drivers/net/geneve.c

   1 /*
   2  * GENEVE: Generic Network Virtualization Encapsulation
   3  *
   4  * Copyright (c) 2015 Red Hat, Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/etherdevice.h>
  16 #include <linux/hash.h>
  17 #include <net/dst_metadata.h>
  18 #include <net/gro_cells.h>
  19 #include <net/rtnetlink.h>
  20 #include <net/geneve.h>
  21 #include <net/protocol.h>
  22
  23 #define GENEVE_NETDEV_VER       "0.6"
  24
  25 #define GENEVE_UDP_PORT         6081
  26
  27 #define GENEVE_N_VID            (1u << 24)
  28 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  29
  30 #define VNI_HASH_BITS           10
  31 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  32
  33 static bool log_ecn_error = true;
  34 module_param(log_ecn_error, bool, 0644);
  35 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  36
  37 #define GENEVE_VER 0
  38 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  39
  40 /* per-network namespace private data for this module */
  41 struct geneve_net {
  42         struct list_head        geneve_list;
  43         struct list_head        sock_list;
  44 };
  45
  46 static unsigned int geneve_net_id;
  47
  48 struct geneve_dev_node {
  49         struct hlist_node hlist;
  50         struct geneve_dev *geneve;
  51 };
  52
  53 /* Pseudo network device */
  54 struct geneve_dev {
  55         struct geneve_dev_node hlist4;  /* vni hash table for IPv4 socket */
  56 #if IS_ENABLED(CONFIG_IPV6)
  57         struct geneve_dev_node hlist6;  /* vni hash table for IPv6 socket */
  58 #endif
  59         struct net         *net;        /* netns for packet i/o */
  60         struct net_device  *dev;        /* netdev for geneve tunnel */
  61         struct ip_tunnel_info info;
  62         struct geneve_sock __rcu *sock4;        /* IPv4 socket used for geneve tunnel */
  63 #if IS_ENABLED(CONFIG_IPV6)
  64         struct geneve_sock __rcu *sock6;        /* IPv6 socket used for geneve tunnel */
  65 #endif
  66         struct list_head   next;        /* geneve's per namespace list */
  67         struct gro_cells   gro_cells;
  68         bool               collect_md;
  69         bool               use_udp6_rx_checksums;
  70 };
  71
  72 struct geneve_sock {
  73         bool                    collect_md;
  74         struct list_head        list;
  75         struct socket           *sock;
  76         struct rcu_head         rcu;
  77         int                     refcnt;
  78         struct hlist_head       vni_list[VNI_HASH_SIZE];
  79 };
  80
  81 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  82 {
  83         __u32 vnid;
  84
  85         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  86         return hash_32(vnid, VNI_HASH_BITS);
  87 }
  88
  89 static __be64 vni_to_tunnel_id(const __u8 *vni)
  90 {
  91 #ifdef __BIG_ENDIAN
  92         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
  93 #else
  94         return (__force __be64)(((__force u64)vni[0] << 40) |
  95                                 ((__force u64)vni[1] << 48) |
  96                                 ((__force u64)vni[2] << 56));
  97 #endif
  98 }
  99
 100 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 101 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 102 {
 103 #ifdef __BIG_ENDIAN
 104         vni[0] = (__force __u8)(tun_id >> 16);
 105         vni[1] = (__force __u8)(tun_id >> 8);
 106         vni[2] = (__force __u8)tun_id;
 107 #else
 108         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 109         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 110         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 111 #endif
 112 }
 113
 114 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
 115 {
 116         return !memcmp(vni, &tun_id[5], 3);
 117 }
 118
 119 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 120 {
 121         return gs->sock->sk->sk_family;
 122 }
 123
 124 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 125                                         __be32 addr, u8 vni[])
 126 {
 127         struct hlist_head *vni_list_head;
 128         struct geneve_dev_node *node;
 129         __u32 hash;
 130
 131         /* Find the device for this VNI */
 132         hash = geneve_net_vni_hash(vni);
 133         vni_list_head = &gs->vni_list[hash];
 134         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 135                 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
 136                     addr == node->geneve->info.key.u.ipv4.dst)
 137                         return node->geneve;
 138         }
 139         return NULL;
 140 }
 141
 142 #if IS_ENABLED(CONFIG_IPV6)
 143 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 144                                          struct in6_addr addr6, u8 vni[])
 145 {
 146         struct hlist_head *vni_list_head;
 147         struct geneve_dev_node *node;
 148         __u32 hash;
 149
 150         /* Find the device for this VNI */
 151         hash = geneve_net_vni_hash(vni);
 152         vni_list_head = &gs->vni_list[hash];
 153         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 154                 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
 155                     ipv6_addr_equal(&addr6, &node->geneve->info.key.u.ipv6.dst))
 156                         return node->geneve;
 157         }
 158         return NULL;
 159 }
 160 #endif
 161
 162 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 163 {
 164         return (struct genevehdr *)(udp_hdr(skb) + 1);
 165 }
 166
 167 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 168                                             struct sk_buff *skb)
 169 {
 170         static u8 zero_vni[3];
 171         u8 *vni;
 172
 173         if (geneve_get_sk_family(gs) == AF_INET) {
 174                 struct iphdr *iph;
 175                 __be32 addr;
 176
 177                 iph = ip_hdr(skb); /* outer IP header... */
 178
 179                 if (gs->collect_md) {
 180                         vni = zero_vni;
 181                         addr = 0;
 182                 } else {
 183                         vni = geneve_hdr(skb)->vni;
 184                         addr = iph->saddr;
 185                 }
 186
 187                 return geneve_lookup(gs, addr, vni);
 188 #if IS_ENABLED(CONFIG_IPV6)
 189         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 190                 static struct in6_addr zero_addr6;
 191                 struct ipv6hdr *ip6h;
 192                 struct in6_addr addr6;
 193
 194                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 195
 196                 if (gs->collect_md) {
 197                         vni = zero_vni;
 198                         addr6 = zero_addr6;
 199                 } else {
 200                         vni = geneve_hdr(skb)->vni;
 201                         addr6 = ip6h->saddr;
 202                 }
 203
 204                 return geneve6_lookup(gs, addr6, vni);
 205 #endif
 206         }
 207         return NULL;
 208 }
 209
 210 /* geneve receive/decap routine */
 211 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 212                       struct sk_buff *skb)
 213 {
 214         struct genevehdr *gnvh = geneve_hdr(skb);
 215         struct metadata_dst *tun_dst = NULL;
 216         struct pcpu_sw_netstats *stats;
 217         unsigned int len;
 218         int err = 0;
 219         void *oiph;
 220
 221         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 222                 __be16 flags;
 223
 224                 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
 225                         (gnvh->oam ? TUNNEL_OAM : 0) |
 226                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 227
 228                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 229                                          vni_to_tunnel_id(gnvh->vni),
 230                                          gnvh->opt_len * 4);
 231                 if (!tun_dst) {
 232                         geneve->dev->stats.rx_dropped++;
 233                         goto drop;
 234                 }
 235                 /* Update tunnel dst according to Geneve options. */
 236                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 237                                         gnvh->options, gnvh->opt_len * 4);
 238         } else {
 239                 /* Drop packets w/ critical options,
 240                  * since we don't support any...
