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 49struct 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 92struct 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 101static unsigned int vrf_net_id; 102 103/* per netns vrf data */ 104struct 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 112struct 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 124static void vrf_rx_stats(struct net_device *dev, int len) 125{ 126 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 127 128 u64_stats_update_begin(&dstats->syncp); 129 dstats->rx_packets++; 130 dstats->rx_bytes += len; 131 u64_stats_update_end(&dstats->syncp); 132} 133 134static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 135{ 136 vrf_dev->stats.tx_errors++; 137 kfree_skb(skb); 138} 139 140static void vrf_get_stats64(struct net_device *dev, 141 struct rtnl_link_stats64 *stats) 142{ 143 int i; 144 145 for_each_possible_cpu(i) { 146 const struct pcpu_dstats *dstats; 147 u64 tbytes, tpkts, tdrops, rbytes, rpkts; 148 unsigned int start; 149 150 dstats = per_cpu_ptr(dev->dstats, i); 151 do { 152 start = u64_stats_fetch_begin(&dstats->syncp); 153 tbytes = dstats->tx_bytes; 154 tpkts = dstats->tx_packets; 155 tdrops = dstats->tx_drops; 156 rbytes = dstats->rx_bytes; 157 rpkts = dstats->rx_packets; 158 } while (u64_stats_fetch_retry(&dstats->syncp, start)); 159 stats->tx_bytes += tbytes; 160 stats->tx_packets += tpkts; 161 stats->tx_dropped += tdrops; 162 stats->rx_bytes += rbytes; 163 stats->rx_packets += rpkts; 164 } 165} 166 167static struct vrf_map *netns_vrf_map(struct net *net) 168{ 169 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 170 171 return &nn_vrf->vmap; 172} 173 174static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev) 175{ 176 return netns_vrf_map(dev_net(dev)); 177} 178 179static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me) 180{ 181 struct list_head *me_head = &me->vrf_list; 182 struct net_vrf *vrf; 183 184 if (list_empty(me_head)) 185 return -ENODEV; 186 187 vrf = list_first_entry(me_head, struct net_vrf, me_list); 188 189 return vrf->ifindex; 190} 191 192static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags) 193{ 194 struct vrf_map_elem *me; 195 196 me = kmalloc(sizeof(*me), flags); 197 if (!me) 198 return NULL; 199 200 return me; 201} 202 203static void vrf_map_elem_free(struct vrf_map_elem *me) 204{ 205 kfree(me); 206} 207 208static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id, 209 int ifindex, int users) 210{ 211 me->table_id = table_id; 212 me->ifindex = ifindex; 213 me->users = users; 214 INIT_LIST_HEAD(&me->vrf_list); 215} 216 217static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap, 218 u32 table_id) 219{ 220 struct vrf_map_elem *me; 221 u32 key; 222 223 key = jhash_1word(table_id, HASH_INITVAL); 224 hash_for_each_possible(vmap->ht, me, hnode, key) { 225 if (me->table_id == table_id) 226 return me; 227 } 228 229 return NULL; 230} 231 232static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me) 233{ 234 u32 table_id = me->table_id; 235 u32 key; 236 237 key = jhash_1word(table_id, HASH_INITVAL); 238 hash_add(vmap->ht, &me->hnode, key); 239} 240 241static void vrf_map_del_elem(struct vrf_map_elem *me) 242{ 243 hash_del(&me->hnode); 244} 245 246static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock) 247{ 248 spin_lock(&vmap->vmap_lock); 249} 250 251static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock) 252{ 253 spin_unlock(&vmap->vmap_lock); 254} 255 256/* called with rtnl lock held */ 257static int 258vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack) 259{ 260 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 261 struct net_vrf *vrf = netdev_priv(dev); 262 struct vrf_map_elem *new_me, *me; 263 u32 table_id = vrf->tb_id; 264 bool free_new_me = false; 265 int users; 266 int res; 267 268 /* we pre-allocate elements used in the spin-locked section (so that we 269 * keep the spinlock as short as possible). 270 */ 271 new_me = vrf_map_elem_alloc(GFP_KERNEL); 272 if (!new_me) 273 return -ENOMEM; 274 275 vrf_map_elem_init(new_me, table_id, dev->ifindex, 0); 276 277 vrf_map_lock(vmap); 278 279 me = vrf_map_lookup_elem(vmap, table_id); 280 if (!me) { 281 me = new_me; 282 vrf_map_add_elem(vmap, me); 283 goto link_vrf; 284 } 285 286 /* we already have an entry in the vrf_map, so it means there is (at 287 * least) a vrf registered on the specific table. 288 */ 289 free_new_me = true; 290 if (vmap->strict_mode) { 291 /* vrfs cannot share the same table */ 292 NL_SET_ERR_MSG(extack, "Table is used by another VRF"); 293 res = -EBUSY; 294 goto unlock; 295 } 296 297link_vrf: 298 users = ++me->users; 299 if (users == 2) 300 ++vmap->shared_tables; 301 302 list_add(&vrf->me_list, &me->vrf_list); 303 304 res = 0; 305 306unlock: 307 vrf_map_unlock(vmap); 308 309 /* clean-up, if needed */ 310 if (free_new_me) 311 vrf_map_elem_free(new_me); 312 313 return res; 314} 315 316/* called with rtnl lock held */ 317static void vrf_map_unregister_dev(struct net_device *dev) 318{ 319 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 320 struct net_vrf *vrf = netdev_priv(dev); 321 u32 table_id = vrf->tb_id; 322 struct vrf_map_elem *me; 323 int users; 324 325 vrf_map_lock(vmap); 326 327 me = vrf_map_lookup_elem(vmap, table_id); 328 if (!me) 329 goto unlock; 330 331 list_del(&vrf->me_list); 332 333 users = --me->users; 334 if (users == 1) { 335 --vmap->shared_tables; 336 } else if (users == 0) { 337 vrf_map_del_elem(me); 338 339 /* no one will refer to this element anymore */ 340 vrf_map_elem_free(me); 341 } 342 343unlock: 344 vrf_map_unlock(vmap); 345} 346 347/* return the vrf device index associated with the table_id */ 348static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id) 349{ 350 struct vrf_map *vmap = netns_vrf_map(net); 351 struct vrf_map_elem *me; 352 int ifindex; 353 354 vrf_map_lock(vmap); 355 356 if (!vmap->strict_mode) { 357 ifindex = -EPERM; 358 goto unlock; 359 } 360 361 me = vrf_map_lookup_elem(vmap, table_id); 362 if (!me) { 363 ifindex = -ENODEV; 364 goto unlock; 365 } 366 367 ifindex = vrf_map_elem_get_vrf_ifindex(me); 368 369unlock: 370 vrf_map_unlock(vmap); 371 372 return ifindex; 373} 374 375/* by default VRF devices do not have a qdisc and are expected 376 * to be created with only a single queue. 