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 /* neighbor handling is done with actual device; do not want
  64  * to flip skb->dev for those ndisc packets. This really fails
  65  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
  66  * a start.
  67  */
  68 #if IS_ENABLED(CONFIG_IPV6)
  69 static bool check_ipv6_frame(const struct sk_buff *skb)
  70 {
  71         const struct ipv6hdr *ipv6h;
  72         struct ipv6hdr _ipv6h;
  73         bool rc = true;
  74
  75         ipv6h = skb_header_pointer(skb, 0, sizeof(_ipv6h), &_ipv6h);
  76         if (!ipv6h)
  77                 goto out;
  78
  79         if (ipv6h->nexthdr == NEXTHDR_ICMP) {
  80                 const struct icmp6hdr *icmph;
  81                 struct icmp6hdr _icmph;
  82
  83                 icmph = skb_header_pointer(skb, sizeof(_ipv6h),
  84                                            sizeof(_icmph), &_icmph);
  85                 if (!icmph)
  86                         goto out;
  87
  88                 switch (icmph->icmp6_type) {
  89                 case NDISC_ROUTER_SOLICITATION:
  90                 case NDISC_ROUTER_ADVERTISEMENT:
  91                 case NDISC_NEIGHBOUR_SOLICITATION:
  92                 case NDISC_NEIGHBOUR_ADVERTISEMENT:
  93                 case NDISC_REDIRECT:
  94                         rc = false;
  95                         break;
  96                 }
  97         }
  98
  99 out:
 100         return rc;
 101 }
 102 #else
 103 static bool check_ipv6_frame(const struct sk_buff *skb)
 104 {
 105         return false;
 106 }
 107 #endif
 108
 109 static bool is_ip_rx_frame(struct sk_buff *skb)
 110 {
 111         switch (skb->protocol) {
 112         case htons(ETH_P_IP):
 113                 return true;
 114         case htons(ETH_P_IPV6):
 115                 return check_ipv6_frame(skb);
 116         }
 117         return false;
 118 }
 119
 120 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
 121 {
 122         vrf_dev->stats.tx_errors++;
 123         kfree_skb(skb);
 124 }
 125
 126 /* note: already called with rcu_read_lock */
 127 static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb)
 128 {
 129         struct sk_buff *skb = *pskb;
 130
 131         if (is_ip_rx_frame(skb)) {
 132                 struct net_device *dev = vrf_master_get_rcu(skb->dev);
 133                 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
 134
 135                 u64_stats_update_begin(&dstats->syncp);
 136                 dstats->rx_pkts++;
 137                 dstats->rx_bytes += skb->len;
 138                 u64_stats_update_end(&dstats->syncp);
 139
 140                 skb->dev = dev;
 141
 142                 return RX_HANDLER_ANOTHER;
 143         }
 144         return RX_HANDLER_PASS;
 145 }
 146
 147 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
 148                                                  struct rtnl_link_stats64 *stats)
 149 {
 150         int i;
 151
 152         for_each_possible_cpu(i) {
 153                 const struct pcpu_dstats *dstats;
 154                 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
 155                 unsigned int start;
 156
 157                 dstats = per_cpu_ptr(dev->dstats, i);
 158                 do {
 159                         start = u64_stats_fetch_begin_irq(&dstats->syncp);
 160                         tbytes = dstats->tx_bytes;
 161                         tpkts = dstats->tx_pkts;
 162                         tdrops = dstats->tx_drps;
 163                         rbytes = dstats->rx_bytes;
 164                         rpkts = dstats->rx_pkts;
 165                 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
 166                 stats->tx_bytes += tbytes;
 167                 stats->tx_packets += tpkts;
 168                 stats->tx_dropped += tdrops;
 169                 stats->rx_bytes += rbytes;
 170                 stats->rx_packets += rpkts;
 171         }
 172         return stats;
 173 }
 174
 175 #if IS_ENABLED(CONFIG_IPV6)
 176 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
 177                                            struct net_device *dev)
 178 {
 179         const struct ipv6hdr *iph = ipv6_hdr(skb);
 180         struct net *net = dev_net(skb->dev);
 181         struct flowi6 fl6 = {
 182                 /* needed to match OIF rule */
 183                 .flowi6_oif = dev->ifindex,
 184                 .flowi6_iif = LOOPBACK_IFINDEX,
