1 /* 2 * vrf.c: device driver to encapsulate a VRF space 3 * 4 * Copyright (c) 2015 Cumulus Networks. All rights reserved. 5 * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com> 6 * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com> 7 * 8 * Based on dummy, team and ipvlan drivers 9 * 10 * This program is free software; you can redistribute it and/or modify 11 * it under the terms of the GNU General Public License as published by 12 * the Free Software Foundation; either version 2 of the License, or 13 * (at your option) any later version. 14 */ 15 16 #include <linux/module.h> 17 #include <linux/kernel.h> 18 #include <linux/netdevice.h> 19 #include <linux/etherdevice.h> 20 #include <linux/ip.h> 21 #include <linux/init.h> 22 #include <linux/moduleparam.h> 23 #include <linux/netfilter.h> 24 #include <linux/rtnetlink.h> 25 #include <net/rtnetlink.h> 26 #include <linux/u64_stats_sync.h> 27 #include <linux/hashtable.h> 28 29 #include <linux/inetdevice.h> 30 #include <net/arp.h> 31 #include <net/ip.h> 32 #include <net/ip_fib.h> 33 #include <net/ip6_fib.h> 34 #include <net/ip6_route.h> 35 #include <net/route.h> 36 #include <net/addrconf.h> 37 #include <net/l3mdev.h> 38 #include <net/fib_rules.h> 39 40 #define RT_FL_TOS(oldflp4) \ 41 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)) 42 43 #define DRV_NAME "vrf" 44 #define DRV_VERSION "1.0" 45 46 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */ 47 static bool add_fib_rules = true; 48 49 struct net_vrf { 50 struct rtable __rcu *rth; 51 struct rtable __rcu *rth_local; 52 struct rt6_info __rcu *rt6; 53 struct rt6_info __rcu *rt6_local; 54 u32 tb_id; 55 }; 56 57 struct pcpu_dstats { 58 u64 tx_pkts; 59 u64 tx_bytes; 60 u64 tx_drps; 61 u64 rx_pkts; 62 u64 rx_bytes; 63 u64 rx_drps; 64 struct u64_stats_sync syncp; 65 }; 66 67 static void vrf_rx_stats(struct net_device *dev, int len) 68 { 69 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 70 71 u64_stats_update_begin(&dstats->syncp); 72 dstats->rx_pkts++; 73 dstats->rx_bytes += len; 74 u64_stats_update_end(&dstats->syncp); 75 } 76 77 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 78 { 79 vrf_dev->stats.tx_errors++; 80 kfree_skb(skb); 81 } 82 83 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev, 84 struct rtnl_link_stats64 *stats) 85 { 86 int i; 87 88 for_each_possible_cpu(i) { 89 const struct pcpu_dstats *dstats; 90 u64 tbytes, tpkts, tdrops, rbytes, rpkts; 91 unsigned int start; 92 93 dstats = per_cpu_ptr(dev->dstats, i); 94 do { 95 start = u64_stats_fetch_begin_irq(&dstats->syncp); 96 tbytes = dstats->tx_bytes; 97 tpkts = dstats->tx_pkts; 98 tdrops = dstats->tx_drps; 99 rbytes = dstats->rx_bytes; 100 rpkts = dstats->rx_pkts; 101 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start)); 102 stats->tx_bytes += tbytes; 103 stats->tx_packets += tpkts; 104 stats->tx_dropped += tdrops; 105 stats->rx_bytes += rbytes; 106 stats->rx_packets += rpkts; 107 } 108 return stats; 109 } 110 111 /* Local traffic destined to local address. Reinsert the packet to rx 112 * path, similar to loopback handling. 113 */ 114 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 115 struct dst_entry *dst) 116 { 117 int len = skb->len; 118 119 skb_orphan(skb); 120 121 skb_dst_set(skb, dst); 122 skb_dst_force(skb); 123 124 /* set pkt_type to avoid skb hitting packet taps twice - 125 * once on Tx and again in Rx processing 126 */ 127 skb->pkt_type = PACKET_LOOPBACK; 128 129 skb->protocol = eth_type_trans(skb, dev); 130 131 if (likely(netif_rx(skb) == NET_RX_SUCCESS)) 132 vrf_rx_stats(dev, len); 133 else 134 this_cpu_inc(dev->dstats->rx_drps); 135 136 return NETDEV_TX_OK; 137 } 138 139 #if IS_ENABLED(CONFIG_IPV6) 140 static int vrf_ip6_local_out(struct net *net, struct sock *sk, 141 struct sk_buff *skb) 142 { 143 int err; 144 145 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, 146 sk, skb, NULL, skb_dst(skb)->dev, dst_output); 147 148 if (likely(err == 1)) 149 err = dst_output(net, sk, skb); 150 151 return err; 152 } 153 154 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 155 struct net_device *dev) 156 { 157 const struct ipv6hdr *iph = ipv6_hdr(skb); 158 struct net *net = dev_net(skb->dev); 159 struct flowi6 fl6 = { 160 /* needed to match OIF rule */ 161 .flowi6_oif = dev->ifindex, 162 .flowi6_iif = LOOPBACK_IFINDEX, 163 .daddr = iph->daddr, 164 .saddr = iph->saddr, 165 .flowlabel = ip6_flowinfo(iph), 