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 #include <net/fib_rules.h> 39 40 #define RT_FL_TOS(oldflp4) \ 41 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)) 42 43 #define DRV_NAME "vrf" 44 #define DRV_VERSION "1.0" 45 46 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */ 47 static bool add_fib_rules = true; 48 49 struct net_vrf { 50 struct rtable __rcu *rth; 51 struct rtable __rcu *rth_local; 52 struct rt6_info __rcu *rt6; 53 struct rt6_info __rcu *rt6_local; 54 u32 tb_id; 55 }; 56 57 struct pcpu_dstats { 58 u64 tx_pkts; 59 u64 tx_bytes; 60 u64 tx_drps; 61 u64 rx_pkts; 62 u64 rx_bytes; 63 u64 rx_drps; 64 struct u64_stats_sync syncp; 65 }; 66 67 static void vrf_rx_stats(struct net_device *dev, int len) 68 { 69 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 70 71 u64_stats_update_begin(&dstats->syncp); 72 dstats->rx_pkts++; 73 dstats->rx_bytes += len; 74 u64_stats_update_end(&dstats->syncp); 75 } 76 77 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 78 { 79 vrf_dev->stats.tx_errors++; 80 kfree_skb(skb); 81 } 82 83 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev, 84 struct rtnl_link_stats64 *stats) 85 { 86 int i; 87 88 for_each_possible_cpu(i) { 89 const struct pcpu_dstats *dstats; 90 u64 tbytes, tpkts, tdrops, rbytes, rpkts; 91 unsigned int start; 92 93 dstats = per_cpu_ptr(dev->dstats, i); 94 do { 95 start = u64_stats_fetch_begin_irq(&dstats->syncp); 96 tbytes = dstats->tx_bytes; 97 tpkts = dstats->tx_pkts; 98 tdrops = dstats->tx_drps; 99 rbytes = dstats->rx_bytes; 100 rpkts = dstats->rx_pkts; 101 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start)); 102 stats->tx_bytes += tbytes; 103 stats->tx_packets += tpkts; 104 stats->tx_dropped += tdrops; 105 stats->rx_bytes += rbytes; 106 stats->rx_packets += rpkts; 107 } 108 return stats; 109 } 110 111 /* Local traffic destined to local address. Reinsert the packet to rx 112 * path, similar to loopback handling. 113 */ 114 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 115 struct dst_entry *dst) 116 { 117 int len = skb->len; 118 119 skb_orphan(skb); 120 121 skb_dst_set(skb, dst); 122 skb_dst_force(skb); 123 124 /* set pkt_type to avoid skb hitting packet taps twice - 125 * once on Tx and again in Rx processing 126 */ 127 skb->pkt_type = PACKET_LOOPBACK; 128 129 skb->protocol = eth_type_trans(skb, dev); 130 131 if (likely(netif_rx(skb) == NET_RX_SUCCESS)) 132 vrf_rx_stats(dev, len); 133 else 134 this_cpu_inc(dev->dstats->rx_drps); 135 136 return NETDEV_TX_OK; 137 } 138 139 #if IS_ENABLED(CONFIG_IPV6) 140 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 141 struct net_device *dev) 142 { 143 const struct ipv6hdr *iph = ipv6_hdr(skb); 144 struct net *net = dev_net(skb->dev); 145 struct flowi6 fl6 = { 146 /* needed to match OIF rule */ 147 .flowi6_oif = dev->ifindex, 148 .flowi6_iif = LOOPBACK_IFINDEX, 149 .daddr = iph->daddr, 150 .saddr = iph->saddr, 151 .flowlabel = ip6_flowinfo(iph), 152 .flowi6_mark = skb->mark, 153 .flowi6_proto = iph->nexthdr, 154 .flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF, 155 }; 156 int ret = NET_XMIT_DROP; 157 struct dst_entry *dst; 158 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 159 160 dst = ip6_route_output(net, NULL, &fl6); 161 if (dst == dst_null) 162 goto err; 163 164 skb_dst_drop(skb); 165 166 /* if dst.dev is loopback or the VRF device again this is locally 167 * originated traffic destined to a local address. Short circuit 168 * to Rx path using our local dst 169 */ 170 if (dst->dev == net->loopback_dev || dst->dev == dev) { 171 struct net_vrf *vrf = netdev_priv(dev); 172 struct rt6_info *rt6_local; 173 174 /* release looked up dst and use cached local dst */ 175 dst_release(dst); 176 177 rcu_read_lock(); 178 179 rt6_local = rcu_dereference(vrf->rt6_local); 180 if (unlikely(!rt6_local)) { 181 rcu_read_unlock(); 182 goto err; 183 } 184 185 /* Ordering issue: cached local dst is created on newlink 186 * before the IPv6 initialization. Using the local dst 187 * requires rt6i_idev to be set so make sure it is. 188 */ 189 if (unlikely(!rt6_local->rt6i_idev)) { 190 rt6_local->rt6i_idev = in6_dev_get(dev); 191 if (!rt6_local->rt6i_idev) { 192 rcu_read_unlock(); 193 goto err; 194 } 195 } 196 197 dst = &rt6_local->dst; 198 dst_hold(dst); 199 200 rcu_read_unlock(); 201 202 return vrf_local_xmit(skb, dev, &rt6_local->dst); 203 } 204 205 skb_dst_set(skb, dst); 206 207 /* strip the ethernet header added for pass through VRF device */ 208 __skb_pull(skb, skb_network_offset(skb)); 209 210 ret = ip6_local_out(net, skb->sk, skb); 211 if (unlikely(net_xmit_eval(ret))) 212 dev->stats.tx_errors++; 213 else 214 ret = NET_XMIT_SUCCESS; 215 216 return ret; 217 err: 218 vrf_tx_error(dev, skb); 219 return NET_XMIT_DROP; 220 } 221 #else 222 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 223 struct net_device *dev) 224 { 225 vrf_tx_error(dev, skb); 226 return NET_XMIT_DROP; 227 } 228 #endif 229 230 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 231 struct net_device *vrf_dev) 232 { 233 struct iphdr *ip4h = ip_hdr(skb); 234 int ret = NET_XMIT_DROP; 235 struct flowi4 fl4 = { 236 /* needed to match OIF rule */ 237 .flowi4_oif = vrf_dev->ifindex, 238 .flowi4_iif = LOOPBACK_IFINDEX, 239 .flowi4_tos = RT_TOS(ip4h->tos), 240 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC | 241 FLOWI_FLAG_SKIP_NH_OIF, 242 .daddr = ip4h->daddr, 243 }; 244 struct net *net = dev_net(vrf_dev); 245 struct rtable *rt; 246 247 rt = ip_route_output_flow(net, &fl4, NULL); 248 if (IS_ERR(rt)) 249 goto err; 250 251 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { 252 ip_rt_put(rt); 253 goto err; 254 } 255 256 skb_dst_drop(skb); 257 258 /* if dst.dev is loopback or the VRF device again this is locally 259 * originated traffic destined to a local address. Short circuit 260 * to Rx path using our local dst 261 */ 262 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) { 263 struct net_vrf *vrf = netdev_priv(vrf_dev); 264 struct rtable *rth_local; 265 struct dst_entry *dst = NULL; 266 267 ip_rt_put(rt); 268 269 rcu_read_lock(); 270 271 rth_local = rcu_dereference(vrf->rth_local); 272 if (likely(rth_local)) { 273 dst = &rth_local->dst; 274 dst_hold(dst); 275 } 276 277 rcu_read_unlock(); 278 279 if (unlikely(!dst)) 280 goto err; 281 282 return vrf_local_xmit(skb, vrf_dev, dst); 283 } 284 285 skb_dst_set(skb, &rt->dst); 286 287 /* strip the ethernet header added for pass through VRF device */ 288 __skb_pull(skb, skb_network_offset(skb)); 289 290 if (!ip4h->saddr) { 291 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 292 RT_SCOPE_LINK); 293 } 294 295 ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 296 if (unlikely(net_xmit_eval(ret))) 297 vrf_dev->stats.tx_errors++; 298 else 299 ret = NET_XMIT_SUCCESS; 300 301 out: 302 return ret; 303 err: 304 vrf_tx_error(vrf_dev, skb); 305 goto out; 306 } 307 308 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 309 { 310 switch (skb->protocol) { 311 case htons(ETH_P_IP): 312 return vrf_process_v4_outbound(skb, dev); 313 case htons(ETH_P_IPV6): 314 return vrf_process_v6_outbound(skb, dev); 