x86/mm/pat: Don't report PAT on CPUs that don't support it
[linux/fpc-iii.git] / drivers / net / vrf.c
blob2e62c4d1b22048f0411238abe24a0615bc51912a
1 /*
2 * vrf.c: device driver to encapsulate a VRF space
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>
8 * Based on dummy, team and ipvlan drivers
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.
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>
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>
41 #define DRV_NAME "vrf"
42 #define DRV_VERSION "1.0"
44 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */
46 static unsigned int vrf_net_id;
48 struct net_vrf {
49 struct rtable __rcu *rth;
50 struct rtable __rcu *rth_local;
51 struct rt6_info __rcu *rt6;
52 struct rt6_info __rcu *rt6_local;
53 u32 tb_id;
56 struct pcpu_dstats {
57 u64 tx_pkts;
58 u64 tx_bytes;
59 u64 tx_drps;
60 u64 rx_pkts;
61 u64 rx_bytes;
62 u64 rx_drps;
63 struct u64_stats_sync syncp;
66 static void vrf_rx_stats(struct net_device *dev, int len)
68 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
70 u64_stats_update_begin(&dstats->syncp);
71 dstats->rx_pkts++;
72 dstats->rx_bytes += len;
73 u64_stats_update_end(&dstats->syncp);
76 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
78 vrf_dev->stats.tx_errors++;
79 kfree_skb(skb);
82 static void vrf_get_stats64(struct net_device *dev,
83 struct rtnl_link_stats64 *stats)
85 int i;
87 for_each_possible_cpu(i) {
88 const struct pcpu_dstats *dstats;
89 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
90 unsigned int start;
92 dstats = per_cpu_ptr(dev->dstats, i);
93 do {
94 start = u64_stats_fetch_begin_irq(&dstats->syncp);
95 tbytes = dstats->tx_bytes;
96 tpkts = dstats->tx_pkts;
97 tdrops = dstats->tx_drps;
98 rbytes = dstats->rx_bytes;
99 rpkts = dstats->rx_pkts;
100 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
101 stats->tx_bytes += tbytes;
102 stats->tx_packets += tpkts;
103 stats->tx_dropped += tdrops;
104 stats->rx_bytes += rbytes;
105 stats->rx_packets += rpkts;
109 /* Local traffic destined to local address. Reinsert the packet to rx
110 * path, similar to loopback handling.
112 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
113 struct dst_entry *dst)
115 int len = skb->len;
117 skb_orphan(skb);
119 skb_dst_set(skb, dst);
120 skb_dst_force(skb);
122 /* set pkt_type to avoid skb hitting packet taps twice -
123 * once on Tx and again in Rx processing
125 skb->pkt_type = PACKET_LOOPBACK;
127 skb->protocol = eth_type_trans(skb, dev);
129 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
130 vrf_rx_stats(dev, len);
131 else
132 this_cpu_inc(dev->dstats->rx_drps);
134 return NETDEV_TX_OK;
137 #if IS_ENABLED(CONFIG_IPV6)
138 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
139 struct sk_buff *skb)
141 int err;
143 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
144 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
146 if (likely(err == 1))
147 err = dst_output(net, sk, skb);
149 return err;
152 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
153 struct net_device *dev)
155 const struct ipv6hdr *iph = ipv6_hdr(skb);
156 struct net *net = dev_net(skb->dev);
157 struct flowi6 fl6 = {
158 /* needed to match OIF rule */
159 .flowi6_oif = dev->ifindex,
160 .flowi6_iif = LOOPBACK_IFINDEX,
161 .daddr = iph->daddr,
162 .saddr = iph->saddr,
163 .flowlabel = ip6_flowinfo(iph),
164 .flowi6_mark = skb->mark,
165 .flowi6_proto = iph->nexthdr,
166 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
168 int ret = NET_XMIT_DROP;
169 struct dst_entry *dst;
170 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
172 dst = ip6_route_output(net, NULL, &fl6);
173 if (dst == dst_null)
174 goto err;
176 skb_dst_drop(skb);
178 /* if dst.dev is loopback or the VRF device again this is locally
179 * originated traffic destined to a local address. Short circuit
180 * to Rx path using our local dst
182 if (dst->dev == net->loopback_dev || dst->dev == dev) {
183 struct net_vrf *vrf = netdev_priv(dev);
184 struct rt6_info *rt6_local;
186 /* release looked up dst and use cached local dst */
187 dst_release(dst);
189 rcu_read_lock();
191 rt6_local = rcu_dereference(vrf->rt6_local);
192 if (unlikely(!rt6_local)) {
193 rcu_read_unlock();
194 goto err;
197 /* Ordering issue: cached local dst is created on newlink
198 * before the IPv6 initialization. Using the local dst
199 * requires rt6i_idev to be set so make sure it is.
