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 DRV_NAME "vrf" 41 #define DRV_VERSION "1.0" 42 43 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */ 44 static bool add_fib_rules = true; 45 46 struct net_vrf { 47 struct rtable __rcu *rth; 48 struct rtable __rcu *rth_local; 49 struct rt6_info __rcu *rt6; 50 struct rt6_info __rcu *rt6_local; 51 u32 tb_id; 52 }; 53 54 struct pcpu_dstats { 55 u64 tx_pkts; 56 u64 tx_bytes; 57 u64 tx_drps; 58 u64 rx_pkts; 59 u64 rx_bytes; 60 u64 rx_drps; 61 struct u64_stats_sync syncp; 62 }; 63 64 static void vrf_rx_stats(struct net_device *dev, int len) 65 { 66 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 67 68 u64_stats_update_begin(&dstats->syncp); 69 dstats->rx_pkts++; 70 dstats->rx_bytes += len; 71 u64_stats_update_end(&dstats->syncp); 72 } 73 74 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 75 { 76 vrf_dev->stats.tx_errors++; 77 kfree_skb(skb); 78 } 79 80 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev, 81 struct rtnl_link_stats64 *stats) 82 { 83 int i; 84 85 for_each_possible_cpu(i) { 86 const struct pcpu_dstats *dstats; 87 u64 tbytes, tpkts, tdrops, rbytes, rpkts; 88 unsigned int start; 89 90 dstats = per_cpu_ptr(dev->dstats, i); 91 do { 92 start = u64_stats_fetch_begin_irq(&dstats->syncp); 93 tbytes = dstats->tx_bytes; 94 tpkts = dstats->tx_pkts; 95 tdrops = dstats->tx_drps; 96 rbytes = dstats->rx_bytes; 97 rpkts = dstats->rx_pkts; 98 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start)); 99 stats->tx_bytes += tbytes; 100 stats->tx_packets += tpkts; 101 stats->tx_dropped += tdrops; 102 stats->rx_bytes += rbytes; 103 stats->rx_packets += rpkts; 104 } 105 return stats; 106 } 107 108 /* Local traffic destined to local address. Reinsert the packet to rx 109 * path, similar to loopback handling. 110 */ 111 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 112 struct dst_entry *dst) 113 { 114 int len = skb->len; 115 116 skb_orphan(skb); 117 118 skb_dst_set(skb, dst); 119 skb_dst_force(skb); 120 121 /* set pkt_type to avoid skb hitting packet taps twice - 122 * once on Tx and again in Rx processing 123 */ 124 skb->pkt_type = PACKET_LOOPBACK; 125 126 skb->protocol = eth_type_trans(skb, dev); 127 128 if (likely(netif_rx(skb) == NET_RX_SUCCESS)) 129 vrf_rx_stats(dev, len); 130 else 131 this_cpu_inc(dev->dstats->rx_drps); 132 133 return NETDEV_TX_OK; 134 } 135 136 #if IS_ENABLED(CONFIG_IPV6) 137 static int vrf_ip6_local_out(struct net *net, struct sock *sk, 138 struct sk_buff *skb) 139 { 140 int err; 141 142 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, 143 sk, skb, NULL, skb_dst(skb)->dev, dst_output); 144 145 if (likely(err == 1)) 146 err = dst_output(net, sk, skb); 147 148 return err; 149 } 150 151 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 152 struct net_device *dev) 153 { 154 const struct ipv6hdr *iph = ipv6_hdr(skb); 155 struct net *net = dev_net(skb->dev); 156 struct flowi6 fl6 = { 157 /* needed to match OIF rule */ 158 .flowi6_oif = dev->ifindex, 159 .flowi6_iif = LOOPBACK_IFINDEX, 160 .daddr = iph->daddr, 161 .saddr = iph->saddr, 162 .flowlabel = ip6_flowinfo(iph), 163 .flowi6_mark = skb->mark, 164 .flowi6_proto = iph->nexthdr, 165 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF, 166 }; 167 int ret = NET_XMIT_DROP; 168 struct dst_entry *dst; 169 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 170 171 dst = ip6_route_output(net, NULL, &fl6); 172 if (dst == dst_null) 173 goto err; 174 175 skb_dst_drop(skb); 176 177 /* if dst.dev is loopback or the VRF device again this is locally 178 * originated traffic destined to a local address. Short circuit 179 * to Rx path using our local dst 180 */ 181 if (dst->dev == net->loopback_dev || dst->dev == dev) { 182 struct net_vrf *vrf = netdev_priv(dev); 183 struct rt6_info *rt6_local; 184 185 /* release looked up dst and use cached local dst */ 186 dst_release(dst); 187 188 rcu_read_lock(); 189 190 rt6_local = rcu_dereference(vrf->rt6_local); 191 if (unlikely(!rt6_local)) { 192 rcu_read_unlock(); 193 goto err; 194 } 195 196 /* Ordering issue: cached local dst is created on newlink 197 * before the IPv6 initialization. Using the local dst 198 * requires rt6i_idev to be set so make sure it is. 199 */ 200 if (unlikely(!rt6_local->rt6i_idev)) { 201 rt6_local->rt6i_idev = in6_dev_get(dev); 202 if (!rt6_local->rt6i_idev) { 203 rcu_read_unlock(); 204 goto err; 205 } 206 } 207 208 dst = &rt6_local->dst; 209 dst_hold(dst); 210 211 rcu_read_unlock(); 212 213 return vrf_local_xmit(skb, dev, &rt6_local->dst); 214 } 215 216 skb_dst_set(skb, dst); 217 218 /* strip the ethernet header added for pass through VRF device */ 219 __skb_pull(skb, skb_network_offset(skb)); 220 221 ret = vrf_ip6_local_out(net, skb->sk, skb); 222 if (unlikely(net_xmit_eval(ret))) 223 dev->stats.tx_errors++; 224 else 225 ret = NET_XMIT_SUCCESS; 226 227 return ret; 228 err: 229 vrf_tx_error(dev, skb); 230 return NET_XMIT_DROP; 231 } 232 #else 233 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 234 struct net_device *dev) 235 { 236 vrf_tx_error(dev, skb); 237 return NET_XMIT_DROP; 238 } 239 #endif 240 241 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 242 static int vrf_ip_local_out(struct net *net, struct sock *sk, 243 struct sk_buff *skb) 244 { 245 int err; 246 247 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 248 skb, NULL, skb_dst(skb)->dev, dst_output); 249 if (likely(err == 1)) 250 err = dst_output(net, sk, skb); 251 252 return err; 253 } 254 255 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 256 struct net_device *vrf_dev) 257 { 258 struct iphdr *ip4h = ip_hdr(skb); 259 int ret = NET_XMIT_DROP; 260 struct flowi4 fl4 = { 261 /* needed to match OIF rule */ 262 .flowi4_oif = vrf_dev->ifindex, 263 .flowi4_iif = LOOPBACK_IFINDEX, 264 .flowi4_tos = RT_TOS(ip4h->tos), 265 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF, 266 .daddr = ip4h->daddr, 267 }; 268 struct net *net = dev_net(vrf_dev); 269 struct rtable *rt; 270 271 rt = ip_route_output_flow(net, &fl4, NULL); 272 if (IS_ERR(rt)) 273 goto err; 274 275 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { 276 ip_rt_put(rt); 277 goto err; 278 } 279 280 skb_dst_drop(skb); 281 282 /* if dst.dev is loopback or the VRF device again this is locally 283 * originated traffic destined to a local address. Short circuit 284 * to Rx path using our local dst 285 */ 286 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) { 287 struct net_vrf *vrf = netdev_priv(vrf_dev); 288 struct rtable *rth_local; 289 struct dst_entry *dst = NULL; 290 291 ip_rt_put(rt); 292 293 rcu_read_lock(); 294 295 rth_local = rcu_dereference(vrf->rth_local); 296 if (likely(rth_local)) { 297 dst = &rth_local->dst; 298 dst_hold(dst); 299 } 300 301 rcu_read_unlock(); 302 303 if (unlikely(!dst)) 304 goto err; 305 306 return vrf_local_xmit(skb, vrf_dev, dst); 307 } 308 309 skb_dst_set(skb, &rt->dst); 310 311 /* strip the ethernet header added for pass through VRF device */ 312 __skb_pull(skb, skb_network_offset(skb)); 313 314 if (!ip4h->saddr) { 315 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 316 RT_SCOPE_LINK); 317 } 318 319 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 320 if (unlikely(net_xmit_eval(ret))) 321 vrf_dev->stats.tx_errors++; 322 else 323 ret = NET_XMIT_SUCCESS; 324 325 out: 326 return ret; 327 err: 328 vrf_tx_error(vrf_dev, skb); 329 goto out; 330 } 331 332 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 333 { 334 switch (skb->protocol) { 335 case htons(ETH_P_IP): 336 return vrf_process_v4_outbound(skb, dev); 337 case htons(ETH_P_IPV6): 338 return vrf_process_v6_outbound(skb, dev); 339 default: 340 vrf_tx_error(dev, skb); 341 return NET_XMIT_DROP; 342 } 343 } 344 345 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 346 { 347 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 348 349 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 350 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 351 352 u64_stats_update_begin(&dstats->syncp); 353 dstats->tx_pkts++; 354 dstats->tx_bytes += skb->len; 355 u64_stats_update_end(&dstats->syncp); 356 } else { 357 this_cpu_inc(dev->dstats->tx_drps); 358 } 359 360 return ret; 361 } 362 363 #if IS_ENABLED(CONFIG_IPV6) 364 /* modelled after ip6_finish_output2 */ 365 static int vrf_finish_output6(struct net *net, struct sock *sk, 366 struct sk_buff *skb) 367 { 368 struct dst_entry *dst = skb_dst(skb); 369 struct net_device *dev = dst->dev; 370 struct neighbour *neigh; 371 struct in6_addr *nexthop; 372 int ret; 373 374 skb->protocol = htons(ETH_P_IPV6); 375 skb->dev = dev; 376 377 rcu_read_lock_bh(); 378 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 379 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 380 if (unlikely(!neigh)) 381 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 382 if (!IS_ERR(neigh)) { 383 ret = dst_neigh_output(dst, neigh, skb); 384 rcu_read_unlock_bh(); 385 return ret; 386 } 387 rcu_read_unlock_bh(); 388 389 IP6_INC_STATS(dev_net(dst->dev), 390 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 391 kfree_skb(skb); 392 return -EINVAL; 393 } 394 395 /* modelled after ip6_output */ 396 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 397 { 398 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 399 net, sk, skb, NULL, skb_dst(skb)->dev, 400 vrf_finish_output6, 401 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 402 } 403 404 /* set dst on skb to send packet to us via dev_xmit path. Allows 405 * packet to go through device based features such as qdisc, netfilter 406 * hooks and packet sockets with skb->dev set to vrf device. 407 */ 408 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 409 struct sock *sk, 410 struct sk_buff *skb) 411 { 412 struct net_vrf *vrf = netdev_priv(vrf_dev); 413 struct dst_entry *dst = NULL; 414 struct rt6_info *rt6; 415 416 /* don't divert link scope packets */ 417 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 418 return skb; 419 420 rcu_read_lock(); 421 422 rt6 = rcu_dereference(vrf->rt6); 423 if (likely(rt6)) { 424 dst = &rt6->dst; 425 dst_hold(dst); 426 } 427 428 rcu_read_unlock(); 429 430 if (unlikely(!dst)) { 431 vrf_tx_error(vrf_dev, skb); 432 return NULL; 433 } 434 435 skb_dst_drop(skb); 436 skb_dst_set(skb, dst); 437 438 return skb; 439 } 440 441 /* holding rtnl */ 442 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 443 { 444 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 445 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local); 446 struct net *net = dev_net(dev); 447 struct dst_entry *dst; 448 449 RCU_INIT_POINTER(vrf->rt6, NULL); 450 RCU_INIT_POINTER(vrf->rt6_local, NULL); 451 synchronize_rcu(); 452 453 /* move dev in dst's to loopback so this VRF device can be deleted 454 * - based on dst_ifdown 455 */ 456 if (rt6) { 457 dst = &rt6->dst; 458 dev_put(dst->dev); 459 dst->dev = net->loopback_dev; 460 dev_hold(dst->dev); 461 dst_release(dst); 462 } 463 464 if (rt6_local) { 465 if (rt6_local->rt6i_idev) 466 in6_dev_put(rt6_local->rt6i_idev); 467 468 dst = &rt6_local->dst; 469 dev_put(dst->dev); 470 dst->dev = net->loopback_dev; 471 dev_hold(dst->dev); 472 dst_release(dst); 473 } 474 } 475 476 static int vrf_rt6_create(struct net_device *dev) 477 { 478 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE; 479 struct net_vrf *vrf = netdev_priv(dev); 480 struct net *net = dev_net(dev); 481 struct fib6_table *rt6i_table; 482 struct rt6_info *rt6, *rt6_local; 483 int rc = -ENOMEM; 484 485 /* IPv6 can be CONFIG enabled and then disabled runtime */ 486 if (!ipv6_mod_enabled()) 487 return 0; 488 489 rt6i_table = fib6_new_table(net, vrf->tb_id); 490 if (!rt6i_table) 491 goto out; 492 493 /* create a dst for routing packets out a VRF device */ 494 rt6 = ip6_dst_alloc(net, dev, flags); 495 if (!rt6) 496 goto out; 497 498 dst_hold(&rt6->dst); 499 500 rt6->rt6i_table = rt6i_table; 501 rt6->dst.output = vrf_output6; 502 503 /* create a dst for local routing - packets sent locally 504 * to local address via the VRF device as a loopback 505 */ 506 rt6_local = ip6_dst_alloc(net, dev, flags); 507 if (!rt6_local) { 508 dst_release(&rt6->dst); 509 goto out; 510 } 511 512 dst_hold(&rt6_local->dst); 513 514 rt6_local->rt6i_idev = in6_dev_get(dev); 515 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL; 516 rt6_local->rt6i_table = rt6i_table; 517 rt6_local->dst.input = ip6_input; 518 519 rcu_assign_pointer(vrf->rt6, rt6); 520 rcu_assign_pointer(vrf->rt6_local, rt6_local); 521 522 rc = 0; 523 out: 524 return rc; 525 } 526 #else 527 