 241                  */
 242                 if (gnvh->critical) {
 243                         geneve->dev->stats.rx_frame_errors++;
 244                         geneve->dev->stats.rx_errors++;
 245                         goto drop;
 246                 }
 247         }
 248
 249         skb_reset_mac_header(skb);
 250         skb->protocol = eth_type_trans(skb, geneve->dev);
 251         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 252
 253         if (tun_dst)
 254                 skb_dst_set(skb, &tun_dst->dst);
 255
 256         /* Ignore packet loops (and multicast echo) */
 257         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
 258                 geneve->dev->stats.rx_errors++;
 259                 goto drop;
 260         }
 261
 262         oiph = skb_network_header(skb);
 263         skb_reset_network_header(skb);
 264
 265         if (geneve_get_sk_family(gs) == AF_INET)
 266                 err = IP_ECN_decapsulate(oiph, skb);
 267 #if IS_ENABLED(CONFIG_IPV6)
 268         else
 269                 err = IP6_ECN_decapsulate(oiph, skb);
 270 #endif
 271
 272         if (unlikely(err)) {
 273                 if (log_ecn_error) {
 274                         if (geneve_get_sk_family(gs) == AF_INET)
 275                                 net_info_ratelimited("non-ECT from %pI4 "
 276                                                      "with TOS=%#x\n",
 277                                                      &((struct iphdr *)oiph)->saddr,
 278                                                      ((struct iphdr *)oiph)->tos);
 279 #if IS_ENABLED(CONFIG_IPV6)
 280                         else
 281                                 net_info_ratelimited("non-ECT from %pI6\n",
 282                                                      &((struct ipv6hdr *)oiph)->saddr);
 283 #endif
 284                 }
 285                 if (err > 1) {
 286                         ++geneve->dev->stats.rx_frame_errors;
 287                         ++geneve->dev->stats.rx_errors;
 288                         goto drop;
 289                 }
 290         }
 291
 292         len = skb->len;
 293         err = gro_cells_receive(&geneve->gro_cells, skb);
 294         if (likely(err == NET_RX_SUCCESS)) {
 295                 stats = this_cpu_ptr(geneve->dev->tstats);
 296                 u64_stats_update_begin(&stats->syncp);
 297                 stats->rx_packets++;
 298                 stats->rx_bytes += len;
 299                 u64_stats_update_end(&stats->syncp);
 300         }
 301         return;
 302 drop:
 303         /* Consume bad packet */
 304         kfree_skb(skb);
 305 }
 306
 307 /* Setup stats when device is created */
 308 static int geneve_init(struct net_device *dev)
 309 {
 310         struct geneve_dev *geneve = netdev_priv(dev);
 311         int err;
 312
 313         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 314         if (!dev->tstats)
 315                 return -ENOMEM;
 316
 317         err = gro_cells_init(&geneve->gro_cells, dev);
 318         if (err) {
 319                 free_percpu(dev->tstats);
 320                 return err;
 321         }
 322
 323         err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL);
 324         if (err) {
 325                 free_percpu(dev->tstats);
 326                 gro_cells_destroy(&geneve->gro_cells);
 327                 return err;
 328         }
 329         return 0;
 330 }
 331
 332 static void geneve_uninit(struct net_device *dev)
 333 {
 334         struct geneve_dev *geneve = netdev_priv(dev);
 335
 336         dst_cache_destroy(&geneve->info.dst_cache);
 337         gro_cells_destroy(&geneve->gro_cells);
 338         free_percpu(dev->tstats);
 339 }
 340
 341 /* Callback from net/ipv4/udp.c to receive packets */
 342 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 343 {
 344         struct genevehdr *geneveh;
 345         struct geneve_dev *geneve;
 346         struct geneve_sock *gs;
 347         int opts_len;
 348
 349         /* Need UDP and Geneve header to be present */
 350         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 351                 goto drop;
 352
 353         /* Return packets with reserved bits set */
 354         geneveh = geneve_hdr(skb);
 355         if (unlikely(geneveh->ver != GENEVE_VER))
 356                 goto drop;
 357
 358         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 359                 goto drop;
 360
 361         gs = rcu_dereference_sk_user_data(sk);
 362         if (!gs)
 363                 goto drop;
 364
 365         geneve = geneve_lookup_skb(gs, skb);
 366         if (!geneve)
 367                 goto drop;
 368
 369         opts_len = geneveh->opt_len * 4;
 370         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 371                                  htons(ETH_P_TEB),
 372                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 373                 geneve->dev->stats.rx_dropped++;
 374                 goto drop;
 375         }
 376
 377         geneve_rx(geneve, gs, skb);
 378         return 0;
 379
 380 drop:
 381         /* Consume bad packet */
 382         kfree_skb(skb);
 383         return 0;
 384 }
 385
 386 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 387                                          __be16 port, bool ipv6_rx_csum)
 388 {
 389         struct socket *sock;
 390         struct udp_port_cfg udp_conf;
 391         int err;
 392
 393         memset(&udp_conf, 0, sizeof(udp_conf));
 394
 395         if (ipv6) {
 396                 udp_conf.family = AF_INET6;
 397                 udp_conf.ipv6_v6only = 1;
 398                 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
 399         } else {
 400                 udp_conf.family = AF_INET;
 401                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 402         }
 403
 404         udp_conf.local_udp_port = port;
 405
 406         /* Open UDP socket */
 407         err = udp_sock_create(net, &udp_conf, &sock);
 408         if (err < 0)
 409                 return ERR_PTR(err);
 410
 411         return sock;
 412 }
 413
 414 static int geneve_hlen(struct genevehdr *gh)
 415 {
 416         return sizeof(*gh) + gh->opt_len * 4;
 417 }
 418
 419 static struct sk_buff **geneve_gro_receive(struct sock *sk,
 420                                            struct sk_buff **head,
 421                                            struct sk_buff *skb)
 422 {
 423         struct sk_buff *p, **pp = NULL;
 424         struct genevehdr *gh, *gh2;
 425         unsigned int hlen, gh_len, off_gnv;
 426         const struct packet_offload *ptype;
 427         __be16 type;
 428         int flush = 1;
 429
 430         off_gnv = skb_gro_offset(skb);
 431         hlen = off_gnv + sizeof(*gh);
 432         gh = skb_gro_header_fast(skb, off_gnv);
 433         if (skb_gro_header_hard(skb, hlen)) {
 434                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 435                 if (unlikely(!gh))
 436                         goto out;
 437         }
 438
 439         if (gh->ver != GENEVE_VER || gh->oam)
 440                 goto out;
 441         gh_len = geneve_hlen(gh);
 442
 443         hlen = off_gnv + gh_len;
 444         if (skb_gro_header_hard(skb, hlen)) {
 445                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 446                 if (unlikely(!gh))
 447                         goto out;
 448         }
 449
 450         for (p = *head; p; p = p->next) {
 451                 if (!NAPI_GRO_CB(p)->same_flow)
 452                         continue;
 453
 454                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 455                 if (gh->opt_len != gh2->opt_len ||
 456                     memcmp(gh, gh2, gh_len)) {
 457                         NAPI_GRO_CB(p)->same_flow = 0;
 458                         continue;
 459                 }
 460         }
 461
 462         type = gh->proto_type;
 463
 464         rcu_read_lock();