315 default: 316 vrf_tx_error(dev, skb); 317 return NET_XMIT_DROP; 318 } 319 } 320 321 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 322 { 323 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 324 325 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 326 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 327 328 u64_stats_update_begin(&dstats->syncp); 329 dstats->tx_pkts++; 330 dstats->tx_bytes += skb->len; 331 u64_stats_update_end(&dstats->syncp); 332 } else { 333 this_cpu_inc(dev->dstats->tx_drps); 334 } 335 336 return ret; 337 } 338 339 #if IS_ENABLED(CONFIG_IPV6) 340 /* modelled after ip6_finish_output2 */ 341 static int vrf_finish_output6(struct net *net, struct sock *sk, 342 struct sk_buff *skb) 343 { 344 struct dst_entry *dst = skb_dst(skb); 345 struct net_device *dev = dst->dev; 346 struct neighbour *neigh; 347 struct in6_addr *nexthop; 348 int ret; 349 350 skb->protocol = htons(ETH_P_IPV6); 351 skb->dev = dev; 352 353 rcu_read_lock_bh(); 354 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 355 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 356 if (unlikely(!neigh)) 357 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 358 if (!IS_ERR(neigh)) { 359 ret = dst_neigh_output(dst, neigh, skb); 360 rcu_read_unlock_bh(); 361 return ret; 362 } 363 rcu_read_unlock_bh(); 364 365 IP6_INC_STATS(dev_net(dst->dev), 366 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 367 kfree_skb(skb); 368 return -EINVAL; 369 } 370 371 /* modelled after ip6_output */ 372 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 373 { 374 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 375 net, sk, skb, NULL, skb_dst(skb)->dev, 376 vrf_finish_output6, 377 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 378 } 379 380 /* holding rtnl */ 381 static void vrf_rt6_release(struct net_vrf *vrf) 382 { 383 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 384 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local); 385 386 RCU_INIT_POINTER(vrf->rt6, NULL); 387 RCU_INIT_POINTER(vrf->rt6_local, NULL); 388 synchronize_rcu(); 389 390 if (rt6) 391 dst_release(&rt6->dst); 392 393 if (rt6_local) { 394 if (rt6_local->rt6i_idev) 395 in6_dev_put(rt6_local->rt6i_idev); 396 397 dst_release(&rt6_local->dst); 398 } 399 } 400 401 static int vrf_rt6_create(struct net_device *dev) 402 { 403 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE; 404 struct net_vrf *vrf = netdev_priv(dev); 405 struct net *net = dev_net(dev); 406 struct fib6_table *rt6i_table; 407 struct rt6_info *rt6, *rt6_local; 408 int rc = -ENOMEM; 409 410 /* IPv6 can be CONFIG enabled and then disabled runtime */ 411 if (!ipv6_mod_enabled()) 412 return 0; 413 414 rt6i_table = fib6_new_table(net, vrf->tb_id); 415 if (!rt6i_table) 416 goto out; 417 418 /* create a dst for routing packets out a VRF device */ 419 rt6 = ip6_dst_alloc(net, dev, flags); 420 if (!rt6) 421 goto out; 422 423 dst_hold(&rt6->dst); 424 425 rt6->rt6i_table = rt6i_table; 426 rt6->dst.output = vrf_output6; 427 428 /* create a dst for local routing - packets sent locally 429 * to local address via the VRF device as a loopback 430 */ 431 rt6_local = ip6_dst_alloc(net, dev, flags); 432 if (!rt6_local) { 433 dst_release(&rt6->dst); 434 goto out; 435 } 436 437 dst_hold(&rt6_local->dst); 438 439 rt6_local->rt6i_idev = in6_dev_get(dev); 440 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL; 441 rt6_local->rt6i_table = rt6i_table; 442 rt6_local->dst.input = ip6_input; 443 444 rcu_assign_pointer(vrf->rt6, rt6); 445 