drivers/net/geneve.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * GENEVE: Generic Network Virtualization Encapsulation
   4  *
   5  * Copyright (c) 2015 Red Hat, Inc.
   6  */
   7
   8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   9
  10 #include <linux/ethtool.h>
  11 #include <linux/kernel.h>
  12 #include <linux/module.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/hash.h>
  15 #include <net/ipv6_stubs.h>
  16 #include <net/dst_metadata.h>
  17 #include <net/gro_cells.h>
  18 #include <net/rtnetlink.h>
  19 #include <net/geneve.h>
  20 #include <net/gro.h>
  21 #include <net/protocol.h>
  22
  23 #define GENEVE_NETDEV_VER       "0.6"
  24
  25 #define GENEVE_N_VID            (1u << 24)
  26 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  27
  28 #define VNI_HASH_BITS           10
  29 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  30
  31 static bool log_ecn_error = true;
  32 module_param(log_ecn_error, bool, 0644);
  33 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  34
  35 #define GENEVE_VER 0
  36 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  37 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
  38 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
  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 struct geneve_config {
  54         struct ip_tunnel_info   info;
  55         bool                    collect_md;
  56         bool                    use_udp6_rx_checksums;
  57         bool                    ttl_inherit;
  58         enum ifla_geneve_df     df;
  59         bool                    inner_proto_inherit;
  60 };
  61
  62 /* Pseudo network device */
  63 struct geneve_dev {
  64         struct geneve_dev_node hlist4;  /* vni hash table for IPv4 socket */
  65 #if IS_ENABLED(CONFIG_IPV6)
  66         struct geneve_dev_node hlist6;  /* vni hash table for IPv6 socket */
  67 #endif
  68         struct net         *net;        /* netns for packet i/o */
  69         struct net_device  *dev;        /* netdev for geneve tunnel */
  70         struct geneve_sock __rcu *sock4;        /* IPv4 socket used for geneve tunnel */
  71 #if IS_ENABLED(CONFIG_IPV6)
  72         struct geneve_sock __rcu *sock6;        /* IPv6 socket used for geneve tunnel */
  73 #endif
  74         struct list_head   next;        /* geneve's per namespace list */
  75         struct gro_cells   gro_cells;
  76         struct geneve_config cfg;
  77 };
  78
  79 struct geneve_sock {
  80         bool                    collect_md;
  81         struct list_head        list;
  82         struct socket           *sock;
  83         struct rcu_head         rcu;
  84         int                     refcnt;
  85         struct hlist_head       vni_list[VNI_HASH_SIZE];
  86 };
  87
  88 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  89 {
  90         __u32 vnid;
  91
  92         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  93         return hash_32(vnid, VNI_HASH_BITS);
  94 }
  95
  96 static __be64 vni_to_tunnel_id(const __u8 *vni)
  97 {
  98 #ifdef __BIG_ENDIAN
  99         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
 100 #else
 101         return (__force __be64)(((__force u64)vni[0] << 40) |
 102                                 ((__force u64)vni[1] << 48) |
 103                                 ((__force u64)vni[2] << 56));
 104 #endif
 105 }
 106
 107 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 108 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 109 {
 110 #ifdef __BIG_ENDIAN
 111         vni[0] = (__force __u8)(tun_id >> 16);
 112         vni[1] = (__force __u8)(tun_id >> 8);
 113         vni[2] = (__force __u8)tun_id;
 114 #else
 115         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 116         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 117         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 118 #endif
 119 }
 120
 121 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
 122 {
 123         return !memcmp(vni, &tun_id[5], 3);
 124 }
 125
 126 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 127 {
 128         return gs->sock->sk->sk_family;
 129 }
 130
 131 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 132                                         __be32 addr, u8 vni[])
 133 {
 134         struct hlist_head *vni_list_head;
 135         struct geneve_dev_node *node;
 136         __u32 hash;
 137
 138         /* Find the device for this VNI */
 139         hash = geneve_net_vni_hash(vni);
 140         vni_list_head = &gs->vni_list[hash];
 141         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 142                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 143                     addr == node->geneve->cfg.info.key.u.ipv4.dst)
 144                         return node->geneve;
 145         }
 146         return NULL;
 147 }
 148
 149 #if IS_ENABLED(CONFIG_IPV6)
 150 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 151                                          struct in6_addr addr6, u8 vni[])
 152 {
 153         struct hlist_head *vni_list_head;
 154         struct geneve_dev_node *node;
 155         __u32 hash;
 156
 157         /* Find the device for this VNI */
 158         hash = geneve_net_vni_hash(vni);
 159         vni_list_head = &gs->vni_list[hash];
 160         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 161                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 162                     ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
 163                         return node->geneve;
 164         }
 165         return NULL;
 166 }
 167 #endif
 168
 169 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 170 {
 171         return (struct genevehdr *)(udp_hdr(skb) + 1);
 172 }
 173
 174 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 175                                             struct sk_buff *skb)
 176 {
 177         static u8 zero_vni[3];
 178         u8 *vni;
 179
 180         if (geneve_get_sk_family(gs) == AF_INET) {
 181                 struct iphdr *iph;
 182                 __be32 addr;
 183
 184                 iph = ip_hdr(skb); /* outer IP header... */
 185
 186                 if (gs->collect_md) {
 187                         vni = zero_vni;
 188                         addr = 0;
 189                 } else {
 190                         vni = geneve_hdr(skb)->vni;
 191                         addr = iph->saddr;
 192                 }
 193
 194                 return geneve_lookup(gs, addr, vni);
 195 #if IS_ENABLED(CONFIG_IPV6)
 196         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 197                 static struct in6_addr zero_addr6;
 198                 struct ipv6hdr *ip6h;
 199                 struct in6_addr addr6;
 200
 201                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 202
 203                 if (gs->collect_md) {
 204                         vni = zero_vni;
 205                         addr6 = zero_addr6;
 206                 } else {
 207                         vni = geneve_hdr(skb)->vni;
 208                         addr6 = ip6h->saddr;
 209                 }
 210
 211                 return geneve6_lookup(gs, addr6, vni);
 212 #endif
 213         }
 214         return NULL;
 215 }
 216
 217 /* geneve receive/decap routine */
 218 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 219                       struct sk_buff *skb)
 220 {
 221         struct genevehdr *gnvh = geneve_hdr(skb);
 222         struct metadata_dst *tun_dst = NULL;
 223         unsigned int len;
 224         int err = 0;
 225         void *oiph;
 226
 227         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 228                 __be16 flags;
 229
 230                 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
 231                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 232
 233                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 234                                          vni_to_tunnel_id(gnvh->vni),
 235                                          gnvh->opt_len * 4);
 236                 if (!tun_dst) {
 237                         geneve->dev->stats.rx_dropped++;
 238                         goto drop;
 239                 }
 240                 /* Update tunnel dst according to Geneve options. */
 241                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 242                                         gnvh->options, gnvh->opt_len * 4,
 243                                         TUNNEL_GENEVE_OPT);
 244         } else {
 245                 /* Drop packets w/ critical options,
 246                  * since we don't support any...
