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/sch_generic.h> 38 #include <net/netns/generic.h> 39 #include <net/netfilter/nf_conntrack.h> 40 41 #define DRV_NAME "vrf" 42 #define DRV_VERSION "1.1" 43 44 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */ 45 46 #define HT_MAP_BITS 4 47 #define HASH_INITVAL ((u32)0xcafef00d) 48 49 struct vrf_map { 50 DECLARE_HASHTABLE(ht, HT_MAP_BITS); 51 spinlock_t vmap_lock; 52 53 /* shared_tables: 54 * count how many distinct tables do not comply with the strict mode 55 * requirement. 56 * shared_tables value must be 0 in order to enable the strict mode. 57 * 58 * example of the evolution of shared_tables: 59 * | time 60 * add vrf0 --> table 100 shared_tables = 0 | t0 61 * add vrf1 --> table 101 shared_tables = 0 | t1 62 * add vrf2 --> table 100 shared_tables = 1 | t2 63 * add vrf3 --> table 100 shared_tables = 1 | t3 64 * add vrf4 --> table 101 shared_tables = 2 v t4 65 * 66 * shared_tables is a "step function" (or "staircase function") 67 * and it is increased by one when the second vrf is associated to a 68 * table. 69 * 70 * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1. 71 * 72 * at t3, another dev (vrf3) is bound to the same table 100 but the 73 * value of shared_tables is still 1. 74 * This means that no matter how many new vrfs will register on the 75 * table 100, the shared_tables will not increase (considering only 76 * table 100). 77 * 78 * at t4, vrf4 is bound to table 101, and shared_tables = 2. 79 * 80 * Looking at the value of shared_tables we can immediately know if 81 * the strict_mode can or cannot be enforced. Indeed, strict_mode 82 * can be enforced iff shared_tables = 0. 83 * 84 * Conversely, shared_tables is decreased when a vrf is de-associated 85 * from a table with exactly two associated vrfs. 86 */ 87 u32 shared_tables; 88 89 bool strict_mode; 90 }; 91 92 struct vrf_map_elem { 93 struct hlist_node hnode; 94 struct list_head vrf_list; /* VRFs registered to this table */ 95 96 u32 table_id; 97 int users; 98 int ifindex; 99 }; 100 101 static unsigned int vrf_net_id; 102 103 /* per netns vrf data */ 104 struct netns_vrf { 105 /* protected by rtnl lock */ 106 bool add_fib_rules; 107 108 struct vrf_map vmap; 109 struct ctl_table_header *ctl_hdr; 110 }; 111 112 struct net_vrf { 113 struct rtable __rcu *rth; 114 struct rt6_info __rcu *rt6; 115 #if IS_ENABLED(CONFIG_IPV6) 116 struct fib6_table *fib6_table; 117 #endif 118 u32 tb_id; 119 120 struct list_head me_list; /* entry in vrf_map_elem */ 121 int ifindex; 122 }; 123 124 struct pcpu_dstats { 125 u64 tx_pkts; 126 u64 tx_bytes; 127 u64 tx_drps; 128 u64 rx_pkts; 129 u64 rx_bytes; 130 u64 rx_drps; 131 struct u64_stats_sync syncp; 132 }; 133 134 static void vrf_rx_stats(struct net_device *dev, int len) 135 { 136 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 137 138 u64_stats_update_begin(&dstats->syncp); 139 dstats->rx_pkts++; 140 dstats->rx_bytes += len; 141 u64_stats_update_end(&dstats->syncp); 142 } 143 144 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 145 { 146 vrf_dev->stats.tx_errors++; 147 kfree_skb(skb); 148 } 149 150 static void vrf_get_stats64(struct net_device *dev, 151 struct rtnl_link_stats64 *stats) 152 { 153 int i; 154 155 for_each_possible_cpu(i) { 156 const struct pcpu_dstats *dstats; 157 u64 tbytes, tpkts, tdrops, rbytes, rpkts; 158 unsigned int start; 159 160 dstats = per_cpu_ptr(dev->dstats, i); 161 do { 162 start = u64_stats_fetch_begin_irq(&dstats->syncp); 163 tbytes = dstats->tx_bytes; 164 tpkts = dstats->tx_pkts; 165 tdrops = dstats->tx_drps; 166 rbytes = dstats->rx_bytes; 167 rpkts = dstats->rx_pkts; 168 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start)); 169 stats->tx_bytes += tbytes; 170 stats->tx_packets += tpkts; 171 stats->tx_dropped += tdrops; 172 stats->rx_bytes += rbytes; 173 stats->rx_packets += rpkts; 174 } 175 } 176 177 static struct vrf_map *netns_vrf_map(struct net *net) 178 { 179 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 180 181 return &nn_vrf->vmap; 182 } 183 184 static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev) 185 { 186 return netns_vrf_map(dev_net(dev)); 187 } 188 189 static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me) 190 { 191 struct list_head *me_head = &me->vrf_list; 192 struct net_vrf *vrf; 193 194 if (list_empty(me_head)) 195 return -ENODEV; 196 197 vrf = list_first_entry(me_head, struct net_vrf, me_list); 198 199 return vrf->ifindex; 200 } 201 202 static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags) 203 { 204 struct vrf_map_elem *me; 205 206 me = kmalloc(sizeof(*me), flags); 207 if (!me) 208 return NULL; 209 210 return me; 211 } 212 213 static void vrf_map_elem_free(struct vrf_map_elem *me) 214 { 215 kfree(me); 216 } 217 218 static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id, 219 int ifindex, int users) 220 { 221 me->table_id = table_id; 222 me->ifindex = ifindex; 223 me->users = users; 224 INIT_LIST_HEAD(&me->vrf_list); 225 } 226 227 static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap, 228 u32 table_id) 229 { 230 struct vrf_map_elem *me; 231 u32 key; 232 233 key = jhash_1word(table_id, HASH_INITVAL); 234 hash_for_each_possible(vmap->ht, me, hnode, key) { 235 if (me->table_id == table_id) 236 return me; 237 } 238 239 return NULL; 240 } 241 242 static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me) 243 { 244 u32 table_id = me->table_id; 245 u32 key; 246 247 key = jhash_1word(table_id, HASH_INITVAL); 248 hash_add(vmap->ht, &me->hnode, key); 249 } 250 251 static void vrf_map_del_elem(struct vrf_map_elem *me) 252 { 253 hash_del(&me->hnode); 254 } 255 256 static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock) 257 { 258 spin_lock(&vmap->vmap_lock); 259 } 260 261 static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock) 262 { 263 spin_unlock(&vmap->vmap_lock); 264 } 265 266 /* called with rtnl lock held */ 267 static int 268 vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack) 269 { 270 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 271 struct net_vrf *vrf = netdev_priv(dev); 272 struct vrf_map_elem *new_me, *me; 273 u32 table_id = vrf->tb_id; 274 bool free_new_me = false; 275 int users; 276 int res; 277 278 /* we pre-allocate elements used in the spin-locked section (so that we 279 * keep the spinlock as short as possible). 280 */ 281 new_me = vrf_map_elem_alloc(GFP_KERNEL); 282 if (!new_me) 283 return -ENOMEM; 284 285 vrf_map_elem_init(new_me, table_id, dev->ifindex, 0); 286 287 vrf_map_lock(vmap); 288 289 me = vrf_map_lookup_elem(vmap, table_id); 290 if (!me) { 291 me = new_me; 292 vrf_map_add_elem(vmap, me); 293 goto link_vrf; 294 } 295 296 /* we already have an entry in the vrf_map, so it means there is (at 297 * least) a vrf registered on the specific table. 