Loading...
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
46static unsigned int vrf_net_id;
47
48struct net_vrf {
49 struct rtable __rcu *rth;
50 struct rt6_info __rcu *rt6;
51 u32 tb_id;
52};
53
54struct pcpu_dstats {
55 u64 tx_pkts;
56 u64 tx_bytes;
57 u64 tx_drps;
58 u64 rx_pkts;
59 u64 rx_bytes;
60 u64 rx_drps;
61 struct u64_stats_sync syncp;
62};
63
64static void vrf_rx_stats(struct net_device *dev, int len)
65{
66 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
67
68 u64_stats_update_begin(&dstats->syncp);
69 dstats->rx_pkts++;
70 dstats->rx_bytes += len;
71 u64_stats_update_end(&dstats->syncp);
72}
73
74static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
75{
76 vrf_dev->stats.tx_errors++;
77 kfree_skb(skb);
78}
79
80static void vrf_get_stats64(struct net_device *dev,
81 struct rtnl_link_stats64 *stats)
82{
83 int i;
84
85 for_each_possible_cpu(i) {
86 const struct pcpu_dstats *dstats;
87 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
88 unsigned int start;
89
90 dstats = per_cpu_ptr(dev->dstats, i);
91 do {
92 start = u64_stats_fetch_begin_irq(&dstats->syncp);
93 tbytes = dstats->tx_bytes;
94 tpkts = dstats->tx_pkts;
95 tdrops = dstats->tx_drps;
96 rbytes = dstats->rx_bytes;
97 rpkts = dstats->rx_pkts;
98 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
99 stats->tx_bytes += tbytes;
100 stats->tx_packets += tpkts;
101 stats->tx_dropped += tdrops;
102 stats->rx_bytes += rbytes;
103 stats->rx_packets += rpkts;
104 }
105}
106
107/* by default VRF devices do not have a qdisc and are expected
108 * to be created with only a single queue.
109 */
110static bool qdisc_tx_is_default(const struct net_device *dev)
111{
112 struct netdev_queue *txq;
113 struct Qdisc *qdisc;
114
115 if (dev->num_tx_queues > 1)
116 return false;
117
118 txq = netdev_get_tx_queue(dev, 0);
119 qdisc = rcu_access_pointer(txq->qdisc);
120
121 return !qdisc->enqueue;
122}
123
124/* Local traffic destined to local address. Reinsert the packet to rx
125 * path, similar to loopback handling.
126 */
127static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
128 struct dst_entry *dst)
129{
130 int len = skb->len;
131
132 skb_orphan(skb);
133
134 skb_dst_set(skb, dst);
135
136 /* set pkt_type to avoid skb hitting packet taps twice -
137 * once on Tx and again in Rx processing
138 */
139 skb->pkt_type = PACKET_LOOPBACK;
140
141 skb->protocol = eth_type_trans(skb, dev);
142
143 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
144 vrf_rx_stats(dev, len);
145 else
146 this_cpu_inc(dev->dstats->rx_drps);
147
148 return NETDEV_TX_OK;
149}
150
151#if IS_ENABLED(CONFIG_IPV6)
152static int vrf_ip6_local_out(struct net *net, struct sock *sk,
153 struct sk_buff *skb)
154{
155 int err;
156
157 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
158 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
159
160 if (likely(err == 1))
161 err = dst_output(net, sk, skb);
162
163 return err;
164}
165
166static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
167 struct net_device *dev)
168{
169 const struct ipv6hdr *iph = ipv6_hdr(skb);
170 struct net *net = dev_net(skb->dev);
171 struct flowi6 fl6 = {
172 /* needed to match OIF rule */
173 .flowi6_oif = dev->ifindex,
174 .flowi6_iif = LOOPBACK_IFINDEX,
175 .daddr = iph->daddr,
176 .saddr = iph->saddr,
177 .flowlabel = ip6_flowinfo(iph),
178 .flowi6_mark = skb->mark,
179 .flowi6_proto = iph->nexthdr,
180 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
181 };
182 int ret = NET_XMIT_DROP;
183 struct dst_entry *dst;
184 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
185
186 dst = ip6_route_output(net, NULL, &fl6);
187 if (dst == dst_null)
188 goto err;
189
190 skb_dst_drop(skb);
191
192 /* if dst.dev is loopback or the VRF device again this is locally
193 * originated traffic destined to a local address. Short circuit
194 * to Rx path
195 */
196 if (dst->dev == dev)
197 return vrf_local_xmit(skb, dev, dst);
198
199 skb_dst_set(skb, dst);
200
201 /* strip the ethernet header added for pass through VRF device */
202 __skb_pull(skb, skb_network_offset(skb));
203
204 ret = vrf_ip6_local_out(net, skb->sk, skb);
205 if (unlikely(net_xmit_eval(ret)))
206 dev->stats.tx_errors++;
207 else
208 ret = NET_XMIT_SUCCESS;
209
210 return ret;
211err:
212 vrf_tx_error(dev, skb);
213 return NET_XMIT_DROP;
214}
215#else
216static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
217 struct net_device *dev)
218{
219 vrf_tx_error(dev, skb);
220 return NET_XMIT_DROP;
221}
222#endif
223
224/* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
