xref: /linux/drivers/net/vrf.c (revision 7aacf86b75bc5523d20fd9127104384fce51ce9c)
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 rtable __rcu	*rth_local;
51 	struct rt6_info	__rcu	*rt6;
52 	struct rt6_info	__rcu	*rt6_local;
53 	u32                     tb_id;
54 };
55 
56 struct pcpu_dstats {
57 	u64			tx_pkts;
58 	u64			tx_bytes;
59 	u64			tx_drps;
60 	u64			rx_pkts;
61 	u64			rx_bytes;
62 	u64			rx_drps;
63 	struct u64_stats_sync	syncp;
64 };
65 
66 static void vrf_rx_stats(struct net_device *dev, int len)
67 {
68 	struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
69 
70 	u64_stats_update_begin(&dstats->syncp);
71 	dstats->rx_pkts++;
72 	dstats->rx_bytes += len;
73 	u64_stats_update_end(&dstats->syncp);
74 }
75 
76 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
77 {
78 	vrf_dev->stats.tx_errors++;
79 	kfree_skb(skb);
80 }
81 
82 static void vrf_get_stats64(struct net_device *dev,
83 			    struct rtnl_link_stats64 *stats)
84 {
85 	int i;
86 
87 	for_each_possible_cpu(i) {
88 		const struct pcpu_dstats *dstats;
89 		u64 tbytes, tpkts, tdrops, rbytes, rpkts;
90 		unsigned int start;
91 
92 		dstats = per_cpu_ptr(dev->dstats, i);
93 		do {
94 			start = u64_stats_fetch_begin_irq(&dstats->syncp);
95 			tbytes = dstats->tx_bytes;
96 			tpkts = dstats->tx_pkts;
97 			tdrops = dstats->tx_drps;
98 			rbytes = dstats->rx_bytes;
99 			rpkts = dstats->rx_pkts;
100 		} while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
101 		stats->tx_bytes += tbytes;
102 		stats->tx_packets += tpkts;
103 		stats->tx_dropped += tdrops;
104 		stats->rx_bytes += rbytes;
105 		stats->rx_packets += rpkts;
106 	}
107 }
108 
109 /* by default VRF devices do not have a qdisc and are expected
110  * to be created with only a single queue.
111  */
112 static bool qdisc_tx_is_default(const struct net_device *dev)
113 {
114 	struct netdev_queue *txq;
115 	struct Qdisc *qdisc;
116 
117 	if (dev->num_tx_queues > 1)
118 		return false;
119 
120 	txq = netdev_get_tx_queue(dev, 0);
121 	qdisc = rcu_access_pointer(txq->qdisc);
122 
123 	return !qdisc->enqueue;
124 }
125 
126 /* Local traffic destined to local address. Reinsert the packet to rx
127  * path, similar to loopback handling.
128  */
129 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
130 			  struct dst_entry *dst)
131 {
132 	int len = skb->len;
133 
134 	skb_orphan(skb);
135 
136 	skb_dst_set(skb, dst);
137 	skb_dst_force(skb);
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 using our local dst
198 	 */
199 	if (dst->dev == net->loopback_dev || dst->dev == dev) {
200 		struct net_vrf *vrf = netdev_priv(dev);
201 		struct rt6_info *rt6_local;
202 
203 		/* release looked up dst and use cached local dst */
204 		dst_release(dst);
205 
206 		rcu_read_lock();
207 
208 		rt6_local = rcu_dereference(vrf->rt6_local);
209 		if (unlikely(!rt6_local)) {
210 			rcu_read_unlock();
211 			goto err;
212 		}
213 
214 		/* Ordering issue: cached local dst is created on newlink
215 		 * before the IPv6 initialization. Using the local dst
216 		 * requires rt6i_idev to be set so make sure it is.
217 		 */
218 		if (unlikely(!rt6_local->rt6i_idev)) {
219 			rt6_local->rt6i_idev = in6_dev_get(dev);
220 			if (!rt6_local->rt6i_idev) {
221 				rcu_read_unlock();
222 				goto err;
223 			}
224 		}
225 
226 		dst = &rt6_local->dst;
227 		dst_hold(dst);
228 
229 		rcu_read_unlock();
230 
231 		return vrf_local_xmit(skb, dev, &rt6_local->dst);
232 	}
233 
234 	skb_dst_set(skb, dst);
235 
236 	/* strip the ethernet header added for pass through VRF device */
237 	__skb_pull(skb, skb_network_offset(skb));
238 
239 	ret = vrf_ip6_local_out(net, skb->sk, skb);
240 	if (unlikely(net_xmit_eval(ret)))
241 		dev->stats.tx_errors++;
242 	else
243 		ret = NET_XMIT_SUCCESS;
244 
245 	return ret;
246 err:
247 	vrf_tx_error(dev, skb);
248 	return NET_XMIT_DROP;
249 }
250 #else
251 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
252 					   struct net_device *dev)
253 {
254 	vrf_tx_error(dev, skb);
255 	return NET_XMIT_DROP;
256 }
257 #endif
258 
259 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
260 static int vrf_ip_local_out(struct net *net, struct sock *sk,
261 			    struct sk_buff *skb)
262 {
263 	int err;
264 
265 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
266 		      skb, NULL, skb_dst(skb)->dev, dst_output);
267 	if (likely(err == 1))
268 		err = dst_output(net, sk, skb);
269 
270 	return err;
271 }
272 
273 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
274 					   struct net_device *vrf_dev)
275 {
276 	struct iphdr *ip4h = ip_hdr(skb);
277 	int ret = NET_XMIT_DROP;
278 	struct flowi4 fl4 = {
279 		/* needed to match OIF rule */