377 */ 378static bool qdisc_tx_is_default(const struct net_device *dev) 379{ 380 struct netdev_queue *txq; 381 struct Qdisc *qdisc; 382 383 if (dev->num_tx_queues > 1) 384 return false; 385 386 txq = netdev_get_tx_queue(dev, 0); 387 qdisc = rcu_access_pointer(txq->qdisc); 388 389 return !qdisc->enqueue; 390} 391 392/* Local traffic destined to local address. Reinsert the packet to rx 393 * path, similar to loopback handling. 394 */ 395static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 396 struct dst_entry *dst) 397{ 398 int len = skb->len; 399 400 skb_orphan(skb); 401 402 skb_dst_set(skb, dst); 403 404 /* set pkt_type to avoid skb hitting packet taps twice - 405 * once on Tx and again in Rx processing 406 */ 407 skb->pkt_type = PACKET_LOOPBACK; 408 409 skb->protocol = eth_type_trans(skb, dev); 410 411 if (likely(__netif_rx(skb) == NET_RX_SUCCESS)) 412 vrf_rx_stats(dev, len); 413 else 414 this_cpu_inc(dev->dstats->rx_drops); 415 416 return NETDEV_TX_OK; 417} 418 419static void vrf_nf_set_untracked(struct sk_buff *skb) 420{ 421 if (skb_get_nfct(skb) == 0) 422 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 423} 424 425static void vrf_nf_reset_ct(struct sk_buff *skb) 426{ 427 if (skb_get_nfct(skb) == IP_CT_UNTRACKED) 428 nf_reset_ct(skb); 429} 430 431#if IS_ENABLED(CONFIG_IPV6) 432static int vrf_ip6_local_out(struct net *net, struct sock *sk, 433 struct sk_buff *skb) 434{ 435 int err; 436 437 vrf_nf_reset_ct(skb); 438 439 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, 440 sk, skb, NULL, skb_dst(skb)->dev, dst_output); 441 442 if (likely(err == 1)) 443 err = dst_output(net, sk, skb); 444 445 return err; 446} 447 448static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 449 struct net_device *dev) 450{ 451 const struct ipv6hdr *iph; 452 struct net *net = dev_net(skb->dev); 453 struct flowi6 fl6; 454 int ret = NET_XMIT_DROP; 455 struct dst_entry *dst; 456 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 457 458 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr))) 459 goto err; 460 461 iph = ipv6_hdr(skb); 462 463 memset(&fl6, 0, sizeof(fl6)); 464 /* needed to match OIF rule */ 465 fl6.flowi6_l3mdev = dev->ifindex; 466 fl6.flowi6_iif = LOOPBACK_IFINDEX; 467 fl6.daddr = iph->daddr; 468 fl6.saddr = iph->saddr; 469 fl6.flowlabel = ip6_flowinfo(iph); 470 fl6.flowi6_mark = skb->mark; 471 fl6.flowi6_proto = iph->nexthdr; 472 473 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL); 474 if (IS_ERR(dst) || dst == dst_null) 475 goto err; 476 477 skb_dst_drop(skb); 478 479 /* if dst.dev is the VRF device again this is locally originated traffic 480 * destined to a local address. Short circuit to Rx path. 481 */ 482 if (dst->dev == dev) 483 return vrf_local_xmit(skb, dev, dst); 484 485 skb_dst_set(skb, dst); 486 487 /* strip the ethernet header added for pass through VRF device */ 488 __skb_pull(skb, skb_network_offset(skb)); 489 490 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb))); 491 ret = vrf_ip6_local_out(net, skb->sk, skb); 492 if (unlikely(net_xmit_eval(ret))) 493 dev->stats.tx_errors++; 494 else 495 ret = NET_XMIT_SUCCESS; 496 497 return ret; 498err: 499 vrf_tx_error(dev, skb); 500 return NET_XMIT_DROP; 501} 502#else 503static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 504 struct net_device *dev) 505{ 506 vrf_tx_error(dev, skb); 507 return NET_XMIT_DROP; 508} 509#endif 510 511/* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 512static int vrf_ip_local_out(struct net *net, struct sock *sk, 513 struct sk_buff *skb) 514{ 515 int err; 516 517 vrf_nf_reset_ct(skb); 518 519 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 520 skb, NULL, skb_dst(skb)->dev, dst_output); 521 if (likely(err == 1)) 522 err = dst_output(net, sk, skb); 523 524 return err; 525} 526 527static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 528 struct net_device *vrf_dev) 529{ 530 struct iphdr *ip4h; 531 int ret = NET_XMIT_DROP; 532 struct flowi4 fl4; 533 struct net *net = dev_net(vrf_dev); 534 struct rtable *rt; 535 536 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr))) 537 goto err; 538 539 ip4h = ip_hdr(skb); 540 541 memset(&fl4, 0, sizeof(fl4)); 542 /* needed to match OIF rule */ 543 fl4.flowi4_l3mdev = vrf_dev->ifindex; 544 fl4.flowi4_iif = LOOPBACK_IFINDEX; 545 fl4.flowi4_tos = RT_TOS(ip4h->tos); 546 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC; 547 fl4.flowi4_proto = ip4h->protocol; 548 fl4.daddr = ip4h->daddr; 549 fl4.saddr = ip4h->saddr; 550 551 rt = ip_route_output_flow(net, &fl4, NULL); 552 if (IS_ERR(rt)) 553 goto err; 554 555 skb_dst_drop(skb); 556 557 /* if dst.dev is the VRF device again this is locally originated traffic 558 * destined to a local address. Short circuit to Rx path. 559 */ 560 if (rt->dst.dev == vrf_dev) 561 return vrf_local_xmit(skb, vrf_dev, &rt->dst); 562 563 skb_dst_set(skb, &rt->dst); 564 565 /* strip the ethernet header added for pass through VRF device */ 566 __skb_pull(skb, skb_network_offset(skb)); 567 568 if (!ip4h->saddr) { 569 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 570 RT_SCOPE_LINK); 571 } 572 573 memset(IPCB(skb), 0, sizeof(*IPCB(skb))); 574 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 575 if (unlikely(net_xmit_eval(ret))) 576 vrf_dev->stats.tx_errors++; 577 else 578 ret = NET_XMIT_SUCCESS; 579 580out: 581 return ret; 582err: 583 vrf_tx_error(vrf_dev, skb); 584 goto out; 585} 586 587static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 588{ 589 switch (skb->protocol) { 590 case htons(ETH_P_IP): 591 return vrf_process_v4_outbound(skb, dev); 592 case htons(ETH_P_IPV6): 593 return vrf_process_v6_outbound(skb, dev); 594 default: 595 vrf_tx_error(dev, skb); 596 return NET_XMIT_DROP; 597 } 598} 599 600static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 601{ 602 int len = skb->len; 603 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 604 605 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 