drivers/net/vrf.c

   1 /*
   2  * vrf.c: device driver to encapsulate a VRF space
   3  *
   4  * Copyright (c) 2015 Cumulus Networks. All rights reserved.
   5  * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com>
   6  * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com>
   7  *
   8  * Based on dummy, team and ipvlan drivers
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  */
  15
  16 #include <linux/module.h>
  17 #include <linux/kernel.h>
  18 #include <linux/netdevice.h>
  19 #include <linux/etherdevice.h>
  20 #include <linux/ip.h>
  21 #include <linux/init.h>
  22 #include <linux/moduleparam.h>
  23 #include <linux/netfilter.h>
  24 #include <linux/rtnetlink.h>
  25 #include <net/rtnetlink.h>
  26 #include <linux/u64_stats_sync.h>
  27 #include <linux/hashtable.h>
  28
  29 #include <linux/inetdevice.h>
  30 #include <net/arp.h>
  31 #include <net/ip.h>
  32 #include <net/ip_fib.h>
  33 #include <net/ip6_route.h>
  34 #include <net/rtnetlink.h>
  35 #include <net/route.h>
  36 #include <net/addrconf.h>
  37 #include <net/l3mdev.h>
  38
  39 #define DRV_NAME        "vrf"
  40 #define DRV_VERSION     "1.0"
  41
  42 #define vrf_is_slave(dev)   ((dev)->flags & IFF_SLAVE)
  43
  44 #define vrf_master_get_rcu(dev) \
  45         ((struct net_device *)rcu_dereference(dev->rx_handler_data))
  46
  47 struct slave {
  48         struct list_head        list;
  49         struct net_device       *dev;
  50 };
  51
  52 struct slave_queue {
  53         struct list_head        all_slaves;
  54 };
  55
  56 struct net_vrf {
  57         struct slave_queue      queue;
  58         struct rtable           *rth;
  59         u32                     tb_id;
  60 };
  61
  62 struct pcpu_dstats {
  63         u64                     tx_pkts;
  64         u64                     tx_bytes;
  65         u64                     tx_drps;
  66         u64                     rx_pkts;
  67         u64                     rx_bytes;
  68         struct u64_stats_sync   syncp;
  69 };
  70
  71 static struct dst_entry *vrf_ip_check(struct dst_entry *dst, u32 cookie)
  72 {
  73         return dst;
  74 }
  75
  76 static int vrf_ip_local_out(struct sk_buff *skb)
  77 {
  78         return ip_local_out(skb);
  79 }
  80
  81 static unsigned int vrf_v4_mtu(const struct dst_entry *dst)
  82 {
  83         /* TO-DO: return max ethernet size? */
  84         return dst->dev->mtu;
  85 }
  86
  87 static void vrf_dst_destroy(struct dst_entry *dst)
  88 {
  89         /* our dst lives forever - or until the device is closed */
  90 }
  91
  92 static unsigned int vrf_default_advmss(const struct dst_entry *dst)
  93 {
  94         return 65535 - 40;
  95 }
  96
  97 static struct dst_ops vrf_dst_ops = {
  98         .family         = AF_INET,
  99         .local_out      = vrf_ip_local_out,
 100         .check          = vrf_ip_check,
 101         .mtu            = vrf_v4_mtu,
 102         .destroy        = vrf_dst_destroy,
 103         .default_advmss = vrf_default_advmss,
 104 };
 105
 106 static bool is_ip_rx_frame(struct sk_buff *skb)
 107 {
 108         switch (skb->protocol) {
 109         case htons(ETH_P_IP):
 110         case htons(ETH_P_IPV6):
 111                 return true;
 112         }
 113         return false;
 114 }
 115
 116 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
 117 {
 118         vrf_dev->stats.tx_errors++;
 119         kfree_skb(skb);
 120 }
 121
 122 /* note: already called with rcu_read_lock */
 123 static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb)
 124 {
 125         struct sk_buff *skb = *pskb;
 126
 127         if (is_ip_rx_frame(skb)) {
 128                 struct net_device *dev = vrf_master_get_rcu(skb->dev);
 129                 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
 130
 131                 u64_stats_update_begin(&dstats->syncp);
 132                 dstats->rx_pkts++;
 133                 dstats->rx_bytes += skb->len;
 134                 u64_stats_update_end(&dstats->syncp);