rcu_assign_pointer(vrf->rt6_local, rt6_local); 446 447 rc = 0; 448 out: 449 return rc; 450 } 451 #else 452 static void vrf_rt6_release(struct net_vrf *vrf) 453 { 454 } 455 456 static int vrf_rt6_create(struct net_device *dev) 457 { 458 return 0; 459 } 460 #endif 461 462 /* modelled after ip_finish_output2 */ 463 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 464 { 465 struct dst_entry *dst = skb_dst(skb); 466 struct rtable *rt = (struct rtable *)dst; 467 struct net_device *dev = dst->dev; 468 unsigned int hh_len = LL_RESERVED_SPACE(dev); 469 struct neighbour *neigh; 470 u32 nexthop; 471 int ret = -EINVAL; 472 473 /* Be paranoid, rather than too clever. */ 474 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 475 struct sk_buff *skb2; 476 477 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 478 if (!skb2) { 479 ret = -ENOMEM; 480 goto err; 481 } 482 if (skb->sk) 483 skb_set_owner_w(skb2, skb->sk); 484 485 consume_skb(skb); 486 skb = skb2; 487 } 488 489 rcu_read_lock_bh(); 490 491 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); 492 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 493 if (unlikely(!neigh)) 494 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 495 if (!IS_ERR(neigh)) 496 ret = dst_neigh_output(dst, neigh, skb); 497 498 rcu_read_unlock_bh(); 499 err: 500 if (unlikely(ret < 0)) 501 vrf_tx_error(skb->dev, skb); 502 return ret; 503 } 504 505 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 506 { 507 struct net_device *dev = skb_dst(skb)->dev; 508 509 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 510 511 skb->dev = dev; 512 skb->protocol = htons(ETH_P_IP); 513 514 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 515 net, sk, skb, NULL, dev, 516 vrf_finish_output, 517 !(IPCB(skb)->flags & IPSKB_REROUTED)); 518 } 519 520 /* holding rtnl */ 521 static void vrf_rtable_release(struct net_vrf *vrf) 522 { 523 struct rtable *rth = rtnl_dereference(vrf->rth); 524 struct rtable *rth_local = rtnl_dereference(vrf->rth_local); 525 526 RCU_INIT_POINTER(vrf->rth, NULL); 527 RCU_INIT_POINTER(vrf->rth_local, NULL); 528 synchronize_rcu(); 529 530 if (rth) 531 dst_release(&rth->dst); 532 533 if (rth_local) 534 dst_release(&rth_local->dst); 535 } 536 537 static int vrf_rtable_create(struct net_device *dev) 538 { 539 struct net_vrf *vrf = netdev_priv(dev); 540 struct rtable *rth, *rth_local; 541 542 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 543 return -ENOMEM; 544 545 /* create a dst for routing packets out through a VRF device */ 546 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0); 547 if (!rth) 548 return -ENOMEM; 549 550 /* create a dst for local ingress routing - packets sent locally 551 * to local address via the VRF device as a loopback 552 */ 553 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0); 554 if (!rth_local) { 555 dst_release(&rth->dst); 556 return -ENOMEM; 557 } 558 559 rth->dst.output = vrf_output; 560 rth->rt_table_id = vrf->tb_id; 561 562 rth_local->rt_table_id = vrf->tb_id; 563 564 rcu_assign_pointer(vrf->rth, rth); 565 rcu_assign_pointer(vrf->rth_local, rth_local); 566 567 return 0; 568 } 569 570 /**************************** device handling ********************/ 571 572 /* cycle interface to flush neighbor cache and move routes across tables */ 573 static void cycle_netdev(struct net_device *dev) 574 { 575 unsigned int flags = dev->flags; 576 int ret; 577 578 if (!netif_running(dev)) 579 return; 580 581 ret = dev_change_flags(dev, flags & ~IFF_UP); 582 if (ret >= 0) 583 ret = dev_change_flags(dev, flags); 584 585 if (ret < 0) { 586 netdev_err(dev, 587 "Failed to cycle device %s; route tables might be wrong!