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_fib.h>
  34 #include <net/ip6_route.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 net_vrf {
  49         struct rtable           *rth;
  50         struct rt6_info         *rt6;
  51         u32                     tb_id;
  52 };
  53
  54 struct pcpu_dstats {
  55         u64                     tx_pkts;
  56         u64                     tx_bytes;
  57         u64                     tx_drps;
  58         u64                     rx_pkts;
  59         u64                     rx_bytes;
  60         struct u64_stats_sync   syncp;
  61 };
  62
  63 static struct dst_entry *vrf_ip_check(struct dst_entry *dst, u32 cookie)
  64 {
  65         return dst;
  66 }
  67
  68 static int vrf_ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
  69 {
  70         return ip_local_out(net, sk, skb);
  71 }
  72
  73 static unsigned int vrf_v4_mtu(const struct dst_entry *dst)
  74 {
  75         /* TO-DO: return max ethernet size? */
  76         return dst->dev->mtu;
  77 }
  78
  79 static void vrf_dst_destroy(struct dst_entry *dst)
  80 {
  81         /* our dst lives forever - or until the device is closed */
  82 }
  83
  84 static unsigned int vrf_default_advmss(const struct dst_entry *dst)
  85 {
  86         return 65535 - 40;
  87 }
  88
  89 static struct dst_ops vrf_dst_ops = {
  90         .family         = AF_INET,
  91         .local_out      = vrf_ip_local_out,
  92         .check          = vrf_ip_check,
  93         .mtu            = vrf_v4_mtu,
  94         .destroy        = vrf_dst_destroy,
  95         .default_advmss = vrf_default_advmss,
  96 };
  97
  98 /* neighbor handling is done with actual device; do not want
  99  * to flip skb->dev for those ndisc packets. This really fails
 100  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
 101  * a start.
 102  */
 103 #if IS_ENABLED(CONFIG_IPV6)
 104 static bool check_ipv6_frame(const struct sk_buff *skb)
 105 {
 106         const struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb->data;
 107         size_t hlen = sizeof(*ipv6h);
 108         bool rc = true;
 109
 110         if (skb->len < hlen)
 111                 goto out;
 112
 113         if (ipv6h->nexthdr == NEXTHDR_ICMP) {
 114                 const struct icmp6hdr *icmph;
 115
 116                 if (skb->len < hlen + sizeof(*icmph))
 117                         goto out;
 118
 119                 icmph = (struct icmp6hdr *)(skb->data + sizeof(*ipv6h));
 120                 switch (icmph->icmp6_type) {
 121                 case NDISC_ROUTER_SOLICITATION:
 122                 case NDISC_ROUTER_ADVERTISEMENT:
 123                 case NDISC_NEIGHBOUR_SOLICITATION:
 124                 case NDISC_NEIGHBOUR_ADVERTISEMENT:
 125                 case NDISC_REDIRECT:
 126                         rc = false;
 127                         break;
 128                 }
 129         }
 130
 131 out:
 132         return rc;
 133 }
 134 #else
 135 static bool check_ipv6_frame(const struct sk_buff *skb)
 136 {
 137         return false;
 138 }
 139 #endif
 140
 141 static bool is_ip_rx_frame(struct sk_buff *skb)
 142 {
 143         switch (skb->protocol) {
 144         case htons(ETH_P_IP):
 145                 return true;
 146         case htons(ETH_P_IPV6):
 147                 return check_ipv6_frame(skb);
 148         }
 149         return false;
 150 }
 151
 152 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
 153 {
 154         vrf_dev->stats.tx_errors++;
 155         kfree_skb(skb);
 156 }
 157
 158 /* note: already called with rcu_read_lock */
 159 static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb)
 160 {
 161         struct sk_buff *skb = *pskb;
 162
 163         if (is_ip_rx_frame(skb)) {
 164                 struct net_device *dev = vrf_master_get_rcu(skb->dev);
 165                 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
 166
 167                 u64_stats_update_begin(&dstats->syncp);
 168                 dstats->rx_pkts++;
 169                 dstats->rx_bytes += skb->len;
 170                 u64_stats_update_end(&dstats->syncp);
 171
 172                 skb->dev = dev;
 173
 174                 return RX_HANDLER_ANOTHER;
 175         }
 176         return RX_HANDLER_PASS;
 177 }
 178
 179 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
 180                                                  struct rtnl_link_stats64 *stats)
 181 {
 182         int i;
 183
 184         for_each_possible_cpu(i) {
