet(skb, xnet);
 755
 756         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 757                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 758         err = skb_cow_head(skb, min_headroom);
 759         if (unlikely(err))
 760                 goto free_dst;
 761
 762         err = udp_tunnel_handle_offloads(skb, udp_sum);
 763         if (err)
 764                 goto free_dst;
 765
 766         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 767         geneve_build_header(gnvh, info);
 768         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 769         return 0;
 770
 771 free_dst:
 772         dst_release(dst);
 773         return err;
 774 }
 775
 776 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 777                                        struct net_device *dev,
 778                                        struct geneve_sock *gs4,
 779                                        struct flowi4 *fl4,
 780                                        const struct ip_tunnel_info *info)
 781 {
 782         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 783         struct geneve_dev *geneve = netdev_priv(dev);
 784         struct dst_cache *dst_cache;
 785         struct rtable *rt = NULL;
 786         __u8 tos;
 787
 788         if (!gs4)
 789                 return ERR_PTR(-EIO);
 790
 791         memset(fl4, 0, sizeof(*fl4));
 792         fl4->flowi4_mark = skb->mark;
 793         fl4->flowi4_proto = IPPROTO_UDP;
 794         fl4->daddr = info->key.u.ipv4.dst;
 795         fl4->saddr = info->key.u.ipv4.src;
 796
 797         tos = info->key.tos;
 798         if ((tos == 1) && !geneve->collect_md) {
 799                 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 800                 use_cache = false;
 801         }
 802         fl4->flowi4_tos = RT_TOS(tos);
 803
 804         dst_cache = (struct dst_cache *)&info->dst_cache;
 805         if (use_cache) {
 806                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 807                 if (rt)
 808                         return rt;
 809         }
 810         rt = ip_route_output_key(geneve->net, fl4);
 811         if (IS_ERR(rt)) {
 812                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 813                 return ERR_PTR(-ENETUNREACH);
 814         }
 815         if (rt->dst.dev == dev) { /* is this necessary? */
 816                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 817                 ip_rt_put(rt);
 818                 return ERR_PTR(-ELOOP);
 819         }
 820         if (use_cache)
 821                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 822         return rt;
 823 }
 824
 825 #if IS_ENABLED(CONFIG_IPV6)
 826 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 827                                            struct net_device *dev,
 828                                            struct geneve_sock *gs6,
 829                                            struct flowi6 *fl6,
 830                                            const struct ip_tunnel_info *info)
 831 {
 832         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 833         struct geneve_dev *geneve = netdev_priv(dev);
 834         struct dst_entry *dst = NULL;
 835         struct dst_cache *dst_cache;
 836         __u8 prio;
 837
 838         if (!gs6)
 839                 return ERR_PTR(-EIO);
 840
 841         memset(fl6, 0, sizeof(*fl6));
 842         fl6->flowi6_mark = skb->mark;
 843         fl6->flowi6_proto = IPPROTO_UDP;
 844         fl6->daddr = info->key.u.ipv6.dst;
 845         fl6->saddr = info->key.u.ipv6.src;
 846         prio = info->key.tos;
 847         if ((prio == 1) && !geneve->collect_md) {
 848                 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 849                 use_cache = false;
 850         }
 851
 852         fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 853                                            info->key.label);
 854         dst_cache = (struct dst_cache *)&info->dst_cache;
 855         if (use_cache) {
 856                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 857                 if (dst)
 858                         return dst;
 859         }
 860         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 861                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 862                 return ERR_PTR(-ENETUNREACH);
 863         }
 864         if (dst->dev == dev) { /* is this necessary? */
 865                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 866                 dst_release(dst);
 867                 return ERR_PTR(-ELOOP);
 868         }
 869
 870         if (use_cache)
 871                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 872         return dst;
 873 }
 874 #endif
 875
 876 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 877                            struct geneve_dev *geneve,
 878                            const struct ip_tunnel_info *info)
 879 {
 880         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 881         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 882         const struct ip_tunnel_key *key = &info->key;
 883         struct rtable *rt;
 884         struct flowi4 fl4;
