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 RT_FL_TOS(oldflp4) \
  40         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
  41
  42 #define DRV_NAME        "vrf"
  43 #define DRV_VERSION     "1.0"
  44
  45 #define vrf_master_get_rcu(dev) \
  46         ((struct net_device *)rcu_dereference(dev->rx_handler_data))
  47
  48 struct slave {
  49         struct list_head        list;
  50         struct net_device       *dev;
  51 };
  52
  53 struct slave_queue {
  54         struct list_head        all_slaves;
  55 };
  56
  57 struct net_vrf {
  58         struct slave_queue      queue;
  59         struct rtable           *rth;
  60         u32                     tb_id;
  61 };
  62
  63 struct pcpu_dstats {
  64         u64                     tx_pkts;
  65         u64                     tx_bytes;
  66         u64                     tx_drps;
  67         u64                     rx_pkts;
  68         u64                     rx_bytes;
  69         struct u64_stats_sync   syncp;
  70 };
  71
  72 static struct dst_entry *vrf_ip_check(struct dst_entry *dst, u32 cookie)
  73 {
  74         return dst;
  75 }
  76
  77 static int vrf_ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
  78 {
  79         return ip_local_out(net, sk, skb);
  80 }
  81
  82 static unsigned int vrf_v4_mtu(const struct dst_entry *dst)
  83 {
  84         /* TO-DO: return max ethernet size? */
  85         return dst->dev->mtu;
  86 }
  87
  88 static void vrf_dst_destroy(struct dst_entry *dst)
  89 {
  90         /* our dst lives forever - or until the device is closed */
  91 }
  92
  93 static unsigned int vrf_default_advmss(const struct dst_entry *dst)
  94 {
  95         return 65535 - 40;
  96 }
  97
  98 static struct dst_ops vrf_dst_ops = {
  99         .family         = AF_INET,
 100         .local_out      = vrf_ip_local_out,
 101         .check          = vrf_ip_check,
 102         .mtu            = vrf_v4_mtu,
 103         .destroy        = vrf_dst_destroy,
 104         .default_advmss = vrf_default_advmss,
 105 };
 106
 107 static bool is_ip_rx_frame(struct sk_buff *skb)
 108 {
 109         switch (skb->protocol) {
 110         case htons(ETH_P_IP):
 111         case htons(ETH_P_IPV6):
 112                 return true;
 113         }
 114         return false;
 115 }
 116
 117 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
 118 {
 119         vrf_dev->stats.tx_errors++;
 120         kfree_skb(skb);
 121 }
 122
 123 /* note: already called with rcu_read_lock */
 124 static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb)
 125 {
 126         struct sk_buff *skb = *pskb;
 127
 128         if (is_ip_rx_frame(skb)) {
 129                 struct net_device *dev = vrf_master_get_rcu(skb->dev);
 130                 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
 131
 132                 u64_stats_update_begin(&dstats->syncp);
 133                 dstats->rx_pkts++;
 134                 dstats->rx_bytes += skb->len;
 135                 u64_stats_update_end(&dstats->syncp);
 136
 137                 skb->dev = dev;
 138
 139                 return RX_HANDLER_ANOTHER;
 140         }
 141         return RX_HANDLER_PASS;
 142 }
 143
 144 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
 145                                                  struct rtnl_link_stats64 *stats)
 146 {
 147         int i;
 148
 149         for_each_possible_cpu(i) {
 150                 const struct pcpu_dstats *dstats;
 151                 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
 152                 unsigned int start;
 153
 154                 dstats = per_cpu_ptr(dev->dstats, i);
 155                 do {
 156                         start = u64_stats_fetch_begin_irq(&dstats->syncp);
 157                         tbytes = dstats->tx_bytes;
 158                         tpkts = dstats->tx_pkts;
 159                         tdrops = dstats->tx_drps;
 160                         rbytes = dstats->rx_bytes;
 161                         rpkts = dstats->rx_pkts;
 162                 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
 163                 stats->tx_bytes += tbytes;
 164                 stats->tx_packets += tpkts;
 165                 stats->tx_dropped += tdrops;
 166                 stats->rx_bytes += rbytes;
 167                 stats->rx_packets += rpkts;
 168         }
 169         return stats;
 170 }
 171
 172 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
 173                                            struct net_device *dev)
 174 {
 175         vrf_tx_error(dev, skb);
 176         return NET_XMIT_DROP;
 177 }
 178
 179 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
 180                             struct net_device *vrf_dev)
 181 {
 182         struct rtable *rt;
 183         int err = 1;
 184
 185         rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
 186         if (IS_ERR(rt))
 187                 goto out;
 188
 189         /* TO-DO: what about broadcast ? */
 190         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 191                 ip_rt_put(rt);
 192                 goto out;
 193         }
 194
 195         skb_dst_drop(skb);
 196         skb_dst_set(skb, &rt->dst);
 197         err = 0;
 198 out:
 199         return err;
 200 }
 201
 202 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
 203                                            struct net_device *vrf_dev)
 204 {
 205         struct iphdr *ip4h = ip_hdr(skb);
 206         int ret = NET_XMIT_DROP;
 207         struct flowi4 fl4 = {
 208                 /* needed to match OIF rule */
