1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * vrf.c: device driver to encapsulate a VRF space 4 * 5 * Copyright (c) 2015 Cumulus Networks. All rights reserved. 6 * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com> 7 * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com> 8 * 9 * Based on dummy, team and ipvlan drivers 10 */ 11 12 #include <linux/ethtool.h> 13 #include <linux/module.h> 14 #include <linux/kernel.h> 15 #include <linux/netdevice.h> 16 #include <linux/etherdevice.h> 17 #include <linux/ip.h> 18 #include <linux/init.h> 19 #include <linux/moduleparam.h> 20 #include <linux/netfilter.h> 21 #include <linux/rtnetlink.h> 22 #include <net/rtnetlink.h> 23 #include <linux/u64_stats_sync.h> 24 #include <linux/hashtable.h> 25 #include <linux/spinlock_types.h> 26 27 #include <linux/inetdevice.h> 28 #include <net/arp.h> 29 #include <net/ip.h> 30 #include <net/ip_fib.h> 31 #include <net/ip6_fib.h> 32 #include <net/ip6_route.h> 33 #include <net/route.h> 34 #include <net/addrconf.h> 35 #include <net/l3mdev.h> 36 #include <net/fib_rules.h> 37 #include <net/netns/generic.h> 38 39 #define DRV_NAME "vrf" 40 #define DRV_VERSION "1.1" 41 42 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */ 43 44 #define HT_MAP_BITS 4 45 #define HASH_INITVAL ((u32)0xcafef00d) 46 47 struct vrf_map { 48 DECLARE_HASHTABLE(ht, HT_MAP_BITS); 49 spinlock_t vmap_lock; 50 51 /* shared_tables: 52 * count how many distinct tables do not comply with the strict mode 53 * requirement. 54 * shared_tables value must be 0 in order to enable the strict mode. 55 * 56 * example of the evolution of shared_tables: 57 * | time 58 * add vrf0 --> table 100 shared_tables = 0 | t0 59 * add vrf1 --> table 101 shared_tables = 0 | t1 60 * add vrf2 --> table 100 shared_tables = 1 | t2 61 * add vrf3 --> table 100 shared_tables = 1 | t3 62 * add vrf4 --> table 101 shared_tables = 2 v t4 63 * 64 * shared_tables is a "step function" (or "staircase function") 65 * and it is increased by one when the second vrf is associated to a 66 * table. 67 * 68 * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1. 69 * 70 * at t3, another dev (vrf3) is bound to the same table 100 but the 71 * value of shared_tables is still 1. 72 * This means that no matter how many new vrfs will register on the 73 * table 100, the shared_tables will not increase (considering only 74 * table 100). 75 * 76 * at t4, vrf4 is bound to table 101, and shared_tables = 2. 77 * 78 * Looking at the value of shared_tables we can immediately know if 79 * the strict_mode can or cannot be enforced. Indeed, strict_mode 80 * can be enforced iff shared_tables = 0. 81 * 82 * Conversely, shared_tables is decreased when a vrf is de-associated 83 * from a table with exactly two associated vrfs. 84 */ 85 u32 shared_tables; 86 87 bool strict_mode; 88 }; 89 90 struct vrf_map_elem { 91 struct hlist_node hnode; 92 struct list_head vrf_list; /* VRFs registered to this table */ 93 94 u32 table_id; 95 int users; 96 int ifindex; 97 }; 98 99 static unsigned int vrf_net_id; 100 101 /* per netns vrf data */ 102 struct netns_vrf { 103 /* protected by rtnl lock */ 104 bool add_fib_rules; 105 106 struct vrf_map vmap; 107 struct ctl_table_header *ctl_hdr; 108 }; 109 110 struct net_vrf { 111 struct rtable __rcu *rth; 112 struct rt6_info __rcu *rt6; 113 #if IS_ENABLED(CONFIG_IPV6) 114 struct fib6_table *fib6_table; 115 #endif 116 u32 tb_id; 117 118 struct list_head me_list; /* entry in vrf_map_elem */ 119 int ifindex; 120 }; 121 122 struct pcpu_dstats { 123 u64 tx_pkts; 124 u64 tx_bytes; 125 u64 tx_drps; 126 u64 rx_pkts; 127 u64 rx_bytes; 128 u64 rx_drps; 129 struct u64_stats_sync syncp; 130 }; 131 132 static void vrf_rx_stats(struct net_device *dev, int len) 133 { 134 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 135 136 u64_stats_update_begin(&dstats->syncp); 137 dstats->rx_pkts++; 138 dstats->rx_bytes += len; 139 u64_stats_update_end(&dstats->syncp); 140 } 141 142 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 143 { 144 vrf_dev->stats.tx_errors++; 145 kfree_skb(skb); 146 } 147 148 static void vrf_get_stats64(struct net_device *dev, 149 struct rtnl_link_stats64 *stats) 150 { 151 int i; 152 153 for_each_possible_cpu(i) { 154 const struct pcpu_dstats *dstats; 155 u64 tbytes, tpkts, tdrops, rbytes, rpkts; 156 unsigned int start; 157 158 dstats = per_cpu_ptr(dev->dstats, i); 159 do { 160 start = u64_stats_fetch_begin_irq(&dstats->syncp); 161 tbytes = dstats->tx_bytes; 162 tpkts = dstats->tx_pkts; 163 tdrops = dstats->tx_drps; 164 rbytes = dstats->rx_bytes; 165 rpkts = dstats->rx_pkts; 166 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start)); 167 stats->tx_bytes += tbytes; 168 stats->tx_packets += tpkts; 169 stats->tx_dropped += tdrops; 170 stats->rx_bytes += rbytes; 171 stats->rx_packets += rpkts; 172 } 173 } 174 175 static struct vrf_map *netns_vrf_map(struct net *net) 176 { 177 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 178 179 return &nn_vrf->vmap; 180 } 181 182 static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev) 183 { 184 return netns_vrf_map(dev_net(dev)); 185 } 186 187 static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me) 188 { 189 struct list_head *me_head = &me->vrf_list; 190 struct net_vrf *vrf; 191 192 if (list_empty(me_head)) 193 return -ENODEV; 194 195 vrf = list_first_entry(me_head, struct net_vrf, me_list); 196 197 return vrf->ifindex; 198 } 199 200 static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags) 201 { 202 struct vrf_map_elem *me; 203 204 me = kmalloc(sizeof(*me), flags); 205 if (!me) 206 return NULL; 207 208 return me; 209 } 210 211 static void vrf_map_elem_free(struct vrf_map_elem *me) 212 { 213 kfree(me); 214 } 215 216 static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id, 217 int ifindex, int users) 218 { 219 me->table_id = table_id; 220 me->ifindex = ifindex; 221 me->users = users; 222 INIT_LIST_HEAD(&me->vrf_list); 223 } 224 225 static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap, 226 u32 table_id) 227 { 228 struct vrf_map_elem *me; 229 u32 key; 230 231 key = jhash_1word(table_id, HASH_INITVAL); 232 hash_for_each_possible(vmap->ht, me, hnode, key) { 233 if (me->table_id == table_id) 234 return me; 235 } 236 237 return NULL; 238 } 239 240 static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me) 241 { 242 u32 table_id = me->table_id; 243 u32 key; 244 245 key = jhash_1word(table_id, HASH_INITVAL); 246 hash_add(vmap->ht, &me->hnode, key); 247 } 248 249 static void vrf_map_del_elem(struct vrf_map_elem *me) 250 { 251 hash_del(&me->hnode); 252 } 253 254 static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock) 255 { 256 spin_lock(&vmap->vmap_lock); 257 } 258 259 static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock) 260 { 261 spin_unlock(&vmap->vmap_lock); 262 } 263 264 /* called with rtnl lock held */ 265 static int 266 vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack) 267 { 268 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 269 struct net_vrf *vrf = netdev_priv(dev); 270 struct vrf_map_elem *new_me, *me; 271 u32 table_id = vrf->tb_id; 272 bool free_new_me = false; 273 int users; 274 int res; 275 276 /* we pre-allocate elements used in the spin-locked section (so that we 277 * keep the spinlock as short as possible). 