drivers/net/geneve.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * GENEVE: Generic Network Virtualization Encapsulation
   4  *
   5  * Copyright (c) 2015 Red Hat, Inc.
   6  */
   7
   8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   9
  10 #include <linux/ethtool.h>
  11 #include <linux/kernel.h>
  12 #include <linux/module.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/hash.h>
  15 #include <net/ipv6_stubs.h>
  16 #include <net/dst_metadata.h>
  17 #include <net/gro_cells.h>
  18 #include <net/rtnetlink.h>
  19 #include <net/geneve.h>
  20 #include <net/protocol.h>
  21
  22 #define GENEVE_NETDEV_VER       "0.6"
  23
  24 #define GENEVE_N_VID            (1u << 24)
  25 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  26
  27 #define VNI_HASH_BITS           10
  28 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  29
  30 static bool log_ecn_error = true;
  31 module_param(log_ecn_error, bool, 0644);
  32 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  33
  34 #define GENEVE_VER 0
  35 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  36 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
  37 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
  38
  39 /* per-network namespace private data for this module */
  40 struct geneve_net {
  41         struct list_head        geneve_list;
  42         struct list_head        sock_list;
  43 };
  44
  45 static unsigned int geneve_net_id;
  46
  47 struct geneve_dev_node {
  48         struct hlist_node hlist;
  49         struct geneve_dev *geneve;
  50 };
  51
  52 struct geneve_config {
  53         struct ip_tunnel_info   info;
  54         bool                    collect_md;
  55         bool                    use_udp6_rx_checksums;
  56         bool                    ttl_inherit;
  57         enum ifla_geneve_df     df;
  58 };
  59
  60 /* Pseudo network device */
  61 struct geneve_dev {
  62         struct geneve_dev_node hlist4;  /* vni hash table for IPv4 socket */
  63 #if IS_ENABLED(CONFIG_IPV6)
  64         struct geneve_dev_node hlist6;  /* vni hash table for IPv6 socket */
  65 #endif
  66         struct net         *net;        /* netns for packet i/o */
  67         struct net_device  *dev;        /* netdev for geneve tunnel */
  68         struct geneve_sock __rcu *sock4;        /* IPv4 socket used for geneve tunnel */
  69 #if IS_ENABLED(CONFIG_IPV6)
  70         struct geneve_sock __rcu *sock6;        /* IPv6 socket used for geneve tunnel */
  71 #endif
  72         struct list_head   next;        /* geneve's per namespace list */
  73         struct gro_cells   gro_cells;
  74         struct geneve_config cfg;
  75 };
  76
  77 struct geneve_sock {
  78         bool                    collect_md;
  79         struct list_head        list;
  80         struct socket           *sock;
  81         struct rcu_head         rcu;
  82         int                     refcnt;
  83         struct hlist_head       vni_list[VNI_HASH_SIZE];
  84 };
  85
  86 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  87 {
  88         __u32 vnid;
  89
  90         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  91         return hash_32(vnid, VNI_HASH_BITS);
  92 }
  93
  94 static __be64 vni_to_tunnel_id(const __u8 *vni)
  95 {
  96 #ifdef __BIG_ENDIAN
  97         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
  98 #else
  99         return (__force __be64)(((__force u64)vni[0] << 40) |
 100                                 ((__force u64)vni[1] << 48) |
 101                                 ((__force u64)vni[2] << 56));
 102 #endif
 103 }
 104
 105 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 106 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 107 {
 108 #ifdef __BIG_ENDIAN
 109         vni[0] = (__force __u8)(tun_id >> 16);
 110         vni[1] = (__force __u8)(tun_id >> 8);
 111         vni[2] = (__force __u8)tun_id;
 112 #else
 113         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 114         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 115         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 116 #endif
 117 }
 118
 119 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
 120 {
 121         return !memcmp(vni, &tun_id[5], 3);
 122 }
 123
 124 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 125 {
 126         return gs->sock->sk->sk_family;
 127 }
 128
 129 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 130                                         __be32 addr, u8 vni[])
 131 {
 132         struct hlist_head *vni_list_head;
 133         struct geneve_dev_node *node;
 134         __u32 hash;
 135
 136         /* Find the device for this VNI */
 137         hash = geneve_net_vni_hash(vni);
 138         vni_list_head = &gs->vni_list[hash];
 139         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 140                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 141                     addr == node->geneve->cfg.info.key.u.ipv4.dst)
 142                         return node->geneve;
 143         }
 144         return NULL;
 145 }
 146
 147 #if IS_ENABLED(CONFIG_IPV6)
 148 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 149                                          struct in6_addr addr6, u8 vni[])
 150 {
 151         struct hlist_head *vni_list_head;
 152         struct geneve_dev_node *node;
 153         __u32 hash;
 154
 155         /* Find the device for this VNI */
 156         hash = geneve_net_vni_hash(vni);
 157         vni_list_head = &gs->vni_list[hash];
 158         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 159                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 160                     ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
 161                         return node->geneve;
 162         }
 163         return NULL;
 164 }
 165 #endif
 166
 167 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 168 {
 169         return (struct genevehdr *)(udp_hdr(skb) + 1);
 170 }
 171
 172 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 173                                             struct sk_buff *skb)
 174 {
 175         static u8 zero_vni[3];
 176         u8 *vni;
 177
 178         if (geneve_get_sk_family(gs) == AF_INET) {
 179                 struct iphdr *iph;
 180                 __be32 addr;
 181
 182                 iph = ip_hdr(skb); /* outer IP header... */
 183
 184                 if (gs->collect_md) {
 185                         vni = zero_vni;
 186                         addr = 0;
 187                 } else {
 188                         vni = geneve_hdr(skb)->vni;
 189                         addr = iph->saddr;
 190                 }
 191
 192                 return geneve_lookup(gs, addr, vni);
 193 #if IS_ENABLED(CONFIG_IPV6)
 194         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 195                 static struct in6_addr zero_addr6;
 196                 struct ipv6hdr *ip6h;
 197                 struct in6_addr addr6;
 198
 199                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 200
 201                 if (gs->collect_md) {
 202                         vni = zero_vni;
 203                         addr6 = zero_addr6;
 204                 } else {
 205                         vni = geneve_hdr(skb)->vni;
 206                         addr6 = ip6h->saddr;
 207                 }
 208
 209                 return geneve6_lookup(gs, addr6, vni);
 210 #endif
 211         }
 212         return NULL;
 213 }
 214
 215 /* geneve receive/decap routine */
 216 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 217                       struct sk_buff *skb)
 218 {
 219         struct genevehdr *gnvh = geneve_hdr(skb);
 220         struct metadata_dst *tun_dst = NULL;
 221         unsigned int len;
 222         int err = 0;
 223         void *oiph;
 224
 225         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 226                 __be16 flags;
 227
 228                 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
 229                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 230
 231                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 232                                          vni_to_tunnel_id(gnvh->vni),
 233                                          gnvh->opt_len * 4);
 234                 if (!tun_dst) {
 235                         geneve->dev->stats.rx_dropped++;
 236                         goto drop;
 237                 }
 238                 /* Update tunnel dst according to Geneve options. */
 239                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 240                                         gnvh->options, gnvh->opt_len * 4,
 241                                         TUNNEL_GENEVE_OPT);
 242         } else {
 243                 /* Drop packets w/ critical options,
 244                  * since we don't support any...
