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_device *dev, 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 struct net *net = dev_net(dev); 386 struct dst_entry *dst; 387 388 RCU_INIT_POINTER(vrf->rt6, NULL); 389 RCU_INIT_POINTER(vrf->rt6_local, NULL); 390 synchronize_rcu(); 391 392 /* move dev in dst's to loopback so this VRF device can be deleted 393 * - based on dst_ifdown 394 */ 395 if (rt6) { 396 dst = &rt6->dst; 397 dev_put(dst->dev); 398 dst->dev = net->loopback_dev; 399 dev_hold(dst->dev); 400 dst_release(dst); 401 } 402 403 if (rt6_local) { 404 if (rt6_local->rt6i_idev) 405 in6_dev_put(rt6_local->rt6i_idev); 406 407 dst = &rt6_local->dst; 408 dev_put(dst->dev); 409 dst->dev = net->loopback_dev; 410 dev_hold(dst->dev); 411 dst_release(dst); 412 } 413 } 414 415 static int vrf_rt6_create(struct net_device *dev) 416 { 417 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE; 418 struct net_vrf *vrf = netdev_priv(dev); 419 struct net *net = dev_net(dev); 420 struct fib6_table *rt6i_table; 421 struct rt6_info *rt6, *rt6_local; 422 int rc = -ENOMEM; 423 424 /* IPv6 can be CONFIG enabled and then disabled runtime */ 425 if (!ipv6_mod_enabled()) 426 return 0; 427 428 rt6i_table = fib6_new_table(net, vrf->tb_id); 429 if (!rt6i_table) 430 goto out; 431 432 /* create a dst for routing packets out a VRF device */ 433 rt6 = ip6_dst_alloc(net, dev, flags); 434 if (!rt6) 435 goto out; 436 437 dst_hold(&rt6->dst); 438 439 rt6->rt6i_table = rt6i_table; 440 rt6->dst.output = vrf_output6; 441 442 /* create a dst for local routing - packets sent locally 443 * to local address via the VRF device as a loopback 444 */ 445 rt6_local = ip6_dst_alloc(net, dev, flags); 446 if (!rt6_local) { 447 dst_release(&rt6->dst); 448 goto out; 449 } 450 451 dst_hold(&rt6_local->dst); 452 453 rt6_local->rt6i_idev = in6_dev_get(dev); 454 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL; 455 rt6_local->rt6i_table = rt6i_table; 456 rt6_local->dst.input = ip6_input; 457 458 rcu_assign_pointer(vrf->rt6, rt6); 459 rcu_assign_pointer(vrf->rt6_local, rt6_local); 460 461 rc = 0; 462 out: 463 return rc; 464 } 465 #else 466 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 467 { 468 } 469 470 static int vrf_rt6_create(struct net_device *dev) 471 { 472 return 0; 473 } 474 #endif 475 476 /* modelled after ip_finish_output2 */ 477 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 478 { 479 struct dst_entry *dst = skb_dst(skb); 480 struct rtable *rt = (struct rtable *)dst; 481 struct net_device *dev = dst->dev; 482 unsigned int hh_len = LL_RESERVED_SPACE(dev); 483 struct neighbour *neigh; 484 u32 nexthop; 485 int ret = -EINVAL; 486 487 /* Be paranoid, rather than too clever. */ 488 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 489 struct sk_buff *skb2; 490 491 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 492 if (!skb2) { 493 ret = -ENOMEM; 494 goto err; 495 } 496 if (skb->sk) 497 skb_set_owner_w(skb2, skb->sk); 498 499 consume_skb(skb); 500 skb = skb2; 501 } 502 503 rcu_read_lock_bh(); 504 505 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); 506 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 507 if (unlikely(!neigh)) 508 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 509 if (!IS_ERR(neigh)) 510 ret = dst_neigh_output(dst, neigh, skb); 511 512 rcu_read_unlock_bh(); 513 err: 514 if (unlikely(ret < 0)) 515 vrf_tx_error(skb->dev, skb); 516 return ret; 517 } 518 519 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 520 { 521 struct net_device *dev = skb_dst(skb)->dev; 522 523 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 524 525 skb->dev = dev; 526 skb->protocol = htons(ETH_P_IP); 527 528 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 529 net, sk, skb, NULL, dev, 530 vrf_finish_output, 531 !