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         NAPI_GRO_CB(skb)->flush |= 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 ipv6 = !!(geneve->info.mode & IP_TUNNEL_INFO_IPV6);
 636         bool metadata = geneve->collect_md;
 637         int ret = 0;
 638
 639 #if IS_ENABLED(CONFIG_IPV6)
 640         if (ipv6 || metadata)
 641                 ret = geneve_sock_add(geneve, true);
 642 #endif
 643         if (!ret && (!ipv6 || metadata))
 644                 ret = geneve_sock_add(geneve, false);
 645         if (ret < 0)
 646                 geneve_sock_release(geneve);
 647
 648         return ret;
 649 }
 650
 651 static int geneve_stop(struct net_device *dev)
 652 {
 653         struct geneve_dev *geneve = netdev_priv(dev);
 654
 655         hlist_del_init_rcu(&geneve->hlist4.hlist);
 656 #if IS_ENABLED(CONFIG_IPV6)
 657         hlist_del_init_rcu(&geneve->hlist6.hlist);
 658 #endif
 659         geneve_sock_release(geneve);
 660         return 0;
 661 }
 662
 663 static void geneve_build_header(struct genevehdr *geneveh,
 664                                 const struct ip_tunnel_info *info)
 665 {
 666         geneveh->ver = GENEVE_VER;
 667         geneveh->opt_len = info->options_len / 4;
 668         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 669         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 670         geneveh->rsvd1 = 0;
 671         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 672         geneveh->proto_type = htons(ETH_P_TEB);
 673         geneveh->rsvd2 = 0;
 674
 675         ip_tunnel_info_opts_get(geneveh->options, info);
 676 }
 677
 678 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 679                             const struct ip_tunnel_info *info,
 680                             bool xnet, int ip_hdr_len)
 681 {
 682         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 683         struct genevehdr *gnvh;
 684         int min_headroom;
 685         int err;
 686
 687         skb_reset_mac_header(skb);
 688         skb_scrub_packet(skb, xnet);
 689
 690         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 691                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 692         err = skb_cow_head(skb, min_headroom);
 693         if (unlikely(err))
 694                 goto free_dst;
 695
 696         err = udp_tunnel_handle_offloads(skb, udp_sum);
 697         if (err)
 698                 goto free_dst;
 699
 700         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 701         geneve_build_header(gnvh, info);
 702         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 703         return 0;
 704
 705 free_dst:
 706         dst_release(dst);
 707         return err;
 708 }
 709
 710 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 711                                        struct net_device *dev,
 712                                        struct geneve_sock *gs4,
 713                                        struct flowi4 *fl4,
 714                                        const struct ip_tunnel_info *info)
 715 {
 716         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 717         struct geneve_dev *geneve = netdev_priv(dev);
 718         struct dst_cache *dst_cache;
 719         struct rtable *rt = NULL;
 720         __u8 tos;
 721
 722         if (!gs4)
 723                 return ERR_PTR(-EIO);
 724
 725         memset(fl4, 0, sizeof(*fl4));
 726         fl4->flowi4_mark = skb->mark;
 727         fl4->flowi4_proto = IPPROTO_UDP;
 728         fl4->daddr = info->key.u.ipv4.dst;
 729         fl4->saddr = info->key.u.ipv4.src;
 730
 731         tos = info->key.tos;
 732         if ((tos == 1) && !geneve->collect_md) {
 733                 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 734                 use_cache = false;
 735         }
 736         fl4->flowi4_tos = RT_TOS(tos);
 737
 738         dst_cache = (struct dst_cache *)&info->dst_cache;
 739         if (use_cache) {
 740                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 741                 if (rt)
 742                         return rt;
 743         }
 744         rt = ip_route_output_key(geneve->net, fl4);
 745         if (IS_ERR(rt)) {
 746                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 747                 return ERR_PTR(-ENETUNREACH);
 748         }
 749         if (rt->dst.dev == dev) { /* is this necessary? */
 750                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 751                 ip_rt_put(rt);
 752                 return ERR_PTR(-ELOOP);
 753         }
 754         if (use_cache)
 755                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 756         return rt;
 757 }
 758
 759 #if IS_ENABLED(CONFIG_IPV6)
 760 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 761                                            struct net_device *dev,
 762                                            struct geneve_sock *gs6,
 763                                            struct flowi6 *fl6,
 764                                            const struct ip_tunnel_info *info)
