Loading...
Note: File does not exist in v3.15.
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
40#define DRV_NAME "vrf"
41#define DRV_VERSION "1.0"
42
43#define FIB_RULE_PREF 1000 /* default preference for FIB rules */
44static bool add_fib_rules = true;
45
46struct net_vrf {
47 struct rtable __rcu *rth;
48 struct rtable __rcu *rth_local;
49 struct rt6_info __rcu *rt6;
50 struct rt6_info __rcu *rt6_local;
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 struct rtnl_link_stats64 *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 return stats;
106}
107
108/* Local traffic destined to local address. Reinsert the packet to rx
109 * path, similar to loopback handling.
110 */
111static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
112 struct dst_entry *dst)
113{
114 int len = skb->len;
115
116 skb_orphan(skb);
117
118 skb_dst_set(skb, dst);
119 skb_dst_force(skb);
120
121 /* set pkt_type to avoid skb hitting packet taps twice -
122 * once on Tx and again in Rx processing
123 */
124 skb->pkt_type = PACKET_LOOPBACK;
125
126 skb->protocol = eth_type_trans(skb, dev);
127
128 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
129 vrf_rx_stats(dev, len);
130 else
131 this_cpu_inc(dev->dstats->rx_drps);
132
133 return NETDEV_TX_OK;
134}
135
136#if IS_ENABLED(CONFIG_IPV6)
137static int vrf_ip6_local_out(struct net *net, struct sock *sk,
138 struct sk_buff *skb)
139{
140 int err;
141
142 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
143 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
144
145 if (likely(err == 1))
146 err = dst_output(net, sk, skb);
147
148 return err;
149}
150
151static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
152 struct net_device *dev)
153{
154 const struct ipv6hdr *iph = ipv6_hdr(skb);
155 struct net *net = dev_net(skb->dev);
156 struct flowi6 fl6 = {
157 /* needed to match OIF rule */
158 .flowi6_oif = dev->ifindex,
159 .flowi6_iif = LOOPBACK_IFINDEX,
160 .daddr = iph->daddr,
161 .saddr = iph->saddr,
162 .flowlabel = ip6_flowinfo(iph),
163 .flowi6_mark = skb->mark,
164 .flowi6_proto = iph->nexthdr,
165 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
166 };
167 int ret = NET_XMIT_DROP;
168 struct dst_entry *dst;
169 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
170
171 dst = ip6_route_output(net, NULL, &fl6);
172 if (dst == dst_null)
173 goto err;
174
175 skb_dst_drop(skb);
176
177 /* if dst.dev is loopback or the VRF device again this is locally
178 * originated traffic destined to a local address. Short circuit
179 * to Rx path using our local dst
180 */
181 if (dst->dev == net->loopback_dev || dst->dev == dev) {
182 struct net_vrf *vrf = netdev_priv(dev);
183 struct rt6_info *rt6_local;
184
185 /* release looked up dst and use cached local dst */
186 dst_release(dst);
187
188 rcu_read_lock();
189
190 rt6_local = rcu_dereference(vrf->rt6_local);
191 if (unlikely(!rt6_local)) {
192 rcu_read_unlock();
193 goto err;
194 }
195
196 /* Ordering issue: cached local dst is created on newlink
197 * before the IPv6 initialization. Using the local dst
198 * requires rt6i_idev to be set so make sure it is.
