xref: /openbmc/linux/drivers/net/vrf.c (revision 9cfc5c90)
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/rtnetlink.h>
36 #include <net/route.h>
37 #include <net/addrconf.h>
38 #include <net/l3mdev.h>
39 
40 #define RT_FL_TOS(oldflp4) \
41 	((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
42 
43 #define DRV_NAME	"vrf"
44 #define DRV_VERSION	"1.0"
45 
46 #define vrf_master_get_rcu(dev) \
47 	((struct net_device *)rcu_dereference(dev->rx_handler_data))
48 
49 struct slave {
50 	struct list_head        list;
51 	struct net_device       *dev;
52 };
53 
54 struct slave_queue {
55 	struct list_head        all_slaves;
56 };
57 
58 struct net_vrf {
59 	struct slave_queue      queue;
60 	struct rtable           *rth;
61 	struct rt6_info		*rt6;
62 	u32                     tb_id;
63 };
64 
65 struct pcpu_dstats {
66 	u64			tx_pkts;
67 	u64			tx_bytes;
68 	u64			tx_drps;
69 	u64			rx_pkts;
70 	u64			rx_bytes;
71 	struct u64_stats_sync	syncp;
72 };
73 
74 static struct dst_entry *vrf_ip_check(struct dst_entry *dst, u32 cookie)
75 {
76 	return dst;
77 }
78 
79 static int vrf_ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
80 {
81 	return ip_local_out(net, sk, skb);
82 }
83 
84 static unsigned int vrf_v4_mtu(const struct dst_entry *dst)
85 {
86 	/* TO-DO: return max ethernet size? */
87 	return dst->dev->mtu;
88 }
89 
90 static void vrf_dst_destroy(struct dst_entry *dst)
91 {
92 	/* our dst lives forever - or until the device is closed */
93 }
94 
95 static unsigned int vrf_default_advmss(const struct dst_entry *dst)
96 {
97 	return 65535 - 40;
98 }
99 
100 static struct dst_ops vrf_dst_ops = {
101 	.family		= AF_INET,
102 	.local_out	= vrf_ip_local_out,
103 	.check		= vrf_ip_check,
104 	.mtu		= vrf_v4_mtu,
105 	.destroy	= vrf_dst_destroy,
106 	.default_advmss	= vrf_default_advmss,
107 };
108 
109 /* neighbor handling is done with actual device; do not want
110  * to flip skb->dev for those ndisc packets. This really fails
111  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
112  * a start.
113  */
114 #if IS_ENABLED(CONFIG_IPV6)
115 static bool check_ipv6_frame(const struct sk_buff *skb)
116 {
117 	const struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb->data;
118 	size_t hlen = sizeof(*ipv6h);
119 	bool rc = true;
120 
121 	if (skb->len < hlen)
122 		goto out;
123 
124 	if (ipv6h->nexthdr == NEXTHDR_ICMP) {
125 		const struct icmp6hdr *icmph;
126 
127 		if (skb->len < hlen + sizeof(*icmph))
128 			goto out;
129 
130 		icmph = (struct icmp6hdr *)(skb->data + sizeof(*ipv6h));
131 		switch (icmph->icmp6_type) {
132 		case NDISC_ROUTER_SOLICITATION:
133 		case NDISC_ROUTER_ADVERTISEMENT:
134 		case NDISC_NEIGHBOUR_SOLICITATION:
135 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
136 		case NDISC_REDIRECT:
137 			rc = false;
138 			break;
139 		}
140 	}
141 
142 out:
143 	return rc;
144 }
145 #else
146 static bool check_ipv6_frame(const struct sk_buff *skb)
147 {
148 	return false;
149 }
150 #endif
151 
152 static bool is_ip_rx_frame(struct sk_buff *skb)
153 {
154 	switch (skb->protocol) {
155 	case htons(ETH_P_IP):
156 		return true;
157 	case htons(ETH_P_IPV6):
158 		return check_ipv6_frame(skb);
159 	}
160 	return false;
161 }
162 
163 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
164 {