166 .flowi6_mark = skb->mark, 167 .flowi6_proto = iph->nexthdr, 168 .flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF, 169 }; 170 int ret = NET_XMIT_DROP; 171 struct dst_entry *dst; 172 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 173 174 dst = ip6_route_output(net, NULL, &fl6); 175 if (dst == dst_null) 176 goto err; 177 178 skb_dst_drop(skb); 179 180 /* if dst.dev is loopback or the VRF device again this is locally 181 * originated traffic destined to a local address. Short circuit 182 * to Rx path using our local dst 183 */ 184 if (dst->dev == net->loopback_dev || dst->dev == dev) { 185 struct net_vrf *vrf = netdev_priv(dev); 186 struct rt6_info *rt6_local; 187 188 /* release looked up dst and use cached local dst */ 189 dst_release(dst); 190 191 rcu_read_lock(); 192 193 rt6_local = rcu_dereference(vrf->rt6_local); 194 if (unlikely(!rt6_local)) { 195 rcu_read_unlock(); 196 goto err; 197 } 198 199 /* Ordering issue: cached local dst is created on newlink 200 * before the IPv6 initialization. Using the local dst 201 * requires rt6i_idev to be set so make sure it is. 202 */ 203 if (unlikely(!rt6_local->rt6i_idev)) { 204 rt6_local->rt6i_idev = in6_dev_get(dev); 205 if (!rt6_local->rt6i_idev) { 206 rcu_read_unlock(); 207 goto err; 208 } 209 } 210 211 dst = &rt6_local->dst; 212 dst_hold(dst); 213 214 rcu_read_unlock(); 215 216 return vrf_local_xmit(skb, dev, &rt6_local->dst); 217 } 218 219 skb_dst_set(skb, dst); 220 221 /* strip the ethernet header added for pass through VRF device */ 222 __skb_pull(skb, skb_network_offset(skb)); 223 224 ret = vrf_ip6_local_out(net, skb->sk, skb); 225 if (unlikely(net_xmit_eval(ret))) 226 dev->stats.tx_errors++; 227 else 228 ret = NET_XMIT_SUCCESS; 229 230 return ret; 231 err: 232 vrf_tx_error(dev, skb); 233 return NET_XMIT_DROP; 234 } 235 #else 236 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 237 struct net_device *dev) 238 { 239 vrf_tx_error(dev, skb); 240 return NET_XMIT_DROP; 241 } 242 #endif 243 244 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 245 static int vrf_ip_local_out(struct net *net, struct sock *sk, 246 struct sk_buff *skb) 247 { 248 int err; 249 250 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 251 skb, NULL, skb_dst(skb)->dev, dst_output); 252 if (likely(err == 1)) 253 err = dst_output(net, sk, skb); 254 255 return err; 256 } 257 258 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 259 struct net_device *vrf_dev) 260 { 261 struct iphdr *ip4h = ip_hdr(skb); 262 int ret = NET_XMIT_DROP; 263 struct flowi4 fl4 = { 264 /* needed to match OIF rule */ 265 .flowi4_oif = vrf_dev->ifindex, 266 .flowi4_iif = LOOPBACK_IFINDEX, 267 .flowi4_tos = RT_TOS(ip4h->tos), 268 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC | 269 FLOWI_FLAG_SKIP_NH_OIF, 270 .daddr = ip4h->daddr, 271 }; 272 struct net *net = dev_net(vrf_dev); 273 struct rtable *rt; 274 275 rt = ip_route_output_flow(net, &fl4, NULL); 276 if (IS_ERR(rt)) 277 goto err; 278 279 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { 280 ip_rt_put(rt); 281 goto err; 282 } 283 284 skb_dst_drop(skb); 285 286 /* if dst.dev is loopback or the VRF device again this is locally 287 * originated traffic destined to a local address. Short circuit 288 * to Rx path using our local dst 289 */ 290 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) { 291 struct net_vrf *vrf = netdev_priv(vrf_dev); 292 struct rtable *rth_local; 293 struct dst_entry *dst = NULL; 294 295 ip_rt_put(rt); 296 297 rcu_read_lock(); 298 299 rth_local = rcu_dereference(vrf->rth_local); 300 if (likely(rth_local)) { 301 dst = &rth_local->dst; 302 dst_hold(dst); 303 } 304 305 rcu_read_unlock(); 306 307 if (unlikely(!dst)) 308 goto err; 309 310 return vrf_local_xmit(skb, vrf_dev, dst); 311 } 312 313 skb_dst_set(skb, &rt->dst); 314 315 /* strip the ethernet header added for pass through VRF device */ 316 __skb_pull(skb, skb_network_offset(skb)); 317 318 if (!ip4h->saddr) { 319 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 320 RT_SCOPE_LINK); 321 } 322 323 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 324 if (unlikely(net_xmit_eval(ret))) 325 vrf_dev->stats.tx_errors++; 326 else 327 ret = NET_XMIT_SUCCESS; 328 329 out: 330 return ret; 331 err: 332 vrf_tx_error(vrf_dev, skb); 333 goto out; 334 } 335 336 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 337 { 338 switch (skb->protocol) { 339 case