(IPCB(skb)->flags & IPSKB_REROUTED)); 532 } 533 534 /* holding rtnl */ 535 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 536 { 537 struct rtable *rth = rtnl_dereference(vrf->rth); 538 struct rtable *rth_local = rtnl_dereference(vrf->rth_local); 539 struct net *net = dev_net(dev); 540 struct dst_entry *dst; 541 542 RCU_INIT_POINTER(vrf->rth, NULL); 543 RCU_INIT_POINTER(vrf->rth_local, NULL); 544 synchronize_rcu(); 545 546 /* move dev in dst's to loopback so this VRF device can be deleted 547 * - based on dst_ifdown 548 */ 549 if (rth) { 550 dst = &rth->dst; 551 dev_put(dst->dev); 552 dst->dev = net->loopback_dev; 553 dev_hold(dst->dev); 554 dst_release(dst); 555 } 556 557 if (rth_local) { 558 dst = &rth_local->dst; 559 dev_put(dst->dev); 560 dst->dev = net->loopback_dev; 561 dev_hold(dst->dev); 562 dst_release(dst); 563 } 564 } 565 566 static int vrf_rtable_create(struct net_device *dev) 567 { 568 struct net_vrf *vrf = netdev_priv(dev); 569 struct rtable *rth, *rth_local; 570 571 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 572 return -ENOMEM; 573 574 /* create a dst for routing packets out through a VRF device */ 575 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0); 576 if (!rth) 577 return -ENOMEM; 578 579 /* create a dst for local ingress routing - packets sent locally 580 * to local address via the VRF device as a loopback 581 */ 582 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0); 583 if (!rth_local) { 584 dst_release(&rth->dst); 585 return -ENOMEM; 586 } 587 588 rth->dst.output = vrf_output; 589 rth->rt_table_id = vrf->tb_id; 590 591 rth_local->rt_table_id = vrf->tb_id; 592 593 rcu_assign_pointer(vrf->rth, rth); 594 rcu_assign_pointer(vrf->rth_local, rth_local); 595 596 return 0; 597 } 598 599 /**************************** device handling ********************/ 600 601 /* cycle interface to flush neighbor cache and move routes across tables */ 602 static void cycle_netdev(struct net_device *dev) 603 { 604 unsigned int flags = dev->flags; 605 int ret; 606 607 if (!netif_running(dev)) 608 return; 609 610 ret = dev_change_flags(dev, flags & ~IFF_UP); 611 if (ret >= 0) 612 ret = dev_change_flags(dev, flags); 613 614 if (ret < 0) { 615 netdev_err(dev, 616 "Failed to cycle device %s; route tables might be wrong!\n", 617 dev->name); 618 } 619 } 620 621 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 622 { 623 int ret; 624 625 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL); 626 if (ret < 0) 627 return ret; 628 629 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 630 cycle_netdev(port_dev); 631 632 return 0; 633 } 634 635 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 636 { 637 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev)) 638 return -EINVAL; 639 640 return do_vrf_add_slave(dev, port_dev); 641 } 642 643 /* inverse of do_vrf_add_slave */ 644 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 645 { 646 netdev_upper_dev_unlink(port_dev, dev); 647 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 648 649 cycle_netdev(port_dev); 650 651 return 0; 652 } 653 654 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 655 { 656 return do_vrf_del_slave(dev, port_dev); 657 } 658 659 static void vrf_dev_uninit(struct net_device *dev) 660 { 661 struct net_vrf *vrf = netdev_priv(dev); 662 struct net_device *port_dev; 663 struct list_head *iter; 664 665 vrf_rtable_release(dev, vrf); 666 vrf_rt6_release(dev, vrf); 667 668 netdev_for_each_lower_dev(dev, port_dev, iter) 669 vrf_del_slave(dev, port_dev); 670 671 free_percpu(dev->dstats); 672 dev->dstats = NULL; 673 } 674 675 static int vrf_dev_init(struct net_device *dev) 676 { 677 struct net_vrf *vrf = netdev_priv(dev); 678 679 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 680 if (!dev->dstats) 681 goto out_nomem; 682 683 /* create the default dst which points back to us */ 684 if (vrf_rtable_create(dev) != 0) 685 goto out_stats; 686 687 if (vrf_rt6_create(dev) != 0) 688 goto out_rth; 689 690 dev->flags = IFF_MASTER | IFF_NOARP; 691 692 /* MTU is irrelevant for VRF device; set to 64k similar to lo */ 693 dev->mtu = 64 * 1024; 694 695 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 696 dev->operstate = IF_OPER_UP; 697 netdev_lockdep_set_classes(dev); 698 return 0; 699 700 out_rth: 701 vrf_rtable_release(dev, vrf); 702 out_stats: 703 free_percpu(dev->dstats); 704 dev->dstats = NULL; 705 out_nomem: 706 return -ENOMEM; 707 } 708 709 static const struct net_device_ops vrf_netdev_ops = { 710 .ndo_init = vrf_dev_init, 711 .ndo_uninit = vrf_dev_uninit, 712 .ndo_start_xmit = vrf_xmit, 713 .ndo_get_stats64 = vrf_get_stats64, 714 .ndo_add_slave = vrf_add_slave, 715 .ndo_del_slave = vrf_del_slave, 716 }; 717 718 static u32 vrf_fib_table(const struct net_device *dev) 719 { 720 struct net_vrf *vrf = netdev_priv(dev); 721 722 return vrf->tb_id; 723 } 724 725 static struct rtable *vrf_get_rtable(const struct net_device *dev, 726 const struct flowi4 *fl4) 727 { 728 struct rtable *rth = NULL; 729 730 if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) { 731 struct net_vrf *vrf = netdev_priv(dev); 732 733 rcu_read_lock(); 734 735 rth = rcu_dereference(vrf->rth); 736 if (likely(rth)) 737 dst_hold(&rth->dst); 738 739 rcu_read_unlock(); 740 } 741 742 return rth; 743 } 744 745 /* called under rcu_read_lock */ 746 static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4) 747 { 748 struct fib_result res = { .tclassid = 0 }; 749 struct net *net = dev_net(dev); 750 u32 orig_tos = fl4->flowi4_tos; 751 u8 flags = fl4->flowi4_flags; 752 u8 scope = fl4->flowi4_scope; 753 u8 tos = RT_FL_TOS(fl4); 754 int rc; 755 756 if (unlikely(!fl4->daddr)) 757 return 0; 758 759 fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF; 760 fl4->flowi4_iif = LOOPBACK_IFINDEX; 761 /* make sure oif is set to VRF device for lookup */ 762 fl4->flowi4_oif = dev->ifindex; 763 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 764 fl4->flowi4_scope = ((tos & RTO_ONLINK) ? 765 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); 766 767 rc = fib_lookup(net, fl4, &res, 0); 768 if (!rc) { 769 if (res.type == RTN_LOCAL) 770 fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr; 771 else 772 fib_select_path(net, &res, fl4, -1); 773 } 774 775 fl4->flowi4_flags = flags; 776 fl4->flowi4_tos = orig_tos; 777 fl4->flowi4_scope = scope; 778 779 return rc; 780 } 781 782 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 783 { 784 return 0; 785 } 786 787 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 788 struct sk_buff *skb, 789 struct net_device *dev) 790 { 791 struct net *net = dev_net(dev); 792 793 nf_reset(skb); 794 795 if (NF_HOOK(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) < 0) 796 skb = NULL; /* kfree_skb(skb) handled by nf code */ 797 798 return skb; 799 } 800 801 #if IS_ENABLED(CONFIG_IPV6) 802 /* neighbor handling is done with actual device; do not want 803 * to flip skb->dev for those ndisc packets. This really fails 804 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 805 * a start. 