drivers/net/geneve.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * GENEVE: Generic Network Virtualization Encapsulation
   4  *
   5  * Copyright (c) 2015 Red Hat, Inc.
   6  */
   7
   8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   9
  10 #include <linux/ethtool.h>
  11 #include <linux/kernel.h>
  12 #include <linux/module.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/hash.h>
  15 #include <net/ipv6_stubs.h>
  16 #include <net/dst_metadata.h>
  17 #include <net/gro_cells.h>
  18 #include <net/rtnetlink.h>
  19 #include <net/geneve.h>
  20 #include <net/gro.h>
  21 #include <net/protocol.h>
  22
  23 #define GENEVE_NETDEV_VER       "0.6"
  24
  25 #define GENEVE_N_VID            (1u << 24)
  26 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  27
  28 #define VNI_HASH_BITS           10
  29 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  30
  31 static bool log_ecn_error = true;
  32 module_param(log_ecn_error, bool, 0644);
  33 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  34
  35 #define GENEVE_VER 0
  36 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  37 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
  38 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
  39
  40 /* per-network namespace private data for this module */
  41 struct geneve_net {
  42         struct list_head        geneve_list;
  43         struct list_head        sock_list;
  44 };
  45
  46 static unsigned int geneve_net_id;
  47
  48 struct geneve_dev_node {
  49         struct hlist_node hlist;
  50         struct geneve_dev *geneve;
  51 };
  52
  53 struct geneve_config {
  54         struct ip_tunnel_info   info;
  55         bool                    collect_md;
  56         bool                    use_udp6_rx_checksums;
  57         bool                    ttl_inherit;
  58         enum ifla_geneve_df     df;
  59         bool                    inner_proto_inherit;
  60 };
  61
  62 /* Pseudo network device */
  63 struct geneve_dev {
  64         struct geneve_dev_node hlist4;  /* vni hash table for IPv4 socket */
  65 #if IS_ENABLED(CONFIG_IPV6)
  66         struct geneve_dev_node hlist6;  /* vni hash table for IPv6 socket */
  67 #endif
  68         struct net         *net;        /* netns for packet i/o */
  69         struct net_device  *dev;        /* netdev for geneve tunnel */
  70         struct geneve_sock __rcu *sock4;        /* IPv4 socket used for geneve tunnel */
  71 #if IS_ENABLED(CONFIG_IPV6)
  72         struct geneve_sock __rcu *sock6;        /* IPv6 socket used for geneve tunnel */
  73 #endif
  74         struct list_head   next;        /* geneve's per namespace list */
  75         struct gro_cells   gro_cells;
  76         struct geneve_config cfg;
  77 };
  78
  79 struct geneve_sock {
  80         bool                    collect_md;
  81         struct list_head        list;
  82         struct socket           *sock;
  83         struct rcu_head         rcu;
  84         int                     refcnt;
  85         struct hlist_head       vni_list[VNI_HASH_SIZE];
  86 };
  87
  88 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  89 {
  90         __u32 vnid;
  91
  92         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  93         return hash_32(vnid, VNI_HASH_BITS);
  94 }
  95
  96 static __be64 vni_to_tunnel_id(const __u8 *vni)
  97 {
  98 #ifdef __BIG_ENDIAN
  99         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
 100 #else
 101         return (__force __be64)(((__force u64)vni[0] << 40) |
 102                                 ((__force u64)vni[1] << 48) |
 103                                 ((__force u64)vni[2] << 56));
 104 #endif
 105 }
 106
 107 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 108 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 109 {
 110 #ifdef __BIG_ENDIAN
 111         vni[0] = (__force __u8)(tun_id >> 16);
 112         vni[1] = (__force __u8)(tun_id >> 8);
 113         vni[2] = (__force __u8)tun_id;
 114 #else
 115         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 116         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 117         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 118 #endif
 119 }
 120
 121 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
 122 {
 123         return !memcmp(vni, &tun_id[5], 3);
 124 }
 125
 126 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 127 {
 128         return gs->sock->sk->sk_family;
 129 }
 130
 131 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 132                                         __be32 addr, u8 vni[])
 133 {
 134         struct hlist_head *vni_list_head;
 135         struct geneve_dev_node *node;
 136         __u32 hash;
 137
 138         /* Find the device for this VNI */
 139         hash = geneve_net_vni_hash(vni);
 140         vni_list_head = &gs->vni_list[hash];
 141         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 142                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 143                     addr == node->geneve->cfg.info.key.u.ipv4.dst)
 144                         return node->geneve;
 145         }
 146         return NULL;
 147 }
 148
 149 #if IS_ENABLED(CONFIG_IPV6)
 150 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 151                                          struct in6_addr addr6, u8 vni[])
 152 {
 153         struct hlist_head *vni_list_head;
 154         struct geneve_dev_node *node;
 155         __u32 hash;
 156
 157         /* Find the device for this VNI */
 158         hash = geneve_net_vni_hash(vni);
 159         vni_list_head = &gs->vni_list[hash];
 160         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 161                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 162                     ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
 163                         return node->geneve;
 164         }
 165         return NULL;
 166 }
 167 #endif
 168
 169 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 170 {
 171         return (struct genevehdr *)(udp_hdr(skb) + 1);
 172 }
 173
 174 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 175                                             struct sk_buff *skb)
 176 {
 177         static u8 zero_vni[3];
 178         u8 *vni;
 179
 180         if (geneve_get_sk_family(gs) == AF_INET) {
 181                 struct iphdr *iph;
 182                 __be32 addr;
 183
 184                 iph = ip_hdr(skb); /* outer IP header... */
 185
 186                 if (gs->collect_md) {
 187                         vni = zero_vni;
 188                         addr = 0;
 189                 } else {
 190                         vni = geneve_hdr(skb)->vni;
 191                         addr = iph->saddr;
 192                 }
 193
 194                 return geneve_lookup(gs, addr, vni);
 195 #if IS_ENABLED(CONFIG_IPV6)
 196         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 197                 static struct in6_addr zero_addr6;
 198                 struct ipv6hdr *ip6h;
 199                 struct in6_addr addr6;
 200
 201                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 202
 203                 if (gs->collect_md) {
 204                         vni = zero_vni;
 205                         addr6 = zero_addr6;
 206                 } else {
 207                         vni = geneve_hdr(skb)->vni;
 208                         addr6 = ip6h->saddr;
 209                 }
 210
 211                 return geneve6_lookup(gs, addr6, vni);
 212 #endif
 213         }
 214         return NULL;
 215 }
 216
 217 /* geneve receive/decap routine */
 218 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 219                       struct sk_buff *skb)
 220 {
 221         struct genevehdr *gnvh = geneve_hdr(skb);
 222         struct metadata_dst *tun_dst = NULL;
 223         unsigned int len;
 224         int nh, err = 0;
 225         void *oiph;
 226
 227         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 228                 __be16 flags;
 229
 230                 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
 231                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 232
 233                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 234                                          vni_to_tunnel_id(gnvh->vni),
 235                                          gnvh->opt_len * 4);
 236                 if (!tun_dst) {
 237                         DEV_STATS_INC(geneve->dev, rx_dropped);
 238                         goto drop;
 239                 }
 240                 /* Update tunnel dst according to Geneve options. */
 241                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 242                                         gnvh->options, gnvh->opt_len * 4,
 243                                         TUNNEL_GENEVE_OPT);
 244         } else {
 245                 /* Drop packets w/ critical options,
 246                  * since we don't support any...