 185                 .daddr = iph->daddr,
 186                 .saddr = iph->saddr,
 187                 .flowlabel = ip6_flowinfo(iph),
 188                 .flowi6_mark = skb->mark,
 189                 .flowi6_proto = iph->nexthdr,
 190                 .flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF,
 191         };
 192         int ret = NET_XMIT_DROP;
 193         struct dst_entry *dst;
 194         struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
 195
 196         dst = ip6_route_output(net, NULL, &fl6);
 197         if (dst == dst_null)
 198                 goto err;
 199
 200         skb_dst_drop(skb);
 201         skb_dst_set(skb, dst);
 202
 203         ret = ip6_local_out(net, skb->sk, skb);
 204         if (unlikely(net_xmit_eval(ret)))
 205                 dev->stats.tx_errors++;
 206         else
 207                 ret = NET_XMIT_SUCCESS;
 208
 209         return ret;
 210 err:
 211         vrf_tx_error(dev, skb);
 212         return NET_XMIT_DROP;
 213 }
 214 #else
 215 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
 216                                            struct net_device *dev)
 217 {
 218         vrf_tx_error(dev, skb);
 219         return NET_XMIT_DROP;
 220 }
 221 #endif
 222
 223 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
 224                             struct net_device *vrf_dev)
 225 {
 226         struct rtable *rt;
 227         int err = 1;
 228
 229         rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
 230         if (IS_ERR(rt))
 231                 goto out;
 232
 233         /* TO-DO: what about broadcast ? */
 234         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 235                 ip_rt_put(rt);
 236                 goto out;
 237         }
 238
 239         skb_dst_drop(skb);
 240         skb_dst_set(skb, &rt->dst);
 241         err = 0;
 242 out:
 243         return err;
 244 }
 245
 246 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
 247                                            struct net_device *vrf_dev)
 248 {
 249         struct iphdr *ip4h = ip_hdr(skb);
 250         int ret = NET_XMIT_DROP;
 251         struct flowi4 fl4 = {
 252                 /* needed to match OIF rule */
 253                 .flowi4_oif = vrf_dev->ifindex,
 254                 .flowi4_iif = LOOPBACK_IFINDEX,
 255                 .flowi4_tos = RT_TOS(ip4h->tos),
 256                 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
 257                                 FLOWI_FLAG_SKIP_NH_OIF,
 258                 .daddr = ip4h->daddr,
 259         };
 260
 261         if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
 262                 goto err;
 263
 264         if (!ip4h->saddr) {
 265                 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
 266                                                RT_SCOPE_LINK);
 267         }
 268
 269         ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 270         if (unlikely(net_xmit_eval(ret)))
 271                 vrf_dev->stats.tx_errors++;
 272         else
 273                 ret = NET_XMIT_SUCCESS;
 274
 275 out:
 276         return ret;
 277 err:
 278         vrf_tx_error(vrf_dev, skb);
 279         goto out;
 280 }
 281
 282 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
 283 {
 284         /* strip the ethernet header added for pass through VRF device */
 285         __skb_pull(skb, skb_network_offset(skb));
 286
 287         switch (skb->protocol) {
 288         case htons(ETH_P_IP):
 289                 return vrf_process_v4_outbound(skb, dev);
 290         case htons(ETH_P_IPV6):
 291                 return vrf_process_v6_outbound(skb, dev);
 292         default:
 293                 vrf_tx_error(dev, skb);
 294                 return NET_XMIT_DROP;
 295         }
 296 }
 297
 298 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
 299 {
 300         netdev_tx_t ret = is_ip_tx_frame(skb, dev);
 301
 302         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 303                 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
 304
 305                 u64_stats_update_begin(&dstats->syncp);
 306                 dstats->tx_pkts++;
 307                 dstats->tx_bytes += skb->len;
 308                 u64_stats_update_end(&dstats->syncp);
 309         } else {
 310                 this_cpu_inc(dev->dstats->tx_drps);
 311         }
 312
 313         return ret;
 314 }
 315