298 */ 299 free_new_me = true; 300 if (vmap->strict_mode) { 301 /* vrfs cannot share the same table */ 302 NL_SET_ERR_MSG(extack, "Table is used by another VRF"); 303 res = -EBUSY; 304 goto unlock; 305 } 306 307 link_vrf: 308 users = ++me->users; 309 if (users == 2) 310 ++vmap->shared_tables; 311 312 list_add(&vrf->me_list, &me->vrf_list); 313 314 res = 0; 315 316 unlock: 317 vrf_map_unlock(vmap); 318 319 /* clean-up, if needed */ 320 if (free_new_me) 321 vrf_map_elem_free(new_me); 322 323 return res; 324 } 325 326 /* called with rtnl lock held */ 327 static void vrf_map_unregister_dev(struct net_device *dev) 328 { 329 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 330 struct net_vrf *vrf = netdev_priv(dev); 331 u32 table_id = vrf->tb_id; 332 struct vrf_map_elem *me; 333 int users; 334 335 vrf_map_lock(vmap); 336 337 me = vrf_map_lookup_elem(vmap, table_id); 338 if (!me) 339 goto unlock; 340 341 list_del(&vrf->me_list); 342 343 users = --me->users; 344 if (users == 1) { 345 --vmap->shared_tables; 346 } else if (users == 0) { 347 vrf_map_del_elem(me); 348 349 /* no one will refer to this element anymore */ 350 vrf_map_elem_free(me); 351 } 352 353 unlock: 354 vrf_map_unlock(vmap); 355 } 356 357 /* return the vrf device index associated with the table_id */ 358 static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id) 359 { 360 struct vrf_map *vmap = netns_vrf_map(net); 361 struct vrf_map_elem *me; 362 int ifindex; 363 364 vrf_map_lock(vmap); 365 366 if (!vmap->strict_mode) { 367 ifindex = -EPERM; 368 goto unlock; 369 } 370 371 me = vrf_map_lookup_elem(vmap, table_id); 372 if (!me) { 373 ifindex = -ENODEV; 374 goto unlock; 375 } 376 377 ifindex = vrf_map_elem_get_vrf_ifindex(me); 378 379 unlock: 380 vrf_map_unlock(vmap); 381 382 return ifindex; 383 } 384 385 /* by default VRF devices do not have a qdisc and are expected 386 * to be created with only a single queue. 387 */ 388 static bool qdisc_tx_is_default(const struct net_device *dev) 389 { 390 struct netdev_queue *txq; 391 struct Qdisc *qdisc; 392 393 if (dev->num_tx_queues > 1) 394 return false; 395 396 txq = netdev_get_tx_queue(dev, 0); 397 qdisc = rcu_access_pointer(txq->qdisc); 398 399 return !qdisc->enqueue; 400 } 401 402 /* Local traffic destined to local address. Reinsert the packet to rx 403 * path, similar to loopback handling. 404 */ 405 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 406 struct dst_entry *dst) 407 { 408 int len = skb->len; 409 410 skb_orphan(skb); 411 412 skb_dst_set(skb, dst); 413 414 /* set pkt_type to avoid skb hitting packet taps twice - 415 * once on Tx and again in Rx processing 416 */ 417 skb->pkt_type = PACKET_LOOPBACK; 418 419 skb->protocol = eth_type_trans(skb, dev); 420 421 if (likely(__netif_rx(skb) == NET_RX_SUCCESS)) 422 vrf_rx_stats(dev, len); 423 else 424 this_cpu_inc(dev->dstats->rx_drps); 425 426 return NETDEV_TX_OK; 427 } 428 429 static void vrf_nf_set_untracked(struct sk_buff *skb) 430 { 431 if (skb_get_nfct(skb) == 0) 432 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 433 } 434 435 static void vrf_nf_reset_ct(struct sk_buff *skb) 436 { 437 if (skb_get_nfct(skb) == IP_CT_UNTRACKED) 438 nf_reset_ct(skb); 439 } 440 441 #if IS_ENABLED(CONFIG_IPV6) 442 static int vrf_ip6_local_out(struct net *net, struct sock *sk, 443 struct sk_buff *skb) 444 { 445 int err; 446 447 vrf_nf_reset_ct(skb); 448 449 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, 450 sk, skb, NULL, skb_dst(skb)->dev, dst_output); 451 452 if (likely(err == 1)) 453 err = dst_output(net, sk, skb); 454 455 return err; 456 } 457 458 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 459 struct net_device *dev) 460 { 461 const struct ipv6hdr *iph; 462 struct net *net = dev_net(skb->dev); 463 struct flowi6 fl6; 464 int ret = NET_XMIT_DROP; 465 struct dst_entry *dst; 466 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 467 468 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr))) 469 goto err; 470 471 iph = ipv6_hdr(skb); 472 473 memset(&fl6, 0, sizeof(fl6)); 474 /* needed to match OIF rule */ 475 fl6.flowi6_l3mdev = dev->ifindex; 476 fl6.flowi6_iif = LOOPBACK_IFINDEX; 477 fl6.daddr = iph->daddr; 478 fl6.saddr = iph->saddr; 479 fl6.flowlabel = ip6_flowinfo(iph); 480 fl6.flowi6_mark = skb->mark; 481 fl6.flowi6_proto = iph->nexthdr; 482 483 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL); 484 if (IS_ERR(dst) || dst == dst_null) 485 goto err; 486 487 skb_dst_drop(skb); 488 489 /* if dst.dev is the VRF device again this is locally originated traffic 490 * destined to a local address. Short circuit to Rx path. 491 */ 492 if (dst->dev == dev) 493 return vrf_local_xmit(skb, dev, dst); 494 495 skb_dst_set(skb, dst); 496 497 /* strip the ethernet header added for pass through VRF device */ 498 __skb_pull(skb, skb_network_offset(skb)); 499 500 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb))); 501 ret = vrf_ip6_local_out(net, skb->sk, skb); 502 if (unlikely(net_xmit_eval(ret))) 503 dev->stats.tx_errors++; 504 else 505 ret = NET_XMIT_SUCCESS; 506 507 return ret; 508 err: 509 vrf_tx_error(dev, skb); 510 return NET_XMIT_DROP; 511 } 512 #else 513 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 514 struct net_device *dev) 515 { 516 vrf_tx_error(dev, skb); 517 return NET_XMIT_DROP; 518 } 519 #endif 520 521 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 522 static int vrf_ip_local_out(struct net *net, struct sock *sk, 523 struct sk_buff *skb) 524 { 525 int err; 526 527 vrf_nf_reset_ct(skb); 528 529 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 530 skb, NULL, skb_dst(skb)->dev, dst_output); 531 if (likely(err == 1)) 532 err = dst_output(net, sk, skb); 533 534 return err; 535 } 536 537 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 538 struct net_device *vrf_dev) 539 { 540 struct iphdr *ip4h; 541 int ret = NET_XMIT_DROP; 542 struct flowi4 fl4; 543 struct net *net = dev_net(vrf_dev); 544 struct rtable *rt; 545 546 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr))) 547 goto err; 548 549 ip4h = ip_hdr(skb); 550 551 memset(&fl4, 0, sizeof(fl4)); 552 /* needed to match OIF rule */ 553 fl4.flowi4_l3mdev = vrf_dev->ifindex; 554 fl4.flowi4_iif = LOOPBACK_IFINDEX; 555 fl4.flowi4_tos = RT_TOS(ip4h->tos); 556 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC; 557 fl4.flowi4_proto = ip4h->protocol; 558 fl4.daddr = ip4h->daddr; 559 fl4.saddr = ip4h->saddr; 560 561 rt = ip_route_output_flow(net, &fl4, NULL); 562 if (IS_ERR(rt)) 563 goto err; 564 565 skb_dst_drop(skb); 566 567 /* if dst.dev is the VRF device again this is locally originated traffic 568 * destined to a local address. Short circuit to Rx path. 569 */ 570 if (rt->dst.dev == vrf_dev) 571 return vrf_local_xmit(skb, vrf_dev, &rt->dst); 572 573 skb_dst_set(skb, &rt->dst); 574 575 /* strip the ethernet header added for pass through VRF device */ 576 __skb_pull(skb, skb_network_offset(skb)); 577 578 if (!ip4h->saddr) { 579 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 580 RT_SCOPE_LINK); 581 } 582 583 memset(IPCB(skb), 0, sizeof(*IPCB(skb))); 584 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 585 if (unlikely(net_xmit_eval(ret))) 586 vrf_dev->stats.tx_errors++; 587 else 588 ret = NET_XMIT_SUCCESS; 589 590 out: 591 return ret; 592 err: 593 vrf_tx_error(vrf_dev, skb); 594 goto out; 595 } 596 597 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 598 { 599 switch (skb->protocol) { 600 case htons(ETH_P_IP): 601 return vrf_process_v4_outbound(skb, dev); 602 case htons(ETH_P_IPV6): 603 return vrf_process_v6_outbound(skb, dev); 604 default: 605 vrf_tx_error(dev, skb); 606 return NET_XMIT_DROP; 607 } 608 } 609 610 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 611 { 612 int len = skb->len; 613 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 614 615 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 616 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 617 618 u64_stats_update_begin(&dstats->syncp); 619 dstats->tx_pkts++; 620 dstats->tx_bytes += len; 621 u64_stats_update_end(&dstats->syncp); 622 } else { 623 this_cpu_inc(dev->dstats->tx_drps); 624 } 625 626 return ret; 627 } 628 629 static void vrf_finish_direct(struct sk_buff *skb) 630 { 631 struct net_device *vrf_dev = skb->dev; 632 633 if (!list_empty(&vrf_dev->ptype_all) && 634 likely(skb_headroom(skb) >= ETH_HLEN)) { 635 struct ethhdr *eth = skb_push(skb, ETH_HLEN); 636 637 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 638 eth_zero_addr(eth->h_dest); 639 eth->h_proto = skb->protocol; 640 641 rcu_read_lock_bh(); 642 dev_queue_xmit_nit(skb, vrf_dev); 643 rcu_read_unlock_bh(); 644 645 skb_pull(skb, ETH_HLEN); 646 } 647 648 vrf_nf_reset_ct(skb); 649 } 650 651 #if IS_ENABLED(CONFIG_IPV6) 652 /* modelled after ip6_finish_output2 */ 653 static int vrf_finish_output6(struct net *net, struct sock *sk, 654 struct sk_buff *skb) 655 { 656 struct dst_entry *dst = skb_dst(skb); 657 struct net_device *dev = dst->dev; 658 const struct in6_addr *nexthop; 659 struct neighbour *neigh; 660 int ret; 661 662 vrf_nf_reset_ct(skb); 663 664 skb->protocol = htons(ETH_P_IPV6); 665 skb->dev = dev; 666 667 rcu_read_lock_bh(); 668 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 669 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 670 if (unlikely(!neigh)) 671 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 672 if (!IS_ERR(neigh)) { 673 sock_confirm_neigh(skb, neigh); 674 ret = neigh_output(neigh, skb, false); 675 rcu_read_unlock_bh(); 676 return ret; 677 } 678 rcu_read_unlock_bh(); 679 680 IP6_INC_STATS(dev_net(dst->dev), 681 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 682 kfree_skb(skb); 683 return -EINVAL; 684 } 685 686 /* modelled after ip6_output */ 687 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 688 { 689 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 690 net, sk, skb, NULL, skb_dst(skb)->dev, 691 vrf_finish_output6, 692 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 693 } 694 695 /* set dst on skb to send packet to us via dev_xmit path. Allows 696 * packet to go through device based features such as qdisc, netfilter 697 * hooks and packet sockets with skb->dev set to vrf device. 698 */ 699 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev, 700 struct sk_buff *skb) 701 { 702 struct net_vrf *vrf = netdev_priv(vrf_dev); 703 struct dst_entry *dst = NULL; 704 struct rt6_info *rt6; 705 706 rcu_read_lock(); 707 708 rt6 = rcu_dereference(vrf->rt6); 709 if (likely(rt6)) { 710 dst = &rt6->dst; 711 dst_hold(dst); 712 } 713 714 rcu_read_unlock(); 715 716 if (unlikely(!dst)) { 717 vrf_tx_error(vrf_dev, skb); 718 return NULL; 719 } 720 721 skb_dst_drop(skb); 722 skb_dst_set(skb, dst); 723 724 return skb; 725 } 726 727 static int vrf_output6_direct_finish(struct net *net, struct sock *sk, 728 struct sk_buff *skb) 729 { 730 vrf_finish_direct(skb); 731 732 return vrf_ip6_local_out(net, sk, skb); 733 } 734 735 static int vrf_output6_direct(struct net *net, struct sock *sk, 736 struct sk_buff *skb) 737 { 738 int err = 1; 739 740 skb->protocol = htons(ETH_P_IPV6); 741 742 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 743 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb, 744 NULL, skb->dev, vrf_output6_direct_finish); 745 746 if (likely(err == 1)) 747 vrf_finish_direct(skb); 748 749 return err; 750 } 751 752 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk, 753 struct sk_buff *skb) 754 { 755 int err; 756 757 err = vrf_output6_direct(net, sk, skb); 758 if (likely(err == 1)) 759 err = vrf_ip6_local_out(net, sk, skb); 760 761 return err; 762 } 763 764 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev, 765 struct sock *sk, 766 struct sk_buff *skb) 767 { 768 struct net *net = dev_net(vrf_dev); 769 int err; 770 771 skb->dev = vrf_dev; 772 773 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, 774 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish); 775 776 if (likely(err == 1)) 777 err = vrf_output6_direct(net, sk, skb); 778 779 if (likely(err == 1)) 780 return skb; 781 782 return NULL; 783 } 784 785 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 786 struct sock *sk, 787 struct sk_buff *skb) 788 { 789 /* don't divert link scope packets */ 790 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 791 return skb; 792 793 vrf_nf_set_untracked(skb); 794 795 if (qdisc_tx_is_default(vrf_dev) || 796 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED) 797 return vrf_ip6_out_direct(vrf_dev, sk, skb); 798 799 return vrf_ip6_out_redirect(vrf_dev, skb); 800 } 801 802 /* holding rtnl */ 803 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 804 { 805 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 806 struct net *net = dev_net(dev); 807 struct dst_entry *dst; 808 809 RCU_INIT_POINTER(vrf->rt6, NULL); 810 synchronize_rcu(); 811 812 /* move dev in dst's to loopback so this VRF device can be deleted 813 * - based on dst_ifdown 814 */ 815 if (rt6) { 816 dst = &rt6->dst; 817 netdev_ref_replace(dst->dev, net->loopback_dev, 