xref: /linux/drivers/net/vrf.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
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 
136 	/* set pkt_type to avoid skb hitting packet taps twice -
137 	 * once on Tx and again in Rx processing
138 	 */
139 	skb->pkt_type = PACKET_LOOPBACK;
140 
141 	skb->protocol = eth_type_trans(skb, dev);
142 
143 	if (likely(netif_rx(skb) == NET_RX_SUCCESS))
144 		vrf_rx_stats(dev, len);
145 	else
146 		this_cpu_inc(dev->dstats->rx_drps);
147 
148 	return NETDEV_TX_OK;
149 }
150 
151 #if IS_ENABLED(CONFIG_IPV6)
152 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
153 			     struct sk_buff *skb)
154 {
155 	int err;
156 
157 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
158 		      sk, skb, NULL, skb_dst(skb)->dev, dst_output);
159 
160 	if (likely(err == 1))
161 		err = dst_output(net, sk, skb);
162 
163 	return err;
164 }
165 
166 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
167 					   struct net_device *dev)
168 {
169 	const struct ipv6hdr *iph = ipv6_hdr(skb);
170 	struct net *net = dev_net(skb->dev);
171 	struct flowi6 fl6 = {
172 		/* needed to match OIF rule */
173 		.flowi6_oif = dev->ifindex,
174 		.flowi6_iif = LOOPBACK_IFINDEX,
175 		.daddr = iph->daddr,
176 		.saddr = iph->saddr,
177 		.flowlabel = ip6_flowinfo(iph),
178 		.flowi6_mark = skb->mark,
179 		.flowi6_proto = iph->nexthdr,
180 		.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
181 	};
182 	int ret = NET_XMIT_DROP;
183 	struct dst_entry *dst;
184 	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
185 
186 	dst = ip6_route_output(net, NULL, &fl6);
187 	if (dst == dst_null)
188 		goto err;
189 
190 	skb_dst_drop(skb);
191 
192 	/* if dst.dev is loopback or the VRF device again this is locally
193 	 * originated traffic destined to a local address. Short circuit
194 	 * to Rx path
195 	 */
196 	if (dst->dev == dev)
197 		return vrf_local_xmit(skb, dev, dst);
198 
199 	skb_dst_set(skb, dst);
200 
201 	/* strip the ethernet header added for pass through VRF device */
202 	__skb_pull(skb, skb_network_offset(skb));
203 
204 	ret = vrf_ip6_local_out(net, skb->sk, skb);
205 	if (unlikely(net_xmit_eval(ret)))
206 		dev->stats.tx_errors++;
207 	else
208 		ret = NET_XMIT_SUCCESS;
209 
210 	return ret;
211 err:
212 	vrf_tx_error(dev, skb);
213 	return NET_XMIT_DROP;
214 }
215 #else
216 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
217 					   struct net_device *dev)
218 {
219 	vrf_tx_error(dev, skb);
220 	return NET_XMIT_DROP;
221 }
222 #endif
223 
224 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
225 static int vrf_ip_local_out(struct net *net, struct sock *sk,
226 			    struct sk_buff *skb)
227 {
228 	int err;
229 
230 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
231 		      skb, NULL, skb_dst(skb)->dev, dst_output);
232 	if (likely(err == 1))
233 		err = dst_output(net, sk, skb);
234 
235 	return err;
236 }
237 
238 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
239 					   struct net_device *vrf_dev)
240 {
241 	struct iphdr *ip4h = ip_hdr(skb);
242 	int ret = NET_XMIT_DROP;
243 	struct flowi4 fl4 = {
244 		/* needed to match OIF rule */
245 		.flowi4_oif = vrf_dev->ifindex,
246 		.flowi4_iif = LOOPBACK_IFINDEX,
247 		.flowi4_tos = RT_TOS(ip4h->tos),
248 		.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
249 		.flowi4_proto = ip4h->protocol,
250 		.daddr = ip4h->daddr,
251 		.saddr = ip4h->saddr,
252 	};
253 	struct net *net = dev_net(vrf_dev);
254 	struct rtable *rt;
255 
256 	rt = ip_route_output_flow(net, &fl4, NULL);
257 	if (IS_ERR(rt))
258 		goto err;
259 
260 	skb_dst_drop(skb);
261 
262 	/* if dst.dev is loopback or the VRF device again this is locally
263 	 * originated traffic destined to a local address. Short circuit
264 	 * to Rx path
265 	 */
266 	if (rt->dst.dev == vrf_dev)
267 		return vrf_local_xmit(skb, vrf_dev, &rt->dst);
268 
269 	skb_dst_set(skb, &rt->dst);
270 
271 	/* strip the ethernet header added for pass through VRF device */
272 	__skb_pull(skb, skb_network_offset(skb));
273 
274 	if (!ip4h->saddr) {
275 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
276 					       RT_SCOPE_LINK);
277 	}
278 
279 	ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
280 	if (unlikely(net_xmit_eval(ret)))
281 		vrf_dev->stats.tx_errors++;