280 		.flowi4_oif = vrf_dev->ifindex,
281 		.flowi4_iif = LOOPBACK_IFINDEX,
282 		.flowi4_tos = RT_TOS(ip4h->tos),
283 		.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
284 		.flowi4_proto = ip4h->protocol,
285 		.daddr = ip4h->daddr,
286 		.saddr = ip4h->saddr,
287 	};
288 	struct net *net = dev_net(vrf_dev);
289 	struct rtable *rt;
290 
291 	rt = ip_route_output_flow(net, &fl4, NULL);
292 	if (IS_ERR(rt))
293 		goto err;
294 
295 	skb_dst_drop(skb);
296 
297 	/* if dst.dev is loopback or the VRF device again this is locally
298 	 * originated traffic destined to a local address. Short circuit
299 	 * to Rx path using our local dst
300 	 */
301 	if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) {
302 		struct net_vrf *vrf = netdev_priv(vrf_dev);
303 		struct rtable *rth_local;
304 		struct dst_entry *dst = NULL;
305 
306 		ip_rt_put(rt);
307 
308 		rcu_read_lock();
309 
310 		rth_local = rcu_dereference(vrf->rth_local);
311 		if (likely(rth_local)) {
312 			dst = &rth_local->dst;
313 			dst_hold(dst);
314 		}
315 
316 		rcu_read_unlock();
317 
318 		if (unlikely(!dst))
319 			goto err;
320 
321 		return vrf_local_xmit(skb, vrf_dev, dst);
322 	}
323 
324 	skb_dst_set(skb, &rt->dst);
325 
326 	/* strip the ethernet header added for pass through VRF device */
327 	__skb_pull(skb, skb_network_offset(skb));
328 
329 	if (!ip4h->saddr) {
330 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
331 					       RT_SCOPE_LINK);
332 	}
333 
334 	ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
335 	if (unlikely(net_xmit_eval(ret)))
336 		vrf_dev->stats.tx_errors++;
337 	else
338 		ret = NET_XMIT_SUCCESS;
339 
340 out:
341 	return ret;
342 err:
343 	vrf_tx_error(vrf_dev, skb);
344 	goto out;
345 }
346 
347 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
348 {
349 	switch (skb->protocol) {
350 	case htons(ETH_P_IP):
351 		return vrf_process_v4_outbound(skb, dev);
352 	case htons(ETH_P_IPV6):
353 		return vrf_process_v6_outbound(skb, dev);
354 	default:
355 		vrf_tx_error(dev, skb);
356 		return NET_XMIT_DROP;
357 	}
358 }
359 
360 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
361 {
362 	int len = skb->len;
363 	netdev_tx_t ret = is_ip_tx_frame(skb, dev);
364 
365 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
366 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
367 
368 		u64_stats_update_begin(&dstats->syncp);
369 		dstats->tx_pkts++;
370 		dstats->tx_bytes += len;
371 		u64_stats_update_end(&dstats->syncp);
372 	} else {
373 		this_cpu_inc(dev->dstats->tx_drps);
374 	}
375 
376 	return ret;
377 }
378 
379 static int vrf_finish_direct(struct net *net, struct sock *sk,
380 			     struct sk_buff *skb)
381 {
382 	struct net_device *vrf_dev = skb->dev;
383 
384 	if (!list_empty(&vrf_dev->ptype_all) &&
385 	    likely(skb_headroom(skb) >= ETH_HLEN)) {
386 		struct ethhdr *eth = skb_push(skb, ETH_HLEN);
387 
388 		ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
389 		eth_zero_addr(eth->h_dest);
390 		eth->h_proto = skb->protocol;
391 
392 		rcu_read_lock_bh();
393 		dev_queue_xmit_nit(skb, vrf_dev);
394 		rcu_read_unlock_bh();
395 
396 		skb_pull(skb, ETH_HLEN);
397 	}
398 
399 	return 1;
400 }
401 
402 #if IS_ENABLED(CONFIG_IPV6)
403 /* modelled after ip6_finish_output2 */
404 static int vrf_finish_output6(struct net *net, struct sock *sk,
405 			      struct sk_buff *skb)
406 {
407 	struct dst_entry *dst = skb_dst(skb);
408 	struct net_device *dev = dst->dev;
409 	struct neighbour *neigh;
410 	struct in6_addr *nexthop;
411 	int ret;
412 
413 	nf_reset(skb);
414 
415 	skb->protocol = htons(ETH_P_IPV6);
416 	skb->dev = dev;
417 
418 	rcu_read_lock_bh();
419 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
420 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
421 	if (unlikely(!neigh))
422 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
423 	if (!IS_ERR(neigh)) {
424 		sock_confirm_neigh(skb, neigh);
425 		ret = neigh_output(neigh, skb);
426 		rcu_read_unlock_bh();
427 		return ret;
428 	}
429 	rcu_read_unlock_bh();
430 
431 	IP6_INC_STATS(dev_net(dst->dev),
432 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
433 	kfree_skb(skb);
434 	return -EINVAL;
435 }
436 
437 /* modelled after ip6_output */
438 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
439 {
440 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
441 			    net, sk, skb, NULL, skb_dst(skb)->dev,
442 			    vrf_finish_output6,
443 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
444 }
445 
446 /* set dst on skb to send packet to us via dev_xmit path. Allows
447  * packet to go through device based features such as qdisc, netfilter
448  * hooks and packet sockets with skb->dev set to vrf device.
