drivers/net/geneve.c

   1 /*
   2  * GENEVE: Generic Network Virtualization Encapsulation
   3  *
   4  * Copyright (c) 2015 Red Hat, Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/netdevice.h>
  16 #include <linux/etherdevice.h>
  17 #include <linux/hash.h>
  18 #include <net/rtnetlink.h>
  19 #include <net/geneve.h>
  20
  21 #define GENEVE_NETDEV_VER       "0.6"
  22
  23 #define GENEVE_UDP_PORT         6081
  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 /* per-network namespace private data for this module */
  36 struct geneve_net {
  37         struct list_head  geneve_list;
  38         struct hlist_head vni_list[VNI_HASH_SIZE];
  39 };
  40
  41 /* Pseudo network device */
  42 struct geneve_dev {
  43         struct hlist_node  hlist;       /* vni hash table */
  44         struct net         *net;        /* netns for packet i/o */
  45         struct net_device  *dev;        /* netdev for geneve tunnel */
  46         struct geneve_sock *sock;       /* socket used for geneve tunnel */
  47         u8                 vni[3];      /* virtual network ID for tunnel */
  48         u8                 ttl;         /* TTL override */
  49         u8                 tos;         /* TOS override */
  50         struct sockaddr_in remote;      /* IPv4 address for link partner */
  51         struct list_head   next;        /* geneve's per namespace list */
  52 };
  53
  54 static int geneve_net_id;
  55
  56 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  57 {
  58         __u32 vnid;
  59
  60         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  61         return hash_32(vnid, VNI_HASH_BITS);
  62 }
  63
  64 /* geneve receive/decap routine */
  65 static void geneve_rx(struct geneve_sock *gs, struct sk_buff *skb)
  66 {
  67         struct genevehdr *gnvh = geneve_hdr(skb);
  68         struct geneve_dev *dummy, *geneve = NULL;
  69         struct geneve_net *gn;
  70         struct iphdr *iph = NULL;
  71         struct pcpu_sw_netstats *stats;
  72         struct hlist_head *vni_list_head;
  73         int err = 0;
  74         __u32 hash;
  75
  76         iph = ip_hdr(skb); /* Still outer IP header... */
  77
  78         gn = gs->rcv_data;
  79
  80         /* Find the device for this VNI */
  81         hash = geneve_net_vni_hash(gnvh->vni);
  82         vni_list_head = &gn->vni_list[hash];
  83         hlist_for_each_entry_rcu(dummy, vni_list_head, hlist) {
  84                 if (!memcmp(gnvh->vni, dummy->vni, sizeof(dummy->vni)) &&
  85                     iph->saddr == dummy->remote.sin_addr.s_addr) {
  86                         geneve = dummy;
  87                         break;
  88                 }
  89         }
  90         if (!geneve)
  91                 goto drop;
  92
  93         /* Drop packets w/ critical options,
  94          * since we don't support any...
  95          */
  96         if (gnvh->critical)
  97                 goto drop;
  98
  99         skb_reset_mac_header(skb);
 100         skb_scrub_packet(skb, !net_eq(geneve->net, dev_net(geneve->dev)));
 101         skb->protocol = eth_type_trans(skb, geneve->dev);
 102         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 103
 104         /* Ignore packet loops (and multicast echo) */
 105         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
 106                 goto drop;
 107
 108         skb_reset_network_header(skb);
 109
 110         iph = ip_hdr(skb); /* Now inner IP header... */
 111         err = IP_ECN_decapsulate(iph, skb);
 112
 113         if (unlikely(err)) {
 114                 if (log_ecn_error)
 115                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
 116                                              &iph->saddr, iph->tos);
 117                 if (err > 1) {
 118                         ++geneve->dev->stats.rx_frame_errors;
 119                         ++geneve->dev->stats.rx_errors;
 120                         goto drop;
 121                 }
 122         }
 123
 124         stats = this_cpu_ptr(geneve->dev->tstats);
 125         u64_stats_update_begin(&stats->syncp);
 126         stats->rx_packets++;
 127         stats->rx_bytes += skb->len;
 128         u64_stats_update_end(&stats->syncp);
 129
 130         netif_rx(skb);
 131
 132         return;
 133 drop:
 134         /* Consume bad packet */
 135         kfree_skb(skb);
 136 }
 137
 138 /* Setup stats when device is created */
 139 static int geneve_init(struct net_device *dev)
 140 {
 141         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 142         if (!dev->tstats)