 247                  */
 248                 if (gnvh->critical) {
 249                         geneve->dev->stats.rx_frame_errors++;
 250                         geneve->dev->stats.rx_errors++;
 251                         goto drop;
 252                 }
 253         }
 254
 255         if (tun_dst)
 256                 skb_dst_set(skb, &tun_dst->dst);
 257
 258         if (gnvh->proto_type == htons(ETH_P_TEB)) {
 259                 skb_reset_mac_header(skb);
 260                 skb->protocol = eth_type_trans(skb, geneve->dev);
 261                 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 262
 263                 /* Ignore packet loops (and multicast echo) */
 264                 if (ether_addr_equal(eth_hdr(skb)->h_source,
 265                                      geneve->dev->dev_addr)) {
 266                         geneve->dev->stats.rx_errors++;
 267                         goto drop;
 268                 }
 269         } else {
 270                 skb_reset_mac_header(skb);
 271                 skb->dev = geneve->dev;
 272                 skb->pkt_type = PACKET_HOST;
 273         }
 274
 275         oiph = skb_network_header(skb);
 276         skb_reset_network_header(skb);
 277
 278         if (geneve_get_sk_family(gs) == AF_INET)
 279                 err = IP_ECN_decapsulate(oiph, skb);
 280 #if IS_ENABLED(CONFIG_IPV6)
 281         else
 282                 err = IP6_ECN_decapsulate(oiph, skb);
 283 #endif
 284
 285         if (unlikely(err)) {
 286                 if (log_ecn_error) {
 287                         if (geneve_get_sk_family(gs) == AF_INET)
 288                                 net_info_ratelimited("non-ECT from %pI4 "
 289                                                      "with TOS=%#x\n",
 290                                                      &((struct iphdr *)oiph)->saddr,
 291                                                      ((struct iphdr *)oiph)->tos);
 292 #if IS_ENABLED(CONFIG_IPV6)
 293                         else
 294                                 net_info_ratelimited("non-ECT from %pI6\n",
 295                                                      &((struct ipv6hdr *)oiph)->saddr);
 296 #endif
 297                 }
 298                 if (err > 1) {
 299                         ++geneve->dev->stats.rx_frame_errors;
 300                         ++geneve->dev->stats.rx_errors;
 301                         goto drop;
 302                 }
 303         }
 304
 305         len = skb->len;
 306         err = gro_cells_receive(&geneve->gro_cells, skb);
 307         if (likely(err == NET_RX_SUCCESS))
 308                 dev_sw_netstats_rx_add(geneve->dev, len);
 309
 310         return;
 311 drop:
 312         /* Consume bad packet */
 313         kfree_skb(skb);
 314 }
 315
 316 /* Setup stats when device is created */
 317 static int geneve_init(struct net_device *dev)
 318 {
 319         struct geneve_dev *geneve = netdev_priv(dev);
 320         int err;
 321
 322         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 323         if (!dev->tstats)
 324                 return -ENOMEM;
 325
 326         err = gro_cells_init(&geneve->gro_cells, dev);
 327         if (err) {
 328                 free_percpu(dev->tstats);
 329                 return err;
 330         }
 331
 332         err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
 333         if (err) {
 334                 free_percpu(dev->tstats);
 335                 gro_cells_destroy(&geneve->gro_cells);
 336                 return err;
 337         }
 338         return 0;
 339 }
 340
 341 static void geneve_uninit(struct net_device *dev)
 342 {
 343         struct geneve_dev *geneve = netdev_priv(dev);
 344
 345         dst_cache_destroy(&geneve->cfg.info.dst_cache);
 346         gro_cells_destroy(&geneve->gro_cells);
 347         free_percpu(dev->tstats);
 348 }
 349
 350 /* Callback from net/ipv4/udp.c to receive packets */
 351 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 352 {
 353         struct genevehdr *geneveh;
 354         struct geneve_dev *geneve;
 355         struct geneve_sock *gs;
 356         __be16 inner_proto;
 357         int opts_len;
 358
 359         /* Need UDP and Geneve header to be present */
 360         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 361                 goto drop;
 362
 363         /* Return packets with reserved bits set */
 364         geneveh = geneve_hdr(skb);
 365         if (unlikely(geneveh->ver != GENEVE_VER))
 366                 goto drop;
 367
 368         gs = rcu_dereference_sk_user_data(sk);
 369         if (!gs)
 370                 goto drop;
 371
 372         geneve = geneve_lookup_skb(gs, skb);
 373         if (!geneve)
 374                 goto drop;
 375
 376         inner_proto = geneveh->proto_type;
 377
 378         if (unlikely((!geneve->cfg.inner_proto_inherit &&
 379                       inner_proto != htons(ETH_P_TEB)))) {
 380                 geneve->dev->stats.rx_dropped++;
 381                 goto drop;
 382         }
 383
 384         opts_len = geneveh->opt_len * 4;
 385         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto,
 386                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 387                 geneve->dev->stats.rx_dropped++;
 388                 goto drop;
 389         }
 390
 391         geneve_rx(geneve, gs, skb);
 392         return 0;
 393
 394 drop:
 395         /* Consume bad packet */
 396         kfree_skb(skb);
 397         return 0;
 398 }
 399
 400 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
 401 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
 402 {
 403         struct genevehdr *geneveh;
 404         struct geneve_sock *gs;
 405         u8 zero_vni[3] = { 0 };
 406         u8 *vni = zero_vni;
 407
 408         if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
 409                 return -EINVAL;
 410
 411         geneveh = geneve_hdr(skb);
 412         if (geneveh->ver != GENEVE_VER)
 413                 return -EINVAL;
 414
 415         if (geneveh->proto_type != htons(ETH_P_TEB))
 416                 return -EINVAL;
 417
 418         gs = rcu_dereference_sk_user_data(sk);
 419         if (!gs)
 420                 return -ENOENT;
 421
 422         if (geneve_get_sk_family(gs) == AF_INET) {
 423                 struct iphdr *iph = ip_hdr(skb);
 424                 __be32 addr4 = 0;
 425
 426                 if (!gs->collect_md) {
 427                         vni = geneve_hdr(skb)->vni;
 428                         addr4 = iph->daddr;
 429                 }
 430
 431                 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
 432         }
 433
 434 #if IS_ENABLED(CONFIG_IPV6)
 435         if (geneve_get_sk_family(gs) == AF_INET6) {
 436                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 437                 struct in6_addr addr6;
 438
 439                 memset(&addr6, 0, sizeof(struct in6_addr));
 440
 441                 if (!gs->collect_md) {
 442                         vni = geneve_hdr(skb)->vni;
 443                         addr6 = ip6h->daddr;
 444                 }
 445
 446                 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
 447         }
 448 #endif
 449
 450         return -EPFNOSUPPORT;
 451 }
 452
 453 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 454                                          __be16 port, bool ipv6_rx_csum)
 455 {
 456         struct socket *sock;
 457         struct udp_port_cfg udp_conf;
 458         int err;
 459
 460         memset(&udp_conf, 0, sizeof(udp_conf));
 461
 462         if (ipv6) {
 463                 udp_conf.family = AF_INET6;
 464                 udp_conf.ipv6_v6only = 1;
 465                 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
 466         } else {
 467                 udp_conf.family = AF_INET;
 468                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 469         }
 470
 471         udp_conf.local_udp_port = port;
 472
 473         /* Open UDP socket */
 474         err = udp_sock_create(net, &udp_conf, &sock);
 475         if (err < 0)
 476                 return ERR_PTR(err);
 477
 478         udp_allow_gso(sock->sk);
 479         return sock;
 480 }
 481
 482 static int geneve_hlen(struct genevehdr *gh)
 483 {
 484         return sizeof(*gh) + gh->opt_len * 4;
 485 }
 486
 487 static struct sk_buff *geneve_gro_receive(struct sock *sk,
 488                                           struct list_head *head,
 489                                           struct sk_buff *skb)
 490 {
 491         struct sk_buff *pp = NULL;
 492         struct sk_buff *p;
 493         struct genevehdr *gh, *gh2;
 494         unsigned int hlen, gh_len, off_gnv;
 495         const struct packet_offload *ptype;
 496         __be16 type;
 497         int flush = 1;
 498
 499         off_gnv = skb_gro_offset(skb);
 500         hlen = off_gnv + sizeof(*gh);
 501         gh = skb_gro_header(skb, hlen, off_gnv);
 502         if (unlikely(!gh))
 503                 goto out;
 504
 505         if (gh->ver != GENEVE_VER || gh->oam)
 506                 goto out;
 507         gh_len = geneve_hlen(gh);
 508
 509         hlen = off_gnv + gh_len;
 510         if (skb_gro_header_hard(skb, hlen)) {
 511                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 512                 if (unlikely(!gh))
 513                         goto out;
 514         }
 515
 516         list_for_each_entry(p, head, list) {
 517                 if (!NAPI_GRO_CB(p)->same_flow)
 518                         continue;