 135
 136                 skb->dev = dev;
 137
 138                 return RX_HANDLER_ANOTHER;
 139         }
 140         return RX_HANDLER_PASS;
 141 }
 142
 143 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
 144                                                  struct rtnl_link_stats64 *stats)
 145 {
 146         int i;
 147
 148         for_each_possible_cpu(i) {
 149                 const struct pcpu_dstats *dstats;
 150                 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
 151                 unsigned int start;
 152
 153                 dstats = per_cpu_ptr(dev->dstats, i);
 154                 do {
 155                         start = u64_stats_fetch_begin_irq(&dstats->syncp);
 156                         tbytes = dstats->tx_bytes;
 157                         tpkts = dstats->tx_pkts;
 158                         tdrops = dstats->tx_drps;
 159                         rbytes = dstats->rx_bytes;
 160                         rpkts = dstats->rx_pkts;
 161                 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
 162                 stats->tx_bytes += tbytes;
 163                 stats->tx_packets += tpkts;
 164                 stats->tx_dropped += tdrops;
 165                 stats->rx_bytes += rbytes;
 166                 stats->rx_packets += rpkts;
 167         }
 168         return stats;
 169 }
 170
 171 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
 172                                            struct net_device *dev)
 173 {
 174         vrf_tx_error(dev, skb);
 175         return NET_XMIT_DROP;
 176 }
 177
 178 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
 179                             struct net_device *vrf_dev)
 180 {
 181         struct rtable *rt;
 182         int err = 1;
 183
 184         rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
 185         if (IS_ERR(rt))
 186                 goto out;
 187
 188         /* TO-DO: what about broadcast ? */
 189         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 190                 ip_rt_put(rt);
 191                 goto out;
 192         }
 193
 194         skb_dst_drop(skb);
 195         skb_dst_set(skb, &rt->dst);
 196         err = 0;
 197 out:
 198         return err;
 199 }
 200
 201 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
 202                                            struct net_device *vrf_dev)
 203 {
 204         struct iphdr *ip4h = ip_hdr(skb);
 205         int ret = NET_XMIT_DROP;
 206         struct flowi4 fl4 = {
 207                 /* needed to match OIF rule */
 208                 .flowi4_oif = vrf_dev->ifindex,
 209                 .flowi4_iif = LOOPBACK_IFINDEX,
 210                 .flowi4_tos = RT_TOS(ip4h->tos),
 211                 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
 212                                 FLOWI_FLAG_SKIP_NH_OIF,
 213                 .daddr = ip4h->daddr,
 214         };
 215
 216         if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
 217                 goto err;
 218
 219         if (!ip4h->saddr) {
 220                 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
 221                                                RT_SCOPE_LINK);
 222         }
 223
 224         ret = ip_local_out(skb);
 225         if (unlikely(net_xmit_eval(ret)))
 226                 vrf_dev->stats.tx_errors++;
 227         else
 228                 ret = NET_XMIT_SUCCESS;
 229
 230 out:
 231         return ret;
 232 err:
 233         vrf_tx_error(vrf_dev, skb);
 234         goto out;
 235 }
 236
 237 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
 238 {
 239         /* strip the ethernet header added for pass through VRF device */
 240         __skb_pull(skb, skb_network_offset(skb));
 241
 242         switch (skb->protocol) {
 243         case htons(ETH_P_IP):
 244                 return vrf_process_v4_outbound(skb, dev);
 245         case htons(ETH_P_IPV6):
 246                 return vrf_process_v6_outbound(skb, dev);