606 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 607 608 u64_stats_update_begin(&dstats->syncp); 609 dstats->tx_packets++; 610 dstats->tx_bytes += len; 611 u64_stats_update_end(&dstats->syncp); 612 } else { 613 this_cpu_inc(dev->dstats->tx_drops); 614 } 615 616 return ret; 617} 618 619static void vrf_finish_direct(struct sk_buff *skb) 620{ 621 struct net_device *vrf_dev = skb->dev; 622 623 if (!list_empty(&vrf_dev->ptype_all) && 624 likely(skb_headroom(skb) >= ETH_HLEN)) { 625 struct ethhdr *eth = skb_push(skb, ETH_HLEN); 626 627 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 628 eth_zero_addr(eth->h_dest); 629 eth->h_proto = skb->protocol; 630 631 dev_queue_xmit_nit(skb, vrf_dev); 632 633 skb_pull(skb, ETH_HLEN); 634 } 635 636 vrf_nf_reset_ct(skb); 637} 638 639#if IS_ENABLED(CONFIG_IPV6) 640/* modelled after ip6_finish_output2 */ 641static int vrf_finish_output6(struct net *net, struct sock *sk, 642 struct sk_buff *skb) 643{ 644 struct dst_entry *dst = skb_dst(skb); 645 struct net_device *dev = dst->dev; 646 const struct in6_addr *nexthop; 647 struct neighbour *neigh; 648 int ret; 649 650 vrf_nf_reset_ct(skb); 651 652 skb->protocol = htons(ETH_P_IPV6); 653 skb->dev = dev; 654 655 rcu_read_lock(); 656 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 657 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 658 if (unlikely(!neigh)) 659 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 660 if (!IS_ERR(neigh)) { 661 sock_confirm_neigh(skb, neigh); 662 ret = neigh_output(neigh, skb, false); 663 rcu_read_unlock(); 664 return ret; 665 } 666 rcu_read_unlock(); 667 668 IP6_INC_STATS(dev_net(dst->dev), 669 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 670 kfree_skb(skb); 671 return -EINVAL; 672} 673 674/* modelled after ip6_output */ 675static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 676{ 677 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 678 net, sk, skb, NULL, skb_dst(skb)->dev, 679 vrf_finish_output6, 680 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 681} 682 683/* set dst on skb to send packet to us via dev_xmit path. Allows 684 * packet to go through device based features such as qdisc, netfilter 685 * hooks and packet sockets with skb->dev set to vrf device. 686 */ 687static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev, 688 struct sk_buff *skb) 689{ 690 struct net_vrf *vrf = netdev_priv(vrf_dev); 691 struct dst_entry *dst = NULL; 692 struct rt6_info *rt6; 693 694 rcu_read_lock(); 695 696 rt6 = rcu_dereference(vrf->rt6); 697 if (likely(rt6)) { 698 dst = &rt6->dst; 699 dst_hold(dst); 700 } 701 702 rcu_read_unlock(); 703 704 if (unlikely(!dst)) { 705 vrf_tx_error(vrf_dev, skb); 706 return NULL; 707 } 708 709 skb_dst_drop(skb); 710 skb_dst_set(skb, dst); 711 712 return skb; 713} 714 715static int vrf_output6_direct_finish(struct net *net, struct sock *sk, 716 struct sk_buff *skb) 717{ 718 vrf_finish_direct(skb); 719 720 return vrf_ip6_local_out(net, sk, skb); 721} 722 723static int vrf_output6_direct(struct net *net, struct sock *sk, 724 struct sk_buff *skb) 725{ 726 int err = 1; 727 728 skb->protocol = htons(ETH_P_IPV6); 729 730 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 731 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb, 732 NULL, skb->dev, vrf_output6_direct_finish); 733 734 if (likely(err == 1)) 735 vrf_finish_direct(skb); 736 737 return err; 738} 739 740static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk, 741 struct sk_buff *skb) 742{ 743 int err; 744 745 err = vrf_output6_direct(net, sk, skb); 746 if (likely(err == 1)) 747 err = vrf_ip6_local_out(net, sk, skb); 748 749 return err; 750} 751 752static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev, 753 struct sock *sk, 754 struct sk_buff *skb) 755{ 756 struct net *net = dev_net(vrf_dev); 757 int err; 758 759 skb->dev = vrf_dev; 760 761 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, 762 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish); 763 764 if (likely(err == 1)) 765 err = vrf_output6_direct(net, sk, skb); 766 767 if (likely(err == 1)) 768 return skb; 769 770 return NULL; 771} 772 773static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 774 struct sock *sk, 775 struct sk_buff *skb) 776{ 777 /* don't divert link scope packets */ 778 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 779 return skb; 780 781 vrf_nf_set_untracked(skb); 782 783 if (qdisc_tx_is_default(vrf_dev) || 784 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED) 785 return vrf_ip6_out_direct(vrf_dev, sk, skb); 786 787 return vrf_ip6_out_redirect(vrf_dev, skb); 788} 789 790/* holding rtnl */ 791static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 792{ 793 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 794 struct net *net = dev_net(dev); 795 struct dst_entry *dst; 796 797 RCU_INIT_POINTER(vrf->rt6, NULL); 798 synchronize_rcu(); 799 800 /* move dev in dst's to loopback so this VRF device can be deleted 801 * - based on dst_ifdown 802 */ 803 if (rt6) { 804 dst = &rt6->dst; 805 netdev_ref_replace(dst->dev, net->loopback_dev, 806 &dst->dev_tracker, GFP_KERNEL); 807 dst->dev = net->loopback_dev; 808 dst_release(dst); 809 } 810} 811 812static int vrf_rt6_create(struct net_device *dev) 813{ 814 int flags = DST_NOPOLICY | DST_NOXFRM; 815 struct net_vrf *vrf = netdev_priv(dev); 816 struct net *net = dev_net(dev); 817 struct rt6_info *rt6; 818 int rc = -ENOMEM; 819 820 /* IPv6 can be CONFIG enabled and then disabled runtime */ 821 if (!ipv6_mod_enabled()) 822 return 0; 823 824 vrf->fib6_table = fib6_new_table(net, vrf->tb_id); 825 if (!vrf->fib6_table) 826 goto out; 827 828 /* create a dst for routing packets out a VRF device */ 829 rt6 = ip6_dst_alloc(net, dev, flags); 830 if (!rt6) 831 goto out; 832 833 rt6->dst.output = vrf_output6; 834 835 rcu_assign_pointer(vrf->rt6, rt6); 836 837 rc = 0; 838out: 839 return rc; 840} 841#else 842static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 843 struct sock *sk, 844 struct sk_buff *skb) 845{ 846 return skb; 847} 848 849static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 