 245                  */
 246                 if (gnvh->critical) {
 247                         geneve->dev->stats.rx_frame_errors++;
 248                         geneve->dev->stats.rx_errors++;
 249                         goto drop;
 250                 }
 251         }
 252
 253         skb_reset_mac_header(skb);
 254         skb->protocol = eth_type_trans(skb, geneve->dev);
 255         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 256
 257         if (tun_dst)
 258                 skb_dst_set(skb, &tun_dst->dst);
 259
 260         /* Ignore packet loops (and multicast echo) */
 261         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
 262                 goto rx_error;
 263
 264         switch (skb_protocol(skb, true)) {
 265         case htons(ETH_P_IP):
 266                 if (pskb_may_pull(skb, sizeof(struct iphdr)))
 267                         goto rx_error;
 268                 break;
 269         case htons(ETH_P_IPV6):
 270                 if (pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 271                         goto rx_error;
 272                 break;
 273         default:
 274                 goto rx_error;
 275         }
 276         oiph = skb_network_header(skb);
 277         skb_reset_network_header(skb);
 278
 279         if (geneve_get_sk_family(gs) == AF_INET)
 280                 err = IP_ECN_decapsulate(oiph, skb);
 281 #if IS_ENABLED(CONFIG_IPV6)
 282         else
 283                 err = IP6_ECN_decapsulate(oiph, skb);
 284 #endif
 285
 286         if (unlikely(err)) {
 287                 if (log_ecn_error) {
 288                         if (geneve_get_sk_family(gs) == AF_INET)
 289                                 net_info_ratelimited("non-ECT from %pI4 "
 290                                                      "with TOS=%#x\n",
 291                                                      &((struct iphdr *)oiph)->saddr,
 292                                                      ((struct iphdr *)oiph)->tos);
 293 #if IS_ENABLED(CONFIG_IPV6)
 294                         else
 295                                 net_info_ratelimited("non-ECT from %pI6\n",
 296                                                      &((struct ipv6hdr *)oiph)->saddr);
 297 #endif
 298                 }
 299                 if (err > 1) {
 300                         ++geneve->dev->stats.rx_frame_errors;
 301                         ++geneve->dev->stats.rx_errors;
 302                         goto drop;
 303                 }
 304         }
 305
 306         len = skb->len;
 307         err = gro_cells_receive(&geneve->gro_cells, skb);
 308         if (likely(err == NET_RX_SUCCESS))
 309                 dev_sw_netstats_rx_add(geneve->dev, len);
 310
 311         return;
 312 rx_error:
 313         geneve->dev->stats.rx_errors++;
 314 drop:
 315         /* Consume bad packet */
 316         kfree_skb(skb);
 317 }
 318
 319 /* Setup stats when device is created */
 320 static int geneve_init(struct net_device *dev)
 321 {
 322         struct geneve_dev *geneve = netdev_priv(dev);
 323         int err;
 324
 325         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 326         if (!dev->tstats)
 327                 return -ENOMEM;
 328
 329         err = gro_cells_init(&geneve->gro_cells, dev);
 330         if (err) {
 331                 free_percpu(dev->tstats);
 332                 return err;
 333         }
 334
 335         err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
 336         if (err) {
 337                 free_percpu(dev->tstats);
 338                 gro_cells_destroy(&geneve->gro_cells);
 339                 return err;
 340         }
 341         return 0;
 342 }
 343
 344 static void geneve_uninit(struct net_device *dev)
 345 {
 346         struct geneve_dev *geneve = netdev_priv(dev);
 347
 348         dst_cache_destroy(&geneve->cfg.info.dst_cache);
 349         gro_cells_destroy(&geneve->gro_cells);
 350         free_percpu(dev->tstats);
 351 }
 352
 353 /* Callback from net/ipv4/udp.c to receive packets */
 354 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 355 {
 356         struct genevehdr *geneveh;
 357         struct geneve_dev *geneve;
 358         struct geneve_sock *gs;
 359         int opts_len;
 360
 361         /* Need UDP and Geneve header to be present */
 362         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 363                 goto drop;
 364
 365         /* Return packets with reserved bits set */
 366         geneveh = geneve_hdr(skb);
 367         if (unlikely(geneveh->ver != GENEVE_VER))
 368                 goto drop;
 369
 370         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 371                 goto drop;
 372
 373         gs = rcu_dereference_sk_user_data(sk);
 374         if (!gs)
 375                 goto drop;
 376
 377         geneve = geneve_lookup_skb(gs, skb);
 378         if (!geneve)
 379                 goto drop;
 380
 381         opts_len = geneveh->opt_len * 4;
 382         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 383                                  htons(ETH_P_TEB),
 384                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 385                 geneve->dev->stats.rx_dropped++;
 386                 goto drop;
 387         }
 388
 389         geneve_rx(geneve, gs, skb);
 390         return 0;
 391
 392 drop:
 393         /* Consume bad packet */
 394         kfree_skb(skb);
 395         return 0;
 396 }
 397
 398 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
 399 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
 400 {
 401         struct genevehdr *geneveh;
 402         struct geneve_sock *gs;
 403         u8 zero_vni[3] = { 0 };
 404         u8 *vni = zero_vni;
 405
 406         if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
 407                 return -EINVAL;
 408
 409         geneveh = geneve_hdr(skb);
 410         if (geneveh->ver != GENEVE_VER)
 411                 return -EINVAL;
 412
 413         if (geneveh->proto_type != htons(ETH_P_TEB))
 414                 return -EINVAL;
 415
 416         gs = rcu_dereference_sk_user_data(sk);
 417         if (!gs)
 418                 return -ENOENT;
 419
 420         if (geneve_get_sk_family(gs) == AF_INET) {
 421                 struct iphdr *iph = ip_hdr(skb);
 422                 __be32 addr4 = 0;
 423
 424                 if (!gs->collect_md) {
 425                         vni = geneve_hdr(skb)->vni;
 426                         addr4 = iph->daddr;
 427                 }
 428
 429                 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
 430         }
 431
 432 #if IS_ENABLED(CONFIG_IPV6)
 433         if (geneve_get_sk_family(gs) == AF_INET6) {
 434                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 435                 struct in6_addr addr6;
 436
 437                 memset(&addr6, 0, sizeof(struct in6_addr));
 438
 439                 if (!gs->collect_md) {
 440                         vni = geneve_hdr(skb)->vni;
 441                         addr6 = ip6h->daddr;
 442                 }
 443
 444                 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
 445         }
 446 #endif
 447
 448         return -EPFNOSUPPORT;
 449 }
 450
 451 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 452                                          __be16 port, bool ipv6_rx_csum)
 453 {
 454         struct socket *sock;
 455         struct udp_port_cfg udp_conf;
 456         int err;
 457
 458         memset(&udp_conf, 0, sizeof(udp_conf));
 459
 460         if (ipv6) {
 461                 udp_conf.family = AF_INET6;
 462                 udp_conf.ipv6_v6only = 1;
 463                 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
 464         } else {
 465                 udp_conf.family = AF_INET;
 466                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 467         }
 468
 469         udp_conf.local_udp_port = port;
 470
 471         /* Open UDP socket */
 472         err = udp_sock_create(net, &udp_conf, &sock);
 473         if (err < 0)
 474                 return ERR_PTR(err);
 475
 476         return sock;
 477 }
 478
 479 static int geneve_hlen(struct genevehdr *gh)
 480 {
 481         return sizeof(*gh) + gh->opt_len * 4;
 482 }
 483
 484 static struct sk_buff *geneve_gro_receive(struct sock *sk,
 485                                           struct list_head *head,
 486                                           struct sk_buff *skb)
 487 {
 488         struct sk_buff *pp = NULL;
 489         struct sk_buff *p;
 490         struct genevehdr *gh, *gh2;
 491         unsigned int hlen, gh_len, off_gnv;
 492         const struct packet_offload *ptype;
 493         __be16 type;
 494         int flush = 1;
 495
 496         off_gnv = skb_gro_offset(skb);
 497         hlen = off_gnv + sizeof(*gh);
 498         gh = skb_gro_header_fast(skb, off_gnv);
 499         if (skb_gro_header_hard(skb, hlen)) {
 500                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 501                 if (unlikely(!gh))
 502                         goto out;
 503         }
 504
 505         if (gh->ver != GENEVE_VER || gh->oam)