199 */
200 if (unlikely(!rt6_local->rt6i_idev)) {
201 rt6_local->rt6i_idev = in6_dev_get(dev);
202 if (!rt6_local->rt6i_idev) {
203 rcu_read_unlock();
204 goto err;
205 }
206 }
207
208 dst = &rt6_local->dst;
209 dst_hold(dst);
210
211 rcu_read_unlock();
212
213 return vrf_local_xmit(skb, dev, &rt6_local->dst);
214 }
215
216 skb_dst_set(skb, dst);
217
218 /* strip the ethernet header added for pass through VRF device */
219 __skb_pull(skb, skb_network_offset(skb));
220
221 ret = vrf_ip6_local_out(net, skb->sk, skb);
222 if (unlikely(net_xmit_eval(ret)))
223 dev->stats.tx_errors++;
224 else
225 ret = NET_XMIT_SUCCESS;
226
227 return ret;
228err:
229 vrf_tx_error(dev, skb);
230 return NET_XMIT_DROP;
231}
232#else
233static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
234 struct net_device *dev)
235{
236 vrf_tx_error(dev, skb);
237 return NET_XMIT_DROP;
238}
239#endif
240
241/* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
242static int vrf_ip_local_out(struct net *net, struct sock *sk,
243 struct sk_buff *skb)
244{
245 int err;
246
247 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
248 skb, NULL, skb_dst(skb)->dev, dst_output);
249 if (likely(err == 1))
250 err = dst_output(net, sk, skb);
251
252 return err;
253}
254
255static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
256 struct net_device *vrf_dev)
257{
258 struct iphdr *ip4h = ip_hdr(skb);
259 int ret = NET_XMIT_DROP;
260 struct flowi4 fl4 = {
261 /* needed to match OIF rule */
262 .flowi4_oif = vrf_dev->ifindex,
263 .flowi4_iif = LOOPBACK_IFINDEX,
264 .flowi4_tos = RT_TOS(ip4h->tos),
265 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
266 .flowi4_proto = ip4h->protocol,
267 .daddr = ip4h->daddr,
268 .saddr = ip4h->saddr,
269 };
270 struct net *net = dev_net(vrf_dev);
271 struct rtable *rt;
272
273 rt = ip_route_output_flow(net, &fl4, NULL);
274 if (IS_ERR(rt))
275 goto err;
276
277 skb_dst_drop(skb);
278
279 /* if dst.dev is loopback or the VRF device again this is locally
280 * originated traffic destined to a local address. Short circuit
281 * to Rx path using our local dst
282 */
283 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) {
284 struct net_vrf *vrf = netdev_priv(vrf_dev);
285 struct rtable *rth_local;
286 struct dst_entry *dst = NULL;
287
288 ip_rt_put(rt);
289
290 rcu_read_lock();
291
292 rth_local = rcu_dereference(vrf->rth_local);
293 if (likely(rth_local)) {
294 dst = &rth_local->dst;
295 dst_hold(dst);
296 }
297
298 rcu_read_unlock();
299
300 if (unlikely(!dst))
301 goto err;
302
303 return vrf_local_xmit(skb, vrf_dev, dst);
304 }
305
306 skb_dst_set(skb, &rt->dst);
307
308 /* strip the ethernet header added for pass through VRF device */
309 __skb_pull(skb, skb_network_offset(skb));
310
311 if (!ip4h->saddr) {
312 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
313 RT_SCOPE_LINK);
314 }
315
316 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
317 if (unlikely(net_xmit_eval(ret)))
318 vrf_dev->stats.tx_errors++;
319 else
320 ret = NET_XMIT_SUCCESS;
321
322out:
323 return ret;
324err:
325 vrf_tx_error(vrf_dev, skb);
326 goto out;
327}
328
329static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
330{
331 switch (skb->protocol) {
332 case htons(ETH_P_IP):
333 return vrf_process_v4_outbound(skb, dev);
334 case htons(ETH_P_IPV6):
335 return vrf_process_v6_outbound(skb, dev);
336 default:
337 vrf_tx_error(dev, skb);
338 return NET_XMIT_DROP;
339 }
340}
341
342static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
343{
344 int len = skb->len;
345 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
346
347 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
348 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
349
350 u64_stats_update_begin(&dstats->syncp);
351 dstats->tx_pkts++;
352 dstats->tx_bytes += len;
353 u64_stats_update_end(&dstats->syncp);
354 } else {
355 this_cpu_inc(dev->dstats->tx_drps);
356 }
357
358 return ret;
359}
360
361#if IS_ENABLED(CONFIG_IPV6)
362/* modelled after ip6_finish_output2 */
363static int vrf_finish_output6(struct net *net, struct sock *sk,
364 struct sk_buff *skb)
365{
366 struct dst_entry *dst = skb_dst(skb);
367 struct net_device *dev = dst->dev;
368 struct neighbour *neigh;
369 struct in6_addr *nexthop;
370 int ret;
371
372 nf_reset(skb);
373
374 skb->protocol = htons(ETH_P_IPV6);
375 skb->dev = dev;
376
377 rcu_read_lock_bh();
378 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
379 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
380 if (unlikely(!neigh))
381 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
382 if (!IS_ERR(neigh)) {
383 ret = dst_neigh_output(dst, neigh, skb);
384 rcu_read_unlock_bh();
385 return ret;
386 }
387 rcu_read_unlock_bh();
388
389 IP6_INC_STATS(dev_net(dst->dev),
390 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
391 kfree_skb(skb);
392 return -EINVAL;
393}
394
395/* modelled after ip6_output */
396static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
397{
398 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
399 net, sk, skb, NULL, skb_dst(skb)->dev,
400 vrf_finish_output6,
401 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
402}
403
404/* set dst on skb to send packet to us via dev_xmit path. Allows
405 * packet to go through device based features such as qdisc, netfilter
406 * hooks and packet sockets with skb->dev set to vrf device.