 465         ptype = gro_find_receive_by_type(type);
 466         if (!ptype)
 467                 goto out_unlock;
 468
 469         skb_gro_pull(skb, gh_len);
 470         skb_gro_postpull_rcsum(skb, gh, gh_len);
 471         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 472         flush = 0;
 473
 474 out_unlock:
 475         rcu_read_unlock();
 476 out:
 477         skb_gro_flush_final(skb, pp, flush);
 478
 479         return pp;
 480 }
 481
 482 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 483                                int nhoff)
 484 {
 485         struct genevehdr *gh;
 486         struct packet_offload *ptype;
 487         __be16 type;
 488         int gh_len;
 489         int err = -ENOSYS;
 490
 491         gh = (struct genevehdr *)(skb->data + nhoff);
 492         gh_len = geneve_hlen(gh);
 493         type = gh->proto_type;
 494
 495         rcu_read_lock();
 496         ptype = gro_find_complete_by_type(type);
 497         if (ptype)
 498                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 499
 500         rcu_read_unlock();
 501
 502         skb_set_inner_mac_header(skb, nhoff + gh_len);
 503
 504         return err;
 505 }
 506
 507 /* Create new listen socket if needed */
 508 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 509                                                 bool ipv6, bool ipv6_rx_csum)
 510 {
 511         struct geneve_net *gn = net_generic(net, geneve_net_id);
 512         struct geneve_sock *gs;
 513         struct socket *sock;
 514         struct udp_tunnel_sock_cfg tunnel_cfg;
 515         int h;
 516
 517         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 518         if (!gs)
 519                 return ERR_PTR(-ENOMEM);
 520
 521         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 522         if (IS_ERR(sock)) {
 523                 kfree(gs);
 524                 return ERR_CAST(sock);
 525         }
 526
 527         gs->sock = sock;
 528         gs->refcnt = 1;
 529         for (h = 0; h < VNI_HASH_SIZE; ++h)
 530                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 531
 532         /* Initialize the geneve udp offloads structure */
 533         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 534
 535         /* Mark socket as an encapsulation socket */
 536         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 537         tunnel_cfg.sk_user_data = gs;
 538         tunnel_cfg.encap_type = 1;
 539         tunnel_cfg.gro_receive = geneve_gro_receive;
 540         tunnel_cfg.gro_complete = geneve_gro_complete;
 541         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 542         tunnel_cfg.encap_destroy = NULL;
 543         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 544         list_add(&gs->list, &gn->sock_list);
 545         return gs;
 546 }
 547
 548 static void __geneve_sock_release(struct geneve_sock *gs)
 549 {
 550         if (!gs || --gs->refcnt)
 551                 return;
 552
 553         list_del(&gs->list);
 554         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 555         udp_tunnel_sock_release(gs->sock);
 556         kfree_rcu(gs, rcu);
 557 }
 558
 559 static void geneve_sock_release(struct geneve_dev *geneve)
 560 {
 561         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 562 #if IS_ENABLED(CONFIG_IPV6)
 563         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 564
 565         rcu_assign_pointer(geneve->sock6, NULL);
 566 #endif
 567
 568         rcu_assign_pointer(geneve->sock4, NULL);
 569         synchronize_net();
 570
 571         __geneve_sock_release(gs4);
 572 #if IS_ENABLED(CONFIG_IPV6)
 573         __geneve_sock_release(gs6);
 574 #endif
 575 }
 576
 577 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 578                                             sa_family_t family,
 579                                             __be16 dst_port)
 580 {
 581         struct geneve_sock *gs;
 582
 583         list_for_each_entry(gs, &gn->sock_list, list) {
 584                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 585                     geneve_get_sk_family(gs) == family) {
 586                         return gs;
 587                 }
 588         }
 589         return NULL;
 590 }
 591
 592 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 593 {
 594         struct net *net = geneve->net;
 595         struct geneve_net *gn = net_generic(net, geneve_net_id);
 596         struct geneve_dev_node *node;
 597         struct geneve_sock *gs;
 598         __u8 vni[3];
 599         __u32 hash;
 600
 601         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst);
 602         if (gs) {
 603                 gs->refcnt++;
 604                 goto out;
 605         }
 606
 607         gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6,
 608                                   geneve->use_udp6_rx_checksums);
 609         if (IS_ERR(gs))
 610                 return PTR_ERR(gs);
 611
 612 out:
 613         gs->collect_md = geneve->collect_md;
 614 #if IS_ENABLED(CONFIG_IPV6)
 615         if (ipv6) {
 616                 rcu_assign_pointer(geneve->sock6, gs);
 617                 node = &geneve->hlist6;
 618         } else
 619 #endif
 620         {
 621                 rcu_assign_pointer(geneve->sock4, gs);
 622                 node = &geneve->hlist4;
 623         }
 624         node->geneve = geneve;
 625
 626         tunnel_id_to_vni(geneve->info.key.tun_id, vni);
 627         hash = geneve_net_vni_hash(vni);
 628         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 629         return 0;
 630 }
 631
 632 static int geneve_open(struct net_device *dev)
 633 {
 634         struct geneve_dev *geneve = netdev_priv(dev);
 635         bool metadata = geneve->collect_md;
 636         bool ipv4, ipv6;
 637         int ret = 0;
 638
 639         ipv6 = geneve->info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 640         ipv4 = !ipv6 || metadata;
 641 #if IS_ENABLED(CONFIG_IPV6)
 642         if (ipv6) {
 643                 ret = geneve_sock_add(geneve, true);
 644                 if (ret < 0 && ret != -EAFNOSUPPORT)
 645                         ipv4 = false;
 646         }
 647 #endif
 648         if (ipv4)
 649                 ret = geneve_sock_add(geneve, false);
 650         if (ret < 0)
 651                 geneve_sock_release(geneve);
 652
 653         return ret;
 654 }
 655
 656 static int geneve_stop(struct net_device *dev)
 657 {
 658         struct geneve_dev *geneve = netdev_priv(dev);
 659
 660         hlist_del_init_rcu(&geneve->hlist4.hlist);
 661 #if IS_ENABLED(CONFIG_IPV6)
 662         hlist_del_init_rcu(&geneve->hlist6.hlist);
 663 #endif
 664         geneve_sock_release(geneve);
 665         return 0;
 666 }
 667
 668 static void geneve_build_header(struct genevehdr *geneveh,
 669                                 const struct ip_tunnel_info *info)
 670 {
 671         geneveh->ver = GENEVE_VER;
 672         geneveh->opt_len = info->options_len / 4;
 673         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 674         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 675         geneveh->rsvd1 = 0;
 676         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 677         geneveh->proto_type = htons(ETH_P_TEB);
 678         geneveh->rsvd2 = 0;
 679
 680         ip_tunnel_info_opts_get(geneveh->options, info);
 681 }
 682
 683 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 684                             const struct ip_tunnel_info *info,
 685                             bool xnet, int ip_hdr_len)
 686 {
 687         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 688         struct genevehdr *gnvh;
 689         int min_headroom;
 690         int err;
 691
 692         skb_reset_mac_header(skb);
 693         skb_scrub_packet(skb, xnet);
 694
 695         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 696                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 697         err = skb_cow_head(skb, min_headroom);
