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 skb_dst_drop(skb); 276 277 /* if dst.dev is loopback or the VRF device again this is locally 278 * originated traffic destined to a local address. Short circuit 279 * to Rx path using our local dst 280 */ 281 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) { 282 struct net_vrf *vrf = netdev_priv(vrf_dev); 283 struct rtable *rth_local; 284 struct dst_entry *dst = NULL; 285 286 ip_rt_put(rt); 287 288 rcu_read_lock(); 289 290 rth_local = rcu_dereference(vrf->rth_local); 291 if (likely(rth_local)) { 292 dst = &rth_local->dst; 293 dst_hold(dst); 294 } 295 296 rcu_read_unlock(); 297 298 if (unlikely(!dst)) 299 goto err; 300 301 return vrf_local_xmit(skb, vrf_dev, dst); 302 } 303 304 skb_dst_set(skb, &rt->dst); 305 306 /* strip the ethernet header added for pass through VRF device */ 307 __skb_pull(skb, skb_network_offset(skb)); 308 309 if (!ip4h->saddr) { 310 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 311 RT_SCOPE_LINK); 312 } 313 314 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 315 if (unlikely(net_xmit_eval(ret))) 316 vrf_dev->stats.tx_errors++; 317 else 318 ret = NET_XMIT_SUCCESS; 319 320 out: 321 return ret; 322 err: 323 vrf_tx_error(vrf_dev, skb); 324 goto out; 325 } 326 327 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 328 { 329 switch (skb->protocol) { 330 case htons(ETH_P_IP): 331 return vrf_process_v4_outbound(skb, dev); 332 case htons(ETH_P_IPV6): 333 return vrf_process_v6_outbound(skb, dev); 334 default: 335 vrf_tx_error(dev, skb); 336 return NET_XMIT_DROP; 337 } 338 } 339 340 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 341 { 342 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 343 344 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 345 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 346 347 u64_stats_update_begin(&dstats->syncp); 348 dstats->tx_pkts++; 349 dstats->tx_bytes += skb->len; 350 u64_stats_update_end(&dstats->syncp); 351 } else { 352 this_cpu_inc(dev->dstats->tx_drps); 353 } 354 355 return ret; 356 } 357 358 #if IS_ENABLED(CONFIG_IPV6) 359 /* modelled after ip6_finish_output2 */ 360 static int vrf_finish_output6(struct net *net, struct sock *sk, 361 struct sk_buff *skb) 362 { 363 struct dst_entry *dst = skb_dst(skb); 364 struct net_device *dev = dst->dev; 365 struct neighbour *neigh; 366 struct in6_addr *nexthop; 367 int ret; 368 369 nf_reset(skb); 370 371 skb->protocol = htons(ETH_P_IPV6); 372 skb->dev = dev; 373 374 rcu_read_lock_bh(); 375 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 376 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 377 if (unlikely(!neigh)) 378 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 379 if (!IS_ERR(neigh)) { 380 ret = dst_neigh_output(dst, neigh, skb); 381 rcu_read_unlock_bh(); 382 return ret; 383 } 384 rcu_read_unlock_bh(); 385 386 IP6_INC_STATS(dev_net(dst->dev), 387 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 388 kfree_skb(skb); 389 return -EINVAL; 390 } 391 392 /* modelled after ip6_output */ 393 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 394 { 395 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 396 net, sk, skb, NULL, skb_dst(skb)->dev, 397 vrf_finish_output6, 398 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 399 } 400 401 /* set dst on skb to send packet to us via dev_xmit path. Allows 402 * packet to go through device based features such as qdisc, netfilter 403 * hooks and packet sockets with skb->dev set to vrf device. 