xref: /linux/drivers/net/vrf.c (revision b24413180f5600bcb3bb70fbed5cf186b60864bd)
1 /*
2  * vrf.c: device driver to encapsulate a VRF space
3  *
4  * Copyright (c) 2015 Cumulus Networks. All rights reserved.
5  * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com>
6  * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com>
7  *
8  * Based on dummy, team and ipvlan drivers
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  */
15 
16 #include <linux/module.h>
17 #include <linux/kernel.h>
18 #include <linux/netdevice.h>
19 #include <linux/etherdevice.h>
20 #include <linux/ip.h>
21 #include <linux/init.h>
22 #include <linux/moduleparam.h>
23 #include <linux/netfilter.h>
24 #include <linux/rtnetlink.h>
25 #include <net/rtnetlink.h>
26 #include <linux/u64_stats_sync.h>
27 #include <linux/hashtable.h>
28 
29 #include <linux/inetdevice.h>
30 #include <net/arp.h>
31 #include <net/ip.h>
32 #include <net/ip_fib.h>
33 #include <net/ip6_fib.h>
34 #include <net/ip6_route.h>
35 #include <net/route.h>
36 #include <net/addrconf.h>
37 #include <net/l3mdev.h>
38 #include <net/fib_rules.h>
39 #include <net/netns/generic.h>
40 
41 #define DRV_NAME	"vrf"
42 #define DRV_VERSION	"1.0"
43 
44 #define FIB_RULE_PREF  1000       /* default preference for FIB rules */
45 
46 static unsigned int vrf_net_id;
47 
48 struct net_vrf {
49 	struct rtable __rcu	*rth;
50 	struct rt6_info	__rcu	*rt6;
51 	u32                     tb_id;
52 };
53 
54 struct pcpu_dstats {
55 	u64			tx_pkts;
56 	u64			tx_bytes;
57 	u64			tx_drps;
58 	u64			rx_pkts;
59 	u64			rx_bytes;
60 	u64			rx_drps;
61 	struct u64_stats_sync	syncp;
62 };
63 
64 static void vrf_rx_stats(struct net_device *dev, int len)
65 {
66 	struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
67 
68 	u64_stats_update_begin(&dstats->syncp);
69 	dstats->rx_pkts++;
70 	dstats->rx_bytes += len;
71 	u64_stats_update_end(&dstats->syncp);
72 }
73 
74 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
75 {
76 	vrf_dev->stats.tx_errors++;
77 	kfree_skb(skb);
78 }
79 
80 static void vrf_get_stats64(struct net_device *dev,
81 			    struct rtnl_link_stats64 *stats)
82 {
83 	int i;
84 
85 	for_each_possible_cpu(i) {
86 		const struct pcpu_dstats *dstats;
87 		u64 tbytes, tpkts, tdrops, rbytes, rpkts;
88 		unsigned int start;
89 
90 		dstats = per_cpu_ptr(dev->dstats, i);
91 		do {
92 			start = u64_stats_fetch_begin_irq(&dstats->syncp);
93 			tbytes = dstats->tx_bytes;
94 			tpkts = dstats->tx_pkts;
95 			tdrops = dstats->tx_drps;
96 			rbytes = dstats->rx_bytes;
97 			rpkts = dstats->rx_pkts;
98 		} while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
99 		stats->tx_bytes += tbytes;
100 		stats->tx_packets += tpkts;
101 		stats->tx_dropped += tdrops;
102 		stats->rx_bytes += rbytes;
103 		stats->rx_packets += rpkts;
104 	}
105 }
106 
107 /* by default VRF devices do not have a qdisc and are expected
108  * to be created with only a single queue.
109  */
110 static bool qdisc_tx_is_default(const struct net_device *dev)
111 {
112 	struct netdev_queue *txq;
113 	struct Qdisc *qdisc;
114 
115 	if (dev->num_tx_queues > 1)
116 		return false;
117 
118 	txq = netdev_get_tx_queue(dev, 0);
119 	qdisc = rcu_access_pointer(txq->qdisc);
120 
121 	return !qdisc->enqueue;
122 }
123 
124 /* Local traffic destined to local address. Reinsert the packet to rx
125  * path, similar to loopback handling.
