xref: /openbmc/linux/drivers/net/vrf.c (revision 04dc82e1)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * vrf.c: device driver to encapsulate a VRF space
4  *
5  * Copyright (c) 2015 Cumulus Networks. All rights reserved.
6  * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com>
7  * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com>
8  *
9  * Based on dummy, team and ipvlan drivers
10  */
11 
12 #include <linux/module.h>
13 #include <linux/kernel.h>
14 #include <linux/netdevice.h>
15 #include <linux/etherdevice.h>
16 #include <linux/ip.h>
17 #include <linux/init.h>
18 #include <linux/moduleparam.h>
19 #include <linux/netfilter.h>
20 #include <linux/rtnetlink.h>
21 #include <net/rtnetlink.h>
22 #include <linux/u64_stats_sync.h>
23 #include <linux/hashtable.h>
24 
25 #include <linux/inetdevice.h>
26 #include <net/arp.h>
27 #include <net/ip.h>
28 #include <net/ip_fib.h>
29 #include <net/ip6_fib.h>
30 #include <net/ip6_route.h>
31 #include <net/route.h>
32 #include <net/addrconf.h>
33 #include <net/l3mdev.h>
34 #include <net/fib_rules.h>
35 #include <net/netns/generic.h>
36 
37 #define DRV_NAME	"vrf"
38 #define DRV_VERSION	"1.0"
39 
40 #define FIB_RULE_PREF  1000       /* default preference for FIB rules */
41 
42 static unsigned int vrf_net_id;
43 
44 struct net_vrf {
45 	struct rtable __rcu	*rth;
46 	struct rt6_info	__rcu	*rt6;
47 #if IS_ENABLED(CONFIG_IPV6)
48 	struct fib6_table	*fib6_table;
49 #endif
50 	u32                     tb_id;
51 };
52 
53 struct pcpu_dstats {
54 	u64			tx_pkts;
55 	u64			tx_bytes;
56 	u64			tx_drps;
57 	u64			rx_pkts;
58 	u64			rx_bytes;
59 	u64			rx_drps;
60 	struct u64_stats_sync	syncp;
61 };
62 
63 static void vrf_rx_stats(struct net_device *dev, int len)
64 {
65 	struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
66 
67 	u64_stats_update_begin(&dstats->syncp);
68 	dstats->rx_pkts++;
69 	dstats->rx_bytes += len;
70 	u64_stats_update_end(&dstats->syncp);
71 }
72 
73 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
74 {
75 	vrf_dev->stats.tx_errors++;
76 	kfree_skb(skb);
77 }
78 
79 static void vrf_get_stats64(struct net_device *dev,
80 			    struct rtnl_link_stats64 *stats)
81 {
82 	int i;
83 
84 	for_each_possible_cpu(i) {
85 		const struct pcpu_dstats *dstats;
86 		u64 tbytes, tpkts, tdrops, rbytes, rpkts;
87 		unsigned int start;
88 
89 		dstats = per_cpu_ptr(dev->dstats, i);
90 		do {
91 			start = u64_stats_fetch_begin_irq(&dstats->syncp);
92 			tbytes = dstats->tx_bytes;
93 			tpkts = dstats->tx_pkts;
94 			tdrops = dstats->tx_drps;
95 			rbytes = dstats->rx_bytes;
96 			rpkts = dstats->rx_pkts;
97 		} while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
98 		stats->tx_bytes += tbytes;
99 		stats->tx_packets += tpkts;
100 		stats->tx_dropped += tdrops;
101 		stats->rx_bytes += rbytes;
102 		stats->rx_packets += rpkts;
103 	}
104 }
105 
106 /* by default VRF devices do not have a qdisc and are expected
107  * to be created with only a single queue.
108  */
109 static bool qdisc_tx_is_default(const struct net_device *dev)
110 {
111 	struct netdev_queue *txq;
112 	struct Qdisc *qdisc;
113 
114 	if (dev->num_tx_queues > 1)
115 		return false;
116 
117 	txq = netdev_get_tx_queue(dev, 0);
118 	qdisc = rcu_access_pointer(txq->qdisc);
119 
120 	return !qdisc->enqueue;
121 }
122 
123 /* Local traffic destined to local address. Reinsert the packet to rx
124  * path, similar to loopback handling.