201 if (unlikely(!rt6_local->rt6i_idev)) {
202 rt6_local->rt6i_idev = in6_dev_get(dev);
203 if (!rt6_local->rt6i_idev) {
204 rcu_read_unlock();
205 goto err;
209 dst = &rt6_local->dst;
210 dst_hold(dst);
212 rcu_read_unlock();
214 return vrf_local_xmit(skb, dev, &rt6_local->dst);
217 skb_dst_set(skb, dst);
219 /* strip the ethernet header added for pass through VRF device */
220 __skb_pull(skb, skb_network_offset(skb));
222 ret = vrf_ip6_local_out(net, skb->sk, skb);
223 if (unlikely(net_xmit_eval(ret)))
224 dev->stats.tx_errors++;
225 else
226 ret = NET_XMIT_SUCCESS;
228 return ret;
229 err:
230 vrf_tx_error(dev, skb);
231 return NET_XMIT_DROP;
233 #else
234 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
235 struct net_device *dev)
237 vrf_tx_error(dev, skb);
238 return NET_XMIT_DROP;
240 #endif
242 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
243 static int vrf_ip_local_out(struct net *net, struct sock *sk,
244 struct sk_buff *skb)
246 int err;
248 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
249 skb, NULL, skb_dst(skb)->dev, dst_output);
250 if (likely(err == 1))
251 err = dst_output(net, sk, skb);
253 return err;
256 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
257 struct net_device *vrf_dev)
259 struct iphdr *ip4h = ip_hdr(skb);
260 int ret = NET_XMIT_DROP;
261 struct flowi4 fl4 = {
262 /* needed to match OIF rule */
263 .flowi4_oif = vrf_dev->ifindex,
264 .flowi4_iif = LOOPBACK_IFINDEX,
265 .flowi4_tos = RT_TOS(ip4h->tos),
266 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
267 .flowi4_proto = ip4h->protocol,
268 .daddr = ip4h->daddr,
269 .saddr = ip4h->saddr,
271 struct net *net = dev_net(vrf_dev);
272 struct rtable *rt;
274 rt = ip_route_output_flow(net, &fl4, NULL);
275 if (IS_ERR(rt))
276 goto err;
278 skb_dst_drop(skb);
280 /* if dst.dev is loopback or the VRF device again this is locally
281 * originated traffic destined to a local address. Short circuit
282 * to Rx path using our local dst
284 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) {
285 struct net_vrf *vrf = netdev_priv(vrf_dev);
286 struct rtable *rth_local;
287 struct dst_entry *dst = NULL;
289 ip_rt_put(rt);
291 rcu_read_lock();
293 rth_local = rcu_dereference(vrf->rth_local);
294 if (likely(rth_local)) {
295 dst = &rth_local->dst;
296 dst_hold(dst);
299 rcu_read_unlock();
301 if (unlikely(!dst))
302 goto err;
304 return vrf_local_xmit(skb, vrf_dev, dst);
307 skb_dst_set(skb, &rt->dst);
309 /* strip the ethernet header added for pass through VRF device */
310 __skb_pull(skb, skb_network_offset(skb));
312 if (!ip4h->saddr) {
313 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
314 RT_SCOPE_LINK);
317 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
318 if (unlikely(net_xmit_eval(ret)))
319 vrf_dev->stats.tx_errors++;
320 else
321 ret = NET_XMIT_SUCCESS;
323 out:
324 return ret;
325 err:
326 vrf_tx_error(vrf_dev, skb);
327 goto out;
330 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
332 switch (skb->protocol) {
333 case htons(ETH_P_IP):
334 return vrf_process_v4_outbound(skb, dev);
335 case htons(ETH_P_IPV6):
336 return vrf_process_v6_outbound(skb, dev);
337 default:
338 vrf_tx_error(dev, skb);
339 return NET_XMIT_DROP;
343 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
345 int len = skb->len;
346 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
348 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
349 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
351 u64_stats_update_begin(&dstats->syncp);
352 dstats->tx_pkts++;
353 dstats->tx_bytes += len;
354 u64_stats_update_end(&dstats->syncp);
355 } else {
356 this_cpu_inc(dev->dstats->tx_drps);
359 return ret;
362 #if IS_ENABLED(CONFIG_IPV6)
363 /* modelled after ip6_finish_output2 */
364 static int vrf_finish_output6(struct net *net, struct sock *sk,
365 struct sk_buff *skb)
367 struct dst_entry *dst = skb_dst(skb);
368 struct net_device *dev = dst->dev;
369 struct neighbour *neigh;
