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 #include <net/inet_dscp.h> 41 42 #define DRV_NAME "vrf" 43 #define DRV_VERSION "1.1" 44 45 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */ 46 47 #define HT_MAP_BITS 4 48 #define HASH_INITVAL ((u32)0xcafef00d) 49 50 struct vrf_map { 51 DECLARE_HASHTABLE(ht, HT_MAP_BITS); 52 spinlock_t vmap_lock; 53 54 /* shared_tables: 55 * count how many distinct tables do not comply with the strict mode 56 * requirement. 57 * shared_tables value must be 0 in order to enable the strict mode. 58 * 59 * example of the evolution of shared_tables: 60 * | time 61 * add vrf0 --> table 100 shared_tables = 0 | t0 62 * add vrf1 --> table 101 shared_tables = 0 | t1 63 * add vrf2 --> table 100 shared_tables = 1 | t2 64 * add vrf3 --> table 100 shared_tables = 1 | t3 65 * add vrf4 --> table 101 shared_tables = 2 v t4 66 * 67 * shared_tables is a "step function" (or "staircase function") 68 * and it is increased by one when the second vrf is associated to a 69 * table. 70 * 71 * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1. 72 * 73 * at t3, another dev (vrf3) is bound to the same table 100 but the 74 * value of shared_tables is still 1. 75 * This means that no matter how many new vrfs will register on the 76 * table 100, the shared_tables will not increase (considering only 77 * table 100). 78 * 79 * at t4, vrf4 is bound to table 101, and shared_tables = 2. 80 * 81 * Looking at the value of shared_tables we can immediately know if 82 * the strict_mode can or cannot be enforced. Indeed, strict_mode 83 * can be enforced iff shared_tables = 0. 84 * 85 * Conversely, shared_tables is decreased when a vrf is de-associated 86 * from a table with exactly two associated vrfs. 87 */ 88 u32 shared_tables; 89 90 bool strict_mode; 91 }; 92 93 struct vrf_map_elem { 94 struct hlist_node hnode; 95 struct list_head vrf_list; /* VRFs registered to this table */ 96 97 u32 table_id; 98 int users; 99 int ifindex; 100 }; 101 102 static unsigned int vrf_net_id; 103 104 /* per netns vrf data */ 105 struct netns_vrf { 106 /* protected by rtnl lock */ 107 bool add_fib_rules; 108 109 struct vrf_map vmap; 110 struct ctl_table_header *ctl_hdr; 111 }; 112 113 struct net_vrf { 114 struct rtable __rcu *rth; 115 struct rt6_info __rcu *rt6; 116 #if IS_ENABLED(CONFIG_IPV6) 117 struct fib6_table *fib6_table; 118 #endif 119 u32 tb_id; 120 121 struct list_head me_list; /* entry in vrf_map_elem */ 122 int ifindex; 123 }; 124 125 static void vrf_rx_stats(struct net_device *dev, int len) 126 { 127 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 128 129 u64_stats_update_begin(&dstats->syncp); 130 u64_stats_inc(&dstats->rx_packets); 131 u64_stats_add(&dstats->rx_bytes, len); 132 u64_stats_update_end(&dstats->syncp); 133 } 134 135 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 136 { 137 vrf_dev->stats.tx_errors++; 138 kfree_skb(skb); 139 } 140 141 static struct vrf_map *netns_vrf_map(struct net *net) 142 { 143 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 144 145 return &nn_vrf->vmap; 146 } 147 148 static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev) 149 { 150 return netns_vrf_map(dev_net(dev)); 151 } 152 153 static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me) 154 { 155 struct list_head *me_head = &me->vrf_list; 156 struct net_vrf *vrf; 157 158 if (list_empty(me_head)) 159 return -ENODEV; 160 161 vrf = list_first_entry(me_head, struct net_vrf, me_list); 162 163 return vrf->ifindex; 164 } 165 166 static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags) 167 { 168 struct vrf_map_elem *me; 169 170 me = kmalloc(sizeof(*me), flags); 171 if (!me) 172 return NULL; 173 174 return me; 175 } 176 177 static void vrf_map_elem_free(struct vrf_map_elem *me) 178 { 179 kfree(me); 180 } 181 182 static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id, 183 int ifindex, int users) 184 { 185 me->table_id = table_id; 186 me->ifindex = ifindex; 187 me->users = users; 188 INIT_LIST_HEAD(&me->vrf_list); 189 } 190 191 static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap, 192 u32 table_id) 193 { 194 struct vrf_map_elem *me; 195 u32 key; 196 197 key = jhash_1word(table_id, HASH_INITVAL); 198 hash_for_each_possible(vmap->ht, me, hnode, key) { 199 if (me->table_id == table_id) 200 return me; 201 } 202 203 return NULL; 204 } 205 206 static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me) 207 { 208 u32 table_id = me->table_id; 209 u32 key; 210 211 key = jhash_1word(table_id, HASH_INITVAL); 212 hash_add(vmap->ht, &me->hnode, key); 213 } 214 215 static void vrf_map_del_elem(struct vrf_map_elem *me) 216 { 217 hash_del(&me->hnode); 218 } 219 220 static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock) 221 { 222 spin_lock(&vmap->vmap_lock); 223 } 224 225 static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock) 226 { 227 spin_unlock(&vmap->vmap_lock); 228 } 229 230 /* called with rtnl lock held */ 231 static int 232 vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack) 233 { 234 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 235 struct net_vrf *vrf = netdev_priv(dev); 236 struct vrf_map_elem *new_me, *me; 237 u32 table_id = vrf->tb_id; 238 bool free_new_me = false; 239 int users; 240 int res; 241 242 /* we pre-allocate elements used in the spin-locked section (so that we 243 * keep the spinlock as short as possible). 244 */ 245 new_me = vrf_map_elem_alloc(GFP_KERNEL); 246 if (!new_me) 247 return -ENOMEM; 248 249 vrf_map_elem_init(new_me, table_id, dev->ifindex, 0); 250 251 vrf_map_lock(vmap); 252 253 me = vrf_map_lookup_elem(vmap, table_id); 254 if (!me) { 255 me = new_me; 256 vrf_map_add_elem(vmap, me); 257 goto link_vrf; 258 } 259 260 /* we already have an entry in the vrf_map, so it means there is (at 261 * least) a vrf registered on the specific table. 