xref: /openbmc/linux/drivers/net/vrf.c (revision 3557b3fd)
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 #if IS_ENABLED(CONFIG_IPV6)
52 	struct fib6_table	*fib6_table;
53 #endif
54 	u32                     tb_id;
55 };
56 
57 struct pcpu_dstats {
58 	u64			tx_pkts;
59 	u64			tx_bytes;
60 	u64			tx_drps;
61 	u64			rx_pkts;
62 	u64			rx_bytes;
63 	u64			rx_drps;
64 	struct u64_stats_sync	syncp;
65 };
66 
67 static void vrf_rx_stats(struct net_device *dev, int len)
68 {
69 	struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
70 
71 	u64_stats_update_begin(&dstats->syncp);
72 	dstats->rx_pkts++;
73 	dstats->rx_bytes += len;
74 	u64_stats_update_end(&dstats->syncp);
75 }
76 
77 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
78 {
79 	vrf_dev->stats.tx_errors++;
80 	kfree_skb(skb);
81 }
82 
83 static void vrf_get_stats64(struct net_device *dev,
84 			    struct rtnl_link_stats64 *stats)
85 {
86 	int i;
87 
88 	for_each_possible_cpu(i) {
89 		const struct pcpu_dstats *dstats;
90 		u64 tbytes, tpkts, tdrops, rbytes, rpkts;
91 		unsigned int start;
92 
93 		dstats = per_cpu_ptr(dev->dstats, i);
94 		do {
95 			start = u64_stats_fetch_begin_irq(&dstats->syncp);
96 			tbytes = dstats->tx_bytes;
97 			tpkts = dstats->tx_pkts;
98 			tdrops = dstats->tx_drps;
99 			rbytes = dstats->rx_bytes;
100 			rpkts = dstats->rx_pkts;
101 		} while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
102 		stats->tx_bytes += tbytes;
103 		stats->tx_packets += tpkts;
104 		stats->tx_dropped += tdrops;
105 		stats->rx_bytes += rbytes;
106 		stats->rx_packets += rpkts;
107 	}
108 }
109 
110 /* by default VRF devices do not have a qdisc and are expected
111  * to be created with only a single queue.
112  */
113 static bool qdisc_tx_is_default(const struct net_device *dev)
114 {
115 	struct netdev_queue *txq;
116 	struct Qdisc *qdisc;
117 
118 	if (dev->num_tx_queues > 1)
119 		return false;
120 
121 	txq = netdev_get_tx_queue(dev, 0);
122 	qdisc = rcu_access_pointer(txq->qdisc);
123 
124 	return !qdisc->enqueue;
125 }
126 
127 /* Local traffic destined to local address. Reinsert the packet to rx
128  * path, similar to loopback handling.
129  */
130 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
131 			  struct dst_entry *dst)
132 {
133 	int len = skb->len;
134 
135 	skb_orphan(skb);
136 
137 	skb_dst_set(skb, dst);
138 
139 	/* set pkt_type to avoid skb hitting packet taps twice -
140 	 * once on Tx and again in Rx processing
141 	 */
142 	skb->pkt_type = PACKET_LOOPBACK;
143 
144 	skb->protocol = eth_type_trans(skb, dev);
145 
146 	if (likely(netif_rx(skb) == NET_RX_SUCCESS))
147 		vrf_rx_stats(dev, len);
148 	else
149 		this_cpu_inc(dev->dstats->rx_drps);
150 
151 	return NETDEV_TX_OK;
152 }
153 
154 #if IS_ENABLED(CONFIG_IPV6)
155 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
156 			     struct sk_buff *skb)
157 {
158 	int err;
159 
160 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
161 		      sk, skb, NULL, skb_dst(skb)->dev, dst_output);
162 
163 	if (likely(err == 1))
164 		err = dst_output(net, sk, skb);
165 
166 	return err;
167 }
168 
169 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
170 					   struct net_device *dev)
171 {
172 	const struct ipv6hdr *iph = ipv6_hdr(skb);
173 	struct net *net = dev_net(skb->dev);
174 	struct flowi6 fl6 = {
175 		/* needed to match OIF rule */
176 		.flowi6_oif = dev->ifindex,
177 		.flowi6_iif = LOOPBACK_IFINDEX,
178 		.daddr = iph->daddr,
179 		.saddr = iph->saddr,
180 		.flowlabel = ip6_flowinfo(iph),
181 		.flowi6_mark = skb->mark,
182 		.flowi6_proto = iph->nexthdr,
183 		.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
184 	};
185 	int ret = NET_XMIT_DROP;
186 	struct dst_entry *dst;
187 	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
188 
189 	dst = ip6_route_output(net, NULL, &fl6);
190 	if (dst == dst_null)
191 		goto err;
192 
193 	skb_dst_drop(skb);
194 
195 	/* if dst.dev is loopback or the VRF device again this is locally
196 	 * originated traffic destined to a local address. Short circuit
197 	 * to Rx path
198 	 */
199 	if (dst->dev == dev)
200 		return vrf_local_xmit(skb, dev, dst);
201 
202 	skb_dst_set(skb, dst);
203 
204 	/* strip the ethernet header added for pass through VRF device */
205 	__skb_pull(skb, skb_network_offset(skb));
206 
207 	ret = vrf_ip6_local_out(net, skb->sk, skb);