165 	vrf_dev->stats.tx_errors++;
166 	kfree_skb(skb);
167 }
168 
169 /* note: already called with rcu_read_lock */
170 static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb)
171 {
172 	struct sk_buff *skb = *pskb;
173 
174 	if (is_ip_rx_frame(skb)) {
175 		struct net_device *dev = vrf_master_get_rcu(skb->dev);
176 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
177 
178 		u64_stats_update_begin(&dstats->syncp);
179 		dstats->rx_pkts++;
180 		dstats->rx_bytes += skb->len;
181 		u64_stats_update_end(&dstats->syncp);
182 
183 		skb->dev = dev;
184 
185 		return RX_HANDLER_ANOTHER;
186 	}
187 	return RX_HANDLER_PASS;
188 }
189 
190 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
191 						 struct rtnl_link_stats64 *stats)
192 {
193 	int i;
194 
195 	for_each_possible_cpu(i) {
196 		const struct pcpu_dstats *dstats;
197 		u64 tbytes, tpkts, tdrops, rbytes, rpkts;
198 		unsigned int start;
199 
200 		dstats = per_cpu_ptr(dev->dstats, i);
201 		do {
202 			start = u64_stats_fetch_begin_irq(&dstats->syncp);
203 			tbytes = dstats->tx_bytes;
204 			tpkts = dstats->tx_pkts;
205 			tdrops = dstats->tx_drps;
206 			rbytes = dstats->rx_bytes;
207 			rpkts = dstats->rx_pkts;
208 		} while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
209 		stats->tx_bytes += tbytes;
210 		stats->tx_packets += tpkts;
211 		stats->tx_dropped += tdrops;
212 		stats->rx_bytes += rbytes;
213 		stats->rx_packets += rpkts;
214 	}
215 	return stats;
216 }
217 
218 #if IS_ENABLED(CONFIG_IPV6)
219 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
220 					   struct net_device *dev)
221 {
222 	const struct ipv6hdr *iph = ipv6_hdr(skb);
223 	struct net *net = dev_net(skb->dev);
224 	struct flowi6 fl6 = {
225 		/* needed to match OIF rule */
226 		.flowi6_oif = dev->ifindex,
227 		.flowi6_iif = LOOPBACK_IFINDEX,
228 		.daddr = iph->daddr,
229 		.saddr = iph->saddr,
230 		.flowlabel = ip6_flowinfo(iph),
231 		.flowi6_mark = skb->mark,
232 		.flowi6_proto = iph->nexthdr,
233 		.flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF,
234 	};
235 	int ret = NET_XMIT_DROP;
236 	struct dst_entry *dst;
237 	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
238 
239 	dst = ip6_route_output(net, NULL, &fl6);
240 	if (dst == dst_null)
241 		goto err;
242 
243 	skb_dst_drop(skb);
244 	skb_dst_set(skb, dst);
245 
246 	ret = ip6_local_out(net, skb->sk, skb);
247 	if (unlikely(net_xmit_eval(ret)))
248 		dev->stats.tx_errors++;
249 	else
250 		ret = NET_XMIT_SUCCESS;
251 
252 	return ret;
253 err:
254 	vrf_tx_error(dev, skb);
255 	return NET_XMIT_DROP;
256 }
257 #else
258 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
259 					   struct net_device *dev)
260 {
261 	vrf_tx_error(dev, skb);
262 	return NET_XMIT_DROP;
263 }
264 #endif
265 
266 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
267 			    struct net_device *vrf_dev)
268 {
269 	struct rtable *rt;
270 	int err = 1;
271 
272 	rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
273 	if (IS_ERR(rt))
274 		goto out;
275 
276 	/* TO-DO: what about broadcast ? */
277 	if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
278 		ip_rt_put(rt);
279 		goto out;
280 	}
281 
282 	skb_dst_drop(skb);
283 	skb_dst_set(skb, &rt->dst);
284 	err = 0;
285 out:
286 	return err;
287 }
288 