850{ 851} 852 853static int vrf_rt6_create(struct net_device *dev) 854{ 855 return 0; 856} 857#endif 858 859/* modelled after ip_finish_output2 */ 860static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 861{ 862 struct dst_entry *dst = skb_dst(skb); 863 struct rtable *rt = (struct rtable *)dst; 864 struct net_device *dev = dst->dev; 865 unsigned int hh_len = LL_RESERVED_SPACE(dev); 866 struct neighbour *neigh; 867 bool is_v6gw = false; 868 869 vrf_nf_reset_ct(skb); 870 871 /* Be paranoid, rather than too clever. */ 872 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 873 skb = skb_expand_head(skb, hh_len); 874 if (!skb) { 875 dev->stats.tx_errors++; 876 return -ENOMEM; 877 } 878 } 879 880 rcu_read_lock(); 881 882 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 883 if (!IS_ERR(neigh)) { 884 int ret; 885 886 sock_confirm_neigh(skb, neigh); 887 /* if crossing protocols, can not use the cached header */ 888 ret = neigh_output(neigh, skb, is_v6gw); 889 rcu_read_unlock(); 890 return ret; 891 } 892 893 rcu_read_unlock(); 894 vrf_tx_error(skb->dev, skb); 895 return -EINVAL; 896} 897 898static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 899{ 900 struct net_device *dev = skb_dst(skb)->dev; 901 902 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 903 904 skb->dev = dev; 905 skb->protocol = htons(ETH_P_IP); 906 907 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 908 net, sk, skb, NULL, dev, 909 vrf_finish_output, 910 !(IPCB(skb)->flags & IPSKB_REROUTED)); 911} 912 913/* set dst on skb to send packet to us via dev_xmit path. Allows 914 * packet to go through device based features such as qdisc, netfilter 915 * hooks and packet sockets with skb->dev set to vrf device. 916 */ 917static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev, 918 struct sk_buff *skb) 919{ 920 struct net_vrf *vrf = netdev_priv(vrf_dev); 921 struct dst_entry *dst = NULL; 922 struct rtable *rth; 923 924 rcu_read_lock(); 925 926 rth = rcu_dereference(vrf->rth); 927 if (likely(rth)) { 928 dst = &rth->dst; 929 dst_hold(dst); 930 } 931 932 rcu_read_unlock(); 933 934 if (unlikely(!dst)) { 935 vrf_tx_error(vrf_dev, skb); 936 return NULL; 937 } 938 939 skb_dst_drop(skb); 940 skb_dst_set(skb, dst); 941 942 return skb; 943} 944 945static int vrf_output_direct_finish(struct net *net, struct sock *sk, 946 struct sk_buff *skb) 947{ 948 vrf_finish_direct(skb); 949 950 return vrf_ip_local_out(net, sk, skb); 951} 952 953static int vrf_output_direct(struct net *net, struct sock *sk, 954 struct sk_buff *skb) 955{ 956 int err = 1; 957 958 skb->protocol = htons(ETH_P_IP); 959 960 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 961 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb, 962 NULL, skb->dev, vrf_output_direct_finish); 963 964 if (likely(err == 1)) 965 vrf_finish_direct(skb); 966 967 return err; 968} 969 970static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk, 971 struct sk_buff *skb) 972{ 973 int err; 974 975 err = vrf_output_direct(net, sk, skb); 976 if (likely(err == 1)) 977 err = vrf_ip_local_out(net, sk, skb); 978 979 return err; 980} 981 982static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev, 983 struct sock *sk, 984 struct sk_buff *skb) 985{ 986 struct net *net = dev_net(vrf_dev); 987 int err; 988 989 skb->dev = vrf_dev; 990 991 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 992 skb, NULL, vrf_dev, vrf_ip_out_direct_finish); 993 994 if (likely(err == 1)) 995 err = vrf_output_direct(net, sk, skb); 996 997 if (likely(err == 1)) 998 return skb; 999 1000 return NULL; 1001} 1002 1003static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 1004 struct sock *sk, 1005 struct sk_buff *skb) 1006{ 1007 /* don't divert multicast or local broadcast */ 1008 if (ipv4_is_multicast(ip_hdr(skb)->daddr) || 1009 ipv4_is_lbcast(ip_hdr(skb)->daddr)) 1010 return skb; 1011 1012 vrf_nf_set_untracked(skb); 1013 1014 if (qdisc_tx_is_default(vrf_dev) || 1015 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED) 1016 return vrf_ip_out_direct(vrf_dev, sk, skb); 1017 1018 return vrf_ip_out_redirect(vrf_dev, skb); 1019} 1020 1021/* called with rcu lock held */ 1022static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 1023 struct sock *sk, 1024 struct sk_buff *skb, 1025 u16 proto) 1026{ 1027 switch (proto) { 1028 case AF_INET: 1029 return vrf_ip_out(vrf_dev, sk, skb); 1030 case AF_INET6: 1031 return vrf_ip6_out(vrf_dev, sk, skb); 1032 } 1033 1034 return skb; 1035} 1036 1037/* holding rtnl */ 1038static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 1039{ 1040 struct rtable *rth = rtnl_dereference(vrf->rth); 1041 struct net *net = dev_net(dev); 1042 struct dst_entry *dst; 1043 1044 RCU_INIT_POINTER(vrf->rth, NULL); 1045 synchronize_rcu(); 1046 1047 /* move dev in dst's to loopback so this VRF device can be deleted 1048 * - based on dst_ifdown 1049 */ 1050 if (rth) { 1051 dst = &rth->dst; 1052 netdev_ref_replace(dst->dev, net->loopback_dev, 1053 &dst->dev_tracker, GFP_KERNEL); 1054 dst->dev = net->loopback_dev; 1055 dst_release(dst); 1056 } 1057} 1058 1059static int vrf_rtable_create(struct net_device *dev) 1060{ 1061 struct net_vrf *vrf = netdev_priv(dev); 1062 struct rtable *rth; 1063 1064 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 1065 return -ENOMEM; 1066 1067 /* create a dst for routing packets out through a VRF device */ 1068 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1); 1069 if (!rth) 1070 return -ENOMEM; 1071 1072 rth->dst.output = vrf_output; 1073 1074 rcu_assign_pointer(vrf->rth, rth); 1075 1076 return 0; 1077} 1078 1079/**************************** device handling ********************/ 1080 1081/* cycle interface to flush neighbor cache and move routes across tables */ 1082static void cycle_netdev(struct net_device *dev, 1083 struct netlink_ext_ack *extack) 1084{ 1085 unsigned int flags = dev->flags; 1086 int ret; 1087 1088 if (!netif_running(dev)) 1089 return; 1090 1091 ret = dev_change_flags(dev, flags & ~IFF_UP, extack); 1092 if (ret >= 0) 1093 ret = dev_change_flags(dev, flags, extack); 1094 1095 if (ret < 0) { 1096 netdev_err(dev, 1097 "Failed to cycle device %s; route tables might be wrong!