449  */
450 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
451 					    struct sk_buff *skb)
452 {
453 	struct net_vrf *vrf = netdev_priv(vrf_dev);
454 	struct dst_entry *dst = NULL;
455 	struct rt6_info *rt6;
456 
457 	rcu_read_lock();
458 
459 	rt6 = rcu_dereference(vrf->rt6);
460 	if (likely(rt6)) {
461 		dst = &rt6->dst;
462 		dst_hold(dst);
463 	}
464 
465 	rcu_read_unlock();
466 
467 	if (unlikely(!dst)) {
468 		vrf_tx_error(vrf_dev, skb);
469 		return NULL;
470 	}
471 
472 	skb_dst_drop(skb);
473 	skb_dst_set(skb, dst);
474 
475 	return skb;
476 }
477 
478 static int vrf_output6_direct(struct net *net, struct sock *sk,
479 			      struct sk_buff *skb)
480 {
481 	skb->protocol = htons(ETH_P_IPV6);
482 
483 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
484 			    net, sk, skb, NULL, skb->dev,
485 			    vrf_finish_direct,
486 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
487 }
488 
489 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
490 					  struct sock *sk,
491 					  struct sk_buff *skb)
492 {
493 	struct net *net = dev_net(vrf_dev);
494 	int err;
495 
496 	skb->dev = vrf_dev;
497 
498 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
499 		      skb, NULL, vrf_dev, vrf_output6_direct);
500 
501 	if (likely(err == 1))
502 		err = vrf_output6_direct(net, sk, skb);
503 
504 	/* reset skb device */
505 	if (likely(err == 1))
506 		nf_reset(skb);
507 	else
508 		skb = NULL;
509 
510 	return skb;
511 }
512 
513 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
514 				   struct sock *sk,
515 				   struct sk_buff *skb)
516 {
517 	/* don't divert link scope packets */
518 	if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
519 		return skb;
520 
521 	if (qdisc_tx_is_default(vrf_dev))
522 		return vrf_ip6_out_direct(vrf_dev, sk, skb);
523 
524 	return vrf_ip6_out_redirect(vrf_dev, skb);
525 }
526 
527 /* holding rtnl */
528 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
529 {
530 	struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
531 	struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local);
532 	struct net *net = dev_net(dev);
533 	struct dst_entry *dst;
534 
535 	RCU_INIT_POINTER(vrf->rt6, NULL);
536 	RCU_INIT_POINTER(vrf->rt6_local, NULL);
537 	synchronize_rcu();
538 
539 	/* move dev in dst's to loopback so this VRF device can be deleted
540 	 * - based on dst_ifdown
541 	 */
542 	if (rt6) {
543 		dst = &rt6->dst;
544 		dev_put(dst->dev);
545 		dst->dev = net->loopback_dev;
546 		dev_hold(dst->dev);
547 		dst_release(dst);
548 	}
549 
550 	if (rt6_local) {
551 		if (rt6_local->rt6i_idev) {
552 			in6_dev_put(rt6_local->rt6i_idev);
553 			rt6_local->rt6i_idev = NULL;
554 		}
555 
556 		dst = &rt6_local->dst;
557 		dev_put(dst->dev);
558 		dst->dev = net->loopback_dev;
559 		dev_hold(dst->dev);
560 		dst_release(dst);
561 	}
562 }
563 
564 static int vrf_rt6_create(struct net_device *dev)
565 {
566 	int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
567 	struct net_vrf *vrf = netdev_priv(dev);
568 	struct net *net = dev_net(dev);
569 	struct fib6_table *rt6i_table;
570 	struct rt6_info *rt6, *rt6_local;
571 	int rc = -ENOMEM;
572 
573 	/* IPv6 can be CONFIG enabled and then disabled runtime */
574 	if (!ipv6_mod_enabled())
575 		return 0;
576 
577 	rt6i_table = fib6_new_table(net, vrf->tb_id);
578 	if (!rt6i_table)
579 		goto out;
580 
581 	/* create a dst for routing packets out a VRF device */
582 	rt6 = ip6_dst_alloc(net, dev, flags);
583 	if (!rt6)
584 		goto out;
585 
586 	rt6->rt6i_table = rt6i_table;
587 	rt6->dst.output	= vrf_output6;
588 
589 	/* create a dst for local routing - packets sent locally
590 	 * to local address via the VRF device as a loopback
591 	 */
592 	rt6_local = ip6_dst_alloc(net, dev, flags);
593 	if (!rt6_local) {
594 		dst_release(&rt6->dst);
595 		goto out;
596 	}
597 
598 	rt6_local->rt6i_idev  = in6_dev_get(dev);
599 	rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL;
600 	rt6_local->rt6i_table = rt6i_table;
601 	rt6_local->dst.input  = ip6_input;
602 
603 	rcu_assign_pointer(vrf->rt6, rt6);
604 	rcu_assign_pointer(vrf->rt6_local, rt6_local);
605 
606 	rc = 0;
607 out:
608 	return rc;
609 }
610 #else
611 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
612 				   struct sock *sk,
613 				   struct sk_buff *skb)
614 {
615 	return skb;
616 }
617 
618 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
619 {
620 }
621 
622 static int vrf_rt6_create(struct net_device *dev)
623 {
624 	return 0;
625 }
626 #endif
627 
628 /* modelled after ip_finish_output2 */
629 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
630 {
631 	struct dst_entry *dst = skb_dst(skb);
632 	struct rtable *rt = (struct rtable *)dst;
633 	struct net_device *dev = dst->dev;
634 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
635 	struct neighbour *neigh;
636 	u32 nexthop;
637 	int ret = -EINVAL;
638 
639 	nf_reset(skb);
640 
641 	/* Be paranoid, rather than too clever. */
642 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
643 		struct sk_buff *skb2;
644 
645 		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
646 		if (!skb2) {
647 			ret = -ENOMEM;
648 			goto err;
649 		}
650 		if (skb->sk)
651 			skb_set_owner_w(skb2, skb->sk);
652 
653 		consume_skb(skb);
654 		skb = skb2;
655 	}
656 
657 	rcu_read_lock_bh();
658 
659 	nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
660 	neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
661 	if (unlikely(!neigh))
662 		neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
663 	if (!IS_ERR(neigh)) {
664 		sock_confirm_neigh(skb, neigh);
665 		ret = neigh_output(neigh, skb);
666 	}
667 
668 	rcu_read_unlock_bh();
669 err:
670 	if (unlikely(ret < 0))
671 		vrf_tx_error(skb->dev, skb);
672 	return ret;
673 }
674 
675 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
676 {
677 	struct net_device *dev = skb_dst(skb)->dev;
678 
679 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
680 
681 	skb->dev = dev;
682 	skb->protocol = htons(ETH_P_IP);
683 
684 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
685 			    net, sk, skb, NULL, dev,
686 			    vrf_finish_output,
687 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
688 }
689 
690 /* set dst on skb to send packet to us via dev_xmit path. Allows
691  * packet to go through device based features such as qdisc, netfilter
692  * hooks and packet sockets with skb->dev set to vrf device.
