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 struct net_vrf {
  46         struct rtable __rcu     *rth;
  47         struct rt6_info __rcu   *rt6;
  48         u32                     tb_id;
  49 };
  50
  51 struct pcpu_dstats {
  52         u64                     tx_pkts;
  53         u64                     tx_bytes;
  54         u64                     tx_drps;
  55         u64                     rx_pkts;
  56         u64                     rx_bytes;
  57         struct u64_stats_sync   syncp;
  58 };
  59
  60 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
  61 {
  62         vrf_dev->stats.tx_errors++;
  63         kfree_skb(skb);
  64 }
  65
  66 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
  67                                                  struct rtnl_link_stats64 *stats)
  68 {
  69         int i;
  70
  71         for_each_possible_cpu(i) {
  72                 const struct pcpu_dstats *dstats;
  73                 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
  74                 unsigned int start;
  75
  76                 dstats = per_cpu_ptr(dev->dstats, i);
  77                 do {
  78                         start = u64_stats_fetch_begin_irq(&dstats->syncp);
  79                         tbytes = dstats->tx_bytes;
  80                         tpkts = dstats->tx_pkts;
  81                         tdrops = dstats->tx_drps;
  82                         rbytes = dstats->rx_bytes;
  83                         rpkts = dstats->rx_pkts;
  84                 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
  85                 stats->tx_bytes += tbytes;
  86                 stats->tx_packets += tpkts;
  87                 stats->tx_dropped += tdrops;
  88                 stats->rx_bytes += rbytes;
  89                 stats->rx_packets += rpkts;
  90         }
  91         return stats;
  92 }
  93
  94 #if IS_ENABLED(CONFIG_IPV6)
  95 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
  96                                            struct net_device *dev)
  97 {
  98         const struct ipv6hdr *iph = ipv6_hdr(skb);
  99         struct net *net = dev_net(skb->dev);
 100         struct flowi6 fl6 = {
 101                 /* needed to match OIF rule */
 102                 .flowi6_oif = dev->ifindex,
 103                 .flowi6_iif = LOOPBACK_IFINDEX,
 104                 .daddr = iph->daddr,
 105                 .saddr = iph->saddr,
 106                 .flowlabel = ip6_flowinfo(iph),
 107                 .flowi6_mark = skb->mark,
 108                 .flowi6_proto = iph->nexthdr,
 109                 .flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF,
 110         };
 111         int ret = NET_XMIT_DROP;
 112         struct dst_entry *dst;
 113         struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
 114
 115         dst = ip6_route_output(net, NULL, &fl6);
 116         if (dst == dst_null)
 117                 goto err;
 118
 119         skb_dst_drop(skb);
 120         skb_dst_set(skb, dst);
 121
 122         ret = ip6_local_out(net, skb->sk, skb);
 123         if (unlikely(net_xmit_eval(ret)))
 124                 dev->stats.tx_errors++;
 125         else
 126                 ret = NET_XMIT_SUCCESS;
 127
 128         return ret;
 129 err:
 130         vrf_tx_error(dev, skb);
 131         return NET_XMIT_DROP;
 132 }
 133 #else
 134 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
 135                                            struct net_device *dev)
 136 {
 137         vrf_tx_error(dev, skb);
 138         return NET_XMIT_DROP;
 139 }
 140 #endif
 141
 142 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
 143                             struct net_device *vrf_dev)
 144 {
 145         struct rtable *rt;
 146         int err = 1;
 147
 148         rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
 149         if (IS_ERR(rt))
 150                 goto out;
 151
 152         /* TO-DO: what about broadcast ? */
 153         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 154                 ip_rt_put(rt);
 155                 goto out;
 156         }
 157
 158         skb_dst_drop(skb);
 159         skb_dst_set(skb, &rt->dst);
 160         err = 0;
 161 out:
 162         return err;
 163 }
 164
 165 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
 166                                            struct net_device *vrf_dev)
 167 {
 168         struct iphdr *ip4h = ip_hdr(skb);
 169         int ret = NET_XMIT_DROP;
 170         struct flowi4 fl4 = {
 171                 /* needed to match OIF rule */