 247                  */
 248                 if (gnvh->critical) {
 249                         DEV_STATS_INC(geneve->dev, rx_frame_errors);
 250                         DEV_STATS_INC(geneve->dev, rx_errors);
 251                         goto drop;
 252                 }
 253         }
 254
 255         if (tun_dst)
 256                 skb_dst_set(skb, &tun_dst->dst);
 257
 258         if (gnvh->proto_type == htons(ETH_P_TEB)) {
 259                 skb_reset_mac_header(skb);
 260                 skb->protocol = eth_type_trans(skb, geneve->dev);
 261                 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 262
 263                 /* Ignore packet loops (and multicast echo) */
 264                 if (ether_addr_equal(eth_hdr(skb)->h_source,
 265                                      geneve->dev->dev_addr)) {
 266                         DEV_STATS_INC(geneve->dev, rx_errors);
 267                         goto drop;
 268                 }
 269         } else {
 270                 skb_reset_mac_header(skb);
 271                 skb->dev = geneve->dev;
 272                 skb->pkt_type = PACKET_HOST;
 273         }
 274
 275         /* Save offset of outer header relative to skb->head,
 276          * because we are going to reset the network header to the inner header
 277          * and might change skb->head.
 278          */
 279         nh = skb_network_header(skb) - skb->head;
 280
 281         skb_reset_network_header(skb);
 282
 283         if (!pskb_inet_may_pull(skb)) {
 284                 DEV_STATS_INC(geneve->dev, rx_length_errors);
 285                 DEV_STATS_INC(geneve->dev, rx_errors);
 286                 goto drop;
 287         }
 288
 289         /* Get the outer header. */
 290         oiph = skb->head + nh;
 291
 292         if (geneve_get_sk_family(gs) == AF_INET)
 293                 err = IP_ECN_decapsulate(oiph, skb);
 294 #if IS_ENABLED(CONFIG_IPV6)
 295         else
 296                 err = IP6_ECN_decapsulate(oiph, skb);
 297 #endif
 298
 299         if (unlikely(err)) {
 300                 if (log_ecn_error) {
 301                         if (geneve_get_sk_family(gs) == AF_INET)
 302                                 net_info_ratelimited("non-ECT from %pI4 "
 303                                                      "with TOS=%#x\n",
 304                                                      &((struct iphdr *)oiph)->saddr,
 305                                                      ((struct iphdr *)oiph)->tos);
 306 #if IS_ENABLED(CONFIG_IPV6)
 307                         else
 308                                 net_info_ratelimited("non-ECT from %pI6\n",
 309                                                      &((struct ipv6hdr *)oiph)->saddr);
 310 #endif
 311                 }
 312                 if (err > 1) {
 313                         DEV_STATS_INC(geneve->dev, rx_frame_errors);
 314                         DEV_STATS_INC(geneve->dev, rx_errors);
 315                         goto drop;
 316                 }
 317         }
 318
 319         len = skb->len;
 320         err = gro_cells_receive(&geneve->gro_cells, skb);
 321         if (likely(err == NET_RX_SUCCESS))
 322                 dev_sw_netstats_rx_add(geneve->dev, len);
 323
 324         return;
 325 drop:
 326         /* Consume bad packet */
 327         kfree_skb(skb);
 328 }
 329
 330 /* Setup stats when device is created */
 331 static int geneve_init(struct net_device *dev)
 332 {
 333         struct geneve_dev *geneve = netdev_priv(dev);
 334         int err;
 335
 336         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 337         if (!dev->tstats)
 338                 return -ENOMEM;
 339
 340         err = gro_cells_init(&geneve->gro_cells, dev);
 341         if (err) {
 342                 free_percpu(dev->tstats);
 343                 return err;
 344         }
 345
 346         err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
 347         if (err) {
 348                 free_percpu(dev->tstats);
 349                 gro_cells_destroy(&geneve->gro_cells);
 350                 return err;
 351         }
 352         netdev_lockdep_set_classes(dev);
 353         return 0;
 354 }
 355
 356 static void geneve_uninit(struct net_device *dev)
 357 {
 358         struct geneve_dev *geneve = netdev_priv(dev);
 359
 360         dst_cache_destroy(&geneve->cfg.info.dst_cache);
 361         gro_cells_destroy(&geneve->gro_cells);
 362         free_percpu(dev->tstats);
 363 }
 364
 365 /* Callback from net/ipv4/udp.c to receive packets */
 366 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 367 {
 368         struct genevehdr *geneveh;
 369         struct geneve_dev *geneve;
 370         struct geneve_sock *gs;
 371         __be16 inner_proto;
 372         int opts_len;
 373
 374         /* Need UDP and Geneve header to be present */
 375         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 376                 goto drop;
 377
 378         /* Return packets with reserved bits set */
 379         geneveh = geneve_hdr(skb);
 380         if (unlikely(geneveh->ver != GENEVE_VER))
 381                 goto drop;
 382
 383         gs = rcu_dereference_sk_user_data(sk);
 384         if (!gs)
 385                 goto drop;
 386
 387         geneve = geneve_lookup_skb(gs, skb);
 388         if (!geneve)
 389                 goto drop;
 390
 391         inner_proto = geneveh->proto_type;
 392
 393         if (unlikely((!geneve->cfg.inner_proto_inherit &&
 394                       inner_proto != htons(ETH_P_TEB)))) {
 395                 DEV_STATS_INC(geneve->dev, rx_dropped);
 396                 goto drop;
 397         }
 398
 399         opts_len = geneveh->opt_len * 4;
 400         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto,
 401                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 402                 DEV_STATS_INC(geneve->dev, rx_dropped);
 403                 goto drop;
 404         }
 405
 406         geneve_rx(geneve, gs, skb);
 407         return 0;
 408
 409 drop:
 410         /* Consume bad packet */
 411         kfree_skb(skb);
 412         return 0;
 413 }
 414
 415 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
 416 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
 417 {
 418         struct genevehdr *geneveh;
 419         struct geneve_sock *gs;
 420         u8 zero_vni[3] = { 0 };
 421         u8 *vni = zero_vni;
 422
 423         if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
 424                 return -EINVAL;
 425
 426         geneveh = geneve_hdr(skb);
 427         if (geneveh->ver != GENEVE_VER)
 428                 return -EINVAL;
 429
 430         if (geneveh->proto_type != htons(ETH_P_TEB))
 431                 return -EINVAL;
 432
 433         gs = rcu_dereference_sk_user_data(sk);
 434         if (!gs)
 435                 return -ENOENT;
 436
 437         if (geneve_get_sk_family(gs) == AF_INET) {
 438                 struct iphdr *iph = ip_hdr(skb);
 439                 __be32 addr4 = 0;
 440
 441                 if (!gs->collect_md) {
 442                         vni = geneve_hdr(skb)->vni;
 443                         addr4 = iph->daddr;
 444                 }
 445
 446                 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
 447         }
 448
 449 #if IS_ENABLED(CONFIG_IPV6)
 450         if (geneve_get_sk_family(gs) == AF_INET6) {
 451                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 452                 struct in6_addr addr6;
 453
 454                 memset(&addr6, 0, sizeof(struct in6_addr));
 455
 456                 if (!gs->collect_md) {
 457                         vni = geneve_hdr(skb)->vni;
 458                         addr6 = ip6h->daddr;
 459                 }
 460
 461                 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
 462         }
 463 #endif
 464
 465         return -EPFNOSUPPORT;
 466 }
 467
 468 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 469                                          __be16 port, bool ipv6_rx_csum)
 470 {
 471         struct socket *sock;
 472         struct udp_port_cfg udp_conf;
 473         int err;
 474
 475         memset(&udp_conf, 0, sizeof(udp_conf));
 476
 477         if (ipv6) {
 478                 udp_conf.family = AF_INET6;
 479                 udp_conf.ipv6_v6only = 1;
 480                 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
 481         } else {
 482                 udp_conf.family = AF_INET;
 483                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 484         }
 485
 486         udp_conf.local_udp_port = port;
 487
 488         /* Open UDP socket */