693  */
694 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
695 					   struct sk_buff *skb)
696 {
697 	struct net_vrf *vrf = netdev_priv(vrf_dev);
698 	struct dst_entry *dst = NULL;
699 	struct rtable *rth;
700 
701 	rcu_read_lock();
702 
703 	rth = rcu_dereference(vrf->rth);
704 	if (likely(rth)) {
705 		dst = &rth->dst;
706 		dst_hold(dst);
707 	}
708 
709 	rcu_read_unlock();
710 
711 	if (unlikely(!dst)) {
712 		vrf_tx_error(vrf_dev, skb);
713 		return NULL;
714 	}
715 
716 	skb_dst_drop(skb);
717 	skb_dst_set(skb, dst);
718 
719 	return skb;
720 }
721 
722 static int vrf_output_direct(struct net *net, struct sock *sk,
723 			     struct sk_buff *skb)
724 {
725 	skb->protocol = htons(ETH_P_IP);
726 
727 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
728 			    net, sk, skb, NULL, skb->dev,
729 			    vrf_finish_direct,
730 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
731 }
732 
733 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
734 					 struct sock *sk,
735 					 struct sk_buff *skb)
736 {
737 	struct net *net = dev_net(vrf_dev);
738 	int err;
739 
740 	skb->dev = vrf_dev;
741 
742 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
743 		      skb, NULL, vrf_dev, vrf_output_direct);
744 
745 	if (likely(err == 1))
746 		err = vrf_output_direct(net, sk, skb);
747 
748 	/* reset skb device */
749 	if (likely(err == 1))
750 		nf_reset(skb);
751 	else
752 		skb = NULL;
753 
754 	return skb;
755 }
756 
757 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
758 				  struct sock *sk,
759 				  struct sk_buff *skb)
760 {
761 	/* don't divert multicast */
762 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
763 		return skb;
764 
765 	if (qdisc_tx_is_default(vrf_dev))
766 		return vrf_ip_out_direct(vrf_dev, sk, skb);
767 
768 	return vrf_ip_out_redirect(vrf_dev, skb);
769 }
770 
771 /* called with rcu lock held */
772 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
773 				  struct sock *sk,
774 				  struct sk_buff *skb,
775 				  u16 proto)
776 {
777 	switch (proto) {
778 	case AF_INET:
779 		return vrf_ip_out(vrf_dev, sk, skb);
780 	case AF_INET6:
781 		return vrf_ip6_out(vrf_dev, sk, skb);
782 	}
783 
784 	return skb;
785 }
786 
787 /* holding rtnl */
788 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
789 {
790 	struct rtable *rth = rtnl_dereference(vrf->rth);
791 	struct rtable *rth_local = rtnl_dereference(vrf->rth_local);
792 	struct net *net = dev_net(dev);
793 	struct dst_entry *dst;
794 
795 	RCU_INIT_POINTER(vrf->rth, NULL);
796 	RCU_INIT_POINTER(vrf->rth_local, NULL);
797 	synchronize_rcu();
798 
799 	/* move dev in dst's to loopback so this VRF device can be deleted
800 	 * - based on dst_ifdown
801 	 */
802 	if (rth) {
803 		dst = &rth->dst;
804 		dev_put(dst->dev);
805 		dst->dev = net->loopback_dev;
806 		dev_hold(dst->dev);
807 		dst_release(dst);
808 	}
809 
810 	if (rth_local) {
811 		dst = &rth_local->dst;
812 		dev_put(dst->dev);
813 		dst->dev = net->loopback_dev;
814 		dev_hold(dst->dev);
815 		dst_release(dst);
816 	}
817 }
818 
819 static int vrf_rtable_create(struct net_device *dev)
820 {
821 	struct net_vrf *vrf = netdev_priv(dev);
822 	struct rtable *rth, *rth_local;
823 
824 	if (!fib_new_table(dev_net(dev), vrf->tb_id))
825 		return -ENOMEM;
826 
827 	/* create a dst for routing packets out through a VRF device */
828 	rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
829 	if (!rth)
830 		return -ENOMEM;
831 
832 	/* create a dst for local ingress routing - packets sent locally
833 	 * to local address via the VRF device as a loopback
834 	 */
835 	rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0);
836 	if (!rth_local) {
837 		dst_release(&rth->dst);
838 		return -ENOMEM;
839 	}
840 
841 	rth->dst.output	= vrf_output;
842 	rth->rt_table_id = vrf->tb_id;
843 
844 	rth_local->rt_table_id = vrf->tb_id;
845 
846 	rcu_assign_pointer(vrf->rth, rth);
847 	rcu_assign_pointer(vrf->rth_local, rth_local);
848 
849 	return 0;
850 }
851 
852 /**************************** device handling ********************/
853 
854 /* cycle interface to flush neighbor cache and move routes across tables */
855 static void cycle_netdev(struct net_device *dev)
856 {
857 	unsigned int flags = dev->flags;
858 	int ret;
859 
860 	if (!netif_running(dev))
861 		return;
862 
863 	ret = dev_change_flags(dev, flags & ~IFF_UP);
864 	if (ret >= 0)
865 		ret = dev_change_flags(dev, flags);
866 
867 	if (ret < 0) {
868 		netdev_err(dev,
869 			   "Failed to cycle device %s; route tables might be wrong!\n",
870 			   dev->name);
871 	}
872 }
873 
874 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
875 {
876 	int ret;
877 
878 	/* do not allow loopback device to be enslaved to a VRF.