 247         default:
 248                 vrf_tx_error(dev, skb);
 249                 return NET_XMIT_DROP;
 250         }
 251 }
 252
 253 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
 254 {
 255         netdev_tx_t ret = is_ip_tx_frame(skb, dev);
 256
 257         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 258                 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
 259
 260                 u64_stats_update_begin(&dstats->syncp);
 261                 dstats->tx_pkts++;
 262                 dstats->tx_bytes += skb->len;
 263                 u64_stats_update_end(&dstats->syncp);
 264         } else {
 265                 this_cpu_inc(dev->dstats->tx_drps);
 266         }
 267
 268         return ret;
 269 }
 270
 271 /* modelled after ip_finish_output2 */
 272 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 273 {
 274         struct dst_entry *dst = skb_dst(skb);
 275         struct rtable *rt = (struct rtable *)dst;
 276         struct net_device *dev = dst->dev;
 277         unsigned int hh_len = LL_RESERVED_SPACE(dev);
 278         struct neighbour *neigh;
 279         u32 nexthop;
 280         int ret = -EINVAL;
 281
 282         /* Be paranoid, rather than too clever. */
 283         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
 284                 struct sk_buff *skb2;
 285
 286                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
 287                 if (!skb2) {
 288                         ret = -ENOMEM;
 289                         goto err;
 290                 }
 291                 if (skb->sk)
 292                         skb_set_owner_w(skb2, skb->sk);
 293
 294                 consume_skb(skb);
 295                 skb = skb2;
 296         }
 297
 298         rcu_read_lock_bh();
 299
 300         nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
 301         neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
 302         if (unlikely(!neigh))
 303                 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
 304         if (!IS_ERR(neigh))
 305                 ret = dst_neigh_output(dst, neigh, skb);
 306
 307         rcu_read_unlock_bh();
 308 err:
 309         if (unlikely(ret < 0))
 310                 vrf_tx_error(skb->dev, skb);
 311         return ret;
 312 }
 313
 314 static int vrf_output(struct sock *sk, struct sk_buff *skb)
 315 {
 316         struct net_device *dev = skb_dst(skb)->dev;
 317         struct net *net = dev_net(dev);
 318
 319         IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
 320
 321         skb->dev = dev;
 322         skb->protocol = htons(ETH_P_IP);
 323
 324         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
 325                             net, sk, skb, NULL, dev,
 326                             vrf_finish_output,
 327                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 328 }
 329
 330 static void vrf_rtable_destroy(struct net_vrf *vrf)
 331 {
 332         struct dst_entry *dst = (struct dst_entry *)vrf->rth;
 333
 334         dst_destroy(dst);
 335         vrf->rth = NULL;
 336 }
 337
 338 static struct rtable *vrf_rtable_create(struct net_device *dev)
 339 {
 340         struct net_vrf *vrf = netdev_priv(dev);
 341         struct rtable *rth;
 342
 343         rth = dst_alloc(&vrf_dst_ops, dev, 2,
 344                         DST_OBSOLETE_NONE,
 345                         (DST_HOST | DST_NOPOLICY | DST_NOXFRM));
 346         if (rth) {
 347                 rth->dst.output = vrf_output;
 348                 rth->rt_genid   = rt_genid_ipv4(dev_net(dev));
 349                 rth->rt_flags   = 0;
 350                 rth->rt_type    = RTN_UNICAST;
 351                 rth->rt_is_input = 0;
 352                 rth->rt_iif     = 0;
 353                 rth->rt_pmtu    = 0;
 354                 rth->rt_gateway = 0;
 355                 rth->rt_uses_gateway = 0;
 356                 rth->rt_table_id = vrf->tb_id;
 357                 INIT_LIST_HEAD(&rth->rt_uncached);