 765 {
 766         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 767         struct geneve_dev *geneve = netdev_priv(dev);
 768         struct dst_entry *dst = NULL;
 769         struct dst_cache *dst_cache;
 770         __u8 prio;
 771
 772         if (!gs6)
 773                 return ERR_PTR(-EIO);
 774
 775         memset(fl6, 0, sizeof(*fl6));
 776         fl6->flowi6_mark = skb->mark;
 777         fl6->flowi6_proto = IPPROTO_UDP;
 778         fl6->daddr = info->key.u.ipv6.dst;
 779         fl6->saddr = info->key.u.ipv6.src;
 780         prio = info->key.tos;
 781         if ((prio == 1) && !geneve->collect_md) {
 782                 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 783                 use_cache = false;
 784         }
 785
 786         fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 787                                            info->key.label);
 788         dst_cache = (struct dst_cache *)&info->dst_cache;
 789         if (use_cache) {
 790                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 791                 if (dst)
 792                         return dst;
 793         }
 794         if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
 795                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 796                 return ERR_PTR(-ENETUNREACH);
 797         }
 798         if (dst->dev == dev) { /* is this necessary? */
 799                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 800                 dst_release(dst);
 801                 return ERR_PTR(-ELOOP);
 802         }
 803
 804         if (use_cache)
 805                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 806         return dst;
 807 }
 808 #endif
 809
 810 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 811                            struct geneve_dev *geneve,
 812                            const struct ip_tunnel_info *info)
 813 {
 814         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 815         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 816         const struct ip_tunnel_key *key = &info->key;
 817         struct rtable *rt;
 818         struct flowi4 fl4;
 819         __u8 tos, ttl;
 820         __be16 sport;
 821         __be16 df;
 822         int err;
 823
 824         rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
 825         if (IS_ERR(rt))
 826                 return PTR_ERR(rt);
 827
 828         if (skb_dst(skb)) {
 829                 int mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr) -
 830                           GENEVE_BASE_HLEN - info->options_len - 14;
 831
 832                 skb_dst_update_pmtu(skb, mtu);
 833         }
 834
 835         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 836         if (geneve->collect_md) {
 837                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 838                 ttl = key->ttl;
 839         } else {
 840                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 841                 ttl = key->ttl ? : ip4_dst_hoplimit(&rt->dst);
 842         }
 843         df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 844
 845         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
 846         if (unlikely(err))
 847                 return err;
 848
 849         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 850                             tos, ttl, df, sport, geneve->info.key.tp_dst,
 851                             !net_eq(geneve->net, dev_net(geneve->dev)),
 852                             !(info->key.tun_flags & TUNNEL_CSUM));
 853         return 0;
 854 }
 855
 856 #if IS_ENABLED(CONFIG_IPV6)
 857 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 858                             struct geneve_dev *geneve,
 859                             const struct ip_tunnel_info *info)
 860 {
 861         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 862         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 863         const struct ip_tunnel_key *key = &info->key;
 864         struct dst_entry *dst = NULL;
 865         struct flowi6 fl6;
 866         __u8 prio, ttl;
 867         __be16 sport;
 868         int err;
 869
 870         dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
 871         if (IS_ERR(dst))
 872                 return PTR_ERR(dst);
 873
 874         if (skb_dst(skb)) {
 875                 int mtu = dst_mtu(dst) - sizeof(struct ipv6hdr) -
 876                           GENEVE_BASE_HLEN - info->options_len - 14;
 877
 878                 skb_dst_update_pmtu(skb, mtu);
 879         }
 880
 881         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 882         if (geneve->collect_md) {
 883                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 884                 ttl = key->ttl;
 885         } else {
 886                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
 887                                            ip_hdr(skb), skb);
 888                 ttl = key->ttl ? : ip6_dst_hoplimit(dst);
 889         }