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 528 struct sock *sk, 529 struct sk_buff *skb) 530 { 531 return skb; 532 } 533 534 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 535 { 536 } 537 538 static int vrf_rt6_create(struct net_device *dev) 539 { 540 return 0; 541 } 542 #endif 543 544 /* modelled after ip_finish_output2 */ 545 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 546 { 547 struct dst_entry *dst = skb_dst(skb); 548 struct rtable *rt = (struct rtable *)dst; 549 struct net_device *dev = dst->dev; 550 unsigned int hh_len = LL_RESERVED_SPACE(dev); 551 struct neighbour *neigh; 552 u32 nexthop; 553 int ret = -EINVAL; 554 555 /* Be paranoid, rather than too clever. */ 556 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 557 struct sk_buff *skb2; 558 559 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 560 if (!skb2) { 561 ret = -ENOMEM; 562 goto err; 563 } 564 if (skb->sk) 565 skb_set_owner_w(skb2, skb->sk); 566 567 consume_skb(skb); 568 skb = skb2; 569 } 570 571 rcu_read_lock_bh(); 572 573 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); 574 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 575 if (unlikely(!neigh)) 576 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 577 if (!IS_ERR(neigh)) 578 ret = dst_neigh_output(dst, neigh, skb); 579 580 rcu_read_unlock_bh(); 581 err: 582 if (unlikely(ret < 0)) 583 vrf_tx_error(skb->dev, skb); 584 return ret; 585 } 586 587 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 588 { 589 struct net_device *dev = skb_dst(skb)->dev; 590 591 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 592 593 skb->dev = dev; 594 skb->protocol = htons(ETH_P_IP); 595 596 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 597 net, sk, skb, NULL, dev, 598 vrf_finish_output, 599 !(IPCB(skb)->flags & IPSKB_REROUTED)); 600 } 601 602 /* set dst on skb to send packet to us via dev_xmit path. Allows 603 * packet to go through device based features such as qdisc, netfilter 604 * hooks and packet sockets with skb->dev set to vrf device. 605 */ 606 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 607 struct sock *sk, 608 struct sk_buff *skb) 609 { 610 struct net_vrf *vrf = netdev_priv(vrf_dev); 611 struct dst_entry *dst = NULL; 612 struct rtable *rth; 613 614 rcu_read_lock(); 615 616 rth = rcu_dereference(vrf->rth); 617 if (likely(rth)) { 618 dst = &rth->dst; 619 dst_hold(dst); 620 } 621 622 rcu_read_unlock(); 623 624 if (unlikely(!dst)) { 625 vrf_tx_error(vrf_dev, skb); 626 return NULL; 627 } 628 629 skb_dst_drop(skb); 630 skb_dst_set(skb, dst); 631 632 return skb; 633 } 634 635 /* called with rcu lock held */ 636 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 637 struct sock *sk, 638 struct sk_buff *skb, 639 u16 proto) 640 { 641 switch (proto) { 642 case AF_INET: 643 return vrf_ip_out(vrf_dev, sk, skb); 644 case AF_INET6: 645 return vrf_ip6_out(vrf_dev, sk, skb); 646 } 647 648 return skb; 649 } 650 651 /* holding rtnl */ 652 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 653 { 654 struct rtable *rth = rtnl_dereference(vrf->rth); 655 struct rtable *rth_local = rtnl_dereference(vrf->rth_local); 656 struct net *net = dev_net(dev); 657 struct dst_entry *dst; 658 659 RCU_INIT_POINTER(vrf->rth, NULL); 660 RCU_INIT_POINTER(vrf->rth_local, NULL); 661 synchronize_rcu(); 662 663 /* move dev in dst's to loopback so this VRF device can be deleted 664 * - based on dst_ifdown 665 */ 666 if (rth) { 667 dst = &rth->dst; 668 dev_put(dst->dev); 669 dst->dev = net->loopback_dev; 670 dev_hold(dst->dev); 671 dst_release(dst); 672 } 673 674 if (rth_local) { 675 dst = &rth_local->dst; 676 dev_put(dst->dev); 677 dst->dev = net->loopback_dev; 678 dev_hold(dst->dev); 679 dst_release(dst); 680 } 681 } 682 683 static int vrf_rtable_create(struct net_device *dev) 684 { 685 struct net_vrf *vrf = netdev_priv(dev); 686 struct rtable *rth, *rth_local; 687 688 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 689 return -ENOMEM; 690 691 /* create a dst for routing packets out through a VRF device */ 692 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0); 693 if (!rth) 694 return -ENOMEM; 695 696 /* create a dst for local ingress routing - packets sent locally 697 * to local address via the VRF device as a loopback 698 */ 699 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0); 700 if (!rth_local) { 701 dst_release(&rth->dst); 702 return -ENOMEM; 703 } 704 705 rth->dst.output = vrf_output; 706 rth->rt_table_id = vrf->tb_id; 707 708 rth_local->rt_table_id = vrf->tb_id; 709 710 rcu_assign_pointer(vrf->rth, rth); 711 rcu_assign_pointer(vrf->rth_local, rth_local); 712 713 return 0; 714 } 715 716 /**************************** device handling ********************/ 717 718 /* cycle interface to flush neighbor cache and move routes across tables */ 719 static void cycle_netdev(struct net_device *dev) 720 { 721 unsigned int flags = dev->flags; 722 int ret; 723 724 if (!netif_running(dev)) 725 return; 726 727 ret = dev_change_flags(dev, flags & ~IFF_UP); 728 if (ret >= 0) 729 ret = dev_change_flags(dev, flags); 730 731 if (ret < 0) { 732 netdev_err(dev, 733 "Failed to cycle device %s; route tables might be wrong!\n", 734 dev->name); 735 } 736 } 737 738 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 739 { 740 int ret; 741 742 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL); 743 if (ret < 0) 744 return ret; 745 746 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 747 cycle_netdev(port_dev); 748 749 return 0; 750 } 751 752 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 753 { 754 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev)) 755 return -EINVAL; 756 757 return do_vrf_add_slave(dev, port_dev); 758 } 759 760 /* inverse of do_vrf_add_slave */ 761 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 762 { 763 netdev_upper_dev_unlink(port_dev, dev); 764 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 765 766 cycle_netdev(port_dev); 767 768 return 0; 769 } 770 771 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 772 { 773 return do_vrf_del_slave(dev, port_dev); 774 } 775 776 static void vrf_dev_uninit(struct net_device *dev) 777 { 778 struct net_vrf *vrf = netdev_priv(dev); 779 struct net_device *port_dev; 780 struct list_head *iter; 781 782 vrf_rtable_release(dev, vrf); 783 vrf_rt6_release(dev, vrf); 784 785 netdev_for_each_lower_dev(dev, port_dev, iter) 786 vrf_del_slave(dev, port_dev); 787 788 free_percpu(dev->dstats); 789 dev->dstats = NULL; 790 } 791 792 static int vrf_dev_init(struct net_device *dev) 793 { 794 struct net_vrf *vrf = netdev_priv(dev); 795 796 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 797 if (!dev->dstats) 798 goto out_nomem; 799 800 /* create the default dst which points back to us */ 801 if (vrf_rtable_create(dev) != 0) 802 goto out_stats; 803 804 if (vrf_rt6_create(dev) != 0) 805 goto out_rth; 806 807 dev->flags = IFF_MASTER | IFF_NOARP; 808 809 /* MTU is irrelevant for VRF device; set to 64k similar to lo */ 810 dev->mtu = 64 * 1024; 811 812 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 813 dev->operstate = IF_OPER_UP; 814 netdev_lockdep_set_classes(dev); 815 return 0; 816 817 out_rth: 818 vrf_rtable_release(dev, vrf); 819 out_stats: 820 free_percpu(dev->dstats); 821 dev->dstats = NULL; 822 out_nomem: 823 return -ENOMEM; 824 } 825 826 static const struct net_device_ops vrf_netdev_ops = { 827 .ndo_init = vrf_dev_init, 828 .ndo_uninit = vrf_dev_uninit, 829 .ndo_start_xmit = vrf_xmit, 830 .ndo_get_stats64 = vrf_get_stats64, 831 .ndo_add_slave = vrf_add_slave, 832 .ndo_del_slave = vrf_del_slave, 833 }; 834 835 static u32 vrf_fib_table(const struct net_device *dev) 836 { 837 struct net_vrf *vrf = netdev_priv(dev); 838 839 return vrf->tb_id; 840 } 841 842 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 843 { 844 return 0; 845 } 846 847 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 848 struct sk_buff *skb, 849 struct net_device *dev) 850 { 851 struct net *net = dev_net(dev); 852 853 nf_reset(skb); 854 855 if (NF_HOOK(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) < 0) 856 skb = NULL; /* kfree_skb(skb) handled by nf code */ 857 858 return skb; 859 } 860 861 #if IS_ENABLED(CONFIG_IPV6) 862 /* neighbor handling is done with actual device; do not want 863 * to flip skb->dev for those ndisc packets. This really fails 864 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 865 * a start. 