 519
 520                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 521                 if (gh->opt_len != gh2->opt_len ||
 522                     memcmp(gh, gh2, gh_len)) {
 523                         NAPI_GRO_CB(p)->same_flow = 0;
 524                         continue;
 525                 }
 526         }
 527
 528         skb_gro_pull(skb, gh_len);
 529         skb_gro_postpull_rcsum(skb, gh, gh_len);
 530         type = gh->proto_type;
 531         if (likely(type == htons(ETH_P_TEB)))
 532                 return call_gro_receive(eth_gro_receive, head, skb);
 533
 534         ptype = gro_find_receive_by_type(type);
 535         if (!ptype)
 536                 goto out;
 537
 538         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 539         flush = 0;
 540
 541 out:
 542         skb_gro_flush_final(skb, pp, flush);
 543
 544         return pp;
 545 }
 546
 547 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 548                                int nhoff)
 549 {
 550         struct genevehdr *gh;
 551         struct packet_offload *ptype;
 552         __be16 type;
 553         int gh_len;
 554         int err = -ENOSYS;
 555
 556         gh = (struct genevehdr *)(skb->data + nhoff);
 557         gh_len = geneve_hlen(gh);
 558         type = gh->proto_type;
 559
 560         /* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */
 561         if (likely(type == htons(ETH_P_TEB)))
 562                 return eth_gro_complete(skb, nhoff + gh_len);
 563
 564         ptype = gro_find_complete_by_type(type);
 565         if (ptype)
 566                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 567
 568         skb_set_inner_mac_header(skb, nhoff + gh_len);
 569
 570         return err;
 571 }
 572
 573 /* Create new listen socket if needed */
 574 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 575                                                 bool ipv6, bool ipv6_rx_csum)
 576 {
 577         struct geneve_net *gn = net_generic(net, geneve_net_id);
 578         struct geneve_sock *gs;
 579         struct socket *sock;
 580         struct udp_tunnel_sock_cfg tunnel_cfg;
 581         int h;
 582
 583         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 584         if (!gs)
 585                 return ERR_PTR(-ENOMEM);
 586
 587         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 588         if (IS_ERR(sock)) {
 589                 kfree(gs);
 590                 return ERR_CAST(sock);
 591         }
 592
 593         gs->sock = sock;
 594         gs->refcnt = 1;
 595         for (h = 0; h < VNI_HASH_SIZE; ++h)
 596                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 597
 598         /* Initialize the geneve udp offloads structure */
 599         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 600
 601         /* Mark socket as an encapsulation socket */
 602         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 603         tunnel_cfg.sk_user_data = gs;
 604         tunnel_cfg.encap_type = 1;
 605         tunnel_cfg.gro_receive = geneve_gro_receive;
 606         tunnel_cfg.gro_complete = geneve_gro_complete;
 607         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 608         tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
 609         tunnel_cfg.encap_destroy = NULL;
 610         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 611         list_add(&gs->list, &gn->sock_list);
 612         return gs;
 613 }
 614
 615 static void __geneve_sock_release(struct geneve_sock *gs)
 616 {
 617         if (!gs || --gs->refcnt)
 618                 return;
 619
 620         list_del(&gs->list);
 621         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 622         udp_tunnel_sock_release(gs->sock);
 623         kfree_rcu(gs, rcu);
 624 }
 625
 626 static void geneve_sock_release(struct geneve_dev *geneve)
 627 {
 628         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 629 #if IS_ENABLED(CONFIG_IPV6)
 630         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 631
 632         rcu_assign_pointer(geneve->sock6, NULL);
 633 #endif
 634
 635         rcu_assign_pointer(geneve->sock4, NULL);
 636         synchronize_net();
 637
 638         __geneve_sock_release(gs4);
 639 #if IS_ENABLED(CONFIG_IPV6)
 640         __geneve_sock_release(gs6);
 641 #endif
 642 }
 643
 644 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 645                                             sa_family_t family,
 646                                             __be16 dst_port)
 647 {
 648         struct geneve_sock *gs;
 649
 650         list_for_each_entry(gs, &gn->sock_list, list) {
 651                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 652                     geneve_get_sk_family(gs) == family) {
 653                         return gs;
 654                 }
 655         }
 656         return NULL;
 657 }
 658
 659 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 660 {
 661         struct net *net = geneve->net;
 662         struct geneve_net *gn = net_generic(net, geneve_net_id);
 663         struct geneve_dev_node *node;
 664         struct geneve_sock *gs;
 665         __u8 vni[3];
 666         __u32 hash;
 667
 668         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
 669         if (gs) {
 670                 gs->refcnt++;
 671                 goto out;
 672         }
 673
 674         gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
 675                                   geneve->cfg.use_udp6_rx_checksums);
 676         if (IS_ERR(gs))
 677                 return PTR_ERR(gs);
 678
 679 out:
 680         gs->collect_md = geneve->cfg.collect_md;
 681 #if IS_ENABLED(CONFIG_IPV6)
 682         if (ipv6) {
 683                 rcu_assign_pointer(geneve->sock6, gs);
 684                 node = &geneve->hlist6;
 685         } else
 686 #endif
 687         {
 688                 rcu_assign_pointer(geneve->sock4, gs);
 689                 node = &geneve->hlist4;
 690         }
 691         node->geneve = geneve;
 692
 693         tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
 694         hash = geneve_net_vni_hash(vni);
 695         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 696         return 0;
 697 }
 698
 699 static int geneve_open(struct net_device *dev)
 700 {
 701         struct geneve_dev *geneve = netdev_priv(dev);
 702         bool metadata = geneve->cfg.collect_md;
 703         bool ipv4, ipv6;
 704         int ret = 0;
 705
 706         ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 707         ipv4 = !ipv6 || metadata;
 708 #if IS_ENABLED(CONFIG_IPV6)
 709         if (ipv6) {
 710                 ret = geneve_sock_add(geneve, true);
 711                 if (ret < 0 && ret != -EAFNOSUPPORT)
 712                         ipv4 = false;
 713         }
 714 #endif
 715         if (ipv4)
 716                 ret = geneve_sock_add(geneve, false);
 717         if (ret < 0)
 718                 geneve_sock_release(geneve);
 719
 720         return ret;
 721 }
 722
 723 static int geneve_stop(struct net_device *dev)
 724 {
 725         struct geneve_dev *geneve = netdev_priv(dev);
 726
 727         hlist_del_init_rcu(&geneve->hlist4.hlist);
 728 #if IS_ENABLED(CONFIG_IPV6)
 729         hlist_del_init_rcu(&geneve->hlist6.hlist);
 730 #endif
 731         geneve_sock_release(geneve);
 732         return 0;
 733 }
 734
 735 static void geneve_build_header(struct genevehdr *geneveh,
 736                                 const struct ip_tunnel_info *info,
 737                                 __be16 inner_proto)
 738 {
 739         geneveh->ver = GENEVE_VER;
 740         geneveh->opt_len = info->options_len / 4;
 741         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 742         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 743         geneveh->rsvd1 = 0;
 744         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 745         geneveh->proto_type = inner_proto;
 746         geneveh->rsvd2 = 0;
 747
 748         if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
 749                 ip_tunnel_info_opts_get(geneveh->options, info);
 750 }
 751
 752 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 753                             const struct ip_tunnel_info *info,
 754                             bool xnet, int ip_hdr_len,
 755                             bool inner_proto_inherit)
 756 {
 757         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 758         struct genevehdr *gnvh;
 759         __be16 inner_proto;
 760         int min_headroom;
 761         int err;
 762
 763         skb_reset_mac_header(skb);
 764         skb_scrub_packet(skb, xnet);
 765
 766         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 767                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 768         err = skb_cow_head(skb, min_headroom);
 769         if (unlikely(err))
 770                 goto free_dst;
 771
 772         err = udp_tunnel_handle_offloads(skb, udp_sum);
 773         if (err)
 774                 goto free_dst;
 775
 776         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 777         inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB);
 778         geneve_build_header(gnvh, info, inner_proto);
 779         skb_set_inner_protocol(skb, inner_proto);
 780         return 0;
 781
 782 free_dst:
 783         dst_release(dst);
 784         return err;
 785 }
 786
 787 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 788                                        struct net_device *dev,
