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/kernel.h>
  11 #include <linux/module.h>
  12 #include <linux/etherdevice.h>
  13 #include <linux/hash.h>
  14 #include <net/ipv6_stubs.h>
  15 #include <net/dst_metadata.h>
  16 #include <net/gro_cells.h>
  17 #include <net/rtnetlink.h>
  18 #include <net/geneve.h>
  19 #include <net/protocol.h>
  20
  21 #define GENEVE_NETDEV_VER       "0.6"
  22
  23 #define GENEVE_N_VID            (1u << 24)
  24 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  25
  26 #define VNI_HASH_BITS           10
  27 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  28
  29 static bool log_ecn_error = true;
  30 module_param(log_ecn_error, bool, 0644);
  31 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  32
  33 #define GENEVE_VER 0
  34 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  35 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
  36 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
  37
  38 /* per-network namespace private data for this module */
  39 struct geneve_net {
  40         struct list_head        geneve_list;
  41         struct list_head        sock_list;
  42 };
  43
  44 static unsigned int geneve_net_id;
  45
  46 struct geneve_dev_node {
  47         struct hlist_node hlist;
  48         struct geneve_dev *geneve;
  49 };
  50
  51 struct geneve_config {
  52         struct ip_tunnel_info   info;
  53         bool                    collect_md;
  54         bool                    use_udp6_rx_checksums;
  55         bool                    ttl_inherit;
  56         enum ifla_geneve_df     df;
  57 };
  58
  59 /* Pseudo network device */
  60 struct geneve_dev {
  61         struct geneve_dev_node hlist4;  /* vni hash table for IPv4 socket */
  62 #if IS_ENABLED(CONFIG_IPV6)
  63         struct geneve_dev_node hlist6;  /* vni hash table for IPv6 socket */
  64 #endif
  65         struct net         *net;        /* netns for packet i/o */
  66         struct net_device  *dev;        /* netdev for geneve tunnel */
  67         struct geneve_sock __rcu *sock4;        /* IPv4 socket used for geneve tunnel */
  68 #if IS_ENABLED(CONFIG_IPV6)
  69         struct geneve_sock __rcu *sock6;        /* IPv6 socket used for geneve tunnel */
  70 #endif
  71         struct list_head   next;        /* geneve's per namespace list */
  72         struct gro_cells   gro_cells;
  73         struct geneve_config cfg;
  74 };
  75
  76 struct geneve_sock {
  77         bool                    collect_md;
  78         struct list_head        list;
  79         struct socket           *sock;
  80         struct rcu_head         rcu;
  81         int                     refcnt;
  82         struct hlist_head       vni_list[VNI_HASH_SIZE];
  83 };
  84
  85 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  86 {
  87         __u32 vnid;
  88
  89         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  90         return hash_32(vnid, VNI_HASH_BITS);
  91 }
  92
  93 static __be64 vni_to_tunnel_id(const __u8 *vni)
  94 {
  95 #ifdef __BIG_ENDIAN
  96         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
  97 #else
  98         return (__force __be64)(((__force u64)vni[0] << 40) |
  99                                 ((__force u64)vni[1] << 48) |
 100                                 ((__force u64)vni[2] << 56));
 101 #endif
 102 }
 103
 104 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 105 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 106 {
 107 #ifdef __BIG_ENDIAN
 108         vni[0] = (__force __u8)(tun_id >> 16);
 109         vni[1] = (__force __u8)(tun_id >> 8);
 110         vni[2] = (__force __u8)tun_id;
 111 #else
 112         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 113         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 114         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 115 #endif
 116 }
 117
 118 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
 119 {
 120         return !memcmp(vni, &tun_id[5], 3);
 121 }
 122
 123 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 124 {
 125         return gs->sock->sk->sk_family;
 126 }
 127
 128 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 129                                         __be32 addr, u8 vni[])
 130 {
 131         struct hlist_head *vni_list_head;
 132         struct geneve_dev_node *node;
 133         __u32 hash;
 134
 135         /* Find the device for this VNI */
 136         hash = geneve_net_vni_hash(vni);
 137         vni_list_head = &gs->vni_list[hash];
 138         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 139                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 140                     addr == node->geneve->cfg.info.key.u.ipv4.dst)
 141                         return node->geneve;
 142         }
 143         return NULL;
 144 }
 145
 146 #if IS_ENABLED(CONFIG_IPV6)
 147 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 148                                          struct in6_addr addr6, u8 vni[])
 149 {
 150         struct hlist_head *vni_list_head;
 151         struct geneve_dev_node *node;
 152         __u32 hash;
 153
 154         /* Find the device for this VNI */
 155         hash = geneve_net_vni_hash(vni);
 156         vni_list_head = &gs->vni_list[hash];
 157         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 158                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 159                     ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
 160                         return node->geneve;
 161         }
 162         return NULL;
 163 }
 164 #endif
 165
 166 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 167 {
 168         return (struct genevehdr *)(udp_hdr(skb) + 1);
 169 }
 170
 171 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 172                                             struct sk_buff *skb)
 173 {
 174         static u8 zero_vni[3];
 175         u8 *vni;
 176
 177         if (geneve_get_sk_family(gs) == AF_INET) {
 178                 struct iphdr *iph;
 179                 __be32 addr;
 180
 181                 iph = ip_hdr(skb); /* outer IP header... */
 182
 183                 if (gs->collect_md) {
 184                         vni = zero_vni;
 185                         addr = 0;
 186                 } else {
 187                         vni = geneve_hdr(skb)->vni;
 188                         addr = iph->saddr;
 189                 }
 190
 191                 return geneve_lookup(gs, addr, vni);
 192 #if IS_ENABLED(CONFIG_IPV6)
 193         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 194                 static struct in6_addr zero_addr6;
 195                 struct ipv6hdr *ip6h;
 196                 struct in6_addr addr6;
 197
 198                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 199
 200                 if (gs->collect_md) {
 201                         vni = zero_vni;
 202                         addr6 = zero_addr6;
 203                 } else {
 204                         vni = geneve_hdr(skb)->vni;
 205                         addr6 = ip6h->saddr;
 206                 }
 207
 208                 return geneve6_lookup(gs, addr6, vni);
 209 #endif
 210         }
 211         return NULL;
 212 }
 213
 214 /* geneve receive/decap routine */
 215 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 216                       struct sk_buff *skb)
 217 {
 218         struct genevehdr *gnvh = geneve_hdr(skb);
 219         struct metadata_dst *tun_dst = NULL;
 220         unsigned int len;
 221         int err = 0;
 222         void *oiph;
 223
 224         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 225                 __be16 flags;
 226
 227                 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
 228                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 229
 230                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 231                                          vni_to_tunnel_id(gnvh->vni),
 232                                          gnvh->opt_len * 4);
 233                 if (!tun_dst) {
 234                         geneve->dev->stats.rx_dropped++;
 235                         goto drop;
 236                 }
 237                 /* Update tunnel dst according to Geneve options. */
 238                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 239                                         gnvh->options, gnvh->opt_len * 4,
 240                                         TUNNEL_GENEVE_OPT);
 241         } else {
 242                 /* Drop packets w/ critical options,
 243                  * since we don't support any...