 489         err = udp_sock_create(net, &udp_conf, &sock);
 490         if (err < 0)
 491                 return ERR_PTR(err);
 492
 493         udp_allow_gso(sock->sk);
 494         return sock;
 495 }
 496
 497 static int geneve_hlen(struct genevehdr *gh)
 498 {
 499         return sizeof(*gh) + gh->opt_len * 4;
 500 }
 501
 502 static struct sk_buff *geneve_gro_receive(struct sock *sk,
 503                                           struct list_head *head,
 504                                           struct sk_buff *skb)
 505 {
 506         struct sk_buff *pp = NULL;
 507         struct sk_buff *p;
 508         struct genevehdr *gh, *gh2;
 509         unsigned int hlen, gh_len, off_gnv;
 510         const struct packet_offload *ptype;
 511         __be16 type;
 512         int flush = 1;
 513
 514         off_gnv = skb_gro_offset(skb);
 515         hlen = off_gnv + sizeof(*gh);
 516         gh = skb_gro_header(skb, hlen, off_gnv);
 517         if (unlikely(!gh))
 518                 goto out;
 519
 520         if (gh->ver != GENEVE_VER || gh->oam)
 521                 goto out;
 522         gh_len = geneve_hlen(gh);
 523
 524         hlen = off_gnv + gh_len;
 525         if (skb_gro_header_hard(skb, hlen)) {
 526                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 527                 if (unlikely(!gh))
 528                         goto out;
 529         }
 530
 531         list_for_each_entry(p, head, list) {
 532                 if (!NAPI_GRO_CB(p)->same_flow)
 533                         continue;
 534
 535                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 536                 if (gh->opt_len != gh2->opt_len ||
 537                     memcmp(gh, gh2, gh_len)) {
 538                         NAPI_GRO_CB(p)->same_flow = 0;
 539                         continue;
 540                 }
 541         }
 542
 543         skb_gro_pull(skb, gh_len);
 544         skb_gro_postpull_rcsum(skb, gh, gh_len);
 545         type = gh->proto_type;
 546         if (likely(type == htons(ETH_P_TEB)))
 547                 return call_gro_receive(eth_gro_receive, head, skb);
 548
 549         ptype = gro_find_receive_by_type(type);
 550         if (!ptype)
 551                 goto out;
 552
 553         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 554         flush = 0;
 555
 556 out:
 557         skb_gro_flush_final(skb, pp, flush);
 558
 559         return pp;
 560 }
 561
 562 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 563                                int nhoff)
 564 {
 565         struct genevehdr *gh;
 566         struct packet_offload *ptype;
 567         __be16 type;
 568         int gh_len;
 569         int err = -ENOSYS;
 570
 571         gh = (struct genevehdr *)(skb->data + nhoff);
 572         gh_len = geneve_hlen(gh);
 573         type = gh->proto_type;
 574
 575         /* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */
 576         if (likely(type == htons(ETH_P_TEB)))
 577                 return eth_gro_complete(skb, nhoff + gh_len);
 578
 579         ptype = gro_find_complete_by_type(type);
 580         if (ptype)
 581                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 582
 583         skb_set_inner_mac_header(skb, nhoff + gh_len);
 584
 585         return err;
 586 }
 587
 588 /* Create new listen socket if needed */
 589 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 590                                                 bool ipv6, bool ipv6_rx_csum)
 591 {
 592         struct geneve_net *gn = net_generic(net, geneve_net_id);
 593         struct geneve_sock *gs;
 594         struct socket *sock;
 595         struct udp_tunnel_sock_cfg tunnel_cfg;
 596         int h;
 597
 598         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 599         if (!gs)
 600                 return ERR_PTR(-ENOMEM);
 601
 602         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 603         if (IS_ERR(sock)) {
 604                 kfree(gs);
 605                 return ERR_CAST(sock);
 606         }
 607
 608         gs->sock = sock;
 609         gs->refcnt = 1;
 610         for (h = 0; h < VNI_HASH_SIZE; ++h)
 611                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 612
 613         /* Initialize the geneve udp offloads structure */
 614         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 615
 616         /* Mark socket as an encapsulation socket */
 617         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 618         tunnel_cfg.sk_user_data = gs;
 619         tunnel_cfg.encap_type = 1;
 620         tunnel_cfg.gro_receive = geneve_gro_receive;
 621         tunnel_cfg.gro_complete = geneve_gro_complete;
 622         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 623         tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
 624         tunnel_cfg.encap_destroy = NULL;
 625         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 626         list_add(&gs->list, &gn->sock_list);
 627         return gs;
 628 }
 629
 630 static void __geneve_sock_release(struct geneve_sock *gs)
 631 {
 632         if (!gs || --gs->refcnt)
 633                 return;
 634
 635         list_del(&gs->list);
 636         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 637         udp_tunnel_sock_release(gs->sock);
 638         kfree_rcu(gs, rcu);
 639 }
 640
 641 static void geneve_sock_release(struct geneve_dev *geneve)
 642 {
 643         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 644 #if IS_ENABLED(CONFIG_IPV6)
 645         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 646
 647         rcu_assign_pointer(geneve->sock6, NULL);
 648 #endif
 649
 650         rcu_assign_pointer(geneve->sock4, NULL);
 651         synchronize_net();
 652
 653         __geneve_sock_release(gs4);
 654 #if IS_ENABLED(CONFIG_IPV6)
 655         __geneve_sock_release(gs6);
 656 #endif
 657 }
 658
 659 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 660                                             sa_family_t family,
 661                                             __be16 dst_port)
 662 {
 663         struct geneve_sock *gs;
 664
 665         list_for_each_entry(gs, &gn->sock_list, list) {
 666                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 667                     geneve_get_sk_family(gs) == family) {
 668                         return gs;
 669                 }
 670         }
 671         return NULL;
 672 }
 673
 674 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 675 {
 676         struct net *net = geneve->net;
 677         struct geneve_net *gn = net_generic(net, geneve_net_id);
 678         struct geneve_dev_node *node;
 679         struct geneve_sock *gs;
 680         __u8 vni[3];
 681         __u32 hash;
 682
 683         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
 684         if (gs) {
 685                 gs->refcnt++;
 686                 goto out;
 687         }
 688
 689         gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
 690                                   geneve->cfg.use_udp6_rx_checksums);
 691         if (IS_ERR(gs))
 692                 return PTR_ERR(gs);
 693
 694 out:
 695         gs->collect_md = geneve->cfg.collect_md;
 696 #if IS_ENABLED(CONFIG_IPV6)
 697         if (ipv6) {
 698                 rcu_assign_pointer(geneve->sock6, gs);
 699                 node = &geneve->hlist6;
 700         } else
 701 #endif
 702         {
 703                 rcu_assign_pointer(geneve->sock4, gs);
 704                 node = &geneve->hlist4;
 705         }
 706         node->geneve = geneve;
 707
 708         tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
 709         hash = geneve_net_vni_hash(vni);
 710         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 711         return 0;
 712 }
 713
 714 static int geneve_open(struct net_device *dev)
 715 {
 716         struct geneve_dev *geneve = netdev_priv(dev);
 717         bool metadata = geneve->cfg.collect_md;
 718         bool ipv4, ipv6;
 719         int ret = 0;
 720
 721         ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 722         ipv4 = !ipv6 || metadata;
 723 #if IS_ENABLED(CONFIG_IPV6)
 724         if (ipv6) {
 725                 ret = geneve_sock_add(geneve, true);
 726                 if (ret < 0 && ret != -EAFNOSUPPORT)
 727                         ipv4 = false;
 728         }
 729 #endif
 730         if (ipv4)
 731                 ret = geneve_sock_add(geneve, false);
 732         if (ret < 0)
 733                 geneve_sock_release(geneve);
 734
 735         return ret;
 736 }
 737
 738 static int geneve_stop(struct net_device *dev)
 739 {
 740         struct geneve_dev *geneve = netdev_priv(dev);