289 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
290 					   struct net_device *vrf_dev)
291 {
292 	struct iphdr *ip4h = ip_hdr(skb);
293 	int ret = NET_XMIT_DROP;
294 	struct flowi4 fl4 = {
295 		/* needed to match OIF rule */
296 		.flowi4_oif = vrf_dev->ifindex,
297 		.flowi4_iif = LOOPBACK_IFINDEX,
298 		.flowi4_tos = RT_TOS(ip4h->tos),
299 		.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
300 				FLOWI_FLAG_SKIP_NH_OIF,
301 		.daddr = ip4h->daddr,
302 	};
303 
304 	if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
305 		goto err;
306 
307 	if (!ip4h->saddr) {
308 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
309 					       RT_SCOPE_LINK);
310 	}
311 
312 	ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
313 	if (unlikely(net_xmit_eval(ret)))
314 		vrf_dev->stats.tx_errors++;
315 	else
316 		ret = NET_XMIT_SUCCESS;
317 
318 out:
319 	return ret;
320 err:
321 	vrf_tx_error(vrf_dev, skb);
322 	goto out;
323 }
324 
325 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
326 {
327 	/* strip the ethernet header added for pass through VRF device */
328 	__skb_pull(skb, skb_network_offset(skb));
329 
330 	switch (skb->protocol) {
331 	case htons(ETH_P_IP):
332 		return vrf_process_v4_outbound(skb, dev);
333 	case htons(ETH_P_IPV6):
334 		return vrf_process_v6_outbound(skb, dev);
335 	default:
336 		vrf_tx_error(dev, skb);
337 		return NET_XMIT_DROP;
338 	}
339 }
340 
341 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
342 {
343 	netdev_tx_t ret = is_ip_tx_frame(skb, dev);
344 
345 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
346 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
347 
348 		u64_stats_update_begin(&dstats->syncp);
349 		dstats->tx_pkts++;
350 		dstats->tx_bytes += skb->len;
351 		u64_stats_update_end(&dstats->syncp);
352 	} else {
353 		this_cpu_inc(dev->dstats->tx_drps);
354 	}
355 
356 	return ret;
357 }
358 
359 #if IS_ENABLED(CONFIG_IPV6)
360 static struct dst_entry *vrf_ip6_check(struct dst_entry *dst, u32 cookie)
361 {
362 	return dst;
363 }
364 
365 static struct dst_ops vrf_dst_ops6 = {
366 	.family		= AF_INET6,
367 	.local_out	= ip6_local_out,
368 	.check		= vrf_ip6_check,
369 	.mtu		= vrf_v4_mtu,
370 	.destroy	= vrf_dst_destroy,
371 	.default_advmss	= vrf_default_advmss,
372 };
373 
374 static int init_dst_ops6_kmem_cachep(void)
375 {
376 	vrf_dst_ops6.kmem_cachep = kmem_cache_create("vrf_ip6_dst_cache",
377 						     sizeof(struct rt6_info),
378 						     0,
379 						     SLAB_HWCACHE_ALIGN,
380 						     NULL);
381 
382 	if (!vrf_dst_ops6.kmem_cachep)
383 		return -ENOMEM;
384 
385 	return 0;
386 }
387 
388 static void free_dst_ops6_kmem_cachep(void)
389 {
390 	kmem_cache_destroy(vrf_dst_ops6.kmem_cachep);
391 }
392 
393 static int vrf_input6(struct sk_buff *skb)
394 {
395 	skb->dev->stats.rx_errors++;
396 	kfree_skb(skb);
397 	return 0;
398 }
399 
400 /* modelled after ip6_finish_output2 */
401 static int vrf_finish_output6(struct net *net, struct sock *sk,
402 			      struct sk_buff *skb)
403 {
404 	struct dst_entry *dst = skb_dst(skb);
405 	struct net_device *dev = dst->dev;
406 	struct neighbour *neigh;
407 	struct in6_addr *nexthop;
408 	int ret;
409 
410 	skb->protocol = htons(ETH_P_IPV6);
411 	skb->dev = dev;
412 
413 	rcu_read_lock_bh();