 698         if (unlikely(err))
 699                 goto free_dst;
 700
 701         err = udp_tunnel_handle_offloads(skb, udp_sum);
 702         if (err)
 703                 goto free_dst;
 704
 705         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 706         geneve_build_header(gnvh, info);
 707         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 708         return 0;
 709
 710 free_dst:
 711         dst_release(dst);
 712         return err;
 713 }
 714
 715 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 716                                        struct net_device *dev,
 717                                        struct geneve_sock *gs4,
 718                                        struct flowi4 *fl4,
 719                                        const struct ip_tunnel_info *info,
 720                                        __be16 dport, __be16 sport)
 721 {
 722         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 723         struct geneve_dev *geneve = netdev_priv(dev);
 724         struct dst_cache *dst_cache;
 725         struct rtable *rt = NULL;
 726         __u8 tos;
 727
 728         if (!gs4)
 729                 return ERR_PTR(-EIO);
 730
 731         memset(fl4, 0, sizeof(*fl4));
 732         fl4->flowi4_mark = skb->mark;
 733         fl4->flowi4_proto = IPPROTO_UDP;
 734         fl4->daddr = info->key.u.ipv4.dst;
 735         fl4->saddr = info->key.u.ipv4.src;
 736         fl4->fl4_dport = dport;
 737         fl4->fl4_sport = sport;
 738
 739         tos = info->key.tos;
 740         if ((tos == 1) && !geneve->collect_md) {
 741                 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 742                 use_cache = false;
 743         }
 744         fl4->flowi4_tos = RT_TOS(tos);
 745
 746         dst_cache = (struct dst_cache *)&info->dst_cache;
 747         if (use_cache) {
 748                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 749                 if (rt)
 750                         return rt;
 751         }
 752         rt = ip_route_output_key(geneve->net, fl4);
 753         if (IS_ERR(rt)) {
 754                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 755                 return ERR_PTR(-ENETUNREACH);
 756         }
 757         if (rt->dst.dev == dev) { /* is this necessary? */
 758                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 759                 ip_rt_put(rt);
 760                 return ERR_PTR(-ELOOP);
 761         }
 762         if (use_cache)
 763                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 764         return rt;
 765 }
 766
 767 #if IS_ENABLED(CONFIG_IPV6)
 768 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 769                                            struct net_device *dev,
 770                                            struct geneve_sock *gs6,
 771                                            struct flowi6 *fl6,
 772                                            const struct ip_tunnel_info *info,
 773                                            __be16 dport, __be16 sport)
 774 {
 775         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 776         struct geneve_dev *geneve = netdev_priv(dev);
 777         struct dst_entry *dst = NULL;
 778         struct dst_cache *dst_cache;
 779         __u8 prio;
 780
 781         if (!gs6)
 782                 return ERR_PTR(-EIO);
 783
 784         memset(fl6, 0, sizeof(*fl6));
 785         fl6->flowi6_mark = skb->mark;
 786         fl6->flowi6_proto = IPPROTO_UDP;
 787         fl6->daddr = info->key.u.ipv6.dst;
 788         fl6->saddr = info->key.u.ipv6.src;
 789         fl6->fl6_dport = dport;
 790         fl6->fl6_sport = sport;
 791
 792         prio = info->key.tos;
 793         if ((prio == 1) && !geneve->collect_md) {
 794                 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 795                 use_cache = false;
 796         }
 797
 798         fl6->flowlabel = ip6_make_flowinfo(prio, info->key.label);
 799         dst_cache = (struct dst_cache *)&info->dst_cache;
 800         if (use_cache) {
 801                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 802                 if (dst)
 803                         return dst;
 804         }
 805         dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
 806                                               NULL);
 807         if (IS_ERR(dst)) {
 808                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 809                 return ERR_PTR(-ENETUNREACH);
 810         }
 811         if (dst->dev == dev) { /* is this necessary? */
 812                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 813                 dst_release(dst);
 814                 return ERR_PTR(-ELOOP);
 815         }
 816
 817         if (use_cache)
 818                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 819         return dst;
 820 }
 821 #endif
 822
 823 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 824                            struct geneve_dev *geneve,
 825                            const struct ip_tunnel_info *info)
 826 {
 827         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 828         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 829         const struct ip_tunnel_key *key = &info->key;
 830         struct rtable *rt;
 831         struct flowi4 fl4;
 832         __u8 tos, ttl;
 833         __be16 sport;
 834         __be16 df;
 835         int err;
 836
 837         if (!pskb_network_may_pull(skb, sizeof(struct iphdr)))
 838                 return -EINVAL;
 839
 840         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 841         rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
 842                               geneve->info.key.tp_dst, sport);
 843         if (IS_ERR(rt))
 844                 return PTR_ERR(rt);
 845
 846         if (skb_dst(skb)) {
 847                 int mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr) -
 848                           GENEVE_BASE_HLEN - info->options_len - 14;
 849
 850                 skb_dst_update_pmtu(skb, mtu);
 851         }
 852
 853         if (geneve->collect_md) {
 854                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 855                 ttl = key->ttl;
 856         } else {
 857                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 858                 ttl = key->ttl ? : ip4_dst_hoplimit(&rt->dst);
 859         }
 860         df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 861
 862         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
 863         if (unlikely(err))
 864                 return err;
 865
 866         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 867                             tos, ttl, df, sport, geneve->info.key.tp_dst,
 868                             !net_eq(geneve->net, dev_net(geneve->dev)),
 869                             !(info->key.tun_flags & TUNNEL_CSUM));
 870         return 0;
 871 }
 872
 873 #if IS_ENABLED(CONFIG_IPV6)
 874 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 875                             struct geneve_dev *geneve,
 876                             const struct ip_tunnel_info *info)
 877 {
 878         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 879         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 880         const struct ip_tunnel_key *key = &info->key;
 881         struct dst_entry *dst = NULL;
 882         struct flowi6 fl6;
 883         __u8 prio, ttl;
 884         __be16 sport;
 885         int err;
 886
 887         if (!pskb_network_may_pull(skb, sizeof(struct ipv6hdr)))
 888                 return -EINVAL;
 889
 890         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 891         dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
 892                                 geneve->info.key.tp_dst, sport);
 893         if (IS_ERR(dst))
 894                 return PTR_ERR(dst);