 172                 .flowi4_oif = vrf_dev->ifindex,
 173                 .flowi4_iif = LOOPBACK_IFINDEX,
 174                 .flowi4_tos = RT_TOS(ip4h->tos),
 175                 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
 176                                 FLOWI_FLAG_SKIP_NH_OIF,
 177                 .daddr = ip4h->daddr,
 178         };
 179
 180         if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
 181                 goto err;
 182
 183         if (!ip4h->saddr) {
 184                 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
 185                                                RT_SCOPE_LINK);
 186         }
 187
 188         ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 189         if (unlikely(net_xmit_eval(ret)))
 190                 vrf_dev->stats.tx_errors++;
 191         else
 192                 ret = NET_XMIT_SUCCESS;
 193
 194 out:
 195         return ret;
 196 err:
 197         vrf_tx_error(vrf_dev, skb);
 198         goto out;
 199 }
 200
 201 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
 202 {
 203         /* strip the ethernet header added for pass through VRF device */
 204         __skb_pull(skb, skb_network_offset(skb));
 205
 206         switch (skb->protocol) {
 207         case htons(ETH_P_IP):
 208                 return vrf_process_v4_outbound(skb, dev);
 209         case htons(ETH_P_IPV6):
 210                 return vrf_process_v6_outbound(skb, dev);
 211         default:
 212                 vrf_tx_error(dev, skb);
 213                 return NET_XMIT_DROP;
 214         }
 215 }
 216
 217 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
 218 {
 219         netdev_tx_t ret = is_ip_tx_frame(skb, dev);
 220
 221         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 222                 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
 223
 224                 u64_stats_update_begin(&dstats->syncp);
 225                 dstats->tx_pkts++;
 226                 dstats->tx_bytes += skb->len;
 227                 u64_stats_update_end(&dstats->syncp);
 228         } else {
 229                 this_cpu_inc(dev->dstats->tx_drps);
 230         }
 231
 232         return ret;
 233 }
 234
 235 #if IS_ENABLED(CONFIG_IPV6)
 236 /* modelled after ip6_finish_output2 */
 237 static int vrf_finish_output6(struct net *net, struct sock *sk,
 238                               struct sk_buff *skb)
 239 {
 240         struct dst_entry *dst = skb_dst(skb);
 241         struct net_device *dev = dst->dev;
 242         struct neighbour *neigh;
 243         struct in6_addr *nexthop;
 244         int ret;
 245
 246         skb->protocol = htons(ETH_P_IPV6);
 247         skb->dev = dev;
 248
 249         rcu_read_lock_bh();
 250         nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
 251         neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
 252         if (unlikely(!neigh))
 253                 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
 254         if (!IS_ERR(neigh)) {
 255                 ret = dst_neigh_output(dst, neigh, skb);
 256                 rcu_read_unlock_bh();
 257                 return ret;
 258         }
 259         rcu_read_unlock_bh();
 260
 261         IP6_INC_STATS(dev_net(dst->dev),
 262                       ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
 263         kfree_skb(skb);
 264         return -EINVAL;
 265 }
 266
 267 /* modelled after ip6_output */
 268 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
 269 {
 270         return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
 271                             net, sk, skb, NULL, skb_dst(skb)->dev,
 272                             vrf_finish_output6,
 273                             !(IP6CB(skb)->flags & IP6SKB_REROUTED));
 274 }
 275
 276 /* holding rtnl */
 277 static void vrf_rt6_release(struct net_vrf *vrf)
 278 {
 279         struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
 280
 281         rcu_assign_pointer(vrf->rt6, NULL);
 282
 283         if (rt6)
 284                 dst_release(&rt6->dst);
 285 }
 286
 287 static int vrf_rt6_create(struct net_device *dev)
 288 {
 289         struct net_vrf *vrf = netdev_priv(dev);
 290         struct net *net = dev_net(dev);
 291         struct fib6_table *rt6i_table;
 292         struct rt6_info *rt6;
 293         int rc = -ENOMEM;
 294
 295         rt6i_table = fib6_new_table(net, vrf->tb_id);
 296         if (!rt6i_table)
 297                 goto out;
 298
 299         rt6 = ip6_dst_alloc(net, dev,