225static int vrf_ip_local_out(struct net *net, struct sock *sk,
226 struct sk_buff *skb)
227{
228 int err;
229
230 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
231 skb, NULL, skb_dst(skb)->dev, dst_output);
232 if (likely(err == 1))
233 err = dst_output(net, sk, skb);
234
235 return err;
236}
237
238static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
239 struct net_device *vrf_dev)
240{
241 struct iphdr *ip4h = ip_hdr(skb);
242 int ret = NET_XMIT_DROP;
243 struct flowi4 fl4 = {
244 /* needed to match OIF rule */
245 .flowi4_oif = vrf_dev->ifindex,
246 .flowi4_iif = LOOPBACK_IFINDEX,
247 .flowi4_tos = RT_TOS(ip4h->tos),
248 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
249 .flowi4_proto = ip4h->protocol,
250 .daddr = ip4h->daddr,
251 .saddr = ip4h->saddr,
252 };
253 struct net *net = dev_net(vrf_dev);
254 struct rtable *rt;
255
256 rt = ip_route_output_flow(net, &fl4, NULL);
257 if (IS_ERR(rt))
258 goto err;
259
260 skb_dst_drop(skb);
261
262 /* if dst.dev is loopback or the VRF device again this is locally
263 * originated traffic destined to a local address. Short circuit
264 * to Rx path
265 */
266 if (rt->dst.dev == vrf_dev)
267 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
268
269 skb_dst_set(skb, &rt->dst);
270
271 /* strip the ethernet header added for pass through VRF device */
272 __skb_pull(skb, skb_network_offset(skb));
273
274 if (!ip4h->saddr) {
275 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
276 RT_SCOPE_LINK);
277 }
278
279 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
280 if (unlikely(net_xmit_eval(ret)))
281 vrf_dev->stats.tx_errors++;
282 else
283 ret = NET_XMIT_SUCCESS;
284
285out:
286 return ret;
287err:
288 vrf_tx_error(vrf_dev, skb);
289 goto out;
290}
291
292static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
293{
294 switch (skb->protocol) {
295 case htons(ETH_P_IP):
296 return vrf_process_v4_outbound(skb, dev);
297 case htons(ETH_P_IPV6):
298 return vrf_process_v6_outbound(skb, dev);
299 default:
300 vrf_tx_error(dev, skb);
301 return NET_XMIT_DROP;
302 }
303}
304
305static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
306{
307 int len = skb->len;
308 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
309
310 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
311 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
312
313 u64_stats_update_begin(&dstats->syncp);
314 dstats->tx_pkts++;
315 dstats->tx_bytes += len;
316 u64_stats_update_end(&dstats->syncp);
317 } else {
318 this_cpu_inc(dev->dstats->tx_drps);
319 }
320
321 return ret;
322}
323
324static int vrf_finish_direct(struct net *net, struct sock *sk,
325 struct sk_buff *skb)
326{
327 struct net_device *vrf_dev = skb->dev;
328
329 if (!list_empty(&vrf_dev->ptype_all) &&
330 likely(skb_headroom(skb) >= ETH_HLEN)) {
331 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
332
333 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
334 eth_zero_addr(eth->h_dest);
335 eth->h_proto = skb->protocol;
336
337 rcu_read_lock_bh();
338 dev_queue_xmit_nit(skb, vrf_dev);
339 rcu_read_unlock_bh();
340
341 skb_pull(skb, ETH_HLEN);
342 }
343
344 return 1;
345}
346
347#if IS_ENABLED(CONFIG_IPV6)
348/* modelled after ip6_finish_output2 */
349static int vrf_finish_output6(struct net *net, struct sock *sk,
350 struct sk_buff *skb)
351{
352 struct dst_entry *dst = skb_dst(skb);
353 struct net_device *dev = dst->dev;
354 struct neighbour *neigh;
355 struct in6_addr *nexthop;
356 int ret;
357
358 nf_reset(skb);
359
360 skb->protocol = htons(ETH_P_IPV6);
361 skb->dev = dev;
362
363 rcu_read_lock_bh();
364 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
365 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
366 if (unlikely(!neigh))
367 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
368 if (!IS_ERR(neigh)) {
369 sock_confirm_neigh(skb, neigh);
370 ret = neigh_output(neigh, skb);
371 rcu_read_unlock_bh();
372 return ret;
373 }
374 rcu_read_unlock_bh();
375
376 IP6_INC_STATS(dev_net(dst->dev),
377 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
378 kfree_skb(skb);
379 return -EINVAL;
380}
381
382/* modelled after ip6_output */
383static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
384{
385 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
386 net, sk, skb, NULL, skb_dst(skb)->dev,
387 vrf_finish_output6,
388 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
389}
390
391/* set dst on skb to send packet to us via dev_xmit path. Allows
392 * packet to go through device based features such as qdisc, netfilter
393 * hooks and packet sockets with skb->dev set to vrf device.