 185                 const struct pcpu_dstats *dstats;
 186                 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
 187                 unsigned int start;
 188
 189                 dstats = per_cpu_ptr(dev->dstats, i);
 190                 do {
 191                         start = u64_stats_fetch_begin_irq(&dstats->syncp);
 192                         tbytes = dstats->tx_bytes;
 193                         tpkts = dstats->tx_pkts;
 194                         tdrops = dstats->tx_drps;
 195                         rbytes = dstats->rx_bytes;
 196                         rpkts = dstats->rx_pkts;
 197                 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
 198                 stats->tx_bytes += tbytes;
 199                 stats->tx_packets += tpkts;
 200                 stats->tx_dropped += tdrops;
 201                 stats->rx_bytes += rbytes;
 202                 stats->rx_packets += rpkts;
 203         }
 204         return stats;
 205 }
 206
 207 #if IS_ENABLED(CONFIG_IPV6)
 208 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
 209                                            struct net_device *dev)
 210 {
 211         const struct ipv6hdr *iph = ipv6_hdr(skb);
 212         struct net *net = dev_net(skb->dev);
 213         struct flowi6 fl6 = {
 214                 /* needed to match OIF rule */
 215                 .flowi6_oif = dev->ifindex,
 216                 .flowi6_iif = LOOPBACK_IFINDEX,
 217                 .daddr = iph->daddr,
 218                 .saddr = iph->saddr,
 219                 .flowlabel = ip6_flowinfo(iph),
 220                 .flowi6_mark = skb->mark,
 221                 .flowi6_proto = iph->nexthdr,
 222                 .flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF,
 223         };
 224         int ret = NET_XMIT_DROP;
 225         struct dst_entry *dst;
 226         struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
 227
 228         dst = ip6_route_output(net, NULL, &fl6);
 229         if (dst == dst_null)
 230                 goto err;
 231
 232         skb_dst_drop(skb);
 233         skb_dst_set(skb, dst);
 234
 235         ret = ip6_local_out(net, skb->sk, skb);
 236         if (unlikely(net_xmit_eval(ret)))
 237                 dev->stats.tx_errors++;
 238         else
 239                 ret = NET_XMIT_SUCCESS;
 240
 241         return ret;
 242 err:
 243         vrf_tx_error(dev, skb);
 244         return NET_XMIT_DROP;
 245 }
 246 #else
 247 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
 248                                            struct net_device *dev)
 249 {
 250         vrf_tx_error(dev, skb);
 251         return NET_XMIT_DROP;
 252 }
 253 #endif
 254
 255 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
 256                             struct net_device *vrf_dev)
 257 {
 258         struct rtable *rt;
 259         int err = 1;
 260
 261         rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
 262         if (IS_ERR(rt))
 263                 goto out;
 264
 265         /* TO-DO: what about broadcast ? */
 266         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 267                 ip_rt_put(rt);
 268                 goto out;
 269         }
 270
 271         skb_dst_drop(skb);
 272         skb_dst_set(skb, &rt->dst);
 273         err = 0;
 274 out:
 275         return err;
 276 }
 277
 278 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
 279                                            struct net_device *vrf_dev)
 280 {
 281         struct iphdr *ip4h = ip_hdr(skb);
 282         int ret = NET_XMIT_DROP;
 283         struct flowi4 fl4 = {
 284                 /* needed to match OIF rule */
 285                 .flowi4_oif = vrf_dev->ifindex,
 286                 .flowi4_iif = LOOPBACK_IFINDEX,
 287                 .flowi4_tos = RT_TOS(ip4h->tos),
 288                 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
 289                                 FLOWI_FLAG_SKIP_NH_OIF,
 290                 .daddr = ip4h->daddr,
 291         };
 292
 293         if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
 294                 goto err;
 295
 296         if (!ip4h->saddr) {
 297                 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
 298                                                RT_SCOPE_LINK);
 299         }
 300
 301         ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 302         if (unlikely(net_xmit_eval(ret)))
 303                 vrf_dev->stats.tx_errors++;
 304         else
 305                 ret = NET_XMIT_SUCCESS;