 741
 742         hlist_del_init_rcu(&geneve->hlist4.hlist);
 743 #if IS_ENABLED(CONFIG_IPV6)
 744         hlist_del_init_rcu(&geneve->hlist6.hlist);
 745 #endif
 746         geneve_sock_release(geneve);
 747         return 0;
 748 }
 749
 750 static void geneve_build_header(struct genevehdr *geneveh,
 751                                 const struct ip_tunnel_info *info,
 752                                 __be16 inner_proto)
 753 {
 754         geneveh->ver = GENEVE_VER;
 755         geneveh->opt_len = info->options_len / 4;
 756         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 757         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 758         geneveh->rsvd1 = 0;
 759         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 760         geneveh->proto_type = inner_proto;
 761         geneveh->rsvd2 = 0;
 762
 763         if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
 764                 ip_tunnel_info_opts_get(geneveh->options, info);
 765 }
 766
 767 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 768                             const struct ip_tunnel_info *info,
 769                             bool xnet, int ip_hdr_len,
 770                             bool inner_proto_inherit)
 771 {
 772         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 773         struct genevehdr *gnvh;
 774         __be16 inner_proto;
 775         int min_headroom;
 776         int err;
 777
 778         skb_reset_mac_header(skb);
 779         skb_scrub_packet(skb, xnet);
 780
 781         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 782                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 783         err = skb_cow_head(skb, min_headroom);
 784         if (unlikely(err))
 785                 goto free_dst;
 786
 787         err = udp_tunnel_handle_offloads(skb, udp_sum);
 788         if (err)
 789                 goto free_dst;
 790
 791         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 792         inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB);
 793         geneve_build_header(gnvh, info, inner_proto);
 794         skb_set_inner_protocol(skb, inner_proto);
 795         return 0;
 796
 797 free_dst:
 798         dst_release(dst);
 799         return err;
 800 }
 801
 802 static u8 geneve_get_dsfield(struct sk_buff *skb, struct net_device *dev,
 803                              const struct ip_tunnel_info *info,
 804                              bool *use_cache)
 805 {
 806         struct geneve_dev *geneve = netdev_priv(dev);
 807         u8 dsfield;
 808
 809         dsfield = info->key.tos;
 810         if (dsfield == 1 && !geneve->cfg.collect_md) {
 811                 dsfield = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 812                 *use_cache = false;
 813         }
 814
 815         return dsfield;
 816 }
 817
 818 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 819                            struct geneve_dev *geneve,
 820                            const struct ip_tunnel_info *info)
 821 {
 822         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 823         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 824         const struct ip_tunnel_key *key = &info->key;
 825         struct rtable *rt;
 826         bool use_cache;
 827         __u8 tos, ttl;
 828         __be16 df = 0;
 829         __be32 saddr;
 830         __be16 sport;
 831         int err;
 832
 833         if (!skb_vlan_inet_prepare(skb))
 834                 return -EINVAL;
 835
 836         if (!gs4)
 837                 return -EIO;
 838
 839         use_cache = ip_tunnel_dst_cache_usable(skb, info);
 840         tos = geneve_get_dsfield(skb, dev, info, &use_cache);
 841         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 842
 843         rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
 844                                    &info->key,
 845                                    sport, geneve->cfg.info.key.tp_dst, tos,
 846                                    use_cache ?
 847                                    (struct dst_cache *)&info->dst_cache : NULL);
 848         if (IS_ERR(rt))
 849                 return PTR_ERR(rt);
 850
 851         err = skb_tunnel_check_pmtu(skb, &rt->dst,
 852                                     GENEVE_IPV4_HLEN + info->options_len,
 853                                     netif_is_any_bridge_port(dev));
 854         if (err < 0) {
 855                 dst_release(&rt->dst);
 856                 return err;
 857         } else if (err) {
 858                 struct ip_tunnel_info *info;
 859
 860                 info = skb_tunnel_info(skb);
 861                 if (info) {
 862                         struct ip_tunnel_info *unclone;
 863
 864                         unclone = skb_tunnel_info_unclone(skb);
 865                         if (unlikely(!unclone)) {
 866                                 dst_release(&rt->dst);
 867                                 return -ENOMEM;
 868                         }
 869
 870                         unclone->key.u.ipv4.dst = saddr;
 871                         unclone->key.u.ipv4.src = info->key.u.ipv4.dst;
 872                 }
 873
 874                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 875                         dst_release(&rt->dst);
 876                         return -EINVAL;
 877                 }
 878
 879                 skb->protocol = eth_type_trans(skb, geneve->dev);
 880                 __netif_rx(skb);
 881                 dst_release(&rt->dst);
 882                 return -EMSGSIZE;
 883         }
 884
 885         tos = ip_tunnel_ecn_encap(tos, ip_hdr(skb), skb);
 886         if (geneve->cfg.collect_md) {
 887                 ttl = key->ttl;
 888
 889                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 890         } else {
 891                 if (geneve->cfg.ttl_inherit)
 892                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 893                 else
 894                         ttl = key->ttl;
 895                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 896
 897                 if (geneve->cfg.df == GENEVE_DF_SET) {
 898                         df = htons(IP_DF);
 899                 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
 900                         struct ethhdr *eth = eth_hdr(skb);
 901
 902                         if (ntohs(eth->h_proto) == ETH_P_IPV6) {
 903                                 df = htons(IP_DF);
 904                         } else if (ntohs(eth->h_proto) == ETH_P_IP) {
 905                                 struct iphdr *iph = ip_hdr(skb);
 906
 907                                 if (iph->frag_off & htons(IP_DF))
 908                                         df = htons(IP_DF);
 909                         }
 910                 }
 911         }
 912
 913         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
 914                                geneve->cfg.inner_proto_inherit);
 915         if (unlikely(err))
 916                 return err;
 917
 918         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, saddr, info->key.u.ipv4.dst,
 919                             tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
 920                             !net_eq(geneve->net, dev_net(geneve->dev)),
 921                             !(info->key.tun_flags & TUNNEL_CSUM));
 922         return 0;
 923 }
 924
 925 #if IS_ENABLED(CONFIG_IPV6)
 926 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 927                             struct geneve_dev *geneve,
 928                             const struct ip_tunnel_info *info)
 929 {
 930         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 931         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 932         const struct ip_tunnel_key *key = &info->key;
 933         struct dst_entry *dst = NULL;
 934         struct in6_addr saddr;
 935         bool use_cache;
 936         __u8 prio, ttl;
 937         __be16 sport;
 938         int err;
 939
 940         if (!skb_vlan_inet_prepare(skb))
 941                 return -EINVAL;
 942
 943         if (!gs6)
 944                 return -EIO;
 945
 946         use_cache = ip_tunnel_dst_cache_usable(skb, info);
 947         prio = geneve_get_dsfield(skb, dev, info, &use_cache);
 948         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 949
 950         dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
 951                                      &saddr, key, sport,
 952                                      geneve->cfg.info.key.tp_dst, prio,
 953                                      use_cache ?