278 */ 279 new_me = vrf_map_elem_alloc(GFP_KERNEL); 280 if (!new_me) 281 return -ENOMEM; 282 283 vrf_map_elem_init(new_me, table_id, dev->ifindex, 0); 284 285 vrf_map_lock(vmap); 286 287 me = vrf_map_lookup_elem(vmap, table_id); 288 if (!me) { 289 me = new_me; 290 vrf_map_add_elem(vmap, me); 291 goto link_vrf; 292 } 293 294 /* we already have an entry in the vrf_map, so it means there is (at 295 * least) a vrf registered on the specific table. 296 */ 297 free_new_me = true; 298 if (vmap->strict_mode) { 299 /* vrfs cannot share the same table */ 300 NL_SET_ERR_MSG(extack, "Table is used by another VRF"); 301 res = -EBUSY; 302 goto unlock; 303 } 304 305 link_vrf: 306 users = ++me->users; 307 if (users == 2) 308 ++vmap->shared_tables; 309 310 list_add(&vrf->me_list, &me->vrf_list); 311 312 res = 0; 313 314 unlock: 315 vrf_map_unlock(vmap); 316 317 /* clean-up, if needed */ 318 if (free_new_me) 319 vrf_map_elem_free(new_me); 320 321 return res; 322 } 323 324 /* called with rtnl lock held */ 325 static void vrf_map_unregister_dev(struct net_device *dev) 326 { 327 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 328 struct net_vrf *vrf = netdev_priv(dev); 329 u32 table_id = vrf->tb_id; 330 struct vrf_map_elem *me; 331 int users; 332 333 vrf_map_lock(vmap); 334 335 me = vrf_map_lookup_elem(vmap, table_id); 336 if (!me) 337 goto unlock; 338 339 list_del(&vrf->me_list); 340 341 users = --me->users; 342 if (users == 1) { 343 --vmap->shared_tables; 344 } else if (users == 0) { 345 vrf_map_del_elem(me); 346 347 /* no one will refer to this element anymore */ 348 vrf_map_elem_free(me); 349 } 350 351 unlock: 352 vrf_map_unlock(vmap); 353 } 354 355 /* return the vrf device index associated with the table_id */ 356 static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id) 357 { 358 struct vrf_map *vmap = netns_vrf_map(net); 359 struct vrf_map_elem *me; 360 int ifindex; 361 362 vrf_map_lock(vmap); 363 364 if (!vmap->strict_mode) { 365 ifindex = -EPERM; 366 goto unlock; 367 } 368 369 me = vrf_map_lookup_elem(vmap, table_id); 370 if (!me) { 371 ifindex = -ENODEV; 372 goto unlock; 373 } 374 375 ifindex = vrf_map_elem_get_vrf_ifindex(me); 376 377 unlock: 378 vrf_map_unlock(vmap); 379 380 return ifindex; 381 } 382 383 /* by default VRF devices do not have a qdisc and are expected 384 * to be created with only a single queue. 385 */ 386 static bool qdisc_tx_is_default(const struct net_device *dev) 387 { 388 struct netdev_queue *txq; 389 struct Qdisc *qdisc; 390 391 if (dev->num_tx_queues > 1) 392 return false; 393 394 txq = netdev_get_tx_queue(dev, 0); 395 qdisc = rcu_access_pointer(txq->qdisc); 396 397 return !qdisc->enqueue; 398 } 399 400 /* Local traffic destined to local address. Reinsert the packet to rx 401 * path, similar to loopback handling. 402 */ 403 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 404 struct dst_entry *dst) 405 { 406 int len = skb->len; 407 408 skb_orphan(skb); 409 410 skb_dst_set(skb, dst); 411 412 /* set pkt_type to avoid skb hitting packet taps twice - 413 * once on Tx and again in Rx processing 414 */ 415 skb->pkt_type = PACKET_LOOPBACK; 416 417 skb->protocol = eth_type_trans(skb, dev); 418 419 if (likely(netif_rx(skb) == NET_RX_SUCCESS)) 420 vrf_rx_stats(dev, len); 421 else 422 this_cpu_inc(dev->dstats->rx_drps); 423 424 return NETDEV_TX_OK; 425 } 426 427 #if IS_ENABLED(CONFIG_IPV6) 428 static int vrf_ip6_local_out(struct net *net, struct sock *sk, 429 struct sk_buff *skb) 430 { 431 int err; 432 433 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, 434 sk, skb, NULL, skb_dst(skb)->dev, dst_output); 435 436 if (likely(err == 1)) 437 err = dst_output(net, sk, skb); 438 439 return err; 440 } 441 442 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 443 struct net_device *dev) 444 { 445 const struct ipv6hdr *iph; 446 struct net *net = dev_net(skb->dev); 447 struct flowi6 fl6; 448 int ret = NET_XMIT_DROP; 449 struct dst_entry *dst; 450 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 451 452 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr))) 453 goto err; 454 455 iph = ipv6_hdr(skb); 456 457 memset(&fl6, 0, sizeof(fl6)); 458 /* needed to match OIF rule */ 459 fl6.flowi6_oif = dev->ifindex; 460 fl6.flowi6_iif = LOOPBACK_IFINDEX; 461 fl6.daddr = iph->daddr; 462 fl6.saddr = iph->saddr; 463 fl6.flowlabel = ip6_flowinfo(iph); 464 fl6.flowi6_mark = skb->mark; 465 fl6.flowi6_proto = iph->nexthdr; 466 fl6.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF; 467 468 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL); 469 if (IS_ERR(dst) || dst == dst_null) 470 goto err; 471 472 skb_dst_drop(skb); 473 474 /* if dst.dev is the VRF device again this is locally originated traffic 475 * destined to a local address. Short circuit to Rx path. 476 */ 477 if (dst->dev == dev) 478 return vrf_local_xmit(skb, dev, dst); 479 480 skb_dst_set(skb, dst); 481 482 /* strip the ethernet header added for pass through VRF device */ 483 __skb_pull(skb, skb_network_offset(skb)); 484 485 ret = vrf_ip6_local_out(net, skb->sk, skb); 486 if (unlikely(net_xmit_eval(ret))) 487 dev->stats.tx_errors++; 488 else 489 ret = NET_XMIT_SUCCESS; 490 491 return ret; 492 err: 493 vrf_tx_error(dev, skb); 494 return NET_XMIT_DROP; 495 } 496 #else 497 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 498 struct net_device *dev) 499 { 500 vrf_tx_error(dev, skb); 501 return NET_XMIT_DROP; 502 } 503 #endif 504 505 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 506 static int vrf_ip_local_out(struct net *net, struct sock *sk, 507 struct sk_buff *skb) 508 { 509 int err; 510 511 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 512 skb, NULL, skb_dst(skb)->dev, dst_output); 513 if (likely(err == 1)) 514 err = dst_output(net, sk, skb); 515 516 return err; 517 } 518 519 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 520 struct net_device *vrf_dev) 521 { 522 struct iphdr *ip4h; 523 int ret = NET_XMIT_DROP; 