208 	if (unlikely(net_xmit_eval(ret)))
209 		dev->stats.tx_errors++;
210 	else
211 		ret = NET_XMIT_SUCCESS;
212 
213 	return ret;
214 err:
215 	vrf_tx_error(dev, skb);
216 	return NET_XMIT_DROP;
217 }
218 #else
219 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
220 					   struct net_device *dev)
221 {
222 	vrf_tx_error(dev, skb);
223 	return NET_XMIT_DROP;
224 }
225 #endif
226 
227 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
228 static int vrf_ip_local_out(struct net *net, struct sock *sk,
229 			    struct sk_buff *skb)
230 {
231 	int err;
232 
233 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
234 		      skb, NULL, skb_dst(skb)->dev, dst_output);
235 	if (likely(err == 1))
236 		err = dst_output(net, sk, skb);
237 
238 	return err;
239 }
240 
241 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
242 					   struct net_device *vrf_dev)
243 {
244 	struct iphdr *ip4h = ip_hdr(skb);
245 	int ret = NET_XMIT_DROP;
246 	struct flowi4 fl4 = {
247 		/* needed to match OIF rule */
248 		.flowi4_oif = vrf_dev->ifindex,
249 		.flowi4_iif = LOOPBACK_IFINDEX,
250 		.flowi4_tos = RT_TOS(ip4h->tos),
251 		.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
252 		.flowi4_proto = ip4h->protocol,
253 		.daddr = ip4h->daddr,
254 		.saddr = ip4h->saddr,
255 	};
256 	struct net *net = dev_net(vrf_dev);
257 	struct rtable *rt;
258 
259 	rt = ip_route_output_flow(net, &fl4, NULL);
260 	if (IS_ERR(rt))
261 		goto err;
262 
263 	skb_dst_drop(skb);
264 
265 	/* if dst.dev is loopback or the VRF device again this is locally
266 	 * originated traffic destined to a local address. Short circuit
267 	 * to Rx path
268 	 */
269 	if (rt->dst.dev == vrf_dev)
270 		return vrf_local_xmit(skb, vrf_dev, &rt->dst);
271 
272 	skb_dst_set(skb, &rt->dst);
273 
274 	/* strip the ethernet header added for pass through VRF device */
275 	__skb_pull(skb, skb_network_offset(skb));
276 
277 	if (!ip4h->saddr) {
278 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
279 					       RT_SCOPE_LINK);
280 	}
281 
282 	ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
283 	if (unlikely(net_xmit_eval(ret)))
284 		vrf_dev->stats.tx_errors++;
285 	else
286 		ret = NET_XMIT_SUCCESS;
287 
288 out:
289 	return ret;
290 err:
291 	vrf_tx_error(vrf_dev, skb);
292 	goto out;
293 }
294 
295 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
296 {
297 	switch (skb->protocol) {
298 	case htons(ETH_P_IP):
299 		return vrf_process_v4_outbound(skb, dev);
300 	case htons(ETH_P_IPV6):
301 		return vrf_process_v6_outbound(skb, dev);
302 	default:
303 		vrf_tx_error(dev, skb);
304 		return NET_XMIT_DROP;
305 	}
306 }
307 
308 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
309 {
310 	int len = skb->len;
311 	netdev_tx_t ret = is_ip_tx_frame(skb, dev);
312 
313 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
314 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
315 
316 		u64_stats_update_begin(&dstats->syncp);
317 		dstats->tx_pkts++;
318 		dstats->tx_bytes += len;
319 		u64_stats_update_end(&dstats->syncp);
320 	} else {
321 		this_cpu_inc(dev->dstats->tx_drps);
322 	}
323 
324 	return ret;
325 }
326 
327 static int vrf_finish_direct(struct net *net, struct sock *sk,
328 			     struct sk_buff *skb)
329 {
330 	struct net_device *vrf_dev = skb->dev;
331 
332 	if (!list_empty(&vrf_dev->ptype_all) &&
333 	    likely(skb_headroom(skb) >= ETH_HLEN)) {
334 		struct ethhdr *eth = skb_push(skb, ETH_HLEN);
335 
336 		ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
337 		eth_zero_addr(eth->h_dest);
338 		eth->h_proto = skb->protocol;
339 
340 		rcu_read_lock_bh();
341 		dev_queue_xmit_nit(skb, vrf_dev);
342 		rcu_read_unlock_bh();
343 
344 		skb_pull(skb, ETH_HLEN);
345 	}
346 
347 	return 1;
348 }
349 
350 #if IS_ENABLED(CONFIG_IPV6)
351 /* modelled after ip6_finish_output2 */
352 static int vrf_finish_output6(struct net *net, struct sock *sk,
353 			      struct sk_buff *skb)
354 {
355 	struct dst_entry *dst = skb_dst(skb);
356 	struct net_device *dev = dst->dev;
357 	struct neighbour *neigh;
358 	struct in6_addr *nexthop;
359 	int ret;
360 
361 	nf_reset(skb);
362 
363 	skb->protocol = htons(ETH_P_IPV6);
364 	skb->dev = dev;
365 
366 	rcu_read_lock_bh();
367 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
368 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