 885         __u8 tos, ttl;
 886         __be16 df = 0;
 887         __be16 sport;
 888         int err;
 889
 890         rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
 891         if (IS_ERR(rt))
 892                 return PTR_ERR(rt);
 893
 894         skb_tunnel_check_pmtu(skb, &rt->dst,
 895                               GENEVE_IPV4_HLEN + info->options_len);
 896
 897         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 898         if (geneve->collect_md) {
 899                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 900                 ttl = key->ttl;
 901
 902                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 903         } else {
 904                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 905                 if (geneve->ttl_inherit)
 906                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 907                 else
 908                         ttl = key->ttl;
 909                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 910
 911                 if (geneve->df == GENEVE_DF_SET) {
 912                         df = htons(IP_DF);
 913                 } else if (geneve->df == GENEVE_DF_INHERIT) {
 914                         struct ethhdr *eth = eth_hdr(skb);
 915
 916                         if (ntohs(eth->h_proto) == ETH_P_IPV6) {
 917                                 df = htons(IP_DF);
 918                         } else if (ntohs(eth->h_proto) == ETH_P_IP) {
 919                                 struct iphdr *iph = ip_hdr(skb);
 920
 921                                 if (iph->frag_off & htons(IP_DF))
 922                                         df = htons(IP_DF);
 923                         }
 924                 }
 925         }
 926
 927         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
 928         if (unlikely(err))
 929                 return err;
 930
 931         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 932                             tos, ttl, df, sport, geneve->info.key.tp_dst,
 933                             !net_eq(geneve->net, dev_net(geneve->dev)),
 934                             !(info->key.tun_flags & TUNNEL_CSUM));
 935         return 0;
 936 }
 937
 938 #if IS_ENABLED(CONFIG_IPV6)
 939 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 940                             struct geneve_dev *geneve,
 941                             const struct ip_tunnel_info *info)
 942 {
 943         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 944         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 945         const struct ip_tunnel_key *key = &info->key;
 946         struct dst_entry *dst = NULL;
 947         struct flowi6 fl6;
 948         __u8 prio, ttl;
 949         __be16 sport;
 950         int err;
 951
 952         dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
 953         if (IS_ERR(dst))
 954                 return PTR_ERR(dst);
 955
 956         skb_tunnel_check_pmtu(skb, dst, GENEVE_IPV6_HLEN + info->options_len);
 957
 958         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 959         if (geneve->collect_md) {
 960                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 961                 ttl = key->ttl;
 962         } else {
 963                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
 964                                            ip_hdr(skb), skb);
 965                 if (geneve->ttl_inherit)
 966                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 967                 else
 968                         ttl = key->ttl;
 969                 ttl = ttl ? : ip6_dst_hoplimit(dst);
 970         }
 971         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
 972         if (unlikely(err))
 973                 return err;
 974
 975         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
 976                              &fl6.saddr, &fl6.daddr, prio, ttl,
 977                              info->key.label, sport, geneve->info.key.tp_dst,
 978                              !(info->key.tun_flags & TUNNEL_CSUM));
 979         return 0;
 980 }
 981 #endif
 982
 983 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
 984 {
 985         struct geneve_dev *geneve = netdev_priv(dev);
 986         struct ip_tunnel_info *info = NULL;
 987         int err;
 988
 989         if (geneve->collect_md) {
 990                 info = skb_tunnel_info(skb);
 991                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 992                         err = -EINVAL;
 993                         netdev_dbg(dev, "no tunnel metadata\n");
 994                         goto tx_error;
 995                 }
 996         } else {
 997                 info = &geneve->info;
 998         }
 999
1000         rcu_read_lock();
1001 #if IS_ENABLED(CONFIG_IPV6)
1002         if (info->mode & IP_TUNNEL_INFO_IPV6)
1003                 err = geneve6_xmit_skb(skb, dev, geneve, info);
1004         else
1005 #endif
1006                 err = geneve_xmit_skb(skb, dev, geneve, info);
1007         rcu_read_unlock();
1008
1009         if (likely(!err))
1010                 return NETDEV_TX_OK;
1011 tx_error:
1012         dev_kfree_skb(skb);
1013
1014         if (err == -ELOOP)
