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 rtable __rcu *rth_local; 48 struct rt6_info __rcu *rt6; 49 struct rt6_info __rcu *rt6_local; 50 u32 tb_id; 51 }; 52 53 struct pcpu_dstats { 54 u64 tx_pkts; 55 u64 tx_bytes; 56 u64 tx_drps; 57 u64 rx_pkts; 58 u64 rx_bytes; 59 u64 rx_drps; 60 struct u64_stats_sync syncp; 61 }; 62 63 static void vrf_rx_stats(struct net_device *dev, int len) 64 { 65 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 66 67 u64_stats_update_begin(&dstats->syncp); 68 dstats->rx_pkts++; 69 dstats->rx_bytes += len; 70 u64_stats_update_end(&dstats->syncp); 71 } 72 73 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 74 { 75 vrf_dev->stats.tx_errors++; 76 kfree_skb(skb); 77 } 78 79 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev, 80 struct rtnl_link_stats64 *stats) 81 { 82 int i; 83 84 for_each_possible_cpu(i) { 85 const struct pcpu_dstats *dstats; 86 u64 tbytes, tpkts, tdrops, rbytes, rpkts; 87 unsigned int start; 88 89 dstats = per_cpu_ptr(dev->dstats, i); 90 do { 91 start = u64_stats_fetch_begin_irq(&dstats->syncp); 92 tbytes = dstats->tx_bytes; 93 tpkts = dstats->tx_pkts; 94 tdrops = dstats->tx_drps; 95 rbytes = dstats->rx_bytes; 96 rpkts = dstats->rx_pkts; 97 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start)); 98 stats->tx_bytes += tbytes; 99 stats->tx_packets += tpkts; 100 stats->tx_dropped += tdrops; 101 stats->rx_bytes += rbytes; 102 stats->rx_packets += rpkts; 103 } 104 return stats; 105 } 106 107 /* Local traffic destined to local address. Reinsert the packet to rx 108 * path, similar to loopback handling. 109 */ 110 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 111 struct dst_entry *dst) 112 { 113 int len = skb->len; 114 115 skb_orphan(skb); 116 117 skb_dst_set(skb, dst); 118 skb_dst_force(skb); 119 120 /* set pkt_type to avoid skb hitting packet taps twice - 121 * once on Tx and again in Rx processing 122 */ 123 skb->pkt_type = PACKET_LOOPBACK; 124 125 skb->protocol = eth_type_trans(skb, dev); 126 127 if (likely(netif_rx(skb) == NET_RX_SUCCESS)) 128 vrf_rx_stats(dev, len); 129 else 130 this_cpu_inc(dev->dstats->rx_drps); 131 132 return NETDEV_TX_OK; 133 } 134 135 #if IS_ENABLED(CONFIG_IPV6) 136 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 137 struct net_device *dev) 138 { 139 const struct ipv6hdr *iph = ipv6_hdr(skb); 140 struct net *net = dev_net(skb->dev); 141 struct flowi6 fl6 = { 142 /* needed to match OIF rule */ 143 .flowi6_oif = dev->ifindex, 144 .flowi6_iif = LOOPBACK_IFINDEX, 145 .daddr = iph->daddr, 146 .saddr = iph->saddr, 147 .flowlabel = ip6_flowinfo(iph), 148 .flowi6_mark = skb->mark, 149 .flowi6_proto = iph->nexthdr, 150 .flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF, 151 }; 152 int ret = NET_XMIT_DROP; 153 struct dst_entry *dst; 154 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 155 156 dst = ip6_route_output(net, NULL, &fl6); 157 if (dst == dst_null) 158 goto err; 159 160 skb_dst_drop(skb); 161 162 /* if dst.dev is loopback or the VRF device again this is locally 163 * originated traffic destined to a local address. Short circuit 164 * to Rx path using our local dst 165 */ 166 if (dst->dev == net->loopback_dev || dst->dev == dev) { 167 struct net_vrf *vrf = netdev_priv(dev); 168 struct rt6_info *rt6_local; 169 170 /* release looked up dst and use cached local dst */ 171 dst_release(dst); 172 173 rcu_read_lock(); 174 175 rt6_local = rcu_dereference(vrf->rt6_local); 176 if (unlikely(!rt6_local)) { 177 rcu_read_unlock(); 178 goto err; 179 } 180 181 /* Ordering issue: cached local dst is created on newlink 182 * before the IPv6 initialization. Using the local dst 183 * requires rt6i_idev