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 nh, 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                         DEV_STATS_INC(geneve->dev, 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                         DEV_STATS_INC(geneve->dev, rx_frame_errors);
 250                         DEV_STATS_INC(geneve->dev, 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                         DEV_STATS_INC(geneve->dev, 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         /* Save offset of outer header relative to skb->head,
 276          * because we are going to reset the network header to the inner header
 277          * and might change skb->head.
 278          */
 279         nh = skb_network_header(skb) - skb->head;
 280
 281         skb_reset_network_header(skb);
 282
 283         if (!pskb_inet_may_pull(skb)) {
 284                 DEV_STATS_INC(geneve->dev, rx_length_errors);
 285                 DEV_STATS_INC(geneve->dev, rx_errors);
 286                 goto drop;
 287         }
 288
 289         /* Get the outer header. */
 290         oiph = skb->head + nh;
 291
 292         if (geneve_get_sk_family(gs) == AF_INET)
 293                 err = IP_ECN_decapsulate(oiph, skb);
 294 #if IS_ENABLED(CONFIG_IPV6)
 295         else
 296                 err = IP6_ECN_decapsulate(oiph, skb);
 297 #endif
 298
 299         if (unlikely(err)) {
 300                 if (log_ecn_error) {
 301                         if (geneve_get_sk_family(gs) == AF_INET)
 302                                 net_info_ratelimited("non-ECT from %pI4 "
 303                                                      "with TOS=%#x\n",
 304                                                      &((struct iphdr *)oiph)->saddr,
 305                                                      ((struct iphdr *)oiph)->tos);
 306 #if IS_ENABLED(CONFIG_IPV6)
 307                         else
 308                                 net_info_ratelimited("non-ECT from %pI6\n",
 309                                                      &((struct ipv6hdr *)oiph)->saddr);
 310 #endif
 311                 }
 312                 if (err > 1) {
 313                         DEV_STATS_INC(geneve->dev, rx_frame_errors);
 314                         DEV_STATS_INC(geneve->dev, rx_errors);
 315                         goto drop;
 316                 }
 317         }
 318
 319         len = skb->len;
 320         err = gro_cells_receive(&geneve->gro_cells, skb);
 321         if (likely(err == NET_RX_SUCCESS))
 322                 dev_sw_netstats_rx_add(geneve->dev, len);
 323
 324         return;
 325 drop:
 326         /* Consume bad packet */
 327         kfree_skb(skb);
 328 }
 329
 330 /* Setup stats when device is created */
 331 static int geneve_init(struct net_device *dev)
 332 {
 333         struct geneve_dev *geneve = netdev_priv(dev);
 334         int err;
 335
 336         err = gro_cells_init(&geneve->gro_cells, dev);
 337         if (err)
 338                 return err;
 339
 340         err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
 341         if (err) {
 342                 gro_cells_destroy(&geneve->gro_cells);
 343                 return err;
 344         }
 345         netdev_lockdep_set_classes(dev);
 346         return 0;
 347 }
 348
 349 static void geneve_uninit(struct net_device *dev)
 350 {
 351         struct geneve_dev *geneve = netdev_priv(dev);
 352
 353         dst_cache_destroy(&geneve->cfg.info.dst_cache);
 354         gro_cells_destroy(&geneve->gro_cells);
 355 }
 356
 357 /* Callback from net/ipv4/udp.c to receive packets */
 358 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 359 {
 360         struct genevehdr *geneveh;
 361         struct geneve_dev *geneve;
 362         struct geneve_sock *gs;
 363         __be16 inner_proto;
 364         int opts_len;
 365
 366         /* Need UDP and Geneve header to be present */
 367         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 368                 goto drop;
 369
 370         /* Return packets with reserved bits set */
 371         geneveh = geneve_hdr(skb);
 372         if (unlikely(geneveh->ver != GENEVE_VER))
 373                 goto drop;
 374
 375         gs = rcu_dereference_sk_user_data(sk);
 376         if (!gs)
 377                 goto drop;
 378
 379         geneve = geneve_lookup_skb(gs, skb);
 380         if (!geneve)
 381                 goto drop;
 382
 383         inner_proto = geneveh->proto_type;
 384
 385         if (unlikely((!geneve->cfg.inner_proto_inherit &&
 386                       inner_proto != htons(ETH_P_TEB)))) {
 387                 DEV_STATS_INC(geneve->dev, rx_dropped);
 388                 goto drop;
 389         }
 390
 391         opts_len = geneveh->opt_len * 4;
 392         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto,
 393                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 394                 DEV_STATS_INC(geneve->dev, rx_dropped);
 395                 goto drop;
 396         }
 397
 398         geneve_rx(geneve, gs, skb);
 399         return 0;
 400
 401 drop:
 402         /* Consume bad packet */
 403         kfree_skb(skb);
 404         return 0;
 405 }
 406
 407 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
 408 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
 409 {
 410         struct genevehdr *geneveh;
 411         struct geneve_sock *gs;
 412         u8 zero_vni[3] = { 0 };
 413         u8 *vni = zero_vni;
 414
 415         if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
 416                 return -EINVAL;
 417
 418         geneveh = geneve_hdr(skb);
 419         if (geneveh->ver != GENEVE_VER)
 420                 return -EINVAL;
 421
 422         if (geneveh->proto_type != htons(ETH_P_TEB))
 423                 return -EINVAL;
 424
 425         gs = rcu_dereference_sk_user_data(sk);
 426         if (!gs)
 427                 return -ENOENT;
 428
 429         if (geneve_get_sk_family(gs) == AF_INET) {
 430                 struct iphdr *iph = ip_hdr(skb);
 431                 __be32 addr4 = 0;
 432
 433                 if (!gs->collect_md) {
 434                         vni = geneve_hdr(skb)->vni;
 435                         addr4 = iph->daddr;
 436                 }
 437
 438                 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
 439         }
 440
 441 #if IS_ENABLED(CONFIG_IPV6)
 442         if (geneve_get_sk_family(gs) == AF_INET6) {
 443                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 444                 struct in6_addr addr6;
 445
 446                 memset(&addr6, 0, sizeof(struct in6_addr));
 447
 448                 if (!gs->collect_md) {
 449                         vni = geneve_hdr(skb)->vni;
 450                         addr6 = ip6h->daddr;
 451                 }
 452
 453                 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
 454         }
 455 #endif
 456
 457         return -EPFNOSUPPORT;
 458 }
 459
 460 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 461                                          __be16 port, bool ipv6_rx_csum)
 462 {
 463         struct socket *sock;
 464         struct udp_port_cfg udp_conf;
 465         int err;
 466
 467         memset(&udp_conf, 0, sizeof(udp_conf));
 468
 469         if (ipv6) {
 470                 udp_conf.family = AF_INET6;
 471                 udp_conf.ipv6_v6only = 1;
 472                 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
 473         } else {
 474                 udp_conf.family = AF_INET;
 475                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 476         }
 477
 478         udp_conf.local_udp_port = port;
 479
 480         /* Open UDP socket */
 481         err = udp_sock_create(net, &udp_conf, &sock);
 482         if (err < 0)
 483                 return ERR_PTR(err);
 484
 485         udp_allow_gso(sock->sk);
 486         return sock;
 487 }
 488
 489 static int geneve_hlen(struct genevehdr *gh)
 490 {
 491         return sizeof(*gh) + gh->opt_len * 4;
 492 }
 493
 494 static struct sk_buff *geneve_gro_receive(struct sock *sk,
 495                                           struct list_head *head,
 496                                           struct sk_buff *skb)
 497 {
 498         struct sk_buff *pp = NULL;
 499         struct sk_buff *p;
 500         struct genevehdr *gh, *gh2;