 244                  */
 245                 if (gnvh->critical) {
 246                         geneve->dev->stats.rx_frame_errors++;
 247                         geneve->dev->stats.rx_errors++;
 248                         goto drop;
 249                 }
 250         }
 251
 252         skb_reset_mac_header(skb);
 253         skb->protocol = eth_type_trans(skb, geneve->dev);
 254         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 255
 256         if (tun_dst)
 257                 skb_dst_set(skb, &tun_dst->dst);
 258
 259         /* Ignore packet loops (and multicast echo) */
 260         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
 261                 goto rx_error;
 262
 263         switch (skb_protocol(skb, true)) {
 264         case htons(ETH_P_IP):
 265                 if (pskb_may_pull(skb, sizeof(struct iphdr)))
 266                         goto rx_error;
 267                 break;
 268         case htons(ETH_P_IPV6):
 269                 if (pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 270                         goto rx_error;
 271                 break;
 272         default:
 273                 goto rx_error;
 274         }
 275         oiph = skb_network_header(skb);
 276         skb_reset_network_header(skb);
 277
 278         if (geneve_get_sk_family(gs) == AF_INET)
 279                 err = IP_ECN_decapsulate(oiph, skb);
 280 #if IS_ENABLED(CONFIG_IPV6)
 281         else
 282                 err = IP6_ECN_decapsulate(oiph, skb);
 283 #endif
 284
 285         if (unlikely(err)) {
 286                 if (log_ecn_error) {
 287                         if (geneve_get_sk_family(gs) == AF_INET)
 288                                 net_info_ratelimited("non-ECT from %pI4 "
 289                                                      "with TOS=%#x\n",
 290                                                      &((struct iphdr *)oiph)->saddr,
 291                                                      ((struct iphdr *)oiph)->tos);
 292 #if IS_ENABLED(CONFIG_IPV6)
 293                         else
 294                                 net_info_ratelimited("non-ECT from %pI6\n",
 295                                                      &((struct ipv6hdr *)oiph)->saddr);
 296 #endif
 297                 }
 298                 if (err > 1) {
 299                         ++geneve->dev->stats.rx_frame_errors;
 300                         ++geneve->dev->stats.rx_errors;
 301                         goto drop;
 302                 }
 303         }
 304
 305         len = skb->len;
 306         err = gro_cells_receive(&geneve->gro_cells, skb);
 307         if (likely(err == NET_RX_SUCCESS))
 308                 dev_sw_netstats_rx_add(geneve->dev, len);
 309
 310         return;
 311 rx_error:
 312         geneve->dev->stats.rx_errors++;
 313 drop:
 314         /* Consume bad packet */
 315         kfree_skb(skb);
 316 }
 317
 318 /* Setup stats when device is created */
 319 static int geneve_init(struct net_device *dev)
 320 {
 321         struct geneve_dev *geneve = netdev_priv(dev);
 322         int err;
 323
 324         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 325         if (!dev->tstats)
 326                 return -ENOMEM;
 327
 328         err = gro_cells_init(&geneve->gro_cells, dev);
 329         if (err) {
 330                 free_percpu(dev->tstats);
 331                 return err;
 332         }
 333
 334         err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
 335         if (err) {
 336                 free_percpu(dev->tstats);
 337                 gro_cells_destroy(&geneve->gro_cells);
 338                 return err;
 339         }
 340         return 0;
 341 }
 342
 343 static void geneve_uninit(struct net_device *dev)
 344 {
 345         struct geneve_dev *geneve = netdev_priv(dev);
 346
 347         dst_cache_destroy(&geneve->cfg.info.dst_cache);
 348         gro_cells_destroy(&geneve->gro_cells);
 349         free_percpu(dev->tstats);
 350 }
 351
 352 /* Callback from net/ipv4/udp.c to receive packets */
 353 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 354 {
 355         struct genevehdr *geneveh;
 356         struct geneve_dev *geneve;
 357         struct geneve_sock *gs;
 358         int opts_len;
 359
 360         /* Need UDP and Geneve header to be present */
 361         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 362                 goto drop;
 363
 364         /* Return packets with reserved bits set */
 365         geneveh = geneve_hdr(skb);
 366         if (unlikely(geneveh->ver != GENEVE_VER))
 367                 goto drop;
 368
 369         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 370                 goto drop;
 371
 372         gs = rcu_dereference_sk_user_data(sk);
 373         if (!gs)
 374                 goto drop;
 375
 376         geneve = geneve_lookup_skb(gs, skb);
 377         if (!geneve)
 378                 goto drop;
 379
 380         opts_len = geneveh->opt_len * 4;
 381         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 382                                  htons(ETH_P_TEB),
 383                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 384                 geneve->dev->stats.rx_dropped++;
 385                 goto drop;
 386         }
 387
 388         geneve_rx(geneve, gs, skb);
 389         return 0;
 390
 391 drop:
 392         /* Consume bad packet */
 393         kfree_skb(skb);
 394         return 0;
 395 }
 396
 397 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
 398 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
 399 {
 400         struct genevehdr *geneveh;
 401         struct geneve_sock *gs;
 402         u8 zero_vni[3] = { 0 };
 403         u8 *vni = zero_vni;
 404
 405         if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
 406                 return -EINVAL;
 407
 408         geneveh = geneve_hdr(skb);
 409         if (geneveh->ver != GENEVE_VER)
 410                 return -EINVAL;
 411
 412         if (geneveh->proto_type != htons(ETH_P_TEB))
 413                 return -EINVAL;
 414
 415         gs = rcu_dereference_sk_user_data(sk);
 416         if (!gs)
 417                 return -ENOENT;
 418
 419         if (geneve_get_sk_family(gs) == AF_INET) {
 420                 struct iphdr *iph = ip_hdr(skb);
 421                 __be32 addr4 = 0;
 422
 423                 if (!gs->collect_md) {
 424                         vni = geneve_hdr(skb)->vni;
 425                         addr4 = iph->daddr;
 426                 }
 427
 428                 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
 429         }
 430
 431 #if IS_ENABLED(CONFIG_IPV6)
 432         if (geneve_get_sk_family(gs) == AF_INET6) {
 433                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 434                 struct in6_addr addr6;
 435
 436                 memset(&addr6, 0, sizeof(struct in6_addr));
 437
 438                 if (!gs->collect_md) {
 439                         vni = geneve_hdr(skb)->vni;
 440                         addr6 = ip6h->daddr;
 441                 }
 442
 443                 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
 444         }
 445 #endif
 446
 447         return -EPFNOSUPPORT;
 448 }
 449
 450 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 451                                          __be16 port, bool ipv6_rx_csum)
 452 {
 453         struct socket *sock;
 454         struct udp_port_cfg udp_conf;
 455         int err;
 456
 457         memset(&udp_conf, 0, sizeof(udp_conf));
 458
 459         if (ipv6) {
 460                 udp_conf.family = AF_INET6;
 461                 udp_conf.ipv6_v6only = 1;
 462                 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
 463         } else {
 464                 udp_conf.family = AF_INET;
 465                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 466         }
 467
 468         udp_conf.local_udp_port = port;
 469
 470         /* Open UDP socket */
 471         err = udp_sock_create(net, &udp_conf, &sock);
 472         if (err < 0)
 473                 return ERR_PTR(err);
 474
 475         return sock;
 476 }
 477
 478 static int geneve_hlen(struct genevehdr *gh)
 479 {
 480         return sizeof(*gh) + gh->opt_len * 4;
 481 }
 482
 483 static struct sk_buff *geneve_gro_receive(struct sock *sk,
 484                                           struct list_head *head,
 485                                           struct sk_buff *skb)
 486 {
 487         struct sk_buff *pp = NULL;
 488         struct sk_buff *p;
 489         struct genevehdr *gh, *gh2;
 490         unsigned int hlen, gh_len, off_gnv;
 491         const struct packet_offload *ptype;
 492         __be16 type;
 493         int flush = 1;
 494
 495         off_gnv = skb_gro_offset(skb);
 496         hlen = off_gnv + sizeof(*gh);
 497         gh = skb_gro_header_fast(skb, off_gnv);
 498         if (skb_gro_header_hard(skb, hlen)) {
 499                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 500                 if (unlikely(!gh))
 501                         goto out;
 502         }
 503
 504         if (gh->ver != GENEVE_VER || gh->oam)