\n", 1098 dev->name); 1099 } 1100} 1101 1102static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1103 struct netlink_ext_ack *extack) 1104{ 1105 int ret; 1106 1107 /* do not allow loopback device to be enslaved to a VRF. 1108 * The vrf device acts as the loopback for the vrf. 1109 */ 1110 if (port_dev == dev_net(dev)->loopback_dev) { 1111 NL_SET_ERR_MSG(extack, 1112 "Can not enslave loopback device to a VRF"); 1113 return -EOPNOTSUPP; 1114 } 1115 1116 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 1117 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack); 1118 if (ret < 0) 1119 goto err; 1120 1121 cycle_netdev(port_dev, extack); 1122 1123 return 0; 1124 1125err: 1126 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1127 return ret; 1128} 1129 1130static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1131 struct netlink_ext_ack *extack) 1132{ 1133 if (netif_is_l3_master(port_dev)) { 1134 NL_SET_ERR_MSG(extack, 1135 "Can not enslave an L3 master device to a VRF"); 1136 return -EINVAL; 1137 } 1138 1139 if (netif_is_l3_slave(port_dev)) 1140 return -EINVAL; 1141 1142 return do_vrf_add_slave(dev, port_dev, extack); 1143} 1144 1145/* inverse of do_vrf_add_slave */ 1146static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1147{ 1148 netdev_upper_dev_unlink(port_dev, dev); 1149 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1150 1151 cycle_netdev(port_dev, NULL); 1152 1153 return 0; 1154} 1155 1156static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1157{ 1158 return do_vrf_del_slave(dev, port_dev); 1159} 1160 1161static void vrf_dev_uninit(struct net_device *dev) 1162{ 1163 struct net_vrf *vrf = netdev_priv(dev); 1164 1165 vrf_rtable_release(dev, vrf); 1166 vrf_rt6_release(dev, vrf); 1167} 1168 1169static int vrf_dev_init(struct net_device *dev) 1170{ 1171 struct net_vrf *vrf = netdev_priv(dev); 1172 1173 /* create the default dst which points back to us */ 1174 if (vrf_rtable_create(dev) != 0) 1175 goto out_nomem; 1176 1177 if (vrf_rt6_create(dev) != 0) 1178 goto out_rth; 1179 1180 dev->flags = IFF_MASTER | IFF_NOARP; 1181 1182 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 1183 dev->operstate = IF_OPER_UP; 1184 netdev_lockdep_set_classes(dev); 1185 return 0; 1186 1187out_rth: 1188 vrf_rtable_release(dev, vrf); 1189out_nomem: 1190 return -ENOMEM; 1191} 1192 1193static 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 1203static 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 1210static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 1211{ 1212 kfree_skb(skb); 1213 return 0; 1214} 1215 1216static 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 1228static 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 skb_reset_mac_len(skb); 1247 1248 /* we set the ethernet destination and the source addresses to the 1249 * address of the VRF device. 1250 */ 1251 ether_addr_copy(eth->h_dest, vrf_dev->dev_addr); 1252 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 1253 eth->h_proto = htons(proto); 1254 1255 /* the destination address of the Ethernet frame corresponds to the 1256 * address set on the VRF interface; therefore, the packet is intended 1257 * to be processed locally. 1258 */ 1259 skb->protocol = eth->h_proto; 1260 skb->pkt_type = PACKET_HOST; 1261 1262 skb_postpush_rcsum(skb, skb->data, ETH_HLEN); 1263 1264 skb_pull_inline(skb, ETH_HLEN); 1265 1266 return 0; 1267} 1268 1269/* prepare and add the mac header to the packet if it was not set previously. 1270 * In this way, packet sniffers such as tcpdump can parse the packet correctly. 1271 * If the mac header was already set, the original mac header is left 1272 * untouched and the function returns immediately. 1273 */ 1274static int vrf_add_mac_header_if_unset(struct sk_buff *skb, 1275 struct net_device *vrf_dev, 1276 u16 proto, struct net_device *orig_dev) 1277{ 1278 if (skb_mac_header_was_set(skb) && dev_has_header(orig_dev)) 1279 return 0; 1280 1281 return vrf_prepare_mac_header(skb, vrf_dev, proto); 1282} 1283 1284#if IS_ENABLED(CONFIG_IPV6) 1285/* neighbor handling is done with actual device; do not want 1286 * to flip skb->dev for those ndisc packets. This really fails 1287 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 1288 * a start. 1289 */ 1290static bool ipv6_ndisc_frame(const struct sk_buff *skb) 1291{ 1292 const struct ipv6hdr *iph = ipv6_hdr(skb); 1293 bool rc = false; 1294 1295 if (iph->nexthdr == NEXTHDR_ICMP) { 1296 const struct icmp6hdr *icmph; 1297 struct icmp6hdr _icmph; 1298 1299 icmph = skb_header_pointer(skb, sizeof(*iph), 1300 sizeof(_icmph), &_icmph); 1301 if (!icmph) 1302 goto out; 1303 1304 switch (icmph->icmp6_type) { 1305 case NDISC_ROUTER_SOLICITATION: 1306 case NDISC_ROUTER_ADVERTISEMENT: 1307 case NDISC_NEIGHBOUR_SOLICITATION: 1308 case NDISC_NEIGHBOUR_ADVERTISEMENT: 1309 case NDISC_REDIRECT: 1310 rc = true; 1311 break; 1312 } 1313 } 1314 1315out: 1316 return rc; 1317} 1318 1319static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 1320 const struct net_device *dev, 1321 struct flowi6 *fl6, 1322 int ifindex, 1323 const struct sk_buff *skb, 1324 int flags) 1325{ 1326 struct net_vrf *vrf = netdev_priv(dev); 1327 1328 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags); 1329} 1330 1331static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 1332 int ifindex) 1333{ 1334 const struct ipv6hdr *iph = ipv6_hdr(skb); 1335 struct flowi6 fl6 = { 1336 .flowi6_iif = ifindex, 1337 .flowi6_mark = skb->mark, 1338 .flowi6_proto = iph->nexthdr, 1339 .daddr = iph->daddr, 1340 .saddr = iph->saddr, 1341 .flowlabel = ip6_flowinfo(iph), 1342 }; 1343 struct net *net = dev_net(vrf_dev); 1344 struct rt6_info *rt6; 1345 1346 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb, 1347 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 1348 if (unlikely(!rt6)) 1349 return; 1350 1351 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 1352 return; 1353 1354 skb_dst_set(skb, &rt6->dst); 1355} 1356 1357static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1358 struct sk_buff *skb) 1359{ 1360 int orig_iif = skb->skb_iif; 1361 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 1362 bool is_ndisc = ipv6_ndisc_frame(skb); 1363 1364 /* loopback, multicast & non-ND link-local traffic; do not push through 1365 * packet taps again. Reset pkt_type for upper layers to process skb. 1366 * For non-loopback strict packets, determine the dst using the original 1367 * ifindex. 