 501         unsigned int hlen, gh_len, off_gnv;
 502         const struct packet_offload *ptype;
 503         __be16 type;
 504         int flush = 1;
 505
 506         off_gnv = skb_gro_offset(skb);
 507         hlen = off_gnv + sizeof(*gh);
 508         gh = skb_gro_header(skb, hlen, off_gnv);
 509         if (unlikely(!gh))
 510                 goto out;
 511
 512         if (gh->ver != GENEVE_VER || gh->oam)
 513                 goto out;
 514         gh_len = geneve_hlen(gh);
 515
 516         hlen = off_gnv + gh_len;
 517         if (!skb_gro_may_pull(skb, hlen)) {
 518                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 519                 if (unlikely(!gh))
 520                         goto out;
 521         }
 522
 523         list_for_each_entry(p, head, list) {
 524                 if (!NAPI_GRO_CB(p)->same_flow)
 525                         continue;
 526
 527                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 528                 if (gh->opt_len != gh2->opt_len ||
 529                     memcmp(gh, gh2, gh_len)) {
 530                         NAPI_GRO_CB(p)->same_flow = 0;
 531                         continue;
 532                 }
 533         }
 534
 535         skb_gro_pull(skb, gh_len);
 536         skb_gro_postpull_rcsum(skb, gh, gh_len);
 537         type = gh->proto_type;
 538         if (likely(type == htons(ETH_P_TEB)))
 539                 return call_gro_receive(eth_gro_receive, head, skb);
 540
 541         ptype = gro_find_receive_by_type(type);
 542         if (!ptype)
 543                 goto out;
 544
 545         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 546         flush = 0;
 547
 548 out:
 549         skb_gro_flush_final(skb, pp, flush);
 550
 551         return pp;
 552 }
 553
 554 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 555                                int nhoff)
 556 {
 557         struct genevehdr *gh;
 558         struct packet_offload *ptype;
 559         __be16 type;
 560         int gh_len;
 561         int err = -ENOSYS;
 562
 563         gh = (struct genevehdr *)(skb->data + nhoff);
 564         gh_len = geneve_hlen(gh);
 565         type = gh->proto_type;
 566
 567         /* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */
 568         if (likely(type == htons(ETH_P_TEB)))
 569                 return eth_gro_complete(skb, nhoff + gh_len);
 570
 571         ptype = gro_find_complete_by_type(type);
 572         if (ptype)
 573                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 574
 575         skb_set_inner_mac_header(skb, nhoff + gh_len);
 576
 577         return err;
 578 }
 579
 580 /* Create new listen socket if needed */
 581 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 582                                                 bool ipv6, bool ipv6_rx_csum)
 583 {
 584         struct geneve_net *gn = net_generic(net, geneve_net_id);
 585         struct geneve_sock *gs;
 586         struct socket *sock;
 587         struct udp_tunnel_sock_cfg tunnel_cfg;
 588         int h;
 589
 590         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 591         if (!gs)
 592                 return ERR_PTR(-ENOMEM);
 593
 594         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 595         if (IS_ERR(sock)) {
 596                 kfree(gs);
 597                 return ERR_CAST(sock);
 598         }
 599
 600         gs->sock = sock;
 601         gs->refcnt = 1;
 602         for (h = 0; h < VNI_HASH_SIZE; ++h)
 603                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 604
 605         /* Initialize the geneve udp offloads structure */
 606         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 607
 608         /* Mark socket as an encapsulation socket */
 609         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 610         tunnel_cfg.sk_user_data = gs;
 611         tunnel_cfg.encap_type = 1;
 612         tunnel_cfg.gro_receive = geneve_gro_receive;
 613         tunnel_cfg.gro_complete = geneve_gro_complete;
 614         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 615         tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
 616         tunnel_cfg.encap_destroy = NULL;
 617         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 618         list_add(&gs->list, &gn->sock_list);
 619         return gs;
 620 }
 621
 622 static void __geneve_sock_release(struct geneve_sock *gs)
 623 {
 624         if (!gs || --gs->refcnt)
 625                 return;
 626
 627         list_del(&gs->list);
 628         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 629         udp_tunnel_sock_release(gs->sock);
 630         kfree_rcu(gs, rcu);
 631 }
 632
 633 static void geneve_sock_release(struct geneve_dev *geneve)
 634 {
 635         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 636 #if IS_ENABLED(CONFIG_IPV6)
 637         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 638
 639         rcu_assign_pointer(geneve->sock6, NULL);
 640 #endif
 641
 642         rcu_assign_pointer(geneve->sock4, NULL);
 643         synchronize_net();
 644
 645         __geneve_sock_release(gs4);
 646 #if IS_ENABLED(CONFIG_IPV6)
 647         __geneve_sock_release(gs6);
 648 #endif
 649 }
 650
 651 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 652                                             sa_family_t family,
 653                                             __be16 dst_port)
 654 {
 655         struct geneve_sock *gs;
 656
 657         list_for_each_entry(gs, &gn->sock_list, list) {
 658                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 659                     geneve_get_sk_family(gs) == family) {
 660                         return gs;
 661                 }
 662         }
 663         return NULL;
 664 }
 665
 666 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 667 {
 668         struct net *net = geneve->net;
 669         struct geneve_net *gn = net_generic(net, geneve_net_id);
 670         struct geneve_dev_node *node;
 671         struct geneve_sock *gs;
 672         __u8 vni[3];
 673         __u32 hash;
 674
 675         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
 676         if (gs) {
 677                 gs->refcnt++;
 678                 goto out;
 679         }
 680
 681         gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
 682                                   geneve->cfg.use_udp6_rx_checksums);
 683         if (IS_ERR(gs))
 684                 return PTR_ERR(gs);
 685
 686 out:
 687         gs->collect_md = geneve->cfg.collect_md;
 688 #if IS_ENABLED(CONFIG_IPV6)
 689         if (ipv6) {
 690                 rcu_assign_pointer(geneve->sock6, gs);
 691                 node = &geneve->hlist6;
 692         } else
 693 #endif
 694         {
 695                 rcu_assign_pointer(geneve->sock4, gs);
 696                 node = &geneve->hlist4;
 697         }
 698         node->geneve = geneve;
 699
 700         tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
 701         hash = geneve_net_vni_hash(vni);
 702         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 703         return 0;
 704 }
 705
 706 static int geneve_open(struct net_device *dev)
 707 {
 708         struct geneve_dev *geneve = netdev_priv(dev);
 709         bool metadata = geneve->cfg.collect_md;
 710         bool ipv4, ipv6;
 711         int ret = 0;
 712
 713         ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 714         ipv4 = !ipv6 || metadata;
 715 #if IS_ENABLED(CONFIG_IPV6)
 716         if (ipv6) {
 717                 ret = geneve_sock_add(geneve, true);
 718                 if (ret < 0 && ret != -EAFNOSUPPORT)
 719                         ipv4 = false;
 720         }
 721 #endif
 722         if (ipv4)
 723                 ret = geneve_sock_add(geneve, false);
 724         if (ret < 0)
 725                 geneve_sock_release(geneve);
 726
 727         return ret;
 728 }
 729
 730 static int geneve_stop(struct net_device *dev)
 731 {
 732         struct geneve_dev *geneve = netdev_priv(dev);
 733
 734         hlist_del_init_rcu(&geneve->hlist4.hlist);
 735 #if IS_ENABLED(CONFIG_IPV6)
 736         hlist_del_init_rcu(&geneve->hlist6.hlist);
 737 #endif
 738         geneve_sock_release(geneve);
 739         return 0;
 740 }
 741
 742 static void geneve_build_header(struct genevehdr *geneveh,