htons(ETH_P_IP): 340 return vrf_process_v4_outbound(skb, dev); 341 case htons(ETH_P_IPV6): 342 return vrf_process_v6_outbound(skb, dev); 343 default: 344 vrf_tx_error(dev, skb); 345 return NET_XMIT_DROP; 346 } 347 } 348 349 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 350 { 351 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 352 353 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 354 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 355 356 u64_stats_update_begin(&dstats->syncp); 357 dstats->tx_pkts++; 358 dstats->tx_bytes += skb->len; 359 u64_stats_update_end(&dstats->syncp); 360 } else { 361 this_cpu_inc(dev->dstats->tx_drps); 362 } 363 364 return ret; 365 } 366 367 #if IS_ENABLED(CONFIG_IPV6) 368 /* modelled after ip6_finish_output2 */ 369 static int vrf_finish_output6(struct net *net, struct sock *sk, 370 struct sk_buff *skb) 371 { 372 struct dst_entry *dst = skb_dst(skb); 373 struct net_device *dev = dst->dev; 374 struct neighbour *neigh; 375 struct in6_addr *nexthop; 376 int ret; 377 378 skb->protocol = htons(ETH_P_IPV6); 379 skb->dev = dev; 380 381 rcu_read_lock_bh(); 382 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 383 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 384 if (unlikely(!neigh)) 385 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 386 if (!IS_ERR(neigh)) { 387 ret = dst_neigh_output(dst, neigh, skb); 388 rcu_read_unlock_bh(); 389 return ret; 390 } 391 rcu_read_unlock_bh(); 392 393 IP6_INC_STATS(dev_net(dst->dev), 394 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 395 kfree_skb(skb); 396 return -EINVAL; 397 } 398 399 /* modelled after ip6_output */ 400 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 401 { 402 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 403 net, sk, skb, NULL, skb_dst(skb)->dev, 404 vrf_finish_output6, 405 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 406 } 407 408 /* set dst on skb to send packet to us via dev_xmit path. Allows 409 * packet to go through device based features such as qdisc, netfilter 410 * hooks and packet sockets with skb->dev set to vrf device. 411 */ 412 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 413 struct sock *sk, 414 struct sk_buff *skb) 415 { 416 struct net_vrf *vrf = netdev_priv(vrf_dev); 417 struct dst_entry *dst = NULL; 418 struct rt6_info *rt6; 419 420 /* don't divert link scope packets */ 421 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 422 return skb; 423 424 rcu_read_lock(); 425 426 rt6 = rcu_dereference(vrf->rt6); 427 if (likely(rt6)) { 428 dst = &rt6->dst; 429 dst_hold(dst); 430 } 431 432 rcu_read_unlock(); 433 434 if (unlikely(!dst)) { 435 vrf_tx_error(vrf_dev, skb); 436 return NULL; 437 } 438 439 skb_dst_drop(skb); 440 skb_dst_set(skb, dst); 441 442 return skb; 443 } 444 445 /* holding rtnl */ 446 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 447 { 448 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 449 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local); 450 struct net *net = dev_net(dev); 451 struct dst_entry *dst; 452 453 RCU_INIT_POINTER(vrf->rt6, NULL); 454 RCU_INIT_POINTER(vrf->rt6_local, NULL); 455 synchronize_rcu(); 456 457 /* move dev in dst's to loopback so this VRF device can be deleted 458 * - based on dst_ifdown 459 */ 460 if (rt6) { 461 dst = &rt6->dst; 462 dev_put(dst->dev); 463 dst->dev = net->loopback_dev; 464 dev_hold(dst->dev); 465 dst_release(dst); 466 } 467 468 if (rt6_local) { 469 if (rt6_local->rt6i_idev) 470 in6_dev_put(rt6_local->rt6i_idev); 471 472 dst = &rt6_local->dst; 473 dev_put(dst->dev); 474 dst->dev = net->loopback_dev; 475 dev_hold(dst->dev); 476 dst_release(dst); 477 } 478 } 479 480 static int vrf_rt6_create(struct net_device *dev) 481 { 482 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE; 483 struct net_vrf *vrf = netdev_priv(dev); 484 struct net *net = dev_net(dev); 485 struct fib6_table *rt6i_table; 486 struct rt6_info *rt6, *rt6_local; 487 int rc = -ENOMEM; 488 489 /* IPv6 can be CONFIG enabled and then disabled runtime */ 490 if (!ipv6_mod_enabled()) 491 return 0; 492 493 rt6i_table = fib6_new_table(net, vrf->tb_id); 494 if (!rt6i_table) 495 goto out; 496 497 /* create a dst for routing packets out a VRF device */ 498 rt6 = ip6_dst_alloc(net, dev, flags); 499 if (!rt6) 500 goto out; 501 502 dst_hold(&rt6->dst); 503 504 rt6->rt6i_table = rt6i_table; 505 rt6->dst.output = vrf_output6; 