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 .flowi4_proto = ip4h->protocol, 267 .daddr = ip4h->daddr, 268 .saddr = ip4h->saddr, 269 }; 270 struct net *net = dev_net(vrf_dev); 271 struct rtable *rt; 272 273 rt = ip_route_output_flow(net, &fl4, NULL); 274 if (IS_ERR(rt)) 275 goto err; 276 277 skb_dst_drop(skb); 278 279 /* if dst.dev is loopback or the VRF device again this is locally 280 * originated traffic destined to a local address. Short circuit 281 * to Rx path using our local dst 282 */ 283 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) { 284 struct net_vrf *vrf = netdev_priv(vrf_dev); 285 struct rtable *rth_local; 286 struct dst_entry *dst = NULL; 287 288 ip_rt_put(rt); 289 290 rcu_read_lock(); 291 292 rth_local = rcu_dereference(vrf->rth_local); 293 if (likely(rth_local)) { 294 dst = &rth_local->dst; 295 dst_hold(dst); 296 } 297 298 rcu_read_unlock(); 299 300 if (unlikely(!dst)) 301 goto err; 302 303 return vrf_local_xmit(skb, vrf_dev, dst); 304 } 305 306 skb_dst_set(skb, &rt->dst); 307 308 /* strip the ethernet header added for pass through VRF device */ 309 __skb_pull(skb, skb_network_offset(skb)); 310 311 if (!ip4h->saddr) { 312 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 313 RT_SCOPE_LINK); 314 } 315 316 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 317 if (unlikely(net_xmit_eval(ret))) 318 vrf_dev->stats.tx_errors++; 319 else 320 ret = NET_XMIT_SUCCESS; 321 322 out: 323 return ret; 324 err: 325 vrf_tx_error(vrf_dev, skb); 326 goto out; 327 } 328 329 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 330 { 331 switch (skb->protocol) { 332 case htons(ETH_P_IP): 333 return vrf_process_v4_outbound(skb, dev); 334 case htons(ETH_P_IPV6): 335 return vrf_process_v6_outbound(skb, dev); 336 default: 337 vrf_tx_error(dev, skb); 338 return NET_XMIT_DROP; 339 } 340 } 341 342 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 343 { 344 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 345 346 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 347 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 348 349 u64_stats_update_begin(&dstats->syncp); 350 dstats->tx_pkts++; 351 dstats->tx_bytes += skb->len; 352 u64_stats_update_end(&dstats->syncp); 353 } else { 354 this_cpu_inc(dev->dstats->tx_drps); 355 } 356 357 return ret; 358 } 359 360 #if IS_ENABLED(CONFIG_IPV6) 361 /* modelled after ip6_finish_output2 */ 362 static int vrf_finish_output6(struct net *net, struct sock *sk, 363 struct sk_buff *skb) 364 { 365 struct dst_entry *dst = skb_dst(skb); 366 struct net_device *dev = dst->dev; 367 struct neighbour *neigh; 368 struct in6_addr *nexthop; 369 int ret; 370 371 nf_reset(skb); 372 373 skb->protocol = htons(ETH_P_IPV6); 374 skb->dev = dev; 375 376 rcu_read_lock_bh(); 377 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 378 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 379 if (unlikely(!neigh)) 380 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 381 if (!IS_ERR(neigh)) { 382 ret = dst_neigh_output(dst, neigh, skb); 383 rcu_read_unlock_bh(); 384 return ret; 385 } 386 rcu_read_unlock_bh(); 387 388 IP6_INC_STATS(dev_net(dst->dev), 389 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 390 kfree_skb(skb); 391 return -EINVAL; 392 } 393 394 /* modelled after ip6_output */ 395 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 396 { 397 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 398 net, sk, skb, NULL, skb_dst(skb)->dev, 399 vrf_finish_output6, 400 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 401 } 402 403 /* set dst on skb to send packet to us via dev_xmit path. Allows 404 * packet to go through device based features such as qdisc, netfilter 405 * hooks and packet sockets with skb->dev set to vrf device. 