\n", 588 dev->name); 589 } 590 } 591 592 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 593 { 594 int ret; 595 596 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL); 597 if (ret < 0) 598 return ret; 599 600 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 601 cycle_netdev(port_dev); 602 603 return 0; 604 } 605 606 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 607 { 608 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev)) 609 return -EINVAL; 610 611 return do_vrf_add_slave(dev, port_dev); 612 } 613 614 /* inverse of do_vrf_add_slave */ 615 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 616 { 617 netdev_upper_dev_unlink(port_dev, dev); 618 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 619 620 cycle_netdev(port_dev); 621 622 return 0; 623 } 624 625 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 626 { 627 return do_vrf_del_slave(dev, port_dev); 628 } 629 630 static void vrf_dev_uninit(struct net_device *dev) 631 { 632 struct net_vrf *vrf = netdev_priv(dev); 633 struct net_device *port_dev; 634 struct list_head *iter; 635 636 vrf_rtable_release(vrf); 637 vrf_rt6_release(vrf); 638 639 netdev_for_each_lower_dev(dev, port_dev, iter) 640 vrf_del_slave(dev, port_dev); 641 642 free_percpu(dev->dstats); 643 dev->dstats = NULL; 644 } 645 646 static int vrf_dev_init(struct net_device *dev) 647 { 648 struct net_vrf *vrf = netdev_priv(dev); 649 650 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 651 if (!dev->dstats) 652 goto out_nomem; 653 654 /* create the default dst which points back to us */ 655 if (vrf_rtable_create(dev) != 0) 656 goto out_stats; 657 658 if (vrf_rt6_create(dev) != 0) 659 goto out_rth; 660 661 dev->flags = IFF_MASTER | IFF_NOARP; 662 663 /* MTU is irrelevant for VRF device; set to 64k similar to lo */ 664 dev->mtu = 64 * 1024; 665 666 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 667 dev->operstate = IF_OPER_UP; 668 netdev_lockdep_set_classes(dev); 669 return 0; 670 671 out_rth: 672 vrf_rtable_release(vrf); 673 out_stats: 674 free_percpu(dev->dstats); 675 dev->dstats = NULL; 676 out_nomem: 677 return -ENOMEM; 678 } 679 680 static const struct net_device_ops vrf_netdev_ops = { 681 .ndo_init = vrf_dev_init, 682 .ndo_uninit = vrf_dev_uninit, 683 .ndo_start_xmit = vrf_xmit, 684 .ndo_get_stats64 = vrf_get_stats64, 685 .ndo_add_slave = vrf_add_slave, 686 .ndo_del_slave = vrf_del_slave, 687 }; 688 689 static u32 vrf_fib_table(const struct net_device *dev) 690 { 691 struct net_vrf *vrf = netdev_priv(dev); 692 693 return vrf->tb_id; 694 } 695 696 static struct rtable *vrf_get_rtable(const struct net_device *dev, 697 const struct flowi4 *fl4) 698 { 699 struct rtable *rth = NULL; 700 701 if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) { 702 struct net_vrf *vrf = netdev_priv(dev); 703 704 rcu_read_lock(); 705 706 rth = rcu_dereference(vrf->rth); 707 if (likely(rth)) 708 dst_hold(&rth->dst); 709 710 rcu_read_unlock(); 711 } 712 713 return rth; 714 } 715 716 /* called under rcu_read_lock */ 717 static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4) 718 { 719 struct fib_result res = { .tclassid = 0 }; 720 struct net *net = dev_net(dev); 721 u32 orig_tos = fl4->flowi4_tos; 722 u8 flags = fl4->flowi4_flags; 723 u8 scope = fl4->flowi4_scope; 724 u8 tos = RT_FL_TOS(fl4); 725 int rc; 726 727 if (unlikely(!fl4->daddr)) 728 return 0; 729 730 fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF; 731 fl4->flowi4_iif = LOOPBACK_IFINDEX; 732 /* make sure oif is set to VRF device for lookup */ 733 fl4->flowi4_oif = dev->ifindex; 734 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 735 fl4->flowi4_scope = ((tos & RTO_ONLINK) ? 