1015                 dev->stats.collisions++;
1016         else if (err == -ENETUNREACH)
1017                 dev->stats.tx_carrier_errors++;
1018
1019         dev->stats.tx_errors++;
1020         return NETDEV_TX_OK;
1021 }
1022
1023 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1024 {
1025         if (new_mtu > dev->max_mtu)
1026                 new_mtu = dev->max_mtu;
1027         else if (new_mtu < dev->min_mtu)
1028                 new_mtu = dev->min_mtu;
1029
1030         dev->mtu = new_mtu;
1031         return 0;
1032 }
1033
1034 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1035 {
1036         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1037         struct geneve_dev *geneve = netdev_priv(dev);
1038
1039         if (ip_tunnel_info_af(info) == AF_INET) {
1040                 struct rtable *rt;
1041                 struct flowi4 fl4;
1042                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1043
1044                 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
1045                 if (IS_ERR(rt))
1046                         return PTR_ERR(rt);
1047
1048                 ip_rt_put(rt);
1049                 info->key.u.ipv4.src = fl4.saddr;
1050 #if IS_ENABLED(CONFIG_IPV6)
1051         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1052                 struct dst_entry *dst;
1053                 struct flowi6 fl6;
1054                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1055
1056                 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
1057                 if (IS_ERR(dst))
1058                         return PTR_ERR(dst);
1059
1060                 dst_release(dst);
1061                 info->key.u.ipv6.src = fl6.saddr;
1062 #endif
1063         } else {
1064                 return -EINVAL;
1065         }
1066
1067         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1068                                              1, USHRT_MAX, true);
1069         info->key.tp_dst = geneve->info.key.tp_dst;
1070         return 0;
1071 }
1072
1073 static const struct net_device_ops geneve_netdev_ops = {
1074         .ndo_init               = geneve_init,
1075         .ndo_uninit             = geneve_uninit,
1076         .ndo_open               = geneve_open,
1077         .ndo_stop               = geneve_stop,
1078         .ndo_start_xmit         = geneve_xmit,
1079         .ndo_get_stats64        = ip_tunnel_get_stats64,
1080         .ndo_change_mtu         = geneve_change_mtu,
1081         .ndo_validate_addr      = eth_validate_addr,
1082         .ndo_set_mac_address    = eth_mac_addr,
1083         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1084 };
1085
1086 static void geneve_get_drvinfo(struct net_device *dev,
1087                                struct ethtool_drvinfo *drvinfo)
1088 {
1089         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1090         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1091 }
1092
1093 static const struct ethtool_ops geneve_ethtool_ops = {
1094         .get_drvinfo    = geneve_get_drvinfo,
1095         .get_link       = ethtool_op_get_link,
1096 };
1097
1098 /* Info for udev, that this is a virtual tunnel endpoint */
1099 static struct device_type geneve_type = {
1100         .name = "geneve",
1101 };
1102
1103 /* Calls the ndo_udp_tunnel_add of the caller in order to
1104  * supply the listening GENEVE udp ports. Callers are expected
1105  * to implement the ndo_udp_tunnel_add.
1106  */
1107 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1108 {
1109         struct net *net = dev_net(dev);
1110         struct geneve_net *gn = net_generic(net, geneve_net_id);
1111         struct geneve_sock *gs;
1112
1113         rcu_read_lock();
1114         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1115                 if (push) {
1116                         udp_tunnel_push_rx_port(dev, gs->sock,
1117                                                 UDP_TUNNEL_TYPE_GENEVE);
1118                 } else {
1119                         udp_tunnel_drop_rx_port(dev, gs->sock,
1120                                                 UDP_TUNNEL_TYPE_GENEVE);
1121                 }
1122         }
1123         rcu_read_unlock();
1124 }
1125
1126 /* Initialize the device structure. */
1127 static void geneve_setup(struct net_device *dev)
1128 {
1129         ether_setup(dev);
1130
1131         dev->netdev_ops = &geneve_netdev_ops;
1132         dev->ethtool_ops = &geneve_ethtool_ops;
1133         dev->needs_free_netdev = true;
1134
1135         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1136
1137         dev->features    |= NETIF_F_LLTX;
1138         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1139         dev->features    |= NETIF_F_RXCSUM;
1140         dev->features    |= NETIF_F_GSO_SOFTWARE;
1141
1142         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1143         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1144
1145         /* MTU range: 68 - (something less than 65535) */
1146         dev->min_mtu = ETH_MIN_MTU;
1147         /* The max_mtu calculation does not take account of GENEVE
1148          * options, to avoid excluding potentially valid
1149          * configurations. This will be further reduced by IPvX hdr size.
