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/netdev_lock.h>
  22 #include <net/protocol.h>
  23
  24 #define GENEVE_NETDEV_VER       "0.6"
  25
  26 #define GENEVE_N_VID            (1u << 24)
  27 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  28
  29 #define VNI_HASH_BITS           10
  30 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  31
  32 static bool log_ecn_error = true;
  33 module_param(log_ecn_error, bool, 0644);
  34 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  35
  36 #define GENEVE_VER 0
  37 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  38 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
  39 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
  40
  41 /* per-network namespace private data for this module */
  42 struct geneve_net {
  43         struct list_head        geneve_list;
  44         struct list_head        sock_list;
  45 };
  46
  47 static unsigned int geneve_net_id;
  48
  49 struct geneve_dev_node {
  50         struct hlist_node hlist;
  51         struct geneve_dev *geneve;
  52 };
  53
  54 struct geneve_config {
  55         struct ip_tunnel_info   info;
  56         bool                    collect_md;
  57         bool                    use_udp6_rx_checksums;
  58         bool                    ttl_inherit;
  59         enum ifla_geneve_df     df;
  60         bool                    inner_proto_inherit;
  61         u16                     port_min;
  62         u16                     port_max;
  63 };
  64
  65 /* Pseudo network device */
  66 struct geneve_dev {
  67         struct geneve_dev_node hlist4;  /* vni hash table for IPv4 socket */
  68 #if IS_ENABLED(CONFIG_IPV6)
  69         struct geneve_dev_node hlist6;  /* vni hash table for IPv6 socket */
  70 #endif
  71         struct net         *net;        /* netns for packet i/o */
  72         struct net_device  *dev;        /* netdev for geneve tunnel */
  73         struct geneve_sock __rcu *sock4;        /* IPv4 socket used for geneve tunnel */
  74 #if IS_ENABLED(CONFIG_IPV6)
  75         struct geneve_sock __rcu *sock6;        /* IPv6 socket used for geneve tunnel */
  76 #endif
  77         struct list_head   next;        /* geneve's per namespace list */
  78         struct gro_cells   gro_cells;
  79         struct geneve_config cfg;
  80 };
  81
  82 struct geneve_sock {
  83         bool                    collect_md;
  84         struct list_head        list;
  85         struct socket           *sock;
  86         struct rcu_head         rcu;
  87         int                     refcnt;
  88         struct hlist_head       vni_list[VNI_HASH_SIZE];
  89 };
  90
  91 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  92 {
  93         __u32 vnid;
  94
  95         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  96         return hash_32(vnid, VNI_HASH_BITS);
  97 }
  98
  99 static __be64 vni_to_tunnel_id(const __u8 *vni)
 100 {
 101 #ifdef __BIG_ENDIAN
 102         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
 103 #else
 104         return (__force __be64)(((__force u64)vni[0] << 40) |
 105                                 ((__force u64)vni[1] << 48) |
 106                                 ((__force u64)vni[2] << 56));
 107 #endif
 108 }
 109
 110 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 111 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 112 {
 113 #ifdef __BIG_ENDIAN
 114         vni[0] = (__force __u8)(tun_id >> 16);
 115         vni[1] = (__force __u8)(tun_id >> 8);
 116         vni[2] = (__force __u8)tun_id;
 117 #else
 118         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 119         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 120         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 121 #endif
 122 }
 123
 124 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
 125 {
 126         return !memcmp(vni, &tun_id[5], 3);
 127 }
 128
 129 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 130 {
 131         return gs->sock->sk->sk_family;
 132 }
 133
 134 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 135                                         __be32 addr, u8 vni[])
 136 {
 137         struct hlist_head *vni_list_head;
 138         struct geneve_dev_node *node;
 139         __u32 hash;
 140
 141         /* Find the device for this VNI */
 142         hash = geneve_net_vni_hash(vni);
 143         vni_list_head = &gs->vni_list[hash];
 144         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 145                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 146                     addr == node->geneve->cfg.info.key.u.ipv4.dst)
 147                         return node->geneve;
 148         }
 149         return NULL;
 150 }
 151
 152 #if IS_ENABLED(CONFIG_IPV6)
 153 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 154                                          struct in6_addr addr6, u8 vni[])
 155 {
 156         struct hlist_head *vni_list_head;
 157         struct geneve_dev_node *node;
 158         __u32 hash;
 159
 160         /* Find the device for this VNI */
 161         hash = geneve_net_vni_hash(vni);
 162         vni_list_head = &gs->vni_list[hash];
 163         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 164                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 165                     ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
 166                         return node->geneve;
 167         }
 168         return NULL;
 169 }
 170 #endif
 171
 172 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 173 {
 174         return (struct genevehdr *)(udp_hdr(skb) + 1);
 175 }
 176
 177 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 178                                             struct sk_buff *skb)
 179 {
 180         static u8 zero_vni[3];
 181         u8 *vni;
 182
 183         if (geneve_get_sk_family(gs) == AF_INET) {
 184                 struct iphdr *iph;
 185                 __be32 addr;
 186
 187                 iph = ip_hdr(skb); /* outer IP header... */
 188
 189                 if (gs->collect_md) {
 190                         vni = zero_vni;
 191                         addr = 0;
 192                 } else {
 193                         vni = geneve_hdr(skb)->vni;
 194                         addr = iph->saddr;
 195                 }
 196
 197                 return geneve_lookup(gs, addr, vni);
 198 #if IS_ENABLED(CONFIG_IPV6)
 199         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 200                 static struct in6_addr zero_addr6;
 201                 struct ipv6hdr *ip6h;
 202                 struct in6_addr addr6;
 203
 204                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 205
 206                 if (gs->collect_md) {
 207                         vni = zero_vni;
 208                         addr6 = zero_addr6;
 209                 } else {
 210                         vni = geneve_hdr(skb)->vni;
 211                         addr6 = ip6h->saddr;
 212                 }
 213
 214                 return geneve6_lookup(gs, addr6, vni);
 215 #endif
 216         }
 217         return NULL;
 218 }
 219
 220 /* geneve receive/decap routine */
 221 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 222                       struct sk_buff *skb)
 223 {
 224         struct genevehdr *gnvh = geneve_hdr(skb);
 225         struct metadata_dst *tun_dst = NULL;
 226         unsigned int len;
 227         int nh, err = 0;
 228         void *oiph;
 229
 230         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 231                 IP_TUNNEL_DECLARE_FLAGS(flags) = { };
 232
 233                 __set_bit(IP_TUNNEL_KEY_BIT, flags);
 234                 __assign_bit(IP_TUNNEL_OAM_BIT, flags, gnvh->oam);
 235                 __assign_bit(IP_TUNNEL_CRIT_OPT_BIT, flags, gnvh->critical);
 236
 237                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 238                                          vni_to_tunnel_id(gnvh->vni),
 239                                          gnvh->opt_len * 4);
 240                 if (!tun_dst) {
 241                         dev_dstats_rx_dropped(geneve->dev);
 242                         goto drop;
 243                 }
 244                 /* Update tunnel dst according to Geneve options. */
 245                 ip_tunnel_flags_zero(flags);
 246                 __set_bit(IP_TUNNEL_GENEVE_OPT_BIT, flags);
 247                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 248                                         gnvh->options, gnvh->opt_len * 4,
 249                                         flags);
 250         } else {
 251                 /* Drop packets w/ critical options,
 252                  * since we don't support any...
 253                  */
 254                 if (gnvh->critical) {
 255                         DEV_STATS_INC(geneve->dev, rx_frame_errors);
 256                         DEV_STATS_INC(geneve->dev, rx_errors);
 257                         goto drop;
 258                 }
 259         }
 260
 261         if (tun_dst)
 262                 skb_dst_set(skb, &tun_dst->dst);
 263
 264         if (gnvh->proto_type == htons(ETH_P_TEB)) {
 265                 skb_reset_mac_header(skb);
 266                 skb->protocol = eth_type_trans(skb, geneve->dev);
 267                 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 268
 269                 /* Ignore packet loops (and multicast echo) */
 270                 if (ether_addr_equal(eth_hdr(skb)->h_source,
 271                                      geneve->dev->dev_addr)) {
 272                         DEV_STATS_INC(geneve->dev, rx_errors);
 273                         goto drop;
 274                 }
 275         } else {
 276                 skb_reset_mac_header(skb);
 277                 skb->dev = geneve->dev;
 278                 skb->pkt_type = PACKET_HOST;
 279         }
 280
 281         /* Save offset of outer header relative to skb->head,
 282          * because we are going to reset the network header to the inner header
 283          * and might change skb->head.