 300                             DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE);
 301         if (!rt6)
 302                 goto out;
 303
 304         dst_hold(&rt6->dst);
 305
 306         rt6->rt6i_table = rt6i_table;
 307         rt6->dst.output = vrf_output6;
 308         rcu_assign_pointer(vrf->rt6, rt6);
 309
 310         rc = 0;
 311 out:
 312         return rc;
 313 }
 314 #else
 315 static void vrf_rt6_release(struct net_vrf *vrf)
 316 {
 317 }
 318
 319 static int vrf_rt6_create(struct net_device *dev)
 320 {
 321         return 0;
 322 }
 323 #endif
 324
 325 /* modelled after ip_finish_output2 */
 326 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 327 {
 328         struct dst_entry *dst = skb_dst(skb);
 329         struct rtable *rt = (struct rtable *)dst;
 330         struct net_device *dev = dst->dev;
 331         unsigned int hh_len = LL_RESERVED_SPACE(dev);
 332         struct neighbour *neigh;
 333         u32 nexthop;
 334         int ret = -EINVAL;
 335
 336         /* Be paranoid, rather than too clever. */
 337         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
 338                 struct sk_buff *skb2;
 339
 340                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
 341                 if (!skb2) {
 342                         ret = -ENOMEM;
 343                         goto err;
 344                 }
 345                 if (skb->sk)
 346                         skb_set_owner_w(skb2, skb->sk);
 347
 348                 consume_skb(skb);
 349                 skb = skb2;
 350         }
 351
 352         rcu_read_lock_bh();
 353
 354         nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
 355         neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
 356         if (unlikely(!neigh))
 357                 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
 358         if (!IS_ERR(neigh))
 359                 ret = dst_neigh_output(dst, neigh, skb);
 360
 361         rcu_read_unlock_bh();
 362 err:
 363         if (unlikely(ret < 0))
 364                 vrf_tx_error(skb->dev, skb);
 365         return ret;
 366 }
 367
 368 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 369 {
 370         struct net_device *dev = skb_dst(skb)->dev;
 371
 372         IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
 373
 374         skb->dev = dev;
 375         skb->protocol = htons(ETH_P_IP);
 376
 377         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
 378                             net, sk, skb, NULL, dev,
 379                             vrf_finish_output,
 380                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 381 }
 382
 383 /* holding rtnl */
 384 static void vrf_rtable_release(struct net_vrf *vrf)
 385 {
 386         struct rtable *rth = rtnl_dereference(vrf->rth);
 387
 388         rcu_assign_pointer(vrf->rth, NULL);
 389
 390         if (rth)
 391                 dst_release(&rth->dst);
 392 }
 393
 394 static int vrf_rtable_create(struct net_device *dev)
 395 {
 396         struct net_vrf *vrf = netdev_priv(dev);
 397         struct rtable *rth;
 398
 399         if (!fib_new_table(dev_net(dev), vrf->tb_id))
 400                 return -ENOMEM;
 401
 402         rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
 403         if (!rth)
 404                 return -ENOMEM;
 405
 406         rth->dst.output = vrf_output;
 407         rth->rt_table_id = vrf->tb_id;
 408
 409         rcu_assign_pointer(vrf->rth, rth);
 410
 411         return 0;
 412 }
 413
 414 /**************************** device handling ********************/
 415
 416 /* cycle interface to flush neighbor cache and move routes across tables */
 417 static void cycle_netdev(struct net_device *dev)
 418 {
 419         unsigned int flags = dev->flags;
 420         int ret;
 421
 422         if (!netif_running(dev))
 423                 return;
 424
 425         ret = dev_change_flags(dev, flags & ~IFF_UP);
 426         if (ret >= 0)
 427                 ret = dev_change_flags(dev, flags);
 428
 429         if (ret < 0) {
 430                 netdev_err(dev,
 431                            "Failed to cycle device %s; route tables might be wrong!\n",
 432                            dev->name);
 433         }
 434 }
 435
 436 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
 437 {
 438         int ret;
 439
 440         ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
 441         if (ret < 0)
 442                 return ret;
 443
 444         port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