 743                                 const struct ip_tunnel_info *info,
 744                                 __be16 inner_proto)
 745 {
 746         geneveh->ver = GENEVE_VER;
 747         geneveh->opt_len = info->options_len / 4;
 748         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 749         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 750         geneveh->rsvd1 = 0;
 751         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 752         geneveh->proto_type = inner_proto;
 753         geneveh->rsvd2 = 0;
 754
 755         if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
 756                 ip_tunnel_info_opts_get(geneveh->options, info);
 757 }
 758
 759 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 760                             const struct ip_tunnel_info *info,
 761                             bool xnet, int ip_hdr_len,
 762                             bool inner_proto_inherit)
 763 {
 764         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 765         struct genevehdr *gnvh;
 766         __be16 inner_proto;
 767         int min_headroom;
 768         int err;
 769
 770         skb_reset_mac_header(skb);
 771         skb_scrub_packet(skb, xnet);
 772
 773         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 774                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 775         err = skb_cow_head(skb, min_headroom);
 776         if (unlikely(err))
 777                 goto free_dst;
 778
 779         err = udp_tunnel_handle_offloads(skb, udp_sum);
 780         if (err)
 781                 goto free_dst;
 782
 783         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 784         inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB);
 785         geneve_build_header(gnvh, info, inner_proto);
 786         skb_set_inner_protocol(skb, inner_proto);
 787         return 0;
 788
 789 free_dst:
 790         dst_release(dst);
 791         return err;
 792 }
 793
 794 static u8 geneve_get_dsfield(struct sk_buff *skb, struct net_device *dev,
 795                              const struct ip_tunnel_info *info,
 796                              bool *use_cache)
 797 {
 798         struct geneve_dev *geneve = netdev_priv(dev);
 799         u8 dsfield;
 800
 801         dsfield = info->key.tos;
 802         if (dsfield == 1 && !geneve->cfg.collect_md) {
 803                 dsfield = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 804                 *use_cache = false;
 805         }
 806
 807         return dsfield;
 808 }
 809
 810 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 811                            struct geneve_dev *geneve,
 812                            const struct ip_tunnel_info *info)
 813 {
 814         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 815         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 816         const struct ip_tunnel_key *key = &info->key;
 817         struct rtable *rt;
 818         bool use_cache;
 819         __u8 tos, ttl;
 820         __be16 df = 0;
 821         __be32 saddr;
 822         __be16 sport;
 823         int err;
 824
 825         if (!pskb_inet_may_pull(skb))
 826                 return -EINVAL;
 827
 828         if (!gs4)
 829                 return -EIO;
 830
 831         use_cache = ip_tunnel_dst_cache_usable(skb, info);
 832         tos = geneve_get_dsfield(skb, dev, info, &use_cache);
 833         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 834
 835         rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
 836                                    &info->key,
 837                                    sport, geneve->cfg.info.key.tp_dst, tos,
 838                                    use_cache ?
 839                                    (struct dst_cache *)&info->dst_cache : NULL);
 840         if (IS_ERR(rt))
 841                 return PTR_ERR(rt);
 842
 843         err = skb_tunnel_check_pmtu(skb, &rt->dst,
 844                                     GENEVE_IPV4_HLEN + info->options_len,
 845                                     netif_is_any_bridge_port(dev));
 846         if (err < 0) {
 847                 dst_release(&rt->dst);
 848                 return err;
 849         } else if (err) {
 850                 struct ip_tunnel_info *info;
 851
 852                 info = skb_tunnel_info(skb);
 853                 if (info) {
 854                         struct ip_tunnel_info *unclone;
 855
 856                         unclone = skb_tunnel_info_unclone(skb);
 857                         if (unlikely(!unclone)) {
 858                                 dst_release(&rt->dst);
 859                                 return -ENOMEM;
 860                         }
 861
 862                         unclone->key.u.ipv4.dst = saddr;
 863                         unclone->key.u.ipv4.src = info->key.u.ipv4.dst;
 864                 }
 865
 866                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 867                         dst_release(&rt->dst);
 868                         return -EINVAL;
 869                 }
 870
 871                 skb->protocol = eth_type_trans(skb, geneve->dev);
 872                 __netif_rx(skb);
 873                 dst_release(&rt->dst);
 874                 return -EMSGSIZE;
 875         }
 876
 877         tos = ip_tunnel_ecn_encap(tos, ip_hdr(skb), skb);
 878         if (geneve->cfg.collect_md) {
 879                 ttl = key->ttl;
 880
 881                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 882         } else {
 883                 if (geneve->cfg.ttl_inherit)
 884                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 885                 else
 886                         ttl = key->ttl;
 887                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 888
 889                 if (geneve->cfg.df == GENEVE_DF_SET) {
 890                         df = htons(IP_DF);
 891                 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
 892                         struct ethhdr *eth = eth_hdr(skb);
 893
 894                         if (ntohs(eth->h_proto) == ETH_P_IPV6) {
 895                                 df = htons(IP_DF);
 896                         } else if (ntohs(eth->h_proto) == ETH_P_IP) {
 897                                 struct iphdr *iph = ip_hdr(skb);
 898
 899                                 if (iph->frag_off & htons(IP_DF))
 900                                         df = htons(IP_DF);
 901                         }
 902                 }
 903         }
 904
 905         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
 906                                geneve->cfg.inner_proto_inherit);
 907         if (unlikely(err))
 908                 return err;
 909
 910         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, saddr, info->key.u.ipv4.dst,
 911                             tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
 912                             !net_eq(geneve->net, dev_net(geneve->dev)),
 913                             !(info->key.tun_flags & TUNNEL_CSUM));
 914         return 0;
 915 }
 916
 917 #if IS_ENABLED(CONFIG_IPV6)
 918 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 919                             struct geneve_dev *geneve,
 920                             const struct ip_tunnel_info *info)
 921 {
 922         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 923         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
 924         const struct ip_tunnel_key *key = &info->key;
 925         struct dst_entry *dst = NULL;
 926         struct in6_addr saddr;
 927         bool use_cache;
 928         __u8 prio, ttl;
 929         __be16 sport;
 930         int err;
 931
 932         if (!pskb_inet_may_pull(skb))
 933                 return -EINVAL;
 934
 935         if (!gs6)
 936                 return -EIO;
 937
 938         use_cache = ip_tunnel_dst_cache_usable(skb, info);
 939         prio = geneve_get_dsfield(skb, dev, info, &use_cache);
 940         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 941
 942         dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
 943                                      &saddr, key, sport,
 944                                      geneve->cfg.info.key.tp_dst, prio,
 945                                      use_cache ?
