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 #include <net/netns/generic.h> 40 41 #define DRV_NAME "vrf" 42 #define DRV_VERSION "1.0" 43 44 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */ 45 46 static unsigned int vrf_net_id; 47 48 struct net_vrf { 49 struct rtable __rcu *rth; 50 struct rtable __rcu *rth_local; 51 struct rt6_info __rcu *rt6; 52 struct rt6_info __rcu *rt6_local; 53 u32 tb_id; 54 }; 55 56 struct pcpu_dstats { 57 u64 tx_pkts; 58 u64 tx_bytes; 59 u64 tx_drps; 60 u64 rx_pkts; 61 u64 rx_bytes; 62 u64 rx_drps; 63 struct u64_stats_sync syncp; 64 }; 65 66 static void vrf_rx_stats(struct net_device *dev, int len) 67 { 68 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 69 70 u64_stats_update_begin(&dstats->syncp); 71 dstats->rx_pkts++; 72 dstats->rx_bytes += len; 73 u64_stats_update_end(&dstats->syncp); 74 } 75 76 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 77 { 78 vrf_dev->stats.tx_errors++; 79 kfree_skb(skb); 80 } 81 82 static void vrf_get_stats64(struct net_device *dev, 83 struct rtnl_link_stats64 *stats) 84 { 85 int i; 86 87 for_each_possible_cpu(i) { 88 const struct pcpu_dstats *dstats; 89 u64 tbytes, tpkts, tdrops, rbytes, rpkts; 90 unsigned int start; 91 92 dstats = per_cpu_ptr(dev->dstats, i); 93 do { 94 start = u64_stats_fetch_begin_irq(&dstats->syncp); 95 tbytes = dstats->tx_bytes; 96 tpkts = dstats->tx_pkts; 97 tdrops = dstats->tx_drps; 98 rbytes = dstats->rx_bytes; 99 rpkts = dstats->rx_pkts; 100 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start)); 101 stats->tx_bytes += tbytes; 102 stats->tx_packets += tpkts; 103 stats->tx_dropped += tdrops; 104 stats->rx_bytes += rbytes; 105 stats->rx_packets += rpkts; 106 } 107 } 108 109 /* by default VRF devices do not have a qdisc and are expected 110 * to be created with only a single queue. 111 */ 112 static bool qdisc_tx_is_default(const struct net_device *dev) 113 { 114 struct netdev_queue *txq; 115 struct Qdisc *qdisc; 116 117 if (dev->num_tx_queues > 1) 118 return false; 119 120 txq = netdev_get_tx_queue(dev, 0); 121 qdisc = rcu_access_pointer(txq->qdisc); 122 123 return !qdisc->enqueue; 124 } 125 126 /* Local traffic destined to local address. Reinsert the packet to rx 127 * path, similar to loopback handling. 128 */ 129 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 130 struct dst_entry *dst) 131 { 132 int len = skb->len; 133 134 skb_orphan(skb); 135 136 skb_dst_set(skb, dst); 137 skb_dst_force(skb); 138 139 /* set pkt_type to avoid skb hitting packet taps twice - 140 * once on Tx and again in Rx processing 141 */ 142 skb->pkt_type = PACKET_LOOPBACK; 143 144 skb->protocol = eth_type_trans(skb, dev); 145 146 if (likely(netif_rx(skb) == NET_RX_SUCCESS)) 147 vrf_rx_stats(dev, len); 148 else 149 this_cpu_inc(dev->dstats->rx_drps); 150 151 return NETDEV_TX_OK; 152 } 153 154 #if IS_ENABLED(CONFIG_IPV6) 155 static int vrf_ip6_local_out(struct net *net, struct sock *sk, 156 struct sk_buff *skb) 157 { 158 int err; 159 160 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, 161 sk, skb, NULL, skb_dst(skb)->dev, dst_output); 162 163 if (likely(err == 1)) 164 err = dst_output(net, sk, skb); 165 166 return err; 167 } 168 169 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 170 struct net_device *dev) 171 { 172 const struct ipv6hdr *iph = ipv6_hdr(skb); 173 struct net *net = dev_net(skb->dev); 174 struct flowi6 fl6 = { 175 /* needed to match OIF rule */ 176 .flowi6_oif = dev->ifindex, 177 .flowi6_iif = LOOPBACK_IFINDEX, 178 .daddr = iph->daddr, 179 .saddr = iph->saddr, 180 .flowlabel = ip6_flowinfo(iph), 181 .flowi6_mark = skb->mark, 182 .flowi6_proto = iph->nexthdr, 183 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF, 184 }; 185 int ret = NET_XMIT_DROP; 186 struct dst_entry *dst; 187 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 188 189 dst = ip6_route_output(net, NULL, &fl6); 190 if (dst == dst_null) 191 goto err; 192 193 skb_dst_drop(skb); 194 195 /* if dst.dev is loopback or the VRF device again this is locally 196 * originated traffic destined to a local address. Short circuit 197 * to Rx path using our local dst 198 */ 199 if (dst->dev == net->loopback_dev || dst->dev == dev) { 200 struct net_vrf *vrf = netdev_priv(dev); 201 struct rt6_info *rt6_local; 202 203 /* release looked up dst and use cached local dst */ 204 dst_release(dst); 205 206 rcu_read_lock(); 207 208 rt6_local = rcu_dereference(vrf->rt6_local); 209 if (unlikely(!rt6_local)) { 210 rcu_read_unlock(); 211 goto err; 212 } 213 214 /* Ordering issue: cached local dst is created on newlink 215 * before the IPv6 initialization. Using the local dst 216 * requires rt6i_idev to be set so make sure it is. 217 */ 218 if (unlikely(!rt6_local->rt6i_idev)) { 219 rt6_local->rt6i_idev = in6_dev_get(dev); 220 if (!rt6_local->rt6i_idev) { 221 rcu_read_unlock(); 