 505                 goto out;
 506         gh_len = geneve_hlen(gh);
 507
 508         hlen = off_gnv + gh_len;
 509         if (skb_gro_header_hard(skb, hlen)) {
 510                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 511                 if (unlikely(!gh))
 512                         goto out;
 513         }
 514
 515         list_for_each_entry(p, head, list) {
 516                 if (!NAPI_GRO_CB(p)->same_flow)
 517                         continue;
 518
 519                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 520                 if (gh->opt_len != gh2->opt_len ||
 521                     memcmp(gh, gh2, gh_len)) {
 522                         NAPI_GRO_CB(p)->same_flow = 0;
 523                         continue;
 524                 }
 525         }
 526
 527         type = gh->proto_type;
 528
 529         rcu_read_lock();
 530         ptype = gro_find_receive_by_type(type);
 531         if (!ptype)
 532                 goto out_unlock;
 533
 534         skb_gro_pull(skb, gh_len);
 535         skb_gro_postpull_rcsum(skb, gh, gh_len);
 536         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 537         flush = 0;
 538
 539 out_unlock:
 540         rcu_read_unlock();
 541 out:
 542         skb_gro_flush_final(skb, pp, flush);
 543
 544         return pp;
 545 }
 546
 547 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 548                                int nhoff)
 549 {
 550         struct genevehdr *gh;
 551         struct packet_offload *ptype;
 552         __be16 type;
 553         int gh_len;
 554         int err = -ENOSYS;
 555
 556         gh = (struct genevehdr *)(skb->data + nhoff);
 557         gh_len = geneve_hlen(gh);
 558         type = gh->proto_type;
 559
 560         rcu_read_lock();
 561         ptype = gro_find_complete_by_type(type);
 562         if (ptype)
 563                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 564
 565         rcu_read_unlock();
 566
 567         skb_set_inner_mac_header(skb, nhoff + gh_len);
 568
 569         return err;
 570 }
 571
 572 /* Create new listen socket if needed */
 573 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 574                                                 bool ipv6, bool ipv6_rx_csum)
 575 {
 576         struct geneve_net *gn = net_generic(net, geneve_net_id);
 577         struct geneve_sock *gs;
 578         struct socket *sock;
 579         struct udp_tunnel_sock_cfg tunnel_cfg;
 580         int h;
 581
 582         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 583         if (!gs)
 584                 return ERR_PTR(-ENOMEM);
 585
 586         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 587         if (IS_ERR(sock)) {
 588                 kfree(gs);
 589                 return ERR_CAST(sock);
 590         }
 591
 592         gs->sock = sock;
 593         gs->refcnt = 1;
 594         for (h = 0; h < VNI_HASH_SIZE; ++h)
 595                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 596
 597         /* Initialize the geneve udp offloads structure */
 598         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 599
 600         /* Mark socket as an encapsulation socket */
 601         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 602         tunnel_cfg.sk_user_data = gs;
 603         tunnel_cfg.encap_type = 1;
 604         tunnel_cfg.gro_receive = geneve_gro_receive;
 605         tunnel_cfg.gro_complete = geneve_gro_complete;
 606         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 607         tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
 608         tunnel_cfg.encap_destroy = NULL;
 609         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 610         list_add(&gs->list, &gn->sock_list);
 611         return gs;
 612 }
 613
 614 static void __geneve_sock_release(struct geneve_sock *gs)
 615 {
 616         if (!gs || --gs->refcnt)
 617                 return;
 618
 619         list_del(&gs->list);
 620         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 621         udp_tunnel_sock_release(gs->sock);
 622         kfree_rcu(gs, rcu);
 623 }
 624
 625 static void geneve_sock_release(struct geneve_dev *geneve)
 626 {
 627         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 628 #if IS_ENABLED(CONFIG_IPV6)
 629         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 630
 631         rcu_assign_pointer(geneve->sock6, NULL);
 632 #endif
 633
 634         rcu_assign_pointer(geneve->sock4, NULL);
 635         synchronize_net();
 636
 637         __geneve_sock_release(gs4);
 638 #if IS_ENABLED(CONFIG_IPV6)
 639         __geneve_sock_release(gs6);
 640 #endif
 641 }
 642
 643 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 644                                             sa_family_t family,
 645                                             __be16 dst_port)
 646 {
 647         struct geneve_sock *gs;
 648
 649         list_for_each_entry(gs, &gn->sock_list, list) {
 650                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 651                     geneve_get_sk_family(gs) == family) {
 652                         return gs;
 653                 }
 654         }
 655         return NULL;
 656 }
 657
 658 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 659 {
 660         struct net *net = geneve->net;
 661         struct geneve_net *gn = net_generic(net, geneve_net_id);
 662         struct geneve_dev_node *node;
 663         struct geneve_sock *gs;
 664         __u8 vni[3];
 665         __u32 hash;
 666
 667         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
 668         if (gs) {
 669                 gs->refcnt++;
 670                 goto out;
 671         }
 672
 673         gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
 674                                   geneve->cfg.use_udp6_rx_checksums);
 675         if (IS_ERR(gs))
 676                 return PTR_ERR(gs);
 677
 678 out:
 679         gs->collect_md = geneve->cfg.collect_md;
 680 #if IS_ENABLED(CONFIG_IPV6)
 681         if (ipv6) {
 682                 rcu_assign_pointer(geneve->sock6, gs);
 683                 node = &geneve->hlist6;
 684         } else
 685 #endif
 686         {
 687                 rcu_assign_pointer(geneve->sock4, gs);
 688                 node = &geneve->hlist4;
 689         }
 690         node->geneve = geneve;
 691
 692         tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
 693         hash = geneve_net_vni_hash(vni);
 694         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 695         return 0;
 696 }
 697
 698 static int geneve_open(struct net_device *dev)
 699 {
 700         struct geneve_dev *geneve = netdev_priv(dev);
 701         bool metadata = geneve->cfg.collect_md;
 702         bool ipv4, ipv6;
 703         int ret = 0;
 704
 705         ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 706         ipv4 = !ipv6 || metadata;
 707 #if IS_ENABLED(CONFIG_IPV6)
 708         if (ipv6) {
 709                 ret = geneve_sock_add(geneve, true);
 710                 if (ret < 0 && ret != -EAFNOSUPPORT)
 711                         ipv4 = false;
 712         }
 713 #endif
 714         if (ipv4)
 715                 ret = geneve_sock_add(geneve, false);
 716         if (ret < 0)
 717                 geneve_sock_release(geneve);
 718
 719         return ret;
 720 }
 721
 722 static int geneve_stop(struct net_device *dev)
 723 {
 724         struct geneve_dev *geneve = netdev_priv(dev);
 725
 726         hlist_del_init_rcu(&geneve->hlist4.hlist);
 727 #if IS_ENABLED(CONFIG_IPV6)
 728         hlist_del_init_rcu(&geneve->hlist6.hlist);
 729 #endif
 730         geneve_sock_release(geneve);
 731         return 0;
 732 }
 733
 734 static void geneve_build_header(struct genevehdr *geneveh,
 735                                 const struct ip_tunnel_info *info)
 736 {
 737         geneveh->ver = GENEVE_VER;
 738         geneveh->opt_len = info->options_len / 4;
 739         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 740         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 741         geneveh->rsvd1 = 0;
 742         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 743         geneveh->proto_type = htons(ETH_P_TEB);
 744         geneveh->rsvd2 = 0;
 745
 746         if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
 747                 ip_tunnel_info_opts_get(geneveh->options, info);
 748 }
 749
 750 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 751                             const struct ip_tunnel_info *info,
 752                             bool xnet, int ip_hdr_len)
 753 {
 754         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 755         struct genevehdr *gnvh;
 756         int min_headroom;
 757         int err;
 758
 759         skb_reset_mac_header(skb);
 760         skb_scrub_packet(skb, xnet);
 761
 762         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 763                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 764         err = skb_cow_head(skb, min_headroom);
 765         if (unlikely(err))
 766                 goto free_dst;
 767
 768         err = udp_tunnel_handle_offloads(skb, udp_sum);