506 507 /* create a dst for local routing - packets sent locally 508 * to local address via the VRF device as a loopback 509 */ 510 rt6_local = ip6_dst_alloc(net, dev, flags); 511 if (!rt6_local) { 512 dst_release(&rt6->dst); 513 goto out; 514 } 515 516 dst_hold(&rt6_local->dst); 517 518 rt6_local->rt6i_idev = in6_dev_get(dev); 519 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL; 520 rt6_local->rt6i_table = rt6i_table; 521 rt6_local->dst.input = ip6_input; 522 523 rcu_assign_pointer(vrf->rt6, rt6); 524 rcu_assign_pointer(vrf->rt6_local, rt6_local); 525 526 rc = 0; 527 out: 528 return rc; 529 } 530 #else 531 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 532 struct sock *sk, 533 struct sk_buff *skb) 534 { 535 return skb; 536 } 537 538 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 539 { 540 } 541 542 static int vrf_rt6_create(struct net_device *dev) 543 { 544 return 0; 545 } 546 #endif 547 548 /* modelled after ip_finish_output2 */ 549 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 550 { 551 struct dst_entry *dst = skb_dst(skb); 552 struct rtable *rt = (struct rtable *)dst; 553 struct net_device *dev = dst->dev; 554 unsigned int hh_len = LL_RESERVED_SPACE(dev); 555 struct neighbour *neigh; 556 u32 nexthop; 557 int ret = -EINVAL; 558 559 /* Be paranoid, rather than too clever. */ 560 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 561 struct sk_buff *skb2; 562 563 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 564 if (!skb2) { 565 ret = -ENOMEM; 566 goto err; 567 } 568 if (skb->sk) 569 skb_set_owner_w(skb2, skb->sk); 570 571 consume_skb(skb); 572 skb = skb2; 573 } 574 575 rcu_read_lock_bh(); 576 577 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); 578 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 579 if (unlikely(!neigh)) 580 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 581 if (!IS_ERR(neigh)) 582 ret = dst_neigh_output(dst, neigh, skb); 583 584 rcu_read_unlock_bh(); 585 err: 586 if (unlikely(ret < 0)) 587 vrf_tx_error(skb->dev, skb); 588 return ret; 589 } 590 591 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 592 { 593 struct net_device *dev = skb_dst(skb)->dev; 594 595 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 596 597 skb->dev = dev; 598 skb->protocol = htons(ETH_P_IP); 599 600 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 601 net, sk, skb, NULL, dev, 602 vrf_finish_output, 603 !(IPCB(skb)->flags & IPSKB_REROUTED)); 604 } 605 606 /* set dst on skb to send packet to us via dev_xmit path. Allows 607 * packet to go through device based features such as qdisc, netfilter 608 * hooks and packet sockets with skb->dev set to vrf device. 609 */ 610 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 611 struct sock *sk, 612 struct sk_buff *skb) 613 { 614 struct net_vrf *vrf = netdev_priv(vrf_dev); 615 struct dst_entry *dst = NULL; 616 struct rtable *rth; 617 618 rcu_read_lock(); 619 620 rth = rcu_dereference(vrf->rth); 621 if (likely(rth)) { 622 dst = &rth->dst; 623 dst_hold(dst); 624 } 625 626 rcu_read_unlock(); 627 628 if (unlikely(!dst)) { 629 vrf_tx_error(vrf_dev, skb); 630 return NULL; 631 } 632 633 skb_dst_drop(skb); 634 skb_dst_set(skb, dst); 635 636 return skb; 637 } 638 639 /* called with rcu lock held */ 640 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 641 struct sock *sk, 642 struct sk_buff *skb, 643 u16 proto) 644 { 645 switch (proto) { 646 case AF_INET: 647 return vrf_ip_out(vrf_dev, sk, skb); 648 case AF_INET6: 649 return vrf_ip6_out(vrf_dev, sk, skb); 650 } 651 652 return skb; 653 } 654 655 /* holding rtnl */ 656 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 657 { 658 struct rtable *rth = rtnl_dereference(vrf->rth); 659 struct rtable *rth_local = rtnl_dereference(vrf->rth_local); 660 struct net *net = dev_net(dev); 661 struct dst_entry *dst; 662 663 RCU_INIT_POINTER(vrf->rth, NULL); 664 RCU_INIT_POINTER(vrf->rth_local, NULL); 665 synchronize_rcu(); 666 667 /* move dev in dst's to loopback so this VRF device can be deleted 668 * - based on dst_ifdown 669 */ 670 if (rth) { 671 dst = &rth->dst; 672 dev_put(dst->dev); 673 dst->dev = net->loopback_dev; 674 dev_hold(dst->dev); 675 dst_release(dst); 676 } 677 678 if (rth_local) { 679 dst = &rth_local->dst; 680 dev_put(dst->dev); 681 dst->dev = net->loopback_dev; 682 dev_hold(dst->dev); 683 dst_release(dst); 684 } 685 } 686 687 static int vrf_rtable_create(struct