1150          */
1151         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1152
1153         netif_keep_dst(dev);
1154         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1155         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1156         eth_hw_addr_random(dev);
1157 }
1158
1159 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1160         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1161         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1162         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1163         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1164         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1165         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1166         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1167         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1168         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1169         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1170         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1171         [IFLA_GENEVE_TTL_INHERIT]       = { .type = NLA_U8 },
1172         [IFLA_GENEVE_DF]                = { .type = NLA_U8 },
1173 };
1174
1175 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1176                            struct netlink_ext_ack *extack)
1177 {
1178         if (tb[IFLA_ADDRESS]) {
1179                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1180                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1181                                             "Provided link layer address is not Ethernet");
1182                         return -EINVAL;
1183                 }
1184
1185                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1186                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1187                                             "Provided Ethernet address is not unicast");
1188                         return -EADDRNOTAVAIL;
1189                 }
1190         }
1191
1192         if (!data) {
1193                 NL_SET_ERR_MSG(extack,
1194                                "Not enough attributes provided to perform the operation");
1195                 return -EINVAL;
1196         }
1197
1198         if (data[IFLA_GENEVE_ID]) {
1199                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1200
1201                 if (vni >= GENEVE_N_VID) {
1202                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1203                                             "Geneve ID must be lower than 16777216");
1204                         return -ERANGE;
1205                 }
1206         }
1207
1208         if (data[IFLA_GENEVE_DF]) {
1209                 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1210
1211                 if (df < 0 || df > GENEVE_DF_MAX) {
1212                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_GENEVE_DF],
1213                                             "Invalid DF attribute");
1214                         return -EINVAL;
1215                 }
1216         }
1217
1218         return 0;
1219 }
1220
1221 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1222                                           const struct ip_tunnel_info *info,
1223                                           bool *tun_on_same_port,
1224                                           bool *tun_collect_md)
1225 {
1226         struct geneve_dev *geneve, *t = NULL;
1227
1228         *tun_on_same_port = false;
1229         *tun_collect_md = false;
1230         list_for_each_entry(geneve, &gn->geneve_list, next) {
1231                 if (info->key.tp_dst == geneve->info.key.tp_dst) {
1232                         *tun_collect_md = geneve->collect_md;
1233                         *tun_on_same_port = true;
1234                 }
1235                 if (info->key.tun_id == geneve->info.key.tun_id &&
1236                     info->key.tp_dst == geneve->info.key.tp_dst &&
1237                     !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1238                         t = geneve;
1239         }
1240         return t;
1241 }
1242
1243 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1244 {
1245         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1246                  info->key.ttl || info->key.label || info->key.tp_src ||
1247                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1248 }
1249
1250 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1251                                   struct ip_tunnel_info *b)
1252 {
1253         if (ip_tunnel_info_af(a) == AF_INET)
1254                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1255         else
1256                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1257 }
1258
1259 static int geneve_configure(struct net *net, struct net_device *dev,
1260                             struct netlink_ext_ack *extack,
1261                             const struct ip_tunnel_info *info,
1262                             bool metadata, bool ipv6_rx_csum,
1263                             bool ttl_inherit, enum ifla_geneve_df df)
1264 {
1265         struct geneve_net *gn = net_generic(net, geneve_net_id);
1266         struct geneve_dev *t, *geneve = netdev_priv(dev);
1267         bool tun_collect_md, tun_on_same_port;
1268         int err, encap_len;
1269
1270         if (metadata && !is_tnl_info_zero(info)) {
1271                 NL_SET_ERR_MSG(extack,
1272                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1273                 return -EINVAL;
1274         }
1275
1276         geneve->net = net;
1277         geneve->dev = dev;
1278
1279         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1280         if (t)
1281                 return -EBUSY;
1282
1283         /* make enough headroom for basic scenario */
1284         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1285         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1286                 encap_len += sizeof(struct iphdr);
1287                 dev->max_mtu -= sizeof(struct iphdr);
1288         } else {
1289                 encap_len += sizeof(struct ipv6hdr);
1290                 dev->max_mtu -= sizeof(struct ipv6hdr);
1291         }
1292         dev->needed_headroom = encap_len + ETH_HLEN;
1293
1294         if (metadata) {
1295                 if (tun_on_same_port) {
1296                         NL_SET_ERR_MSG(extack,
1297                                        "There can be only one externally controlled device on a destination port");
1298                         return -EPERM;
1299                 }