736 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); 737 738 rc = fib_lookup(net, fl4, &res, 0); 739 if (!rc) { 740 if (res.type == RTN_LOCAL) 741 fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr; 742 else 743 fib_select_path(net, &res, fl4, -1); 744 } 745 746 fl4->flowi4_flags = flags; 747 fl4->flowi4_tos = orig_tos; 748 fl4->flowi4_scope = scope; 749 750 return rc; 751 } 752 753 #if IS_ENABLED(CONFIG_IPV6) 754 /* neighbor handling is done with actual device; do not want 755 * to flip skb->dev for those ndisc packets. This really fails 756 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 757 * a start. 758 */ 759 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 760 { 761 const struct ipv6hdr *iph = ipv6_hdr(skb); 762 bool rc = false; 763 764 if (iph->nexthdr == NEXTHDR_ICMP) { 765 const struct icmp6hdr *icmph; 766 struct icmp6hdr _icmph; 767 768 icmph = skb_header_pointer(skb, sizeof(*iph), 769 sizeof(_icmph), &_icmph); 770 if (!icmph) 771 goto out; 772 773 switch (icmph->icmp6_type) { 774 case NDISC_ROUTER_SOLICITATION: 775 case NDISC_ROUTER_ADVERTISEMENT: 776 case NDISC_NEIGHBOUR_SOLICITATION: 777 case NDISC_NEIGHBOUR_ADVERTISEMENT: 778 case NDISC_REDIRECT: 779 rc = true; 780 break; 781 } 782 } 783 784 out: 785 return rc; 786 } 787 788 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 789 struct sk_buff *skb) 790 { 791 /* loopback traffic; do not push through packet taps again. 792 * Reset pkt_type for upper layers to process skb 793 */ 794 if (skb->pkt_type == PACKET_LOOPBACK) { 795 skb->dev = vrf_dev; 796 skb->skb_iif = vrf_dev->ifindex; 797 skb->pkt_type = PACKET_HOST; 798 goto out; 799 } 800 801 /* if packet is NDISC keep the ingress interface */ 802 if (!ipv6_ndisc_frame(skb)) { 803 skb->dev = vrf_dev; 804 skb->skb_iif = vrf_dev->ifindex; 805 806 skb_push(skb, skb->mac_len); 807 dev_queue_xmit_nit(skb, vrf_dev); 808 skb_pull(skb, skb->mac_len); 809 810 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 811 } 812 813 out: 814 return skb; 815 } 816 817 #else 818 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 819 struct sk_buff *skb) 820 { 821 return skb; 822 } 823 #endif 824 825 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 826 struct sk_buff *skb) 827 { 828 skb->dev = vrf_dev; 829 skb->skb_iif = vrf_dev->ifindex; 830 831 /* loopback traffic; do not push through packet taps again. 832 * Reset pkt_type for upper layers to process skb 833 */ 834 if (skb->pkt_type == PACKET_LOOPBACK) { 835 skb->pkt_type = PACKET_HOST; 836 goto out; 837 } 838 839 skb_push(skb, skb->mac_len); 840 dev_queue_xmit_nit(skb, vrf_dev); 841 skb_pull(skb, skb->mac_len); 842 843 out: 844 return skb; 845 } 846 847 /* called with rcu lock held */ 848 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 849 struct sk_buff *skb, 850 u16 proto) 851 { 852 switch (proto) { 853 case AF_INET: 854 return vrf_ip_rcv(vrf_dev, skb); 855 case AF_INET6: 856 return vrf_ip6_rcv(vrf_dev, skb); 857 } 858 859 return skb; 860 } 861 862 #if IS_ENABLED(CONFIG_IPV6) 863 static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev, 864 struct flowi6 *fl6) 865 { 866 struct dst_entry *dst = NULL; 867 868 if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) { 869 struct net_vrf *vrf = netdev_priv(dev); 870 struct rt6_info *rt; 871 872 rcu_read_lock(); 