394 */
395static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
396 struct sk_buff *skb)
397{
398 struct net_vrf *vrf = netdev_priv(vrf_dev);
399 struct dst_entry *dst = NULL;
400 struct rt6_info *rt6;
401
402 rcu_read_lock();
403
404 rt6 = rcu_dereference(vrf->rt6);
405 if (likely(rt6)) {
406 dst = &rt6->dst;
407 dst_hold(dst);
408 }
409
410 rcu_read_unlock();
411
412 if (unlikely(!dst)) {
413 vrf_tx_error(vrf_dev, skb);
414 return NULL;
415 }
416
417 skb_dst_drop(skb);
418 skb_dst_set(skb, dst);
419
420 return skb;
421}
422
423static int vrf_output6_direct(struct net *net, struct sock *sk,
424 struct sk_buff *skb)
425{
426 skb->protocol = htons(ETH_P_IPV6);
427
428 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
429 net, sk, skb, NULL, skb->dev,
430 vrf_finish_direct,
431 !(IPCB(skb)->flags & IPSKB_REROUTED));
432}
433
434static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
435 struct sock *sk,
436 struct sk_buff *skb)
437{
438 struct net *net = dev_net(vrf_dev);
439 int err;
440
441 skb->dev = vrf_dev;
442
443 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
444 skb, NULL, vrf_dev, vrf_output6_direct);
445
446 if (likely(err == 1))
447 err = vrf_output6_direct(net, sk, skb);
448
449 /* reset skb device */
450 if (likely(err == 1))
451 nf_reset(skb);
452 else
453 skb = NULL;
454
455 return skb;
456}
457
458static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
459 struct sock *sk,
460 struct sk_buff *skb)
461{
462 /* don't divert link scope packets */
463 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
464 return skb;
465
466 if (qdisc_tx_is_default(vrf_dev))
467 return vrf_ip6_out_direct(vrf_dev, sk, skb);
468
469 return vrf_ip6_out_redirect(vrf_dev, skb);
470}
471
472/* holding rtnl */
473static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
474{
475 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
476 struct net *net = dev_net(dev);
477 struct dst_entry *dst;
478
479 RCU_INIT_POINTER(vrf->rt6, NULL);
480 synchronize_rcu();
481
482 /* move dev in dst's to loopback so this VRF device can be deleted
483 * - based on dst_ifdown
484 */
485 if (rt6) {
486 dst = &rt6->dst;
487 dev_put(dst->dev);
488 dst->dev = net->loopback_dev;
489 dev_hold(dst->dev);
490 dst_release(dst);
491 }
492}
493
494static int vrf_rt6_create(struct net_device *dev)
495{
496 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
497 struct net_vrf *vrf = netdev_priv(dev);
498 struct net *net = dev_net(dev);
499 struct fib6_table *rt6i_table;
500 struct rt6_info *rt6;
501 int rc = -ENOMEM;
502
503 /* IPv6 can be CONFIG enabled and then disabled runtime */
504 if (!ipv6_mod_enabled())
505 return 0;
506
507 rt6i_table = fib6_new_table(net, vrf->tb_id);
508 if (!rt6i_table)
509 goto out;
510
511 /* create a dst for routing packets out a VRF device */
512 rt6 = ip6_dst_alloc(net, dev, flags);
513 if (!rt6)
514 goto out;
515
516 rt6->rt6i_table = rt6i_table;
517 rt6->dst.output = vrf_output6;
518
519 rcu_assign_pointer(vrf->rt6, rt6);
520
521 rc = 0;
522out:
523 return rc;
524}
525#else
526static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
527 struct sock *sk,
528 struct sk_buff *skb)
529{
530 return skb;
531}
532
533static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
534{
535}
536
537static int vrf_rt6_create(struct net_device *dev)
538{
539 return 0;
540}
541#endif
542
543/* modelled after ip_finish_output2 */
544static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
545{
546 struct dst_entry *dst = skb_dst(skb);
547 struct rtable *rt = (struct rtable *)dst;
548 struct net_device *dev = dst->dev;
549 unsigned int hh_len = LL_RESERVED_SPACE(dev);
550 struct neighbour *neigh;
551 u32 nexthop;
552 int ret = -EINVAL;
553
554 nf_reset(skb);
555
556 /* Be paranoid, rather than too clever. */
557 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
558 struct sk_buff *skb2;
559
560 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
561 if (!skb2) {
562 ret = -ENOMEM;
563 goto err;
564 }
565 if (skb->sk)
566 skb_set_owner_w(skb2, skb->sk);
567
568 consume_skb(skb);
569 skb = skb2;
570 }
571
572 rcu_read_lock_bh();
573
574 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
575 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
576 if (unlikely(!neigh))
577 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
578 if (!IS_ERR(neigh)) {
579 sock_confirm_neigh(skb, neigh);
580 ret = neigh_output(neigh, skb);
581 rcu_read_unlock_bh();
582 return ret;
583 }
584
585 rcu_read_unlock_bh();
586err:
587 vrf_tx_error(skb->dev, skb);
588 return ret;
589}
590
591static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
592{
593 struct net_device *dev = skb_dst(skb)->dev;
594
595 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
596
597 skb->dev = dev;
598 skb->protocol = htons(ETH_P_IP);
599
600 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
601 net, sk, skb, NULL, dev,
602 vrf_finish_output,
603 !(IPCB(skb)->flags & IPSKB_REROUTED));
604}
605
606/* set dst on skb to send packet to us via dev_xmit path. Allows
607 * packet to go through device based features such as qdisc, netfilter
608 * hooks and packet sockets with skb->dev set to vrf device.