406 */ 407 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 408 struct sock *sk, 409 struct sk_buff *skb) 410 { 411 struct net_vrf *vrf = netdev_priv(vrf_dev); 412 struct dst_entry *dst = NULL; 413 struct rt6_info *rt6; 414 415 /* don't divert link scope packets */ 416 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 417 return skb; 418 419 rcu_read_lock(); 420 421 rt6 = rcu_dereference(vrf->rt6); 422 if (likely(rt6)) { 423 dst = &rt6->dst; 424 dst_hold(dst); 425 } 426 427 rcu_read_unlock(); 428 429 if (unlikely(!dst)) { 430 vrf_tx_error(vrf_dev, skb); 431 return NULL; 432 } 433 434 skb_dst_drop(skb); 435 skb_dst_set(skb, dst); 436 437 return skb; 438 } 439 440 /* holding rtnl */ 441 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 442 { 443 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 444 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local); 445 struct net *net = dev_net(dev); 446 struct dst_entry *dst; 447 448 RCU_INIT_POINTER(vrf->rt6, NULL); 449 RCU_INIT_POINTER(vrf->rt6_local, NULL); 450 synchronize_rcu(); 451 452 /* move dev in dst's to loopback so this VRF device can be deleted 453 * - based on dst_ifdown 454 */ 455 if (rt6) { 456 dst = &rt6->dst; 457 dev_put(dst->dev); 458 dst->dev = net->loopback_dev; 459 dev_hold(dst->dev); 460 dst_release(dst); 461 } 462 463 if (rt6_local) { 464 if (rt6_local->rt6i_idev) 465 in6_dev_put(rt6_local->rt6i_idev); 466 467 dst = &rt6_local->dst; 468 dev_put(dst->dev); 469 dst->dev = net->loopback_dev; 470 dev_hold(dst->dev); 471 dst_release(dst); 472 } 473 } 474 475 static int vrf_rt6_create(struct net_device *dev) 476 { 477 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE; 478 struct net_vrf *vrf = netdev_priv(dev); 479 struct net *net = dev_net(dev); 480 struct fib6_table *rt6i_table; 481 struct rt6_info *rt6, *rt6_local; 482 int rc = -ENOMEM; 483 484 /* IPv6 can be CONFIG enabled and then disabled runtime */ 485 if (!ipv6_mod_enabled()) 486 return 0; 487 488 rt6i_table = fib6_new_table(net, vrf->tb_id); 489 if (!rt6i_table) 490 goto out; 491 492 /* create a dst for routing packets out a VRF device */ 493 rt6 = ip6_dst_alloc(net, dev, flags); 494 if (!rt6) 495 goto out; 496 497 dst_hold(&rt6->dst); 498 499 rt6->rt6i_table = rt6i_table; 500 rt6->dst.output = vrf_output6; 501 502 /* create a dst for local routing - packets sent locally 503 * to local address via the VRF device as a loopback 504 */ 505 rt6_local = ip6_dst_alloc(net, dev, flags); 506 if (!rt6_local) { 507 dst_release(&rt6->dst); 508 goto out; 509 } 510 511 dst_hold(&rt6_local->dst); 512 513 rt6_local->rt6i_idev = in6_dev_get(dev); 514 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL; 515 rt6_local->rt6i_table = rt6i_table; 516 rt6_local->dst.input = ip6_input; 517 518 rcu_assign_pointer(vrf->rt6, rt6); 519 rcu_assign_pointer(vrf->rt6_local, rt6_local); 520 521 rc = 0; 522 out: 523 return rc; 524 } 525 #else 526 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 527 struct sock *sk, 528 struct sk_buff *skb) 529 { 530 return skb; 531 } 532 533 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 534 { 535 } 536 537 static int vrf_rt6_create(struct net_device *dev) 538 { 539 return 0; 540 } 541 #endif 542 543 /* modelled after ip_finish_output2 */ 544 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 545 { 546 struct dst_entry *dst = skb_dst(skb); 547 struct rtable *rt = (struct rtable *)dst; 548 struct net_device *dev = dst->dev; 549 unsigned int hh_len = LL_RESERVED_SPACE(dev); 550 struct neighbour *neigh; 551 u32 nexthop; 552 int ret = -EINVAL; 553 554 nf_reset(skb); 555 556 /* Be paranoid, rather than too clever. */ 557 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 558 struct sk_buff *skb2; 559 560 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 561 if (!skb2) { 562 ret = -ENOMEM; 563 goto err; 564 } 565 if (skb->sk) 566 skb_set_owner_w(skb2, skb->sk); 567 568 consume_skb(skb); 569 skb = skb2; 570 } 571 572 rcu_read_lock_bh(); 573 574 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); 575 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 576 if (unlikely(!neigh)) 577 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 578 if (!IS_ERR(neigh)) 579 ret = dst_neigh_output(dst, neigh, skb); 580 581 rcu_read_unlock_bh(); 582 err: 583 if (unlikely(ret < 0)) 584 vrf_tx_error(skb->dev, skb); 585 return ret; 586 } 587 588 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 589 { 590 struct net_device *dev = skb_dst(skb)->dev; 591 592 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 593 594 skb->dev = dev; 595 skb->protocol = htons(ETH_P_IP); 596 597 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 598 net, sk, skb, NULL, dev, 599 vrf_finish_output, 600 !(IPCB(skb)->flags & IPSKB_REROUTED)); 601 } 602 603 /* set dst on skb to send packet to us via dev_xmit path. Allows 604 * packet to go through device based features such as qdisc, netfilter 605 * hooks and packet sockets with skb->dev set to vrf device. 606 */ 607 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 608 struct sock *sk, 609 struct sk_buff *skb) 610 { 611 struct net_vrf *vrf = netdev_priv(vrf_dev); 612 struct dst_entry *dst = NULL; 613 struct rtable *rth; 614 615 /* don't divert multicast */ 616 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 617 return skb; 618 619 rcu_read_lock(); 620 621 rth = rcu_dereference(vrf->rth); 622 if (likely(rth)) { 623 dst = &rth->dst; 624 dst_hold(dst); 625 } 626 627 rcu_read_unlock(); 628 629 if (unlikely(!dst)) { 630 vrf_tx_error(vrf_dev, skb); 631 return NULL; 632 } 633 634 skb_dst_drop(skb); 635 skb_dst_set(skb, dst); 636 637 return skb; 638 } 639 640 /* called with rcu lock held */ 641 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 642 struct sock *sk, 643 struct sk_buff *skb, 644 u16 proto) 645 { 646 switch (proto) { 647 case AF_INET: 648 return vrf_ip_out(vrf_dev, sk, skb); 649 case AF_INET6: 650 return vrf_ip6_out(vrf_dev, sk, skb); 651 } 652 653 return skb; 654 } 655 656 /* holding rtnl */ 657 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 658 { 659 struct rtable *rth = rtnl_dereference(vrf->rth); 660 struct rtable *rth_local = rtnl_dereference(vrf->rth_local); 661 struct net *net = dev_net(dev); 662 struct dst_entry *dst; 663 664 RCU_INIT_POINTER(vrf->rth, NULL); 665 RCU_INIT_POINTER(vrf->rth_local, NULL); 666 synchronize_rcu(); 667 668 /* move dev in dst's to loopback so this VRF device can be deleted 669 * - based on dst_ifdown 670 */ 671 if (rth) { 672 dst = &rth->dst; 673 dev_put(dst->dev); 674 dst->dev = net->loopback_dev; 675 dev_hold(dst->dev); 676 dst_release(dst); 677 } 678 679 if (rth_local) { 680 dst = &rth_local->dst; 681 dev_put(dst->dev); 682 dst->dev = net->loopback_dev; 683 dev_hold(dst->dev); 684 dst_release(dst); 685 } 686 } 687 688 static int vrf_rtable_create(struct net_device *dev) 689 { 690 