 506                 goto out;
 507         gh_len = geneve_hlen(gh);
 508
 509         hlen = off_gnv + gh_len;
 510         if (skb_gro_header_hard(skb, hlen)) {
 511                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 512                 if (unlikely(!gh))
 513                         goto out;
 514         }
 515
 516         list_for_each_entry(p, head, list) {
 517                 if (!NAPI_GRO_CB(p)->same_flow)
 518                         continue;
 519
 520                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 521                 if (gh->opt_len != gh2->opt_len ||
 522                     memcmp(gh, gh2, gh_len)) {
 523                         NAPI_GRO_CB(p)->same_flow = 0;
 524                         continue;
 525                 }
 526         }
 527
 528         type = gh->proto_type;
 529
 530         rcu_read_lock();
 531         ptype = gro_find_receive_by_type(type);
 532         if (!ptype)
 533                 goto out_unlock;
 534
 535         skb_gro_pull(skb, gh_len);
 536         skb_gro_postpull_rcsum(skb, gh, gh_len);
 537         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 538         flush = 0;
 539
 540 out_unlock:
 541         rcu_read_unlock();
 542 out:
 543         skb_gro_flush_final(skb, pp, flush);
 544
 545         return pp;
 546 }
 547
 548 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 549                                int nhoff)
 550 {
 551         struct genevehdr *gh;
 552         struct packet_offload *ptype;
 553         __be16 type;
 554         int gh_len;
 555         int err = -ENOSYS;
 556
 557         gh = (struct genevehdr *)(skb->data + nhoff);
 558         gh_len = geneve_hlen(gh);
 559         type = gh->proto_type;
 560
 561         rcu_read_lock();
 562         ptype = gro_find_complete_by_type(type);
 563         if (ptype)
 564                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 565
 566         rcu_read_unlock();
 567
 568         skb_set_inner_mac_header(skb, nhoff + gh_len);
 569
 570         return err;
 571 }
 572
 573 /* Create new listen socket if needed */
 574 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 575                                                 bool ipv6, bool ipv6_rx_csum)
 576 {
 577         struct geneve_net *gn = net_generic(net, geneve_net_id);
 578         struct geneve_sock *gs;
 579         struct socket *sock;
 580         struct udp_tunnel_sock_cfg tunnel_cfg;
 581         int h;
 582
 583         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 584         if (!gs)
 585                 return ERR_PTR(-ENOMEM);
 586
 587         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 588         if (IS_ERR(sock)) {
 589                 kfree(gs);
 590                 return ERR_CAST(sock);
 591         }
 592
 593         gs->sock = sock;
 594         gs->refcnt = 1;
 595         for (h = 0; h < VNI_HASH_SIZE; ++h)
 596                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 597
 598         /* Initialize the geneve udp offloads structure */
 599         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 600
 601         /* Mark socket as an encapsulation socket */
 602         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 603         tunnel_cfg.sk_user_data = gs;
 604         tunnel_cfg.encap_type = 1;
 605         tunnel_cfg.gro_receive = geneve_gro_receive;
 606         tunnel_cfg.gro_complete = geneve_gro_complete;
 607         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 608         tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
 609         tunnel_cfg.encap_destroy = NULL;
 610         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 611         list_add(&gs->list, &gn->sock_list);
 612         return gs;
 613 }
 614
 615 static void __geneve_sock_release(struct geneve_sock *gs)
 616 {
 617         if (!gs || --gs->refcnt)
 618                 return;
 619
 620         list_del(&gs->list);
 621         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 622         udp_tunnel_sock_release(gs->sock);
 623         kfree_rcu(gs, rcu);
 624 }
 625
 626 static void geneve_sock_release(struct geneve_dev *geneve)
 627 {
 628         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 629 #if IS_ENABLED(CONFIG_IPV6)
 630         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 631
 632         rcu_assign_pointer(geneve->sock6, NULL);
 633 #endif
 634
 635         rcu_assign_pointer(geneve->sock4, NULL);
 636         synchronize_net();
 637
 638         __geneve_sock_release(gs4);
 639 #if IS_ENABLED(CONFIG_IPV6)
 640         __geneve_sock_release(gs6);
 641 #endif
 642 }
 643
 644 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 645                                             sa_family_t family,
 646                                             __be16 dst_port)
 647 {
 648         struct geneve_sock *gs;
 649
 650         list_for_each_entry(gs, &gn->sock_list, list) {
 651                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 652                     geneve_get_sk_family(gs) == family) {
 653                         return gs;
 654                 }
 655         }
 656         return NULL;
 657 }
 658
 659 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 660 {
 661         struct net *net = geneve->net;
 662         struct geneve_net *gn = net_generic(net, geneve_net_id);
 663         struct geneve_dev_node *node;
 664         struct geneve_sock *gs;
 665         __u8 vni[3];
 666         __u32 hash;
 667
 668         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
 669         if (gs) {
 670                 gs->refcnt++;
 671                 goto out;
 672         }
 673
 674         gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
 675                                   geneve->cfg.use_udp6_rx_checksums);
 676         if (IS_ERR(gs))
 677                 return PTR_ERR(gs);
 678
 679 out:
 680         gs->collect_md = geneve->cfg.collect_md;
 681 #if IS_ENABLED(CONFIG_IPV6)
 682         if (ipv6) {
 683                 rcu_assign_pointer(geneve->sock6, gs);
 684                 node = &geneve->hlist6;
 685         } else
 686 #endif
 687         {
 688                 rcu_assign_pointer(geneve->sock4, gs);
 689                 node = &geneve->hlist4;
 690         }
 691         node->geneve = geneve;
 692
 693         tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
 694         hash = geneve_net_vni_hash(vni);
 695         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 696         return 0;
 697 }
 698
 699 static int geneve_open(struct net_device *dev)
 700 {
 701         struct geneve_dev *geneve = netdev_priv(dev);
 702         bool metadata = geneve->cfg.collect_md;
 703         bool ipv4, ipv6;
 704         int ret = 0;
 705
 706         ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 707         ipv4 = !ipv6 || metadata;
 708 #if IS_ENABLED(CONFIG_IPV6)
 709         if (ipv6) {
 710                 ret = geneve_sock_add(geneve, true);
 711                 if (ret < 0 && ret != -EAFNOSUPPORT)
 712                         ipv4 = false;
 713         }
 714 #endif
 715         if (ipv4)
 716                 ret = geneve_sock_add(geneve, false);
 717         if (ret < 0)
 718                 geneve_sock_release(geneve);
 719
 720         return ret;
 721 }
 722
 723 static int geneve_stop(struct net_device *dev)
 724 {
 725         struct geneve_dev *geneve = netdev_priv(dev);
 726
 727         hlist_del_init_rcu(&geneve->hlist4.hlist);
 728 #if IS_ENABLED(CONFIG_IPV6)
 729         hlist_del_init_rcu(&geneve->hlist6.hlist);
 730 #endif
 731         geneve_sock_release(geneve);
 732         return 0;
 733 }
 734
 735 static void geneve_build_header(struct genevehdr *geneveh,
 736                                 const struct ip_tunnel_info *info)
 737 {
 738         geneveh->ver = GENEVE_VER;
 739         geneveh->opt_len = info->options_len / 4;
 740         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 741         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 742         geneveh->rsvd1 = 0;
 743         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 744         geneveh->proto_type = htons(ETH_P_TEB);
 745         geneveh->rsvd2 = 0;
 746
 747         if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
 748                 ip_tunnel_info_opts_get(geneveh->options, info);
 749 }
 750
 751 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 752                             const struct ip_tunnel_info *info,
 753                             bool xnet, int ip_hdr_len)
 754 {
 755         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 756         struct genevehdr *gnvh;
 757         int min_headroom;
 758         int err;
 759
 760         skb_reset_mac_header(skb);
 761         skb_scrub_packet(skb, xnet);
 762
 763         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 764                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 765         err = skb_cow_head(skb, min_headroom);
 766         if (unlikely(err))
 767                 goto free_dst;
 768
 769         err = udp_tunnel_handle_offloads(skb, udp_sum);