 890         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
 891         if (unlikely(err))
 892                 return err;
 893
 894         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
 895                              &fl6.saddr, &fl6.daddr, prio, ttl,
 896                              info->key.label, sport, geneve->info.key.tp_dst,
 897                              !(info->key.tun_flags & TUNNEL_CSUM));
 898         return 0;
 899 }
 900 #endif
 901
 902 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
 903 {
 904         struct geneve_dev *geneve = netdev_priv(dev);
 905         struct ip_tunnel_info *info = NULL;
 906         int err;
 907
 908         if (geneve->collect_md) {
 909                 info = skb_tunnel_info(skb);
 910                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
 911                         err = -EINVAL;
 912                         netdev_dbg(dev, "no tunnel metadata\n");
 913                         goto tx_error;
 914                 }
 915         } else {
 916                 info = &geneve->info;
 917         }
 918
 919         rcu_read_lock();
 920 #if IS_ENABLED(CONFIG_IPV6)
 921         if (info->mode & IP_TUNNEL_INFO_IPV6)
 922                 err = geneve6_xmit_skb(skb, dev, geneve, info);
 923         else
 924 #endif
 925                 err = geneve_xmit_skb(skb, dev, geneve, info);
 926         rcu_read_unlock();
 927
 928         if (likely(!err))
 929                 return NETDEV_TX_OK;
 930 tx_error:
 931         dev_kfree_skb(skb);
 932
 933         if (err == -ELOOP)
 934                 dev->stats.collisions++;
 935         else if (err == -ENETUNREACH)
 936                 dev->stats.tx_carrier_errors++;
 937
 938         dev->stats.tx_errors++;
 939         return NETDEV_TX_OK;
 940 }
 941
 942 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
 943 {
 944         /* Only possible if called internally, ndo_change_mtu path's new_mtu
 945          * is guaranteed to be between dev->min_mtu and dev->max_mtu.
 946          */
 947         if (new_mtu > dev->max_mtu)
 948                 new_mtu = dev->max_mtu;
 949
 950         dev->mtu = new_mtu;
 951         return 0;
 952 }
 953
 954 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
 955 {
 956         struct ip_tunnel_info *info = skb_tunnel_info(skb);
 957         struct geneve_dev *geneve = netdev_priv(dev);
 958
 959         if (ip_tunnel_info_af(info) == AF_INET) {
 960                 struct rtable *rt;
 961                 struct flowi4 fl4;
 962                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 963
 964                 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
 965                 if (IS_ERR(rt))
 966                         return PTR_ERR(rt);
 967
 968                 ip_rt_put(rt);
 969                 info->key.u.ipv4.src = fl4.saddr;
 970 #if IS_ENABLED(CONFIG_IPV6)
 971         } else if (ip_tunnel_info_af(info) == AF_INET6) {
 972                 struct dst_entry *dst;
 973                 struct flowi6 fl6;
 974                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 975
 976                 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
 977                 if (IS_ERR(dst))
 978                         return PTR_ERR(dst);
 979
 980                 dst_release(dst);
 981                 info->key.u.ipv6.src = fl6.saddr;
 982 #endif
 983         } else {
 984                 return -EINVAL;
 985         }
 986
 987         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
 988                                              1, USHRT_MAX, true);
 989         info->key.tp_dst = geneve->info.key.tp_dst;
 990         return 0;
 991 }
 992
 993 static const struct net_device_ops geneve_netdev_ops = {
 994         .ndo_init               = geneve_init,
 995         .ndo_uninit             = geneve_uninit,
 996         .ndo_open               = geneve_open,
 997         .ndo_stop               = geneve_stop,
 998         .ndo_start_xmit         = geneve_xmit,
 999         .ndo_get_stats64        = ip_tunnel_get_stats64,
1000         .ndo_change_mtu         = geneve_change_mtu,
1001         .ndo_validate_addr      = eth_validate_addr,
1002         .ndo_set_mac_address    = eth_mac_addr,
1003         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1004 };
1005
1006 static void geneve_get_drvinfo(struct net_device *dev,
1007                                struct ethtool_drvinfo *drvinfo)
1008 {
1009         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1010         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1011 }
1012
1013 static const struct ethtool_ops geneve_ethtool_ops = {
1014         .get_drvinfo    = geneve_get_drvinfo,
1015         .get_link       = ethtool_op_get_link,
1016 };
1017
1018 /* Info for udev, that this is a virtual tunnel endpoint */
1019 static struct device_type geneve_type = {
1020         .name = "geneve",
1021 };
1022
1023 /* Calls the ndo_udp_tunnel_add of the caller in order to
1024  * supply the listening GENEVE udp ports. Callers are expected
1025  * to implement the ndo_udp_tunnel_add.