 284          */
 285         nh = skb_network_header(skb) - skb->head;
 286
 287         skb_reset_network_header(skb);
 288
 289         if (!pskb_inet_may_pull(skb)) {
 290                 DEV_STATS_INC(geneve->dev, rx_length_errors);
 291                 DEV_STATS_INC(geneve->dev, rx_errors);
 292                 goto drop;
 293         }
 294
 295         /* Get the outer header. */
 296         oiph = skb->head + nh;
 297
 298         if (geneve_get_sk_family(gs) == AF_INET)
 299                 err = IP_ECN_decapsulate(oiph, skb);
 300 #if IS_ENABLED(CONFIG_IPV6)
 301         else
 302                 err = IP6_ECN_decapsulate(oiph, skb);
 303 #endif
 304
 305         if (unlikely(err)) {
 306                 if (log_ecn_error) {
 307                         if (geneve_get_sk_family(gs) == AF_INET)
 308                                 net_info_ratelimited("non-ECT from %pI4 "
 309                                                      "with TOS=%#x\n",
 310                                                      &((struct iphdr *)oiph)->saddr,
 311                                                      ((struct iphdr *)oiph)->tos);
 312 #if IS_ENABLED(CONFIG_IPV6)
 313                         else
 314                                 net_info_ratelimited("non-ECT from %pI6\n",
 315                                                      &((struct ipv6hdr *)oiph)->saddr);
 316 #endif
 317                 }
 318                 if (err > 1) {
 319                         DEV_STATS_INC(geneve->dev, rx_frame_errors);
 320                         DEV_STATS_INC(geneve->dev, rx_errors);
 321                         goto drop;
 322                 }
 323         }
 324
 325         len = skb->len;
 326         err = gro_cells_receive(&geneve->gro_cells, skb);
 327         if (likely(err == NET_RX_SUCCESS))
 328                 dev_dstats_rx_add(geneve->dev, len);
 329
 330         return;
 331 drop:
 332         /* Consume bad packet */
 333         kfree_skb(skb);
 334 }
 335
 336 /* Setup stats when device is created */
 337 static int geneve_init(struct net_device *dev)
 338 {
 339         struct geneve_dev *geneve = netdev_priv(dev);
 340         int err;
 341
 342         err = gro_cells_init(&geneve->gro_cells, dev);
 343         if (err)
 344                 return err;
 345
 346         err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
 347         if (err) {
 348                 gro_cells_destroy(&geneve->gro_cells);
 349                 return err;
 350         }
 351         netdev_lockdep_set_classes(dev);
 352         return 0;
 353 }
 354
 355 static void geneve_uninit(struct net_device *dev)
 356 {
 357         struct geneve_dev *geneve = netdev_priv(dev);
 358
 359         dst_cache_destroy(&geneve->cfg.info.dst_cache);
 360         gro_cells_destroy(&geneve->gro_cells);
 361 }
 362
 363 /* Callback from net/ipv4/udp.c to receive packets */
 364 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 365 {
 366         struct genevehdr *geneveh;
 367         struct geneve_dev *geneve;
 368         struct geneve_sock *gs;
 369         __be16 inner_proto;
 370         int opts_len;
 371
 372         /* Need UDP and Geneve header to be present */
 373         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 374                 goto drop;
 375
 376         /* Return packets with reserved bits set */
 377         geneveh = geneve_hdr(skb);
 378         if (unlikely(geneveh->ver != GENEVE_VER))
 379                 goto drop;
 380
 381         gs = rcu_dereference_sk_user_data(sk);
 382         if (!gs)
 383                 goto drop;
 384
 385         geneve = geneve_lookup_skb(gs, skb);
 386         if (!geneve)
 387                 goto drop;
 388
 389         inner_proto = geneveh->proto_type;
 390
 391         if (unlikely((!geneve->cfg.inner_proto_inherit &&
 392                       inner_proto != htons(ETH_P_TEB)))) {
 393                 dev_dstats_rx_dropped(geneve->dev);
 394                 goto drop;
 395         }
 396
 397         opts_len = geneveh->opt_len * 4;
 398         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto,
 399                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 400                 dev_dstats_rx_dropped(geneve->dev);
 401                 goto drop;
 402         }
 403
 404         geneve_rx(geneve, gs, skb);
 405         return 0;
 406
 407 drop:
 408         /* Consume bad packet */
 409         kfree_skb(skb);
 410         return 0;
 411 }
 412
 413 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
 414 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
 415 {
 416         struct genevehdr *geneveh;
 417         struct geneve_sock *gs;
 418         u8 zero_vni[3] = { 0 };
 419         u8 *vni = zero_vni;
 420
 421         if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
 422                 return -EINVAL;
 423
 424         geneveh = geneve_hdr(skb);
 425         if (geneveh->ver != GENEVE_VER)
 426                 return -EINVAL;
 427
 428         if (geneveh->proto_type != htons(ETH_P_TEB))
 429                 return -EINVAL;
 430
 431         gs = rcu_dereference_sk_user_data(sk);
 432         if (!gs)
 433                 return -ENOENT;
 434
 435         if (geneve_get_sk_family(gs) == AF_INET) {
 436                 struct iphdr *iph = ip_hdr(skb);
 437                 __be32 addr4 = 0;
 438
 439                 if (!gs->collect_md) {
 440                         vni = geneve_hdr(skb)->vni;
 441                         addr4 = iph->daddr;
 442                 }
 443
 444                 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
 445         }
 446
 447 #if IS_ENABLED(CONFIG_IPV6)
 448         if (geneve_get_sk_family(gs) == AF_INET6) {
 449                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 450                 struct in6_addr addr6;
 451
 452                 memset(&addr6, 0, sizeof(struct in6_addr));
 453
 454                 if (!gs->collect_md) {
 455                         vni = geneve_hdr(skb)->vni;
 456                         addr6 = ip6h->daddr;
 457                 }
 458
 459                 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
 460         }
 461 #endif
 462
 463         return -EPFNOSUPPORT;
 464 }
 465
 466 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 467                                          __be16 port, bool ipv6_rx_csum)
 468 {
 469         struct socket *sock;
 470         struct udp_port_cfg udp_conf;
 471         int err;
 472
 473         memset(&udp_conf, 0, sizeof(udp_conf));
 474
 475         if (ipv6) {
 476                 udp_conf.family = AF_INET6;
 477                 udp_conf.ipv6_v6only = 1;
 478                 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
 479         } else {
 480                 udp_conf.family = AF_INET;
 481                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 482         }
 483
 484         udp_conf.local_udp_port = port;
 485
 486         /* Open UDP socket */
 487         err = udp_sock_create(net, &udp_conf, &sock);
 488         if (err < 0)
 489                 return ERR_PTR(err);
 490
 491         udp_allow_gso(sock->sk);
 492         return sock;
 493 }
 494
 495 static int geneve_hlen(struct genevehdr *gh)
 496 {
 497         return sizeof(*gh) + gh->opt_len * 4;
 498 }
 499
 500 static struct sk_buff *geneve_gro_receive(struct sock *sk,
 501                                           struct list_head *head,
 502                                           struct sk_buff *skb)
 503 {
 504         struct sk_buff *pp = NULL;
 505         struct sk_buff *p;
 506         struct genevehdr *gh, *gh2;
 507         unsigned int hlen, gh_len, off_gnv;
 508         const struct packet_offload *ptype;
 509         __be16 type;
 510         int flush = 1;
 511
 512         off_gnv = skb_gro_offset(skb);
 513         hlen = off_gnv + sizeof(*gh);
 514         gh = skb_gro_header(skb, hlen, off_gnv);
 515         if (unlikely(!gh))
 516                 goto out;
 517
 518         if (gh->ver != GENEVE_VER || gh->oam)
 519                 goto out;
 520         gh_len = geneve_hlen(gh);
 521
 522         hlen = off_gnv + gh_len;
 523         if (!skb_gro_may_pull(skb, hlen)) {
 524                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 525                 if (unlikely(!gh))