struct net_vrf *vrf = netdev_priv(dev); 691 struct rtable *rth, *rth_local; 692 693 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 694 return -ENOMEM; 695 696 /* create a dst for routing packets out through a VRF device */ 697 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0); 698 if (!rth) 699 return -ENOMEM; 700 701 /* create a dst for local ingress routing - packets sent locally 702 * to local address via the VRF device as a loopback 703 */ 704 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0); 705 if (!rth_local) { 706 dst_release(&rth->dst); 707 return -ENOMEM; 708 } 709 710 rth->dst.output = vrf_output; 711 rth->rt_table_id = vrf->tb_id; 712 713 rth_local->rt_table_id = vrf->tb_id; 714 715 rcu_assign_pointer(vrf->rth, rth); 716 rcu_assign_pointer(vrf->rth_local, rth_local); 717 718 return 0; 719 } 720 721 /**************************** device handling ********************/ 722 723 /* cycle interface to flush neighbor cache and move routes across tables */ 724 static void cycle_netdev(struct net_device *dev) 725 { 726 unsigned int flags = dev->flags; 727 int ret; 728 729 if (!netif_running(dev)) 730 return; 731 732 ret = dev_change_flags(dev, flags & ~IFF_UP); 733 if (ret >= 0) 734 ret = dev_change_flags(dev, flags); 735 736 if (ret < 0) { 737 netdev_err(dev, 738 "Failed to cycle device %s; route tables might be wrong!\n", 739 dev->name); 740 } 741 } 742 743 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 744 { 745 int ret; 746 747 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL); 748 if (ret < 0) 749 return ret; 750 751 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 752 cycle_netdev(port_dev); 753 754 return 0; 755 } 756 757 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 758 { 759 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev)) 760 return -EINVAL; 761 762 return do_vrf_add_slave(dev, port_dev); 763 } 764 765 /* inverse of do_vrf_add_slave */ 766 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 767 { 768 netdev_upper_dev_unlink(port_dev, dev); 769 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 770 771 cycle_netdev(port_dev); 772 773 return 0; 774 } 775 776 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 777 { 778 return do_vrf_del_slave(dev, port_dev); 779 } 780 781 static void vrf_dev_uninit(struct net_device *dev) 782 { 783 struct net_vrf *vrf = netdev_priv(dev); 784 struct net_device *port_dev; 785 struct list_head *iter; 786 787 vrf_rtable_release(dev, vrf); 788 vrf_rt6_release(dev, vrf); 789 790 netdev_for_each_lower_dev(dev, port_dev, iter) 791 vrf_del_slave(dev, port_dev); 792 793 free_percpu(dev->dstats); 794 dev->dstats = NULL; 795 } 796 797 static int vrf_dev_init(struct net_device *dev) 798 { 799 struct net_vrf *vrf = netdev_priv(dev); 800 801 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 802 if (!dev->dstats) 803 goto out_nomem; 804 805 /* create the default dst which points back to us */ 806 if (vrf_rtable_create(dev) != 0) 807 goto out_stats; 808 809 if (vrf_rt6_create(dev) != 0) 810 goto out_rth; 811 812 dev->flags = IFF_MASTER | IFF_NOARP; 813 814 /* MTU is irrelevant for VRF device; set to 64k similar to lo */ 815 dev->mtu = 64 * 1024; 816 817 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 818 dev->operstate = IF_OPER_UP; 819 netdev_lockdep_set_classes(dev); 820 return 0; 821 822 out_rth: 823 vrf_rtable_release(dev, vrf); 824 out_stats: 825 free_percpu(dev->dstats); 826 dev->dstats = NULL; 827 out_nomem: 828 return -ENOMEM; 829 } 830 831 static const struct