net_device *dev) 688 { 689 struct net_vrf *vrf = netdev_priv(dev); 690 struct rtable *rth, *rth_local; 691 692 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 693 return -ENOMEM; 694 695 /* create a dst for routing packets out through a VRF device */ 696 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0); 697 if (!rth) 698 return -ENOMEM; 699 700 /* create a dst for local ingress routing - packets sent locally 701 * to local address via the VRF device as a loopback 702 */ 703 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0); 704 if (!rth_local) { 705 dst_release(&rth->dst); 706 return -ENOMEM; 707 } 708 709 rth->dst.output = vrf_output; 710 rth->rt_table_id = vrf->tb_id; 711 712 rth_local->rt_table_id = vrf->tb_id; 713 714 rcu_assign_pointer(vrf->rth, rth); 715 rcu_assign_pointer(vrf->rth_local, rth_local); 716 717 return 0; 718 } 719 720 /**************************** device handling ********************/ 721 722 /* cycle interface to flush neighbor cache and move routes across tables */ 723 static void cycle_netdev(struct net_device *dev) 724 { 725 unsigned int flags = dev->flags; 726 int ret; 727 728 if (!netif_running(dev)) 729 return; 730 731 ret = dev_change_flags(dev, flags & ~IFF_UP); 732 if (ret >= 0) 733 ret = dev_change_flags(dev, flags); 734 735 if (ret < 0) { 736 netdev_err(dev, 737 "Failed to cycle device %s; route tables might be wrong!\n", 738 dev->name); 739 } 740 } 741 742 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 743 { 744 int ret; 745 746 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL); 747 if (ret < 0) 748 return ret; 749 750 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 751 cycle_netdev(port_dev); 752 753 return 0; 754 } 755 756 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 757 { 758 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev)) 759 return -EINVAL; 760 761 return do_vrf_add_slave(dev, port_dev); 762 } 763 764 /* inverse of do_vrf_add_slave */ 765 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 766 { 767 netdev_upper_dev_unlink(port_dev, dev); 768 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 769 770 cycle_netdev(port_dev); 771 772 return 0; 773 } 774 775 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 776 { 777 return do_vrf_del_slave(dev, port_dev); 778 } 779 780 static void vrf_dev_uninit(struct net_device *dev) 781 { 782 struct net_vrf *vrf = netdev_priv(dev); 783 struct net_device *port_dev; 784 struct list_head *iter; 785 786 vrf_rtable_release(dev, vrf); 787 vrf_rt6_release(dev, vrf); 788 789 netdev_for_each_lower_dev(dev, port_dev, iter) 790 vrf_del_slave(dev, port_dev); 791 792 free_percpu(dev->dstats); 793 dev->dstats = NULL; 794 } 795 796 static int vrf_dev_init(struct net_device *dev) 797 { 798 struct net_vrf *vrf = netdev_priv(dev); 799 800 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 801 if (!dev->dstats) 802 goto out_nomem; 803 804 /* create the default dst which points back to us */ 805 if (vrf_rtable_create(dev) != 0) 806 goto out_stats; 807 808 if (vrf_rt6_create(dev) != 0) 809 goto out_rth; 810 811 dev->flags = IFF_MASTER | IFF_NOARP; 812 813 /* MTU is irrelevant for VRF device; set to 64k similar to lo */ 814 dev->mtu = 64 * 1024; 815 816 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 817 dev->operstate = IF_OPER_UP; 818 netdev_lockdep_set_classes(dev); 819 return 0; 820 821 out_rth: 822 vrf_rtable_release(dev, vrf); 823 out_stats: 824 free_percpu(dev->dstats); 825 dev->dstats = NULL; 826 out_nomem: 827 return -ENOMEM; 828 } 829 830 static const struct net_device_ops vrf_netdev_ops = { 831 .ndo_init = vrf_dev_init, 832 .ndo_uninit = vrf_dev_uninit, 833 .ndo_start_xmit = vrf_xmit, 834 .ndo_get_stats64 = vrf_get_stats64, 835 .ndo_add_slave = vrf_add_slave, 836 .ndo_del_slave = vrf_del_slave, 837 }; 838 839 static u32 vrf_fib_table(const struct net_device *dev) 840 { 841 struct net_vrf *vrf = netdev_priv(dev); 842 843 return vrf->tb_id; 844 } 845 846 static struct rtable *vrf_get_rtable(const struct net_device *dev, 847 const struct flowi4 *fl4) 848 { 849 struct rtable *rth = NULL; 850 851 if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) { 852 struct net_vrf *vrf = netdev_priv(dev); 853 854 rcu_read_lock(); 855 856 rth = rcu_dereference(vrf->rth); 857 if (likely(rth)) 858 dst_hold(&rth->dst); 859 860 rcu_read_unlock(); 861 } 862 863 return rth; 864 } 865 866 