to be set so make sure it is. 184 */ 185 if (unlikely(!rt6_local->rt6i_idev)) { 186 rt6_local->rt6i_idev = in6_dev_get(dev); 187 if (!rt6_local->rt6i_idev) { 188 rcu_read_unlock(); 189 goto err; 190 } 191 } 192 193 dst = &rt6_local->dst; 194 dst_hold(dst); 195 196 rcu_read_unlock(); 197 198 return vrf_local_xmit(skb, dev, &rt6_local->dst); 199 } 200 201 skb_dst_set(skb, dst); 202 203 /* strip the ethernet header added for pass through VRF device */ 204 __skb_pull(skb, skb_network_offset(skb)); 205 206 ret = ip6_local_out(net, skb->sk, skb); 207 if (unlikely(net_xmit_eval(ret))) 208 dev->stats.tx_errors++; 209 else 210 ret = NET_XMIT_SUCCESS; 211 212 return ret; 213 err: 214 vrf_tx_error(dev, skb); 215 return NET_XMIT_DROP; 216 } 217 #else 218 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 219 struct net_device *dev) 220 { 221 vrf_tx_error(dev, skb); 222 return NET_XMIT_DROP; 223 } 224 #endif 225 226 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 227 struct net_device *vrf_dev) 228 { 229 struct iphdr *ip4h = ip_hdr(skb); 230 int ret = NET_XMIT_DROP; 231 struct flowi4 fl4 = { 232 /* needed to match OIF rule */ 233 .flowi4_oif = vrf_dev->ifindex, 234 .flowi4_iif = LOOPBACK_IFINDEX, 235 .flowi4_tos = RT_TOS(ip4h->tos), 236 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC | 237 FLOWI_FLAG_SKIP_NH_OIF, 238 .daddr = ip4h->daddr, 239 }; 240 struct net *net = dev_net(vrf_dev); 241 struct rtable *rt; 242 243 rt = ip_route_output_flow(net, &fl4, NULL); 244 if (IS_ERR(rt)) 245 goto err; 246 247 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { 248 ip_rt_put(rt); 249 goto err; 250 } 251 252 skb_dst_drop(skb); 253 254 /* if dst.dev is loopback or the VRF device again this is locally 255 * originated traffic destined to a local address. Short circuit 256 * to Rx path using our local dst 257 */ 258 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) { 259 struct net_vrf *vrf = netdev_priv(vrf_dev); 260 struct rtable *rth_local; 261 struct dst_entry *dst = NULL; 262 263 ip_rt_put(rt); 264 265 rcu_read_lock(); 266 267 rth_local = rcu_dereference(vrf->rth_local); 268 if (likely(rth_local)) { 269 dst = &rth_local->dst; 270 dst_hold(dst); 271 } 272 273 rcu_read_unlock(); 274 275 if (unlikely(!dst)) 276 goto err; 277 278 return vrf_local_xmit(skb, vrf_dev, dst); 279 } 280 281 skb_dst_set(skb, &rt->dst); 282 283 /* strip the ethernet header added for pass through VRF device */ 284 __skb_pull(skb, skb_network_offset(skb)); 285 286 if (!ip4h->saddr) { 287 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 288 RT_SCOPE_LINK); 289 } 290 291 ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 292 if (unlikely(net_xmit_eval(ret))) 293 vrf_dev->stats.tx_errors++; 294 else 295 ret = NET_XMIT_SUCCESS; 296 297 out: 298 return ret; 299 err: 300 vrf_tx_error(vrf_dev, skb); 301 goto out; 302 } 303 304 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 305 { 306 switch (skb->protocol) { 307 case htons(ETH_P_IP): 308 return vrf_process_v4_outbound(skb, dev); 309 case htons(ETH_P_IPV6): 310 return vrf_process_v6_outbound(skb, dev); 311 default: 312 vrf_tx_error(dev, skb); 313 return NET_XMIT_DROP; 314 } 315 } 316 317 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 318 { 319 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 320 321 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 322 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 323 324 u64_stats_update_begin(&dstats->syncp); 325 dstats->tx_pkts++; 326 dstats->tx_bytes += skb->len; 327 u64_stats_update_end(&dstats->syncp); 328 } else { 329 this_cpu_inc(dev->dstats->tx_drps); 330 } 331 332 return ret; 333 } 334 335 #if