 526                         goto out;
 527         }
 528
 529         list_for_each_entry(p, head, list) {
 530                 if (!NAPI_GRO_CB(p)->same_flow)
 531                         continue;
 532
 533                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 534                 if (gh->opt_len != gh2->opt_len ||
 535                     memcmp(gh, gh2, gh_len)) {
 536                         NAPI_GRO_CB(p)->same_flow = 0;
 537                         continue;
 538                 }
 539         }
 540
 541         skb_gro_pull(skb, gh_len);
 542         skb_gro_postpull_rcsum(skb, gh, gh_len);
 543         type = gh->proto_type;
 544         if (likely(type == htons(ETH_P_TEB)))
 545                 return call_gro_receive(eth_gro_receive, head, skb);
 546
 547         ptype = gro_find_receive_by_type(type);
 548         if (!ptype)
 549                 goto out;
 550
 551         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 552         flush = 0;
 553
 554 out:
 555         skb_gro_flush_final(skb, pp, flush);
 556
 557         return pp;
 558 }
 559
 560 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 561                                int nhoff)
 562 {
 563         struct genevehdr *gh;
 564         struct packet_offload *ptype;
 565         __be16 type;
 566         int gh_len;
 567         int err = -ENOSYS;
 568
 569         gh = (struct genevehdr *)(skb->data + nhoff);
 570         gh_len = geneve_hlen(gh);
 571         type = gh->proto_type;
 572
 573         /* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */
 574         if (likely(type == htons(ETH_P_TEB)))
 575                 return eth_gro_complete(skb, nhoff + gh_len);
 576
 577         ptype = gro_find_complete_by_type(type);
 578         if (ptype)
 579                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 580
 581         skb_set_inner_mac_header(skb, nhoff + gh_len);
 582
 583         return err;
 584 }
 585
 586 /* Create new listen socket if needed */
 587 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 588                                                 bool ipv6, bool ipv6_rx_csum)
 589 {
 590         struct geneve_net *gn = net_generic(net, geneve_net_id);
 591         struct geneve_sock *gs;
 592         struct socket *sock;
 593         struct udp_tunnel_sock_cfg tunnel_cfg;
 594         int h;
 595
 596         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 597         if (!gs)
 598                 return ERR_PTR(-ENOMEM);
 599
 600         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 601         if (IS_ERR(sock)) {
 602                 kfree(gs);
 603                 return ERR_CAST(sock);
 604         }
 605
 606         gs->sock = sock;
 607         gs->refcnt = 1;
 608         for (h = 0; h < VNI_HASH_SIZE; ++h)
 609                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 610
 611         /* Initialize the geneve udp offloads structure */
 612         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 613
 614         /* Mark socket as an encapsulation socket */
 615         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 616         tunnel_cfg.sk_user_data = gs;
 617         tunnel_cfg.encap_type = 1;
 618         tunnel_cfg.gro_receive = geneve_gro_receive;
 619         tunnel_cfg.gro_complete = geneve_gro_complete;
 620         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 621         tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
 622         tunnel_cfg.encap_destroy = NULL;
 623         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 624         list_add(&gs->list, &gn->sock_list);
 625         return gs;
 626 }
 627
 628 static void __geneve_sock_release(struct geneve_sock *gs)
 629 {
 630         if (!gs || --gs->refcnt)
 631                 return;
 632
 633         list_del(&gs->list);
 634         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 635         udp_tunnel_sock_release(gs->sock);
 636         kfree_rcu(gs, rcu);
 637 }
 638
 639 static void geneve_sock_release(struct geneve_dev *geneve)
 640 {
 641         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 642 #if IS_ENABLED(CONFIG_IPV6)
 643         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 644
 645         rcu_assign_pointer(geneve->sock6, NULL);
 646 #endif
 647
 648         rcu_assign_pointer(geneve->sock4, NULL);
 649         synchronize_net();
 650
 651         __geneve_sock_release(gs4);
 652 #if IS_ENABLED(CONFIG_IPV6)
 653         __geneve_sock_release(gs6);
 654 #endif
 655 }
 656
 657 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 658                                             sa_family_t family,
 659                                             __be16 dst_port)
 660 {
 661         struct geneve_sock *gs;
 662
 663         list_for_each_entry(gs, &gn->sock_list, list) {
 664                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 665                     geneve_get_sk_family(gs) == family) {
 666                         return gs;
 667                 }
 668         }
 669         return NULL;
 670 }
 671
 672 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 673 {
 674         struct net *net = geneve->net;
 675         struct geneve_net *gn = net_generic(net, geneve_net_id);
 676         struct geneve_dev_node *node;
 677         struct geneve_sock *gs;
 678         __u8 vni[3];
 679         __u32 hash;
 680
 681         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
 682         if (gs) {
 683                 gs->refcnt++;
 684                 goto out;
 685         }
 686
 687         gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
 688                                   geneve->cfg.use_udp6_rx_checksums);
 689         if (IS_ERR(gs))
 690                 return PTR_ERR(gs);
 691
 692 out:
 693         gs->collect_md = geneve->cfg.collect_md;
 694 #if IS_ENABLED(CONFIG_IPV6)
 695         if (ipv6) {
 696                 rcu_assign_pointer(geneve->sock6, gs);
 697                 node = &geneve->hlist6;
 698         } else
 699 #endif
 700         {
 701                 rcu_assign_pointer(geneve->sock4, gs);
 702                 node = &geneve->hlist4;
 703         }
 704         node->geneve = geneve;
 705
 706         tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
 707         hash = geneve_net_vni_hash(vni);
 708         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 709         return 0;
 710 }
 711
 712 static int geneve_open(struct net_device *dev)
 713 {
 714         struct geneve_dev *geneve = netdev_priv(dev);
 715         bool metadata = geneve->cfg.collect_md;
 716         bool ipv4, ipv6;
 717         int ret = 0;
 718
 719         ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 720         ipv4 = !ipv6 || metadata;
 721 #if IS_ENABLED(CONFIG_IPV6)
 722         if (ipv6) {
 723                 ret = geneve_sock_add(geneve, true);
 724                 if (ret < 0 && ret != -EAFNOSUPPORT)
 725                         ipv4 = false;
 726         }
 727 #endif
 728         if (ipv4)
 729                 ret = geneve_sock_add(geneve, false);
 730         if (ret < 0)
 731                 geneve_sock_release(geneve);
 732
 733         return ret;
 734 }
 735
 736 static int geneve_stop(struct net_device *dev)
 737 {
 738         struct geneve_dev *geneve = netdev_priv(dev);
 739
 740         hlist_del_init_rcu(&geneve->hlist4.hlist);
 741 #if IS_ENABLED(CONFIG_IPV6)
 742         hlist_del_init_rcu(&geneve->hlist6.hlist);
 743 #endif
 744         geneve_sock_release(geneve);
 745         return 0;
 746 }
 747
 748 static void geneve_build_header(struct genevehdr *geneveh,
 749                                 const struct ip_tunnel_info *info,
 750                                 __be16 inner_proto)
 751 {
 752         geneveh->ver = GENEVE_VER;
 753         geneveh->opt_len = info->options_len / 4;
 754         geneveh->oam = test_bit(IP_TUNNEL_OAM_BIT, info->key.tun_flags);
 755         geneveh->critical = test_bit(IP_TUNNEL_CRIT_OPT_BIT,
 756                                      info->key.tun_flags);
 757         geneveh->rsvd1 = 0;
 758         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 759         geneveh->proto_type = inner_proto;
 760         geneveh->rsvd2 = 0;
 761