609 */
610static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
611 struct sk_buff *skb)
612{
613 struct net_vrf *vrf = netdev_priv(vrf_dev);
614 struct dst_entry *dst = NULL;
615 struct rtable *rth;
616
617 rcu_read_lock();
618
619 rth = rcu_dereference(vrf->rth);
620 if (likely(rth)) {
621 dst = &rth->dst;
622 dst_hold(dst);
623 }
624
625 rcu_read_unlock();
626
627 if (unlikely(!dst)) {
628 vrf_tx_error(vrf_dev, skb);
629 return NULL;
630 }
631
632 skb_dst_drop(skb);
633 skb_dst_set(skb, dst);
634
635 return skb;
636}
637
638static int vrf_output_direct(struct net *net, struct sock *sk,
639 struct sk_buff *skb)
640{
641 skb->protocol = htons(ETH_P_IP);
642
643 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
644 net, sk, skb, NULL, skb->dev,
645 vrf_finish_direct,
646 !(IPCB(skb)->flags & IPSKB_REROUTED));
647}
648
649static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
650 struct sock *sk,
651 struct sk_buff *skb)
652{
653 struct net *net = dev_net(vrf_dev);
654 int err;
655
656 skb->dev = vrf_dev;
657
658 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
659 skb, NULL, vrf_dev, vrf_output_direct);
660
661 if (likely(err == 1))
662 err = vrf_output_direct(net, sk, skb);
663
664 /* reset skb device */
665 if (likely(err == 1))
666 nf_reset(skb);
667 else
668 skb = NULL;
669
670 return skb;
671}
672
673static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
674 struct sock *sk,
675 struct sk_buff *skb)
676{
677 /* don't divert multicast or local broadcast */
678 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
679 ipv4_is_lbcast(ip_hdr(skb)->daddr))
680 return skb;
681
682 if (qdisc_tx_is_default(vrf_dev))
683 return vrf_ip_out_direct(vrf_dev, sk, skb);
684
685 return vrf_ip_out_redirect(vrf_dev, skb);
686}
687
688/* called with rcu lock held */
689static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
690 struct sock *sk,
691 struct sk_buff *skb,
692 u16 proto)
693{
694 switch (proto) {
695 case AF_INET:
696 return vrf_ip_out(vrf_dev, sk, skb);
697 case AF_INET6:
698 return vrf_ip6_out(vrf_dev, sk, skb);
699 }
700
701 return skb;
702}
703
704/* holding rtnl */
705static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
706{
707 struct rtable *rth = rtnl_dereference(vrf->rth);
708 struct net *net = dev_net(dev);
709 struct dst_entry *dst;
710
711 RCU_INIT_POINTER(vrf->rth, NULL);
712 synchronize_rcu();
713
714 /* move dev in dst's to loopback so this VRF device can be deleted
715 * - based on dst_ifdown
716 */
717 if (rth) {
718 dst = &rth->dst;
719 dev_put(dst->dev);
720 dst->dev = net->loopback_dev;
721 dev_hold(dst->dev);
722 dst_release(dst);
723 }
724}
725
726static int vrf_rtable_create(struct net_device *dev)
727{
728 struct net_vrf *vrf = netdev_priv(dev);
729 struct rtable *rth;
730
731 if (!fib_new_table(dev_net(dev), vrf->tb_id))
732 return -ENOMEM;
733
734 /* create a dst for routing packets out through a VRF device */
735 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
736 if (!rth)
737 return -ENOMEM;
738
739 rth->dst.output = vrf_output;
740
741 rcu_assign_pointer(vrf->rth, rth);
742
743 return 0;
744}
745
746/**************************** device handling ********************/
747
748/* cycle interface to flush neighbor cache and move routes across tables */
749static void cycle_netdev(struct net_device *dev)
750{
751 unsigned int flags = dev->flags;
752 int ret;
753
754 if (!netif_running(dev))
755 return;
756
757 ret = dev_change_flags(dev, flags & ~IFF_UP);
758 if (ret >= 0)
759 ret = dev_change_flags(dev, flags);
760
761 if (ret < 0) {
762 netdev_err(dev,
763 "Failed to cycle device %s; route tables might be wrong!\n",
764 dev->name);
765 }
766}
767
768static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
769 struct netlink_ext_ack *extack)
770{
771 int ret;
772
773 /* do not allow loopback device to be enslaved to a VRF.
774 * The vrf device acts as the loopback for the vrf.