282 	else
283 		ret = NET_XMIT_SUCCESS;
284 
285 out:
286 	return ret;
287 err:
288 	vrf_tx_error(vrf_dev, skb);
289 	goto out;
290 }
291 
292 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
293 {
294 	switch (skb->protocol) {
295 	case htons(ETH_P_IP):
296 		return vrf_process_v4_outbound(skb, dev);
297 	case htons(ETH_P_IPV6):
298 		return vrf_process_v6_outbound(skb, dev);
299 	default:
300 		vrf_tx_error(dev, skb);
301 		return NET_XMIT_DROP;
302 	}
303 }
304 
305 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
306 {
307 	int len = skb->len;
308 	netdev_tx_t ret = is_ip_tx_frame(skb, dev);
309 
310 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
311 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
312 
313 		u64_stats_update_begin(&dstats->syncp);
314 		dstats->tx_pkts++;
315 		dstats->tx_bytes += len;
316 		u64_stats_update_end(&dstats->syncp);
317 	} else {
318 		this_cpu_inc(dev->dstats->tx_drps);
319 	}
320 
321 	return ret;
322 }
323 
324 static int vrf_finish_direct(struct net *net, struct sock *sk,
325 			     struct sk_buff *skb)
326 {
327 	struct net_device *vrf_dev = skb->dev;
328 
329 	if (!list_empty(&vrf_dev->ptype_all) &&
330 	    likely(skb_headroom(skb) >= ETH_HLEN)) {
331 		struct ethhdr *eth = skb_push(skb, ETH_HLEN);
332 
333 		ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
334 		eth_zero_addr(eth->h_dest);
335 		eth->h_proto = skb->protocol;
336 
337 		rcu_read_lock_bh();
338 		dev_queue_xmit_nit(skb, vrf_dev);
339 		rcu_read_unlock_bh();
340 
341 		skb_pull(skb, ETH_HLEN);
342 	}
343 
344 	return 1;
345 }
346 
347 #if IS_ENABLED(CONFIG_IPV6)
348 /* modelled after ip6_finish_output2 */
349 static int vrf_finish_output6(struct net *net, struct sock *sk,
350 			      struct sk_buff *skb)
351 {
352 	struct dst_entry *dst = skb_dst(skb);
353 	struct net_device *dev = dst->dev;
354 	struct neighbour *neigh;
355 	struct in6_addr *nexthop;
356 	int ret;
357 
358 	nf_reset(skb);
359 
360 	skb->protocol = htons(ETH_P_IPV6);
361 	skb->dev = dev;
362 
363 	rcu_read_lock_bh();
364 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
365 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
366 	if (unlikely(!neigh))
367 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
368 	if (!IS_ERR(neigh)) {
369 		sock_confirm_neigh(skb, neigh);
370 		ret = neigh_output(neigh, skb);
371 		rcu_read_unlock_bh();
372 		return ret;
373 	}
374 	rcu_read_unlock_bh();
375 
376 	IP6_INC_STATS(dev_net(dst->dev),
377 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
378 	kfree_skb(skb);
379 	return -EINVAL;
380 }
381 
382 /* modelled after ip6_output */
383 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
384 {
385 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
386 			    net, sk, skb, NULL, skb_dst(skb)->dev,
387 			    vrf_finish_output6,
388 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
389 }
390 
391 /* set dst on skb to send packet to us via dev_xmit path. Allows
392  * packet to go through device based features such as qdisc, netfilter
393  * hooks and packet sockets with skb->dev set to vrf device.
394  */
395 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
396 					    struct sk_buff *skb)
397 {
398 	struct net_vrf *vrf = netdev_priv(vrf_dev);
399 	struct dst_entry *dst = NULL;
400 	struct rt6_info *rt6;
401 
402 	rcu_read_lock();
403 
404 	rt6 = rcu_dereference(vrf->rt6);
405 	if (likely(rt6)) {
406 		dst = &rt6->dst;
407 		dst_hold(dst);
408 	}
409 
410 	rcu_read_unlock();
411 
412 	if (unlikely(!dst)) {
413 		vrf_tx_error(vrf_dev, skb);
414 		return NULL;
415 	}
416 
417 	skb_dst_drop(skb);
418 	skb_dst_set(skb, dst);
419 
420 	return skb;
421 }
422 
423 static int vrf_output6_direct(struct net *net, struct sock *sk,
424 			      struct sk_buff *skb)
425 {
426 	skb->protocol = htons(ETH_P_IPV6);
427 
428 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
429 			    net, sk, skb, NULL, skb->dev,
430 			    vrf_finish_direct,
431 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
432 }
433 
434 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
435 					  struct sock *sk,
436 					  struct sk_buff *skb)
437 {
438 	struct net *net = dev_net(vrf_dev);
439 	int err;
440 
441 	skb->dev = vrf_dev;
442 
443 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