1300         } else {
1301                 if (tun_collect_md) {
1302                         NL_SET_ERR_MSG(extack,
1303                                        "There already exists an externally controlled device on this destination port");
1304                         return -EPERM;
1305                 }
1306         }
1307
1308         dst_cache_reset(&geneve->info.dst_cache);
1309         geneve->info = *info;
1310         geneve->collect_md = metadata;
1311         geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1312         geneve->ttl_inherit = ttl_inherit;
1313         geneve->df = df;
1314
1315         err = register_netdevice(dev);
1316         if (err)
1317                 return err;
1318
1319         list_add(&geneve->next, &gn->geneve_list);
1320         return 0;
1321 }
1322
1323 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1324 {
1325         memset(info, 0, sizeof(*info));
1326         info->key.tp_dst = htons(dst_port);
1327 }
1328
1329 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1330                           struct netlink_ext_ack *extack,
1331                           struct ip_tunnel_info *info, bool *metadata,
1332                           bool *use_udp6_rx_checksums, bool *ttl_inherit,
1333                           enum ifla_geneve_df *df, bool changelink)
1334 {
1335         int attrtype;
1336
1337         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1338                 NL_SET_ERR_MSG(extack,
1339                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1340                 return -EINVAL;
1341         }
1342
1343         if (data[IFLA_GENEVE_REMOTE]) {
1344                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1345                         attrtype = IFLA_GENEVE_REMOTE;
1346                         goto change_notsup;
1347                 }
1348
1349                 info->key.u.ipv4.dst =
1350                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1351
1352                 if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) {
1353                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1354                                             "Remote IPv4 address cannot be Multicast");
1355                         return -EINVAL;
1356                 }
1357         }
1358
1359         if (data[IFLA_GENEVE_REMOTE6]) {
1360 #if IS_ENABLED(CONFIG_IPV6)
1361                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1362                         attrtype = IFLA_GENEVE_REMOTE6;
1363                         goto change_notsup;
1364                 }
1365
1366                 info->mode = IP_TUNNEL_INFO_IPV6;
1367                 info->key.u.ipv6.dst =
1368                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1369
1370                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1371                     IPV6_ADDR_LINKLOCAL) {
1372                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1373                                             "Remote IPv6 address cannot be link-local");
1374                         return -EINVAL;
1375                 }
1376                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1377                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1378                                             "Remote IPv6 address cannot be Multicast");
1379                         return -EINVAL;
1380                 }
1381                 info->key.tun_flags |= TUNNEL_CSUM;
1382                 *use_udp6_rx_checksums = true;
1383 #else
1384                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1385                                     "IPv6 support not enabled in the kernel");
1386                 return -EPFNOSUPPORT;
1387 #endif
1388         }
1389
1390         if (data[IFLA_GENEVE_ID]) {
1391                 __u32 vni;
1392                 __u8 tvni[3];
1393                 __be64 tunid;
1394
1395                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1396                 tvni[0] = (vni & 0x00ff0000) >> 16;
1397                 tvni[1] = (vni & 0x0000ff00) >> 8;
1398                 tvni[2] =  vni & 0x000000ff;
1399
1400                 tunid = vni_to_tunnel_id(tvni);
1401                 if (changelink && (tunid != info->key.tun_id)) {
1402                         attrtype = IFLA_GENEVE_ID;
1403                         goto change_notsup;
1404                 }
1405                 info->key.tun_id = tunid;
1406         }
1407
1408         if (data[IFLA_GENEVE_TTL_INHERIT]) {
1409                 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1410                         *ttl_inherit = true;
1411                 else
1412                         *ttl_inherit = false;
1413         } else if (data[IFLA_GENEVE_TTL]) {
1414                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1415                 *ttl_inherit = false;
1416         }
1417
1418         if (data[IFLA_GENEVE_TOS])
1419                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1420
1421         if (data[IFLA_GENEVE_DF])
1422                 *df = nla_get_u8(data[IFLA_GENEVE_DF]);
1423
1424         if (data[IFLA_GENEVE_LABEL]) {
1425                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1426                                   IPV6_FLOWLABEL_MASK;
1427                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1428                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1429                                             "Label attribute only applies for IPv6 Geneve devices");
1430                         return -EINVAL;
1431                 }
1432         }
1433
1434         if (data[IFLA_GENEVE_PORT]) {
1435                 if (changelink) {
1436                         attrtype = IFLA_GENEVE_PORT;
1437                         goto change_notsup;
1438                 }
1439                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1440         }
1441
1442         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1443                 if (changelink) {
1444                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1445                         goto change_notsup;
1446                 }
1447                 *metadata = true;
1448         }
1449
1450         if (data[IFLA_GENEVE_UDP_CSUM]) {
1451                 if (changelink) {
1452                         attrtype = IFLA_GENEVE_UDP_CSUM;
1453                         goto change_notsup;
1454                 }
1455                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1456                         info->key.tun_flags |= TUNNEL_CSUM;
1457         }
1458