369 	if (unlikely(!neigh))
370 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
371 	if (!IS_ERR(neigh)) {
372 		sock_confirm_neigh(skb, neigh);
373 		ret = neigh_output(neigh, skb, false);
374 		rcu_read_unlock_bh();
375 		return ret;
376 	}
377 	rcu_read_unlock_bh();
378 
379 	IP6_INC_STATS(dev_net(dst->dev),
380 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
381 	kfree_skb(skb);
382 	return -EINVAL;
383 }
384 
385 /* modelled after ip6_output */
386 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
387 {
388 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
389 			    net, sk, skb, NULL, skb_dst(skb)->dev,
390 			    vrf_finish_output6,
391 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
392 }
393 
394 /* set dst on skb to send packet to us via dev_xmit path. Allows
395  * packet to go through device based features such as qdisc, netfilter
396  * hooks and packet sockets with skb->dev set to vrf device.
397  */
398 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
399 					    struct sk_buff *skb)
400 {
401 	struct net_vrf *vrf = netdev_priv(vrf_dev);
402 	struct dst_entry *dst = NULL;
403 	struct rt6_info *rt6;
404 
405 	rcu_read_lock();
406 
407 	rt6 = rcu_dereference(vrf->rt6);
408 	if (likely(rt6)) {
409 		dst = &rt6->dst;
410 		dst_hold(dst);
411 	}
412 
413 	rcu_read_unlock();
414 
415 	if (unlikely(!dst)) {
416 		vrf_tx_error(vrf_dev, skb);
417 		return NULL;
418 	}
419 
420 	skb_dst_drop(skb);
421 	skb_dst_set(skb, dst);
422 
423 	return skb;
424 }
425 
426 static int vrf_output6_direct(struct net *net, struct sock *sk,
427 			      struct sk_buff *skb)
428 {
429 	skb->protocol = htons(ETH_P_IPV6);
430 
431 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
432 			    net, sk, skb, NULL, skb->dev,
433 			    vrf_finish_direct,
434 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
435 }
436 
437 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
438 					  struct sock *sk,
439 					  struct sk_buff *skb)
440 {
441 	struct net *net = dev_net(vrf_dev);
442 	int err;
443 
444 	skb->dev = vrf_dev;
445 
446 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
447 		      skb, NULL, vrf_dev, vrf_output6_direct);
448 
449 	if (likely(err == 1))
450 		err = vrf_output6_direct(net, sk, skb);
451 
452 	/* reset skb device */
453 	if (likely(err == 1))
454 		nf_reset(skb);
455 	else
456 		skb = NULL;
457 
458 	return skb;
459 }
460 
461 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
462 				   struct sock *sk,
463 				   struct sk_buff *skb)
464 {
465 	/* don't divert link scope packets */
466 	if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
467 		return skb;
468 
469 	if (qdisc_tx_is_default(vrf_dev))
470 		return vrf_ip6_out_direct(vrf_dev, sk, skb);
471 
472 	return vrf_ip6_out_redirect(vrf_dev, skb);
473 }
474 
475 /* holding rtnl */
476 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
477 {
478 	struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
479 	struct net *net = dev_net(dev);
480 	struct dst_entry *dst;
481 
482 	RCU_INIT_POINTER(vrf->rt6, NULL);
483 	synchronize_rcu();
484 
485 	/* move dev in dst's to loopback so this VRF device can be deleted
486 	 * - based on dst_ifdown
487 	 */
488 	if (rt6) {
489 		dst = &rt6->dst;
490 		dev_put(dst->dev);
491 		dst->dev = net->loopback_dev;
492 		dev_hold(dst->dev);
493 		dst_release(dst);
494 	}
495 }
496 
497 static int vrf_rt6_create(struct net_device *dev)
498 {
499 	int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
500 	struct net_vrf *vrf = netdev_priv(dev);
501 	struct net *net = dev_net(dev);
502 	struct rt6_info *rt6;
503 	int rc = -ENOMEM;
504 
505 	/* IPv6 can be CONFIG enabled and then disabled runtime */
506 	if (!ipv6_mod_enabled())
507 		return 0;
508 
509 	vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
510 	if (!vrf->fib6_table)
511 		goto out;
512 
513 	/* create a dst for routing packets out a VRF device */
514 	rt6 = ip6_dst_alloc(net, dev, flags);
515 	if (!rt6)
516 		goto out;
517 
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 	bool is_v6gw = false;
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 	neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
576 	if (!IS_ERR(neigh)) {
577 		sock_confirm_neigh(skb, neigh);
578 		/* if crossing protocols, can not use the cached header */