262 */ 263 free_new_me = true; 264 if (vmap->strict_mode) { 265 /* vrfs cannot share the same table */ 266 NL_SET_ERR_MSG(extack, "Table is used by another VRF"); 267 res = -EBUSY; 268 goto unlock; 269 } 270 271 link_vrf: 272 users = ++me->users; 273 if (users == 2) 274 ++vmap->shared_tables; 275 276 list_add(&vrf->me_list, &me->vrf_list); 277 278 res = 0; 279 280 unlock: 281 vrf_map_unlock(vmap); 282 283 /* clean-up, if needed */ 284 if (free_new_me) 285 vrf_map_elem_free(new_me); 286 287 return res; 288 } 289 290 /* called with rtnl lock held */ 291 static void vrf_map_unregister_dev(struct net_device *dev) 292 { 293 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 294 struct net_vrf *vrf = netdev_priv(dev); 295 u32 table_id = vrf->tb_id; 296 struct vrf_map_elem *me; 297 int users; 298 299 vrf_map_lock(vmap); 300 301 me = vrf_map_lookup_elem(vmap, table_id); 302 if (!me) 303 goto unlock; 304 305 list_del(&vrf->me_list); 306 307 users = --me->users; 308 if (users == 1) { 309 --vmap->shared_tables; 310 } else if (users == 0) { 311 vrf_map_del_elem(me); 312 313 /* no one will refer to this element anymore */ 314 vrf_map_elem_free(me); 315 } 316 317 unlock: 318 vrf_map_unlock(vmap); 319 } 320 321 /* return the vrf device index associated with the table_id */ 322 static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id) 323 { 324 struct vrf_map *vmap = netns_vrf_map(net); 325 struct vrf_map_elem *me; 326 int ifindex; 327 328 vrf_map_lock(vmap); 329 330 if (!vmap->strict_mode) { 331 ifindex = -EPERM; 332 goto unlock; 333 } 334 335 me = vrf_map_lookup_elem(vmap, table_id); 336 if (!me) { 337 ifindex = -ENODEV; 338 goto unlock; 339 } 340 341 ifindex = vrf_map_elem_get_vrf_ifindex(me); 342 343 unlock: 344 vrf_map_unlock(vmap); 345 346 return ifindex; 347 } 348 349 /* by default VRF devices do not have a qdisc and are expected 350 * to be created with only a single queue. 351 */ 352 static bool qdisc_tx_is_default(const struct net_device *dev) 353 { 354 struct netdev_queue *txq; 355 struct Qdisc *qdisc; 356 357 if (dev->num_tx_queues > 1) 358 return false; 359 360 txq = netdev_get_tx_queue(dev, 0); 361 qdisc = rcu_access_pointer(txq->qdisc); 362 363 return !qdisc->enqueue; 364 } 365 366 /* Local traffic destined to local address. Reinsert the packet to rx 367 * path, similar to loopback handling. 368 */ 369 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 370 struct dst_entry *dst) 371 { 372 int len = skb->len; 373 374 skb_orphan(skb); 375 376 skb_dst_set(skb, dst); 377 378 /* set pkt_type to avoid skb hitting packet taps twice - 379 * once on Tx and again in Rx processing 380 */ 381 skb->pkt_type = PACKET_LOOPBACK; 382 383 skb->protocol = eth_type_trans(skb, dev); 384 385 if (likely(__netif_rx(skb) == NET_RX_SUCCESS)) { 386 vrf_rx_stats(dev, len); 387 } else { 388 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 389 390 u64_stats_update_begin(&dstats->syncp); 391 u64_stats_inc(&dstats->rx_drops); 392 u64_stats_update_end(&dstats->syncp); 393 } 394 395 return NETDEV_TX_OK; 396 } 397 398 static void vrf_nf_set_untracked(struct sk_buff *skb) 399 { 400 if (skb_get_nfct(skb) == 0) 401 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 402 } 403 404 static void vrf_nf_reset_ct(struct sk_buff *skb) 405 { 406 if (skb_get_nfct(skb) == IP_CT_UNTRACKED) 407 nf_reset_ct(skb); 408 } 409 410 #if IS_ENABLED(CONFIG_IPV6) 411 static int vrf_ip6_local_out(struct net *net, struct sock *sk, 412 struct sk_buff *skb) 413 { 414 int err; 415 416 vrf_nf_reset_ct(skb); 417 418 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, 419 sk, skb, NULL, skb_dst(skb)->dev, dst_output); 420 421 if (likely(err == 1)) 422 err = dst_output(net, sk, skb); 423 424 return err; 425 } 426 427 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 428 struct net_device *dev) 429 { 430 const struct ipv6hdr *iph; 431 struct net *net = dev_net(skb->dev); 432 struct flowi6 fl6; 433 int ret = NET_XMIT_DROP; 434 struct dst_entry *dst; 435 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 436 437 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr))) 438 goto err; 439 440 iph = ipv6_hdr(skb); 441 442 memset(&fl6, 0, sizeof(fl6)); 443 /* needed to match OIF rule */ 444 fl6.flowi6_l3mdev = dev->ifindex; 445 fl6.flowi6_iif = LOOPBACK_IFINDEX; 446 fl6.daddr = iph->daddr; 447 fl6.saddr = iph->saddr; 448 fl6.flowlabel = ip6_flowinfo(iph); 449 fl6.flowi6_mark = skb->mark; 450 fl6.flowi6_proto = iph->nexthdr; 451 452 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL); 453 if (IS_ERR(dst) || dst == dst_null) 454 goto err; 455 456 skb_dst_drop(skb); 457 458 /* if dst.dev is the VRF device again this is locally originated traffic 459 * destined to a local address. Short circuit to Rx path. 460 */ 461 if (dst->dev == dev) 462 return vrf_local_xmit(skb, dev, dst); 463 464 skb_dst_set(skb, dst); 465 466 /* strip the ethernet header added for pass through VRF device */ 467 __skb_pull(skb, skb_network_offset(skb)); 468 469 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb))); 470 ret = vrf_ip6_local_out(net, skb->sk, skb); 471 if (unlikely(net_xmit_eval(ret))) 472 dev->stats.tx_errors++; 473 else 474 ret = NET_XMIT_SUCCESS; 475 476 return ret; 477 err: 478 vrf_tx_error(dev, skb); 479 return NET_XMIT_DROP; 480 } 481 #else 482 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 483 struct net_device *dev) 484 { 485 vrf_tx_error(dev, skb); 486 return NET_XMIT_DROP; 487 } 488 #endif 489 490 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 491 static int vrf_ip_local_out(struct net *net, struct sock *sk, 492 struct sk_buff *skb) 493 { 494 int err; 495 496 vrf_nf_reset_ct(skb); 497 498 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 499 skb, NULL, skb_dst(skb)->dev, dst_output); 500 if (likely(err == 1)) 501 err = dst_output(net, sk, skb); 502 503 return err; 504 } 505 506 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 507 struct net_device *vrf_dev) 508 { 509 struct iphdr *ip4h; 510 int ret = NET_XMIT_DROP; 511 struct flowi4 fl4; 512 struct net *net = dev_net(vrf_dev); 513 struct rtable *rt; 514 515 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr))) 516 goto err; 517 518 ip4h = ip_hdr(skb); 519 520 memset(&fl4, 0, sizeof(fl4)); 521 /* needed to match OIF rule */ 522 fl4.flowi4_l3mdev = vrf_dev->ifindex; 523 fl4.flowi4_iif = LOOPBACK_IFINDEX; 524 fl4.flowi4_tos = ip4h->tos & INET_DSCP_MASK; 525 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC; 526 fl4.flowi4_proto = ip4h->protocol; 527 fl4.daddr = ip4h->daddr; 528 fl4.saddr = ip4h->saddr; 529 530 rt = ip_route_output_flow(net, &fl4, NULL); 531 if (IS_ERR(rt)) 532 goto err; 533 534 skb_dst_drop(skb); 535 536 /* if dst.dev is the VRF device again this is locally originated traffic 537 * destined to a local address. Short circuit to Rx path. 538 */ 539 if (rt->dst.dev == vrf_dev) 540 return vrf_local_xmit(skb, vrf_dev, &rt->dst); 541 542 skb_dst_set(skb, &rt->dst); 543 544 /* strip the ethernet header added for pass through VRF device */ 545 __skb_pull(skb, skb_network_offset(skb)); 