125  */
126 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
127 			  struct dst_entry *dst)
128 {
129 	int len = skb->len;
130 
131 	skb_orphan(skb);
132 
133 	skb_dst_set(skb, dst);
134 
135 	/* set pkt_type to avoid skb hitting packet taps twice -
136 	 * once on Tx and again in Rx processing
137 	 */
138 	skb->pkt_type = PACKET_LOOPBACK;
139 
140 	skb->protocol = eth_type_trans(skb, dev);
141 
142 	if (likely(netif_rx(skb) == NET_RX_SUCCESS))
143 		vrf_rx_stats(dev, len);
144 	else
145 		this_cpu_inc(dev->dstats->rx_drps);
146 
147 	return NETDEV_TX_OK;
148 }
149 
150 #if IS_ENABLED(CONFIG_IPV6)
151 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
152 			     struct sk_buff *skb)
153 {
154 	int err;
155 
156 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
157 		      sk, skb, NULL, skb_dst(skb)->dev, dst_output);
158 
159 	if (likely(err == 1))
160 		err = dst_output(net, sk, skb);
161 
162 	return err;
163 }
164 
165 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
166 					   struct net_device *dev)
167 {
168 	const struct ipv6hdr *iph = ipv6_hdr(skb);
169 	struct net *net = dev_net(skb->dev);
170 	struct flowi6 fl6 = {
171 		/* needed to match OIF rule */
172 		.flowi6_oif = dev->ifindex,
173 		.flowi6_iif = LOOPBACK_IFINDEX,
174 		.daddr = iph->daddr,
175 		.saddr = iph->saddr,
176 		.flowlabel = ip6_flowinfo(iph),
177 		.flowi6_mark = skb->mark,
178 		.flowi6_proto = iph->nexthdr,
179 		.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
180 	};
181 	int ret = NET_XMIT_DROP;
182 	struct dst_entry *dst;
183 	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
184 
185 	dst = ip6_route_output(net, NULL, &fl6);
186 	if (dst == dst_null)
187 		goto err;
188 
189 	skb_dst_drop(skb);
190 
191 	/* if dst.dev is loopback or the VRF device again this is locally
192 	 * originated traffic destined to a local address. Short circuit
193 	 * to Rx path
194 	 */
195 	if (dst->dev == dev)
196 		return vrf_local_xmit(skb, dev, dst);
197 
198 	skb_dst_set(skb, dst);
199 
200 	/* strip the ethernet header added for pass through VRF device */
201 	__skb_pull(skb, skb_network_offset(skb));
202 
203 	ret = vrf_ip6_local_out(net, skb->sk, skb);
204 	if (unlikely(net_xmit_eval(ret)))
205 		dev->stats.tx_errors++;
206 	else
207 		ret = NET_XMIT_SUCCESS;
208 
209 	return ret;
210 err:
211 	vrf_tx_error(dev, skb);
212 	return NET_XMIT_DROP;
213 }
214 #else
215 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
216 					   struct net_device *dev)
217 {
218 	vrf_tx_error(dev, skb);
219 	return NET_XMIT_DROP;
220 }
221 #endif
222 
223 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
224 static int vrf_ip_local_out(struct net *net, struct sock *sk,
225 			    struct sk_buff *skb)
226 {
227 	int err;
228 
229 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
230 		      skb, NULL, skb_dst(skb)->dev, dst_output);
231 	if (likely(err == 1))
232 		err = dst_output(net, sk, skb);
233 
234 	return err;
235 }
236 
237 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
238 					   struct net_device *vrf_dev)
239 {
240 	struct iphdr *ip4h = ip_hdr(skb);
241 	int ret = NET_XMIT_DROP;
242 	struct flowi4 fl4 = {
243 		/* needed to match OIF rule */
244 		.flowi4_oif = vrf_dev->ifindex,
245 		.flowi4_iif = LOOPBACK_IFINDEX,
246 		.flowi4_tos = RT_TOS(ip4h->tos),
247 		.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
248 		.flowi4_proto = ip4h->protocol,
249 		.daddr = ip4h->daddr,
250 		.saddr = ip4h->saddr,
251 	};
252 	struct net *net = dev_net(vrf_dev);
253 	struct rtable *rt;
254 
255 	rt = ip_route_output_flow(net, &fl4, NULL);
256 	if (IS_ERR(rt))
257 		goto err;
258 
259 	skb_dst_drop(skb);
260 
261 	/* if dst.dev is loopback or the VRF device again this is locally
262 	 * originated traffic destined to a local address. Short circuit
263 	 * to Rx path
264 	 */
265 	if (rt->dst.dev == vrf_dev)
266 		return vrf_local_xmit(skb, vrf_dev, &rt->dst);
267 
268 	skb_dst_set(skb, &rt->dst);
269 
270 	/* strip the ethernet header added for pass through VRF device */
271 	__skb_pull(skb, skb_network_offset(skb));
272 
273 	if (!ip4h->saddr) {
274 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
275 					       RT_SCOPE_LINK);
276 	}
277 
278 	ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
279 	if (unlikely(net_xmit_eval(ret)))
280 		vrf_dev->stats.tx_errors++;
281 	else
282 		ret = NET_XMIT_SUCCESS;
283 
284 out:
285 	return ret;
286 err:
287 	vrf_tx_error(vrf_dev, skb);
288 	goto out;
289 }
290 
291 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
292 {
293 	switch (skb->protocol) {
294 	case htons(ETH_P_IP):
295 		return vrf_process_v4_outbound(skb, dev);
296 	case htons(ETH_P_IPV6):
297 		return vrf_process_v6_outbound(skb, dev);
298 	default:
299 		vrf_tx_error(dev, skb);
300 		return NET_XMIT_DROP;
301 	}
302 }
303 
304 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
305 {
306 	int len = skb->len;
307 	netdev_tx_t ret = is_ip_tx_frame(skb, dev);
308 
309 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
310 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
311 
312 		u64_stats_update_begin(&dstats->syncp);
313 		dstats->tx_pkts++;
314 		dstats->tx_bytes += len;
315 		u64_stats_update_end(&dstats->syncp);
316 	} else {
317 		this_cpu_inc(dev->dstats->tx_drps);
318 	}
319 
320 	return ret;
321 }
322 
323 static int vrf_finish_direct(struct net *net, struct sock *sk,
324 			     struct sk_buff *skb)
325 {
326 	struct net_device *vrf_dev = skb->dev;
327 
328 	if (!list_empty(&vrf_dev->ptype_all) &&
329 	    likely(skb_headroom(skb) >= ETH_HLEN)) {
330 		struct ethhdr *eth = skb_push(skb, ETH_HLEN);
331 
332 		ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
333 		eth_zero_addr(eth->h_dest);
334 		eth->h_proto = skb->protocol;
335 
336 		rcu_read_lock_bh();
337 		dev_queue_xmit_nit(skb, vrf_dev);
338 		rcu_read_unlock_bh();
339 
340 		skb_pull(skb, ETH_HLEN);
341 	}
342 
343 	return 1;
344 }
345 
346 #if IS_ENABLED(CONFIG_IPV6)
347 /* modelled after ip6_finish_output2 */
348 static int vrf_finish_output6(struct net *net, struct sock *sk,
349 			      struct sk_buff *skb)
350 {
351 	struct dst_entry *dst = skb_dst(skb);
352 	struct net_device *dev = dst->dev;
353 	struct neighbour *neigh;
354 	struct in6_addr *nexthop;
355 	int ret;
356 
357 	nf_reset(skb);
358 
359 	skb->protocol = htons(ETH_P_IPV6);
360 	skb->dev = dev;
361 
362 	rcu_read_lock_bh();
363 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
364 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
365 	if (unlikely(!neigh))
366 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
367 	if (!IS_ERR(neigh)) {
368 		sock_confirm_neigh(skb, neigh);
369 		ret = neigh_output(neigh, skb, false);
370 		rcu_read_unlock_bh();
371 		return ret;
372 	}
373 	rcu_read_unlock_bh();
374 
375 	IP6_INC_STATS(dev_net(dst->dev),
376 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
377 	kfree_skb(skb);
378 	return -EINVAL;
379 }
380 
381 /* modelled after ip6_output */
382 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
383 {
384 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
385 			    net, sk, skb, NULL, skb_dst(skb)->dev,
386 			    vrf_finish_output6,
387 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
388 }
389 
390 /* set dst on skb to send packet to us via dev_xmit path. Allows
391  * packet to go through device based features such as qdisc, netfilter
392  * hooks and packet sockets with skb->dev set to vrf device.
