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