1026  */
1027 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1028 {
1029         struct net *net = dev_net(dev);
1030         struct geneve_net *gn = net_generic(net, geneve_net_id);
1031         struct geneve_sock *gs;
1032
1033         rcu_read_lock();
1034         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1035                 if (push) {
1036                         udp_tunnel_push_rx_port(dev, gs->sock,
1037                                                 UDP_TUNNEL_TYPE_GENEVE);
1038                 } else {
1039                         udp_tunnel_drop_rx_port(dev, gs->sock,
1040                                                 UDP_TUNNEL_TYPE_GENEVE);
1041                 }
1042         }
1043         rcu_read_unlock();
1044 }
1045
1046 /* Initialize the device structure. */
1047 static void geneve_setup(struct net_device *dev)
1048 {
1049         ether_setup(dev);
1050
1051         dev->netdev_ops = &geneve_netdev_ops;
1052         dev->ethtool_ops = &geneve_ethtool_ops;
1053         dev->needs_free_netdev = true;
1054
1055         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1056
1057         dev->features    |= NETIF_F_LLTX;
1058         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1059         dev->features    |= NETIF_F_RXCSUM;
1060         dev->features    |= NETIF_F_GSO_SOFTWARE;
1061
1062         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1063         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1064
1065         /* MTU range: 68 - (something less than 65535) */
1066         dev->min_mtu = ETH_MIN_MTU;
1067         /* The max_mtu calculation does not take account of GENEVE
1068          * options, to avoid excluding potentially valid
1069          * configurations. This will be further reduced by IPvX hdr size.
1070          */
1071         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1072
1073         netif_keep_dst(dev);
1074         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1075         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1076         eth_hw_addr_random(dev);
1077 }
1078
1079 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1080         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1081         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1082         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1083         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1084         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1085         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1086         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1087         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1088         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1089         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1090         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1091 };
1092
1093 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1094                            struct netlink_ext_ack *extack)
1095 {
1096         if (tb[IFLA_ADDRESS]) {
1097                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1098                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1099                                             "Provided link layer address is not Ethernet");
1100                         return -EINVAL;
1101                 }
1102
1103                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1104                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1105                                             "Provided Ethernet address is not unicast");
1106                         return -EADDRNOTAVAIL;
1107                 }
1108         }
1109
1110         if (!data) {
1111                 NL_SET_ERR_MSG(extack,
1112                                "Not enough attributes provided to perform the operation");
1113                 return -EINVAL;
1114         }
1115
1116         if (data[IFLA_GENEVE_ID]) {
1117                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1118
1119                 if (vni >= GENEVE_N_VID) {
1120                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1121                                             "Geneve ID must be lower than 16777216");
1122                         return -ERANGE;
1123                 }
1124         }
1125
1126         return 0;
1127 }
1128
1129 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1130                                           const struct ip_tunnel_info *info,
1131                                           bool *tun_on_same_port,
1132                                           bool *tun_collect_md)
1133 {
1134         struct geneve_dev *geneve, *t = NULL;
1135
1136         *tun_on_same_port = false;
1137         *tun_collect_md = false;
1138         list_for_each_entry(geneve, &gn->geneve_list, next) {
1139                 if (info->key.tp_dst == geneve->info.key.tp_dst) {
1140                         *tun_collect_md = geneve->collect_md;
1141                         *tun_on_same_port = true;
1142                 }
1143                 if (info->key.tun_id == geneve->info.key.tun_id &&
1144                     info->key.tp_dst == geneve->info.key.tp_dst &&
1145                     !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1146                         t = geneve;
1147         }
1148         return t;
1149 }
1150
1151 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1152 {
1153         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1154                  info->key.ttl || info->key.label || info->key.tp_src ||
1155                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1156 }
1157
1158 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1159                                   struct ip_tunnel_info *b)
1160 {
1161         if (ip_tunnel_info_af(a) == AF_INET)
1162                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1163         else
1164                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1165 }
1166
1167 static int geneve_configure(struct net *net, struct net_device *dev,
1168                             struct netlink_ext_ack *extack,
1169                             const struct ip_tunnel_info *info,