IS_ENABLED(CONFIG_IPV6) 336 /* modelled after ip6_finish_output2 */ 337 static int vrf_finish_output6(struct net *net, struct sock *sk, 338 struct sk_buff *skb) 339 { 340 struct dst_entry *dst = skb_dst(skb); 341 struct net_device *dev = dst->dev; 342 struct neighbour *neigh; 343 struct in6_addr *nexthop; 344 int ret; 345 346 skb->protocol = htons(ETH_P_IPV6); 347 skb->dev = dev; 348 349 rcu_read_lock_bh(); 350 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 351 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 352 if (unlikely(!neigh)) 353 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 354 if (!IS_ERR(neigh)) { 355 ret = dst_neigh_output(dst, neigh, skb); 356 rcu_read_unlock_bh(); 357 return ret; 358 } 359 rcu_read_unlock_bh(); 360 361 IP6_INC_STATS(dev_net(dst->dev), 362 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 363 kfree_skb(skb); 364 return -EINVAL; 365 } 366 367 /* modelled after ip6_output */ 368 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 369 { 370 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 371 net, sk, skb, NULL, skb_dst(skb)->dev, 372 vrf_finish_output6, 373 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 374 } 375 376 /* holding rtnl */ 377 static void vrf_rt6_release(struct net_vrf *vrf) 378 { 379 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 380 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local); 381 382 RCU_INIT_POINTER(vrf->rt6, NULL); 383 RCU_INIT_POINTER(vrf->rt6_local, NULL); 384 synchronize_rcu(); 385 386 if (rt6) 387 dst_release(&rt6->dst); 388 389 if (rt6_local) { 390 if (rt6_local->rt6i_idev) 391 in6_dev_put(rt6_local->rt6i_idev); 392 393 dst_release(&rt6_local->dst); 394 } 395 } 396 397 static int vrf_rt6_create(struct net_device *dev) 398 { 399 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE; 400 struct net_vrf *vrf = netdev_priv(dev); 401 struct net *net = dev_net(dev); 402 struct fib6_table *rt6i_table; 403 struct rt6_info *rt6, *rt6_local; 404 int rc = -ENOMEM; 405 406 rt6i_table = fib6_new_table(net, vrf->tb_id); 407 if (!rt6i_table) 408 goto out; 409 410 /* create a dst for routing packets out a VRF device */ 411 rt6 = ip6_dst_alloc(net, dev, flags); 412 if (!rt6) 413 goto out; 414 415 dst_hold(&rt6->dst); 416 417 rt6->rt6i_table = rt6i_table; 418 rt6->dst.output = vrf_output6; 419 420 /* create a dst for local routing - packets sent locally 421 * to local address via the VRF device as a loopback 422 */ 423 rt6_local = ip6_dst_alloc(net, dev, flags); 424 if (!rt6_local) { 425 dst_release(&rt6->dst); 426 goto out; 427 } 428 429 dst_hold(&rt6_local->dst); 430 431 rt6_local->rt6i_idev = in6_dev_get(dev); 432 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL; 433 rt6_local->rt6i_table = rt6i_table; 434 rt6_local->dst.input = ip6_input; 435 436 rcu_assign_pointer(vrf->rt6, rt6); 437 rcu_assign_pointer(vrf->rt6_local, rt6_local); 438 439 rc = 0; 440 out: 441 return rc; 442 } 443 #else 444 static void vrf_rt6_release(struct net_vrf *vrf) 445 { 446 } 447 448 static int vrf_rt6_create(struct net_device *dev) 449 { 450 return 0; 451 } 452 #endif 453 454 /* modelled after ip_finish_output2 */ 455 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 456 { 457 struct dst_entry *dst = skb_dst(skb); 458 struct rtable *rt = (struct rtable *)dst; 459 struct net_device *dev = dst->dev; 460 unsigned int hh_len = LL_RESERVED_SPACE(dev); 461 struct neighbour *neigh; 462 u32 nexthop; 463 int ret = -EINVAL; 464 465 /* Be paranoid, rather than too clever. */ 466 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 467 struct sk_buff *skb2; 468 469 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 470 if (!skb2) { 