 770         if (err)
 771                 goto free_dst;
 772
 773         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 774         geneve_build_header(gnvh, info);
 775         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 776         return 0;
 777
 778 free_dst:
 779         dst_release(dst);
 780         return err;
 781 }
 782
 783 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 784                                        struct net_device *dev,
 785                                        struct geneve_sock *gs4,
 786                                        struct flowi4 *fl4,
 787                                        const struct ip_tunnel_info *info,
 788                                        __be16 dport, __be16 sport)
 789 {
 790         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 791         struct geneve_dev *geneve = netdev_priv(dev);
 792         struct dst_cache *dst_cache;
 793         struct rtable *rt = NULL;
 794         __u8 tos;
 795
 796         if (!gs4)
 797                 return ERR_PTR(-EIO);
 798
 799         memset(fl4, 0, sizeof(*fl4));
 800         fl4->flowi4_mark = skb->mark;
 801         fl4->flowi4_proto = IPPROTO_UDP;
 802         fl4->daddr = info->key.u.ipv4.dst;
 803         fl4->saddr = info->key.u.ipv4.src;
 804         fl4->fl4_dport = dport;
 805         fl4->fl4_sport = sport;
 806
 807         tos = info->key.tos;
 808         if ((tos == 1) && !geneve->cfg.collect_md) {
 809                 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 810                 use_cache = false;
 811         }
 812         fl4->flowi4_tos = RT_TOS(tos);
 813
 814         dst_cache = (struct dst_cache *)&info->dst_cache;
 815         if (use_cache) {
 816                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 817                 if (rt)
 818                         return rt;
 819         }
 820         rt = ip_route_output_key(geneve->net, fl4);
 821         if (IS_ERR(rt)) {
 822                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 823                 return ERR_PTR(-ENETUNREACH);
 824         }
 825         if (rt->dst.dev == dev) { /* is this necessary? */
 826                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 827                 ip_rt_put(rt);
 828                 return ERR_PTR(-ELOOP);
 829         }
 830         if (use_cache)
 831                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 832         return rt;
 833 }
 834
 835 #if IS_ENABLED(CONFIG_IPV6)
 836 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 837                                            struct net_device *dev,
 838                                            struct geneve_sock *gs6,
 839                                            struct flowi6 *fl6,
 840                                            const struct ip_tunnel_info *info,
 841                                            __be16 dport, __be16 sport)
 842 {
 843         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 844         struct geneve_dev *geneve = netdev_priv(dev);
 845         struct dst_entry *dst = NULL;
 846         struct dst_cache *dst_cache;
 847         __u8 prio;
 848
 849         if (!gs6)
 850                 return ERR_PTR(-EIO);
 851
 852         memset(fl6, 0, sizeof(*fl6));
 853         fl6->flowi6_mark = skb->mark;
 854         fl6->flowi6_proto = IPPROTO_UDP;
 855         fl6->daddr = info->key.u.ipv6.dst;
 856         fl6->saddr = info->key.u.ipv6.src;
 857         fl6->fl6_dport = dport;
 858         fl6->fl6_sport = sport;
 859
 860         prio = info->key.tos;
 861         if ((prio == 1) && !geneve->cfg.collect_md) {
 862                 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 863                 use_cache = false;
 864         }
 865
 866         fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 867                                            info->key.label);
 868         dst_cache = (struct dst_cache *)&info->dst_cache;
 869         if (use_cache) {
 870                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 871                 if (dst)
 872                         return dst;
 873         }
 874         dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
 875                                               NULL);
 876         if (IS_ERR(dst)) {
 877                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 878                 return ERR_PTR(-ENETUNREACH);
 879         }
 880         if (dst->dev == dev) { /* is this necessary? */
 881                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 882                 dst_release(dst);
 883                 return ERR_PTR(-ELOOP);
 884         }
 885
 886         if (use_cache)
 887                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 888         return dst;
 889 }
 890 #endif
 891
 892 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 893                            struct geneve_dev *geneve,
 894                            const struct ip_tunnel_info *info)
 895 {
 896         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 897         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 898         const struct ip_tunnel_key *key = &info->key;
 899         struct rtable *rt;
 900         struct flowi4 fl4;
 901         __u8 tos, ttl;
 902         __be16 df = 0;
 903         __be16 sport;
 904         int err;
 905
 906         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 907         rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
 908                               geneve->cfg.info.key.tp_dst, sport);
 909         if (IS_ERR(rt))
 910                 return PTR_ERR(rt);
 911
 912         err = skb_tunnel_check_pmtu(skb, &rt->dst,
 913                                     GENEVE_IPV4_HLEN + info->options_len,
 914                                     netif_is_any_bridge_port(dev));
 915         if (err < 0) {
 916                 dst_release(&rt->dst);
 917                 return err;
 918         } else if (err) {
 919                 struct ip_tunnel_info *info;
 920
 921                 info = skb_tunnel_info(skb);
 922                 if (info) {
 923                         info->key.u.ipv4.dst = fl4.saddr;
 924                         info->key.u.ipv4.src = fl4.daddr;
 925                 }
 926
 927                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 928                         dst_release(&rt->dst);
 929                         return -EINVAL;
 930                 }
 931
 932                 skb->protocol = eth_type_trans(skb, geneve->dev);
 933                 netif_rx(skb);
 934                 dst_release(&rt->dst);
 935                 return -EMSGSIZE;
 936         }
 937
 938         if (geneve->cfg.collect_md) {
 939                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 940                 ttl = key->ttl;
 941
 942                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 943         } else {
 944                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 945                 if (geneve->cfg.ttl_inherit)
 946                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 947                 else
 948                         ttl = key->ttl;
 949                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 950
 951                 if (geneve->cfg.df == GENEVE_DF_SET) {
 952                         df = htons(IP_DF);
 953                 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
 954                         struct ethhdr *eth = eth_hdr(skb);
 955
 956                         if (ntohs(eth->h_proto) == ETH_P_IPV6) {
 957                                 df = htons(IP_DF);
 958                         } else if (ntohs(eth->h_proto) == ETH_P_IP) {
 959                                 struct iphdr *iph = ip_hdr(skb);
 960
 961                                 if (iph->frag_off & htons(IP_DF))
 962                                         df = htons(IP_DF);
 963                         }
 964                 }
 965         }
 966
 967         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
 968         if (unlikely(err))
 969                 return err;
 970
 971         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 972                             tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
 973                             !net_eq(geneve->net, dev_net(geneve->dev)),
 974                             !(info->key.tun_flags & TUNNEL_CSUM));
 975         return 0;
 976 }
 977
 978 #if IS_ENABLED(CONFIG_IPV6)
 979 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 980                             struct geneve_dev *geneve,
 981                             const struct ip_tunnel_info *info)
 982 {
 983         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 984         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 985         const struct ip_tunnel_key *key = &info->key;
 986         struct dst_entry *dst = NULL;
 987         struct flowi6 fl6;
 988         __u8 prio, ttl;
 989         __be16 sport;
 990         int err;
 991
