xref: /linux/drivers/net/vrf.c (revision 0e1368a28dd5231ae0dbe240dfe0ff2657de5647)
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/flow.h>
30 #include <net/ip.h>
31 #include <net/ip_fib.h>
32 #include <net/ip6_fib.h>
33 #include <net/ip6_route.h>
34 #include <net/route.h>
35 #include <net/addrconf.h>
36 #include <net/l3mdev.h>
37 #include <net/fib_rules.h>
38 #include <net/netdev_lock.h>
39 #include <net/sch_generic.h>
40 #include <net/netns/generic.h>
41 #include <net/netfilter/nf_conntrack.h>
42 
43 #define DRV_NAME	"vrf"
44 #define DRV_VERSION	"1.1"
45 
46 #define FIB_RULE_PREF  1000       /* default preference for FIB rules */
47 
48 #define HT_MAP_BITS	4
49 #define HASH_INITVAL	((u32)0xcafef00d)
50 
51 struct  vrf_map {
52 	DECLARE_HASHTABLE(ht, HT_MAP_BITS);
53 	spinlock_t vmap_lock;
54 
55 	/* shared_tables:
56 	 * count how many distinct tables do not comply with the strict mode
57 	 * requirement.
58 	 * shared_tables value must be 0 in order to enable the strict mode.
59 	 *
60 	 * example of the evolution of shared_tables:
61 	 *                                                        | time
62 	 * add  vrf0 --> table 100        shared_tables = 0       | t0
63 	 * add  vrf1 --> table 101        shared_tables = 0       | t1
64 	 * add  vrf2 --> table 100        shared_tables = 1       | t2
65 	 * add  vrf3 --> table 100        shared_tables = 1       | t3
66 	 * add  vrf4 --> table 101        shared_tables = 2       v t4
67 	 *
68 	 * shared_tables is a "step function" (or "staircase function")
69 	 * and it is increased by one when the second vrf is associated to a
70 	 * table.
71 	 *
72 	 * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1.
73 	 *
74 	 * at t3, another dev (vrf3) is bound to the same table 100 but the
75 	 * value of shared_tables is still 1.
76 	 * This means that no matter how many new vrfs will register on the
77 	 * table 100, the shared_tables will not increase (considering only
78 	 * table 100).
79 	 *
80 	 * at t4, vrf4 is bound to table 101, and shared_tables = 2.
81 	 *
82 	 * Looking at the value of shared_tables we can immediately know if
83 	 * the strict_mode can or cannot be enforced. Indeed, strict_mode
84 	 * can be enforced iff shared_tables = 0.
85 	 *
86 	 * Conversely, shared_tables is decreased when a vrf is de-associated
87 	 * from a table with exactly two associated vrfs.
88 	 */
89 	u32 shared_tables;
90 
91 	bool strict_mode;
92 };
93 
94 struct vrf_map_elem {
95 	struct hlist_node hnode;
96 	struct list_head vrf_list;  /* VRFs registered to this table */
97 
98 	u32 table_id;
99 	int users;
100 	int ifindex;
101 };
102 
103 static unsigned int vrf_net_id;
104 
105 /* per netns vrf data */
106 struct netns_vrf {
107 	/* protected by rtnl lock */
108 	bool add_fib_rules;
109 
110 	struct vrf_map vmap;
111 	struct ctl_table_header	*ctl_hdr;
112 };
113 
114 struct net_vrf {
115 	struct rtable		*rth;
116 	struct rt6_info		*rt6;
117 #if IS_ENABLED(CONFIG_IPV6)
118 	struct fib6_table	*fib6_table;
119 #endif
120 	u32                     tb_id;
121 
122 	struct list_head	me_list;   /* entry in vrf_map_elem */
123 	int			ifindex;
124 };
125 
126 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
127 {
128 	vrf_dev->stats.tx_errors++;
129 	kfree_skb(skb);
130 }
131 
132 static struct vrf_map *netns_vrf_map(struct net *net)
133 {
134 	struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
135 
136 	return &nn_vrf->vmap;
137 }
138 
139 static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev)
140 {
141 	return netns_vrf_map(dev_net(dev));
142 }
143 
144 static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me)
145 {
146 	struct list_head *me_head = &me->vrf_list;
147 	struct net_vrf *vrf;
148 
149 	if (list_empty(me_head))
150 		return -ENODEV;
151 
152 	vrf = list_first_entry(me_head, struct net_vrf, me_list);
153 
154 	return vrf->ifindex;
155 }
156 
157 static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags)
158 {
159 	struct vrf_map_elem *me;
160 
161 	me = kmalloc_obj(*me, flags);
162 	if (!me)
163 		return NULL;
164 
165 	return me;
166 }
167 
168 static void vrf_map_elem_free(struct vrf_map_elem *me)
169 {
170 	kfree(me);
171 }
172 
173 static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id,
174 			      int ifindex, int users)
175 {
176 	me->table_id = table_id;
177 	me->ifindex = ifindex;
178 	me->users = users;
179 	INIT_LIST_HEAD(&me->vrf_list);
180 }
181 
182 static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap,
183 						u32 table_id)
184 {
185 	struct vrf_map_elem *me;
186 	u32 key;
187 
188 	key = jhash_1word(table_id, HASH_INITVAL);
189 	hash_for_each_possible(vmap->ht, me, hnode, key) {
190 		if (me->table_id == table_id)
191 			return me;
192 	}
193 
194 	return NULL;
195 }
196 
197 static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me)
198 {
199 	u32 table_id = me->table_id;
200 	u32 key;
201 
202 	key = jhash_1word(table_id, HASH_INITVAL);
203 	hash_add(vmap->ht, &me->hnode, key);
204 }
205 
206 static void vrf_map_del_elem(struct vrf_map_elem *me)
207 {
208 	hash_del(&me->hnode);
209 }
210 
211 static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock)
212 {
213 	spin_lock(&vmap->vmap_lock);
214 }
215 
216 static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock)
217 {
218 	spin_unlock(&vmap->vmap_lock);
219 }
220 
221 /* called with rtnl lock held */
222 static int
223 vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack)
224 {
225 	struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
226 	struct net_vrf *vrf = netdev_priv(dev);
227 	struct vrf_map_elem *new_me, *me;
228 	u32 table_id = vrf->tb_id;
229 	bool free_new_me = false;
230 	int users;
231 	int res;
232 
233 	/* we pre-allocate elements used in the spin-locked section (so that we
234 	 * keep the spinlock as short as possible).
235 	 */
236 	new_me = vrf_map_elem_alloc(GFP_KERNEL);
237 	if (!new_me)
238 		return -ENOMEM;
239 
240 	vrf_map_elem_init(new_me, table_id, dev->ifindex, 0);
241 
242 	vrf_map_lock(vmap);
243 
244 	me = vrf_map_lookup_elem(vmap, table_id);
245 	if (!me) {
246 		me = new_me;
247 		vrf_map_add_elem(vmap, me);
248 		goto link_vrf;
249 	}
250 
251 	/* we already have an entry in the vrf_map, so it means there is (at
252 	 * least) a vrf registered on the specific table.