1459         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1460 #if IS_ENABLED(CONFIG_IPV6)
1461                 if (changelink) {
1462                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1463                         goto change_notsup;
1464                 }
1465                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1466                         info->key.tun_flags &= ~TUNNEL_CSUM;
1467 #else
1468                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1469                                     "IPv6 support not enabled in the kernel");
1470                 return -EPFNOSUPPORT;
1471 #endif
1472         }
1473
1474         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1475 #if IS_ENABLED(CONFIG_IPV6)
1476                 if (changelink) {
1477                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1478                         goto change_notsup;
1479                 }
1480                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1481                         *use_udp6_rx_checksums = false;
1482 #else
1483                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1484                                     "IPv6 support not enabled in the kernel");
1485                 return -EPFNOSUPPORT;
1486 #endif
1487         }
1488
1489         return 0;
1490 change_notsup:
1491         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1492                             "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1493         return -EOPNOTSUPP;
1494 }
1495
1496 static void geneve_link_config(struct net_device *dev,
1497                                struct ip_tunnel_info *info, struct nlattr *tb[])
1498 {
1499         struct geneve_dev *geneve = netdev_priv(dev);
1500         int ldev_mtu = 0;
1501
1502         if (tb[IFLA_MTU]) {
1503                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1504                 return;
1505         }
1506
1507         switch (ip_tunnel_info_af(info)) {
1508         case AF_INET: {
1509                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1510                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1511
1512                 if (!IS_ERR(rt) && rt->dst.dev) {
1513                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1514                         ip_rt_put(rt);
1515                 }
1516                 break;
1517         }
1518 #if IS_ENABLED(CONFIG_IPV6)
1519         case AF_INET6: {
1520                 struct rt6_info *rt;
1521
1522                 if (!__in6_dev_get(dev))
1523                         break;
1524
1525                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1526                                 NULL, 0);
1527
1528                 if (rt && rt->dst.dev)
1529                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1530                 ip6_rt_put(rt);
1531                 break;
1532         }
1533 #endif
1534         }
1535
1536         if (ldev_mtu <= 0)
1537                 return;
1538
1539         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1540 }
1541
1542 static int geneve_newlink(struct net *net, struct net_device *dev,
1543                           struct nlattr *tb[], struct nlattr *data[],
1544                           struct netlink_ext_ack *extack)
1545 {
1546         enum ifla_geneve_df df = GENEVE_DF_UNSET;
1547         bool use_udp6_rx_checksums = false;
1548         struct ip_tunnel_info info;
1549         bool ttl_inherit = false;
1550         bool metadata = false;
1551         int err;
1552
1553         init_tnl_info(&info, GENEVE_UDP_PORT);
1554         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1555                              &use_udp6_rx_checksums, &ttl_inherit, &df, false);
1556         if (err)
1557                 return err;
1558
1559         err = geneve_configure(net, dev, extack, &info, metadata,
1560                                use_udp6_rx_checksums, ttl_inherit, df);
1561         if (err)
1562                 return err;
1563
1564         geneve_link_config(dev, &info, tb);
1565
1566         return 0;
1567 }
1568
1569 /* Quiesces the geneve device data path for both TX and RX.
1570  *
1571  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1572  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1573  * to complete for the existing set of in-flight packets to be transmitted,
1574  * then we would have quiesced the transmit data path. All the future packets
1575  * will get dropped until we unquiesce the data path.
1576  *
1577  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1578  * if we set that to NULL under RCU and wait for synchronize_net() to
1579  * complete, then we would have quiesced the receive data path.
1580  */
1581 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1582                            struct geneve_sock **gs6)
1583 {
1584         *gs4 = rtnl_dereference(geneve->sock4);
1585         rcu_assign_pointer(geneve->sock4, NULL);
1586         if (*gs4)
1587                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1588 #if IS_ENABLED(CONFIG_IPV6)
1589         *gs6 = rtnl_dereference(geneve->sock6);
1590         rcu_assign_pointer(geneve->sock6, NULL);
1591         if (*gs6)
1592                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1593 #else
1594         *gs6 = NULL;
1595 #endif
1596         synchronize_net();
1597 }
1598
1599 /* Resumes the geneve device data path for both TX and RX. */
1600 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1601                              struct geneve_sock __maybe_unused *gs6)
1602 {
1603         rcu_assign_pointer(geneve->sock4, gs4);
1604         if (gs4)
1605                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1606 #if IS_ENABLED(CONFIG_IPV6)
1607         rcu_assign_pointer(geneve->sock6, gs6);
1608         if (gs6)
1609                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1610 #endif
1611         synchronize_net();
1612 }
1613
1614 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1615                              struct nlattr *data[],
1616                              struct netlink_ext_ack *extack)
1617 {
1618         struct geneve_dev *geneve = netdev_priv(dev);
1619         struct geneve_sock *gs4, *gs6;
1620         struct ip_tunnel_info info;
1621         bool metadata;
1622         bool use_udp6_rx_checksums;
1623         enum ifla_geneve_df df;
1624         bool ttl_inherit;
1625         int err;
1626
1627         /* If the geneve device is configured for metadata (or externally
1628          * controlled, for example, OVS), then nothing can be changed.