net_device_ops vrf_netdev_ops = { 832 .ndo_init = vrf_dev_init, 833 .ndo_uninit = vrf_dev_uninit, 834 .ndo_start_xmit = vrf_xmit, 835 .ndo_get_stats64 = vrf_get_stats64, 836 .ndo_add_slave = vrf_add_slave, 837 .ndo_del_slave = vrf_del_slave, 838 }; 839 840 static u32 vrf_fib_table(const struct net_device *dev) 841 { 842 struct net_vrf *vrf = netdev_priv(dev); 843 844 return vrf->tb_id; 845 } 846 847 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 848 { 849 return 0; 850 } 851 852 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 853 struct sk_buff *skb, 854 struct net_device *dev) 855 { 856 struct net *net = dev_net(dev); 857 858 if (NF_HOOK(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) < 0) 859 skb = NULL; /* kfree_skb(skb) handled by nf code */ 860 861 return skb; 862 } 863 864 #if IS_ENABLED(CONFIG_IPV6) 865 /* neighbor handling is done with actual device; do not want 866 * to flip skb->dev for those ndisc packets. This really fails 867 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 868 * a start. 869 */ 870 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 871 { 872 const struct ipv6hdr *iph = ipv6_hdr(skb); 873 bool rc = false; 874 875 if (iph->nexthdr == NEXTHDR_ICMP) { 876 const struct icmp6hdr *icmph; 877 struct icmp6hdr _icmph; 878 879 icmph = skb_header_pointer(skb, sizeof(*iph), 880 sizeof(_icmph), &_icmph); 881 if (!icmph) 882 goto out; 883 884 switch (icmph->icmp6_type) { 885 case NDISC_ROUTER_SOLICITATION: 886 case NDISC_ROUTER_ADVERTISEMENT: 887 case NDISC_NEIGHBOUR_SOLICITATION: 888 case NDISC_NEIGHBOUR_ADVERTISEMENT: 889 case NDISC_REDIRECT: 890 rc = true; 891 break; 892 } 893 } 894 895 out: 896 return rc; 897 } 898 899 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 900 const struct net_device *dev, 901 struct flowi6 *fl6, 902 int ifindex, 903 int flags) 904 { 905 struct net_vrf *vrf = netdev_priv(dev); 906 struct fib6_table *table = NULL; 907 struct rt6_info *rt6; 908 909 rcu_read_lock(); 910 911 /* fib6_table does not have a refcnt and can not be freed */ 912 rt6 = rcu_dereference(vrf->rt6); 913 if (likely(rt6)) 914 table = rt6->rt6i_table; 915 916 rcu_read_unlock(); 917 918 if (!table) 919 return NULL; 920 921 return ip6_pol_route(net, table, ifindex, fl6, flags); 922 } 923 924 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 925 int ifindex) 926 { 927 const struct ipv6hdr *iph = ipv6_hdr(skb); 928 struct flowi6 fl6 = { 929 .daddr = iph->daddr, 930 .saddr = iph->saddr, 931 .flowlabel = ip6_flowinfo(iph), 932 .flowi6_mark = skb->mark, 933 .flowi6_proto = iph->nexthdr, 934 .flowi6_iif = ifindex, 935 }; 936 struct net *net = dev_net(vrf_dev); 937 struct rt6_info *rt6; 938 939 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, 940 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 941 if (unlikely(!rt6)) 942 return; 943 944 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 945 return; 946 947 skb_dst_set(skb, &rt6->dst); 948 } 949 950 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 951 struct sk_buff *skb) 952 { 953 int orig_iif = skb->skb_iif; 954 bool need_strict; 955 956 /* loopback traffic; do not push through packet taps again. 