126  */
127 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
128 			  struct dst_entry *dst)
129 {
130 	int len = skb->len;
131 
132 	skb_orphan(skb);
133 
134 	skb_dst_set(skb, dst);
135 	skb_dst_force(skb);
136 
137 	/* set pkt_type to avoid skb hitting packet taps twice -
138 	 * once on Tx and again in Rx processing
139 	 */
140 	skb->pkt_type = PACKET_LOOPBACK;
141 
142 	skb->protocol = eth_type_trans(skb, dev);
143 
144 	if (likely(netif_rx(skb) == NET_RX_SUCCESS))
145 		vrf_rx_stats(dev, len);
146 	else
147 		this_cpu_inc(dev->dstats->rx_drps);
148 
149 	return NETDEV_TX_OK;
150 }
151 
152 #if IS_ENABLED(CONFIG_IPV6)
153 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
154 			     struct sk_buff *skb)
155 {
156 	int err;
157 
158 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
159 		      sk, skb, NULL, skb_dst(skb)->dev, dst_output);
160 
161 	if (likely(err == 1))
162 		err = dst_output(net, sk, skb);
163 
164 	return err;
165 }
166 
167 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
168 					   struct net_device *dev)
169 {
170 	const struct ipv6hdr *iph = ipv6_hdr(skb);
171 	struct net *net = dev_net(skb->dev);
172 	struct flowi6 fl6 = {
173 		/* needed to match OIF rule */
174 		.flowi6_oif = dev->ifindex,
175 		.flowi6_iif = LOOPBACK_IFINDEX,
176 		.daddr = iph->daddr,
177 		.saddr = iph->saddr,
178 		.flowlabel = ip6_flowinfo(iph),
179 		.flowi6_mark = skb->mark,
180 		.flowi6_proto = iph->nexthdr,
181 		.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
182 	};
183 	int ret = NET_XMIT_DROP;
184 	struct dst_entry *dst;
185 	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
186 
187 	dst = ip6_route_output(net, NULL, &fl6);
188 	if (dst == dst_null)
189 		goto err;
190 
191 	skb_dst_drop(skb);
192 
193 	/* if dst.dev is loopback or the VRF device again this is locally
194 	 * originated traffic destined to a local address. Short circuit
195 	 * to Rx path
196 	 */
197 	if (dst->dev == dev)
198 		return vrf_local_xmit(skb, dev, dst);
199 
200 	skb_dst_set(skb, dst);
201 
202 	/* strip the ethernet header added for pass through VRF device */
203 	__skb_pull(skb, skb_network_offset(skb));
204 
205 	ret = vrf_ip6_local_out(net, skb->sk, skb);
206 	if (unlikely(net_xmit_eval(ret)))
207 		dev->stats.tx_errors++;
208 	else
209 		ret = NET_XMIT_SUCCESS;
210 
211 	return ret;
212 err:
213 	vrf_tx_error(dev, skb);
214 	return NET_XMIT_DROP;
215 }
216 #else
217 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
218 					   struct net_device *dev)
219 {
220 	vrf_tx_error(dev, skb);
221 	return NET_XMIT_DROP;
222 }
223 #endif
224 
225 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
226 static int vrf_ip_local_out(struct net *net, struct sock *sk,
227 			    struct sk_buff *skb)
228 {
229 	int err;
230 
231 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
232 		      skb, NULL, skb_dst(skb)->dev, dst_output);
233 	if (likely(err == 1))
234 		err = dst_output(net, sk, skb);
235 
236 	return err;
237 }
238 
239 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
240 					   struct net_device *vrf_dev)
241 {
242 	struct iphdr *ip4h = ip_hdr(skb);
243 	int ret = NET_XMIT_DROP;
244 	struct flowi4 fl4 = {
245 		/* needed to match OIF rule */
246 		.flowi4_oif = vrf_dev->ifindex,
247 		.flowi4_iif = LOOPBACK_IFINDEX,
248 		.flowi4_tos = RT_TOS(ip4h->tos),
249 		.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
250 		.flowi4_proto = ip4h->protocol,
251 		.daddr = ip4h->daddr,
252 		.saddr = ip4h->saddr,
253 	};
254 	struct net *net = dev_net(vrf_dev);
255 	struct rtable *rt;
256 
257 	rt = ip_route_output_flow(net, &fl4, NULL);
258 	if (IS_ERR(rt))
259 		goto err;
260 
261 	skb_dst_drop(skb);
262 
263 	/* if dst.dev is loopback or the VRF device again this is locally
264 	 * originated traffic destined to a local address. Short circuit
265 	 * to Rx path
266 	 */
267 	if (rt->dst.dev == vrf_dev)
268 		return vrf_local_xmit(skb, vrf_dev, &rt->dst);
269 
270 	skb_dst_set(skb, &rt->dst);
271 
272 	/* strip the ethernet header added for pass through VRF device */
273 	__skb_pull(skb, skb_network_offset(skb));
274 
275 	if (!ip4h->saddr) {
276 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
277 					       RT_SCOPE_LINK);
278 	}
279 
280 	ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
281 	if (unlikely(net_xmit_eval(ret)))
282 		vrf_dev->stats.tx_errors++;
283 	else
284 		ret = NET_XMIT_SUCCESS;
285 
286 out:
287 	return ret;
288 err:
289 	vrf_tx_error(vrf_dev, skb);
290 	goto out;
291 }
292 
293 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
294 {
295 	switch (skb->protocol) {
296 	case htons(ETH_P_IP):
297 		return vrf_process_v4_outbound(skb, dev);
298 	case htons(ETH_P_IPV6):
299 		return vrf_process_v6_outbound(skb, dev);
300 	default:
301 		vrf_tx_error(dev, skb);
302 		return NET_XMIT_DROP;
303 	}
304 }
305 
306 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
307 {
308 	int len = skb->len;
309 	netdev_tx_t ret = is_ip_tx_frame(skb, dev);
310 
311 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
312 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
313 
314 		u64_stats_update_begin(&dstats->syncp);
315 		dstats->tx_pkts++;
316 		dstats->tx_bytes += len;
317 		u64_stats_update_end(&dstats->syncp);
318 	} else {
319 		this_cpu_inc(dev->dstats->tx_drps);
320 	}
321 
322 	return ret;
323 }
324 
325 static int vrf_finish_direct(struct net *net, struct sock *sk,
326 			     struct sk_buff *skb)
327 {
328 	struct net_device *vrf_dev = skb->dev;
329 
330 	if (!list_empty(&vrf_dev->ptype_all) &&
331 	    likely(skb_headroom(skb) >= ETH_HLEN)) {
332 		struct ethhdr *eth = skb_push(skb, ETH_HLEN);
333 