 316 #if IS_ENABLED(CONFIG_IPV6)
 317 /* modelled after ip6_finish_output2 */
 318 static int vrf_finish_output6(struct net *net, struct sock *sk,
 319                               struct sk_buff *skb)
 320 {
 321         struct dst_entry *dst = skb_dst(skb);
 322         struct net_device *dev = dst->dev;
 323         struct neighbour *neigh;
 324         struct in6_addr *nexthop;
 325         int ret;
 326
 327         skb->protocol = htons(ETH_P_IPV6);
 328         skb->dev = dev;
 329
 330         rcu_read_lock_bh();
 331         nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
 332         neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
 333         if (unlikely(!neigh))
 334                 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
 335         if (!IS_ERR(neigh)) {
 336                 ret = dst_neigh_output(dst, neigh, skb);
 337                 rcu_read_unlock_bh();
 338                 return ret;
 339         }
 340         rcu_read_unlock_bh();
 341
 342         IP6_INC_STATS(dev_net(dst->dev),
 343                       ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
 344         kfree_skb(skb);
 345         return -EINVAL;
 346 }
 347
 348 /* modelled after ip6_output */
 349 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
 350 {
 351         return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
 352                             net, sk, skb, NULL, skb_dst(skb)->dev,
 353                             vrf_finish_output6,
 354                             !(IP6CB(skb)->flags & IP6SKB_REROUTED));
 355 }
 356
 357 static void vrf_rt6_release(struct net_vrf *vrf)
 358 {
 359         dst_release(&vrf->rt6->dst);
 360         vrf->rt6 = NULL;
 361 }
 362
 363 static int vrf_rt6_create(struct net_device *dev)
 364 {
 365         struct net_vrf *vrf = netdev_priv(dev);
 366         struct net *net = dev_net(dev);
 367         struct fib6_table *rt6i_table;
 368         struct rt6_info *rt6;
 369         int rc = -ENOMEM;
 370
 371         rt6i_table = fib6_new_table(net, vrf->tb_id);
 372         if (!rt6i_table)
 373                 goto out;
 374
 375         rt6 = ip6_dst_alloc(net, dev,
 376                             DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE);
 377         if (!rt6)
 378                 goto out;
 379
 380         dst_hold(&rt6->dst);
 381
 382         rt6->rt6i_table = rt6i_table;
 383         rt6->dst.output = vrf_output6;
 384         vrf->rt6 = rt6;
 385         rc = 0;
 386 out:
 387         return rc;
 388 }
 389 #else
 390 static void vrf_rt6_release(struct net_vrf *vrf)
 391 {
 392 }
 393
 394 static int vrf_rt6_create(struct net_device *dev)
 395 {
 396         return 0;
 397 }
 398 #endif
 399
 400 /* modelled after ip_finish_output2 */
 401 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 402 {
 403         struct dst_entry *dst = skb_dst(skb);
 404         struct rtable *rt = (struct rtable *)dst;
 405         struct net_device *dev = dst->dev;
 406         unsigned int hh_len = LL_RESERVED_SPACE(dev);
 407         struct neighbour *neigh;
 408         u32 nexthop;
 409         int ret = -EINVAL;
 410
 411         /* Be paranoid, rather than too clever. */
 412         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
 413                 struct sk_buff *skb2;
 414
 415                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
 416                 if (!skb2) {
 417                         ret = -ENOMEM;
 418                         goto err;
 419                 }
 420                 if (skb->sk)
 421                         skb_set_owner_w(skb2, skb->sk);
 422
 423                 consume_skb(skb);
 424                 skb = skb2;
 425         }
 426
 427         rcu_read_lock_bh();
 428
 429         nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
 430         neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
 431         if (unlikely(!neigh))
 432                 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
 433         if (!IS_ERR(neigh))
 434                 ret = dst_neigh_output(dst, neigh, skb);
 435
 436         rcu_read_unlock_bh();
 437 err:
 438         if (unlikely(ret < 0))
 439                 vrf_tx_error(skb->dev, skb);
 440         return ret;
 441 }
 442
 443 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 444 {
 445         struct net_device *dev = skb_dst(skb)->dev;
 446
 447         IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