957 * Reset pkt_type for upper layers to process skb 958 */ 959 if (skb->pkt_type == PACKET_LOOPBACK) { 960 skb->dev = vrf_dev; 961 skb->skb_iif = vrf_dev->ifindex; 962 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 963 skb->pkt_type = PACKET_HOST; 964 goto out; 965 } 966 967 /* if packet is NDISC or addressed to multicast or link-local 968 * then keep the ingress interface 969 */ 970 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 971 if (!ipv6_ndisc_frame(skb) && !need_strict) { 972 vrf_rx_stats(vrf_dev, skb->len); 973 skb->dev = vrf_dev; 974 skb->skb_iif = vrf_dev->ifindex; 975 976 skb_push(skb, skb->mac_len); 977 dev_queue_xmit_nit(skb, vrf_dev); 978 skb_pull(skb, skb->mac_len); 979 980 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 981 } 982 983 if (need_strict) 984 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 985 986 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 987 out: 988 return skb; 989 } 990 991 #else 992 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 993 struct sk_buff *skb) 994 { 995 return skb; 996 } 997 #endif 998 999 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 1000 struct sk_buff *skb) 1001 { 1002 skb->dev = vrf_dev; 1003 skb->skb_iif = vrf_dev->ifindex; 1004 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1005 1006 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1007 goto out; 1008 1009 /* loopback traffic; do not push through packet taps again. 1010 * Reset pkt_type for upper layers to process skb 1011 */ 1012 if (skb->pkt_type == PACKET_LOOPBACK) { 1013 skb->pkt_type = PACKET_HOST; 1014 goto out; 1015 } 1016 1017 vrf_rx_stats(vrf_dev, skb->len); 1018 1019 skb_push(skb, skb->mac_len); 1020 dev_queue_xmit_nit(skb, vrf_dev); 1021 skb_pull(skb, skb->mac_len); 1022 1023 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1024 out: 1025 return skb; 1026 } 1027 1028 /* called with rcu lock held */ 1029 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1030 struct sk_buff *skb, 1031 u16 proto) 1032 { 1033 switch (proto) { 1034 case AF_INET: 1035 return vrf_ip_rcv(vrf_dev, skb); 1036 case AF_INET6: 1037 return vrf_ip6_rcv(vrf_dev, skb); 1038 } 1039 1040 return skb; 1041 } 1042 1043 #if IS_ENABLED(CONFIG_IPV6) 1044 /* send to link-local or multicast address via interface enslaved to 1045 * VRF device. Force lookup to VRF table without changing flow struct 1046 */ 1047 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1048 struct flowi6 *fl6) 1049 { 1050 struct net *net = dev_net(dev); 1051 int flags = RT6_LOOKUP_F_IFACE; 1052 struct dst_entry *dst = NULL; 1053 struct rt6_info *rt; 1054 1055 /* VRF device does not have a link-local address and 1056 * sending packets to link-local or mcast addresses over 1057 * a VRF device does not make sense 1058 */ 1059 if (fl6->flowi6_oif == dev->ifindex) { 1060 dst = &net->ipv6.ip6_null_entry->dst; 1061 dst_hold(dst); 1062 return dst; 1063 } 1064 1065 if (!ipv6_addr_any(&fl6->saddr)) 1066 flags |= RT6_LOOKUP_F_HAS_SADDR; 1067 1068 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags); 1069 if (rt) 1070 dst = &rt->dst; 1071 1072 return dst; 1073 } 1074 #endif 1075 1076 static const struct l3mdev_ops vrf_l3mdev_ops = { 1077 .l3mdev_fib_table = vrf_fib_table, 1078 .l3mdev_l3_rcv = vrf_l3_rcv, 1079 .l3mdev_l3_out = vrf_l3_out, 1080 #if IS_ENABLED(CONFIG_IPV6) 1081 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1082 #endif 1083 }; 1084 1085 static void vrf_get_drvinfo(struct net_device *dev, 1086 struct ethtool_drvinfo *info) 1087 { 1088 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 1089 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 1090 } 1091 1092 static const struct ethtool_ops vrf_ethtool_ops = { 1093 .get_drvinfo = vrf_get_drvinfo, 1094 }; 1095 1096 static inline size_t vrf_fib_rule_nl_size(void) 1097 { 1098 size_t sz; 1099 1100 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1101 