222 goto err; 223 } 224 } 225 226 dst = &rt6_local->dst; 227 dst_hold(dst); 228 229 rcu_read_unlock(); 230 231 return vrf_local_xmit(skb, dev, &rt6_local->dst); 232 } 233 234 skb_dst_set(skb, dst); 235 236 /* strip the ethernet header added for pass through VRF device */ 237 __skb_pull(skb, skb_network_offset(skb)); 238 239 ret = vrf_ip6_local_out(net, skb->sk, skb); 240 if (unlikely(net_xmit_eval(ret))) 241 dev->stats.tx_errors++; 242 else 243 ret = NET_XMIT_SUCCESS; 244 245 return ret; 246 err: 247 vrf_tx_error(dev, skb); 248 return NET_XMIT_DROP; 249 } 250 #else 251 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 252 struct net_device *dev) 253 { 254 vrf_tx_error(dev, skb); 255 return NET_XMIT_DROP; 256 } 257 #endif 258 259 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 260 static int vrf_ip_local_out(struct net *net, struct sock *sk, 261 struct sk_buff *skb) 262 { 263 int err; 264 265 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 266 skb, NULL, skb_dst(skb)->dev, dst_output); 267 if (likely(err == 1)) 268 err = dst_output(net, sk, skb); 269 270 return err; 271 } 272 273 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 274 struct net_device *vrf_dev) 275 { 276 struct iphdr *ip4h = ip_hdr(skb); 277 int ret = NET_XMIT_DROP; 278 struct flowi4 fl4 = { 279 /* needed to match OIF rule */ 280 .flowi4_oif = vrf_dev->ifindex, 281 .flowi4_iif = LOOPBACK_IFINDEX, 282 .flowi4_tos = RT_TOS(ip4h->tos), 283 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF, 284 .flowi4_proto = ip4h->protocol, 285 .daddr = ip4h->daddr, 286 .saddr = ip4h->saddr, 287 }; 288 struct net *net = dev_net(vrf_dev); 289 struct rtable *rt; 290 291 rt = ip_route_output_flow(net, &fl4, NULL); 292 if (IS_ERR(rt)) 293 goto err; 294 295 skb_dst_drop(skb); 296 297 /* if dst.dev is loopback or the VRF device again this is locally 298 * originated traffic destined to a local address. Short circuit 299 * to Rx path using our local dst 300 */ 301 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) { 302 struct net_vrf *vrf = netdev_priv(vrf_dev); 303 struct rtable *rth_local; 304 struct dst_entry *dst = NULL; 305 306 ip_rt_put(rt); 307 308 rcu_read_lock(); 309 310 rth_local = rcu_dereference(vrf->rth_local); 311 if (likely(rth_local)) { 312 dst = &rth_local->dst; 313 dst_hold(dst); 314 } 315 316 rcu_read_unlock(); 317 318 if (unlikely(!dst)) 319 goto err; 320 321 return vrf_local_xmit(skb, vrf_dev, dst); 322 } 323 324 skb_dst_set(skb, &rt->dst); 325 326 /* strip the ethernet header added for pass through VRF device */ 327 __skb_pull(skb, skb_network_offset(skb)); 328 329 if (!ip4h->saddr) { 330 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 331 RT_SCOPE_LINK); 332 } 333 334 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 335 if (unlikely(net_xmit_eval(ret))) 336 vrf_dev->stats.tx_errors++; 337 else 338 ret = NET_XMIT_SUCCESS; 339 340 out: 341 return ret; 342 err: 343 vrf_tx_error(vrf_dev, skb); 344 goto out; 345 } 346 347 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 348 { 349 switch (skb->protocol) { 350 case htons(ETH_P_IP): 351 return vrf_process_v4_outbound(skb, dev); 352 case htons(ETH_P_IPV6): 353 return vrf_process_v6_outbound(skb, dev); 354 default: 355 vrf_tx_error(dev, skb); 356 return NET_XMIT_DROP; 357 } 358 } 359 360 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 361 { 362 int len = skb->len; 363 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 364 365 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 366 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 367 368 u64_stats_update_begin(&dstats->syncp); 369 dstats->tx_pkts++; 370 dstats->tx_bytes += len; 371 u64_stats_update_end(&dstats->syncp); 372 } else { 373 this_cpu_inc(dev->dstats->tx_drps); 374 } 375 376 return ret; 377 } 378 379 static int vrf_finish_direct(struct net *net, struct sock *sk, 380 struct sk_buff *skb) 381 { 382 struct net_device *vrf_dev = skb->dev; 383 384 if (!list_empty(&vrf_dev->ptype_all) && 385 likely(skb_headroom(skb) >= ETH_HLEN)) { 386 struct ethhdr *eth = skb_push(skb, ETH_HLEN); 387 388 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 389 eth_zero_addr(eth->h_dest); 390 eth->h_proto = skb->protocol; 391 392 rcu_read_lock_bh(); 393 dev_queue_xmit_nit(skb, vrf_dev); 394 rcu_read_unlock_bh(); 395 396 skb_pull(skb, ETH_HLEN); 397 } 398 399 return 1; 400 } 401 402 #if IS_ENABLED(CONFIG_IPV6) 403 /* modelled after ip6_finish_output2 */ 404 static int vrf_finish_output6(struct net *net, struct sock *sk, 405 struct sk_buff *skb) 406 { 407 struct dst_entry *dst = skb_dst(skb); 408 struct net_device *dev = dst->dev; 409 struct neighbour *neigh; 410 struct in6_addr *nexthop; 411 int ret; 412 413 nf_reset(skb); 414 415 skb->protocol = htons(ETH_P_IPV6); 416 skb->dev = dev; 417 418 rcu_read_lock_bh(); 419 