414 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
415 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
416 	if (unlikely(!neigh))
417 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
418 	if (!IS_ERR(neigh)) {
419 		ret = dst_neigh_output(dst, neigh, skb);
420 		rcu_read_unlock_bh();
421 		return ret;
422 	}
423 	rcu_read_unlock_bh();
424 
425 	IP6_INC_STATS(dev_net(dst->dev),
426 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
427 	kfree_skb(skb);
428 	return -EINVAL;
429 }
430 
431 /* modelled after ip6_output */
432 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
433 {
434 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
435 			    net, sk, skb, NULL, skb_dst(skb)->dev,
436 			    vrf_finish_output6,
437 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
438 }
439 
440 static void vrf_rt6_destroy(struct net_vrf *vrf)
441 {
442 	dst_destroy(&vrf->rt6->dst);
443 	free_percpu(vrf->rt6->rt6i_pcpu);
444 	vrf->rt6 = NULL;
445 }
446 
447 static int vrf_rt6_create(struct net_device *dev)
448 {
449 	struct net_vrf *vrf = netdev_priv(dev);
450 	struct dst_entry *dst;
451 	struct rt6_info *rt6;
452 	int cpu;
453 	int rc = -ENOMEM;
454 
455 	rt6 = dst_alloc(&vrf_dst_ops6, dev, 0,
456 			DST_OBSOLETE_NONE,
457 			(DST_HOST | DST_NOPOLICY | DST_NOXFRM));
458 	if (!rt6)
459 		goto out;
460 
461 	dst = &rt6->dst;
462 
463 	rt6->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, GFP_KERNEL);
464 	if (!rt6->rt6i_pcpu) {
465 		dst_destroy(dst);
466 		goto out;
467 	}
468 	for_each_possible_cpu(cpu) {
469 		struct rt6_info **p = per_cpu_ptr(rt6->rt6i_pcpu, cpu);
470 		*p =  NULL;
471 	}
472 
473 	memset(dst + 1, 0, sizeof(*rt6) - sizeof(*dst));
474 
475 	INIT_LIST_HEAD(&rt6->rt6i_siblings);
476 	INIT_LIST_HEAD(&rt6->rt6i_uncached);
477 
478 	rt6->dst.input	= vrf_input6;
479 	rt6->dst.output	= vrf_output6;
480 
481 	rt6->rt6i_table = fib6_get_table(dev_net(dev), vrf->tb_id);
482 
483 	atomic_set(&rt6->dst.__refcnt, 2);
484 
485 	vrf->rt6 = rt6;
486 	rc = 0;
487 out:
488 	return rc;
489 }
490 #else
491 static int init_dst_ops6_kmem_cachep(void)
492 {
493 	return 0;
494 }
495 
496 static void free_dst_ops6_kmem_cachep(void)
497 {
498 }
499 
500 static void vrf_rt6_destroy(struct net_vrf *vrf)
501 {
502 }
503 
504 static int vrf_rt6_create(struct net_device *dev)
505 {
506 	return 0;
507 }
508 #endif
509 
510 /* modelled after ip_finish_output2 */
511 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
512 {
513 	struct dst_entry *dst = skb_dst(skb);
514 	struct rtable *rt = (struct rtable *)dst;
515 	struct net_device *dev = dst->dev;
516 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
517 	struct neighbour *neigh;
518 	u32 nexthop;
519 	int ret = -EINVAL;
520 
521 	/* Be paranoid, rather than too clever. */
522 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
523 		struct sk_buff *skb2;
524 
525 		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
526 		if (!skb2) {
527 			ret = -ENOMEM;
528 			goto err;
529 		}
530 		if (skb->sk)
531 			skb_set_owner_w(skb2, skb->sk);
532 
533 		consume_skb(skb);
534 		skb = skb2;
535 	}
536 
537 	rcu_read_lock_bh();
538 
539 	nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
540 	neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
541 	if (unlikely(!neigh))
