2 * vrf.c: device driver to encapsulate a VRF space
4 * Copyright (c) 2015 Cumulus Networks. All rights reserved.
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 DRV_NAME "vrf"
41 #define DRV_VERSION "1.0"
43 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */
44 static bool add_fib_rules = true;
47 struct rtable __rcu *rth;
48 struct rtable __rcu *rth_local;
49 struct rt6_info __rcu *rt6;
50 struct rt6_info __rcu *rt6_local;
61 struct u64_stats_sync syncp;
64 static void vrf_rx_stats(struct net_device *dev, int len)
66 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
68 u64_stats_update_begin(&dstats->syncp);
70 dstats->rx_bytes += len;
71 u64_stats_update_end(&dstats->syncp);
74 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
76 vrf_dev->stats.tx_errors++;
80 static void vrf_get_stats64(struct net_device *dev,
81 struct rtnl_link_stats64 *stats)
85 for_each_possible_cpu(i) {
86 const struct pcpu_dstats *dstats;
87 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
90 dstats = per_cpu_ptr(dev->dstats, i);
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;
107 /* Local traffic destined to local address. Reinsert the packet to rx
108 * path, similar to loopback handling.
110 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
111 struct dst_entry *dst)
117 skb_dst_set(skb, dst);
120 /* set pkt_type to avoid skb hitting packet taps twice -
121 * once on Tx and again in Rx processing
123 skb->pkt_type = PACKET_LOOPBACK;
125 skb->protocol = eth_type_trans(skb, dev);
127 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
128 vrf_rx_stats(dev, len);
130 this_cpu_inc(dev->dstats->rx_drps);
135 #if IS_ENABLED(CONFIG_IPV6)
136 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
141 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
142 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
144 if (likely(err == 1))
145 err = dst_output(net, sk, skb);
150 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
151 struct net_device *dev)
153 const struct ipv6hdr *iph = ipv6_hdr(skb);
154 struct net *net = dev_net(skb->dev);
155 struct flowi6 fl6 = {
156 /* needed to match OIF rule */
157 .flowi6_oif = dev->ifindex,
158 .flowi6_iif = LOOPBACK_IFINDEX,
161 .flowlabel = ip6_flowinfo(iph),
162 .flowi6_mark = skb->mark,
163 .flowi6_proto = iph->nexthdr,
164 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
166 int ret = NET_XMIT_DROP;
167 struct dst_entry *dst;
168 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
170 dst = ip6_route_output(net, NULL, &fl6);
176 /* if dst.dev is loopback or the VRF device again this is locally
177 * originated traffic destined to a local address. Short circuit
178 * to Rx path using our local dst
180 if (dst->dev == net->loopback_dev || dst->dev == dev) {
181 struct net_vrf *vrf = netdev_priv(dev);
182 struct rt6_info *rt6_local;
184 /* release looked up dst and use cached local dst */
189 rt6_local = rcu_dereference(vrf->rt6_local);
190 if (unlikely(!rt6_local)) {
195 /* Ordering issue: cached local dst is created on newlink
196 * before the IPv6 initialization. Using the local dst
197 * requires rt6i_idev to be set so make sure it is.
