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>
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>
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>
40 #define RT_FL_TOS(oldflp4) \
41 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
43 #define DRV_NAME "vrf"
44 #define DRV_VERSION "1.0"
46 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */
47 static bool add_fib_rules = true;
50 struct rtable __rcu *rth;
51 struct rtable __rcu *rth_local;
52 struct rt6_info __rcu *rt6;
53 struct rt6_info __rcu *rt6_local;
64 struct u64_stats_sync syncp;
67 static void vrf_rx_stats(struct net_device *dev, int len)
69 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
71 u64_stats_update_begin(&dstats->syncp);
73 dstats->rx_bytes += len;
74 u64_stats_update_end(&dstats->syncp);
77 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
79 vrf_dev->stats.tx_errors++;
83 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
84 struct rtnl_link_stats64 *stats)
88 for_each_possible_cpu(i) {
89 const struct pcpu_dstats *dstats;
90 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
93 dstats = per_cpu_ptr(dev->dstats, i);
95 start = u64_stats_fetch_begin_irq(&dstats->syncp);
96 tbytes = dstats->tx_bytes;
97 tpkts = dstats->tx_pkts;
98 tdrops = dstats->tx_drps;
99 rbytes = dstats->rx_bytes;
100 rpkts = dstats->rx_pkts;
101 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
102 stats->tx_bytes += tbytes;
103 stats->tx_packets += tpkts;
104 stats->tx_dropped += tdrops;
105 stats->rx_bytes += rbytes;
106 stats->rx_packets += rpkts;
111 /* Local traffic destined to local address. Reinsert the packet to rx
112 * path, similar to loopback handling.
114 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
115 struct dst_entry *dst)
121 skb_dst_set(skb, dst);
124 /* set pkt_type to avoid skb hitting packet taps twice -
125 * once on Tx and again in Rx processing
127 skb->pkt_type = PACKET_LOOPBACK;
129 skb->protocol = eth_type_trans(skb, dev);
131 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
132 vrf_rx_stats(dev, len);
134 this_cpu_inc(dev->dstats->rx_drps);
139 #if IS_ENABLED(CONFIG_IPV6)
140 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
145 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
146 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
148 if (likely(err == 1))
149 err = dst_output(net, sk, skb);
154 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
155 struct net_device *dev)
157 const struct ipv6hdr *iph = ipv6_hdr(skb);
158 struct net *net = dev_net(skb->dev);
159 struct flowi6 fl6 = {
160 /* needed to match OIF rule */
161 .flowi6_oif = dev->ifindex,
162 .flowi6_iif = LOOPBACK_IFINDEX,
165 .flowlabel = ip6_flowinfo(iph),
166 .flowi6_mark = skb->mark,
167 .flowi6_proto = iph->nexthdr,
168 .flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF,
170 int ret = NET_XMIT_DROP;
171 struct dst_entry *dst;
172 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
174 dst = ip6_route_output(net, NULL, &fl6);
180 /* if dst.dev is loopback or the VRF device again this is locally
181 * originated traffic destined to a local address. Short circuit
182 * to Rx path using our local dst
184 if (dst->dev == net->loopback_dev || dst->dev == dev) {
185 struct net_vrf *vrf = netdev_priv(dev);
186 struct rt6_info *rt6_local;
188 /* release looked up dst and use cached local dst */
193 rt6_local = rcu_dereference(vrf->rt6_local);
194 if (unlikely(!rt6_local)) {
199 /* Ordering issue: cached local dst is created on newlink
200 * before the IPv6 initialization. Using the local dst
201 * requires rt6i_idev to be set so make sure it is.