404 */ 405 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 406 struct sock *sk, 407 struct sk_buff *skb) 408 { 409 struct net_vrf *vrf = netdev_priv(vrf_dev); 410 struct dst_entry *dst = NULL; 411 struct rt6_info *rt6; 412 413 /* don't divert link scope packets */ 414 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 415 return skb; 416 417 rcu_read_lock(); 418 419 rt6 = rcu_dereference(vrf->rt6); 420 if (likely(rt6)) { 421 dst = &rt6->dst; 422 dst_hold(dst); 423 } 424 425 rcu_read_unlock(); 426 427 if (unlikely(!dst)) { 428 vrf_tx_error(vrf_dev, skb); 429 return NULL; 430 } 431 432 skb_dst_drop(skb); 433 skb_dst_set(skb, dst); 434 435 return skb; 436 } 437 438 /* holding rtnl */ 439 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 440 { 441 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 442 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local); 443 struct net *net = dev_net(dev); 444 struct dst_entry *dst; 445 446 RCU_INIT_POINTER(vrf->rt6, NULL); 447 RCU_INIT_POINTER(vrf->rt6_local, NULL); 448 synchronize_rcu(); 449 450 /* move dev in dst's to loopback so this VRF device can be deleted 451 * - based on dst_ifdown 452 */ 453 if (rt6) { 454 dst = &rt6->dst; 455 dev_put(dst->dev); 456 dst->dev = net->loopback_dev; 457 dev_hold(dst->dev); 458 dst_release(dst); 459 } 460 461 if (rt6_local) { 462 if (rt6_local->rt6i_idev) 463 in6_dev_put(rt6_local->rt6i_idev); 464 465 dst = &rt6_local->dst; 466 dev_put(dst->dev); 467 dst->dev = net->loopback_dev; 468 dev_hold(dst->dev); 469 dst_release(dst); 470 } 471 } 472 473 static int vrf_rt6_create(struct net_device *dev) 474 { 475 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE; 476 struct net_vrf *vrf = netdev_priv(dev); 477 struct net *net = dev_net(dev); 478 struct fib6_table *rt6i_table; 479 struct rt6_info *rt6, *rt6_local; 480 int rc = -ENOMEM; 481 482 /* IPv6 can be CONFIG enabled and then disabled runtime */ 483 if (!ipv6_mod_enabled()) 484 return 0; 485 486 rt6i_table = fib6_new_table(net, vrf->tb_id); 487 if (!rt6i_table) 488 goto out; 489 490 /* create a dst for routing packets out a VRF device */ 491 rt6 = ip6_dst_alloc(net, dev, flags); 492 if (!rt6) 493 goto out; 494 495 dst_hold(&rt6->dst); 496 497 rt6->rt6i_table = rt6i_table; 498 rt6->dst.output = vrf_output6; 499 500 /* create a dst for local routing - packets sent locally 501 * to local address via the VRF device as a loopback 502 */ 503 rt6_local = ip6_dst_alloc(net, dev, flags); 504 if (!rt6_local) { 505 dst_release(&rt6->dst); 506 goto out; 507 } 508 509 dst_hold(&rt6_local->dst); 510 511 rt6_local->rt6i_idev = in6_dev_get(dev); 512 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL; 513 rt6_local->rt6i_table = rt6i_table; 514 rt6_local->dst.input = ip6_input; 515 516 rcu_assign_pointer(vrf->rt6, rt6); 517 rcu_assign_pointer(vrf->rt6_local, rt6_local); 518 519 rc = 0; 520 out: 521 return rc; 522 } 523 #else 524 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 525 struct sock *sk, 526 struct sk_buff *skb) 527 { 528 return skb; 529 } 530 531 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 532 { 533 } 534 535 static int vrf_rt6_create(struct net_device *dev) 536 { 537 return 0; 538 } 539 #endif 540 541 /* modelled after ip_finish_output2 */ 542 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 543 { 544 struct dst_entry *dst = skb_dst(skb); 545 struct rtable *rt = (struct rtable *)dst; 546 struct net_device *dev = dst->dev; 547 unsigned int hh_len = LL_RESERVED_SPACE(dev); 548 struct neighbour *neigh; 549 u32 nexthop; 550 int ret = -EINVAL; 551 552 nf_reset(skb); 553 554 /* Be paranoid, rather than too clever. */ 555 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 556 struct sk_buff *skb2; 557 558 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 559 if (!skb2) { 560 ret = -ENOMEM; 561 goto err; 562 } 563 if (skb->sk) 564 skb_set_owner_w(skb2, skb->sk); 565 566 consume_skb(skb); 567 skb = skb2; 568 } 569 570 rcu_read_lock_bh(); 571 572 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); 573 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 574 if (unlikely(!neigh)) 575 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 576 if (!IS_ERR(neigh)) 577 ret = dst_neigh_output(dst, neigh, skb); 578 579 rcu_read_unlock_bh(); 580 err: 581 if (unlikely(ret < 0)) 582 vrf_tx_error(skb->dev, skb); 583 return ret; 584 } 585 586 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 587 { 588 struct net_device *dev = skb_dst(skb)->dev; 589 590 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 591 592 skb->dev = dev; 593 skb->protocol = htons(ETH_P_IP); 594 595 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 596 net, sk, skb, NULL, dev, 597 vrf_finish_output, 598 !