xref: /linux/drivers/net/vrf.c (revision 37f0e658eeeac720f3d558cf5aaf9edf0705ff23)
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 
39 #define RT_FL_TOS(oldflp4) \
40 	((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
41 
42 #define DRV_NAME	"vrf"
43 #define DRV_VERSION	"1.0"
44 
45 #define vrf_master_get_rcu(dev) \
46 	((struct net_device *)rcu_dereference(dev->rx_handler_data))
47 
48 struct net_vrf {
49 	struct rtable           *rth;
50 	struct rt6_info		*rt6;
51 	u32                     tb_id;
52 };
53 
54 struct pcpu_dstats {
55 	u64			tx_pkts;
56 	u64			tx_bytes;
57 	u64			tx_drps;
58 	u64			rx_pkts;
59 	u64			rx_bytes;
60 	struct u64_stats_sync	syncp;
61 };
62 
63 /* neighbor handling is done with actual device; do not want
64  * to flip skb->dev for those ndisc packets. This really fails
65  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
66  * a start.
67  */
68 #if IS_ENABLED(CONFIG_IPV6)
69 static bool check_ipv6_frame(const struct sk_buff *skb)
70 {
71 	const struct ipv6hdr *ipv6h;
72 	struct ipv6hdr _ipv6h;
73 	bool rc = true;
74 
75 	ipv6h = skb_header_pointer(skb, 0, sizeof(_ipv6h), &_ipv6h);
76 	if (!ipv6h)
77 		goto out;
78 
79 	if (ipv6h->nexthdr == NEXTHDR_ICMP) {
80 		const struct icmp6hdr *icmph;
81 		struct icmp6hdr _icmph;
82 
83 		icmph = skb_header_pointer(skb, sizeof(_ipv6h),
84 					   sizeof(_icmph), &_icmph);
85 		if (!icmph)
86 			goto out;
87 
88 		switch (icmph->icmp6_type) {
89 		case NDISC_ROUTER_SOLICITATION:
90 		case NDISC_ROUTER_ADVERTISEMENT:
91 		case NDISC_NEIGHBOUR_SOLICITATION:
92 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
93 		case NDISC_REDIRECT:
94 			rc = false;
95 			break;
96 		}
97 	}
98 
99 out:
100 	return rc;
101 }
102 #else
103 static bool check_ipv6_frame(const struct sk_buff *skb)
104 {
105 	return false;
106 }
107 #endif
108 
109 static bool is_ip_rx_frame(struct sk_buff *skb)
110 {
111 	switch (skb->protocol) {
112 	case htons(ETH_P_IP):
113 		return true;
114 	case htons(ETH_P_IPV6):
115 		return check_ipv6_frame(skb);
116 	}
117 	return false;
118 }
119 
120 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
121 {
122 	vrf_dev->stats.tx_errors++;
123 	kfree_skb(skb);
124 }
125 
126 /* note: already called with rcu_read_lock */
127 static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb)
128 {
129 	struct sk_buff *skb = *pskb;
130 
131 	if (is_ip_rx_frame(skb)) {
132 		struct net_device *dev = vrf_master_get_rcu(skb->dev);
133 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
134 
135 		u64_stats_update_begin(&dstats->syncp);
136 		dstats->rx_pkts++;
137 		dstats->rx_bytes += skb->len;
138 		u64_stats_update_end(&dstats->syncp);
139 
140 		skb->dev = dev;
141 
142 		return RX_HANDLER_ANOTHER;
143 	}
144 	return RX_HANDLER_PASS;
145 }
146 
147 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
148 						 struct rtnl_link_stats64 *stats)
149 {
150 	int i;
151 
152 	for_each_possible_cpu(i) {
153 		const struct pcpu_dstats *dstats;
154 		u64 tbytes, tpkts, tdrops, rbytes, rpkts;
155 		unsigned int start;
156 
157 		dstats = per_cpu_ptr(dev->dstats, i);
158 		do {
159 			start = u64_stats_fetch_begin_irq(&dstats->syncp);
160 			tbytes = dstats->tx_bytes;
161 			tpkts = dstats->tx_pkts;
162 			tdrops = dstats->tx_drps;
163 			rbytes = dstats->rx_bytes;
164 			rpkts = dstats->rx_pkts;
165 		} while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
166 		stats->tx_bytes += tbytes;
167 		stats->tx_packets += tpkts;
168 		stats->tx_dropped += tdrops;
169 		stats->rx_bytes += rbytes;