806 */ 807 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 808 { 809 const struct ipv6hdr *iph = ipv6_hdr(skb); 810 bool rc = false; 811 812 if (iph->nexthdr == NEXTHDR_ICMP) { 813 const struct icmp6hdr *icmph; 814 struct icmp6hdr _icmph; 815 816 icmph = skb_header_pointer(skb, sizeof(*iph), 817 sizeof(_icmph), &_icmph); 818 if (!icmph) 819 goto out; 820 821 switch (icmph->icmp6_type) { 822 case NDISC_ROUTER_SOLICITATION: 823 case NDISC_ROUTER_ADVERTISEMENT: 824 case NDISC_NEIGHBOUR_SOLICITATION: 825 case NDISC_NEIGHBOUR_ADVERTISEMENT: 826 case NDISC_REDIRECT: 827 rc = true; 828 break; 829 } 830 } 831 832 out: 833 return rc; 834 } 835 836 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 837 const struct net_device *dev, 838 struct flowi6 *fl6, 839 int ifindex, 840 int flags) 841 { 842 struct net_vrf *vrf = netdev_priv(dev); 843 struct fib6_table *table = NULL; 844 struct rt6_info *rt6; 845 846 rcu_read_lock(); 847 848 /* fib6_table does not have a refcnt and can not be freed */ 849 rt6 = rcu_dereference(vrf->rt6); 850 if (likely(rt6)) 851 table = rt6->rt6i_table; 852 853 rcu_read_unlock(); 854 855 if (!table) 856 return NULL; 857 858 return ip6_pol_route(net, table, ifindex, fl6, flags); 859 } 860 861 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 862 int ifindex) 863 { 864 const struct ipv6hdr *iph = ipv6_hdr(skb); 865 struct flowi6 fl6 = { 866 .daddr = iph->daddr, 867 .saddr = iph->saddr, 868 .flowlabel = ip6_flowinfo(iph), 869 .flowi6_mark = skb->mark, 870 .flowi6_proto = iph->nexthdr, 871 .flowi6_iif = ifindex, 872 }; 873 struct net *net = dev_net(vrf_dev); 874 struct rt6_info *rt6; 875 876 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, 877 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 878 if (unlikely(!rt6)) 879 return; 880 881 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 882 return; 883 884 skb_dst_set(skb, &rt6->dst); 885 } 886 887 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 888 struct sk_buff *skb) 889 { 890 int orig_iif = skb->skb_iif; 891 bool need_strict; 892 893 /* loopback traffic; do not push through packet taps again. 894 * Reset pkt_type for upper layers to process skb 895 */ 896 if (skb->pkt_type == PACKET_LOOPBACK) { 897 skb->dev = vrf_dev; 898 skb->skb_iif = vrf_dev->ifindex; 899 skb->pkt_type = PACKET_HOST; 900 goto out; 901 } 902 903 /* if packet is NDISC or addressed to multicast or link-local 904 * then keep the ingress interface 905 */ 906 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 907 if (!ipv6_ndisc_frame(skb) && !need_strict) { 908 skb->dev = vrf_dev; 909 skb->skb_iif = vrf_dev->ifindex; 910 911 skb_push(skb, skb->mac_len); 912 dev_queue_xmit_nit(skb, vrf_dev); 913 skb_pull(skb, skb->mac_len); 914 915 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 916 } 917 918 if (need_strict) 919 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 920 921 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 922 out: 923 return skb; 924 } 925 926 #else 927 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 928 struct sk_buff *skb) 929 { 930 return skb; 931 } 932 #endif 933 934 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 935 struct sk_buff *skb) 936 { 937 skb->dev = vrf_dev; 938 skb->skb_iif = vrf_dev->ifindex; 939 940 /* loopback traffic; do not push through packet taps again. 