579 		ret = neigh_output(neigh, skb, is_v6gw);
580 		rcu_read_unlock_bh();
581 		return ret;
582 	}
583 
584 	rcu_read_unlock_bh();
585 err:
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 or local broadcast */
677 	if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
678 	    ipv4_is_lbcast(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 
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 			 struct netlink_ext_ack *extack)
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, extack);
758 	if (ret >= 0)
759 		ret = dev_change_flags(dev, flags, extack);
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 			    struct netlink_ext_ack *extack)
770 {
771 	int ret;
772 
773 	/* do not allow loopback device to be enslaved to a VRF.
774 	 * The vrf device acts as the loopback for the vrf.
775 	 */
776 	if (port_dev == dev_net(dev)->loopback_dev) {
777 		NL_SET_ERR_MSG(extack,
778 			       "Can not enslave loopback device to a VRF");
779 		return -EOPNOTSUPP;
780 	}
781 
782 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
783 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
784 	if (ret < 0)
785 		goto err;
786 
787 	cycle_netdev(port_dev, extack);
788 
789 	return 0;
790 
791 err:
792 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
793 	return ret;
794 }
795 
796 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
797 			 struct netlink_ext_ack *extack)
798 {
799 	if (netif_is_l3_master(port_dev)) {
800 		NL_SET_ERR_MSG(extack,
801 			       "Can not enslave an L3 master device to a VRF");
802 		return -EINVAL;
803 	}
804 
805 	if (netif_is_l3_slave(port_dev))
806 		return -EINVAL;
807 
808 	return do_vrf_add_slave(dev, port_dev, extack);
809 }
810 
811 /* inverse of do_vrf_add_slave */
812 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
813 {
814 	netdev_upper_dev_unlink(port_dev, dev);
815 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
816 
817 	cycle_netdev(port_dev, NULL);
818 
819 	return 0;
820 }
821 
822 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
823 {
824 	return do_vrf_del_slave(dev, port_dev);
825 }
826 
827 static void vrf_dev_uninit(struct net_device *dev)
828 {
829 	struct net_vrf *vrf = netdev_priv(dev);
830 
831 	vrf_rtable_release(dev, vrf);
832 	vrf_rt6_release(dev, vrf);
833 
834 	free_percpu(dev->dstats);
835 	dev->dstats = NULL;
836 }
837 
838 static int vrf_dev_init(struct net_device *dev)
839 {
840 	struct net_vrf *vrf = netdev_priv(dev);
841 
842 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
843 	if (!dev->dstats)
844 		goto out_nomem;
845 
846 	/* create the default dst which points back to us */
847 	if (vrf_rtable_create(dev) != 0)
848 		goto out_stats;
849 
850 	if (vrf_rt6_create(dev) != 0)
851 		goto out_rth;
852 
853 	dev->flags = IFF_MASTER | IFF_NOARP;
854 
855 	/* MTU is irrelevant for VRF device; set to 64k similar to lo */
856 	dev->mtu = 64 * 1024;
857 
858 	/* similarly, oper state is irrelevant; set to up to avoid confusion */
859 	dev->operstate = IF_OPER_UP;
860 	netdev_lockdep_set_classes(dev);
861 	return 0;
862 
863 out_rth:
864 	vrf_rtable_release(dev, vrf);
865 out_stats:
866 	free_percpu(dev->dstats);
867 	dev->dstats = NULL;
868 out_nomem:
869 	return -ENOMEM;
870 }
871 
872 static const struct net_device_ops vrf_netdev_ops = {
873 	.ndo_init		= vrf_dev_init,
874 	.ndo_uninit		= vrf_dev_uninit,
875 	.ndo_start_xmit		= vrf_xmit,
876 	.ndo_get_stats64	= vrf_get_stats64,
877 	.ndo_add_slave		= vrf_add_slave,
878 	.ndo_del_slave		= vrf_del_slave,
879 };
880 
881 static u32 vrf_fib_table(const struct net_device *dev)
882 {
883 	struct net_vrf *vrf = netdev_priv(dev);
884 
885 	return vrf->tb_id;
886 }
887 
888 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
889 {
890 	kfree_skb(skb);
891 	return 0;
892 }
893 
894 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
895 				      struct sk_buff *skb,
896 				      struct net_device *dev)
897 {
898 	struct net *net = dev_net(dev);
899 
900 	if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
901 		skb = NULL;    /* kfree_skb(skb) handled by nf code */
902 
903 	return skb;
904 }
905 
906 #if IS_ENABLED(CONFIG_IPV6)
907 /* neighbor handling is done with actual device; do not want
908  * to flip skb->dev for those ndisc packets. This really fails
909  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
910  * a start.