199 if (unlikely(!rt6_local->rt6i_idev)) {
200 rt6_local->rt6i_idev = in6_dev_get(dev);
201 if (!rt6_local->rt6i_idev) {
207 dst = &rt6_local->dst;
212 return vrf_local_xmit(skb, dev, &rt6_local->dst);
215 skb_dst_set(skb, dst);
217 /* strip the ethernet header added for pass through VRF device */
218 __skb_pull(skb, skb_network_offset(skb));
220 ret = vrf_ip6_local_out(net, skb->sk, skb);
221 if (unlikely(net_xmit_eval(ret)))
222 dev->stats.tx_errors++;
224 ret = NET_XMIT_SUCCESS;
228 vrf_tx_error(dev, skb);
229 return NET_XMIT_DROP;
232 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
233 struct net_device *dev)
235 vrf_tx_error(dev, skb);
236 return NET_XMIT_DROP;
240 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
241 static int vrf_ip_local_out(struct net *net, struct sock *sk,
246 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
247 skb, NULL, skb_dst(skb)->dev, dst_output);
248 if (likely(err == 1))
249 err = dst_output(net, sk, skb);
254 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
255 struct net_device *vrf_dev)
257 struct iphdr *ip4h = ip_hdr(skb);
258 int ret = NET_XMIT_DROP;
259 struct flowi4 fl4 = {
260 /* needed to match OIF rule */
261 .flowi4_oif = vrf_dev->ifindex,
262 .flowi4_iif = LOOPBACK_IFINDEX,
263 .flowi4_tos = RT_TOS(ip4h->tos),
264 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
265 .flowi4_proto = ip4h->protocol,
266 .daddr = ip4h->daddr,
267 .saddr = ip4h->saddr,
269 struct net *net = dev_net(vrf_dev);
272 rt = ip_route_output_flow(net, &fl4, NULL);
278 /* if dst.dev is loopback or the VRF device again this is locally
279 * originated traffic destined to a local address. Short circuit
280 * to Rx path using our local dst
282 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) {
283 struct net_vrf *vrf = netdev_priv(vrf_dev);
284 struct rtable *rth_local;
285 struct dst_entry *dst = NULL;
291 rth_local = rcu_dereference(vrf->rth_local);
292 if (likely(rth_local)) {
293 dst = &rth_local->dst;
302 return vrf_local_xmit(skb, vrf_dev, dst);
305 skb_dst_set(skb, &rt->dst);
307 /* strip the ethernet header added for pass through VRF device */
308 __skb_pull(skb, skb_network_offset(skb));
311 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
315 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
316 if (unlikely(net_xmit_eval(ret)))
317 vrf_dev->stats.tx_errors++;
319 ret = NET_XMIT_SUCCESS;
324 vrf_tx_error(vrf_dev, skb);
328 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
330 switch (skb->protocol) {
331 case htons(ETH_P_IP):
332 return vrf_process_v4_outbound(skb, dev);
333 case htons(ETH_P_IPV6):
334 return vrf_process_v6_outbound(skb, dev);
336 vrf_tx_error(dev, skb);
337 return NET_XMIT_DROP;
341 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
344 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
346 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
347 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
349 u64_stats_update_begin(&dstats->syncp);
351 dstats->tx_bytes += len;
352 u64_stats_update_end(&dstats->syncp);
354 this_cpu_inc(dev->dstats->tx_drps);
360 #if IS_ENABLED(CONFIG_IPV6)
361 /* modelled after ip6_finish_output2 */
362 static int vrf_finish_output6(struct net *net, struct sock *sk,
365 struct dst_entry *dst = skb_dst(skb);
366 struct net_device *dev = dst->dev;
367 struct neighbour *neigh;
368 struct in6_addr *nexthop;
373 skb->protocol = htons(ETH_P_IPV6);
377 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
378 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
379 if (unlikely(!neigh))
380 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
381 if (!IS_ERR(neigh)) {
382 sock_confirm_neigh(skb, neigh);
383 ret = neigh_output(neigh, skb);
384 rcu_read_unlock_bh();
387 rcu_read_unlock_bh();
389 IP6_INC_STATS(dev_net(dst->dev),
390 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
395 /* modelled after ip6_output */
396 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
398 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
399 net, sk, skb, NULL, skb_dst(skb)->dev,
401 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
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.