 448
 449         skb->dev = dev;
 450         skb->protocol = htons(ETH_P_IP);
 451
 452         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
 453                             net, sk, skb, NULL, dev,
 454                             vrf_finish_output,
 455                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 456 }
 457
 458 static void vrf_rtable_release(struct net_vrf *vrf)
 459 {
 460         struct dst_entry *dst = (struct dst_entry *)vrf->rth;
 461
 462         dst_release(dst);
 463         vrf->rth = NULL;
 464 }
 465
 466 static struct rtable *vrf_rtable_create(struct net_device *dev)
 467 {
 468         struct net_vrf *vrf = netdev_priv(dev);
 469         struct rtable *rth;
 470
 471         if (!fib_new_table(dev_net(dev), vrf->tb_id))
 472                 return NULL;
 473
 474         rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
 475         if (rth) {
 476                 rth->dst.output = vrf_output;
 477                 rth->rt_table_id = vrf->tb_id;
 478         }
 479
 480         return rth;
 481 }
 482
 483 /**************************** device handling ********************/
 484
 485 /* cycle interface to flush neighbor cache and move routes across tables */
 486 static void cycle_netdev(struct net_device *dev)
 487 {
 488         unsigned int flags = dev->flags;
 489         int ret;
 490
 491         if (!netif_running(dev))
 492                 return;
 493
 494         ret = dev_change_flags(dev, flags & ~IFF_UP);
 495         if (ret >= 0)
 496                 ret = dev_change_flags(dev, flags);
 497
 498         if (ret < 0) {
 499                 netdev_err(dev,
 500                            "Failed to cycle device %s; route tables might be wrong!\n",
 501                            dev->name);
 502         }
 503 }
 504
 505 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
 506 {
 507         int ret;
 508
 509         /* register the packet handler for slave ports */
 510         ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev);
 511         if (ret) {
 512                 netdev_err(port_dev,
 513                            "Device %s failed to register rx_handler\n",
 514                            port_dev->name);
 515                 goto out_fail;
 516         }
 517
 518         ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
 519         if (ret < 0)
 520                 goto out_unregister;
 521
 522         port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
 523         cycle_netdev(port_dev);
 524
 525         return 0;
 526
 527 out_unregister:
 528         netdev_rx_handler_unregister(port_dev);
 529 out_fail:
 530         return ret;
 531 }
 532
 533 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
 534 {
 535         if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
 536                 return -EINVAL;
 537
 538         return do_vrf_add_slave(dev, port_dev);
 539 }
 540
 541 /* inverse of do_vrf_add_slave */
 542 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
 543 {
 544         netdev_upper_dev_unlink(port_dev, dev);
 545         port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
 546
 547         netdev_rx_handler_unregister(port_dev);
 548
 549         cycle_netdev(port_dev);
 550
 551         return 0;
 552 }
 553
 554 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
 555 {
 556         return do_vrf_del_slave(dev, port_dev);
 557 }
 558
 559 static void vrf_dev_uninit(struct net_device *dev)
 560 {
 561         struct net_vrf *vrf = netdev_priv(dev);
 562         struct net_device *port_dev;
 563         struct list_head *iter;
 564
 565         vrf_rtable_release(vrf);
 566         vrf_rt6_release(vrf);
 567
 568         netdev_for_each_lower_dev(dev, port_dev, iter)
 569                 vrf_del_slave(dev, port_dev);
 570
 571         free_percpu(dev->dstats);
 572         dev->dstats = NULL;
 573 }
 574
 575 static int vrf_dev_init(struct net_device *dev)
 576 {
 577         struct net_vrf *vrf = netdev_priv(dev);
 578
 579         dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
 580         if (!dev->dstats)
 581                 goto out_nomem;
 582
 583         /* create the default dst which points back to us */
 584         vrf->rth = vrf_rtable_create(dev);
 585         if (!vrf->rth)
 586                 goto out_stats;
 587
 588         if (vrf_rt6_create(dev) != 0)
 589                 goto out_rth;
 590