393  */
394 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
395 					    struct sk_buff *skb)
396 {
397 	struct net_vrf *vrf = netdev_priv(vrf_dev);
398 	struct dst_entry *dst = NULL;
399 	struct rt6_info *rt6;
400 
401 	rcu_read_lock();
402 
403 	rt6 = rcu_dereference(vrf->rt6);
404 	if (likely(rt6)) {
405 		dst = &rt6->dst;
406 		dst_hold(dst);
407 	}
408 
409 	rcu_read_unlock();
410 
411 	if (unlikely(!dst)) {
412 		vrf_tx_error(vrf_dev, skb);
413 		return NULL;
414 	}
415 
416 	skb_dst_drop(skb);
417 	skb_dst_set(skb, dst);
418 
419 	return skb;
420 }
421 
422 static int vrf_output6_direct(struct net *net, struct sock *sk,
423 			      struct sk_buff *skb)
424 {
425 	skb->protocol = htons(ETH_P_IPV6);
426 
427 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
428 			    net, sk, skb, NULL, skb->dev,
429 			    vrf_finish_direct,
430 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
431 }
432 
433 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
434 					  struct sock *sk,
435 					  struct sk_buff *skb)
436 {
437 	struct net *net = dev_net(vrf_dev);
438 	int err;
439 
440 	skb->dev = vrf_dev;
441 
442 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
443 		      skb, NULL, vrf_dev, vrf_output6_direct);
444 
445 	if (likely(err == 1))
446 		err = vrf_output6_direct(net, sk, skb);
447 
448 	/* reset skb device */
449 	if (likely(err == 1))
450 		nf_reset(skb);
451 	else
452 		skb = NULL;
453 
454 	return skb;
455 }
456 
457 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
458 				   struct sock *sk,
459 				   struct sk_buff *skb)
460 {
461 	/* don't divert link scope packets */
462 	if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
463 		return skb;
464 
465 	if (qdisc_tx_is_default(vrf_dev))
466 		return vrf_ip6_out_direct(vrf_dev, sk, skb);
467 
468 	return vrf_ip6_out_redirect(vrf_dev, skb);
469 }
470 
471 /* holding rtnl */
472 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
473 {
474 	struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
475 	struct net *net = dev_net(dev);
476 	struct dst_entry *dst;
477 
478 	RCU_INIT_POINTER(vrf->rt6, NULL);
479 	synchronize_rcu();
480 
481 	/* move dev in dst's to loopback so this VRF device can be deleted
482 	 * - based on dst_ifdown
483 	 */
484 	if (rt6) {
485 		dst = &rt6->dst;
486 		dev_put(dst->dev);
487 		dst->dev = net->loopback_dev;
488 		dev_hold(dst->dev);
489 		dst_release(dst);
490 	}
491 }
492 
493 static int vrf_rt6_create(struct net_device *dev)
494 {
495 	int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
496 	struct net_vrf *vrf = netdev_priv(dev);
497 	struct net *net = dev_net(dev);
498 	struct rt6_info *rt6;
499 	int rc = -ENOMEM;
500 
501 	/* IPv6 can be CONFIG enabled and then disabled runtime */
502 	if (!ipv6_mod_enabled())
503 		return 0;
504 
505 	vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
506 	if (!vrf->fib6_table)
507 		goto out;
508 
509 	/* create a dst for routing packets out a VRF device */
510 	rt6 = ip6_dst_alloc(net, dev, flags);
511 	if (!rt6)
512 		goto out;
513 
514 	rt6->dst.output	= vrf_output6;
515 
516 	rcu_assign_pointer(vrf->rt6, rt6);
517 
518 	rc = 0;
519 out:
520 	return rc;
521 }
522 #else
523 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
524 				   struct sock *sk,
525 				   struct sk_buff *skb)
526 {
527 	return skb;
528 }
529 
530 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
531 {
532 }
533 
534 static int vrf_rt6_create(struct net_device *dev)
535 {
536 	return 0;
537 }
538 #endif
539 
540 /* modelled after ip_finish_output2 */
541 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
542 {
543 	struct dst_entry *dst = skb_dst(skb);
544 	struct rtable *rt = (struct rtable *)dst;
545 	struct net_device *dev = dst->dev;
546 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
547 	struct neighbour *neigh;
548 	bool is_v6gw = false;
549 	int ret = -EINVAL;
550 
551 	nf_reset(skb);
552 
553 	/* Be paranoid, rather than too clever. */
554 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
555 		struct sk_buff *skb2;
556 
557 		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
558 		if (!skb2) {
559 			ret = -ENOMEM;
560 			goto err;
561 		}
562 		if (skb->sk)
563 			skb_set_owner_w(skb2, skb->sk);
564 
565 		consume_skb(skb);
566 		skb = skb2;
567 	}
568 
569 	rcu_read_lock_bh();
570 
571 	neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
572 	if (!IS_ERR(neigh)) {
573 		sock_confirm_neigh(skb, neigh);
574 		/* if crossing protocols, can not use the cached header */
575 		ret = neigh_output(neigh, skb, is_v6gw);
576 		rcu_read_unlock_bh();
577 		return ret;
578 	}
579 
580 	rcu_read_unlock_bh();
581 err:
582 	vrf_tx_error(skb->dev, skb);
583 	return ret;
584 }
585 
586 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
587 {
588 	struct net_device *dev = skb_dst(skb)->dev;
589 
590 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
591 
592 	skb->dev = dev;
593 	skb->protocol = htons(ETH_P_IP);
594 
595 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
596 			    net, sk, skb, NULL, dev,
597 			    vrf_finish_output,
598 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
599 }
600 
601 /* set dst on skb to send packet to us via dev_xmit path. Allows
602  * packet to go through device based features such as qdisc, netfilter
603  * hooks and packet sockets with skb->dev set to vrf device.