370 struct in6_addr *nexthop;
371 int ret;
373 nf_reset(skb);
375 skb->protocol = htons(ETH_P_IPV6);
376 skb->dev = dev;
378 rcu_read_lock_bh();
379 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
380 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
381 if (unlikely(!neigh))
382 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
383 if (!IS_ERR(neigh)) {
384 sock_confirm_neigh(skb, neigh);
385 ret = neigh_output(neigh, skb);
386 rcu_read_unlock_bh();
387 return ret;
389 rcu_read_unlock_bh();
391 IP6_INC_STATS(dev_net(dst->dev),
392 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
393 kfree_skb(skb);
394 return -EINVAL;
397 /* modelled after ip6_output */
398 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
400 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
401 net, sk, skb, NULL, skb_dst(skb)->dev,
402 vrf_finish_output6,
403 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
406 /* set dst on skb to send packet to us via dev_xmit path. Allows
407 * packet to go through device based features such as qdisc, netfilter
408 * hooks and packet sockets with skb->dev set to vrf device.
410 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
411 struct sock *sk,
412 struct sk_buff *skb)
414 struct net_vrf *vrf = netdev_priv(vrf_dev);
415 struct dst_entry *dst = NULL;
416 struct rt6_info *rt6;
418 /* don't divert link scope packets */
419 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
420 return skb;
422 rcu_read_lock();
424 rt6 = rcu_dereference(vrf->rt6);
425 if (likely(rt6)) {
426 dst = &rt6->dst;
427 dst_hold(dst);
430 rcu_read_unlock();
432 if (unlikely(!dst)) {
433 vrf_tx_error(vrf_dev, skb);
434 return NULL;
437 skb_dst_drop(skb);
438 skb_dst_set(skb, dst);
440 return skb;
443 /* holding rtnl */
444 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
446 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
447 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local);
448 struct net *net = dev_net(dev);
449 struct dst_entry *dst;
451 RCU_INIT_POINTER(vrf->rt6, NULL);
452 RCU_INIT_POINTER(vrf->rt6_local, NULL);
453 synchronize_rcu();
455 /* move dev in dst's to loopback so this VRF device can be deleted
456 * - based on dst_ifdown
458 if (rt6) {
459 dst = &rt6->dst;
460 dev_put(dst->dev);
461 dst->dev = net->loopback_dev;
462 dev_hold(dst->dev);
463 dst_release(dst);
466 if (rt6_local) {
467 if (rt6_local->rt6i_idev) {
468 in6_dev_put(rt6_local->rt6i_idev);
469 rt6_local->rt6i_idev = NULL;
472 dst = &rt6_local->dst;
473 dev_put(dst->dev);
474 dst->dev = net->loopback_dev;
475 dev_hold(dst->dev);
476 dst_release(dst);
480 static int vrf_rt6_create(struct net_device *dev)
482 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE;
483 struct net_vrf *vrf = netdev_priv(dev);
484 struct net *net = dev_net(dev);
485 struct fib6_table *rt6i_table;
486 struct rt6_info *rt6, *rt6_local;
487 int rc = -ENOMEM;
489 /* IPv6 can be CONFIG enabled and then disabled runtime */
490 if (!ipv6_mod_enabled())
491 return 0;
493 rt6i_table = fib6_new_table(net, vrf->tb_id);
494 if (!rt6i_table)
495 goto out;
497 /* create a dst for routing packets out a VRF device */
498 rt6 = ip6_dst_alloc(net, dev, flags);
499 if (!rt6)
500 goto out;
502 dst_hold(&rt6->dst);
504 rt6->rt6i_table = rt6i_table;
505 rt6->dst.output = vrf_output6;
507 /* create a dst for local routing - packets sent locally
508 * to local address via the VRF device as a loopback
510 rt6_local = ip6_dst_alloc(net, dev, flags);
511 if (!rt6_local) {
512 dst_release(&rt6->dst);
513 goto out;
516 dst_hold(&rt6_local->dst);
518 rt6_local->rt6i_idev = in6_dev_get(dev);
519 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL;
520 rt6_local->rt6i_table = rt6i_table;
521 rt6_local->dst.input = ip6_input;
523 rcu_assign_pointer(vrf->rt6, rt6);
524 rcu_assign_pointer(vrf->rt6_local, rt6_local);
526 rc = 0;
527 out:
528 return rc;
530 #else
531 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