 358                 rth->rt_uncached_list = NULL;
 359         }
 360
 361         return rth;
 362 }
 363
 364 /**************************** device handling ********************/
 365
 366 /* cycle interface to flush neighbor cache and move routes across tables */
 367 static void cycle_netdev(struct net_device *dev)
 368 {
 369         unsigned int flags = dev->flags;
 370         int ret;
 371
 372         if (!netif_running(dev))
 373                 return;
 374
 375         ret = dev_change_flags(dev, flags & ~IFF_UP);
 376         if (ret >= 0)
 377                 ret = dev_change_flags(dev, flags);
 378
 379         if (ret < 0) {
 380                 netdev_err(dev,
 381                            "Failed to cycle device %s; route tables might be wrong!\n",
 382                            dev->name);
 383         }
 384 }
 385
 386 static struct slave *__vrf_find_slave_dev(struct slave_queue *queue,
 387                                           struct net_device *dev)
 388 {
 389         struct list_head *head = &queue->all_slaves;
 390         struct slave *slave;
 391
 392         list_for_each_entry(slave, head, list) {
 393                 if (slave->dev == dev)
 394                         return slave;
 395         }
 396
 397         return NULL;
 398 }
 399
 400 /* inverse of __vrf_insert_slave */
 401 static void __vrf_remove_slave(struct slave_queue *queue, struct slave *slave)
 402 {
 403         list_del(&slave->list);
 404 }
 405
 406 static void __vrf_insert_slave(struct slave_queue *queue, struct slave *slave)
 407 {
 408         list_add(&slave->list, &queue->all_slaves);
 409 }
 410
 411 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
 412 {
 413         struct slave *slave = kzalloc(sizeof(*slave), GFP_KERNEL);
 414         struct net_vrf *vrf = netdev_priv(dev);
 415         struct slave_queue *queue = &vrf->queue;
 416         int ret = -ENOMEM;
 417
 418         if (!slave)
 419                 goto out_fail;
 420
 421         slave->dev = port_dev;
 422
 423         /* register the packet handler for slave ports */
 424         ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev);
 425         if (ret) {
 426                 netdev_err(port_dev,
 427                            "Device %s failed to register rx_handler\n",
 428                            port_dev->name);
 429                 goto out_fail;
 430         }
 431
 432         ret = netdev_master_upper_dev_link(port_dev, dev);
 433         if (ret < 0)
 434                 goto out_unregister;
 435
 436         port_dev->flags |= IFF_SLAVE;
 437         __vrf_insert_slave(queue, slave);
 438         cycle_netdev(port_dev);
 439
 440         return 0;
 441
 442 out_unregister:
 443         netdev_rx_handler_unregister(port_dev);
 444 out_fail:
 445         kfree(slave);
 446         return ret;
 447 }
 448
 449 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
 450 {
 451         if (netif_is_l3_master(port_dev) || vrf_is_slave(port_dev))
 452                 return -EINVAL;
 453
 454         return do_vrf_add_slave(dev, port_dev);
 455 }
 456
 457 /* inverse of do_vrf_add_slave */
 458 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
 459 {
 460         struct net_vrf *vrf = netdev_priv(dev);
 461         struct slave_queue *queue = &vrf->queue;
 462         struct slave *slave;
 463
 464         netdev_upper_dev_unlink(port_dev, dev);
 465         port_dev->flags &= ~IFF_SLAVE;
 466
 467         netdev_rx_handler_unregister(port_dev);
 468
 469         cycle_netdev(port_dev);
 470
 471         slave = __vrf_find_slave_dev(queue, port_dev);
 472         if (slave)
 473                 __vrf_remove_slave(queue, slave);
 474
 475         kfree(slave);
 476
 477         return 0;
 478 }
 479
 480 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
 481 {
 482         return do_vrf_del_slave(dev, port_dev);
 483 }
 484
 485 static void vrf_dev_uninit(struct net_device *dev)
 486 {