408 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
412 struct net_vrf *vrf = netdev_priv(vrf_dev);
413 struct dst_entry *dst = NULL;
414 struct rt6_info *rt6;
416 /* don't divert link scope packets */
417 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
422 rt6 = rcu_dereference(vrf->rt6);
430 if (unlikely(!dst)) {
431 vrf_tx_error(vrf_dev, skb);
436 skb_dst_set(skb, dst);
442 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
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;
449 RCU_INIT_POINTER(vrf->rt6, NULL);
450 RCU_INIT_POINTER(vrf->rt6_local, NULL);
453 /* move dev in dst's to loopback so this VRF device can be deleted
454 * - based on dst_ifdown
459 dst->dev = net->loopback_dev;
465 if (rt6_local->rt6i_idev) {
466 in6_dev_put(rt6_local->rt6i_idev);
467 rt6_local->rt6i_idev = NULL;
470 dst = &rt6_local->dst;
472 dst->dev = net->loopback_dev;
478 static int vrf_rt6_create(struct net_device *dev)
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;
487 /* IPv6 can be CONFIG enabled and then disabled runtime */
488 if (!ipv6_mod_enabled())
491 rt6i_table = fib6_new_table(net, vrf->tb_id);
495 /* create a dst for routing packets out a VRF device */
496 rt6 = ip6_dst_alloc(net, dev, flags);
502 rt6->rt6i_table = rt6i_table;
503 rt6->dst.output = vrf_output6;
505 /* create a dst for local routing - packets sent locally
506 * to local address via the VRF device as a loopback
508 rt6_local = ip6_dst_alloc(net, dev, flags);
510 dst_release(&rt6->dst);
514 dst_hold(&rt6_local->dst);
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;
521 rcu_assign_pointer(vrf->rt6, rt6);
522 rcu_assign_pointer(vrf->rt6_local, rt6_local);
529 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
536 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
540 static int vrf_rt6_create(struct net_device *dev)
546 /* modelled after ip_finish_output2 */
547 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
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;
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);
581 if (!IS_ERR(neigh)) {
582 sock_confirm_neigh(skb, neigh);
583 ret = neigh_output(neigh, skb);
586 rcu_read_unlock_bh();
588 if (unlikely(ret < 0))
589 vrf_tx_error(skb->dev, skb);
593 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
595 struct net_device *dev = skb_dst(skb)->dev;
597 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
600 skb->protocol = htons(ETH_P_IP);
602 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
603 net, sk, skb, NULL, dev,
605 !(IPCB(skb)->flags & IPSKB_REROUTED));
608 /* set dst on skb to send packet to us via dev_xmit path. Allows
609 * packet to go through device based features such as qdisc, netfilter
610 * hooks and packet sockets with skb->dev set to vrf device.
612 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
616 struct net_vrf *vrf = netdev_priv(vrf_dev);
617 struct dst_entry *dst = NULL;
620 /* don't divert multicast */
621 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
626 rth = rcu_dereference(vrf->rth);
634 if (unlikely(!dst)) {
635 vrf_tx_error(vrf_dev, skb);
640 skb_dst_set(skb, dst);
645 /* called with rcu lock held */
646 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
653 return vrf_ip_out(vrf_dev, sk, skb);
655 return vrf_ip6_out(vrf_dev, sk, skb);
662 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
664 struct rtable *rth = rtnl_dereference(vrf->rth);
665 struct rtable *rth_local = rtnl_dereference(vrf->rth_local);
666 struct net *net = dev_net(dev);
667 struct dst_entry *dst;
669 RCU_INIT_POINTER(vrf->rth, NULL);
670 RCU_INIT_POINTER(vrf->rth_local, NULL);
673 /* move dev in dst's to loopback so this VRF device can be deleted
674 * - based on dst_ifdown
679 dst->dev = net->loopback_dev;
685 dst = &rth_local->dst;
687 dst->dev = net->loopback_dev;
693 static int vrf_rtable_create(struct net_device *dev)
695 struct net_vrf *vrf = netdev_priv(dev);
696 struct rtable *rth, *rth_local;
698 if (!fib_new_table(dev_net(dev), vrf->tb_id))
701 /* create a dst for routing packets out through a VRF device */
702 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
706 /* create a dst for local ingress routing - packets sent locally
707 * to local address via the VRF device as a loopback
709 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0);
711 dst_release(&rth->dst);
715 rth->dst.output = vrf_output;
716 rth->rt_table_id = vrf->tb_id;
718 rth_local->rt_table_id = vrf->tb_id;
720 rcu_assign_pointer(vrf->rth, rth);
721 rcu_assign_pointer(vrf->rth_local, rth_local);