818 &dst->dev_tracker, GFP_KERNEL); 819 dst->dev = net->loopback_dev; 820 dst_release(dst); 821 } 822 } 823 824 static int vrf_rt6_create(struct net_device *dev) 825 { 826 int flags = DST_NOPOLICY | DST_NOXFRM; 827 struct net_vrf *vrf = netdev_priv(dev); 828 struct net *net = dev_net(dev); 829 struct rt6_info *rt6; 830 int rc = -ENOMEM; 831 832 /* IPv6 can be CONFIG enabled and then disabled runtime */ 833 if (!ipv6_mod_enabled()) 834 return 0; 835 836 vrf->fib6_table = fib6_new_table(net, vrf->tb_id); 837 if (!vrf->fib6_table) 838 goto out; 839 840 /* create a dst for routing packets out a VRF device */ 841 rt6 = ip6_dst_alloc(net, dev, flags); 842 if (!rt6) 843 goto out; 844 845 rt6->dst.output = vrf_output6; 846 847 rcu_assign_pointer(vrf->rt6, rt6); 848 849 rc = 0; 850 out: 851 return rc; 852 } 853 #else 854 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 855 struct sock *sk, 856 struct sk_buff *skb) 857 { 858 return skb; 859 } 860 861 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 862 { 863 } 864 865 static int vrf_rt6_create(struct net_device *dev) 866 { 867 return 0; 868 } 869 #endif 870 871 /* modelled after ip_finish_output2 */ 872 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 873 { 874 struct dst_entry *dst = skb_dst(skb); 875 struct rtable *rt = (struct rtable *)dst; 876 struct net_device *dev = dst->dev; 877 unsigned int hh_len = LL_RESERVED_SPACE(dev); 878 struct neighbour *neigh; 879 bool is_v6gw = false; 880 881 vrf_nf_reset_ct(skb); 882 883 /* Be paranoid, rather than too clever. */ 884 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 885 skb = skb_expand_head(skb, hh_len); 886 if (!skb) { 887 dev->stats.tx_errors++; 888 return -ENOMEM; 889 } 890 } 891 892 rcu_read_lock_bh(); 893 894 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 895 if (!IS_ERR(neigh)) { 896 int ret; 897 898 sock_confirm_neigh(skb, neigh); 899 /* if crossing protocols, can not use the cached header */ 900 ret = neigh_output(neigh, skb, is_v6gw); 901 rcu_read_unlock_bh(); 902 return ret; 903 } 904 905 rcu_read_unlock_bh(); 906 vrf_tx_error(skb->dev, skb); 907 return -EINVAL; 908 } 909 910 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 911 { 912 struct net_device *dev = skb_dst(skb)->dev; 913 914 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 915 916 skb->dev = dev; 917 skb->protocol = htons(ETH_P_IP); 918 919 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 920 net, sk, skb, NULL, dev, 921 vrf_finish_output, 922 !(IPCB(skb)->flags & IPSKB_REROUTED)); 923 } 924 925 /* set dst on skb to send packet to us via dev_xmit path. Allows 926 * packet to go through device based features such as qdisc, netfilter 927 * hooks and packet sockets with skb->dev set to vrf device. 928 */ 929 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev, 930 struct sk_buff *skb) 931 { 932 struct net_vrf *vrf = netdev_priv(vrf_dev); 933 struct dst_entry *dst = NULL; 934 struct rtable *rth; 935 936 rcu_read_lock(); 937 938 rth = rcu_dereference(vrf->rth); 939 if (likely(rth)) { 940 dst = &rth->dst; 941 dst_hold(dst); 942 } 943 944 rcu_read_unlock(); 945 946 if (unlikely(!dst)) { 947 vrf_tx_error(vrf_dev, skb); 948 return NULL; 949 } 950 951 skb_dst_drop(skb); 952 skb_dst_set(skb, dst); 953 954 return skb; 955 } 956 957 static int vrf_output_direct_finish(struct net *net, struct sock *sk, 958 struct sk_buff *skb) 959 { 960 vrf_finish_direct(skb); 961 962 return vrf_ip_local_out(net, sk, skb); 963 } 964 965 static int vrf_output_direct(struct net *net, struct sock *sk, 966 struct sk_buff *skb) 967 { 968 int err = 1; 969 970 skb->protocol = htons(ETH_P_IP); 971 972 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 973 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb, 974 NULL, skb->dev, vrf_output_direct_finish); 975 976 if (likely(err == 1)) 977 vrf_finish_direct(skb); 978 979 return err; 980 } 981 982 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk, 983 struct sk_buff *skb) 984 { 985 int err; 986 987 err = vrf_output_direct(net, sk, skb); 988 if (likely(err == 1)) 989 err = vrf_ip_local_out(net, sk, skb); 990 991 return err; 992 } 993 994 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev, 995 struct sock *sk, 996 struct sk_buff *skb) 997 { 998 struct net *net = dev_net(vrf_dev); 999 int err; 1000 1001 skb->dev = vrf_dev; 1002 1003 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 1004 skb, NULL, vrf_dev, vrf_ip_out_direct_finish); 1005 1006 if (likely(err == 1)) 1007 err = vrf_output_direct(net, sk, skb); 1008 1009 if (likely(err == 1)) 1010 return skb; 1011 1012 return NULL; 1013 } 1014 1015 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 1016 struct sock *sk, 1017 struct sk_buff *skb) 1018 { 1019 /* don't divert multicast or local broadcast */ 1020 if (ipv4_is_multicast(ip_hdr(skb)->daddr) || 1021 ipv4_is_lbcast(ip_hdr(skb)->daddr)) 1022 return skb; 1023 1024 vrf_nf_set_untracked(skb); 1025 1026 if (qdisc_tx_is_default(vrf_dev) || 1027 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED) 1028 return vrf_ip_out_direct(vrf_dev, sk, skb); 1029 1030 return vrf_ip_out_redirect(vrf_dev, skb); 1031 } 1032 1033 /* called with rcu lock held */ 1034 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 1035 struct sock *sk, 1036 struct sk_buff *skb, 1037 u16 proto) 1038 { 1039 switch (proto) { 1040 case AF_INET: 1041 return vrf_ip_out(vrf_dev, sk, skb); 1042 case AF_INET6: 1043 return vrf_ip6_out(vrf_dev, sk, skb); 1044 } 1045 1046 return skb; 1047 } 1048 1049 /* holding rtnl */ 1050 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 1051 { 1052 struct rtable *rth = rtnl_dereference(vrf->rth); 1053 struct net *net = dev_net(dev); 1054 struct dst_entry *dst; 1055 1056 RCU_INIT_POINTER(vrf->rth, NULL); 1057 synchronize_rcu(); 1058 1059 /* move dev in dst's to loopback so this VRF device can be deleted 1060 * - based on dst_ifdown 1061 */ 1062 if (rth) { 1063 dst = &rth->dst; 1064 netdev_ref_replace(dst->dev, net->loopback_dev, 1065 &dst->dev_tracker, GFP_KERNEL); 1066 dst->dev = net->loopback_dev; 1067 dst_release(dst); 1068 } 1069 } 1070 1071 static int vrf_rtable_create(struct net_device *dev) 1072 { 1073 struct net_vrf *vrf = netdev_priv(dev); 1074 struct rtable *rth; 1075 1076 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 1077 return -ENOMEM; 1078 1079 /* create a dst for routing packets out through a VRF device */ 1080 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1); 1081 if (!rth) 1082 return -ENOMEM; 1083 1084 rth->dst.output = vrf_output; 1085 1086 rcu_assign_pointer(vrf->rth, rth); 1087 1088 return 0; 1089 } 1090 1091 /**************************** device handling ********************/ 1092 1093 /* cycle interface to flush neighbor cache and move routes across tables */ 1094 static void cycle_netdev(struct net_device *dev, 1095 struct netlink_ext_ack *extack) 1096 { 1097 unsigned int flags = dev->flags; 1098 int ret; 1099 1100 if (!netif_running(dev)) 1101 return; 1102 1103 ret = dev_change_flags(dev, flags & ~IFF_UP, extack); 1104 if (ret >= 0) 1105 ret = dev_change_flags(dev, flags, extack); 1106 1107 if (ret < 0) { 1108 netdev_err(dev, 1109 "Failed to cycle device %s; route tables might be wrong!