471 ret = -ENOMEM; 472 goto err; 473 } 474 if (skb->sk) 475 skb_set_owner_w(skb2, skb->sk); 476 477 consume_skb(skb); 478 skb = skb2; 479 } 480 481 rcu_read_lock_bh(); 482 483 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); 484 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 485 if (unlikely(!neigh)) 486 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 487 if (!IS_ERR(neigh)) 488 ret = dst_neigh_output(dst, neigh, skb); 489 490 rcu_read_unlock_bh(); 491 err: 492 if (unlikely(ret < 0)) 493 vrf_tx_error(skb->dev, skb); 494 return ret; 495 } 496 497 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 498 { 499 struct net_device *dev = skb_dst(skb)->dev; 500 501 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 502 503 skb->dev = dev; 504 skb->protocol = htons(ETH_P_IP); 505 506 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 507 net, sk, skb, NULL, dev, 508 vrf_finish_output, 509 !(IPCB(skb)->flags & IPSKB_REROUTED)); 510 } 511 512 /* holding rtnl */ 513 static void vrf_rtable_release(struct net_vrf *vrf) 514 { 515 struct rtable *rth = rtnl_dereference(vrf->rth); 516 struct rtable *rth_local = rtnl_dereference(vrf->rth_local); 517 518 RCU_INIT_POINTER(vrf->rth, NULL); 519 RCU_INIT_POINTER(vrf->rth_local, NULL); 520 synchronize_rcu(); 521 522 if (rth) 523 dst_release(&rth->dst); 524 525 if (rth_local) 526 dst_release(&rth_local->dst); 527 } 528 529 static int vrf_rtable_create(struct net_device *dev) 530 { 531 struct net_vrf *vrf = netdev_priv(dev); 532 struct rtable *rth, *rth_local; 533 534 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 535 return -ENOMEM; 536 537 /* create a dst for routing packets out through a VRF device */ 538 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0); 539 if (!rth) 540 return -ENOMEM; 541 542 /* create a dst for local ingress routing - packets sent locally 543 * to local address via the VRF device as a loopback 544 */ 545 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0); 546 if (!rth_local) { 547 dst_release(&rth->dst); 548 return -ENOMEM; 549 } 550 551 rth->dst.output = vrf_output; 552 rth->rt_table_id = vrf->tb_id; 553 554 rth_local->rt_table_id = vrf->tb_id; 555 556 rcu_assign_pointer(vrf->rth, rth); 557 rcu_assign_pointer(vrf->rth_local, rth_local); 558 559 return 0; 560 } 561 562 /**************************** device handling ********************/ 563 564 /* cycle interface to flush neighbor cache and move routes across tables */ 565 static void cycle_netdev(struct net_device *dev) 566 { 567 unsigned int flags = dev->flags; 568 int ret; 569 570 if (!netif_running(dev)) 571 return; 572 573 ret = dev_change_flags(dev, flags & ~IFF_UP); 574 if (ret >= 0) 575 ret = dev_change_flags(dev, flags); 576 577 if (ret < 0) { 578 netdev_err(dev, 579 "Failed to cycle device %s; route tables might be wrong!\n", 580 dev->name); 581 } 582 } 583 584 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 585 { 586 int ret; 587 588 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL); 589 if (ret < 0) 590 return ret; 591 592 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 593 cycle_netdev(port_dev); 594 595 return 0; 596 } 597 598 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 599 { 600 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev)) 601 return -EINVAL; 602 603 return do_vrf_add_slave(dev, port_dev); 604 } 605 606 /* inverse of do_vrf_add_slave */ 607 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 608 { 609 netdev_upper_dev_unlink(port_dev, dev); 610 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 611 612 cycle_netdev(port_dev); 613 614 return 0; 615 } 616 617 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 