726 /**************************** device handling ********************/
728 /* cycle interface to flush neighbor cache and move routes across tables */
729 static void cycle_netdev(struct net_device *dev)
731 unsigned int flags = dev->flags;
734 if (!netif_running(dev))
737 ret = dev_change_flags(dev, flags & ~IFF_UP);
739 ret = dev_change_flags(dev, flags);
743 "Failed to cycle device %s; route tables might be wrong!\n",
748 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
752 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
756 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
757 cycle_netdev(port_dev);
762 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
764 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
767 return do_vrf_add_slave(dev, port_dev);
770 /* inverse of do_vrf_add_slave */
771 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
773 netdev_upper_dev_unlink(port_dev, dev);
774 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
776 cycle_netdev(port_dev);
781 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
783 return do_vrf_del_slave(dev, port_dev);
786 static void vrf_dev_uninit(struct net_device *dev)
788 struct net_vrf *vrf = netdev_priv(dev);
789 struct net_device *port_dev;
790 struct list_head *iter;
792 vrf_rtable_release(dev, vrf);
793 vrf_rt6_release(dev, vrf);
795 netdev_for_each_lower_dev(dev, port_dev, iter)
796 vrf_del_slave(dev, port_dev);
798 free_percpu(dev->dstats);
802 static int vrf_dev_init(struct net_device *dev)
804 struct net_vrf *vrf = netdev_priv(dev);
806 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
810 /* create the default dst which points back to us */
811 if (vrf_rtable_create(dev) != 0)
814 if (vrf_rt6_create(dev) != 0)
817 dev->flags = IFF_MASTER | IFF_NOARP;
819 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
820 dev->mtu = 64 * 1024;
822 /* similarly, oper state is irrelevant; set to up to avoid confusion */
823 dev->operstate = IF_OPER_UP;
824 netdev_lockdep_set_classes(dev);
828 vrf_rtable_release(dev, vrf);
830 free_percpu(dev->dstats);
836 static const struct net_device_ops vrf_netdev_ops = {
837 .ndo_init = vrf_dev_init,
838 .ndo_uninit = vrf_dev_uninit,
839 .ndo_start_xmit = vrf_xmit,
840 .ndo_get_stats64 = vrf_get_stats64,
841 .ndo_add_slave = vrf_add_slave,
842 .ndo_del_slave = vrf_del_slave,
845 static u32 vrf_fib_table(const struct net_device *dev)
847 struct net_vrf *vrf = netdev_priv(dev);
852 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
857 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
859 struct net_device *dev)
861 struct net *net = dev_net(dev);
863 if (NF_HOOK(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) < 0)
864 skb = NULL; /* kfree_skb(skb) handled by nf code */
869 #if IS_ENABLED(CONFIG_IPV6)
870 /* neighbor handling is done with actual device; do not want
871 * to flip skb->dev for those ndisc packets. This really fails
872 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
875 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
877 const struct ipv6hdr *iph = ipv6_hdr(skb);
880 if (iph->nexthdr == NEXTHDR_ICMP) {
881 const struct icmp6hdr *icmph;
882 struct icmp6hdr _icmph;
884 icmph = skb_header_pointer(skb, sizeof(*iph),
885 sizeof(_icmph), &_icmph);
889 switch (icmph->icmp6_type) {
890 case NDISC_ROUTER_SOLICITATION:
891 case NDISC_ROUTER_ADVERTISEMENT:
892 case NDISC_NEIGHBOUR_SOLICITATION:
893 case NDISC_NEIGHBOUR_ADVERTISEMENT:
904 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
905 const struct net_device *dev,
910 struct net_vrf *vrf = netdev_priv(dev);
911 struct fib6_table *table = NULL;
912 struct rt6_info *rt6;
916 /* fib6_table does not have a refcnt and can not be freed */
917 rt6 = rcu_dereference(vrf->rt6);
919 table = rt6->rt6i_table;
926 return ip6_pol_route(net, table, ifindex, fl6, flags);
929 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
932 const struct ipv6hdr *iph = ipv6_hdr(skb);
933 struct flowi6 fl6 = {
936 .flowlabel = ip6_flowinfo(iph),
937 .flowi6_mark = skb->mark,
938 .flowi6_proto = iph->nexthdr,
939 .flowi6_iif = ifindex,
941 struct net *net = dev_net(vrf_dev);
942 struct rt6_info *rt6;
944 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
945 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
949 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
952 skb_dst_set(skb, &rt6->dst);
955 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
958 int orig_iif = skb->skb_iif;
961 /* loopback traffic; do not push through packet taps again.