444 		      skb, NULL, vrf_dev, vrf_output6_direct);
445 
446 	if (likely(err == 1))
447 		err = vrf_output6_direct(net, sk, skb);
448 
449 	/* reset skb device */
450 	if (likely(err == 1))
451 		nf_reset(skb);
452 	else
453 		skb = NULL;
454 
455 	return skb;
456 }
457 
458 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
459 				   struct sock *sk,
460 				   struct sk_buff *skb)
461 {
462 	/* don't divert link scope packets */
463 	if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
464 		return skb;
465 
466 	if (qdisc_tx_is_default(vrf_dev))
467 		return vrf_ip6_out_direct(vrf_dev, sk, skb);
468 
469 	return vrf_ip6_out_redirect(vrf_dev, skb);
470 }
471 
472 /* holding rtnl */
473 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
474 {
475 	struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
476 	struct net *net = dev_net(dev);
477 	struct dst_entry *dst;
478 
479 	RCU_INIT_POINTER(vrf->rt6, NULL);
480 	synchronize_rcu();
481 
482 	/* move dev in dst's to loopback so this VRF device can be deleted
483 	 * - based on dst_ifdown
484 	 */
485 	if (rt6) {
486 		dst = &rt6->dst;
487 		dev_put(dst->dev);
488 		dst->dev = net->loopback_dev;
489 		dev_hold(dst->dev);
490 		dst_release(dst);
491 	}
492 }
493 
494 static int vrf_rt6_create(struct net_device *dev)
495 {
496 	int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
497 	struct net_vrf *vrf = netdev_priv(dev);
498 	struct net *net = dev_net(dev);
499 	struct fib6_table *rt6i_table;
500 	struct rt6_info *rt6;
501 	int rc = -ENOMEM;
502 
503 	/* IPv6 can be CONFIG enabled and then disabled runtime */
504 	if (!ipv6_mod_enabled())
505 		return 0;
506 
507 	rt6i_table = fib6_new_table(net, vrf->tb_id);
508 	if (!rt6i_table)
509 		goto out;
510 
511 	/* create a dst for routing packets out a VRF device */
512 	rt6 = ip6_dst_alloc(net, dev, flags);
513 	if (!rt6)
514 		goto out;
515 
516 	rt6->rt6i_table = rt6i_table;
517 	rt6->dst.output	= vrf_output6;
518 
519 	rcu_assign_pointer(vrf->rt6, rt6);
520 
521 	rc = 0;
522 out:
523 	return rc;
524 }
525 #else
526 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
527 				   struct sock *sk,
528 				   struct sk_buff *skb)
529 {
530 	return skb;
531 }
532 
533 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
534 {
535 }
536 
537 static int vrf_rt6_create(struct net_device *dev)
538 {
539 	return 0;
540 }
541 #endif
542 
543 /* modelled after ip_finish_output2 */
544 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
545 {
546 	struct dst_entry *dst = skb_dst(skb);
547 	struct rtable *rt = (struct rtable *)dst;
548 	struct net_device *dev = dst->dev;
549 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
550 	struct neighbour *neigh;
551 	u32 nexthop;
552 	int ret = -EINVAL;
553 
554 	nf_reset(skb);
555 
556 	/* Be paranoid, rather than too clever. */
557 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
558 		struct sk_buff *skb2;
559 
560 		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
561 		if (!skb2) {
562 			ret = -ENOMEM;
563 			goto err;
564 		}
565 		if (skb->sk)
566 			skb_set_owner_w(skb2, skb->sk);
567 
568 		consume_skb(skb);
569 		skb = skb2;
570 	}
571 
572 	rcu_read_lock_bh();
573 
574 	nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
575 	neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
576 	if (unlikely(!neigh))
577 		neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
578 	if (!IS_ERR(neigh)) {
579 		sock_confirm_neigh(skb, neigh);
580 		ret = neigh_output(neigh, skb);
581 	}
582 
583 	rcu_read_unlock_bh();
584 err:
585 	if (unlikely(ret < 0))
586 		vrf_tx_error(skb->dev, skb);
587 	return ret;
588 }
589 
590 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
591 {
592 	struct net_device *dev = skb_dst(skb)->dev;
593 
594 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
595 
596 	skb->dev = dev;
597 	skb->protocol = htons(ETH_P_IP);
598 
599 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
600 			    net, sk, skb, NULL, dev,
601 			    vrf_finish_output,
602 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
603 }
604 
605 /* set dst on skb to send packet to us via dev_xmit path. Allows
606  * packet to go through device based features such as qdisc, netfilter
607  * hooks and packet sockets with skb->dev set to vrf device.