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 867 { 868 return 0; 869 } 870 871 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 872 struct sk_buff *skb, 873 struct net_device *dev) 874 { 875 struct net *net = dev_net(dev); 876 877 nf_reset(skb); 878 879 if (NF_HOOK(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) < 0) 880 skb = NULL; /* kfree_skb(skb) handled by nf code */ 881 882 return skb; 883 } 884 885 #if IS_ENABLED(CONFIG_IPV6) 886 /* neighbor handling is done with actual device; do not want 887 * to flip skb->dev for those ndisc packets. This really fails 888 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 889 * a start. 890 */ 891 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 892 { 893 const struct ipv6hdr *iph = ipv6_hdr(skb); 894 bool rc = false; 895 896 if (iph->nexthdr == NEXTHDR_ICMP) { 897 const struct icmp6hdr *icmph; 898 struct icmp6hdr _icmph; 899 900 icmph = skb_header_pointer(skb, sizeof(*iph), 901 sizeof(_icmph), &_icmph); 902 if (!icmph) 903 goto out; 904 905 switch (icmph->icmp6_type) { 906 case NDISC_ROUTER_SOLICITATION: 907 case NDISC_ROUTER_ADVERTISEMENT: 908 case NDISC_NEIGHBOUR_SOLICITATION: 909 case NDISC_NEIGHBOUR_ADVERTISEMENT: 910 case NDISC_REDIRECT: 911 rc = true; 912 break; 913 } 914 } 915 916 out: 917 return rc; 918 } 919 920 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 921 const struct net_device *dev, 922 struct flowi6 *fl6, 923 int ifindex, 924 int flags) 925 { 926 struct net_vrf *vrf = netdev_priv(dev); 927 struct fib6_table *table = NULL; 928 struct rt6_info *rt6; 929 930 rcu_read_lock(); 931 932 /* fib6_table does not have a refcnt and can not be freed */ 933 rt6 = rcu_dereference(vrf->rt6); 934 if (likely(rt6)) 935 table = rt6->rt6i_table; 936 937 rcu_read_unlock(); 938 939 if (!table) 940 return NULL; 941 942 return ip6_pol_route(net, table, ifindex, fl6, flags); 943 } 944 945 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 946 int ifindex) 947 { 948 const struct ipv6hdr *iph = ipv6_hdr(skb); 949 struct flowi6 fl6 = { 950 .daddr = iph->daddr, 951 .saddr = iph->saddr, 952 .flowlabel = ip6_flowinfo(iph), 953 .flowi6_mark = skb->mark, 954 .flowi6_proto = iph->nexthdr, 955 .flowi6_iif = ifindex, 956 }; 957 struct net *net = dev_net(vrf_dev); 958 struct rt6_info *rt6; 959 960 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, 961 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 962 if (unlikely(!rt6)) 963 return; 964 965 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 966 return; 967 968 skb_dst_set(skb, &rt6->dst); 969 } 970 971 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 972 struct sk_buff *skb) 973 { 974 int orig_iif = skb->skb_iif; 975 bool need_strict; 976 977 /* loopback traffic; do not push through packet taps again. 978 * Reset pkt_type for upper layers to process skb 979 */ 980 if (skb->pkt_type == PACKET_LOOPBACK) { 981 skb->dev = vrf_dev; 982 skb->skb_iif = vrf_dev->ifindex; 983 skb->pkt_type = PACKET_HOST; 984 goto out; 985 } 986 987 /* if packet is NDISC or addressed to multicast or link-local 988 * then keep the ingress interface 989 */ 990 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 991 if (!ipv6_ndisc_frame(skb) && !need_strict) { 992 skb->dev = vrf_dev; 993 skb->skb_iif = vrf_dev->ifindex; 994 995 skb_push(skb, skb->mac_len); 996 dev_queue_xmit_nit(skb, vrf_dev); 997 skb_pull(skb, skb->mac_len); 998 999 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1000 } 1001 1002 if (need_strict) 1003 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1004 1005 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 1006 out: 1007 return skb; 1008 } 1009 1010 #else 1011 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1012 struct sk_buff *skb) 1013 { 1014 return skb; 1015 } 1016 #endif 1017 1018 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 1019 struct sk_buff *skb) 1020 { 1021 skb->dev = vrf_dev; 1022 skb->skb_iif = vrf_dev->ifindex; 1023 1024 /* loopback traffic; do not push through packet taps again. 