 591         dev->flags = IFF_MASTER | IFF_NOARP;
 592
 593         return 0;
 594
 595 out_rth:
 596         vrf_rtable_release(vrf);
 597 out_stats:
 598         free_percpu(dev->dstats);
 599         dev->dstats = NULL;
 600 out_nomem:
 601         return -ENOMEM;
 602 }
 603
 604 static const struct net_device_ops vrf_netdev_ops = {
 605         .ndo_init               = vrf_dev_init,
 606         .ndo_uninit             = vrf_dev_uninit,
 607         .ndo_start_xmit         = vrf_xmit,
 608         .ndo_get_stats64        = vrf_get_stats64,
 609         .ndo_add_slave          = vrf_add_slave,
 610         .ndo_del_slave          = vrf_del_slave,
 611 };
 612
 613 static u32 vrf_fib_table(const struct net_device *dev)
 614 {
 615         struct net_vrf *vrf = netdev_priv(dev);
 616
 617         return vrf->tb_id;
 618 }
 619
 620 static struct rtable *vrf_get_rtable(const struct net_device *dev,
 621                                      const struct flowi4 *fl4)
 622 {
 623         struct rtable *rth = NULL;
 624
 625         if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
 626                 struct net_vrf *vrf = netdev_priv(dev);
 627
 628                 rth = vrf->rth;
 629                 dst_hold(&rth->dst);
 630         }
 631
 632         return rth;
 633 }
 634
 635 /* called under rcu_read_lock */
 636 static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4)
 637 {
 638         struct fib_result res = { .tclassid = 0 };
 639         struct net *net = dev_net(dev);
 640         u32 orig_tos = fl4->flowi4_tos;
 641         u8 flags = fl4->flowi4_flags;
 642         u8 scope = fl4->flowi4_scope;
 643         u8 tos = RT_FL_TOS(fl4);
 644         int rc;
 645
 646         if (unlikely(!fl4->daddr))
 647                 return 0;
 648
 649         fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF;
 650         fl4->flowi4_iif = LOOPBACK_IFINDEX;
 651         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
 652         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
 653                              RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
 654
 655         rc = fib_lookup(net, fl4, &res, 0);
 656         if (!rc) {
 657                 if (res.type == RTN_LOCAL)
 658                         fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr;
 659                 else
 660                         fib_select_path(net, &res, fl4, -1);
 661         }
 662
 663         fl4->flowi4_flags = flags;
 664         fl4->flowi4_tos = orig_tos;
 665         fl4->flowi4_scope = scope;
 666
 667         return rc;
 668 }
 669
 670 #if IS_ENABLED(CONFIG_IPV6)
 671 static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
 672                                          const struct flowi6 *fl6)
 673 {
 674         struct rt6_info *rt = NULL;
 675
 676         if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) {
 677                 struct net_vrf *vrf = netdev_priv(dev);
 678
 679                 rt = vrf->rt6;
 680                 dst_hold(&rt->dst);
 681         }
 682
 683         return (struct dst_entry *)rt;
 684 }
 685 #endif
 686
 687 static const struct l3mdev_ops vrf_l3mdev_ops = {
 688         .l3mdev_fib_table       = vrf_fib_table,
 689         .l3mdev_get_rtable      = vrf_get_rtable,
 690         .l3mdev_get_saddr       = vrf_get_saddr,
 691 #if IS_ENABLED(CONFIG_IPV6)
 692         .l3mdev_get_rt6_dst     = vrf_get_rt6_dst,
 693 #endif
 694 };
 695
 696 static void vrf_get_drvinfo(struct net_device *dev,
 697                             struct ethtool_drvinfo *info)
 698 {
 699         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
 700         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
 701 }
 702
 703 static const struct ethtool_ops vrf_ethtool_ops = {
 704         .get_drvinfo    = vrf_get_drvinfo,
 705 };
 706
 707 static void vrf_setup(struct net_device *dev)
 708 {
 709         ether_setup(dev);
 710
 711         /* Initialize the device structure. */
 712         dev->netdev_ops = &vrf_netdev_ops;
 713         dev->l3mdev_ops = &vrf_l3mdev_ops;
 714         dev->ethtool_ops = &vrf_ethtool_ops;
 715         dev->destructor = free_netdev;
 716
 717         /* Fill in device structure with ethernet-generic values. */
 718         eth_hw_addr_random(dev);
 719
 720         /* don't acquire vrf device's netif_tx_lock when transmitting */
 721         dev->features |= NETIF_F_LLTX;
 722