 769         if (err)
 770                 goto free_dst;
 771
 772         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 773         geneve_build_header(gnvh, info);
 774         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 775         return 0;
 776
 777 free_dst:
 778         dst_release(dst);
 779         return err;
 780 }
 781
 782 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 783                                        struct net_device *dev,
 784                                        struct geneve_sock *gs4,
 785                                        struct flowi4 *fl4,
 786                                        const struct ip_tunnel_info *info,
 787                                        __be16 dport, __be16 sport)
 788 {
 789         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 790         struct geneve_dev *geneve = netdev_priv(dev);
 791         struct dst_cache *dst_cache;
 792         struct rtable *rt = NULL;
 793         __u8 tos;
 794
 795         if (!gs4)
 796                 return ERR_PTR(-EIO);
 797
 798         memset(fl4, 0, sizeof(*fl4));
 799         fl4->flowi4_mark = skb->mark;
 800         fl4->flowi4_proto = IPPROTO_UDP;
 801         fl4->daddr = info->key.u.ipv4.dst;
 802         fl4->saddr = info->key.u.ipv4.src;
 803         fl4->fl4_dport = dport;
 804         fl4->fl4_sport = sport;
 805
 806         tos = info->key.tos;
 807         if ((tos == 1) && !geneve->cfg.collect_md) {
 808                 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 809                 use_cache = false;
 810         }
 811         fl4->flowi4_tos = RT_TOS(tos);
 812
 813         dst_cache = (struct dst_cache *)&info->dst_cache;
 814         if (use_cache) {
 815                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 816                 if (rt)
 817                         return rt;
 818         }
 819         rt = ip_route_output_key(geneve->net, fl4);
 820         if (IS_ERR(rt)) {
 821                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 822                 return ERR_PTR(-ENETUNREACH);
 823         }
 824         if (rt->dst.dev == dev) { /* is this necessary? */
 825                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 826                 ip_rt_put(rt);
 827                 return ERR_PTR(-ELOOP);
 828         }
 829         if (use_cache)
 830                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 831         return rt;
 832 }
 833
 834 #if IS_ENABLED(CONFIG_IPV6)
 835 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 836                                            struct net_device *dev,
 837                                            struct geneve_sock *gs6,
 838                                            struct flowi6 *fl6,
 839                                            const struct ip_tunnel_info *info,
 840                                            __be16 dport, __be16 sport)
 841 {
 842         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 843         struct geneve_dev *geneve = netdev_priv(dev);
 844         struct dst_entry *dst = NULL;
 845         struct dst_cache *dst_cache;
 846         __u8 prio;
 847
 848         if (!gs6)
 849                 return ERR_PTR(-EIO);
 850
 851         memset(fl6, 0, sizeof(*fl6));
 852         fl6->flowi6_mark = skb->mark;
 853         fl6->flowi6_proto = IPPROTO_UDP;
 854         fl6->daddr = info->key.u.ipv6.dst;
 855         fl6->saddr = info->key.u.ipv6.src;
 856         fl6->fl6_dport = dport;
 857         fl6->fl6_sport = sport;
 858
 859         prio = info->key.tos;
 860         if ((prio == 1) && !geneve->cfg.collect_md) {
 861                 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 862                 use_cache = false;
 863         }
 864
 865         fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 866                                            info->key.label);
 867         dst_cache = (struct dst_cache *)&info->dst_cache;
 868         if (use_cache) {
 869                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 870                 if (dst)
 871                         return dst;
 872         }
 873         dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
 874                                               NULL);
 875         if (IS_ERR(dst)) {
 876                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 877                 return ERR_PTR(-ENETUNREACH);
 878         }
 879         if (dst->dev == dev) { /* is this necessary? */
 880                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 881                 dst_release(dst);
 882                 return ERR_PTR(-ELOOP);
 883         }
 884
 885         if (use_cache)
 886                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 887         return dst;
 888 }
 889 #endif
 890
 891 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 892                            struct geneve_dev *geneve,
 893                            const struct ip_tunnel_info *info)
 894 {
 895         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 896         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 897         const struct ip_tunnel_key *key = &info->key;
 898         struct rtable *rt;
 899         struct flowi4 fl4;
 900         __u8 tos, ttl;
 901         __be16 df = 0;
 902         __be16 sport;
 903         int err;
 904
 905         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 906         rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
 907                               geneve->cfg.info.key.tp_dst, sport);
 908         if (IS_ERR(rt))
 909                 return PTR_ERR(rt);
 910
 911         err = skb_tunnel_check_pmtu(skb, &rt->dst,
 912                                     GENEVE_IPV4_HLEN + info->options_len,
 913                                     netif_is_any_bridge_port(dev));
 914         if (err < 0) {
 915                 dst_release(&rt->dst);
 916                 return err;
 917         } else if (err) {
 918                 struct ip_tunnel_info *info;
 919
 920                 info = skb_tunnel_info(skb);
 921                 if (info) {
 922                         info->key.u.ipv4.dst = fl4.saddr;
 923                         info->key.u.ipv4.src = fl4.daddr;
 924                 }
 925
 926                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 927                         dst_release(&rt->dst);
 928                         return -EINVAL;
 929                 }
 930
 931                 skb->protocol = eth_type_trans(skb, geneve->dev);
 932                 netif_rx(skb);
 933                 dst_release(&rt->dst);
 934                 return -EMSGSIZE;
 935         }
 936
 937         if (geneve->cfg.collect_md) {
 938                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 939                 ttl = key->ttl;
 940
 941                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 942         } else {
 943                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 944                 if (geneve->cfg.ttl_inherit)
 945                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 946                 else
 947                         ttl = key->ttl;
 948                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 949
 950                 if (geneve->cfg.df == GENEVE_DF_SET) {
 951                         df = htons(IP_DF);
 952                 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
 953                         struct ethhdr *eth = eth_hdr(skb);
 954
 955                         if (ntohs(eth->h_proto) == ETH_P_IPV6) {
 956                                 df = htons(IP_DF);
 957                         } else if (ntohs(eth->h_proto) == ETH_P_IP) {
 958                                 struct iphdr *iph = ip_hdr(skb);
 959
 960                                 if (iph->frag_off & htons(IP_DF))
 961                                         df = htons(IP_DF);
 962                         }
 963                 }
 964         }
 965
 966         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
 967         if (unlikely(err))
 968                 return err;
 969
 970         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 971                             tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
 972                             !net_eq(geneve->net, dev_net(geneve->dev)),
 973                             !(info->key.tun_flags & TUNNEL_CSUM));
 974         return 0;
 975 }
 976
 977 #if IS_ENABLED(CONFIG_IPV6)
 978 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 979                             struct geneve_dev *geneve,
 980                             const struct ip_tunnel_info *info)
 981 {
 982         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 983         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 984         const struct ip_tunnel_key *key = &info->key;
 985         struct dst_entry *dst = NULL;
 986         struct flowi6 fl6;
 987         __u8 prio, ttl;
 988         __be16 sport;
 989         int err;
 990