879 	 * The vrf device acts as the loopback for the vrf.
880 	 */
881 	if (port_dev == dev_net(dev)->loopback_dev)
882 		return -EOPNOTSUPP;
883 
884 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
885 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
886 	if (ret < 0)
887 		goto err;
888 
889 	cycle_netdev(port_dev);
890 
891 	return 0;
892 
893 err:
894 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
895 	return ret;
896 }
897 
898 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
899 {
900 	if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
901 		return -EINVAL;
902 
903 	return do_vrf_add_slave(dev, port_dev);
904 }
905 
906 /* inverse of do_vrf_add_slave */
907 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
908 {
909 	netdev_upper_dev_unlink(port_dev, dev);
910 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
911 
912 	cycle_netdev(port_dev);
913 
914 	return 0;
915 }
916 
917 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
918 {
919 	return do_vrf_del_slave(dev, port_dev);
920 }
921 
922 static void vrf_dev_uninit(struct net_device *dev)
923 {
924 	struct net_vrf *vrf = netdev_priv(dev);
925 
926 	vrf_rtable_release(dev, vrf);
927 	vrf_rt6_release(dev, vrf);
928 
929 	free_percpu(dev->dstats);
930 	dev->dstats = NULL;
931 }
932 
933 static int vrf_dev_init(struct net_device *dev)
934 {
935 	struct net_vrf *vrf = netdev_priv(dev);
936 
937 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
938 	if (!dev->dstats)
939 		goto out_nomem;
940 
941 	/* create the default dst which points back to us */
942 	if (vrf_rtable_create(dev) != 0)
943 		goto out_stats;
944 
945 	if (vrf_rt6_create(dev) != 0)
946 		goto out_rth;
947 
948 	dev->flags = IFF_MASTER | IFF_NOARP;
949 
950 	/* MTU is irrelevant for VRF device; set to 64k similar to lo */
951 	dev->mtu = 64 * 1024;
952 
953 	/* similarly, oper state is irrelevant; set to up to avoid confusion */
954 	dev->operstate = IF_OPER_UP;
955 	netdev_lockdep_set_classes(dev);
956 	return 0;
957 
958 out_rth:
959 	vrf_rtable_release(dev, vrf);
960 out_stats:
961 	free_percpu(dev->dstats);
962 	dev->dstats = NULL;
963 out_nomem:
964 	return -ENOMEM;
965 }
966 
967 static const struct net_device_ops vrf_netdev_ops = {
968 	.ndo_init		= vrf_dev_init,
969 	.ndo_uninit		= vrf_dev_uninit,
970 	.ndo_start_xmit		= vrf_xmit,
971 	.ndo_get_stats64	= vrf_get_stats64,
972 	.ndo_add_slave		= vrf_add_slave,
973 	.ndo_del_slave		= vrf_del_slave,
974 };
975 
976 static u32 vrf_fib_table(const struct net_device *dev)
977 {
978 	struct net_vrf *vrf = netdev_priv(dev);
979 
980 	return vrf->tb_id;
981 }
982 
983 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
984 {
985 	kfree_skb(skb);
986 	return 0;
987 }
988 
989 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
990 				      struct sk_buff *skb,
991 				      struct net_device *dev)
992 {
993 	struct net *net = dev_net(dev);
994 
995 	if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
996 		skb = NULL;    /* kfree_skb(skb) handled by nf code */
997 
998 	return skb;
999 }
1000 
1001 #if IS_ENABLED(CONFIG_IPV6)
1002 /* neighbor handling is done with actual device; do not want
1003  * to flip skb->dev for those ndisc packets. This really fails
1004  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1005  * a start.