941 * Reset pkt_type for upper layers to process skb 942 */ 943 if (skb->pkt_type == PACKET_LOOPBACK) { 944 skb->pkt_type = PACKET_HOST; 945 goto out; 946 } 947 948 skb_push(skb, skb->mac_len); 949 dev_queue_xmit_nit(skb, vrf_dev); 950 skb_pull(skb, skb->mac_len); 951 952 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 953 out: 954 return skb; 955 } 956 957 /* called with rcu lock held */ 958 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 959 struct sk_buff *skb, 960 u16 proto) 961 { 962 switch (proto) { 963 case AF_INET: 964 return vrf_ip_rcv(vrf_dev, skb); 965 case AF_INET6: 966 return vrf_ip6_rcv(vrf_dev, skb); 967 } 968 969 return skb; 970 } 971 972 #if IS_ENABLED(CONFIG_IPV6) 973 static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev, 974 struct flowi6 *fl6) 975 { 976 bool need_strict = rt6_need_strict(&fl6->daddr); 977 struct net_vrf *vrf = netdev_priv(dev); 978 struct net *net = dev_net(dev); 979 struct dst_entry *dst = NULL; 980 struct rt6_info *rt; 981 982 /* send to link-local or multicast address */ 983 if (need_strict) { 984 int flags = RT6_LOOKUP_F_IFACE; 985 986 /* VRF device does not have a link-local address and 987 * sending packets to link-local or mcast addresses over 988 * a VRF device does not make sense 989 */ 990 if (fl6->flowi6_oif == dev->ifindex) { 991 struct dst_entry *dst = &net->ipv6.ip6_null_entry->dst; 992 993 dst_hold(dst); 994 return dst; 995 } 996 997 if (!ipv6_addr_any(&fl6->saddr)) 998 flags |= RT6_LOOKUP_F_HAS_SADDR; 999 1000 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags); 1001 if (rt) 1002 dst = &rt->dst; 1003 1004 } else if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) { 1005 1006 rcu_read_lock(); 1007 1008 rt = rcu_dereference(vrf->rt6); 1009 if (likely(rt)) { 1010 dst = &rt->dst; 1011 dst_hold(dst); 1012 } 1013 1014 rcu_read_unlock(); 1015 } 1016 1017 /* make sure oif is set to VRF device for lookup */ 1018 if (!need_strict) 1019 fl6->flowi6_oif = dev->ifindex; 1020 1021 return dst; 1022 } 1023 1024 /* called under rcu_read_lock */ 1025 static int vrf_get_saddr6(struct net_device *dev, const struct sock *sk, 1026 struct flowi6 *fl6) 1027 { 1028 struct net *net = dev_net(dev); 1029 struct dst_entry *dst; 1030 struct rt6_info *rt; 1031 int err; 1032 1033 if (rt6_need_strict(&fl6->daddr)) { 1034 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, 1035 RT6_LOOKUP_F_IFACE); 1036 if (unlikely(!rt)) 1037 return 0; 1038 1039 dst = &rt->dst; 1040 } else { 1041 __u8 flags = fl6->flowi6_flags; 1042 1043 fl6->flowi6_flags |= FLOWI_FLAG_L3MDEV_SRC; 1044 fl6->flowi6_flags |= FLOWI_FLAG_SKIP_NH_OIF; 1045 1046 dst = ip6_route_output(net, sk, fl6); 1047 rt = (struct rt6_info *)dst; 1048 1049 fl6->flowi6_flags = flags; 1050 } 1051 1052 err = dst->error; 1053 if (!err) { 1054 err = ip6_route_get_saddr(net, rt, &fl6->daddr, 1055 sk ? inet6_sk(sk)->srcprefs : 0, 1056 &fl6->saddr); 1057 } 1058 1059 dst_release(dst); 1060 1061 return err; 1062 } 1063 #endif 1064 1065 static const struct l3mdev_ops vrf_l3mdev_ops = { 1066 .l3mdev_fib_table = vrf_fib_table, 1067 .l3mdev_get_rtable = vrf_get_rtable, 1068 .l3mdev_get_saddr = vrf_get_saddr, 1069 .l3mdev_l3_rcv = vrf_l3_rcv, 1070 #if IS_ENABLED(CONFIG_IPV6) 1071 .l3mdev_get_rt6_dst = vrf_get_rt6_dst, 1072 .l3mdev_get_saddr6 = vrf_get_saddr6, 1073 #endif 1074 }; 1075 1076 static void vrf_get_drvinfo(struct net_device *dev, 1077 struct ethtool_drvinfo *info) 1078 { 1079 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 