532 struct sock *sk,
533 struct sk_buff *skb)
535 return skb;
538 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
542 static int vrf_rt6_create(struct net_device *dev)
544 return 0;
546 #endif
548 /* modelled after ip_finish_output2 */
549 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
551 struct dst_entry *dst = skb_dst(skb);
552 struct rtable *rt = (struct rtable *)dst;
553 struct net_device *dev = dst->dev;
554 unsigned int hh_len = LL_RESERVED_SPACE(dev);
555 struct neighbour *neigh;
556 u32 nexthop;
557 int ret = -EINVAL;
559 nf_reset(skb);
561 /* Be paranoid, rather than too clever. */
562 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
563 struct sk_buff *skb2;
565 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
566 if (!skb2) {
567 ret = -ENOMEM;
568 goto err;
570 if (skb->sk)
571 skb_set_owner_w(skb2, skb->sk);
573 consume_skb(skb);
574 skb = skb2;
577 rcu_read_lock_bh();
579 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
580 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
581 if (unlikely(!neigh))
582 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
583 if (!IS_ERR(neigh)) {
584 sock_confirm_neigh(skb, neigh);
585 ret = neigh_output(neigh, skb);
588 rcu_read_unlock_bh();
589 err:
590 if (unlikely(ret < 0))
591 vrf_tx_error(skb->dev, skb);
592 return ret;
595 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
597 struct net_device *dev = skb_dst(skb)->dev;
599 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
601 skb->dev = dev;
602 skb->protocol = htons(ETH_P_IP);
604 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
605 net, sk, skb, NULL, dev,
606 vrf_finish_output,
607 !(IPCB(skb)->flags & IPSKB_REROUTED));
610 /* set dst on skb to send packet to us via dev_xmit path. Allows
611 * packet to go through device based features such as qdisc, netfilter
612 * hooks and packet sockets with skb->dev set to vrf device.
614 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
615 struct sock *sk,
616 struct sk_buff *skb)
618 struct net_vrf *vrf = netdev_priv(vrf_dev);
619 struct dst_entry *dst = NULL;
620 struct rtable *rth;
622 /* don't divert multicast */
623 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
624 return skb;
626 rcu_read_lock();
628 rth = rcu_dereference(vrf->rth);
629 if (likely(rth)) {
630 dst = &rth->dst;
631 dst_hold(dst);
634 rcu_read_unlock();
636 if (unlikely(!dst)) {
637 vrf_tx_error(vrf_dev, skb);
638 return NULL;
641 skb_dst_drop(skb);
642 skb_dst_set(skb, dst);
644 return skb;
647 /* called with rcu lock held */
648 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
649 struct sock *sk,
650 struct sk_buff *skb,
651 u16 proto)
653 switch (proto) {
654 case AF_INET:
655 return vrf_ip_out(vrf_dev, sk, skb);
656 case AF_INET6:
657 return vrf_ip6_out(vrf_dev, sk, skb);
660 return skb;
663 /* holding rtnl */
664 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
666 struct rtable *rth = rtnl_dereference(vrf->rth);
667 struct rtable *rth_local = rtnl_dereference(vrf->rth_local);
668 struct net *net = dev_net(dev);
669 struct dst_entry *dst;
671 RCU_INIT_POINTER(vrf->rth, NULL);
672 RCU_INIT_POINTER(vrf->rth_local, NULL);
673 synchronize_rcu();
675 /* move dev in dst's to loopback so this VRF device can be deleted
676 * - based on dst_ifdown
678 if (rth) {
679 dst = &rth->dst;
680 dev_put(dst->dev);
681 dst->dev = net->loopback_dev;
682 dev_hold(dst->dev);
683 dst_release(dst);
686 if (rth_local) {
687 dst = &rth_local->dst;
688 dev_put(dst->dev);
689 dst->dev = net->loopback_dev;
690 dev_hold(dst->dev);
691 dst_release(dst);
695 static int vrf_rtable_create(struct net_device *dev)
697 struct net_vrf *vrf = netdev_priv(dev);
698 struct rtable *rth, *rth_local;
700 if (!fib_new_table(dev_net(dev), vrf->tb_id))
701 return -ENOMEM;
703 /* create a dst for routing packets out through a VRF device */
704 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
705 if (!rth)