542 		neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
543 	if (!IS_ERR(neigh))
544 		ret = dst_neigh_output(dst, neigh, skb);
545 
546 	rcu_read_unlock_bh();
547 err:
548 	if (unlikely(ret < 0))
549 		vrf_tx_error(skb->dev, skb);
550 	return ret;
551 }
552 
553 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
554 {
555 	struct net_device *dev = skb_dst(skb)->dev;
556 
557 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
558 
559 	skb->dev = dev;
560 	skb->protocol = htons(ETH_P_IP);
561 
562 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
563 			    net, sk, skb, NULL, dev,
564 			    vrf_finish_output,
565 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
566 }
567 
568 static void vrf_rtable_destroy(struct net_vrf *vrf)
569 {
570 	struct dst_entry *dst = (struct dst_entry *)vrf->rth;
571 
572 	dst_destroy(dst);
573 	vrf->rth = NULL;
574 }
575 
576 static struct rtable *vrf_rtable_create(struct net_device *dev)
577 {
578 	struct net_vrf *vrf = netdev_priv(dev);
579 	struct rtable *rth;
580 
581 	rth = dst_alloc(&vrf_dst_ops, dev, 2,
582 			DST_OBSOLETE_NONE,
583 			(DST_HOST | DST_NOPOLICY | DST_NOXFRM));
584 	if (rth) {
585 		rth->dst.output	= vrf_output;
586 		rth->rt_genid	= rt_genid_ipv4(dev_net(dev));
587 		rth->rt_flags	= 0;
588 		rth->rt_type	= RTN_UNICAST;
589 		rth->rt_is_input = 0;
590 		rth->rt_iif	= 0;
591 		rth->rt_pmtu	= 0;
592 		rth->rt_gateway	= 0;
593 		rth->rt_uses_gateway = 0;
594 		rth->rt_table_id = vrf->tb_id;
595 		INIT_LIST_HEAD(&rth->rt_uncached);
596 		rth->rt_uncached_list = NULL;
597 	}
598 
599 	return rth;
600 }
601 
602 /**************************** device handling ********************/
603 
604 /* cycle interface to flush neighbor cache and move routes across tables */
605 static void cycle_netdev(struct net_device *dev)
606 {
607 	unsigned int flags = dev->flags;
608 	int ret;
609 
610 	if (!netif_running(dev))
611 		return;
612 
613 	ret = dev_change_flags(dev, flags & ~IFF_UP);
614 	if (ret >= 0)
615 		ret = dev_change_flags(dev, flags);
616 
617 	if (ret < 0) {
618 		netdev_err(dev,
619 			   "Failed to cycle device %s; route tables might be wrong!\n",
620 			   dev->name);
621 	}
622 }
623 
624 static struct slave *__vrf_find_slave_dev(struct slave_queue *queue,
625 					  struct net_device *dev)
626 {
627 	struct list_head *head = &queue->all_slaves;
628 	struct slave *slave;
629 
630 	list_for_each_entry(slave, head, list) {
631 		if (slave->dev == dev)
632 			return slave;
633 	}
634 
635 	return NULL;
636 }
637 
638 /* inverse of __vrf_insert_slave */
639 static void __vrf_remove_slave(struct slave_queue *queue, struct slave *slave)
640 {
641 	list_del(&slave->list);
642 }
643 
644 static void __vrf_insert_slave(struct slave_queue *queue, struct slave *slave)
645 {
646 	list_add(&slave->list, &queue->all_slaves);
647 }
648 
649 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
650 {
651 	struct slave *slave = kzalloc(sizeof(*slave), GFP_KERNEL);
652 	struct net_vrf *vrf = netdev_priv(dev);
653 	struct slave_queue *queue = &vrf->queue;
654 	int ret = -ENOMEM;
655 
656 	if (!slave)
657 		goto out_fail;
658 
659 	slave->dev = port_dev;
660 
661 	/* register the packet handler for slave ports */
662 	ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev);
663 	if (ret) {