\n", 1110 dev->name); 1111 } 1112 } 1113 1114 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1115 struct netlink_ext_ack *extack) 1116 { 1117 int ret; 1118 1119 /* do not allow loopback device to be enslaved to a VRF. 1120 * The vrf device acts as the loopback for the vrf. 1121 */ 1122 if (port_dev == dev_net(dev)->loopback_dev) { 1123 NL_SET_ERR_MSG(extack, 1124 "Can not enslave loopback device to a VRF"); 1125 return -EOPNOTSUPP; 1126 } 1127 1128 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 1129 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack); 1130 if (ret < 0) 1131 goto err; 1132 1133 cycle_netdev(port_dev, extack); 1134 1135 return 0; 1136 1137 err: 1138 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1139 return ret; 1140 } 1141 1142 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1143 struct netlink_ext_ack *extack) 1144 { 1145 if (netif_is_l3_master(port_dev)) { 1146 NL_SET_ERR_MSG(extack, 1147 "Can not enslave an L3 master device to a VRF"); 1148 return -EINVAL; 1149 } 1150 1151 if (netif_is_l3_slave(port_dev)) 1152 return -EINVAL; 1153 1154 return do_vrf_add_slave(dev, port_dev, extack); 1155 } 1156 1157 /* inverse of do_vrf_add_slave */ 1158 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1159 { 1160 netdev_upper_dev_unlink(port_dev, dev); 1161 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1162 1163 cycle_netdev(port_dev, NULL); 1164 1165 return 0; 1166 } 1167 1168 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1169 { 1170 return do_vrf_del_slave(dev, port_dev); 1171 } 1172 1173 static void vrf_dev_uninit(struct net_device *dev) 1174 { 1175 struct net_vrf *vrf = netdev_priv(dev); 1176 1177 vrf_rtable_release(dev, vrf); 1178 vrf_rt6_release(dev, vrf); 1179 1180 free_percpu(dev->dstats); 1181 dev->dstats = NULL; 1182 } 1183 1184 static int vrf_dev_init(struct net_device *dev) 1185 { 1186 struct net_vrf *vrf = netdev_priv(dev); 1187 1188 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 1189 if (!dev->dstats) 1190 goto out_nomem; 1191 1192 /* create the default dst which points back to us */ 1193 if (vrf_rtable_create(dev) != 0) 1194 goto out_stats; 1195 1196 if (vrf_rt6_create(dev) != 0) 1197 goto out_rth; 1198 1199 dev->flags = IFF_MASTER | IFF_NOARP; 1200 1201 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 1202 dev->operstate = IF_OPER_UP; 1203 netdev_lockdep_set_classes(dev); 1204 return 0; 1205 1206 out_rth: 1207 vrf_rtable_release(dev, vrf); 1208 out_stats: 1209 free_percpu(dev->dstats); 1210 dev->dstats = NULL; 1211 out_nomem: 1212 return -ENOMEM; 1213 } 1214 1215 static const struct net_device_ops vrf_netdev_ops = { 1216 .ndo_init = vrf_dev_init, 1217 .ndo_uninit = vrf_dev_uninit, 1218 .ndo_start_xmit = vrf_xmit, 1219 .ndo_set_mac_address = eth_mac_addr, 1220 .ndo_get_stats64 = vrf_get_stats64, 1221 .ndo_add_slave = vrf_add_slave, 1222 .ndo_del_slave = vrf_del_slave, 1223 }; 1224 1225 static u32 vrf_fib_table(const struct net_device *dev) 1226 { 1227 struct net_vrf *vrf = netdev_priv(dev); 1228 1229 return vrf->tb_id; 1230 } 1231 1232 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 1233 { 1234 kfree_skb(skb); 1235 return 0; 1236 } 1237 1238 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 1239 struct sk_buff *skb, 1240 struct net_device *dev) 1241 { 1242 struct net *net = dev_net(dev); 1243 1244 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1) 1245 skb = NULL; /* kfree_skb(skb) handled by nf code */ 1246 1247 return skb; 1248 } 1249 1250 static int vrf_prepare_mac_header(struct sk_buff *skb, 1251 struct net_device *vrf_dev, u16 proto) 1252 { 1253 struct ethhdr *eth; 1254 int err; 1255 1256 /* in general, we do not know if there is enough space in the head of 1257 * the packet for hosting the mac header. 1258 */ 1259 err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev)); 1260 if (unlikely(err)) 1261 /* no space in the skb head */ 1262 return -ENOBUFS; 1263 1264 __skb_push(skb, ETH_HLEN); 1265 eth = (struct ethhdr *)skb->data; 1266 1267 skb_reset_mac_header(skb); 1268 skb_reset_mac_len(skb); 1269 1270 /* we set the ethernet destination and the source addresses to the 1271 * address of the VRF device. 1272 */ 1273 ether_addr_copy(eth->h_dest, vrf_dev->dev_addr); 1274 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 1275 eth->h_proto = htons(proto); 1276 1277 /* the destination address of the Ethernet frame corresponds to the 1278 * address set on the VRF interface; therefore, the packet is intended 1279 * to be processed locally. 1280 */ 1281 skb->protocol = eth->h_proto; 1282 skb->pkt_type = PACKET_HOST; 1283 1284 skb_postpush_rcsum(skb, skb->data, ETH_HLEN); 1285 1286 skb_pull_inline(skb, ETH_HLEN); 1287 1288 return 0; 1289 } 1290 1291 /* prepare and add the mac header to the packet if it was not set previously. 1292 * In this way, packet sniffers such as tcpdump can parse the packet correctly. 1293 * If the mac header was already set, the original mac header is left 1294 * untouched and the function returns immediately. 1295 */ 1296 static int vrf_add_mac_header_if_unset(struct sk_buff *skb, 1297 struct net_device *vrf_dev, 1298 u16 proto, struct net_device *orig_dev) 1299 { 1300 if (skb_mac_header_was_set(skb) && dev_has_header(orig_dev)) 1301 return 0; 1302 1303 return vrf_prepare_mac_header(skb, vrf_dev, proto); 1304 } 1305 1306 #if IS_ENABLED(CONFIG_IPV6) 1307 /* neighbor handling is done with actual device; do not want 1308 * to flip skb->dev for those ndisc packets. This really fails 1309 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 1310 * a start. 