 789                                        struct geneve_sock *gs4,
 790                                        struct flowi4 *fl4,
 791                                        const struct ip_tunnel_info *info,
 792                                        __be16 dport, __be16 sport,
 793                                        __u8 *full_tos)
 794 {
 795         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 796         struct geneve_dev *geneve = netdev_priv(dev);
 797         struct dst_cache *dst_cache;
 798         struct rtable *rt = NULL;
 799         __u8 tos;
 800
 801         if (!gs4)
 802                 return ERR_PTR(-EIO);
 803
 804         memset(fl4, 0, sizeof(*fl4));
 805         fl4->flowi4_mark = skb->mark;
 806         fl4->flowi4_proto = IPPROTO_UDP;
 807         fl4->daddr = info->key.u.ipv4.dst;
 808         fl4->saddr = info->key.u.ipv4.src;
 809         fl4->fl4_dport = dport;
 810         fl4->fl4_sport = sport;
 811         fl4->flowi4_flags = info->key.flow_flags;
 812
 813         tos = info->key.tos;
 814         if ((tos == 1) && !geneve->cfg.collect_md) {
 815                 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 816                 use_cache = false;
 817         }
 818         fl4->flowi4_tos = RT_TOS(tos);
 819         if (full_tos)
 820                 *full_tos = tos;
 821
 822         dst_cache = (struct dst_cache *)&info->dst_cache;
 823         if (use_cache) {
 824                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 825                 if (rt)
 826                         return rt;
 827         }
 828         rt = ip_route_output_key(geneve->net, fl4);
 829         if (IS_ERR(rt)) {
 830                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 831                 return ERR_PTR(-ENETUNREACH);
 832         }
 833         if (rt->dst.dev == dev) { /* is this necessary? */
 834                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 835                 ip_rt_put(rt);
 836                 return ERR_PTR(-ELOOP);
 837         }
 838         if (use_cache)
 839                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 840         return rt;
 841 }
 842
 843 #if IS_ENABLED(CONFIG_IPV6)
 844 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 845                                            struct net_device *dev,
 846                                            struct geneve_sock *gs6,
 847                                            struct flowi6 *fl6,
 848                                            const struct ip_tunnel_info *info,
 849                                            __be16 dport, __be16 sport)
 850 {
 851         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 852         struct geneve_dev *geneve = netdev_priv(dev);
 853         struct dst_entry *dst = NULL;
 854         struct dst_cache *dst_cache;
 855         __u8 prio;
 856
 857         if (!gs6)
 858                 return ERR_PTR(-EIO);
 859
 860         memset(fl6, 0, sizeof(*fl6));
 861         fl6->flowi6_mark = skb->mark;
 862         fl6->flowi6_proto = IPPROTO_UDP;
 863         fl6->daddr = info->key.u.ipv6.dst;
 864         fl6->saddr = info->key.u.ipv6.src;
 865         fl6->fl6_dport = dport;
 866         fl6->fl6_sport = sport;
 867
 868         prio = info->key.tos;
 869         if ((prio == 1) && !geneve->cfg.collect_md) {
 870                 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 871                 use_cache = false;
 872         }
 873
 874         fl6->flowlabel = ip6_make_flowinfo(prio, info->key.label);
 875         dst_cache = (struct dst_cache *)&info->dst_cache;
 876         if (use_cache) {
 877                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 878                 if (dst)
 879                         return dst;
 880         }
 881         dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
 882                                               NULL);
 883         if (IS_ERR(dst)) {
 884                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 885                 return ERR_PTR(-ENETUNREACH);
 886         }
 887         if (dst->dev == dev) { /* is this necessary? */
 888                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 889                 dst_release(dst);
 890                 return ERR_PTR(-ELOOP);
 891         }
 892
 893         if (use_cache)
 894                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 895         return dst;
 896 }
 897 #endif
 898
 899 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 900                            struct geneve_dev *geneve,
 901                            const struct ip_tunnel_info *info)
 902 {
 903         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 904         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 905         const struct ip_tunnel_key *key = &info->key;
 906         struct rtable *rt;
 907         struct flowi4 fl4;
 908         __u8 full_tos;
 909         __u8 tos, ttl;
 910         __be16 df = 0;
 911         __be16 sport;
 912         int err;
 913
 914         if (!pskb_inet_may_pull(skb))
 915                 return -EINVAL;
 916
 917         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 918         rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
 919                               geneve->cfg.info.key.tp_dst, sport, &full_tos);
 920         if (IS_ERR(rt))
 921                 return PTR_ERR(rt);
 922
 923         err = skb_tunnel_check_pmtu(skb, &rt->dst,
 924                                     GENEVE_IPV4_HLEN + info->options_len,
 925                                     netif_is_any_bridge_port(dev));
 926         if (err < 0) {
 927                 dst_release(&rt->dst);
 928                 return err;
 929         } else if (err) {
 930                 struct ip_tunnel_info *info;
 931
 932                 info = skb_tunnel_info(skb);
 933                 if (info) {
 934                         struct ip_tunnel_info *unclone;
 935
 936                         unclone = skb_tunnel_info_unclone(skb);
 937                         if (unlikely(!unclone)) {
 938                                 dst_release(&rt->dst);
 939                                 return -ENOMEM;
 940                         }
 941
 942                         unclone->key.u.ipv4.dst = fl4.saddr;
 943                         unclone->key.u.ipv4.src = fl4.daddr;
 944                 }
 945
 946                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 947                         dst_release(&rt->dst);
 948                         return -EINVAL;
 949                 }
 950
 951                 skb->protocol = eth_type_trans(skb, geneve->dev);
 952                 __netif_rx(skb);
 953                 dst_release(&rt->dst);
 954                 return -EMSGSIZE;
 955         }
 956
 957         if (geneve->cfg.collect_md) {
 958                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 959                 ttl = key->ttl;
 960
 961                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 962         } else {
 963                 tos = ip_tunnel_ecn_encap(full_tos, ip_hdr(skb), skb);
 964                 if (geneve->cfg.ttl_inherit)
 965                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 966                 else
 967                         ttl = key->ttl;
 968                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 969
 970                 if (geneve->cfg.df == GENEVE_DF_SET) {
 971                         df = htons(IP_DF);
 972                 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
 973                         struct ethhdr *eth = eth_hdr(skb);
 974
 975                         if (ntohs(eth->h_proto) == ETH_P_IPV6) {
 976                                 df = htons(IP_DF);
 977                         } else if (ntohs(eth->h_proto) == ETH_P_IP) {
 978                                 struct iphdr *iph = ip_hdr(skb);
 979
 980                                 if (iph->frag_off & htons(IP_DF))
 981                                         df = htons(IP_DF);
 982                         }
 983                 }
 984         }
 985
 986         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
 987                                geneve->cfg.inner_proto_inherit);
 988         if (unlikely(err))
 989                 return err;
 990
 991         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 992                             tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
 993                             !net_eq(geneve->net, dev_net(geneve->dev)),
 994                             !(info->key.tun_flags & TUNNEL_CSUM));
 995         return 0;
 996 }
 997
 998 #if IS_ENABLED(CONFIG_IPV6)
 999 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
1000                             struct geneve_dev *geneve,
1001                             const struct ip_tunnel_info *info)
1002 {
1003         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
1004         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1005         const struct ip_tunnel_key *key = &info->key;
1006         struct dst_entry *dst = NULL;
1007         struct flowi6 fl6;
1008         __u8 prio, ttl;
1009         __be16 sport;
1010         int err;
1011
1012         if (!pskb_inet_may_pull(skb))
1013                 return -EINVAL;
1014
1015         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1016         dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1017                                 geneve->cfg.info.key.tp_dst, sport);