 992         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 993         dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
 994                                 geneve->cfg.info.key.tp_dst, sport);
 995         if (IS_ERR(dst))
 996                 return PTR_ERR(dst);
 997
 998         err = skb_tunnel_check_pmtu(skb, dst,
 999                                     GENEVE_IPV6_HLEN + info->options_len,
1000                                     netif_is_any_bridge_port(dev));
1001         if (err < 0) {
1002                 dst_release(dst);
1003                 return err;
1004         } else if (err) {
1005                 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1006
1007                 if (info) {
1008                         info->key.u.ipv6.dst = fl6.saddr;
1009                         info->key.u.ipv6.src = fl6.daddr;
1010                 }
1011
1012                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1013                         dst_release(dst);
1014                         return -EINVAL;
1015                 }
1016
1017                 skb->protocol = eth_type_trans(skb, geneve->dev);
1018                 netif_rx(skb);
1019                 dst_release(dst);
1020                 return -EMSGSIZE;
1021         }
1022
1023         if (geneve->cfg.collect_md) {
1024                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1025                 ttl = key->ttl;
1026         } else {
1027                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1028                                            ip_hdr(skb), skb);
1029                 if (geneve->cfg.ttl_inherit)
1030                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1031                 else
1032                         ttl = key->ttl;
1033                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1034         }
1035         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
1036         if (unlikely(err))
1037                 return err;
1038
1039         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1040                              &fl6.saddr, &fl6.daddr, prio, ttl,
1041                              info->key.label, sport, geneve->cfg.info.key.tp_dst,
1042                              !(info->key.tun_flags & TUNNEL_CSUM));
1043         return 0;
1044 }
1045 #endif
1046
1047 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1048 {
1049         struct geneve_dev *geneve = netdev_priv(dev);
1050         struct ip_tunnel_info *info = NULL;
1051         int err;
1052
1053         if (geneve->cfg.collect_md) {
1054                 info = skb_tunnel_info(skb);
1055                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1056                         netdev_dbg(dev, "no tunnel metadata\n");
1057                         dev_kfree_skb(skb);
1058                         dev->stats.tx_dropped++;
1059                         return NETDEV_TX_OK;
1060                 }
1061         } else {
1062                 info = &geneve->cfg.info;
1063         }
1064
1065         rcu_read_lock();
1066 #if IS_ENABLED(CONFIG_IPV6)
1067         if (info->mode & IP_TUNNEL_INFO_IPV6)
1068                 err = geneve6_xmit_skb(skb, dev, geneve, info);
1069         else
1070 #endif
1071                 err = geneve_xmit_skb(skb, dev, geneve, info);
1072         rcu_read_unlock();
1073
1074         if (likely(!err))
1075                 return NETDEV_TX_OK;
1076
1077         if (err != -EMSGSIZE)
1078                 dev_kfree_skb(skb);
1079
1080         if (err == -ELOOP)
1081                 dev->stats.collisions++;
1082         else if (err == -ENETUNREACH)
1083                 dev->stats.tx_carrier_errors++;
1084
1085         dev->stats.tx_errors++;
1086         return NETDEV_TX_OK;
1087 }
1088
1089 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1090 {
1091         if (new_mtu > dev->max_mtu)
1092                 new_mtu = dev->max_mtu;
1093         else if (new_mtu < dev->min_mtu)
1094                 new_mtu = dev->min_mtu;
1095
1096         dev->mtu = new_mtu;
1097         return 0;
1098 }
1099
1100 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1101 {
1102         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1103         struct geneve_dev *geneve = netdev_priv(dev);
1104         __be16 sport;
1105
1106         if (ip_tunnel_info_af(info) == AF_INET) {
1107                 struct rtable *rt;
1108                 struct flowi4 fl4;
1109
1110                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1111                 sport = udp_flow_src_port(geneve->net, skb,
1112                                           1, USHRT_MAX, true);
1113
1114                 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1115                                       geneve->cfg.info.key.tp_dst, sport);
1116                 if (IS_ERR(rt))
1117                         return PTR_ERR(rt);
1118
1119                 ip_rt_put(rt);
1120                 info->key.u.ipv4.src = fl4.saddr;
1121 #if IS_ENABLED(CONFIG_IPV6)
1122         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1123                 struct dst_entry *dst;
1124                 struct flowi6 fl6;
1125
1126                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1127                 sport = udp_flow_src_port(geneve->net, skb,
1128                                           1, USHRT_MAX, true);
1129
1130                 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1131                                         geneve->cfg.info.key.tp_dst, sport);
1132                 if (IS_ERR(dst))
1133                         return PTR_ERR(dst);
1134
1135                 dst_release(dst);
1136                 info->key.u.ipv6.src = fl6.saddr;
1137 #endif
1138         } else {
1139                 return -EINVAL;
1140         }
1141
1142         info->key.tp_src = sport;
1143         info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1144         return 0;
1145 }
1146
1147 static const struct net_device_ops geneve_netdev_ops = {
1148         .ndo_init               = geneve_init,
1149         .ndo_uninit             = geneve_uninit,
1150         .ndo_open               = geneve_open,
1151         .ndo_stop               = geneve_stop,
1152         .ndo_start_xmit         = geneve_xmit,
1153         .ndo_get_stats64        = dev_get_tstats64,
1154         .ndo_change_mtu         = geneve_change_mtu,
1155         .ndo_validate_addr      = eth_validate_addr,
1156         .ndo_set_mac_address    = eth_mac_addr,
1157         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1158 };
1159
1160 static void geneve_get_drvinfo(struct net_device *dev,
1161                                struct ethtool_drvinfo *drvinfo)
1162 {
1163         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1164         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1165 }
1166
1167 static const struct ethtool_ops geneve_ethtool_ops = {
1168         .get_drvinfo    = geneve_get_drvinfo,
1169         .get_link       = ethtool_op_get_link,
1170 };
1171
1172 /* Info for udev, that this is a virtual tunnel endpoint */
1173 static struct device_type geneve_type = {
1174         .name = "geneve",
1175 };
1176
1177 /* Calls the ndo_udp_tunnel_add of the caller in order to
1178  * supply the listening GENEVE udp ports. Callers are expected
1179  * to implement the ndo_udp_tunnel_add.
1180  */
1181 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1182 {
1183         struct net *net = dev_net(dev);
1184         struct geneve_net *gn = net_generic(net, geneve_net_id);
1185         struct geneve_sock *gs;
1186
1187         rcu_read_lock();
1188         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1189                 if (push) {
1190                         udp_tunnel_push_rx_port(dev, gs->sock,
1191                                                 UDP_TUNNEL_TYPE_GENEVE);
1192                 } else {
1193                         udp_tunnel_drop_rx_port(dev, gs->sock,
1194                                                 UDP_TUNNEL_TYPE_GENEVE);
1195                 }
1196         }
1197         rcu_read_unlock();
1198 }
1199
1200 /* Initialize the device structure. */
1201 static void geneve_setup(struct net_device *dev)
1202 {
1203         ether_setup(dev);
1204
1205         dev->netdev_ops = &geneve_netdev_ops;
1206         dev->ethtool_ops = &geneve_ethtool_ops;
1207         dev->needs_free_netdev = true;
1208
1209         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1210
1211         dev->features    |= NETIF_F_LLTX;
1212         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1213         dev->features    |= NETIF_F_RXCSUM;
1214         dev->features    |= NETIF_F_GSO_SOFTWARE;
1215
1216         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1217         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1218
1219         /* MTU range: 68 - (something less than 65535) */
1220         dev->min_mtu = ETH_MIN_MTU;
1221         /* The max_mtu calculation does not take account of GENEVE
1222          * options, to avoid excluding potentially valid
1223          * configurations. This will be further reduced by IPvX hdr size.