1368 */ 1369 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) { 1370 skb->dev = vrf_dev; 1371 skb->skb_iif = vrf_dev->ifindex; 1372 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1373 1374 if (skb->pkt_type == PACKET_LOOPBACK) 1375 skb->pkt_type = PACKET_HOST; 1376 else 1377 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1378 1379 goto out; 1380 } 1381 1382 /* if packet is NDISC then keep the ingress interface */ 1383 if (!is_ndisc) { 1384 struct net_device *orig_dev = skb->dev; 1385 1386 vrf_rx_stats(vrf_dev, skb->len); 1387 skb->dev = vrf_dev; 1388 skb->skb_iif = vrf_dev->ifindex; 1389 1390 if (!list_empty(&vrf_dev->ptype_all)) { 1391 int err; 1392 1393 err = vrf_add_mac_header_if_unset(skb, vrf_dev, 1394 ETH_P_IPV6, 1395 orig_dev); 1396 if (likely(!err)) { 1397 skb_push(skb, skb->mac_len); 1398 dev_queue_xmit_nit(skb, vrf_dev); 1399 skb_pull(skb, skb->mac_len); 1400 } 1401 } 1402 1403 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1404 } 1405 1406 if (need_strict) 1407 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1408 1409 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 1410out: 1411 return skb; 1412} 1413 1414#else 1415static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1416 struct sk_buff *skb) 1417{ 1418 return skb; 1419} 1420#endif 1421 1422static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 1423 struct sk_buff *skb) 1424{ 1425 struct net_device *orig_dev = skb->dev; 1426 1427 skb->dev = vrf_dev; 1428 skb->skb_iif = vrf_dev->ifindex; 1429 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1430 1431 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1432 goto out; 1433 1434 /* loopback traffic; do not push through packet taps again. 1435 * Reset pkt_type for upper layers to process skb 1436 */ 1437 if (skb->pkt_type == PACKET_LOOPBACK) { 1438 skb->pkt_type = PACKET_HOST; 1439 goto out; 1440 } 1441 1442 vrf_rx_stats(vrf_dev, skb->len); 1443 1444 if (!list_empty(&vrf_dev->ptype_all)) { 1445 int err; 1446 1447 err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP, 1448 orig_dev); 1449 if (likely(!err)) { 1450 skb_push(skb, skb->mac_len); 1451 dev_queue_xmit_nit(skb, vrf_dev); 1452 skb_pull(skb, skb->mac_len); 1453 } 1454 } 1455 1456 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1457out: 1458 return skb; 1459} 1460 1461/* called with rcu lock held */ 1462static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1463 struct sk_buff *skb, 1464 u16 proto) 1465{ 1466 switch (proto) { 1467 case AF_INET: 1468 return vrf_ip_rcv(vrf_dev, skb); 1469 case AF_INET6: 1470 return vrf_ip6_rcv(vrf_dev, skb); 1471 } 1472 1473 return skb; 1474} 1475 1476#if IS_ENABLED(CONFIG_IPV6) 1477/* send to link-local or multicast address via interface enslaved to 1478 * VRF device. Force lookup to VRF table without changing flow struct 1479 * Note: Caller to this function must hold rcu_read_lock() and no refcnt 1480 * is taken on the dst by this function. 1481 */ 1482static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1483 struct flowi6 *fl6) 1484{ 1485 struct net *net = dev_net(dev); 1486 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF; 1487 struct dst_entry *dst = NULL; 1488 struct rt6_info *rt; 1489 1490 /* VRF device does not have a link-local address and 1491 * sending packets to link-local or mcast addresses over 1492 * a VRF device does not make sense 1493 */ 1494 if (fl6->flowi6_oif == dev->ifindex) { 1495 dst = &net->ipv6.ip6_null_entry->dst; 1496 return dst; 1497 } 1498 1499 if (!ipv6_addr_any(&fl6->saddr)) 1500 flags |= RT6_LOOKUP_F_HAS_SADDR; 1501 1502 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags); 1503 if (rt) 1504 dst = &rt->dst; 1505 1506 return dst; 1507} 1508#endif 1509 1510static const struct l3mdev_ops vrf_l3mdev_ops = { 1511 .l3mdev_fib_table = vrf_fib_table, 1512 .l3mdev_l3_rcv = vrf_l3_rcv, 1513 .l3mdev_l3_out = vrf_l3_out, 1514#if IS_ENABLED(CONFIG_IPV6) 1515 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1516#endif 1517}; 1518 1519static void vrf_get_drvinfo(struct net_device *dev, 1520 struct ethtool_drvinfo *info) 1521{ 1522 strscpy(info->driver, DRV_NAME, sizeof(info->driver)); 1523 strscpy(info->version, DRV_VERSION, sizeof(info->version)); 1524} 1525 1526static const struct ethtool_ops vrf_ethtool_ops = { 1527 .get_drvinfo = vrf_get_drvinfo, 1528}; 1529 1530static inline size_t vrf_fib_rule_nl_size(void) 1531{ 1532 size_t sz; 1533 1534 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1535 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1536 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1537 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */ 1538 1539 return sz; 1540} 1541 1542static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1543{ 1544 struct fib_rule_hdr *frh; 1545 struct nlmsghdr *nlh; 1546 struct sk_buff *skb; 1547 int err; 1548 1549 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) && 1550 !ipv6_mod_enabled()) 1551 return 0; 1552 1553 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1554 if (!skb) 1555 return -ENOMEM; 1556 1557 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1558 if (!nlh) 1559 goto nla_put_failure; 1560 1561 /* rule only needs to appear once */ 1562 nlh->nlmsg_flags |= NLM_F_EXCL; 1563 1564 frh = nlmsg_data(nlh); 1565 memset(frh, 0, sizeof(*frh)); 1566 frh->family = family; 1567 frh->action = FR_ACT_TO_TBL; 1568 1569 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL)) 1570 goto nla_put_failure; 1571 1572 if (nla_put_u8(skb, FRA_L3MDEV, 1)) 1573 goto nla_put_failure; 1574 1575 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1576 goto nla_put_failure; 1577 1578 nlmsg_end(skb, nlh); 1579 1580 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1581 