775 */
776 if (port_dev == dev_net(dev)->loopback_dev) {
777 NL_SET_ERR_MSG(extack,
778 "Can not enslave loopback device to a VRF");
779 return -EOPNOTSUPP;
780 }
781
782 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
783 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
784 if (ret < 0)
785 goto err;
786
787 cycle_netdev(port_dev);
788
789 return 0;
790
791err:
792 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
793 return ret;
794}
795
796static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
797 struct netlink_ext_ack *extack)
798{
799 if (netif_is_l3_master(port_dev)) {
800 NL_SET_ERR_MSG(extack,
801 "Can not enslave an L3 master device to a VRF");
802 return -EINVAL;
803 }
804
805 if (netif_is_l3_slave(port_dev))
806 return -EINVAL;
807
808 return do_vrf_add_slave(dev, port_dev, extack);
809}
810
811/* inverse of do_vrf_add_slave */
812static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
813{
814 netdev_upper_dev_unlink(port_dev, dev);
815 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
816
817 cycle_netdev(port_dev);
818
819 return 0;
820}
821
822static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
823{
824 return do_vrf_del_slave(dev, port_dev);
825}
826
827static void vrf_dev_uninit(struct net_device *dev)
828{
829 struct net_vrf *vrf = netdev_priv(dev);
830
831 vrf_rtable_release(dev, vrf);
832 vrf_rt6_release(dev, vrf);
833
834 free_percpu(dev->dstats);
835 dev->dstats = NULL;
836}
837
838static int vrf_dev_init(struct net_device *dev)
839{
840 struct net_vrf *vrf = netdev_priv(dev);
841
842 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
843 if (!dev->dstats)
844 goto out_nomem;
845
846 /* create the default dst which points back to us */
847 if (vrf_rtable_create(dev) != 0)
848 goto out_stats;
849
850 if (vrf_rt6_create(dev) != 0)
851 goto out_rth;
852
853 dev->flags = IFF_MASTER | IFF_NOARP;
854
855 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
856 dev->mtu = 64 * 1024;
857
858 /* similarly, oper state is irrelevant; set to up to avoid confusion */
859 dev->operstate = IF_OPER_UP;
860 netdev_lockdep_set_classes(dev);
861 return 0;
862
863out_rth:
864 vrf_rtable_release(dev, vrf);
865out_stats:
866 free_percpu(dev->dstats);
867 dev->dstats = NULL;
868out_nomem:
869 return -ENOMEM;
870}
871
872static const struct net_device_ops vrf_netdev_ops = {
873 .ndo_init = vrf_dev_init,
874 .ndo_uninit = vrf_dev_uninit,
875 .ndo_start_xmit = vrf_xmit,
876 .ndo_get_stats64 = vrf_get_stats64,
877 .ndo_add_slave = vrf_add_slave,
878 .ndo_del_slave = vrf_del_slave,
879};
880
881static u32 vrf_fib_table(const struct net_device *dev)
882{
883 struct net_vrf *vrf = netdev_priv(dev);
884
885 return vrf->tb_id;
886}
887
888static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
889{
890 kfree_skb(skb);
891 return 0;
892}
893
894static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
895 struct sk_buff *skb,
896 struct net_device *dev)
897{
898 struct net *net = dev_net(dev);
899
900 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
901 skb = NULL; /* kfree_skb(skb) handled by nf code */
902
903 return skb;
904}
905
906#if IS_ENABLED(CONFIG_IPV6)
907/* neighbor handling is done with actual device; do not want
908 * to flip skb->dev for those ndisc packets. This really fails
909 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
910 * a start.
911 */
912static bool ipv6_ndisc_frame(const struct sk_buff *skb)
913{
914 const struct ipv6hdr *iph = ipv6_hdr(skb);
915 bool rc = false;
916
917 if (iph->nexthdr == NEXTHDR_ICMP) {
918 const struct icmp6hdr *icmph;
919 struct icmp6hdr _icmph;
920
921 icmph = skb_header_pointer(skb, sizeof(*iph),
922 sizeof(_icmph), &_icmph);
923 if (!icmph)
924 goto out;
925
926 switch (icmph->icmp6_type) {
927 case NDISC_ROUTER_SOLICITATION:
928 case NDISC_ROUTER_ADVERTISEMENT:
929 case NDISC_NEIGHBOUR_SOLICITATION:
930 case NDISC_NEIGHBOUR_ADVERTISEMENT:
931 case NDISC_REDIRECT:
932 rc = true;
933 break;
934 }
935 }
936
937out:
938 return rc;
939}
940
941static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
942 const struct net_device *dev,
943 struct flowi6 *fl6,
944 int ifindex,
945 const struct sk_buff *skb,
946 int flags)
947{
948 struct net_vrf *vrf = netdev_priv(dev);
949 struct fib6_table *table = NULL;
950 struct rt6_info *rt6;
951
952 rcu_read_lock();
953
954 /* fib6_table does not have a refcnt and can not be freed */
955 rt6 = rcu_dereference(vrf->rt6);
956 if (likely(rt6))
957 table = rt6->rt6i_table;
958
959 rcu_read_unlock();
960
961 if (!table)
962 return NULL;
963
964 return ip6_pol_route(net, table, ifindex, fl6, skb, flags);
965}
966
967static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
968 int ifindex)
969{
970 const struct ipv6hdr *iph = ipv6_hdr(skb);
971 struct flowi6 fl6 = {
972 .flowi6_iif = ifindex,
973 .flowi6_mark = skb->mark,
974 .flowi6_proto = iph->nexthdr,
975 .daddr = iph->daddr,
976 .saddr = iph->saddr,
977 .flowlabel = ip6_flowinfo(iph),
978 };
979 struct net *net = dev_net(vrf_dev);
980 struct rt6_info *rt6;
981
982 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
983 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
984 if (unlikely(!rt6))
985 return;
986
987 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
988 return;
989
990 skb_dst_set(skb, &rt6->dst);
991}
992
993static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
994 struct sk_buff *skb)
995{
996 int orig_iif = skb->skb_iif;
997 bool need_strict;
998
999 /* loopback traffic; do not push through packet taps again.