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 420 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 421 if (unlikely(!neigh)) 422 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 423 if (!IS_ERR(neigh)) { 424 sock_confirm_neigh(skb, neigh); 425 ret = neigh_output(neigh, skb); 426 rcu_read_unlock_bh(); 427 return ret; 428 } 429 rcu_read_unlock_bh(); 430 431 IP6_INC_STATS(dev_net(dst->dev), 432 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 433 kfree_skb(skb); 434 return -EINVAL; 435 } 436 437 /* modelled after ip6_output */ 438 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 439 { 440 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 441 net, sk, skb, NULL, skb_dst(skb)->dev, 442 vrf_finish_output6, 443 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 444 } 445 446 /* set dst on skb to send packet to us via dev_xmit path. Allows 447 * packet to go through device based features such as qdisc, netfilter 448 * hooks and packet sockets with skb->dev set to vrf device. 449 */ 450 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev, 451 struct sk_buff *skb) 452 { 453 struct net_vrf *vrf = netdev_priv(vrf_dev); 454 struct dst_entry *dst = NULL; 455 struct rt6_info *rt6; 456 457 rcu_read_lock(); 458 459 rt6 = rcu_dereference(vrf->rt6); 460 if (likely(rt6)) { 461 dst = &rt6->dst; 462 dst_hold(dst); 463 } 464 465 rcu_read_unlock(); 466 467 if (unlikely(!dst)) { 468 vrf_tx_error(vrf_dev, skb); 469 return NULL; 470 } 471 472 skb_dst_drop(skb); 473 skb_dst_set(skb, dst); 474 475 return skb; 476 } 477 478 static int vrf_output6_direct(struct net *net, struct sock *sk, 479 struct sk_buff *skb) 480 { 481 skb->protocol = htons(ETH_P_IPV6); 482 483 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 484 net, sk, skb, NULL, skb->dev, 485 vrf_finish_direct, 486 !(IPCB(skb)->flags & IPSKB_REROUTED)); 487 } 488 489 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev, 490 struct sock *sk, 491 struct sk_buff *skb) 492 { 493 struct net *net = dev_net(vrf_dev); 494 int err; 495 496 skb->dev = vrf_dev; 497 498 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, 499 skb, NULL, vrf_dev, vrf_output6_direct); 500 501 if (likely(err == 1)) 502 err = vrf_output6_direct(net, sk, skb); 503 504 /* reset skb device */ 505 if (likely(err == 1)) 506 nf_reset(skb); 507 else 508 skb = NULL; 509 510 return skb; 511 } 512 513 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 514 struct sock *sk, 515 struct sk_buff *skb) 516 { 517 /* don't divert link scope packets */ 518 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 519 return skb; 520 521 if (qdisc_tx_is_default(vrf_dev)) 522 return vrf_ip6_out_direct(vrf_dev, sk, skb); 523 524 return vrf_ip6_out_redirect(vrf_dev, skb); 525 } 526 527 /* holding rtnl */ 528 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 529 { 530 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 531 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local); 532 struct net *net = dev_net(dev); 533 struct dst_entry *dst; 534 535 RCU_INIT_POINTER(vrf->rt6, NULL); 536 RCU_INIT_POINTER(vrf->rt6_local, NULL); 537 synchronize_rcu(); 538 539 /* move dev in dst's to loopback so this VRF device can be deleted 540 * - based on dst_ifdown 541 */ 542 if (rt6) { 543 dst = &rt6->dst; 544 dev_put(dst->dev); 545 dst->dev = net->loopback_dev; 546 dev_hold(dst->dev); 547 dst_release(dst); 548 } 549 550 if (rt6_local) { 551 if (rt6_local->rt6i_idev) { 552 in6_dev_put(rt6_local->rt6i_idev); 553 rt6_local->rt6i_idev = NULL; 554 } 555 556 dst = &rt6_local->dst; 557 dev_put(dst->dev); 558 dst->dev = net->loopback_dev; 559 dev_hold(dst->dev); 560 dst_release(dst); 561 } 562 } 563 564 static int vrf_rt6_create(struct net_device *dev) 565 { 566 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM; 567 struct net_vrf *vrf = netdev_priv(dev); 568 struct net *net = dev_net(dev); 569 struct fib6_table *rt6i_table; 570 struct rt6_info *rt6, *rt6_local; 571 int rc = -ENOMEM; 572 573 /* IPv6 can be CONFIG enabled and then disabled runtime */ 574 if (!ipv6_mod_enabled()) 575 return 0; 576 577 rt6i_table = fib6_new_table(net, vrf->tb_id); 578 if (!rt6i_table) 579 goto out; 580 581 /* create a dst for routing packets out a VRF device */ 582 rt6 = ip6_dst_alloc(net, dev, flags); 583 if (!rt6) 584 goto out; 585 586 rt6->rt6i_table = rt6i_table; 587 rt6->dst.output = vrf_output6; 588 589 /* create a dst for local routing - packets sent locally 590 * to local address via the VRF device as a loopback 591 */ 592 rt6_local = ip6_dst_alloc(net, dev, flags); 593 if (!rt6_local) { 594 dst_release(&rt6->dst); 595 goto out; 596 } 597 598 rt6_local->rt6i_idev = in6_dev_get(dev); 599 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL; 600 rt6_local->rt6i_table = rt6i_table; 601 rt6_local->dst.input = ip6_input; 602 603 rcu_assign_pointer(vrf->rt6, rt6); 