334 		ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
335 		eth_zero_addr(eth->h_dest);
336 		eth->h_proto = skb->protocol;
337 
338 		rcu_read_lock_bh();
339 		dev_queue_xmit_nit(skb, vrf_dev);
340 		rcu_read_unlock_bh();
341 
342 		skb_pull(skb, ETH_HLEN);
343 	}
344 
345 	return 1;
346 }
347 
348 #if IS_ENABLED(CONFIG_IPV6)
349 /* modelled after ip6_finish_output2 */
350 static int vrf_finish_output6(struct net *net, struct sock *sk,
351 			      struct sk_buff *skb)
352 {
353 	struct dst_entry *dst = skb_dst(skb);
354 	struct net_device *dev = dst->dev;
355 	struct neighbour *neigh;
356 	struct in6_addr *nexthop;
357 	int ret;
358 
359 	nf_reset(skb);
360 
361 	skb->protocol = htons(ETH_P_IPV6);
362 	skb->dev = dev;
363 
364 	rcu_read_lock_bh();
365 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
366 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
367 	if (unlikely(!neigh))
368 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
369 	if (!IS_ERR(neigh)) {
370 		sock_confirm_neigh(skb, neigh);
371 		ret = neigh_output(neigh, skb);
372 		rcu_read_unlock_bh();
373 		return ret;
374 	}
375 	rcu_read_unlock_bh();
376 
377 	IP6_INC_STATS(dev_net(dst->dev),
378 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
379 	kfree_skb(skb);
380 	return -EINVAL;
381 }
382 
383 /* modelled after ip6_output */
384 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
385 {
386 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
387 			    net, sk, skb, NULL, skb_dst(skb)->dev,
388 			    vrf_finish_output6,
389 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
390 }
391 
392 /* set dst on skb to send packet to us via dev_xmit path. Allows
393  * packet to go through device based features such as qdisc, netfilter
394  * hooks and packet sockets with skb->dev set to vrf device.
395  */
396 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
397 					    struct sk_buff *skb)
398 {
399 	struct net_vrf *vrf = netdev_priv(vrf_dev);
400 	struct dst_entry *dst = NULL;
401 	struct rt6_info *rt6;
402 
403 	rcu_read_lock();
404 
405 	rt6 = rcu_dereference(vrf->rt6);
406 	if (likely(rt6)) {
407 		dst = &rt6->dst;
408 		dst_hold(dst);
409 	}
410 
411 	rcu_read_unlock();
412 
413 	if (unlikely(!dst)) {
414 		vrf_tx_error(vrf_dev, skb);
415 		return NULL;
416 	}
417 
418 	skb_dst_drop(skb);
419 	skb_dst_set(skb, dst);
420 
421 	return skb;
422 }
423 
424 static int vrf_output6_direct(struct net *net, struct sock *sk,
425 			      struct sk_buff *skb)
426 {
427 	skb->protocol = htons(ETH_P_IPV6);
428 
429 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
430 			    net, sk, skb, NULL, skb->dev,
431 			    vrf_finish_direct,
432 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
433 }
434 
435 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
436 					  struct sock *sk,
437 					  struct sk_buff *skb)
438 {
439 	struct net *net = dev_net(vrf_dev);
440 	int err;
441 
442 	skb->dev = vrf_dev;
443 
444 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
445 		      skb, NULL, vrf_dev, vrf_output6_direct);
446 
447 	if (likely(err == 1))
448 		err = vrf_output6_direct(net, sk, skb);
449 
450 	/* reset skb device */
451 	if (likely(err == 1))
452 		nf_reset(skb);
453 	else
454 		skb = NULL;
455 
456 	return skb;
457 }
458 
459 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
460 				   struct sock *sk,
461 				   struct sk_buff *skb)
462 {
463 	/* don't divert link scope packets */
464 	if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
465 		return skb;
466 
467 	if (qdisc_tx_is_default(vrf_dev))
468 		return vrf_ip6_out_direct(vrf_dev, sk, skb);
469 
470 	return vrf_ip6_out_redirect(vrf_dev, skb);
471 }
472 
473 /* holding rtnl */
474 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
475 {
476 	struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
477 	struct net *net = dev_net(dev);
478 	struct dst_entry *dst;
479 
480 	RCU_INIT_POINTER(vrf->rt6, NULL);
481 	synchronize_rcu();
482 
483 	/* move dev in dst's to loopback so this VRF device can be deleted
484 	 * - based on dst_ifdown
485 	 */
486 	if (rt6) {
487 		dst = &rt6->dst;
488 		dev_put(dst->dev);
489 		dst->dev = net->loopback_dev;
490 		dev_hold(dst->dev);
491 		dst_release(dst);
492 	}
493 }
494 
495 static int vrf_rt6_create(struct net_device *dev)
496 {
497 	int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
498 	struct net_vrf *vrf = netdev_priv(dev);
499 	struct net *net = dev_net(dev);
500 	struct fib6_table *rt6i_table;
501 	struct rt6_info *rt6;
502 	int rc = -ENOMEM;
503 
504 	/* IPv6 can be CONFIG enabled and then disabled runtime */
505 	if (!ipv6_mod_enabled())
506 		return 0;
507 
508 	rt6i_table = fib6_new_table(net, vrf->tb_id);
509 	if (!rt6i_table)
510 		goto out;
511 
512 	/* create a dst for routing packets out a VRF device */
513 	rt6 = ip6_dst_alloc(net, dev, flags);
514 	if (!rt6)
515 		goto out;
516 
517 	rt6->rt6i_table = rt6i_table;
518 	rt6->dst.output	= vrf_output6;
519 
520 	rcu_assign_pointer(vrf->rt6, rt6);
521 
522 	rc = 0;
523 out:
524 	return rc;
525 }
526 #else
527 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
528 				   struct sock *sk,
529 				   struct sk_buff *skb)
530 {
531 	return skb;
532 }
533 
534 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
535 {
536 }
537 
538 static int vrf_rt6_create(struct net_device *dev)
539 {
540 	return 0;
541 }
542 #endif
543 
544 /* modelled after ip_finish_output2 */
545 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
546 {
547 	struct dst_entry *dst = skb_dst(skb);
548 	struct rtable *rt = (struct rtable *)dst;
549 	struct net_device *dev = dst->dev;
550 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