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1102 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1103 1104 return sz; 1105 } 1106 1107 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1108 { 1109 struct fib_rule_hdr *frh; 1110 struct nlmsghdr *nlh; 1111 struct sk_buff *skb; 1112 int err; 1113 1114 if (family == AF_INET6 && !ipv6_mod_enabled()) 1115 return 0; 1116 1117 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1118 if (!skb) 1119 return -ENOMEM; 1120 1121 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1122 if (!nlh) 1123 goto nla_put_failure; 1124 1125 /* rule only needs to appear once */ 1126 nlh->nlmsg_flags &= NLM_F_EXCL; 1127 1128 frh = nlmsg_data(nlh); 1129 memset(frh, 0, sizeof(*frh)); 1130 frh->family = family; 1131 frh->action = FR_ACT_TO_TBL; 1132 1133 if (nla_put_u32(skb, FRA_L3MDEV, 1)) 1134 goto nla_put_failure; 1135 1136 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1137 goto nla_put_failure; 1138 1139 nlmsg_end(skb, nlh); 1140 1141 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1142 skb->sk = dev_net(dev)->rtnl; 1143 if (add_it) { 1144 err = fib_nl_newrule(skb, nlh); 1145 if (err == -EEXIST) 1146 err = 0; 1147 } else { 1148 err = fib_nl_delrule(skb, nlh); 1149 if (err == -ENOENT) 1150 err = 0; 1151 } 1152 nlmsg_free(skb); 1153 1154 return err; 1155 1156 nla_put_failure: 1157 nlmsg_free(skb); 1158 1159 return -EMSGSIZE; 1160 } 1161 1162 static int vrf_add_fib_rules(const struct net_device *dev) 1163 { 1164 int err; 1165 1166 err = vrf_fib_rule(dev, AF_INET, true); 1167 if (err < 0) 1168 goto out_err; 1169 1170 err = vrf_fib_rule(dev, AF_INET6, true); 1171 if (err < 0) 1172 goto ipv6_err; 1173 1174 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1175 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1176 if (err < 0) 1177 goto ipmr_err; 1178 #endif 1179 1180 return 0; 1181 1182 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1183 ipmr_err: 1184 vrf_fib_rule(dev, AF_INET6, false); 1185 #endif 1186 1187 ipv6_err: 1188 vrf_fib_rule(dev, AF_INET, false); 1189 1190 out_err: 1191 netdev_err(dev, "Failed to add FIB rules.\n"); 1192 return err; 1193 } 1194 1195 static void vrf_setup(struct net_device *dev) 1196 { 1197 ether_setup(dev); 1198 1199 /* Initialize the device structure. */ 1200 dev->netdev_ops = &vrf_netdev_ops; 1201 dev->l3mdev_ops = &vrf_l3mdev_ops; 1202 dev->ethtool_ops = &vrf_ethtool_ops; 1203 dev->destructor = free_netdev; 1204 1205 /* Fill in device structure with ethernet-generic values. */ 1206 eth_hw_addr_random(dev); 1207 1208 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1209 dev->features |= NETIF_F_LLTX; 1210 1211 /* don't allow vrf devices to change network namespaces. */ 1212 dev->features |= NETIF_F_NETNS_LOCAL; 1213 1214 /* does not make sense for a VLAN to be added to a vrf device */ 1215 dev->features |= NETIF_F_VLAN_CHALLENGED; 1216 1217 /* enable offload features */ 1218 dev->features |= NETIF_F_GSO_SOFTWARE; 1219 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM; 1220 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1221 1222 dev->hw_features = dev->features; 1223 dev->hw_enc_features = dev->features; 1224 1225 /* default to no qdisc; user can add if desired */ 1226 dev->priv_flags |= IFF_NO_QUEUE; 1227 } 1228 1229 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) 1230 { 1231 if (tb[IFLA_ADDRESS]) { 1232 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1233 return -EINVAL; 1234 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1235 return -EADDRNOTAVAIL; 1236 } 