253 	 */
254 	free_new_me = true;
255 	if (vmap->strict_mode) {
256 		/* vrfs cannot share the same table */
257 		NL_SET_ERR_MSG(extack, "Table is used by another VRF");
258 		res = -EBUSY;
259 		goto unlock;
260 	}
261 
262 link_vrf:
263 	users = ++me->users;
264 	if (users == 2)
265 		++vmap->shared_tables;
266 
267 	list_add(&vrf->me_list, &me->vrf_list);
268 
269 	res = 0;
270 
271 unlock:
272 	vrf_map_unlock(vmap);
273 
274 	/* clean-up, if needed */
275 	if (free_new_me)
276 		vrf_map_elem_free(new_me);
277 
278 	return res;
279 }
280 
281 /* called with rtnl lock held */
282 static void vrf_map_unregister_dev(struct net_device *dev)
283 {
284 	struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
285 	struct net_vrf *vrf = netdev_priv(dev);
286 	u32 table_id = vrf->tb_id;
287 	struct vrf_map_elem *me;
288 	int users;
289 
290 	vrf_map_lock(vmap);
291 
292 	me = vrf_map_lookup_elem(vmap, table_id);
293 	if (!me)
294 		goto unlock;
295 
296 	list_del(&vrf->me_list);
297 
298 	users = --me->users;
299 	if (users == 1) {
300 		--vmap->shared_tables;
301 	} else if (users == 0) {
302 		vrf_map_del_elem(me);
303 
304 		/* no one will refer to this element anymore */
305 		vrf_map_elem_free(me);
306 	}
307 
308 unlock:
309 	vrf_map_unlock(vmap);
310 }
311 
312 /* return the vrf device index associated with the table_id */
313 static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id)
314 {
315 	struct vrf_map *vmap = netns_vrf_map(net);
316 	struct vrf_map_elem *me;
317 	int ifindex;
318 
319 	vrf_map_lock(vmap);
320 
321 	if (!vmap->strict_mode) {
322 		ifindex = -EPERM;
323 		goto unlock;
324 	}
325 
326 	me = vrf_map_lookup_elem(vmap, table_id);
327 	if (!me) {
328 		ifindex = -ENODEV;
329 		goto unlock;
330 	}
331 
332 	ifindex = vrf_map_elem_get_vrf_ifindex(me);
333 
334 unlock:
335 	vrf_map_unlock(vmap);
336 
337 	return ifindex;
338 }
339 
340 /* by default VRF devices do not have a qdisc and are expected
341  * to be created with only a single queue.
342  */
343 static bool qdisc_tx_is_default(const struct net_device *dev)
344 {
345 	struct netdev_queue *txq;
346 
347 	if (dev->num_tx_queues > 1)
348 		return false;
349 
350 	txq = netdev_get_tx_queue(dev, 0);
351 
352 	return qdisc_txq_has_no_queue(txq);
353 }
354 
355 /* Local traffic destined to local address. Reinsert the packet to rx
356  * path, similar to loopback handling.
357  */
358 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
359 			  struct dst_entry *dst)
360 {
361 	unsigned int len = skb->len;
362 
363 	skb_orphan(skb);
364 
365 	skb_dst_set(skb, dst);
366 
367 	/* set pkt_type to avoid skb hitting packet taps twice -
368 	 * once on Tx and again in Rx processing
369 	 */
370 	skb->pkt_type = PACKET_LOOPBACK;
371 
372 	skb->protocol = eth_type_trans(skb, dev);
373 
374 	if (likely(__netif_rx(skb) == NET_RX_SUCCESS))
375 		dev_dstats_rx_add(dev, len);
376 	else
377 		dev_dstats_rx_dropped(dev);
378 
379 	return NETDEV_TX_OK;
380 }
381 
382 static void vrf_nf_set_untracked(struct sk_buff *skb)
383 {
384 	if (skb_get_nfct(skb) == 0)
385 		nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
386 }
387 
388 static void vrf_nf_reset_ct(struct sk_buff *skb)
389 {
390 	if (skb_get_nfct(skb) == IP_CT_UNTRACKED)
391 		nf_reset_ct(skb);
392 }
393 
394 #if IS_ENABLED(CONFIG_IPV6)
395 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
396 			     struct sk_buff *skb)
397 {
398 	int err;
399 
400 	vrf_nf_reset_ct(skb);
401 
402 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
403 		      sk, skb, NULL, skb_dst(skb)->dev, dst_output);
404 
405 	if (likely(err == 1))
406 		err = dst_output(net, sk, skb);
407 
408 	return err;
409 }
410 
411 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
412 					   struct net_device *dev)
413 {
414 	const struct ipv6hdr *iph;
415 	struct net *net = dev_net(skb->dev);
416 	struct flowi6 fl6;
417 	int ret = NET_XMIT_DROP;
418 	struct dst_entry *dst;
419 	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
420 
421 	if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
422 		goto err;
423 
424 	iph = ipv6_hdr(skb);
425 
426 	memset(&fl6, 0, sizeof(fl6));
427 	/* needed to match OIF rule */
428 	fl6.flowi6_l3mdev = dev->ifindex;
429 	fl6.flowi6_iif = LOOPBACK_IFINDEX;
430 	fl6.daddr = iph->daddr;
431 	fl6.saddr = iph->saddr;
432 	fl6.flowlabel = ip6_flowinfo(iph);
433 	fl6.flowi6_mark = skb->mark;
434 	fl6.flowi6_proto = iph->nexthdr;
435 
436 	dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL);
437 	if (IS_ERR(dst) || dst == dst_null)
438 		goto err;
439 
440 	skb_dst_drop(skb);
441 
442 	/* if dst.dev is the VRF device again this is locally originated traffic
443 	 * destined to a local address. Short circuit to Rx path.
444 	 */
445 	if (dst->dev == dev)
446 		return vrf_local_xmit(skb, dev, dst);
447 
448 	skb_dst_set(skb, dst);
449 
450 	/* strip the ethernet header added for pass through VRF device */
451 	__skb_pull(skb, skb_network_offset(skb));
452 
453 	memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
454 	ret = vrf_ip6_local_out(net, skb->sk, skb);
455 	if (unlikely(net_xmit_eval(ret)))
456 		dev->stats.tx_errors++;
457 	else
458 		ret = NET_XMIT_SUCCESS;
459 
460 	return ret;
461 err:
462 	vrf_tx_error(dev, skb);
463 	return NET_XMIT_DROP;
464 }
465 #else
466 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
467 					   struct net_device *dev)
468 {
469 	vrf_tx_error(dev, skb);
470 	return NET_XMIT_DROP;
471 }
472 #endif
473 
474 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
475 static int vrf_ip_local_out(struct net *net, struct sock *sk,
476 			    struct sk_buff *skb)
477 {
478 	int err;
479 
480 	vrf_nf_reset_ct(skb);
481 
482 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
483 		      skb, NULL, skb_dst(skb)->dev, dst_output);
484 	if (likely(err == 1))
485 		err = dst_output(net, sk, skb);
486 
487 	return err;
488 }
489 
490 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
491 					   struct net_device *vrf_dev)
492 {
493 	struct iphdr *ip4h;
494 	int ret = NET_XMIT_DROP;
495 	struct flowi4 fl4;
496 	struct net *net = dev_net(vrf_dev);
497 	struct rtable *rt;
498 
499 	if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
500 		goto err;
501 
502 	ip4h = ip_hdr(skb);
503 
504 	memset(&fl4, 0, sizeof(fl4));
505 	/* needed to match OIF rule */
506 	fl4.flowi4_l3mdev = vrf_dev->ifindex;
507 	fl4.flowi4_iif = LOOPBACK_IFINDEX;
508 	fl4.flowi4_dscp = ip4h_dscp(ip4h);
509 	fl4.flowi4_flags = FLOWI_FLAG_ANYSRC;
510 	fl4.flowi4_proto = ip4h->protocol;
511 	fl4.daddr = ip4h->daddr;
512 	fl4.saddr = ip4h->saddr;
513 
514 	rt = ip_route_output_flow(net, &fl4, NULL);
515 	if (IS_ERR(rt))
516 		goto err;
517 
518 	skb_dst_drop(skb);
519 
520 	/* if dst.dev is the VRF device again this is locally originated traffic
521 	 * destined to a local address. Short circuit to Rx path.