551 	struct neighbour *neigh;
552 	u32 nexthop;
553 	int ret = -EINVAL;
554 
555 	nf_reset(skb);
556 
557 	/* Be paranoid, rather than too clever. */
558 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
559 		struct sk_buff *skb2;
560 
561 		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
562 		if (!skb2) {
563 			ret = -ENOMEM;
564 			goto err;
565 		}
566 		if (skb->sk)
567 			skb_set_owner_w(skb2, skb->sk);
568 
569 		consume_skb(skb);
570 		skb = skb2;
571 	}
572 
573 	rcu_read_lock_bh();
574 
575 	nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
576 	neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
577 	if (unlikely(!neigh))
578 		neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
579 	if (!IS_ERR(neigh)) {
580 		sock_confirm_neigh(skb, neigh);
581 		ret = neigh_output(neigh, skb);
582 	}
583 
584 	rcu_read_unlock_bh();
585 err:
586 	if (unlikely(ret < 0))
587 		vrf_tx_error(skb->dev, skb);
588 	return ret;
589 }
590 
591 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
592 {
593 	struct net_device *dev = skb_dst(skb)->dev;
594 
595 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
596 
597 	skb->dev = dev;
598 	skb->protocol = htons(ETH_P_IP);
599 
600 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
601 			    net, sk, skb, NULL, dev,
602 			    vrf_finish_output,
603 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
604 }
605 
606 /* set dst on skb to send packet to us via dev_xmit path. Allows
607  * packet to go through device based features such as qdisc, netfilter
608  * hooks and packet sockets with skb->dev set to vrf device.
609  */
610 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
611 					   struct sk_buff *skb)
612 {
613 	struct net_vrf *vrf = netdev_priv(vrf_dev);
614 	struct dst_entry *dst = NULL;
615 	struct rtable *rth;
616 
617 	rcu_read_lock();
618 
619 	rth = rcu_dereference(vrf->rth);
620 	if (likely(rth)) {
621 		dst = &rth->dst;
622 		dst_hold(dst);
623 	}
624 
625 	rcu_read_unlock();
626 
627 	if (unlikely(!dst)) {
628 		vrf_tx_error(vrf_dev, skb);
629 		return NULL;
630 	}
631 
632 	skb_dst_drop(skb);
633 	skb_dst_set(skb, dst);
634 
635 	return skb;
636 }
637 
638 static int vrf_output_direct(struct net *net, struct sock *sk,
639 			     struct sk_buff *skb)
640 {
641 	skb->protocol = htons(ETH_P_IP);
642 
643 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
644 			    net, sk, skb, NULL, skb->dev,
645 			    vrf_finish_direct,
646 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
647 }
648 
649 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
650 					 struct sock *sk,
651 					 struct sk_buff *skb)
652 {
653 	struct net *net = dev_net(vrf_dev);
654 	int err;
655 
656 	skb->dev = vrf_dev;
657 
658 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
659 		      skb, NULL, vrf_dev, vrf_output_direct);
660 
661 	if (likely(err == 1))
662 		err = vrf_output_direct(net, sk, skb);
663 
664 	/* reset skb device */
665 	if (likely(err == 1))
666 		nf_reset(skb);
667 	else
668 		skb = NULL;
669 
670 	return skb;
671 }
672 
673 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
674 				  struct sock *sk,
675 				  struct sk_buff *skb)
676 {
677 	/* don't divert multicast */
678 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
679 		return skb;
680 
681 	if (qdisc_tx_is_default(vrf_dev))
682 		return vrf_ip_out_direct(vrf_dev, sk, skb);
683 
684 	return vrf_ip_out_redirect(vrf_dev, skb);
685 }
686 
687 /* called with rcu lock held */
688 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
689 				  struct sock *sk,
690 				  struct sk_buff *skb,
691 				  u16 proto)
692 {
693 	switch (proto) {
694 	case AF_INET:
695 		return vrf_ip_out(vrf_dev, sk, skb);
696 	case AF_INET6:
697 		return vrf_ip6_out(vrf_dev, sk, skb);
698 	}
699 
700 	return skb;
701 }
702 
703 /* holding rtnl */
704 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
705 {
706 	struct rtable *rth = rtnl_dereference(vrf->rth);
707 	struct net *net = dev_net(dev);
708 	struct dst_entry *dst;
709 
710 	RCU_INIT_POINTER(vrf->rth, NULL);
711 	synchronize_rcu();
712 
713 	/* move dev in dst's to loopback so this VRF device can be deleted
714 	 * - based on dst_ifdown
715 	 */
716 	if (rth) {
717 		dst = &rth->dst;
718 		dev_put(dst->dev);
719 		dst->dev = net->loopback_dev;
720 		dev_hold(dst->dev);
721 		dst_release(dst);
722 	}
723 }
724 
725 static int vrf_rtable_create(struct net_device *dev)
726 {
727 	struct net_vrf *vrf = netdev_priv(dev);
728 	struct rtable *rth;
729 
730 	if (!fib_new_table(dev_net(dev), vrf->tb_id))
731 		return -ENOMEM;
732 
733 	/* create a dst for routing packets out through a VRF device */
734 	rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
735 	if (!rth)
736 		return -ENOMEM;
737 
738 	rth->dst.output	= vrf_output;
739 	rth->rt_table_id = vrf->tb_id;
740 
741 	rcu_assign_pointer(vrf->rth, rth);
742 
743 	return 0;
744 }
745 
746 /**************************** device handling ********************/
747 
748 /* cycle interface to flush neighbor cache and move routes across tables */
749 static void cycle_netdev(struct net_device *dev)
750 {
751 	unsigned int flags = dev->flags;
752 	int ret;
753 
754 	if (!netif_running(dev))
755 		return;
756 
757 	ret = dev_change_flags(dev, flags & ~IFF_UP);
758 	if (ret >= 0)
759 		ret = dev_change_flags(dev, flags);
760 
761 	if (ret < 0) {
762 		netdev_err(dev,
763 			   "Failed to cycle device %s; route tables might be wrong!\n",
764 			   dev->name);
765 	}
766 }
767 
768 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
769 {
770 	int ret;
771 
772 	/* do not allow loopback device to be enslaved to a VRF.