 143                 return -ENOMEM;
 144
 145         return 0;
 146 }
 147
 148 static void geneve_uninit(struct net_device *dev)
 149 {
 150         free_percpu(dev->tstats);
 151 }
 152
 153 static int geneve_open(struct net_device *dev)
 154 {
 155         struct geneve_dev *geneve = netdev_priv(dev);
 156         struct net *net = geneve->net;
 157         struct geneve_net *gn = net_generic(geneve->net, geneve_net_id);
 158         struct geneve_sock *gs;
 159
 160         gs = geneve_sock_add(net, htons(GENEVE_UDP_PORT), geneve_rx, gn,
 161                              false, false);
 162         if (IS_ERR(gs))
 163                 return PTR_ERR(gs);
 164
 165         geneve->sock = gs;
 166
 167         return 0;
 168 }
 169
 170 static int geneve_stop(struct net_device *dev)
 171 {
 172         struct geneve_dev *geneve = netdev_priv(dev);
 173         struct geneve_sock *gs = geneve->sock;
 174
 175         geneve_sock_release(gs);
 176
 177         return 0;
 178 }
 179
 180 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
 181 {
 182         struct geneve_dev *geneve = netdev_priv(dev);
 183         struct geneve_sock *gs = geneve->sock;
 184         struct rtable *rt = NULL;
 185         const struct iphdr *iip; /* interior IP header */
 186         struct flowi4 fl4;
 187         int err;
 188         __be16 sport;
 189         __u8 tos, ttl;
 190
 191         iip = ip_hdr(skb);
 192
 193         skb_reset_mac_header(skb);
 194
 195         /* TODO: port min/max limits should be configurable */
 196         sport = udp_flow_src_port(dev_net(dev), skb, 0, 0, true);
 197
 198         tos = geneve->tos;
 199         if (tos == 1)
 200                 tos = ip_tunnel_get_dsfield(iip, skb);
 201
 202         memset(&fl4, 0, sizeof(fl4));
 203         fl4.flowi4_tos = RT_TOS(tos);
 204         fl4.daddr = geneve->remote.sin_addr.s_addr;
 205         rt = ip_route_output_key(geneve->net, &fl4);
 206         if (IS_ERR(rt)) {
 207                 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
 208                 dev->stats.tx_carrier_errors++;
 209                 goto tx_error;
 210         }
 211         if (rt->dst.dev == dev) { /* is this necessary? */
 212                 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
 213                 dev->stats.collisions++;
 214                 goto rt_tx_error;
 215         }
 216
 217         tos = ip_tunnel_ecn_encap(tos, iip, skb);
 218
 219         ttl = geneve->ttl;
 220         if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 221                 ttl = 1;
 222
 223         ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 224
 225         /* no need to handle local destination and encap bypass...yet... */
 226
 227         err = geneve_xmit_skb(gs, rt, skb, fl4.saddr, fl4.daddr,
 228                               tos, ttl, 0, sport, htons(GENEVE_UDP_PORT), 0,
 229                               geneve->vni, 0, NULL, false,
 230                               !net_eq(geneve->net, dev_net(geneve->dev)));
 231         if (err < 0)
 232                 ip_rt_put(rt);
 233
 234         iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
 235
 236         return NETDEV_TX_OK;
 237
 238 rt_tx_error:
 239         ip_rt_put(rt);
 240 tx_error:
 241         dev->stats.tx_errors++;
 242         dev_kfree_skb(skb);
 243         return NETDEV_TX_OK;
 244 }
 245
 246 static const struct net_device_ops geneve_netdev_ops = {
 247         .ndo_init               = geneve_init,
 248         .ndo_uninit             = geneve_uninit,
 249         .ndo_open               = geneve_open,
 250         .ndo_stop               = geneve_stop,
 251         .ndo_start_xmit         = geneve_xmit,
 252         .ndo_get_stats64        = ip_tunnel_get_stats64,
 253         .ndo_change_mtu         = eth_change_mtu,
 254         .ndo_validate_addr      = eth_validate_addr,
 255         .ndo_set_mac_address    = eth_mac_addr,
 256 };
 257
 258 static void geneve_get_drvinfo(struct net_device *dev,
 259                                struct ethtool_drvinfo *drvinfo)
 260 {
 261         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
 262         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
 263 }
 264
 265 static const struct ethtool_ops geneve_ethtool_ops = {
 266         .get_drvinfo    = geneve_get_drvinfo,
 267         .get_link       = ethtool_op_get_link,
 268 };
 269
 270 /* Info for udev, that this is a virtual tunnel endpoint */
 271 static struct device_type geneve_type = {
 272         .name = "geneve",