 306
 307 out:
 308         return ret;
 309 err:
 310         vrf_tx_error(vrf_dev, skb);
 311         goto out;
 312 }
 313
 314 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
 315 {
 316         /* strip the ethernet header added for pass through VRF device */
 317         __skb_pull(skb, skb_network_offset(skb));
 318
 319         switch (skb->protocol) {
 320         case htons(ETH_P_IP):
 321                 return vrf_process_v4_outbound(skb, dev);
 322         case htons(ETH_P_IPV6):
 323                 return vrf_process_v6_outbound(skb, dev);
 324         default:
 325                 vrf_tx_error(dev, skb);
 326                 return NET_XMIT_DROP;
 327         }
 328 }
 329
 330 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
 331 {
 332         netdev_tx_t ret = is_ip_tx_frame(skb, dev);
 333
 334         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 335                 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
 336
 337                 u64_stats_update_begin(&dstats->syncp);
 338                 dstats->tx_pkts++;
 339                 dstats->tx_bytes += skb->len;
 340                 u64_stats_update_end(&dstats->syncp);
 341         } else {
 342                 this_cpu_inc(dev->dstats->tx_drps);
 343         }
 344
 345         return ret;
 346 }
 347
 348 #if IS_ENABLED(CONFIG_IPV6)
 349 static struct dst_entry *vrf_ip6_check(struct dst_entry *dst, u32 cookie)
 350 {
 351         return dst;
 352 }
 353
 354 static struct dst_ops vrf_dst_ops6 = {
 355         .family         = AF_INET6,
 356         .local_out      = ip6_local_out,
 357         .check          = vrf_ip6_check,
 358         .mtu            = vrf_v4_mtu,
 359         .destroy        = vrf_dst_destroy,
 360         .default_advmss = vrf_default_advmss,
 361 };
 362
 363 static int init_dst_ops6_kmem_cachep(void)
 364 {
 365         vrf_dst_ops6.kmem_cachep = kmem_cache_create("vrf_ip6_dst_cache",
 366                                                      sizeof(struct rt6_info),
 367                                                      0,
 368                                                      SLAB_HWCACHE_ALIGN,
 369                                                      NULL);
 370
 371         if (!vrf_dst_ops6.kmem_cachep)
 372                 return -ENOMEM;
 373
 374         return 0;
 375 }
 376
 377 static void free_dst_ops6_kmem_cachep(void)
 378 {
 379         kmem_cache_destroy(vrf_dst_ops6.kmem_cachep);
 380 }
 381
 382 static int vrf_input6(struct sk_buff *skb)
 383 {
 384         skb->dev->stats.rx_errors++;
 385         kfree_skb(skb);
 386         return 0;
 387 }
 388
 389 /* modelled after ip6_finish_output2 */
 390 static int vrf_finish_output6(struct net *net, struct sock *sk,
 391                               struct sk_buff *skb)
 392 {
 393         struct dst_entry *dst = skb_dst(skb);
 394         struct net_device *dev = dst->dev;
 395         struct neighbour *neigh;
 396         struct in6_addr *nexthop;
 397         int ret;
 398
 399         skb->protocol = htons(ETH_P_IPV6);
 400         skb->dev = dev;
 401
 402         rcu_read_lock_bh();
 403         nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
 404         neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
 405         if (unlikely(!neigh))
 406                 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
 407         if (!IS_ERR(neigh)) {
 408                 ret = dst_neigh_output(dst, neigh, skb);
 409                 rcu_read_unlock_bh();
 410                 return ret;
 411         }
 412         rcu_read_unlock_bh();
 413
 414         IP6_INC_STATS(dev_net(dst->dev),
 415                       ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
 416         kfree_skb(skb);
 417         return -EINVAL;
 418 }
 419
 420 /* modelled after ip6_output */
 421 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
 422 {
 423         return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
 424                             net, sk, skb, NULL, skb_dst(skb)->dev,
 425                             vrf_finish_output6,
 426                             !(IP6CB(skb)->flags & IP6SKB_REROUTED));
 427 }
 428
 429 static void vrf_rt6_destroy(struct net_vrf *vrf)
 430 {
 431         dst_destroy(&vrf->rt6->dst);
 432         free_percpu(vrf->rt6->rt6i_pcpu);
 433         vrf->rt6 = NULL;
 434 }
 435
 436 static int vrf_rt6_create(struct net_device *dev)
 437 {
 438         struct net_vrf *vrf = netdev_priv(dev);