618 { 619 return do_vrf_del_slave(dev, port_dev); 620 } 621 622 static void vrf_dev_uninit(struct net_device *dev) 623 { 624 struct net_vrf *vrf = netdev_priv(dev); 625 struct net_device *port_dev; 626 struct list_head *iter; 627 628 vrf_rtable_release(vrf); 629 vrf_rt6_release(vrf); 630 631 netdev_for_each_lower_dev(dev, port_dev, iter) 632 vrf_del_slave(dev, port_dev); 633 634 free_percpu(dev->dstats); 635 dev->dstats = NULL; 636 } 637 638 static int vrf_dev_init(struct net_device *dev) 639 { 640 struct net_vrf *vrf = netdev_priv(dev); 641 642 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 643 if (!dev->dstats) 644 goto out_nomem; 645 646 /* create the default dst which points back to us */ 647 if (vrf_rtable_create(dev) != 0) 648 goto out_stats; 649 650 if (vrf_rt6_create(dev) != 0) 651 goto out_rth; 652 653 dev->flags = IFF_MASTER | IFF_NOARP; 654 655 /* MTU is irrelevant for VRF device; set to 64k similar to lo */ 656 dev->mtu = 64 * 1024; 657 658 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 659 dev->operstate = IF_OPER_UP; 660 661 return 0; 662 663 out_rth: 664 vrf_rtable_release(vrf); 665 out_stats: 666 free_percpu(dev->dstats); 667 dev->dstats = NULL; 668 out_nomem: 669 return -ENOMEM; 670 } 671 672 static const struct net_device_ops vrf_netdev_ops = { 673 .ndo_init = vrf_dev_init, 674 .ndo_uninit = vrf_dev_uninit, 675 .ndo_start_xmit = vrf_xmit, 676 .ndo_get_stats64 = vrf_get_stats64, 677 .ndo_add_slave = vrf_add_slave, 678 .ndo_del_slave = vrf_del_slave, 679 }; 680 681 static u32 vrf_fib_table(const struct net_device *dev) 682 { 683 struct net_vrf *vrf = netdev_priv(dev); 684 685 return vrf->tb_id; 686 } 687 688 static struct rtable *vrf_get_rtable(const struct net_device *dev, 689 const struct flowi4 *fl4) 690 { 691 struct rtable *rth = NULL; 692 693 if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) { 694 struct net_vrf *vrf = netdev_priv(dev); 695 696 rcu_read_lock(); 697 698 rth = rcu_dereference(vrf->rth); 699 if (likely(rth)) 700 dst_hold(&rth->dst); 701 702 rcu_read_unlock(); 703 } 704 705 return rth; 706 } 707 708 /* called under rcu_read_lock */ 709 static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4) 710 { 711 struct fib_result res = { .tclassid = 0 }; 712 struct net *net = dev_net(dev); 713 u32 orig_tos = fl4->flowi4_tos; 714 u8 flags = fl4->flowi4_flags; 715 u8 scope = fl4->flowi4_scope; 716 u8 tos = RT_FL_TOS(fl4); 717 int rc; 718 719 if (unlikely(!fl4->daddr)) 720 return 0; 721 722 fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF; 723 fl4->flowi4_iif = LOOPBACK_IFINDEX; 724 /* make sure oif is set to VRF device for lookup */ 725 fl4->flowi4_oif = dev->ifindex; 726 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 727 fl4->flowi4_scope = ((tos & RTO_ONLINK) ? 728 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); 729 730 rc = fib_lookup(net, fl4, &res, 0); 731 if (!rc) { 732 if (res.type == RTN_LOCAL) 733 fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr; 734 else 735 fib_select_path(net, &res, fl4, -1); 736 } 737 738 fl4->flowi4_flags = flags; 739 fl4->flowi4_tos = orig_tos; 740 fl4->flowi4_scope = scope; 741 742 return rc; 743 } 744 745 #if IS_ENABLED(CONFIG_IPV6) 746 /* neighbor handling is done with actual device; do not want 747 * to flip skb->dev for those ndisc packets. This really fails 748 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 749 * a start. 