 209                 .flowi4_oif = vrf_dev->ifindex,
 210                 .flowi4_iif = LOOPBACK_IFINDEX,
 211                 .flowi4_tos = RT_TOS(ip4h->tos),
 212                 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
 213                                 FLOWI_FLAG_SKIP_NH_OIF,
 214                 .daddr = ip4h->daddr,
 215         };
 216
 217         if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
 218                 goto err;
 219
 220         if (!ip4h->saddr) {
 221                 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
 222                                                RT_SCOPE_LINK);
 223         }
 224
 225         ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 226         if (unlikely(net_xmit_eval(ret)))
 227                 vrf_dev->stats.tx_errors++;
 228         else
 229                 ret = NET_XMIT_SUCCESS;
 230
 231 out:
 232         return ret;
 233 err:
 234         vrf_tx_error(vrf_dev, skb);
 235         goto out;
 236 }
 237
 238 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
 239 {
 240         /* strip the ethernet header added for pass through VRF device */
 241         __skb_pull(skb, skb_network_offset(skb));
 242
 243         switch (skb->protocol) {
 244         case htons(ETH_P_IP):
 245                 return vrf_process_v4_outbound(skb, dev);
 246         case htons(ETH_P_IPV6):
 247                 return vrf_process_v6_outbound(skb, dev);
 248         default:
 249                 vrf_tx_error(dev, skb);
 250                 return NET_XMIT_DROP;
 251         }
 252 }
 253
 254 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
 255 {
 256         netdev_tx_t ret = is_ip_tx_frame(skb, dev);
 257
 258         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 259                 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
 260
 261                 u64_stats_update_begin(&dstats->syncp);
 262                 dstats->tx_pkts++;
 263                 dstats->tx_bytes += skb->len;
 264                 u64_stats_update_end(&dstats->syncp);
 265         } else {
 266                 this_cpu_inc(dev->dstats->tx_drps);
 267         }
 268
 269         return ret;
 270 }
 271
 272 /* modelled after ip_finish_output2 */
 273 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 274 {
 275         struct dst_entry *dst = skb_dst(skb);
 276         struct rtable *rt = (struct rtable *)dst;
 277         struct net_device *dev = dst->dev;
 278         unsigned int hh_len = LL_RESERVED_SPACE(dev);
 279         struct neighbour *neigh;
 280         u32 nexthop;
 281         int ret = -EINVAL;
 282
 283         /* Be paranoid, rather than too clever. */
 284         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
 285                 struct sk_buff *skb2;
 286
 287                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
 288                 if (!skb2) {
 289                         ret = -ENOMEM;
 290                         goto err;
 291                 }
 292                 if (skb->sk)
 293                         skb_set_owner_w(skb2, skb->sk);
 294
 295                 consume_skb(skb);
 296                 skb = skb2;
 297         }
 298
 299         rcu_read_lock_bh();
 300
 301         nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
 302         neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
 303         if (unlikely(!neigh))
 304                 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
 305         if (!IS_ERR(neigh))
 306                 ret = dst_neigh_output(dst, neigh, skb);
 307
 308         rcu_read_unlock_bh();
 309 err:
 310         if (unlikely(ret < 0))
 311                 vrf_tx_error(skb->dev, skb);
 312         return ret;
 313 }
 314
 315 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 316 {
 317         struct net_device *dev = skb_dst(skb)->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->priv_flags |= IFF_L3MDEV_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) || netif_is_l3_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->priv_flags &= ~IFF_L3MDEV_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 /* called under rcu_read_lock */
 559 static void vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4)
 560 {
 561         struct fib_result res = { .tclassid = 0 };
 562         struct net *net = dev_net(dev);
 563         u32 orig_tos = fl4->flowi4_tos;
 564         u8 flags = fl4->flowi4_flags;
 565         u8 scope = fl4->flowi4_scope;
 566         u8 tos = RT_FL_TOS(fl4);
 567
 568         if (unlikely(!fl4->daddr))
 569                 return;
 570
 571         fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF;
 572         fl4->flowi4_iif = LOOPBACK_IFINDEX;
 573         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
 574         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
 575                              RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
 576
 577         if (!fib_lookup(net, fl4, &res, 0)) {
 578                 if (res.type == RTN_LOCAL)
 579                         fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr;
 580                 else
 581                         fib_select_path(net, &res, fl4, -1);
 582         }
 583
 584         fl4->flowi4_flags = flags;
 585         fl4->flowi4_tos = orig_tos;
 586         fl4->flowi4_scope = scope;
 587 }
 588
 589 static const struct l3mdev_ops vrf_l3mdev_ops = {
 590         .l3mdev_fib_table       = vrf_fib_table,
 591         .l3mdev_get_rtable      = vrf_get_rtable,
 592         .l3mdev_get_saddr       = vrf_get_saddr,