 487         struct net_vrf *vrf = netdev_priv(dev);
 488         struct slave_queue *queue = &vrf->queue;
 489         struct list_head *head = &queue->all_slaves;
 490         struct slave *slave, *next;
 491
 492         vrf_rtable_destroy(vrf);
 493
 494         list_for_each_entry_safe(slave, next, head, list)
 495                 vrf_del_slave(dev, slave->dev);
 496
 497         free_percpu(dev->dstats);
 498         dev->dstats = NULL;
 499 }
 500
 501 static int vrf_dev_init(struct net_device *dev)
 502 {
 503         struct net_vrf *vrf = netdev_priv(dev);
 504
 505         INIT_LIST_HEAD(&vrf->queue.all_slaves);
 506
 507         dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
 508         if (!dev->dstats)
 509                 goto out_nomem;
 510
 511         /* create the default dst which points back to us */
 512         vrf->rth = vrf_rtable_create(dev);
 513         if (!vrf->rth)
 514                 goto out_stats;
 515
 516         dev->flags = IFF_MASTER | IFF_NOARP;
 517
 518         return 0;
 519
 520 out_stats:
 521         free_percpu(dev->dstats);
 522         dev->dstats = NULL;
 523 out_nomem:
 524         return -ENOMEM;
 525 }
 526
 527 static const struct net_device_ops vrf_netdev_ops = {
 528         .ndo_init               = vrf_dev_init,
 529         .ndo_uninit             = vrf_dev_uninit,
 530         .ndo_start_xmit         = vrf_xmit,
 531         .ndo_get_stats64        = vrf_get_stats64,
 532         .ndo_add_slave          = vrf_add_slave,
 533         .ndo_del_slave          = vrf_del_slave,
 534 };
 535
 536 static u32 vrf_fib_table(const struct net_device *dev)
 537 {
 538         struct net_vrf *vrf = netdev_priv(dev);
 539
 540         return vrf->tb_id;
 541 }
 542
 543 static struct rtable *vrf_get_rtable(const struct net_device *dev,
 544                                      const struct flowi4 *fl4)
 545 {
 546         struct rtable *rth = NULL;
 547
 548         if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
 549                 struct net_vrf *vrf = netdev_priv(dev);
 550
 551                 rth = vrf->rth;
 552                 atomic_inc(&rth->dst.__refcnt);
 553         }
 554
 555         return rth;
 556 }
 557
 558 static const struct l3mdev_ops vrf_l3mdev_ops = {
 559         .l3mdev_fib_table       = vrf_fib_table,
 560         .l3mdev_get_rtable      = vrf_get_rtable,
 561 };
 562
 563 static void vrf_get_drvinfo(struct net_device *dev,
 564                             struct ethtool_drvinfo *info)
 565 {
 566         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
 567         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
 568 }
 569
 570 static const struct ethtool_ops vrf_ethtool_ops = {
 571         .get_drvinfo    = vrf_get_drvinfo,
 572 };
 573
 574 static void vrf_setup(struct net_device *dev)
 575 {
 576         ether_setup(dev);
 577
 578         /* Initialize the device structure. */
 579         dev->netdev_ops = &vrf_netdev_ops;
 580         dev->l3mdev_ops = &vrf_l3mdev_ops;
 581         dev->ethtool_ops = &vrf_ethtool_ops;
 582         dev->destructor = free_netdev;
 583
 584         /* Fill in device structure with ethernet-generic values. */
 585         eth_hw_addr_random(dev);
 586
 587         /* don't acquire vrf device's netif_tx_lock when transmitting */
 588         dev->features |= NETIF_F_LLTX;
 589
 590         /* don't allow vrf devices to change network namespaces. */
 591         dev->features |= NETIF_F_NETNS_LOCAL;
 592 }
 593
 594 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
 595 {
 596         if (tb[IFLA_ADDRESS]) {
 597                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 598                         return -EINVAL;
 599                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 600                         return -EADDRNOTAVAIL;
 601         }
 602         return 0;
 603 }
 604
 605 static void vrf_dellink(struct net_device *dev, struct list_head *head)
 606 {