522 	 */
523 	if (rt->dst.dev == vrf_dev)
524 		return vrf_local_xmit(skb, vrf_dev, &rt->dst);
525 
526 	skb_dst_set(skb, &rt->dst);
527 
528 	/* strip the ethernet header added for pass through VRF device */
529 	__skb_pull(skb, skb_network_offset(skb));
530 
531 	if (!ip4h->saddr) {
532 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
533 					       RT_SCOPE_LINK);
534 	}
535 
536 	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
537 	ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
538 	if (unlikely(net_xmit_eval(ret)))
539 		vrf_dev->stats.tx_errors++;
540 	else
541 		ret = NET_XMIT_SUCCESS;
542 
543 out:
544 	return ret;
545 err:
546 	vrf_tx_error(vrf_dev, skb);
547 	goto out;
548 }
549 
550 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
551 {
552 	switch (skb->protocol) {
553 	case htons(ETH_P_IP):
554 		return vrf_process_v4_outbound(skb, dev);
555 	case htons(ETH_P_IPV6):
556 		return vrf_process_v6_outbound(skb, dev);
557 	default:
558 		vrf_tx_error(dev, skb);
559 		return NET_XMIT_DROP;
560 	}
561 }
562 
563 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
564 {
565 	unsigned int len = skb->len;
566 	netdev_tx_t ret;
567 
568 	ret = is_ip_tx_frame(skb, dev);
569 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN))
570 		dev_dstats_tx_add(dev, len);
571 	else
572 		dev_dstats_tx_dropped(dev);
573 
574 	return ret;
575 }
576 
577 static void vrf_finish_direct(struct sk_buff *skb)
578 {
579 	struct net_device *vrf_dev = skb->dev;
580 
581 	if (!list_empty(&vrf_dev->ptype_all) &&
582 	    likely(skb_headroom(skb) >= ETH_HLEN)) {
583 		struct ethhdr *eth = skb_push(skb, ETH_HLEN);
584 
585 		ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
586 		eth_zero_addr(eth->h_dest);
587 		eth->h_proto = skb->protocol;
588 
589 		rcu_read_lock_bh();
590 		dev_queue_xmit_nit(skb, vrf_dev);
591 		rcu_read_unlock_bh();
592 
593 		skb_pull(skb, ETH_HLEN);
594 	}
595 
596 	vrf_nf_reset_ct(skb);
597 }
598 
599 #if IS_ENABLED(CONFIG_IPV6)
600 /* modelled after ip6_finish_output2 */
601 static int vrf_finish_output6(struct net *net, struct sock *sk,
602 			      struct sk_buff *skb)
603 {
604 	struct dst_entry *dst = skb_dst(skb);
605 	struct net_device *dev = dst->dev;
606 	const struct in6_addr *nexthop;
607 	struct neighbour *neigh;
608 	int ret;
609 
610 	vrf_nf_reset_ct(skb);
611 
612 	skb->protocol = htons(ETH_P_IPV6);
613 	skb->dev = dev;
614 
615 	rcu_read_lock();
616 	nexthop = rt6_nexthop(dst_rt6_info(dst), &ipv6_hdr(skb)->daddr);
617 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
618 	if (unlikely(!neigh))
619 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
620 	if (!IS_ERR(neigh)) {
621 		sock_confirm_neigh(skb, neigh);
622 		ret = neigh_output(neigh, skb, false);
623 		rcu_read_unlock();
624 		return ret;
625 	}
626 	rcu_read_unlock();
627 
628 	IP6_INC_STATS(dev_net(dst->dev),
629 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
630 	kfree_skb(skb);
631 	return -EINVAL;
632 }
633 
634 /* modelled after ip6_output */
635 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
636 {
637 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
638 			    net, sk, skb, NULL, skb_dst(skb)->dev,
639 			    vrf_finish_output6,
640 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
641 }
642 
643 /* set dst on skb to send packet to us via dev_xmit path. Allows
644  * packet to go through device based features such as qdisc, netfilter
645  * hooks and packet sockets with skb->dev set to vrf device.
646  */
647 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
648 					    struct sk_buff *skb)
649 {
650 	struct net_vrf *vrf = netdev_priv(vrf_dev);
651 	struct rt6_info *rt6;
652 
653 	rt6 = vrf->rt6;
654 	dst_hold(&rt6->dst);
655 
656 	skb_dst_drop(skb);
657 	skb_dst_set(skb, &rt6->dst);
658 
659 	return skb;
660 }
661 
662 static int vrf_output6_direct_finish(struct net *net, struct sock *sk,
663 				     struct sk_buff *skb)
664 {
665 	vrf_finish_direct(skb);
666 
667 	return vrf_ip6_local_out(net, sk, skb);
668 }
669 
670 static int vrf_output6_direct(struct net *net, struct sock *sk,
671 			      struct sk_buff *skb)
672 {
673 	int err = 1;
674 
675 	skb->protocol = htons(ETH_P_IPV6);
676 
677 	if (!(IPCB(skb)->flags & IPSKB_REROUTED))
678 		err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
679 			      NULL, skb->dev, vrf_output6_direct_finish);
680 
681 	if (likely(err == 1))
682 		vrf_finish_direct(skb);
683 
684 	return err;
685 }
686 
687 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk,
688 				     struct sk_buff *skb)
689 {
690 	int err;
691 
692 	err = vrf_output6_direct(net, sk, skb);
693 	if (likely(err == 1))
694 		err = vrf_ip6_local_out(net, sk, skb);
695 
696 	return err;
697 }
698 
699 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
700 					  struct sock *sk,
701 					  struct sk_buff *skb)
702 {
703 	struct net *net = dev_net(vrf_dev);
704 	int err;
705 
706 	skb->dev = vrf_dev;
707 
708 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
709 		      skb, NULL, vrf_dev, vrf_ip6_out_direct_finish);
710 
711 	if (likely(err == 1))
712 		err = vrf_output6_direct(net, sk, skb);
713 
714 	if (likely(err == 1))
715 		return skb;
716 
717 	return NULL;
718 }
719 
720 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
721 				   struct sock *sk,
722 				   struct sk_buff *skb)
723 {
724 	/* don't divert link scope packets */
725 	if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
726 		return skb;
727 
728 	vrf_nf_set_untracked(skb);
729 
730 	if (qdisc_tx_is_default(vrf_dev) ||
731 	    IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
732 		return vrf_ip6_out_direct(vrf_dev, sk, skb);
733 
734 	return vrf_ip6_out_redirect(vrf_dev, skb);
735 }
736 
737 /* holding rtnl */
738 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
739 {
740 	struct rt6_info *rt6 = vrf->rt6;
741 
742 	if (rt6) {
743 		dst_dev_put(&rt6->dst);
744 		dst_release(&rt6->dst);
745 	}
746 }
747 
748 static int vrf_rt6_create(struct net_device *dev)
749 {
750 	int flags = DST_NOPOLICY | DST_NOXFRM;
751 	struct net_vrf *vrf = netdev_priv(dev);
752 	struct net *net = dev_net(dev);
753 	struct rt6_info *rt6;
754 	int rc = -ENOMEM;
755 
756 	/* IPv6 can be CONFIG enabled and then disabled runtime */
757 	if (!ipv6_mod_enabled())
758 		return 0;
759 
760 	vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
761 	if (!vrf->fib6_table)
762 		goto out;
763 
764 	/* create a dst for routing packets out a VRF device */