(IPCB(skb)->flags & IPSKB_REROUTED)); 599 } 600 601 /* set dst on skb to send packet to us via dev_xmit path. Allows 602 * packet to go through device based features such as qdisc, netfilter 603 * hooks and packet sockets with skb->dev set to vrf device. 604 */ 605 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 606 struct sock *sk, 607 struct sk_buff *skb) 608 { 609 struct net_vrf *vrf = netdev_priv(vrf_dev); 610 struct dst_entry *dst = NULL; 611 struct rtable *rth; 612 613 /* don't divert multicast */ 614 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 615 return skb; 616 617 rcu_read_lock(); 618 619 rth = rcu_dereference(vrf->rth); 620 if (likely(rth)) { 621 dst = &rth->dst; 622 dst_hold(dst); 623 } 624 625 rcu_read_unlock(); 626 627 if (unlikely(!dst)) { 628 vrf_tx_error(vrf_dev, skb); 629 return NULL; 630 } 631 632 skb_dst_drop(skb); 633 skb_dst_set(skb, dst); 634 635 return skb; 636 } 637 638 /* called with rcu lock held */ 639 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 640 struct sock *sk, 641 struct sk_buff *skb, 642 u16 proto) 643 { 644 switch (proto) { 645 case AF_INET: 646 return vrf_ip_out(vrf_dev, sk, skb); 647 case AF_INET6: 648 return vrf_ip6_out(vrf_dev, sk, skb); 649 } 650 651 return skb; 652 } 653 654 /* holding rtnl */ 655 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 656 { 657 struct rtable *rth = rtnl_dereference(vrf->rth); 658 struct rtable *rth_local = rtnl_dereference(vrf->rth_local); 659 struct net *net = dev_net(dev); 660 struct dst_entry *dst; 661 662 RCU_INIT_POINTER(vrf->rth, NULL); 663 RCU_INIT_POINTER(vrf->rth_local, NULL); 664 synchronize_rcu(); 665 666 /* move dev in dst's to loopback so this VRF device can be deleted 667 * - based on dst_ifdown 668 */ 669 if (rth) { 670 dst = &rth->dst; 671 dev_put(dst->dev); 672 dst->dev = net->loopback_dev; 673 dev_hold(dst->dev); 674 dst_release(dst); 675 } 676 677 if (rth_local) { 678 dst = &rth_local->dst; 679 dev_put(dst->dev); 680 dst->dev = net->loopback_dev; 681 dev_hold(dst->dev); 682 dst_release(dst); 683 } 684 } 685 686 static int vrf_rtable_create(struct net_device *dev) 687 { 688 struct net_vrf *vrf = netdev_priv(dev); 689 struct rtable *rth, *rth_local; 690 691 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 692 return -ENOMEM; 693 694 /* create a dst for routing packets out through a VRF device */ 695 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0); 696 if (!rth) 697 return -ENOMEM; 698 699 /* create a dst for local ingress routing - packets sent locally 700 * to local address via the VRF device as a loopback 701 */ 702 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0); 703 if (!rth_local) { 704 dst_release(&rth->dst); 705 return -ENOMEM; 706 } 707 708 rth->dst.output = vrf_output; 709 rth->rt_table_id = vrf->tb_id; 710 711 rth_local->rt_table_id = vrf->tb_id; 712 713 rcu_assign_pointer(vrf->rth, rth); 714 rcu_assign_pointer(vrf->rth_local, rth_local); 715 716 return 0; 717 } 718 719 /**************************** device handling ********************/ 720 721 /* cycle interface to flush neighbor cache and move routes across tables */ 722 static void cycle_netdev(struct net_device *dev) 723 { 724 unsigned int flags = dev->flags; 725 int ret; 726 727 if (!netif_running(dev)) 728 return; 729 730 ret = dev_change_flags(dev, flags & ~IFF_UP); 731 if (ret >= 0) 732 ret = dev_change_flags(dev, flags); 733 734 if (ret < 0) { 735 netdev_err(dev, 736 "Failed to cycle device %s; route tables might be wrong!