546 547 if (!ip4h->saddr) { 548 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 549 RT_SCOPE_LINK); 550 } 551 552 memset(IPCB(skb), 0, sizeof(*IPCB(skb))); 553 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 554 if (unlikely(net_xmit_eval(ret))) 555 vrf_dev->stats.tx_errors++; 556 else 557 ret = NET_XMIT_SUCCESS; 558 559 out: 560 return ret; 561 err: 562 vrf_tx_error(vrf_dev, skb); 563 goto out; 564 } 565 566 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 567 { 568 switch (skb->protocol) { 569 case htons(ETH_P_IP): 570 return vrf_process_v4_outbound(skb, dev); 571 case htons(ETH_P_IPV6): 572 return vrf_process_v6_outbound(skb, dev); 573 default: 574 vrf_tx_error(dev, skb); 575 return NET_XMIT_DROP; 576 } 577 } 578 579 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 580 { 581 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 582 583 int len = skb->len; 584 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 585 586 u64_stats_update_begin(&dstats->syncp); 587 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 588 589 u64_stats_inc(&dstats->tx_packets); 590 u64_stats_add(&dstats->tx_bytes, len); 591 } else { 592 u64_stats_inc(&dstats->tx_drops); 593 } 594 u64_stats_update_end(&dstats->syncp); 595 596 return ret; 597 } 598 599 static void vrf_finish_direct(struct sk_buff *skb) 600 { 601 struct net_device *vrf_dev = skb->dev; 602 603 if (!list_empty(&vrf_dev->ptype_all) && 604 likely(skb_headroom(skb) >= ETH_HLEN)) { 605 struct ethhdr *eth = skb_push(skb, ETH_HLEN); 606 607 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 608 eth_zero_addr(eth->h_dest); 609 eth->h_proto = skb->protocol; 610 611 dev_queue_xmit_nit(skb, vrf_dev); 612 613 skb_pull(skb, ETH_HLEN); 614 } 615 616 vrf_nf_reset_ct(skb); 617 } 618 619 #if IS_ENABLED(CONFIG_IPV6) 620 /* modelled after ip6_finish_output2 */ 621 static int vrf_finish_output6(struct net *net, struct sock *sk, 622 struct sk_buff *skb) 623 { 624 struct dst_entry *dst = skb_dst(skb); 625 struct net_device *dev = dst->dev; 626 const struct in6_addr *nexthop; 627 struct neighbour *neigh; 628 int ret; 629 630 vrf_nf_reset_ct(skb); 631 632 skb->protocol = htons(ETH_P_IPV6); 633 skb->dev = dev; 634 635 rcu_read_lock(); 636 nexthop = rt6_nexthop(dst_rt6_info(dst), &ipv6_hdr(skb)->daddr); 637 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 638 if (unlikely(!neigh)) 639 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 640 if (!IS_ERR(neigh)) { 641 sock_confirm_neigh(skb, neigh); 642 ret = neigh_output(neigh, skb, false); 643 rcu_read_unlock(); 644 return ret; 645 } 646 rcu_read_unlock(); 647 648 IP6_INC_STATS(dev_net(dst->dev), 649 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 650 kfree_skb(skb); 651 return -EINVAL; 652 } 653 654 /* modelled after ip6_output */ 655 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 656 { 657 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 658 net, sk, skb, NULL, skb_dst(skb)->dev, 659 vrf_finish_output6, 660 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 661 } 662 663 /* set dst on skb to send packet to us via dev_xmit path. Allows 664 * packet to go through device based features such as qdisc, netfilter 665 * hooks and packet sockets with skb->dev set to vrf device. 666 */ 667 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev, 668 struct sk_buff *skb) 669 { 670 struct net_vrf *vrf = netdev_priv(vrf_dev); 671 struct dst_entry *dst = NULL; 672 struct rt6_info *rt6; 673 674 rcu_read_lock(); 675 676 rt6 = rcu_dereference(vrf->rt6); 677 if (likely(rt6)) { 678 dst = &rt6->dst; 679 dst_hold(dst); 680 } 681 682 rcu_read_unlock(); 683 684 if (unlikely(!dst)) { 685 vrf_tx_error(vrf_dev, skb); 686 return NULL; 687 } 688 689 skb_dst_drop(skb); 690 skb_dst_set(skb, dst); 691 692 return skb; 693 } 694 695 static int vrf_output6_direct_finish(struct net *net, struct sock *sk, 696 struct sk_buff *skb) 697 { 698 vrf_finish_direct(skb); 699 700 return vrf_ip6_local_out(net, sk, skb); 701 } 702 703 static int vrf_output6_direct(struct net *net, struct sock *sk, 704 struct sk_buff *skb) 705 { 706 int err = 1; 707 708 skb->protocol = htons(ETH_P_IPV6); 709 710 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 711 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb, 712 NULL, skb->dev, vrf_output6_direct_finish); 713 714 if (likely(err == 1)) 715 vrf_finish_direct(skb); 716 717 return err; 718 } 719 720 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk, 721 struct sk_buff *skb) 722 { 723 int err; 724 725 err = vrf_output6_direct(net, sk, skb); 726 if (likely(err == 1)) 727 err = vrf_ip6_local_out(net, sk, skb); 728 729 return err; 730 } 731 732 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev, 733 struct sock *sk, 734 struct sk_buff *skb) 735 { 736 struct net *net = dev_net(vrf_dev); 737 int err; 738 739 skb->dev = vrf_dev; 740 741 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, 742 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish); 743 744 if (likely(err == 1)) 745 err = vrf_output6_direct(net, sk, skb); 746 747 if (likely(err == 1)) 748 return skb; 749 750 return NULL; 751 } 752 753 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 754 struct sock *sk, 755 struct sk_buff *skb) 756 { 757 /* don't divert link scope packets */ 758 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 759 return skb; 760 761 vrf_nf_set_untracked(skb); 762 763 if (qdisc_tx_is_default(vrf_dev) || 764 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED) 765 return vrf_ip6_out_direct(vrf_dev, sk, skb); 766 767 return vrf_ip6_out_redirect(vrf_dev, skb); 768 } 769 770 /* holding rtnl */ 771 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 772 { 773 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 774 struct net *net = dev_net(dev); 775 struct dst_entry *dst; 776 777 RCU_INIT_POINTER(vrf->rt6, NULL); 778 synchronize_rcu(); 779 780 /* move dev in dst's to loopback so this VRF device can be deleted 781 * - based on dst_ifdown 782 */ 783 if (rt6) { 784 dst = &rt6->dst; 785 netdev_ref_replace(dst->dev, net->loopback_dev, 786 &dst->dev_tracker, GFP_KERNEL); 787 dst->dev = net->loopback_dev; 788 dst_release(dst); 789 } 790 } 791 792 static int vrf_rt6_create(struct net_device *dev) 793 { 794 int flags = DST_NOPOLICY | DST_NOXFRM; 795 struct net_vrf *vrf = netdev_priv(dev); 796 struct net *net = dev_net(dev); 797 struct rt6_info *rt6; 798 int rc = -ENOMEM; 799 800 /* IPv6 can be CONFIG