 991         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 992         dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
 993                                 geneve->cfg.info.key.tp_dst, sport);
 994         if (IS_ERR(dst))
 995                 return PTR_ERR(dst);
 996
 997         err = skb_tunnel_check_pmtu(skb, dst,
 998                                     GENEVE_IPV6_HLEN + info->options_len,
 999                                     netif_is_any_bridge_port(dev));
1000         if (err < 0) {
1001                 dst_release(dst);
1002                 return err;
1003         } else if (err) {
1004                 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1005
1006                 if (info) {
1007                         info->key.u.ipv6.dst = fl6.saddr;
1008                         info->key.u.ipv6.src = fl6.daddr;
1009                 }
1010
1011                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1012                         dst_release(dst);
1013                         return -EINVAL;
1014                 }
1015
1016                 skb->protocol = eth_type_trans(skb, geneve->dev);
1017                 netif_rx(skb);
1018                 dst_release(dst);
1019                 return -EMSGSIZE;
1020         }
1021
1022         if (geneve->cfg.collect_md) {
1023                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1024                 ttl = key->ttl;
1025         } else {
1026                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1027                                            ip_hdr(skb), skb);
1028                 if (geneve->cfg.ttl_inherit)
1029                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1030                 else
1031                         ttl = key->ttl;
1032                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1033         }
1034         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
1035         if (unlikely(err))
1036                 return err;
1037
1038         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1039                              &fl6.saddr, &fl6.daddr, prio, ttl,
1040                              info->key.label, sport, geneve->cfg.info.key.tp_dst,
1041                              !(info->key.tun_flags & TUNNEL_CSUM));
1042         return 0;
1043 }
1044 #endif
1045
1046 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1047 {
1048         struct geneve_dev *geneve = netdev_priv(dev);
1049         struct ip_tunnel_info *info = NULL;
1050         int err;
1051
1052         if (geneve->cfg.collect_md) {
1053                 info = skb_tunnel_info(skb);
1054                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1055                         netdev_dbg(dev, "no tunnel metadata\n");
1056                         dev_kfree_skb(skb);
1057                         dev->stats.tx_dropped++;
1058                         return NETDEV_TX_OK;
1059                 }
1060         } else {
1061                 info = &geneve->cfg.info;
1062         }
1063
1064         rcu_read_lock();
1065 #if IS_ENABLED(CONFIG_IPV6)
1066         if (info->mode & IP_TUNNEL_INFO_IPV6)
1067                 err = geneve6_xmit_skb(skb, dev, geneve, info);
1068         else
1069 #endif
1070                 err = geneve_xmit_skb(skb, dev, geneve, info);
1071         rcu_read_unlock();
1072
1073         if (likely(!err))
1074                 return NETDEV_TX_OK;
1075
1076         if (err != -EMSGSIZE)
1077                 dev_kfree_skb(skb);
1078
1079         if (err == -ELOOP)
1080                 dev->stats.collisions++;
1081         else if (err == -ENETUNREACH)
1082                 dev->stats.tx_carrier_errors++;
1083
1084         dev->stats.tx_errors++;
1085         return NETDEV_TX_OK;
1086 }
1087
1088 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1089 {
1090         if (new_mtu > dev->max_mtu)
1091                 new_mtu = dev->max_mtu;
1092         else if (new_mtu < dev->min_mtu)
1093                 new_mtu = dev->min_mtu;
1094
1095         dev->mtu = new_mtu;
1096         return 0;
1097 }
1098
1099 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1100 {
1101         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1102         struct geneve_dev *geneve = netdev_priv(dev);
1103         __be16 sport;
1104
1105         if (ip_tunnel_info_af(info) == AF_INET) {
1106                 struct rtable *rt;
1107                 struct flowi4 fl4;
1108
1109                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1110                 sport = udp_flow_src_port(geneve->net, skb,
1111                                           1, USHRT_MAX, true);
1112
1113                 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1114                                       geneve->cfg.info.key.tp_dst, sport);
1115                 if (IS_ERR(rt))
1116                         return PTR_ERR(rt);
1117
1118                 ip_rt_put(rt);
1119                 info->key.u.ipv4.src = fl4.saddr;
1120 #if IS_ENABLED(CONFIG_IPV6)
1121         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1122                 struct dst_entry *dst;
1123                 struct flowi6 fl6;
1124
1125                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1126                 sport = udp_flow_src_port(geneve->net, skb,
1127                                           1, USHRT_MAX, true);
1128
1129                 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1130                                         geneve->cfg.info.key.tp_dst, sport);
1131                 if (IS_ERR(dst))
1132                         return PTR_ERR(dst);
1133
1134                 dst_release(dst);
1135                 info->key.u.ipv6.src = fl6.saddr;
1136 #endif
1137         } else {
1138                 return -EINVAL;
1139         }
1140
1141         info->key.tp_src = sport;
1142         info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1143         return 0;
1144 }
1145
1146 static const struct net_device_ops geneve_netdev_ops = {
1147         .ndo_init               = geneve_init,
1148         .ndo_uninit             = geneve_uninit,
1149         .ndo_open               = geneve_open,
1150         .ndo_stop               = geneve_stop,
1151         .ndo_start_xmit         = geneve_xmit,
1152         .ndo_get_stats64        = ip_tunnel_get_stats64,
1153         .ndo_change_mtu         = geneve_change_mtu,
1154         .ndo_validate_addr      = eth_validate_addr,
1155         .ndo_set_mac_address    = eth_mac_addr,
1156         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1157 };
1158
1159 static void geneve_get_drvinfo(struct net_device *dev,
1160                                struct ethtool_drvinfo *drvinfo)
1161 {
1162         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1163         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1164 }
1165
1166 static const struct ethtool_ops geneve_ethtool_ops = {
1167         .get_drvinfo    = geneve_get_drvinfo,
1168         .get_link       = ethtool_op_get_link,
1169 };
1170
1171 /* Info for udev, that this is a virtual tunnel endpoint */
1172 static struct device_type geneve_type = {
1173         .name = "geneve",
1174 };
1175
1176 /* Calls the ndo_udp_tunnel_add of the caller in order to
1177  * supply the listening GENEVE udp ports. Callers are expected
1178  * to implement the ndo_udp_tunnel_add.
1179  */
1180 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1181 {
1182         struct net *net = dev_net(dev);
1183         struct geneve_net *gn = net_generic(net, geneve_net_id);
1184         struct geneve_sock *gs;
1185
1186         rcu_read_lock();
1187         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1188                 if (push) {
1189                         udp_tunnel_push_rx_port(dev, gs->sock,
1190                                                 UDP_TUNNEL_TYPE_GENEVE);
1191                 } else {
1192                         udp_tunnel_drop_rx_port(dev, gs->sock,
1193                                                 UDP_TUNNEL_TYPE_GENEVE);
1194                 }
1195         }
1196         rcu_read_unlock();
1197 }
1198
1199 /* Initialize the device structure. */
1200 static void geneve_setup(struct net_device *dev)
1201 {
1202         ether_setup(dev);
1203
1204         dev->netdev_ops = &geneve_netdev_ops;
1205         dev->ethtool_ops = &geneve_ethtool_ops;
1206         dev->needs_free_netdev = true;
1207
1208         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1209
1210         dev->features    |= NETIF_F_LLTX;
1211         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1212         dev->features    |= NETIF_F_RXCSUM;
1213         dev->features    |= NETIF_F_GSO_SOFTWARE;
1214
1215         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1216         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1217
1218         /* MTU range: 68 - (something less than 65535) */
1219         dev->min_mtu = ETH_MIN_MTU;
1220         /* The max_mtu calculation does not take account of GENEVE
1221          * options, to avoid excluding potentially valid
1222          * configurations. This will be further reduced by IPvX hdr size.