 895
 896         if (skb_dst(skb)) {
 897                 int mtu = dst_mtu(dst) - sizeof(struct ipv6hdr) -
 898                           GENEVE_BASE_HLEN - info->options_len - 14;
 899
 900                 skb_dst_update_pmtu(skb, mtu);
 901         }
 902
 903         if (geneve->collect_md) {
 904                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 905                 ttl = key->ttl;
 906         } else {
 907                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
 908                                            ip_hdr(skb), skb);
 909                 ttl = key->ttl ? : ip6_dst_hoplimit(dst);
 910         }
 911         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
 912         if (unlikely(err))
 913                 return err;
 914
 915         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
 916                              &fl6.saddr, &fl6.daddr, prio, ttl,
 917                              info->key.label, sport, geneve->info.key.tp_dst,
 918                              !(info->key.tun_flags & TUNNEL_CSUM));
 919         return 0;
 920 }
 921 #endif
 922
 923 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
 924 {
 925         struct geneve_dev *geneve = netdev_priv(dev);
 926         struct ip_tunnel_info *info = NULL;
 927         int err;
 928
 929         if (geneve->collect_md) {
 930                 info = skb_tunnel_info(skb);
 931                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 932                         netdev_dbg(dev, "no tunnel metadata\n");
 933                         dev_kfree_skb(skb);
 934                         dev->stats.tx_dropped++;
 935                         return NETDEV_TX_OK;
 936                 }
 937         } else {
 938                 info = &geneve->info;
 939         }
 940
 941         rcu_read_lock();
 942 #if IS_ENABLED(CONFIG_IPV6)
 943         if (info->mode & IP_TUNNEL_INFO_IPV6)
 944                 err = geneve6_xmit_skb(skb, dev, geneve, info);
 945         else
 946 #endif
 947                 err = geneve_xmit_skb(skb, dev, geneve, info);
 948         rcu_read_unlock();
 949
 950         if (likely(!err))
 951                 return NETDEV_TX_OK;
 952
 953         dev_kfree_skb(skb);
 954
 955         if (err == -ELOOP)
 956                 dev->stats.collisions++;
 957         else if (err == -ENETUNREACH)
 958                 dev->stats.tx_carrier_errors++;
 959
 960         dev->stats.tx_errors++;
 961         return NETDEV_TX_OK;
 962 }
 963
 964 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
 965 {
 966         /* Only possible if called internally, ndo_change_mtu path's new_mtu
 967          * is guaranteed to be between dev->min_mtu and dev->max_mtu.
 968          */
 969         if (new_mtu > dev->max_mtu)
 970                 new_mtu = dev->max_mtu;
 971
 972         dev->mtu = new_mtu;
 973         return 0;
 974 }
 975
 976 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
 977 {
 978         struct ip_tunnel_info *info = skb_tunnel_info(skb);
 979         struct geneve_dev *geneve = netdev_priv(dev);
 980         __be16 sport;
 981
 982         if (ip_tunnel_info_af(info) == AF_INET) {
 983                 struct rtable *rt;
 984                 struct flowi4 fl4;
 985
 986                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 987                 sport = udp_flow_src_port(geneve->net, skb,
 988                                           1, USHRT_MAX, true);
 989
 990                 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
 991                                       geneve->info.key.tp_dst, sport);
 992                 if (IS_ERR(rt))
 993                         return PTR_ERR(rt);
 994
 995                 ip_rt_put(rt);
 996                 info->key.u.ipv4.src = fl4.saddr;
 997 #if IS_ENABLED(CONFIG_IPV6)
 998         } else if (ip_tunnel_info_af(info) == AF_INET6) {
 999                 struct dst_entry *dst;
1000                 struct flowi6 fl6;
1001
1002                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1003                 sport = udp_flow_src_port(geneve->net, skb,
1004                                           1, USHRT_MAX, true);
1005
1006                 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1007                                         geneve->info.key.tp_dst, sport);
1008                 if (IS_ERR(dst))
1009                         return PTR_ERR(dst);
1010
1011                 dst_release(dst);
1012                 info->key.u.ipv6.src = fl6.saddr;
1013 #endif
1014         } else {
1015                 return -EINVAL;
1016         }
1017
1018         info->key.tp_src = sport;
1019         info->key.tp_dst = geneve->info.key.tp_dst;
1020         return 0;
1021 }
1022
1023 static const struct net_device_ops geneve_netdev_ops = {
1024         .ndo_init               = geneve_init,
1025         .ndo_uninit             = geneve_uninit,
1026         .ndo_open               = geneve_open,
1027         .ndo_stop               = geneve_stop,
1028         .ndo_start_xmit         = geneve_xmit,
1029         .ndo_get_stats64        = ip_tunnel_get_stats64,
1030         .ndo_change_mtu         = geneve_change_mtu,
1031         .ndo_validate_addr      = eth_validate_addr,
1032         .ndo_set_mac_address    = eth_mac_addr,
1033         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1034 };
1035
1036 static void geneve_get_drvinfo(struct net_device *dev,
1037                                struct ethtool_drvinfo *drvinfo)
1038 {
1039         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1040         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1041 }
1042
1043 static const struct ethtool_ops geneve_ethtool_ops = {
1044         .get_drvinfo    = geneve_get_drvinfo,
1045         .get_link       = ethtool_op_get_link,
1046 };
1047
1048 /* Info for udev, that this is a virtual tunnel endpoint */
1049 static struct device_type geneve_type = {
1050         .name = "geneve",
1051 };
1052
1053 /* Calls the ndo_udp_tunnel_add of the caller in order to
1054  * supply the listening GENEVE udp ports. Callers are expected
1055  * to implement the ndo_udp_tunnel_add.
1056  */
1057 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1058 {
1059         struct net *net = dev_net(dev);
1060         struct geneve_net *gn = net_generic(net, geneve_net_id);
1061         struct geneve_sock *gs;
1062
1063         rcu_read_lock();
1064         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1065                 if (push) {
1066                         udp_tunnel_push_rx_port(dev, gs->sock,
1067                                                 UDP_TUNNEL_TYPE_GENEVE);
1068                 } else {
1069                         udp_tunnel_drop_rx_port(dev, gs->sock,
1070                                                 UDP_TUNNEL_TYPE_GENEVE);
1071                 }
1072         }
1073         rcu_read_unlock();
1074 }
1075
1076 /* Initialize the device structure. */
1077 static void geneve_setup(struct net_device *dev)
1078 {
1079         ether_setup(dev);
1080
1081         dev->netdev_ops = &geneve_netdev_ops;
1082         dev->ethtool_ops = &geneve_ethtool_ops;
1083         dev->needs_free_netdev = true;
1084
1085         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1086
1087         dev->features    |= NETIF_F_LLTX;
1088         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1089         dev->features    |= NETIF_F_RXCSUM;
1090         dev->features    |= NETIF_F_GSO_SOFTWARE;
1091
1092         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1093         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1094
1095         /* MTU range: 68 - (something less than 65535) */
1096         dev->min_mtu = ETH_MIN_MTU;
1097         /* The max_mtu calculation does not take account of GENEVE
1098          * options, to avoid excluding potentially valid
1099          * configurations. This will be further reduced by IPvX hdr size.