962 * Reset pkt_type for upper layers to process skb
964 if (skb->pkt_type == PACKET_LOOPBACK) {
966 skb->skb_iif = vrf_dev->ifindex;
967 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
968 skb->pkt_type = PACKET_HOST;
972 /* if packet is NDISC or addressed to multicast or link-local
973 * then keep the ingress interface
975 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
976 if (!ipv6_ndisc_frame(skb) && !need_strict) {
977 vrf_rx_stats(vrf_dev, skb->len);
979 skb->skb_iif = vrf_dev->ifindex;
981 skb_push(skb, skb->mac_len);
982 dev_queue_xmit_nit(skb, vrf_dev);
983 skb_pull(skb, skb->mac_len);
985 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
989 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
991 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
997 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1004 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1005 struct sk_buff *skb)
1008 skb->skb_iif = vrf_dev->ifindex;
1009 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1011 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1014 /* loopback traffic; do not push through packet taps again.
1015 * Reset pkt_type for upper layers to process skb
1017 if (skb->pkt_type == PACKET_LOOPBACK) {
1018 skb->pkt_type = PACKET_HOST;
1022 vrf_rx_stats(vrf_dev, skb->len);
1024 skb_push(skb, skb->mac_len);
1025 dev_queue_xmit_nit(skb, vrf_dev);
1026 skb_pull(skb, skb->mac_len);
1028 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1033 /* called with rcu lock held */
1034 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1035 struct sk_buff *skb,
1040 return vrf_ip_rcv(vrf_dev, skb);
1042 return vrf_ip6_rcv(vrf_dev, skb);
1048 #if IS_ENABLED(CONFIG_IPV6)
1049 /* send to link-local or multicast address via interface enslaved to
1050 * VRF device. Force lookup to VRF table without changing flow struct
1052 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1055 struct net *net = dev_net(dev);
1056 int flags = RT6_LOOKUP_F_IFACE;
1057 struct dst_entry *dst = NULL;
1058 struct rt6_info *rt;
1060 /* VRF device does not have a link-local address and
1061 * sending packets to link-local or mcast addresses over
1062 * a VRF device does not make sense
1064 if (fl6->flowi6_oif == dev->ifindex) {
1065 dst = &net->ipv6.ip6_null_entry->dst;
1070 if (!ipv6_addr_any(&fl6->saddr))
1071 flags |= RT6_LOOKUP_F_HAS_SADDR;
1073 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1081 static const struct l3mdev_ops vrf_l3mdev_ops = {
1082 .l3mdev_fib_table = vrf_fib_table,
1083 .l3mdev_l3_rcv = vrf_l3_rcv,
1084 .l3mdev_l3_out = vrf_l3_out,
1085 #if IS_ENABLED(CONFIG_IPV6)
1086 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1090 static void vrf_get_drvinfo(struct net_device *dev,
1091 struct ethtool_drvinfo *info)
1093 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1094 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1097 static const struct ethtool_ops vrf_ethtool_ops = {
1098 .get_drvinfo = vrf_get_drvinfo,
1101 static inline size_t vrf_fib_rule_nl_size(void)
1105 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1106 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1107 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1112 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1114 struct fib_rule_hdr *frh;
1115 struct nlmsghdr *nlh;
1116 struct sk_buff *skb;
1119 if (family == AF_INET6 && !ipv6_mod_enabled())
1122 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1126 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1128 goto nla_put_failure;
1130 /* rule only needs to appear once */
1131 nlh->nlmsg_flags |= NLM_F_EXCL;
1133 frh = nlmsg_data(nlh);
1134 memset(frh, 0, sizeof(*frh));
1135 frh->family = family;
1136 frh->action = FR_ACT_TO_TBL;
1138 if (nla_put_u32(skb, FRA_L3MDEV, 1))
1139 goto nla_put_failure;
1141 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1142 goto nla_put_failure;
1144 nlmsg_end(skb, nlh);
1146 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1147 skb->sk = dev_net(dev)->rtnl;
1149 err = fib_nl_newrule(skb, nlh);
1153 err = fib_nl_delrule(skb, nlh);
1167 static int vrf_add_fib_rules(const struct net_device *dev)
1171 err = vrf_fib_rule(dev, AF_INET, true);
1175 err = vrf_fib_rule(dev, AF_INET6, true);