604  */
605 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
606 					   struct sk_buff *skb)
607 {
608 	struct net_vrf *vrf = netdev_priv(vrf_dev);
609 	struct dst_entry *dst = NULL;
610 	struct rtable *rth;
611 
612 	rcu_read_lock();
613 
614 	rth = rcu_dereference(vrf->rth);
615 	if (likely(rth)) {
616 		dst = &rth->dst;
617 		dst_hold(dst);
618 	}
619 
620 	rcu_read_unlock();
621 
622 	if (unlikely(!dst)) {
623 		vrf_tx_error(vrf_dev, skb);
624 		return NULL;
625 	}
626 
627 	skb_dst_drop(skb);
628 	skb_dst_set(skb, dst);
629 
630 	return skb;
631 }
632 
633 static int vrf_output_direct(struct net *net, struct sock *sk,
634 			     struct sk_buff *skb)
635 {
636 	skb->protocol = htons(ETH_P_IP);
637 
638 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
639 			    net, sk, skb, NULL, skb->dev,
640 			    vrf_finish_direct,
641 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
642 }
643 
644 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
645 					 struct sock *sk,
646 					 struct sk_buff *skb)
647 {
648 	struct net *net = dev_net(vrf_dev);
649 	int err;
650 
651 	skb->dev = vrf_dev;
652 
653 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
654 		      skb, NULL, vrf_dev, vrf_output_direct);
655 
656 	if (likely(err == 1))
657 		err = vrf_output_direct(net, sk, skb);
658 
659 	/* reset skb device */
660 	if (likely(err == 1))
661 		nf_reset(skb);
662 	else
663 		skb = NULL;
664 
665 	return skb;
666 }
667 
668 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
669 				  struct sock *sk,
670 				  struct sk_buff *skb)
671 {
672 	/* don't divert multicast or local broadcast */
673 	if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
674 	    ipv4_is_lbcast(ip_hdr(skb)->daddr))
675 		return skb;
676 
677 	if (qdisc_tx_is_default(vrf_dev))
678 		return vrf_ip_out_direct(vrf_dev, sk, skb);
679 
680 	return vrf_ip_out_redirect(vrf_dev, skb);
681 }
682 
683 /* called with rcu lock held */
684 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
685 				  struct sock *sk,
686 				  struct sk_buff *skb,
687 				  u16 proto)
688 {
689 	switch (proto) {
690 	case AF_INET:
691 		return vrf_ip_out(vrf_dev, sk, skb);
692 	case AF_INET6:
693 		return vrf_ip6_out(vrf_dev, sk, skb);
694 	}
695 
696 	return skb;
697 }
698 
699 /* holding rtnl */
700 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
701 {
702 	struct rtable *rth = rtnl_dereference(vrf->rth);
703 	struct net *net = dev_net(dev);
704 	struct dst_entry *dst;
705 
706 	RCU_INIT_POINTER(vrf->rth, NULL);
707 	synchronize_rcu();
708 
709 	/* move dev in dst's to loopback so this VRF device can be deleted
710 	 * - based on dst_ifdown
711 	 */
712 	if (rth) {
713 		dst = &rth->dst;
714 		dev_put(dst->dev);
715 		dst->dev = net->loopback_dev;
716 		dev_hold(dst->dev);
717 		dst_release(dst);
718 	}
719 }
720 
721 static int vrf_rtable_create(struct net_device *dev)
722 {
723 	struct net_vrf *vrf = netdev_priv(dev);
724 	struct rtable *rth;
725 
726 	if (!fib_new_table(dev_net(dev), vrf->tb_id))
727 		return -ENOMEM;
728 
729 	/* create a dst for routing packets out through a VRF device */
730 	rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
731 	if (!rth)
732 		return -ENOMEM;
733 
734 	rth->dst.output	= vrf_output;
735 
736 	rcu_assign_pointer(vrf->rth, rth);
737 
738 	return 0;
739 }
740 
741 /**************************** device handling ********************/
742 
743 /* cycle interface to flush neighbor cache and move routes across tables */
744 static void cycle_netdev(struct net_device *dev,
745 			 struct netlink_ext_ack *extack)
746 {
747 	unsigned int flags = dev->flags;
748 	int ret;
749 
750 	if (!netif_running(dev))
751 		return;
752 
753 	ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
754 	if (ret >= 0)
755 		ret = dev_change_flags(dev, flags, extack);
756 
757 	if (ret < 0) {
758 		netdev_err(dev,
759 			   "Failed to cycle device %s; route tables might be wrong!\n",
760 			   dev->name);
761 	}
762 }
763 
764 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
765 			    struct netlink_ext_ack *extack)
766 {
767 	int ret;
768 
769 	/* do not allow loopback device to be enslaved to a VRF.