 762         if (test_bit(IP_TUNNEL_GENEVE_OPT_BIT, info->key.tun_flags))
 763                 ip_tunnel_info_opts_get(geneveh->options, info);
 764 }
 765
 766 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 767                             const struct ip_tunnel_info *info,
 768                             bool xnet, int ip_hdr_len,
 769                             bool inner_proto_inherit)
 770 {
 771         bool udp_sum = test_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags);
 772         struct genevehdr *gnvh;
 773         __be16 inner_proto;
 774         int min_headroom;
 775         int err;
 776
 777         skb_reset_mac_header(skb);
 778         skb_scrub_packet(skb, xnet);
 779
 780         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 781                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 782         err = skb_cow_head(skb, min_headroom);
 783         if (unlikely(err))
 784                 goto free_dst;
 785
 786         err = udp_tunnel_handle_offloads(skb, udp_sum);
 787         if (err)
 788                 goto free_dst;
 789
 790         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 791         inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB);
 792         geneve_build_header(gnvh, info, inner_proto);
 793         skb_set_inner_protocol(skb, inner_proto);
 794         return 0;
 795
 796 free_dst:
 797         dst_release(dst);
 798         return err;
 799 }
 800
 801 static u8 geneve_get_dsfield(struct sk_buff *skb, struct net_device *dev,
 802                              const struct ip_tunnel_info *info,
 803                              bool *use_cache)
 804 {
 805         struct geneve_dev *geneve = netdev_priv(dev);
 806         u8 dsfield;
 807
 808         dsfield = info->key.tos;
 809         if (dsfield == 1 && !geneve->cfg.collect_md) {
 810                 dsfield = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 811                 *use_cache = false;
 812         }
 813
 814         return dsfield;
 815 }
 816
 817 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 818                            struct geneve_dev *geneve,
 819                            const struct ip_tunnel_info *info)
 820 {
 821         bool inner_proto_inherit = geneve->cfg.inner_proto_inherit;
 822         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 823         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 824         const struct ip_tunnel_key *key = &info->key;
 825         struct rtable *rt;
 826         bool use_cache;
 827         __u8 tos, ttl;
 828         __be16 df = 0;
 829         __be32 saddr;
 830         __be16 sport;
 831         int err;
 832
 833         if (skb_vlan_inet_prepare(skb, inner_proto_inherit))
 834                 return -EINVAL;
 835
 836         if (!gs4)
 837                 return -EIO;
 838
 839         use_cache = ip_tunnel_dst_cache_usable(skb, info);
 840         tos = geneve_get_dsfield(skb, dev, info, &use_cache);
 841         sport = udp_flow_src_port(geneve->net, skb,
 842                                   geneve->cfg.port_min,
 843                                   geneve->cfg.port_max, true);
 844
 845         rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
 846                                    &info->key,
 847                                    sport, geneve->cfg.info.key.tp_dst, tos,
 848                                    use_cache ?
 849                                    (struct dst_cache *)&info->dst_cache : NULL);
 850         if (IS_ERR(rt))
 851                 return PTR_ERR(rt);
 852
 853         err = skb_tunnel_check_pmtu(skb, &rt->dst,
 854                                     GENEVE_IPV4_HLEN + info->options_len,
 855                                     netif_is_any_bridge_port(dev));
 856         if (err < 0) {
 857                 dst_release(&rt->dst);
 858                 return err;
 859         } else if (err) {
 860                 struct ip_tunnel_info *info;
 861
 862                 info = skb_tunnel_info(skb);
 863                 if (info) {
 864                         struct ip_tunnel_info *unclone;
 865
 866                         unclone = skb_tunnel_info_unclone(skb);
 867                         if (unlikely(!unclone)) {
 868                                 dst_release(&rt->dst);
 869                                 return -ENOMEM;
 870                         }
 871
 872                         unclone->key.u.ipv4.dst = saddr;
 873                         unclone->key.u.ipv4.src = info->key.u.ipv4.dst;
 874                 }
 875
 876                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 877                         dst_release(&rt->dst);
 878                         return -EINVAL;
 879                 }
 880
 881                 skb->protocol = eth_type_trans(skb, geneve->dev);
 882                 __netif_rx(skb);
 883                 dst_release(&rt->dst);
 884                 return -EMSGSIZE;
 885         }
 886
 887         tos = ip_tunnel_ecn_encap(tos, ip_hdr(skb), skb);
 888         if (geneve->cfg.collect_md) {
 889                 ttl = key->ttl;
 890
 891                 df = test_bit(IP_TUNNEL_DONT_FRAGMENT_BIT, key->tun_flags) ?
 892                      htons(IP_DF) : 0;
 893         } else {
 894                 if (geneve->cfg.ttl_inherit)
 895                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 896                 else
 897                         ttl = key->ttl;
 898                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 899
 900                 if (geneve->cfg.df == GENEVE_DF_SET) {
 901                         df = htons(IP_DF);
 902                 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
 903                         struct ethhdr *eth = skb_eth_hdr(skb);
 904
 905                         if (ntohs(eth->h_proto) == ETH_P_IPV6) {
 906                                 df = htons(IP_DF);
 907                         } else if (ntohs(eth->h_proto) == ETH_P_IP) {
 908                                 struct iphdr *iph = ip_hdr(skb);
 909
 910                                 if (iph->frag_off & htons(IP_DF))
 911                                         df = htons(IP_DF);
 912                         }
 913                 }
 914         }
 915
 916         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
 917                                inner_proto_inherit);
 918         if (unlikely(err))
 919                 return err;
 920
 921         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, saddr, info->key.u.ipv4.dst,
 922                             tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
 923                             !net_eq(geneve->net, dev_net(geneve->dev)),
 924                             !test_bit(IP_TUNNEL_CSUM_BIT,
 925                                       info->key.tun_flags));
 926         return 0;
 927 }
 928
 929 #if IS_ENABLED(CONFIG_IPV6)
 930 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 931                             struct geneve_dev *geneve,
 932                             const struct ip_tunnel_info *info)
 933 {
 934         bool inner_proto_inherit = geneve->cfg.inner_proto_inherit;
 935         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 936         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 937         const struct ip_tunnel_key *key = &info->key;
 938         struct dst_entry *dst = NULL;
 939         struct in6_addr saddr;
 940         bool use_cache;
 941         __u8 prio, ttl;
 942         __be16 sport;
 943         int err;
 944
 945         if (skb_vlan_inet_prepare(skb, inner_proto_inherit))
 946                 return -EINVAL;
 947
 948         if (!gs6)
 949                 return -EIO;
 950
 951         use_cache = ip_tunnel_dst_cache_usable(skb, info);
 952         prio = geneve_get_dsfield(skb, dev, info, &use_cache);
 953         sport = udp_flow_src_port(geneve->net, skb,
 954                                   geneve->cfg.port_min,
 955                                   geneve->cfg.port_max, true);
 956
 957         dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
 958                                      &saddr, key, sport,
 959                                      geneve->cfg.info.key.tp_dst, prio,
 960                                      use_cache ?