911  */
912 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
913 {
914 	const struct ipv6hdr *iph = ipv6_hdr(skb);
915 	bool rc = false;
916 
917 	if (iph->nexthdr == NEXTHDR_ICMP) {
918 		const struct icmp6hdr *icmph;
919 		struct icmp6hdr _icmph;
920 
921 		icmph = skb_header_pointer(skb, sizeof(*iph),
922 					   sizeof(_icmph), &_icmph);
923 		if (!icmph)
924 			goto out;
925 
926 		switch (icmph->icmp6_type) {
927 		case NDISC_ROUTER_SOLICITATION:
928 		case NDISC_ROUTER_ADVERTISEMENT:
929 		case NDISC_NEIGHBOUR_SOLICITATION:
930 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
931 		case NDISC_REDIRECT:
932 			rc = true;
933 			break;
934 		}
935 	}
936 
937 out:
938 	return rc;
939 }
940 
941 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
942 					     const struct net_device *dev,
943 					     struct flowi6 *fl6,
944 					     int ifindex,
945 					     const struct sk_buff *skb,
946 					     int flags)
947 {
948 	struct net_vrf *vrf = netdev_priv(dev);
949 
950 	return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
951 }
952 
953 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
954 			      int ifindex)
955 {
956 	const struct ipv6hdr *iph = ipv6_hdr(skb);
957 	struct flowi6 fl6 = {
958 		.flowi6_iif     = ifindex,
959 		.flowi6_mark    = skb->mark,
960 		.flowi6_proto   = iph->nexthdr,
961 		.daddr          = iph->daddr,
962 		.saddr          = iph->saddr,
963 		.flowlabel      = ip6_flowinfo(iph),
964 	};
965 	struct net *net = dev_net(vrf_dev);
966 	struct rt6_info *rt6;
967 
968 	rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
969 				   RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
970 	if (unlikely(!rt6))
971 		return;
972 
973 	if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
974 		return;
975 
976 	skb_dst_set(skb, &rt6->dst);
977 }
978 
979 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
980 				   struct sk_buff *skb)
981 {
982 	int orig_iif = skb->skb_iif;
983 	bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
984 	bool is_ndisc = ipv6_ndisc_frame(skb);
985 
986 	/* loopback, multicast & non-ND link-local traffic; do not push through
987 	 * packet taps again. Reset pkt_type for upper layers to process skb
988 	 */
989 	if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
990 		skb->dev = vrf_dev;
991 		skb->skb_iif = vrf_dev->ifindex;
992 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
993 		if (skb->pkt_type == PACKET_LOOPBACK)
994 			skb->pkt_type = PACKET_HOST;
995 		goto out;
996 	}
997 
998 	/* if packet is NDISC then keep the ingress interface */
999 	if (!is_ndisc) {
1000 		vrf_rx_stats(vrf_dev, skb->len);
1001 		skb->dev = vrf_dev;
1002 		skb->skb_iif = vrf_dev->ifindex;
1003 
1004 		if (!list_empty(&vrf_dev->ptype_all)) {
1005 			skb_push(skb, skb->mac_len);
1006 			dev_queue_xmit_nit(skb, vrf_dev);
1007 			skb_pull(skb, skb->mac_len);
1008 		}
1009 
1010 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1011 	}
1012 
1013 	if (need_strict)
1014 		vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1015 
1016 	skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1017 out:
1018 	return skb;
1019 }
1020 
1021 #else
1022 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1023 				   struct sk_buff *skb)
1024 {
1025 	return skb;
1026 }
1027 #endif
1028 
1029 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1030 				  struct sk_buff *skb)
1031 {
1032 	skb->dev = vrf_dev;
1033 	skb->skb_iif = vrf_dev->ifindex;
1034 	IPCB(skb)->flags |= IPSKB_L3SLAVE;
1035 
1036 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1037 		goto out;
1038 
1039 	/* loopback traffic; do not push through packet taps again.