1000 * Reset pkt_type for upper layers to process skb
1001 */
1002 if (skb->pkt_type == PACKET_LOOPBACK) {
1003 skb->dev = vrf_dev;
1004 skb->skb_iif = vrf_dev->ifindex;
1005 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1006 skb->pkt_type = PACKET_HOST;
1007 goto out;
1008 }
1009
1010 /* if packet is NDISC or addressed to multicast or link-local
1011 * then keep the ingress interface
1012 */
1013 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1014 if (!ipv6_ndisc_frame(skb) && !need_strict) {
1015 vrf_rx_stats(vrf_dev, skb->len);
1016 skb->dev = vrf_dev;
1017 skb->skb_iif = vrf_dev->ifindex;
1018
1019 if (!list_empty(&vrf_dev->ptype_all)) {
1020 skb_push(skb, skb->mac_len);
1021 dev_queue_xmit_nit(skb, vrf_dev);
1022 skb_pull(skb, skb->mac_len);
1023 }
1024
1025 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1026 }
1027
1028 if (need_strict)
1029 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1030
1031 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1032out:
1033 return skb;
1034}
1035
1036#else
1037static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1038 struct sk_buff *skb)
1039{
1040 return skb;
1041}
1042#endif
1043
1044static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1045 struct sk_buff *skb)
1046{
1047 skb->dev = vrf_dev;
1048 skb->skb_iif = vrf_dev->ifindex;
1049 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1050
1051 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1052 goto out;
1053
1054 /* loopback traffic; do not push through packet taps again.
1055 * Reset pkt_type for upper layers to process skb
1056 */
1057 if (skb->pkt_type == PACKET_LOOPBACK) {
1058 skb->pkt_type = PACKET_HOST;
1059 goto out;
1060 }
1061
1062 vrf_rx_stats(vrf_dev, skb->len);
1063
1064 if (!list_empty(&vrf_dev->ptype_all)) {
1065 skb_push(skb, skb->mac_len);
1066 dev_queue_xmit_nit(skb, vrf_dev);
1067 skb_pull(skb, skb->mac_len);
1068 }
1069
1070 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1071out:
1072 return skb;
1073}
1074
1075/* called with rcu lock held */
1076static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1077 struct sk_buff *skb,
1078 u16 proto)
1079{
1080 switch (proto) {
1081 case AF_INET:
1082 return vrf_ip_rcv(vrf_dev, skb);
1083 case AF_INET6:
1084 return vrf_ip6_rcv(vrf_dev, skb);
1085 }
1086
1087 return skb;
1088}
1089
1090#if IS_ENABLED(CONFIG_IPV6)
1091/* send to link-local or multicast address via interface enslaved to
1092 * VRF device. Force lookup to VRF table without changing flow struct
1093 */
1094static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1095 struct flowi6 *fl6)
1096{
1097 struct net *net = dev_net(dev);
1098 int flags = RT6_LOOKUP_F_IFACE;
1099 struct dst_entry *dst = NULL;
1100 struct rt6_info *rt;
1101
1102 /* VRF device does not have a link-local address and
1103 * sending packets to link-local or mcast addresses over
1104 * a VRF device does not make sense
1105 */
1106 if (fl6->flowi6_oif == dev->ifindex) {
1107 dst = &net->ipv6.ip6_null_entry->dst;
1108 dst_hold(dst);
1109 return dst;
1110 }
1111
1112 if (!ipv6_addr_any(&fl6->saddr))
1113 flags |= RT6_LOOKUP_F_HAS_SADDR;
1114
1115 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1116 if (rt)
1117 dst = &rt->dst;
1118
1119 return dst;
1120}
1121#endif
1122
1123static const struct l3mdev_ops vrf_l3mdev_ops = {
1124 .l3mdev_fib_table = vrf_fib_table,
1125 .l3mdev_l3_rcv = vrf_l3_rcv,
1126 .l3mdev_l3_out = vrf_l3_out,
1127#if IS_ENABLED(CONFIG_IPV6)
1128 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1129#endif
1130};
1131
1132static void vrf_get_drvinfo(struct net_device *dev,
1133 struct ethtool_drvinfo *info)
1134{
1135 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1136 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1137}
1138
1139static const struct ethtool_ops vrf_ethtool_ops = {
1140 .get_drvinfo = vrf_get_drvinfo,
1141};
1142
1143static inline size_t vrf_fib_rule_nl_size(void)
1144{
1145 size_t sz;
1146
1147 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1148 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1149 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1150 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */
1151
1152 return sz;
1153}
1154
1155static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1156{
1157 struct fib_rule_hdr *frh;
1158 struct nlmsghdr *nlh;
1159 struct sk_buff *skb;
1160 int err;
1161
1162 if (family == AF_INET6 && !ipv6_mod_enabled())
1163 return 0;
1164
1165 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1166 if (!skb)
1167 return -ENOMEM;
1168
1169 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1170 if (!nlh)
1171 goto nla_put_failure;
1172
1173 /* rule only needs to appear once */
1174 nlh->nlmsg_flags |= NLM_F_EXCL;
1175
1176 frh = nlmsg_data(nlh);
1177 memset(frh, 0, sizeof(*frh));
1178 frh->family = family;
1179 frh->action = FR_ACT_TO_TBL;
1180
1181 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1182 goto nla_put_failure;
1183
1184 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1185 goto nla_put_failure;
1186
1187 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1188 goto nla_put_failure;
1189
1190 nlmsg_end(skb, nlh);
1191
1192 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1193 skb->sk = dev_net(dev)->rtnl;
1194 if (add_it) {
1195 err = fib_nl_newrule(skb, nlh, NULL);
1196 if (err == -EEXIST)
1197 err = 0;
1198 } else {
1199 err = fib_nl_delrule(skb, nlh, NULL);
1200 if (err == -ENOENT)
1201 err = 0;
1202 }
1203 nlmsg_free(skb);
1204
1205 return err;
1206
1207nla_put_failure:
1208 nlmsg_free(skb);
1209
1210 return -EMSGSIZE;
1211}
1212
1213static int vrf_add_fib_rules(const struct net_device *dev)
1214{
1215 int err;
1216
1217 err = vrf_fib_rule(dev, AF_INET, true);
1218 if (err < 0)
1219 goto out_err;
1220