866 */ 867 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 868 { 869 const struct ipv6hdr *iph = ipv6_hdr(skb); 870 bool rc = false; 871 872 if (iph->nexthdr == NEXTHDR_ICMP) { 873 const struct icmp6hdr *icmph; 874 struct icmp6hdr _icmph; 875 876 icmph = skb_header_pointer(skb, sizeof(*iph), 877 sizeof(_icmph), &_icmph); 878 if (!icmph) 879 goto out; 880 881 switch (icmph->icmp6_type) { 882 case NDISC_ROUTER_SOLICITATION: 883 case NDISC_ROUTER_ADVERTISEMENT: 884 case NDISC_NEIGHBOUR_SOLICITATION: 885 case NDISC_NEIGHBOUR_ADVERTISEMENT: 886 case NDISC_REDIRECT: 887 rc = true; 888 break; 889 } 890 } 891 892 out: 893 return rc; 894 } 895 896 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 897 const struct net_device *dev, 898 struct flowi6 *fl6, 899 int ifindex, 900 int flags) 901 { 902 struct net_vrf *vrf = netdev_priv(dev); 903 struct fib6_table *table = NULL; 904 struct rt6_info *rt6; 905 906 rcu_read_lock(); 907 908 /* fib6_table does not have a refcnt and can not be freed */ 909 rt6 = rcu_dereference(vrf->rt6); 910 if (likely(rt6)) 911 table = rt6->rt6i_table; 912 913 rcu_read_unlock(); 914 915 if (!table) 916 return NULL; 917 918 return ip6_pol_route(net, table, ifindex, fl6, flags); 919 } 920 921 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 922 int ifindex) 923 { 924 const struct ipv6hdr *iph = ipv6_hdr(skb); 925 struct flowi6 fl6 = { 926 .daddr = iph->daddr, 927 .saddr = iph->saddr, 928 .flowlabel = ip6_flowinfo(iph), 929 .flowi6_mark = skb->mark, 930 .flowi6_proto = iph->nexthdr, 931 .flowi6_iif = ifindex, 932 }; 933 struct net *net = dev_net(vrf_dev); 934 struct rt6_info *rt6; 935 936 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, 937 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 938 if (unlikely(!rt6)) 939 return; 940 941 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 942 return; 943 944 skb_dst_set(skb, &rt6->dst); 945 } 946 947 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 948 struct sk_buff *skb) 949 { 950 int orig_iif = skb->skb_iif; 951 bool need_strict; 952 953 /* loopback traffic; do not push through packet taps again. 954 * Reset pkt_type for upper layers to process skb 955 */ 956 if (skb->pkt_type == PACKET_LOOPBACK) { 957 skb->dev = vrf_dev; 958 skb->skb_iif = vrf_dev->ifindex; 959 skb->pkt_type = PACKET_HOST; 960 goto out; 961 } 962 963 /* if packet is NDISC or addressed to multicast or link-local 964 * then keep the ingress interface 965 */ 966 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 967 if (!ipv6_ndisc_frame(skb) && !need_strict) { 968 skb->dev = vrf_dev; 969 skb->skb_iif = vrf_dev->ifindex; 970 971 skb_push(skb, skb->mac_len); 972 dev_queue_xmit_nit(skb, vrf_dev); 973 skb_pull(skb, skb->mac_len); 974 975 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 976 } 977 978 if (need_strict) 979 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 980 981 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 982 out: 983 return skb; 984 } 985 986 #else 987 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 988 struct sk_buff *skb) 989 { 990 return skb; 991 } 992 #endif 993 994 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 995 struct sk_buff *skb) 996 { 997 skb->dev = vrf_dev; 998 skb->skb_iif = vrf_dev->ifindex; 999 1000 /* loopback traffic; do not push through packet taps again. 1001 * Reset pkt_type for upper layers to process skb 1002 */ 1003 if (skb->pkt_type == PACKET_LOOPBACK) { 1004 skb->pkt_type = PACKET_HOST; 1005 goto out; 1006 } 1007 1008 skb_push(skb, skb->mac_len); 1009 dev_queue_xmit_nit(skb, vrf_dev); 1010 skb_pull(skb, skb->mac_len); 1011 1012 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1013 out: 1014 return skb; 1015 } 1016 1017 /* called with rcu lock held */ 1018 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1019 struct sk_buff *skb, 1020 u16 proto) 1021 { 1022 switch (proto) { 1023 case AF_INET: 1024 return vrf_ip_rcv(vrf_dev, skb); 1025 case AF_INET6: 1026 return vrf_ip6_rcv(vrf_dev, skb); 1027 } 1028 1029 return skb; 1030 } 1031 1032 #if IS_ENABLED(CONFIG_IPV6) 1033 /* send to link-local or multicast address via