407 */
408static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
409 struct sock *sk,
410 struct sk_buff *skb)
411{
412 struct net_vrf *vrf = netdev_priv(vrf_dev);
413 struct dst_entry *dst = NULL;
414 struct rt6_info *rt6;
415
416 /* don't divert link scope packets */
417 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
418 return skb;
419
420 rcu_read_lock();
421
422 rt6 = rcu_dereference(vrf->rt6);
423 if (likely(rt6)) {
424 dst = &rt6->dst;
425 dst_hold(dst);
426 }
427
428 rcu_read_unlock();
429
430 if (unlikely(!dst)) {
431 vrf_tx_error(vrf_dev, skb);
432 return NULL;
433 }
434
435 skb_dst_drop(skb);
436 skb_dst_set(skb, dst);
437
438 return skb;
439}
440
441/* holding rtnl */
442static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
443{
444 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
445 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local);
446 struct net *net = dev_net(dev);
447 struct dst_entry *dst;
448
449 RCU_INIT_POINTER(vrf->rt6, NULL);
450 RCU_INIT_POINTER(vrf->rt6_local, NULL);
451 synchronize_rcu();
452
453 /* move dev in dst's to loopback so this VRF device can be deleted
454 * - based on dst_ifdown
455 */
456 if (rt6) {
457 dst = &rt6->dst;
458 dev_put(dst->dev);
459 dst->dev = net->loopback_dev;
460 dev_hold(dst->dev);
461 dst_release(dst);
462 }
463
464 if (rt6_local) {
465 if (rt6_local->rt6i_idev) {
466 in6_dev_put(rt6_local->rt6i_idev);
467 rt6_local->rt6i_idev = NULL;
468 }
469
470 dst = &rt6_local->dst;
471 dev_put(dst->dev);
472 dst->dev = net->loopback_dev;
473 dev_hold(dst->dev);
474 dst_release(dst);
475 }
476}
477
478static int vrf_rt6_create(struct net_device *dev)
479{
480 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE;
481 struct net_vrf *vrf = netdev_priv(dev);
482 struct net *net = dev_net(dev);
483 struct fib6_table *rt6i_table;
484 struct rt6_info *rt6, *rt6_local;
485 int rc = -ENOMEM;
486
487 /* IPv6 can be CONFIG enabled and then disabled runtime */
488 if (!ipv6_mod_enabled())
489 return 0;
490
491 rt6i_table = fib6_new_table(net, vrf->tb_id);
492 if (!rt6i_table)
493 goto out;
494
495 /* create a dst for routing packets out a VRF device */
496 rt6 = ip6_dst_alloc(net, dev, flags);
497 if (!rt6)
498 goto out;
499
500 dst_hold(&rt6->dst);
501
502 rt6->rt6i_table = rt6i_table;
503 rt6->dst.output = vrf_output6;
504
505 /* create a dst for local routing - packets sent locally
506 * to local address via the VRF device as a loopback
507 */
508 rt6_local = ip6_dst_alloc(net, dev, flags);
509 if (!rt6_local) {
510 dst_release(&rt6->dst);
511 goto out;
512 }
513
514 dst_hold(&rt6_local->dst);
515
516 rt6_local->rt6i_idev = in6_dev_get(dev);
517 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL;
518 rt6_local->rt6i_table = rt6i_table;
519 rt6_local->dst.input = ip6_input;
520
521 rcu_assign_pointer(vrf->rt6, rt6);
522 rcu_assign_pointer(vrf->rt6_local, rt6_local);
523
524 rc = 0;
525out:
526 return rc;
527}
528#else
529static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
530 struct sock *sk,
531 struct sk_buff *skb)
532{
533 return skb;
534}
535
536static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
537{
538}
539
540static int vrf_rt6_create(struct net_device *dev)
541{
542 return 0;
543}
544#endif
545
546/* modelled after ip_finish_output2 */
547static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
548{
549 struct dst_entry *dst = skb_dst(skb);
550 struct rtable *rt = (struct rtable *)dst;
551 struct net_device *dev = dst->dev;
552 unsigned int hh_len = LL_RESERVED_SPACE(dev);
553 struct neighbour *neigh;
554 u32 nexthop;
555 int ret = -EINVAL;