765 	rt6 = ip6_dst_alloc(net, dev, flags);
766 	if (!rt6)
767 		goto out;
768 
769 	rt6->dst.output	= vrf_output6;
770 
771 	vrf->rt6 = rt6;
772 
773 	rc = 0;
774 out:
775 	return rc;
776 }
777 #else
778 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
779 				   struct sock *sk,
780 				   struct sk_buff *skb)
781 {
782 	return skb;
783 }
784 
785 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
786 {
787 }
788 
789 static int vrf_rt6_create(struct net_device *dev)
790 {
791 	return 0;
792 }
793 #endif
794 
795 /* modelled after ip_finish_output2 */
796 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
797 {
798 	struct dst_entry *dst = skb_dst(skb);
799 	struct rtable *rt = dst_rtable(dst);
800 	struct net_device *dev = dst->dev;
801 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
802 	struct neighbour *neigh;
803 	bool is_v6gw = false;
804 
805 	vrf_nf_reset_ct(skb);
806 
807 	/* Be paranoid, rather than too clever. */
808 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
809 		skb = skb_expand_head(skb, hh_len);
810 		if (!skb) {
811 			dev->stats.tx_errors++;
812 			return -ENOMEM;
813 		}
814 	}
815 
816 	rcu_read_lock();
817 
818 	neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
819 	if (!IS_ERR(neigh)) {
820 		int ret;
821 
822 		sock_confirm_neigh(skb, neigh);
823 		/* if crossing protocols, can not use the cached header */
824 		ret = neigh_output(neigh, skb, is_v6gw);
825 		rcu_read_unlock();
826 		return ret;
827 	}
828 
829 	rcu_read_unlock();
830 	vrf_tx_error(skb->dev, skb);
831 	return -EINVAL;
832 }
833 
834 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
835 {
836 	struct net_device *dev = skb_dst(skb)->dev;
837 
838 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
839 
840 	skb->dev = dev;
841 	skb->protocol = htons(ETH_P_IP);
842 
843 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
844 			    net, sk, skb, NULL, dev,
845 			    vrf_finish_output,
846 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
847 }
848 
849 /* set dst on skb to send packet to us via dev_xmit path. Allows
850  * packet to go through device based features such as qdisc, netfilter
851  * hooks and packet sockets with skb->dev set to vrf device.
852  */
853 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
854 					   struct sk_buff *skb)
855 {
856 	struct net_vrf *vrf = netdev_priv(vrf_dev);
857 	struct rtable *rth;
858 
859 	rth = vrf->rth;
860 	dst_hold(&rth->dst);
861 
862 	skb_dst_drop(skb);
863 	skb_dst_set(skb, &rth->dst);
864 
865 	return skb;
866 }
867 
868 static int vrf_output_direct_finish(struct net *net, struct sock *sk,
869 				    struct sk_buff *skb)
870 {
871 	vrf_finish_direct(skb);
872 
873 	return vrf_ip_local_out(net, sk, skb);
874 }
875 
876 static int vrf_output_direct(struct net *net, struct sock *sk,
877 			     struct sk_buff *skb)
878 {
879 	int err = 1;
880 
881 	skb->protocol = htons(ETH_P_IP);
882 
883 	if (!(IPCB(skb)->flags & IPSKB_REROUTED))
884 		err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
885 			      NULL, skb->dev, vrf_output_direct_finish);
886 
887 	if (likely(err == 1))
888 		vrf_finish_direct(skb);
889 
890 	return err;
891 }
892 
893 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk,
894 				    struct sk_buff *skb)
895 {
896 	int err;
897 
898 	err = vrf_output_direct(net, sk, skb);
899 	if (likely(err == 1))
900 		err = vrf_ip_local_out(net, sk, skb);
901 
902 	return err;
903 }
904 
905 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
906 					 struct sock *sk,
907 					 struct sk_buff *skb)
908 {
909 	struct net *net = dev_net(vrf_dev);
910 	int err;
911 
912 	skb->dev = vrf_dev;
913 
914 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
915 		      skb, NULL, vrf_dev, vrf_ip_out_direct_finish);
916 
917 	if (likely(err == 1))
918 		err = vrf_output_direct(net, sk, skb);
919 
920 	if (likely(err == 1))
921 		return skb;
922 
923 	return NULL;
924 }
925 
926 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
927 				  struct sock *sk,
928 				  struct sk_buff *skb)
929 {
930 	/* don't divert multicast or local broadcast */
931 	if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
932 	    ipv4_is_lbcast(ip_hdr(skb)->daddr))
933 		return skb;
934 
935 	vrf_nf_set_untracked(skb);
936 
937 	if (qdisc_tx_is_default(vrf_dev) ||
938 	    IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
939 		return vrf_ip_out_direct(vrf_dev, sk, skb);
940 
941 	return vrf_ip_out_redirect(vrf_dev, skb);
942 }
943 
944 /* called with rcu lock held */
945 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
946 				  struct sock *sk,
947 				  struct sk_buff *skb,
948 				  u16 proto)
949 {
950 	switch (proto) {
951 	case AF_INET:
952 		return vrf_ip_out(vrf_dev, sk, skb);
953 	case AF_INET6:
954 		return vrf_ip6_out(vrf_dev, sk, skb);
955 	}
956 
957 	return skb;
958 }
959 
960 /* holding rtnl */
961 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
962 {
963 	struct rtable *rth = vrf->rth;
964 
965 	dst_dev_put(&rth->dst);
966 	dst_release(&rth->dst);
967 }
968 
969 static int vrf_rtable_create(struct net_device *dev)
970 {
971 	struct net_vrf *vrf = netdev_priv(dev);
972 	struct rtable *rth;
973 
974 	if (!fib_new_table(dev_net(dev), vrf->tb_id))
975 		return -ENOMEM;
976 
977 	/* create a dst for routing packets out through a VRF device */
978 	rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1);
979 	if (!rth)
980 		return -ENOMEM;
981 
982 	rth->dst.output	= vrf_output;
983 
984 	vrf->rth = rth;
985 
986 	return 0;
987 }
988 
989 /**************************** device handling ********************/
990 
991 /* cycle interface to flush neighbor cache and move routes across tables */
992 static void cycle_netdev(struct net_device *dev,
993 			 struct netlink_ext_ack *extack)
994 {
995 	unsigned int flags = dev->flags;
996 	int ret;
997 
998 	if (!netif_running(dev))
999 		return;
1000 
1001 	ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
1002 	if (ret >= 0)
1003 		ret = dev_change_flags(dev, flags, extack);
1004 
1005 	if (ret < 0) {
1006 		netdev_err(dev,
1007 			   "Failed to cycle device %s; route tables might be wrong!\n",
1008 			   dev->name);
1009 	}
1010 }
1011 
1012 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1013 			    struct netlink_ext_ack *extack)
1014 {
1015 	int ret;
1016 
1017 	/* do not allow loopback device to be enslaved to a VRF.
1018 	 * The vrf device acts as the loopback for the vrf.