1080 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 1081 } 1082 1083 static const struct ethtool_ops vrf_ethtool_ops = { 1084 .get_drvinfo = vrf_get_drvinfo, 1085 }; 1086 1087 static inline size_t vrf_fib_rule_nl_size(void) 1088 { 1089 size_t sz; 1090 1091 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1092 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1093 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1094 1095 return sz; 1096 } 1097 1098 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1099 { 1100 struct fib_rule_hdr *frh; 1101 struct nlmsghdr *nlh; 1102 struct sk_buff *skb; 1103 int err; 1104 1105 if (family == AF_INET6 && !ipv6_mod_enabled()) 1106 return 0; 1107 1108 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1109 if (!skb) 1110 return -ENOMEM; 1111 1112 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1113 if (!nlh) 1114 goto nla_put_failure; 1115 1116 /* rule only needs to appear once */ 1117 nlh->nlmsg_flags &= NLM_F_EXCL; 1118 1119 frh = nlmsg_data(nlh); 1120 memset(frh, 0, sizeof(*frh)); 1121 frh->family = family; 1122 frh->action = FR_ACT_TO_TBL; 1123 1124 if (nla_put_u32(skb, FRA_L3MDEV, 1)) 1125 goto nla_put_failure; 1126 1127 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1128 goto nla_put_failure; 1129 1130 nlmsg_end(skb, nlh); 1131 1132 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1133 skb->sk = dev_net(dev)->rtnl; 1134 if (add_it) { 1135 err = fib_nl_newrule(skb, nlh); 1136 if (err == -EEXIST) 1137 err = 0; 1138 } else { 1139 err = fib_nl_delrule(skb, nlh); 1140 if (err == -ENOENT) 1141 err = 0; 1142 } 1143 nlmsg_free(skb); 1144 1145 return err; 1146 1147 nla_put_failure: 1148 nlmsg_free(skb); 1149 1150 return -EMSGSIZE; 1151 } 1152 1153 static int vrf_add_fib_rules(const struct net_device *dev) 1154 { 1155 int err; 1156 1157 err = vrf_fib_rule(dev, AF_INET, true); 1158 if (err < 0) 1159 goto out_err; 1160 1161 err = vrf_fib_rule(dev, AF_INET6, true); 1162 if (err < 0) 1163 goto ipv6_err; 1164 1165 return 0; 1166 1167 ipv6_err: 1168 vrf_fib_rule(dev, AF_INET, false); 1169 1170 out_err: 1171 netdev_err(dev, "Failed to add FIB rules.\n"); 1172 return err; 1173 } 1174 1175 static void vrf_setup(struct net_device *dev) 1176 { 1177 ether_setup(dev); 1178 1179 /* Initialize the device structure. */ 1180 dev->netdev_ops = &vrf_netdev_ops; 1181 dev->l3mdev_ops = &vrf_l3mdev_ops; 1182 dev->ethtool_ops = &vrf_ethtool_ops; 1183 dev->destructor = free_netdev; 1184 1185 /* Fill in device structure with ethernet-generic values. */ 1186 eth_hw_addr_random(dev); 1187 1188 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1189 dev->features |= NETIF_F_LLTX; 1190 1191 /* don't allow vrf devices to change network namespaces. */ 1192 dev->features |= NETIF_F_NETNS_LOCAL; 1193 1194 /* does not make sense for a VLAN to be added to a vrf device */ 1195 dev->features |= NETIF_F_VLAN_CHALLENGED; 1196 1197 /* enable offload features */ 1198 dev->features |= NETIF_F_GSO_SOFTWARE; 1199 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM; 1200 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1201 1202 dev->hw_features = dev->features; 1203 dev->hw_enc_features = dev->features; 1204 1205 /* default to no qdisc; user can add if desired */ 1206 dev->priv_flags |= IFF_NO_QUEUE; 1207 } 1208 1209 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) 1210 { 1211 if (tb[IFLA_ADDRESS]) { 1212 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1213 return -EINVAL; 1214 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1215 