1179 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1180 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1187 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1189 vrf_fib_rule(dev, AF_INET6, false);
1193 vrf_fib_rule(dev, AF_INET, false);
1196 netdev_err(dev, "Failed to add FIB rules.\n");
1200 static void vrf_setup(struct net_device *dev)
1204 /* Initialize the device structure. */
1205 dev->netdev_ops = &vrf_netdev_ops;
1206 dev->l3mdev_ops = &vrf_l3mdev_ops;
1207 dev->ethtool_ops = &vrf_ethtool_ops;
1208 dev->destructor = free_netdev;
1210 /* Fill in device structure with ethernet-generic values. */
1211 eth_hw_addr_random(dev);
1213 /* don't acquire vrf device's netif_tx_lock when transmitting */
1214 dev->features |= NETIF_F_LLTX;
1216 /* don't allow vrf devices to change network namespaces. */
1217 dev->features |= NETIF_F_NETNS_LOCAL;
1219 /* does not make sense for a VLAN to be added to a vrf device */
1220 dev->features |= NETIF_F_VLAN_CHALLENGED;
1222 /* enable offload features */
1223 dev->features |= NETIF_F_GSO_SOFTWARE;
1224 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1225 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1227 dev->hw_features = dev->features;
1228 dev->hw_enc_features = dev->features;
1230 /* default to no qdisc; user can add if desired */
1231 dev->priv_flags |= IFF_NO_QUEUE;
1234 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
1236 if (tb[IFLA_ADDRESS]) {
1237 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1239 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1240 return -EADDRNOTAVAIL;
1245 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1247 unregister_netdevice_queue(dev, head);
1250 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1251 struct nlattr *tb[], struct nlattr *data[])
1253 struct net_vrf *vrf = netdev_priv(dev);
1256 if (!data || !data[IFLA_VRF_TABLE])
1259 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1260 if (vrf->tb_id == RT_TABLE_UNSPEC)
1263 dev->priv_flags |= IFF_L3MDEV_MASTER;
1265 err = register_netdevice(dev);
1269 if (add_fib_rules) {
1270 err = vrf_add_fib_rules(dev);
1272 unregister_netdevice(dev);
1275 add_fib_rules = false;
1282 static size_t vrf_nl_getsize(const struct net_device *dev)
1284 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1287 static int vrf_fillinfo(struct sk_buff *skb,
1288 const struct net_device *dev)
1290 struct net_vrf *vrf = netdev_priv(dev);
1292 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1295 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1296 const struct net_device *slave_dev)
1298 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1301 static int vrf_fill_slave_info(struct sk_buff *skb,
1302 const struct net_device *vrf_dev,
1303 const struct net_device *slave_dev)
1305 struct net_vrf *vrf = netdev_priv(vrf_dev);
1307 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1313 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1314 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1317 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1319 .priv_size = sizeof(struct net_vrf),
1321 .get_size = vrf_nl_getsize,
1322 .policy = vrf_nl_policy,
1323 .validate = vrf_validate,
1324 .fill_info = vrf_fillinfo,
1326 .get_slave_size = vrf_get_slave_size,
1327 .fill_slave_info = vrf_fill_slave_info,
1329 .newlink = vrf_newlink,
1330 .dellink = vrf_dellink,
1332 .maxtype = IFLA_VRF_MAX,
1335 static int vrf_device_event(struct notifier_block *unused,
1336 unsigned long event, void *ptr)
1338 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1340 /* only care about unregister events to drop slave references */
1341 if (event == NETDEV_UNREGISTER) {
1342 struct net_device *vrf_dev;
1344 if (!netif_is_l3_slave(dev))
1347 vrf_dev = netdev_master_upper_dev_get(dev);
1348 vrf_del_slave(vrf_dev, dev);
1354 static struct notifier_block vrf_notifier_block __read_mostly = {
1355 .notifier_call = vrf_device_event,
1358 static int __init vrf_init_module(void)
1362 register_netdevice_notifier(&vrf_notifier_block);
1364 rc = rtnl_link_register(&vrf_link_ops);
1371 unregister_netdevice_notifier(&vrf_notifier_block);
1375 module_init(vrf_init_module);
1376 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1377 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1378 MODULE_LICENSE("GPL");
1379 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1380 MODULE_VERSION(DRV_VERSION);