770 	 * The vrf device acts as the loopback for the vrf.
771 	 */
772 	if (port_dev == dev_net(dev)->loopback_dev) {
773 		NL_SET_ERR_MSG(extack,
774 			       "Can not enslave loopback device to a VRF");
775 		return -EOPNOTSUPP;
776 	}
777 
778 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
779 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
780 	if (ret < 0)
781 		goto err;
782 
783 	cycle_netdev(port_dev, extack);
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 			 struct netlink_ext_ack *extack)
794 {
795 	if (netif_is_l3_master(port_dev)) {
796 		NL_SET_ERR_MSG(extack,
797 			       "Can not enslave an L3 master device to a VRF");
798 		return -EINVAL;
799 	}
800 
801 	if (netif_is_l3_slave(port_dev))
802 		return -EINVAL;
803 
804 	return do_vrf_add_slave(dev, port_dev, extack);
805 }
806 
807 /* inverse of do_vrf_add_slave */
808 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
809 {
810 	netdev_upper_dev_unlink(port_dev, dev);
811 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
812 
813 	cycle_netdev(port_dev, NULL);
814 
815 	return 0;
816 }
817 
818 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
819 {
820 	return do_vrf_del_slave(dev, port_dev);
821 }
822 
823 static void vrf_dev_uninit(struct net_device *dev)
824 {
825 	struct net_vrf *vrf = netdev_priv(dev);
826 
827 	vrf_rtable_release(dev, vrf);
828 	vrf_rt6_release(dev, vrf);
829 
830 	free_percpu(dev->dstats);
831 	dev->dstats = NULL;
832 }
833 
834 static int vrf_dev_init(struct net_device *dev)
835 {
836 	struct net_vrf *vrf = netdev_priv(dev);
837 
838 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
839 	if (!dev->dstats)
840 		goto out_nomem;
841 
842 	/* create the default dst which points back to us */
843 	if (vrf_rtable_create(dev) != 0)
844 		goto out_stats;
845 
846 	if (vrf_rt6_create(dev) != 0)
847 		goto out_rth;
848 
849 	dev->flags = IFF_MASTER | IFF_NOARP;
850 
851 	/* MTU is irrelevant for VRF device; set to 64k similar to lo */
852 	dev->mtu = 64 * 1024;
853 
854 	/* similarly, oper state is irrelevant; set to up to avoid confusion */
855 	dev->operstate = IF_OPER_UP;
856 	netdev_lockdep_set_classes(dev);
857 	return 0;
858 
859 out_rth:
860 	vrf_rtable_release(dev, vrf);
861 out_stats:
862 	free_percpu(dev->dstats);
863 	dev->dstats = NULL;
864 out_nomem:
865 	return -ENOMEM;
866 }
867 
868 static const struct net_device_ops vrf_netdev_ops = {
869 	.ndo_init		= vrf_dev_init,
870 	.ndo_uninit		= vrf_dev_uninit,
871 	.ndo_start_xmit		= vrf_xmit,
872 	.ndo_set_mac_address	= eth_mac_addr,
873 	.ndo_get_stats64	= vrf_get_stats64,
874 	.ndo_add_slave		= vrf_add_slave,
875 	.ndo_del_slave		= vrf_del_slave,
876 };
877 
878 static u32 vrf_fib_table(const struct net_device *dev)
879 {
880 	struct net_vrf *vrf = netdev_priv(dev);
881 
882 	return vrf->tb_id;
883 }
884 
885 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
886 {
887 	kfree_skb(skb);
888 	return 0;
889 }
890 
891 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
892 				      struct sk_buff *skb,
893 				      struct net_device *dev)
894 {
895 	struct net *net = dev_net(dev);
896 
897 	if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
898 		skb = NULL;    /* kfree_skb(skb) handled by nf code */
899 
900 	return skb;
901 }
902 
903 #if IS_ENABLED(CONFIG_IPV6)
904 /* neighbor handling is done with actual device; do not want
905  * to flip skb->dev for those ndisc packets. This really fails
906  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
907  * a start.