enabled and then disabled runtime */ 801 if (!ipv6_mod_enabled()) 802 return 0; 803 804 vrf->fib6_table = fib6_new_table(net, vrf->tb_id); 805 if (!vrf->fib6_table) 806 goto out; 807 808 /* create a dst for routing packets out a VRF device */ 809 rt6 = ip6_dst_alloc(net, dev, flags); 810 if (!rt6) 811 goto out; 812 813 rt6->dst.output = vrf_output6; 814 815 rcu_assign_pointer(vrf->rt6, rt6); 816 817 rc = 0; 818 out: 819 return rc; 820 } 821 #else 822 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 823 struct sock *sk, 824 struct sk_buff *skb) 825 { 826 return skb; 827 } 828 829 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 830 { 831 } 832 833 static int vrf_rt6_create(struct net_device *dev) 834 { 835 return 0; 836 } 837 #endif 838 839 /* modelled after ip_finish_output2 */ 840 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 841 { 842 struct dst_entry *dst = skb_dst(skb); 843 struct rtable *rt = dst_rtable(dst); 844 struct net_device *dev = dst->dev; 845 unsigned int hh_len = LL_RESERVED_SPACE(dev); 846 struct neighbour *neigh; 847 bool is_v6gw = false; 848 849 vrf_nf_reset_ct(skb); 850 851 /* Be paranoid, rather than too clever. */ 852 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 853 skb = skb_expand_head(skb, hh_len); 854 if (!skb) { 855 dev->stats.tx_errors++; 856 return -ENOMEM; 857 } 858 } 859 860 rcu_read_lock(); 861 862 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 863 if (!IS_ERR(neigh)) { 864 int ret; 865 866 sock_confirm_neigh(skb, neigh); 867 /* if crossing protocols, can not use the cached header */ 868 ret = neigh_output(neigh, skb, is_v6gw); 869 rcu_read_unlock(); 870 return ret; 871 } 872 873 rcu_read_unlock(); 874 vrf_tx_error(skb->dev, skb); 875 return -EINVAL; 876 } 877 878 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 879 { 880 struct net_device *dev = skb_dst(skb)->dev; 881 882 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 883 884 skb->dev = dev; 885 skb->protocol = htons(ETH_P_IP); 886 887 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 888 net, sk, skb, NULL, dev, 889 vrf_finish_output, 890 !(IPCB(skb)->flags & IPSKB_REROUTED)); 891 } 892 893 /* set dst on skb to send packet to us via dev_xmit path. Allows 894 * packet to go through device based features such as qdisc, netfilter 895 * hooks and packet sockets with skb->dev set to vrf device. 896 */ 897 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev, 898 struct sk_buff *skb) 899 { 900 struct net_vrf *vrf = netdev_priv(vrf_dev); 901 struct dst_entry *dst = NULL; 902 struct rtable *rth; 903 904 rcu_read_lock(); 905 906 rth = rcu_dereference(vrf->rth); 907 if (likely(rth)) { 908 dst = &rth->dst; 909 dst_hold(dst); 910 } 911 912 rcu_read_unlock(); 913 914 if (unlikely(!dst)) { 915 vrf_tx_error(vrf_dev, skb); 916 return NULL; 917 } 918 919 skb_dst_drop(skb); 920 skb_dst_set(skb, dst); 921 922 return skb; 923 } 924 925 static int vrf_output_direct_finish(struct net *net, struct sock *sk, 926 struct sk_buff *skb) 927 { 928 vrf_finish_direct(skb); 929 930 return vrf_ip_local_out(net, sk, skb); 931 } 932 933 static int vrf_output_direct(struct net *net, struct sock *sk, 934 struct sk_buff *skb) 935 { 936 int err = 1; 937 938 skb->protocol = htons(ETH_P_IP); 939 940 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 941 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb, 942 NULL, skb->dev, vrf_output_direct_finish); 943 944 if (likely(err == 1)) 945 vrf_finish_direct(skb); 946 947 return err; 948 } 949 950 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk, 951 struct sk_buff *skb) 952 { 953 int err; 954 955 err = vrf_output_direct(net, sk, skb); 956 if (likely(err == 1)) 957 err = vrf_ip_local_out(net, sk, skb); 958 959 return err; 960 } 961 962 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev, 963 struct sock *sk, 964 struct sk_buff *skb) 965 { 966 struct net *net = dev_net(vrf_dev); 967 int err; 968 969 skb->dev = vrf_dev; 970 971 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 972 skb, NULL, vrf_dev, vrf_ip_out_direct_finish); 973 974 if (likely(err == 1)) 975 err = vrf_output_direct(net, sk, skb); 976 977 if (likely(err == 1)) 978 return skb; 979 980 return NULL; 981 } 982 983 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 984 struct sock *sk, 985 struct sk_buff *skb) 986 { 987 /* don't divert multicast or local broadcast */ 988 if (ipv4_is_multicast(ip_hdr(skb)->daddr) || 989 ipv4_is_lbcast(ip_hdr(skb)->daddr)) 990 return skb; 991 992 vrf_nf_set_untracked(skb); 993 994 if (qdisc_tx_is_default(vrf_dev) || 995 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED) 996 return vrf_ip_out_direct(vrf_dev, sk, skb); 997 998 return vrf_ip_out_redirect(vrf_dev, skb); 999 } 1000 1001 /* called with rcu lock held */ 1002 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 1003 struct sock *sk, 1004 struct sk_buff *skb, 1005 u16 proto) 1006 { 1007 switch (proto) { 1008 case AF_INET: 1009 return vrf_ip_out(vrf_dev, sk, skb); 1010 case AF_INET6: 1011 return vrf_ip6_out(vrf_dev, sk, skb); 1012 } 1013 1014 return skb; 1015 } 1016 1017 /* holding rtnl */ 1018 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 1019 { 1020 struct rtable *rth = rtnl_dereference(vrf->rth); 1021 struct net *net = dev_net(dev); 1022 struct dst_entry *dst; 1023 1024 RCU_INIT_POINTER(vrf->rth, NULL); 1025 synchronize_rcu(); 1026 1027 /* move dev in dst's to loopback so this VRF device can be deleted 1028 * - based on dst_ifdown 1029 */ 1030 if (rth) { 1031 dst = &rth->dst; 1032 netdev_ref_replace(dst->dev, net->loopback_dev, 1033 &dst->dev_tracker, GFP_KERNEL); 1034 dst->dev = net->loopback_dev; 1035 dst_release(dst); 1036 } 1037 } 1038 1039 static int vrf_rtable_create(struct net_device *dev) 1040 { 1041 struct net_vrf *vrf = netdev_priv(dev); 1042 struct rtable *rth; 1043 1044 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 1045 return -ENOMEM; 1046 1047 /* create a dst for routing packets out through a VRF device */ 1048 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1); 1049 if (!rth) 1050 return -ENOMEM; 1051 1052 rth->dst.output = vrf_output; 1053 1054 rcu_assign_pointer(vrf->rth, rth); 1055 1056 return 0; 1057 } 1058 1059 /**************************** device handling ********************/ 1060 1061 /* cycle interface to flush neighbor cache and move routes across tables */ 1062 static void cycle_netdev(struct net_device *dev, 1063 struct netlink_ext_ack *extack) 1064 { 1065 unsigned int flags = dev->flags; 1066 int ret; 1067 1068 if (!netif_running(dev)) 1069 return; 1070 1071 ret = dev_change_flags(dev, flags & ~IFF_UP, extack); 1072 if (ret >= 0) 1073 ret = dev_change_flags(dev, flags, extack); 1074 1075 if (ret < 0) { 1076 netdev_err(dev, 1077 "Failed to cycle device %s; route tables might be wrong!