170 		stats->rx_packets += rpkts;
171 	}
172 	return stats;
173 }
174 
175 #if IS_ENABLED(CONFIG_IPV6)
176 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
177 					   struct net_device *dev)
178 {
179 	const struct ipv6hdr *iph = ipv6_hdr(skb);
180 	struct net *net = dev_net(skb->dev);
181 	struct flowi6 fl6 = {
182 		/* needed to match OIF rule */
183 		.flowi6_oif = dev->ifindex,
184 		.flowi6_iif = LOOPBACK_IFINDEX,
185 		.daddr = iph->daddr,
186 		.saddr = iph->saddr,
187 		.flowlabel = ip6_flowinfo(iph),
188 		.flowi6_mark = skb->mark,
189 		.flowi6_proto = iph->nexthdr,
190 		.flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF,
191 	};
192 	int ret = NET_XMIT_DROP;
193 	struct dst_entry *dst;
194 	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
195 
196 	dst = ip6_route_output(net, NULL, &fl6);
197 	if (dst == dst_null)
198 		goto err;
199 
200 	skb_dst_drop(skb);
201 	skb_dst_set(skb, dst);
202 
203 	ret = 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 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
224 			    struct net_device *vrf_dev)
225 {
226 	struct rtable *rt;
227 	int err = 1;
228 
229 	rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
230 	if (IS_ERR(rt))
231 		goto out;
232 
233 	/* TO-DO: what about broadcast ? */
234 	if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
235 		ip_rt_put(rt);
236 		goto out;
237 	}
238 
239 	skb_dst_drop(skb);
240 	skb_dst_set(skb, &rt->dst);
241 	err = 0;
242 out:
243 	return err;
244 }
245 
246 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
247 					   struct net_device *vrf_dev)
248 {
249 	struct iphdr *ip4h = ip_hdr(skb);
250 	int ret = NET_XMIT_DROP;
251 	struct flowi4 fl4 = {
252 		/* needed to match OIF rule */
253 		.flowi4_oif = vrf_dev->ifindex,
254 		.flowi4_iif = LOOPBACK_IFINDEX,
255 		.flowi4_tos = RT_TOS(ip4h->tos),
256 		.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
257 				FLOWI_FLAG_SKIP_NH_OIF,
258 		.daddr = ip4h->daddr,
259 	};
260 
261 	if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
262 		goto err;
263 
264 	if (!ip4h->saddr) {
265 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
266 					       RT_SCOPE_LINK);
267 	}
268 
269 	ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
270 	if (unlikely(net_xmit_eval(ret)))
271 		vrf_dev->stats.tx_errors++;
272 	else
273 		ret = NET_XMIT_SUCCESS;
274 
275 out:
276 	return ret;
277 err:
278 	vrf_tx_error(vrf_dev, skb);
279 	goto out;
280 }
281 
282 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
283 {
284 	/* strip the ethernet header added for pass through VRF device */
285 	__skb_pull(skb, skb_network_offset(skb));
286 
287 	switch (skb->protocol) {
288 	case htons(ETH_P_IP):
289 		return vrf_process_v4_outbound(skb, dev);
290 	case htons(ETH_P_IPV6):
291 		return vrf_process_v6_outbound(skb, dev);
292 	default:
293 		vrf_tx_error(dev, skb);
294 		return NET_XMIT_DROP;
295 	}
296 }
297 
298 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
299 {
300 	netdev_tx_t ret = is_ip_tx_frame(skb, dev);
301 
302 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
303 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
304 
305 		u64_stats_update_begin(&dstats->syncp);
306 		dstats->tx_pkts++;
307 		dstats->tx_bytes += skb->len;
308 		u64_stats_update_end(&dstats->syncp);
309 	} else {
310 		this_cpu_inc(dev->dstats->tx_drps);
311 	}
312 
313 	return ret;
314 }
315 
316 #if IS_ENABLED(CONFIG_IPV6)
317 /* modelled after ip6_finish_output2 */