1223          */
1224         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1225
1226         netif_keep_dst(dev);
1227         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1228         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1229         eth_hw_addr_random(dev);
1230 }
1231
1232 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1233         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1234         [IFLA_GENEVE_REMOTE]            = { .len = sizeof_field(struct iphdr, daddr) },
1235         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1236         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1237         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1238         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1239         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1240         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1241         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1242         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1243         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1244         [IFLA_GENEVE_TTL_INHERIT]       = { .type = NLA_U8 },
1245         [IFLA_GENEVE_DF]                = { .type = NLA_U8 },
1246 };
1247
1248 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1249                            struct netlink_ext_ack *extack)
1250 {
1251         if (tb[IFLA_ADDRESS]) {
1252                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1253                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1254                                             "Provided link layer address is not Ethernet");
1255                         return -EINVAL;
1256                 }
1257
1258                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1259                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1260                                             "Provided Ethernet address is not unicast");
1261                         return -EADDRNOTAVAIL;
1262                 }
1263         }
1264
1265         if (!data) {
1266                 NL_SET_ERR_MSG(extack,
1267                                "Not enough attributes provided to perform the operation");
1268                 return -EINVAL;
1269         }
1270
1271         if (data[IFLA_GENEVE_ID]) {
1272                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1273
1274                 if (vni >= GENEVE_N_VID) {
1275                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1276                                             "Geneve ID must be lower than 16777216");
1277                         return -ERANGE;
1278                 }
1279         }
1280
1281         if (data[IFLA_GENEVE_DF]) {
1282                 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1283
1284                 if (df < 0 || df > GENEVE_DF_MAX) {
1285                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1286                                             "Invalid DF attribute");
1287                         return -EINVAL;
1288                 }
1289         }
1290
1291         return 0;
1292 }
1293
1294 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1295                                           const struct ip_tunnel_info *info,
1296                                           bool *tun_on_same_port,
1297                                           bool *tun_collect_md)
1298 {
1299         struct geneve_dev *geneve, *t = NULL;
1300
1301         *tun_on_same_port = false;
1302         *tun_collect_md = false;
1303         list_for_each_entry(geneve, &gn->geneve_list, next) {
1304                 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1305                         *tun_collect_md = geneve->cfg.collect_md;
1306                         *tun_on_same_port = true;
1307                 }
1308                 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1309                     info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1310                     !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1311                         t = geneve;
1312         }
1313         return t;
1314 }
1315
1316 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1317 {
1318         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1319                  info->key.ttl || info->key.label || info->key.tp_src ||
1320                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1321 }
1322
1323 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1324                                   struct ip_tunnel_info *b)
1325 {
1326         if (ip_tunnel_info_af(a) == AF_INET)
1327                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1328         else
1329                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1330 }
1331
1332 static int geneve_configure(struct net *net, struct net_device *dev,
1333                             struct netlink_ext_ack *extack,
1334                             const struct geneve_config *cfg)
1335 {
1336         struct geneve_net *gn = net_generic(net, geneve_net_id);
1337         struct geneve_dev *t, *geneve = netdev_priv(dev);
1338         const struct ip_tunnel_info *info = &cfg->info;
1339         bool tun_collect_md, tun_on_same_port;
1340         int err, encap_len;
1341
1342         if (cfg->collect_md && !is_tnl_info_zero(info)) {
1343                 NL_SET_ERR_MSG(extack,
1344                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1345                 return -EINVAL;
1346         }
1347
1348         geneve->net = net;
1349         geneve->dev = dev;
1350
1351         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1352         if (t)
1353                 return -EBUSY;
1354
1355         /* make enough headroom for basic scenario */
1356         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1357         if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1358                 encap_len += sizeof(struct iphdr);
1359                 dev->max_mtu -= sizeof(struct iphdr);
1360         } else {
1361                 encap_len += sizeof(struct ipv6hdr);
1362                 dev->max_mtu -= sizeof(struct ipv6hdr);
1363         }
1364         dev->needed_headroom = encap_len + ETH_HLEN;
1365
1366         if (cfg->collect_md) {
1367                 if (tun_on_same_port) {
1368                         NL_SET_ERR_MSG(extack,
1369                                        "There can be only one externally controlled device on a destination port");
1370                         return -EPERM;
1371                 }
1372         } else {
1373                 if (tun_collect_md) {
1374                         NL_SET_ERR_MSG(extack,
1375                                        "There already exists an externally controlled device on this destination port");
1376                         return -EPERM;
1377                 }
1378         }
1379
1380         dst_cache_reset(&geneve->cfg.info.dst_cache);
1381         memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1382
1383         err = register_netdevice(dev);
1384         if (err)
1385                 return err;
1386
1387         list_add(&geneve->next, &gn->geneve_list);
1388         return 0;
1389 }
1390
1391 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1392 {
1393         memset(info, 0, sizeof(*info));
1394         info->key.tp_dst = htons(dst_port);
1395 }
1396
1397 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1398                           struct netlink_ext_ack *extack,
1399                           struct geneve_config *cfg, bool changelink)
1400 {
1401         struct ip_tunnel_info *info = &cfg->info;
1402         int attrtype;
1403
1404         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1405                 NL_SET_ERR_MSG(extack,
1406                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1407                 return -EINVAL;
1408         }
1409
1410         if (data[IFLA_GENEVE_REMOTE]) {
1411                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1412                         attrtype = IFLA_GENEVE_REMOTE;
1413                         goto change_notsup;
1414                 }
1415
1416                 info->key.u.ipv4.dst =
1417                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1418
1419                 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1420                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1421                                             "Remote IPv4 address cannot be Multicast");
1422                         return -EINVAL;
1423                 }
1424         }
1425
1426         if (data[IFLA_GENEVE_REMOTE6]) {
1427 #if IS_ENABLED(CONFIG_IPV6)
1428                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1429                         attrtype = IFLA_GENEVE_REMOTE6;
1430                         goto change_notsup;
1431                 }
1432
1433                 info->mode = IP_TUNNEL_INFO_IPV6;
1434                 info->key.u.ipv6.dst =
1435                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1436
1437                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1438                     IPV6_ADDR_LINKLOCAL) {
1439                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1440                                             "Remote IPv6 address cannot be link-local");
1441                         return -EINVAL;
1442                 }
1443                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1444                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1445                                             "Remote IPv6 address cannot be Multicast");