1221 err = vrf_fib_rule(dev, AF_INET6, true);
1222 if (err < 0)
1223 goto ipv6_err;
1224
1225#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1226 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1227 if (err < 0)
1228 goto ipmr_err;
1229#endif
1230
1231 return 0;
1232
1233#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1234ipmr_err:
1235 vrf_fib_rule(dev, AF_INET6, false);
1236#endif
1237
1238ipv6_err:
1239 vrf_fib_rule(dev, AF_INET, false);
1240
1241out_err:
1242 netdev_err(dev, "Failed to add FIB rules.\n");
1243 return err;
1244}
1245
1246static void vrf_setup(struct net_device *dev)
1247{
1248 ether_setup(dev);
1249
1250 /* Initialize the device structure. */
1251 dev->netdev_ops = &vrf_netdev_ops;
1252 dev->l3mdev_ops = &vrf_l3mdev_ops;
1253 dev->ethtool_ops = &vrf_ethtool_ops;
1254 dev->needs_free_netdev = true;
1255
1256 /* Fill in device structure with ethernet-generic values. */
1257 eth_hw_addr_random(dev);
1258
1259 /* don't acquire vrf device's netif_tx_lock when transmitting */
1260 dev->features |= NETIF_F_LLTX;
1261
1262 /* don't allow vrf devices to change network namespaces. */
1263 dev->features |= NETIF_F_NETNS_LOCAL;
1264
1265 /* does not make sense for a VLAN to be added to a vrf device */
1266 dev->features |= NETIF_F_VLAN_CHALLENGED;
1267
1268 /* enable offload features */
1269 dev->features |= NETIF_F_GSO_SOFTWARE;
1270 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1271 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1272
1273 dev->hw_features = dev->features;
1274 dev->hw_enc_features = dev->features;
1275
1276 /* default to no qdisc; user can add if desired */
1277 dev->priv_flags |= IFF_NO_QUEUE;
1278}
1279
1280static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1281 struct netlink_ext_ack *extack)
1282{
1283 if (tb[IFLA_ADDRESS]) {
1284 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1285 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1286 return -EINVAL;
1287 }
1288 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1289 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1290 return -EADDRNOTAVAIL;
1291 }
1292 }
1293 return 0;
1294}
1295
1296static void vrf_dellink(struct net_device *dev, struct list_head *head)
1297{
1298 struct net_device *port_dev;
1299 struct list_head *iter;
1300
1301 netdev_for_each_lower_dev(dev, port_dev, iter)
1302 vrf_del_slave(dev, port_dev);
1303
1304 unregister_netdevice_queue(dev, head);
1305}
1306
1307static int vrf_newlink(struct net *src_net, struct net_device *dev,
1308 struct nlattr *tb[], struct nlattr *data[],
1309 struct netlink_ext_ack *extack)
1310{
1311 struct net_vrf *vrf = netdev_priv(dev);
1312 bool *add_fib_rules;
1313 struct net *net;
1314 int err;
1315
1316 if (!data || !data[IFLA_VRF_TABLE]) {
1317 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1318 return -EINVAL;
1319 }
1320
1321 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1322 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1323 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1324 "Invalid VRF table id");
1325 return -EINVAL;
1326 }
1327
1328 dev->priv_flags |= IFF_L3MDEV_MASTER;
1329
1330 err = register_netdevice(dev);
1331 if (err)
1332 goto out;
1333
1334 net = dev_net(dev);
1335 add_fib_rules = net_generic(net, vrf_net_id);
1336 if (*add_fib_rules) {
1337 err = vrf_add_fib_rules(dev);
1338 if (err) {
1339 unregister_netdevice(dev);
1340 goto out;
1341 }
1342 *add_fib_rules = false;
1343 }
1344
1345out:
1346 return err;
1347}
1348
1349static size_t vrf_nl_getsize(const struct net_device *dev)
1350{
1351 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1352}
1353
1354static int vrf_fillinfo(struct sk_buff *skb,
1355 const struct net_device *dev)
1356{
1357 struct net_vrf *vrf = netdev_priv(dev);
1358
1359 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1360}
1361
1362static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1363 const struct net_device *slave_dev)
1364{
1365 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1366}
1367
1368static int vrf_fill_slave_info(struct sk_buff *skb,
1369 const struct net_device *vrf_dev,
1370 const struct net_device *slave_dev)
1371{
1372 struct net_vrf *vrf = netdev_priv(vrf_dev);
1373
1374 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1375 return -EMSGSIZE;
1376
1377 return 0;
1378}
1379
1380static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1381 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1382};
1383
1384static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1385 .kind = DRV_NAME,
1386 .priv_size = sizeof(struct net_vrf),
1387
1388 .get_size = vrf_nl_getsize,
1389 .policy = vrf_nl_policy,
1390 .validate = vrf_validate,
1391 .fill_info = vrf_fillinfo,
1392
1393 .get_slave_size = vrf_get_slave_size,
1394 .fill_slave_info = vrf_fill_slave_info,
1395
1396 .newlink = vrf_newlink,
1397 .dellink = vrf_dellink,
1398 .setup = vrf_setup,
1399 .maxtype = IFLA_VRF_MAX,
1400};
1401
1402static int vrf_device_event(struct notifier_block *unused,
1403 unsigned long event, void *ptr)
1404{
1405 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1406
1407 /* only care about unregister events to drop slave references */
1408 if (event == NETDEV_UNREGISTER) {
1409 struct net_device *vrf_dev;
1410
1411 if (!netif_is_l3_slave(dev))
1412 goto out;
1413
1414 vrf_dev = netdev_master_upper_dev_get(dev);
1415 vrf_del_slave(vrf_dev, dev);
1416 }
1417out:
1418 return NOTIFY_DONE;
1419}
1420
1421static struct notifier_block vrf_notifier_block __read_mostly = {
1422 .notifier_call = vrf_device_event,
1423};
1424
1425/* Initialize per network namespace state */
1426static int __net_init vrf_netns_init(struct net *net)
1427{
1428 bool *add_fib_rules = net_generic(net, vrf_net_id);
1429
1430 *add_fib_rules = true;
1431
1432 return 0;