706 return -ENOMEM;
708 /* create a dst for local ingress routing - packets sent locally
709 * to local address via the VRF device as a loopback
711 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0);
712 if (!rth_local) {
713 dst_release(&rth->dst);
714 return -ENOMEM;
717 rth->dst.output = vrf_output;
718 rth->rt_table_id = vrf->tb_id;
720 rth_local->rt_table_id = vrf->tb_id;
722 rcu_assign_pointer(vrf->rth, rth);
723 rcu_assign_pointer(vrf->rth_local, rth_local);
725 return 0;
728 /**************************** device handling ********************/
730 /* cycle interface to flush neighbor cache and move routes across tables */
731 static void cycle_netdev(struct net_device *dev)
733 unsigned int flags = dev->flags;
734 int ret;
736 if (!netif_running(dev))
737 return;
739 ret = dev_change_flags(dev, flags & ~IFF_UP);
740 if (ret >= 0)
741 ret = dev_change_flags(dev, flags);
743 if (ret < 0) {
744 netdev_err(dev,
745 "Failed to cycle device %s; route tables might be wrong!\n",
746 dev->name);
750 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
752 int ret;
754 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
755 if (ret < 0)
756 return ret;
758 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
759 cycle_netdev(port_dev);
761 return 0;
764 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
766 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
767 return -EINVAL;
769 return do_vrf_add_slave(dev, port_dev);
772 /* inverse of do_vrf_add_slave */
773 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
775 netdev_upper_dev_unlink(port_dev, dev);
776 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
778 cycle_netdev(port_dev);
780 return 0;
783 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
785 return do_vrf_del_slave(dev, port_dev);
788 static void vrf_dev_uninit(struct net_device *dev)
790 struct net_vrf *vrf = netdev_priv(dev);
791 struct net_device *port_dev;
792 struct list_head *iter;
794 vrf_rtable_release(dev, vrf);
795 vrf_rt6_release(dev, vrf);
797 netdev_for_each_lower_dev(dev, port_dev, iter)
798 vrf_del_slave(dev, port_dev);
800 free_percpu(dev->dstats);
801 dev->dstats = NULL;
804 static int vrf_dev_init(struct net_device *dev)
806 struct net_vrf *vrf = netdev_priv(dev);
808 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
809 if (!dev->dstats)
810 goto out_nomem;
812 /* create the default dst which points back to us */
813 if (vrf_rtable_create(dev) != 0)
814 goto out_stats;
816 if (vrf_rt6_create(dev) != 0)
817 goto out_rth;
819 dev->flags = IFF_MASTER | IFF_NOARP;
821 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
822 dev->mtu = 64 * 1024;
824 /* similarly, oper state is irrelevant; set to up to avoid confusion */
825 dev->operstate = IF_OPER_UP;
826 netdev_lockdep_set_classes(dev);
827 return 0;
829 out_rth:
830 vrf_rtable_release(dev, vrf);
831 out_stats:
832 free_percpu(dev->dstats);
833 dev->dstats = NULL;
834 out_nomem:
835 return -ENOMEM;
838 static const struct net_device_ops vrf_netdev_ops = {
839 .ndo_init = vrf_dev_init,
840 .ndo_uninit = vrf_dev_uninit,
841 .ndo_start_xmit = vrf_xmit,
842 .ndo_get_stats64 = vrf_get_stats64,
843 .ndo_add_slave = vrf_add_slave,
844 .ndo_del_slave = vrf_del_slave,
847 static u32 vrf_fib_table(const struct net_device *dev)
849 struct net_vrf *vrf = netdev_priv(dev);
851 return vrf->tb_id;
854 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
856 kfree_skb(skb);
857 return 0;
860 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
861 struct sk_buff *skb,
862 struct net_device *dev)
864 struct net *net = dev_net(dev);
866 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
867 skb = NULL; /* kfree_skb(skb) handled by nf code */
869 return skb;
872 #if IS_ENABLED(CONFIG_IPV6)
873 /* neighbor handling is done with actual device; do not want
874 * to flip skb->dev for those ndisc packets. This really fails
875 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
876 * a start.