556
557 nf_reset(skb);
558
559 /* Be paranoid, rather than too clever. */
560 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
561 struct sk_buff *skb2;
562
563 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
564 if (!skb2) {
565 ret = -ENOMEM;
566 goto err;
567 }
568 if (skb->sk)
569 skb_set_owner_w(skb2, skb->sk);
570
571 consume_skb(skb);
572 skb = skb2;
573 }
574
575 rcu_read_lock_bh();
576
577 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
578 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
579 if (unlikely(!neigh))
580 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
581 if (!IS_ERR(neigh))
582 ret = dst_neigh_output(dst, neigh, skb);
583
584 rcu_read_unlock_bh();
585err:
586 if (unlikely(ret < 0))
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(struct net_device *vrf_dev,
611 struct sock *sk,
612 struct sk_buff *skb)
613{
614 struct net_vrf *vrf = netdev_priv(vrf_dev);
615 struct dst_entry *dst = NULL;
616 struct rtable *rth;
617
618 /* don't divert multicast */
619 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
620 return skb;
621
622 rcu_read_lock();
623
624 rth = rcu_dereference(vrf->rth);
625 if (likely(rth)) {
626 dst = &rth->dst;
627 dst_hold(dst);
628 }
629
630 rcu_read_unlock();
631
632 if (unlikely(!dst)) {
633 vrf_tx_error(vrf_dev, skb);
634 return NULL;
635 }
636
637 skb_dst_drop(skb);
638 skb_dst_set(skb, dst);
639
640 return skb;
641}
642
643/* called with rcu lock held */
644static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
645 struct sock *sk,
646 struct sk_buff *skb,
647 u16 proto)
648{
649 switch (proto) {
650 case AF_INET:
651 return vrf_ip_out(vrf_dev, sk, skb);
652 case AF_INET6:
653 return vrf_ip6_out(vrf_dev, sk, skb);
654 }
655
656 return skb;
657}
658
659/* holding rtnl */
660static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
661{
662 struct rtable *rth = rtnl_dereference(vrf->rth);
663 struct rtable *rth_local = rtnl_dereference(vrf->rth_local);
664 struct net *net = dev_net(dev);
665 struct dst_entry *dst;
666
667 RCU_INIT_POINTER(vrf->rth, NULL);
668 RCU_INIT_POINTER(vrf->rth_local, NULL);
669 synchronize_rcu();
670
671 /* move dev in dst's to loopback so this VRF device can be deleted
672 * - based on dst_ifdown
673 */
674 if (rth) {
675 dst = &rth->dst;
676 dev_put(dst->dev);
677 dst->dev = net->loopback_dev;
678 dev_hold(dst->dev);
679 dst_release(dst);
680 }
681
682 if (rth_local) {
683 dst = &rth_local->dst;
684 dev_put(dst->dev);
685 dst->dev = net->loopback_dev;
686 dev_hold(dst->dev);
687 dst_release(dst);
688 }
689}
690
691static int vrf_rtable_create(struct net_device *dev)
692{
693 struct net_vrf *vrf = netdev_priv(dev);
694 struct rtable *rth, *rth_local;
695
696 if (!fib_new_table(dev_net(dev), vrf->tb_id))
697 return -ENOMEM;
698
699 /* create a dst for routing packets out through a VRF device */
700 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
701 if (!rth)
702 return -ENOMEM;
703
704 /* create a dst for local ingress routing - packets sent locally
705 * to local address via the VRF device as a loopback
706 */
707 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0);
708 if (!rth_local) {
709 dst_release(&rth->dst);
710 return -ENOMEM;
711 }
712
713 rth->dst.output = vrf_output;
714 rth->rt_table_id = vrf->tb_id;
715
716 rth_local->rt_table_id = vrf->tb_id;
717
718 rcu_assign_pointer(vrf->rth, rth);
719 rcu_assign_pointer(vrf->rth_local, rth_local);
720
721 return 0;
722}
723
724/**************************** device handling ********************/
725
726/* cycle interface to flush neighbor cache and move routes across tables */