1170                             bool metadata, bool ipv6_rx_csum)
1171 {
1172         struct geneve_net *gn = net_generic(net, geneve_net_id);
1173         struct geneve_dev *t, *geneve = netdev_priv(dev);
1174         bool tun_collect_md, tun_on_same_port;
1175         int err, encap_len;
1176
1177         if (metadata && !is_tnl_info_zero(info)) {
1178                 NL_SET_ERR_MSG(extack,
1179                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1180                 return -EINVAL;
1181         }
1182
1183         geneve->net = net;
1184         geneve->dev = dev;
1185
1186         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1187         if (t)
1188                 return -EBUSY;
1189
1190         /* make enough headroom for basic scenario */
1191         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1192         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1193                 encap_len += sizeof(struct iphdr);
1194                 dev->max_mtu -= sizeof(struct iphdr);
1195         } else {
1196                 encap_len += sizeof(struct ipv6hdr);
1197                 dev->max_mtu -= sizeof(struct ipv6hdr);
1198         }
1199         dev->needed_headroom = encap_len + ETH_HLEN;
1200
1201         if (metadata) {
1202                 if (tun_on_same_port) {
1203                         NL_SET_ERR_MSG(extack,
1204                                        "There can be only one externally controlled device on a destination port");
1205                         return -EPERM;
1206                 }
1207         } else {
1208                 if (tun_collect_md) {
1209                         NL_SET_ERR_MSG(extack,
1210                                        "There already exists an externally controlled device on this destination port");
1211                         return -EPERM;
1212                 }
1213         }
1214
1215         dst_cache_reset(&geneve->info.dst_cache);
1216         geneve->info = *info;
1217         geneve->collect_md = metadata;
1218         geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1219
1220         err = register_netdevice(dev);
1221         if (err)
1222                 return err;
1223
1224         list_add(&geneve->next, &gn->geneve_list);
1225         return 0;
1226 }
1227
1228 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1229 {
1230         memset(info, 0, sizeof(*info));
1231         info->key.tp_dst = htons(dst_port);
1232 }
1233
1234 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1235                           struct netlink_ext_ack *extack,
1236                           struct ip_tunnel_info *info, bool *metadata,
1237                           bool *use_udp6_rx_checksums, bool changelink)
1238 {
1239         int attrtype;
1240
1241         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1242                 NL_SET_ERR_MSG(extack,
1243                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1244                 return -EINVAL;
1245         }
1246
1247         if (data[IFLA_GENEVE_REMOTE]) {
1248                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1249                         attrtype = IFLA_GENEVE_REMOTE;
1250                         goto change_notsup;
1251                 }
1252
1253                 info->key.u.ipv4.dst =
1254                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1255
1256                 if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) {
1257                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1258                                             "Remote IPv4 address cannot be Multicast");
1259                         return -EINVAL;
1260                 }
1261         }
1262
1263         if (data[IFLA_GENEVE_REMOTE6]) {
1264  #if IS_ENABLED(CONFIG_IPV6)
1265                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1266                         attrtype = IFLA_GENEVE_REMOTE6;
1267                         goto change_notsup;
1268                 }
1269
1270                 info->mode = IP_TUNNEL_INFO_IPV6;
1271                 info->key.u.ipv6.dst =
1272                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1273
1274                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1275                     IPV6_ADDR_LINKLOCAL) {
1276                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1277                                             "Remote IPv6 address cannot be link-local");
1278                         return -EINVAL;
1279                 }
1280                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1281                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1282                                             "Remote IPv6 address cannot be Multicast");
1283                         return -EINVAL;
1284                 }
1285                 info->key.tun_flags |= TUNNEL_CSUM;
1286                 *use_udp6_rx_checksums = true;
1287 #else
1288                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1289                                     "IPv6 support not enabled in the kernel");
1290                 return -EPFNOSUPPORT;
1291 #endif
1292         }
1293
1294         if (data[IFLA_GENEVE_ID]) {
1295                 __u32 vni;
1296                 __u8 tvni[3];
1297                 __be64 tunid;
1298
1299                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1300                 tvni[0] = (vni & 0x00ff0000) >> 16;
1301                 tvni[1] = (vni & 0x0000ff00) >> 8;
1302                 tvni[2] =  vni & 0x000000ff;
1303
1304                 tunid = vni_to_tunnel_id(tvni);
1305                 if (changelink && (tunid != info->key.tun_id)) {
1306                         attrtype = IFLA_GENEVE_ID;
1307                         goto change_notsup;
1308                 }
1309                 info->key.tun_id = tunid;
1310         }
1311
1312         if (data[IFLA_GENEVE_TTL])
1313                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1314
1315         if (data[IFLA_GENEVE_TOS])
1316                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1317
1318         if (data[IFLA_GENEVE_LABEL]) {