1018         if (IS_ERR(dst))
1019                 return PTR_ERR(dst);
1020
1021         err = skb_tunnel_check_pmtu(skb, dst,
1022                                     GENEVE_IPV6_HLEN + info->options_len,
1023                                     netif_is_any_bridge_port(dev));
1024         if (err < 0) {
1025                 dst_release(dst);
1026                 return err;
1027         } else if (err) {
1028                 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1029
1030                 if (info) {
1031                         struct ip_tunnel_info *unclone;
1032
1033                         unclone = skb_tunnel_info_unclone(skb);
1034                         if (unlikely(!unclone)) {
1035                                 dst_release(dst);
1036                                 return -ENOMEM;
1037                         }
1038
1039                         unclone->key.u.ipv6.dst = fl6.saddr;
1040                         unclone->key.u.ipv6.src = fl6.daddr;
1041                 }
1042
1043                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1044                         dst_release(dst);
1045                         return -EINVAL;
1046                 }
1047
1048                 skb->protocol = eth_type_trans(skb, geneve->dev);
1049                 __netif_rx(skb);
1050                 dst_release(dst);
1051                 return -EMSGSIZE;
1052         }
1053
1054         if (geneve->cfg.collect_md) {
1055                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1056                 ttl = key->ttl;
1057         } else {
1058                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1059                                            ip_hdr(skb), skb);
1060                 if (geneve->cfg.ttl_inherit)
1061                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1062                 else
1063                         ttl = key->ttl;
1064                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1065         }
1066         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
1067                                geneve->cfg.inner_proto_inherit);
1068         if (unlikely(err))
1069                 return err;
1070
1071         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1072                              &fl6.saddr, &fl6.daddr, prio, ttl,
1073                              info->key.label, sport, geneve->cfg.info.key.tp_dst,
1074                              !(info->key.tun_flags & TUNNEL_CSUM));
1075         return 0;
1076 }
1077 #endif
1078
1079 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1080 {
1081         struct geneve_dev *geneve = netdev_priv(dev);
1082         struct ip_tunnel_info *info = NULL;
1083         int err;
1084
1085         if (geneve->cfg.collect_md) {
1086                 info = skb_tunnel_info(skb);
1087                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1088                         netdev_dbg(dev, "no tunnel metadata\n");
1089                         dev_kfree_skb(skb);
1090                         dev->stats.tx_dropped++;
1091                         return NETDEV_TX_OK;
1092                 }
1093         } else {
1094                 info = &geneve->cfg.info;
1095         }
1096
1097         rcu_read_lock();
1098 #if IS_ENABLED(CONFIG_IPV6)
1099         if (info->mode & IP_TUNNEL_INFO_IPV6)
1100                 err = geneve6_xmit_skb(skb, dev, geneve, info);
1101         else
1102 #endif
1103                 err = geneve_xmit_skb(skb, dev, geneve, info);
1104         rcu_read_unlock();
1105
1106         if (likely(!err))
1107                 return NETDEV_TX_OK;
1108
1109         if (err != -EMSGSIZE)
1110                 dev_kfree_skb(skb);
1111
1112         if (err == -ELOOP)
1113                 dev->stats.collisions++;
1114         else if (err == -ENETUNREACH)
1115                 dev->stats.tx_carrier_errors++;
1116
1117         dev->stats.tx_errors++;
1118         return NETDEV_TX_OK;
1119 }
1120
1121 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1122 {
1123         if (new_mtu > dev->max_mtu)
1124                 new_mtu = dev->max_mtu;
1125         else if (new_mtu < dev->min_mtu)
1126                 new_mtu = dev->min_mtu;
1127
1128         dev->mtu = new_mtu;
1129         return 0;
1130 }
1131
1132 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1133 {
1134         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1135         struct geneve_dev *geneve = netdev_priv(dev);
1136         __be16 sport;
1137
1138         if (ip_tunnel_info_af(info) == AF_INET) {
1139                 struct rtable *rt;
1140                 struct flowi4 fl4;
1141
1142                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1143                 sport = udp_flow_src_port(geneve->net, skb,
1144                                           1, USHRT_MAX, true);
1145
1146                 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1147                                       geneve->cfg.info.key.tp_dst, sport, NULL);
1148                 if (IS_ERR(rt))
1149                         return PTR_ERR(rt);
1150
1151                 ip_rt_put(rt);
1152                 info->key.u.ipv4.src = fl4.saddr;
1153 #if IS_ENABLED(CONFIG_IPV6)
1154         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1155                 struct dst_entry *dst;
1156                 struct flowi6 fl6;
1157
1158                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1159                 sport = udp_flow_src_port(geneve->net, skb,
1160                                           1, USHRT_MAX, true);
1161
1162                 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1163                                         geneve->cfg.info.key.tp_dst, sport);
1164                 if (IS_ERR(dst))
1165                         return PTR_ERR(dst);
1166
1167                 dst_release(dst);
1168                 info->key.u.ipv6.src = fl6.saddr;
1169 #endif
1170         } else {
1171                 return -EINVAL;
1172         }
1173
1174         info->key.tp_src = sport;
1175         info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1176         return 0;
1177 }
1178
1179 static const struct net_device_ops geneve_netdev_ops = {
1180         .ndo_init               = geneve_init,
1181         .ndo_uninit             = geneve_uninit,
1182         .ndo_open               = geneve_open,
1183         .ndo_stop               = geneve_stop,
1184         .ndo_start_xmit         = geneve_xmit,
1185         .ndo_get_stats64        = dev_get_tstats64,
1186         .ndo_change_mtu         = geneve_change_mtu,
1187         .ndo_validate_addr      = eth_validate_addr,
1188         .ndo_set_mac_address    = eth_mac_addr,
1189         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1190 };
1191
1192 static void geneve_get_drvinfo(struct net_device *dev,
1193                                struct ethtool_drvinfo *drvinfo)
1194 {
1195         strscpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1196         strscpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1197 }
1198
1199 static const struct ethtool_ops geneve_ethtool_ops = {
1200         .get_drvinfo    = geneve_get_drvinfo,
1201         .get_link       = ethtool_op_get_link,
1202 };
1203
1204 /* Info for udev, that this is a virtual tunnel endpoint */
1205 static struct device_type geneve_type = {
1206         .name = "geneve",
1207 };
1208
1209 /* Calls the ndo_udp_tunnel_add of the caller in order to
1210  * supply the listening GENEVE udp ports. Callers are expected
1211  * to implement the ndo_udp_tunnel_add.
1212  */
1213 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1214 {
1215         struct net *net = dev_net(dev);
1216         struct geneve_net *gn = net_generic(net, geneve_net_id);
1217         struct geneve_sock *gs;
1218
1219         rcu_read_lock();
1220         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1221                 if (push) {
1222                         udp_tunnel_push_rx_port(dev, gs->sock,
1223                                                 UDP_TUNNEL_TYPE_GENEVE);
1224                 } else {
1225                         udp_tunnel_drop_rx_port(dev, gs->sock,
1226                                                 UDP_TUNNEL_TYPE_GENEVE);
1227                 }
1228         }
1229         rcu_read_unlock();
1230 }
1231
1232 /* Initialize the device structure. */
1233 static void geneve_setup(struct net_device *dev)
1234 {
1235         ether_setup(dev);
1236
1237         dev->netdev_ops = &geneve_netdev_ops;
1238         dev->ethtool_ops = &geneve_ethtool_ops;
1239         dev->needs_free_netdev = true;
1240
1241         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1242
1243         dev->features    |= NETIF_F_LLTX;
1244         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1245         dev->features    |= NETIF_F_RXCSUM;
1246         dev->features    |= NETIF_F_GSO_SOFTWARE;
1247
1248         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1249         dev->hw_features |= NETIF_F_RXCSUM;
1250         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1251
1252         /* MTU range: 68 - (something less than 65535) */
1253         dev->min_mtu = ETH_MIN_MTU;
1254         /* The max_mtu calculation does not take account of GENEVE
1255          * options, to avoid excluding potentially valid
1256          * configurations. This will be further reduced by IPvX hdr size.