1311 */ 1312 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 1313 { 1314 const struct ipv6hdr *iph = ipv6_hdr(skb); 1315 bool rc = false; 1316 1317 if (iph->nexthdr == NEXTHDR_ICMP) { 1318 const struct icmp6hdr *icmph; 1319 struct icmp6hdr _icmph; 1320 1321 icmph = skb_header_pointer(skb, sizeof(*iph), 1322 sizeof(_icmph), &_icmph); 1323 if (!icmph) 1324 goto out; 1325 1326 switch (icmph->icmp6_type) { 1327 case NDISC_ROUTER_SOLICITATION: 1328 case NDISC_ROUTER_ADVERTISEMENT: 1329 case NDISC_NEIGHBOUR_SOLICITATION: 1330 case NDISC_NEIGHBOUR_ADVERTISEMENT: 1331 case NDISC_REDIRECT: 1332 rc = true; 1333 break; 1334 } 1335 } 1336 1337 out: 1338 return rc; 1339 } 1340 1341 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 1342 const struct net_device *dev, 1343 struct flowi6 *fl6, 1344 int ifindex, 1345 const struct sk_buff *skb, 1346 int flags) 1347 { 1348 struct net_vrf *vrf = netdev_priv(dev); 1349 1350 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags); 1351 } 1352 1353 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 1354 int ifindex) 1355 { 1356 const struct ipv6hdr *iph = ipv6_hdr(skb); 1357 struct flowi6 fl6 = { 1358 .flowi6_iif = ifindex, 1359 .flowi6_mark = skb->mark, 1360 .flowi6_proto = iph->nexthdr, 1361 .daddr = iph->daddr, 1362 .saddr = iph->saddr, 1363 .flowlabel = ip6_flowinfo(iph), 1364 }; 1365 struct net *net = dev_net(vrf_dev); 1366 struct rt6_info *rt6; 1367 1368 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb, 1369 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 1370 if (unlikely(!rt6)) 1371 return; 1372 1373 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 1374 return; 1375 1376 skb_dst_set(skb, &rt6->dst); 1377 } 1378 1379 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1380 struct sk_buff *skb) 1381 { 1382 int orig_iif = skb->skb_iif; 1383 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 1384 bool is_ndisc = ipv6_ndisc_frame(skb); 1385 1386 /* loopback, multicast & non-ND link-local traffic; do not push through 1387 * packet taps again. Reset pkt_type for upper layers to process skb. 1388 * For strict packets with a source LLA, determine the dst using the 1389 * original ifindex. 1390 */ 1391 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) { 1392 skb->dev = vrf_dev; 1393 skb->skb_iif = vrf_dev->ifindex; 1394 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1395 1396 if (skb->pkt_type == PACKET_LOOPBACK) 1397 skb->pkt_type = PACKET_HOST; 1398 else if (ipv6_addr_type(&ipv6_hdr(skb)->saddr) & IPV6_ADDR_LINKLOCAL) 1399 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1400 1401 goto out; 1402 } 1403 1404 /* if packet is NDISC then keep the ingress interface */ 1405 if (!is_ndisc) { 1406 struct net_device *orig_dev = skb->dev; 1407 1408 vrf_rx_stats(vrf_dev, skb->len); 1409 skb->dev = vrf_dev; 1410 skb->skb_iif = vrf_dev->ifindex; 1411 1412 if (!list_empty(&vrf_dev->ptype_all)) { 1413 int err; 1414 1415 err = vrf_add_mac_header_if_unset(skb, vrf_dev, 1416 ETH_P_IPV6, 1417 orig_dev); 1418 if (likely(!err)) { 1419 skb_push(skb, skb->mac_len); 1420 dev_queue_xmit_nit(skb, vrf_dev); 1421 skb_pull(skb, skb->mac_len); 1422 } 1423 } 1424 1425 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1426 } 1427 1428 if (need_strict) 1429 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1430 1431 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 1432 out: 1433 return skb; 1434 } 1435 1436 #else 1437 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1438 struct sk_buff *skb) 1439 { 1440 return skb; 1441 } 1442 #endif 1443 1444 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 1445 struct sk_buff *skb) 1446 { 1447 struct net_device *orig_dev = skb->dev; 1448 1449 skb->dev = vrf_dev; 1450 skb->skb_iif = vrf_dev->ifindex; 1451 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1452 1453 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1454 goto out; 1455 1456 /* loopback traffic; do not push through packet taps again. 1457 * Reset pkt_type for upper layers to process skb 1458 */ 1459 if (skb->pkt_type == PACKET_LOOPBACK) { 1460 skb->pkt_type = PACKET_HOST; 1461 goto out; 1462 } 1463 1464 vrf_rx_stats(vrf_dev, skb->len); 1465 1466 if (!list_empty(&vrf_dev->ptype_all)) { 1467 int err; 1468 1469 err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP, 1470 orig_dev); 1471 if (likely(!err)) { 1472 skb_push(skb, skb->mac_len); 1473 dev_queue_xmit_nit(skb, vrf_dev); 1474 skb_pull(skb, skb->mac_len); 1475 } 1476 } 1477 1478 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1479 out: 1480 return skb; 1481 } 1482 1483 /* called with rcu lock held */ 1484 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1485 struct sk_buff *skb, 1486 u16 proto) 1487 { 1488 switch (proto) { 1489 case AF_INET: 1490 return vrf_ip_rcv(vrf_dev, skb); 1491 case AF_INET6: 1492 return vrf_ip6_rcv(vrf_dev, skb); 1493 } 1494 1495 return skb; 1496 } 1497 1498 #if IS_ENABLED(CONFIG_IPV6) 1499 /* send to link-local or multicast address via interface enslaved to 1500 * VRF device. Force lookup to VRF table without changing flow struct 1501 * Note: Caller to this function must hold rcu_read_lock() and no refcnt 1502 * is taken on the dst by this function. 1503 */ 1504 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1505 struct flowi6 *fl6) 1506 { 1507 struct net *net = dev_net(dev); 1508 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF; 1509 struct dst_entry *dst = NULL; 1510 struct rt6_info *rt; 1511 1512 /* VRF device does not have a link-local address and 1513 * sending packets to link-local or mcast addresses over 1514 * a VRF device does not make sense 1515 */ 1516 if (fl6->flowi6_oif == dev->ifindex) { 1517 dst = &net->ipv6.ip6_null_entry->dst; 1518 return dst; 1519 } 1520 1521 if (!ipv6_addr_any(&fl6->saddr)) 1522 flags |= RT6_LOOKUP_F_HAS_SADDR; 1523 1524 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags); 1525 if (rt) 1526 dst = &rt->dst; 1527 1528 return dst; 1529 } 1530 #endif 1531 1532 static const struct l3mdev_ops vrf_l3mdev_ops = { 1533 .l3mdev_fib_table = vrf_fib_table, 1534 .l3mdev_l3_rcv = vrf_l3_rcv, 1535 .l3mdev_l3_out = vrf_l3_out, 1536 #if IS_ENABLED(CONFIG_IPV6) 1537 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1538 #endif 1539 }; 1540 1541 static void vrf_get_drvinfo(struct net_device *dev, 1542 struct ethtool_drvinfo *info) 1543 { 1544 strscpy(info->driver, DRV_NAME, sizeof(info->driver)); 1545 strscpy(info->version, DRV_VERSION, sizeof(info->version)); 1546 } 1547 1548 static const struct ethtool_ops vrf_ethtool_ops = { 1549 .get_drvinfo = vrf_get_drvinfo, 1550 }; 1551 1552 static inline size_t vrf_fib_rule_nl_size(void) 1553 { 1554 size_t sz; 1555 1556 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1557 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1558 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1559 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */ 1560 1561 return sz; 1562 } 1563 1564 