 954                                      (struct dst_cache *)&info->dst_cache : NULL);
 955         if (IS_ERR(dst))
 956                 return PTR_ERR(dst);
 957
 958         err = skb_tunnel_check_pmtu(skb, dst,
 959                                     GENEVE_IPV6_HLEN + info->options_len,
 960                                     netif_is_any_bridge_port(dev));
 961         if (err < 0) {
 962                 dst_release(dst);
 963                 return err;
 964         } else if (err) {
 965                 struct ip_tunnel_info *info = skb_tunnel_info(skb);
 966
 967                 if (info) {
 968                         struct ip_tunnel_info *unclone;
 969
 970                         unclone = skb_tunnel_info_unclone(skb);
 971                         if (unlikely(!unclone)) {
 972                                 dst_release(dst);
 973                                 return -ENOMEM;
 974                         }
 975
 976                         unclone->key.u.ipv6.dst = saddr;
 977                         unclone->key.u.ipv6.src = info->key.u.ipv6.dst;
 978                 }
 979
 980                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 981                         dst_release(dst);
 982                         return -EINVAL;
 983                 }
 984
 985                 skb->protocol = eth_type_trans(skb, geneve->dev);
 986                 __netif_rx(skb);
 987                 dst_release(dst);
 988                 return -EMSGSIZE;
 989         }
 990
 991         prio = ip_tunnel_ecn_encap(prio, ip_hdr(skb), skb);
 992         if (geneve->cfg.collect_md) {
 993                 ttl = key->ttl;
 994         } else {
 995                 if (geneve->cfg.ttl_inherit)
 996                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 997                 else
 998                         ttl = key->ttl;
 999                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1000         }
1001         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
1002                                geneve->cfg.inner_proto_inherit);
1003         if (unlikely(err))
1004                 return err;
1005
1006         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1007                              &saddr, &key->u.ipv6.dst, prio, ttl,
1008                              info->key.label, sport, geneve->cfg.info.key.tp_dst,
1009                              !(info->key.tun_flags & TUNNEL_CSUM));
1010         return 0;
1011 }
1012 #endif
1013
1014 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1015 {
1016         struct geneve_dev *geneve = netdev_priv(dev);
1017         struct ip_tunnel_info *info = NULL;
1018         int err;
1019
1020         if (geneve->cfg.collect_md) {
1021                 info = skb_tunnel_info(skb);
1022                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1023                         netdev_dbg(dev, "no tunnel metadata\n");
1024                         dev_kfree_skb(skb);
1025                         DEV_STATS_INC(dev, tx_dropped);
1026                         return NETDEV_TX_OK;
1027                 }
1028         } else {
1029                 info = &geneve->cfg.info;
1030         }
1031
1032         rcu_read_lock();
1033 #if IS_ENABLED(CONFIG_IPV6)
1034         if (info->mode & IP_TUNNEL_INFO_IPV6)
1035                 err = geneve6_xmit_skb(skb, dev, geneve, info);
1036         else
1037 #endif
1038                 err = geneve_xmit_skb(skb, dev, geneve, info);
1039         rcu_read_unlock();
1040
1041         if (likely(!err))
1042                 return NETDEV_TX_OK;
1043
1044         if (err != -EMSGSIZE)
1045                 dev_kfree_skb(skb);
1046
1047         if (err == -ELOOP)
1048                 DEV_STATS_INC(dev, collisions);
1049         else if (err == -ENETUNREACH)
1050                 DEV_STATS_INC(dev, tx_carrier_errors);
1051
1052         DEV_STATS_INC(dev, tx_errors);
1053         return NETDEV_TX_OK;
1054 }
1055
1056 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1057 {
1058         if (new_mtu > dev->max_mtu)
1059                 new_mtu = dev->max_mtu;
1060         else if (new_mtu < dev->min_mtu)
1061                 new_mtu = dev->min_mtu;
1062
1063         dev->mtu = new_mtu;
1064         return 0;
1065 }
1066
1067 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1068 {
1069         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1070         struct geneve_dev *geneve = netdev_priv(dev);
1071         __be16 sport;
1072
1073         if (ip_tunnel_info_af(info) == AF_INET) {
1074                 struct rtable *rt;
1075                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1076                 bool use_cache;
1077                 __be32 saddr;
1078                 u8 tos;
1079
1080                 if (!gs4)
1081                         return -EIO;
1082
1083                 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1084                 tos = geneve_get_dsfield(skb, dev, info, &use_cache);
1085                 sport = udp_flow_src_port(geneve->net, skb,
1086                                           1, USHRT_MAX, true);
1087
1088                 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
1089                                            &info->key,
1090                                            sport, geneve->cfg.info.key.tp_dst,
1091                                            tos,
1092                                            use_cache ? &info->dst_cache : NULL);
1093                 if (IS_ERR(rt))
1094                         return PTR_ERR(rt);
1095
1096                 ip_rt_put(rt);
1097                 info->key.u.ipv4.src = saddr;
1098 #if IS_ENABLED(CONFIG_IPV6)
1099         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1100                 struct dst_entry *dst;
1101                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1102                 struct in6_addr saddr;
1103                 bool use_cache;
1104                 u8 prio;
1105
1106                 if (!gs6)
1107                         return -EIO;
1108
1109                 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1110                 prio = geneve_get_dsfield(skb, dev, info, &use_cache);
1111                 sport = udp_flow_src_port(geneve->net, skb,
1112                                           1, USHRT_MAX, true);
1113
1114                 dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
1115                                              &saddr, &info->key, sport,
1116                                              geneve->cfg.info.key.tp_dst, prio,
1117                                              use_cache ? &info->dst_cache : NULL);
1118                 if (IS_ERR(dst))
1119                         return PTR_ERR(dst);
1120
1121                 dst_release(dst);
1122                 info->key.u.ipv6.src = saddr;
1123 #endif
1124         } else {
1125                 return -EINVAL;
1126         }
1127
1128         info->key.tp_src = sport;
1129         info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1130         return 0;
1131 }
1132
1133 static const struct net_device_ops geneve_netdev_ops = {
1134         .ndo_init               = geneve_init,
1135         .ndo_uninit             = geneve_uninit,
1136         .ndo_open               = geneve_open,
1137         .ndo_stop               = geneve_stop,
1138         .ndo_start_xmit         = geneve_xmit,
1139         .ndo_get_stats64        = dev_get_tstats64,
1140         .ndo_change_mtu         = geneve_change_mtu,
1141         .ndo_validate_addr      = eth_validate_addr,
1142         .ndo_set_mac_address    = eth_mac_addr,
1143         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1144 };
1145
1146 static void geneve_get_drvinfo(struct net_device *dev,
1147                                struct ethtool_drvinfo *drvinfo)
1148 {
1149         strscpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1150         strscpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1151 }
1152
1153 static const struct ethtool_ops geneve_ethtool_ops = {
1154         .get_drvinfo    = geneve_get_drvinfo,
1155         .get_link       = ethtool_op_get_link,
1156 };
1157
1158 /* Info for udev, that this is a virtual tunnel endpoint */
1159 static struct device_type geneve_type = {
1160         .name = "geneve",
1161 };
1162
1163 /* Calls the ndo_udp_tunnel_add of the caller in order to
1164  * supply the listening GENEVE udp ports. Callers are expected
1165  * to implement the ndo_udp_tunnel_add.
1166  */
1167 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1168 {
1169         struct net *net = dev_net(dev);
1170         struct geneve_net *gn = net_generic(net, geneve_net_id);
1171         struct geneve_sock *gs;
1172
1173         rcu_read_lock();
1174         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1175                 if (push) {
1176                         udp_tunnel_push_rx_port(dev, gs->sock,
1177                                                 UDP_TUNNEL_TYPE_GENEVE);
1178                 } else {
1179                         udp_tunnel_drop_rx_port(dev, gs->sock,
1180                                                 UDP_TUNNEL_TYPE_GENEVE);
1181                 }
1182         }
1183         rcu_read_unlock();
1184 }
1185
1186 /* Initialize the device structure. */
1187 static void geneve_setup(struct net_device *dev)
1188 {
1189         ether_setup(dev);
1190
1191         dev->netdev_ops = &geneve_netdev_ops;
1192         dev->ethtool_ops = &geneve_ethtool_ops;
1193         dev->needs_free_netdev = true;
1194
1195         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1196
1197         dev->features    |= NETIF_F_LLTX;
1198         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1199         dev->features    |= NETIF_F_RXCSUM;
1200         dev->features    |= NETIF_F_GSO_SOFTWARE;
1201
1202         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1203         dev->hw_features |= NETIF_F_RXCSUM;
1204         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1205
1206         /* MTU range: 68 - (something less than 65535) */
1207         dev->min_mtu = ETH_MIN_MTU;
1208         /* The max_mtu calculation does not take account of GENEVE
1209          * options, to avoid excluding potentially valid
1210          * configurations. This will be further reduced by IPvX hdr size.