524 struct flowi4 fl4; 525 struct net *net = dev_net(vrf_dev); 526 struct rtable *rt; 527 528 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr))) 529 goto err; 530 531 ip4h = ip_hdr(skb); 532 533 memset(&fl4, 0, sizeof(fl4)); 534 /* needed to match OIF rule */ 535 fl4.flowi4_oif = vrf_dev->ifindex; 536 fl4.flowi4_iif = LOOPBACK_IFINDEX; 537 fl4.flowi4_tos = RT_TOS(ip4h->tos); 538 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF; 539 fl4.flowi4_proto = ip4h->protocol; 540 fl4.daddr = ip4h->daddr; 541 fl4.saddr = ip4h->saddr; 542 543 rt = ip_route_output_flow(net, &fl4, NULL); 544 if (IS_ERR(rt)) 545 goto err; 546 547 skb_dst_drop(skb); 548 549 /* if dst.dev is the VRF device again this is locally originated traffic 550 * destined to a local address. Short circuit to Rx path. 551 */ 552 if (rt->dst.dev == vrf_dev) 553 return vrf_local_xmit(skb, vrf_dev, &rt->dst); 554 555 skb_dst_set(skb, &rt->dst); 556 557 /* strip the ethernet header added for pass through VRF device */ 558 __skb_pull(skb, skb_network_offset(skb)); 559 560 if (!ip4h->saddr) { 561 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 562 RT_SCOPE_LINK); 563 } 564 565 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 566 if (unlikely(net_xmit_eval(ret))) 567 vrf_dev->stats.tx_errors++; 568 else 569 ret = NET_XMIT_SUCCESS; 570 571 out: 572 return ret; 573 err: 574 vrf_tx_error(vrf_dev, skb); 575 goto out; 576 } 577 578 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 579 { 580 switch (skb->protocol) { 581 case htons(ETH_P_IP): 582 return vrf_process_v4_outbound(skb, dev); 583 case htons(ETH_P_IPV6): 584 return vrf_process_v6_outbound(skb, dev); 585 default: 586 vrf_tx_error(dev, skb); 587 return NET_XMIT_DROP; 588 } 589 } 590 591 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 592 { 593 int len = skb->len; 594 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 595 596 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 597 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 598 599 u64_stats_update_begin(&dstats->syncp); 600 dstats->tx_pkts++; 601 dstats->tx_bytes += len; 602 u64_stats_update_end(&dstats->syncp); 603 } else { 604 this_cpu_inc(dev->dstats->tx_drps); 605 } 606 607 return ret; 608 } 609 610 static void vrf_finish_direct(struct sk_buff *skb) 611 { 612 struct net_device *vrf_dev = skb->dev; 613 614 if (!list_empty(&vrf_dev->ptype_all) && 615 likely(skb_headroom(skb) >= ETH_HLEN)) { 616 struct ethhdr *eth = skb_push(skb, ETH_HLEN); 617 618 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 619 eth_zero_addr(eth->h_dest); 620 eth->h_proto = skb->protocol; 621 622 rcu_read_lock_bh(); 623 dev_queue_xmit_nit(skb, vrf_dev); 624 rcu_read_unlock_bh(); 625 626 skb_pull(skb, ETH_HLEN); 627 } 628 629 /* reset skb device */ 630 nf_reset_ct(skb); 631 } 632 633 #if IS_ENABLED(CONFIG_IPV6) 634 /* modelled after ip6_finish_output2 */ 635 static int vrf_finish_output6(struct net *net, struct sock *sk, 636 struct sk_buff *skb) 637 { 638 struct dst_entry *dst = skb_dst(skb); 639 struct net_device *dev = dst->dev; 640 const struct in6_addr *nexthop; 641 struct neighbour *neigh; 642 int ret; 643 644 nf_reset_ct(skb); 645 646 skb->protocol = htons(ETH_P_IPV6); 647 skb->dev = dev; 648 649 rcu_read_lock_bh(); 650 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 651 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 652 if (unlikely(!neigh)) 653 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 654 if (!IS_ERR(neigh)) { 655 sock_confirm_neigh(skb, neigh); 656 ret = neigh_output(neigh, skb, false); 657 rcu_read_unlock_bh(); 658 return ret; 659 } 660 rcu_read_unlock_bh(); 661 662 IP6_INC_STATS(dev_net(dst->dev), 663 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 664 kfree_skb(skb); 665 return -EINVAL; 666 } 667 668 /* modelled after ip6_output */ 669 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 670 { 671 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 672 net, sk, skb, NULL, skb_dst(skb)->dev, 673 vrf_finish_output6, 674 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 675 } 676 677 /* set dst on skb to send packet to us via dev_xmit path. Allows 678 * packet to go through device based features such as qdisc, netfilter 679 * hooks and packet sockets with skb->dev set to vrf device. 680 */ 681 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev, 682 struct sk_buff *skb) 683 { 684 struct net_vrf *vrf = netdev_priv(vrf_dev); 685 struct dst_entry *dst = NULL; 686 struct rt6_info *rt6; 687 688 rcu_read_lock(); 689 690 rt6 = rcu_dereference(vrf->rt6); 691 if (likely(rt6)) { 692 dst = &rt6->dst; 693 dst_hold(dst); 694 } 695 696 rcu_read_unlock(); 697 698 if (unlikely(!dst)) { 699 vrf_tx_error(vrf_dev, skb); 700 return NULL; 701 } 702 703 skb_dst_drop(skb); 704 skb_dst_set(skb, dst); 705 706 return skb; 707 } 708 709 static int vrf_output6_direct_finish(struct net *net, struct sock *sk, 710 struct sk_buff *skb) 711 { 712 vrf_finish_direct(skb); 713 714 return vrf_ip6_local_out(net, sk, skb); 715 } 716 717 static int vrf_output6_direct(struct net *net, struct sock *sk, 718 struct sk_buff *skb) 719 { 720 int err = 1; 721 722 skb->protocol = htons(ETH_P_IPV6); 723 724 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 725 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb, 726 NULL, skb->dev, vrf_output6_direct_finish); 727 728 if (likely(err == 1)) 729 vrf_finish_direct(skb); 730 731 return err; 732 } 733 734 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk, 735 struct sk_buff *skb) 736 { 737 int err; 738 739 err = vrf_output6_direct(net, sk, skb); 740 if (likely(err == 1)) 741 err = vrf_ip6_local_out(net, sk, skb); 742 743 return err; 744 } 745 746 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev, 747 struct sock *sk, 748 struct sk_buff *skb) 749 { 750 struct net *net = dev_net(vrf_dev); 751 int err; 752 753 skb->dev = vrf_dev; 754 755 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, 756 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish); 757 758 if (likely(err == 1)) 759 err = vrf_output6_direct(net, sk, skb); 760 761 if (likely(err == 1)) 762 return skb; 763 764 return NULL; 765 } 766 767 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 768 struct sock *sk, 769 struct sk_buff *skb) 770 { 771 /* don't divert link scope packets */ 772 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 773 return skb; 774 775 if (qdisc_tx_is_default(vrf_dev) || 776 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED) 777 return vrf_ip6_out_direct(vrf_dev, sk, skb); 778 779 return vrf_ip6_out_redirect(vrf_dev, skb); 780 } 781 782 /* holding