 961                                      (struct dst_cache *)&info->dst_cache : NULL);
 962         if (IS_ERR(dst))
 963                 return PTR_ERR(dst);
 964
 965         err = skb_tunnel_check_pmtu(skb, dst,
 966                                     GENEVE_IPV6_HLEN + info->options_len,
 967                                     netif_is_any_bridge_port(dev));
 968         if (err < 0) {
 969                 dst_release(dst);
 970                 return err;
 971         } else if (err) {
 972                 struct ip_tunnel_info *info = skb_tunnel_info(skb);
 973
 974                 if (info) {
 975                         struct ip_tunnel_info *unclone;
 976
 977                         unclone = skb_tunnel_info_unclone(skb);
 978                         if (unlikely(!unclone)) {
 979                                 dst_release(dst);
 980                                 return -ENOMEM;
 981                         }
 982
 983                         unclone->key.u.ipv6.dst = saddr;
 984                         unclone->key.u.ipv6.src = info->key.u.ipv6.dst;
 985                 }
 986
 987                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 988                         dst_release(dst);
 989                         return -EINVAL;
 990                 }
 991
 992                 skb->protocol = eth_type_trans(skb, geneve->dev);
 993                 __netif_rx(skb);
 994                 dst_release(dst);
 995                 return -EMSGSIZE;
 996         }
 997
 998         prio = ip_tunnel_ecn_encap(prio, ip_hdr(skb), skb);
 999         if (geneve->cfg.collect_md) {
1000                 ttl = key->ttl;
1001         } else {
1002                 if (geneve->cfg.ttl_inherit)
1003                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1004                 else
1005                         ttl = key->ttl;
1006                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1007         }
1008         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
1009                                inner_proto_inherit);
1010         if (unlikely(err))
1011                 return err;
1012
1013         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1014                              &saddr, &key->u.ipv6.dst, prio, ttl,
1015                              info->key.label, sport, geneve->cfg.info.key.tp_dst,
1016                              !test_bit(IP_TUNNEL_CSUM_BIT,
1017                                        info->key.tun_flags));
1018         return 0;
1019 }
1020 #endif
1021
1022 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1023 {
1024         struct geneve_dev *geneve = netdev_priv(dev);
1025         struct ip_tunnel_info *info = NULL;
1026         int err;
1027
1028         if (geneve->cfg.collect_md) {
1029                 info = skb_tunnel_info(skb);
1030                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1031                         netdev_dbg(dev, "no tunnel metadata\n");
1032                         dev_kfree_skb(skb);
1033                         dev_dstats_tx_dropped(dev);
1034                         return NETDEV_TX_OK;
1035                 }
1036         } else {
1037                 info = &geneve->cfg.info;
1038         }
1039
1040         rcu_read_lock();
1041 #if IS_ENABLED(CONFIG_IPV6)
1042         if (info->mode & IP_TUNNEL_INFO_IPV6)
1043                 err = geneve6_xmit_skb(skb, dev, geneve, info);
1044         else
1045 #endif
1046                 err = geneve_xmit_skb(skb, dev, geneve, info);
1047         rcu_read_unlock();
1048
1049         if (likely(!err))
1050                 return NETDEV_TX_OK;
1051
1052         if (err != -EMSGSIZE)
1053                 dev_kfree_skb(skb);
1054
1055         if (err == -ELOOP)
1056                 DEV_STATS_INC(dev, collisions);
1057         else if (err == -ENETUNREACH)
1058                 DEV_STATS_INC(dev, tx_carrier_errors);
1059
1060         DEV_STATS_INC(dev, tx_errors);
1061         return NETDEV_TX_OK;
1062 }
1063
1064 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1065 {
1066         if (new_mtu > dev->max_mtu)
1067                 new_mtu = dev->max_mtu;
1068         else if (new_mtu < dev->min_mtu)
1069                 new_mtu = dev->min_mtu;
1070
1071         WRITE_ONCE(dev->mtu, new_mtu);
1072         return 0;
1073 }
1074
1075 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1076 {
1077         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1078         struct geneve_dev *geneve = netdev_priv(dev);
1079         __be16 sport;
1080
1081         if (ip_tunnel_info_af(info) == AF_INET) {
1082                 struct rtable *rt;
1083                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1084                 bool use_cache;
1085                 __be32 saddr;
1086                 u8 tos;
1087
1088                 if (!gs4)
1089                         return -EIO;
1090
1091                 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1092                 tos = geneve_get_dsfield(skb, dev, info, &use_cache);
1093                 sport = udp_flow_src_port(geneve->net, skb,
1094                                           geneve->cfg.port_min,
1095                                           geneve->cfg.port_max, true);
1096
1097                 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
1098                                            &info->key,
1099                                            sport, geneve->cfg.info.key.tp_dst,
1100                                            tos,
1101                                            use_cache ? &info->dst_cache : NULL);
1102                 if (IS_ERR(rt))
1103                         return PTR_ERR(rt);
1104
1105                 ip_rt_put(rt);
1106                 info->key.u.ipv4.src = saddr;
1107 #if IS_ENABLED(CONFIG_IPV6)
1108         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1109                 struct dst_entry *dst;
1110                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1111                 struct in6_addr saddr;
1112                 bool use_cache;
1113                 u8 prio;
1114
1115                 if (!gs6)
1116                         return -EIO;
1117
1118                 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1119                 prio = geneve_get_dsfield(skb, dev, info, &use_cache);
1120                 sport = udp_flow_src_port(geneve->net, skb,
1121                                           geneve->cfg.port_min,
1122                                           geneve->cfg.port_max, true);
1123
1124                 dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
1125                                              &saddr, &info->key, sport,
1126                                              geneve->cfg.info.key.tp_dst, prio,
1127                                              use_cache ? &info->dst_cache : NULL);
1128                 if (IS_ERR(dst))
1129                         return PTR_ERR(dst);
1130
1131                 dst_release(dst);
1132                 info->key.u.ipv6.src = saddr;
1133 #endif
1134         } else {
1135                 return -EINVAL;
1136         }
1137
1138         info->key.tp_src = sport;
1139         info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1140         return 0;
1141 }
1142
1143 static const struct net_device_ops geneve_netdev_ops = {
1144         .ndo_init               = geneve_init,
1145         .ndo_uninit             = geneve_uninit,
1146         .ndo_open               = geneve_open,
1147         .ndo_stop               = geneve_stop,
1148         .ndo_start_xmit         = geneve_xmit,
1149         .ndo_change_mtu         = geneve_change_mtu,
1150         .ndo_validate_addr      = eth_validate_addr,
1151         .ndo_set_mac_address    = eth_mac_addr,
1152         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1153 };
1154
1155 static void geneve_get_drvinfo(struct net_device *dev,
1156                                struct ethtool_drvinfo *drvinfo)
1157 {
1158         strscpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1159         strscpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1160 }
1161
1162 static const struct ethtool_ops geneve_ethtool_ops = {
1163         .get_drvinfo    = geneve_get_drvinfo,
1164         .get_link       = ethtool_op_get_link,
1165 };
1166
1167 /* Info for udev, that this is a virtual tunnel endpoint */
1168 static const struct device_type geneve_type = {
1169         .name = "geneve",
1170 };
1171
1172 /* Calls the ndo_udp_tunnel_add of the caller in order to
1173  * supply the listening GENEVE udp ports. Callers are expected
1174  * to implement the ndo_udp_tunnel_add.
1175  */
1176 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1177 {
1178         struct net *net = dev_net(dev);
1179         struct geneve_net *gn = net_generic(net, geneve_net_id);
1180         struct geneve_sock *gs;
1181
1182         rcu_read_lock();
1183         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1184                 if (push) {
1185                         udp_tunnel_push_rx_port(dev, gs->sock,
1186                                                 UDP_TUNNEL_TYPE_GENEVE);
1187                 } else {
1188                         udp_tunnel_drop_rx_port(dev, gs->sock,
1189                                                 UDP_TUNNEL_TYPE_GENEVE);
1190                 }
1191         }
1192         rcu_read_unlock();
1193 }
1194
1195 /* Initialize the device structure. */
1196 static void geneve_setup(struct net_device *dev)
1197 {
1198         ether_setup(dev);
1199
1200         dev->netdev_ops = &geneve_netdev_ops;
1201         dev->ethtool_ops = &geneve_ethtool_ops;
1202         dev->needs_free_netdev = true;
1203
1204         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1205
1206         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1207         dev->features    |= NETIF_F_RXCSUM;
1208         dev->features    |= NETIF_F_GSO_SOFTWARE;
1209
1210         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1211         dev->hw_features |= NETIF_F_RXCSUM;
1212         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1213
1214         dev->pcpu_stat_type = NETDEV_PCPU_STAT_DSTATS;
1215         /* MTU range: 68 - (something less than 65535) */
1216         dev->min_mtu = ETH_MIN_MTU;
1217         /* The max_mtu calculation does not take account of GENEVE
1218          * options, to avoid excluding potentially valid
1219          * configurations. This will be further reduced by IPvX hdr size.