608  */
609 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
610 					   struct sk_buff *skb)
611 {
612 	struct net_vrf *vrf = netdev_priv(vrf_dev);
613 	struct dst_entry *dst = NULL;
614 	struct rtable *rth;
615 
616 	rcu_read_lock();
617 
618 	rth = rcu_dereference(vrf->rth);
619 	if (likely(rth)) {
620 		dst = &rth->dst;
621 		dst_hold(dst);
622 	}
623 
624 	rcu_read_unlock();
625 
626 	if (unlikely(!dst)) {
627 		vrf_tx_error(vrf_dev, skb);
628 		return NULL;
629 	}
630 
631 	skb_dst_drop(skb);
632 	skb_dst_set(skb, dst);
633 
634 	return skb;
635 }
636 
637 static int vrf_output_direct(struct net *net, struct sock *sk,
638 			     struct sk_buff *skb)
639 {
640 	skb->protocol = htons(ETH_P_IP);
641 
642 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
643 			    net, sk, skb, NULL, skb->dev,
644 			    vrf_finish_direct,
645 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
646 }
647 
648 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
649 					 struct sock *sk,
650 					 struct sk_buff *skb)
651 {
652 	struct net *net = dev_net(vrf_dev);
653 	int err;
654 
655 	skb->dev = vrf_dev;
656 
657 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
658 		      skb, NULL, vrf_dev, vrf_output_direct);
659 
660 	if (likely(err == 1))
661 		err = vrf_output_direct(net, sk, skb);
662 
663 	/* reset skb device */
664 	if (likely(err == 1))
665 		nf_reset(skb);
666 	else
667 		skb = NULL;
668 
669 	return skb;
670 }
671 
672 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
673 				  struct sock *sk,
674 				  struct sk_buff *skb)
675 {
676 	/* don't divert multicast */
677 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
678 		return skb;
679 
680 	if (qdisc_tx_is_default(vrf_dev))
681 		return vrf_ip_out_direct(vrf_dev, sk, skb);
682 
683 	return vrf_ip_out_redirect(vrf_dev, skb);
684 }
685 
686 /* called with rcu lock held */
687 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
688 				  struct sock *sk,
689 				  struct sk_buff *skb,
690 				  u16 proto)
691 {
692 	switch (proto) {
693 	case AF_INET:
694 		return vrf_ip_out(vrf_dev, sk, skb);
695 	case AF_INET6:
696 		return vrf_ip6_out(vrf_dev, sk, skb);
697 	}
698 
699 	return skb;
700 }
701 
702 /* holding rtnl */
703 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
704 {
705 	struct rtable *rth = rtnl_dereference(vrf->rth);
706 	struct net *net = dev_net(dev);
707 	struct dst_entry *dst;
708 
709 	RCU_INIT_POINTER(vrf->rth, NULL);
710 	synchronize_rcu();
711 
712 	/* move dev in dst's to loopback so this VRF device can be deleted
713 	 * - based on dst_ifdown
714 	 */
715 	if (rth) {
716 		dst = &rth->dst;
717 		dev_put(dst->dev);
718 		dst->dev = net->loopback_dev;
719 		dev_hold(dst->dev);
720 		dst_release(dst);
721 	}
722 }
723 
724 static int vrf_rtable_create(struct net_device *dev)
725 {
726 	struct net_vrf *vrf = netdev_priv(dev);
727 	struct rtable *rth;
728 
729 	if (!fib_new_table(dev_net(dev), vrf->tb_id))
730 		return -ENOMEM;
731 
732 	/* create a dst for routing packets out through a VRF device */
733 	rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
734 	if (!rth)
735 		return -ENOMEM;
736 
737 	rth->dst.output	= vrf_output;
738 	rth->rt_table_id = vrf->tb_id;
739 
740 	rcu_assign_pointer(vrf->rth, rth);
741 
742 	return 0;
743 }
744 
745 /**************************** device handling ********************/
746 
747 /* cycle interface to flush neighbor cache and move routes across tables */
748 static void cycle_netdev(struct net_device *dev)
749 {
750 	unsigned int flags = dev->flags;
751 	int ret;
752 
753 	if (!netif_running(dev))
754 		return;
755 
756 	ret = dev_change_flags(dev, flags & ~IFF_UP);
757 	if (ret >= 0)
758 		ret = dev_change_flags(dev, flags);
759 
760 	if (ret < 0) {
761 		netdev_err(dev,
762 			   "Failed to cycle device %s; route tables might be wrong!\n",
763 			   dev->name);
764 	}
765 }
766 
767 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
768 			    struct netlink_ext_ack *extack)
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 		NL_SET_ERR_MSG(extack,
777 			       "Can not enslave loopback device to a VRF");
778 		return -EOPNOTSUPP;
779 	}
780 
781 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
782 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
783 	if (ret < 0)
784 		goto err;
785 
786 	cycle_netdev(port_dev);
787 
788 	return 0;
789 
790 err:
791 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
792 	return ret;
793 }
794 
795 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
796 			 struct netlink_ext_ack *extack)
797 {
798 	if (netif_is_l3_master(port_dev)) {
799 		NL_SET_ERR_MSG(extack,
800 			       "Can not enslave an L3 master device to a VRF");
801 		return -EINVAL;
802 	}
803 
804 	if (netif_is_l3_slave(port_dev))
805 		return -EINVAL;
806 
807 	return do_vrf_add_slave(dev, port_dev, extack);
808 }
809 
810 /* inverse of do_vrf_add_slave */