 439         struct dst_entry *dst;
 440         struct rt6_info *rt6;
 441         int cpu;
 442         int rc = -ENOMEM;
 443
 444         rt6 = dst_alloc(&vrf_dst_ops6, dev, 0,
 445                         DST_OBSOLETE_NONE,
 446                         (DST_HOST | DST_NOPOLICY | DST_NOXFRM));
 447         if (!rt6)
 448                 goto out;
 449
 450         dst = &rt6->dst;
 451
 452         rt6->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, GFP_KERNEL);
 453         if (!rt6->rt6i_pcpu) {
 454                 dst_destroy(dst);
 455                 goto out;
 456         }
 457         for_each_possible_cpu(cpu) {
 458                 struct rt6_info **p = per_cpu_ptr(rt6->rt6i_pcpu, cpu);
 459                 *p =  NULL;
 460         }
 461
 462         memset(dst + 1, 0, sizeof(*rt6) - sizeof(*dst));
 463
 464         INIT_LIST_HEAD(&rt6->rt6i_siblings);
 465         INIT_LIST_HEAD(&rt6->rt6i_uncached);
 466
 467         rt6->dst.input  = vrf_input6;
 468         rt6->dst.output = vrf_output6;
 469
 470         rt6->rt6i_table = fib6_get_table(dev_net(dev), vrf->tb_id);
 471
 472         atomic_set(&rt6->dst.__refcnt, 2);
 473
 474         vrf->rt6 = rt6;
 475         rc = 0;
 476 out:
 477         return rc;
 478 }
 479 #else
 480 static int init_dst_ops6_kmem_cachep(void)
 481 {
 482         return 0;
 483 }
 484
 485 static void free_dst_ops6_kmem_cachep(void)
 486 {
 487 }
 488
 489 static void vrf_rt6_destroy(struct net_vrf *vrf)
 490 {
 491 }
 492
 493 static int vrf_rt6_create(struct net_device *dev)
 494 {
 495         return 0;
 496 }
 497 #endif
 498
 499 /* modelled after ip_finish_output2 */
 500 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 501 {
 502         struct dst_entry *dst = skb_dst(skb);
 503         struct rtable *rt = (struct rtable *)dst;
 504         struct net_device *dev = dst->dev;
 505         unsigned int hh_len = LL_RESERVED_SPACE(dev);
 506         struct neighbour *neigh;
 507         u32 nexthop;
 508         int ret = -EINVAL;
 509
 510         /* Be paranoid, rather than too clever. */
 511         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
 512                 struct sk_buff *skb2;
 513
 514                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
 515                 if (!skb2) {
 516                         ret = -ENOMEM;
 517                         goto err;
 518                 }
 519                 if (skb->sk)
 520                         skb_set_owner_w(skb2, skb->sk);
 521
 522                 consume_skb(skb);
 523                 skb = skb2;
 524         }
 525
 526         rcu_read_lock_bh();
 527
 528         nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
 529         neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
 530         if (unlikely(!neigh))
 531                 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
 532         if (!IS_ERR(neigh))
 533                 ret = dst_neigh_output(dst, neigh, skb);
 534
 535         rcu_read_unlock_bh();
 536 err:
 537         if (unlikely(ret < 0))
 538                 vrf_tx_error(skb->dev, skb);
 539         return ret;
 540 }
 541
 542 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 543 {
 544         struct net_device *dev = skb_dst(skb)->dev;
 545
 546         IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
 547
 548         skb->dev = dev;
 549         skb->protocol = htons(ETH_P_IP);
 550
 551         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
 552                             net, sk, skb, NULL, dev,
 553                             vrf_finish_output,
 554                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 555 }
 556
 557 static void vrf_rtable_destroy(struct net_vrf *vrf)
 558 {
 559         struct dst_entry *dst = (struct dst_entry *)vrf->rth;
 560
 561         dst_destroy(dst);
 562         vrf->rth = NULL;
 563 }
 564
 565 static struct rtable *vrf_rtable_create(struct net_device *dev)
 566 {
 567         struct net_vrf *vrf = netdev_priv(dev);
 568         struct rtable *rth;
 569
 570         rth = dst_alloc(&vrf_dst_ops, dev, 2,
 571                         DST_OBSOLETE_NONE,
 572                         (DST_HOST | DST_NOPOLICY | DST_NOXFRM));
 573         if (rth) {
 574                 rth->dst.output = vrf_output;
 575                 rth->rt_genid   = rt_genid_ipv4(dev_net(dev));
 576                 rth->rt_flags   = 0;
 577                 rth->rt_type    = RTN_UNICAST;
 578                 rth->rt_is_input = 0;