773 	 * The vrf device acts as the loopback for the vrf.
774 	 */
775 	if (port_dev == dev_net(dev)->loopback_dev)
776 		return -EOPNOTSUPP;
777 
778 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
779 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
780 	if (ret < 0)
781 		goto err;
782 
783 	cycle_netdev(port_dev);
784 
785 	return 0;
786 
787 err:
788 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
789 	return ret;
790 }
791 
792 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
793 {
794 	if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
795 		return -EINVAL;
796 
797 	return do_vrf_add_slave(dev, port_dev);
798 }
799 
800 /* inverse of do_vrf_add_slave */
801 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
802 {
803 	netdev_upper_dev_unlink(port_dev, dev);
804 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
805 
806 	cycle_netdev(port_dev);
807 
808 	return 0;
809 }
810 
811 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
812 {
813 	return do_vrf_del_slave(dev, port_dev);
814 }
815 
816 static void vrf_dev_uninit(struct net_device *dev)
817 {
818 	struct net_vrf *vrf = netdev_priv(dev);
819 
820 	vrf_rtable_release(dev, vrf);
821 	vrf_rt6_release(dev, vrf);
822 
823 	free_percpu(dev->dstats);
824 	dev->dstats = NULL;
825 }
826 
827 static int vrf_dev_init(struct net_device *dev)
828 {
829 	struct net_vrf *vrf = netdev_priv(dev);
830 
831 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
832 	if (!dev->dstats)
833 		goto out_nomem;
834 
835 	/* create the default dst which points back to us */
836 	if (vrf_rtable_create(dev) != 0)
837 		goto out_stats;
838 
839 	if (vrf_rt6_create(dev) != 0)
840 		goto out_rth;
841 
842 	dev->flags = IFF_MASTER | IFF_NOARP;
843 
844 	/* MTU is irrelevant for VRF device; set to 64k similar to lo */
845 	dev->mtu = 64 * 1024;
846 
847 	/* similarly, oper state is irrelevant; set to up to avoid confusion */
848 	dev->operstate = IF_OPER_UP;
849 	netdev_lockdep_set_classes(dev);
850 	return 0;
851 
852 out_rth:
853 	vrf_rtable_release(dev, vrf);
854 out_stats:
855 	free_percpu(dev->dstats);
856 	dev->dstats = NULL;
857 out_nomem:
858 	return -ENOMEM;
859 }
860 
861 static const struct net_device_ops vrf_netdev_ops = {
862 	.ndo_init		= vrf_dev_init,
863 	.ndo_uninit		= vrf_dev_uninit,
864 	.ndo_start_xmit		= vrf_xmit,
865 	.ndo_get_stats64	= vrf_get_stats64,
866 	.ndo_add_slave		= vrf_add_slave,
867 	.ndo_del_slave		= vrf_del_slave,
868 };
869 
870 static u32 vrf_fib_table(const struct net_device *dev)
871 {
872 	struct net_vrf *vrf = netdev_priv(dev);
873 
874 	return vrf->tb_id;
875 }
876 
877 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
878 {
879 	kfree_skb(skb);
880 	return 0;
881 }
882 
883 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
884 				      struct sk_buff *skb,
885 				      struct net_device *dev)
886 {
887 	struct net *net = dev_net(dev);
888 
889 	if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
890 		skb = NULL;    /* kfree_skb(skb) handled by nf code */
891 
892 	return skb;
893 }
894 
895 #if IS_ENABLED(CONFIG_IPV6)
896 /* neighbor handling is done with actual device; do not want
897  * to flip skb->dev for those ndisc packets. This really fails
898  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
899  * a start.