1040 	 * Reset pkt_type for upper layers to process skb
1041 	 */
1042 	if (skb->pkt_type == PACKET_LOOPBACK) {
1043 		skb->pkt_type = PACKET_HOST;
1044 		goto out;
1045 	}
1046 
1047 	vrf_rx_stats(vrf_dev, skb->len);
1048 
1049 	if (!list_empty(&vrf_dev->ptype_all)) {
1050 		skb_push(skb, skb->mac_len);
1051 		dev_queue_xmit_nit(skb, vrf_dev);
1052 		skb_pull(skb, skb->mac_len);
1053 	}
1054 
1055 	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1056 out:
1057 	return skb;
1058 }
1059 
1060 /* called with rcu lock held */
1061 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1062 				  struct sk_buff *skb,
1063 				  u16 proto)
1064 {
1065 	switch (proto) {
1066 	case AF_INET:
1067 		return vrf_ip_rcv(vrf_dev, skb);
1068 	case AF_INET6:
1069 		return vrf_ip6_rcv(vrf_dev, skb);
1070 	}
1071 
1072 	return skb;
1073 }
1074 
1075 #if IS_ENABLED(CONFIG_IPV6)
1076 /* send to link-local or multicast address via interface enslaved to
1077  * VRF device. Force lookup to VRF table without changing flow struct
1078  */
1079 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1080 					      struct flowi6 *fl6)
1081 {
1082 	struct net *net = dev_net(dev);
1083 	int flags = RT6_LOOKUP_F_IFACE;
1084 	struct dst_entry *dst = NULL;
1085 	struct rt6_info *rt;
1086 
1087 	/* VRF device does not have a link-local address and
1088 	 * sending packets to link-local or mcast addresses over
1089 	 * a VRF device does not make sense
1090 	 */
1091 	if (fl6->flowi6_oif == dev->ifindex) {
1092 		dst = &net->ipv6.ip6_null_entry->dst;
1093 		dst_hold(dst);
1094 		return dst;
1095 	}
1096 
1097 	if (!ipv6_addr_any(&fl6->saddr))
1098 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1099 
1100 	rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1101 	if (rt)
1102 		dst = &rt->dst;
1103 
1104 	return dst;
1105 }
1106 #endif
1107 
1108 static const struct l3mdev_ops vrf_l3mdev_ops = {
1109 	.l3mdev_fib_table	= vrf_fib_table,
1110 	.l3mdev_l3_rcv		= vrf_l3_rcv,
1111 	.l3mdev_l3_out		= vrf_l3_out,
1112 #if IS_ENABLED(CONFIG_IPV6)
1113 	.l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1114 #endif
1115 };
1116 
1117 static void vrf_get_drvinfo(struct net_device *dev,
1118 			    struct ethtool_drvinfo *info)
1119 {
1120 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1121 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1122 }
1123 
1124 static const struct ethtool_ops vrf_ethtool_ops = {
1125 	.get_drvinfo	= vrf_get_drvinfo,
1126 };
1127 
1128 static inline size_t vrf_fib_rule_nl_size(void)
1129 {
1130 	size_t sz;
1131 
1132 	sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1133 	sz += nla_total_size(sizeof(u8));	/* FRA_L3MDEV */
1134 	sz += nla_total_size(sizeof(u32));	/* FRA_PRIORITY */
1135 	sz += nla_total_size(sizeof(u8));       /* FRA_PROTOCOL */
1136 
1137 	return sz;
1138 }
1139 
1140 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1141 {
1142 	struct fib_rule_hdr *frh;
1143 	struct nlmsghdr *nlh;
1144 	struct sk_buff *skb;
1145 	int err;
1146 
1147 	if (family == AF_INET6 && !ipv6_mod_enabled())
1148 		return 0;
1149 
1150 	skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1151 	if (!skb)
1152 		return -ENOMEM;
1153 
1154 	nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1155 	if (!nlh)
1156 		goto nla_put_failure;
1157 
1158 	/* rule only needs to appear once */
1159 	nlh->nlmsg_flags |= NLM_F_EXCL;