1319                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1320                                   IPV6_FLOWLABEL_MASK;
1321                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1322                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1323                                             "Label attribute only applies for IPv6 Geneve devices");
1324                         return -EINVAL;
1325                 }
1326         }
1327
1328         if (data[IFLA_GENEVE_PORT]) {
1329                 if (changelink) {
1330                         attrtype = IFLA_GENEVE_PORT;
1331                         goto change_notsup;
1332                 }
1333                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1334         }
1335
1336         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1337                 if (changelink) {
1338                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1339                         goto change_notsup;
1340                 }
1341                 *metadata = true;
1342         }
1343
1344         if (data[IFLA_GENEVE_UDP_CSUM]) {
1345                 if (changelink) {
1346                         attrtype = IFLA_GENEVE_UDP_CSUM;
1347                         goto change_notsup;
1348                 }
1349                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1350                         info->key.tun_flags |= TUNNEL_CSUM;
1351         }
1352
1353         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1354 #if IS_ENABLED(CONFIG_IPV6)
1355                 if (changelink) {
1356                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1357                         goto change_notsup;
1358                 }
1359                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1360                         info->key.tun_flags &= ~TUNNEL_CSUM;
1361 #else
1362                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1363                                     "IPv6 support not enabled in the kernel");
1364                 return -EPFNOSUPPORT;
1365 #endif
1366         }
1367
1368         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1369 #if IS_ENABLED(CONFIG_IPV6)
1370                 if (changelink) {
1371                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1372                         goto change_notsup;
1373                 }
1374                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1375                         *use_udp6_rx_checksums = false;
1376 #else
1377                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1378                                     "IPv6 support not enabled in the kernel");
1379                 return -EPFNOSUPPORT;
1380 #endif
1381         }
1382
1383         return 0;
1384 change_notsup:
1385         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1386                             "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1387         return -EOPNOTSUPP;
1388 }
1389
1390 static int geneve_newlink(struct net *net, struct net_device *dev,
1391                           struct nlattr *tb[], struct nlattr *data[],
1392                           struct netlink_ext_ack *extack)
1393 {
1394         bool use_udp6_rx_checksums = false;
1395         struct ip_tunnel_info info;
1396         bool metadata = false;
1397         int err;
1398
1399         init_tnl_info(&info, GENEVE_UDP_PORT);
1400         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1401                              &use_udp6_rx_checksums, false);
1402         if (err)
1403                 return err;
1404
1405         return geneve_configure(net, dev, extack, &info, metadata,
1406                                 use_udp6_rx_checksums);
1407 }
1408
1409 /* Quiesces the geneve device data path for both TX and RX.
1410  *
1411  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1412  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1413  * to complete for the existing set of in-flight packets to be transmitted,
1414  * then we would have quiesced the transmit data path. All the future packets
1415  * will get dropped until we unquiesce the data path.
1416  *
1417  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1418  * if we set that to NULL under RCU and wait for synchronize_net() to
1419  * complete, then we would have quiesced the receive data path.
1420  */
1421 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1422                            struct geneve_sock **gs6)
1423 {
1424         *gs4 = rtnl_dereference(geneve->sock4);
1425         rcu_assign_pointer(geneve->sock4, NULL);
1426         if (*gs4)
1427                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1428 #if IS_ENABLED(CONFIG_IPV6)
1429         *gs6 = rtnl_dereference(geneve->sock6);
1430         rcu_assign_pointer(geneve->sock6, NULL);
1431         if (*gs6)
1432                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1433 #else
1434         *gs6 = NULL;
1435 #endif
1436         synchronize_net();
1437 }
1438
1439 /* Resumes the geneve device data path for both TX and RX. */
1440 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1441                              struct geneve_sock __maybe_unused *gs6)
1442 {
1443         rcu_assign_pointer(geneve->sock4, gs4);
1444         if (gs4)
1445                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1446 #if IS_ENABLED(CONFIG_IPV6)
1447         rcu_assign_pointer(geneve->sock6, gs6);
1448         if (gs6)
1449                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1450 #endif
1451         synchronize_net();
1452 }
1453
1454 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1455                              struct nlattr *data[],
1456                              struct netlink_ext_ack *extack)
1457 {
1458         struct geneve_dev *geneve = netdev_priv(dev);
1459         struct geneve_sock *gs4, *gs6;
1460         struct ip_tunnel_info info;
1461         bool metadata;
1462         bool use_udp6_rx_checksums;
1463         int err;
1464
1465         /* If the geneve device is configured for metadata (or externally
1466          * controlled, for example, OVS), then nothing can be changed.