203 if (unlikely(!rt6_local->rt6i_idev)) {
204 rt6_local->rt6i_idev = in6_dev_get(dev);
205 if (!rt6_local->rt6i_idev) {
211 dst = &rt6_local->dst;
216 return vrf_local_xmit(skb, dev, &rt6_local->dst);
219 skb_dst_set(skb, dst);
221 /* strip the ethernet header added for pass through VRF device */
222 __skb_pull(skb, skb_network_offset(skb));
224 ret = vrf_ip6_local_out(net, skb->sk, skb);
225 if (unlikely(net_xmit_eval(ret)))
226 dev->stats.tx_errors++;
228 ret = NET_XMIT_SUCCESS;
232 vrf_tx_error(dev, skb);
233 return NET_XMIT_DROP;
236 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
237 struct net_device *dev)
239 vrf_tx_error(dev, skb);
240 return NET_XMIT_DROP;
244 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
245 static int vrf_ip_local_out(struct net *net, struct sock *sk,
250 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
251 skb, NULL, skb_dst(skb)->dev, dst_output);
252 if (likely(err == 1))
253 err = dst_output(net, sk, skb);
258 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
259 struct net_device *vrf_dev)
261 struct iphdr *ip4h = ip_hdr(skb);
262 int ret = NET_XMIT_DROP;
263 struct flowi4 fl4 = {
264 /* needed to match OIF rule */
265 .flowi4_oif = vrf_dev->ifindex,
266 .flowi4_iif = LOOPBACK_IFINDEX,
267 .flowi4_tos = RT_TOS(ip4h->tos),
268 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
269 FLOWI_FLAG_SKIP_NH_OIF,
270 .daddr = ip4h->daddr,
272 struct net *net = dev_net(vrf_dev);
275 rt = ip_route_output_flow(net, &fl4, NULL);
279 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
286 /* if dst.dev is loopback or the VRF device again this is locally
287 * originated traffic destined to a local address. Short circuit
288 * to Rx path using our local dst
290 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) {
291 struct net_vrf *vrf = netdev_priv(vrf_dev);
292 struct rtable *rth_local;
293 struct dst_entry *dst = NULL;
299 rth_local = rcu_dereference(vrf->rth_local);
300 if (likely(rth_local)) {
301 dst = &rth_local->dst;
310 return vrf_local_xmit(skb, vrf_dev, dst);
313 skb_dst_set(skb, &rt->dst);
315 /* strip the ethernet header added for pass through VRF device */
316 __skb_pull(skb, skb_network_offset(skb));
319 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
323 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
324 if (unlikely(net_xmit_eval(ret)))
325 vrf_dev->stats.tx_errors++;
327 ret = NET_XMIT_SUCCESS;
332 vrf_tx_error(vrf_dev, skb);
336 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
338 switch (skb->protocol) {
339 case htons(ETH_P_IP):
340 return vrf_process_v4_outbound(skb, dev);
341 case htons(ETH_P_IPV6):
342 return vrf_process_v6_outbound(skb, dev);
344 vrf_tx_error(dev, skb);
345 return NET_XMIT_DROP;
349 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
351 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
353 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
354 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
356 u64_stats_update_begin(&dstats->syncp);
358 dstats->tx_bytes += skb->len;
359 u64_stats_update_end(&dstats->syncp);
361 this_cpu_inc(dev->dstats->tx_drps);
367 #if IS_ENABLED(CONFIG_IPV6)
368 /* modelled after ip6_finish_output2 */
369 static int vrf_finish_output6(struct net *net, struct sock *sk,
372 struct dst_entry *dst = skb_dst(skb);
373 struct net_device *dev = dst->dev;
374 struct neighbour *neigh;
375 struct in6_addr *nexthop;
378 skb->protocol = htons(ETH_P_IPV6);
382 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
383 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
384 if (unlikely(!neigh))
385 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
386 if (!IS_ERR(neigh)) {
387 ret = dst_neigh_output(dst, neigh, skb);
388 rcu_read_unlock_bh();
391 rcu_read_unlock_bh();
393 IP6_INC_STATS(dev_net(dst->dev),
394 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
399 /* modelled after ip6_output */
400 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
402 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
403 net, sk, skb, NULL, skb_dst(skb)->dev,
405 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
408 /* set dst on skb to send packet to us via dev_xmit path. Allows
409 * packet to go through device based features such as qdisc, netfilter
410 * hooks and packet sockets with skb->dev set to vrf device.