1160 
1161 	frh = nlmsg_data(nlh);
1162 	memset(frh, 0, sizeof(*frh));
1163 	frh->family = family;
1164 	frh->action = FR_ACT_TO_TBL;
1165 
1166 	if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1167 		goto nla_put_failure;
1168 
1169 	if (nla_put_u8(skb, FRA_L3MDEV, 1))
1170 		goto nla_put_failure;
1171 
1172 	if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1173 		goto nla_put_failure;
1174 
1175 	nlmsg_end(skb, nlh);
1176 
1177 	/* fib_nl_{new,del}rule handling looks for net from skb->sk */
1178 	skb->sk = dev_net(dev)->rtnl;
1179 	if (add_it) {
1180 		err = fib_nl_newrule(skb, nlh, NULL);
1181 		if (err == -EEXIST)
1182 			err = 0;
1183 	} else {
1184 		err = fib_nl_delrule(skb, nlh, NULL);
1185 		if (err == -ENOENT)
1186 			err = 0;
1187 	}
1188 	nlmsg_free(skb);
1189 
1190 	return err;
1191 
1192 nla_put_failure:
1193 	nlmsg_free(skb);
1194 
1195 	return -EMSGSIZE;
1196 }
1197 
1198 static int vrf_add_fib_rules(const struct net_device *dev)
1199 {
1200 	int err;
1201 
1202 	err = vrf_fib_rule(dev, AF_INET,  true);
1203 	if (err < 0)
1204 		goto out_err;
1205 
1206 	err = vrf_fib_rule(dev, AF_INET6, true);
1207 	if (err < 0)
1208 		goto ipv6_err;
1209 
1210 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1211 	err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1212 	if (err < 0)
1213 		goto ipmr_err;
1214 #endif
1215 
1216 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1217 	err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1218 	if (err < 0)
1219 		goto ip6mr_err;
1220 #endif
1221 
1222 	return 0;
1223 
1224 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1225 ip6mr_err:
1226 	vrf_fib_rule(dev, RTNL_FAMILY_IPMR,  false);
1227 #endif
1228 
1229 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1230 ipmr_err:
1231 	vrf_fib_rule(dev, AF_INET6,  false);
1232 #endif
1233 
1234 ipv6_err:
1235 	vrf_fib_rule(dev, AF_INET,  false);
1236 
1237 out_err:
1238 	netdev_err(dev, "Failed to add FIB rules.\n");
1239 	return err;
1240 }
1241 
1242 static void vrf_setup(struct net_device *dev)
1243 {
1244 	ether_setup(dev);
1245 
1246 	/* Initialize the device structure. */
1247 	dev->netdev_ops = &vrf_netdev_ops;
1248 	dev->l3mdev_ops = &vrf_l3mdev_ops;
1249 	dev->ethtool_ops = &vrf_ethtool_ops;
1250 	dev->needs_free_netdev = true;
1251 
1252 	/* Fill in device structure with ethernet-generic values. */
1253 	eth_hw_addr_random(dev);
1254 
1255 	/* don't acquire vrf device's netif_tx_lock when transmitting */
1256 	dev->features |= NETIF_F_LLTX;
1257 
1258 	/* don't allow vrf devices to change network namespaces. */
1259 	dev->features |= NETIF_F_NETNS_LOCAL;
1260 
1261 	/* does not make sense for a VLAN to be added to a vrf device */
1262 	dev->features   |= NETIF_F_VLAN_CHALLENGED;
1263 
1264 	/* enable offload features */
1265 	dev->features   |= NETIF_F_GSO_SOFTWARE;
1266 	dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1267 	dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1268 
1269 	dev->hw_features = dev->features;
1270 	dev->hw_enc_features = dev->features;
1271 
1272 	/* default to no qdisc; user can add if desired */
1273 	dev->priv_flags |= IFF_NO_QUEUE;
1274 	dev->priv_flags |= IFF_NO_RX_HANDLER;
1275 
1276 	/* VRF devices do not care about MTU, but if the MTU is set
1277 	 * too low then the ipv4 and ipv6 protocols are disabled
1278 	 * which breaks networking.