1211          */
1212         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1213
1214         netif_keep_dst(dev);
1215         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1216         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1217         eth_hw_addr_random(dev);
1218 }
1219
1220 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1221         [IFLA_GENEVE_UNSPEC]            = { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT },
1222         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1223         [IFLA_GENEVE_REMOTE]            = { .len = sizeof_field(struct iphdr, daddr) },
1224         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1225         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1226         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1227         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1228         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1229         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1230         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1231         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1232         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1233         [IFLA_GENEVE_TTL_INHERIT]       = { .type = NLA_U8 },
1234         [IFLA_GENEVE_DF]                = { .type = NLA_U8 },
1235         [IFLA_GENEVE_INNER_PROTO_INHERIT]       = { .type = NLA_FLAG },
1236 };
1237
1238 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1239                            struct netlink_ext_ack *extack)
1240 {
1241         if (tb[IFLA_ADDRESS]) {
1242                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1243                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1244                                             "Provided link layer address is not Ethernet");
1245                         return -EINVAL;
1246                 }
1247
1248                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1249                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1250                                             "Provided Ethernet address is not unicast");
1251                         return -EADDRNOTAVAIL;
1252                 }
1253         }
1254
1255         if (!data) {
1256                 NL_SET_ERR_MSG(extack,
1257                                "Not enough attributes provided to perform the operation");
1258                 return -EINVAL;
1259         }
1260
1261         if (data[IFLA_GENEVE_ID]) {
1262                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1263
1264                 if (vni >= GENEVE_N_VID) {
1265                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1266                                             "Geneve ID must be lower than 16777216");
1267                         return -ERANGE;
1268                 }
1269         }
1270
1271         if (data[IFLA_GENEVE_DF]) {
1272                 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1273
1274                 if (df < 0 || df > GENEVE_DF_MAX) {
1275                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1276                                             "Invalid DF attribute");
1277                         return -EINVAL;
1278                 }
1279         }
1280
1281         return 0;
1282 }
1283
1284 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1285                                           const struct ip_tunnel_info *info,
1286                                           bool *tun_on_same_port,
1287                                           bool *tun_collect_md)
1288 {
1289         struct geneve_dev *geneve, *t = NULL;
1290
1291         *tun_on_same_port = false;
1292         *tun_collect_md = false;
1293         list_for_each_entry(geneve, &gn->geneve_list, next) {
1294                 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1295                         *tun_collect_md = geneve->cfg.collect_md;
1296                         *tun_on_same_port = true;
1297                 }
1298                 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1299                     info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1300                     !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1301                         t = geneve;
1302         }
1303         return t;
1304 }
1305
1306 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1307 {
1308         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1309                  info->key.ttl || info->key.label || info->key.tp_src ||
1310                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1311 }
1312
1313 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1314                                   struct ip_tunnel_info *b)
1315 {
1316         if (ip_tunnel_info_af(a) == AF_INET)
1317                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1318         else
1319                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1320 }
1321
1322 static int geneve_configure(struct net *net, struct net_device *dev,
1323                             struct netlink_ext_ack *extack,
1324                             const struct geneve_config *cfg)
1325 {
1326         struct geneve_net *gn = net_generic(net, geneve_net_id);
1327         struct geneve_dev *t, *geneve = netdev_priv(dev);
1328         const struct ip_tunnel_info *info = &cfg->info;
1329         bool tun_collect_md, tun_on_same_port;
1330         int err, encap_len;
1331
1332         if (cfg->collect_md && !is_tnl_info_zero(info)) {
1333                 NL_SET_ERR_MSG(extack,
1334                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1335                 return -EINVAL;
1336         }
1337
1338         geneve->net = net;
1339         geneve->dev = dev;
1340
1341         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1342         if (t)
1343                 return -EBUSY;
1344
1345         /* make enough headroom for basic scenario */
1346         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1347         if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1348                 encap_len += sizeof(struct iphdr);
1349                 dev->max_mtu -= sizeof(struct iphdr);
1350         } else {
1351                 encap_len += sizeof(struct ipv6hdr);
1352                 dev->max_mtu -= sizeof(struct ipv6hdr);
1353         }
1354         dev->needed_headroom = encap_len + ETH_HLEN;
1355
1356         if (cfg->collect_md) {
1357                 if (tun_on_same_port) {
1358                         NL_SET_ERR_MSG(extack,
1359                                        "There can be only one externally controlled device on a destination port");
1360                         return -EPERM;
1361                 }
1362         } else {
1363                 if (tun_collect_md) {
1364                         NL_SET_ERR_MSG(extack,
1365                                        "There already exists an externally controlled device on this destination port");
1366                         return -EPERM;
1367                 }
1368         }
1369
1370         dst_cache_reset(&geneve->cfg.info.dst_cache);
1371         memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1372
1373         if (geneve->cfg.inner_proto_inherit) {
1374                 dev->header_ops = NULL;
1375                 dev->type = ARPHRD_NONE;
1376                 dev->hard_header_len = 0;
1377                 dev->addr_len = 0;
1378                 dev->flags = IFF_POINTOPOINT | IFF_NOARP;
1379         }
1380
1381         err = register_netdevice(dev);
1382         if (err)
1383                 return err;
1384
1385         list_add(&geneve->next, &gn->geneve_list);
1386         return 0;
1387 }
1388
1389 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1390 {
1391         memset(info, 0, sizeof(*info));
1392         info->key.tp_dst = htons(dst_port);
1393 }
1394
1395 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1396                           struct netlink_ext_ack *extack,
1397                           struct geneve_config *cfg, bool changelink)
1398 {
1399         struct ip_tunnel_info *info = &cfg->info;
1400         int attrtype;
1401
1402         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1403                 NL_SET_ERR_MSG(extack,
1404                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1405                 return -EINVAL;
1406         }
1407
1408         if (data[IFLA_GENEVE_REMOTE]) {
1409                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1410                         attrtype = IFLA_GENEVE_REMOTE;
1411                         goto change_notsup;
1412                 }
1413
1414                 info->key.u.ipv4.dst =
1415                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1416
1417                 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1418                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1419                                             "Remote IPv4 address cannot be Multicast");
1420                         return -EINVAL;
1421                 }
1422         }
1423
1424         if (data[IFLA_GENEVE_REMOTE6]) {
1425 #if IS_ENABLED(CONFIG_IPV6)
1426                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1427                         attrtype = IFLA_GENEVE_REMOTE6;
1428                         goto change_notsup;
1429                 }
1430
1431                 info->mode = IP_TUNNEL_INFO_IPV6;
1432                 info->key.u.ipv6.dst =
1433                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1434
1435                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1436                     IPV6_ADDR_LINKLOCAL) {
1437                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1438                                             "Remote IPv6 address cannot be link-local");
1439                         return -EINVAL;
1440                 }
1441                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1442                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1443                                             "Remote IPv6 address cannot be Multicast");