1100          */
1101         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1102
1103         netif_keep_dst(dev);
1104         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1105         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1106         eth_hw_addr_random(dev);
1107 }
1108
1109 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1110         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1111         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1112         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1113         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1114         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1115         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1116         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1117         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1118         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1119         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1120         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1121 };
1122
1123 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1124                            struct netlink_ext_ack *extack)
1125 {
1126         if (tb[IFLA_ADDRESS]) {
1127                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1128                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1129                                             "Provided link layer address is not Ethernet");
1130                         return -EINVAL;
1131                 }
1132
1133                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1134                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1135                                             "Provided Ethernet address is not unicast");
1136                         return -EADDRNOTAVAIL;
1137                 }
1138         }
1139
1140         if (!data) {
1141                 NL_SET_ERR_MSG(extack,
1142                                "Not enough attributes provided to perform the operation");
1143                 return -EINVAL;
1144         }
1145
1146         if (data[IFLA_GENEVE_ID]) {
1147                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1148
1149                 if (vni >= GENEVE_N_VID) {
1150                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1151                                             "Geneve ID must be lower than 16777216");
1152                         return -ERANGE;
1153                 }
1154         }
1155
1156         return 0;
1157 }
1158
1159 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1160                                           const struct ip_tunnel_info *info,
1161                                           bool *tun_on_same_port,
1162                                           bool *tun_collect_md)
1163 {
1164         struct geneve_dev *geneve, *t = NULL;
1165
1166         *tun_on_same_port = false;
1167         *tun_collect_md = false;
1168         list_for_each_entry(geneve, &gn->geneve_list, next) {
1169                 if (info->key.tp_dst == geneve->info.key.tp_dst) {
1170                         *tun_collect_md = geneve->collect_md;
1171                         *tun_on_same_port = true;
1172                 }
1173                 if (info->key.tun_id == geneve->info.key.tun_id &&
1174                     info->key.tp_dst == geneve->info.key.tp_dst &&
1175                     !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1176                         t = geneve;
1177         }
1178         return t;
1179 }
1180
1181 static bool is_all_zero(const u8 *fp, size_t size)
1182 {
1183         int i;
1184
1185         for (i = 0; i < size; i++)
1186                 if (fp[i])
1187                         return false;
1188         return true;
1189 }
1190
1191 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1192 {
1193         if (info->key.tun_id || info->key.tun_flags || info->key.tos ||
1194             info->key.ttl || info->key.label || info->key.tp_src ||
1195             !is_all_zero((const u8 *)&info->key.u, sizeof(info->key.u)))
1196                 return false;
1197         else
1198                 return true;
1199 }
1200
1201 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1202                                   struct ip_tunnel_info *b)
1203 {
1204         if (ip_tunnel_info_af(a) == AF_INET)
1205                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1206         else
1207                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1208 }
1209
1210 static int geneve_configure(struct net *net, struct net_device *dev,
1211                             struct netlink_ext_ack *extack,
1212                             const struct ip_tunnel_info *info,
1213                             bool metadata, bool ipv6_rx_csum)
1214 {
1215         struct geneve_net *gn = net_generic(net, geneve_net_id);
1216         struct geneve_dev *t, *geneve = netdev_priv(dev);
1217         bool tun_collect_md, tun_on_same_port;
1218         int err, encap_len;
1219
1220         if (metadata && !is_tnl_info_zero(info)) {
1221                 NL_SET_ERR_MSG(extack,
1222                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1223                 return -EINVAL;
1224         }
1225
1226         geneve->net = net;
1227         geneve->dev = dev;
1228
1229         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1230         if (t)
1231                 return -EBUSY;
1232
1233         /* make enough headroom for basic scenario */
1234         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1235         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1236                 encap_len += sizeof(struct iphdr);
1237                 dev->max_mtu -= sizeof(struct iphdr);
1238         } else {
1239                 encap_len += sizeof(struct ipv6hdr);
1240                 dev->max_mtu -= sizeof(struct ipv6hdr);
1241         }
1242         dev->needed_headroom = encap_len + ETH_HLEN;
1243
1244         if (metadata) {
1245                 if (tun_on_same_port) {
1246                         NL_SET_ERR_MSG(extack,
1247                                        "There can be only one externally controlled device on a destination port");
1248                         return -EPERM;
1249                 }
1250         } else {
1251                 if (tun_collect_md) {
1252                         NL_SET_ERR_MSG(extack,
1253                                        "There already exists an externally controlled device on this destination port");
1254                         return -EPERM;
1255                 }
1256         }
1257
1258         dst_cache_reset(&geneve->info.dst_cache);
1259         geneve->info = *info;
1260         geneve->collect_md = metadata;
1261         geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1262
1263         err = register_netdevice(dev);
1264         if (err)
1265                 return err;
1266
1267         list_add(&geneve->next, &gn->geneve_list);
1268         return 0;
1269 }
1270
1271 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1272 {
1273         memset(info, 0, sizeof(*info));
1274         info->key.tp_dst = htons(dst_port);
1275 }
1276
1277 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1278                           struct netlink_ext_ack *extack,
1279                           struct ip_tunnel_info *info, bool *metadata,
1280                           bool *use_udp6_rx_checksums, bool changelink)
1281 {
1282         int attrtype;
1283
1284         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1285                 NL_SET_ERR_MSG(extack,
1286                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1287                 return -EINVAL;
1288         }
1289
1290         if (data[IFLA_GENEVE_REMOTE]) {
1291                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1292                         attrtype = IFLA_GENEVE_REMOTE;
1293                         goto change_notsup;
1294                 }
1295
1296                 info->key.u.ipv4.dst =
1297                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1298
1299                 if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) {
1300                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1301                                             "Remote IPv4 address cannot be Multicast");
1302                         return -EINVAL;
1303                 }
1304         }
1305
1306         if (data[IFLA_GENEVE_REMOTE6]) {
1307  #if IS_ENABLED(CONFIG_IPV6)