return -EADDRNOTAVAIL; 1216 } 1217 return 0; 1218 } 1219 1220 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1221 { 1222 unregister_netdevice_queue(dev, head); 1223 } 1224 1225 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1226 struct nlattr *tb[], struct nlattr *data[]) 1227 { 1228 struct net_vrf *vrf = netdev_priv(dev); 1229 int err; 1230 1231 if (!data || !data[IFLA_VRF_TABLE]) 1232 return -EINVAL; 1233 1234 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1235 1236 dev->priv_flags |= IFF_L3MDEV_MASTER; 1237 1238 err = register_netdevice(dev); 1239 if (err) 1240 goto out; 1241 1242 if (add_fib_rules) { 1243 err = vrf_add_fib_rules(dev); 1244 if (err) { 1245 unregister_netdevice(dev); 1246 goto out; 1247 } 1248 add_fib_rules = false; 1249 } 1250 1251 out: 1252 return err; 1253 } 1254 1255 static size_t vrf_nl_getsize(const struct net_device *dev) 1256 { 1257 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1258 } 1259 1260 static int vrf_fillinfo(struct sk_buff *skb, 1261 const struct net_device *dev) 1262 { 1263 struct net_vrf *vrf = netdev_priv(dev); 1264 1265 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1266 } 1267 1268 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1269 const struct net_device *slave_dev) 1270 { 1271 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1272 } 1273 1274 static int vrf_fill_slave_info(struct sk_buff *skb, 1275 const struct net_device *vrf_dev, 1276 const struct net_device *slave_dev) 1277 { 1278 struct net_vrf *vrf = netdev_priv(vrf_dev); 1279 1280 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1281 return -EMSGSIZE; 1282 1283 return 0; 1284 } 1285 1286 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1287 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1288 }; 1289 1290 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1291 .kind = DRV_NAME, 1292 .priv_size = sizeof(struct net_vrf), 1293 1294 .get_size = vrf_nl_getsize, 1295 .policy = vrf_nl_policy, 1296 .validate = vrf_validate, 1297 .fill_info = vrf_fillinfo, 1298 1299 .get_slave_size = vrf_get_slave_size, 1300 .fill_slave_info = vrf_fill_slave_info, 1301 1302 .newlink = vrf_newlink, 1303 .dellink = vrf_dellink, 1304 .setup = vrf_setup, 1305 .maxtype = IFLA_VRF_MAX, 1306 }; 1307 1308 static int vrf_device_event(struct notifier_block *unused, 1309 unsigned long event, void *ptr) 1310 { 1311 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1312 1313 /* only care about unregister events to drop slave references */ 1314 if (event == NETDEV_UNREGISTER) { 1315 struct net_device *vrf_dev; 1316 1317 if (!netif_is_l3_slave(dev)) 1318 goto out; 1319 1320 vrf_dev = netdev_master_upper_dev_get(dev); 1321 vrf_del_slave(vrf_dev, dev); 1322 } 1323 out: 1324 return NOTIFY_DONE; 1325 } 1326 1327 static struct notifier_block vrf_notifier_block __read_mostly = { 1328 .notifier_call = vrf_device_event, 1329 }; 1330 1331 static int __init vrf_init_module(void) 1332 { 1333 int rc; 1334 1335 register_netdevice_notifier(&vrf_notifier_block); 1336 1337 rc = rtnl_link_register(&vrf_link_ops); 1338 if (rc < 0) 1339 goto error; 1340 1341 return 0; 1342 1343 error: 1344 unregister_netdevice_notifier(&vrf_notifier_block); 1345 return rc; 1346 } 1347 1348 module_init(vrf_init_module); 1349 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 1350 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 1351 MODULE_LICENSE("GPL"); 1352 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 1353 MODULE_VERSION(DRV_VERSION); 1354