604 rcu_assign_pointer(vrf->rt6_local, rt6_local); 605 606 rc = 0; 607 out: 608 return rc; 609 } 610 #else 611 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 612 struct sock *sk, 613 struct sk_buff *skb) 614 { 615 return skb; 616 } 617 618 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 619 { 620 } 621 622 static int vrf_rt6_create(struct net_device *dev) 623 { 624 return 0; 625 } 626 #endif 627 628 /* modelled after ip_finish_output2 */ 629 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 630 { 631 struct dst_entry *dst = skb_dst(skb); 632 struct rtable *rt = (struct rtable *)dst; 633 struct net_device *dev = dst->dev; 634 unsigned int hh_len = LL_RESERVED_SPACE(dev); 635 struct neighbour *neigh; 636 u32 nexthop; 637 int ret = -EINVAL; 638 639 nf_reset(skb); 640 641 /* Be paranoid, rather than too clever. */ 642 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 643 struct sk_buff *skb2; 644 645 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 646 if (!skb2) { 647 ret = -ENOMEM; 648 goto err; 649 } 650 if (skb->sk) 651 skb_set_owner_w(skb2, skb->sk); 652 653 consume_skb(skb); 654 skb = skb2; 655 } 656 657 rcu_read_lock_bh(); 658 659 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); 660 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 661 if (unlikely(!neigh)) 662 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 663 if (!IS_ERR(neigh)) { 664 sock_confirm_neigh(skb, neigh); 665 ret = neigh_output(neigh, skb); 666 } 667 668 rcu_read_unlock_bh(); 669 err: 670 if (unlikely(ret < 0)) 671 vrf_tx_error(skb->dev, skb); 672 return ret; 673 } 674 675 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 676 { 677 struct net_device *dev = skb_dst(skb)->dev; 678 679 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 680 681 skb->dev = dev; 682 skb->protocol = htons(ETH_P_IP); 683 684 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 685 net, sk, skb, NULL, dev, 686 vrf_finish_output, 687 !(IPCB(skb)->flags & IPSKB_REROUTED)); 688 } 689 690 /* set dst on skb to send packet to us via dev_xmit path. Allows 691 * packet to go through device based features such as qdisc, netfilter 692 * hooks and packet sockets with skb->dev set to vrf device. 693 */ 694 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev, 695 struct sk_buff *skb) 696 { 697 struct net_vrf *vrf = netdev_priv(vrf_dev); 698 struct dst_entry *dst = NULL; 699 struct rtable *rth; 700 701 rcu_read_lock(); 702 703 rth = rcu_dereference(vrf->rth); 704 if (likely(rth)) { 705 dst = &rth->dst; 706 dst_hold(dst); 707 } 708 709 rcu_read_unlock(); 710 711 if (unlikely(!dst)) { 712 vrf_tx_error(vrf_dev, skb); 713 return NULL; 714 } 715 716 skb_dst_drop(skb); 717 skb_dst_set(skb, dst); 718 719 return skb; 720 } 721 722 static int vrf_output_direct(struct net *net, struct sock *sk, 723 struct sk_buff *skb) 724 { 725 skb->protocol = htons(ETH_P_IP); 726 727 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 728 net, sk, skb, NULL, skb->dev, 729 vrf_finish_direct, 730 !(IPCB(skb)->flags & IPSKB_REROUTED)); 731 } 732 733 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev, 734 struct sock *sk, 735 struct sk_buff *skb) 736 { 737 struct net *net = dev_net(vrf_dev); 738 int err; 739 740 skb->dev = vrf_dev; 741 742 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 743 skb, NULL, vrf_dev, vrf_output_direct); 744 745 if (likely(err == 1)) 746 err = vrf_output_direct(net, sk, skb); 747 748 /* reset skb device */ 749 if (likely(err == 1)) 750 nf_reset(skb); 751 else 752 skb = NULL; 753 754 return skb; 755 } 756 757 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 758 struct sock *sk, 759 struct sk_buff *skb) 760 { 761 /* don't divert multicast */ 762 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 763 return skb; 764 765 if (qdisc_tx_is_default(vrf_dev)) 766 return vrf_ip_out_direct(vrf_dev, sk, skb); 767 768 return vrf_ip_out_redirect(vrf_dev, skb); 769 } 770 771 /* called with rcu lock held */ 772 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 773 struct sock *sk, 774 struct sk_buff *skb, 775 u16 proto) 776 { 777 switch (proto) { 778 case AF_INET: 779 return vrf_ip_out(vrf_dev, sk, skb); 780 case AF_INET6: 781 return vrf_ip6_out(vrf_dev, sk, skb); 782 } 783 784 return skb; 785 } 786 787 /* holding rtnl */ 788 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 789 { 790 struct rtable *rth = rtnl_dereference(vrf->rth); 791 struct rtable *rth_local = rtnl_dereference(vrf->rth_local); 792 struct net *net = dev_net(dev); 793 struct dst_entry *dst; 794 795 RCU_INIT_POINTER(vrf->rth, NULL); 796 RCU_INIT_POINTER(vrf->rth_local, NULL); 797 synchronize_rcu(); 798 799 /* move dev in dst's to loopback so this VRF device can be deleted 800 * - based on dst_ifdown 801 */ 802 if (rth) { 803 dst = &rth->dst; 804 dev_put(dst->dev); 