 723         /* don't allow vrf devices to change network namespaces. */
 724         dev->features |= NETIF_F_NETNS_LOCAL;
 725 }
 726
 727 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
 728 {
 729         if (tb[IFLA_ADDRESS]) {
 730                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 731                         return -EINVAL;
 732                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 733                         return -EADDRNOTAVAIL;
 734         }
 735         return 0;
 736 }
 737
 738 static void vrf_dellink(struct net_device *dev, struct list_head *head)
 739 {
 740         unregister_netdevice_queue(dev, head);
 741 }
 742
 743 static int vrf_newlink(struct net *src_net, struct net_device *dev,
 744                        struct nlattr *tb[], struct nlattr *data[])
 745 {
 746         struct net_vrf *vrf = netdev_priv(dev);
 747
 748         if (!data || !data[IFLA_VRF_TABLE])
 749                 return -EINVAL;
 750
 751         vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
 752
 753         dev->priv_flags |= IFF_L3MDEV_MASTER;
 754
 755         return register_netdevice(dev);
 756 }
 757
 758 static size_t vrf_nl_getsize(const struct net_device *dev)
 759 {
 760         return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
 761 }
 762
 763 static int vrf_fillinfo(struct sk_buff *skb,
 764                         const struct net_device *dev)
 765 {
 766         struct net_vrf *vrf = netdev_priv(dev);
 767
 768         return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
 769 }
 770
 771 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
 772                                  const struct net_device *slave_dev)
 773 {
 774         return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
 775 }
 776
 777 static int vrf_fill_slave_info(struct sk_buff *skb,
 778                                const struct net_device *vrf_dev,
 779                                const struct net_device *slave_dev)
 780 {
 781         struct net_vrf *vrf = netdev_priv(vrf_dev);
 782
 783         if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
 784                 return -EMSGSIZE;
 785
 786         return 0;
 787 }
 788
 789 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
 790         [IFLA_VRF_TABLE] = { .type = NLA_U32 },
 791 };
 792
 793 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
 794         .kind           = DRV_NAME,
 795         .priv_size      = sizeof(struct net_vrf),
 796
 797         .get_size       = vrf_nl_getsize,
 798         .policy         = vrf_nl_policy,
 799         .validate       = vrf_validate,
 800         .fill_info      = vrf_fillinfo,
 801
 802         .get_slave_size  = vrf_get_slave_size,
 803         .fill_slave_info = vrf_fill_slave_info,
 804
 805         .newlink        = vrf_newlink,
 806         .dellink        = vrf_dellink,
 807         .setup          = vrf_setup,
 808         .maxtype        = IFLA_VRF_MAX,
 809 };
 810
 811 static int vrf_device_event(struct notifier_block *unused,
 812                             unsigned long event, void *ptr)
 813 {
 814         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 815
 816         /* only care about unregister events to drop slave references */
 817         if (event == NETDEV_UNREGISTER) {
 818                 struct net_device *vrf_dev;
 819
 820                 if (!netif_is_l3_slave(dev))
 821                         goto out;
 822
 823                 vrf_dev = netdev_master_upper_dev_get(dev);
 824                 vrf_del_slave(vrf_dev, dev);
 825         }
 826 out:
 827         return NOTIFY_DONE;
 828 }
 829
 830 static struct notifier_block vrf_notifier_block __read_mostly = {
 831         .notifier_call = vrf_device_event,
 832 };
 833
 834 static int __init vrf_init_module(void)
 835 {
 836         int rc;
 837
 838         register_netdevice_notifier(&vrf_notifier_block);
 839
 840         rc = rtnl_link_register(&vrf_link_ops);
 841         if (rc < 0)
 842                 goto error;
 843
 844         return 0;
 845
 846 error:
 847         unregister_netdevice_notifier(&vrf_notifier_block);
 848         return rc;
 849 }
 850
 851 module_init(vrf_init_module);
 852 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
 853 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
 854 MODULE_LICENSE("GPL");
 855 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
 856 MODULE_VERSION(DRV_VERSION);