rtnl */ 783 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 784 { 785 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 786 struct net *net = dev_net(dev); 787 struct dst_entry *dst; 788 789 RCU_INIT_POINTER(vrf->rt6, NULL); 790 synchronize_rcu(); 791 792 /* move dev in dst's to loopback so this VRF device can be deleted 793 * - based on dst_ifdown 794 */ 795 if (rt6) { 796 dst = &rt6->dst; 797 dev_put(dst->dev); 798 dst->dev = net->loopback_dev; 799 dev_hold(dst->dev); 800 dst_release(dst); 801 } 802 } 803 804 static int vrf_rt6_create(struct net_device *dev) 805 { 806 int flags = DST_NOPOLICY | DST_NOXFRM; 807 struct net_vrf *vrf = netdev_priv(dev); 808 struct net *net = dev_net(dev); 809 struct rt6_info *rt6; 810 int rc = -ENOMEM; 811 812 /* IPv6 can be CONFIG enabled and then disabled runtime */ 813 if (!ipv6_mod_enabled()) 814 return 0; 815 816 vrf->fib6_table = fib6_new_table(net, vrf->tb_id); 817 if (!vrf->fib6_table) 818 goto out; 819 820 /* create a dst for routing packets out a VRF device */ 821 rt6 = ip6_dst_alloc(net, dev, flags); 822 if (!rt6) 823 goto out; 824 825 rt6->dst.output = vrf_output6; 826 827 rcu_assign_pointer(vrf->rt6, rt6); 828 829 rc = 0; 830 out: 831 return rc; 832 } 833 #else 834 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 835 struct sock *sk, 836 struct sk_buff *skb) 837 { 838 return skb; 839 } 840 841 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 842 { 843 } 844 845 static int vrf_rt6_create(struct net_device *dev) 846 { 847 return 0; 848 } 849 #endif 850 851 /* modelled after ip_finish_output2 */ 852 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 853 { 854 struct dst_entry *dst = skb_dst(skb); 855 struct rtable *rt = (struct rtable *)dst; 856 struct net_device *dev = dst->dev; 857 unsigned int hh_len = LL_RESERVED_SPACE(dev); 858 struct neighbour *neigh; 859 bool is_v6gw = false; 860 861 nf_reset_ct(skb); 862 863 /* Be paranoid, rather than too clever. */ 864 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 865 skb = skb_expand_head(skb, hh_len); 866 if (!skb) { 867 dev->stats.tx_errors++; 868 return -ENOMEM; 869 } 870 } 871 872 rcu_read_lock_bh(); 873 874 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 875 if (!IS_ERR(neigh)) { 876 int ret; 877 878 sock_confirm_neigh(skb, neigh); 879 /* if crossing protocols, can not use the cached header */ 880 ret = neigh_output(neigh, skb, is_v6gw); 881 rcu_read_unlock_bh(); 882 return ret; 883 } 884 885 rcu_read_unlock_bh(); 886 vrf_tx_error(skb->dev, skb); 887 return -EINVAL; 888 } 889 890 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 891 { 892 struct net_device *dev = skb_dst(skb)->dev; 893 894 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 895 896 skb->dev = dev; 897 skb->protocol = htons(ETH_P_IP); 898 899 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 900 net, sk, skb, NULL, dev, 901 vrf_finish_output, 902 !(IPCB(skb)->flags & IPSKB_REROUTED)); 903 } 904 905 /* set dst on skb to send packet to us via dev_xmit path. Allows 906 * packet to go through device based features such as qdisc, netfilter 907 * hooks and packet sockets with skb->dev set to vrf device. 908 */ 909 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev, 910 struct sk_buff *skb) 911 { 912 struct net_vrf *vrf = netdev_priv(vrf_dev); 913 struct dst_entry *dst = NULL; 914 struct rtable *rth; 915 916 rcu_read_lock(); 917 918 rth = rcu_dereference(vrf->rth); 919 if (likely(rth)) { 920 dst = &rth->dst; 921 dst_hold(dst); 922 } 923 924 rcu_read_unlock(); 925 926 if (unlikely(!dst)) { 927 vrf_tx_error(vrf_dev, skb); 928 return NULL; 929 } 930 931 skb_dst_drop(skb); 932 skb_dst_set(skb, dst); 933 934 return skb; 935 } 936 937 static int vrf_output_direct_finish(struct net *net, struct sock *sk, 938 struct sk_buff *skb) 939 { 940 vrf_finish_direct(skb); 941 942 return vrf_ip_local_out(net, sk, skb); 943 } 944 945 static int vrf_output_direct(struct net *net, struct sock *sk, 946 struct sk_buff *skb) 947 { 948 int err = 1; 949 950 skb->protocol = htons(ETH_P_IP); 951 952 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 953 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb, 954 NULL, skb->dev, vrf_output_direct_finish); 955 956 if (likely(err == 1)) 957 vrf_finish_direct(skb); 958 959 return err; 960 } 961 962 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk, 963 struct sk_buff *skb) 964 { 965 int err; 966 967 err = vrf_output_direct(net, sk, skb); 968 if (likely(err == 1)) 969 err = vrf_ip_local_out(net, sk, skb); 970 971 return err; 972 } 973 974 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev, 975 struct sock *sk, 976 struct sk_buff *skb) 977 { 978 struct net *net = dev_net(vrf_dev); 979 int err; 980 981 skb->dev = vrf_dev; 982 983 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 984 skb, NULL, vrf_dev, vrf_ip_out_direct_finish); 985 986 if (likely(err == 1)) 987 err = vrf_output_direct(net, sk, skb); 988 989 if (likely(err == 1)) 990 return skb; 991 992 return NULL; 993 } 994 995 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 996 struct sock *sk, 997 struct sk_buff *skb) 998 { 999 /* don't divert multicast or local broadcast */ 1000 if (ipv4_is_multicast(ip_hdr(skb)->daddr) || 1001 ipv4_is_lbcast(ip_hdr(skb)->daddr)) 1002 return skb; 1003 1004 if (qdisc_tx_is_default(vrf_dev) || 1005 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED) 1006 return vrf_ip_out_direct(vrf_dev, sk, skb); 1007 1008 return vrf_ip_out_redirect(vrf_dev, skb); 1009 } 1010 1011 /* called with rcu lock held */ 1012 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 1013 struct sock *sk, 1014 struct sk_buff *skb, 1015 u16 proto) 1016 { 1017 switch (proto) { 1018 case AF_INET: 1019 return vrf_ip_out(vrf_dev, sk, skb); 1020 case AF_INET6: 1021 return vrf_ip6_out(vrf_dev, sk, skb); 1022 } 1023 1024 return skb; 1025 } 1026 1027 /* holding rtnl */ 1028 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 1029 { 1030 struct rtable *rth = rtnl_dereference(vrf->rth); 1031 struct net *net = dev_net(dev); 1032 struct dst_entry *dst; 1033 1034 RCU_INIT_POINTER(vrf->rth, NULL); 1035 synchronize_rcu(); 1036 1037 /* move dev in dst's to loopback so this VRF device can be deleted 1038 * - based on dst_ifdown 1039 */ 1040 if (rth) { 1041 dst = &rth->dst; 1042 dev_put(dst->dev); 1043 dst->dev = net->loopback_dev; 1044 dev_hold(dst->dev); 1045 dst_release(dst); 1046 } 1047 } 1048 1049 static int vrf_rtable_create(struct net_device *dev) 1050 { 1051 struct net_vrf *vrf = netdev_priv(dev); 1052 struct rtable *rth; 1053 1054 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 1055 return -ENOMEM; 1056 1057 /* create a dst for routing packets out through a VRF device */ 1058 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1); 1059 if (!rth) 1060 return -ENOMEM; 1061 1062 rth->dst.output = vrf_output; 1063 1064 rcu_assign_pointer(vrf->rth, rth); 1065 1066 return 0; 1067 } 1068 1069 /**************************** device handling ********************/ 1070 1071 /* cycle interface to flush neighbor cache and move routes across tables */ 1072 static void cycle_netdev(struct net_device *dev, 1073 struct netlink_ext_ack *extack) 1074 { 1075 unsigned int flags = dev->flags; 1076 int ret; 1077 1078 if (!netif_running(dev)) 1079 return; 1080 1081 ret = dev_change_flags(dev, flags & ~IFF_UP, extack); 1082 if (ret >= 0) 1083 ret = dev_change_flags(dev, flags, extack); 1084 1085 if (ret < 0) { 1086 netdev_err(dev, 1087 "Failed to cycle device %s; route tables might be wrong!