805 dst->dev = net->loopback_dev; 806 dev_hold(dst->dev); 807 dst_release(dst); 808 } 809 810 if (rth_local) { 811 dst = &rth_local->dst; 812 dev_put(dst->dev); 813 dst->dev = net->loopback_dev; 814 dev_hold(dst->dev); 815 dst_release(dst); 816 } 817 } 818 819 static int vrf_rtable_create(struct net_device *dev) 820 { 821 struct net_vrf *vrf = netdev_priv(dev); 822 struct rtable *rth, *rth_local; 823 824 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 825 return -ENOMEM; 826 827 /* create a dst for routing packets out through a VRF device */ 828 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0); 829 if (!rth) 830 return -ENOMEM; 831 832 /* create a dst for local ingress routing - packets sent locally 833 * to local address via the VRF device as a loopback 834 */ 835 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0); 836 if (!rth_local) { 837 dst_release(&rth->dst); 838 return -ENOMEM; 839 } 840 841 rth->dst.output = vrf_output; 842 rth->rt_table_id = vrf->tb_id; 843 844 rth_local->rt_table_id = vrf->tb_id; 845 846 rcu_assign_pointer(vrf->rth, rth); 847 rcu_assign_pointer(vrf->rth_local, rth_local); 848 849 return 0; 850 } 851 852 /**************************** device handling ********************/ 853 854 /* cycle interface to flush neighbor cache and move routes across tables */ 855 static void cycle_netdev(struct net_device *dev) 856 { 857 unsigned int flags = dev->flags; 858 int ret; 859 860 if (!netif_running(dev)) 861 return; 862 863 ret = dev_change_flags(dev, flags & ~IFF_UP); 864 if (ret >= 0) 865 ret = dev_change_flags(dev, flags); 866 867 if (ret < 0) { 868 netdev_err(dev, 869 "Failed to cycle device %s; route tables might be wrong!\n", 870 dev->name); 871 } 872 } 873 874 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 875 { 876 int ret; 877 878 /* do not allow loopback device to be enslaved to a VRF. 879 * The vrf device acts as the loopback for the vrf. 880 */ 881 if (port_dev == dev_net(dev)->loopback_dev) 882 return -EOPNOTSUPP; 883 884 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 885 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL); 886 if (ret < 0) 887 goto err; 888 889 cycle_netdev(port_dev); 890 891 return 0; 892 893 err: 894 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 895 return ret; 896 } 897 898 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 899 { 900 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev)) 901 return -EINVAL; 902 903 return do_vrf_add_slave(dev, port_dev); 904 } 905 906 /* inverse of do_vrf_add_slave */ 907 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 908 { 909 netdev_upper_dev_unlink(port_dev, dev); 910 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 911 912 cycle_netdev(port_dev); 913 914 return 0; 915 } 916 917 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 918 { 919 return do_vrf_del_slave(dev, port_dev); 920 } 921 922 static void vrf_dev_uninit(struct net_device *dev) 923 { 924 struct net_vrf *vrf = netdev_priv(dev); 925 struct net_device *port_dev; 926 struct list_head *iter; 927 928 vrf_rtable_release(dev, vrf); 929 vrf_rt6_release(dev, vrf); 930 931 netdev_for_each_lower_dev(dev, port_dev, iter) 932 vrf_del_slave(dev, port_dev); 933 934 free_percpu(dev->dstats); 935 dev->dstats = NULL; 936 } 937 938 static int vrf_dev_init(struct net_device *dev) 939 { 940 struct net_vrf *vrf = netdev_priv(dev); 941 942 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 943 if (!dev->dstats) 944 goto out_nomem; 945 946 /* create the default dst which points back to us */ 947 if (vrf_rtable_create(dev) != 0) 948 goto out_stats; 949 950 if (vrf_rt6_create(dev) != 0) 951 goto out_rth; 952 953 dev->flags = IFF_MASTER | IFF_NOARP; 954 955 /* MTU is irrelevant for VRF device; set to 64k similar to lo */ 956 dev->mtu = 64 * 1024; 957 958 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 959 dev->operstate = IF_OPER_UP; 960 netdev_lockdep_set_classes(dev); 961 return 0; 962 963 out_rth: 964 vrf_rtable_release(dev, vrf); 965 out_stats: 966 free_percpu(dev->dstats); 967 dev->dstats = NULL; 968 out_nomem: 969 return -ENOMEM; 970 } 971 972 static const struct net_device_ops vrf_netdev_ops = { 973 .ndo_init = vrf_dev_init, 974 .ndo_uninit = vrf_dev_uninit, 975 .ndo_start_xmit = vrf_xmit, 976 .ndo_get_stats64 = vrf_get_stats64, 977 .ndo_add_slave = vrf_add_slave, 978 .ndo_del_slave = vrf_del_slave, 979 }; 980 981 static u32 vrf_fib_table(const struct net_device *dev) 982 { 983 struct net_vrf *vrf = netdev_priv(dev); 984 985 return vrf->tb_id; 986 } 987 988 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 989 { 990 kfree_skb(skb); 991 return 0; 992 } 993 994 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 995 struct sk_buff *skb, 996 struct net_device *dev) 997 { 998 struct net *net = dev_net(dev); 999 1000 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1) 1001 skb = NULL; /* kfree_skb(skb) handled by nf code */ 1002 1003 return skb; 1004 } 1005 1006 #if IS_ENABLED(CONFIG_IPV6) 1007 /* neighbor handling is done with actual device; do not want 1008 * to flip skb->dev for those ndisc packets. This really fails 1009 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 1010 * a start. 