318 static int vrf_finish_output6(struct net *net, struct sock *sk,
319 			      struct sk_buff *skb)
320 {
321 	struct dst_entry *dst = skb_dst(skb);
322 	struct net_device *dev = dst->dev;
323 	struct neighbour *neigh;
324 	struct in6_addr *nexthop;
325 	int ret;
326 
327 	skb->protocol = htons(ETH_P_IPV6);
328 	skb->dev = dev;
329 
330 	rcu_read_lock_bh();
331 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
332 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
333 	if (unlikely(!neigh))
334 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
335 	if (!IS_ERR(neigh)) {
336 		ret = dst_neigh_output(dst, neigh, skb);
337 		rcu_read_unlock_bh();
338 		return ret;
339 	}
340 	rcu_read_unlock_bh();
341 
342 	IP6_INC_STATS(dev_net(dst->dev),
343 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
344 	kfree_skb(skb);
345 	return -EINVAL;
346 }
347 
348 /* modelled after ip6_output */
349 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
350 {
351 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
352 			    net, sk, skb, NULL, skb_dst(skb)->dev,
353 			    vrf_finish_output6,
354 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
355 }
356 
357 static void vrf_rt6_release(struct net_vrf *vrf)
358 {
359 	dst_release(&vrf->rt6->dst);
360 	vrf->rt6 = NULL;
361 }
362 
363 static int vrf_rt6_create(struct net_device *dev)
364 {
365 	struct net_vrf *vrf = netdev_priv(dev);
366 	struct net *net = dev_net(dev);
367 	struct rt6_info *rt6;
368 	int rc = -ENOMEM;
369 
370 	rt6 = ip6_dst_alloc(net, dev,
371 			    DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE);
372 	if (!rt6)
373 		goto out;
374 
375 	rt6->dst.output	= vrf_output6;
376 	rt6->rt6i_table = fib6_get_table(net, vrf->tb_id);
377 	dst_hold(&rt6->dst);
378 	vrf->rt6 = rt6;
379 	rc = 0;
380 out:
381 	return rc;
382 }
383 #else
384 static void vrf_rt6_release(struct net_vrf *vrf)
385 {
386 }
387 
388 static int vrf_rt6_create(struct net_device *dev)
389 {
390 	return 0;
391 }
392 #endif
393 
394 /* modelled after ip_finish_output2 */
395 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
396 {
397 	struct dst_entry *dst = skb_dst(skb);
398 	struct rtable *rt = (struct rtable *)dst;
399 	struct net_device *dev = dst->dev;
400 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
401 	struct neighbour *neigh;
402 	u32 nexthop;
403 	int ret = -EINVAL;
404 
405 	/* Be paranoid, rather than too clever. */
406 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
407 		struct sk_buff *skb2;
408 
409 		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
410 		if (!skb2) {
411 			ret = -ENOMEM;
412 			goto err;
413 		}
414 		if (skb->sk)
415 			skb_set_owner_w(skb2, skb->sk);
416 
417 		consume_skb(skb);
418 		skb = skb2;
419 	}
420 
421 	rcu_read_lock_bh();
422 
423 	nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
424 	neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
425 	if (unlikely(!neigh))
426 		neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
427 	if (!IS_ERR(neigh))
428 		ret = dst_neigh_output(dst, neigh, skb);
429 
430 	rcu_read_unlock_bh();
431 err:
432 	if (unlikely(ret < 0))
433 		vrf_tx_error(skb->dev, skb);
434 	return ret;
435 }
436 
437 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
438 {
439 	struct net_device *dev = skb_dst(skb)->dev;
440 
441 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
442 
443 	skb->dev = dev;
444 	skb->protocol = htons(ETH_P_IP);
445 
446 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
447 			    net, sk, skb, NULL, dev,
448 			    vrf_finish_output,