811 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
812 {
813 	netdev_upper_dev_unlink(port_dev, dev);
814 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
815 
816 	cycle_netdev(port_dev);
817 
818 	return 0;
819 }
820 
821 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
822 {
823 	return do_vrf_del_slave(dev, port_dev);
824 }
825 
826 static void vrf_dev_uninit(struct net_device *dev)
827 {
828 	struct net_vrf *vrf = netdev_priv(dev);
829 
830 	vrf_rtable_release(dev, vrf);
831 	vrf_rt6_release(dev, vrf);
832 
833 	free_percpu(dev->dstats);
834 	dev->dstats = NULL;
835 }
836 
837 static int vrf_dev_init(struct net_device *dev)
838 {
839 	struct net_vrf *vrf = netdev_priv(dev);
840 
841 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
842 	if (!dev->dstats)
843 		goto out_nomem;
844 
845 	/* create the default dst which points back to us */
846 	if (vrf_rtable_create(dev) != 0)
847 		goto out_stats;
848 
849 	if (vrf_rt6_create(dev) != 0)
850 		goto out_rth;
851 
852 	dev->flags = IFF_MASTER | IFF_NOARP;
853 
854 	/* MTU is irrelevant for VRF device; set to 64k similar to lo */
855 	dev->mtu = 64 * 1024;
856 
857 	/* similarly, oper state is irrelevant; set to up to avoid confusion */
858 	dev->operstate = IF_OPER_UP;
859 	netdev_lockdep_set_classes(dev);
860 	return 0;
861 
862 out_rth:
863 	vrf_rtable_release(dev, vrf);
864 out_stats:
865 	free_percpu(dev->dstats);
866 	dev->dstats = NULL;
867 out_nomem:
868 	return -ENOMEM;
869 }
870 
871 static const struct net_device_ops vrf_netdev_ops = {
872 	.ndo_init		= vrf_dev_init,
873 	.ndo_uninit		= vrf_dev_uninit,
874 	.ndo_start_xmit		= vrf_xmit,
875 	.ndo_get_stats64	= vrf_get_stats64,
876 	.ndo_add_slave		= vrf_add_slave,
877 	.ndo_del_slave		= vrf_del_slave,
878 };
879 
880 static u32 vrf_fib_table(const struct net_device *dev)
881 {
882 	struct net_vrf *vrf = netdev_priv(dev);
883 
884 	return vrf->tb_id;
885 }
886 
887 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
888 {
889 	kfree_skb(skb);
890 	return 0;
891 }
892 
893 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
894 				      struct sk_buff *skb,
895 				      struct net_device *dev)
896 {
897 	struct net *net = dev_net(dev);
898 
899 	if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
900 		skb = NULL;    /* kfree_skb(skb) handled by nf code */
901 
902 	return skb;
903 }
904 
905 #if IS_ENABLED(CONFIG_IPV6)
906 /* neighbor handling is done with actual device; do not want
907  * to flip skb->dev for those ndisc packets. This really fails
908  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
909  * a start.
910  */
911 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
912 {
913 	const struct ipv6hdr *iph = ipv6_hdr(skb);
914 	bool rc = false;
915 
916 	if (iph->nexthdr == NEXTHDR_ICMP) {
917 		const struct icmp6hdr *icmph;
918 		struct icmp6hdr _icmph;
919 
920 		icmph = skb_header_pointer(skb, sizeof(*iph),
921 					   sizeof(_icmph), &_icmph);
922 		if (!icmph)
923 			goto out;
924 
925 		switch (icmph->icmp6_type) {
926 		case NDISC_ROUTER_SOLICITATION:
927 		case NDISC_ROUTER_ADVERTISEMENT:
928 		case NDISC_NEIGHBOUR_SOLICITATION:
929 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
930 		case NDISC_REDIRECT:
931 			rc = true;
932 			break;
933 		}
934 	}
935 
936 out:
937 	return rc;
938 }
939 
940 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
941 					     const struct net_device *dev,
942 					     struct flowi6 *fl6,
943 					     int ifindex,
944 					     int flags)
945 {
946 	struct net_vrf *vrf = netdev_priv(dev);
947 	struct fib6_table *table = NULL;
948 	struct rt6_info *rt6;
949 
950 	rcu_read_lock();
951 
952 	/* fib6_table does not have a refcnt and can not be freed */
953 	rt6 = rcu_dereference(vrf->rt6);
954 	if (likely(rt6))
955 		table = rt6->rt6i_table;
956 
957 	rcu_read_unlock();
958 
959 	if (!table)
960 		return NULL;
961 
962 	return ip6_pol_route(net, table, ifindex, fl6, flags);
963 }
964 
965 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
966 			      int ifindex)
967 {
968 	const struct ipv6hdr *iph = ipv6_hdr(skb);
969 	struct flowi6 fl6 = {
970 		.flowi6_iif     = ifindex,
971 		.flowi6_mark    = skb->mark,
972 		.flowi6_proto   = iph->nexthdr,
973 		.daddr          = iph->daddr,
974 		.saddr          = iph->saddr,
975 		.flowlabel      = ip6_flowinfo(iph),
976 	};
977 	struct net *net = dev_net(vrf_dev);
978 	struct rt6_info *rt6;
979 
980 	rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
981 				   RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
982 	if (unlikely(!rt6))
983 		return;
984 
985 	if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
986 		return;
987 
988 	skb_dst_set(skb, &rt6->dst);
989 }
990 
991 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
992 				   struct sk_buff *skb)
993 {
994 	int orig_iif = skb->skb_iif;
995 	bool need_strict;
996 
997 	/* loopback traffic; do not push through packet taps again.