908  */
909 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
910 {
911 	const struct ipv6hdr *iph = ipv6_hdr(skb);
912 	bool rc = false;
913 
914 	if (iph->nexthdr == NEXTHDR_ICMP) {
915 		const struct icmp6hdr *icmph;
916 		struct icmp6hdr _icmph;
917 
918 		icmph = skb_header_pointer(skb, sizeof(*iph),
919 					   sizeof(_icmph), &_icmph);
920 		if (!icmph)
921 			goto out;
922 
923 		switch (icmph->icmp6_type) {
924 		case NDISC_ROUTER_SOLICITATION:
925 		case NDISC_ROUTER_ADVERTISEMENT:
926 		case NDISC_NEIGHBOUR_SOLICITATION:
927 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
928 		case NDISC_REDIRECT:
929 			rc = true;
930 			break;
931 		}
932 	}
933 
934 out:
935 	return rc;
936 }
937 
938 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
939 					     const struct net_device *dev,
940 					     struct flowi6 *fl6,
941 					     int ifindex,
942 					     const struct sk_buff *skb,
943 					     int flags)
944 {
945 	struct net_vrf *vrf = netdev_priv(dev);
946 
947 	return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
948 }
949 
950 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
951 			      int ifindex)
952 {
953 	const struct ipv6hdr *iph = ipv6_hdr(skb);
954 	struct flowi6 fl6 = {
955 		.flowi6_iif     = ifindex,
956 		.flowi6_mark    = skb->mark,
957 		.flowi6_proto   = iph->nexthdr,
958 		.daddr          = iph->daddr,
959 		.saddr          = iph->saddr,
960 		.flowlabel      = ip6_flowinfo(iph),
961 	};
962 	struct net *net = dev_net(vrf_dev);
963 	struct rt6_info *rt6;
964 
965 	rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
966 				   RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
967 	if (unlikely(!rt6))
968 		return;
969 
970 	if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
971 		return;
972 
973 	skb_dst_set(skb, &rt6->dst);
974 }
975 
976 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
977 				   struct sk_buff *skb)
978 {
979 	int orig_iif = skb->skb_iif;
980 	bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
981 	bool is_ndisc = ipv6_ndisc_frame(skb);
982 
983 	/* loopback, multicast & non-ND link-local traffic; do not push through
984 	 * packet taps again. Reset pkt_type for upper layers to process skb
985 	 */
986 	if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
987 		skb->dev = vrf_dev;
988 		skb->skb_iif = vrf_dev->ifindex;
989 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
990 		if (skb->pkt_type == PACKET_LOOPBACK)
991 			skb->pkt_type = PACKET_HOST;
992 		goto out;
993 	}
994 
995 	/* if packet is NDISC then keep the ingress interface */
996 	if (!is_ndisc) {
997 		vrf_rx_stats(vrf_dev, skb->len);
998 		skb->dev = vrf_dev;
999 		skb->skb_iif = vrf_dev->ifindex;
1000 
1001 		if (!list_empty(&vrf_dev->ptype_all)) {
1002 			skb_push(skb, skb->mac_len);
1003 			dev_queue_xmit_nit(skb, vrf_dev);
1004 			skb_pull(skb, skb->mac_len);
1005 		}
1006 
1007 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1008 	}
1009 
1010 	if (need_strict)
1011 		vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1012 
1013 	skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1014 out:
1015 	return skb;
1016 }
1017 
1018 #else
1019 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1020 				   struct sk_buff *skb)
1021 {
1022 	return skb;
1023 }
1024 #endif
1025 
1026 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1027 				  struct sk_buff *skb)
1028 {
1029 	skb->dev = vrf_dev;
1030 	skb->skb_iif = vrf_dev->ifindex;
1031 	IPCB(skb)->flags |= IPSKB_L3SLAVE;
1032 
1033 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1034 		goto out;
1035 
1036 	/* loopback traffic; do not push through packet taps again.
1037 	 * Reset pkt_type for upper layers to process skb
1038 	 */
1039 	if (skb->pkt_type == PACKET_LOOPBACK) {
1040 		skb->pkt_type = PACKET_HOST;
1041 		goto out;
1042 	}
1043 
1044 	vrf_rx_stats(vrf_dev, skb->len);
1045 
1046 	if (!list_empty(&vrf_dev->ptype_all)) {
1047 		skb_push(skb, skb->mac_len);
1048 		dev_queue_xmit_nit(skb, vrf_dev);
1049 		skb_pull(skb, skb->mac_len);
1050 	}
1051 
1052 	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1053 out:
1054 	return skb;
1055 }
1056 
1057 /* called with rcu lock held */
1058 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1059 				  struct sk_buff *skb,
1060 				  u16 proto)
1061 {
1062 	switch (proto) {
1063 	case AF_INET:
1064 		return vrf_ip_rcv(vrf_dev, skb);
1065 	case AF_INET6:
1066 		return vrf_ip6_rcv(vrf_dev, skb);
1067 	}
1068 
1069 	return skb;
1070 }
1071 
1072 #if IS_ENABLED(CONFIG_IPV6)
1073 /* send to link-local or multicast address via interface enslaved to
1074  * VRF device. Force lookup to VRF table without changing flow struct
1075  */
1076 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1077 					      struct flowi6 *fl6)
1078 {
1079 	struct net *net = dev_net(dev);
1080 	int flags = RT6_LOOKUP_F_IFACE;
1081 	struct dst_entry *dst = NULL;
1082 	struct rt6_info *rt;
1083 
1084 	/* VRF device does not have a link-local address and
1085 	 * sending packets to link-local or mcast addresses over
1086 	 * a VRF device does not make sense
1087 	 */
1088 	if (fl6->flowi6_oif == dev->ifindex) {
1089 		dst = &net->ipv6.ip6_null_entry->dst;
1090 		dst_hold(dst);