900  */
901 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
902 {
903 	const struct ipv6hdr *iph = ipv6_hdr(skb);
904 	bool rc = false;
905 
906 	if (iph->nexthdr == NEXTHDR_ICMP) {
907 		const struct icmp6hdr *icmph;
908 		struct icmp6hdr _icmph;
909 
910 		icmph = skb_header_pointer(skb, sizeof(*iph),
911 					   sizeof(_icmph), &_icmph);
912 		if (!icmph)
913 			goto out;
914 
915 		switch (icmph->icmp6_type) {
916 		case NDISC_ROUTER_SOLICITATION:
917 		case NDISC_ROUTER_ADVERTISEMENT:
918 		case NDISC_NEIGHBOUR_SOLICITATION:
919 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
920 		case NDISC_REDIRECT:
921 			rc = true;
922 			break;
923 		}
924 	}
925 
926 out:
927 	return rc;
928 }
929 
930 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
931 					     const struct net_device *dev,
932 					     struct flowi6 *fl6,
933 					     int ifindex,
934 					     int flags)
935 {
936 	struct net_vrf *vrf = netdev_priv(dev);
937 	struct fib6_table *table = NULL;
938 	struct rt6_info *rt6;
939 
940 	rcu_read_lock();
941 
942 	/* fib6_table does not have a refcnt and can not be freed */
943 	rt6 = rcu_dereference(vrf->rt6);
944 	if (likely(rt6))
945 		table = rt6->rt6i_table;
946 
947 	rcu_read_unlock();
948 
949 	if (!table)
950 		return NULL;
951 
952 	return ip6_pol_route(net, table, ifindex, fl6, flags);
953 }
954 
955 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
956 			      int ifindex)
957 {
958 	const struct ipv6hdr *iph = ipv6_hdr(skb);
959 	struct flowi6 fl6 = {
960 		.flowi6_iif     = ifindex,
961 		.flowi6_mark    = skb->mark,
962 		.flowi6_proto   = iph->nexthdr,
963 		.daddr          = iph->daddr,
964 		.saddr          = iph->saddr,
965 		.flowlabel      = ip6_flowinfo(iph),
966 	};
967 	struct net *net = dev_net(vrf_dev);
968 	struct rt6_info *rt6;
969 
970 	rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
971 				   RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
972 	if (unlikely(!rt6))
973 		return;
974 
975 	if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
976 		return;
977 
978 	skb_dst_set(skb, &rt6->dst);
979 }
980 
981 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
982 				   struct sk_buff *skb)
983 {
984 	int orig_iif = skb->skb_iif;
985 	bool need_strict;
986 
987 	/* loopback traffic; do not push through packet taps again.
988 	 * Reset pkt_type for upper layers to process skb
989 	 */
990 	if (skb->pkt_type == PACKET_LOOPBACK) {
991 		skb->dev = vrf_dev;
992 		skb->skb_iif = vrf_dev->ifindex;
993 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
994 		skb->pkt_type = PACKET_HOST;
995 		goto out;
996 	}
997 
998 	/* if packet is NDISC or addressed to multicast or link-local
999 	 * then keep the ingress interface
1000 	 */
1001 	need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1002 	if (!ipv6_ndisc_frame(skb) && !need_strict) {
1003 		vrf_rx_stats(vrf_dev, skb->len);
1004 		skb->dev = vrf_dev;
1005 		skb->skb_iif = vrf_dev->ifindex;
1006 
1007 		if (!list_empty(&vrf_dev->ptype_all)) {
1008 			skb_push(skb, skb->mac_len);
1009 			dev_queue_xmit_nit(skb, vrf_dev);
1010 			skb_pull(skb, skb->mac_len);
1011 		}
1012 
1013 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1014 	}
1015 
1016 	if (need_strict)
1017 		vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1018 
1019 	skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1020 out:
1021 	return skb;
1022 }
1023 
1024 #else
1025 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1026 				   struct sk_buff *skb)
1027 {
1028 	return skb;
1029 }
1030 #endif
1031 
1032 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1033 				  struct sk_buff *skb)
1034 {
1035 	skb->dev = vrf_dev;
1036 	skb->skb_iif = vrf_dev->ifindex;
1037 	IPCB(skb)->flags |= IPSKB_L3SLAVE;
1038 
1039 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1040 		goto out;
1041 
1042 	/* loopback traffic; do not push through packet taps again.
1043 	 * Reset pkt_type for upper layers to process skb
1044 	 */
1045 	if (skb->pkt_type == PACKET_LOOPBACK) {
1046 		skb->pkt_type = PACKET_HOST;
1047 		goto out;
1048 	}
1049 
1050 	vrf_rx_stats(vrf_dev, skb->len);
1051 
1052 	if (!list_empty(&vrf_dev->ptype_all)) {
1053 		skb_push(skb, skb->mac_len);
1054 		dev_queue_xmit_nit(skb, vrf_dev);
1055 		skb_pull(skb, skb->mac_len);
1056 	}
1057 
1058 	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1059 out:
1060 	return skb;
1061 }
1062 
1063 /* called with rcu lock held */
1064 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1065 				  struct sk_buff *skb,
1066 				  u16 proto)
1067 {
1068 	switch (proto) {
1069 	case AF_INET:
1070 		return vrf_ip_rcv(vrf_dev, skb);
1071 	case AF_INET6:
1072 		return vrf_ip6_rcv(vrf_dev, skb);
1073 	}
1074 
1075 	return skb;
1076 }
1077 
1078 #if IS_ENABLED(CONFIG_IPV6)
1079 /* send to link-local or multicast address via interface enslaved to
1080  * VRF device. Force lookup to VRF table without changing flow struct
1081  */
1082 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1083 					      struct flowi6 *fl6)
1084 {