1467          */
1468         if (geneve->collect_md)
1469                 return -EOPNOTSUPP;
1470
1471         /* Start with the existing info. */
1472         memcpy(&info, &geneve->info, sizeof(info));
1473         metadata = geneve->collect_md;
1474         use_udp6_rx_checksums = geneve->use_udp6_rx_checksums;
1475         err = geneve_nl2info(tb, data, extack, &info, &metadata,
1476                              &use_udp6_rx_checksums, true);
1477         if (err)
1478                 return err;
1479
1480         if (!geneve_dst_addr_equal(&geneve->info, &info))
1481                 dst_cache_reset(&info.dst_cache);
1482
1483         geneve_quiesce(geneve, &gs4, &gs6);
1484         geneve->info = info;
1485         geneve->collect_md = metadata;
1486         geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
1487         geneve_unquiesce(geneve, gs4, gs6);
1488
1489         return 0;
1490 }
1491
1492 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1493 {
1494         struct geneve_dev *geneve = netdev_priv(dev);
1495
1496         list_del(&geneve->next);
1497         unregister_netdevice_queue(dev, head);
1498 }
1499
1500 static size_t geneve_get_size(const struct net_device *dev)
1501 {
1502         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1503                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1504                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1505                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1506                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1507                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1508                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1509                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1510                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1511                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1512                 0;
1513 }
1514
1515 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1516 {
1517         struct geneve_dev *geneve = netdev_priv(dev);
1518         struct ip_tunnel_info *info = &geneve->info;
1519         bool metadata = geneve->collect_md;
1520         __u8 tmp_vni[3];
1521         __u32 vni;
1522
1523         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1524         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1525         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1526                 goto nla_put_failure;
1527
1528         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1529                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1530                                     info->key.u.ipv4.dst))
1531                         goto nla_put_failure;
1532                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1533                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1534                         goto nla_put_failure;
1535
1536 #if IS_ENABLED(CONFIG_IPV6)
1537         } else if (!metadata) {
1538                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1539                                      &info->key.u.ipv6.dst))
1540                         goto nla_put_failure;
1541                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1542                                !(info->key.tun_flags & TUNNEL_CSUM)))
1543                         goto nla_put_failure;
1544 #endif
1545         }
1546
1547         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1548             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1549             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1550                 goto nla_put_failure;
1551
1552         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1553                 goto nla_put_failure;
1554
1555         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1556                 goto nla_put_failure;
1557
1558 #if IS_ENABLED(CONFIG_IPV6)
1559         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1560                        !geneve->use_udp6_rx_checksums))
1561                 goto nla_put_failure;
1562 #endif
1563
1564         return 0;
1565
1566 nla_put_failure:
1567         return -EMSGSIZE;
1568 }
1569
1570 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1571         .kind           = "geneve",
1572         .maxtype        = IFLA_GENEVE_MAX,
1573         .policy         = geneve_policy,
1574         .priv_size      = sizeof(struct geneve_dev),
1575         .setup          = geneve_setup,
1576         .validate       = geneve_validate,
1577         .newlink        = geneve_newlink,
1578         .changelink     = geneve_changelink,
1579         .dellink        = geneve_dellink,
1580         .get_size       = geneve_get_size,
1581         .fill_info      = geneve_fill_info,
1582 };
1583
1584 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1585                                         u8 name_assign_type, u16 dst_port)
1586 {
1587         struct nlattr *tb[IFLA_MAX + 1];
1588         struct ip_tunnel_info info;
1589         struct net_device *dev;
1590         LIST_HEAD(list_kill);
1591         int err;
1592
1593         memset(tb, 0, sizeof(tb));
1594         dev = rtnl_create_link(net, name, name_assign_type,
1595                                &geneve_link_ops, tb);
1596         if (IS_ERR(dev))
1597                 return dev;
1598
1599         init_tnl_info(&info, dst_port);
1600         err = geneve_configure(net, dev, NULL, &info, true, true);
1601         if (err) {
1602                 free_netdev(dev);
1603                 return ERR_PTR(err);
1604         }
1605
1606         /* openvswitch users expect packet sizes to be unrestricted,
1607          * so set the largest MTU we can.