\n", 1078 dev->name); 1079 } 1080 } 1081 1082 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1083 struct netlink_ext_ack *extack) 1084 { 1085 int ret; 1086 1087 /* do not allow loopback device to be enslaved to a VRF. 1088 * The vrf device acts as the loopback for the vrf. 1089 */ 1090 if (port_dev == dev_net(dev)->loopback_dev) { 1091 NL_SET_ERR_MSG(extack, 1092 "Can not enslave loopback device to a VRF"); 1093 return -EOPNOTSUPP; 1094 } 1095 1096 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 1097 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack); 1098 if (ret < 0) 1099 goto err; 1100 1101 cycle_netdev(port_dev, extack); 1102 1103 return 0; 1104 1105 err: 1106 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1107 return ret; 1108 } 1109 1110 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1111 struct netlink_ext_ack *extack) 1112 { 1113 if (netif_is_l3_master(port_dev)) { 1114 NL_SET_ERR_MSG(extack, 1115 "Can not enslave an L3 master device to a VRF"); 1116 return -EINVAL; 1117 } 1118 1119 if (netif_is_l3_slave(port_dev)) 1120 return -EINVAL; 1121 1122 return do_vrf_add_slave(dev, port_dev, extack); 1123 } 1124 1125 /* inverse of do_vrf_add_slave */ 1126 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1127 { 1128 netdev_upper_dev_unlink(port_dev, dev); 1129 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1130 1131 cycle_netdev(port_dev, NULL); 1132 1133 return 0; 1134 } 1135 1136 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1137 { 1138 return do_vrf_del_slave(dev, port_dev); 1139 } 1140 1141 static void vrf_dev_uninit(struct net_device *dev) 1142 { 1143 struct net_vrf *vrf = netdev_priv(dev); 1144 1145 vrf_rtable_release(dev, vrf); 1146 vrf_rt6_release(dev, vrf); 1147 } 1148 1149 static int vrf_dev_init(struct net_device *dev) 1150 { 1151 struct net_vrf *vrf = netdev_priv(dev); 1152 1153 /* create the default dst which points back to us */ 1154 if (vrf_rtable_create(dev) != 0) 1155 goto out_nomem; 1156 1157 if (vrf_rt6_create(dev) != 0) 1158 goto out_rth; 1159 1160 dev->flags = IFF_MASTER | IFF_NOARP; 1161 1162 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 1163 dev->operstate = IF_OPER_UP; 1164 netdev_lockdep_set_classes(dev); 1165 return 0; 1166 1167 out_rth: 1168 vrf_rtable_release(dev, vrf); 1169 out_nomem: 1170 return -ENOMEM; 1171 } 1172 1173 static const struct net_device_ops vrf_netdev_ops = { 1174 .ndo_init = vrf_dev_init, 1175 .ndo_uninit = vrf_dev_uninit, 1176 .ndo_start_xmit = vrf_xmit, 1177 .ndo_set_mac_address = eth_mac_addr, 1178 .ndo_add_slave = vrf_add_slave, 1179 .ndo_del_slave = vrf_del_slave, 1180 }; 1181 1182 static u32 vrf_fib_table(const struct net_device *dev) 1183 { 1184 struct net_vrf *vrf = netdev_priv(dev); 1185 1186 return vrf->tb_id; 1187 } 1188 1189 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 1190 { 1191 kfree_skb(skb); 1192 return 0; 1193 } 1194 1195 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 1196 struct sk_buff *skb, 1197 struct net_device *dev) 1198 { 1199 struct net *net = dev_net(dev); 1200 1201 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1) 1202 skb = NULL; /* kfree_skb(skb) handled by nf code */ 1203 1204 return skb; 1205 } 1206 1207 static int vrf_prepare_mac_header(struct sk_buff *skb, 1208 struct net_device *vrf_dev, u16 proto) 1209 { 1210 struct ethhdr *eth; 1211 int err; 1212 1213 /* in general, we do not know if there is enough space in the head of 1214 * the packet for hosting the mac header. 1215 */ 1216 err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev)); 1217 if (unlikely(err)) 1218 /* no space in the skb head */ 1219 return -ENOBUFS; 1220 1221 __skb_push(skb, ETH_HLEN); 1222 eth = (struct ethhdr *)skb->data; 1223 1224 skb_reset_mac_header(skb); 1225 skb_reset_mac_len(skb); 1226 1227 /* we set the ethernet destination and the source addresses to the 1228 * address of the VRF device. 1229 */ 1230 ether_addr_copy(eth->h_dest, vrf_dev->dev_addr); 1231 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 1232 eth->h_proto = htons(proto); 1233 1234 /* the destination address of the Ethernet frame corresponds to the 1235 * address set on the VRF interface; therefore, the packet is intended 1236 * to be processed locally. 1237 */ 1238 skb->protocol = eth->h_proto; 1239 skb->pkt_type = PACKET_HOST; 1240 1241 skb_postpush_rcsum(skb, skb->data, ETH_HLEN); 1242 1243 skb_pull_inline(skb, ETH_HLEN); 1244 1245 return 0; 1246 } 1247 1248 /* prepare and add the mac header to the packet if it was not set previously. 1249 * In this way, packet sniffers such as tcpdump can parse the packet correctly. 1250 * If the mac header was already set, the original mac header is left 1251 * untouched and the function returns immediately. 1252 */ 1253 static int vrf_add_mac_header_if_unset(struct sk_buff *skb, 1254 struct net_device *vrf_dev, 1255 u16 proto, struct net_device *orig_dev) 1256 { 1257 if (skb_mac_header_was_set(skb) && dev_has_header(orig_dev)) 1258 return 0; 1259 1260 return vrf_prepare_mac_header(skb, vrf_dev, proto); 1261 } 1262 1263 #if IS_ENABLED(CONFIG_IPV6) 1264 /* neighbor handling is done with actual device; do not want 1265 * to flip skb->dev for those ndisc packets. This really fails 1266 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 1267 * a start. 