 946                                      (struct dst_cache *)&info->dst_cache : NULL);
 947         if (IS_ERR(dst))
 948                 return PTR_ERR(dst);
 949
 950         err = skb_tunnel_check_pmtu(skb, dst,
 951                                     GENEVE_IPV6_HLEN + info->options_len,
 952                                     netif_is_any_bridge_port(dev));
 953         if (err < 0) {
 954                 dst_release(dst);
 955                 return err;
 956         } else if (err) {
 957                 struct ip_tunnel_info *info = skb_tunnel_info(skb);
 958
 959                 if (info) {
 960                         struct ip_tunnel_info *unclone;
 961
 962                         unclone = skb_tunnel_info_unclone(skb);
 963                         if (unlikely(!unclone)) {
 964                                 dst_release(dst);
 965                                 return -ENOMEM;
 966                         }
 967
 968                         unclone->key.u.ipv6.dst = saddr;
 969                         unclone->key.u.ipv6.src = info->key.u.ipv6.dst;
 970                 }
 971
 972                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 973                         dst_release(dst);
 974                         return -EINVAL;
 975                 }
 976
 977                 skb->protocol = eth_type_trans(skb, geneve->dev);
 978                 __netif_rx(skb);
 979                 dst_release(dst);
 980                 return -EMSGSIZE;
 981         }
 982
 983         prio = ip_tunnel_ecn_encap(prio, ip_hdr(skb), skb);
 984         if (geneve->cfg.collect_md) {
 985                 ttl = key->ttl;
 986         } else {
 987                 if (geneve->cfg.ttl_inherit)
 988                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 989                 else
 990                         ttl = key->ttl;
 991                 ttl = ttl ? : ip6_dst_hoplimit(dst);
 992         }
 993         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
 994                                geneve->cfg.inner_proto_inherit);
 995         if (unlikely(err))
 996                 return err;
 997
 998         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
 999                              &saddr, &key->u.ipv6.dst, prio, ttl,
1000                              info->key.label, sport, geneve->cfg.info.key.tp_dst,
1001                              !(info->key.tun_flags & TUNNEL_CSUM));
1002         return 0;
1003 }
1004 #endif
1005
1006 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1007 {
1008         struct geneve_dev *geneve = netdev_priv(dev);
1009         struct ip_tunnel_info *info = NULL;
1010         int err;
1011
1012         if (geneve->cfg.collect_md) {
1013                 info = skb_tunnel_info(skb);
1014                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1015                         netdev_dbg(dev, "no tunnel metadata\n");
1016                         dev_kfree_skb(skb);
1017                         DEV_STATS_INC(dev, tx_dropped);
1018                         return NETDEV_TX_OK;
1019                 }
1020         } else {
1021                 info = &geneve->cfg.info;
1022         }
1023
1024         rcu_read_lock();
1025 #if IS_ENABLED(CONFIG_IPV6)
1026         if (info->mode & IP_TUNNEL_INFO_IPV6)
1027                 err = geneve6_xmit_skb(skb, dev, geneve, info);
1028         else
1029 #endif
1030                 err = geneve_xmit_skb(skb, dev, geneve, info);
1031         rcu_read_unlock();
1032
1033         if (likely(!err))
1034                 return NETDEV_TX_OK;
1035
1036         if (err != -EMSGSIZE)
1037                 dev_kfree_skb(skb);
1038
1039         if (err == -ELOOP)
1040                 DEV_STATS_INC(dev, collisions);
1041         else if (err == -ENETUNREACH)
1042                 DEV_STATS_INC(dev, tx_carrier_errors);
1043
1044         DEV_STATS_INC(dev, tx_errors);
1045         return NETDEV_TX_OK;
1046 }
1047
1048 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1049 {
1050         if (new_mtu > dev->max_mtu)
1051                 new_mtu = dev->max_mtu;
1052         else if (new_mtu < dev->min_mtu)
1053                 new_mtu = dev->min_mtu;
1054
1055         dev->mtu = new_mtu;
1056         return 0;
1057 }
1058
1059 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1060 {
1061         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1062         struct geneve_dev *geneve = netdev_priv(dev);
1063         __be16 sport;
1064
1065         if (ip_tunnel_info_af(info) == AF_INET) {
1066                 struct rtable *rt;
1067                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1068                 bool use_cache;
1069                 __be32 saddr;
1070                 u8 tos;
1071
1072                 if (!gs4)
1073                         return -EIO;
1074
1075                 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1076                 tos = geneve_get_dsfield(skb, dev, info, &use_cache);
1077                 sport = udp_flow_src_port(geneve->net, skb,
1078                                           1, USHRT_MAX, true);
1079
1080                 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
1081                                            &info->key,
1082                                            sport, geneve->cfg.info.key.tp_dst,
1083                                            tos,
1084                                            use_cache ? &info->dst_cache : NULL);
1085                 if (IS_ERR(rt))
1086                         return PTR_ERR(rt);
1087
1088                 ip_rt_put(rt);
1089                 info->key.u.ipv4.src = saddr;
1090 #if IS_ENABLED(CONFIG_IPV6)
1091         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1092                 struct dst_entry *dst;
1093                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1094                 struct in6_addr saddr;
1095                 bool use_cache;
1096                 u8 prio;
1097
1098                 if (!gs6)
1099                         return -EIO;
1100
1101                 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1102                 prio = geneve_get_dsfield(skb, dev, info, &use_cache);
1103                 sport = udp_flow_src_port(geneve->net, skb,
1104                                           1, USHRT_MAX, true);
1105
1106                 dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
1107                                              &saddr, &info->key, sport,
1108                                              geneve->cfg.info.key.tp_dst, prio,
1109                                              use_cache ? &info->dst_cache : NULL);
1110                 if (IS_ERR(dst))
1111                         return PTR_ERR(dst);
1112
1113                 dst_release(dst);
1114                 info->key.u.ipv6.src = saddr;
1115 #endif
1116         } else {
1117                 return -EINVAL;
1118         }
1119
1120         info->key.tp_src = sport;
1121         info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1122         return 0;
1123 }
1124
1125 static const struct net_device_ops geneve_netdev_ops = {
1126         .ndo_init               = geneve_init,
1127         .ndo_uninit             = geneve_uninit,
1128         .ndo_open               = geneve_open,
1129         .ndo_stop               = geneve_stop,
1130         .ndo_start_xmit         = geneve_xmit,
1131         .ndo_change_mtu         = geneve_change_mtu,
1132         .ndo_validate_addr      = eth_validate_addr,
1133         .ndo_set_mac_address    = eth_mac_addr,
1134         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1135 };
1136
1137 static void geneve_get_drvinfo(struct net_device *dev,
1138                                struct ethtool_drvinfo *drvinfo)
1139 {
1140         strscpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1141         strscpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1142 }
1143
1144 static const struct ethtool_ops geneve_ethtool_ops = {
1145         .get_drvinfo    = geneve_get_drvinfo,
1146         .get_link       = ethtool_op_get_link,
1147 };
1148
1149 /* Info for udev, that this is a virtual tunnel endpoint */
1150 static const struct device_type geneve_type = {
1151         .name = "geneve",
1152 };
1153