1224          */
1225         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1226
1227         netif_keep_dst(dev);
1228         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1229         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1230         eth_hw_addr_random(dev);
1231 }
1232
1233 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1234         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1235         [IFLA_GENEVE_REMOTE]            = { .len = sizeof_field(struct iphdr, daddr) },
1236         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1237         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1238         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1239         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1240         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1241         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1242         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1243         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1244         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1245         [IFLA_GENEVE_TTL_INHERIT]       = { .type = NLA_U8 },
1246         [IFLA_GENEVE_DF]                = { .type = NLA_U8 },
1247 };
1248
1249 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1250                            struct netlink_ext_ack *extack)
1251 {
1252         if (tb[IFLA_ADDRESS]) {
1253                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1254                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1255                                             "Provided link layer address is not Ethernet");
1256                         return -EINVAL;
1257                 }
1258
1259                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1260                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1261                                             "Provided Ethernet address is not unicast");
1262                         return -EADDRNOTAVAIL;
1263                 }
1264         }
1265
1266         if (!data) {
1267                 NL_SET_ERR_MSG(extack,
1268                                "Not enough attributes provided to perform the operation");
1269                 return -EINVAL;
1270         }
1271
1272         if (data[IFLA_GENEVE_ID]) {
1273                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1274
1275                 if (vni >= GENEVE_N_VID) {
1276                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1277                                             "Geneve ID must be lower than 16777216");
1278                         return -ERANGE;
1279                 }
1280         }
1281
1282         if (data[IFLA_GENEVE_DF]) {
1283                 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1284
1285                 if (df < 0 || df > GENEVE_DF_MAX) {
1286                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1287                                             "Invalid DF attribute");
1288                         return -EINVAL;
1289                 }
1290         }
1291
1292         return 0;
1293 }
1294
1295 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1296                                           const struct ip_tunnel_info *info,
1297                                           bool *tun_on_same_port,
1298                                           bool *tun_collect_md)
1299 {
1300         struct geneve_dev *geneve, *t = NULL;
1301
1302         *tun_on_same_port = false;
1303         *tun_collect_md = false;
1304         list_for_each_entry(geneve, &gn->geneve_list, next) {
1305                 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1306                         *tun_collect_md = geneve->cfg.collect_md;
1307                         *tun_on_same_port = true;
1308                 }
1309                 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1310                     info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1311                     !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1312                         t = geneve;
1313         }
1314         return t;
1315 }
1316
1317 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1318 {
1319         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1320                  info->key.ttl || info->key.label || info->key.tp_src ||
1321                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1322 }
1323
1324 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1325                                   struct ip_tunnel_info *b)
1326 {
1327         if (ip_tunnel_info_af(a) == AF_INET)
1328                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1329         else
1330                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1331 }
1332
1333 static int geneve_configure(struct net *net, struct net_device *dev,
1334                             struct netlink_ext_ack *extack,
1335                             const struct geneve_config *cfg)
1336 {
1337         struct geneve_net *gn = net_generic(net, geneve_net_id);
1338         struct geneve_dev *t, *geneve = netdev_priv(dev);
1339         const struct ip_tunnel_info *info = &cfg->info;
1340         bool tun_collect_md, tun_on_same_port;
1341         int err, encap_len;
1342
1343         if (cfg->collect_md && !is_tnl_info_zero(info)) {
1344                 NL_SET_ERR_MSG(extack,
1345                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1346                 return -EINVAL;
1347         }
1348
1349         geneve->net = net;
1350         geneve->dev = dev;
1351
1352         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1353         if (t)
1354                 return -EBUSY;
1355
1356         /* make enough headroom for basic scenario */
1357         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1358         if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1359                 encap_len += sizeof(struct iphdr);
1360                 dev->max_mtu -= sizeof(struct iphdr);
1361         } else {
1362                 encap_len += sizeof(struct ipv6hdr);
1363                 dev->max_mtu -= sizeof(struct ipv6hdr);
1364         }
1365         dev->needed_headroom = encap_len + ETH_HLEN;
1366
1367         if (cfg->collect_md) {
1368                 if (tun_on_same_port) {
1369                         NL_SET_ERR_MSG(extack,
1370                                        "There can be only one externally controlled device on a destination port");
1371                         return -EPERM;
1372                 }
1373         } else {
1374                 if (tun_collect_md) {
1375                         NL_SET_ERR_MSG(extack,
1376                                        "There already exists an externally controlled device on this destination port");
1377                         return -EPERM;
1378                 }
1379         }
1380
1381         dst_cache_reset(&geneve->cfg.info.dst_cache);
1382         memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1383
1384         err = register_netdevice(dev);
1385         if (err)
1386                 return err;
1387
1388         list_add(&geneve->next, &gn->geneve_list);
1389         return 0;
1390 }
1391
1392 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1393 {
1394         memset(info, 0, sizeof(*info));
1395         info->key.tp_dst = htons(dst_port);
1396 }
1397
1398 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1399                           struct netlink_ext_ack *extack,
1400                           struct geneve_config *cfg, bool changelink)
1401 {
1402         struct ip_tunnel_info *info = &cfg->info;
1403         int attrtype;
1404
1405         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1406                 NL_SET_ERR_MSG(extack,
1407                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1408                 return -EINVAL;
1409         }
1410
1411         if (data[IFLA_GENEVE_REMOTE]) {
1412                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1413                         attrtype = IFLA_GENEVE_REMOTE;
1414                         goto change_notsup;
1415                 }
1416
1417                 info->key.u.ipv4.dst =
1418                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1419
1420                 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1421                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1422                                             "Remote IPv4 address cannot be Multicast");
1423                         return -EINVAL;
1424                 }
1425         }
1426
1427         if (data[IFLA_GENEVE_REMOTE6]) {
1428 #if IS_ENABLED(CONFIG_IPV6)
1429                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1430                         attrtype = IFLA_GENEVE_REMOTE6;
1431                         goto change_notsup;
1432                 }
1433
1434                 info->mode = IP_TUNNEL_INFO_IPV6;
1435                 info->key.u.ipv6.dst =
1436                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1437
1438                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1439                     IPV6_ADDR_LINKLOCAL) {
1440                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1441                                             "Remote IPv6 address cannot be link-local");
1442                         return -EINVAL;
1443                 }
1444                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1445                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1446                                             "Remote IPv6 address cannot be Multicast");