 593 };
 594
 595 static void vrf_get_drvinfo(struct net_device *dev,
 596                             struct ethtool_drvinfo *info)
 597 {
 598         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
 599         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
 600 }
 601
 602 static const struct ethtool_ops vrf_ethtool_ops = {
 603         .get_drvinfo    = vrf_get_drvinfo,
 604 };
 605
 606 static void vrf_setup(struct net_device *dev)
 607 {
 608         ether_setup(dev);
 609
 610         /* Initialize the device structure. */
 611         dev->netdev_ops = &vrf_netdev_ops;
 612         dev->l3mdev_ops = &vrf_l3mdev_ops;
 613         dev->ethtool_ops = &vrf_ethtool_ops;
 614         dev->destructor = free_netdev;
 615
 616         /* Fill in device structure with ethernet-generic values. */
 617         eth_hw_addr_random(dev);
 618
 619         /* don't acquire vrf device's netif_tx_lock when transmitting */
 620         dev->features |= NETIF_F_LLTX;
 621
 622         /* don't allow vrf devices to change network namespaces. */
 623         dev->features |= NETIF_F_NETNS_LOCAL;
 624 }
 625
 626 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
 627 {
 628         if (tb[IFLA_ADDRESS]) {
 629                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 630                         return -EINVAL;
 631                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 632                         return -EADDRNOTAVAIL;
 633         }
 634         return 0;
 635 }
 636
 637 static void vrf_dellink(struct net_device *dev, struct list_head *head)
 638 {
 639         unregister_netdevice_queue(dev, head);
 640 }
 641
 642 static int vrf_newlink(struct net *src_net, struct net_device *dev,
 643                        struct nlattr *tb[], struct nlattr *data[])
 644 {
 645         struct net_vrf *vrf = netdev_priv(dev);
 646         int err;
 647
 648         if (!data || !data[IFLA_VRF_TABLE])
 649                 return -EINVAL;
 650
 651         vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
 652
 653         dev->priv_flags |= IFF_L3MDEV_MASTER;
 654
 655         err = register_netdevice(dev);
 656         if (err < 0)
 657                 goto out_fail;
 658
 659         return 0;
 660
 661 out_fail:
 662         free_netdev(dev);
 663         return err;
 664 }
 665
 666 static size_t vrf_nl_getsize(const struct net_device *dev)
 667 {
 668         return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
 669 }
 670
 671 static int vrf_fillinfo(struct sk_buff *skb,
 672                         const struct net_device *dev)
 673 {
 674         struct net_vrf *vrf = netdev_priv(dev);
 675
 676         return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
 677 }
 678
 679 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
 680         [IFLA_VRF_TABLE] = { .type = NLA_U32 },
 681 };
 682
 683 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
 684         .kind           = DRV_NAME,
 685         .priv_size      = sizeof(struct net_vrf),
 686
 687         .get_size       = vrf_nl_getsize,
 688         .policy         = vrf_nl_policy,
 689         .validate       = vrf_validate,
 690         .fill_info      = vrf_fillinfo,
 691
 692         .newlink        = vrf_newlink,
 693         .dellink        = vrf_dellink,
 694         .setup          = vrf_setup,
 695         .maxtype        = IFLA_VRF_MAX,
 696 };
 697
 698 static int vrf_device_event(struct notifier_block *unused,
 699                             unsigned long event, void *ptr)
 700 {
 701         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 702
 703         /* only care about unregister events to drop slave references */
 704         if (event == NETDEV_UNREGISTER) {
 705                 struct net_device *vrf_dev;
 706
 707                 if (!netif_is_l3_slave(dev))
 708                         goto out;
 709
 710                 vrf_dev = netdev_master_upper_dev_get(dev);
 711                 vrf_del_slave(vrf_dev, dev);
 712         }
 713 out:
 714         return NOTIFY_DONE;
 715 }
 716
 717 static struct notifier_block vrf_notifier_block __read_mostly = {
 718         .notifier_call = vrf_device_event,
 719 };
 720
 721 static int __init vrf_init_module(void)
 722 {
 723         int rc;
 724
 725         vrf_dst_ops.kmem_cachep =
 726                 kmem_cache_create("vrf_ip_dst_cache",
 727                                   sizeof(struct rtable), 0,
 728                                   SLAB_HWCACHE_ALIGN,
 729                                   NULL);
 730
 731         if (!vrf_dst_ops.kmem_cachep)
 732                 return -ENOMEM;
 733
 734         register_netdevice_notifier(&vrf_notifier_block);
 735
 736         rc = rtnl_link_register(&vrf_link_ops);
 737         if (rc < 0)
 738                 goto error;
 739
 740         return 0;
 741
 742 error:
 743         unregister_netdevice_notifier(&vrf_notifier_block);
 744         kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
 745         return rc;
 746 }
 747
 748 static void __exit vrf_cleanup_module(void)
 749 {
 750         rtnl_link_unregister(&vrf_link_ops);
 751         unregister_netdevice_notifier(&vrf_notifier_block);
 752         kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
 753 }
 754
 755 module_init(vrf_init_module);
 756 module_exit(vrf_cleanup_module);
 757 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
 758 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
 759 MODULE_LICENSE("GPL");
 760 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
 761 MODULE_VERSION(DRV_VERSION);