727static void cycle_netdev(struct net_device *dev)
728{
729 unsigned int flags = dev->flags;
730 int ret;
731
732 if (!netif_running(dev))
733 return;
734
735 ret = dev_change_flags(dev, flags & ~IFF_UP);
736 if (ret >= 0)
737 ret = dev_change_flags(dev, flags);
738
739 if (ret < 0) {
740 netdev_err(dev,
741 "Failed to cycle device %s; route tables might be wrong!\n",
742 dev->name);
743 }
744}
745
746static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
747{
748 int ret;
749
750 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
751 if (ret < 0)
752 return ret;
753
754 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
755 cycle_netdev(port_dev);
756
757 return 0;
758}
759
760static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
761{
762 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
763 return -EINVAL;
764
765 return do_vrf_add_slave(dev, port_dev);
766}
767
768/* inverse of do_vrf_add_slave */
769static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
770{
771 netdev_upper_dev_unlink(port_dev, dev);
772 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
773
774 cycle_netdev(port_dev);
775
776 return 0;
777}
778
779static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
780{
781 return do_vrf_del_slave(dev, port_dev);
782}
783
784static void vrf_dev_uninit(struct net_device *dev)
785{
786 struct net_vrf *vrf = netdev_priv(dev);
787 struct net_device *port_dev;
788 struct list_head *iter;
789
790 vrf_rtable_release(dev, vrf);
791 vrf_rt6_release(dev, vrf);
792
793 netdev_for_each_lower_dev(dev, port_dev, iter)
794 vrf_del_slave(dev, port_dev);
795
796 free_percpu(dev->dstats);
797 dev->dstats = NULL;
798}
799
800static int vrf_dev_init(struct net_device *dev)
801{
802 struct net_vrf *vrf = netdev_priv(dev);
803
804 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
805 if (!dev->dstats)
806 goto out_nomem;
807
808 /* create the default dst which points back to us */
809 if (vrf_rtable_create(dev) != 0)
810 goto out_stats;
811
812 if (vrf_rt6_create(dev) != 0)
813 goto out_rth;
814
815 dev->flags = IFF_MASTER | IFF_NOARP;
816
817 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
818 dev->mtu = 64 * 1024;
819
820 /* similarly, oper state is irrelevant; set to up to avoid confusion */
821 dev->operstate = IF_OPER_UP;
822 netdev_lockdep_set_classes(dev);
823 return 0;
824
825out_rth:
826 vrf_rtable_release(dev, vrf);
827out_stats:
828 free_percpu(dev->dstats);
829 dev->dstats = NULL;
830out_nomem:
831 return -ENOMEM;
832}
833
834static const struct net_device_ops vrf_netdev_ops = {
835 .ndo_init = vrf_dev_init,
836 .ndo_uninit = vrf_dev_uninit,
837 .ndo_start_xmit = vrf_xmit,
838 .ndo_get_stats64 = vrf_get_stats64,
839 .ndo_add_slave = vrf_add_slave,
840 .ndo_del_slave = vrf_del_slave,
841};
842
843static u32 vrf_fib_table(const struct net_device *dev)
844{
845 struct net_vrf *vrf = netdev_priv(dev);
846
847 return vrf->tb_id;
848}
849
850static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
851{
852 return 0;
853}
854
855static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
856 struct sk_buff *skb,
857 struct net_device *dev)
858{
859 struct net *net = dev_net(dev);
860
861 if (NF_HOOK(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) < 0)
862 skb = NULL; /* kfree_skb(skb) handled by nf code */
863
864 return skb;
865}
866
867#if IS_ENABLED(CONFIG_IPV6)
868/* neighbor handling is done with actual device; do not want
869 * to flip skb->dev for those ndisc packets. This really fails
870 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
871 * a start.