 607         unregister_netdevice_queue(dev, head);
 608 }
 609
 610 static int vrf_newlink(struct net *src_net, struct net_device *dev,
 611                        struct nlattr *tb[], struct nlattr *data[])
 612 {
 613         struct net_vrf *vrf = netdev_priv(dev);
 614         int err;
 615
 616         if (!data || !data[IFLA_VRF_TABLE])
 617                 return -EINVAL;
 618
 619         vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
 620
 621         dev->priv_flags |= IFF_L3MDEV_MASTER;
 622
 623         err = register_netdevice(dev);
 624         if (err < 0)
 625                 goto out_fail;
 626
 627         return 0;
 628
 629 out_fail:
 630         free_netdev(dev);
 631         return err;
 632 }
 633
 634 static size_t vrf_nl_getsize(const struct net_device *dev)
 635 {
 636         return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
 637 }
 638
 639 static int vrf_fillinfo(struct sk_buff *skb,
 640                         const struct net_device *dev)
 641 {
 642         struct net_vrf *vrf = netdev_priv(dev);
 643
 644         return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
 645 }
 646
 647 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
 648         [IFLA_VRF_TABLE] = { .type = NLA_U32 },
 649 };
 650
 651 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
 652         .kind           = DRV_NAME,
 653         .priv_size      = sizeof(struct net_vrf),
 654
 655         .get_size       = vrf_nl_getsize,
 656         .policy         = vrf_nl_policy,
 657         .validate       = vrf_validate,
 658         .fill_info      = vrf_fillinfo,
 659
 660         .newlink        = vrf_newlink,
 661         .dellink        = vrf_dellink,
 662         .setup          = vrf_setup,
 663         .maxtype        = IFLA_VRF_MAX,
 664 };
 665
 666 static int vrf_device_event(struct notifier_block *unused,
 667                             unsigned long event, void *ptr)
 668 {
 669         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 670
 671         /* only care about unregister events to drop slave references */
 672         if (event == NETDEV_UNREGISTER) {
 673                 struct net_device *vrf_dev;
 674
 675                 if (!vrf_is_slave(dev) || netif_is_l3_master(dev))
 676                         goto out;
 677
 678                 vrf_dev = netdev_master_upper_dev_get(dev);
 679                 vrf_del_slave(vrf_dev, dev);
 680         }
 681 out:
 682         return NOTIFY_DONE;
 683 }
 684
 685 static struct notifier_block vrf_notifier_block __read_mostly = {
 686         .notifier_call = vrf_device_event,
 687 };
 688
 689 static int __init vrf_init_module(void)
 690 {
 691         int rc;
 692
 693         vrf_dst_ops.kmem_cachep =
 694                 kmem_cache_create("vrf_ip_dst_cache",
 695                                   sizeof(struct rtable), 0,
 696                                   SLAB_HWCACHE_ALIGN,
 697                                   NULL);
 698
 699         if (!vrf_dst_ops.kmem_cachep)
 700                 return -ENOMEM;
 701
 702         register_netdevice_notifier(&vrf_notifier_block);
 703
 704         rc = rtnl_link_register(&vrf_link_ops);
 705         if (rc < 0)
 706                 goto error;
 707
 708         return 0;
 709
 710 error:
 711         unregister_netdevice_notifier(&vrf_notifier_block);
 712         kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
 713         return rc;
 714 }
 715
 716 static void __exit vrf_cleanup_module(void)
 717 {
 718         rtnl_link_unregister(&vrf_link_ops);
 719         unregister_netdevice_notifier(&vrf_notifier_block);
 720         kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
 721 }
 722
 723 module_init(vrf_init_module);
 724 module_exit(vrf_cleanup_module);
 725 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
 726 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
 727 MODULE_LICENSE("GPL");
 728 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
 729 MODULE_VERSION(DRV_VERSION);