 579                 rth->rt_iif     = 0;
 580                 rth->rt_pmtu    = 0;
 581                 rth->rt_gateway = 0;
 582                 rth->rt_uses_gateway = 0;
 583                 rth->rt_table_id = vrf->tb_id;
 584                 INIT_LIST_HEAD(&rth->rt_uncached);
 585                 rth->rt_uncached_list = NULL;
 586         }
 587
 588         return rth;
 589 }
 590
 591 /**************************** device handling ********************/
 592
 593 /* cycle interface to flush neighbor cache and move routes across tables */
 594 static void cycle_netdev(struct net_device *dev)
 595 {
 596         unsigned int flags = dev->flags;
 597         int ret;
 598
 599         if (!netif_running(dev))
 600                 return;
 601
 602         ret = dev_change_flags(dev, flags & ~IFF_UP);
 603         if (ret >= 0)
 604                 ret = dev_change_flags(dev, flags);
 605
 606         if (ret < 0) {
 607                 netdev_err(dev,
 608                            "Failed to cycle device %s; route tables might be wrong!\n",
 609                            dev->name);
 610         }
 611 }
 612
 613 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
 614 {
 615         int ret;
 616
 617         /* register the packet handler for slave ports */
 618         ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev);
 619         if (ret) {
 620                 netdev_err(port_dev,
 621                            "Device %s failed to register rx_handler\n",
 622                            port_dev->name);
 623                 goto out_fail;
 624         }
 625
 626         ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
 627         if (ret < 0)
 628                 goto out_unregister;
 629
 630         port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
 631         cycle_netdev(port_dev);
 632
 633         return 0;
 634
 635 out_unregister:
 636         netdev_rx_handler_unregister(port_dev);
 637 out_fail:
 638         return ret;
 639 }
 640
 641 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
 642 {
 643         if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
 644                 return -EINVAL;
 645
 646         return do_vrf_add_slave(dev, port_dev);
 647 }
 648
 649 /* inverse of do_vrf_add_slave */
 650 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
 651 {
 652         netdev_upper_dev_unlink(port_dev, dev);
 653         port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
 654
 655         netdev_rx_handler_unregister(port_dev);
 656
 657         cycle_netdev(port_dev);
 658
 659         return 0;
 660 }
 661
 662 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
 663 {
 664         return do_vrf_del_slave(dev, port_dev);
 665 }
 666
 667 static void vrf_dev_uninit(struct net_device *dev)
 668 {
 669         struct net_vrf *vrf = netdev_priv(dev);
 670         struct net_device *port_dev;
 671         struct list_head *iter;
 672
 673         vrf_rtable_destroy(vrf);
 674         vrf_rt6_destroy(vrf);
 675
 676         netdev_for_each_lower_dev(dev, port_dev, iter)
 677                 vrf_del_slave(dev, port_dev);
 678
 679         free_percpu(dev->dstats);
 680         dev->dstats = NULL;
 681 }
 682
 683 static int vrf_dev_init(struct net_device *dev)
 684 {
 685         struct net_vrf *vrf = netdev_priv(dev);
 686
 687         dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
 688         if (!dev->dstats)
 689                 goto out_nomem;
 690
 691         /* create the default dst which points back to us */
 692         vrf->rth = vrf_rtable_create(dev);
 693         if (!vrf->rth)
 694                 goto out_stats;
 695
 696         if (vrf_rt6_create(dev) != 0)
 697                 goto out_rth;
 698
 699         dev->flags = IFF_MASTER | IFF_NOARP;
 700
 701         return 0;
 702
 703 out_rth:
 704         vrf_rtable_destroy(vrf);
 705 out_stats:
 706         free_percpu(dev->dstats);
 707         dev->dstats = NULL;
 708 out_nomem:
 709         return -ENOMEM;
 710 }
 711
 712 static const struct net_device_ops vrf_netdev_ops = {
 713         .ndo_init               = vrf_dev_init,
 714         .ndo_uninit             = vrf_dev_uninit,
 715         .ndo_start_xmit         = vrf_xmit,
 716         .ndo_get_stats64        = vrf_get_stats64,
 717         .ndo_add_slave          = vrf_add_slave,
 718         .ndo_del_slave          = vrf_del_slave,
 719 };
 720
 721 static u32 vrf_fib_table(const struct net_device *dev)
 722 {
 723         struct net_vrf *vrf = netdev_priv(dev);