1446                         return -EINVAL;
1447                 }
1448                 info->key.tun_flags |= TUNNEL_CSUM;
1449                 cfg->use_udp6_rx_checksums = true;
1450 #else
1451                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1452                                     "IPv6 support not enabled in the kernel");
1453                 return -EPFNOSUPPORT;
1454 #endif
1455         }
1456
1457         if (data[IFLA_GENEVE_ID]) {
1458                 __u32 vni;
1459                 __u8 tvni[3];
1460                 __be64 tunid;
1461
1462                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1463                 tvni[0] = (vni & 0x00ff0000) >> 16;
1464                 tvni[1] = (vni & 0x0000ff00) >> 8;
1465                 tvni[2] =  vni & 0x000000ff;
1466
1467                 tunid = vni_to_tunnel_id(tvni);
1468                 if (changelink && (tunid != info->key.tun_id)) {
1469                         attrtype = IFLA_GENEVE_ID;
1470                         goto change_notsup;
1471                 }
1472                 info->key.tun_id = tunid;
1473         }
1474
1475         if (data[IFLA_GENEVE_TTL_INHERIT]) {
1476                 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1477                         cfg->ttl_inherit = true;
1478                 else
1479                         cfg->ttl_inherit = false;
1480         } else if (data[IFLA_GENEVE_TTL]) {
1481                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1482                 cfg->ttl_inherit = false;
1483         }
1484
1485         if (data[IFLA_GENEVE_TOS])
1486                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1487
1488         if (data[IFLA_GENEVE_DF])
1489                 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1490
1491         if (data[IFLA_GENEVE_LABEL]) {
1492                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1493                                   IPV6_FLOWLABEL_MASK;
1494                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1495                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1496                                             "Label attribute only applies for IPv6 Geneve devices");
1497                         return -EINVAL;
1498                 }
1499         }
1500
1501         if (data[IFLA_GENEVE_PORT]) {
1502                 if (changelink) {
1503                         attrtype = IFLA_GENEVE_PORT;
1504                         goto change_notsup;
1505                 }
1506                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1507         }
1508
1509         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1510                 if (changelink) {
1511                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1512                         goto change_notsup;
1513                 }
1514                 cfg->collect_md = true;
1515         }
1516
1517         if (data[IFLA_GENEVE_UDP_CSUM]) {
1518                 if (changelink) {
1519                         attrtype = IFLA_GENEVE_UDP_CSUM;
1520                         goto change_notsup;
1521                 }
1522                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1523                         info->key.tun_flags |= TUNNEL_CSUM;
1524         }
1525
1526         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1527 #if IS_ENABLED(CONFIG_IPV6)
1528                 if (changelink) {
1529                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1530                         goto change_notsup;
1531                 }
1532                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1533                         info->key.tun_flags &= ~TUNNEL_CSUM;
1534 #else
1535                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1536                                     "IPv6 support not enabled in the kernel");
1537                 return -EPFNOSUPPORT;
1538 #endif
1539         }
1540
1541         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1542 #if IS_ENABLED(CONFIG_IPV6)
1543                 if (changelink) {
1544                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1545                         goto change_notsup;
1546                 }
1547                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1548                         cfg->use_udp6_rx_checksums = false;
1549 #else
1550                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1551                                     "IPv6 support not enabled in the kernel");
1552                 return -EPFNOSUPPORT;
1553 #endif
1554         }
1555
1556         return 0;
1557 change_notsup:
1558         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1559                             "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1560         return -EOPNOTSUPP;
1561 }
1562
1563 static void geneve_link_config(struct net_device *dev,
1564                                struct ip_tunnel_info *info, struct nlattr *tb[])
1565 {
1566         struct geneve_dev *geneve = netdev_priv(dev);
1567         int ldev_mtu = 0;
1568
1569         if (tb[IFLA_MTU]) {
1570                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1571                 return;
1572         }
1573
1574         switch (ip_tunnel_info_af(info)) {
1575         case AF_INET: {
1576                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1577                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1578
1579                 if (!IS_ERR(rt) && rt->dst.dev) {
1580                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1581                         ip_rt_put(rt);
1582                 }
1583                 break;
1584         }
1585 #if IS_ENABLED(CONFIG_IPV6)
1586         case AF_INET6: {
1587                 struct rt6_info *rt;
1588
1589                 if (!__in6_dev_get(dev))
1590                         break;
1591
1592                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1593                                 NULL, 0);
1594
1595                 if (rt && rt->dst.dev)
1596                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1597                 ip6_rt_put(rt);
1598                 break;
1599         }
1600 #endif
1601         }
1602
1603         if (ldev_mtu <= 0)
1604                 return;
1605
1606         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1607 }
1608
1609 static int geneve_newlink(struct net *net, struct net_device *dev,
1610                           struct nlattr *tb[], struct nlattr *data[],
1611                           struct netlink_ext_ack *extack)
1612 {
1613         struct geneve_config cfg = {
1614                 .df = GENEVE_DF_UNSET,
1615                 .use_udp6_rx_checksums = false,
1616                 .ttl_inherit = false,
1617                 .collect_md = false,
1618         };
1619         int err;
1620
1621         init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1622         err = geneve_nl2info(tb, data, extack, &cfg, false);
1623         if (err)
1624                 return err;
1625
1626         err = geneve_configure(net, dev, extack, &cfg);
1627         if (err)
1628                 return err;
1629
1630         geneve_link_config(dev, &cfg.info, tb);
1631
1632         return 0;
1633 }
1634
1635 /* Quiesces the geneve device data path for both TX and RX.
1636  *
1637  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1638  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1639  * to complete for the existing set of in-flight packets to be transmitted,
1640  * then we would have quiesced the transmit data path. All the future packets
1641  * will get dropped until we unquiesce the data path.
1642  *
1643  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1644  * if we set that to NULL under RCU and wait for synchronize_net() to
1645  * complete, then we would have quiesced the receive data path.
1646  */
1647 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1648                            struct geneve_sock **gs6)
1649 {
1650         *gs4 = rtnl_dereference(geneve->sock4);
1651         rcu_assign_pointer(geneve->sock4, NULL);
1652         if (*gs4)
1653                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1654 #if IS_ENABLED(CONFIG_IPV6)
1655         *gs6 = rtnl_dereference(geneve->sock6);
1656         rcu_assign_pointer(geneve->sock6, NULL);
1657         if (*gs6)
1658                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1659 #else
1660         *gs6 = NULL;
1661 #endif
1662         synchronize_net();
1663 }
1664
1665 /* Resumes the geneve device data path for both TX and RX. */
1666 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1667                              struct geneve_sock __maybe_unused *gs6)
1668 {
1669         rcu_assign_pointer(geneve->sock4, gs4);
1670         if (gs4)
1671                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1672 #if IS_ENABLED(CONFIG_IPV6)
1673         rcu_assign_pointer(geneve->sock6, gs6);
1674         if (gs6)
1675                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1676 #endif
1677         synchronize_net();
1678 }
1679
1680 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1681                              struct nlattr *data[],
1682                              struct netlink_ext_ack *extack)
1683 {
1684         struct geneve_dev *geneve = netdev_priv(dev);
1685         struct geneve_sock *gs4, *gs6;
1686         struct geneve_config cfg;
1687         int err;
1688
1689         /* If the geneve device is configured for metadata (or externally
1690          * controlled, for example, OVS), then nothing can be changed.