1154 /* Calls the ndo_udp_tunnel_add of the caller in order to
1155  * supply the listening GENEVE udp ports. Callers are expected
1156  * to implement the ndo_udp_tunnel_add.
1157  */
1158 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1159 {
1160         struct net *net = dev_net(dev);
1161         struct geneve_net *gn = net_generic(net, geneve_net_id);
1162         struct geneve_sock *gs;
1163
1164         rcu_read_lock();
1165         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1166                 if (push) {
1167                         udp_tunnel_push_rx_port(dev, gs->sock,
1168                                                 UDP_TUNNEL_TYPE_GENEVE);
1169                 } else {
1170                         udp_tunnel_drop_rx_port(dev, gs->sock,
1171                                                 UDP_TUNNEL_TYPE_GENEVE);
1172                 }
1173         }
1174         rcu_read_unlock();
1175 }
1176
1177 /* Initialize the device structure. */
1178 static void geneve_setup(struct net_device *dev)
1179 {
1180         ether_setup(dev);
1181
1182         dev->netdev_ops = &geneve_netdev_ops;
1183         dev->ethtool_ops = &geneve_ethtool_ops;
1184         dev->needs_free_netdev = true;
1185
1186         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1187
1188         dev->features    |= NETIF_F_LLTX;
1189         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1190         dev->features    |= NETIF_F_RXCSUM;
1191         dev->features    |= NETIF_F_GSO_SOFTWARE;
1192
1193         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1194         dev->hw_features |= NETIF_F_RXCSUM;
1195         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1196
1197         dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
1198         /* MTU range: 68 - (something less than 65535) */
1199         dev->min_mtu = ETH_MIN_MTU;
1200         /* The max_mtu calculation does not take account of GENEVE
1201          * options, to avoid excluding potentially valid
1202          * configurations. This will be further reduced by IPvX hdr size.
1203          */
1204         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1205
1206         netif_keep_dst(dev);
1207         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1208         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1209         eth_hw_addr_random(dev);
1210 }
1211
1212 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1213         [IFLA_GENEVE_UNSPEC]            = { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT },
1214         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1215         [IFLA_GENEVE_REMOTE]            = { .len = sizeof_field(struct iphdr, daddr) },
1216         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1217         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1218         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1219         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1220         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1221         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1222         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1223         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1224         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1225         [IFLA_GENEVE_TTL_INHERIT]       = { .type = NLA_U8 },
1226         [IFLA_GENEVE_DF]                = { .type = NLA_U8 },
1227         [IFLA_GENEVE_INNER_PROTO_INHERIT]       = { .type = NLA_FLAG },
1228 };
1229
1230 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1231                            struct netlink_ext_ack *extack)
1232 {
1233         if (tb[IFLA_ADDRESS]) {
1234                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1235                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1236                                             "Provided link layer address is not Ethernet");
1237                         return -EINVAL;
1238                 }
1239
1240                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1241                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1242                                             "Provided Ethernet address is not unicast");
1243                         return -EADDRNOTAVAIL;
1244                 }
1245         }
1246
1247         if (!data) {
1248                 NL_SET_ERR_MSG(extack,
1249                                "Not enough attributes provided to perform the operation");
1250                 return -EINVAL;
1251         }
1252
1253         if (data[IFLA_GENEVE_ID]) {
1254                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1255
1256                 if (vni >= GENEVE_N_VID) {
1257                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1258                                             "Geneve ID must be lower than 16777216");
1259                         return -ERANGE;
1260                 }
1261         }
1262
1263         if (data[IFLA_GENEVE_DF]) {
1264                 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1265
1266                 if (df < 0 || df > GENEVE_DF_MAX) {
1267                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1268                                             "Invalid DF attribute");
1269                         return -EINVAL;
1270                 }
1271         }
1272
1273         return 0;
1274 }
1275
1276 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1277                                           const struct ip_tunnel_info *info,
1278                                           bool *tun_on_same_port,
1279                                           bool *tun_collect_md)
1280 {
1281         struct geneve_dev *geneve, *t = NULL;
1282
1283         *tun_on_same_port = false;
1284         *tun_collect_md = false;
1285         list_for_each_entry(geneve, &gn->geneve_list, next) {
1286                 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1287                         *tun_collect_md = geneve->cfg.collect_md;
1288                         *tun_on_same_port = true;
1289                 }
1290                 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1291                     info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1292                     !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1293                         t = geneve;
1294         }
1295         return t;
1296 }
1297
1298 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1299 {
1300         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1301                  info->key.ttl || info->key.label || info->key.tp_src ||
1302                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1303 }
1304
1305 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1306                                   struct ip_tunnel_info *b)
1307 {
1308         if (ip_tunnel_info_af(a) == AF_INET)
1309                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1310         else
1311                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1312 }
1313
1314 static int geneve_configure(struct net *net, struct net_device *dev,
1315                             struct netlink_ext_ack *extack,
1316                             const struct geneve_config *cfg)
1317 {
1318         struct geneve_net *gn = net_generic(net, geneve_net_id);
1319         struct geneve_dev *t, *geneve = netdev_priv(dev);
1320         const struct ip_tunnel_info *info = &cfg->info;
1321         bool tun_collect_md, tun_on_same_port;
1322         int err, encap_len;
1323
1324         if (cfg->collect_md && !is_tnl_info_zero(info)) {
1325                 NL_SET_ERR_MSG(extack,
1326                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1327                 return -EINVAL;
1328         }
1329
1330         geneve->net = net;
1331         geneve->dev = dev;
1332
1333         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1334         if (t)
1335                 return -EBUSY;
1336
1337         /* make enough headroom for basic scenario */
1338         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1339         if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1340                 encap_len += sizeof(struct iphdr);
1341                 dev->max_mtu -= sizeof(struct iphdr);