1629          */
1630         if (geneve->collect_md)
1631                 return -EOPNOTSUPP;
1632
1633         /* Start with the existing info. */
1634         memcpy(&info, &geneve->info, sizeof(info));
1635         metadata = geneve->collect_md;
1636         use_udp6_rx_checksums = geneve->use_udp6_rx_checksums;
1637         ttl_inherit = geneve->ttl_inherit;
1638         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1639                              &use_udp6_rx_checksums, &ttl_inherit, &df, true);
1640         if (err)
1641                 return err;
1642
1643         if (!geneve_dst_addr_equal(&geneve->info, &info)) {
1644                 dst_cache_reset(&info.dst_cache);
1645                 geneve_link_config(dev, &info, tb);
1646         }
1647
1648         geneve_quiesce(geneve, &gs4, &gs6);
1649         geneve->info = info;
1650         geneve->collect_md = metadata;
1651         geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
1652         geneve->ttl_inherit = ttl_inherit;
1653         geneve_unquiesce(geneve, gs4, gs6);
1654
1655         return 0;
1656 }
1657
1658 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1659 {
1660         struct geneve_dev *geneve = netdev_priv(dev);
1661
1662         list_del(&geneve->next);
1663         unregister_netdevice_queue(dev, head);
1664 }
1665
1666 static size_t geneve_get_size(const struct net_device *dev)
1667 {
1668         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1669                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1670                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1671                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1672                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_DF */
1673                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1674                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1675                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1676                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1677                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1678                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1679                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1680                 0;
1681 }
1682
1683 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1684 {
1685         struct geneve_dev *geneve = netdev_priv(dev);
1686         struct ip_tunnel_info *info = &geneve->info;
1687         bool ttl_inherit = geneve->ttl_inherit;
1688         bool metadata = geneve->collect_md;
1689         __u8 tmp_vni[3];
1690         __u32 vni;
1691
1692         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1693         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1694         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1695                 goto nla_put_failure;
1696
1697         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1698                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1699                                     info->key.u.ipv4.dst))
1700                         goto nla_put_failure;
1701                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1702                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1703                         goto nla_put_failure;
1704
1705 #if IS_ENABLED(CONFIG_IPV6)
1706         } else if (!metadata) {
1707                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1708                                      &info->key.u.ipv6.dst))
1709                         goto nla_put_failure;
1710                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1711                                !(info->key.tun_flags & TUNNEL_CSUM)))
1712                         goto nla_put_failure;
1713 #endif
1714         }
1715
1716         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1717             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1718             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1719                 goto nla_put_failure;
1720
1721         if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->df))
1722                 goto nla_put_failure;
1723
1724         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1725                 goto nla_put_failure;
1726
1727         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1728                 goto nla_put_failure;
1729
1730 #if IS_ENABLED(CONFIG_IPV6)
1731         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1732                        !geneve->use_udp6_rx_checksums))
1733                 goto nla_put_failure;
1734 #endif
1735
1736         if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1737                 goto nla_put_failure;
1738
1739         return 0;
1740
1741 nla_put_failure:
1742         return -EMSGSIZE;
1743 }
1744
1745 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1746         .kind           = "geneve",
1747         .maxtype        = IFLA_GENEVE_MAX,
1748         .policy         = geneve_policy,
1749         .priv_size      = sizeof(struct geneve_dev),
1750         .setup          = geneve_setup,
1751         .validate       = geneve_validate,
1752         .newlink        = geneve_newlink,
1753         .changelink     = geneve_changelink,
1754         .dellink        = geneve_dellink,
1755         .get_size       = geneve_get_size,
1756         .fill_info      = geneve_fill_info,
1757 };
1758
1759 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1760                                         u8 name_assign_type, u16 dst_port)
1761 {
1762         struct nlattr *tb[IFLA_MAX + 1];
1763         struct ip_tunnel_info info;
1764         struct net_device *dev;
1765         LIST_HEAD(list_kill);
1766         int err;
1767
1768         memset(tb, 0, sizeof(tb));
1769         dev = rtnl_create_link(net, name, name_assign_type,
1770                                &geneve_link_ops, tb, NULL);
1771         if (IS_ERR(dev))
1772                 return dev;
1773
1774         init_tnl_info(&info, dst_port);
1775         err = geneve_configure(net, dev, NULL, &info,
1776                                true, true, false, GENEVE_DF_UNSET);
1777         if (err) {
1778                 free_netdev(dev);
1779                 return ERR_PTR(err);
1780         }
1781
1782         /* openvswitch users expect packet sizes to be unrestricted,
1783          * so set the largest MTU we can.