\n", 737 dev->name); 738 } 739 } 740 741 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 742 { 743 int ret; 744 745 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL); 746 if (ret < 0) 747 return ret; 748 749 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 750 cycle_netdev(port_dev); 751 752 return 0; 753 } 754 755 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 756 { 757 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev)) 758 return -EINVAL; 759 760 return do_vrf_add_slave(dev, port_dev); 761 } 762 763 /* inverse of do_vrf_add_slave */ 764 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 765 { 766 netdev_upper_dev_unlink(port_dev, dev); 767 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 768 769 cycle_netdev(port_dev); 770 771 return 0; 772 } 773 774 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 775 { 776 return do_vrf_del_slave(dev, port_dev); 777 } 778 779 static void vrf_dev_uninit(struct net_device *dev) 780 { 781 struct net_vrf *vrf = netdev_priv(dev); 782 struct net_device *port_dev; 783 struct list_head *iter; 784 785 vrf_rtable_release(dev, vrf); 786 vrf_rt6_release(dev, vrf); 787 788 netdev_for_each_lower_dev(dev, port_dev, iter) 789 vrf_del_slave(dev, port_dev); 790 791 free_percpu(dev->dstats); 792 dev->dstats = NULL; 793 } 794 795 static int vrf_dev_init(struct net_device *dev) 796 { 797 struct net_vrf *vrf = netdev_priv(dev); 798 799 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 800 if (!dev->dstats) 801 goto out_nomem; 802 803 /* create the default dst which points back to us */ 804 if (vrf_rtable_create(dev) != 0) 805 goto out_stats; 806 807 if (vrf_rt6_create(dev) != 0) 808 goto out_rth; 809 810 dev->flags = IFF_MASTER | IFF_NOARP; 811 812 /* MTU is irrelevant for VRF device; set to 64k similar to lo */ 813 dev->mtu = 64 * 1024; 814 815 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 816 dev->operstate = IF_OPER_UP; 817 netdev_lockdep_set_classes(dev); 818 return 0; 819 820 out_rth: 821 vrf_rtable_release(dev, vrf); 822 out_stats: 823 free_percpu(dev->dstats); 824 dev->dstats = NULL; 825 out_nomem: 826 return -ENOMEM; 827 } 828 829 static const struct net_device_ops vrf_netdev_ops = { 830 .ndo_init = vrf_dev_init, 831 .ndo_uninit = vrf_dev_uninit, 832 .ndo_start_xmit = vrf_xmit, 833 .ndo_get_stats64 = vrf_get_stats64, 834 .ndo_add_slave = vrf_add_slave, 835 .ndo_del_slave = vrf_del_slave, 836 }; 837 838 static u32 vrf_fib_table(const struct net_device *dev) 839 { 840 struct net_vrf *vrf = netdev_priv(dev); 841 842 return vrf->tb_id; 843 } 844 845 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 846 { 847 return 0; 848 } 849 850 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 851 struct sk_buff *skb, 852 struct net_device *dev) 853 { 854 struct net *net = dev_net(dev); 855 856 if (NF_HOOK(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) < 0) 857 skb = NULL; /* kfree_skb(skb) handled by nf code */ 858 859 return skb; 860 } 861 862 #if IS_ENABLED(CONFIG_IPV6) 863 /* neighbor handling is done with actual device; do not want 864 * to flip skb->dev for those ndisc packets. This really fails 865 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 866 * a start. 867 */ 868 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 869 { 870 const struct ipv6hdr *iph = ipv6_hdr(skb); 871 bool rc = false; 872 873 if (iph->nexthdr == NEXTHDR_ICMP) { 874 const struct icmp6hdr *icmph; 875 struct icmp6hdr _icmph; 876 877 icmph = skb_header_pointer(skb, sizeof(*iph), 878 sizeof(_icmph), &_icmph); 879 if (!icmph) 880 goto out; 881 882 switch (icmph->icmp6_type) { 883 case NDISC_ROUTER_SOLICITATION: 884 case NDISC_ROUTER_ADVERTISEMENT: 885 case NDISC_NEIGHBOUR_SOLICITATION: 886 case NDISC_NEIGHBOUR_ADVERTISEMENT: 887 case NDISC_REDIRECT: 888 rc = true; 889 break; 890 } 891 } 892 893 out: 894 return rc; 895 } 896 897 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 898 const struct net_device *dev, 899 struct flowi6 *fl6, 900 int ifindex, 901 int flags) 902 { 903 struct net_vrf *vrf = netdev_priv(dev); 904 struct fib6_table *table = NULL; 905 struct rt6_info *rt6; 906 907 rcu_read_lock(); 908 909 /* fib6_table does not have a refcnt and can not be freed */ 910 