1085 	struct net *net = dev_net(dev);
1086 	int flags = RT6_LOOKUP_F_IFACE;
1087 	struct dst_entry *dst = NULL;
1088 	struct rt6_info *rt;
1089 
1090 	/* VRF device does not have a link-local address and
1091 	 * sending packets to link-local or mcast addresses over
1092 	 * a VRF device does not make sense
1093 	 */
1094 	if (fl6->flowi6_oif == dev->ifindex) {
1095 		dst = &net->ipv6.ip6_null_entry->dst;
1096 		dst_hold(dst);
1097 		return dst;
1098 	}
1099 
1100 	if (!ipv6_addr_any(&fl6->saddr))
1101 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1102 
1103 	rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1104 	if (rt)
1105 		dst = &rt->dst;
1106 
1107 	return dst;
1108 }
1109 #endif
1110 
1111 static const struct l3mdev_ops vrf_l3mdev_ops = {
1112 	.l3mdev_fib_table	= vrf_fib_table,
1113 	.l3mdev_l3_rcv		= vrf_l3_rcv,
1114 	.l3mdev_l3_out		= vrf_l3_out,
1115 #if IS_ENABLED(CONFIG_IPV6)
1116 	.l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1117 #endif
1118 };
1119 
1120 static void vrf_get_drvinfo(struct net_device *dev,
1121 			    struct ethtool_drvinfo *info)
1122 {
1123 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1124 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1125 }
1126 
1127 static const struct ethtool_ops vrf_ethtool_ops = {
1128 	.get_drvinfo	= vrf_get_drvinfo,
1129 };
1130 
1131 static inline size_t vrf_fib_rule_nl_size(void)
1132 {
1133 	size_t sz;
1134 
1135 	sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1136 	sz += nla_total_size(sizeof(u8));	/* FRA_L3MDEV */
1137 	sz += nla_total_size(sizeof(u32));	/* FRA_PRIORITY */
1138 
1139 	return sz;
1140 }
1141 
1142 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1143 {
1144 	struct fib_rule_hdr *frh;
1145 	struct nlmsghdr *nlh;
1146 	struct sk_buff *skb;
1147 	int err;
1148 
1149 	if (family == AF_INET6 && !ipv6_mod_enabled())
1150 		return 0;
1151 
1152 	skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1153 	if (!skb)
1154 		return -ENOMEM;
1155 
1156 	nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1157 	if (!nlh)
1158 		goto nla_put_failure;
1159 
1160 	/* rule only needs to appear once */
1161 	nlh->nlmsg_flags |= NLM_F_EXCL;
1162 
1163 	frh = nlmsg_data(nlh);
1164 	memset(frh, 0, sizeof(*frh));
1165 	frh->family = family;
1166 	frh->action = FR_ACT_TO_TBL;
1167 
1168 	if (nla_put_u32(skb, FRA_L3MDEV, 1))
1169 		goto nla_put_failure;
1170 
1171 	if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1172 		goto nla_put_failure;
1173 
1174 	nlmsg_end(skb, nlh);
1175 
1176 	/* fib_nl_{new,del}rule handling looks for net from skb->sk */
1177 	skb->sk = dev_net(dev)->rtnl;
1178 	if (add_it) {
1179 		err = fib_nl_newrule(skb, nlh, NULL);
1180 		if (err == -EEXIST)
1181 			err = 0;
1182 	} else {
1183 		err = fib_nl_delrule(skb, nlh, NULL);
1184 		if (err == -ENOENT)
1185 			err = 0;
1186 	}
1187 	nlmsg_free(skb);
1188 
1189 	return err;
1190 
1191 nla_put_failure:
1192 	nlmsg_free(skb);
1193 
1194 	return -EMSGSIZE;
1195 }
1196 
1197 static int vrf_add_fib_rules(const struct net_device *dev)
1198 {
1199 	int err;
1200 
1201 	err = vrf_fib_rule(dev, AF_INET,  true);
1202 	if (err < 0)
1203 		goto out_err;
1204 
1205 	err = vrf_fib_rule(dev, AF_INET6, true);
1206 	if (err < 0)
1207 		goto ipv6_err;
1208 
1209 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1210 	err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1211 	if (err < 0)
1212 		goto ipmr_err;
1213 #endif
1214 
1215 	return 0;
1216 
1217 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1218 ipmr_err:
1219 	vrf_fib_rule(dev, AF_INET6,  false);
1220 #endif
1221 
1222 ipv6_err:
1223 	vrf_fib_rule(dev, AF_INET,  false);
1224 
1225 out_err:
1226 	netdev_err(dev, "Failed to add FIB rules.\n");
1227 	return err;
1228 }
1229 
1230 static void vrf_setup(struct net_device *dev)
1231 {
1232 	ether_setup(dev);
1233 
1234 	/* Initialize the device structure. */
1235 	dev->netdev_ops = &vrf_netdev_ops;
1236 	dev->l3mdev_ops = &vrf_l3mdev_ops;
1237 	dev->ethtool_ops = &vrf_ethtool_ops;
1238 	dev->needs_free_netdev = true;
1239 
1240 	/* Fill in device structure with ethernet-generic values. */
1241 	eth_hw_addr_random(dev);
1242 
1243 	/* don't acquire vrf device's netif_tx_lock when transmitting */
1244 	dev->features |= NETIF_F_LLTX;
1245 
1246 	/* don't allow vrf devices to change network namespaces. */
1247 	dev->features |= NETIF_F_NETNS_LOCAL;
1248 
1249 	/* does not make sense for a VLAN to be added to a vrf device */
1250 	dev->features   |= NETIF_F_VLAN_CHALLENGED;
1251 
1252 	/* enable offload features */
1253 	dev->features   |= NETIF_F_GSO_SOFTWARE;
1254 	dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1255 	dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1256 
1257 	dev->hw_features = dev->features;
1258 	dev->hw_enc_features = dev->features;
1259 
1260 	/* default to no qdisc; user can add if desired */
1261 	dev->priv_flags |= IFF_NO_QUEUE;
1262 }
1263 
1264 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1265 			struct netlink_ext_ack *extack)
1266 {
1267 	if (tb[IFLA_ADDRESS]) {