1220          */
1221         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1222
1223         netif_keep_dst(dev);
1224         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1225         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1226         dev->lltx = true;
1227         eth_hw_addr_random(dev);
1228 }
1229
1230 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1231         [IFLA_GENEVE_UNSPEC]            = { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT },
1232         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1233         [IFLA_GENEVE_REMOTE]            = { .len = sizeof_field(struct iphdr, daddr) },
1234         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1235         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1236         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1237         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1238         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1239         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1240         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1241         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1242         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1243         [IFLA_GENEVE_TTL_INHERIT]       = { .type = NLA_U8 },
1244         [IFLA_GENEVE_DF]                = { .type = NLA_U8 },
1245         [IFLA_GENEVE_INNER_PROTO_INHERIT]       = { .type = NLA_FLAG },
1246         [IFLA_GENEVE_PORT_RANGE]        = NLA_POLICY_EXACT_LEN(sizeof(struct ifla_geneve_port_range)),
1247 };
1248
1249 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1250                            struct netlink_ext_ack *extack)
1251 {
1252         if (tb[IFLA_ADDRESS]) {
1253                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1254                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1255                                             "Provided link layer address is not Ethernet");
1256                         return -EINVAL;
1257                 }
1258
1259                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1260                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1261                                             "Provided Ethernet address is not unicast");
1262                         return -EADDRNOTAVAIL;
1263                 }
1264         }
1265
1266         if (!data) {
1267                 NL_SET_ERR_MSG(extack,
1268                                "Not enough attributes provided to perform the operation");
1269                 return -EINVAL;
1270         }
1271
1272         if (data[IFLA_GENEVE_ID]) {
1273                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1274
1275                 if (vni >= GENEVE_N_VID) {
1276                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1277                                             "Geneve ID must be lower than 16777216");
1278                         return -ERANGE;
1279                 }
1280         }
1281
1282         if (data[IFLA_GENEVE_DF]) {
1283                 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1284
1285                 if (df < 0 || df > GENEVE_DF_MAX) {
1286                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1287                                             "Invalid DF attribute");
1288                         return -EINVAL;
1289                 }
1290         }
1291
1292         if (data[IFLA_GENEVE_PORT_RANGE]) {
1293                 const struct ifla_geneve_port_range *p;
1294
1295                 p = nla_data(data[IFLA_GENEVE_PORT_RANGE]);
1296                 if (ntohs(p->high) < ntohs(p->low)) {
1297                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_PORT_RANGE],
1298                                             "Invalid source port range");
1299                         return -EINVAL;
1300                 }
1301         }
1302
1303         return 0;
1304 }
1305
1306 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1307                                           const struct ip_tunnel_info *info,
1308                                           bool *tun_on_same_port,
1309                                           bool *tun_collect_md)
1310 {
1311         struct geneve_dev *geneve, *t = NULL;
1312
1313         *tun_on_same_port = false;
1314         *tun_collect_md = false;
1315         list_for_each_entry(geneve, &gn->geneve_list, next) {
1316                 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1317                         *tun_collect_md = geneve->cfg.collect_md;
1318                         *tun_on_same_port = true;
1319                 }
1320                 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1321                     info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1322                     !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1323                         t = geneve;
1324         }
1325         return t;
1326 }
1327
1328 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1329 {
1330         return !(info->key.tun_id || info->key.tos ||
1331                  !ip_tunnel_flags_empty(info->key.tun_flags) ||
1332                  info->key.ttl || info->key.label || info->key.tp_src ||
1333                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1334 }
1335
1336 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1337                                   struct ip_tunnel_info *b)
1338 {
1339         if (ip_tunnel_info_af(a) == AF_INET)
1340                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1341         else
1342                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1343 }
1344
1345 static int geneve_configure(struct net *net, struct net_device *dev,
1346                             struct netlink_ext_ack *extack,
1347                             const struct geneve_config *cfg)
1348 {
1349         struct geneve_net *gn = net_generic(net, geneve_net_id);
1350         struct geneve_dev *t, *geneve = netdev_priv(dev);
1351         const struct ip_tunnel_info *info = &cfg->info;
1352         bool tun_collect_md, tun_on_same_port;
1353         int err, encap_len;
1354
1355         if (cfg->collect_md && !is_tnl_info_zero(info)) {
1356                 NL_SET_ERR_MSG(extack,
1357                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1358                 return -EINVAL;
1359         }
1360
1361         geneve->net = net;
1362         geneve->dev = dev;
1363
1364         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1365         if (t)
1366                 return -EBUSY;
1367
1368         /* make enough headroom for basic scenario */
1369         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1370         if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1371                 encap_len += sizeof(struct iphdr);
1372                 dev->max_mtu -= sizeof(struct iphdr);
1373         } else {
1374                 encap_len += sizeof(struct ipv6hdr);
1375                 dev->max_mtu -= sizeof(struct ipv6hdr);
1376         }
1377         dev->needed_headroom = encap_len + ETH_HLEN;
1378
1379         if (cfg->collect_md) {
1380                 if (tun_on_same_port) {
1381                         NL_SET_ERR_MSG(extack,
1382                                        "There can be only one externally controlled device on a destination port");
1383                         return -EPERM;
1384                 }
1385         } else {
1386                 if (tun_collect_md) {
1387                         NL_SET_ERR_MSG(extack,
1388                                        "There already exists an externally controlled device on this destination port");
1389                         return -EPERM;
1390                 }
1391         }
1392
1393         dst_cache_reset(&geneve->cfg.info.dst_cache);
1394         memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1395
1396         if (geneve->cfg.inner_proto_inherit) {
1397                 dev->header_ops = NULL;
1398                 dev->type = ARPHRD_NONE;
1399                 dev->hard_header_len = 0;
1400                 dev->addr_len = 0;
1401                 dev->flags = IFF_POINTOPOINT | IFF_NOARP;
1402         }
1403
1404         err = register_netdevice(dev);
1405         if (err)
1406                 return err;
1407
1408         list_add(&geneve->next, &gn->geneve_list);
1409         return 0;
1410 }
1411
1412 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1413 {
1414         memset(info, 0, sizeof(*info));
1415         info->key.tp_dst = htons(dst_port);
1416 }
1417
1418 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1419                           struct netlink_ext_ack *extack,
1420                           struct geneve_config *cfg, bool changelink)
1421 {
1422         struct ip_tunnel_info *info = &cfg->info;
1423         int attrtype;
1424
1425         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1426                 NL_SET_ERR_MSG(extack,
1427                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1428                 return -EINVAL;
1429         }
1430
1431         if (data[IFLA_GENEVE_REMOTE]) {
1432                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1433                         attrtype = IFLA_GENEVE_REMOTE;
1434                         goto change_notsup;
1435                 }
1436
1437                 info->key.u.ipv4.dst =
1438                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1439
1440                 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1441                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1442                                             "Remote IPv4 address cannot be Multicast");
1443                         return -EINVAL;
1444                 }
1445         }
1446
1447         if (data[IFLA_GENEVE_REMOTE6]) {
1448 #if IS_ENABLED(CONFIG_IPV6)
1449                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1450                         attrtype = IFLA_GENEVE_REMOTE6;
1451                         goto change_notsup;
1452                 }
1453
1454                 info->mode = IP_TUNNEL_INFO_IPV6;
1455                 info->key.u.ipv6.dst =
1456                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1457
1458                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1459                     IPV6_ADDR_LINKLOCAL) {
1460                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1461                                             "Remote IPv6 address cannot be link-local");
1462                         return -EINVAL;
1463                 }
1464                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1465                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1466                                             "Remote IPv6 address cannot be Multicast");