1308                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1309                         attrtype = IFLA_GENEVE_REMOTE6;
1310                         goto change_notsup;
1311                 }
1312
1313                 info->mode = IP_TUNNEL_INFO_IPV6;
1314                 info->key.u.ipv6.dst =
1315                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1316
1317                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1318                     IPV6_ADDR_LINKLOCAL) {
1319                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1320                                             "Remote IPv6 address cannot be link-local");
1321                         return -EINVAL;
1322                 }
1323                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1324                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1325                                             "Remote IPv6 address cannot be Multicast");
1326                         return -EINVAL;
1327                 }
1328                 info->key.tun_flags |= TUNNEL_CSUM;
1329                 *use_udp6_rx_checksums = true;
1330 #else
1331                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1332                                     "IPv6 support not enabled in the kernel");
1333                 return -EPFNOSUPPORT;
1334 #endif
1335         }
1336
1337         if (data[IFLA_GENEVE_ID]) {
1338                 __u32 vni;
1339                 __u8 tvni[3];
1340                 __be64 tunid;
1341
1342                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1343                 tvni[0] = (vni & 0x00ff0000) >> 16;
1344                 tvni[1] = (vni & 0x0000ff00) >> 8;
1345                 tvni[2] =  vni & 0x000000ff;
1346
1347                 tunid = vni_to_tunnel_id(tvni);
1348                 if (changelink && (tunid != info->key.tun_id)) {
1349                         attrtype = IFLA_GENEVE_ID;
1350                         goto change_notsup;
1351                 }
1352                 info->key.tun_id = tunid;
1353         }
1354
1355         if (data[IFLA_GENEVE_TTL])
1356                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1357
1358         if (data[IFLA_GENEVE_TOS])
1359                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1360
1361         if (data[IFLA_GENEVE_LABEL]) {
1362                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1363                                   IPV6_FLOWLABEL_MASK;
1364                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1365                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1366                                             "Label attribute only applies for IPv6 Geneve devices");
1367                         return -EINVAL;
1368                 }
1369         }
1370
1371         if (data[IFLA_GENEVE_PORT]) {
1372                 if (changelink) {
1373                         attrtype = IFLA_GENEVE_PORT;
1374                         goto change_notsup;
1375                 }
1376                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1377         }
1378
1379         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1380                 if (changelink) {
1381                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1382                         goto change_notsup;
1383                 }
1384                 *metadata = true;
1385         }
1386
1387         if (data[IFLA_GENEVE_UDP_CSUM]) {
1388                 if (changelink) {
1389                         attrtype = IFLA_GENEVE_UDP_CSUM;
1390                         goto change_notsup;
1391                 }
1392                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1393                         info->key.tun_flags |= TUNNEL_CSUM;
1394         }
1395
1396         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1397 #if IS_ENABLED(CONFIG_IPV6)
1398                 if (changelink) {
1399                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1400                         goto change_notsup;
1401                 }
1402                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1403                         info->key.tun_flags &= ~TUNNEL_CSUM;
1404 #else
1405                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1406                                     "IPv6 support not enabled in the kernel");
1407                 return -EPFNOSUPPORT;
1408 #endif
1409         }
1410
1411         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1412 #if IS_ENABLED(CONFIG_IPV6)
1413                 if (changelink) {
1414                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1415                         goto change_notsup;
1416                 }
1417                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1418                         *use_udp6_rx_checksums = false;
1419 #else
1420                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1421                                     "IPv6 support not enabled in the kernel");
1422                 return -EPFNOSUPPORT;
1423 #endif
1424         }
1425
1426         return 0;
1427 change_notsup:
1428         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1429                             "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1430         return -EOPNOTSUPP;
1431 }
1432
1433 static int geneve_newlink(struct net *net, struct net_device *dev,
1434                           struct nlattr *tb[], struct nlattr *data[],
1435                           struct netlink_ext_ack *extack)
1436 {
1437         bool use_udp6_rx_checksums = false;
1438         struct ip_tunnel_info info;
1439         bool metadata = false;
1440         int err;
1441
1442         init_tnl_info(&info, GENEVE_UDP_PORT);
1443         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1444                              &use_udp6_rx_checksums, false);
1445         if (err)
1446                 return err;
1447
1448         return geneve_configure(net, dev, extack, &info, metadata,
1449                                 use_udp6_rx_checksums);
1450 }
1451
1452 /* Quiesces the geneve device data path for both TX and RX.
1453  *
1454  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1455  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1456  * to complete for the existing set of in-flight packets to be transmitted,
1457  * then we would have quiesced the transmit data path. All the future packets
1458  * will get dropped until we unquiesce the data path.
1459  *
1460  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1461  * if we set that to NULL under RCU and wait for synchronize_net() to
1462  * complete, then we would have quiesced the receive data path.
1463  */
1464 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1465                            struct geneve_sock **gs6)
1466 {
1467         *gs4 = rtnl_dereference(geneve->sock4);
1468         rcu_assign_pointer(geneve->sock4, NULL);
1469         if (*gs4)
1470                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1471 #if IS_ENABLED(CONFIG_IPV6)
1472         *gs6 = rtnl_dereference(geneve->sock6);
1473         rcu_assign_pointer(geneve->sock6, NULL);
1474         if (*gs6)
1475                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1476 #else
1477         *gs6 = NULL;
1478 #endif
1479         synchronize_net();
1480 }
1481
1482 /* Resumes the geneve device data path for both TX and RX. */
1483 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1484                              struct geneve_sock __maybe_unused *gs6)
1485 {
1486         rcu_assign_pointer(geneve->sock4, gs4);
1487         if (gs4)
1488                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1489 #if IS_ENABLED(CONFIG_IPV6)
1490         rcu_assign_pointer(geneve->sock6, gs6);
1491         if (gs6)
1492                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1493 #endif
1494         synchronize_net();
1495 }
1496
1497 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1498                              struct nlattr *data[],
1499                              struct netlink_ext_ack *extack)
1500 {
1501         struct geneve_dev *geneve = netdev_priv(dev);
1502         struct geneve_sock *gs4, *gs6;
1503         struct ip_tunnel_info info;
1504         bool metadata;
1505         bool use_udp6_rx_checksums;
1506         int err;
1507
1508         /* If the geneve device is configured for metadata (or externally
1509          * controlled, for example, OVS), then nothing can be changed.