skb->sk = dev_net(dev)->rtnl; 1582 if (add_it) { 1583 err = fib_nl_newrule(skb, nlh, NULL); 1584 if (err == -EEXIST) 1585 err = 0; 1586 } else { 1587 err = fib_nl_delrule(skb, nlh, NULL); 1588 if (err == -ENOENT) 1589 err = 0; 1590 } 1591 nlmsg_free(skb); 1592 1593 return err; 1594 1595nla_put_failure: 1596 nlmsg_free(skb); 1597 1598 return -EMSGSIZE; 1599} 1600 1601static int vrf_add_fib_rules(const struct net_device *dev) 1602{ 1603 int err; 1604 1605 err = vrf_fib_rule(dev, AF_INET, true); 1606 if (err < 0) 1607 goto out_err; 1608 1609 err = vrf_fib_rule(dev, AF_INET6, true); 1610 if (err < 0) 1611 goto ipv6_err; 1612 1613#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1614 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1615 if (err < 0) 1616 goto ipmr_err; 1617#endif 1618 1619#if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1620 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true); 1621 if (err < 0) 1622 goto ip6mr_err; 1623#endif 1624 1625 return 0; 1626 1627#if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1628ip6mr_err: 1629 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false); 1630#endif 1631 1632#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1633ipmr_err: 1634 vrf_fib_rule(dev, AF_INET6, false); 1635#endif 1636 1637ipv6_err: 1638 vrf_fib_rule(dev, AF_INET, false); 1639 1640out_err: 1641 netdev_err(dev, "Failed to add FIB rules.\n"); 1642 return err; 1643} 1644 1645static void vrf_setup(struct net_device *dev) 1646{ 1647 ether_setup(dev); 1648 1649 /* Initialize the device structure. */ 1650 dev->netdev_ops = &vrf_netdev_ops; 1651 dev->l3mdev_ops = &vrf_l3mdev_ops; 1652 dev->ethtool_ops = &vrf_ethtool_ops; 1653 dev->needs_free_netdev = true; 1654 1655 /* Fill in device structure with ethernet-generic values. */ 1656 eth_hw_addr_random(dev); 1657 1658 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1659 dev->features |= NETIF_F_LLTX; 1660 1661 /* don't allow vrf devices to change network namespaces. */ 1662 dev->features |= NETIF_F_NETNS_LOCAL; 1663 1664 /* does not make sense for a VLAN to be added to a vrf device */ 1665 dev->features |= NETIF_F_VLAN_CHALLENGED; 1666 1667 /* enable offload features */ 1668 dev->features |= NETIF_F_GSO_SOFTWARE; 1669 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC; 1670 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1671 1672 dev->hw_features = dev->features; 1673 dev->hw_enc_features = dev->features; 1674 1675 /* default to no qdisc; user can add if desired */ 1676 dev->priv_flags |= IFF_NO_QUEUE; 1677 dev->priv_flags |= IFF_NO_RX_HANDLER; 1678 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1679 1680 /* VRF devices do not care about MTU, but if the MTU is set 1681 * too low then the ipv4 and ipv6 protocols are disabled 1682 * which breaks networking. 1683 */ 1684 dev->min_mtu = IPV6_MIN_MTU; 1685 dev->max_mtu = IP6_MAX_MTU; 1686 dev->mtu = dev->max_mtu; 1687 1688 dev->pcpu_stat_type = NETDEV_PCPU_STAT_DSTATS; 1689} 1690 1691static int vrf_validate(struct nlattr *tb[], struct nlattr *data[], 1692 struct netlink_ext_ack *extack) 1693{ 1694 if (tb[IFLA_ADDRESS]) { 1695 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1696 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1697 return -EINVAL; 1698 } 1699 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1700 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1701 return -EADDRNOTAVAIL; 1702 } 1703 } 1704 return 0; 1705} 1706 1707static void vrf_dellink(struct net_device *dev, struct list_head *head) 1708{ 1709 struct net_device *port_dev; 1710 struct list_head *iter; 1711 1712 netdev_for_each_lower_dev(dev, port_dev, iter) 1713 vrf_del_slave(dev, port_dev); 1714 1715 vrf_map_unregister_dev(dev); 1716 1717 unregister_netdevice_queue(dev, head); 1718} 1719 1720static int vrf_newlink(struct net *src_net, struct net_device *dev, 1721 struct nlattr *tb[], struct nlattr *data[], 1722 struct netlink_ext_ack *extack) 1723{ 1724 struct net_vrf *vrf = netdev_priv(dev); 1725 struct netns_vrf *nn_vrf; 1726 bool *add_fib_rules; 1727 struct net *net; 1728 int err; 1729 1730 if (!data || !data[IFLA_VRF_TABLE]) { 1731 NL_SET_ERR_MSG(extack, "VRF table id is missing"); 1732 return -EINVAL; 1733 } 1734 1735 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1736 if (vrf->tb_id == RT_TABLE_UNSPEC) { 1737 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE], 1738 "Invalid VRF table id"); 1739 return -EINVAL; 1740 } 1741 1742 dev->priv_flags |= IFF_L3MDEV_MASTER; 1743 1744 err = register_netdevice(dev); 1745 if (err) 1746 goto out; 1747 1748 /* mapping between table_id and vrf; 1749 * note: such binding could not be done in the dev init function 1750 * because dev->ifindex id is not available yet. 1751 */ 1752 vrf->ifindex = dev->ifindex; 1753 1754 err = vrf_map_register_dev(dev, extack); 1755 if (err) { 1756 unregister_netdevice(dev); 1757 goto out; 1758 } 1759 1760 net = dev_net(dev); 1761 nn_vrf = net_generic(net, vrf_net_id); 1762 1763 add_fib_rules = &nn_vrf->add_fib_rules; 1764 if (*add_fib_rules) { 1765 err = vrf_add_fib_rules(dev); 1766 if (err) { 1767 vrf_map_unregister_dev(dev); 1768 unregister_netdevice(dev); 1769 goto out; 1770 } 1771 *add_fib_rules = false; 1772 } 1773 1774out: 1775 return err; 1776} 1777 1778static size_t vrf_nl_getsize(const struct net_device *dev) 1779{ 1780 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1781} 1782 1783static int vrf_fillinfo(struct sk_buff *skb, 1784 const struct net_device *dev) 1785{ 1786 struct net_vrf *vrf = netdev_priv(dev); 1787 1788 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1789} 1790 1791static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1792 const struct net_device *slave_dev) 1793{ 1794 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1795} 1796 1797static int vrf_fill_slave_info(struct sk_buff *skb, 1798 const struct net_device *vrf_dev, 1799 const struct net_device *slave_dev) 1800{ 1801 struct net_vrf *vrf = netdev_priv(vrf_dev); 1802 1803 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1804 return -EMSGSIZE; 1805 