1467                         return -EINVAL;
1468                 }
1469                 __set_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags);
1470                 cfg->use_udp6_rx_checksums = true;
1471 #else
1472                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1473                                     "IPv6 support not enabled in the kernel");
1474                 return -EPFNOSUPPORT;
1475 #endif
1476         }
1477
1478         if (data[IFLA_GENEVE_ID]) {
1479                 __u32 vni;
1480                 __u8 tvni[3];
1481                 __be64 tunid;
1482
1483                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1484                 tvni[0] = (vni & 0x00ff0000) >> 16;
1485                 tvni[1] = (vni & 0x0000ff00) >> 8;
1486                 tvni[2] =  vni & 0x000000ff;
1487
1488                 tunid = vni_to_tunnel_id(tvni);
1489                 if (changelink && (tunid != info->key.tun_id)) {
1490                         attrtype = IFLA_GENEVE_ID;
1491                         goto change_notsup;
1492                 }
1493                 info->key.tun_id = tunid;
1494         }
1495
1496         if (data[IFLA_GENEVE_TTL_INHERIT]) {
1497                 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1498                         cfg->ttl_inherit = true;
1499                 else
1500                         cfg->ttl_inherit = false;
1501         } else if (data[IFLA_GENEVE_TTL]) {
1502                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1503                 cfg->ttl_inherit = false;
1504         }
1505
1506         if (data[IFLA_GENEVE_TOS])
1507                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1508
1509         if (data[IFLA_GENEVE_DF])
1510                 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1511
1512         if (data[IFLA_GENEVE_LABEL]) {
1513                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1514                                   IPV6_FLOWLABEL_MASK;
1515                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1516                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1517                                             "Label attribute only applies for IPv6 Geneve devices");
1518                         return -EINVAL;
1519                 }
1520         }
1521
1522         if (data[IFLA_GENEVE_PORT]) {
1523                 if (changelink) {
1524                         attrtype = IFLA_GENEVE_PORT;
1525                         goto change_notsup;
1526                 }
1527                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1528         }
1529
1530         if (data[IFLA_GENEVE_PORT_RANGE]) {
1531                 const struct ifla_geneve_port_range *p;
1532
1533                 if (changelink) {
1534                         attrtype = IFLA_GENEVE_PORT_RANGE;
1535                         goto change_notsup;
1536                 }
1537                 p = nla_data(data[IFLA_GENEVE_PORT_RANGE]);
1538                 cfg->port_min = ntohs(p->low);
1539                 cfg->port_max = ntohs(p->high);
1540         }
1541
1542         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1543                 if (changelink) {
1544                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1545                         goto change_notsup;
1546                 }
1547                 cfg->collect_md = true;
1548         }
1549
1550         if (data[IFLA_GENEVE_UDP_CSUM]) {
1551                 if (changelink) {
1552                         attrtype = IFLA_GENEVE_UDP_CSUM;
1553                         goto change_notsup;
1554                 }
1555                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1556                         __set_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags);
1557         }
1558
1559         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1560 #if IS_ENABLED(CONFIG_IPV6)
1561                 if (changelink) {
1562                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1563                         goto change_notsup;
1564                 }
1565                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1566                         __clear_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags);
1567 #else
1568                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1569                                     "IPv6 support not enabled in the kernel");
1570                 return -EPFNOSUPPORT;
1571 #endif
1572         }
1573
1574         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1575 #if IS_ENABLED(CONFIG_IPV6)
1576                 if (changelink) {
1577                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1578                         goto change_notsup;
1579                 }
1580                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1581                         cfg->use_udp6_rx_checksums = false;
1582 #else
1583                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1584                                     "IPv6 support not enabled in the kernel");
1585                 return -EPFNOSUPPORT;
1586 #endif
1587         }
1588
1589         if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) {
1590                 if (changelink) {
1591                         attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT;
1592                         goto change_notsup;
1593                 }
1594                 cfg->inner_proto_inherit = true;
1595         }
1596
1597         return 0;
1598 change_notsup:
1599         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1600                             "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported");
1601         return -EOPNOTSUPP;
1602 }
1603
1604 static void geneve_link_config(struct net_device *dev,
1605                                struct ip_tunnel_info *info, struct nlattr *tb[])
1606 {
1607         struct geneve_dev *geneve = netdev_priv(dev);
1608         int ldev_mtu = 0;
1609
1610         if (tb[IFLA_MTU]) {
1611                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1612                 return;
1613         }
1614
1615         switch (ip_tunnel_info_af(info)) {
1616         case AF_INET: {
1617                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1618                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1619
1620                 if (!IS_ERR(rt) && rt->dst.dev) {
1621                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1622                         ip_rt_put(rt);
1623                 }
1624                 break;
1625         }
1626 #if IS_ENABLED(CONFIG_IPV6)
1627         case AF_INET6: {
1628                 struct rt6_info *rt;
1629
1630                 if (!__in6_dev_get(dev))
1631                         break;
1632
1633                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1634                                 NULL, 0);
1635
1636                 if (rt && rt->dst.dev)
1637                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1638                 ip6_rt_put(rt);
1639                 break;
1640         }
1641 #endif
1642         }
1643
1644         if (ldev_mtu <= 0)
1645                 return;
1646
1647         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1648 }
1649
1650 static int geneve_newlink(struct net_device *dev,
1651                           struct rtnl_newlink_params *params,
1652                           struct netlink_ext_ack *extack)
1653 {
1654         struct net *link_net = rtnl_newlink_link_net(params);
1655         struct nlattr **data = params->data;
1656         struct nlattr **tb = params->tb;
1657         struct geneve_config cfg = {
1658                 .df = GENEVE_DF_UNSET,
1659                 .use_udp6_rx_checksums = false,
1660                 .ttl_inherit = false,
1661                 .collect_md = false,
1662                 .port_min = 1,
1663                 .port_max = USHRT_MAX,
1664         };
1665         int err;
1666
1667         init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1668         err = geneve_nl2info(tb, data, extack, &cfg, false);
1669         if (err)
1670                 return err;
1671
1672         err = geneve_configure(link_net, dev, extack, &cfg);
1673         if (err)
1674                 return err;
1675
1676         geneve_link_config(dev, &cfg.info, tb);
1677
1678         return 0;
1679 }
1680
1681 /* Quiesces the geneve device data path for both TX and RX.
1682  *
1683  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1684  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1685  * to complete for the existing set of in-flight packets to be transmitted,
1686  * then we would have quiesced the transmit data path. All the future packets
1687  * will get dropped until we unquiesce the data path.
1688  *
1689  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1690  * if we set that to NULL under RCU and wait for synchronize_net() to
1691  * complete, then we would have quiesced the receive data path.
1692  */
1693 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1694                            struct geneve_sock **gs6)
1695 {
1696         *gs4 = rtnl_dereference(geneve->sock4);
1697         rcu_assign_pointer(geneve->sock4, NULL);
1698         if (*gs4)
1699                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1700 #if IS_ENABLED(CONFIG_IPV6)
1701         *gs6 = rtnl_dereference(geneve->sock6);
1702         rcu_assign_pointer(geneve->sock6, NULL);
1703         if (*gs6)
1704                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1705 #else
1706         *gs6 = NULL;
1707 #endif
1708         synchronize_net();
1709 }
1710
1711 /* Resumes the geneve device data path for both TX and RX. */
1712 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1713                              struct geneve_sock __maybe_unused *gs6)
1714 {
1715         rcu_assign_pointer(geneve->sock4, gs4);
1716         if (gs4)
1717                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1718 #if IS_ENABLED(CONFIG_IPV6)
1719         rcu_assign_pointer(geneve->sock6, gs6);
1720         if (gs6)
1721                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1722 #endif
1723         synchronize_net();
1724 }
1725
1726 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1727                              struct nlattr *data[],
1728                              struct netlink_ext_ack *extack)
1729 {
1730         struct geneve_dev *geneve = netdev_priv(dev);
1731         struct geneve_sock *gs4, *gs6;
1732         struct geneve_config cfg;
1733         int err;
1734
1735         /* If the geneve device is configured for metadata (or externally
1736          * controlled, for example, OVS), then nothing can be changed.