1268 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1269 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1270 			return -EINVAL;
1271 		}
1272 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1273 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1274 			return -EADDRNOTAVAIL;
1275 		}
1276 	}
1277 	return 0;
1278 }
1279 
1280 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1281 {
1282 	struct net_device *port_dev;
1283 	struct list_head *iter;
1284 
1285 	netdev_for_each_lower_dev(dev, port_dev, iter)
1286 		vrf_del_slave(dev, port_dev);
1287 
1288 	unregister_netdevice_queue(dev, head);
1289 }
1290 
1291 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1292 		       struct nlattr *tb[], struct nlattr *data[],
1293 		       struct netlink_ext_ack *extack)
1294 {
1295 	struct net_vrf *vrf = netdev_priv(dev);
1296 	bool *add_fib_rules;
1297 	struct net *net;
1298 	int err;
1299 
1300 	if (!data || !data[IFLA_VRF_TABLE]) {
1301 		NL_SET_ERR_MSG(extack, "VRF table id is missing");
1302 		return -EINVAL;
1303 	}
1304 
1305 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1306 	if (vrf->tb_id == RT_TABLE_UNSPEC) {
1307 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1308 				    "Invalid VRF table id");
1309 		return -EINVAL;
1310 	}
1311 
1312 	dev->priv_flags |= IFF_L3MDEV_MASTER;
1313 
1314 	err = register_netdevice(dev);
1315 	if (err)
1316 		goto out;
1317 
1318 	net = dev_net(dev);
1319 	add_fib_rules = net_generic(net, vrf_net_id);
1320 	if (*add_fib_rules) {
1321 		err = vrf_add_fib_rules(dev);
1322 		if (err) {
1323 			unregister_netdevice(dev);
1324 			goto out;
1325 		}
1326 		*add_fib_rules = false;
1327 	}
1328 
1329 out:
1330 	return err;
1331 }
1332 
1333 static size_t vrf_nl_getsize(const struct net_device *dev)
1334 {
1335 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
1336 }
1337 
1338 static int vrf_fillinfo(struct sk_buff *skb,
1339 			const struct net_device *dev)
1340 {
1341 	struct net_vrf *vrf = netdev_priv(dev);
1342 
1343 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1344 }
1345 
1346 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1347 				 const struct net_device *slave_dev)
1348 {
1349 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
1350 }
1351 
1352 static int vrf_fill_slave_info(struct sk_buff *skb,
1353 			       const struct net_device *vrf_dev,
1354 			       const struct net_device *slave_dev)
1355 {
1356 	struct net_vrf *vrf = netdev_priv(vrf_dev);
1357 
1358 	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1359 		return -EMSGSIZE;
1360 
1361 	return 0;
1362 }
1363 
1364 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1365 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
1366 };
1367 
1368 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1369 	.kind		= DRV_NAME,
1370 	.priv_size	= sizeof(struct net_vrf),
1371 
1372 	.get_size	= vrf_nl_getsize,
1373 	.policy		= vrf_nl_policy,
1374 	.validate	= vrf_validate,
1375 	.fill_info	= vrf_fillinfo,
1376 
1377 	.get_slave_size  = vrf_get_slave_size,
1378 	.fill_slave_info = vrf_fill_slave_info,
1379 
1380 	.newlink	= vrf_newlink,
1381 	.dellink	= vrf_dellink,
1382 	.setup		= vrf_setup,
1383 	.maxtype	= IFLA_VRF_MAX,
1384 };
1385 
1386 static int vrf_device_event(struct notifier_block *unused,
1387 			    unsigned long event, void *ptr)
1388 {
1389 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1390 
1391 	/* only care about unregister events to drop slave references */
1392 	if (event == NETDEV_UNREGISTER) {
1393 		struct net_device *vrf_dev;
1394 
1395 		if (!netif_is_l3_slave(dev))
1396 			goto out;
1397 
1398 		vrf_dev = netdev_master_upper_dev_get(dev);
1399 		vrf_del_slave(vrf_dev, dev);
1400 	}
1401 out:
1402 	return NOTIFY_DONE;
1403 }
1404 
1405 static struct notifier_block vrf_notifier_block __read_mostly = {
1406 	.notifier_call = vrf_device_event,
1407 };
1408 
1409 /* Initialize per network namespace state */
1410 static int __net_init vrf_netns_init(struct net *net)
1411 {
1412 	bool *add_fib_rules = net_generic(net, vrf_net_id);
1413 
1414 	*add_fib_rules = true;
1415 
1416 	return 0;
1417 }
1418 
1419 static struct pernet_operations vrf_net_ops __net_initdata = {
1420 	.init = vrf_netns_init,
1421 	.id   = &vrf_net_id,
1422 	.size = sizeof(bool),
1423 };
1424 
1425 static int __init vrf_init_module(void)
1426 {
1427 	int rc;
1428 
1429 	register_netdevice_notifier(&vrf_notifier_block);
1430 
1431 	rc = register_pernet_subsys(&vrf_net_ops);
1432 	if (rc < 0)
1433 		goto error;
1434 
1435 	rc = rtnl_link_register(&vrf_link_ops);
1436 	if (rc < 0) {
1437 		unregister_pernet_subsys(&vrf_net_ops);
1438 		goto error;
1439 	}
1440 
1441 	return 0;
1442 
1443 error:
1444 	unregister_netdevice_notifier(&vrf_notifier_block);
1445 	return rc;
1446 }
1447 
1448 module_init(vrf_init_module);
1449 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1450 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1451 MODULE_LICENSE("GPL");
1452 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1453 MODULE_VERSION(DRV_VERSION);
1454