872 */
873static bool ipv6_ndisc_frame(const struct sk_buff *skb)
874{
875 const struct ipv6hdr *iph = ipv6_hdr(skb);
876 bool rc = false;
877
878 if (iph->nexthdr == NEXTHDR_ICMP) {
879 const struct icmp6hdr *icmph;
880 struct icmp6hdr _icmph;
881
882 icmph = skb_header_pointer(skb, sizeof(*iph),
883 sizeof(_icmph), &_icmph);
884 if (!icmph)
885 goto out;
886
887 switch (icmph->icmp6_type) {
888 case NDISC_ROUTER_SOLICITATION:
889 case NDISC_ROUTER_ADVERTISEMENT:
890 case NDISC_NEIGHBOUR_SOLICITATION:
891 case NDISC_NEIGHBOUR_ADVERTISEMENT:
892 case NDISC_REDIRECT:
893 rc = true;
894 break;
895 }
896 }
897
898out:
899 return rc;
900}
901
902static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
903 const struct net_device *dev,
904 struct flowi6 *fl6,
905 int ifindex,
906 int flags)
907{
908 struct net_vrf *vrf = netdev_priv(dev);
909 struct fib6_table *table = NULL;
910 struct rt6_info *rt6;
911
912 rcu_read_lock();
913
914 /* fib6_table does not have a refcnt and can not be freed */
915 rt6 = rcu_dereference(vrf->rt6);
916 if (likely(rt6))
917 table = rt6->rt6i_table;
918
919 rcu_read_unlock();
920
921 if (!table)
922 return NULL;
923
924 return ip6_pol_route(net, table, ifindex, fl6, flags);
925}
926
927static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
928 int ifindex)
929{
930 const struct ipv6hdr *iph = ipv6_hdr(skb);
931 struct flowi6 fl6 = {
932 .daddr = iph->daddr,
933 .saddr = iph->saddr,
934 .flowlabel = ip6_flowinfo(iph),
935 .flowi6_mark = skb->mark,
936 .flowi6_proto = iph->nexthdr,
937 .flowi6_iif = ifindex,
938 };
939 struct net *net = dev_net(vrf_dev);
940 struct rt6_info *rt6;
941
942 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
943 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
944 if (unlikely(!rt6))
945 return;
946
947 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
948 return;
949
950 skb_dst_set(skb, &rt6->dst);
951}
952
953static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
954 struct sk_buff *skb)
955{
956 int orig_iif = skb->skb_iif;
957 bool need_strict;
958
959 /* loopback traffic; do not push through packet taps again.
960 * Reset pkt_type for upper layers to process skb
961 */
962 if (skb->pkt_type == PACKET_LOOPBACK) {
963 skb->dev = vrf_dev;
964 skb->skb_iif = vrf_dev->ifindex;
965 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
966 skb->pkt_type = PACKET_HOST;
967 goto out;
968 }
969
970 /* if packet is NDISC or addressed to multicast or link-local
971 * then keep the ingress interface
972 */
973 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
974 if (!ipv6_ndisc_frame(skb) && !need_strict) {
975 vrf_rx_stats(vrf_dev, skb->len);
976 skb->dev = vrf_dev;
977 skb->skb_iif = vrf_dev->ifindex;
978
979 skb_push(skb, skb->mac_len);
980 dev_queue_xmit_nit(skb, vrf_dev);
981 skb_pull(skb, skb->mac_len);
982
983 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
984 }
985
986 if (need_strict)
987 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
988
989 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
990out:
991 return skb;
992}
993
994#else
995static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
996 struct sk_buff *skb)
997{
998 return skb;
999}
1000#endif
1001
1002static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1003 struct sk_buff *skb)
1004{
1005 skb->dev = vrf_dev;
1006 skb->skb_iif = vrf_dev->ifindex;
1007 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1008
1009 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1010 goto out;
1011
1012 /* loopback traffic; do not push through packet taps again.
1013 * Reset pkt_type for upper layers to process skb
1014 */
1015 if (skb->pkt_type == PACKET_LOOPBACK) {
1016 skb->pkt_type = PACKET_HOST;
1017 goto out;
1018 }
1019
1020 vrf_rx_stats(vrf_dev, skb->len);
1021
1022 skb_push(skb, skb->mac_len);
1023 dev_queue_xmit_nit(skb, vrf_dev);
1024 skb_pull(skb, skb->mac_len);
1025
1026 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1027out:
1028 return skb;
1029}
1030
1031/* called with rcu lock held */
1032static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1033 struct sk_buff *skb,
1034 u16 proto)
1035{
1036 switch (proto) {
1037 case AF_INET:
1038 return vrf_ip_rcv(vrf_dev, skb);
1039 case AF_INET6:
1040 return vrf_ip6_rcv(vrf_dev, skb);
1041 }
1042
1043 return skb;
1044}
1045
1046#if IS_ENABLED(CONFIG_IPV6)