412 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
416 struct net_vrf *vrf = netdev_priv(vrf_dev);
417 struct dst_entry *dst = NULL;
418 struct rt6_info *rt6;
420 /* don't divert link scope packets */
421 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
426 rt6 = rcu_dereference(vrf->rt6);
434 if (unlikely(!dst)) {
435 vrf_tx_error(vrf_dev, skb);
440 skb_dst_set(skb, dst);
446 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
448 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
449 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local);
450 struct net *net = dev_net(dev);
451 struct dst_entry *dst;
453 RCU_INIT_POINTER(vrf->rt6, NULL);
454 RCU_INIT_POINTER(vrf->rt6_local, NULL);
457 /* move dev in dst's to loopback so this VRF device can be deleted
458 * - based on dst_ifdown
463 dst->dev = net->loopback_dev;
469 if (rt6_local->rt6i_idev)
470 in6_dev_put(rt6_local->rt6i_idev);
472 dst = &rt6_local->dst;
474 dst->dev = net->loopback_dev;
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;
489 /* IPv6 can be CONFIG enabled and then disabled runtime */
490 if (!ipv6_mod_enabled())
493 rt6i_table = fib6_new_table(net, vrf->tb_id);
497 /* create a dst for routing packets out a VRF device */
498 rt6 = ip6_dst_alloc(net, dev, flags);
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);
512 dst_release(&rt6->dst);
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);
531 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
538 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
542 static int vrf_rt6_create(struct net_device *dev)
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;
559 /* Be paranoid, rather than too clever. */
560 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
561 struct sk_buff *skb2;
563 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
569 skb_set_owner_w(skb2, skb->sk);
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);
582 ret = dst_neigh_output(dst, neigh, skb);
584 rcu_read_unlock_bh();
586 if (unlikely(ret < 0))
587 vrf_tx_error(skb->dev, skb);
591 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
593 struct net_device *dev = skb_dst(skb)->dev;
595 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
598 skb->protocol = htons(ETH_P_IP);
600 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
601 net, sk, skb, NULL, dev,
603 !(IPCB(skb)->flags & IPSKB_REROUTED));
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.
610 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
614 struct net_vrf *vrf = netdev_priv(vrf_dev);
615 struct dst_entry *dst = NULL;
620 rth = rcu_dereference(vrf->rth);
628 if (unlikely(!dst)) {
629 vrf_tx_error(vrf_dev, skb);
634 skb_dst_set(skb, dst);
639 /* called with rcu lock held */
640 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
647 return vrf_ip_out(vrf_dev, sk, skb);
649 return vrf_ip6_out(vrf_dev, sk, skb);
656 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
658 struct rtable *rth = rtnl_dereference(vrf->rth);
659 struct rtable *rth_local = rtnl_dereference(vrf->rth_local);
660 struct net *net = dev_net(dev);
661 struct dst_entry *dst;
663 RCU_INIT_POINTER(vrf->rth, NULL);
664 RCU_INIT_POINTER(vrf->rth_local, NULL);
667 /* move dev in dst's to loopback so this VRF device can be deleted
668 * - based on dst_ifdown
673 dst->dev = net->loopback_dev;
679 dst = &rth_local->dst;
681 dst->dev = net->loopback_dev;
687 static int vrf_rtable_create(struct net_device *dev)
689 struct net_vrf *vrf = netdev_priv(dev);
690 struct rtable *rth, *rth_local;
692 if (!fib_new_table(dev_net(dev), vrf->tb_id))
695 /* create a dst for routing packets out through a VRF device */
696 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
700 /* create a dst for local ingress routing - packets sent locally
701 * to local address via the VRF device as a loopback
703 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0);
705 dst_release(&rth->dst);
709 rth->dst.output = vrf_output;
710 rth->rt_table_id = vrf->tb_id;
712 rth_local->rt_table_id = vrf->tb_id;
714 rcu_assign_pointer(vrf->rth, rth);
715 rcu_assign_pointer(vrf->rth_local, rth_local);
720 /**************************** device handling ********************/
722 /* cycle interface to flush neighbor cache and move routes across tables */
723 static void cycle_netdev(struct net_device *dev)
725 unsigned int flags = dev->flags;
728 if (!netif_running(dev))
731 ret = dev_change_flags(dev, flags & ~IFF_UP);
733 ret = dev_change_flags(dev, flags);
737 "Failed to cycle device %s; route tables might be wrong!\n",
742 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
746 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
750 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
751 cycle_netdev(port_dev);
756 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
758 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
761 return do_vrf_add_slave(dev, port_dev);
764 /* inverse of do_vrf_add_slave */