1237 return 0; 1238 } 1239 1240 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1241 { 1242 unregister_netdevice_queue(dev, head); 1243 } 1244 1245 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1246 struct nlattr *tb[], struct nlattr *data[]) 1247 { 1248 struct net_vrf *vrf = netdev_priv(dev); 1249 int err; 1250 1251 if (!data || !data[IFLA_VRF_TABLE]) 1252 return -EINVAL; 1253 1254 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1255 if (vrf->tb_id == RT_TABLE_UNSPEC) 1256 return -EINVAL; 1257 1258 dev->priv_flags |= IFF_L3MDEV_MASTER; 1259 1260 err = register_netdevice(dev); 1261 if (err) 1262 goto out; 1263 1264 if (add_fib_rules) { 1265 err = vrf_add_fib_rules(dev); 1266 if (err) { 1267 unregister_netdevice(dev); 1268 goto out; 1269 } 1270 add_fib_rules = false; 1271 } 1272 1273 out: 1274 return err; 1275 } 1276 1277 static size_t vrf_nl_getsize(const struct net_device *dev) 1278 { 1279 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1280 } 1281 1282 static int vrf_fillinfo(struct sk_buff *skb, 1283 const struct net_device *dev) 1284 { 1285 struct net_vrf *vrf = netdev_priv(dev); 1286 1287 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1288 } 1289 1290 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1291 const struct net_device *slave_dev) 1292 { 1293 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1294 } 1295 1296 static int vrf_fill_slave_info(struct sk_buff *skb, 1297 const struct net_device *vrf_dev, 1298 const struct net_device *slave_dev) 1299 { 1300 struct net_vrf *vrf = netdev_priv(vrf_dev); 1301 1302 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1303 return -EMSGSIZE; 1304 1305 return 0; 1306 } 1307 1308 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1309 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1310 }; 1311 1312 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1313 .kind = DRV_NAME, 1314 .priv_size = sizeof(struct net_vrf), 1315 1316 .get_size = vrf_nl_getsize, 1317 .policy = vrf_nl_policy, 1318 .validate = vrf_validate, 1319 .fill_info = vrf_fillinfo, 1320 1321 .get_slave_size = vrf_get_slave_size, 1322 .fill_slave_info = vrf_fill_slave_info, 1323 1324 .newlink = vrf_newlink, 1325 .dellink = vrf_dellink, 1326 .setup = vrf_setup, 1327 .maxtype = IFLA_VRF_MAX, 1328 }; 1329 1330 static int vrf_device_event(struct notifier_block *unused, 1331 unsigned long event, void *ptr) 1332 { 1333 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1334 1335 /* only care about unregister events to drop slave references */ 1336 if (event == NETDEV_UNREGISTER) { 1337 struct net_device *vrf_dev; 1338 1339 if (!netif_is_l3_slave(dev)) 1340 goto out; 1341 1342 vrf_dev = netdev_master_upper_dev_get(dev); 1343 vrf_del_slave(vrf_dev, dev); 1344 } 1345 out: 1346 return NOTIFY_DONE; 1347 } 1348 1349 static struct notifier_block vrf_notifier_block __read_mostly = { 1350 .notifier_call = vrf_device_event, 1351 }; 1352 1353 static int __init vrf_init_module(void) 1354 { 1355 int rc; 1356 1357 register_netdevice_notifier(&vrf_notifier_block); 1358 1359 rc = rtnl_link_register(&vrf_link_ops); 1360 if (rc < 0) 1361 goto error; 1362 1363 return 0; 1364 1365 error: 1366 unregister_netdevice_notifier(&vrf_notifier_block); 1367 return rc; 1368 } 1369 1370 module_init(vrf_init_module); 1371 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 1372 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 1373 MODULE_LICENSE("GPL"); 1374 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 1375 MODULE_VERSION(DRV_VERSION); 1376