 445         cycle_netdev(port_dev);
 446
 447         return 0;
 448 }
 449
 450 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
 451 {
 452         if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
 453                 return -EINVAL;
 454
 455         return do_vrf_add_slave(dev, port_dev);
 456 }
 457
 458 /* inverse of do_vrf_add_slave */
 459 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
 460 {
 461         netdev_upper_dev_unlink(port_dev, dev);
 462         port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
 463
 464         cycle_netdev(port_dev);
 465
 466         return 0;
 467 }
 468
 469 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
 470 {
 471         return do_vrf_del_slave(dev, port_dev);
 472 }
 473
 474 static void vrf_dev_uninit(struct net_device *dev)
 475 {
 476         struct net_vrf *vrf = netdev_priv(dev);
 477         struct net_device *port_dev;
 478         struct list_head *iter;
 479
 480         vrf_rtable_release(vrf);
 481         vrf_rt6_release(vrf);
 482
 483         netdev_for_each_lower_dev(dev, port_dev, iter)
 484                 vrf_del_slave(dev, port_dev);
 485
 486         free_percpu(dev->dstats);
 487         dev->dstats = NULL;
 488 }
 489
 490 static int vrf_dev_init(struct net_device *dev)
 491 {
 492         struct net_vrf *vrf = netdev_priv(dev);
 493
 494         dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
 495         if (!dev->dstats)
 496                 goto out_nomem;
 497
 498         /* create the default dst which points back to us */
 499         if (vrf_rtable_create(dev) != 0)
 500                 goto out_stats;
 501
 502         if (vrf_rt6_create(dev) != 0)
 503                 goto out_rth;
 504
 505         dev->flags = IFF_MASTER | IFF_NOARP;
 506
 507         return 0;
 508
 509 out_rth:
 510         vrf_rtable_release(vrf);
 511 out_stats:
 512         free_percpu(dev->dstats);
 513         dev->dstats = NULL;
 514 out_nomem:
 515         return -ENOMEM;
 516 }
 517
 518 static const struct net_device_ops vrf_netdev_ops = {
 519         .ndo_init               = vrf_dev_init,
 520         .ndo_uninit             = vrf_dev_uninit,
 521         .ndo_start_xmit         = vrf_xmit,
 522         .ndo_get_stats64        = vrf_get_stats64,
 523         .ndo_add_slave          = vrf_add_slave,
 524         .ndo_del_slave          = vrf_del_slave,
 525 };
 526
 527 static u32 vrf_fib_table(const struct net_device *dev)
 528 {
 529         struct net_vrf *vrf = netdev_priv(dev);
 530
 531         return vrf->tb_id;
 532 }
 533
 534 static struct rtable *vrf_get_rtable(const struct net_device *dev,
 535                                      const struct flowi4 *fl4)
 536 {
 537         struct rtable *rth = NULL;
 538
 539         if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
 540                 struct net_vrf *vrf = netdev_priv(dev);
 541
 542                 rcu_read_lock();
 543
 544                 rth = rcu_dereference(vrf->rth);
 545                 if (likely(rth))
 546                         dst_hold(&rth->dst);
 547
 548                 rcu_read_unlock();
 549         }
 550
 551         return rth;
 552 }
 553
 554 /* called under rcu_read_lock */
 555 static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4)
 556 {
 557         struct fib_result res = { .tclassid = 0 };
 558         struct net *net = dev_net(dev);
 559         u32 orig_tos = fl4->flowi4_tos;
 560         u8 flags = fl4->flowi4_flags;
 561         u8 scope = fl4->flowi4_scope;
 562         u8 tos = RT_FL_TOS(fl4);
 563         int rc;
 564
 565         if (unlikely(!fl4->daddr))
 566                 return 0;
 567
 568         fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF;
 569         fl4->flowi4_iif = LOOPBACK_IFINDEX;
 570         /* make sure oif is set to VRF device for lookup */
 571         fl4->flowi4_oif = dev->ifindex;
 572         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
 573         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
 574                              RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
 575
 576         rc = fib_lookup(net, fl4, &res, 0);
 577         if (!rc) {
 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         return rc;
 589 }
 590
 591 #if IS_ENABLED(CONFIG_IPV6)
 592 /* neighbor handling is done with actual device; do not want
 593  * to flip skb->dev for those ndisc packets. This really fails
 594  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
 595  * a start.