1019 	 */
1020 	if (port_dev == dev_net(dev)->loopback_dev) {
1021 		NL_SET_ERR_MSG(extack,
1022 			       "Can not enslave loopback device to a VRF");
1023 		return -EOPNOTSUPP;
1024 	}
1025 
1026 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
1027 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
1028 	if (ret < 0)
1029 		goto err;
1030 
1031 	cycle_netdev(port_dev, extack);
1032 
1033 	return 0;
1034 
1035 err:
1036 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1037 	synchronize_net();
1038 	return ret;
1039 }
1040 
1041 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1042 			 struct netlink_ext_ack *extack)
1043 {
1044 	if (netif_is_l3_master(port_dev)) {
1045 		NL_SET_ERR_MSG(extack,
1046 			       "Can not enslave an L3 master device to a VRF");
1047 		return -EINVAL;
1048 	}
1049 
1050 	if (netif_is_l3_slave(port_dev))
1051 		return -EINVAL;
1052 
1053 	return do_vrf_add_slave(dev, port_dev, extack);
1054 }
1055 
1056 /* inverse of do_vrf_add_slave */
1057 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev,
1058 			    bool needs_sync)
1059 {
1060 	netdev_upper_dev_unlink(port_dev, dev);
1061 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1062 	/* Make sure that concurrent RCU readers that identified the device
1063 	 * as a VRF port see a VRF master or no master at all.
1064 	 */
1065 	if (needs_sync)
1066 		synchronize_net();
1067 
1068 	cycle_netdev(port_dev, NULL);
1069 
1070 	return 0;
1071 }
1072 
1073 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1074 {
1075 	return do_vrf_del_slave(dev, port_dev, true);
1076 }
1077 
1078 static void vrf_dev_uninit(struct net_device *dev)
1079 {
1080 	struct net_vrf *vrf = netdev_priv(dev);
1081 
1082 	vrf_rtable_release(dev, vrf);
1083 	vrf_rt6_release(dev, vrf);
1084 }
1085 
1086 static int vrf_dev_init(struct net_device *dev)
1087 {
1088 	struct net_vrf *vrf = netdev_priv(dev);
1089 
1090 	/* create the default dst which points back to us */
1091 	if (vrf_rtable_create(dev) != 0)
1092 		goto out_nomem;
1093 
1094 	if (vrf_rt6_create(dev) != 0)
1095 		goto out_rth;
1096 
1097 	dev->flags = IFF_MASTER | IFF_NOARP;
1098 
1099 	/* similarly, oper state is irrelevant; set to up to avoid confusion */
1100 	dev->operstate = IF_OPER_UP;
1101 	netdev_lockdep_set_classes(dev);
1102 	return 0;
1103 
1104 out_rth:
1105 	vrf_rtable_release(dev, vrf);
1106 out_nomem:
1107 	return -ENOMEM;
1108 }
1109 
1110 static const struct net_device_ops vrf_netdev_ops = {
1111 	.ndo_init		= vrf_dev_init,
1112 	.ndo_uninit		= vrf_dev_uninit,
1113 	.ndo_start_xmit		= vrf_xmit,
1114 	.ndo_set_mac_address	= eth_mac_addr,
1115 	.ndo_add_slave		= vrf_add_slave,
1116 	.ndo_del_slave		= vrf_del_slave,
1117 };
1118 
1119 static u32 vrf_fib_table(const struct net_device *dev)
1120 {
1121 	struct net_vrf *vrf = netdev_priv(dev);
1122 
1123 	return vrf->tb_id;
1124 }
1125 
1126 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
1127 {
1128 	kfree_skb(skb);
1129 	return 0;
1130 }
1131 
1132 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
1133 				      struct sk_buff *skb,
1134 				      struct net_device *dev)
1135 {
1136 	struct net *net = dev_net(dev);
1137 
1138 	if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
1139 		skb = NULL;    /* kfree_skb(skb) handled by nf code */
1140 
1141 	return skb;
1142 }
1143 
1144 static int vrf_prepare_mac_header(struct sk_buff *skb,
1145 				  struct net_device *vrf_dev, u16 proto)
1146 {
1147 	struct ethhdr *eth;
1148 	int err;
1149 
1150 	/* in general, we do not know if there is enough space in the head of
1151 	 * the packet for hosting the mac header.
1152 	 */
1153 	err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev));
1154 	if (unlikely(err))
1155 		/* no space in the skb head */
1156 		return -ENOBUFS;
1157 
1158 	__skb_push(skb, ETH_HLEN);
1159 	eth = (struct ethhdr *)skb->data;
1160 
1161 	skb_reset_mac_header(skb);
1162 	skb_reset_mac_len(skb);
1163 
1164 	/* we set the ethernet destination and the source addresses to the
1165 	 * address of the VRF device.
1166 	 */
1167 	ether_addr_copy(eth->h_dest, vrf_dev->dev_addr);
1168 	ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
1169 	eth->h_proto = htons(proto);
1170 
1171 	/* the destination address of the Ethernet frame corresponds to the
1172 	 * address set on the VRF interface; therefore, the packet is intended
1173 	 * to be processed locally.
1174 	 */
1175 	skb->protocol = eth->h_proto;
1176 	skb->pkt_type = PACKET_HOST;
1177 
1178 	skb_postpush_rcsum(skb, skb->data, ETH_HLEN);
1179 
1180 	skb_pull_inline(skb, ETH_HLEN);
1181 
1182 	return 0;
1183 }
1184 
1185 /* prepare and add the mac header to the packet if it was not set previously.
1186  * In this way, packet sniffers such as tcpdump can parse the packet correctly.
1187  * If the mac header was already set, the original mac header is left
1188  * untouched and the function returns immediately.
1189  */
1190 static int vrf_add_mac_header_if_unset(struct sk_buff *skb,
1191 				       struct net_device *vrf_dev,
1192 				       u16 proto, struct net_device *orig_dev)
1193 {
1194 	if (skb_mac_header_was_set(skb) && dev_has_header(orig_dev))
1195 		return 0;
1196 
1197 	return vrf_prepare_mac_header(skb, vrf_dev, proto);
1198 }
1199 
1200 #if IS_ENABLED(CONFIG_IPV6)
1201 /* neighbor handling is done with actual device; do not want
1202  * to flip skb->dev for those ndisc packets. This really fails
1203  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1204  * a start.
1205  */
1206 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
1207 {
1208 	const struct ipv6hdr *iph = ipv6_hdr(skb);
1209 	bool rc = false;
1210 
1211 	if (iph->nexthdr == NEXTHDR_ICMP) {
1212 		const struct icmp6hdr *icmph;
1213 		struct icmp6hdr _icmph;
1214 
1215 		icmph = skb_header_pointer(skb, sizeof(*iph),
1216 					   sizeof(_icmph), &_icmph);
1217 		if (!icmph)
1218 			goto out;
1219 
1220 		switch (icmph->icmp6_type) {
1221 		case NDISC_ROUTER_SOLICITATION:
1222 		case NDISC_ROUTER_ADVERTISEMENT:
1223 		case NDISC_NEIGHBOUR_SOLICITATION:
1224 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
1225 		case NDISC_REDIRECT:
1226 			rc = true;
1227 			break;
1228 		}
1229 	}
1230 
1231 out:
1232 	return rc;
1233 }
1234 
1235 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
1236 					     const struct net_device *dev,
1237 					     struct flowi6 *fl6,
1238 					     int ifindex,
1239 					     const struct sk_buff *skb,
1240 					     int flags)
1241 {
1242 	struct net_vrf *vrf = netdev_priv(dev);
1243 
1244 	return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
1245 }
1246 
1247 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1248 			      int ifindex)
1249 {
1250 	const struct ipv6hdr *iph = ipv6_hdr(skb);
1251 	struct flowi6 fl6 = {
1252 		.flowi6_iif     = ifindex,
1253 		.flowi6_mark    = skb->mark,
1254 		.flowi6_proto   = iph->nexthdr,
1255 		.daddr          = iph->daddr,
1256 		.saddr          = iph->saddr,
1257 		.flowlabel      = ip6_flowinfo(iph),
1258 	};
1259 	struct net *net = dev_net(vrf_dev);
1260 	struct rt6_info *rt6;
1261 
1262 	skb_dst_drop(skb);
1263 
1264 	rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
1265 				   RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1266 	if (unlikely(!rt6))
1267 		return;
1268 
1269 	if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1270 		return;
1271 
1272 	skb_dst_set(skb, &rt6->dst);
1273 }
1274 
1275 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1276 				   struct sk_buff *skb)
1277 {
1278 	int orig_iif = skb->skb_iif;
1279 	bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1280 	bool is_ndisc = ipv6_ndisc_frame(skb);
1281 
1282 	/* loopback, multicast & non-ND link-local traffic; do not push through
1283 	 * packet taps again. Reset pkt_type for upper layers to process skb.