1342         } else {
1343                 encap_len += sizeof(struct ipv6hdr);
1344                 dev->max_mtu -= sizeof(struct ipv6hdr);
1345         }
1346         dev->needed_headroom = encap_len + ETH_HLEN;
1347
1348         if (cfg->collect_md) {
1349                 if (tun_on_same_port) {
1350                         NL_SET_ERR_MSG(extack,
1351                                        "There can be only one externally controlled device on a destination port");
1352                         return -EPERM;
1353                 }
1354         } else {
1355                 if (tun_collect_md) {
1356                         NL_SET_ERR_MSG(extack,
1357                                        "There already exists an externally controlled device on this destination port");
1358                         return -EPERM;
1359                 }
1360         }
1361
1362         dst_cache_reset(&geneve->cfg.info.dst_cache);
1363         memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1364
1365         if (geneve->cfg.inner_proto_inherit) {
1366                 dev->header_ops = NULL;
1367                 dev->type = ARPHRD_NONE;
1368                 dev->hard_header_len = 0;
1369                 dev->addr_len = 0;
1370                 dev->flags = IFF_POINTOPOINT | IFF_NOARP;
1371         }
1372
1373         err = register_netdevice(dev);
1374         if (err)
1375                 return err;
1376
1377         list_add(&geneve->next, &gn->geneve_list);
1378         return 0;
1379 }
1380
1381 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1382 {
1383         memset(info, 0, sizeof(*info));
1384         info->key.tp_dst = htons(dst_port);
1385 }
1386
1387 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1388                           struct netlink_ext_ack *extack,
1389                           struct geneve_config *cfg, bool changelink)
1390 {
1391         struct ip_tunnel_info *info = &cfg->info;
1392         int attrtype;
1393
1394         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1395                 NL_SET_ERR_MSG(extack,
1396                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1397                 return -EINVAL;
1398         }
1399
1400         if (data[IFLA_GENEVE_REMOTE]) {
1401                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1402                         attrtype = IFLA_GENEVE_REMOTE;
1403                         goto change_notsup;
1404                 }
1405
1406                 info->key.u.ipv4.dst =
1407                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1408
1409                 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1410                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1411                                             "Remote IPv4 address cannot be Multicast");
1412                         return -EINVAL;
1413                 }
1414         }
1415
1416         if (data[IFLA_GENEVE_REMOTE6]) {
1417 #if IS_ENABLED(CONFIG_IPV6)
1418                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1419                         attrtype = IFLA_GENEVE_REMOTE6;
1420                         goto change_notsup;
1421                 }
1422
1423                 info->mode = IP_TUNNEL_INFO_IPV6;
1424                 info->key.u.ipv6.dst =
1425                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1426
1427                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1428                     IPV6_ADDR_LINKLOCAL) {
1429                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1430                                             "Remote IPv6 address cannot be link-local");
1431                         return -EINVAL;
1432                 }
1433                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1434                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1435                                             "Remote IPv6 address cannot be Multicast");
1436                         return -EINVAL;
1437                 }
1438                 info->key.tun_flags |= TUNNEL_CSUM;
1439                 cfg->use_udp6_rx_checksums = true;
1440 #else
1441                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1442                                     "IPv6 support not enabled in the kernel");
1443                 return -EPFNOSUPPORT;
1444 #endif
1445         }
1446
1447         if (data[IFLA_GENEVE_ID]) {
1448                 __u32 vni;
1449                 __u8 tvni[3];
1450                 __be64 tunid;
1451
1452                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1453                 tvni[0] = (vni & 0x00ff0000) >> 16;
1454                 tvni[1] = (vni & 0x0000ff00) >> 8;
1455                 tvni[2] =  vni & 0x000000ff;
1456
1457                 tunid = vni_to_tunnel_id(tvni);
1458                 if (changelink && (tunid != info->key.tun_id)) {
1459                         attrtype = IFLA_GENEVE_ID;
1460                         goto change_notsup;
1461                 }
1462                 info->key.tun_id = tunid;
1463         }
1464
1465         if (data[IFLA_GENEVE_TTL_INHERIT]) {
1466                 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1467                         cfg->ttl_inherit = true;
1468                 else
1469                         cfg->ttl_inherit = false;
1470         } else if (data[IFLA_GENEVE_TTL]) {
1471                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1472                 cfg->ttl_inherit = false;
1473         }
1474
1475         if (data[IFLA_GENEVE_TOS])
1476                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1477
1478         if (data[IFLA_GENEVE_DF])
1479                 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1480
1481         if (data[IFLA_GENEVE_LABEL]) {
1482                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1483                                   IPV6_FLOWLABEL_MASK;
1484                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1485                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1486                                             "Label attribute only applies for IPv6 Geneve devices");
1487                         return -EINVAL;
1488                 }
1489         }
1490
1491         if (data[IFLA_GENEVE_PORT]) {
1492                 if (changelink) {
1493                         attrtype = IFLA_GENEVE_PORT;
1494                         goto change_notsup;
1495                 }
1496                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1497         }
1498
1499         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1500                 if (changelink) {
1501                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1502                         goto change_notsup;
1503                 }
1504                 cfg->collect_md = true;
1505         }
1506
1507         if (data[IFLA_GENEVE_UDP_CSUM]) {
1508                 if (changelink) {
1509                         attrtype = IFLA_GENEVE_UDP_CSUM;
1510                         goto change_notsup;
1511                 }
1512                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1513                         info->key.tun_flags |= TUNNEL_CSUM;
1514         }
1515
1516         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1517 #if IS_ENABLED(CONFIG_IPV6)
1518                 if (changelink) {
1519                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1520                         goto change_notsup;
1521                 }
1522                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1523                         info->key.tun_flags &= ~TUNNEL_CSUM;
1524 #else
1525                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1526                                     "IPv6 support not enabled in the kernel");
1527                 return -EPFNOSUPPORT;
1528 #endif
1529         }
1530
1531         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1532 #if IS_ENABLED(CONFIG_IPV6)
1533                 if (changelink) {
1534                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1535                         goto change_notsup;
1536                 }
1537                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1538                         cfg->use_udp6_rx_checksums = false;