1011 */ 1012 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 1013 { 1014 const struct ipv6hdr *iph = ipv6_hdr(skb); 1015 bool rc = false; 1016 1017 if (iph->nexthdr == NEXTHDR_ICMP) { 1018 const struct icmp6hdr *icmph; 1019 struct icmp6hdr _icmph; 1020 1021 icmph = skb_header_pointer(skb, sizeof(*iph), 1022 sizeof(_icmph), &_icmph); 1023 if (!icmph) 1024 goto out; 1025 1026 switch (icmph->icmp6_type) { 1027 case NDISC_ROUTER_SOLICITATION: 1028 case NDISC_ROUTER_ADVERTISEMENT: 1029 case NDISC_NEIGHBOUR_SOLICITATION: 1030 case NDISC_NEIGHBOUR_ADVERTISEMENT: 1031 case NDISC_REDIRECT: 1032 rc = true; 1033 break; 1034 } 1035 } 1036 1037 out: 1038 return rc; 1039 } 1040 1041 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 1042 const struct net_device *dev, 1043 struct flowi6 *fl6, 1044 int ifindex, 1045 int flags) 1046 { 1047 struct net_vrf *vrf = netdev_priv(dev); 1048 struct fib6_table *table = NULL; 1049 struct rt6_info *rt6; 1050 1051 rcu_read_lock(); 1052 1053 /* fib6_table does not have a refcnt and can not be freed */ 1054 rt6 = rcu_dereference(vrf->rt6); 1055 if (likely(rt6)) 1056 table = rt6->rt6i_table; 1057 1058 rcu_read_unlock(); 1059 1060 if (!table) 1061 return NULL; 1062 1063 return ip6_pol_route(net, table, ifindex, fl6, flags); 1064 } 1065 1066 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 1067 int ifindex) 1068 { 1069 const struct ipv6hdr *iph = ipv6_hdr(skb); 1070 struct flowi6 fl6 = { 1071 .daddr = iph->daddr, 1072 .saddr = iph->saddr, 1073 .flowlabel = ip6_flowinfo(iph), 1074 .flowi6_mark = skb->mark, 1075 .flowi6_proto = iph->nexthdr, 1076 .flowi6_iif = ifindex, 1077 }; 1078 struct net *net = dev_net(vrf_dev); 1079 struct rt6_info *rt6; 1080 1081 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, 1082 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 1083 if (unlikely(!rt6)) 1084 return; 1085 1086 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 1087 return; 1088 1089 skb_dst_set(skb, &rt6->dst); 1090 } 1091 1092 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1093 struct sk_buff *skb) 1094 { 1095 int orig_iif = skb->skb_iif; 1096 bool need_strict; 1097 1098 /* loopback traffic; do not push through packet taps again. 1099 * Reset pkt_type for upper layers to process skb 1100 */ 1101 if (skb->pkt_type == PACKET_LOOPBACK) { 1102 skb->dev = vrf_dev; 1103 skb->skb_iif = vrf_dev->ifindex; 1104 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1105 skb->pkt_type = PACKET_HOST; 1106 goto out; 1107 } 1108 1109 /* if packet is NDISC or addressed to multicast or link-local 1110 * then keep the ingress interface 1111 */ 1112 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 1113 if (!ipv6_ndisc_frame(skb) && !need_strict) { 1114 vrf_rx_stats(vrf_dev, skb->len); 1115 skb->dev = vrf_dev; 1116 skb->skb_iif = vrf_dev->ifindex; 1117 1118 if (!list_empty(&vrf_dev->ptype_all)) { 1119 skb_push(skb, skb->mac_len); 1120 dev_queue_xmit_nit(skb, vrf_dev); 1121 skb_pull(skb, skb->mac_len); 1122 } 1123 1124 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1125 } 1126 1127 if (need_strict) 1128 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1129 1130 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 1131 out: 1132 return skb; 1133 } 1134 1135 #else 1136 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1137 struct sk_buff *skb) 1138 { 1139 return skb; 1140 } 1141 #endif 1142 1143 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 1144 struct sk_buff *skb) 1145 { 1146 skb->dev = vrf_dev; 1147 skb->skb_iif = vrf_dev->ifindex; 1148 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1149 1150 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1151 goto out; 1152 1153 /* loopback traffic; do not push through packet taps again. 