1284 	 * For non-loopback strict packets, determine the dst using the original
1285 	 * ifindex.
1286 	 */
1287 	if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
1288 		skb->dev = vrf_dev;
1289 		skb->skb_iif = vrf_dev->ifindex;
1290 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1291 
1292 		if (skb->pkt_type == PACKET_LOOPBACK)
1293 			skb->pkt_type = PACKET_HOST;
1294 		else
1295 			vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1296 
1297 		goto out;
1298 	}
1299 
1300 	/* if packet is NDISC then keep the ingress interface */
1301 	if (!is_ndisc) {
1302 		struct net_device *orig_dev = skb->dev;
1303 
1304 		dev_dstats_rx_add(vrf_dev, skb->len);
1305 		skb->dev = vrf_dev;
1306 		skb->skb_iif = vrf_dev->ifindex;
1307 
1308 		if (!list_empty(&vrf_dev->ptype_all)) {
1309 			int err;
1310 
1311 			err = vrf_add_mac_header_if_unset(skb, vrf_dev,
1312 							  ETH_P_IPV6,
1313 							  orig_dev);
1314 			if (likely(!err)) {
1315 				skb_push(skb, skb->mac_len);
1316 				dev_queue_xmit_nit(skb, vrf_dev);
1317 				skb_pull(skb, skb->mac_len);
1318 			}
1319 		}
1320 
1321 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1322 	}
1323 
1324 	if (need_strict)
1325 		vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1326 
1327 	skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1328 out:
1329 	return skb;
1330 }
1331 
1332 #else
1333 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1334 				   struct sk_buff *skb)
1335 {
1336 	return skb;
1337 }
1338 #endif
1339 
1340 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1341 				  struct sk_buff *skb)
1342 {
1343 	struct net_device *orig_dev = skb->dev;
1344 
1345 	skb->dev = vrf_dev;
1346 	skb->skb_iif = vrf_dev->ifindex;
1347 	IPCB(skb)->flags |= IPSKB_L3SLAVE;
1348 
1349 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1350 		goto out;
1351 
1352 	/* loopback traffic; do not push through packet taps again.
1353 	 * Reset pkt_type for upper layers to process skb
1354 	 */
1355 	if (skb->pkt_type == PACKET_LOOPBACK) {
1356 		skb->pkt_type = PACKET_HOST;
1357 		goto out;
1358 	}
1359 
1360 	dev_dstats_rx_add(vrf_dev, skb->len);
1361 
1362 	if (!list_empty(&vrf_dev->ptype_all)) {
1363 		int err;
1364 
1365 		err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP,
1366 						  orig_dev);
1367 		if (likely(!err)) {
1368 			skb_push(skb, skb->mac_len);
1369 			dev_queue_xmit_nit(skb, vrf_dev);
1370 			skb_pull(skb, skb->mac_len);
1371 		}
1372 	}
1373 
1374 	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1375 out:
1376 	return skb;
1377 }
1378 
1379 /* called with rcu lock held */
1380 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1381 				  struct sk_buff *skb,
1382 				  u16 proto)
1383 {
1384 	switch (proto) {
1385 	case AF_INET:
1386 		return vrf_ip_rcv(vrf_dev, skb);
1387 	case AF_INET6:
1388 		return vrf_ip6_rcv(vrf_dev, skb);
1389 	}
1390 
1391 	return skb;
1392 }
1393 
1394 #if IS_ENABLED(CONFIG_IPV6)
1395 /* send to link-local or multicast address via interface enslaved to
1396  * VRF device. Force lookup to VRF table without changing flow struct
1397  * Note: Caller to this function must hold rcu_read_lock() and no refcnt
1398  * is taken on the dst by this function.
1399  */
1400 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1401 					      struct flowi6 *fl6)
1402 {
1403 	struct net *net = dev_net(dev);
1404 	int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;
1405 	struct dst_entry *dst = NULL;
1406 	struct rt6_info *rt;
1407 
1408 	/* VRF device does not have a link-local address and
1409 	 * sending packets to link-local or mcast addresses over
1410 	 * a VRF device does not make sense
1411 	 */
1412 	if (fl6->flowi6_oif == dev->ifindex) {
1413 		dst = &net->ipv6.ip6_null_entry->dst;
1414 		return dst;
1415 	}
1416 
1417 	if (!ipv6_addr_any(&fl6->saddr))
1418 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1419 
1420 	rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1421 	if (rt)
1422 		dst = &rt->dst;
1423 
1424 	return dst;
1425 }
1426 #endif
1427 
1428 static const struct l3mdev_ops vrf_l3mdev_ops = {
1429 	.l3mdev_fib_table	= vrf_fib_table,
1430 	.l3mdev_l3_rcv		= vrf_l3_rcv,
1431 	.l3mdev_l3_out		= vrf_l3_out,
1432 #if IS_ENABLED(CONFIG_IPV6)
1433 	.l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1434 #endif
1435 };
1436 
1437 static void vrf_get_drvinfo(struct net_device *dev,
1438 			    struct ethtool_drvinfo *info)
1439 {
1440 	strscpy(info->driver, DRV_NAME, sizeof(info->driver));
1441 	strscpy(info->version, DRV_VERSION, sizeof(info->version));
1442 }
1443 
1444 static const struct ethtool_ops vrf_ethtool_ops = {
1445 	.get_drvinfo	= vrf_get_drvinfo,
1446 };
1447 
1448 static inline size_t vrf_fib_rule_nl_size(void)
1449 {
1450 	size_t sz;
1451 
1452 	sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1453 	sz += nla_total_size(sizeof(u8));	/* FRA_L3MDEV */
1454 	sz += nla_total_size(sizeof(u32));	/* FRA_PRIORITY */
1455 	sz += nla_total_size(sizeof(u8));       /* FRA_PROTOCOL */
1456 
1457 	return sz;
1458 }
1459 
1460 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1461 {
1462 	struct fib_rule_hdr *frh;
1463 	struct nlmsghdr *nlh;
1464 	struct sk_buff *skb;
1465 	int err;
1466 
1467 	if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&
1468 	    !ipv6_mod_enabled())
1469 		return 0;
1470 
1471 	skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1472 	if (!skb)
1473 		return -ENOMEM;
1474 
1475 	nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1476 	if (!nlh)