765 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
767 netdev_upper_dev_unlink(port_dev, dev);
768 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
770 cycle_netdev(port_dev);
775 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
777 return do_vrf_del_slave(dev, port_dev);
780 static void vrf_dev_uninit(struct net_device *dev)
782 struct net_vrf *vrf = netdev_priv(dev);
783 struct net_device *port_dev;
784 struct list_head *iter;
786 vrf_rtable_release(dev, vrf);
787 vrf_rt6_release(dev, vrf);
789 netdev_for_each_lower_dev(dev, port_dev, iter)
790 vrf_del_slave(dev, port_dev);
792 free_percpu(dev->dstats);
796 static int vrf_dev_init(struct net_device *dev)
798 struct net_vrf *vrf = netdev_priv(dev);
800 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
804 /* create the default dst which points back to us */
805 if (vrf_rtable_create(dev) != 0)
808 if (vrf_rt6_create(dev) != 0)
811 dev->flags = IFF_MASTER | IFF_NOARP;
813 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
814 dev->mtu = 64 * 1024;
816 /* similarly, oper state is irrelevant; set to up to avoid confusion */
817 dev->operstate = IF_OPER_UP;
818 netdev_lockdep_set_classes(dev);
822 vrf_rtable_release(dev, vrf);
824 free_percpu(dev->dstats);
830 static const struct net_device_ops vrf_netdev_ops = {
831 .ndo_init = vrf_dev_init,
832 .ndo_uninit = vrf_dev_uninit,
833 .ndo_start_xmit = vrf_xmit,
834 .ndo_get_stats64 = vrf_get_stats64,
835 .ndo_add_slave = vrf_add_slave,
836 .ndo_del_slave = vrf_del_slave,
839 static u32 vrf_fib_table(const struct net_device *dev)
841 struct net_vrf *vrf = netdev_priv(dev);
846 static struct rtable *vrf_get_rtable(const struct net_device *dev,
847 const struct flowi4 *fl4)
849 struct rtable *rth = NULL;
851 if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
852 struct net_vrf *vrf = netdev_priv(dev);
856 rth = rcu_dereference(vrf->rth);
866 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
871 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
873 struct net_device *dev)
875 struct net *net = dev_net(dev);
879 if (NF_HOOK(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) < 0)
880 skb = NULL; /* kfree_skb(skb) handled by nf code */
885 #if IS_ENABLED(CONFIG_IPV6)
886 /* neighbor handling is done with actual device; do not want
887 * to flip skb->dev for those ndisc packets. This really fails
888 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
891 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
893 const struct ipv6hdr *iph = ipv6_hdr(skb);
896 if (iph->nexthdr == NEXTHDR_ICMP) {
897 const struct icmp6hdr *icmph;
898 struct icmp6hdr _icmph;
900 icmph = skb_header_pointer(skb, sizeof(*iph),
901 sizeof(_icmph), &_icmph);
905 switch (icmph->icmp6_type) {
906 case NDISC_ROUTER_SOLICITATION:
907 case NDISC_ROUTER_ADVERTISEMENT:
908 case NDISC_NEIGHBOUR_SOLICITATION:
909 case NDISC_NEIGHBOUR_ADVERTISEMENT:
920 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
921 const struct net_device *dev,
926 struct net_vrf *vrf = netdev_priv(dev);
927 struct fib6_table *table = NULL;
928 struct rt6_info *rt6;
932 /* fib6_table does not have a refcnt and can not be freed */
933 rt6 = rcu_dereference(vrf->rt6);
935 table = rt6->rt6i_table;
942 return ip6_pol_route(net, table, ifindex, fl6, flags);
945 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
948 const struct ipv6hdr *iph = ipv6_hdr(skb);
949 struct flowi6 fl6 = {
952 .flowlabel = ip6_flowinfo(iph),
953 .flowi6_mark = skb->mark,
954 .flowi6_proto = iph->nexthdr,
955 .flowi6_iif = ifindex,
957 struct net *net = dev_net(vrf_dev);
958 struct rt6_info *rt6;
960 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
961 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
965 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
968 skb_dst_set(skb, &rt6->dst);
971 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
974 int orig_iif = skb->skb_iif;
977 /* loopback traffic; do not push through packet taps again.
978 * Reset pkt_type for upper layers to process skb
980 if (skb->pkt_type == PACKET_LOOPBACK) {
982 skb->skb_iif = vrf_dev->ifindex;
983 skb->pkt_type = PACKET_HOST;
987 /* if packet is NDISC or addressed to multicast or link-local
988 * then keep the ingress interface
990 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
991 if (!ipv6_ndisc_frame(skb) && !need_strict) {
993 skb->skb_iif = vrf_dev->ifindex;
995 skb_push(skb, skb->mac_len);
996 dev_queue_xmit_nit(skb, vrf_dev);
997 skb_pull(skb, skb->mac_len);
999 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1003 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1005 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1011 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1012 struct sk_buff *skb)
1018 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1019 struct sk_buff *skb)
1022 skb->skb_iif = vrf_dev->ifindex;
1024 /* loopback traffic; do not push through packet taps again.