 273 };
 274
 275 /* Initialize the device structure. */
 276 static void geneve_setup(struct net_device *dev)
 277 {
 278         ether_setup(dev);
 279
 280         dev->netdev_ops = &geneve_netdev_ops;
 281         dev->ethtool_ops = &geneve_ethtool_ops;
 282         dev->destructor = free_netdev;
 283
 284         SET_NETDEV_DEVTYPE(dev, &geneve_type);
 285
 286         dev->tx_queue_len = 0;
 287         dev->features    |= NETIF_F_LLTX;
 288         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
 289         dev->features    |= NETIF_F_RXCSUM;
 290         dev->features    |= NETIF_F_GSO_SOFTWARE;
 291
 292         dev->vlan_features = dev->features;
 293         dev->features    |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
 294
 295         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
 296         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
 297         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
 298
 299         netif_keep_dst(dev);
 300         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 301 }
 302
 303 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
 304         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
 305         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
 306         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
 307         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
 308 };
 309
 310 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
 311 {
 312         if (tb[IFLA_ADDRESS]) {
 313                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 314                         return -EINVAL;
 315
 316                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 317                         return -EADDRNOTAVAIL;
 318         }
 319
 320         if (!data)
 321                 return -EINVAL;
 322
 323         if (data[IFLA_GENEVE_ID]) {
 324                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
 325
 326                 if (vni >= GENEVE_VID_MASK)
 327                         return -ERANGE;
 328         }
 329
 330         return 0;
 331 }
 332
 333 static int geneve_newlink(struct net *net, struct net_device *dev,
 334                          struct nlattr *tb[], struct nlattr *data[])
 335 {
 336         struct geneve_net *gn = net_generic(net, geneve_net_id);
 337         struct geneve_dev *dummy, *geneve = netdev_priv(dev);
 338         struct hlist_head *vni_list_head;
 339         struct sockaddr_in remote;      /* IPv4 address for link partner */
 340         __u32 vni, hash;
 341         int err;
 342
 343         if (!data[IFLA_GENEVE_ID] || !data[IFLA_GENEVE_REMOTE])
 344                 return -EINVAL;
 345
 346         geneve->net = net;
 347         geneve->dev = dev;
 348
 349         vni = nla_get_u32(data[IFLA_GENEVE_ID]);
 350         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
 351         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
 352         geneve->vni[2] =  vni & 0x000000ff;
 353
 354         geneve->remote.sin_addr.s_addr =
 355                 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
 356         if (IN_MULTICAST(ntohl(geneve->remote.sin_addr.s_addr)))
 357                 return -EINVAL;
 358
 359         remote = geneve->remote;
 360         hash = geneve_net_vni_hash(geneve->vni);
 361         vni_list_head = &gn->vni_list[hash];
 362         hlist_for_each_entry_rcu(dummy, vni_list_head, hlist) {
 363                 if (!memcmp(geneve->vni, dummy->vni, sizeof(dummy->vni)) &&
 364                     !memcmp(&remote, &dummy->remote, sizeof(dummy->remote)))
 365                         return -EBUSY;
 366         }
 367
 368         if (tb[IFLA_ADDRESS] == NULL)
 369                 eth_hw_addr_random(dev);
 370
 371         err = register_netdevice(dev);
 372         if (err)
 373                 return err;
 374
 375         if (data[IFLA_GENEVE_TTL])
 376                 geneve->ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
 377
 378         if (data[IFLA_GENEVE_TOS])
 379                 geneve->tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
 380
 381         list_add(&geneve->next, &gn->geneve_list);
 382
 383         hlist_add_head_rcu(&geneve->hlist, &gn->vni_list[hash]);
 384
 385         return 0;
 386 }
 387
 388 static void geneve_dellink(struct net_device *dev, struct list_head *head)
 389 {
 390         struct geneve_dev *geneve = netdev_priv(dev);
 391
 392         if (!hlist_unhashed(&geneve->hlist))
 393                 hlist_del_rcu(&geneve->hlist);