664 		netdev_err(port_dev,
665 			   "Device %s failed to register rx_handler\n",
666 			   port_dev->name);
667 		goto out_fail;
668 	}
669 
670 	ret = netdev_master_upper_dev_link(port_dev, dev);
671 	if (ret < 0)
672 		goto out_unregister;
673 
674 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
675 	__vrf_insert_slave(queue, slave);
676 	cycle_netdev(port_dev);
677 
678 	return 0;
679 
680 out_unregister:
681 	netdev_rx_handler_unregister(port_dev);
682 out_fail:
683 	kfree(slave);
684 	return ret;
685 }
686 
687 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
688 {
689 	if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
690 		return -EINVAL;
691 
692 	return do_vrf_add_slave(dev, port_dev);
693 }
694 
695 /* inverse of do_vrf_add_slave */
696 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
697 {
698 	struct net_vrf *vrf = netdev_priv(dev);
699 	struct slave_queue *queue = &vrf->queue;
700 	struct slave *slave;
701 
702 	netdev_upper_dev_unlink(port_dev, dev);
703 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
704 
705 	netdev_rx_handler_unregister(port_dev);
706 
707 	cycle_netdev(port_dev);
708 
709 	slave = __vrf_find_slave_dev(queue, port_dev);
710 	if (slave)
711 		__vrf_remove_slave(queue, slave);
712 
713 	kfree(slave);
714 
715 	return 0;
716 }
717 
718 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
719 {
720 	return do_vrf_del_slave(dev, port_dev);
721 }
722 
723 static void vrf_dev_uninit(struct net_device *dev)
724 {
725 	struct net_vrf *vrf = netdev_priv(dev);
726 	struct slave_queue *queue = &vrf->queue;
727 	struct list_head *head = &queue->all_slaves;
728 	struct slave *slave, *next;
729 
730 	vrf_rtable_destroy(vrf);
731 	vrf_rt6_destroy(vrf);
732 
733 	list_for_each_entry_safe(slave, next, head, list)
734 		vrf_del_slave(dev, slave->dev);
735 
736 	free_percpu(dev->dstats);
737 	dev->dstats = NULL;
738 }
739 
740 static int vrf_dev_init(struct net_device *dev)
741 {
742 	struct net_vrf *vrf = netdev_priv(dev);
743 
744 	INIT_LIST_HEAD(&vrf->queue.all_slaves);
745 
746 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
747 	if (!dev->dstats)
748 		goto out_nomem;
749 
750 	/* create the default dst which points back to us */
751 	vrf->rth = vrf_rtable_create(dev);
752 	if (!vrf->rth)
753 		goto out_stats;
754 
755 	if (vrf_rt6_create(dev) != 0)
756 		goto out_rth;
757 
758 	dev->flags = IFF_MASTER | IFF_NOARP;
759 
760 	return 0;
761 
762 out_rth:
763 	vrf_rtable_destroy(vrf);
764 out_stats:
765 	free_percpu(dev->dstats);
766 	dev->dstats = NULL;
767 out_nomem:
768 	return -ENOMEM;
769 }
770 
771 static const struct net_device_ops vrf_netdev_ops = {
772 	.ndo_init		= vrf_dev_init,
773 	.ndo_uninit		= vrf_dev_uninit,
774 	.ndo_start_xmit		= vrf_xmit,
775 	.ndo_get_stats64	= vrf_get_stats64,
776 	.ndo_add_slave		= vrf_add_slave,
777 	.ndo_del_slave		= vrf_del_slave,
778 };
779 
780 static u32 vrf_fib_table(const struct net_device *dev)
781 {
782 	struct net_vrf *vrf = netdev_priv(dev);
783 
784 	return vrf->tb_id;
785 }
786 
787 static struct rtable *vrf_get_rtable(const struct net_device *dev,
788 				     const struct flowi4 *fl4)
789 {
790 	struct rtable *rth = NULL;
791 
792 	if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
793 		struct net_vrf *vrf = netdev_priv(dev);