998 	 * Reset pkt_type for upper layers to process skb
999 	 */
1000 	if (skb->pkt_type == PACKET_LOOPBACK) {
1001 		skb->dev = vrf_dev;
1002 		skb->skb_iif = vrf_dev->ifindex;
1003 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1004 		skb->pkt_type = PACKET_HOST;
1005 		goto out;
1006 	}
1007 
1008 	/* if packet is NDISC or addressed to multicast or link-local
1009 	 * then keep the ingress interface
1010 	 */
1011 	need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1012 	if (!ipv6_ndisc_frame(skb) && !need_strict) {
1013 		vrf_rx_stats(vrf_dev, skb->len);
1014 		skb->dev = vrf_dev;
1015 		skb->skb_iif = vrf_dev->ifindex;
1016 
1017 		if (!list_empty(&vrf_dev->ptype_all)) {
1018 			skb_push(skb, skb->mac_len);
1019 			dev_queue_xmit_nit(skb, vrf_dev);
1020 			skb_pull(skb, skb->mac_len);
1021 		}
1022 
1023 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1024 	}
1025 
1026 	if (need_strict)
1027 		vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1028 
1029 	skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1030 out:
1031 	return skb;
1032 }
1033 
1034 #else
1035 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1036 				   struct sk_buff *skb)
1037 {
1038 	return skb;
1039 }
1040 #endif
1041 
1042 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1043 				  struct sk_buff *skb)
1044 {
1045 	skb->dev = vrf_dev;
1046 	skb->skb_iif = vrf_dev->ifindex;
1047 	IPCB(skb)->flags |= IPSKB_L3SLAVE;
1048 
1049 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1050 		goto out;
1051 
1052 	/* loopback traffic; do not push through packet taps again.
1053 	 * Reset pkt_type for upper layers to process skb
1054 	 */
1055 	if (skb->pkt_type == PACKET_LOOPBACK) {
1056 		skb->pkt_type = PACKET_HOST;
1057 		goto out;
1058 	}
1059 
1060 	vrf_rx_stats(vrf_dev, skb->len);
1061 
1062 	if (!list_empty(&vrf_dev->ptype_all)) {
1063 		skb_push(skb, skb->mac_len);
1064 		dev_queue_xmit_nit(skb, vrf_dev);
1065 		skb_pull(skb, skb->mac_len);
1066 	}
1067 
1068 	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1069 out:
1070 	return skb;
1071 }
1072 
1073 /* called with rcu lock held */
1074 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1075 				  struct sk_buff *skb,
1076 				  u16 proto)
1077 {
1078 	switch (proto) {
1079 	case AF_INET:
1080 		return vrf_ip_rcv(vrf_dev, skb);
1081 	case AF_INET6:
1082 		return vrf_ip6_rcv(vrf_dev, skb);
1083 	}
1084 
1085 	return skb;
1086 }
1087 
1088 #if IS_ENABLED(CONFIG_IPV6)
1089 /* send to link-local or multicast address via interface enslaved to
1090  * VRF device. Force lookup to VRF table without changing flow struct
1091  */
1092 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1093 					      struct flowi6 *fl6)
1094 {
1095 	struct net *net = dev_net(dev);
1096 	int flags = RT6_LOOKUP_F_IFACE;
1097 	struct dst_entry *dst = NULL;
1098 	struct rt6_info *rt;
1099 
1100 	/* VRF device does not have a link-local address and
1101 	 * sending packets to link-local or mcast addresses over
1102 	 * a VRF device does not make sense
1103 	 */
1104 	if (fl6->flowi6_oif == dev->ifindex) {
1105 		dst = &net->ipv6.ip6_null_entry->dst;
1106 		dst_hold(dst);
1107 		return dst;
1108 	}
1109 
1110 	if (!ipv6_addr_any(&fl6->saddr))
1111 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1112 
1113 	rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1114 	if (rt)
1115 		dst = &rt->dst;
1116 
1117 	return dst;
1118 }
1119 #endif
1120 
1121 static const struct l3mdev_ops vrf_l3mdev_ops = {
1122 	.l3mdev_fib_table	= vrf_fib_table,
1123 	.l3mdev_l3_rcv		= vrf_l3_rcv,
1124 	.l3mdev_l3_out		= vrf_l3_out,
1125 #if IS_ENABLED(CONFIG_IPV6)
1126 	.l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1127 #endif
1128 };
1129 
1130 static void vrf_get_drvinfo(struct net_device *dev,
1131 			    struct ethtool_drvinfo *info)
1132 {
1133 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1134 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1135 }
1136 
1137 static const struct ethtool_ops vrf_ethtool_ops = {
1138 	.get_drvinfo	= vrf_get_drvinfo,
1139 };
1140 
1141 static inline size_t vrf_fib_rule_nl_size(void)
1142 {
1143 	size_t sz;
1144 
1145 	sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1146 	sz += nla_total_size(sizeof(u8));	/* FRA_L3MDEV */