rt6 = rcu_dereference(vrf->rt6); 911 if (likely(rt6)) 912 table = rt6->rt6i_table; 913 914 rcu_read_unlock(); 915 916 if (!table) 917 return NULL; 918 919 return ip6_pol_route(net, table, ifindex, fl6, flags); 920 } 921 922 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 923 int ifindex) 924 { 925 const struct ipv6hdr *iph = ipv6_hdr(skb); 926 struct flowi6 fl6 = { 927 .daddr = iph->daddr, 928 .saddr = iph->saddr, 929 .flowlabel = ip6_flowinfo(iph), 930 .flowi6_mark = skb->mark, 931 .flowi6_proto = iph->nexthdr, 932 .flowi6_iif = ifindex, 933 }; 934 struct net *net = dev_net(vrf_dev); 935 struct rt6_info *rt6; 936 937 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, 938 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 939 if (unlikely(!rt6)) 940 return; 941 942 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 943 return; 944 945 skb_dst_set(skb, &rt6->dst); 946 } 947 948 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 949 struct sk_buff *skb) 950 { 951 int orig_iif = skb->skb_iif; 952 bool need_strict; 953 954 /* loopback traffic; do not push through packet taps again. 955 * Reset pkt_type for upper layers to process skb 956 */ 957 if (skb->pkt_type == PACKET_LOOPBACK) { 958 skb->dev = vrf_dev; 959 skb->skb_iif = vrf_dev->ifindex; 960 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 961 skb->pkt_type = PACKET_HOST; 962 goto out; 963 } 964 965 /* if packet is NDISC or addressed to multicast or link-local 966 * then keep the ingress interface 967 */ 968 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 969 if (!ipv6_ndisc_frame(skb) && !need_strict) { 970 skb->dev = vrf_dev; 971 skb->skb_iif = vrf_dev->ifindex; 972 973 skb_push(skb, skb->mac_len); 974 dev_queue_xmit_nit(skb, vrf_dev); 975 skb_pull(skb, skb->mac_len); 976 977 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 978 } 979 980 if (need_strict) 981 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 982 983 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 984 out: 985 return skb; 986 } 987 988 #else 989 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 990 struct sk_buff *skb) 991 { 992 return skb; 993 } 994 #endif 995 996 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 997 struct sk_buff *skb) 998 { 999 skb->dev = vrf_dev; 1000 skb->skb_iif = vrf_dev->ifindex; 1001 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1002 1003 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1004 goto out; 1005 1006 /* loopback traffic; do not push through packet taps again. 1007 * Reset pkt_type for upper layers to process skb 1008 */ 1009 if (skb->pkt_type == PACKET_LOOPBACK) { 1010 skb->pkt_type = PACKET_HOST; 1011 goto out; 1012 } 1013 1014 skb_push(skb, skb->mac_len); 1015 dev_queue_xmit_nit(skb, vrf_dev); 1016 skb_pull(skb, skb->mac_len); 1017 1018 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1019 out: 1020 return skb; 1021 } 1022 1023 /* called with rcu lock held */ 1024 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1025 struct sk_buff *skb, 1026 u16 proto) 1027 { 1028 switch (proto) { 1029 case AF_INET: 1030 return vrf_ip_rcv(vrf_dev, skb); 1031 case AF_INET6: 1032 return vrf_ip6_rcv(vrf_dev, skb); 1033 } 1034 1035 return skb; 1036 } 1037 1038 #if IS_ENABLED(CONFIG_IPV6) 1039 /* send to link-local or multicast address via interface enslaved to 1040 * VRF device. Force lookup to VRF table without changing flow struct 1041 */ 1042 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1043 struct flowi6 *fl6) 1044 { 1045 struct net *net = dev_net(dev); 1046 int flags = RT6_LOOKUP_F_IFACE; 1047 struct dst_entry *dst = NULL; 1048 struct rt6_info *rt; 1049 1050 /* VRF device does not have a link-local address and 1051 * sending packets to link-local or mcast addresses over 1052 * a VRF device does not make sense 1053 */ 1054 if (fl6->flowi6_oif == dev->ifindex) { 1055 dst = &net->ipv6.ip6_null_entry->dst; 1056 dst_hold(dst); 1057 return dst; 1058 } 1059 1060 if (!ipv6_addr_any(&fl6->saddr)) 