794 
795 		rth = vrf->rth;
796 		atomic_inc(&rth->dst.__refcnt);
797 	}
798 
799 	return rth;
800 }
801 
802 /* called under rcu_read_lock */
803 static void vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4)
804 {
805 	struct fib_result res = { .tclassid = 0 };
806 	struct net *net = dev_net(dev);
807 	u32 orig_tos = fl4->flowi4_tos;
808 	u8 flags = fl4->flowi4_flags;
809 	u8 scope = fl4->flowi4_scope;
810 	u8 tos = RT_FL_TOS(fl4);
811 
812 	if (unlikely(!fl4->daddr))
813 		return;
814 
815 	fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF;
816 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
817 	fl4->flowi4_tos = tos & IPTOS_RT_MASK;
818 	fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
819 			     RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
820 
821 	if (!fib_lookup(net, fl4, &res, 0)) {
822 		if (res.type == RTN_LOCAL)
823 			fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr;
824 		else
825 			fib_select_path(net, &res, fl4, -1);
826 	}
827 
828 	fl4->flowi4_flags = flags;
829 	fl4->flowi4_tos = orig_tos;
830 	fl4->flowi4_scope = scope;
831 }
832 
833 #if IS_ENABLED(CONFIG_IPV6)
834 static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
835 					 const struct flowi6 *fl6)
836 {
837 	struct rt6_info *rt = NULL;
838 
839 	if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) {
840 		struct net_vrf *vrf = netdev_priv(dev);
841 
842 		rt = vrf->rt6;
843 		atomic_inc(&rt->dst.__refcnt);
844 	}
845 
846 	return (struct dst_entry *)rt;
847 }
848 #endif
849 
850 static const struct l3mdev_ops vrf_l3mdev_ops = {
851 	.l3mdev_fib_table	= vrf_fib_table,
852 	.l3mdev_get_rtable	= vrf_get_rtable,
853 	.l3mdev_get_saddr	= vrf_get_saddr,
854 #if IS_ENABLED(CONFIG_IPV6)
855 	.l3mdev_get_rt6_dst	= vrf_get_rt6_dst,
856 #endif
857 };
858 
859 static void vrf_get_drvinfo(struct net_device *dev,
860 			    struct ethtool_drvinfo *info)
861 {
862 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
863 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
864 }
865 
866 static const struct ethtool_ops vrf_ethtool_ops = {
867 	.get_drvinfo	= vrf_get_drvinfo,
868 };
869 
870 static void vrf_setup(struct net_device *dev)
871 {
872 	ether_setup(dev);
873 
874 	/* Initialize the device structure. */
875 	dev->netdev_ops = &vrf_netdev_ops;
876 	dev->l3mdev_ops = &vrf_l3mdev_ops;
877 	dev->ethtool_ops = &vrf_ethtool_ops;
878 	dev->destructor = free_netdev;
879 
880 	/* Fill in device structure with ethernet-generic values. */
881 	eth_hw_addr_random(dev);
882 
883 	/* don't acquire vrf device's netif_tx_lock when transmitting */
884 	dev->features |= NETIF_F_LLTX;
885 
886 	/* don't allow vrf devices to change network namespaces. */
887 	dev->features |= NETIF_F_NETNS_LOCAL;
888 }
889 
890 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
891 {
892 	if (tb[IFLA_ADDRESS]) {
893 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
894 			return -EINVAL;
895 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
896 			return -EADDRNOTAVAIL;
897 	}
898 	return 0;
899 }
900 
901 static void vrf_dellink(struct net_device *dev, struct list_head *head)
902 {
903 	unregister_netdevice_queue(dev, head);
904 }
905 
906 static int vrf_newlink(struct net *src_net, struct net_device *dev,