1447                         return -EINVAL;
1448                 }
1449                 info->key.tun_flags |= TUNNEL_CSUM;
1450                 cfg->use_udp6_rx_checksums = true;
1451 #else
1452                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1453                                     "IPv6 support not enabled in the kernel");
1454                 return -EPFNOSUPPORT;
1455 #endif
1456         }
1457
1458         if (data[IFLA_GENEVE_ID]) {
1459                 __u32 vni;
1460                 __u8 tvni[3];
1461                 __be64 tunid;
1462
1463                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1464                 tvni[0] = (vni & 0x00ff0000) >> 16;
1465                 tvni[1] = (vni & 0x0000ff00) >> 8;
1466                 tvni[2] =  vni & 0x000000ff;
1467
1468                 tunid = vni_to_tunnel_id(tvni);
1469                 if (changelink && (tunid != info->key.tun_id)) {
1470                         attrtype = IFLA_GENEVE_ID;
1471                         goto change_notsup;
1472                 }
1473                 info->key.tun_id = tunid;
1474         }
1475
1476         if (data[IFLA_GENEVE_TTL_INHERIT]) {
1477                 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1478                         cfg->ttl_inherit = true;
1479                 else
1480                         cfg->ttl_inherit = false;
1481         } else if (data[IFLA_GENEVE_TTL]) {
1482                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1483                 cfg->ttl_inherit = false;
1484         }
1485
1486         if (data[IFLA_GENEVE_TOS])
1487                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1488
1489         if (data[IFLA_GENEVE_DF])
1490                 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1491
1492         if (data[IFLA_GENEVE_LABEL]) {
1493                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1494                                   IPV6_FLOWLABEL_MASK;
1495                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1496                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1497                                             "Label attribute only applies for IPv6 Geneve devices");
1498                         return -EINVAL;
1499                 }
1500         }
1501
1502         if (data[IFLA_GENEVE_PORT]) {
1503                 if (changelink) {
1504                         attrtype = IFLA_GENEVE_PORT;
1505                         goto change_notsup;
1506                 }
1507                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1508         }
1509
1510         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1511                 if (changelink) {
1512                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1513                         goto change_notsup;
1514                 }
1515                 cfg->collect_md = true;
1516         }
1517
1518         if (data[IFLA_GENEVE_UDP_CSUM]) {
1519                 if (changelink) {
1520                         attrtype = IFLA_GENEVE_UDP_CSUM;
1521                         goto change_notsup;
1522                 }
1523                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1524                         info->key.tun_flags |= TUNNEL_CSUM;
1525         }
1526
1527         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1528 #if IS_ENABLED(CONFIG_IPV6)
1529                 if (changelink) {
1530                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1531                         goto change_notsup;
1532                 }
1533                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1534                         info->key.tun_flags &= ~TUNNEL_CSUM;
1535 #else
1536                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1537                                     "IPv6 support not enabled in the kernel");
1538                 return -EPFNOSUPPORT;
1539 #endif
1540         }
1541
1542         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1543 #if IS_ENABLED(CONFIG_IPV6)
1544                 if (changelink) {
1545                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1546                         goto change_notsup;
1547                 }
1548                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1549                         cfg->use_udp6_rx_checksums = false;
1550 #else
1551                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1552                                     "IPv6 support not enabled in the kernel");
1553                 return -EPFNOSUPPORT;
1554 #endif
1555         }
1556
1557         return 0;
1558 change_notsup:
1559         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1560                             "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1561         return -EOPNOTSUPP;
1562 }
1563
1564 static void geneve_link_config(struct net_device *dev,
1565                                struct ip_tunnel_info *info, struct nlattr *tb[])
1566 {
1567         struct geneve_dev *geneve = netdev_priv(dev);
1568         int ldev_mtu = 0;
1569
1570         if (tb[IFLA_MTU]) {
1571                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1572                 return;
1573         }
1574
1575         switch (ip_tunnel_info_af(info)) {
1576         case AF_INET: {
1577                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1578                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1579
1580                 if (!IS_ERR(rt) && rt->dst.dev) {
1581                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1582                         ip_rt_put(rt);
1583                 }
1584                 break;
1585         }
1586 #if IS_ENABLED(CONFIG_IPV6)
1587         case AF_INET6: {
1588                 struct rt6_info *rt;
1589
1590                 if (!__in6_dev_get(dev))
1591                         break;
1592
1593                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1594                                 NULL, 0);
1595
1596                 if (rt && rt->dst.dev)
1597                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1598                 ip6_rt_put(rt);
1599                 break;
1600         }
1601 #endif
1602         }
1603
1604         if (ldev_mtu <= 0)
1605                 return;
1606
1607         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1608 }
1609
1610 static int geneve_newlink(struct net *net, struct net_device *dev,
1611                           struct nlattr *tb[], struct nlattr *data[],
1612                           struct netlink_ext_ack *extack)
1613 {
1614         struct geneve_config cfg = {
1615                 .df = GENEVE_DF_UNSET,
1616                 .use_udp6_rx_checksums = false,
1617                 .ttl_inherit = false,
1618                 .collect_md = false,
1619         };
1620         int err;
1621
1622         init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1623         err = geneve_nl2info(tb, data, extack, &cfg, false);
1624         if (err)
1625                 return err;
1626
1627         err = geneve_configure(net, dev, extack, &cfg);
1628         if (err)
1629                 return err;
1630
1631         geneve_link_config(dev, &cfg.info, tb);
1632
1633         return 0;
1634 }
1635
1636 /* Quiesces the geneve device data path for both TX and RX.
1637  *
1638  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1639  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1640  * to complete for the existing set of in-flight packets to be transmitted,
1641  * then we would have quiesced the transmit data path. All the future packets
1642  * will get dropped until we unquiesce the data path.
1643  *
1644  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1645  * if we set that to NULL under RCU and wait for synchronize_net() to
1646  * complete, then we would have quiesced the receive data path.
1647  */
1648 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1649                            struct geneve_sock **gs6)
1650 {
1651         *gs4 = rtnl_dereference(geneve->sock4);
1652         rcu_assign_pointer(geneve->sock4, NULL);
1653         if (*gs4)
1654                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1655 #if IS_ENABLED(CONFIG_IPV6)
1656         *gs6 = rtnl_dereference(geneve->sock6);
1657         rcu_assign_pointer(geneve->sock6, NULL);
1658         if (*gs6)
1659                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1660 #else
1661         *gs6 = NULL;
1662 #endif
1663         synchronize_net();
1664 }
1665
1666 /* Resumes the geneve device data path for both TX and RX. */
1667 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1668                              struct geneve_sock __maybe_unused *gs6)
1669 {
1670         rcu_assign_pointer(geneve->sock4, gs4);
1671         if (gs4)
1672                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1673 #if IS_ENABLED(CONFIG_IPV6)
1674         rcu_assign_pointer(geneve->sock6, gs6);
1675         if (gs6)
1676                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1677 #endif
1678         synchronize_net();
1679 }
1680
1681 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1682                              struct nlattr *data[],
1683                              struct netlink_ext_ack *extack)
1684 {
1685         struct geneve_dev *geneve = netdev_priv(dev);
1686         struct geneve_sock *gs4, *gs6;
1687         struct geneve_config cfg;
1688         int err;
1689
1690         /* If the geneve device is configured for metadata (or externally
1691          * controlled, for example, OVS), then nothing can be changed.