1608          */
1609         err = geneve_change_mtu(dev, IP_MAX_MTU);
1610         if (err)
1611                 goto err;
1612
1613         err = rtnl_configure_link(dev, NULL);
1614         if (err < 0)
1615                 goto err;
1616
1617         return dev;
1618 err:
1619         geneve_dellink(dev, &list_kill);
1620         unregister_netdevice_many(&list_kill);
1621         return ERR_PTR(err);
1622 }
1623 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1624
1625 static int geneve_netdevice_event(struct notifier_block *unused,
1626                                   unsigned long event, void *ptr)
1627 {
1628         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1629
1630         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1631             event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1632                 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1633         } else if (event == NETDEV_UNREGISTER) {
1634                 geneve_offload_rx_ports(dev, false);
1635         } else if (event == NETDEV_REGISTER) {
1636                 geneve_offload_rx_ports(dev, true);
1637         }
1638
1639         return NOTIFY_DONE;
1640 }
1641
1642 static struct notifier_block geneve_notifier_block __read_mostly = {
1643         .notifier_call = geneve_netdevice_event,
1644 };
1645
1646 static __net_init int geneve_init_net(struct net *net)
1647 {
1648         struct geneve_net *gn = net_generic(net, geneve_net_id);
1649
1650         INIT_LIST_HEAD(&gn->geneve_list);
1651         INIT_LIST_HEAD(&gn->sock_list);
1652         return 0;
1653 }
1654
1655 static void __net_exit geneve_exit_net(struct net *net)
1656 {
1657         struct geneve_net *gn = net_generic(net, geneve_net_id);
1658         struct geneve_dev *geneve, *next;
1659         struct net_device *dev, *aux;
1660         LIST_HEAD(list);
1661
1662         rtnl_lock();
1663
1664         /* gather any geneve devices that were moved into this ns */
1665         for_each_netdev_safe(net, dev, aux)
1666                 if (dev->rtnl_link_ops == &geneve_link_ops)
1667                         unregister_netdevice_queue(dev, &list);
1668
1669         /* now gather any other geneve devices that were created in this ns */
1670         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1671                 /* If geneve->dev is in the same netns, it was already added
1672                  * to the list by the previous loop.
1673                  */
1674                 if (!net_eq(dev_net(geneve->dev), net))
1675                         unregister_netdevice_queue(geneve->dev, &list);
1676         }
1677
1678         /* unregister the devices gathered above */
1679         unregister_netdevice_many(&list);
1680         rtnl_unlock();
1681         WARN_ON_ONCE(!list_empty(&gn->sock_list));
1682 }
1683
1684 static struct pernet_operations geneve_net_ops = {
1685         .init = geneve_init_net,
1686         .exit = geneve_exit_net,
1687         .id   = &geneve_net_id,
1688         .size = sizeof(struct geneve_net),
1689 };
1690
1691 static int __init geneve_init_module(void)
1692 {
1693         int rc;
1694
1695         rc = register_pernet_subsys(&geneve_net_ops);
1696         if (rc)
1697                 goto out1;
1698
1699         rc = register_netdevice_notifier(&geneve_notifier_block);
1700         if (rc)
1701                 goto out2;
1702
1703         rc = rtnl_link_register(&geneve_link_ops);
1704         if (rc)
1705                 goto out3;
1706
1707         return 0;
1708 out3:
1709         unregister_netdevice_notifier(&geneve_notifier_block);
1710 out2:
1711         unregister_pernet_subsys(&geneve_net_ops);
1712 out1:
1713         return rc;
1714 }
1715 late_initcall(geneve_init_module);
1716
1717 static void __exit geneve_cleanup_module(void)
1718 {
1719         rtnl_link_unregister(&geneve_link_ops);
1720         unregister_netdevice_notifier(&geneve_notifier_block);
1721         unregister_pernet_subsys(&geneve_net_ops);
1722 }
1723 module_exit(geneve_cleanup_module);
1724
1725 MODULE_LICENSE("GPL");
1726 MODULE_VERSION(GENEVE_NETDEV_VER);
1727 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1728 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1729 MODULE_ALIAS_RTNL_LINK("geneve");