1025 * Reset pkt_type for upper layers to process skb 1026 */ 1027 if (skb->pkt_type == PACKET_LOOPBACK) { 1028 skb->pkt_type = PACKET_HOST; 1029 goto out; 1030 } 1031 1032 skb_push(skb, skb->mac_len); 1033 dev_queue_xmit_nit(skb, vrf_dev); 1034 skb_pull(skb, skb->mac_len); 1035 1036 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1037 out: 1038 return skb; 1039 } 1040 1041 /* called with rcu lock held */ 1042 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1043 struct sk_buff *skb, 1044 u16 proto) 1045 { 1046 switch (proto) { 1047 case AF_INET: 1048 return vrf_ip_rcv(vrf_dev, skb); 1049 case AF_INET6: 1050 return vrf_ip6_rcv(vrf_dev, skb); 1051 } 1052 1053 return skb; 1054 } 1055 1056 #if IS_ENABLED(CONFIG_IPV6) 1057 /* send to link-local or multicast address via interface enslaved to 1058 * VRF device. Force lookup to VRF table without changing flow struct 1059 */ 1060 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1061 struct flowi6 *fl6) 1062 { 1063 struct net *net = dev_net(dev); 1064 int flags = RT6_LOOKUP_F_IFACE; 1065 struct dst_entry *dst = NULL; 1066 struct rt6_info *rt; 1067 1068 /* VRF device does not have a link-local address and 1069 * sending packets to link-local or mcast addresses over 1070 * a VRF device does not make sense 1071 */ 1072 if (fl6->flowi6_oif == dev->ifindex) { 1073 dst = &net->ipv6.ip6_null_entry->dst; 1074 dst_hold(dst); 1075 return dst; 1076 } 1077 1078 if (!ipv6_addr_any(&fl6->saddr)) 1079 flags |= RT6_LOOKUP_F_HAS_SADDR; 1080 1081 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags); 1082 if (rt) 1083 dst = &rt->dst; 1084 1085 return dst; 1086 } 1087 #endif 1088 1089 static const struct l3mdev_ops vrf_l3mdev_ops = { 1090 .l3mdev_fib_table = vrf_fib_table, 1091 .l3mdev_get_rtable = vrf_get_rtable, 1092 .l3mdev_l3_rcv = vrf_l3_rcv, 1093 .l3mdev_l3_out = vrf_l3_out, 1094 #if IS_ENABLED(CONFIG_IPV6) 1095 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1096 #endif 1097 }; 1098 1099 static void vrf_get_drvinfo(struct net_device *dev, 1100 struct ethtool_drvinfo *info) 1101 { 1102 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 1103 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 1104 } 1105 1106 static const struct ethtool_ops vrf_ethtool_ops = { 1107 .get_drvinfo = vrf_get_drvinfo, 1108 }; 1109 1110 static inline size_t vrf_fib_rule_nl_size(void) 1111 { 1112 size_t sz; 1113 1114 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1115 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1116 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1117 1118 return sz; 1119 } 1120 1121 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1122 { 1123 struct fib_rule_hdr *frh; 1124 struct nlmsghdr *nlh; 1125 struct sk_buff *skb; 1126 int err; 1127 1128 if (family == AF_INET6 && !ipv6_mod_enabled()) 1129 return 0; 1130 1131 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1132 if (!skb) 1133 return -ENOMEM; 1134 1135 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1136 if (!nlh) 1137 goto nla_put_failure; 1138 1139 /* rule only needs to appear once */ 1140 nlh->nlmsg_flags &= NLM_F_EXCL; 1141 1142 frh = nlmsg_data(nlh); 1143 memset(frh, 0, sizeof(*frh)); 1144 frh->family = family; 1145 frh->action = FR_ACT_TO_TBL; 1146 1147 if (nla_put_u32(skb, FRA_L3MDEV, 1)) 1148 goto nla_put_failure; 1149 1150 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1151 goto nla_put_failure; 1152 1153 nlmsg_end(skb, nlh); 1154 1155 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1156 skb->sk = dev_net(dev)->rtnl; 1157 if (add_it) { 1158 err = fib_nl_newrule(skb, nlh); 1159 if (err == -EEXIST) 1160 err = 0; 1161 } else { 1162 err = fib_nl_delrule(skb, nlh); 1163 if (err == -ENOENT) 1164 err = 0; 1165 } 1166 nlmsg_free(skb); 1167 1168 return err; 1169 1170 nla_put_failure: 1171 nlmsg_free(skb); 1172 1173 return -EMSGSIZE; 1174 } 1175 1176 static int vrf_add_fib_rules(const struct net_device *dev) 1177 { 1178 int err; 1179 1180 err = vrf_fib_rule(dev, AF_INET, true); 1181 if (err < 0) 1182 goto out_err; 1183 1184 err = vrf_fib_rule(dev, AF_INET6, true); 1185 if (err < 0) 1186 goto ipv6_err; 1187 1188 return 0; 1189 1190 ipv6_err: 1191 vrf_fib_rule(dev, AF_INET, false); 1192 1193 out_err: 1194 netdev_err(dev, "Failed to add FIB rules.