1091 		return dst;
1092 	}
1093 
1094 	if (!ipv6_addr_any(&fl6->saddr))
1095 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1096 
1097 	rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1098 	if (rt)
1099 		dst = &rt->dst;
1100 
1101 	return dst;
1102 }
1103 #endif
1104 
1105 static const struct l3mdev_ops vrf_l3mdev_ops = {
1106 	.l3mdev_fib_table	= vrf_fib_table,
1107 	.l3mdev_l3_rcv		= vrf_l3_rcv,
1108 	.l3mdev_l3_out		= vrf_l3_out,
1109 #if IS_ENABLED(CONFIG_IPV6)
1110 	.l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1111 #endif
1112 };
1113 
1114 static void vrf_get_drvinfo(struct net_device *dev,
1115 			    struct ethtool_drvinfo *info)
1116 {
1117 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1118 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1119 }
1120 
1121 static const struct ethtool_ops vrf_ethtool_ops = {
1122 	.get_drvinfo	= vrf_get_drvinfo,
1123 };
1124 
1125 static inline size_t vrf_fib_rule_nl_size(void)
1126 {
1127 	size_t sz;
1128 
1129 	sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1130 	sz += nla_total_size(sizeof(u8));	/* FRA_L3MDEV */
1131 	sz += nla_total_size(sizeof(u32));	/* FRA_PRIORITY */
1132 	sz += nla_total_size(sizeof(u8));       /* FRA_PROTOCOL */
1133 
1134 	return sz;
1135 }
1136 
1137 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1138 {
1139 	struct fib_rule_hdr *frh;
1140 	struct nlmsghdr *nlh;
1141 	struct sk_buff *skb;
1142 	int err;
1143 
1144 	if (family == AF_INET6 && !ipv6_mod_enabled())
1145 		return 0;
1146 
1147 	skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1148 	if (!skb)
1149 		return -ENOMEM;
1150 
1151 	nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1152 	if (!nlh)
1153 		goto nla_put_failure;
1154 
1155 	/* rule only needs to appear once */
1156 	nlh->nlmsg_flags |= NLM_F_EXCL;
1157 
1158 	frh = nlmsg_data(nlh);
1159 	memset(frh, 0, sizeof(*frh));
1160 	frh->family = family;
1161 	frh->action = FR_ACT_TO_TBL;
1162 
1163 	if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1164 		goto nla_put_failure;
1165 
1166 	if (nla_put_u8(skb, FRA_L3MDEV, 1))
1167 		goto nla_put_failure;
1168 
1169 	if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1170 		goto nla_put_failure;
1171 
1172 	nlmsg_end(skb, nlh);
1173 
1174 	/* fib_nl_{new,del}rule handling looks for net from skb->sk */
1175 	skb->sk = dev_net(dev)->rtnl;
1176 	if (add_it) {
1177 		err = fib_nl_newrule(skb, nlh, NULL);
1178 		if (err == -EEXIST)
1179 			err = 0;
1180 	} else {
1181 		err = fib_nl_delrule(skb, nlh, NULL);
1182 		if (err == -ENOENT)
1183 			err = 0;
1184 	}
1185 	nlmsg_free(skb);
1186 
1187 	return err;
1188 
1189 nla_put_failure:
1190 	nlmsg_free(skb);
1191 
1192 	return -EMSGSIZE;
1193 }
1194 
1195 static int vrf_add_fib_rules(const struct net_device *dev)
1196 {
1197 	int err;
1198 
1199 	err = vrf_fib_rule(dev, AF_INET,  true);
1200 	if (err < 0)
1201 		goto out_err;
1202 
1203 	err = vrf_fib_rule(dev, AF_INET6, true);
1204 	if (err < 0)
1205 		goto ipv6_err;
1206 
1207 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1208 	err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1209 	if (err < 0)
1210 		goto ipmr_err;
1211 #endif
1212 
1213 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1214 	err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1215 	if (err < 0)
1216 		goto ip6mr_err;
1217 #endif
1218 
1219 	return 0;
1220 
1221 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1222 ip6mr_err:
1223 	vrf_fib_rule(dev, RTNL_FAMILY_IPMR,  false);
1224 #endif
1225 
1226 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1227 ipmr_err:
1228 	vrf_fib_rule(dev, AF_INET6,  false);
1229 #endif
1230 
1231 ipv6_err:
1232 	vrf_fib_rule(dev, AF_INET,  false);
1233 
1234 out_err:
1235 	netdev_err(dev, "Failed to add FIB rules.\n");
1236 	return err;
1237 }
1238 
1239 static void vrf_setup(struct net_device *dev)
1240 {
1241 	ether_setup(dev);
1242 
1243 	/* Initialize the device structure. */
1244 	dev->netdev_ops = &vrf_netdev_ops;
1245 	dev->l3mdev_ops = &vrf_l3mdev_ops;
1246 	dev->ethtool_ops = &vrf_ethtool_ops;
1247 	dev->needs_free_netdev = true;
1248 
1249 	/* Fill in device structure with ethernet-generic values. */
1250 	eth_hw_addr_random(dev);
1251 
1252 	/* don't acquire vrf device's netif_tx_lock when transmitting */
1253 	dev->features |= NETIF_F_LLTX;
1254 
1255 	/* don't allow vrf devices to change network namespaces. */
1256 	dev->features |= NETIF_F_NETNS_LOCAL;
1257 
1258 	/* does not make sense for a VLAN to be added to a vrf device */
1259 	dev->features   |= NETIF_F_VLAN_CHALLENGED;
1260 
1261 	/* enable offload features */
1262 	dev->features   |= NETIF_F_GSO_SOFTWARE;
1263 	dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1264 	dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1265 
1266 	dev->hw_features = dev->features;
1267 	dev->hw_enc_features = dev->features;
1268 
1269 	/* default to no qdisc; user can add if desired */
1270 	dev->priv_flags |= IFF_NO_QUEUE;
1271 	dev->priv_flags |= IFF_NO_RX_HANDLER;
1272 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1273 
1274 	/* VRF devices do not care about MTU, but if the MTU is set
1275 	 * too low then the ipv4 and ipv6 protocols are disabled
1276 	 * which breaks networking.