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1565 { 1566 struct fib_rule_hdr *frh; 1567 struct nlmsghdr *nlh; 1568 struct sk_buff *skb; 1569 int err; 1570 1571 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) && 1572 !ipv6_mod_enabled()) 1573 return 0; 1574 1575 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1576 if (!skb) 1577 return -ENOMEM; 1578 1579 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1580 if (!nlh) 1581 goto nla_put_failure; 1582 1583 /* rule only needs to appear once */ 1584 nlh->nlmsg_flags |= NLM_F_EXCL; 1585 1586 frh = nlmsg_data(nlh); 1587 memset(frh, 0, sizeof(*frh)); 1588 frh->family = family; 1589 frh->action = FR_ACT_TO_TBL; 1590 1591 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL)) 1592 goto nla_put_failure; 1593 1594 if (nla_put_u8(skb, FRA_L3MDEV, 1)) 1595 goto nla_put_failure; 1596 1597 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1598 goto nla_put_failure; 1599 1600 nlmsg_end(skb, nlh); 1601 1602 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1603 skb->sk = dev_net(dev)->rtnl; 1604 if (add_it) { 1605 err = fib_nl_newrule(skb, nlh, NULL); 1606 if (err == -EEXIST) 1607 err = 0; 1608 } else { 1609 err = fib_nl_delrule(skb, nlh, NULL); 1610 if (err == -ENOENT) 1611 err = 0; 1612 } 1613 nlmsg_free(skb); 1614 1615 return err; 1616 1617 nla_put_failure: 1618 nlmsg_free(skb); 1619 1620 return -EMSGSIZE; 1621 } 1622 1623 static int vrf_add_fib_rules(const struct net_device *dev) 1624 { 1625 int err; 1626 1627 err = vrf_fib_rule(dev, AF_INET, true); 1628 if (err < 0) 1629 goto out_err; 1630 1631 err = vrf_fib_rule(dev, AF_INET6, true); 1632 if (err < 0) 1633 goto ipv6_err; 1634 1635 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1636 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1637 if (err < 0) 1638 goto ipmr_err; 1639 #endif 1640 1641 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1642 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true); 1643 if (err < 0) 1644 goto ip6mr_err; 1645 #endif 1646 1647 return 0; 1648 1649 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1650 ip6mr_err: 1651 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false); 1652 #endif 1653 1654 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1655 ipmr_err: 1656 vrf_fib_rule(dev, AF_INET6, false); 1657 #endif 1658 1659 ipv6_err: 1660 vrf_fib_rule(dev, AF_INET, false); 1661 1662 out_err: 1663 netdev_err(dev, "Failed to add FIB rules.\n"); 1664 return err; 1665 } 1666 1667 static void vrf_setup(struct net_device *dev) 1668 { 1669 ether_setup(dev); 1670 1671 /* Initialize the device structure. */ 1672 dev->netdev_ops = &vrf_netdev_ops; 1673 dev->l3mdev_ops = &vrf_l3mdev_ops; 1674 dev->ethtool_ops = &vrf_ethtool_ops; 1675 dev->needs_free_netdev = true; 1676 1677 /* Fill in device structure with ethernet-generic values. */ 1678 eth_hw_addr_random(dev); 1679 1680 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1681 dev->features |= NETIF_F_LLTX; 1682 1683 /* don't allow vrf devices to change network namespaces. */ 1684 dev->features |= NETIF_F_NETNS_LOCAL; 1685 1686 /* does not make sense for a VLAN to be added to a vrf device */ 1687 dev->features |= NETIF_F_VLAN_CHALLENGED; 1688 1689 /* enable offload features */ 1690 dev->features |= NETIF_F_GSO_SOFTWARE; 1691 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC; 1692 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1693 1694 dev->hw_features = dev->features; 1695 dev->hw_enc_features = dev->features; 1696 1697 /* default to no qdisc; user can add if desired */ 1698 dev->priv_flags |= IFF_NO_QUEUE; 1699 dev->priv_flags |= IFF_NO_RX_HANDLER; 1700 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1701 1702 /* VRF devices do not care about MTU, but if the MTU is set 1703 * too low then the ipv4 and ipv6 protocols are disabled 1704 * which breaks networking. 1705 */ 1706 dev->min_mtu = IPV6_MIN_MTU; 1707 dev->max_mtu = IP6_MAX_MTU; 1708 dev->mtu = dev->max_mtu; 1709 } 1710 1711 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[], 1712 struct netlink_ext_ack *extack) 1713 { 1714 if (tb[IFLA_ADDRESS]) { 1715 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1716 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1717 return -EINVAL; 1718 } 1719 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1720 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1721 return -EADDRNOTAVAIL; 1722 } 1723 } 1724 return 0; 1725 } 1726 1727 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1728 { 1729 struct net_device *port_dev; 1730 struct list_head *iter; 1731 1732 netdev_for_each_lower_dev(dev, port_dev, iter) 1733 vrf_del_slave(dev, port_dev); 1734 1735 vrf_map_unregister_dev(dev); 1736 1737 unregister_netdevice_queue(dev, head); 1738 } 1739 1740 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1741 struct nlattr *tb[], struct nlattr *data[], 1742 struct netlink_ext_ack *extack) 1743 { 1744 struct net_vrf *vrf = netdev_priv(dev); 1745 struct netns_vrf *nn_vrf; 1746 bool *add_fib_rules; 1747 struct net *net; 1748 int err; 1749 1750 if (!data || !data[IFLA_VRF_TABLE]) { 1751 NL_SET_ERR_MSG(extack, "VRF table id is missing"); 1752 return -EINVAL; 1753 } 1754 1755 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1756 if (vrf->tb_id == RT_TABLE_UNSPEC) { 1757 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE], 1758 "Invalid VRF table id"); 1759 return -EINVAL; 1760 } 1761 1762 dev->priv_flags |= IFF_L3MDEV_MASTER; 1763 1764 err = register_netdevice(dev); 1765 if (err) 1766 goto out; 1767 1768 /* mapping between table_id and vrf; 1769 * note: such binding could not be done in the dev init function 1770 * because dev->ifindex id is not available yet. 1771 */ 1772 vrf->ifindex = dev->ifindex; 1773 1774 err = vrf_map_register_dev(dev, extack); 1775 if (err) { 1776 unregister_netdevice(dev); 1777 goto out; 1778 } 1779 1780 net = dev_net(dev); 1781 nn_vrf = net_generic(net, vrf_net_id); 1782 1783 add_fib_rules = &nn_vrf->add_fib_rules; 1784 if (*add_fib_rules) { 1785 err = vrf_add_fib_rules(dev); 1786 if (err) { 1787 vrf_map_unregister_dev(dev); 1788 unregister_netdevice(dev); 1789 goto out; 1790 } 1791 *add_fib_rules = false; 1792 } 1793 1794 out: 1795 return err; 1796 } 1797 1798 static size_t vrf_nl_getsize(const struct net_device *dev) 1799 { 1800 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1801 } 1802 1803 static int vrf_fillinfo(struct sk_buff *skb, 1804 const struct net_device *dev) 1805 { 1806 struct net_vrf *vrf = netdev_priv(dev); 1807 1808 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1809 } 1810 1811 