907 		       struct nlattr *tb[], struct nlattr *data[])
908 {
909 	struct net_vrf *vrf = netdev_priv(dev);
910 	int err;
911 
912 	if (!data || !data[IFLA_VRF_TABLE])
913 		return -EINVAL;
914 
915 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
916 
917 	dev->priv_flags |= IFF_L3MDEV_MASTER;
918 
919 	err = register_netdevice(dev);
920 	if (err < 0)
921 		goto out_fail;
922 
923 	return 0;
924 
925 out_fail:
926 	free_netdev(dev);
927 	return err;
928 }
929 
930 static size_t vrf_nl_getsize(const struct net_device *dev)
931 {
932 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
933 }
934 
935 static int vrf_fillinfo(struct sk_buff *skb,
936 			const struct net_device *dev)
937 {
938 	struct net_vrf *vrf = netdev_priv(dev);
939 
940 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
941 }
942 
943 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
944 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
945 };
946 
947 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
948 	.kind		= DRV_NAME,
949 	.priv_size	= sizeof(struct net_vrf),
950 
951 	.get_size	= vrf_nl_getsize,
952 	.policy		= vrf_nl_policy,
953 	.validate	= vrf_validate,
954 	.fill_info	= vrf_fillinfo,
955 
956 	.newlink	= vrf_newlink,
957 	.dellink	= vrf_dellink,
958 	.setup		= vrf_setup,
959 	.maxtype	= IFLA_VRF_MAX,
960 };
961 
962 static int vrf_device_event(struct notifier_block *unused,
963 			    unsigned long event, void *ptr)
964 {
965 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
966 
967 	/* only care about unregister events to drop slave references */
968 	if (event == NETDEV_UNREGISTER) {
969 		struct net_device *vrf_dev;
970 
971 		if (!netif_is_l3_slave(dev))
972 			goto out;
973 
974 		vrf_dev = netdev_master_upper_dev_get(dev);
975 		vrf_del_slave(vrf_dev, dev);
976 	}
977 out:
978 	return NOTIFY_DONE;
979 }
980 
981 static struct notifier_block vrf_notifier_block __read_mostly = {
982 	.notifier_call = vrf_device_event,
983 };
984 
985 static int __init vrf_init_module(void)
986 {
987 	int rc;
988 
989 	vrf_dst_ops.kmem_cachep =
990 		kmem_cache_create("vrf_ip_dst_cache",
991 				  sizeof(struct rtable), 0,
992 				  SLAB_HWCACHE_ALIGN,
993 				  NULL);
994 
995 	if (!vrf_dst_ops.kmem_cachep)
996 		return -ENOMEM;
997 
998 	rc = init_dst_ops6_kmem_cachep();
999 	if (rc != 0)
1000 		goto error2;
1001 
1002 	register_netdevice_notifier(&vrf_notifier_block);
1003 
1004 	rc = rtnl_link_register(&vrf_link_ops);
1005 	if (rc < 0)
1006 		goto error;
1007 
1008 	return 0;
1009 
1010 error:
1011 	unregister_netdevice_notifier(&vrf_notifier_block);
1012 	free_dst_ops6_kmem_cachep();
1013 error2:
1014 	kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
1015 	return rc;
1016 }
1017 
1018 static void __exit vrf_cleanup_module(void)
1019 {
1020 	rtnl_link_unregister(&vrf_link_ops);
1021 	unregister_netdevice_notifier(&vrf_notifier_block);
1022 	kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
1023 	free_dst_ops6_kmem_cachep();
1024 }
1025 
1026 module_init(vrf_init_module);
1027 module_exit(vrf_cleanup_module);
1028 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1029 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1030 MODULE_LICENSE("GPL");
1031 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1032 MODULE_VERSION(DRV_VERSION);
1033