1444                         return -EINVAL;
1445                 }
1446                 info->key.tun_flags |= TUNNEL_CSUM;
1447                 cfg->use_udp6_rx_checksums = true;
1448 #else
1449                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1450                                     "IPv6 support not enabled in the kernel");
1451                 return -EPFNOSUPPORT;
1452 #endif
1453         }
1454
1455         if (data[IFLA_GENEVE_ID]) {
1456                 __u32 vni;
1457                 __u8 tvni[3];
1458                 __be64 tunid;
1459
1460                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1461                 tvni[0] = (vni & 0x00ff0000) >> 16;
1462                 tvni[1] = (vni & 0x0000ff00) >> 8;
1463                 tvni[2] =  vni & 0x000000ff;
1464
1465                 tunid = vni_to_tunnel_id(tvni);
1466                 if (changelink && (tunid != info->key.tun_id)) {
1467                         attrtype = IFLA_GENEVE_ID;
1468                         goto change_notsup;
1469                 }
1470                 info->key.tun_id = tunid;
1471         }
1472
1473         if (data[IFLA_GENEVE_TTL_INHERIT]) {
1474                 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1475                         cfg->ttl_inherit = true;
1476                 else
1477                         cfg->ttl_inherit = false;
1478         } else if (data[IFLA_GENEVE_TTL]) {
1479                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1480                 cfg->ttl_inherit = false;
1481         }
1482
1483         if (data[IFLA_GENEVE_TOS])
1484                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1485
1486         if (data[IFLA_GENEVE_DF])
1487                 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1488
1489         if (data[IFLA_GENEVE_LABEL]) {
1490                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1491                                   IPV6_FLOWLABEL_MASK;
1492                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1493                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1494                                             "Label attribute only applies for IPv6 Geneve devices");
1495                         return -EINVAL;
1496                 }
1497         }
1498
1499         if (data[IFLA_GENEVE_PORT]) {
1500                 if (changelink) {
1501                         attrtype = IFLA_GENEVE_PORT;
1502                         goto change_notsup;
1503                 }
1504                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1505         }
1506
1507         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1508                 if (changelink) {
1509                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1510                         goto change_notsup;
1511                 }
1512                 cfg->collect_md = true;
1513         }
1514
1515         if (data[IFLA_GENEVE_UDP_CSUM]) {
1516                 if (changelink) {
1517                         attrtype = IFLA_GENEVE_UDP_CSUM;
1518                         goto change_notsup;
1519                 }
1520                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1521                         info->key.tun_flags |= TUNNEL_CSUM;
1522         }
1523
1524         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1525 #if IS_ENABLED(CONFIG_IPV6)
1526                 if (changelink) {
1527                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1528                         goto change_notsup;
1529                 }
1530                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1531                         info->key.tun_flags &= ~TUNNEL_CSUM;
1532 #else
1533                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1534                                     "IPv6 support not enabled in the kernel");
1535                 return -EPFNOSUPPORT;
1536 #endif
1537         }
1538
1539         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1540 #if IS_ENABLED(CONFIG_IPV6)
1541                 if (changelink) {
1542                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1543                         goto change_notsup;
1544                 }
1545                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1546                         cfg->use_udp6_rx_checksums = false;
1547 #else
1548                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1549                                     "IPv6 support not enabled in the kernel");
1550                 return -EPFNOSUPPORT;
1551 #endif
1552         }
1553
1554         if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) {
1555                 if (changelink) {
1556                         attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT;
1557                         goto change_notsup;
1558                 }
1559                 cfg->inner_proto_inherit = true;
1560         }
1561
1562         return 0;
1563 change_notsup:
1564         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1565                             "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported");
1566         return -EOPNOTSUPP;
1567 }
1568
1569 static void geneve_link_config(struct net_device *dev,
1570                                struct ip_tunnel_info *info, struct nlattr *tb[])
1571 {
1572         struct geneve_dev *geneve = netdev_priv(dev);
1573         int ldev_mtu = 0;
1574
1575         if (tb[IFLA_MTU]) {
1576                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1577                 return;
1578         }
1579
1580         switch (ip_tunnel_info_af(info)) {
1581         case AF_INET: {
1582                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1583                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1584
1585                 if (!IS_ERR(rt) && rt->dst.dev) {
1586                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1587                         ip_rt_put(rt);
1588                 }
1589                 break;
1590         }
1591 #if IS_ENABLED(CONFIG_IPV6)
1592         case AF_INET6: {
1593                 struct rt6_info *rt;
1594
1595                 if (!__in6_dev_get(dev))
1596                         break;
1597
1598                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1599                                 NULL, 0);
1600
1601                 if (rt && rt->dst.dev)
1602                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1603                 ip6_rt_put(rt);
1604                 break;
1605         }
1606 #endif
1607         }
1608
1609         if (ldev_mtu <= 0)
1610                 return;
1611
1612         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1613 }
1614
1615 static int geneve_newlink(struct net *net, struct net_device *dev,
1616                           struct nlattr *tb[], struct nlattr *data[],
1617                           struct netlink_ext_ack *extack)
1618 {
1619         struct geneve_config cfg = {
1620                 .df = GENEVE_DF_UNSET,
1621                 .use_udp6_rx_checksums = false,
1622                 .ttl_inherit = false,
1623                 .collect_md = false,
1624         };
1625         int err;
1626
1627         init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1628         err = geneve_nl2info(tb, data, extack, &cfg, false);
1629         if (err)
1630                 return err;
1631
1632         err = geneve_configure(net, dev, extack, &cfg);
1633         if (err)
1634                 return err;
1635
1636         geneve_link_config(dev, &cfg.info, tb);
1637
1638         return 0;
1639 }
1640
1641 /* Quiesces the geneve device data path for both TX and RX.
1642  *
1643  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1644  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1645  * to complete for the existing set of in-flight packets to be transmitted,
1646  * then we would have quiesced the transmit data path. All the future packets
1647  * will get dropped until we unquiesce the data path.
1648  *
1649  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1650  * if we set that to NULL under RCU and wait for synchronize_net() to
1651  * complete, then we would have quiesced the receive data path.
1652  */
1653 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1654                            struct geneve_sock **gs6)
1655 {
1656         *gs4 = rtnl_dereference(geneve->sock4);
1657         rcu_assign_pointer(geneve->sock4, NULL);
1658         if (*gs4)
1659                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1660 #if IS_ENABLED(CONFIG_IPV6)
1661         *gs6 = rtnl_dereference(geneve->sock6);
1662         rcu_assign_pointer(geneve->sock6, NULL);
1663         if (*gs6)
1664                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1665 #else
1666         *gs6 = NULL;
1667 #endif
1668         synchronize_net();
1669 }
1670
1671 /* Resumes the geneve device data path for both TX and RX. */
1672 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1673                              struct geneve_sock __maybe_unused *gs6)
1674 {
1675         rcu_assign_pointer(geneve->sock4, gs4);
1676         if (gs4)
1677                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1678 #if IS_ENABLED(CONFIG_IPV6)
1679         rcu_assign_pointer(geneve->sock6, gs6);
1680         if (gs6)
1681                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1682 #endif
1683         synchronize_net();
1684 }
1685
1686 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1687                              struct nlattr *data[],
1688                              struct netlink_ext_ack *extack)
1689 {
1690         struct geneve_dev *geneve = netdev_priv(dev);
1691         struct geneve_sock *gs4, *gs6;
1692         struct geneve_config cfg;
1693         int err;
1694
1695         /* If the geneve device is configured for metadata (or externally
1696          * controlled, for example, OVS), then nothing can be changed.