1539 #else
1540                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1541                                     "IPv6 support not enabled in the kernel");
1542                 return -EPFNOSUPPORT;
1543 #endif
1544         }
1545
1546         if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) {
1547                 if (changelink) {
1548                         attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT;
1549                         goto change_notsup;
1550                 }
1551                 cfg->inner_proto_inherit = true;
1552         }
1553
1554         return 0;
1555 change_notsup:
1556         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1557                             "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported");
1558         return -EOPNOTSUPP;
1559 }
1560
1561 static void geneve_link_config(struct net_device *dev,
1562                                struct ip_tunnel_info *info, struct nlattr *tb[])
1563 {
1564         struct geneve_dev *geneve = netdev_priv(dev);
1565         int ldev_mtu = 0;
1566
1567         if (tb[IFLA_MTU]) {
1568                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1569                 return;
1570         }
1571
1572         switch (ip_tunnel_info_af(info)) {
1573         case AF_INET: {
1574                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1575                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1576
1577                 if (!IS_ERR(rt) && rt->dst.dev) {
1578                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1579                         ip_rt_put(rt);
1580                 }
1581                 break;
1582         }
1583 #if IS_ENABLED(CONFIG_IPV6)
1584         case AF_INET6: {
1585                 struct rt6_info *rt;
1586
1587                 if (!__in6_dev_get(dev))
1588                         break;
1589
1590                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1591                                 NULL, 0);
1592
1593                 if (rt && rt->dst.dev)
1594                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1595                 ip6_rt_put(rt);
1596                 break;
1597         }
1598 #endif
1599         }
1600
1601         if (ldev_mtu <= 0)
1602                 return;
1603
1604         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1605 }
1606
1607 static int geneve_newlink(struct net *net, struct net_device *dev,
1608                           struct nlattr *tb[], struct nlattr *data[],
1609                           struct netlink_ext_ack *extack)
1610 {
1611         struct geneve_config cfg = {
1612                 .df = GENEVE_DF_UNSET,
1613                 .use_udp6_rx_checksums = false,
1614                 .ttl_inherit = false,
1615                 .collect_md = false,
1616         };
1617         int err;
1618
1619         init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1620         err = geneve_nl2info(tb, data, extack, &cfg, false);
1621         if (err)
1622                 return err;
1623
1624         err = geneve_configure(net, dev, extack, &cfg);
1625         if (err)
1626                 return err;
1627
1628         geneve_link_config(dev, &cfg.info, tb);
1629
1630         return 0;
1631 }
1632
1633 /* Quiesces the geneve device data path for both TX and RX.
1634  *
1635  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1636  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1637  * to complete for the existing set of in-flight packets to be transmitted,
1638  * then we would have quiesced the transmit data path. All the future packets
1639  * will get dropped until we unquiesce the data path.
1640  *
1641  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1642  * if we set that to NULL under RCU and wait for synchronize_net() to
1643  * complete, then we would have quiesced the receive data path.
1644  */
1645 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1646                            struct geneve_sock **gs6)
1647 {
1648         *gs4 = rtnl_dereference(geneve->sock4);
1649         rcu_assign_pointer(geneve->sock4, NULL);
1650         if (*gs4)
1651                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1652 #if IS_ENABLED(CONFIG_IPV6)
1653         *gs6 = rtnl_dereference(geneve->sock6);
1654         rcu_assign_pointer(geneve->sock6, NULL);
1655         if (*gs6)
1656                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1657 #else
1658         *gs6 = NULL;
1659 #endif
1660         synchronize_net();
1661 }
1662
1663 /* Resumes the geneve device data path for both TX and RX. */
1664 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1665                              struct geneve_sock __maybe_unused *gs6)
1666 {
1667         rcu_assign_pointer(geneve->sock4, gs4);
1668         if (gs4)
1669                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1670 #if IS_ENABLED(CONFIG_IPV6)
1671         rcu_assign_pointer(geneve->sock6, gs6);
1672         if (gs6)
1673                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1674 #endif
1675         synchronize_net();
1676 }
1677
1678 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1679                              struct nlattr *data[],
1680                              struct netlink_ext_ack *extack)
1681 {
1682         struct geneve_dev *geneve = netdev_priv(dev);
1683         struct geneve_sock *gs4, *gs6;
1684         struct geneve_config cfg;
1685         int err;
1686
1687         /* If the geneve device is configured for metadata (or externally
1688          * controlled, for example, OVS), then nothing can be changed.
1689          */
1690         if (geneve->cfg.collect_md)
1691                 return -EOPNOTSUPP;
1692
1693         /* Start with the existing info. */
1694         memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1695         err = geneve_nl2info(tb, data, extack, &cfg, true);
1696         if (err)
1697                 return err;
1698
1699         if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1700                 dst_cache_reset(&cfg.info.dst_cache);
1701                 geneve_link_config(dev, &cfg.info, tb);
1702         }
1703
1704         geneve_quiesce(geneve, &gs4, &gs6);
1705         memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1706         geneve_unquiesce(geneve, gs4, gs6);
1707
1708         return 0;
1709 }
1710
1711 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1712 {
1713         struct geneve_dev *geneve = netdev_priv(dev);
1714
1715         list_del(&geneve->next);
1716         unregister_netdevice_queue(dev, head);
1717 }
1718
1719 static size_t geneve_get_size(const struct net_device *dev)
1720 {
1721         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1722                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1723                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1724                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1725                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_DF */
1726                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1727                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1728                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1729                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1730                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1731                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1732                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1733                 nla_total_size(0) +      /* IFLA_GENEVE_INNER_PROTO_INHERIT */
1734                 0;
1735 }
1736
1737 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1738 {
1739         struct geneve_dev *geneve = netdev_priv(dev);
1740         struct ip_tunnel_info *info = &geneve->cfg.info;
1741         bool ttl_inherit = geneve->cfg.ttl_inherit;
1742         bool metadata = geneve->cfg.collect_md;
1743         __u8 tmp_vni[3];
1744         __u32 vni;
1745
1746         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1747         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1748         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1749                 goto nla_put_failure;