878 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
880 const struct ipv6hdr *iph = ipv6_hdr(skb);
881 bool rc = false;
883 if (iph->nexthdr == NEXTHDR_ICMP) {
884 const struct icmp6hdr *icmph;
885 struct icmp6hdr _icmph;
887 icmph = skb_header_pointer(skb, sizeof(*iph),
888 sizeof(_icmph), &_icmph);
889 if (!icmph)
890 goto out;
892 switch (icmph->icmp6_type) {
893 case NDISC_ROUTER_SOLICITATION:
894 case NDISC_ROUTER_ADVERTISEMENT:
895 case NDISC_NEIGHBOUR_SOLICITATION:
896 case NDISC_NEIGHBOUR_ADVERTISEMENT:
897 case NDISC_REDIRECT:
898 rc = true;
899 break;
903 out:
904 return rc;
907 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
908 const struct net_device *dev,
909 struct flowi6 *fl6,
910 int ifindex,
911 int flags)
913 struct net_vrf *vrf = netdev_priv(dev);
914 struct fib6_table *table = NULL;
915 struct rt6_info *rt6;
917 rcu_read_lock();
919 /* fib6_table does not have a refcnt and can not be freed */
920 rt6 = rcu_dereference(vrf->rt6);
921 if (likely(rt6))
922 table = rt6->rt6i_table;
924 rcu_read_unlock();
926 if (!table)
927 return NULL;
929 return ip6_pol_route(net, table, ifindex, fl6, flags);
932 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
933 int ifindex)
935 const struct ipv6hdr *iph = ipv6_hdr(skb);
936 struct flowi6 fl6 = {
937 .daddr = iph->daddr,
938 .saddr = iph->saddr,
939 .flowlabel = ip6_flowinfo(iph),
940 .flowi6_mark = skb->mark,
941 .flowi6_proto = iph->nexthdr,
942 .flowi6_iif = ifindex,
944 struct net *net = dev_net(vrf_dev);
945 struct rt6_info *rt6;
947 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
948 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
949 if (unlikely(!rt6))
950 return;
952 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
953 return;
955 skb_dst_set(skb, &rt6->dst);
958 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
959 struct sk_buff *skb)
961 int orig_iif = skb->skb_iif;
962 bool need_strict;
964 /* loopback traffic; do not push through packet taps again.
965 * Reset pkt_type for upper layers to process skb
967 if (skb->pkt_type == PACKET_LOOPBACK) {
968 skb->dev = vrf_dev;
969 skb->skb_iif = vrf_dev->ifindex;
970 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
971 skb->pkt_type = PACKET_HOST;
972 goto out;
975 /* if packet is NDISC or addressed to multicast or link-local
976 * then keep the ingress interface
978 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
979 if (!ipv6_ndisc_frame(skb) && !need_strict) {
980 vrf_rx_stats(vrf_dev, skb->len);
981 skb->dev = vrf_dev;
982 skb->skb_iif = vrf_dev->ifindex;
984 skb_push(skb, skb->mac_len);
985 dev_queue_xmit_nit(skb, vrf_dev);
986 skb_pull(skb, skb->mac_len);
988 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
991 if (need_strict)
992 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
994 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
995 out:
996 return skb;
999 #else
1000 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1001 struct sk_buff *skb)
1003 return skb;
1005 #endif
1007 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1008 struct sk_buff *skb)
1010 skb->dev = vrf_dev;
1011 skb->skb_iif = vrf_dev->ifindex;
1012 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1014 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1015 goto out;
1017 /* loopback traffic; do not push through packet taps again.