1154 * Reset pkt_type for upper layers to process skb 1155 */ 1156 if (skb->pkt_type == PACKET_LOOPBACK) { 1157 skb->pkt_type = PACKET_HOST; 1158 goto out; 1159 } 1160 1161 vrf_rx_stats(vrf_dev, skb->len); 1162 1163 if (!list_empty(&vrf_dev->ptype_all)) { 1164 skb_push(skb, skb->mac_len); 1165 dev_queue_xmit_nit(skb, vrf_dev); 1166 skb_pull(skb, skb->mac_len); 1167 } 1168 1169 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1170 out: 1171 return skb; 1172 } 1173 1174 /* called with rcu lock held */ 1175 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1176 struct sk_buff *skb, 1177 u16 proto) 1178 { 1179 switch (proto) { 1180 case AF_INET: 1181 return vrf_ip_rcv(vrf_dev, skb); 1182 case AF_INET6: 1183 return vrf_ip6_rcv(vrf_dev, skb); 1184 } 1185 1186 return skb; 1187 } 1188 1189 #if IS_ENABLED(CONFIG_IPV6) 1190 /* send to link-local or multicast address via interface enslaved to 1191 * VRF device. Force lookup to VRF table without changing flow struct 1192 */ 1193 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1194 struct flowi6 *fl6) 1195 { 1196 struct net *net = dev_net(dev); 1197 int flags = RT6_LOOKUP_F_IFACE; 1198 struct dst_entry *dst = NULL; 1199 struct rt6_info *rt; 1200 1201 /* VRF device does not have a link-local address and 1202 * sending packets to link-local or mcast addresses over 1203 * a VRF device does not make sense 1204 */ 1205 if (fl6->flowi6_oif == dev->ifindex) { 1206 dst = &net->ipv6.ip6_null_entry->dst; 1207 dst_hold(dst); 1208 return dst; 1209 } 1210 1211 if (!ipv6_addr_any(&fl6->saddr)) 1212 flags |= RT6_LOOKUP_F_HAS_SADDR; 1213 1214 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags); 1215 if (rt) 1216 dst = &rt->dst; 1217 1218 return dst; 1219 } 1220 #endif 1221 1222 static const struct l3mdev_ops vrf_l3mdev_ops = { 1223 .l3mdev_fib_table = vrf_fib_table, 1224 .l3mdev_l3_rcv = vrf_l3_rcv, 1225 .l3mdev_l3_out = vrf_l3_out, 1226 #if IS_ENABLED(CONFIG_IPV6) 1227 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1228 #endif 1229 }; 1230 1231 static void vrf_get_drvinfo(struct net_device *dev, 1232 struct ethtool_drvinfo *info) 1233 { 1234 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 1235 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 1236 } 1237 1238 static const struct ethtool_ops vrf_ethtool_ops = { 1239 .get_drvinfo = vrf_get_drvinfo, 1240 }; 1241 1242 static inline size_t vrf_fib_rule_nl_size(void) 1243 { 1244 size_t sz; 1245 1246 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1247 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1248 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1249 1250 return sz; 1251 } 1252 1253 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1254 { 1255 struct fib_rule_hdr *frh; 1256 struct nlmsghdr *nlh; 1257 struct sk_buff *skb; 1258 int err; 1259 1260 if (family == AF_INET6 && !ipv6_mod_enabled()) 1261 return 0; 1262 1263 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1264 if (!skb) 1265 return -ENOMEM; 1266 1267 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1268 if (!nlh) 1269 goto nla_put_failure; 1270 1271 /* rule only needs to appear once */ 1272 nlh->nlmsg_flags |= NLM_F_EXCL; 1273 1274 frh = nlmsg_data(nlh); 1275 memset(frh, 0, sizeof(*frh)); 1276 frh->family = family; 1277 frh->action = FR_ACT_TO_TBL; 1278 1279 if (nla_put_u32(skb, FRA_L3MDEV, 1)) 1280 goto nla_put_failure; 1281 1282 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1283 goto nla_put_failure; 1284 1285 nlmsg_end(skb, nlh); 1286 1287 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1288 skb->sk = dev_net(dev)->rtnl; 1289 if (add_it) { 1290 err = fib_nl_newrule(skb, nlh, NULL); 1291 if (err == -EEXIST) 1292 err = 0; 1293 } else { 1294 err = fib_nl_delrule(skb, nlh, NULL); 1295 if (err == -ENOENT) 1296 err = 0; 1297 } 1298 nlmsg_free(skb); 1299 1300 return err; 1301 1302 nla_put_failure: 1303 nlmsg_free(skb); 1304 1305 return -EMSGSIZE; 1306 } 1307 1308 static int vrf_add_fib_rules(const struct net_device *dev) 1309 { 1310 int err; 1311 1312 err = vrf_fib_rule(dev, AF_INET, true); 1313 if (err < 0) 1314 goto out_err; 1315 1316 err = vrf_fib_rule(dev, AF_INET6, true); 1317 if (err < 0) 1318 goto ipv6_err; 1319 1320 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1321 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1322 if (err < 0) 1323 goto ipmr_err; 1324 #endif 1325 1326 return 0; 1327 1328 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1329 ipmr_err: 1330 vrf_fib_rule(dev, AF_INET6, false); 1331 #endif 1332 1333 ipv6_err: 1334 vrf_fib_rule(dev, AF_INET, false); 1335 1336 out_err: 1337 netdev_err(dev, "Failed to add FIB rules.\n"); 1338 return err; 1339 } 1340 1341 static void vrf_setup(struct net_device *dev) 1342 { 1343 ether_setup(dev); 1344 1345 /* Initialize the device structure. */ 1346 dev->netdev_ops = &vrf_netdev_ops; 1347 dev->l3mdev_ops = &vrf_l3mdev_ops; 1348 dev->ethtool_ops = &vrf_ethtool_ops; 1349 dev->needs_free_netdev = true; 1350 1351 /* Fill in device structure with ethernet-generic values. */ 1352 eth_hw_addr_random(dev); 1353 1354 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1355 dev->features |= NETIF_F_LLTX; 1356 1357 /* don't allow vrf devices to change network namespaces. */ 1358 dev->features |= NETIF_F_NETNS_LOCAL; 1359 1360 /* does not make sense for a VLAN to be added to a vrf device */ 1361 dev->features |= NETIF_F_VLAN_CHALLENGED; 1362 1363 /* enable offload features */ 1364 dev->features |= NETIF_F_GSO_SOFTWARE; 1365 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM; 