1257          */
1258         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1259
1260         netif_keep_dst(dev);
1261         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1262         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1263         eth_hw_addr_random(dev);
1264 }
1265
1266 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1267         [IFLA_GENEVE_UNSPEC]            = { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT },
1268         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1269         [IFLA_GENEVE_REMOTE]            = { .len = sizeof_field(struct iphdr, daddr) },
1270         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1271         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1272         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1273         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1274         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1275         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1276         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1277         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1278         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1279         [IFLA_GENEVE_TTL_INHERIT]       = { .type = NLA_U8 },
1280         [IFLA_GENEVE_DF]                = { .type = NLA_U8 },
1281         [IFLA_GENEVE_INNER_PROTO_INHERIT]       = { .type = NLA_FLAG },
1282 };
1283
1284 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1285                            struct netlink_ext_ack *extack)
1286 {
1287         if (tb[IFLA_ADDRESS]) {
1288                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1289                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1290                                             "Provided link layer address is not Ethernet");
1291                         return -EINVAL;
1292                 }
1293
1294                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1295                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1296                                             "Provided Ethernet address is not unicast");
1297                         return -EADDRNOTAVAIL;
1298                 }
1299         }
1300
1301         if (!data) {
1302                 NL_SET_ERR_MSG(extack,
1303                                "Not enough attributes provided to perform the operation");
1304                 return -EINVAL;
1305         }
1306
1307         if (data[IFLA_GENEVE_ID]) {
1308                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1309
1310                 if (vni >= GENEVE_N_VID) {
1311                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1312                                             "Geneve ID must be lower than 16777216");
1313                         return -ERANGE;
1314                 }
1315         }
1316
1317         if (data[IFLA_GENEVE_DF]) {
1318                 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1319
1320                 if (df < 0 || df > GENEVE_DF_MAX) {
1321                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1322                                             "Invalid DF attribute");
1323                         return -EINVAL;
1324                 }
1325         }
1326
1327         return 0;
1328 }
1329
1330 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1331                                           const struct ip_tunnel_info *info,
1332                                           bool *tun_on_same_port,
1333                                           bool *tun_collect_md)
1334 {
1335         struct geneve_dev *geneve, *t = NULL;
1336
1337         *tun_on_same_port = false;
1338         *tun_collect_md = false;
1339         list_for_each_entry(geneve, &gn->geneve_list, next) {
1340                 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1341                         *tun_collect_md = geneve->cfg.collect_md;
1342                         *tun_on_same_port = true;
1343                 }
1344                 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1345                     info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1346                     !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1347                         t = geneve;
1348         }
1349         return t;
1350 }
1351
1352 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1353 {
1354         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1355                  info->key.ttl || info->key.label || info->key.tp_src ||
1356                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1357 }
1358
1359 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1360                                   struct ip_tunnel_info *b)
1361 {
1362         if (ip_tunnel_info_af(a) == AF_INET)
1363                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1364         else
1365                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1366 }
1367
1368 static int geneve_configure(struct net *net, struct net_device *dev,
1369                             struct netlink_ext_ack *extack,
1370                             const struct geneve_config *cfg)
1371 {
1372         struct geneve_net *gn = net_generic(net, geneve_net_id);
1373         struct geneve_dev *t, *geneve = netdev_priv(dev);
1374         const struct ip_tunnel_info *info = &cfg->info;
1375         bool tun_collect_md, tun_on_same_port;
1376         int err, encap_len;
1377
1378         if (cfg->collect_md && !is_tnl_info_zero(info)) {
1379                 NL_SET_ERR_MSG(extack,
1380                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1381                 return -EINVAL;
1382         }
1383
1384         geneve->net = net;
1385         geneve->dev = dev;
1386
1387         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1388         if (t)
1389                 return -EBUSY;
1390
1391         /* make enough headroom for basic scenario */
1392         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1393         if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1394                 encap_len += sizeof(struct iphdr);
1395                 dev->max_mtu -= sizeof(struct iphdr);
1396         } else {
1397                 encap_len += sizeof(struct ipv6hdr);
1398                 dev->max_mtu -= sizeof(struct ipv6hdr);
1399         }
1400         dev->needed_headroom = encap_len + ETH_HLEN;
1401
1402         if (cfg->collect_md) {
1403                 if (tun_on_same_port) {
1404                         NL_SET_ERR_MSG(extack,
1405                                        "There can be only one externally controlled device on a destination port");
1406                         return -EPERM;
1407                 }
1408         } else {
1409                 if (tun_collect_md) {
1410                         NL_SET_ERR_MSG(extack,
1411                                        "There already exists an externally controlled device on this destination port");
1412                         return -EPERM;
1413                 }
1414         }
1415
1416         dst_cache_reset(&geneve->cfg.info.dst_cache);
1417         memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1418
1419         if (geneve->cfg.inner_proto_inherit) {
1420                 dev->header_ops = NULL;
1421                 dev->type = ARPHRD_NONE;
1422                 dev->hard_header_len = 0;
1423                 dev->addr_len = 0;
1424                 dev->flags = IFF_POINTOPOINT | IFF_NOARP;
1425         }
1426
1427         err = register_netdevice(dev);
1428         if (err)
1429                 return err;
1430
1431         list_add(&geneve->next, &gn->geneve_list);
1432         return 0;
1433 }
1434
1435 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1436 {
1437         memset(info, 0, sizeof(*info));
1438         info->key.tp_dst = htons(dst_port);
1439 }
1440
1441 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1442                           struct netlink_ext_ack *extack,
1443                           struct geneve_config *cfg, bool changelink)
1444 {
1445         struct ip_tunnel_info *info = &cfg->info;
1446         int attrtype;
1447
1448         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1449                 NL_SET_ERR_MSG(extack,
1450                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1451                 return -EINVAL;
1452         }
1453
1454         if (data[IFLA_GENEVE_REMOTE]) {
1455                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1456                         attrtype = IFLA_GENEVE_REMOTE;
1457                         goto change_notsup;
1458                 }
1459
1460                 info->key.u.ipv4.dst =
1461                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1462
1463                 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1464                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1465                                             "Remote IPv4 address cannot be Multicast");
1466                         return -EINVAL;
1467                 }
1468         }
1469
1470         if (data[IFLA_GENEVE_REMOTE6]) {
1471 #if IS_ENABLED(CONFIG_IPV6)
1472                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1473                         attrtype = IFLA_GENEVE_REMOTE6;
1474                         goto change_notsup;
1475                 }
1476
1477                 info->mode = IP_TUNNEL_INFO_IPV6;
1478                 info->key.u.ipv6.dst =
1479                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1480
1481                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1482                     IPV6_ADDR_LINKLOCAL) {
1483                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1484                                             "Remote IPv6 address cannot be link-local");
1485                         return -EINVAL;
1486                 }
1487                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1488                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1489                                             "Remote IPv6 address cannot be Multicast");