\n", 1088 dev->name); 1089 } 1090 } 1091 1092 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1093 struct netlink_ext_ack *extack) 1094 { 1095 int ret; 1096 1097 /* do not allow loopback device to be enslaved to a VRF. 1098 * The vrf device acts as the loopback for the vrf. 1099 */ 1100 if (port_dev == dev_net(dev)->loopback_dev) { 1101 NL_SET_ERR_MSG(extack, 1102 "Can not enslave loopback device to a VRF"); 1103 return -EOPNOTSUPP; 1104 } 1105 1106 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 1107 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack); 1108 if (ret < 0) 1109 goto err; 1110 1111 cycle_netdev(port_dev, extack); 1112 1113 return 0; 1114 1115 err: 1116 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1117 return ret; 1118 } 1119 1120 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1121 struct netlink_ext_ack *extack) 1122 { 1123 if (netif_is_l3_master(port_dev)) { 1124 NL_SET_ERR_MSG(extack, 1125 "Can not enslave an L3 master device to a VRF"); 1126 return -EINVAL; 1127 } 1128 1129 if (netif_is_l3_slave(port_dev)) 1130 return -EINVAL; 1131 1132 return do_vrf_add_slave(dev, port_dev, extack); 1133 } 1134 1135 /* inverse of do_vrf_add_slave */ 1136 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1137 { 1138 netdev_upper_dev_unlink(port_dev, dev); 1139 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1140 1141 cycle_netdev(port_dev, NULL); 1142 1143 return 0; 1144 } 1145 1146 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1147 { 1148 return do_vrf_del_slave(dev, port_dev); 1149 } 1150 1151 static void vrf_dev_uninit(struct net_device *dev) 1152 { 1153 struct net_vrf *vrf = netdev_priv(dev); 1154 1155 vrf_rtable_release(dev, vrf); 1156 vrf_rt6_release(dev, vrf); 1157 1158 free_percpu(dev->dstats); 1159 dev->dstats = NULL; 1160 } 1161 1162 static int vrf_dev_init(struct net_device *dev) 1163 { 1164 struct net_vrf *vrf = netdev_priv(dev); 1165 1166 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 1167 if (!dev->dstats) 1168 goto out_nomem; 1169 1170 /* create the default dst which points back to us */ 1171 if (vrf_rtable_create(dev) != 0) 1172 goto out_stats; 1173 1174 if (vrf_rt6_create(dev) != 0) 1175 goto out_rth; 1176 1177 dev->flags = IFF_MASTER | IFF_NOARP; 1178 1179 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 1180 dev->operstate = IF_OPER_UP; 1181 netdev_lockdep_set_classes(dev); 1182 return 0; 1183 1184 out_rth: 1185 vrf_rtable_release(dev, vrf); 1186 out_stats: 1187 free_percpu(dev->dstats); 1188 dev->dstats = NULL; 1189 out_nomem: 1190 return -ENOMEM; 1191 } 1192 1193 static const struct net_device_ops vrf_netdev_ops = { 1194 .ndo_init = vrf_dev_init, 1195 .ndo_uninit = vrf_dev_uninit, 1196 .ndo_start_xmit = vrf_xmit, 1197 .ndo_set_mac_address = eth_mac_addr, 1198 .ndo_get_stats64 = vrf_get_stats64, 1199 .ndo_add_slave = vrf_add_slave, 1200 .ndo_del_slave = vrf_del_slave, 1201 }; 1202 1203 static u32 vrf_fib_table(const struct net_device *dev) 1204 { 1205 struct net_vrf *vrf = netdev_priv(dev); 1206 1207 return vrf->tb_id; 1208 } 1209 1210 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 1211 { 1212 kfree_skb(skb); 1213 return 0; 1214 } 1215 1216 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 1217 struct sk_buff *skb, 1218 struct net_device *dev) 1219 { 1220 struct net *net = dev_net(dev); 1221 1222 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1) 1223 skb = NULL; /* kfree_skb(skb) handled by nf code */ 1224 1225 return skb; 1226 } 1227 1228 static int vrf_prepare_mac_header(struct sk_buff *skb, 1229 struct net_device *vrf_dev, u16 proto) 1230 { 1231 struct ethhdr *eth; 1232 int err; 1233 1234 /* in general, we do not know if there is enough space in the head of 1235 * the packet for hosting the mac header. 1236 */ 1237 err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev)); 1238 if (unlikely(err)) 1239 /* no space in the skb head */ 1240 return -ENOBUFS; 1241 1242 __skb_push(skb, ETH_HLEN); 1243 eth = (struct ethhdr *)skb->data; 1244 1245 skb_reset_mac_header(skb); 1246 1247 /* we set the ethernet destination and the source addresses to the 1248 * address of the VRF device. 1249 */ 1250 ether_addr_copy(eth->h_dest, vrf_dev->dev_addr); 1251 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 1252 eth->h_proto = htons(proto); 1253 1254 /* the destination address of the Ethernet frame corresponds to the 1255 * address set on the VRF interface; therefore, the packet is intended 1256 * to be processed locally. 1257 */ 1258 skb->protocol = eth->h_proto; 1259 skb->pkt_type = PACKET_HOST; 1260 1261 skb_postpush_rcsum(skb, skb->data, ETH_HLEN); 1262 1263 skb_pull_inline(skb, ETH_HLEN); 1264 1265 return 0; 1266 } 1267 1268 /* prepare and add the mac header to the packet if it was not set previously. 1269 * In this way, packet sniffers such as tcpdump can parse the packet correctly. 1270 * If the mac header was already set, the original mac header is left 1271 * untouched and the function returns immediately. 1272 */ 1273 static int vrf_add_mac_header_if_unset(struct sk_buff *skb, 1274 struct net_device *vrf_dev, 1275 u16 proto) 1276 { 1277 if (skb_mac_header_was_set(skb)) 1278 return 0; 1279 1280 return vrf_prepare_mac_header(skb, vrf_dev, proto); 1281 } 1282 1283 #if IS_ENABLED(CONFIG_IPV6) 1284 /* neighbor handling is done with actual device; do not want 1285 * to flip skb->dev for those ndisc packets. This really fails 1286 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 1287 * a start. 