1806 return 0; 1807} 1808 1809static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1810 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1811}; 1812 1813static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1814 .kind = DRV_NAME, 1815 .priv_size = sizeof(struct net_vrf), 1816 1817 .get_size = vrf_nl_getsize, 1818 .policy = vrf_nl_policy, 1819 .validate = vrf_validate, 1820 .fill_info = vrf_fillinfo, 1821 1822 .get_slave_size = vrf_get_slave_size, 1823 .fill_slave_info = vrf_fill_slave_info, 1824 1825 .newlink = vrf_newlink, 1826 .dellink = vrf_dellink, 1827 .setup = vrf_setup, 1828 .maxtype = IFLA_VRF_MAX, 1829}; 1830 1831static int vrf_device_event(struct notifier_block *unused, 1832 unsigned long event, void *ptr) 1833{ 1834 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1835 1836 /* only care about unregister events to drop slave references */ 1837 if (event == NETDEV_UNREGISTER) { 1838 struct net_device *vrf_dev; 1839 1840 if (!netif_is_l3_slave(dev)) 1841 goto out; 1842 1843 vrf_dev = netdev_master_upper_dev_get(dev); 1844 vrf_del_slave(vrf_dev, dev); 1845 } 1846out: 1847 return NOTIFY_DONE; 1848} 1849 1850static struct notifier_block vrf_notifier_block __read_mostly = { 1851 .notifier_call = vrf_device_event, 1852}; 1853 1854static int vrf_map_init(struct vrf_map *vmap) 1855{ 1856 spin_lock_init(&vmap->vmap_lock); 1857 hash_init(vmap->ht); 1858 1859 vmap->strict_mode = false; 1860 1861 return 0; 1862} 1863 1864#ifdef CONFIG_SYSCTL 1865static bool vrf_strict_mode(struct vrf_map *vmap) 1866{ 1867 bool strict_mode; 1868 1869 vrf_map_lock(vmap); 1870 strict_mode = vmap->strict_mode; 1871 vrf_map_unlock(vmap); 1872 1873 return strict_mode; 1874} 1875 1876static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode) 1877{ 1878 bool *cur_mode; 1879 int res = 0; 1880 1881 vrf_map_lock(vmap); 1882 1883 cur_mode = &vmap->strict_mode; 1884 if (*cur_mode == new_mode) 1885 goto unlock; 1886 1887 if (*cur_mode) { 1888 /* disable strict mode */ 1889 *cur_mode = false; 1890 } else { 1891 if (vmap->shared_tables) { 1892 /* we cannot allow strict_mode because there are some 1893 * vrfs that share one or more tables. 1894 */ 1895 res = -EBUSY; 1896 goto unlock; 1897 } 1898 1899 /* no tables are shared among vrfs, so we can go back 1900 * to 1:1 association between a vrf with its table. 1901 */ 1902 *cur_mode = true; 1903 } 1904 1905unlock: 1906 vrf_map_unlock(vmap); 1907 1908 return res; 1909} 1910 1911static int vrf_shared_table_handler(struct ctl_table *table, int write, 1912 void *buffer, size_t *lenp, loff_t *ppos) 1913{ 1914 struct net *net = (struct net *)table->extra1; 1915 struct vrf_map *vmap = netns_vrf_map(net); 1916 int proc_strict_mode = 0; 1917 struct ctl_table tmp = { 1918 .procname = table->procname, 1919 .data = &proc_strict_mode, 1920 .maxlen = sizeof(int), 1921 .mode = table->mode, 1922 .extra1 = SYSCTL_ZERO, 1923 .extra2 = SYSCTL_ONE, 1924 }; 1925 int ret; 1926 1927 if (!write) 1928 proc_strict_mode = vrf_strict_mode(vmap); 1929 1930 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 1931 1932 if (write && ret == 0) 1933 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode); 1934 1935 return ret; 1936} 1937 1938static const struct ctl_table vrf_table[] = { 1939 { 1940 .procname = "strict_mode", 1941 .data = NULL, 1942 .maxlen = sizeof(int), 1943 .mode = 0644, 1944 .proc_handler = vrf_shared_table_handler, 1945 /* set by the vrf_netns_init */ 1946 .extra1 = NULL, 1947 }, 1948}; 1949 1950static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1951{ 1952 struct ctl_table *table; 1953 1954 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL); 1955 if (!table) 1956 return -ENOMEM; 1957 1958 /* init the extra1 parameter with the reference to current netns */ 1959 table[0].extra1 = net; 1960 1961 nn_vrf->ctl_hdr = register_net_sysctl_sz(net, "net/vrf", table, 1962 ARRAY_SIZE(vrf_table)); 1963 if (!nn_vrf->ctl_hdr) { 1964 kfree(table); 1965 return -ENOMEM; 1966 } 1967 1968 return 0; 1969} 1970 1971static void vrf_netns_exit_sysctl(struct net *net) 1972{ 1973 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1974 struct ctl_table *table; 1975 1976 table = nn_vrf->ctl_hdr->ctl_table_arg; 1977 unregister_net_sysctl_table(nn_vrf->ctl_hdr); 1978 kfree(table); 1979} 1980#else 1981static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1982{ 1983 return 0; 1984} 1985 1986static void vrf_netns_exit_sysctl(struct net *net) 1987{ 1988} 1989#endif 1990 1991/* Initialize per network namespace state */ 1992static int __net_init vrf_netns_init(struct net *net) 1993{ 1994 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1995 1996 nn_vrf->add_fib_rules = true; 1997 vrf_map_init(&nn_vrf->vmap); 1998 1999 return vrf_netns_init_sysctl(net, nn_vrf); 2000} 2001 2002static void __net_exit vrf_netns_exit(struct net *net) 2003{ 2004 vrf_netns_exit_sysctl(net); 2005} 2006 2007static struct pernet_operations vrf_net_ops __net_initdata = { 2008 .init = vrf_netns_init, 2009 .exit = vrf_netns_exit, 2010 .id = &vrf_net_id, 2011 .size = sizeof(struct netns_vrf), 2012}; 2013 2014static int __init vrf_init_module(void) 2015{ 2016 int rc; 2017 2018 register_netdevice_notifier(&vrf_notifier_block); 2019 2020 rc = register_pernet_subsys(&vrf_net_ops); 2021 if (rc < 0) 2022 goto error; 2023 2024 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF, 2025 vrf_ifindex_lookup_by_table_id); 2026 if (rc < 0) 2027 goto unreg_pernet; 2028 2029 rc = rtnl_link_register(&vrf_link_ops); 2030 if (rc < 0) 2031 goto table_lookup_unreg; 2032 2033 return 0; 2034 2035table_lookup_unreg: 2036 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF, 2037 vrf_ifindex_lookup_by_table_id); 2038 2039unreg_pernet: 2040 unregister_pernet_subsys(&vrf_net_ops); 2041 2042error: 2043 unregister_netdevice_notifier(&vrf_notifier_block); 2044 return rc; 2045} 2046 2047module_init(vrf_init_module); 2048MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 2049MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 2050MODULE_LICENSE("GPL"); 2051MODULE_ALIAS_RTNL_LINK(DRV_NAME); 2052MODULE_VERSION(DRV_VERSION); 2053