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1812 const struct net_device *slave_dev) 1813 { 1814 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1815 } 1816 1817 static int vrf_fill_slave_info(struct sk_buff *skb, 1818 const struct net_device *vrf_dev, 1819 const struct net_device *slave_dev) 1820 { 1821 struct net_vrf *vrf = netdev_priv(vrf_dev); 1822 1823 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1824 return -EMSGSIZE; 1825 1826 return 0; 1827 } 1828 1829 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1830 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1831 }; 1832 1833 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1834 .kind = DRV_NAME, 1835 .priv_size = sizeof(struct net_vrf), 1836 1837 .get_size = vrf_nl_getsize, 1838 .policy = vrf_nl_policy, 1839 .validate = vrf_validate, 1840 .fill_info = vrf_fillinfo, 1841 1842 .get_slave_size = vrf_get_slave_size, 1843 .fill_slave_info = vrf_fill_slave_info, 1844 1845 .newlink = vrf_newlink, 1846 .dellink = vrf_dellink, 1847 .setup = vrf_setup, 1848 .maxtype = IFLA_VRF_MAX, 1849 }; 1850 1851 static int vrf_device_event(struct notifier_block *unused, 1852 unsigned long event, void *ptr) 1853 { 1854 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1855 1856 /* only care about unregister events to drop slave references */ 1857 if (event == NETDEV_UNREGISTER) { 1858 struct net_device *vrf_dev; 1859 1860 if (!netif_is_l3_slave(dev)) 1861 goto out; 1862 1863 vrf_dev = netdev_master_upper_dev_get(dev); 1864 vrf_del_slave(vrf_dev, dev); 1865 } 1866 out: 1867 return NOTIFY_DONE; 1868 } 1869 1870 static struct notifier_block vrf_notifier_block __read_mostly = { 1871 .notifier_call = vrf_device_event, 1872 }; 1873 1874 static int vrf_map_init(struct vrf_map *vmap) 1875 { 1876 spin_lock_init(&vmap->vmap_lock); 1877 hash_init(vmap->ht); 1878 1879 vmap->strict_mode = false; 1880 1881 return 0; 1882 } 1883 1884 #ifdef CONFIG_SYSCTL 1885 static bool vrf_strict_mode(struct vrf_map *vmap) 1886 { 1887 bool strict_mode; 1888 1889 vrf_map_lock(vmap); 1890 strict_mode = vmap->strict_mode; 1891 vrf_map_unlock(vmap); 1892 1893 return strict_mode; 1894 } 1895 1896 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode) 1897 { 1898 bool *cur_mode; 1899 int res = 0; 1900 1901 vrf_map_lock(vmap); 1902 1903 cur_mode = &vmap->strict_mode; 1904 if (*cur_mode == new_mode) 1905 goto unlock; 1906 1907 if (*cur_mode) { 1908 /* disable strict mode */ 1909 *cur_mode = false; 1910 } else { 1911 if (vmap->shared_tables) { 1912 /* we cannot allow strict_mode because there are some 1913 * vrfs that share one or more tables. 1914 */ 1915 res = -EBUSY; 1916 goto unlock; 1917 } 1918 1919 /* no tables are shared among vrfs, so we can go back 1920 * to 1:1 association between a vrf with its table. 1921 */ 1922 *cur_mode = true; 1923 } 1924 1925 unlock: 1926 vrf_map_unlock(vmap); 1927 1928 return res; 1929 } 1930 1931 static int vrf_shared_table_handler(struct ctl_table *table, int write, 1932 void *buffer, size_t *lenp, loff_t *ppos) 1933 { 1934 struct net *net = (struct net *)table->extra1; 1935 struct vrf_map *vmap = netns_vrf_map(net); 1936 int proc_strict_mode = 0; 1937 struct ctl_table tmp = { 1938 .procname = table->procname, 1939 .data = &proc_strict_mode, 1940 .maxlen = sizeof(int), 1941 .mode = table->mode, 1942 .extra1 = SYSCTL_ZERO, 1943 .extra2 = SYSCTL_ONE, 1944 }; 1945 int ret; 1946 1947 if (!write) 1948 proc_strict_mode = vrf_strict_mode(vmap); 1949 1950 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 1951 1952 if (write && ret == 0) 1953 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode); 1954 1955 return ret; 1956 } 1957 1958 static const struct ctl_table vrf_table[] = { 1959 { 1960 .procname = "strict_mode", 1961 .data = NULL, 1962 .maxlen = sizeof(int), 1963 .mode = 0644, 1964 .proc_handler = vrf_shared_table_handler, 1965 /* set by the vrf_netns_init */ 1966 .extra1 = NULL, 1967 }, 1968 { }, 1969 }; 1970 1971 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1972 { 1973 struct ctl_table *table; 1974 1975 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL); 1976 if (!table) 1977 return -ENOMEM; 1978 1979 /* init the extra1 parameter with the reference to current netns */ 1980 table[0].extra1 = net; 1981 1982 nn_vrf->ctl_hdr = register_net_sysctl(net, "net/vrf", table); 1983 if (!nn_vrf->ctl_hdr) { 1984 kfree(table); 1985 return -ENOMEM; 1986 } 1987 1988 return 0; 1989 } 1990 1991 static void vrf_netns_exit_sysctl(struct net *net) 1992 { 1993 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1994 struct ctl_table *table; 1995 1996 table = nn_vrf->ctl_hdr->ctl_table_arg; 1997 unregister_net_sysctl_table(nn_vrf->ctl_hdr); 1998 kfree(table); 1999 } 2000 #else 2001 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 2002 { 2003 return 0; 2004 } 2005 2006 static void vrf_netns_exit_sysctl(struct net *net) 2007 { 2008 } 2009 #endif 2010 2011 /* Initialize per network namespace state */ 2012 static int __net_init vrf_netns_init(struct net *net) 2013 { 2014 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 2015 2016 nn_vrf->add_fib_rules = true; 2017 vrf_map_init(&nn_vrf->vmap); 2018 2019 return vrf_netns_init_sysctl(net, nn_vrf); 2020 } 2021 2022 static void __net_exit vrf_netns_exit(struct net *net) 2023 { 2024 vrf_netns_exit_sysctl(net); 2025 } 2026 2027 static struct pernet_operations vrf_net_ops __net_initdata = { 2028 .init = vrf_netns_init, 2029 .exit = vrf_netns_exit, 2030 .id = &vrf_net_id, 2031 .size = sizeof(struct netns_vrf), 2032 }; 2033 2034 static int __init vrf_init_module(void) 2035 { 2036 int rc; 2037 2038 register_netdevice_notifier(&vrf_notifier_block); 2039 2040 rc = register_pernet_subsys(&vrf_net_ops); 2041 if (rc < 0) 2042 goto error; 2043 2044 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF, 2045 vrf_ifindex_lookup_by_table_id); 2046 if (rc < 0) 2047 goto unreg_pernet; 2048 2049 rc = rtnl_link_register(&vrf_link_ops); 2050 if (rc < 0) 2051 goto table_lookup_unreg; 2052 2053 return 0; 2054 2055 table_lookup_unreg: 2056 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF, 2057 vrf_ifindex_lookup_by_table_id); 2058 2059 unreg_pernet: 2060 unregister_pernet_subsys(&vrf_net_ops); 2061 2062 error: 2063 unregister_netdevice_notifier(&vrf_notifier_block); 2064 return rc; 2065 } 2066 2067 module_init(vrf_init_module); 2068 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 2069 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 2070 MODULE_LICENSE("GPL"); 2071 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 2072 MODULE_VERSION(DRV_VERSION); 2073