449 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
450 }
451 
452 static void vrf_rtable_release(struct net_vrf *vrf)
453 {
454 	struct dst_entry *dst = (struct dst_entry *)vrf->rth;
455 
456 	dst_release(dst);
457 	vrf->rth = NULL;
458 }
459 
460 static struct rtable *vrf_rtable_create(struct net_device *dev)
461 {
462 	struct net_vrf *vrf = netdev_priv(dev);
463 	struct rtable *rth;
464 
465 	rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
466 	if (rth) {
467 		rth->dst.output	= vrf_output;
468 		rth->rt_table_id = vrf->tb_id;
469 	}
470 
471 	return rth;
472 }
473 
474 /**************************** device handling ********************/
475 
476 /* cycle interface to flush neighbor cache and move routes across tables */
477 static void cycle_netdev(struct net_device *dev)
478 {
479 	unsigned int flags = dev->flags;
480 	int ret;
481 
482 	if (!netif_running(dev))
483 		return;
484 
485 	ret = dev_change_flags(dev, flags & ~IFF_UP);
486 	if (ret >= 0)
487 		ret = dev_change_flags(dev, flags);
488 
489 	if (ret < 0) {
490 		netdev_err(dev,
491 			   "Failed to cycle device %s; route tables might be wrong!\n",
492 			   dev->name);
493 	}
494 }
495 
496 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
497 {
498 	int ret;
499 
500 	/* register the packet handler for slave ports */
501 	ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev);
502 	if (ret) {
503 		netdev_err(port_dev,
504 			   "Device %s failed to register rx_handler\n",
505 			   port_dev->name);
506 		goto out_fail;
507 	}
508 
509 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
510 	if (ret < 0)
511 		goto out_unregister;
512 
513 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
514 	cycle_netdev(port_dev);
515 
516 	return 0;
517 
518 out_unregister:
519 	netdev_rx_handler_unregister(port_dev);
520 out_fail:
521 	return ret;
522 }
523 
524 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
525 {
526 	if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
527 		return -EINVAL;
528 
529 	return do_vrf_add_slave(dev, port_dev);
530 }
531 
532 /* inverse of do_vrf_add_slave */
533 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
534 {
535 	netdev_upper_dev_unlink(port_dev, dev);
536 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
537 
538 	netdev_rx_handler_unregister(port_dev);
539 
540 	cycle_netdev(port_dev);
541 
542 	return 0;
543 }
544 
545 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
546 {
547 	return do_vrf_del_slave(dev, port_dev);
548 }
549 
550 static void vrf_dev_uninit(struct net_device *dev)
551 {
552 	struct net_vrf *vrf = netdev_priv(dev);
553 	struct net_device *port_dev;
554 	struct list_head *iter;
555 
556 	vrf_rtable_release(vrf);
557 	vrf_rt6_release(vrf);
558 
559 	netdev_for_each_lower_dev(dev, port_dev, iter)
560 		vrf_del_slave(dev, port_dev);
561 
562 	free_percpu(dev->dstats);
563 	dev->dstats = NULL;
564 }
565 
566 static int vrf_dev_init(struct net_device *dev)
567 {
568 	struct net_vrf *vrf = netdev_priv(dev);
569 
570 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
571 	if (!dev->dstats)
572 		goto out_nomem;
573 
574 	/* create the default dst which points back to us */
575 	vrf->rth = vrf_rtable_create(dev);
576 	if (!vrf->rth)
577 		goto out_stats;
578 
579 	if (vrf_rt6_create(dev) != 0)
580 		goto out_rth;
581 
582 	dev->flags = IFF_MASTER | IFF_NOARP;
583 
584 	return 0;
585 
586 out_rth:
587 	vrf_rtable_release(vrf);
588 out_stats:
589 	free_percpu(dev->dstats);
590 	dev->dstats = NULL;
591 out_nomem:
592 	return -ENOMEM;
593 }
594 
595 static const struct net_device_ops vrf_netdev_ops = {
596 	.ndo_init		= vrf_dev_init,
597 	.ndo_uninit		= vrf_dev_uninit,
598 	.ndo_start_xmit		= vrf_xmit,
599 	.ndo_get_stats64	= vrf_get_stats64,
600 	.ndo_add_slave		= vrf_add_slave,
601 	.ndo_del_slave		= vrf_del_slave,
602 };
603 
604 static u32 vrf_fib_table(const struct net_device *dev)
605 {
606 	struct net_vrf *vrf = netdev_priv(dev);
607 
608 	return vrf->tb_id;
609 }
610 
611 static struct rtable *vrf_get_rtable(const struct net_device *dev,
612 				     const struct flowi4 *fl4)
613 {
614 	struct rtable *rth = NULL;
615 
616 	if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
617 		struct net_vrf *vrf = netdev_priv(dev);
618 
619 		rth = vrf->rth;
620 		dst_hold(&rth->dst);
621 	}
622 
623 	return rth;
624 }
625 
626 /* called under rcu_read_lock */
627 static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4)
628 {
629 	struct fib_result res = { .tclassid = 0 };
630 	struct net *net = dev_net(dev);
631 	u32 orig_tos = fl4->flowi4_tos;
632 	u8 flags = fl4->flowi4_flags;
633 	u8 scope = fl4->flowi4_scope;
634 	u8 tos = RT_FL_TOS(fl4);
635 	int rc;
636 
637 	if (unlikely(!fl4->daddr))
638 		return 0;
639 
640 	fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF;
641 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
642 	fl4->flowi4_tos = tos & IPTOS_RT_MASK;
643 	fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
644 			     RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
645 
646 	rc = fib_lookup(net, fl4, &res, 0);
647 	if (!rc) {
648 		if (res.type == RTN_LOCAL)
649 			fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr;
650 		else
651 			fib_select_path(net, &res, fl4, -1);
652 	}
653 
654 	fl4->flowi4_flags = flags;
655 	fl4->flowi4_tos = orig_tos;
656 	fl4->flowi4_scope = scope;
657 
658 	return rc;
659 }
660 
661 #if IS_ENABLED(CONFIG_IPV6)
662 static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
663 					 const struct flowi6 *fl6)
664 {
665 	struct rt6_info *rt = NULL;
666 
667 	if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) {
668 		struct net_vrf *vrf = netdev_priv(dev);
669 
670 		rt = vrf->rt6;
671 		dst_hold(&rt->dst);
672 	}
673 
674 	return (struct dst_entry *)rt;
675 }
676 #endif
677 
678 static const struct l3mdev_ops vrf_l3mdev_ops = {
679 	.l3mdev_fib_table	= vrf_fib_table,
680 	.l3mdev_get_rtable	= vrf_get_rtable,
681 	.l3mdev_get_saddr	= vrf_get_saddr,
682 #if IS_ENABLED(CONFIG_IPV6)
683 	.l3mdev_get_rt6_dst	= vrf_get_rt6_dst,
684 #endif
685 };
686 
687 static void vrf_get_drvinfo(struct net_device *dev,
688 			    struct ethtool_drvinfo *info)
689 {
690 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
691 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
692 }
693 
694 static const struct ethtool_ops vrf_ethtool_ops = {
695 	.get_drvinfo	= vrf_get_drvinfo,
696 };
697 
698 static void vrf_setup(struct net_device *dev)
699 {
700 	ether_setup(dev);
701 
702 	/* Initialize the device structure. */
703 	dev->netdev_ops = &vrf_netdev_ops;
704 	dev->l3mdev_ops = &vrf_l3mdev_ops;
705 	dev->ethtool_ops = &vrf_ethtool_ops;
706 	dev->destructor = free_netdev;
707 
708 	/* Fill in device structure with ethernet-generic values. */
709 	eth_hw_addr_random(dev);
710 
711 	/* don't acquire vrf device's netif_tx_lock when transmitting */
712 	dev->features |= NETIF_F_LLTX;
713 
714 	/* don't allow vrf devices to change network namespaces. */
715 	dev->features |= NETIF_F_NETNS_LOCAL;
716 }
717 