1490                         return -EINVAL;
1491                 }
1492                 info->key.tun_flags |= TUNNEL_CSUM;
1493                 cfg->use_udp6_rx_checksums = true;
1494 #else
1495                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1496                                     "IPv6 support not enabled in the kernel");
1497                 return -EPFNOSUPPORT;
1498 #endif
1499         }
1500
1501         if (data[IFLA_GENEVE_ID]) {
1502                 __u32 vni;
1503                 __u8 tvni[3];
1504                 __be64 tunid;
1505
1506                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1507                 tvni[0] = (vni & 0x00ff0000) >> 16;
1508                 tvni[1] = (vni & 0x0000ff00) >> 8;
1509                 tvni[2] =  vni & 0x000000ff;
1510
1511                 tunid = vni_to_tunnel_id(tvni);
1512                 if (changelink && (tunid != info->key.tun_id)) {
1513                         attrtype = IFLA_GENEVE_ID;
1514                         goto change_notsup;
1515                 }
1516                 info->key.tun_id = tunid;
1517         }
1518
1519         if (data[IFLA_GENEVE_TTL_INHERIT]) {
1520                 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1521                         cfg->ttl_inherit = true;
1522                 else
1523                         cfg->ttl_inherit = false;
1524         } else if (data[IFLA_GENEVE_TTL]) {
1525                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1526                 cfg->ttl_inherit = false;
1527         }
1528
1529         if (data[IFLA_GENEVE_TOS])
1530                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1531
1532         if (data[IFLA_GENEVE_DF])
1533                 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1534
1535         if (data[IFLA_GENEVE_LABEL]) {
1536                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1537                                   IPV6_FLOWLABEL_MASK;
1538                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1539                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1540                                             "Label attribute only applies for IPv6 Geneve devices");
1541                         return -EINVAL;
1542                 }
1543         }
1544
1545         if (data[IFLA_GENEVE_PORT]) {
1546                 if (changelink) {
1547                         attrtype = IFLA_GENEVE_PORT;
1548                         goto change_notsup;
1549                 }
1550                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1551         }
1552
1553         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1554                 if (changelink) {
1555                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1556                         goto change_notsup;
1557                 }
1558                 cfg->collect_md = true;
1559         }
1560
1561         if (data[IFLA_GENEVE_UDP_CSUM]) {
1562                 if (changelink) {
1563                         attrtype = IFLA_GENEVE_UDP_CSUM;
1564                         goto change_notsup;
1565                 }
1566                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1567                         info->key.tun_flags |= TUNNEL_CSUM;
1568         }
1569
1570         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1571 #if IS_ENABLED(CONFIG_IPV6)
1572                 if (changelink) {
1573                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1574                         goto change_notsup;
1575                 }
1576                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1577                         info->key.tun_flags &= ~TUNNEL_CSUM;
1578 #else
1579                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1580                                     "IPv6 support not enabled in the kernel");
1581                 return -EPFNOSUPPORT;
1582 #endif
1583         }
1584
1585         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1586 #if IS_ENABLED(CONFIG_IPV6)
1587                 if (changelink) {
1588                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1589                         goto change_notsup;
1590                 }
1591                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1592                         cfg->use_udp6_rx_checksums = false;
1593 #else
1594                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1595                                     "IPv6 support not enabled in the kernel");
1596                 return -EPFNOSUPPORT;
1597 #endif
1598         }
1599
1600         if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) {
1601                 if (changelink) {
1602                         attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT;
1603                         goto change_notsup;
1604                 }
1605                 cfg->inner_proto_inherit = true;
1606         }
1607
1608         return 0;
1609 change_notsup:
1610         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1611                             "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported");
1612         return -EOPNOTSUPP;
1613 }
1614
1615 static void geneve_link_config(struct net_device *dev,
1616                                struct ip_tunnel_info *info, struct nlattr *tb[])
1617 {
1618         struct geneve_dev *geneve = netdev_priv(dev);
1619         int ldev_mtu = 0;
1620
1621         if (tb[IFLA_MTU]) {
1622                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1623                 return;
1624         }
1625
1626         switch (ip_tunnel_info_af(info)) {
1627         case AF_INET: {
1628                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1629                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1630
1631                 if (!IS_ERR(rt) && rt->dst.dev) {
1632                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1633                         ip_rt_put(rt);
1634                 }
1635                 break;
1636         }
1637 #if IS_ENABLED(CONFIG_IPV6)
1638         case AF_INET6: {
1639                 struct rt6_info *rt;
1640
1641                 if (!__in6_dev_get(dev))
1642                         break;
1643
1644                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1645                                 NULL, 0);
1646
1647                 if (rt && rt->dst.dev)
1648                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1649                 ip6_rt_put(rt);
1650                 break;
1651         }
1652 #endif
1653         }
1654
1655         if (ldev_mtu <= 0)
1656                 return;
1657
1658         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1659 }
1660
1661 static int geneve_newlink(struct net *net, struct net_device *dev,
1662                           struct nlattr *tb[], struct nlattr *data[],
1663                           struct netlink_ext_ack *extack)
1664 {
1665         struct geneve_config cfg = {
1666                 .df = GENEVE_DF_UNSET,
1667                 .use_udp6_rx_checksums = false,
1668                 .ttl_inherit = false,
1669                 .collect_md = false,
1670         };
1671         int err;
1672
1673         init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1674         err = geneve_nl2info(tb, data, extack, &cfg, false);
1675         if (err)
1676                 return err;
1677
1678         err = geneve_configure(net, dev, extack, &cfg);
1679         if (err)
1680                 return err;
1681
1682         geneve_link_config(dev, &cfg.info, tb);
1683
1684         return 0;
1685 }
1686
1687 /* Quiesces the geneve device data path for both TX and RX.
1688  *
1689  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1690  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1691  * to complete for the existing set of in-flight packets to be transmitted,
1692  * then we would have quiesced the transmit data path. All the future packets
1693  * will get dropped until we unquiesce the data path.
1694  *
1695  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1696  * if we set that to NULL under RCU and wait for synchronize_net() to
1697  * complete, then we would have quiesced the receive data path.
1698  */
1699 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1700                            struct geneve_sock **gs6)
1701 {
1702         *gs4 = rtnl_dereference(geneve->sock4);
1703         rcu_assign_pointer(geneve->sock4, NULL);
1704         if (*gs4)
1705                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1706 #if IS_ENABLED(CONFIG_IPV6)
1707         *gs6 = rtnl_dereference(geneve->sock6);
1708         rcu_assign_pointer(geneve->sock6, NULL);
1709         if (*gs6)
1710                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1711 #else
1712         *gs6 = NULL;
1713 #endif
1714         synchronize_net();
1715 }
1716
1717 /* Resumes the geneve device data path for both TX and RX. */
1718 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1719                              struct geneve_sock __maybe_unused *gs6)
1720 {
1721         rcu_assign_pointer(geneve->sock4, gs4);
1722         if (gs4)
1723                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1724 #if IS_ENABLED(CONFIG_IPV6)
1725         rcu_assign_pointer(geneve->sock6, gs6);
1726         if (gs6)
1727                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1728 #endif
1729         synchronize_net();
1730 }
1731
1732 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1733                              struct nlattr *data[],
1734                              struct netlink_ext_ack *extack)
1735 {
1736         struct geneve_dev *geneve = netdev_priv(dev);
1737         struct geneve_sock *gs4, *gs6;
1738         struct geneve_config cfg;
1739         int err;
1740
1741         /* If the geneve device is configured for metadata (or externally
1742          * controlled, for example, OVS), then nothing can be changed.