1018 * Reset pkt_type for upper layers to process skb
1020 if (skb->pkt_type == PACKET_LOOPBACK) {
1021 skb->pkt_type = PACKET_HOST;
1022 goto out;
1025 vrf_rx_stats(vrf_dev, skb->len);
1027 skb_push(skb, skb->mac_len);
1028 dev_queue_xmit_nit(skb, vrf_dev);
1029 skb_pull(skb, skb->mac_len);
1031 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1032 out:
1033 return skb;
1036 /* called with rcu lock held */
1037 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1038 struct sk_buff *skb,
1039 u16 proto)
1041 switch (proto) {
1042 case AF_INET:
1043 return vrf_ip_rcv(vrf_dev, skb);
1044 case AF_INET6:
1045 return vrf_ip6_rcv(vrf_dev, skb);
1048 return skb;
1051 #if IS_ENABLED(CONFIG_IPV6)
1052 /* send to link-local or multicast address via interface enslaved to
1053 * VRF device. Force lookup to VRF table without changing flow struct
1055 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1056 struct flowi6 *fl6)
1058 struct net *net = dev_net(dev);
1059 int flags = RT6_LOOKUP_F_IFACE;
1060 struct dst_entry *dst = NULL;
1061 struct rt6_info *rt;
1063 /* VRF device does not have a link-local address and
1064 * sending packets to link-local or mcast addresses over
1065 * a VRF device does not make sense
1067 if (fl6->flowi6_oif == dev->ifindex) {
1068 dst = &net->ipv6.ip6_null_entry->dst;
1069 dst_hold(dst);
1070 return dst;
1073 if (!ipv6_addr_any(&fl6->saddr))
1074 flags |= RT6_LOOKUP_F_HAS_SADDR;
1076 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1077 if (rt)
1078 dst = &rt->dst;
1080 return dst;
1082 #endif
1084 static const struct l3mdev_ops vrf_l3mdev_ops = {
1085 .l3mdev_fib_table = vrf_fib_table,
1086 .l3mdev_l3_rcv = vrf_l3_rcv,
1087 .l3mdev_l3_out = vrf_l3_out,
1088 #if IS_ENABLED(CONFIG_IPV6)
1089 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1090 #endif
1093 static void vrf_get_drvinfo(struct net_device *dev,
1094 struct ethtool_drvinfo *info)
1096 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1097 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1100 static const struct ethtool_ops vrf_ethtool_ops = {
1101 .get_drvinfo = vrf_get_drvinfo,
1104 static inline size_t vrf_fib_rule_nl_size(void)
1106 size_t sz;
1108 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1109 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1110 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1112 return sz;
1115 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1117 struct fib_rule_hdr *frh;
1118 struct nlmsghdr *nlh;
1119 struct sk_buff *skb;
1120 int err;
1122 if (family == AF_INET6 && !ipv6_mod_enabled())
1123 return 0;
1125 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1126 if (!skb)
1127 return -ENOMEM;
1129 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1130 if (!nlh)
1131 goto nla_put_failure;
1133 /* rule only needs to appear once */
1134 nlh->nlmsg_flags |= NLM_F_EXCL;
1136 frh = nlmsg_data(nlh);
1137 memset(frh, 0, sizeof(*frh));
1138 frh->family = family;
1139 frh->action = FR_ACT_TO_TBL;
1141 if (nla_put_u32(skb, FRA_L3MDEV, 1))
1142 goto nla_put_failure;
1144 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1145 goto nla_put_failure;
1147 nlmsg_end(skb, nlh);
1149 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1150 skb->sk = dev_net(dev)->rtnl;
1151 if (add_it) {
1152 err = fib_nl_newrule(skb, nlh);
1153 if (err == -EEXIST)
1154 err = 0;
1155 } else {
1156 err = fib_nl_delrule(skb, nlh);
1157 if (err == -ENOENT)
1158 err = 0;
1160 nlmsg_free(skb);
1162 return err;
1164 nla_put_failure:
1165 nlmsg_free(skb);
1167 return -EMSGSIZE;
1170 static int vrf_add_fib_rules(const struct net_device *dev)
1172 int err;
1174 err = vrf_fib_rule(dev, AF_INET, true);
1175 if (err < 0)
1176 goto out_err;
1178 err = vrf_fib_rule(dev, AF_INET6, true);
1179 if (err < 0)
1180 goto ipv6_err;
1182 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1183 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1184 if (err < 0)
1185 goto ipmr_err;
1186 #endif
1188 return 0;
1190 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1191 ipmr_err:
1192 vrf_fib_rule(dev, AF_INET6, false);
1193 #endif
1195 ipv6_err:
1196 vrf_fib_rule(dev, AF_INET, false);
1198 out_err:
1199 netdev_err(dev, "Failed to add FIB rules.\n");
1200 return err;
1203 static void vrf_setup(struct net_device *dev)
1205 ether_setup(dev);
1207 /* Initialize the device structure. */
1208 dev->netdev_ops = &vrf_netdev_ops;
1209 dev->l3mdev_ops = &vrf_l3mdev_ops;
1210 dev->ethtool_ops = &vrf_ethtool_ops;
1211 dev->needs_free_netdev = true;
1213 /* Fill in device structure with ethernet-generic values. */
1214 eth_hw_addr_random(dev);
1216 /* don't acquire vrf device's netif_tx_lock when transmitting */