1277 	 */
1278 	dev->min_mtu = IPV6_MIN_MTU;
1279 	dev->max_mtu = ETH_MAX_MTU;
1280 }
1281 
1282 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1283 			struct netlink_ext_ack *extack)
1284 {
1285 	if (tb[IFLA_ADDRESS]) {
1286 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1287 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1288 			return -EINVAL;
1289 		}
1290 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1291 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1292 			return -EADDRNOTAVAIL;
1293 		}
1294 	}
1295 	return 0;
1296 }
1297 
1298 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1299 {
1300 	struct net_device *port_dev;
1301 	struct list_head *iter;
1302 
1303 	netdev_for_each_lower_dev(dev, port_dev, iter)
1304 		vrf_del_slave(dev, port_dev);
1305 
1306 	unregister_netdevice_queue(dev, head);
1307 }
1308 
1309 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1310 		       struct nlattr *tb[], struct nlattr *data[],
1311 		       struct netlink_ext_ack *extack)
1312 {
1313 	struct net_vrf *vrf = netdev_priv(dev);
1314 	bool *add_fib_rules;
1315 	struct net *net;
1316 	int err;
1317 
1318 	if (!data || !data[IFLA_VRF_TABLE]) {
1319 		NL_SET_ERR_MSG(extack, "VRF table id is missing");
1320 		return -EINVAL;
1321 	}
1322 
1323 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1324 	if (vrf->tb_id == RT_TABLE_UNSPEC) {
1325 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1326 				    "Invalid VRF table id");
1327 		return -EINVAL;
1328 	}
1329 
1330 	dev->priv_flags |= IFF_L3MDEV_MASTER;
1331 
1332 	err = register_netdevice(dev);
1333 	if (err)
1334 		goto out;
1335 
1336 	net = dev_net(dev);
1337 	add_fib_rules = net_generic(net, vrf_net_id);
1338 	if (*add_fib_rules) {
1339 		err = vrf_add_fib_rules(dev);
1340 		if (err) {
1341 			unregister_netdevice(dev);
1342 			goto out;
1343 		}
1344 		*add_fib_rules = false;
1345 	}
1346 
1347 out:
1348 	return err;
1349 }
1350 
1351 static size_t vrf_nl_getsize(const struct net_device *dev)
1352 {
1353 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
1354 }
1355 
1356 static int vrf_fillinfo(struct sk_buff *skb,
1357 			const struct net_device *dev)
1358 {
1359 	struct net_vrf *vrf = netdev_priv(dev);
1360 
1361 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1362 }
1363 
1364 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1365 				 const struct net_device *slave_dev)
1366 {
1367 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
1368 }
1369 
1370 static int vrf_fill_slave_info(struct sk_buff *skb,
1371 			       const struct net_device *vrf_dev,
1372 			       const struct net_device *slave_dev)
1373 {
1374 	struct net_vrf *vrf = netdev_priv(vrf_dev);
1375 
1376 	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1377 		return -EMSGSIZE;
1378 
1379 	return 0;
1380 }
1381 
1382 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1383 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
1384 };
1385 
1386 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1387 	.kind		= DRV_NAME,
1388 	.priv_size	= sizeof(struct net_vrf),
1389 
1390 	.get_size	= vrf_nl_getsize,
1391 	.policy		= vrf_nl_policy,
1392 	.validate	= vrf_validate,
1393 	.fill_info	= vrf_fillinfo,
1394 
1395 	.get_slave_size  = vrf_get_slave_size,
1396 	.fill_slave_info = vrf_fill_slave_info,
1397 
1398 	.newlink	= vrf_newlink,
1399 	.dellink	= vrf_dellink,
1400 	.setup		= vrf_setup,
1401 	.maxtype	= IFLA_VRF_MAX,
1402 };
1403 
1404 static int vrf_device_event(struct notifier_block *unused,
1405 			    unsigned long event, void *ptr)
1406 {
1407 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1408 
1409 	/* only care about unregister events to drop slave references */
1410 	if (event == NETDEV_UNREGISTER) {
1411 		struct net_device *vrf_dev;
1412 
1413 		if (!netif_is_l3_slave(dev))
1414 			goto out;
1415 
1416 		vrf_dev = netdev_master_upper_dev_get(dev);
1417 		vrf_del_slave(vrf_dev, dev);
1418 	}
1419 out:
1420 	return NOTIFY_DONE;
1421 }
1422 
1423 static struct notifier_block vrf_notifier_block __read_mostly = {
1424 	.notifier_call = vrf_device_event,
1425 };
1426 
1427 /* Initialize per network namespace state */
1428 static int __net_init vrf_netns_init(struct net *net)
1429 {
1430 	bool *add_fib_rules = net_generic(net, vrf_net_id);
1431 
1432 	*add_fib_rules = true;
1433 
1434 	return 0;
1435 }
1436 
1437 static struct pernet_operations vrf_net_ops __net_initdata = {
1438 	.init = vrf_netns_init,
1439 	.id   = &vrf_net_id,
1440 	.size = sizeof(bool),
1441 };
1442 
1443 static int __init vrf_init_module(void)
1444 {
1445 	int rc;
1446 
1447 	register_netdevice_notifier(&vrf_notifier_block);
1448 
1449 	rc = register_pernet_subsys(&vrf_net_ops);
1450 	if (rc < 0)
1451 		goto error;
1452 
1453 	rc = rtnl_link_register(&vrf_link_ops);
1454 	if (rc < 0) {
1455 		unregister_pernet_subsys(&vrf_net_ops);
1456 		goto error;
1457 	}
1458 
1459 	return 0;
1460 
1461 error:
1462 	unregister_netdevice_notifier(&vrf_notifier_block);
1463 	return rc;
1464 }
1465 
1466 module_init(vrf_init_module);
1467 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1468 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1469 MODULE_LICENSE("GPL");
1470 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1471 MODULE_VERSION(DRV_VERSION);
1472