1697          */
1698         if (geneve->cfg.collect_md)
1699                 return -EOPNOTSUPP;
1700
1701         /* Start with the existing info. */
1702         memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1703         err = geneve_nl2info(tb, data, extack, &cfg, true);
1704         if (err)
1705                 return err;
1706
1707         if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1708                 dst_cache_reset(&cfg.info.dst_cache);
1709                 geneve_link_config(dev, &cfg.info, tb);
1710         }
1711
1712         geneve_quiesce(geneve, &gs4, &gs6);
1713         memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1714         geneve_unquiesce(geneve, gs4, gs6);
1715
1716         return 0;
1717 }
1718
1719 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1720 {
1721         struct geneve_dev *geneve = netdev_priv(dev);
1722
1723         list_del(&geneve->next);
1724         unregister_netdevice_queue(dev, head);
1725 }
1726
1727 static size_t geneve_get_size(const struct net_device *dev)
1728 {
1729         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1730                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1731                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1732                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1733                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_DF */
1734                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1735                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1736                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1737                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1738                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1739                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1740                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1741                 nla_total_size(0) +      /* IFLA_GENEVE_INNER_PROTO_INHERIT */
1742                 0;
1743 }
1744
1745 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1746 {
1747         struct geneve_dev *geneve = netdev_priv(dev);
1748         struct ip_tunnel_info *info = &geneve->cfg.info;
1749         bool ttl_inherit = geneve->cfg.ttl_inherit;
1750         bool metadata = geneve->cfg.collect_md;
1751         __u8 tmp_vni[3];
1752         __u32 vni;
1753
1754         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1755         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1756         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1757                 goto nla_put_failure;
1758
1759         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1760                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1761                                     info->key.u.ipv4.dst))
1762                         goto nla_put_failure;
1763                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1764                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1765                         goto nla_put_failure;
1766
1767 #if IS_ENABLED(CONFIG_IPV6)
1768         } else if (!metadata) {
1769                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1770                                      &info->key.u.ipv6.dst))
1771                         goto nla_put_failure;
1772                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1773                                !(info->key.tun_flags & TUNNEL_CSUM)))
1774                         goto nla_put_failure;
1775 #endif
1776         }
1777
1778         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1779             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1780             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1781                 goto nla_put_failure;
1782
1783         if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1784                 goto nla_put_failure;
1785
1786         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1787                 goto nla_put_failure;
1788
1789         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1790                 goto nla_put_failure;
1791
1792 #if IS_ENABLED(CONFIG_IPV6)
1793         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1794                        !geneve->cfg.use_udp6_rx_checksums))
1795                 goto nla_put_failure;
1796 #endif
1797
1798         if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1799                 goto nla_put_failure;
1800
1801         if (geneve->cfg.inner_proto_inherit &&
1802             nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT))
1803                 goto nla_put_failure;
1804
1805         return 0;
1806
1807 nla_put_failure:
1808         return -EMSGSIZE;
1809 }
1810
1811 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1812         .kind           = "geneve",
1813         .maxtype        = IFLA_GENEVE_MAX,
1814         .policy         = geneve_policy,
1815         .priv_size      = sizeof(struct geneve_dev),
1816         .setup          = geneve_setup,
1817         .validate       = geneve_validate,
1818         .newlink        = geneve_newlink,
1819         .changelink     = geneve_changelink,
1820         .dellink        = geneve_dellink,
1821         .get_size       = geneve_get_size,
1822         .fill_info      = geneve_fill_info,
1823 };
1824
1825 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1826                                         u8 name_assign_type, u16 dst_port)
1827 {
1828         struct nlattr *tb[IFLA_MAX + 1];
1829         struct net_device *dev;
1830         LIST_HEAD(list_kill);
1831         int err;
1832         struct geneve_config cfg = {
1833                 .df = GENEVE_DF_UNSET,
1834                 .use_udp6_rx_checksums = true,
1835                 .ttl_inherit = false,
1836                 .collect_md = true,
1837         };
1838
1839         memset(tb, 0, sizeof(tb));
1840         dev = rtnl_create_link(net, name, name_assign_type,
1841                                &geneve_link_ops, tb, NULL);
1842         if (IS_ERR(dev))
1843                 return dev;
1844
1845         init_tnl_info(&cfg.info, dst_port);
1846         err = geneve_configure(net, dev, NULL, &cfg);
1847         if (err) {
1848                 free_netdev(dev);
1849                 return ERR_PTR(err);
1850         }
1851
1852         /* openvswitch users expect packet sizes to be unrestricted,
1853          * so set the largest MTU we can.
1854          */
1855         err = geneve_change_mtu(dev, IP_MAX_MTU);
1856         if (err)
1857                 goto err;
1858
1859         err = rtnl_configure_link(dev, NULL, 0, NULL);
1860         if (err < 0)
1861                 goto err;
1862
1863         return dev;
1864 err:
1865         geneve_dellink(dev, &list_kill);
1866         unregister_netdevice_many(&list_kill);
1867         return ERR_PTR(err);
1868 }
1869 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1870
1871 static int geneve_netdevice_event(struct notifier_block *unused,
1872                                   unsigned long event, void *ptr)
1873 {
1874         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1875
1876         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1877                 geneve_offload_rx_ports(dev, true);
1878         else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1879                 geneve_offload_rx_ports(dev, false);
1880
1881         return NOTIFY_DONE;
1882 }
1883
1884 static struct notifier_block geneve_notifier_block __read_mostly = {
1885         .notifier_call = geneve_netdevice_event,
1886 };
1887
1888 static __net_init int geneve_init_net(struct net *net)
1889 {
1890         struct geneve_net *gn = net_generic(net, geneve_net_id);
1891
1892         INIT_LIST_HEAD(&gn->geneve_list);
1893         INIT_LIST_HEAD(&gn->sock_list);
1894         return 0;
1895 }
1896
1897 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1898 {
1899         struct geneve_net *gn = net_generic(net, geneve_net_id);
1900         struct geneve_dev *geneve, *next;
1901         struct net_device *dev, *aux;
1902
1903         /* gather any geneve devices that were moved into this ns */
1904         for_each_netdev_safe(net, dev, aux)
1905                 if (dev->rtnl_link_ops == &geneve_link_ops)
1906                         unregister_netdevice_queue(dev, head);
1907
1908         /* now gather any other geneve devices that were created in this ns */
1909         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1910                 /* If geneve->dev is in the same netns, it was already added
1911                  * to the list by the previous loop.
1912                  */
1913                 if (!net_eq(dev_net(geneve->dev), net))
1914                         unregister_netdevice_queue(geneve->dev, head);
1915         }
1916 }
1917
1918 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1919 {
1920         struct net *net;
1921         LIST_HEAD(list);
1922
1923         rtnl_lock();
1924         list_for_each_entry(net, net_list, exit_list)
1925                 geneve_destroy_tunnels(net, &list);
1926
1927         /* unregister the devices gathered above */
1928         unregister_netdevice_many(&list);
1929         rtnl_unlock();
1930
1931         list_for_each_entry(net, net_list, exit_list) {
1932                 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1933
1934                 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1935         }
1936 }
1937
1938 static struct pernet_operations geneve_net_ops = {
1939         .init = geneve_init_net,
1940         .exit_batch = geneve_exit_batch_net,
1941         .id   = &geneve_net_id,
1942         .size = sizeof(struct geneve_net),
1943 };
1944
1945 static int __init geneve_init_module(void)
1946 {
1947         int rc;
1948
1949         rc = register_pernet_subsys(&geneve_net_ops);
1950         if (rc)
1951                 goto out1;
1952
1953         rc = register_netdevice_notifier(&geneve_notifier_block);
1954         if (rc)
1955                 goto out2;
1956
1957         rc = rtnl_link_register(&geneve_link_ops);
1958         if (rc)
1959                 goto out3;
1960
1961         return 0;
1962 out3:
1963         unregister_netdevice_notifier(&geneve_notifier_block);
1964 out2:
1965         unregister_pernet_subsys(&geneve_net_ops);
1966 out1:
1967         return rc;
1968 }
1969 late_initcall(geneve_init_module);
1970
1971 static void __exit geneve_cleanup_module(void)
1972 {
1973         rtnl_link_unregister(&geneve_link_ops);
1974         unregister_netdevice_notifier(&geneve_notifier_block);
1975         unregister_pernet_subsys(&geneve_net_ops);
1976 }
1977 module_exit(geneve_cleanup_module);
1978
1979 MODULE_LICENSE("GPL");
1980 MODULE_VERSION(GENEVE_NETDEV_VER);
1981 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1982 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1983 MODULE_ALIAS_RTNL_LINK("geneve");