1477 		goto nla_put_failure;
1478 
1479 	/* rule only needs to appear once */
1480 	nlh->nlmsg_flags |= NLM_F_EXCL;
1481 
1482 	frh = nlmsg_data(nlh);
1483 	memset(frh, 0, sizeof(*frh));
1484 	frh->family = family;
1485 	frh->action = FR_ACT_TO_TBL;
1486 
1487 	if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1488 		goto nla_put_failure;
1489 
1490 	if (nla_put_u8(skb, FRA_L3MDEV, 1))
1491 		goto nla_put_failure;
1492 
1493 	if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1494 		goto nla_put_failure;
1495 
1496 	nlmsg_end(skb, nlh);
1497 
1498 	if (add_it) {
1499 		err = fib_newrule(dev_net(dev), skb, nlh, NULL, true);
1500 		if (err == -EEXIST)
1501 			err = 0;
1502 	} else {
1503 		err = fib_delrule(dev_net(dev), skb, nlh, NULL, true);
1504 		if (err == -ENOENT)
1505 			err = 0;
1506 	}
1507 	nlmsg_free(skb);
1508 
1509 	return err;
1510 
1511 nla_put_failure:
1512 	nlmsg_free(skb);
1513 
1514 	return -EMSGSIZE;
1515 }
1516 
1517 static int vrf_add_fib_rules(const struct net_device *dev)
1518 {
1519 	int err;
1520 
1521 	err = vrf_fib_rule(dev, AF_INET,  true);
1522 	if (err < 0)
1523 		goto out_err;
1524 
1525 	err = vrf_fib_rule(dev, AF_INET6, true);
1526 	if (err < 0)
1527 		goto ipv6_err;
1528 
1529 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1530 	err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1531 	if (err < 0)
1532 		goto ipmr_err;
1533 #endif
1534 
1535 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1536 	err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1537 	if (err < 0)
1538 		goto ip6mr_err;
1539 #endif
1540 
1541 	return 0;
1542 
1543 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1544 ip6mr_err:
1545 	vrf_fib_rule(dev, RTNL_FAMILY_IPMR,  false);
1546 #endif
1547 
1548 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1549 ipmr_err:
1550 	vrf_fib_rule(dev, AF_INET6,  false);
1551 #endif
1552 
1553 ipv6_err:
1554 	vrf_fib_rule(dev, AF_INET,  false);
1555 
1556 out_err:
1557 	netdev_err(dev, "Failed to add FIB rules.\n");
1558 	return err;
1559 }
1560 
1561 static void vrf_setup(struct net_device *dev)
1562 {
1563 	ether_setup(dev);
1564 
1565 	/* Initialize the device structure. */
1566 	dev->netdev_ops = &vrf_netdev_ops;
1567 	dev->l3mdev_ops = &vrf_l3mdev_ops;
1568 	dev->ethtool_ops = &vrf_ethtool_ops;
1569 	dev->needs_free_netdev = true;
1570 
1571 	/* Fill in device structure with ethernet-generic values. */
1572 	eth_hw_addr_random(dev);
1573 
1574 	/* don't acquire vrf device's netif_tx_lock when transmitting */
1575 	dev->lltx = true;
1576 
1577 	/* don't allow vrf devices to change network namespaces. */
1578 	dev->netns_immutable = true;
1579 
1580 	/* does not make sense for a VLAN to be added to a vrf device */
1581 	dev->features   |= NETIF_F_VLAN_CHALLENGED;
1582 
1583 	/* enable offload features */
1584 	dev->features   |= NETIF_F_GSO_SOFTWARE;
1585 	dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1586 	dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1587 
1588 	dev->hw_features = dev->features;
1589 	dev->hw_enc_features = dev->features;
1590 
1591 	/* default to no qdisc; user can add if desired */
1592 	dev->priv_flags |= IFF_NO_QUEUE;
1593 	dev->priv_flags |= IFF_NO_RX_HANDLER;
1594 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1595 
1596 	/* VRF devices do not care about MTU, but if the MTU is set
1597 	 * too low then the ipv4 and ipv6 protocols are disabled
1598 	 * which breaks networking.
1599 	 */
1600 	dev->min_mtu = IPV6_MIN_MTU;
1601 	dev->max_mtu = IP6_MAX_MTU;
1602 	dev->mtu = dev->max_mtu;
1603 
1604 	dev->pcpu_stat_type = NETDEV_PCPU_STAT_DSTATS;
1605 }
1606 
1607 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1608 			struct netlink_ext_ack *extack)
1609 {
1610 	if (tb[IFLA_ADDRESS]) {
1611 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1612 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1613 			return -EINVAL;
1614 		}
1615 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1616 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1617 			return -EADDRNOTAVAIL;
1618 		}
1619 	}
1620 	return 0;
1621 }
1622 
1623 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1624 {
1625 	struct net_device *port_dev;
1626 	struct list_head *iter;
1627 
1628 	netdev_for_each_lower_dev(dev, port_dev, iter)
1629 		do_vrf_del_slave(dev, port_dev, false);
1630 
1631 	vrf_map_unregister_dev(dev);
1632 
1633 	unregister_netdevice_queue(dev, head);
1634 }
1635 
1636 static int vrf_newlink(struct net_device *dev,
1637 		       struct rtnl_newlink_params *params,
1638 		       struct netlink_ext_ack *extack)
1639 {
1640 	struct net_vrf *vrf = netdev_priv(dev);
1641 	struct nlattr **data = params->data;
1642 	struct netns_vrf *nn_vrf;
1643 	bool *add_fib_rules;
1644 	struct net *net;
1645 	int err;
1646 
1647 	if (!data || !data[IFLA_VRF_TABLE]) {
1648 		NL_SET_ERR_MSG(extack, "VRF table id is missing");
1649 		return -EINVAL;
1650 	}
1651 
1652 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1653 	if (vrf->tb_id == RT_TABLE_UNSPEC) {
1654 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1655 				    "Invalid VRF table id");
1656 		return -EINVAL;
1657 	}
1658 
1659 	dev->priv_flags |= IFF_L3MDEV_MASTER;
1660 
1661 	err = register_netdevice(dev);
1662 	if (err)
1663 		goto out;
1664 
1665 	/* mapping between table_id and vrf;
1666 	 * note: such binding could not be done in the dev init function
1667 	 * because dev->ifindex id is not available yet.