1268 */ 1269 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 1270 { 1271 const struct ipv6hdr *iph = ipv6_hdr(skb); 1272 bool rc = false; 1273 1274 if (iph->nexthdr == NEXTHDR_ICMP) { 1275 const struct icmp6hdr *icmph; 1276 struct icmp6hdr _icmph; 1277 1278 icmph = skb_header_pointer(skb, sizeof(*iph), 1279 sizeof(_icmph), &_icmph); 1280 if (!icmph) 1281 goto out; 1282 1283 switch (icmph->icmp6_type) { 1284 case NDISC_ROUTER_SOLICITATION: 1285 case NDISC_ROUTER_ADVERTISEMENT: 1286 case NDISC_NEIGHBOUR_SOLICITATION: 1287 case NDISC_NEIGHBOUR_ADVERTISEMENT: 1288 case NDISC_REDIRECT: 1289 rc = true; 1290 break; 1291 } 1292 } 1293 1294 out: 1295 return rc; 1296 } 1297 1298 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 1299 const struct net_device *dev, 1300 struct flowi6 *fl6, 1301 int ifindex, 1302 const struct sk_buff *skb, 1303 int flags) 1304 { 1305 struct net_vrf *vrf = netdev_priv(dev); 1306 1307 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags); 1308 } 1309 1310 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 1311 int ifindex) 1312 { 1313 const struct ipv6hdr *iph = ipv6_hdr(skb); 1314 struct flowi6 fl6 = { 1315 .flowi6_iif = ifindex, 1316 .flowi6_mark = skb->mark, 1317 .flowi6_proto = iph->nexthdr, 1318 .daddr = iph->daddr, 1319 .saddr = iph->saddr, 1320 .flowlabel = ip6_flowinfo(iph), 1321 }; 1322 struct net *net = dev_net(vrf_dev); 1323 struct rt6_info *rt6; 1324 1325 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb, 1326 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 1327 if (unlikely(!rt6)) 1328 return; 1329 1330 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 1331 return; 1332 1333 skb_dst_set(skb, &rt6->dst); 1334 } 1335 1336 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1337 struct sk_buff *skb) 1338 { 1339 int orig_iif = skb->skb_iif; 1340 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 1341 bool is_ndisc = ipv6_ndisc_frame(skb); 1342 1343 /* loopback, multicast & non-ND link-local traffic; do not push through 1344 * packet taps again. Reset pkt_type for upper layers to process skb. 1345 * For non-loopback strict packets, determine the dst using the original 1346 * ifindex. 1347 */ 1348 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) { 1349 skb->dev = vrf_dev; 1350 skb->skb_iif = vrf_dev->ifindex; 1351 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1352 1353 if (skb->pkt_type == PACKET_LOOPBACK) 1354 skb->pkt_type = PACKET_HOST; 1355 else 1356 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1357 1358 goto out; 1359 } 1360 1361 /* if packet is NDISC then keep the ingress interface */ 1362 if (!is_ndisc) { 1363 struct net_device *orig_dev = skb->dev; 1364 1365 vrf_rx_stats(vrf_dev, skb->len); 1366 skb->dev = vrf_dev; 1367 skb->skb_iif = vrf_dev->ifindex; 1368 1369 if (!list_empty(&vrf_dev->ptype_all)) { 1370 int err; 1371 1372 err = vrf_add_mac_header_if_unset(skb, vrf_dev, 1373 ETH_P_IPV6, 1374 orig_dev); 1375 if (likely(!err)) { 1376 skb_push(skb, skb->mac_len); 1377 dev_queue_xmit_nit(skb, vrf_dev); 1378 skb_pull(skb, skb->mac_len); 1379 } 1380 } 1381 1382 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1383 } 1384 1385 if (need_strict) 1386 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1387 1388 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 1389 out: 1390 return skb; 1391 } 1392 1393 #else 1394 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1395 struct sk_buff *skb) 1396 { 1397 return skb; 1398 } 1399 #endif 1400 1401 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 1402 struct sk_buff *skb) 1403 { 1404 struct net_device *orig_dev = skb->dev; 1405 1406 skb->dev = vrf_dev; 1407 skb->skb_iif = vrf_dev->ifindex; 1408 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1409 1410 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1411 goto out; 1412 1413 /* loopback traffic; do not push through packet taps again. 1414 * Reset pkt_type for upper layers to process skb 1415 */ 1416 if (skb->pkt_type == PACKET_LOOPBACK) { 1417 skb->pkt_type = PACKET_HOST; 1418 goto out; 1419 } 1420 1421 vrf_rx_stats(vrf_dev, skb->len); 1422 1423 if (!list_empty(&vrf_dev->ptype_all)) { 1424 int err; 1425 1426 err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP, 1427 orig_dev); 1428 if (likely(!err)) { 1429 skb_push(skb, skb->mac_len); 1430 dev_queue_xmit_nit(skb, vrf_dev); 1431 skb_pull(skb, skb->mac_len); 1432 } 1433 } 1434 1435 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1436 out: 1437 return skb; 1438 } 1439 1440 /* called with rcu lock held */ 1441 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1442 struct sk_buff *skb, 1443 u16 proto) 1444 { 1445 switch (proto) { 1446 case AF_INET: 1447 return vrf_ip_rcv(vrf_dev, skb); 1448 case AF_INET6: 1449 return vrf_ip6_rcv(vrf_dev, skb); 1450 } 1451 1452 return skb; 1453 } 1454 1455 #if IS_ENABLED(CONFIG_IPV6) 1456 /* send to link-local or multicast address via interface enslaved to 1457 * VRF device. Force lookup to VRF table without changing flow struct 1458 * Note: Caller to this function must hold rcu_read_lock() and no refcnt 1459 * is taken on the dst by this function. 1460 */ 1461 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1462 struct flowi6 *fl6) 1463 { 1464 struct net *net = dev_net(dev); 1465 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF; 1466 struct dst_entry *dst = NULL; 1467 struct rt6_info *rt; 1468 1469 /* VRF device does not have a link-local address and 1470 * sending packets to link-local or mcast addresses over 1471 * a VRF device does not make sense 1472 */ 1473 if (fl6->flowi6_oif == dev->ifindex) { 1474 dst = &net->ipv6.ip6_null_entry->dst; 1475 return dst; 1476 } 1477 1478 if (!ipv6_addr_any(&fl6->saddr)) 1479 flags |= RT6_LOOKUP_F_HAS_SADDR; 1480 1481 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags); 1482 if (rt) 1483 dst = &rt->dst; 1484 1485 return dst; 1486 } 1487 #endif 1488 1489 static const struct l3mdev_ops vrf_l3mdev_ops = { 1490 .l3mdev_fib_table = vrf_fib_table, 1491 .l3mdev_l3_rcv = vrf_l3_rcv, 1492 .l3mdev_l3_out = vrf_l3_out, 1493 #if IS_ENABLED(CONFIG_IPV6) 1494 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1495 #endif 1496 }; 1497 1498 static void vrf_get_drvinfo(struct net_device *dev, 1499 struct ethtool_drvinfo *info) 1500 { 1501 strscpy(info->driver, DRV_NAME, sizeof(info->driver)); 1502 strscpy(info->version, DRV_VERSION, sizeof(info->version)); 1503 } 1504 1505 static const struct ethtool_ops vrf_ethtool_ops = { 1506 .get_drvinfo = vrf_get_drvinfo, 1507 }; 1508 1509 static inline size_t vrf_fib_rule_nl_size(void) 1510 { 1511 size_t sz; 1512 1513 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1514 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1515 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1516 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */ 1517 1518 return sz; 1519 } 1520 1521 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1522 { 1523 struct fib_rule_hdr *frh; 1524 struct nlmsghdr *nlh; 1525 struct sk_buff *skb; 1526 int err; 1527 1528 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) && 1529 !ipv6_mod_enabled()) 1530 return 0; 1531 1532 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1533 if (!skb) 1534 return -ENOMEM; 