1288 */ 1289 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 1290 { 1291 const struct ipv6hdr *iph = ipv6_hdr(skb); 1292 bool rc = false; 1293 1294 if (iph->nexthdr == NEXTHDR_ICMP) { 1295 const struct icmp6hdr *icmph; 1296 struct icmp6hdr _icmph; 1297 1298 icmph = skb_header_pointer(skb, sizeof(*iph), 1299 sizeof(_icmph), &_icmph); 1300 if (!icmph) 1301 goto out; 1302 1303 switch (icmph->icmp6_type) { 1304 case NDISC_ROUTER_SOLICITATION: 1305 case NDISC_ROUTER_ADVERTISEMENT: 1306 case NDISC_NEIGHBOUR_SOLICITATION: 1307 case NDISC_NEIGHBOUR_ADVERTISEMENT: 1308 case NDISC_REDIRECT: 1309 rc = true; 1310 break; 1311 } 1312 } 1313 1314 out: 1315 return rc; 1316 } 1317 1318 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 1319 const struct net_device *dev, 1320 struct flowi6 *fl6, 1321 int ifindex, 1322 const struct sk_buff *skb, 1323 int flags) 1324 { 1325 struct net_vrf *vrf = netdev_priv(dev); 1326 1327 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags); 1328 } 1329 1330 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 1331 int ifindex) 1332 { 1333 const struct ipv6hdr *iph = ipv6_hdr(skb); 1334 struct flowi6 fl6 = { 1335 .flowi6_iif = ifindex, 1336 .flowi6_mark = skb->mark, 1337 .flowi6_proto = iph->nexthdr, 1338 .daddr = iph->daddr, 1339 .saddr = iph->saddr, 1340 .flowlabel = ip6_flowinfo(iph), 1341 }; 1342 struct net *net = dev_net(vrf_dev); 1343 struct rt6_info *rt6; 1344 1345 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb, 1346 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 1347 if (unlikely(!rt6)) 1348 return; 1349 1350 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 1351 return; 1352 1353 skb_dst_set(skb, &rt6->dst); 1354 } 1355 1356 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1357 struct sk_buff *skb) 1358 { 1359 int orig_iif = skb->skb_iif; 1360 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 1361 bool is_ndisc = ipv6_ndisc_frame(skb); 1362 1363 nf_reset_ct(skb); 1364 1365 /* loopback, multicast & non-ND link-local traffic; do not push through 1366 * packet taps again. Reset pkt_type for upper layers to process skb. 1367 * For strict packets with a source LLA, determine the dst using the 1368 * original ifindex. 1369 */ 1370 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) { 1371 skb->dev = vrf_dev; 1372 skb->skb_iif = vrf_dev->ifindex; 1373 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1374 1375 if (skb->pkt_type == PACKET_LOOPBACK) 1376 skb->pkt_type = PACKET_HOST; 1377 else if (ipv6_addr_type(&ipv6_hdr(skb)->saddr) & IPV6_ADDR_LINKLOCAL) 1378 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1379 1380 goto out; 1381 } 1382 1383 /* if packet is NDISC then keep the ingress interface */ 1384 if (!is_ndisc) { 1385 vrf_rx_stats(vrf_dev, skb->len); 1386 skb->dev = vrf_dev; 1387 skb->skb_iif = vrf_dev->ifindex; 1388 1389 if (!list_empty(&vrf_dev->ptype_all)) { 1390 int err; 1391 1392 err = vrf_add_mac_header_if_unset(skb, vrf_dev, 1393 ETH_P_IPV6); 1394 if (likely(!err)) { 1395 skb_push(skb, skb->mac_len); 1396 dev_queue_xmit_nit(skb, vrf_dev); 1397 skb_pull(skb, skb->mac_len); 1398 } 1399 } 1400 1401 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1402 } 1403 1404 if (need_strict) 1405 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1406 1407 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 1408 out: 1409 return skb; 1410 } 1411 1412 #else 1413 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1414 struct sk_buff *skb) 1415 { 1416 return skb; 1417 } 1418 #endif 1419 1420 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 1421 struct sk_buff *skb) 1422 { 1423 skb->dev = vrf_dev; 1424 skb->skb_iif = vrf_dev->ifindex; 1425 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1426 1427 nf_reset_ct(skb); 1428 1429 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1430 goto out; 1431 1432 /* loopback traffic; do not push through packet taps again. 1433 * Reset pkt_type for upper layers to process skb 1434 */ 1435 if (skb->pkt_type == PACKET_LOOPBACK) { 1436 skb->pkt_type = PACKET_HOST; 1437 goto out; 1438 } 1439 1440 vrf_rx_stats(vrf_dev, skb->len); 1441 1442 if (!list_empty(&vrf_dev->ptype_all)) { 1443 int err; 1444 1445 err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP); 1446 if (likely(!err)) { 1447 skb_push(skb, skb->mac_len); 1448 dev_queue_xmit_nit(skb, vrf_dev); 1449 skb_pull(skb, skb->mac_len); 1450 } 1451 } 1452 1453 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1454 out: 1455 return skb; 1456 } 1457 1458 /* called with rcu lock held */ 1459 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1460 struct sk_buff *skb, 1461 u16 proto) 1462 { 1463 switch (proto) { 1464 case AF_INET: 1465 return vrf_ip_rcv(vrf_dev, skb); 1466 case AF_INET6: 1467 return vrf_ip6_rcv(vrf_dev, skb); 1468 } 1469 1470 return skb; 1471 } 1472 1473 #if IS_ENABLED(CONFIG_IPV6) 1474 /* send to link-local or multicast address via interface enslaved to 1475 * VRF device. Force lookup to VRF table without changing flow struct 1476 * Note: Caller to this function must hold rcu_read_lock() and no refcnt 1477 * is taken on the dst by this function. 1478 */ 1479 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1480 struct flowi6 *fl6) 1481 { 1482 struct net *net = dev_net(dev); 1483 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF; 1484 struct dst_entry *dst = NULL; 1485 struct rt6_info *rt; 1486 1487 /* VRF device does not have a link-local address and 1488 * sending packets to link-local or mcast addresses over 1489 * a VRF device does not make sense 1490 */ 1491 if (fl6->flowi6_oif == dev->ifindex) { 1492 dst = &net->ipv6.ip6_null_entry->dst; 1493 return dst; 1494 } 1495 1496 if (!ipv6_addr_any(&fl6->saddr)) 1497 flags |= RT6_LOOKUP_F_HAS_SADDR; 1498 1499 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags); 1500 if (rt) 1501 dst = &rt->dst; 1502 1503 return dst; 1504 } 1505 #endif 1506 1507 static const struct l3mdev_ops vrf_l3mdev_ops = { 1508 .l3mdev_fib_table = vrf_fib_table, 1509 .l3mdev_l3_rcv = vrf_l3_rcv, 1510 .l3mdev_l3_out = vrf_l3_out, 1511 #if IS_ENABLED(CONFIG_IPV6) 1512 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1513 #endif 1514 }; 1515 1516 static void vrf_get_drvinfo(struct net_device *dev, 1517 struct ethtool_drvinfo *info) 1518 { 1519 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 1520 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 1521 } 1522 1523 static const struct ethtool_ops vrf_ethtool_ops = { 1524 .get_drvinfo = vrf_get_drvinfo, 1525 }; 1526 1527 static inline size_t vrf_fib_rule_nl_size(void) 1528 { 1529 size_t sz; 1530 1531 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1532 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1533 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1534 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */ 1535 1536 return sz; 1537 } 1538 1539 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1540 { 1541 struct fib_rule_hdr *frh; 1542 struct nlmsghdr *nlh; 1543 struct sk_buff *skb; 1544 int err; 1545 1546 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) && 1547 !ipv6_mod_enabled()) 1548 return 0; 1549 1550 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1551 if (!skb) 1552 return -ENOMEM; 1553 1554 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1555 if (!nlh) 1556 goto nla_put_failure; 1557 1558 /* rule only needs to appear once */ 1559 nlh->nlmsg_flags |= NLM_F_EXCL; 1560 1561 frh = nlmsg_data(nlh); 1562 memset(frh, 0, sizeof(*frh)); 1563 frh->family = family; 1564 frh->action = FR_ACT_TO_TBL; 1565 1566 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL)) 1567 goto nla_put_failure; 1568 1569 if (nla_put_u8(skb, FRA_L3MDEV, 1)) 1570 goto nla_put_failure; 1571 1572 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1573 goto nla_put_failure; 1574 1575 nlmsg_end(skb, nlh); 1576 1577 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1578 skb->sk = dev_net(dev)->rtnl; 1579 if (add_it) { 1580 err = fib_nl_newrule(skb, nlh, NULL); 1581 if (err == -EEXIST) 1582 err = 0; 1583 } else { 1584 err = fib_nl_delrule(skb, nlh, NULL); 1585 if (err == -ENOENT) 1586 err = 0; 1587 } 1588 nlmsg_free(skb); 1589 1590 return err; 1591 1592 nla_put_failure: 1593 nlmsg_free(skb); 1594 1595 return -EMSGSIZE; 1596 } 1597 1598 static int vrf_add_fib_rules(const struct net_device *dev) 1599 { 1600 int err; 1601 1602 err = vrf_fib_rule(dev, AF_INET, true); 1603 if (err < 0) 1604 goto out_err; 1605 1606 err = vrf_fib_rule(dev, AF_INET6, true); 1607 if (err < 0) 1608 goto ipv6_err; 1609 1610 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1611 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1612 if (err < 0) 1613 goto ipmr_err; 1614 #endif 1615 1616 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1617 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true); 1618 if (err < 0) 1619 goto ip6mr_err; 1620 #endif 1621 1622 return 0; 1623 1624 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1625 ip6mr_err: 1626 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false); 1627 #endif 1628 1629 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1630 ipmr_err: 1631 vrf_fib_rule(dev, AF_INET6, false); 1632 #endif 1633 1634 ipv6_err: 1635 vrf_fib_rule(dev, AF_INET, false); 1636 1637 out_err: 1638 netdev_err(dev, "Failed to add FIB rules.\n"); 1639 return err; 1640 } 1641 1642 static void vrf_setup(struct net_device *dev) 1643 { 1644 ether_setup(dev); 1645 1646 /* Initialize the device structure. */ 1647 dev->netdev_ops = &vrf_netdev_ops; 1648 dev->l3mdev_ops = &vrf_l3mdev_ops; 1649 dev->ethtool_ops = &vrf_ethtool_ops; 1650 dev->needs_free_netdev = true; 1651 1652 /* Fill in device structure with ethernet-generic values. */ 1653 eth_hw_addr_random(dev); 1654 1655 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1656 dev->features |= NETIF_F_LLTX; 1657 1658 /* don't allow vrf devices to change network namespaces. */ 1659 dev->features |= NETIF_F_NETNS_LOCAL; 1660 1661 /* does not make sense for a VLAN to be added to a vrf device */ 1662 dev->features |= NETIF_F_VLAN_CHALLENGED; 1663 1664 /* enable offload features */ 1665 dev->features |= NETIF_F_GSO_SOFTWARE; 1666 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC; 1667 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1668 1669 dev->hw_features = dev->features; 1670 dev->hw_enc_features = dev->features; 1671 1672 /* default to no qdisc; user can add if desired */ 1673 dev->priv_flags |= IFF_NO_QUEUE; 1674 dev->priv_flags |= IFF_NO_RX_HANDLER; 1675 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1676 1677 /* VRF devices do not care about MTU, but if the MTU is set 1678 * too low then the ipv4 and ipv6 protocols are disabled 1679 * which breaks networking. 1680 */ 1681 dev->min_mtu = IPV6_MIN_MTU; 1682 dev->max_mtu = IP6_MAX_MTU; 1683 dev->mtu = dev->max_mtu; 1684 } 1685 1686 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[], 1687 struct netlink_ext_ack *extack) 1688 { 1689 if (tb[IFLA_ADDRESS]) { 1690 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1691 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1692 return -EINVAL; 1693 } 1694 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1695 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1696 return -EADDRNOTAVAIL; 1697 } 1698 } 1699 return 0; 1700 } 1701 1702 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1703 { 1704 struct net_device *port_dev; 1705 struct list_head *iter; 1706 1707 netdev_for_each_lower_dev(dev, port_dev, iter) 1708 vrf_del_slave(dev, port_dev); 1709 1710 vrf_map_unregister_dev(dev); 1711 1712 unregister_netdevice_queue(dev, head); 1713 } 1714 1715 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1716 struct nlattr *tb[], struct nlattr *data[], 1717 struct netlink_ext_ack *extack) 1718 { 1719 struct net_vrf *vrf = netdev_priv(dev); 1720 struct netns_vrf *nn_vrf; 1721 bool *add_fib_rules; 1722 struct net *net; 1723 int err; 1724 1725 if (!data || !data[IFLA_VRF_TABLE]) { 1726 NL_SET_ERR_MSG(extack, "VRF table id is missing"); 1727 return -EINVAL; 1728 } 1729 1730 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1731 if (vrf->tb_id == RT_TABLE_UNSPEC) { 1732 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE], 1733 "Invalid VRF table id"); 1734 return -EINVAL; 1735 } 1736 1737 dev->priv_flags |= IFF_L3MDEV_MASTER; 1738 1739 err = register_netdevice(dev); 1740 if (err) 1741 goto out; 1742 1743 /* mapping between table_id and vrf; 1744 * note: such binding could not be done in the dev init function 1745 * because dev->ifindex id is not available yet. 1746 */ 1747 vrf->ifindex = dev->ifindex; 1748 1749 err = vrf_map_register_dev(dev, extack); 1750 if (err) { 1751 unregister_netdevice(dev); 1752 goto out; 1753 } 1754 1755 net = dev_net(dev); 1756 nn_vrf = net_generic(net, vrf_net_id); 1757 1758 add_fib_rules = &nn_vrf->add_fib_rules; 1759 if (*add_fib_rules) { 1760 err = vrf_add_fib_rules(dev); 1761 if (err) { 1762 vrf_map_unregister_dev(dev); 1763 unregister_netdevice(dev); 1764 goto out; 1765 } 1766 *add_fib_rules = false; 1767 } 1768 1769 out: 1770 return err; 1771 } 1772 1773 static size_t vrf_nl_getsize(const struct net_device *dev) 1774 { 1775 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1776 } 1777 1778 static int vrf_fillinfo(struct sk_buff *skb, 1779 const struct net_device *dev) 1780 { 1781 struct net_vrf *vrf = netdev_priv(dev); 1782 1783 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1784 } 1785 1786 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1787 const struct net_device *slave_dev) 1788 { 1789 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1790 } 1791 1792 static int vrf_fill_slave_info(struct sk_buff *skb, 1793 const struct net_device *vrf_dev, 1794 const struct net_device *slave_dev) 1795 { 1796 struct net_vrf *vrf = netdev_priv(vrf_dev); 1797 1798 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1799 return -EMSGSIZE; 1800 1801 return 0; 1802 } 1803 1804 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1805 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1806 }; 1807 1808 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1809 .kind = DRV_NAME, 1810 .priv_size = sizeof(struct net_vrf), 1811 1812 .get_size = vrf_nl_getsize, 1813 .policy = vrf_nl_policy, 1814 .validate = vrf_validate, 1815 .fill_info = vrf_fillinfo, 1816 1817 .get_slave_size = vrf_get_slave_size, 1818 .fill_slave_info = vrf_fill_slave_info, 1819 1820 .newlink = vrf_newlink, 1821 .dellink = vrf_dellink, 1822 .setup = vrf_setup, 1823 .maxtype = IFLA_VRF_MAX, 1824 }; 1825 1826 static int vrf_device_event(struct notifier_block *unused, 1827 unsigned long event, void *ptr) 1828 { 1829 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1830 1831 /* only care about unregister events to drop slave references */ 1832 if (event == NETDEV_UNREGISTER) { 1833 struct net_device *vrf_dev; 1834 1835 if (!netif_is_l3_slave(dev)) 1836 goto out; 1837 1838 vrf_dev = netdev_master_upper_dev_get(dev); 1839 vrf_del_slave(vrf_dev, dev); 1840 } 1841 out: 1842 return NOTIFY_DONE; 1843 } 1844 1845 static struct notifier_block vrf_notifier_block __read_mostly = { 1846 .notifier_call = vrf_device_event, 1847 }; 1848 1849 static int vrf_map_init(struct vrf_map *vmap) 1850 { 1851 spin_lock_init(&vmap->vmap_lock); 1852 hash_init(vmap->ht); 1853 1854 vmap->strict_mode = false; 1855 1856 return 0; 1857 } 1858 1859 #ifdef CONFIG_SYSCTL 1860 static bool vrf_strict_mode(struct vrf_map *vmap) 1861 { 1862 bool strict_mode; 1863 1864 vrf_map_lock(vmap); 1865 strict_mode = vmap->strict_mode; 1866 vrf_map_unlock(vmap); 1867 1868 return strict_mode; 1869 } 1870 1871 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode) 1872 { 1873 bool *cur_mode; 1874 int res = 0; 1875 1876 vrf_map_lock(vmap); 1877 1878 cur_mode = &vmap->strict_mode; 1879 if (*cur_mode == new_mode) 1880 goto unlock; 1881 1882 if (*cur_mode) { 1883 /* disable strict mode */ 1884 *cur_mode = false; 1885 } else { 1886 if (vmap->shared_tables) { 1887 /* we cannot allow strict_mode because there are some 1888 * vrfs that share one or more tables. 1889 */ 1890 res = -EBUSY; 1891 goto unlock; 1892 } 1893 1894 /* no tables are shared among vrfs, so we can go back 1895 * to 1:1 association between a vrf with its table. 1896 */ 1897 *cur_mode = true; 1898 } 1899 1900 unlock: 1901 vrf_map_unlock(vmap); 1902 1903 return res; 1904 } 1905 1906 static int vrf_shared_table_handler(struct ctl_table *table, int write, 1907 void *buffer, size_t *lenp, loff_t *ppos) 1908 { 1909 struct net *net = (struct net *)table->extra1; 1910 struct vrf_map *vmap = netns_vrf_map(net); 1911 int proc_strict_mode = 0; 1912 struct ctl_table tmp = { 1913 .procname = table->procname, 1914 .data = &proc_strict_mode, 1915 .maxlen = sizeof(int), 1916 .mode = table->mode, 1917 .extra1 = SYSCTL_ZERO, 1918 .extra2 = SYSCTL_ONE, 1919 }; 1920 int ret; 1921 1922 if (!write) 1923 proc_strict_mode = vrf_strict_mode(vmap); 1924 1925 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 1926 1927 if (write && ret == 0) 1928 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode); 1929 1930 return ret; 1931 } 1932 1933 static const struct ctl_table vrf_table[] = { 1934 { 1935 .procname = "strict_mode", 1936 .data = NULL, 1937 .maxlen = sizeof(int), 1938 .mode = 0644, 1939 .proc_handler = vrf_shared_table_handler, 1940 /* set by the vrf_netns_init */ 1941 .extra1 = NULL, 1942 }, 1943 { }, 1944 }; 1945 1946 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1947 { 1948 struct ctl_table *table; 1949 1950 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL); 1951 if (!table) 1952 return -ENOMEM; 1953 1954 /* init the extra1 parameter with the reference to current netns */ 1955 table[0].extra1 = net; 1956 1957 nn_vrf->ctl_hdr = register_net_sysctl(net, "net/vrf", table); 1958 if (!nn_vrf->ctl_hdr) { 1959 kfree(table); 1960 return -ENOMEM; 1961 } 1962 1963 return 0; 1964 } 1965 1966 static void vrf_netns_exit_sysctl(struct net *net) 1967 { 1968 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1969 struct ctl_table *table; 1970 1971 table = nn_vrf->ctl_hdr->ctl_table_arg; 1972 unregister_net_sysctl_table(nn_vrf->ctl_hdr); 1973 kfree(table); 1974 } 1975 #else 1976 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1977 { 1978 return 0; 1979 } 1980 1981 static void vrf_netns_exit_sysctl(struct net *net) 1982 { 1983 } 1984 #endif 1985 1986 /* Initialize per network namespace state */ 1987 static int __net_init vrf_netns_init(struct net *net) 1988 { 1989 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1990 1991 nn_vrf->add_fib_rules = true; 1992 vrf_map_init(&nn_vrf->vmap); 1993 1994 return vrf_netns_init_sysctl(net, nn_vrf); 1995 } 1996 1997 static void __net_exit vrf_netns_exit(struct net *net) 1998 { 1999 vrf_netns_exit_sysctl(net); 2000 } 2001 2002 static struct pernet_operations vrf_net_ops __net_initdata = { 2003 .init = vrf_netns_init, 2004 .exit = vrf_netns_exit, 2005 .id = &vrf_net_id, 2006 .size = sizeof(struct netns_vrf), 2007 }; 2008 2009 static int __init vrf_init_module(void) 2010 { 2011 int rc; 2012 2013 register_netdevice_notifier(&vrf_notifier_block); 2014 2015 rc = register_pernet_subsys(&vrf_net_ops); 2016 if (rc < 0) 2017 goto error; 2018 2019 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF, 2020 vrf_ifindex_lookup_by_table_id); 2021 if (rc < 0) 2022 goto unreg_pernet; 2023 2024 rc = rtnl_link_register(&vrf_link_ops); 2025 if (rc < 0) 2026 goto table_lookup_unreg; 2027 2028 return 0; 2029 2030 table_lookup_unreg: 2031 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF, 2032 vrf_ifindex_lookup_by_table_id); 2033 2034 unreg_pernet: 2035 unregister_pernet_subsys(&vrf_net_ops); 2036 2037 error: 2038 unregister_netdevice_notifier(&vrf_notifier_block); 2039 return rc; 2040 } 2041 2042 module_init(vrf_init_module); 2043 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 2044 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 2045 MODULE_LICENSE("GPL"); 2046 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 2047 MODULE_VERSION(DRV_VERSION); 2048