1217 dev->features |= NETIF_F_LLTX;
1219 /* don't allow vrf devices to change network namespaces. */
1220 dev->features |= NETIF_F_NETNS_LOCAL;
1222 /* does not make sense for a VLAN to be added to a vrf device */
1223 dev->features |= NETIF_F_VLAN_CHALLENGED;
1225 /* enable offload features */
1226 dev->features |= NETIF_F_GSO_SOFTWARE;
1227 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1228 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1230 dev->hw_features = dev->features;
1231 dev->hw_enc_features = dev->features;
1233 /* default to no qdisc; user can add if desired */
1234 dev->priv_flags |= IFF_NO_QUEUE;
1237 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
1239 if (tb[IFLA_ADDRESS]) {
1240 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1241 return -EINVAL;
1242 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1243 return -EADDRNOTAVAIL;
1245 return 0;
1248 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1250 unregister_netdevice_queue(dev, head);
1253 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1254 struct nlattr *tb[], struct nlattr *data[])
1256 struct net_vrf *vrf = netdev_priv(dev);
1257 bool *add_fib_rules;
1258 struct net *net;
1259 int err;
1261 if (!data || !data[IFLA_VRF_TABLE])
1262 return -EINVAL;
1264 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1265 if (vrf->tb_id == RT_TABLE_UNSPEC)
1266 return -EINVAL;
1268 dev->priv_flags |= IFF_L3MDEV_MASTER;
1270 err = register_netdevice(dev);
1271 if (err)
1272 goto out;
1274 net = dev_net(dev);
1275 add_fib_rules = net_generic(net, vrf_net_id);
1276 if (*add_fib_rules) {
1277 err = vrf_add_fib_rules(dev);
1278 if (err) {
1279 unregister_netdevice(dev);
1280 goto out;
1282 *add_fib_rules = false;
1285 out:
1286 return err;
1289 static size_t vrf_nl_getsize(const struct net_device *dev)
1291 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1294 static int vrf_fillinfo(struct sk_buff *skb,
1295 const struct net_device *dev)
1297 struct net_vrf *vrf = netdev_priv(dev);
1299 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1302 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1303 const struct net_device *slave_dev)
1305 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1308 static int vrf_fill_slave_info(struct sk_buff *skb,
1309 const struct net_device *vrf_dev,
1310 const struct net_device *slave_dev)
1312 struct net_vrf *vrf = netdev_priv(vrf_dev);
1314 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1315 return -EMSGSIZE;
1317 return 0;
1320 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1321 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1324 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1325 .kind = DRV_NAME,
1326 .priv_size = sizeof(struct net_vrf),
1328 .get_size = vrf_nl_getsize,
1329 .policy = vrf_nl_policy,
1330 .validate = vrf_validate,
1331 .fill_info = vrf_fillinfo,
1333 .get_slave_size = vrf_get_slave_size,
1334 .fill_slave_info = vrf_fill_slave_info,
1336 .newlink = vrf_newlink,
1337 .dellink = vrf_dellink,
1338 .setup = vrf_setup,
1339 .maxtype = IFLA_VRF_MAX,
1342 static int vrf_device_event(struct notifier_block *unused,
1343 unsigned long event, void *ptr)
1345 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1347 /* only care about unregister events to drop slave references */
1348 if (event == NETDEV_UNREGISTER) {
1349 struct net_device *vrf_dev;
1351 if (!netif_is_l3_slave(dev))
1352 goto out;
1354 vrf_dev = netdev_master_upper_dev_get(dev);
1355 vrf_del_slave(vrf_dev, dev);
1357 out:
1358 return NOTIFY_DONE;
1361 static struct notifier_block vrf_notifier_block __read_mostly = {
1362 .notifier_call = vrf_device_event,
1365 /* Initialize per network namespace state */
1366 static int __net_init vrf_netns_init(struct net *net)
1368 bool *add_fib_rules = net_generic(net, vrf_net_id);
1370 *add_fib_rules = true;
1372 return 0;
1375 static struct pernet_operations vrf_net_ops __net_initdata = {
1376 .init = vrf_netns_init,
1377 .id = &vrf_net_id,
1378 .size = sizeof(bool),
1381 static int __init vrf_init_module(void)
1383 int rc;
1385 register_netdevice_notifier(&vrf_notifier_block);
1387 rc = register_pernet_subsys(&vrf_net_ops);
1388 if (rc < 0)
1389 goto error;
1391 rc = rtnl_link_register(&vrf_link_ops);
1392 if (rc < 0) {
1393 unregister_pernet_subsys(&vrf_net_ops);
1394 goto error;
1397 return 0;
1399 error:
1400 unregister_netdevice_notifier(&vrf_notifier_block);
1401 return rc;
1404 module_init(vrf_init_module);
1405 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1406 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1407 MODULE_LICENSE("GPL");
1408 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1409 MODULE_VERSION(DRV_VERSION);