1047/* send to link-local or multicast address via interface enslaved to
1048 * VRF device. Force lookup to VRF table without changing flow struct
1049 */
1050static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1051 struct flowi6 *fl6)
1052{
1053 struct net *net = dev_net(dev);
1054 int flags = RT6_LOOKUP_F_IFACE;
1055 struct dst_entry *dst = NULL;
1056 struct rt6_info *rt;
1057
1058 /* VRF device does not have a link-local address and
1059 * sending packets to link-local or mcast addresses over
1060 * a VRF device does not make sense
1061 */
1062 if (fl6->flowi6_oif == dev->ifindex) {
1063 dst = &net->ipv6.ip6_null_entry->dst;
1064 dst_hold(dst);
1065 return dst;
1066 }
1067
1068 if (!ipv6_addr_any(&fl6->saddr))
1069 flags |= RT6_LOOKUP_F_HAS_SADDR;
1070
1071 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1072 if (rt)
1073 dst = &rt->dst;
1074
1075 return dst;
1076}
1077#endif
1078
1079static const struct l3mdev_ops vrf_l3mdev_ops = {
1080 .l3mdev_fib_table = vrf_fib_table,
1081 .l3mdev_l3_rcv = vrf_l3_rcv,
1082 .l3mdev_l3_out = vrf_l3_out,
1083#if IS_ENABLED(CONFIG_IPV6)
1084 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1085#endif
1086};
1087
1088static void vrf_get_drvinfo(struct net_device *dev,
1089 struct ethtool_drvinfo *info)
1090{
1091 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1092 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1093}
1094
1095static const struct ethtool_ops vrf_ethtool_ops = {
1096 .get_drvinfo = vrf_get_drvinfo,
1097};
1098
1099static inline size_t vrf_fib_rule_nl_size(void)
1100{
1101 size_t sz;
1102
1103 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1104 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1105 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1106
1107 return sz;
1108}
1109
1110static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1111{
1112 struct fib_rule_hdr *frh;
1113 struct nlmsghdr *nlh;
1114 struct sk_buff *skb;
1115 int err;
1116
1117 if (family == AF_INET6 && !ipv6_mod_enabled())
1118 return 0;
1119
1120 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1121 if (!skb)
1122 return -ENOMEM;
1123
1124 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1125 if (!nlh)
1126 goto nla_put_failure;
1127
1128 /* rule only needs to appear once */
1129 nlh->nlmsg_flags &= NLM_F_EXCL;
1130
1131 frh = nlmsg_data(nlh);
1132 memset(frh, 0, sizeof(*frh));
1133 frh->family = family;
1134 frh->action = FR_ACT_TO_TBL;
1135
1136 if (nla_put_u32(skb, FRA_L3MDEV, 1))
1137 goto nla_put_failure;
1138
1139 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1140 goto nla_put_failure;
1141
1142 nlmsg_end(skb, nlh);
1143
1144 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1145 skb->sk = dev_net(dev)->rtnl;
1146 if (add_it) {
1147 err = fib_nl_newrule(skb, nlh);
1148 if (err == -EEXIST)
1149 err = 0;
1150 } else {
1151 err = fib_nl_delrule(skb, nlh);
1152 if (err == -ENOENT)
1153 err = 0;
1154 }
1155 nlmsg_free(skb);
1156
1157 return err;
1158
1159nla_put_failure:
1160 nlmsg_free(skb);
1161
1162 return -EMSGSIZE;
1163}
1164
1165static int vrf_add_fib_rules(const struct net_device *dev)
1166{
1167 int err;
1168
1169 err = vrf_fib_rule(dev, AF_INET, true);
1170 if (err < 0)
1171 goto out_err;
1172
1173 err = vrf_fib_rule(dev, AF_INET6, true);
1174 if (err < 0)
1175 goto ipv6_err;
1176
1177#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1178 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1179 if (err < 0)
1180 goto ipmr_err;
1181#endif
1182
1183 return 0;
1184
1185#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1186ipmr_err:
1187 vrf_fib_rule(dev, AF_INET6, false);
1188#endif
1189
1190ipv6_err:
1191 vrf_fib_rule(dev, AF_INET, false);
1192
1193out_err:
1194 netdev_err(dev, "Failed to add FIB rules.\n");
1195 return err;
1196}
1197
1198static void vrf_setup(struct net_device *dev)
1199{
1200 ether_setup(dev);
1201
1202 /* Initialize the device structure. */
1203 dev->netdev_ops = &vrf_netdev_ops;
1204 dev->l3mdev_ops = &vrf_l3mdev_ops;
1205 dev->ethtool_ops = &vrf_ethtool_ops;
1206 dev->destructor = free_netdev;
1207
1208 /* Fill in device structure with ethernet-generic values. */
1209 eth_hw_addr_random(dev);
1210
1211 /* don't acquire vrf device's netif_tx_lock when transmitting */
1212 dev->features |= NETIF_F_LLTX;
1213
1214 /* don't allow vrf devices to change network namespaces. */