1750
1751         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1752                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1753                                     info->key.u.ipv4.dst))
1754                         goto nla_put_failure;
1755                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1756                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1757                         goto nla_put_failure;
1758
1759 #if IS_ENABLED(CONFIG_IPV6)
1760         } else if (!metadata) {
1761                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1762                                      &info->key.u.ipv6.dst))
1763                         goto nla_put_failure;
1764                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1765                                !(info->key.tun_flags & TUNNEL_CSUM)))
1766                         goto nla_put_failure;
1767 #endif
1768         }
1769
1770         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1771             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1772             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1773                 goto nla_put_failure;
1774
1775         if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1776                 goto nla_put_failure;
1777
1778         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1779                 goto nla_put_failure;
1780
1781         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1782                 goto nla_put_failure;
1783
1784 #if IS_ENABLED(CONFIG_IPV6)
1785         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1786                        !geneve->cfg.use_udp6_rx_checksums))
1787                 goto nla_put_failure;
1788 #endif
1789
1790         if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1791                 goto nla_put_failure;
1792
1793         if (geneve->cfg.inner_proto_inherit &&
1794             nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT))
1795                 goto nla_put_failure;
1796
1797         return 0;
1798
1799 nla_put_failure:
1800         return -EMSGSIZE;
1801 }
1802
1803 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1804         .kind           = "geneve",
1805         .maxtype        = IFLA_GENEVE_MAX,
1806         .policy         = geneve_policy,
1807         .priv_size      = sizeof(struct geneve_dev),
1808         .setup          = geneve_setup,
1809         .validate       = geneve_validate,
1810         .newlink        = geneve_newlink,
1811         .changelink     = geneve_changelink,
1812         .dellink        = geneve_dellink,
1813         .get_size       = geneve_get_size,
1814         .fill_info      = geneve_fill_info,
1815 };
1816
1817 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1818                                         u8 name_assign_type, u16 dst_port)
1819 {
1820         struct nlattr *tb[IFLA_MAX + 1];
1821         struct net_device *dev;
1822         LIST_HEAD(list_kill);
1823         int err;
1824         struct geneve_config cfg = {
1825                 .df = GENEVE_DF_UNSET,
1826                 .use_udp6_rx_checksums = true,
1827                 .ttl_inherit = false,
1828                 .collect_md = true,
1829         };
1830
1831         memset(tb, 0, sizeof(tb));
1832         dev = rtnl_create_link(net, name, name_assign_type,
1833                                &geneve_link_ops, tb, NULL);
1834         if (IS_ERR(dev))
1835                 return dev;
1836
1837         init_tnl_info(&cfg.info, dst_port);
1838         err = geneve_configure(net, dev, NULL, &cfg);
1839         if (err) {
1840                 free_netdev(dev);
1841                 return ERR_PTR(err);
1842         }
1843
1844         /* openvswitch users expect packet sizes to be unrestricted,
1845          * so set the largest MTU we can.
1846          */
1847         err = geneve_change_mtu(dev, IP_MAX_MTU);
1848         if (err)
1849                 goto err;
1850
1851         err = rtnl_configure_link(dev, NULL, 0, NULL);
1852         if (err < 0)
1853                 goto err;
1854
1855         return dev;
1856 err:
1857         geneve_dellink(dev, &list_kill);
1858         unregister_netdevice_many(&list_kill);
1859         return ERR_PTR(err);
1860 }
1861 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1862
1863 static int geneve_netdevice_event(struct notifier_block *unused,
1864                                   unsigned long event, void *ptr)
1865 {
1866         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1867
1868         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1869                 geneve_offload_rx_ports(dev, true);
1870         else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1871                 geneve_offload_rx_ports(dev, false);
1872
1873         return NOTIFY_DONE;
1874 }
1875
1876 static struct notifier_block geneve_notifier_block __read_mostly = {
1877         .notifier_call = geneve_netdevice_event,
1878 };
1879
1880 static __net_init int geneve_init_net(struct net *net)
1881 {
1882         struct geneve_net *gn = net_generic(net, geneve_net_id);
1883
1884         INIT_LIST_HEAD(&gn->geneve_list);
1885         INIT_LIST_HEAD(&gn->sock_list);
1886         return 0;
1887 }
1888
1889 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1890 {
1891         struct geneve_net *gn = net_generic(net, geneve_net_id);
1892         struct geneve_dev *geneve, *next;
1893         struct net_device *dev, *aux;
1894
1895         /* gather any geneve devices that were moved into this ns */
1896         for_each_netdev_safe(net, dev, aux)
1897                 if (dev->rtnl_link_ops == &geneve_link_ops)
1898                         unregister_netdevice_queue(dev, head);
1899
1900         /* now gather any other geneve devices that were created in this ns */
1901         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1902                 /* If geneve->dev is in the same netns, it was already added
1903                  * to the list by the previous loop.
1904                  */
1905                 if (!net_eq(dev_net(geneve->dev), net))
1906                         unregister_netdevice_queue(geneve->dev, head);
1907         }
1908 }
1909
1910 static void __net_exit geneve_exit_batch_rtnl(struct list_head *net_list,
1911                                               struct list_head *dev_to_kill)
1912 {
1913         struct net *net;
1914
1915         list_for_each_entry(net, net_list, exit_list)
1916                 geneve_destroy_tunnels(net, dev_to_kill);
1917 }
1918
1919 static void __net_exit geneve_exit_net(struct net *net)
1920 {
1921         const struct geneve_net *gn = net_generic(net, geneve_net_id);
1922
1923         WARN_ON_ONCE(!list_empty(&gn->sock_list));
1924 }
1925
1926 static struct pernet_operations geneve_net_ops = {
1927         .init = geneve_init_net,
1928         .exit_batch_rtnl = geneve_exit_batch_rtnl,
1929         .exit = geneve_exit_net,
1930         .id   = &geneve_net_id,
1931         .size = sizeof(struct geneve_net),
1932 };
1933
1934 static int __init geneve_init_module(void)
1935 {
1936         int rc;
1937
1938         rc = register_pernet_subsys(&geneve_net_ops);
1939         if (rc)
1940                 goto out1;
1941
1942         rc = register_netdevice_notifier(&geneve_notifier_block);
1943         if (rc)
1944                 goto out2;
1945
1946         rc = rtnl_link_register(&geneve_link_ops);
1947         if (rc)
1948                 goto out3;
1949
1950         return 0;
1951 out3:
1952         unregister_netdevice_notifier(&geneve_notifier_block);
1953 out2:
1954         unregister_pernet_subsys(&geneve_net_ops);
1955 out1:
1956         return rc;
1957 }
1958 late_initcall(geneve_init_module);
1959
1960 static void __exit geneve_cleanup_module(void)
1961 {
1962         rtnl_link_unregister(&geneve_link_ops);
1963         unregister_netdevice_notifier(&geneve_notifier_block);
1964         unregister_pernet_subsys(&geneve_net_ops);
1965 }
1966 module_exit(geneve_cleanup_module);
1967
1968 MODULE_LICENSE("GPL");
1969 MODULE_VERSION(GENEVE_NETDEV_VER);
1970 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1971 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1972 MODULE_ALIAS_RTNL_LINK("geneve");