 394
 395         list_del(&geneve->next);
 396         unregister_netdevice_queue(dev, head);
 397 }
 398
 399 static size_t geneve_get_size(const struct net_device *dev)
 400 {
 401         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
 402                 nla_total_size(sizeof(struct in_addr)) + /* IFLA_GENEVE_REMOTE */
 403                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
 404                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
 405                 0;
 406 }
 407
 408 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
 409 {
 410         struct geneve_dev *geneve = netdev_priv(dev);
 411         __u32 vni;
 412
 413         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
 414         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
 415                 goto nla_put_failure;
 416
 417         if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
 418                             geneve->remote.sin_addr.s_addr))
 419                 goto nla_put_failure;
 420
 421         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
 422             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos))
 423                 goto nla_put_failure;
 424
 425         return 0;
 426
 427 nla_put_failure:
 428         return -EMSGSIZE;
 429 }
 430
 431 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
 432         .kind           = "geneve",
 433         .maxtype        = IFLA_GENEVE_MAX,
 434         .policy         = geneve_policy,
 435         .priv_size      = sizeof(struct geneve_dev),
 436         .setup          = geneve_setup,
 437         .validate       = geneve_validate,
 438         .newlink        = geneve_newlink,
 439         .dellink        = geneve_dellink,
 440         .get_size       = geneve_get_size,
 441         .fill_info      = geneve_fill_info,
 442 };
 443
 444 static __net_init int geneve_init_net(struct net *net)
 445 {
 446         struct geneve_net *gn = net_generic(net, geneve_net_id);
 447         unsigned int h;
 448
 449         INIT_LIST_HEAD(&gn->geneve_list);
 450
 451         for (h = 0; h < VNI_HASH_SIZE; ++h)
 452                 INIT_HLIST_HEAD(&gn->vni_list[h]);
 453
 454         return 0;
 455 }
 456
 457 static void __net_exit geneve_exit_net(struct net *net)
 458 {
 459         struct geneve_net *gn = net_generic(net, geneve_net_id);
 460         struct geneve_dev *geneve, *next;
 461         struct net_device *dev, *aux;
 462         LIST_HEAD(list);
 463
 464         rtnl_lock();
 465
 466         /* gather any geneve devices that were moved into this ns */
 467         for_each_netdev_safe(net, dev, aux)
 468                 if (dev->rtnl_link_ops == &geneve_link_ops)
 469                         unregister_netdevice_queue(dev, &list);
 470
 471         /* now gather any other geneve devices that were created in this ns */
 472         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
 473                 /* If geneve->dev is in the same netns, it was already added
 474                  * to the list by the previous loop.
 475                  */
 476                 if (!net_eq(dev_net(geneve->dev), net))
 477                         unregister_netdevice_queue(geneve->dev, &list);
 478         }
 479
 480         /* unregister the devices gathered above */
 481         unregister_netdevice_many(&list);
 482         rtnl_unlock();
 483 }
 484
 485 static struct pernet_operations geneve_net_ops = {
 486         .init = geneve_init_net,
 487         .exit = geneve_exit_net,
 488         .id   = &geneve_net_id,
 489         .size = sizeof(struct geneve_net),
 490 };
 491
 492 static int __init geneve_init_module(void)
 493 {
 494         int rc;
 495
 496         rc = register_pernet_subsys(&geneve_net_ops);
 497         if (rc)
 498                 goto out1;
 499
 500         rc = rtnl_link_register(&geneve_link_ops);
 501         if (rc)
 502                 goto out2;
 503
 504         return 0;
 505 out2:
 506         unregister_pernet_subsys(&geneve_net_ops);
 507 out1:
 508         return rc;
 509 }
 510 late_initcall(geneve_init_module);
 511
 512 static void __exit geneve_cleanup_module(void)
 513 {
 514         rtnl_link_unregister(&geneve_link_ops);
 515         unregister_pernet_subsys(&geneve_net_ops);
 516 }
 517 module_exit(geneve_cleanup_module);
 518
 519 MODULE_LICENSE("GPL");
 520 MODULE_VERSION(GENEVE_NETDEV_VER);
 521 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
 522 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
 523 MODULE_ALIAS_RTNL_LINK("geneve");