750 */ 751 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 752 { 753 const struct ipv6hdr *iph = ipv6_hdr(skb); 754 bool rc = false; 755 756 if (iph->nexthdr == NEXTHDR_ICMP) { 757 const struct icmp6hdr *icmph; 758 struct icmp6hdr _icmph; 759 760 icmph = skb_header_pointer(skb, sizeof(*iph), 761 sizeof(_icmph), &_icmph); 762 if (!icmph) 763 goto out; 764 765 switch (icmph->icmp6_type) { 766 case NDISC_ROUTER_SOLICITATION: 767 case NDISC_ROUTER_ADVERTISEMENT: 768 case NDISC_NEIGHBOUR_SOLICITATION: 769 case NDISC_NEIGHBOUR_ADVERTISEMENT: 770 case NDISC_REDIRECT: 771 rc = true; 772 break; 773 } 774 } 775 776 out: 777 return rc; 778 } 779 780 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 781 struct sk_buff *skb) 782 { 783 /* loopback traffic; do not push through packet taps again. 784 * Reset pkt_type for upper layers to process skb 785 */ 786 if (skb->pkt_type == PACKET_LOOPBACK) { 787 skb->dev = vrf_dev; 788 skb->skb_iif = vrf_dev->ifindex; 789 skb->pkt_type = PACKET_HOST; 790 goto out; 791 } 792 793 /* if packet is NDISC keep the ingress interface */ 794 if (!ipv6_ndisc_frame(skb)) { 795 skb->dev = vrf_dev; 796 skb->skb_iif = vrf_dev->ifindex; 797 798 skb_push(skb, skb->mac_len); 799 dev_queue_xmit_nit(skb, vrf_dev); 800 skb_pull(skb, skb->mac_len); 801 802 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 803 } 804 805 out: 806 return skb; 807 } 808 809 #else 810 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 811 struct sk_buff *skb) 812 { 813 return skb; 814 } 815 #endif 816 817 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 818 struct sk_buff *skb) 819 { 820 skb->dev = vrf_dev; 821 skb->skb_iif = vrf_dev->ifindex; 822 823 /* loopback traffic; do not push through packet taps again. 824 * Reset pkt_type for upper layers to process skb 825 */ 826 if (skb->pkt_type == PACKET_LOOPBACK) { 827 skb->pkt_type = PACKET_HOST; 828 goto out; 829 } 830 831 skb_push(skb, skb->mac_len); 832 dev_queue_xmit_nit(skb, vrf_dev); 833 skb_pull(skb, skb->mac_len); 834 835 out: 836 return skb; 837 } 838 839 /* called with rcu lock held */ 840 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 841 struct sk_buff *skb, 842 u16 proto) 843 { 844 switch (proto) { 845 case AF_INET: 846 return vrf_ip_rcv(vrf_dev, skb); 847 case AF_INET6: 848 return vrf_ip6_rcv(vrf_dev, skb); 849 } 850 851 return skb; 852 } 853 854 #if IS_ENABLED(CONFIG_IPV6) 855 static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev, 856 const struct flowi6 *fl6) 857 { 858 struct dst_entry *dst = NULL; 859 860 if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) { 861 struct net_vrf *vrf = netdev_priv(dev); 862 struct rt6_info *rt; 863 864 rcu_read_lock(); 865 866 rt = rcu_dereference(vrf->rt6); 867 if (likely(rt)) { 868 dst = &rt->dst; 869 dst_hold(dst); 870 } 871 872 rcu_read_unlock(); 873 } 874 875 return dst; 876 } 877 #endif 878 879 static const struct l3mdev_ops vrf_l3mdev_ops = { 880 .l3mdev_fib_table = vrf_fib_table, 881 .l3mdev_get_rtable = vrf_get_rtable, 882 .l3mdev_get_saddr = vrf_get_saddr, 883 .l3mdev_l3_rcv = vrf_l3_rcv, 884 #if IS_ENABLED(CONFIG_IPV6) 885 .l3mdev_get_rt6_dst = vrf_get_rt6_dst, 886 #endif 887 }; 888 889 static void vrf_get_drvinfo(struct net_device *dev, 890 struct ethtool_drvinfo *info) 891 { 892 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 893 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 894 } 895 896 static const struct ethtool_ops vrf_ethtool_ops = { 897 .get_drvinfo = vrf_get_drvinfo, 898 }; 899 900 static void vrf_setup(struct net_device *dev) 901 { 902 ether_setup(dev); 903 904 /* Initialize the device structure. */ 905 dev->netdev_ops = &vrf_netdev_ops; 906 dev->l3mdev_ops = &vrf_l3mdev_ops; 907 dev->ethtool_ops = &vrf_ethtool_ops; 908 dev->destructor = free_netdev; 909 910 /* Fill in device structure with ethernet-generic values. */ 911 eth_hw_addr_random(dev); 912 913 /* don't acquire