1692          */
1693         if (geneve->cfg.collect_md)
1694                 return -EOPNOTSUPP;
1695
1696         /* Start with the existing info. */
1697         memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1698         err = geneve_nl2info(tb, data, extack, &cfg, true);
1699         if (err)
1700                 return err;
1701
1702         if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1703                 dst_cache_reset(&cfg.info.dst_cache);
1704                 geneve_link_config(dev, &cfg.info, tb);
1705         }
1706
1707         geneve_quiesce(geneve, &gs4, &gs6);
1708         memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1709         geneve_unquiesce(geneve, gs4, gs6);
1710
1711         return 0;
1712 }
1713
1714 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1715 {
1716         struct geneve_dev *geneve = netdev_priv(dev);
1717
1718         list_del(&geneve->next);
1719         unregister_netdevice_queue(dev, head);
1720 }
1721
1722 static size_t geneve_get_size(const struct net_device *dev)
1723 {
1724         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1725                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1726                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1727                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1728                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_DF */
1729                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1730                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1731                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1732                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1733                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1734                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1735                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1736                 0;
1737 }
1738
1739 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1740 {
1741         struct geneve_dev *geneve = netdev_priv(dev);
1742         struct ip_tunnel_info *info = &geneve->cfg.info;
1743         bool ttl_inherit = geneve->cfg.ttl_inherit;
1744         bool metadata = geneve->cfg.collect_md;
1745         __u8 tmp_vni[3];
1746         __u32 vni;
1747
1748         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1749         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1750         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1751                 goto nla_put_failure;
1752
1753         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1754                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1755                                     info->key.u.ipv4.dst))
1756                         goto nla_put_failure;
1757                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1758                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1759                         goto nla_put_failure;
1760
1761 #if IS_ENABLED(CONFIG_IPV6)
1762         } else if (!metadata) {
1763                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1764                                      &info->key.u.ipv6.dst))
1765                         goto nla_put_failure;
1766                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1767                                !(info->key.tun_flags & TUNNEL_CSUM)))
1768                         goto nla_put_failure;
1769 #endif
1770         }
1771
1772         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1773             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1774             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1775                 goto nla_put_failure;
1776
1777         if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1778                 goto nla_put_failure;
1779
1780         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1781                 goto nla_put_failure;
1782
1783         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1784                 goto nla_put_failure;
1785
1786 #if IS_ENABLED(CONFIG_IPV6)
1787         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1788                        !geneve->cfg.use_udp6_rx_checksums))
1789                 goto nla_put_failure;
1790 #endif
1791
1792         if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1793                 goto nla_put_failure;
1794
1795         return 0;
1796
1797 nla_put_failure:
1798         return -EMSGSIZE;
1799 }
1800
1801 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1802         .kind           = "geneve",
1803         .maxtype        = IFLA_GENEVE_MAX,
1804         .policy         = geneve_policy,
1805         .priv_size      = sizeof(struct geneve_dev),
1806         .setup          = geneve_setup,
1807         .validate       = geneve_validate,
1808         .newlink        = geneve_newlink,
1809         .changelink     = geneve_changelink,
1810         .dellink        = geneve_dellink,
1811         .get_size       = geneve_get_size,
1812         .fill_info      = geneve_fill_info,
1813 };
1814
1815 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1816                                         u8 name_assign_type, u16 dst_port)
1817 {
1818         struct nlattr *tb[IFLA_MAX + 1];
1819         struct net_device *dev;
1820         LIST_HEAD(list_kill);
1821         int err;
1822         struct geneve_config cfg = {
1823                 .df = GENEVE_DF_UNSET,
1824                 .use_udp6_rx_checksums = true,
1825                 .ttl_inherit = false,
1826                 .collect_md = true,
1827         };
1828
1829         memset(tb, 0, sizeof(tb));
1830         dev = rtnl_create_link(net, name, name_assign_type,
1831                                &geneve_link_ops, tb, NULL);
1832         if (IS_ERR(dev))
1833                 return dev;
1834
1835         init_tnl_info(&cfg.info, dst_port);
1836         err = geneve_configure(net, dev, NULL, &cfg);
1837         if (err) {
1838                 free_netdev(dev);
1839                 return ERR_PTR(err);
1840         }
1841
1842         /* openvswitch users expect packet sizes to be unrestricted,
1843          * so set the largest MTU we can.
1844          */
1845         err = geneve_change_mtu(dev, IP_MAX_MTU);
1846         if (err)
1847                 goto err;
1848
1849         err = rtnl_configure_link(dev, NULL);
1850         if (err < 0)
1851                 goto err;
1852
1853         return dev;
1854 err:
1855         geneve_dellink(dev, &list_kill);
1856         unregister_netdevice_many(&list_kill);
1857         return ERR_PTR(err);
1858 }
1859 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1860
1861 static int geneve_netdevice_event(struct notifier_block *unused,
1862                                   unsigned long event, void *ptr)
1863 {
1864         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1865
1866         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1867             event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1868                 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1869         } else if (event == NETDEV_UNREGISTER) {
1870                 if (!dev->udp_tunnel_nic_info)
1871                         geneve_offload_rx_ports(dev, false);
1872         } else if (event == NETDEV_REGISTER) {
1873                 if (!dev->udp_tunnel_nic_info)
1874                         geneve_offload_rx_ports(dev, true);
1875         }
1876
1877         return NOTIFY_DONE;
1878 }
1879
1880 static struct notifier_block geneve_notifier_block __read_mostly = {
1881         .notifier_call = geneve_netdevice_event,
1882 };
1883
1884 static __net_init int geneve_init_net(struct net *net)
1885 {
1886         struct geneve_net *gn = net_generic(net, geneve_net_id);
1887
1888         INIT_LIST_HEAD(&gn->geneve_list);
1889         INIT_LIST_HEAD(&gn->sock_list);
1890         return 0;
1891 }
1892
1893 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1894 {
1895         struct geneve_net *gn = net_generic(net, geneve_net_id);
1896         struct geneve_dev *geneve, *next;
1897         struct net_device *dev, *aux;
1898
1899         /* gather any geneve devices that were moved into this ns */
1900         for_each_netdev_safe(net, dev, aux)
1901                 if (dev->rtnl_link_ops == &geneve_link_ops)
1902                         unregister_netdevice_queue(dev, head);
1903
1904         /* now gather any other geneve devices that were created in this ns */
1905         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1906                 /* If geneve->dev is in the same netns, it was already added
1907                  * to the list by the previous loop.
1908                  */
1909                 if (!net_eq(dev_net(geneve->dev), net))
1910                         unregister_netdevice_queue(geneve->dev, head);
1911         }
1912 }
1913
1914 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1915 {
1916         struct net *net;
1917         LIST_HEAD(list);
1918
1919         rtnl_lock();
1920         list_for_each_entry(net, net_list, exit_list)
1921                 geneve_destroy_tunnels(net, &list);
1922
1923         /* unregister the devices gathered above */
1924         unregister_netdevice_many(&list);
1925         rtnl_unlock();
1926
1927         list_for_each_entry(net, net_list, exit_list) {
1928                 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1929
1930                 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1931         }
1932 }
1933
1934 static struct pernet_operations geneve_net_ops = {
1935         .init = geneve_init_net,
1936         .exit_batch = geneve_exit_batch_net,
1937         .id   = &geneve_net_id,
1938         .size = sizeof(struct geneve_net),
1939 };
1940
1941 static int __init geneve_init_module(void)
1942 {
1943         int rc;
1944
1945         rc = register_pernet_subsys(&geneve_net_ops);
1946         if (rc)
1947                 goto out1;
1948
1949         rc = register_netdevice_notifier(&geneve_notifier_block);
1950         if (rc)
1951                 goto out2;
1952
1953         rc = rtnl_link_register(&geneve_link_ops);
1954         if (rc)
1955                 goto out3;
1956
1957         return 0;
1958 out3:
1959         unregister_netdevice_notifier(&geneve_notifier_block);
1960 out2:
1961         unregister_pernet_subsys(&geneve_net_ops);
1962 out1:
1963         return rc;
1964 }
1965 late_initcall(geneve_init_module);
1966
1967 static void __exit geneve_cleanup_module(void)
1968 {
1969         rtnl_link_unregister(&geneve_link_ops);
1970         unregister_netdevice_notifier(&geneve_notifier_block);
1971         unregister_pernet_subsys(&geneve_net_ops);
1972 }
1973 module_exit(geneve_cleanup_module);
1974
1975 MODULE_LICENSE("GPL");
1976 MODULE_VERSION(GENEVE_NETDEV_VER);
1977 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1978 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1979 MODULE_ALIAS_RTNL_LINK("geneve");