1535 1536 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1537 if (!nlh) 1538 goto nla_put_failure; 1539 1540 /* rule only needs to appear once */ 1541 nlh->nlmsg_flags |= NLM_F_EXCL; 1542 1543 frh = nlmsg_data(nlh); 1544 memset(frh, 0, sizeof(*frh)); 1545 frh->family = family; 1546 frh->action = FR_ACT_TO_TBL; 1547 1548 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL)) 1549 goto nla_put_failure; 1550 1551 if (nla_put_u8(skb, FRA_L3MDEV, 1)) 1552 goto nla_put_failure; 1553 1554 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1555 goto nla_put_failure; 1556 1557 nlmsg_end(skb, nlh); 1558 1559 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1560 skb->sk = dev_net(dev)->rtnl; 1561 if (add_it) { 1562 err = fib_nl_newrule(skb, nlh, NULL); 1563 if (err == -EEXIST) 1564 err = 0; 1565 } else { 1566 err = fib_nl_delrule(skb, nlh, NULL); 1567 if (err == -ENOENT) 1568 err = 0; 1569 } 1570 nlmsg_free(skb); 1571 1572 return err; 1573 1574 nla_put_failure: 1575 nlmsg_free(skb); 1576 1577 return -EMSGSIZE; 1578 } 1579 1580 static int vrf_add_fib_rules(const struct net_device *dev) 1581 { 1582 int err; 1583 1584 err = vrf_fib_rule(dev, AF_INET, true); 1585 if (err < 0) 1586 goto out_err; 1587 1588 err = vrf_fib_rule(dev, AF_INET6, true); 1589 if (err < 0) 1590 goto ipv6_err; 1591 1592 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1593 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1594 if (err < 0) 1595 goto ipmr_err; 1596 #endif 1597 1598 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1599 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true); 1600 if (err < 0) 1601 goto ip6mr_err; 1602 #endif 1603 1604 return 0; 1605 1606 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1607 ip6mr_err: 1608 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false); 1609 #endif 1610 1611 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1612 ipmr_err: 1613 vrf_fib_rule(dev, AF_INET6, false); 1614 #endif 1615 1616 ipv6_err: 1617 vrf_fib_rule(dev, AF_INET, false); 1618 1619 out_err: 1620 netdev_err(dev, "Failed to add FIB rules.\n"); 1621 return err; 1622 } 1623 1624 static void vrf_setup(struct net_device *dev) 1625 { 1626 ether_setup(dev); 1627 1628 /* Initialize the device structure. */ 1629 dev->netdev_ops = &vrf_netdev_ops; 1630 dev->l3mdev_ops = &vrf_l3mdev_ops; 1631 dev->ethtool_ops = &vrf_ethtool_ops; 1632 dev->needs_free_netdev = true; 1633 1634 /* Fill in device structure with ethernet-generic values. */ 1635 eth_hw_addr_random(dev); 1636 1637 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1638 dev->lltx = true; 1639 1640 /* don't allow vrf devices to change network namespaces. */ 1641 dev->netns_local = true; 1642 1643 /* does not make sense for a VLAN to be added to a vrf device */ 1644 dev->features |= NETIF_F_VLAN_CHALLENGED; 1645 1646 /* enable offload features */ 1647 dev->features |= NETIF_F_GSO_SOFTWARE; 1648 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC; 1649 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1650 1651 dev->hw_features = dev->features; 1652 dev->hw_enc_features = dev->features; 1653 1654 /* default to no qdisc; user can add if desired */ 1655 dev->priv_flags |= IFF_NO_QUEUE; 1656 dev->priv_flags |= IFF_NO_RX_HANDLER; 1657 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1658 1659 /* VRF devices do not care about MTU, but if the MTU is set 1660 * too low then the ipv4 and ipv6 protocols are disabled 1661 * which breaks networking. 1662 */ 1663 dev->min_mtu = IPV6_MIN_MTU; 1664 dev->max_mtu = IP6_MAX_MTU; 1665 dev->mtu = dev->max_mtu; 1666 1667 dev->pcpu_stat_type = NETDEV_PCPU_STAT_DSTATS; 1668 } 1669 1670 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[], 1671 struct netlink_ext_ack *extack) 1672 { 1673 if (tb[IFLA_ADDRESS]) { 1674 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1675 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1676 return -EINVAL; 1677 } 1678 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1679 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1680 return -EADDRNOTAVAIL; 1681 } 1682 } 1683 return 0; 1684 } 1685 1686 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1687 { 1688 struct net_device *port_dev; 1689 struct list_head *iter; 1690 1691 netdev_for_each_lower_dev(dev, port_dev, iter) 1692 vrf_del_slave(dev, port_dev); 1693 1694 vrf_map_unregister_dev(dev); 1695 1696 unregister_netdevice_queue(dev, head); 1697 } 1698 1699 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1700 struct nlattr *tb[], struct nlattr *data[], 1701 struct netlink_ext_ack *extack) 1702 { 1703 struct net_vrf *vrf = netdev_priv(dev); 1704 struct netns_vrf *nn_vrf; 1705 bool *add_fib_rules; 1706 struct net *net; 1707 int err; 1708 1709 if (!data || !data[IFLA_VRF_TABLE]) { 1710 NL_SET_ERR_MSG(extack, "VRF table id is missing"); 1711 return -EINVAL; 1712 } 1713 1714 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1715 if (vrf->tb_id == RT_TABLE_UNSPEC) { 1716 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE], 1717 "Invalid VRF table id"); 1718 return -EINVAL; 1719 } 1720 1721 dev->priv_flags |= IFF_L3MDEV_MASTER; 1722 1723 err = register_netdevice(dev); 1724 if (err) 1725 goto out; 1726 1727 /* mapping between table_id and vrf; 1728 * note: such binding could not be done in the dev init function 1729 * because dev->ifindex id is not available yet. 1730 */ 1731 vrf->ifindex = dev->ifindex; 1732 1733 err = vrf_map_register_dev(dev, extack); 1734 if (err) { 1735 unregister_netdevice(dev); 1736 goto out; 1737 } 1738 1739 net = dev_net(dev); 1740 nn_vrf = net_generic(net, vrf_net_id); 1741 1742 add_fib_rules = &nn_vrf->add_fib_rules; 1743 if (*add_fib_rules) { 1744 err = vrf_add_fib_rules(dev); 1745 if (err) { 1746 vrf_map_unregister_dev(dev); 1747 unregister_netdevice(dev); 1748 goto out; 1749 } 1750 *add_fib_rules = false; 1751 } 1752 1753 out: 1754 return err; 1755 } 1756 1757 static size_t vrf_nl_getsize(const struct net_device *dev) 1758 { 1759 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1760 } 1761 1762 static int vrf_fillinfo(struct sk_buff *skb, 1763 const struct net_device *dev) 1764 { 1765 struct net_vrf *vrf = netdev_priv(dev); 1766 1767 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1768 } 1769 1770 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1771 const struct net_device *slave_dev) 1772 { 1773 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1774 } 1775 1776 static