873 874 rt = rcu_dereference(vrf->rt6); 875 if (likely(rt)) { 876 dst = &rt->dst; 877 dst_hold(dst); 878 } 879 880 rcu_read_unlock(); 881 } 882 883 return dst; 884 } 885 #endif 886 887 static const struct l3mdev_ops vrf_l3mdev_ops = { 888 .l3mdev_fib_table = vrf_fib_table, 889 .l3mdev_get_rtable = vrf_get_rtable, 890 .l3mdev_get_saddr = vrf_get_saddr, 891 .l3mdev_l3_rcv = vrf_l3_rcv, 892 #if IS_ENABLED(CONFIG_IPV6) 893 .l3mdev_get_rt6_dst = vrf_get_rt6_dst, 894 #endif 895 }; 896 897 static void vrf_get_drvinfo(struct net_device *dev, 898 struct ethtool_drvinfo *info) 899 { 900 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 901 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 902 } 903 904 static const struct ethtool_ops vrf_ethtool_ops = { 905 .get_drvinfo = vrf_get_drvinfo, 906 }; 907 908 static inline size_t vrf_fib_rule_nl_size(void) 909 { 910 size_t sz; 911 912 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 913 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 914 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 915 916 return sz; 917 } 918 919 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 920 { 921 struct fib_rule_hdr *frh; 922 struct nlmsghdr *nlh; 923 struct sk_buff *skb; 924 int err; 925 926 if (family == AF_INET6 && !ipv6_mod_enabled()) 927 return 0; 928 929 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 930 if (!skb) 931 return -ENOMEM; 932 933 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 934 if (!nlh) 935 goto nla_put_failure; 936 937 /* rule only needs to appear once */ 938 nlh->nlmsg_flags &= NLM_F_EXCL; 939 940 frh = nlmsg_data(nlh); 941 memset(frh, 0, sizeof(*frh)); 942 frh->family = family; 943 frh->action = FR_ACT_TO_TBL; 944 945 if (nla_put_u32(skb, FRA_L3MDEV, 1)) 946 goto nla_put_failure; 947 948 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 949 goto nla_put_failure; 950 951 nlmsg_end(skb, nlh); 952 953 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 954 skb->sk = dev_net(dev)->rtnl; 955 if (add_it) { 956 err = fib_nl_newrule(skb, nlh); 957 if (err == -EEXIST) 958 err = 0; 959 } else { 960 err = fib_nl_delrule(skb, nlh); 961 if (err == -ENOENT) 962 err = 0; 963 } 964 nlmsg_free(skb); 965 966 return err; 967 968 nla_put_failure: 969 nlmsg_free(skb); 970 971 return -EMSGSIZE; 972 } 973 974 static int vrf_add_fib_rules(const struct net_device *dev) 975 { 976 int err; 977 978 err = vrf_fib_rule(dev, AF_INET, true); 979 if (err < 0) 980 goto out_err; 981 982 err = vrf_fib_rule(dev, AF_INET6, true); 983 if (err < 0) 984 goto ipv6_err; 985 986 return 0; 987 988 ipv6_err: 989 vrf_fib_rule(dev, AF_INET, false); 990 991 out_err: 992 netdev_err(dev, "Failed to add FIB rules.\n"); 993 return err; 994 } 995 996 static void vrf_setup(struct net_device *dev) 997 { 998 ether_setup(dev); 999 1000 /* Initialize the device structure. */ 1001 dev->netdev_ops = &vrf_netdev_ops; 1002 dev->l3mdev_ops = &vrf_l3mdev_ops; 1003 dev->ethtool_ops = &vrf_ethtool_ops; 1004 dev->destructor = free_netdev; 1005 1006 /* Fill in device structure with ethernet-generic values. */ 1007 eth_hw_addr_random(dev); 1008 1009 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1010 dev->features |= NETIF_F_LLTX; 1011 1012 /* don't allow vrf devices to change network namespaces. */ 1013 dev->features |= NETIF_F_NETNS_LOCAL; 1014 } 1015 1016 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) 1017 { 1018 if (tb[IFLA_ADDRESS]) { 1019 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1020 return -EINVAL; 1021 