 596  */
 597 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
 598 {
 599         const struct ipv6hdr *iph = ipv6_hdr(skb);
 600         bool rc = false;
 601
 602         if (iph->nexthdr == NEXTHDR_ICMP) {
 603                 const struct icmp6hdr *icmph;
 604                 struct icmp6hdr _icmph;
 605
 606                 icmph = skb_header_pointer(skb, sizeof(*iph),
 607                                            sizeof(_icmph), &_icmph);
 608                 if (!icmph)
 609                         goto out;
 610
 611                 switch (icmph->icmp6_type) {
 612                 case NDISC_ROUTER_SOLICITATION:
 613                 case NDISC_ROUTER_ADVERTISEMENT:
 614                 case NDISC_NEIGHBOUR_SOLICITATION:
 615                 case NDISC_NEIGHBOUR_ADVERTISEMENT:
 616                 case NDISC_REDIRECT:
 617                         rc = true;
 618                         break;
 619                 }
 620         }
 621
 622 out:
 623         return rc;
 624 }
 625
 626 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
 627                                    struct sk_buff *skb)
 628 {
 629         /* if packet is NDISC keep the ingress interface */
 630         if (!ipv6_ndisc_frame(skb)) {
 631                 skb->dev = vrf_dev;
 632                 skb->skb_iif = vrf_dev->ifindex;
 633
 634                 skb_push(skb, skb->mac_len);
 635                 dev_queue_xmit_nit(skb, vrf_dev);
 636                 skb_pull(skb, skb->mac_len);
 637
 638                 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
 639         }
 640
 641         return skb;
 642 }
 643
 644 #else
 645 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
 646                                    struct sk_buff *skb)
 647 {
 648         return skb;
 649 }
 650 #endif
 651
 652 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
 653                                   struct sk_buff *skb)
 654 {
 655         skb->dev = vrf_dev;
 656         skb->skb_iif = vrf_dev->ifindex;
 657
 658         skb_push(skb, skb->mac_len);
 659         dev_queue_xmit_nit(skb, vrf_dev);
 660         skb_pull(skb, skb->mac_len);
 661
 662         return skb;
 663 }
 664
 665 /* called with rcu lock held */
 666 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
 667                                   struct sk_buff *skb,
 668                                   u16 proto)
 669 {
 670         switch (proto) {
 671         case AF_INET:
 672                 return vrf_ip_rcv(vrf_dev, skb);
 673         case AF_INET6:
 674                 return vrf_ip6_rcv(vrf_dev, skb);
 675         }
 676
 677         return skb;
 678 }
 679
 680 #if IS_ENABLED(CONFIG_IPV6)
 681 static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
 682                                          const struct flowi6 *fl6)
 683 {
 684         struct dst_entry *dst = NULL;
 685
 686         if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) {
 687                 struct net_vrf *vrf = netdev_priv(dev);
 688                 struct rt6_info *rt;
 689
 690                 rcu_read_lock();
 691
 692                 rt = rcu_dereference(vrf->rt6);
 693                 if (likely(rt)) {
 694                         dst = &rt->dst;
 695                         dst_hold(dst);
 696                 }
 697
 698                 rcu_read_unlock();
 699         }
 700
 701         return dst;
 702 }
 703 #endif
 704
 705 static const struct l3mdev_ops vrf_l3mdev_ops = {
 706         .l3mdev_fib_table       = vrf_fib_table,
 707         .l3mdev_get_rtable      = vrf_get_rtable,
 708         .l3mdev_get_saddr       = vrf_get_saddr,
 709         .l3mdev_l3_rcv          = vrf_l3_rcv,
 710 #if IS_ENABLED(CONFIG_IPV6)
 711         .l3mdev_get_rt6_dst     = vrf_get_rt6_dst,
 712 #endif
 713 };
 714
 715 static void vrf_get_drvinfo(struct net_device *dev,
 716                             struct ethtool_drvinfo *info)
 717 {
 718         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
 719         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
 720 }
 721
 722 static const struct ethtool_ops vrf_ethtool_ops = {
 723         .get_drvinfo    = vrf_get_drvinfo,
 724 };
 725
 726 static void vrf_setup(struct net_device *dev)
 727 {
 728         ether_setup(dev);
 729
 730         /* Initialize the device structure. */
 731         dev->netdev_ops = &vrf_netdev_ops;
 732         dev->l3mdev_ops = &vrf_l3mdev_ops;
 733         dev->ethtool_ops = &vrf_ethtool_ops;
 734         dev->destructor = free_netdev;
 735
 736         /* Fill in device structure with ethernet-generic values. */