1006  */
1007 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
1008 {
1009 	const struct ipv6hdr *iph = ipv6_hdr(skb);
1010 	bool rc = false;
1011 
1012 	if (iph->nexthdr == NEXTHDR_ICMP) {
1013 		const struct icmp6hdr *icmph;
1014 		struct icmp6hdr _icmph;
1015 
1016 		icmph = skb_header_pointer(skb, sizeof(*iph),
1017 					   sizeof(_icmph), &_icmph);
1018 		if (!icmph)
1019 			goto out;
1020 
1021 		switch (icmph->icmp6_type) {
1022 		case NDISC_ROUTER_SOLICITATION:
1023 		case NDISC_ROUTER_ADVERTISEMENT:
1024 		case NDISC_NEIGHBOUR_SOLICITATION:
1025 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
1026 		case NDISC_REDIRECT:
1027 			rc = true;
1028 			break;
1029 		}
1030 	}
1031 
1032 out:
1033 	return rc;
1034 }
1035 
1036 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
1037 					     const struct net_device *dev,
1038 					     struct flowi6 *fl6,
1039 					     int ifindex,
1040 					     int flags)
1041 {
1042 	struct net_vrf *vrf = netdev_priv(dev);
1043 	struct fib6_table *table = NULL;
1044 	struct rt6_info *rt6;
1045 
1046 	rcu_read_lock();
1047 
1048 	/* fib6_table does not have a refcnt and can not be freed */
1049 	rt6 = rcu_dereference(vrf->rt6);
1050 	if (likely(rt6))
1051 		table = rt6->rt6i_table;
1052 
1053 	rcu_read_unlock();
1054 
1055 	if (!table)
1056 		return NULL;
1057 
1058 	return ip6_pol_route(net, table, ifindex, fl6, flags);
1059 }
1060 
1061 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1062 			      int ifindex)
1063 {
1064 	const struct ipv6hdr *iph = ipv6_hdr(skb);
1065 	struct flowi6 fl6 = {
1066 		.daddr          = iph->daddr,
1067 		.saddr          = iph->saddr,
1068 		.flowlabel      = ip6_flowinfo(iph),
1069 		.flowi6_mark    = skb->mark,
1070 		.flowi6_proto   = iph->nexthdr,
1071 		.flowi6_iif     = ifindex,
1072 	};
1073 	struct net *net = dev_net(vrf_dev);
1074 	struct rt6_info *rt6;
1075 
1076 	rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
1077 				   RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1078 	if (unlikely(!rt6))
1079 		return;
1080 
1081 	if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1082 		return;
1083 
1084 	skb_dst_set(skb, &rt6->dst);
1085 }
1086 
1087 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1088 				   struct sk_buff *skb)
1089 {
1090 	int orig_iif = skb->skb_iif;
1091 	bool need_strict;
1092 
1093 	/* loopback traffic; do not push through packet taps again.
1094 	 * Reset pkt_type for upper layers to process skb
1095 	 */
1096 	if (skb->pkt_type == PACKET_LOOPBACK) {
1097 		skb->dev = vrf_dev;
1098 		skb->skb_iif = vrf_dev->ifindex;
1099 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1100 		skb->pkt_type = PACKET_HOST;
1101 		goto out;
1102 	}
1103 
1104 	/* if packet is NDISC or addressed to multicast or link-local
1105 	 * then keep the ingress interface
1106 	 */
1107 	need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1108 	if (!ipv6_ndisc_frame(skb) && !need_strict) {
1109 		vrf_rx_stats(vrf_dev, skb->len);
1110 		skb->dev = vrf_dev;
1111 		skb->skb_iif = vrf_dev->ifindex;
1112 
1113 		if (!list_empty(&vrf_dev->ptype_all)) {
1114 			skb_push(skb, skb->mac_len);
1115 			dev_queue_xmit_nit(skb, vrf_dev);
1116 			skb_pull(skb, skb->mac_len);
1117 		}
1118 
1119 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1120 	}
1121 
1122 	if (need_strict)
1123 		vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1124 
1125 	skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1126 out:
1127 	return skb;
1128 }
1129 
1130 #else
1131 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1132 				   struct sk_buff *skb)
1133 {
1134 	return skb;
1135 }
1136 #endif
1137 
1138 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1139 				  struct sk_buff *skb)
1140 {
1141 	skb->dev = vrf_dev;
1142 	skb->skb_iif = vrf_dev->ifindex;
1143 	IPCB(skb)->flags |= IPSKB_L3SLAVE;
1144 
1145 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1146 		goto out;
1147 
1148 	/* loopback traffic; do not push through packet taps again.
1149 	 * Reset pkt_type for upper layers to process skb
1150 	 */
1151 	if (skb->pkt_type == PACKET_LOOPBACK) {
1152 		skb->pkt_type = PACKET_HOST;
1153 		goto out;
1154 	}
1155 
1156 	vrf_rx_stats(vrf_dev, skb->len);
1157 
1158 	if (!list_empty(&vrf_dev->ptype_all)) {
1159 		skb_push(skb, skb->mac_len);
1160 		dev_queue_xmit_nit(skb, vrf_dev);
1161 		skb_pull(skb, skb->mac_len);
1162 	}
1163 
1164 	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1165 out:
1166 	return skb;
1167 }
1168 
1169 /* called with rcu lock held */
1170 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1171 				  struct sk_buff *skb,
1172 				  u16 proto)
1173 {
1174 	switch (proto) {
1175 	case AF_INET:
1176 		return vrf_ip_rcv(vrf_dev, skb);
1177 	case AF_INET6:
1178 		return vrf_ip6_rcv(vrf_dev, skb);
1179 	}
1180 
1181 	return skb;
1182 }
1183 
1184 #if IS_ENABLED(CONFIG_IPV6)
1185 /* send to link-local or multicast address via interface enslaved to