1147 	sz += nla_total_size(sizeof(u32));	/* FRA_PRIORITY */
1148 
1149 	return sz;
1150 }
1151 
1152 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1153 {
1154 	struct fib_rule_hdr *frh;
1155 	struct nlmsghdr *nlh;
1156 	struct sk_buff *skb;
1157 	int err;
1158 
1159 	if (family == AF_INET6 && !ipv6_mod_enabled())
1160 		return 0;
1161 
1162 	skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1163 	if (!skb)
1164 		return -ENOMEM;
1165 
1166 	nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1167 	if (!nlh)
1168 		goto nla_put_failure;
1169 
1170 	/* rule only needs to appear once */
1171 	nlh->nlmsg_flags |= NLM_F_EXCL;
1172 
1173 	frh = nlmsg_data(nlh);
1174 	memset(frh, 0, sizeof(*frh));
1175 	frh->family = family;
1176 	frh->action = FR_ACT_TO_TBL;
1177 
1178 	if (nla_put_u8(skb, FRA_L3MDEV, 1))
1179 		goto nla_put_failure;
1180 
1181 	if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1182 		goto nla_put_failure;
1183 
1184 	nlmsg_end(skb, nlh);
1185 
1186 	/* fib_nl_{new,del}rule handling looks for net from skb->sk */
1187 	skb->sk = dev_net(dev)->rtnl;
1188 	if (add_it) {
1189 		err = fib_nl_newrule(skb, nlh, NULL);
1190 		if (err == -EEXIST)
1191 			err = 0;
1192 	} else {
1193 		err = fib_nl_delrule(skb, nlh, NULL);
1194 		if (err == -ENOENT)
1195 			err = 0;
1196 	}
1197 	nlmsg_free(skb);
1198 
1199 	return err;
1200 
1201 nla_put_failure:
1202 	nlmsg_free(skb);
1203 
1204 	return -EMSGSIZE;
1205 }
1206 
1207 static int vrf_add_fib_rules(const struct net_device *dev)
1208 {
1209 	int err;
1210 
1211 	err = vrf_fib_rule(dev, AF_INET,  true);
1212 	if (err < 0)
1213 		goto out_err;
1214 
1215 	err = vrf_fib_rule(dev, AF_INET6, true);
1216 	if (err < 0)
1217 		goto ipv6_err;
1218 
1219 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1220 	err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1221 	if (err < 0)
1222 		goto ipmr_err;
1223 #endif
1224 
1225 	return 0;
1226 
1227 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1228 ipmr_err:
1229 	vrf_fib_rule(dev, AF_INET6,  false);
1230 #endif
1231 
1232 ipv6_err:
1233 	vrf_fib_rule(dev, AF_INET,  false);
1234 
1235 out_err:
1236 	netdev_err(dev, "Failed to add FIB rules.\n");
1237 	return err;
1238 }
1239 
1240 static void vrf_setup(struct net_device *dev)
1241 {
1242 	ether_setup(dev);
1243 
1244 	/* Initialize the device structure. */
1245 	dev->netdev_ops = &vrf_netdev_ops;
1246 	dev->l3mdev_ops = &vrf_l3mdev_ops;
1247 	dev->ethtool_ops = &vrf_ethtool_ops;
1248 	dev->needs_free_netdev = true;
1249 
1250 	/* Fill in device structure with ethernet-generic values. */
1251 	eth_hw_addr_random(dev);
1252 
1253 	/* don't acquire vrf device's netif_tx_lock when transmitting */
1254 	dev->features |= NETIF_F_LLTX;
1255 
1256 	/* don't allow vrf devices to change network namespaces. */
1257 	dev->features |= NETIF_F_NETNS_LOCAL;
1258 
1259 	/* does not make sense for a VLAN to be added to a vrf device */
1260 	dev->features   |= NETIF_F_VLAN_CHALLENGED;
1261 
1262 	/* enable offload features */
1263 	dev->features   |= NETIF_F_GSO_SOFTWARE;
1264 	dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1265 	dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1266 
1267 	dev->hw_features = dev->features;
1268 	dev->hw_enc_features = dev->features;
1269 
1270 	/* default to no qdisc; user can add if desired */
1271 	dev->priv_flags |= IFF_NO_QUEUE;
1272 }
1273 
1274 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1275 			struct netlink_ext_ack *extack)
1276 {
1277 	if (tb[IFLA_ADDRESS]) {
1278 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1279 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1280 			return -EINVAL;
1281 		}
1282 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1283 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1284 			return -EADDRNOTAVAIL;
1285 		}
1286 	}
1287 	return 0;
1288 }
1289 
1290 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1291 {
1292 	struct net_device *port_dev;
1293 	struct list_head *iter;
1294 
1295 	netdev_for_each_lower_dev(dev, port_dev, iter)
1296 		vrf_del_slave(dev, port_dev);
1297 
1298 	unregister_netdevice_queue(dev, head);
1299 }
1300 
1301 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1302 		       struct nlattr *tb[], struct nlattr *data[],
1303 		       struct netlink_ext_ack *extack)