1061 flags |= RT6_LOOKUP_F_HAS_SADDR; 1062 1063 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags); 1064 if (rt) 1065 dst = &rt->dst; 1066 1067 return dst; 1068 } 1069 #endif 1070 1071 static const struct l3mdev_ops vrf_l3mdev_ops = { 1072 .l3mdev_fib_table = vrf_fib_table, 1073 .l3mdev_l3_rcv = vrf_l3_rcv, 1074 .l3mdev_l3_out = vrf_l3_out, 1075 #if IS_ENABLED(CONFIG_IPV6) 1076 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1077 #endif 1078 }; 1079 1080 static void vrf_get_drvinfo(struct net_device *dev, 1081 struct ethtool_drvinfo *info) 1082 { 1083 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 1084 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 1085 } 1086 1087 static const struct ethtool_ops vrf_ethtool_ops = { 1088 .get_drvinfo = vrf_get_drvinfo, 1089 }; 1090 1091 static inline size_t vrf_fib_rule_nl_size(void) 1092 { 1093 size_t sz; 1094 1095 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1096 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1097 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1098 1099 return sz; 1100 } 1101 1102 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1103 { 1104 struct fib_rule_hdr *frh; 1105 struct nlmsghdr *nlh; 1106 struct sk_buff *skb; 1107 int err; 1108 1109 if (family == AF_INET6 && !ipv6_mod_enabled()) 1110 return 0; 1111 1112 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1113 if (!skb) 1114 return -ENOMEM; 1115 1116 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1117 if (!nlh) 1118 goto nla_put_failure; 1119 1120 /* rule only needs to appear once */ 1121 nlh->nlmsg_flags &= NLM_F_EXCL; 1122 1123 frh = nlmsg_data(nlh); 1124 memset(frh, 0, sizeof(*frh)); 1125 frh->family = family; 1126 frh->action = FR_ACT_TO_TBL; 1127 1128 if (nla_put_u32(skb, FRA_L3MDEV, 1)) 1129 goto nla_put_failure; 1130 1131 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1132 goto nla_put_failure; 1133 1134 nlmsg_end(skb, nlh); 1135 1136 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1137 skb->sk = dev_net(dev)->rtnl; 1138 if (add_it) { 1139 err = fib_nl_newrule(skb, nlh); 1140 if (err == -EEXIST) 1141 err = 0; 1142 } else { 1143 err = fib_nl_delrule(skb, nlh); 1144 if (err == -ENOENT) 1145 err = 0; 1146 } 1147 nlmsg_free(skb); 1148 1149 return err; 1150 1151 nla_put_failure: 1152 nlmsg_free(skb); 1153 1154 return -EMSGSIZE; 1155 } 1156 1157 static int vrf_add_fib_rules(const struct net_device *dev) 1158 { 1159 int err; 1160 1161 err = vrf_fib_rule(dev, AF_INET, true); 1162 if (err < 0) 1163 goto out_err; 1164 1165 err = vrf_fib_rule(dev, AF_INET6, true); 1166 if (err < 0) 1167 goto ipv6_err; 1168 1169 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1170 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1171 if (err < 0) 1172 goto ipmr_err; 1173 #endif 1174 1175 return 0; 1176 1177 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1178 ipmr_err: 1179 vrf_fib_rule(dev, AF_INET6, false); 1180 #endif 1181 1182 ipv6_err: 1183 vrf_fib_rule(dev, AF_INET, false); 1184 1185 out_err: 1186 netdev_err(dev, "Failed to add FIB rules.\n"); 1187 return err; 1188 } 1189 1190 static void vrf_setup(struct net_device *dev) 1191 { 1192 ether_setup(dev); 1193 1194 /* Initialize the device structure. */ 1195 dev->netdev_ops = &vrf_netdev_ops; 1196 dev->l3mdev_ops = &vrf_l3mdev_ops; 1197 dev->ethtool_ops = &vrf_ethtool_ops; 1198 dev->destructor = free_netdev; 1199 1200 /* Fill in device structure with ethernet-generic values. */ 1201 eth_hw_addr_random(dev); 1202 1203 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1204 dev->features |= NETIF_F_LLTX; 1205 1206 /* don't allow vrf devices to change network namespaces. */ 1207 dev->features |= NETIF_F_NETNS_LOCAL; 1208 1209 /* does not make sense for a VLAN to be added to a vrf device */ 1210 dev->features |= NETIF_F_VLAN_CHALLENGED; 1211 1212 /* enable offload features */ 1213 dev->features |= NETIF_F_GSO_SOFTWARE; 1214 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM; 1215 