1737          */
1738         if (geneve->cfg.collect_md)
1739                 return -EOPNOTSUPP;
1740
1741         /* Start with the existing info. */
1742         memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1743         err = geneve_nl2info(tb, data, extack, &cfg, true);
1744         if (err)
1745                 return err;
1746
1747         if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1748                 dst_cache_reset(&cfg.info.dst_cache);
1749                 geneve_link_config(dev, &cfg.info, tb);
1750         }
1751
1752         geneve_quiesce(geneve, &gs4, &gs6);
1753         memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1754         geneve_unquiesce(geneve, gs4, gs6);
1755
1756         return 0;
1757 }
1758
1759 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1760 {
1761         struct geneve_dev *geneve = netdev_priv(dev);
1762
1763         list_del(&geneve->next);
1764         unregister_netdevice_queue(dev, head);
1765 }
1766
1767 static size_t geneve_get_size(const struct net_device *dev)
1768 {
1769         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1770                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1771                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1772                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1773                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_DF */
1774                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1775                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1776                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1777                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1778                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1779                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1780                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1781                 nla_total_size(0) +      /* IFLA_GENEVE_INNER_PROTO_INHERIT */
1782                 nla_total_size(sizeof(struct ifla_geneve_port_range)) + /* IFLA_GENEVE_PORT_RANGE */
1783                 0;
1784 }
1785
1786 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1787 {
1788         struct geneve_dev *geneve = netdev_priv(dev);
1789         struct ip_tunnel_info *info = &geneve->cfg.info;
1790         bool ttl_inherit = geneve->cfg.ttl_inherit;
1791         bool metadata = geneve->cfg.collect_md;
1792         struct ifla_geneve_port_range ports = {
1793                 .low    = htons(geneve->cfg.port_min),
1794                 .high   = htons(geneve->cfg.port_max),
1795         };
1796         __u8 tmp_vni[3];
1797         __u32 vni;
1798
1799         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1800         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1801         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1802                 goto nla_put_failure;
1803
1804         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1805                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1806                                     info->key.u.ipv4.dst))
1807                         goto nla_put_failure;
1808                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1809                                test_bit(IP_TUNNEL_CSUM_BIT,
1810                                         info->key.tun_flags)))
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                                !test_bit(IP_TUNNEL_CSUM_BIT,
1820                                          info->key.tun_flags)))
1821                         goto nla_put_failure;
1822 #endif
1823         }
1824
1825         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1826             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1827             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1828                 goto nla_put_failure;
1829
1830         if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1831                 goto nla_put_failure;
1832
1833         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1834                 goto nla_put_failure;
1835
1836         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1837                 goto nla_put_failure;
1838
1839 #if IS_ENABLED(CONFIG_IPV6)
1840         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1841                        !geneve->cfg.use_udp6_rx_checksums))
1842                 goto nla_put_failure;
1843 #endif
1844
1845         if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1846                 goto nla_put_failure;
1847
1848         if (geneve->cfg.inner_proto_inherit &&
1849             nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT))
1850                 goto nla_put_failure;
1851
1852         if (nla_put(skb, IFLA_GENEVE_PORT_RANGE, sizeof(ports), &ports))
1853                 goto nla_put_failure;
1854
1855         return 0;
1856
1857 nla_put_failure:
1858         return -EMSGSIZE;
1859 }
1860
1861 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1862         .kind           = "geneve",
1863         .maxtype        = IFLA_GENEVE_MAX,
1864         .policy         = geneve_policy,
1865         .priv_size      = sizeof(struct geneve_dev),
1866         .setup          = geneve_setup,
1867         .validate       = geneve_validate,
1868         .newlink        = geneve_newlink,
1869         .changelink     = geneve_changelink,
1870         .dellink        = geneve_dellink,
1871         .get_size       = geneve_get_size,
1872         .fill_info      = geneve_fill_info,
1873 };
1874
1875 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1876                                         u8 name_assign_type, u16 dst_port)
1877 {
1878         struct nlattr *tb[IFLA_MAX + 1];
1879         struct net_device *dev;
1880         LIST_HEAD(list_kill);
1881         int err;
1882         struct geneve_config cfg = {
1883                 .df = GENEVE_DF_UNSET,
1884                 .use_udp6_rx_checksums = true,
1885                 .ttl_inherit = false,
1886                 .collect_md = true,
1887                 .port_min = 1,
1888                 .port_max = USHRT_MAX,
1889         };
1890
1891         memset(tb, 0, sizeof(tb));
1892         dev = rtnl_create_link(net, name, name_assign_type,
1893                                &geneve_link_ops, tb, NULL);
1894         if (IS_ERR(dev))
1895                 return dev;
1896
1897         init_tnl_info(&cfg.info, dst_port);
1898         err = geneve_configure(net, dev, NULL, &cfg);
1899         if (err) {
1900                 free_netdev(dev);
1901                 return ERR_PTR(err);
1902         }
1903
1904         /* openvswitch users expect packet sizes to be unrestricted,
1905          * so set the largest MTU we can.
1906          */
1907         err = geneve_change_mtu(dev, IP_MAX_MTU);
1908         if (err)
1909                 goto err;
1910
1911         err = rtnl_configure_link(dev, NULL, 0, NULL);
1912         if (err < 0)
1913                 goto err;
1914
1915         return dev;
1916 err:
1917         geneve_dellink(dev, &list_kill);
1918         unregister_netdevice_many(&list_kill);
1919         return ERR_PTR(err);
1920 }
1921 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1922
1923 static int geneve_netdevice_event(struct notifier_block *unused,
1924                                   unsigned long event, void *ptr)
1925 {
1926         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1927
1928         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1929                 geneve_offload_rx_ports(dev, true);
1930         else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1931                 geneve_offload_rx_ports(dev, false);
1932
1933         return NOTIFY_DONE;
1934 }
1935
1936 static struct notifier_block geneve_notifier_block __read_mostly = {
1937         .notifier_call = geneve_netdevice_event,
1938 };
1939
1940 static __net_init int geneve_init_net(struct net *net)
1941 {
1942         struct geneve_net *gn = net_generic(net, geneve_net_id);
1943
1944         INIT_LIST_HEAD(&gn->geneve_list);
1945         INIT_LIST_HEAD(&gn->sock_list);
1946         return 0;
1947 }
1948
1949 static void __net_exit geneve_exit_rtnl_net(struct net *net,
1950                                             struct list_head *dev_to_kill)
1951 {
1952         struct geneve_net *gn = net_generic(net, geneve_net_id);
1953         struct geneve_dev *geneve, *next;
1954
1955         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next)
1956                 geneve_dellink(geneve->dev, dev_to_kill);
1957 }
1958
1959 static void __net_exit geneve_exit_net(struct net *net)
1960 {
1961         const struct geneve_net *gn = net_generic(net, geneve_net_id);
1962
1963         WARN_ON_ONCE(!list_empty(&gn->sock_list));
1964 }
1965
1966 static struct pernet_operations geneve_net_ops = {
1967         .init = geneve_init_net,
1968         .exit_rtnl = geneve_exit_rtnl_net,
1969         .exit = geneve_exit_net,
1970         .id   = &geneve_net_id,
1971         .size = sizeof(struct geneve_net),
1972 };
1973
1974 static int __init geneve_init_module(void)
1975 {
1976         int rc;
1977
1978         rc = register_pernet_subsys(&geneve_net_ops);
1979         if (rc)
1980                 goto out1;
1981
1982         rc = register_netdevice_notifier(&geneve_notifier_block);
1983         if (rc)
1984                 goto out2;
1985
1986         rc = rtnl_link_register(&geneve_link_ops);
1987         if (rc)
1988                 goto out3;
1989
1990         return 0;
1991 out3:
1992         unregister_netdevice_notifier(&geneve_notifier_block);
1993 out2:
1994         unregister_pernet_subsys(&geneve_net_ops);
1995 out1:
1996         return rc;
1997 }
1998 late_initcall(geneve_init_module);
1999
2000 static void __exit geneve_cleanup_module(void)
2001 {
2002         rtnl_link_unregister(&geneve_link_ops);
2003         unregister_netdevice_notifier(&geneve_notifier_block);
2004         unregister_pernet_subsys(&geneve_net_ops);
2005 }
2006 module_exit(geneve_cleanup_module);
2007
2008 MODULE_LICENSE("GPL");
2009 MODULE_VERSION(GENEVE_NETDEV_VER);
2010 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
2011 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
2012 MODULE_ALIAS_RTNL_LINK("geneve");