1691          */
1692         if (geneve->cfg.collect_md)
1693                 return -EOPNOTSUPP;
1694
1695         /* Start with the existing info. */
1696         memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1697         err = geneve_nl2info(tb, data, extack, &cfg, true);
1698         if (err)
1699                 return err;
1700
1701         if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1702                 dst_cache_reset(&cfg.info.dst_cache);
1703                 geneve_link_config(dev, &cfg.info, tb);
1704         }
1705
1706         geneve_quiesce(geneve, &gs4, &gs6);
1707         memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1708         geneve_unquiesce(geneve, gs4, gs6);
1709
1710         return 0;
1711 }
1712
1713 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1714 {
1715         struct geneve_dev *geneve = netdev_priv(dev);
1716
1717         list_del(&geneve->next);
1718         unregister_netdevice_queue(dev, head);
1719 }
1720
1721 static size_t geneve_get_size(const struct net_device *dev)
1722 {
1723         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1724                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1725                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1726                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1727                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_DF */
1728                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1729                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1730                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1731                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1732                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1733                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1734                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1735                 0;
1736 }
1737
1738 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1739 {
1740         struct geneve_dev *geneve = netdev_priv(dev);
1741         struct ip_tunnel_info *info = &geneve->cfg.info;
1742         bool ttl_inherit = geneve->cfg.ttl_inherit;
1743         bool metadata = geneve->cfg.collect_md;
1744         __u8 tmp_vni[3];
1745         __u32 vni;
1746
1747         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1748         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1749         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1750                 goto nla_put_failure;
1751
1752         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1753                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1754                                     info->key.u.ipv4.dst))
1755                         goto nla_put_failure;
1756                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1757                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1758                         goto nla_put_failure;
1759
1760 #if IS_ENABLED(CONFIG_IPV6)
1761         } else if (!metadata) {
1762                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1763                                      &info->key.u.ipv6.dst))
1764                         goto nla_put_failure;
1765                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1766                                !(info->key.tun_flags & TUNNEL_CSUM)))
1767                         goto nla_put_failure;
1768 #endif
1769         }
1770
1771         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1772             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1773             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1774                 goto nla_put_failure;
1775
1776         if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1777                 goto nla_put_failure;
1778
1779         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1780                 goto nla_put_failure;
1781
1782         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1783                 goto nla_put_failure;
1784
1785 #if IS_ENABLED(CONFIG_IPV6)
1786         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1787                        !geneve->cfg.use_udp6_rx_checksums))
1788                 goto nla_put_failure;
1789 #endif
1790
1791         if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1792                 goto nla_put_failure;
1793
1794         return 0;
1795
1796 nla_put_failure:
1797         return -EMSGSIZE;
1798 }
1799
1800 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1801         .kind           = "geneve",
1802         .maxtype        = IFLA_GENEVE_MAX,
1803         .policy         = geneve_policy,
1804         .priv_size      = sizeof(struct geneve_dev),
1805         .setup          = geneve_setup,
1806         .validate       = geneve_validate,
1807         .newlink        = geneve_newlink,
1808         .changelink     = geneve_changelink,
1809         .dellink        = geneve_dellink,
1810         .get_size       = geneve_get_size,
1811         .fill_info      = geneve_fill_info,
1812 };
1813
1814 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1815                                         u8 name_assign_type, u16 dst_port)
1816 {
1817         struct nlattr *tb[IFLA_MAX + 1];
1818         struct net_device *dev;
1819         LIST_HEAD(list_kill);
1820         int err;
1821         struct geneve_config cfg = {
1822                 .df = GENEVE_DF_UNSET,
1823                 .use_udp6_rx_checksums = true,
1824                 .ttl_inherit = false,
1825                 .collect_md = true,
1826         };
1827
1828         memset(tb, 0, sizeof(tb));
1829         dev = rtnl_create_link(net, name, name_assign_type,
1830                                &geneve_link_ops, tb, NULL);
1831         if (IS_ERR(dev))
1832                 return dev;
1833
1834         init_tnl_info(&cfg.info, dst_port);
1835         err = geneve_configure(net, dev, NULL, &cfg);
1836         if (err) {
1837                 free_netdev(dev);
1838                 return ERR_PTR(err);
1839         }
1840
1841         /* openvswitch users expect packet sizes to be unrestricted,
1842          * so set the largest MTU we can.
1843          */
1844         err = geneve_change_mtu(dev, IP_MAX_MTU);
1845         if (err)
1846                 goto err;
1847
1848         err = rtnl_configure_link(dev, NULL);
1849         if (err < 0)
1850                 goto err;
1851
1852         return dev;
1853 err:
1854         geneve_dellink(dev, &list_kill);
1855         unregister_netdevice_many(&list_kill);
1856         return ERR_PTR(err);
1857 }
1858 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1859
1860 static int geneve_netdevice_event(struct notifier_block *unused,
1861                                   unsigned long event, void *ptr)
1862 {
1863         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1864
1865         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1866             event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1867                 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1868         } else if (event == NETDEV_UNREGISTER) {
1869                 if (!dev->udp_tunnel_nic_info)
1870                         geneve_offload_rx_ports(dev, false);
1871         } else if (event == NETDEV_REGISTER) {
1872                 if (!dev->udp_tunnel_nic_info)
1873                         geneve_offload_rx_ports(dev, true);
1874         }
1875
1876         return NOTIFY_DONE;
1877 }
1878
1879 static struct notifier_block geneve_notifier_block __read_mostly = {
1880         .notifier_call = geneve_netdevice_event,
1881 };
1882
1883 static __net_init int geneve_init_net(struct net *net)
1884 {
1885         struct geneve_net *gn = net_generic(net, geneve_net_id);
1886
1887         INIT_LIST_HEAD(&gn->geneve_list);
1888         INIT_LIST_HEAD(&gn->sock_list);
1889         return 0;
1890 }
1891
1892 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1893 {
1894         struct geneve_net *gn = net_generic(net, geneve_net_id);
1895         struct geneve_dev *geneve, *next;
1896         struct net_device *dev, *aux;
1897
1898         /* gather any geneve devices that were moved into this ns */
1899         for_each_netdev_safe(net, dev, aux)
1900                 if (dev->rtnl_link_ops == &geneve_link_ops)
1901                         unregister_netdevice_queue(dev, head);
1902
1903         /* now gather any other geneve devices that were created in this ns */
1904         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1905                 /* If geneve->dev is in the same netns, it was already added
1906                  * to the list by the previous loop.
1907                  */
1908                 if (!net_eq(dev_net(geneve->dev), net))
1909                         unregister_netdevice_queue(geneve->dev, head);
1910         }
1911 }
1912
1913 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1914 {
1915         struct net *net;
1916         LIST_HEAD(list);
1917
1918         rtnl_lock();
1919         list_for_each_entry(net, net_list, exit_list)
1920                 geneve_destroy_tunnels(net, &list);
1921
1922         /* unregister the devices gathered above */
1923         unregister_netdevice_many(&list);
1924         rtnl_unlock();
1925
1926         list_for_each_entry(net, net_list, exit_list) {
1927                 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1928
1929                 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1930         }
1931 }
1932
1933 static struct pernet_operations geneve_net_ops = {
1934         .init = geneve_init_net,
1935         .exit_batch = geneve_exit_batch_net,
1936         .id   = &geneve_net_id,
1937         .size = sizeof(struct geneve_net),
1938 };
1939
1940 static int __init geneve_init_module(void)
1941 {
1942         int rc;
1943
1944         rc = register_pernet_subsys(&geneve_net_ops);
1945         if (rc)
1946                 goto out1;
1947
1948         rc = register_netdevice_notifier(&geneve_notifier_block);
1949         if (rc)
1950                 goto out2;
1951
1952         rc = rtnl_link_register(&geneve_link_ops);
1953         if (rc)
1954                 goto out3;
1955
1956         return 0;
1957 out3:
1958         unregister_netdevice_notifier(&geneve_notifier_block);
1959 out2:
1960         unregister_pernet_subsys(&geneve_net_ops);
1961 out1:
1962         return rc;
1963 }
1964 late_initcall(geneve_init_module);
1965
1966 static void __exit geneve_cleanup_module(void)
1967 {
1968         rtnl_link_unregister(&geneve_link_ops);
1969         unregister_netdevice_notifier(&geneve_notifier_block);
1970         unregister_pernet_subsys(&geneve_net_ops);
1971 }
1972 module_exit(geneve_cleanup_module);
1973
1974 MODULE_LICENSE("GPL");
1975 MODULE_VERSION(GENEVE_NETDEV_VER);
1976 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1977 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1978 MODULE_ALIAS_RTNL_LINK("geneve");