 724
 725         return vrf->tb_id;
 726 }
 727
 728 static struct rtable *vrf_get_rtable(const struct net_device *dev,
 729                                      const struct flowi4 *fl4)
 730 {
 731         struct rtable *rth = NULL;
 732
 733         if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
 734                 struct net_vrf *vrf = netdev_priv(dev);
 735
 736                 rth = vrf->rth;
 737                 atomic_inc(&rth->dst.__refcnt);
 738         }
 739
 740         return rth;
 741 }
 742
 743 /* called under rcu_read_lock */
 744 static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4)
 745 {
 746         struct fib_result res = { .tclassid = 0 };
 747         struct net *net = dev_net(dev);
 748         u32 orig_tos = fl4->flowi4_tos;
 749         u8 flags = fl4->flowi4_flags;
 750         u8 scope = fl4->flowi4_scope;
 751         u8 tos = RT_FL_TOS(fl4);
 752         int rc;
 753
 754         if (unlikely(!fl4->daddr))
 755                 return 0;
 756
 757         fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF;
 758         fl4->flowi4_iif = LOOPBACK_IFINDEX;
 759         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
 760         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
 761                              RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
 762
 763         rc = fib_lookup(net, fl4, &res, 0);
 764         if (!rc) {
 765                 if (res.type == RTN_LOCAL)
 766                         fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr;
 767                 else
 768                         fib_select_path(net, &res, fl4, -1);
 769         }
 770
 771         fl4->flowi4_flags = flags;
 772         fl4->flowi4_tos = orig_tos;
 773         fl4->flowi4_scope = scope;
 774
 775         return rc;
 776 }
 777
 778 #if IS_ENABLED(CONFIG_IPV6)
 779 static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
 780                                          const struct flowi6 *fl6)
 781 {
 782         struct rt6_info *rt = NULL;
 783
 784         if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) {
 785                 struct net_vrf *vrf = netdev_priv(dev);
 786
 787                 rt = vrf->rt6;
 788                 atomic_inc(&rt->dst.__refcnt);
 789         }
 790
 791         return (struct dst_entry *)rt;
 792 }
 793 #endif
 794
 795 static const struct l3mdev_ops vrf_l3mdev_ops = {
 796         .l3mdev_fib_table       = vrf_fib_table,
 797         .l3mdev_get_rtable      = vrf_get_rtable,
 798         .l3mdev_get_saddr       = vrf_get_saddr,
 799 #if IS_ENABLED(CONFIG_IPV6)
 800         .l3mdev_get_rt6_dst     = vrf_get_rt6_dst,
 801 #endif
 802 };
 803
 804 static void vrf_get_drvinfo(struct net_device *dev,
 805                             struct ethtool_drvinfo *info)
 806 {
 807         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
 808         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
 809 }
 810
 811 static const struct ethtool_ops vrf_ethtool_ops = {
 812         .get_drvinfo    = vrf_get_drvinfo,
 813 };
 814
 815 static void vrf_setup(struct net_device *dev)
 816 {
 817         ether_setup(dev);
 818
 819         /* Initialize the device structure. */
 820         dev->netdev_ops = &vrf_netdev_ops;
 821         dev->l3mdev_ops = &vrf_l3mdev_ops;
 822         dev->ethtool_ops = &vrf_ethtool_ops;
 823         dev->destructor = free_netdev;
 824
 825         /* Fill in device structure with ethernet-generic values. */
 826         eth_hw_addr_random(dev);
 827
 828         /* don't acquire vrf device's netif_tx_lock when transmitting */
 829         dev->features |= NETIF_F_LLTX;
 830
 831         /* don't allow vrf devices to change network namespaces. */
 832         dev->features |= NETIF_F_NETNS_LOCAL;
 833 }
 834
 835 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
 836 {
 837         if (tb[IFLA_ADDRESS]) {
 838                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 839                         return -EINVAL;
 840                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 841                         return -EADDRNOTAVAIL;
 842         }
 843         return 0;
 844 }
 845
 846 static void vrf_dellink(struct net_device *dev, struct list_head *head)
 847 {
 848         unregister_netdevice_queue(dev, head);
 849 }
 850
 851 static int vrf_newlink(struct net *src_net, struct net_device *dev,
 852                        struct nlattr *tb[], struct nlattr *data[])
 853 {