1366 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1367 1368 dev->hw_features = dev->features; 1369 dev->hw_enc_features = dev->features; 1370 1371 /* default to no qdisc; user can add if desired */ 1372 dev->priv_flags |= IFF_NO_QUEUE; 1373 } 1374 1375 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[], 1376 struct netlink_ext_ack *extack) 1377 { 1378 if (tb[IFLA_ADDRESS]) { 1379 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1380 return -EINVAL; 1381 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1382 return -EADDRNOTAVAIL; 1383 } 1384 return 0; 1385 } 1386 1387 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1388 { 1389 unregister_netdevice_queue(dev, head); 1390 } 1391 1392 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1393 struct nlattr *tb[], struct nlattr *data[], 1394 struct netlink_ext_ack *extack) 1395 { 1396 struct net_vrf *vrf = netdev_priv(dev); 1397 bool *add_fib_rules; 1398 struct net *net; 1399 int err; 1400 1401 if (!data || !data[IFLA_VRF_TABLE]) 1402 return -EINVAL; 1403 1404 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1405 if (vrf->tb_id == RT_TABLE_UNSPEC) 1406 return -EINVAL; 1407 1408 dev->priv_flags |= IFF_L3MDEV_MASTER; 1409 1410 err = register_netdevice(dev); 1411 if (err) 1412 goto out; 1413 1414 net = dev_net(dev); 1415 add_fib_rules = net_generic(net, vrf_net_id); 1416 if (*add_fib_rules) { 1417 err = vrf_add_fib_rules(dev); 1418 if (err) { 1419 unregister_netdevice(dev); 1420 goto out; 1421 } 1422 *add_fib_rules = false; 1423 } 1424 1425 out: 1426 return err; 1427 } 1428 1429 static size_t vrf_nl_getsize(const struct net_device *dev) 1430 { 1431 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1432 } 1433 1434 static int vrf_fillinfo(struct sk_buff *skb, 1435 const struct net_device *dev) 1436 { 1437 struct net_vrf *vrf = netdev_priv(dev); 1438 1439 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1440 } 1441 1442 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1443 const struct net_device *slave_dev) 1444 { 1445 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1446 } 1447 1448 static int vrf_fill_slave_info(struct sk_buff *skb, 1449 const struct net_device *vrf_dev, 1450 const struct net_device *slave_dev) 1451 { 1452 struct net_vrf *vrf = netdev_priv(vrf_dev); 1453 1454 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1455 return -EMSGSIZE; 1456 1457 return 0; 1458 } 1459 1460 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1461 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1462 }; 1463 1464 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1465 .kind = DRV_NAME, 1466 .priv_size = sizeof(struct net_vrf), 1467 1468 .get_size = vrf_nl_getsize, 1469 .policy = vrf_nl_policy, 1470 .validate = vrf_validate, 1471 .fill_info = vrf_fillinfo, 1472 1473 .get_slave_size = vrf_get_slave_size, 1474 .fill_slave_info = vrf_fill_slave_info, 1475 1476 .newlink = vrf_newlink, 1477 .dellink = vrf_dellink, 1478 .setup = vrf_setup, 1479 .maxtype = IFLA_VRF_MAX, 1480 }; 1481 1482 static int vrf_device_event(struct notifier_block *unused, 1483 unsigned long event, void *ptr) 1484 { 1485 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1486 1487 /* only care about unregister events to drop slave references */ 1488 if (event == NETDEV_UNREGISTER) { 1489 struct net_device *vrf_dev; 1490 1491 if (!netif_is_l3_slave(dev)) 1492 goto out; 1493 1494 vrf_dev = netdev_master_upper_dev_get(dev); 1495 vrf_del_slave(vrf_dev, dev); 1496 } 1497 out: 1498 return NOTIFY_DONE; 1499 } 1500 1501 static struct notifier_block vrf_notifier_block __read_mostly = { 1502 .notifier_call = vrf_device_event, 1503 }; 1504 1505 /* Initialize per network namespace state */ 1506 static int __net_init vrf_netns_init(struct net *net) 1507 { 1508 bool *add_fib_rules = net_generic(net, vrf_net_id); 1509 1510 *add_fib_rules = true; 1511 1512 return 0; 1513 } 1514 1515 static struct pernet_operations vrf_net_ops __net_initdata = { 1516 .init = vrf_netns_init, 1517 .id = &vrf_net_id, 1518 .size = sizeof(bool), 1519 }; 1520 1521 static int __init vrf_init_module(void) 1522 { 1523 int rc; 1524 1525 register_netdevice_notifier(&vrf_notifier_block); 1526 1527 rc = register_pernet_subsys(&vrf_net_ops); 1528 if (rc < 0) 1529 goto error; 1530 1531 rc = rtnl_link_register(&vrf_link_ops); 1532 if (rc < 0) { 1533 unregister_pernet_subsys(&vrf_net_ops); 1534 goto error; 1535 } 1536 1537 return 0; 1538 1539 error: 1540 unregister_netdevice_notifier(&vrf_notifier_block); 1541 return rc; 1542 } 1543 1544 module_init(vrf_init_module); 1545 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 1546 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 1547 MODULE_LICENSE("GPL"); 1548 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 1549 MODULE_VERSION(DRV_VERSION); 1550