1510          */
1511         if (geneve->collect_md)
1512                 return -EOPNOTSUPP;
1513
1514         /* Start with the existing info. */
1515         memcpy(&info, &geneve->info, sizeof(info));
1516         metadata = geneve->collect_md;
1517         use_udp6_rx_checksums = geneve->use_udp6_rx_checksums;
1518         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1519                              &use_udp6_rx_checksums, true);
1520         if (err)
1521                 return err;
1522
1523         if (!geneve_dst_addr_equal(&geneve->info, &info))
1524                 dst_cache_reset(&info.dst_cache);
1525
1526         geneve_quiesce(geneve, &gs4, &gs6);
1527         geneve->info = info;
1528         geneve->collect_md = metadata;
1529         geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
1530         geneve_unquiesce(geneve, gs4, gs6);
1531
1532         return 0;
1533 }
1534
1535 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1536 {
1537         struct geneve_dev *geneve = netdev_priv(dev);
1538
1539         list_del(&geneve->next);
1540         unregister_netdevice_queue(dev, head);
1541 }
1542
1543 static size_t geneve_get_size(const struct net_device *dev)
1544 {
1545         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1546                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1547                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1548                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1549                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1550                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1551                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1552                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1553                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1554                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1555                 0;
1556 }
1557
1558 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1559 {
1560         struct geneve_dev *geneve = netdev_priv(dev);
1561         struct ip_tunnel_info *info = &geneve->info;
1562         bool metadata = geneve->collect_md;
1563         __u8 tmp_vni[3];
1564         __u32 vni;
1565
1566         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1567         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1568         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1569                 goto nla_put_failure;
1570
1571         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1572                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1573                                     info->key.u.ipv4.dst))
1574                         goto nla_put_failure;
1575                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1576                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1577                         goto nla_put_failure;
1578
1579 #if IS_ENABLED(CONFIG_IPV6)
1580         } else if (!metadata) {
1581                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1582                                      &info->key.u.ipv6.dst))
1583                         goto nla_put_failure;
1584                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1585                                !(info->key.tun_flags & TUNNEL_CSUM)))
1586                         goto nla_put_failure;
1587 #endif
1588         }
1589
1590         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1591             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1592             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1593                 goto nla_put_failure;
1594
1595         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1596                 goto nla_put_failure;
1597
1598         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1599                 goto nla_put_failure;
1600
1601 #if IS_ENABLED(CONFIG_IPV6)
1602         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1603                        !geneve->use_udp6_rx_checksums))
1604                 goto nla_put_failure;
1605 #endif
1606
1607         return 0;
1608
1609 nla_put_failure:
1610         return -EMSGSIZE;
1611 }
1612
1613 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1614         .kind           = "geneve",
1615         .maxtype        = IFLA_GENEVE_MAX,
1616         .policy         = geneve_policy,
1617         .priv_size      = sizeof(struct geneve_dev),
1618         .setup          = geneve_setup,
1619         .validate       = geneve_validate,
1620         .newlink        = geneve_newlink,
1621         .changelink     = geneve_changelink,
1622         .dellink        = geneve_dellink,
1623         .get_size       = geneve_get_size,
1624         .fill_info      = geneve_fill_info,
1625 };
1626
1627 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1628                                         u8 name_assign_type, u16 dst_port)
1629 {
1630         struct nlattr *tb[IFLA_MAX + 1];
1631         struct ip_tunnel_info info;
1632         struct net_device *dev;
1633         LIST_HEAD(list_kill);
1634         int err;
1635
1636         memset(tb, 0, sizeof(tb));
1637         dev = rtnl_create_link(net, name, name_assign_type,
1638                                &geneve_link_ops, tb);
1639         if (IS_ERR(dev))
1640                 return dev;
1641
1642         init_tnl_info(&info, dst_port);
1643         err = geneve_configure(net, dev, NULL, &info, true, true);
1644         if (err) {
1645                 free_netdev(dev);
1646                 return ERR_PTR(err);
1647         }
1648
1649         /* openvswitch users expect packet sizes to be unrestricted,
1650          * so set the largest MTU we can.
1651          */
1652         err = geneve_change_mtu(dev, IP_MAX_MTU);
1653         if (err)
1654                 goto err;
1655
1656         err = rtnl_configure_link(dev, NULL);
1657         if (err < 0)
1658                 goto err;
1659
1660         return dev;
1661 err:
1662         geneve_dellink(dev, &list_kill);
1663         unregister_netdevice_many(&list_kill);
1664         return ERR_PTR(err);
1665 }
1666 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1667
1668 static int geneve_netdevice_event(struct notifier_block *unused,
1669                                   unsigned long event, void *ptr)
1670 {
1671         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1672
1673         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1674             event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1675                 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1676         } else if (event == NETDEV_UNREGISTER) {
1677                 geneve_offload_rx_ports(dev, false);
1678         } else if (event == NETDEV_REGISTER) {
1679                 geneve_offload_rx_ports(dev, true);
1680         }
1681
1682         return NOTIFY_DONE;
1683 }
1684
1685 static struct notifier_block geneve_notifier_block __read_mostly = {
1686         .notifier_call = geneve_netdevice_event,
1687 };
1688
1689 static __net_init int geneve_init_net(struct net *net)
1690 {
1691         struct geneve_net *gn = net_generic(net, geneve_net_id);
1692
1693         INIT_LIST_HEAD(&gn->geneve_list);
1694         INIT_LIST_HEAD(&gn->sock_list);
1695         return 0;
1696 }
1697
1698 static void __net_exit geneve_exit_net(struct net *net)
1699 {
1700         struct geneve_net *gn = net_generic(net, geneve_net_id);
1701         struct geneve_dev *geneve, *next;
1702         struct net_device *dev, *aux;
1703         LIST_HEAD(list);
1704
1705         rtnl_lock();
1706
1707         /* gather any geneve devices that were moved into this ns */
1708         for_each_netdev_safe(net, dev, aux)
1709                 if (dev->rtnl_link_ops == &geneve_link_ops)
1710                         unregister_netdevice_queue(dev, &list);
1711
1712         /* now gather any other geneve devices that were created in this ns */
1713         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1714                 /* If geneve->dev is in the same netns, it was already added
1715                  * to the list by the previous loop.
1716                  */
1717                 if (!net_eq(dev_net(geneve->dev), net))
1718                         unregister_netdevice_queue(geneve->dev, &list);
1719         }
1720
1721         /* unregister the devices gathered above */
1722         unregister_netdevice_many(&list);
1723         rtnl_unlock();
1724 }
1725
1726 static struct pernet_operations geneve_net_ops = {
1727         .init = geneve_init_net,
1728         .exit = geneve_exit_net,
1729         .id   = &geneve_net_id,
1730         .size = sizeof(struct geneve_net),
1731 };
1732
1733 static int __init geneve_init_module(void)
1734 {
1735         int rc;
1736
1737         rc = register_pernet_subsys(&geneve_net_ops);
1738         if (rc)
1739                 goto out1;
1740
1741         rc = register_netdevice_notifier(&geneve_notifier_block);
1742         if (rc)
1743                 goto out2;
1744
1745         rc = rtnl_link_register(&geneve_link_ops);
1746         if (rc)
1747                 goto out3;
1748
1749         return 0;
1750 out3:
1751         unregister_netdevice_notifier(&geneve_notifier_block);
1752 out2:
1753         unregister_pernet_subsys(&geneve_net_ops);
1754 out1:
1755         return rc;
1756 }
1757 late_initcall(geneve_init_module);
1758
1759 static void __exit geneve_cleanup_module(void)
1760 {
1761         rtnl_link_unregister(&geneve_link_ops);
1762         unregister_netdevice_notifier(&geneve_notifier_block);
1763         unregister_pernet_subsys(&geneve_net_ops);
1764 }
1765 module_exit(geneve_cleanup_module);
1766
1767 MODULE_LICENSE("GPL");
1768 MODULE_VERSION(GENEVE_NETDEV_VER);
1769 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1770 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1771 MODULE_ALIAS_RTNL_LINK("geneve");