interface enslaved to 1034 * VRF device. Force lookup to VRF table without changing flow struct 1035 */ 1036 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1037 struct flowi6 *fl6) 1038 { 1039 struct net *net = dev_net(dev); 1040 int flags = RT6_LOOKUP_F_IFACE; 1041 struct dst_entry *dst = NULL; 1042 struct rt6_info *rt; 1043 1044 /* VRF device does not have a link-local address and 1045 * sending packets to link-local or mcast addresses over 1046 * a VRF device does not make sense 1047 */ 1048 if (fl6->flowi6_oif == dev->ifindex) { 1049 dst = &net->ipv6.ip6_null_entry->dst; 1050 dst_hold(dst); 1051 return dst; 1052 } 1053 1054 if (!ipv6_addr_any(&fl6->saddr)) 1055 flags |= RT6_LOOKUP_F_HAS_SADDR; 1056 1057 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags); 1058 if (rt) 1059 dst = &rt->dst; 1060 1061 return dst; 1062 } 1063 #endif 1064 1065 static const struct l3mdev_ops vrf_l3mdev_ops = { 1066 .l3mdev_fib_table = vrf_fib_table, 1067 .l3mdev_l3_rcv = vrf_l3_rcv, 1068 .l3mdev_l3_out = vrf_l3_out, 1069 #if IS_ENABLED(CONFIG_IPV6) 1070 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1071 #endif 1072 }; 1073 1074 static void vrf_get_drvinfo(struct net_device *dev, 1075 struct ethtool_drvinfo *info) 1076 { 1077 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 1078 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 1079 } 1080 1081 static const struct ethtool_ops vrf_ethtool_ops = { 1082 .get_drvinfo = vrf_get_drvinfo, 1083 }; 1084 1085 static inline size_t vrf_fib_rule_nl_size(void) 1086 { 1087 size_t sz; 1088 1089 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1090 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1091 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1092 1093 return sz; 1094 } 1095 1096 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1097 { 1098 struct fib_rule_hdr *frh; 1099 struct nlmsghdr *nlh; 1100 struct sk_buff *skb; 1101 int err; 1102 1103 if (family == AF_INET6 && !ipv6_mod_enabled()) 1104 return 0; 1105 1106 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1107 if (!skb) 1108 return -ENOMEM; 1109 1110 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1111 if (!nlh) 1112 goto nla_put_failure; 1113 1114 /* rule only needs to appear once */ 1115 nlh->nlmsg_flags &= NLM_F_EXCL; 1116 1117 frh = nlmsg_data(nlh); 1118 memset(frh, 0, sizeof(*frh)); 1119 frh->family = family; 1120 frh->action = FR_ACT_TO_TBL; 1121 1122 if (nla_put_u32(skb, FRA_L3MDEV, 1)) 1123 goto nla_put_failure; 1124 1125 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1126 goto nla_put_failure; 1127 1128 nlmsg_end(skb, nlh); 1129 1130 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1131 skb->sk = dev_net(dev)->rtnl; 1132 if (add_it) { 1133 err = fib_nl_newrule(skb, nlh); 1134 if (err == -EEXIST) 1135 err = 0; 1136 } else { 1137 err = fib_nl_delrule(skb, nlh); 1138 if (err == -ENOENT) 1139 err = 0; 1140 } 1141 nlmsg_free(skb); 1142 1143 return err; 1144 1145 nla_put_failure: 1146 nlmsg_free(skb); 1147 1148 return -EMSGSIZE; 1149 } 1150 1151 static int vrf_add_fib_rules(const struct net_device *dev) 1152 { 1153 int err; 1154 1155 err = vrf_fib_rule(dev, AF_INET, true); 1156 if (err < 0) 1157 goto out_err; 1158 1159 err = vrf_fib_rule(dev, AF_INET6, true); 1160 if (err < 0) 1161 goto ipv6_err; 1162 1163 return 0; 1164 1165 ipv6_err: 1166 vrf_fib_rule(dev, AF_INET, false); 1167 1168 out_err: 1169 netdev_err(dev, "Failed to add FIB rules.\n"); 1170 return err; 1171 } 1172 1173 static void vrf_setup(struct net_device *dev) 1174 { 1175 ether_setup(dev); 1176 1177 /* Initialize the device structure. */ 1178 dev->netdev_ops = &vrf_netdev_ops; 1179 dev->l3mdev_ops = &vrf_l3mdev_ops; 1180 dev->ethtool_ops = &vrf_ethtool_ops; 1181 dev->destructor = free_netdev; 1182 1183 /* Fill in device structure with ethernet-generic values. */ 1184 eth_hw_addr_random(dev); 1185 1186 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1187 dev->features |= NETIF_F_LLTX; 1188 1189 /* don't allow vrf devices to change network namespaces. */ 1190 dev->features |= NETIF_F_NETNS_LOCAL; 1191 1192 /* does not make sense for a VLAN to be added to a vrf device */ 1193 dev->features |= NETIF_F_VLAN_CHALLENGED; 1194 1195 /* enable offload features */ 1196 dev->features |= NETIF_F_GSO_SOFTWARE; 1197 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM; 1198 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1199 1200 dev->hw_features = dev->features; 1201 dev->hw_enc_features = dev->features; 1202 1203 /* default to no qdisc; user can add if desired */ 1204 dev->priv_flags |= IFF_NO_QUEUE; 1205 } 1206 1207 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) 1208 { 1209 if (tb[IFLA_ADDRESS]) { 1210 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1211 return -EINVAL; 1212 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1213 return -EADDRNOTAVAIL; 1214 } 1215 return 0; 1216 } 1217 1218 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1219 { 1220 unregister_netdevice_queue(dev, head); 1221 } 1222 1223 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1224 struct nlattr *tb[], struct nlattr *data[]) 1225 { 1226 struct net_vrf *vrf = netdev_priv(dev); 1227 int err; 1228 1229 if (!data || !data[IFLA_VRF_TABLE]) 1230 return -EINVAL; 1231 1232 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1233 1234 dev->priv_flags |= IFF_L3MDEV_MASTER; 1235 1236 err = register_netdevice(dev); 1237 if (err) 1238 goto out; 1239 1240 if (add_fib_rules) { 1241 err = vrf_add_fib_rules(dev); 1242 if (err) { 1243 unregister_netdevice(dev); 1244 goto out; 1245 } 1246 add_fib_rules = false; 1247 } 1248 1249 out: 1250 return err; 1251 } 1252 1253 static size_t vrf_nl_getsize(const struct net_device *dev) 1254 { 1255 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1256 } 1257 1258 static int vrf_fillinfo(struct sk_buff *skb, 1259 const struct net_device *dev) 1260 { 1261 struct net_vrf *vrf = netdev_priv(dev); 1262 1263 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1264 } 1265 1266 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1267 const struct net_device *slave_dev) 1268 { 1269 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1270 } 1271 1272 static int vrf_fill_slave_info(struct sk_buff *skb, 1273 const struct net_device *vrf_dev, 1274 const struct net_device *slave_dev) 1275 { 1276 struct net_vrf *vrf = netdev_priv(vrf_dev); 1277 1278 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1279 return -EMSGSIZE; 1280 1281 return 0; 1282 } 1283 1284 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1285 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1286 }; 1287 1288 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1289 .kind = DRV_NAME, 1290 .priv_size = sizeof(struct net_vrf), 1291 1292 .get_size = vrf_nl_getsize, 1293 .policy = vrf_nl_policy, 1294 .validate = vrf_validate, 1295 .fill_info = vrf_fillinfo, 1296 1297 .get_slave_size = vrf_get_slave_size, 1298 .fill_slave_info = vrf_fill_slave_info, 1299 1300 .newlink = vrf_newlink, 1301 .dellink = vrf_dellink, 1302 .setup = vrf_setup, 1303 .maxtype = IFLA_VRF_MAX, 1304 }; 1305 1306 static int vrf_device_event(struct notifier_block *unused, 1307 unsigned long event, void *ptr) 1308 { 1309 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1310 1311 /* only care about unregister events to drop slave references */ 1312 if (event == NETDEV_UNREGISTER) { 1313 struct net_device *vrf_dev; 1314 1315 if (!netif_is_l3_slave(dev)) 1316 goto out; 1317 1318 vrf_dev = netdev_master_upper_dev_get(dev); 1319 vrf_del_slave(vrf_dev, dev); 1320 } 1321 out: 1322 return NOTIFY_DONE; 1323 } 1324 1325 static struct notifier_block vrf_notifier_block __read_mostly = { 1326 .notifier_call = vrf_device_event, 1327 }; 1328 1329 static int __init vrf_init_module(void) 1330 { 1331 int rc; 1332 1333 register_netdevice_notifier(&vrf_notifier_block); 1334 1335 rc = rtnl_link_register(&vrf_link_ops); 1336 if (rc < 0) 1337 goto error; 1338 1339 return 0; 1340 1341 error: 1342 unregister_netdevice_notifier(&vrf_notifier_block); 1343 return rc; 1344 } 1345 1346 module_init(vrf_init_module); 1347 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 1348 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 1349 MODULE_LICENSE("GPL"); 1350 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 1351 MODULE_VERSION(DRV_VERSION); 1352