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1022 return -EADDRNOTAVAIL; 1023 } 1024 return 0; 1025 } 1026 1027 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1028 { 1029 unregister_netdevice_queue(dev, head); 1030 } 1031 1032 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1033 struct nlattr *tb[], struct nlattr *data[]) 1034 { 1035 struct net_vrf *vrf = netdev_priv(dev); 1036 int err; 1037 1038 if (!data || !data[IFLA_VRF_TABLE]) 1039 return -EINVAL; 1040 1041 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1042 1043 dev->priv_flags |= IFF_L3MDEV_MASTER; 1044 1045 err = register_netdevice(dev); 1046 if (err) 1047 goto out; 1048 1049 if (add_fib_rules) { 1050 err = vrf_add_fib_rules(dev); 1051 if (err) { 1052 unregister_netdevice(dev); 1053 goto out; 1054 } 1055 add_fib_rules = false; 1056 } 1057 1058 out: 1059 return err; 1060 } 1061 1062 static size_t vrf_nl_getsize(const struct net_device *dev) 1063 { 1064 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1065 } 1066 1067 static int vrf_fillinfo(struct sk_buff *skb, 1068 const struct net_device *dev) 1069 { 1070 struct net_vrf *vrf = netdev_priv(dev); 1071 1072 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1073 } 1074 1075 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1076 const struct net_device *slave_dev) 1077 { 1078 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1079 } 1080 1081 static int vrf_fill_slave_info(struct sk_buff *skb, 1082 const struct net_device *vrf_dev, 1083 const struct net_device *slave_dev) 1084 { 1085 struct net_vrf *vrf = netdev_priv(vrf_dev); 1086 1087 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1088 return -EMSGSIZE; 1089 1090 return 0; 1091 } 1092 1093 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1094 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1095 }; 1096 1097 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1098 .kind = DRV_NAME, 1099 .priv_size = sizeof(struct net_vrf), 1100 1101 .get_size = vrf_nl_getsize, 1102 .policy = vrf_nl_policy, 1103 .validate = vrf_validate, 1104 .fill_info = vrf_fillinfo, 1105 1106 .get_slave_size = vrf_get_slave_size, 1107 .fill_slave_info = vrf_fill_slave_info, 1108 1109 .newlink = vrf_newlink, 1110 .dellink = vrf_dellink, 1111 .setup = vrf_setup, 1112 .maxtype = IFLA_VRF_MAX, 1113 }; 1114 1115 static int vrf_device_event(struct notifier_block *unused, 1116 unsigned long event, void *ptr) 1117 { 1118 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1119 1120 /* only care about unregister events to drop slave references */ 1121 if (event == NETDEV_UNREGISTER) { 1122 struct net_device *vrf_dev; 1123 1124 if (!netif_is_l3_slave(dev)) 1125 goto out; 1126 1127 vrf_dev = netdev_master_upper_dev_get(dev); 1128 vrf_del_slave(vrf_dev, dev); 1129 } 1130 out: 1131 return NOTIFY_DONE; 1132 } 1133 1134 static struct notifier_block vrf_notifier_block __read_mostly = { 1135 .notifier_call = vrf_device_event, 1136 }; 1137 1138 static int __init vrf_init_module(void) 1139 { 1140 int rc; 1141 1142 register_netdevice_notifier(&vrf_notifier_block); 1143 1144 rc = rtnl_link_register(&vrf_link_ops); 1145 if (rc < 0) 1146 goto error; 1147 1148 return 0; 1149 1150 error: 1151 unregister_netdevice_notifier(&vrf_notifier_block); 1152 return rc; 1153 } 1154 1155 module_init(vrf_init_module); 1156 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 1157 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 1158 MODULE_LICENSE("GPL"); 1159 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 1160 MODULE_VERSION(DRV_VERSION); 1161