1025 * Reset pkt_type for upper layers to process skb
1027 if (skb->pkt_type == PACKET_LOOPBACK) {
1028 skb->pkt_type = PACKET_HOST;
1032 skb_push(skb, skb->mac_len);
1033 dev_queue_xmit_nit(skb, vrf_dev);
1034 skb_pull(skb, skb->mac_len);
1036 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1041 /* called with rcu lock held */
1042 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1043 struct sk_buff *skb,
1048 return vrf_ip_rcv(vrf_dev, skb);
1050 return vrf_ip6_rcv(vrf_dev, skb);
1056 #if IS_ENABLED(CONFIG_IPV6)
1057 /* send to link-local or multicast address via interface enslaved to
1058 * VRF device. Force lookup to VRF table without changing flow struct
1060 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1063 struct net *net = dev_net(dev);
1064 int flags = RT6_LOOKUP_F_IFACE;
1065 struct dst_entry *dst = NULL;
1066 struct rt6_info *rt;
1068 /* VRF device does not have a link-local address and
1069 * sending packets to link-local or mcast addresses over
1070 * a VRF device does not make sense
1072 if (fl6->flowi6_oif == dev->ifindex) {
1073 dst = &net->ipv6.ip6_null_entry->dst;
1078 if (!ipv6_addr_any(&fl6->saddr))
1079 flags |= RT6_LOOKUP_F_HAS_SADDR;
1081 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1088 /* called under rcu_read_lock */
1089 static int vrf_get_saddr6(struct net_device *dev, const struct sock *sk,
1092 struct net *net = dev_net(dev);
1093 struct dst_entry *dst;
1094 struct rt6_info *rt;
1097 if (rt6_need_strict(&fl6->daddr)) {
1098 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif,
1099 RT6_LOOKUP_F_IFACE);
1105 __u8 flags = fl6->flowi6_flags;
1107 fl6->flowi6_flags |= FLOWI_FLAG_L3MDEV_SRC;
1108 fl6->flowi6_flags |= FLOWI_FLAG_SKIP_NH_OIF;
1110 dst = ip6_route_output(net, sk, fl6);
1111 rt = (struct rt6_info *)dst;
1113 fl6->flowi6_flags = flags;
1118 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
1119 sk ? inet6_sk(sk)->srcprefs : 0,
1129 static const struct l3mdev_ops vrf_l3mdev_ops = {
1130 .l3mdev_fib_table = vrf_fib_table,
1131 .l3mdev_get_rtable = vrf_get_rtable,
1132 .l3mdev_l3_rcv = vrf_l3_rcv,
1133 .l3mdev_l3_out = vrf_l3_out,
1134 #if IS_ENABLED(CONFIG_IPV6)
1135 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1136 .l3mdev_get_saddr6 = vrf_get_saddr6,
1140 static void vrf_get_drvinfo(struct net_device *dev,
1141 struct ethtool_drvinfo *info)
1143 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1144 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1147 static const struct ethtool_ops vrf_ethtool_ops = {
1148 .get_drvinfo = vrf_get_drvinfo,
1151 static inline size_t vrf_fib_rule_nl_size(void)
1155 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1156 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1157 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1162 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1164 struct fib_rule_hdr *frh;
1165 struct nlmsghdr *nlh;
1166 struct sk_buff *skb;
1169 if (family == AF_INET6 && !ipv6_mod_enabled())
1172 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1176 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1178 goto nla_put_failure;
1180 /* rule only needs to appear once */
1181 nlh->nlmsg_flags &= NLM_F_EXCL;
1183 frh = nlmsg_data(nlh);
1184 memset(frh, 0, sizeof(*frh));
1185 frh->family = family;
1186 frh->action = FR_ACT_TO_TBL;
1188 if (nla_put_u32(skb, FRA_L3MDEV, 1))
1189 goto nla_put_failure;
1191 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1192 goto nla_put_failure;
1194 nlmsg_end(skb, nlh);
1196 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1197 skb->sk = dev_net(dev)->rtnl;
1199 err = fib_nl_newrule(skb, nlh);
1203 err = fib_nl_delrule(skb, nlh);
1217 static int vrf_add_fib_rules(const struct net_device *dev)
1221 err = vrf_fib_rule(dev, AF_INET, true);
1225 err = vrf_fib_rule(dev, AF_INET6, true);
1232 vrf_fib_rule(dev, AF_INET, false);
1235 netdev_err(dev, "Failed to add FIB rules.\n");
1239 static void vrf_setup(struct net_device *dev)