vrf device's netif_tx_lock when transmitting */ 914 dev->features |= NETIF_F_LLTX; 915 916 /* don't allow vrf devices to change network namespaces. */ 917 dev->features |= NETIF_F_NETNS_LOCAL; 918 } 919 920 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) 921 { 922 if (tb[IFLA_ADDRESS]) { 923 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 924 return -EINVAL; 925 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 926 return -EADDRNOTAVAIL; 927 } 928 return 0; 929 } 930 931 static void vrf_dellink(struct net_device *dev, struct list_head *head) 932 { 933 unregister_netdevice_queue(dev, head); 934 } 935 936 static int vrf_newlink(struct net *src_net, struct net_device *dev, 937 struct nlattr *tb[], struct nlattr *data[]) 938 { 939 struct net_vrf *vrf = netdev_priv(dev); 940 941 if (!data || !data[IFLA_VRF_TABLE]) 942 return -EINVAL; 943 944 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 945 946 dev->priv_flags |= IFF_L3MDEV_MASTER; 947 948 return register_netdevice(dev); 949 } 950 951 static size_t vrf_nl_getsize(const struct net_device *dev) 952 { 953 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 954 } 955 956 static int vrf_fillinfo(struct sk_buff *skb, 957 const struct net_device *dev) 958 { 959 struct net_vrf *vrf = netdev_priv(dev); 960 961 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 962 } 963 964 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 965 const struct net_device *slave_dev) 966 { 967 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 968 } 969 970 static int vrf_fill_slave_info(struct sk_buff *skb, 971 const struct net_device *vrf_dev, 972 const struct net_device *slave_dev) 973 { 974 struct net_vrf *vrf = netdev_priv(vrf_dev); 975 976 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 977 return -EMSGSIZE; 978 979 return 0; 980 } 981 982 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 983 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 984 }; 985 986 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 987 .kind = DRV_NAME, 988 .priv_size = sizeof(struct net_vrf), 989 990 .get_size = vrf_nl_getsize, 991 .policy = vrf_nl_policy, 992 .validate = vrf_validate, 993 .fill_info = vrf_fillinfo, 994 995 .get_slave_size = vrf_get_slave_size, 996 .fill_slave_info = vrf_fill_slave_info, 997 998 .newlink = vrf_newlink, 999 .dellink = vrf_dellink, 1000 .setup = vrf_setup, 1001 .maxtype = IFLA_VRF_MAX, 1002 }; 1003 1004 static int vrf_device_event(struct notifier_block *unused, 1005 unsigned long event, void *ptr) 1006 { 1007 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1008 1009 /* only care about unregister events to drop slave references */ 1010 if (event == NETDEV_UNREGISTER) { 1011 struct net_device *vrf_dev; 1012 1013 if (!netif_is_l3_slave(dev)) 1014 goto out; 1015 1016 vrf_dev = netdev_master_upper_dev_get(dev); 1017 vrf_del_slave(vrf_dev, dev); 1018 } 1019 out: 1020 return NOTIFY_DONE; 1021 } 1022 1023 static struct notifier_block vrf_notifier_block __read_mostly = { 1024 .notifier_call = vrf_device_event, 1025 }; 1026 1027 static int __init vrf_init_module(void) 1028 { 1029 int rc; 1030 1031 register_netdevice_notifier(&vrf_notifier_block); 1032 1033 rc = rtnl_link_register(&vrf_link_ops); 1034 if (rc < 0) 1035 goto error; 1036 1037 return 0; 1038 1039 error: 1040 unregister_netdevice_notifier(&vrf_notifier_block); 1041 return rc; 1042 } 1043 1044 module_init(vrf_init_module); 1045 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 1046 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 1047 MODULE_LICENSE("GPL"); 1048 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 1049 MODULE_VERSION(DRV_VERSION); 1050