\n"); 1195 return err; 1196 } 1197 1198 static void vrf_setup(struct net_device *dev) 1199 { 1200 ether_setup(dev); 1201 1202 /* Initialize the device structure. */ 1203 dev->netdev_ops = &vrf_netdev_ops; 1204 dev->l3mdev_ops = &vrf_l3mdev_ops; 1205 dev->ethtool_ops = &vrf_ethtool_ops; 1206 dev->destructor = free_netdev; 1207 1208 /* Fill in device structure with ethernet-generic values. */ 1209 eth_hw_addr_random(dev); 1210 1211 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1212 dev->features |= NETIF_F_LLTX; 1213 1214 /* don't allow vrf devices to change network namespaces. */ 1215 dev->features |= NETIF_F_NETNS_LOCAL; 1216 1217 /* does not make sense for a VLAN to be added to a vrf device */ 1218 dev->features |= NETIF_F_VLAN_CHALLENGED; 1219 1220 /* enable offload features */ 1221 dev->features |= NETIF_F_GSO_SOFTWARE; 1222 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM; 1223 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1224 1225 dev->hw_features = dev->features; 1226 dev->hw_enc_features = dev->features; 1227 1228 /* default to no qdisc; user can add if desired */ 1229 dev->priv_flags |= IFF_NO_QUEUE; 1230 } 1231 1232 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) 1233 { 1234 if (tb[IFLA_ADDRESS]) { 1235 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1236 return -EINVAL; 1237 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1238 return -EADDRNOTAVAIL; 1239 } 1240 return 0; 1241 } 1242 1243 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1244 { 1245 unregister_netdevice_queue(dev, head); 1246 } 1247 1248 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1249 struct nlattr *tb[], struct nlattr *data[]) 1250 { 1251 struct net_vrf *vrf = netdev_priv(dev); 1252 int err; 1253 1254 if (!data || !data[IFLA_VRF_TABLE]) 1255 return -EINVAL; 1256 1257 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1258 1259 dev->priv_flags |= IFF_L3MDEV_MASTER; 1260 1261 err = register_netdevice(dev); 1262 if (err) 1263 goto out; 1264 1265 if (add_fib_rules) { 1266 err = vrf_add_fib_rules(dev); 1267 if (err) { 1268 unregister_netdevice(dev); 1269 goto out; 1270 } 1271 add_fib_rules = false; 1272 } 1273 1274 out: 1275 return err; 1276 } 1277 1278 static size_t vrf_nl_getsize(const struct net_device *dev) 1279 { 1280 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1281 } 1282 1283 static int vrf_fillinfo(struct sk_buff *skb, 1284 const struct net_device *dev) 1285 { 1286 struct net_vrf *vrf = netdev_priv(dev); 1287 1288 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1289 } 1290 1291 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1292 const struct net_device *slave_dev) 1293 { 1294 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1295 } 1296 1297 static int vrf_fill_slave_info(struct sk_buff *skb, 1298 const struct net_device *vrf_dev, 1299 const struct net_device *slave_dev) 1300 { 1301 struct net_vrf *vrf = netdev_priv(vrf_dev); 1302 1303 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1304 return -EMSGSIZE; 1305 1306 return 0; 1307 } 1308 1309 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1310 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1311 }; 1312 1313 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1314 .kind = DRV_NAME, 1315 .priv_size = sizeof(struct net_vrf), 1316 1317 .get_size = vrf_nl_getsize, 1318 .policy = vrf_nl_policy, 1319 .validate = vrf_validate, 1320 .fill_info = vrf_fillinfo, 1321 1322 .get_slave_size = vrf_get_slave_size, 1323 .fill_slave_info = vrf_fill_slave_info, 1324 1325 .newlink = vrf_newlink, 1326 .dellink = vrf_dellink, 1327 .setup = vrf_setup, 1328 .maxtype = IFLA_VRF_MAX, 1329 }; 1330 1331 static int vrf_device_event(struct notifier_block *unused, 1332 unsigned long event, void *ptr) 1333 { 1334 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1335 1336 /* only care about unregister events to drop slave references */ 1337 if (event == NETDEV_UNREGISTER) { 1338 struct net_device *vrf_dev; 1339 1340 if (!netif_is_l3_slave(dev)) 1341 goto out; 1342 1343 vrf_dev = netdev_master_upper_dev_get(dev); 1344 vrf_del_slave(vrf_dev, dev); 1345 } 1346 out: 1347 return NOTIFY_DONE; 1348 } 1349 1350 static struct notifier_block vrf_notifier_block __read_mostly = { 1351 .notifier_call = vrf_device_event, 1352 }; 1353 1354 static int __init vrf_init_module(void) 1355 { 1356 int rc; 1357 1358 register_netdevice_notifier(&vrf_notifier_block); 1359 1360 rc = rtnl_link_register(&vrf_link_ops); 1361 if (rc < 0) 1362 goto error; 1363 1364 return 0; 1365 1366 error: 1367 unregister_netdevice_notifier(&vrf_notifier_block); 1368 return rc; 1369 } 1370 1371 module_init(vrf_init_module); 1372 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 1373 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 1374 MODULE_LICENSE("GPL"); 1375 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 1376 MODULE_VERSION(DRV_VERSION); 1377