1243 /* Initialize the device structure. */
1244 dev->netdev_ops = &vrf_netdev_ops;
1245 dev->l3mdev_ops = &vrf_l3mdev_ops;
1246 dev->ethtool_ops = &vrf_ethtool_ops;
1247 dev->destructor = free_netdev;
1249 /* Fill in device structure with ethernet-generic values. */
1250 eth_hw_addr_random(dev);
1252 /* don't acquire vrf device's netif_tx_lock when transmitting */
1253 dev->features |= NETIF_F_LLTX;
1255 /* don't allow vrf devices to change network namespaces. */
1256 dev->features |= NETIF_F_NETNS_LOCAL;
1258 /* does not make sense for a VLAN to be added to a vrf device */
1259 dev->features |= NETIF_F_VLAN_CHALLENGED;
1261 /* enable offload features */
1262 dev->features |= NETIF_F_GSO_SOFTWARE;
1263 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1264 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1266 dev->hw_features = dev->features;
1267 dev->hw_enc_features = dev->features;
1269 /* default to no qdisc; user can add if desired */
1270 dev->priv_flags |= IFF_NO_QUEUE;
1273 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
1275 if (tb[IFLA_ADDRESS]) {
1276 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1278 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1279 return -EADDRNOTAVAIL;
1284 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1286 unregister_netdevice_queue(dev, head);
1289 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1290 struct nlattr *tb[], struct nlattr *data[])
1292 struct net_vrf *vrf = netdev_priv(dev);
1295 if (!data || !data[IFLA_VRF_TABLE])
1298 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1300 dev->priv_flags |= IFF_L3MDEV_MASTER;
1302 err = register_netdevice(dev);
1306 if (add_fib_rules) {
1307 err = vrf_add_fib_rules(dev);
1309 unregister_netdevice(dev);
1312 add_fib_rules = false;
1319 static size_t vrf_nl_getsize(const struct net_device *dev)
1321 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1324 static int vrf_fillinfo(struct sk_buff *skb,
1325 const struct net_device *dev)
1327 struct net_vrf *vrf = netdev_priv(dev);
1329 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1332 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1333 const struct net_device *slave_dev)
1335 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1338 static int vrf_fill_slave_info(struct sk_buff *skb,
1339 const struct net_device *vrf_dev,
1340 const struct net_device *slave_dev)
1342 struct net_vrf *vrf = netdev_priv(vrf_dev);
1344 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1350 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1351 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1354 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1356 .priv_size = sizeof(struct net_vrf),
1358 .get_size = vrf_nl_getsize,
1359 .policy = vrf_nl_policy,
1360 .validate = vrf_validate,
1361 .fill_info = vrf_fillinfo,
1363 .get_slave_size = vrf_get_slave_size,
1364 .fill_slave_info = vrf_fill_slave_info,
1366 .newlink = vrf_newlink,
1367 .dellink = vrf_dellink,
1369 .maxtype = IFLA_VRF_MAX,
1372 static int vrf_device_event(struct notifier_block *unused,
1373 unsigned long event, void *ptr)
1375 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1377 /* only care about unregister events to drop slave references */
1378 if (event == NETDEV_UNREGISTER) {
1379 struct net_device *vrf_dev;
1381 if (!netif_is_l3_slave(dev))
1384 vrf_dev = netdev_master_upper_dev_get(dev);
1385 vrf_del_slave(vrf_dev, dev);
1391 static struct notifier_block vrf_notifier_block __read_mostly = {
1392 .notifier_call = vrf_device_event,
1395 static int __init vrf_init_module(void)
1399 register_netdevice_notifier(&vrf_notifier_block);
1401 rc = rtnl_link_register(&vrf_link_ops);
1408 unregister_netdevice_notifier(&vrf_notifier_block);
1412 module_init(vrf_init_module);
1413 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1414 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1415 MODULE_LICENSE("GPL");
1416 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1417 MODULE_VERSION(DRV_VERSION);