1304 {
1305 	struct net_vrf *vrf = netdev_priv(dev);
1306 	bool *add_fib_rules;
1307 	struct net *net;
1308 	int err;
1309 
1310 	if (!data || !data[IFLA_VRF_TABLE]) {
1311 		NL_SET_ERR_MSG(extack, "VRF table id is missing");
1312 		return -EINVAL;
1313 	}
1314 
1315 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1316 	if (vrf->tb_id == RT_TABLE_UNSPEC) {
1317 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1318 				    "Invalid VRF table id");
1319 		return -EINVAL;
1320 	}
1321 
1322 	dev->priv_flags |= IFF_L3MDEV_MASTER;
1323 
1324 	err = register_netdevice(dev);
1325 	if (err)
1326 		goto out;
1327 
1328 	net = dev_net(dev);
1329 	add_fib_rules = net_generic(net, vrf_net_id);
1330 	if (*add_fib_rules) {
1331 		err = vrf_add_fib_rules(dev);
1332 		if (err) {
1333 			unregister_netdevice(dev);
1334 			goto out;
1335 		}
1336 		*add_fib_rules = false;
1337 	}
1338 
1339 out:
1340 	return err;
1341 }
1342 
1343 static size_t vrf_nl_getsize(const struct net_device *dev)
1344 {
1345 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
1346 }
1347 
1348 static int vrf_fillinfo(struct sk_buff *skb,
1349 			const struct net_device *dev)
1350 {
1351 	struct net_vrf *vrf = netdev_priv(dev);
1352 
1353 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1354 }
1355 
1356 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1357 				 const struct net_device *slave_dev)
1358 {
1359 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
1360 }
1361 
1362 static int vrf_fill_slave_info(struct sk_buff *skb,
1363 			       const struct net_device *vrf_dev,
1364 			       const struct net_device *slave_dev)
1365 {
1366 	struct net_vrf *vrf = netdev_priv(vrf_dev);
1367 
1368 	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1369 		return -EMSGSIZE;
1370 
1371 	return 0;
1372 }
1373 
1374 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1375 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
1376 };
1377 
1378 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1379 	.kind		= DRV_NAME,
1380 	.priv_size	= sizeof(struct net_vrf),
1381 
1382 	.get_size	= vrf_nl_getsize,
1383 	.policy		= vrf_nl_policy,
1384 	.validate	= vrf_validate,
1385 	.fill_info	= vrf_fillinfo,
1386 
1387 	.get_slave_size  = vrf_get_slave_size,
1388 	.fill_slave_info = vrf_fill_slave_info,
1389 
1390 	.newlink	= vrf_newlink,
1391 	.dellink	= vrf_dellink,
1392 	.setup		= vrf_setup,
1393 	.maxtype	= IFLA_VRF_MAX,
1394 };
1395 
1396 static int vrf_device_event(struct notifier_block *unused,
1397 			    unsigned long event, void *ptr)
1398 {
1399 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1400 
1401 	/* only care about unregister events to drop slave references */
1402 	if (event == NETDEV_UNREGISTER) {
1403 		struct net_device *vrf_dev;
1404 
1405 		if (!netif_is_l3_slave(dev))
1406 			goto out;
1407 
1408 		vrf_dev = netdev_master_upper_dev_get(dev);
1409 		vrf_del_slave(vrf_dev, dev);
1410 	}
1411 out:
1412 	return NOTIFY_DONE;
1413 }
1414 
1415 static struct notifier_block vrf_notifier_block __read_mostly = {
1416 	.notifier_call = vrf_device_event,
1417 };
1418 
1419 /* Initialize per network namespace state */
1420 static int __net_init vrf_netns_init(struct net *net)
1421 {
1422 	bool *add_fib_rules = net_generic(net, vrf_net_id);
1423 
1424 	*add_fib_rules = true;
1425 
1426 	return 0;
1427 }
1428 
1429 static struct pernet_operations vrf_net_ops __net_initdata = {
1430 	.init = vrf_netns_init,
1431 	.id   = &vrf_net_id,
1432 	.size = sizeof(bool),
1433 };
1434 
1435 static int __init vrf_init_module(void)
1436 {
1437 	int rc;
1438 
1439 	register_netdevice_notifier(&vrf_notifier_block);
1440 
1441 	rc = register_pernet_subsys(&vrf_net_ops);
1442 	if (rc < 0)
1443 		goto error;
1444 
1445 	rc = rtnl_link_register(&vrf_link_ops);
1446 	if (rc < 0) {
1447 		unregister_pernet_subsys(&vrf_net_ops);
1448 		goto error;
1449 	}
1450 
1451 	return 0;
1452 
1453 error:
1454 	unregister_netdevice_notifier(&vrf_notifier_block);
1455 	return rc;
1456 }
1457 
1458 module_init(vrf_init_module);
1459 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1460 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1461 MODULE_LICENSE("GPL");
1462 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1463 MODULE_VERSION(DRV_VERSION);
1464