int vrf_fill_slave_info(struct sk_buff *skb, 1777 const struct net_device *vrf_dev, 1778 const struct net_device *slave_dev) 1779 { 1780 struct net_vrf *vrf = netdev_priv(vrf_dev); 1781 1782 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1783 return -EMSGSIZE; 1784 1785 return 0; 1786 } 1787 1788 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1789 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1790 }; 1791 1792 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1793 .kind = DRV_NAME, 1794 .priv_size = sizeof(struct net_vrf), 1795 1796 .get_size = vrf_nl_getsize, 1797 .policy = vrf_nl_policy, 1798 .validate = vrf_validate, 1799 .fill_info = vrf_fillinfo, 1800 1801 .get_slave_size = vrf_get_slave_size, 1802 .fill_slave_info = vrf_fill_slave_info, 1803 1804 .newlink = vrf_newlink, 1805 .dellink = vrf_dellink, 1806 .setup = vrf_setup, 1807 .maxtype = IFLA_VRF_MAX, 1808 }; 1809 1810 static int vrf_device_event(struct notifier_block *unused, 1811 unsigned long event, void *ptr) 1812 { 1813 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1814 1815 /* only care about unregister events to drop slave references */ 1816 if (event == NETDEV_UNREGISTER) { 1817 struct net_device *vrf_dev; 1818 1819 if (!netif_is_l3_slave(dev)) 1820 goto out; 1821 1822 vrf_dev = netdev_master_upper_dev_get(dev); 1823 vrf_del_slave(vrf_dev, dev); 1824 } 1825 out: 1826 return NOTIFY_DONE; 1827 } 1828 1829 static struct notifier_block vrf_notifier_block __read_mostly = { 1830 .notifier_call = vrf_device_event, 1831 }; 1832 1833 static int vrf_map_init(struct vrf_map *vmap) 1834 { 1835 spin_lock_init(&vmap->vmap_lock); 1836 hash_init(vmap->ht); 1837 1838 vmap->strict_mode = false; 1839 1840 return 0; 1841 } 1842 1843 #ifdef CONFIG_SYSCTL 1844 static bool vrf_strict_mode(struct vrf_map *vmap) 1845 { 1846 bool strict_mode; 1847 1848 vrf_map_lock(vmap); 1849 strict_mode = vmap->strict_mode; 1850 vrf_map_unlock(vmap); 1851 1852 return strict_mode; 1853 } 1854 1855 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode) 1856 { 1857 bool *cur_mode; 1858 int res = 0; 1859 1860 vrf_map_lock(vmap); 1861 1862 cur_mode = &vmap->strict_mode; 1863 if (*cur_mode == new_mode) 1864 goto unlock; 1865 1866 if (*cur_mode) { 1867 /* disable strict mode */ 1868 *cur_mode = false; 1869 } else { 1870 if (vmap->shared_tables) { 1871 /* we cannot allow strict_mode because there are some 1872 * vrfs that share one or more tables. 1873 */ 1874 res = -EBUSY; 1875 goto unlock; 1876 } 1877 1878 /* no tables are shared among vrfs, so we can go back 1879 * to 1:1 association between a vrf with its table. 1880 */ 1881 *cur_mode = true; 1882 } 1883 1884 unlock: 1885 vrf_map_unlock(vmap); 1886 1887 return res; 1888 } 1889 1890 static int vrf_shared_table_handler(const struct ctl_table *table, int write, 1891 void *buffer, size_t *lenp, loff_t *ppos) 1892 { 1893 struct net *net = (struct net *)table->extra1; 1894 struct vrf_map *vmap = netns_vrf_map(net); 1895 int proc_strict_mode = 0; 1896 struct ctl_table tmp = { 1897 .procname = table->procname, 1898 .data = &proc_strict_mode, 1899 .maxlen = sizeof(int), 1900 .mode = table->mode, 1901 .extra1 = SYSCTL_ZERO, 1902 .extra2 = SYSCTL_ONE, 1903 }; 1904 int ret; 1905 1906 if (!write) 1907 proc_strict_mode = vrf_strict_mode(vmap); 1908 1909 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 1910 1911 if (write && ret == 0) 1912 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode); 1913 1914 return ret; 1915 } 1916 1917 static const struct ctl_table vrf_table[] = { 1918 { 1919 .procname = "strict_mode", 1920 .data = NULL, 1921 .maxlen = sizeof(int), 1922 .mode = 0644, 1923 .proc_handler = vrf_shared_table_handler, 1924 /* set by the vrf_netns_init */ 1925 .extra1 = NULL, 1926 }, 1927 }; 1928 1929 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1930 { 1931 struct ctl_table *table; 1932 1933 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL); 1934 if (!table) 1935 return -ENOMEM; 1936 1937 /* init the extra1 parameter with the reference to current netns */ 1938 table[0].extra1 = net; 1939 1940 nn_vrf->ctl_hdr = register_net_sysctl_sz(net, "net/vrf", table, 1941 ARRAY_SIZE(vrf_table)); 1942 if (!nn_vrf->ctl_hdr) { 1943 kfree(table); 1944 return -ENOMEM; 1945 } 1946 1947 return 0; 1948 } 1949 1950 static void vrf_netns_exit_sysctl(struct net *net) 1951 { 1952 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1953 const struct ctl_table *table; 1954 1955 table = nn_vrf->ctl_hdr->ctl_table_arg; 1956 unregister_net_sysctl_table(nn_vrf->ctl_hdr); 1957 kfree(table); 1958 } 1959 #else 1960 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1961 { 1962 return 0; 1963 } 1964 1965 static void vrf_netns_exit_sysctl(struct net *net) 1966 { 1967 } 1968 #endif 1969 1970 /* Initialize per network namespace state */ 1971 static int __net_init vrf_netns_init(struct net *net) 1972 { 1973 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1974 1975 nn_vrf->add_fib_rules = true; 1976 vrf_map_init(&nn_vrf->vmap); 1977 1978 return vrf_netns_init_sysctl(net, nn_vrf); 1979 } 1980 1981 static void __net_exit vrf_netns_exit(struct net *net) 1982 { 1983 vrf_netns_exit_sysctl(net); 1984 } 1985 1986 static struct pernet_operations vrf_net_ops __net_initdata = { 1987 .init = vrf_netns_init, 1988 .exit = vrf_netns_exit, 1989 .id = &vrf_net_id, 1990 .size = sizeof(struct netns_vrf), 1991 }; 1992 1993 static int __init vrf_init_module(void) 1994 { 1995 int rc; 1996 1997 register_netdevice_notifier(&vrf_notifier_block); 1998 1999 rc = register_pernet_subsys(&vrf_net_ops); 2000 if (rc < 0) 2001 goto error; 2002 2003 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF, 2004 vrf_ifindex_lookup_by_table_id); 2005 if (rc < 0) 2006 goto unreg_pernet; 2007 2008 rc = rtnl_link_register(&vrf_link_ops); 2009 if (rc < 0) 2010 goto table_lookup_unreg; 2011 2012 return 0; 2013 2014 table_lookup_unreg: 2015 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF, 2016 vrf_ifindex_lookup_by_table_id); 2017 2018 unreg_pernet: 2019 unregister_pernet_subsys(&vrf_net_ops); 2020 2021 error: 2022 unregister_netdevice_notifier(&vrf_notifier_block); 2023 return rc; 2024 } 2025 2026 module_init(vrf_init_module); 2027 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 2028 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 2029 MODULE_LICENSE("GPL"); 2030 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 2031 MODULE_VERSION(DRV_VERSION); 2032