1215 dev->features |= NETIF_F_NETNS_LOCAL;
1216
1217 /* does not make sense for a VLAN to be added to a vrf device */
1218 dev->features |= NETIF_F_VLAN_CHALLENGED;
1219
1220 /* enable offload features */
1221 dev->features |= NETIF_F_GSO_SOFTWARE;
1222 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1223 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1224
1225 dev->hw_features = dev->features;
1226 dev->hw_enc_features = dev->features;
1227
1228 /* default to no qdisc; user can add if desired */
1229 dev->priv_flags |= IFF_NO_QUEUE;
1230}
1231
1232static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
1233{
1234 if (tb[IFLA_ADDRESS]) {
1235 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1236 return -EINVAL;
1237 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1238 return -EADDRNOTAVAIL;
1239 }
1240 return 0;
1241}
1242
1243static void vrf_dellink(struct net_device *dev, struct list_head *head)
1244{
1245 unregister_netdevice_queue(dev, head);
1246}
1247
1248static int vrf_newlink(struct net *src_net, struct net_device *dev,
1249 struct nlattr *tb[], struct nlattr *data[])
1250{
1251 struct net_vrf *vrf = netdev_priv(dev);
1252 int err;
1253
1254 if (!data || !data[IFLA_VRF_TABLE])
1255 return -EINVAL;
1256
1257 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1258 if (vrf->tb_id == RT_TABLE_UNSPEC)
1259 return -EINVAL;
1260
1261 dev->priv_flags |= IFF_L3MDEV_MASTER;
1262
1263 err = register_netdevice(dev);
1264 if (err)
1265 goto out;
1266
1267 if (add_fib_rules) {
1268 err = vrf_add_fib_rules(dev);
1269 if (err) {
1270 unregister_netdevice(dev);
1271 goto out;
1272 }
1273 add_fib_rules = false;
1274 }
1275
1276out:
1277 return err;
1278}
1279
1280static size_t vrf_nl_getsize(const struct net_device *dev)
1281{
1282 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1283}
1284
1285static int vrf_fillinfo(struct sk_buff *skb,
1286 const struct net_device *dev)
1287{
1288 struct net_vrf *vrf = netdev_priv(dev);
1289
1290 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1291}
1292
1293static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1294 const struct net_device *slave_dev)
1295{
1296 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1297}
1298
1299static int vrf_fill_slave_info(struct sk_buff *skb,
1300 const struct net_device *vrf_dev,
1301 const struct net_device *slave_dev)
1302{
1303 struct net_vrf *vrf = netdev_priv(vrf_dev);
1304
1305 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1306 return -EMSGSIZE;
1307
1308 return 0;
1309}
1310
1311static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1312 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1313};
1314
1315static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1316 .kind = DRV_NAME,
1317 .priv_size = sizeof(struct net_vrf),
1318
1319 .get_size = vrf_nl_getsize,
1320 .policy = vrf_nl_policy,
1321 .validate = vrf_validate,
1322 .fill_info = vrf_fillinfo,
1323
1324 .get_slave_size = vrf_get_slave_size,
1325 .fill_slave_info = vrf_fill_slave_info,
1326
1327 .newlink = vrf_newlink,
1328 .dellink = vrf_dellink,
1329 .setup = vrf_setup,
1330 .maxtype = IFLA_VRF_MAX,
1331};
1332
1333static int vrf_device_event(struct notifier_block *unused,
1334 unsigned long event, void *ptr)
1335{
1336 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1337
1338 /* only care about unregister events to drop slave references */
1339 if (event == NETDEV_UNREGISTER) {
1340 struct net_device *vrf_dev;
1341
1342 if (!netif_is_l3_slave(dev))
1343 goto out;
1344
1345 vrf_dev = netdev_master_upper_dev_get(dev);
1346 vrf_del_slave(vrf_dev, dev);
1347 }
1348out:
1349 return NOTIFY_DONE;
1350}
1351
1352static struct notifier_block vrf_notifier_block __read_mostly = {
1353 .notifier_call = vrf_device_event,
1354};
1355
1356static int __init vrf_init_module(void)
1357{
1358 int rc;
1359
1360 register_netdevice_notifier(&vrf_notifier_block);
1361
1362 rc = rtnl_link_register(&vrf_link_ops);
1363 if (rc < 0)
1364 goto error;
1365
1366 return 0;
1367
1368error:
1369 unregister_netdevice_notifier(&vrf_notifier_block);
1370 return rc;
1371}
1372
1373module_init(vrf_init_module);
1374MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1375MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1376MODULE_LICENSE("GPL");
1377MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1378MODULE_VERSION(DRV_VERSION);