1743          */
1744         if (geneve->cfg.collect_md)
1745                 return -EOPNOTSUPP;
1746
1747         /* Start with the existing info. */
1748         memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1749         err = geneve_nl2info(tb, data, extack, &cfg, true);
1750         if (err)
1751                 return err;
1752
1753         if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1754                 dst_cache_reset(&cfg.info.dst_cache);
1755                 geneve_link_config(dev, &cfg.info, tb);
1756         }
1757
1758         geneve_quiesce(geneve, &gs4, &gs6);
1759         memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1760         geneve_unquiesce(geneve, gs4, gs6);
1761
1762         return 0;
1763 }
1764
1765 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1766 {
1767         struct geneve_dev *geneve = netdev_priv(dev);
1768
1769         list_del(&geneve->next);
1770         unregister_netdevice_queue(dev, head);
1771 }
1772
1773 static size_t geneve_get_size(const struct net_device *dev)
1774 {
1775         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1776                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1777                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1778                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1779                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_DF */
1780                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1781                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1782                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1783                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1784                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1785                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1786                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1787                 nla_total_size(0) +      /* IFLA_GENEVE_INNER_PROTO_INHERIT */
1788                 0;
1789 }
1790
1791 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1792 {
1793         struct geneve_dev *geneve = netdev_priv(dev);
1794         struct ip_tunnel_info *info = &geneve->cfg.info;
1795         bool ttl_inherit = geneve->cfg.ttl_inherit;
1796         bool metadata = geneve->cfg.collect_md;
1797         __u8 tmp_vni[3];
1798         __u32 vni;
1799
1800         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1801         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1802         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1803                 goto nla_put_failure;
1804
1805         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1806                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1807                                     info->key.u.ipv4.dst))
1808                         goto nla_put_failure;
1809                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1810                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1811                         goto nla_put_failure;
1812
1813 #if IS_ENABLED(CONFIG_IPV6)
1814         } else if (!metadata) {
1815                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1816                                      &info->key.u.ipv6.dst))
1817                         goto nla_put_failure;
1818                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1819                                !(info->key.tun_flags & TUNNEL_CSUM)))
1820                         goto nla_put_failure;
1821 #endif
1822         }
1823
1824         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1825             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1826             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1827                 goto nla_put_failure;
1828
1829         if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1830                 goto nla_put_failure;
1831
1832         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1833                 goto nla_put_failure;
1834
1835         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1836                 goto nla_put_failure;
1837
1838 #if IS_ENABLED(CONFIG_IPV6)
1839         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1840                        !geneve->cfg.use_udp6_rx_checksums))
1841                 goto nla_put_failure;
1842 #endif
1843
1844         if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1845                 goto nla_put_failure;
1846
1847         if (geneve->cfg.inner_proto_inherit &&
1848             nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT))
1849                 goto nla_put_failure;
1850
1851         return 0;
1852
1853 nla_put_failure:
1854         return -EMSGSIZE;
1855 }
1856
1857 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1858         .kind           = "geneve",
1859         .maxtype        = IFLA_GENEVE_MAX,
1860         .policy         = geneve_policy,
1861         .priv_size      = sizeof(struct geneve_dev),
1862         .setup          = geneve_setup,
1863         .validate       = geneve_validate,
1864         .newlink        = geneve_newlink,
1865         .changelink     = geneve_changelink,
1866         .dellink        = geneve_dellink,
1867         .get_size       = geneve_get_size,
1868         .fill_info      = geneve_fill_info,
1869 };
1870
1871 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1872                                         u8 name_assign_type, u16 dst_port)
1873 {
1874         struct nlattr *tb[IFLA_MAX + 1];
1875         struct net_device *dev;
1876         LIST_HEAD(list_kill);
1877         int err;
1878         struct geneve_config cfg = {
1879                 .df = GENEVE_DF_UNSET,
1880                 .use_udp6_rx_checksums = true,
1881                 .ttl_inherit = false,
1882                 .collect_md = true,
1883         };
1884
1885         memset(tb, 0, sizeof(tb));
1886         dev = rtnl_create_link(net, name, name_assign_type,
1887                                &geneve_link_ops, tb, NULL);
1888         if (IS_ERR(dev))
1889                 return dev;
1890
1891         init_tnl_info(&cfg.info, dst_port);
1892         err = geneve_configure(net, dev, NULL, &cfg);
1893         if (err) {
1894                 free_netdev(dev);
1895                 return ERR_PTR(err);
1896         }
1897
1898         /* openvswitch users expect packet sizes to be unrestricted,
1899          * so set the largest MTU we can.
1900          */
1901         err = geneve_change_mtu(dev, IP_MAX_MTU);
1902         if (err)
1903                 goto err;
1904
1905         err = rtnl_configure_link(dev, NULL, 0, NULL);
1906         if (err < 0)
1907                 goto err;
1908
1909         return dev;
1910 err:
1911         geneve_dellink(dev, &list_kill);
1912         unregister_netdevice_many(&list_kill);
1913         return ERR_PTR(err);
1914 }
1915 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1916
1917 static int geneve_netdevice_event(struct notifier_block *unused,
1918                                   unsigned long event, void *ptr)
1919 {
1920         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1921
1922         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1923                 geneve_offload_rx_ports(dev, true);
1924         else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1925                 geneve_offload_rx_ports(dev, false);
1926
1927         return NOTIFY_DONE;
1928 }
1929
1930 static struct notifier_block geneve_notifier_block __read_mostly = {
1931         .notifier_call = geneve_netdevice_event,
1932 };
1933
1934 static __net_init int geneve_init_net(struct net *net)
1935 {
1936         struct geneve_net *gn = net_generic(net, geneve_net_id);
1937
1938         INIT_LIST_HEAD(&gn->geneve_list);
1939         INIT_LIST_HEAD(&gn->sock_list);
1940         return 0;
1941 }
1942
1943 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1944 {
1945         struct geneve_net *gn = net_generic(net, geneve_net_id);
1946         struct geneve_dev *geneve, *next;
1947         struct net_device *dev, *aux;
1948
1949         /* gather any geneve devices that were moved into this ns */
1950         for_each_netdev_safe(net, dev, aux)
1951                 if (dev->rtnl_link_ops == &geneve_link_ops)
1952                         unregister_netdevice_queue(dev, head);
1953
1954         /* now gather any other geneve devices that were created in this ns */
1955         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1956                 /* If geneve->dev is in the same netns, it was already added
1957                  * to the list by the previous loop.
1958                  */
1959                 if (!net_eq(dev_net(geneve->dev), net))
1960                         unregister_netdevice_queue(geneve->dev, head);
1961         }
1962 }
1963
1964 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1965 {
1966         struct net *net;
1967         LIST_HEAD(list);
1968
1969         rtnl_lock();
1970         list_for_each_entry(net, net_list, exit_list)
1971                 geneve_destroy_tunnels(net, &list);
1972
1973         /* unregister the devices gathered above */
1974         unregister_netdevice_many(&list);
1975         rtnl_unlock();
1976
1977         list_for_each_entry(net, net_list, exit_list) {
1978                 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1979
1980                 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1981         }
1982 }
1983
1984 static struct pernet_operations geneve_net_ops = {
1985         .init = geneve_init_net,
1986         .exit_batch = geneve_exit_batch_net,
1987         .id   = &geneve_net_id,
1988         .size = sizeof(struct geneve_net),
1989 };
1990
1991 static int __init geneve_init_module(void)
1992 {
1993         int rc;
1994
1995         rc = register_pernet_subsys(&geneve_net_ops);
1996         if (rc)
1997                 goto out1;
1998
1999         rc = register_netdevice_notifier(&geneve_notifier_block);
2000         if (rc)
2001                 goto out2;
2002
2003         rc = rtnl_link_register(&geneve_link_ops);
2004         if (rc)
2005                 goto out3;
2006
2007         return 0;
2008 out3:
2009         unregister_netdevice_notifier(&geneve_notifier_block);
2010 out2:
2011         unregister_pernet_subsys(&geneve_net_ops);
2012 out1:
2013         return rc;
2014 }
2015 late_initcall(geneve_init_module);
2016
2017 static void __exit geneve_cleanup_module(void)
2018 {
2019         rtnl_link_unregister(&geneve_link_ops);
2020         unregister_netdevice_notifier(&geneve_notifier_block);
2021         unregister_pernet_subsys(&geneve_net_ops);
2022 }
2023 module_exit(geneve_cleanup_module);
2024
2025 MODULE_LICENSE("GPL");
2026 MODULE_VERSION(GENEVE_NETDEV_VER);
2027 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
2028 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
2029 MODULE_ALIAS_RTNL_LINK("geneve");