1668 	 */
1669 	vrf->ifindex = dev->ifindex;
1670 
1671 	err = vrf_map_register_dev(dev, extack);
1672 	if (err) {
1673 		unregister_netdevice(dev);
1674 		goto out;
1675 	}
1676 
1677 	net = dev_net(dev);
1678 	nn_vrf = net_generic(net, vrf_net_id);
1679 
1680 	add_fib_rules = &nn_vrf->add_fib_rules;
1681 	if (*add_fib_rules) {
1682 		err = vrf_add_fib_rules(dev);
1683 		if (err) {
1684 			vrf_map_unregister_dev(dev);
1685 			unregister_netdevice(dev);
1686 			goto out;
1687 		}
1688 		*add_fib_rules = false;
1689 	}
1690 
1691 out:
1692 	return err;
1693 }
1694 
1695 static size_t vrf_nl_getsize(const struct net_device *dev)
1696 {
1697 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
1698 }
1699 
1700 static int vrf_fillinfo(struct sk_buff *skb,
1701 			const struct net_device *dev)
1702 {
1703 	struct net_vrf *vrf = netdev_priv(dev);
1704 
1705 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1706 }
1707 
1708 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1709 				 const struct net_device *slave_dev)
1710 {
1711 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
1712 }
1713 
1714 static int vrf_fill_slave_info(struct sk_buff *skb,
1715 			       const struct net_device *vrf_dev,
1716 			       const struct net_device *slave_dev)
1717 {
1718 	struct net_vrf *vrf = netdev_priv(vrf_dev);
1719 
1720 	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1721 		return -EMSGSIZE;
1722 
1723 	return 0;
1724 }
1725 
1726 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1727 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
1728 };
1729 
1730 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1731 	.kind		= DRV_NAME,
1732 	.priv_size	= sizeof(struct net_vrf),
1733 
1734 	.get_size	= vrf_nl_getsize,
1735 	.policy		= vrf_nl_policy,
1736 	.validate	= vrf_validate,
1737 	.fill_info	= vrf_fillinfo,
1738 
1739 	.get_slave_size  = vrf_get_slave_size,
1740 	.fill_slave_info = vrf_fill_slave_info,
1741 
1742 	.newlink	= vrf_newlink,
1743 	.dellink	= vrf_dellink,
1744 	.setup		= vrf_setup,
1745 	.maxtype	= IFLA_VRF_MAX,
1746 };
1747 
1748 static int vrf_device_event(struct notifier_block *unused,
1749 			    unsigned long event, void *ptr)
1750 {
1751 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1752 
1753 	/* only care about unregister events to drop slave references */
1754 	if (event == NETDEV_UNREGISTER) {
1755 		struct net_device *vrf_dev;
1756 
1757 		if (!netif_is_l3_slave(dev))
1758 			goto out;
1759 
1760 		vrf_dev = netdev_master_upper_dev_get(dev);
1761 		do_vrf_del_slave(vrf_dev, dev, false);
1762 	}
1763 out:
1764 	return NOTIFY_DONE;
1765 }
1766 
1767 static struct notifier_block vrf_notifier_block __read_mostly = {
1768 	.notifier_call = vrf_device_event,
1769 };
1770 
1771 static int vrf_map_init(struct vrf_map *vmap)
1772 {
1773 	spin_lock_init(&vmap->vmap_lock);
1774 	hash_init(vmap->ht);
1775 
1776 	vmap->strict_mode = false;
1777 
1778 	return 0;
1779 }
1780 
1781 #ifdef CONFIG_SYSCTL
1782 static bool vrf_strict_mode(struct vrf_map *vmap)
1783 {
1784 	bool strict_mode;
1785 
1786 	vrf_map_lock(vmap);
1787 	strict_mode = vmap->strict_mode;
1788 	vrf_map_unlock(vmap);
1789 
1790 	return strict_mode;
1791 }
1792 
1793 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode)
1794 {
1795 	bool *cur_mode;
1796 	int res = 0;
1797 
1798 	vrf_map_lock(vmap);
1799 
1800 	cur_mode = &vmap->strict_mode;
1801 	if (*cur_mode == new_mode)
1802 		goto unlock;
1803 
1804 	if (*cur_mode) {
1805 		/* disable strict mode */
1806 		*cur_mode = false;
1807 	} else {
1808 		if (vmap->shared_tables) {
1809 			/* we cannot allow strict_mode because there are some
1810 			 * vrfs that share one or more tables.
1811 			 */
1812 			res = -EBUSY;
1813 			goto unlock;
1814 		}
1815 
1816 		/* no tables are shared among vrfs, so we can go back
1817 		 * to 1:1 association between a vrf with its table.
1818 		 */
1819 		*cur_mode = true;
1820 	}
1821 
1822 unlock:
1823 	vrf_map_unlock(vmap);
1824 
1825 	return res;
1826 }
1827 
1828 static int vrf_shared_table_handler(const struct ctl_table *table, int write,
1829 				    void *buffer, size_t *lenp, loff_t *ppos)
1830 {
1831 	struct net *net = (struct net *)table->extra1;
1832 	struct vrf_map *vmap = netns_vrf_map(net);
1833 	int proc_strict_mode = 0;
1834 	struct ctl_table tmp = {
1835 		.procname	= table->procname,
1836 		.data		= &proc_strict_mode,
1837 		.maxlen		= sizeof(int),
1838 		.mode		= table->mode,
1839 		.extra1		= SYSCTL_ZERO,
1840 		.extra2		= SYSCTL_ONE,
1841 	};
1842 	int ret;
1843 
1844 	if (!write)
1845 		proc_strict_mode = vrf_strict_mode(vmap);
1846 
1847 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1848 
1849 	if (write && ret == 0)
1850 		ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode);
1851 
1852 	return ret;
1853 }
1854 
1855 static const struct ctl_table vrf_table[] = {
1856 	{
1857 		.procname	= "strict_mode",
1858 		.data		= NULL,
1859 		.maxlen		= sizeof(int),
1860 		.mode		= 0644,
1861 		.proc_handler	= vrf_shared_table_handler,
1862 		/* set by the vrf_netns_init */
1863 		.extra1		= NULL,
1864 	},
1865 };
1866 
1867 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
1868 {
1869 	struct ctl_table *table;
1870 
1871 	table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL);
1872 	if (!table)
1873 		return -ENOMEM;
1874 
1875 	/* init the extra1 parameter with the reference to current netns */
1876 	table[0].extra1 = net;
1877 
1878 	nn_vrf->ctl_hdr = register_net_sysctl_sz(net, "net/vrf", table,
1879 						 ARRAY_SIZE(vrf_table));
1880 	if (!nn_vrf->ctl_hdr) {
1881 		kfree(table);
1882 		return -ENOMEM;
1883 	}
1884 
1885 	return 0;
1886 }
1887 
1888 static void vrf_netns_exit_sysctl(struct net *net)
1889 {
1890 	struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1891 	const struct ctl_table *table;
1892 
1893 	table = nn_vrf->ctl_hdr->ctl_table_arg;
1894 	unregister_net_sysctl_table(nn_vrf->ctl_hdr);
1895 	kfree(table);
1896 }
1897 #else
1898 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
1899 {
1900 	return 0;
1901 }
1902 
1903 static void vrf_netns_exit_sysctl(struct net *net)
1904 {
1905 }
1906 #endif
1907 
1908 /* Initialize per network namespace state */
1909 static int __net_init vrf_netns_init(struct net *net)
1910 {
1911 	struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1912 
1913 	nn_vrf->add_fib_rules = true;
1914 	vrf_map_init(&nn_vrf->vmap);
1915 
1916 	return vrf_netns_init_sysctl(net, nn_vrf);
1917 }
1918 
1919 static void __net_exit vrf_netns_exit(struct net *net)
1920 {
1921 	vrf_netns_exit_sysctl(net);
1922 }
1923 
1924 static struct pernet_operations vrf_net_ops __net_initdata = {
1925 	.init = vrf_netns_init,
1926 	.exit = vrf_netns_exit,
1927 	.id   = &vrf_net_id,
1928 	.size = sizeof(struct netns_vrf),
1929 };
1930 
1931 static int __init vrf_init_module(void)
1932 {
1933 	int rc;
1934 
1935 	register_netdevice_notifier(&vrf_notifier_block);
1936 
1937 	rc = register_pernet_subsys(&vrf_net_ops);
1938 	if (rc < 0)
1939 		goto error;
1940 
1941 	rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF,
1942 					  vrf_ifindex_lookup_by_table_id);
1943 	if (rc < 0)
1944 		goto unreg_pernet;
1945 
1946 	rc = rtnl_link_register(&vrf_link_ops);
1947 	if (rc < 0)
1948 		goto table_lookup_unreg;
1949 
1950 	return 0;
1951 
1952 table_lookup_unreg:
1953 	l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF,
1954 				       vrf_ifindex_lookup_by_table_id);
1955 
1956 unreg_pernet:
1957 	unregister_pernet_subsys(&vrf_net_ops);
1958 
1959 error:
1960 	unregister_netdevice_notifier(&vrf_notifier_block);
1961 	return rc;
1962 }
1963 
1964 module_init(vrf_init_module);
1965 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1966 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1967 MODULE_LICENSE("GPL");
1968 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1969 MODULE_VERSION(DRV_VERSION);
1970