718 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
719 {
720 	if (tb[IFLA_ADDRESS]) {
721 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
722 			return -EINVAL;
723 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
724 			return -EADDRNOTAVAIL;
725 	}
726 	return 0;
727 }
728 
729 static void vrf_dellink(struct net_device *dev, struct list_head *head)
730 {
731 	unregister_netdevice_queue(dev, head);
732 }
733 
734 static int vrf_newlink(struct net *src_net, struct net_device *dev,
735 		       struct nlattr *tb[], struct nlattr *data[])
736 {
737 	struct net_vrf *vrf = netdev_priv(dev);
738 
739 	if (!data || !data[IFLA_VRF_TABLE])
740 		return -EINVAL;
741 
742 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
743 
744 	dev->priv_flags |= IFF_L3MDEV_MASTER;
745 
746 	return register_netdevice(dev);
747 }
748 
749 static size_t vrf_nl_getsize(const struct net_device *dev)
750 {
751 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
752 }
753 
754 static int vrf_fillinfo(struct sk_buff *skb,
755 			const struct net_device *dev)
756 {
757 	struct net_vrf *vrf = netdev_priv(dev);
758 
759 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
760 }
761 
762 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
763 				 const struct net_device *slave_dev)
764 {
765 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
766 }
767 
768 static int vrf_fill_slave_info(struct sk_buff *skb,
769 			       const struct net_device *vrf_dev,
770 			       const struct net_device *slave_dev)
771 {
772 	struct net_vrf *vrf = netdev_priv(vrf_dev);
773 
774 	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
775 		return -EMSGSIZE;
776 
777 	return 0;
778 }
779 
780 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
781 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
782 };
783 
784 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
785 	.kind		= DRV_NAME,
786 	.priv_size	= sizeof(struct net_vrf),
787 
788 	.get_size	= vrf_nl_getsize,
789 	.policy		= vrf_nl_policy,
790 	.validate	= vrf_validate,
791 	.fill_info	= vrf_fillinfo,
792 
793 	.get_slave_size  = vrf_get_slave_size,
794 	.fill_slave_info = vrf_fill_slave_info,
795 
796 	.newlink	= vrf_newlink,
797 	.dellink	= vrf_dellink,
798 	.setup		= vrf_setup,
799 	.maxtype	= IFLA_VRF_MAX,
800 };
801 
802 static int vrf_device_event(struct notifier_block *unused,
803 			    unsigned long event, void *ptr)
804 {
805 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
806 
807 	/* only care about unregister events to drop slave references */
808 	if (event == NETDEV_UNREGISTER) {
809 		struct net_device *vrf_dev;
810 
811 		if (!netif_is_l3_slave(dev))
812 			goto out;
813 
814 		vrf_dev = netdev_master_upper_dev_get(dev);
815 		vrf_del_slave(vrf_dev, dev);
816 	}
817 out:
818 	return NOTIFY_DONE;
819 }
820 
821 static struct notifier_block vrf_notifier_block __read_mostly = {
822 	.notifier_call = vrf_device_event,
823 };
824 
825 static int __init vrf_init_module(void)
826 {
827 	int rc;
828 
829 	register_netdevice_notifier(&vrf_notifier_block);
830 
831 	rc = rtnl_link_register(&vrf_link_ops);
832 	if (rc < 0)
833 		goto error;
834 
835 	return 0;
836 
837 error:
838 	unregister_netdevice_notifier(&vrf_notifier_block);
839 	return rc;
840 }
841 
842 module_init(vrf_init_module);
843 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
844 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
845 MODULE_LICENSE("GPL");
846 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
847 MODULE_VERSION(DRV_VERSION);
848