1433}
1434
1435static struct pernet_operations vrf_net_ops __net_initdata = {
1436 .init = vrf_netns_init,
1437 .id = &vrf_net_id,
1438 .size = sizeof(bool),
1439};
1440
1441static int __init vrf_init_module(void)
1442{
1443 int rc;
1444
1445 register_netdevice_notifier(&vrf_notifier_block);
1446
1447 rc = register_pernet_subsys(&vrf_net_ops);
1448 if (rc < 0)
1449 goto error;
1450
1451 rc = rtnl_link_register(&vrf_link_ops);
1452 if (rc < 0) {
1453 unregister_pernet_subsys(&vrf_net_ops);
1454 goto error;
1455 }
1456
1457 return 0;
1458
1459error:
1460 unregister_netdevice_notifier(&vrf_notifier_block);
1461 return rc;
1462}
1463
1464module_init(vrf_init_module);
1465MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1466MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1467MODULE_LICENSE("GPL");
1468MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1469MODULE_VERSION(DRV_VERSION);
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
48struct net_vrf {
49 struct rtable *rth;
50 struct rt6_info *rt6;
51 u32 tb_id;
52};
53
54struct 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)
69static 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
99out:
100 return rc;
101}
102#else
103static bool check_ipv6_frame(const struct sk_buff *skb)
104{
105 return false;
106}
107#endif
108
109static 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
120static 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 */
127static 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
147static 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)
176static 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;
210err:
211 vrf_tx_error(dev, skb);
212 return NET_XMIT_DROP;
213}
214#else
215static 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
223static 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;
242out:
243 return err;
244}
245
246static 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
275out:
276 return ret;
277err:
278 vrf_tx_error(vrf_dev, skb);
279 goto out;
280}
281
282static 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
298static 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 */
318static 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 */
349static 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
357static void vrf_rt6_release(struct net_vrf *vrf)
358{
359 dst_release(&vrf->rt6->dst);
360 vrf->rt6 = NULL;
361}
362
363static 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;
380out:
381 return rc;
382}
383#else
384static void vrf_rt6_release(struct net_vrf *vrf)
385{
386}
387
388static int vrf_rt6_create(struct net_device *dev)
389{
390 return 0;
391}
392#endif
393
394/* modelled after ip_finish_output2 */
395static 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();
431err:
432 if (unlikely(ret < 0))
433 vrf_tx_error(skb->dev, skb);
434 return ret;
435}
436
437static 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
452static 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
460static 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 */
477static 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
496static 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
518out_unregister:
519 netdev_rx_handler_unregister(port_dev);
520out_fail:
521 return ret;
522}
523
524static 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 */
533static 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
545static 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
550static 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
566static 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
586out_rth:
587 vrf_rtable_release(vrf);
588out_stats:
589 free_percpu(dev->dstats);
590 dev->dstats = NULL;
591out_nomem:
592 return -ENOMEM;
593}
594
595static 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
604static 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
611static 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 */
627static 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)
662static 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
678static 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
687static 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
694static const struct ethtool_ops vrf_ethtool_ops = {
695 .get_drvinfo = vrf_get_drvinfo,
696};
697
698static 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
718static 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
729static void vrf_dellink(struct net_device *dev, struct list_head *head)
730{
731 unregister_netdevice_queue(dev, head);
732}
733
734static 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
749static size_t vrf_nl_getsize(const struct net_device *dev)
750{
751 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
752}
753
754static 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
762static 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
768static 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
780static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
781 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
782};
783
784static 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
802static 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 }
817out:
818 return NOTIFY_DONE;
819}
820
821static struct notifier_block vrf_notifier_block __read_mostly = {
822 .notifier_call = vrf_device_event,
823};
824
825static 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
837error:
838 unregister_netdevice_notifier(&vrf_notifier_block);
839 return rc;
840}
841
842module_init(vrf_init_module);
843MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
844MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
845MODULE_LICENSE("GPL");
846MODULE_ALIAS_RTNL_LINK(DRV_NAME);
847MODULE_VERSION(DRV_VERSION);