 854         struct net_vrf *vrf = netdev_priv(dev);
 855
 856         if (!data || !data[IFLA_VRF_TABLE])
 857                 return -EINVAL;
 858
 859         vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
 860
 861         dev->priv_flags |= IFF_L3MDEV_MASTER;
 862
 863         return register_netdevice(dev);
 864 }
 865
 866 static size_t vrf_nl_getsize(const struct net_device *dev)
 867 {
 868         return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
 869 }
 870
 871 static int vrf_fillinfo(struct sk_buff *skb,
 872                         const struct net_device *dev)
 873 {
 874         struct net_vrf *vrf = netdev_priv(dev);
 875
 876         return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
 877 }
 878
 879 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
 880                                  const struct net_device *slave_dev)
 881 {
 882         return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
 883 }
 884
 885 static int vrf_fill_slave_info(struct sk_buff *skb,
 886                                const struct net_device *vrf_dev,
 887                                const struct net_device *slave_dev)
 888 {
 889         struct net_vrf *vrf = netdev_priv(vrf_dev);
 890
 891         if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
 892                 return -EMSGSIZE;
 893
 894         return 0;
 895 }
 896
 897 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
 898         [IFLA_VRF_TABLE] = { .type = NLA_U32 },
 899 };
 900
 901 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
 902         .kind           = DRV_NAME,
 903         .priv_size      = sizeof(struct net_vrf),
 904
 905         .get_size       = vrf_nl_getsize,
 906         .policy         = vrf_nl_policy,
 907         .validate       = vrf_validate,
 908         .fill_info      = vrf_fillinfo,
 909
 910         .get_slave_size  = vrf_get_slave_size,
 911         .fill_slave_info = vrf_fill_slave_info,
 912
 913         .newlink        = vrf_newlink,
 914         .dellink        = vrf_dellink,
 915         .setup          = vrf_setup,
 916         .maxtype        = IFLA_VRF_MAX,
 917 };
 918
 919 static int vrf_device_event(struct notifier_block *unused,
 920                             unsigned long event, void *ptr)
 921 {
 922         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 923
 924         /* only care about unregister events to drop slave references */
 925         if (event == NETDEV_UNREGISTER) {
 926                 struct net_device *vrf_dev;
 927
 928                 if (!netif_is_l3_slave(dev))
 929                         goto out;
 930
 931                 vrf_dev = netdev_master_upper_dev_get(dev);
 932                 vrf_del_slave(vrf_dev, dev);
 933         }
 934 out:
 935         return NOTIFY_DONE;
 936 }
 937
 938 static struct notifier_block vrf_notifier_block __read_mostly = {
 939         .notifier_call = vrf_device_event,
 940 };
 941
 942 static int __init vrf_init_module(void)
 943 {
 944         int rc;
 945
 946         vrf_dst_ops.kmem_cachep =
 947                 kmem_cache_create("vrf_ip_dst_cache",
 948                                   sizeof(struct rtable), 0,
 949                                   SLAB_HWCACHE_ALIGN,
 950                                   NULL);
 951
 952         if (!vrf_dst_ops.kmem_cachep)
 953                 return -ENOMEM;
 954
 955         rc = init_dst_ops6_kmem_cachep();
 956         if (rc != 0)
 957                 goto error2;
 958
 959         register_netdevice_notifier(&vrf_notifier_block);
 960
 961         rc = rtnl_link_register(&vrf_link_ops);
 962         if (rc < 0)
 963                 goto error;
 964
 965         return 0;
 966
 967 error:
 968         unregister_netdevice_notifier(&vrf_notifier_block);
 969         free_dst_ops6_kmem_cachep();
 970 error2:
 971         kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
 972         return rc;
 973 }
 974
 975 static void __exit vrf_cleanup_module(void)
 976 {
 977         rtnl_link_unregister(&vrf_link_ops);
 978         unregister_netdevice_notifier(&vrf_notifier_block);
 979         kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
 980         free_dst_ops6_kmem_cachep();
 981 }
 982
 983 module_init(vrf_init_module);
 984 module_exit(vrf_cleanup_module);
 985 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
 986 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
 987 MODULE_LICENSE("GPL");
 988 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
 989 MODULE_VERSION(DRV_VERSION);