1186  * VRF device. Force lookup to VRF table without changing flow struct
1187  */
1188 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1189 					      struct flowi6 *fl6)
1190 {
1191 	struct net *net = dev_net(dev);
1192 	int flags = RT6_LOOKUP_F_IFACE;
1193 	struct dst_entry *dst = NULL;
1194 	struct rt6_info *rt;
1195 
1196 	/* VRF device does not have a link-local address and
1197 	 * sending packets to link-local or mcast addresses over
1198 	 * a VRF device does not make sense
1199 	 */
1200 	if (fl6->flowi6_oif == dev->ifindex) {
1201 		dst = &net->ipv6.ip6_null_entry->dst;
1202 		dst_hold(dst);
1203 		return dst;
1204 	}
1205 
1206 	if (!ipv6_addr_any(&fl6->saddr))
1207 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1208 
1209 	rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1210 	if (rt)
1211 		dst = &rt->dst;
1212 
1213 	return dst;
1214 }
1215 #endif
1216 
1217 static const struct l3mdev_ops vrf_l3mdev_ops = {
1218 	.l3mdev_fib_table	= vrf_fib_table,
1219 	.l3mdev_l3_rcv		= vrf_l3_rcv,
1220 	.l3mdev_l3_out		= vrf_l3_out,
1221 #if IS_ENABLED(CONFIG_IPV6)
1222 	.l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1223 #endif
1224 };
1225 
1226 static void vrf_get_drvinfo(struct net_device *dev,
1227 			    struct ethtool_drvinfo *info)
1228 {
1229 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1230 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1231 }
1232 
1233 static const struct ethtool_ops vrf_ethtool_ops = {
1234 	.get_drvinfo	= vrf_get_drvinfo,
1235 };
1236 
1237 static inline size_t vrf_fib_rule_nl_size(void)
1238 {
1239 	size_t sz;
1240 
1241 	sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1242 	sz += nla_total_size(sizeof(u8));	/* FRA_L3MDEV */
1243 	sz += nla_total_size(sizeof(u32));	/* FRA_PRIORITY */
1244 
1245 	return sz;
1246 }
1247 
1248 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1249 {
1250 	struct fib_rule_hdr *frh;
1251 	struct nlmsghdr *nlh;
1252 	struct sk_buff *skb;
1253 	int err;
1254 
1255 	if (family == AF_INET6 && !ipv6_mod_enabled())
1256 		return 0;
1257 
1258 	skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1259 	if (!skb)
1260 		return -ENOMEM;
1261 
1262 	nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1263 	if (!nlh)
1264 		goto nla_put_failure;
1265 
1266 	/* rule only needs to appear once */
1267 	nlh->nlmsg_flags |= NLM_F_EXCL;
1268 
1269 	frh = nlmsg_data(nlh);
1270 	memset(frh, 0, sizeof(*frh));
1271 	frh->family = family;
1272 	frh->action = FR_ACT_TO_TBL;
1273 
1274 	if (nla_put_u32(skb, FRA_L3MDEV, 1))
1275 		goto nla_put_failure;
1276 
1277 	if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1278 		goto nla_put_failure;
1279 
1280 	nlmsg_end(skb, nlh);
1281 
1282 	/* fib_nl_{new,del}rule handling looks for net from skb->sk */
1283 	skb->sk = dev_net(dev)->rtnl;
1284 	if (add_it) {
1285 		err = fib_nl_newrule(skb, nlh, NULL);
1286 		if (err == -EEXIST)
1287 			err = 0;
1288 	} else {
1289 		err = fib_nl_delrule(skb, nlh, NULL);
1290 		if (err == -ENOENT)
1291 			err = 0;
1292 	}
1293 	nlmsg_free(skb);
1294 
1295 	return err;
1296 
1297 nla_put_failure:
1298 	nlmsg_free(skb);
1299 
1300 	return -EMSGSIZE;
1301 }
1302 
1303 static int vrf_add_fib_rules(const struct net_device *dev)
1304 {
1305 	int err;
1306 
1307 	err = vrf_fib_rule(dev, AF_INET,  true);
1308 	if (err < 0)
1309 		goto out_err;
1310 
1311 	err = vrf_fib_rule(dev, AF_INET6, true);
1312 	if (err < 0)
1313 		goto ipv6_err;
1314 
1315 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1316 	err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1317 	if (err < 0)
1318 		goto ipmr_err;
1319 #endif
1320 
1321 	return 0;
1322 
1323 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1324 ipmr_err:
1325 	vrf_fib_rule(dev, AF_INET6,  false);
1326 #endif
1327 
1328 ipv6_err:
1329 	vrf_fib_rule(dev, AF_INET,  false);
1330 
1331 out_err:
1332 	netdev_err(dev, "Failed to add FIB rules.\n");
1333 	return err;
1334 }
1335 
1336 static void vrf_setup(struct net_device *dev)
1337 {
1338 	ether_setup(dev);
1339 
1340 	/* Initialize the device structure. */
1341 	dev->netdev_ops = &vrf_netdev_ops;
1342 	dev->l3mdev_ops = &vrf_l3mdev_ops;
1343 	dev->ethtool_ops = &vrf_ethtool_ops;
1344 	dev->needs_free_netdev = true;
1345 
1346 	/* Fill in device structure with ethernet-generic values. */
1347 	eth_hw_addr_random(dev);
1348 
1349 	/* don't acquire vrf device's netif_tx_lock when transmitting */
1350 	dev->features |= NETIF_F_LLTX;
1351 
1352 	/* don't allow vrf devices to change network namespaces. */
1353 	dev->features |= NETIF_F_NETNS_LOCAL;
1354 
1355 	/* does not make sense for a VLAN to be added to a vrf device */
1356 	dev->features   |= NETIF_F_VLAN_CHALLENGED;
1357 
1358 	/* enable offload features */
1359 	dev->features   |= NETIF_F_GSO_SOFTWARE;
1360 	dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1361 	dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1362 
1363 	dev->hw_features = dev->features;
1364 	dev->hw_enc_features = dev->features;
1365 