1279 	 */
1280 	dev->min_mtu = IPV6_MIN_MTU;
1281 	dev->max_mtu = ETH_MAX_MTU;
1282 }
1283 
1284 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1285 			struct netlink_ext_ack *extack)
1286 {
1287 	if (tb[IFLA_ADDRESS]) {
1288 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1289 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1290 			return -EINVAL;
1291 		}
1292 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1293 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1294 			return -EADDRNOTAVAIL;
1295 		}
1296 	}
1297 	return 0;
1298 }
1299 
1300 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1301 {
1302 	struct net_device *port_dev;
1303 	struct list_head *iter;
1304 
1305 	netdev_for_each_lower_dev(dev, port_dev, iter)
1306 		vrf_del_slave(dev, port_dev);
1307 
1308 	unregister_netdevice_queue(dev, head);
1309 }
1310 
1311 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1312 		       struct nlattr *tb[], struct nlattr *data[],
1313 		       struct netlink_ext_ack *extack)
1314 {
1315 	struct net_vrf *vrf = netdev_priv(dev);
1316 	bool *add_fib_rules;
1317 	struct net *net;
1318 	int err;
1319 
1320 	if (!data || !data[IFLA_VRF_TABLE]) {
1321 		NL_SET_ERR_MSG(extack, "VRF table id is missing");
1322 		return -EINVAL;
1323 	}
1324 
1325 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1326 	if (vrf->tb_id == RT_TABLE_UNSPEC) {
1327 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1328 				    "Invalid VRF table id");
1329 		return -EINVAL;
1330 	}
1331 
1332 	dev->priv_flags |= IFF_L3MDEV_MASTER;
1333 
1334 	err = register_netdevice(dev);
1335 	if (err)
1336 		goto out;
1337 
1338 	net = dev_net(dev);
1339 	add_fib_rules = net_generic(net, vrf_net_id);
1340 	if (*add_fib_rules) {
1341 		err = vrf_add_fib_rules(dev);
1342 		if (err) {
1343 			unregister_netdevice(dev);
1344 			goto out;
1345 		}
1346 		*add_fib_rules = false;
1347 	}
1348 
1349 out:
1350 	return err;
1351 }
1352 
1353 static size_t vrf_nl_getsize(const struct net_device *dev)
1354 {
1355 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
1356 }
1357 
1358 static int vrf_fillinfo(struct sk_buff *skb,
1359 			const struct net_device *dev)
1360 {
1361 	struct net_vrf *vrf = netdev_priv(dev);
1362 
1363 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1364 }
1365 
1366 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1367 				 const struct net_device *slave_dev)
1368 {
1369 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
1370 }
1371 
1372 static int vrf_fill_slave_info(struct sk_buff *skb,
1373 			       const struct net_device *vrf_dev,
1374 			       const struct net_device *slave_dev)
1375 {
1376 	struct net_vrf *vrf = netdev_priv(vrf_dev);
1377 
1378 	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1379 		return -EMSGSIZE;
1380 
1381 	return 0;
1382 }
1383 
1384 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1385 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
1386 };
1387 
1388 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1389 	.kind		= DRV_NAME,
1390 	.priv_size	= sizeof(struct net_vrf),
1391 
1392 	.get_size	= vrf_nl_getsize,
1393 	.policy		= vrf_nl_policy,
1394 	.validate	= vrf_validate,
1395 	.fill_info	= vrf_fillinfo,
1396 
1397 	.get_slave_size  = vrf_get_slave_size,
1398 	.fill_slave_info = vrf_fill_slave_info,
1399 
1400 	.newlink	= vrf_newlink,
1401 	.dellink	= vrf_dellink,
1402 	.setup		= vrf_setup,
1403 	.maxtype	= IFLA_VRF_MAX,
1404 };
1405 
1406 static int vrf_device_event(struct notifier_block *unused,
1407 			    unsigned long event, void *ptr)
1408 {
1409 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1410 
1411 	/* only care about unregister events to drop slave references */
1412 	if (event == NETDEV_UNREGISTER) {
1413 		struct net_device *vrf_dev;
1414 
1415 		if (!netif_is_l3_slave(dev))
1416 			goto out;
1417 
1418 		vrf_dev = netdev_master_upper_dev_get(dev);
1419 		vrf_del_slave(vrf_dev, dev);
1420 	}
1421 out:
1422 	return NOTIFY_DONE;
1423 }
1424 
1425 static struct notifier_block vrf_notifier_block __read_mostly = {
1426 	.notifier_call = vrf_device_event,
1427 };
1428 
1429 /* Initialize per network namespace state */
1430 static int __net_init vrf_netns_init(struct net *net)
1431 {
1432 	bool *add_fib_rules = net_generic(net, vrf_net_id);
1433 
1434 	*add_fib_rules = true;
1435 
1436 	return 0;
1437 }
1438 
1439 static struct pernet_operations vrf_net_ops __net_initdata = {
1440 	.init = vrf_netns_init,
1441 	.id   = &vrf_net_id,
1442 	.size = sizeof(bool),
1443 };
1444 
1445 static int __init vrf_init_module(void)
1446 {
1447 	int rc;
1448 
1449 	register_netdevice_notifier(&vrf_notifier_block);
1450 
1451 	rc = register_pernet_subsys(&vrf_net_ops);
1452 	if (rc < 0)
1453 		goto error;
1454 
1455 	rc = rtnl_link_register(&vrf_link_ops);
1456 	if (rc < 0) {
1457 		unregister_pernet_subsys(&vrf_net_ops);
1458 		goto error;
1459 	}
1460 
1461 	return 0;
1462 
1463 error:
1464 	unregister_netdevice_notifier(&vrf_notifier_block);
1465 	return rc;
1466 }
1467 
1468 module_init(vrf_init_module);
1469 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1470 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1471 MODULE_LICENSE("GPL");
1472 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1473 MODULE_VERSION(DRV_VERSION);
1474