 737         eth_hw_addr_random(dev);
 738
 739         /* don't acquire vrf device's netif_tx_lock when transmitting */
 740         dev->features |= NETIF_F_LLTX;
 741
 742         /* don't allow vrf devices to change network namespaces. */
 743         dev->features |= NETIF_F_NETNS_LOCAL;
 744 }
 745
 746 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
 747 {
 748         if (tb[IFLA_ADDRESS]) {
 749                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 750                         return -EINVAL;
 751                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 752                         return -EADDRNOTAVAIL;
 753         }
 754         return 0;
 755 }
 756
 757 static void vrf_dellink(struct net_device *dev, struct list_head *head)
 758 {
 759         unregister_netdevice_queue(dev, head);
 760 }
 761
 762 static int vrf_newlink(struct net *src_net, struct net_device *dev,
 763                        struct nlattr *tb[], struct nlattr *data[])
 764 {
 765         struct net_vrf *vrf = netdev_priv(dev);
 766
 767         if (!data || !data[IFLA_VRF_TABLE])
 768                 return -EINVAL;
 769
 770         vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
 771
 772         dev->priv_flags |= IFF_L3MDEV_MASTER;
 773
 774         return register_netdevice(dev);
 775 }
 776
 777 static size_t vrf_nl_getsize(const struct net_device *dev)
 778 {
 779         return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
 780 }
 781
 782 static int vrf_fillinfo(struct sk_buff *skb,
 783                         const struct net_device *dev)
 784 {
 785         struct net_vrf *vrf = netdev_priv(dev);
 786
 787         return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
 788 }
 789
 790 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
 791                                  const struct net_device *slave_dev)
 792 {
 793         return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
 794 }
 795
 796 static int vrf_fill_slave_info(struct sk_buff *skb,
 797                                const struct net_device *vrf_dev,
 798                                const struct net_device *slave_dev)
 799 {
 800         struct net_vrf *vrf = netdev_priv(vrf_dev);
 801
 802         if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
 803                 return -EMSGSIZE;
 804
 805         return 0;
 806 }
 807
 808 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
 809         [IFLA_VRF_TABLE] = { .type = NLA_U32 },
 810 };
 811
 812 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
 813         .kind           = DRV_NAME,
 814         .priv_size      = sizeof(struct net_vrf),
 815
 816         .get_size       = vrf_nl_getsize,
 817         .policy         = vrf_nl_policy,
 818         .validate       = vrf_validate,
 819         .fill_info      = vrf_fillinfo,
 820
 821         .get_slave_size  = vrf_get_slave_size,
 822         .fill_slave_info = vrf_fill_slave_info,
 823
 824         .newlink        = vrf_newlink,
 825         .dellink        = vrf_dellink,
 826         .setup          = vrf_setup,
 827         .maxtype        = IFLA_VRF_MAX,
 828 };
 829
 830 static int vrf_device_event(struct notifier_block *unused,
 831                             unsigned long event, void *ptr)
 832 {
 833         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 834
 835         /* only care about unregister events to drop slave references */
 836         if (event == NETDEV_UNREGISTER) {
 837                 struct net_device *vrf_dev;
 838
 839                 if (!netif_is_l3_slave(dev))
 840                         goto out;
 841
 842                 vrf_dev = netdev_master_upper_dev_get(dev);
 843                 vrf_del_slave(vrf_dev, dev);
 844         }
 845 out:
 846         return NOTIFY_DONE;
 847 }
 848
 849 static struct notifier_block vrf_notifier_block __read_mostly = {
 850         .notifier_call = vrf_device_event,
 851 };
 852
 853 static int __init vrf_init_module(void)
 854 {
 855         int rc;
 856
 857         register_netdevice_notifier(&vrf_notifier_block);
 858
 859         rc = rtnl_link_register(&vrf_link_ops);
 860         if (rc < 0)
 861                 goto error;
 862
 863         return 0;
 864
 865 error:
 866         unregister_netdevice_notifier(&vrf_notifier_block);
 867         return rc;
 868 }
 869
 870 module_init(vrf_init_module);
 871 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
 872 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
 873 MODULE_LICENSE("GPL");
 874 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
 875 MODULE_VERSION(DRV_VERSION);