1784          */
1785         err = geneve_change_mtu(dev, IP_MAX_MTU);
1786         if (err)
1787                 goto err;
1788
1789         err = rtnl_configure_link(dev, NULL);
1790         if (err < 0)
1791                 goto err;
1792
1793         return dev;
1794 err:
1795         geneve_dellink(dev, &list_kill);
1796         unregister_netdevice_many(&list_kill);
1797         return ERR_PTR(err);
1798 }
1799 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1800
1801 static int geneve_netdevice_event(struct notifier_block *unused,
1802                                   unsigned long event, void *ptr)
1803 {
1804         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1805
1806         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1807             event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1808                 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1809         } else if (event == NETDEV_UNREGISTER) {
1810                 geneve_offload_rx_ports(dev, false);
1811         } else if (event == NETDEV_REGISTER) {
1812                 geneve_offload_rx_ports(dev, true);
1813         }
1814
1815         return NOTIFY_DONE;
1816 }
1817
1818 static struct notifier_block geneve_notifier_block __read_mostly = {
1819         .notifier_call = geneve_netdevice_event,
1820 };
1821
1822 static __net_init int geneve_init_net(struct net *net)
1823 {
1824         struct geneve_net *gn = net_generic(net, geneve_net_id);
1825
1826         INIT_LIST_HEAD(&gn->geneve_list);
1827         INIT_LIST_HEAD(&gn->sock_list);
1828         return 0;
1829 }
1830
1831 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1832 {
1833         struct geneve_net *gn = net_generic(net, geneve_net_id);
1834         struct geneve_dev *geneve, *next;
1835         struct net_device *dev, *aux;
1836
1837         /* gather any geneve devices that were moved into this ns */
1838         for_each_netdev_safe(net, dev, aux)
1839                 if (dev->rtnl_link_ops == &geneve_link_ops)
1840                         unregister_netdevice_queue(dev, head);
1841
1842         /* now gather any other geneve devices that were created in this ns */
1843         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1844                 /* If geneve->dev is in the same netns, it was already added
1845                  * to the list by the previous loop.
1846                  */
1847                 if (!net_eq(dev_net(geneve->dev), net))
1848                         unregister_netdevice_queue(geneve->dev, head);
1849         }
1850
1851         WARN_ON_ONCE(!list_empty(&gn->sock_list));
1852 }
1853
1854 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1855 {
1856         struct net *net;
1857         LIST_HEAD(list);
1858
1859         rtnl_lock();
1860         list_for_each_entry(net, net_list, exit_list)
1861                 geneve_destroy_tunnels(net, &list);
1862
1863         /* unregister the devices gathered above */
1864         unregister_netdevice_many(&list);
1865         rtnl_unlock();
1866 }
1867
1868 static struct pernet_operations geneve_net_ops = {
1869         .init = geneve_init_net,
1870         .exit_batch = geneve_exit_batch_net,
1871         .id   = &geneve_net_id,
1872         .size = sizeof(struct geneve_net),
1873 };
1874
1875 static int __init geneve_init_module(void)
1876 {
1877         int rc;
1878
1879         rc = register_pernet_subsys(&geneve_net_ops);
1880         if (rc)
1881                 goto out1;
1882
1883         rc = register_netdevice_notifier(&geneve_notifier_block);
1884         if (rc)
1885                 goto out2;
1886
1887         rc = rtnl_link_register(&geneve_link_ops);
1888         if (rc)
1889                 goto out3;
1890
1891         return 0;
1892 out3:
1893         unregister_netdevice_notifier(&geneve_notifier_block);
1894 out2:
1895         unregister_pernet_subsys(&geneve_net_ops);
1896 out1:
1897         return rc;
1898 }
1899 late_initcall(geneve_init_module);
1900
1901 static void __exit geneve_cleanup_module(void)
1902 {
1903         rtnl_link_unregister(&geneve_link_ops);
1904         unregister_netdevice_notifier(&geneve_notifier_block);
1905         unregister_pernet_subsys(&geneve_net_ops);
1906 }
1907 module_exit(geneve_cleanup_module);
1908
1909 MODULE_LICENSE("GPL");
1910 MODULE_VERSION(GENEVE_NETDEV_VER);
1911 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1912 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1913 MODULE_ALIAS_RTNL_LINK("geneve");