1366 	/* default to no qdisc; user can add if desired */
1367 	dev->priv_flags |= IFF_NO_QUEUE;
1368 }
1369 
1370 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1371 			struct netlink_ext_ack *extack)
1372 {
1373 	if (tb[IFLA_ADDRESS]) {
1374 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1375 			return -EINVAL;
1376 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1377 			return -EADDRNOTAVAIL;
1378 	}
1379 	return 0;
1380 }
1381 
1382 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1383 {
1384 	struct net_device *port_dev;
1385 	struct list_head *iter;
1386 
1387 	netdev_for_each_lower_dev(dev, port_dev, iter)
1388 		vrf_del_slave(dev, port_dev);
1389 
1390 	unregister_netdevice_queue(dev, head);
1391 }
1392 
1393 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1394 		       struct nlattr *tb[], struct nlattr *data[],
1395 		       struct netlink_ext_ack *extack)
1396 {
1397 	struct net_vrf *vrf = netdev_priv(dev);
1398 	bool *add_fib_rules;
1399 	struct net *net;
1400 	int err;
1401 
1402 	if (!data || !data[IFLA_VRF_TABLE])
1403 		return -EINVAL;
1404 
1405 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1406 	if (vrf->tb_id == RT_TABLE_UNSPEC)
1407 		return -EINVAL;
1408 
1409 	dev->priv_flags |= IFF_L3MDEV_MASTER;
1410 
1411 	err = register_netdevice(dev);
1412 	if (err)
1413 		goto out;
1414 
1415 	net = dev_net(dev);
1416 	add_fib_rules = net_generic(net, vrf_net_id);
1417 	if (*add_fib_rules) {
1418 		err = vrf_add_fib_rules(dev);
1419 		if (err) {
1420 			unregister_netdevice(dev);
1421 			goto out;
1422 		}
1423 		*add_fib_rules = false;
1424 	}
1425 
1426 out:
1427 	return err;
1428 }
1429 
1430 static size_t vrf_nl_getsize(const struct net_device *dev)
1431 {
1432 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
1433 }
1434 
1435 static int vrf_fillinfo(struct sk_buff *skb,
1436 			const struct net_device *dev)
1437 {
1438 	struct net_vrf *vrf = netdev_priv(dev);
1439 
1440 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1441 }
1442 
1443 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1444 				 const struct net_device *slave_dev)
1445 {
1446 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
1447 }
1448 
1449 static int vrf_fill_slave_info(struct sk_buff *skb,
1450 			       const struct net_device *vrf_dev,
1451 			       const struct net_device *slave_dev)
1452 {
1453 	struct net_vrf *vrf = netdev_priv(vrf_dev);
1454 
1455 	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1456 		return -EMSGSIZE;
1457 
1458 	return 0;
1459 }
1460 
1461 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1462 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
1463 };
1464 
1465 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1466 	.kind		= DRV_NAME,
1467 	.priv_size	= sizeof(struct net_vrf),
1468 
1469 	.get_size	= vrf_nl_getsize,
1470 	.policy		= vrf_nl_policy,
1471 	.validate	= vrf_validate,
1472 	.fill_info	= vrf_fillinfo,
1473 
1474 	.get_slave_size  = vrf_get_slave_size,
1475 	.fill_slave_info = vrf_fill_slave_info,
1476 
1477 	.newlink	= vrf_newlink,
1478 	.dellink	= vrf_dellink,
1479 	.setup		= vrf_setup,
1480 	.maxtype	= IFLA_VRF_MAX,
1481 };
1482 
1483 static int vrf_device_event(struct notifier_block *unused,
1484 			    unsigned long event, void *ptr)
1485 {
1486 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1487 
1488 	/* only care about unregister events to drop slave references */
1489 	if (event == NETDEV_UNREGISTER) {
1490 		struct net_device *vrf_dev;
1491 
1492 		if (!netif_is_l3_slave(dev))
1493 			goto out;
1494 
1495 		vrf_dev = netdev_master_upper_dev_get(dev);
1496 		vrf_del_slave(vrf_dev, dev);
1497 	}
1498 out:
1499 	return NOTIFY_DONE;
1500 }
1501 
1502 static struct notifier_block vrf_notifier_block __read_mostly = {
1503 	.notifier_call = vrf_device_event,
1504 };
1505 
1506 /* Initialize per network namespace state */
1507 static int __net_init vrf_netns_init(struct net *net)
1508 {
1509 	bool *add_fib_rules = net_generic(net, vrf_net_id);
1510 
1511 	*add_fib_rules = true;
1512 
1513 	return 0;
1514 }
1515 
1516 static struct pernet_operations vrf_net_ops __net_initdata = {
1517 	.init = vrf_netns_init,
1518 	.id   = &vrf_net_id,
1519 	.size = sizeof(bool),
1520 };
1521 
1522 static int __init vrf_init_module(void)
1523 {
1524 	int rc;
1525 
1526 	register_netdevice_notifier(&vrf_notifier_block);
1527 
1528 	rc = register_pernet_subsys(&vrf_net_ops);
1529 	if (rc < 0)
1530 		goto error;
1531 
1532 	rc = rtnl_link_register(&vrf_link_ops);
1533 	if (rc < 0) {
1534 		unregister_pernet_subsys(&vrf_net_ops);
1535 		goto error;
1536 	}
1537 
1538 	return 0;
1539 
1540 error:
1541 	unregister_netdevice_notifier(&vrf_notifier_block);
1542 	return rc;
1543 }
1544 
1545 module_init(vrf_init_module);
1546 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1547 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1548 MODULE_LICENSE("GPL");
1549 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1550 MODULE_VERSION(DRV_VERSION);
1551