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1216 1217 dev->hw_features = dev->features; 1218 dev->hw_enc_features = dev->features; 1219 1220 /* default to no qdisc; user can add if desired */ 1221 dev->priv_flags |= IFF_NO_QUEUE; 1222 } 1223 1224 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) 1225 { 1226 if (tb[IFLA_ADDRESS]) { 1227 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1228 return -EINVAL; 1229 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1230 return -EADDRNOTAVAIL; 1231 } 1232 return 0; 1233 } 1234 1235 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1236 { 1237 unregister_netdevice_queue(dev, head); 1238 } 1239 1240 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1241 struct nlattr *tb[], struct nlattr *data[]) 1242 { 1243 struct net_vrf *vrf = netdev_priv(dev); 1244 int err; 1245 1246 if (!data || !data[IFLA_VRF_TABLE]) 1247 return -EINVAL; 1248 1249 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1250 1251 dev->priv_flags |= IFF_L3MDEV_MASTER; 1252 1253 err = register_netdevice(dev); 1254 if (err) 1255 goto out; 1256 1257 if (add_fib_rules) { 1258 err = vrf_add_fib_rules(dev); 1259 if (err) { 1260 unregister_netdevice(dev); 1261 goto out; 1262 } 1263 add_fib_rules = false; 1264 } 1265 1266 out: 1267 return err; 1268 } 1269 1270 static size_t vrf_nl_getsize(const struct net_device *dev) 1271 { 1272 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1273 } 1274 1275 static int vrf_fillinfo(struct sk_buff *skb, 1276 const struct net_device *dev) 1277 { 1278 struct net_vrf *vrf = netdev_priv(dev); 1279 1280 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1281 } 1282 1283 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1284 const struct net_device *slave_dev) 1285 { 1286 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1287 } 1288 1289 static int vrf_fill_slave_info(struct sk_buff *skb, 1290 const struct net_device *vrf_dev, 1291 const struct net_device *slave_dev) 1292 { 1293 struct net_vrf *vrf = netdev_priv(vrf_dev); 1294 1295 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1296 return -EMSGSIZE; 1297 1298 return 0; 1299 } 1300 1301 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1302 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1303 }; 1304 1305 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1306 .kind = DRV_NAME, 1307 .priv_size = sizeof(struct net_vrf), 1308 1309 .get_size = vrf_nl_getsize, 1310 .policy = vrf_nl_policy, 1311 .validate = vrf_validate, 1312 .fill_info = vrf_fillinfo, 1313 1314 .get_slave_size = vrf_get_slave_size, 1315 .fill_slave_info = vrf_fill_slave_info, 1316 1317 .newlink = vrf_newlink, 1318 .dellink = vrf_dellink, 1319 .setup = vrf_setup, 1320 .maxtype = IFLA_VRF_MAX, 1321 }; 1322 1323 static int vrf_device_event(struct notifier_block *unused, 1324 unsigned long event, void *ptr) 1325 { 1326 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1327 1328 /* only care about unregister events to drop slave references */ 1329 if (event == NETDEV_UNREGISTER) { 1330 struct net_device *vrf_dev; 1331 1332 if (!netif_is_l3_slave(dev)) 1333 goto out; 1334 1335 vrf_dev = netdev_master_upper_dev_get(dev); 1336 vrf_del_slave(vrf_dev, dev); 1337 } 1338 out: 1339 return NOTIFY_DONE; 1340 } 1341 1342 static struct notifier_block vrf_notifier_block __read_mostly = { 1343 .notifier_call = vrf_device_event, 1344 }; 1345 1346 static int __init vrf_init_module(void) 1347 { 1348 int rc; 1349 1350 register_netdevice_notifier(&vrf_notifier_block); 1351 1352 rc = rtnl_link_register(&vrf_link_ops); 1353 if (rc < 0) 1354 goto error; 1355 1356 return 0; 1357 1358 error: 1359 unregister_netdevice_notifier(&vrf_notifier_block); 1360 return rc; 1361 } 1362 1363 module_init(vrf_init_module); 1364 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 1365 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 1366 MODULE_LICENSE("GPL"); 1367 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 1368 MODULE_VERSION(DRV_VERSION); 1369