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>
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;
49 struct rtable __rcu *rth;
50 struct rt6_info __rcu *rt6;
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 /* by default VRF devices do not have a qdisc and are expected
108 * to be created with only a single queue.
110 static bool qdisc_tx_is_default(const struct net_device *dev)
112 struct netdev_queue *txq;
115 if (dev->num_tx_queues > 1)
118 txq = netdev_get_tx_queue(dev, 0);
119 qdisc = rcu_access_pointer(txq->qdisc);
121 return !qdisc->enqueue;
124 /* Local traffic destined to local address. Reinsert the packet to rx
125 * path, similar to loopback handling.
127 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
128 struct dst_entry *dst)
134 skb_dst_set(skb, dst);
137 /* set pkt_type to avoid skb hitting packet taps twice -
138 * once on Tx and again in Rx processing
140 skb->pkt_type = PACKET_LOOPBACK;
142 skb->protocol = eth_type_trans(skb, dev);
144 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
145 vrf_rx_stats(dev, len);
147 this_cpu_inc(dev->dstats->rx_drps);
152 #if IS_ENABLED(CONFIG_IPV6)
153 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
158 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
159 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
161 if (likely(err == 1))
162 err = dst_output(net, sk, skb);
167 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
168 struct net_device *dev)
170 const struct ipv6hdr *iph = ipv6_hdr(skb);
171 struct net *net = dev_net(skb->dev);
172 struct flowi6 fl6 = {
173 /* needed to match OIF rule */
174 .flowi6_oif = dev->ifindex,
175 .flowi6_iif = LOOPBACK_IFINDEX,
178 .flowlabel = ip6_flowinfo(iph),
179 .flowi6_mark = skb->mark,
180 .flowi6_proto = iph->nexthdr,
181 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
183 int ret = NET_XMIT_DROP;
184 struct dst_entry *dst;
185 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
187 dst = ip6_route_output(net, NULL, &fl6);
193 /* if dst.dev is loopback or the VRF device again this is locally
194 * originated traffic destined to a local address. Short circuit
198 return vrf_local_xmit(skb, dev, dst);
200 skb_dst_set(skb, dst);
202 /* strip the ethernet header added for pass through VRF device */
203 __skb_pull(skb, skb_network_offset(skb));
205 ret = vrf_ip6_local_out(net, skb->sk, skb);
206 if (unlikely(net_xmit_eval(ret)))
207 dev->stats.tx_errors++;
209 ret = NET_XMIT_SUCCESS;
213 vrf_tx_error(dev, skb);
214 return NET_XMIT_DROP;
217 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
218 struct net_device *dev)
220 vrf_tx_error(dev, skb);
221 return NET_XMIT_DROP;
225 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
226 static int vrf_ip_local_out(struct net *net, struct sock *sk,
231 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
232 skb, NULL, skb_dst(skb)->dev, dst_output);
233 if (likely(err == 1))
234 err = dst_output(net, sk, skb);
239 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
240 struct net_device *vrf_dev)
242 struct iphdr *ip4h = ip_hdr(skb);
243 int ret = NET_XMIT_DROP;
244 struct flowi4 fl4 = {
245 /* needed to match OIF rule */
246 .flowi4_oif = vrf_dev->ifindex,
247 .flowi4_iif = LOOPBACK_IFINDEX,
248 .flowi4_tos = RT_TOS(ip4h->tos),
249 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
250 .flowi4_proto = ip4h->protocol,
251 .daddr = ip4h->daddr,
252 .saddr = ip4h->saddr,
254 struct net *net = dev_net(vrf_dev);
257 rt = ip_route_output_flow(net, &fl4, NULL);
263 /* if dst.dev is loopback or the VRF device again this is locally
264 * originated traffic destined to a local address. Short circuit
267 if (rt->dst.dev == vrf_dev)
268 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
270 skb_dst_set(skb, &rt->dst);
272 /* strip the ethernet header added for pass through VRF device */
273 __skb_pull(skb, skb_network_offset(skb));
276 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
280 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
281 if (unlikely(net_xmit_eval(ret)))
282 vrf_dev->stats.tx_errors++;
284 ret = NET_XMIT_SUCCESS;
289 vrf_tx_error(vrf_dev, skb);
293 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
295 switch (skb->protocol) {
296 case htons(ETH_P_IP):
297 return vrf_process_v4_outbound(skb, dev);
298 case htons(ETH_P_IPV6):
299 return vrf_process_v6_outbound(skb, dev);
301 vrf_tx_error(dev, skb);
302 return NET_XMIT_DROP;
306 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
309 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
311 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
312 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
314 u64_stats_update_begin(&dstats->syncp);
316 dstats->tx_bytes += len;
317 u64_stats_update_end(&dstats->syncp);
319 this_cpu_inc(dev->dstats->tx_drps);
325 static int vrf_finish_direct(struct net *net, struct sock *sk,
328 struct net_device *vrf_dev = skb->dev;
330 if (!list_empty(&vrf_dev->ptype_all) &&
331 likely(skb_headroom(skb) >= ETH_HLEN)) {
332 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
334 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
335 eth_zero_addr(eth->h_dest);
336 eth->h_proto = skb->protocol;
339 dev_queue_xmit_nit(skb, vrf_dev);
340 rcu_read_unlock_bh();
342 skb_pull(skb, ETH_HLEN);
348 #if IS_ENABLED(CONFIG_IPV6)
349 /* modelled after ip6_finish_output2 */
350 static int vrf_finish_output6(struct net *net, struct sock *sk,
353 struct dst_entry *dst = skb_dst(skb);
354 struct net_device *dev = dst->dev;
355 struct neighbour *neigh;
356 struct in6_addr *nexthop;
361 skb->protocol = htons(ETH_P_IPV6);
365 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
366 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
367 if (unlikely(!neigh))
368 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
369 if (!IS_ERR(neigh)) {
370 sock_confirm_neigh(skb, neigh);
371 ret = neigh_output(neigh, skb);
372 rcu_read_unlock_bh();
375 rcu_read_unlock_bh();
377 IP6_INC_STATS(dev_net(dst->dev),
378 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
383 /* modelled after ip6_output */
384 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
386 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
387 net, sk, skb, NULL, skb_dst(skb)->dev,
389 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
392 /* set dst on skb to send packet to us via dev_xmit path. Allows
393 * packet to go through device based features such as qdisc, netfilter
394 * hooks and packet sockets with skb->dev set to vrf device.
396 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
399 struct net_vrf *vrf = netdev_priv(vrf_dev);
400 struct dst_entry *dst = NULL;
401 struct rt6_info *rt6;
405 rt6 = rcu_dereference(vrf->rt6);
413 if (unlikely(!dst)) {
414 vrf_tx_error(vrf_dev, skb);
419 skb_dst_set(skb, dst);
424 static int vrf_output6_direct(struct net *net, struct sock *sk,
427 skb->protocol = htons(ETH_P_IPV6);
429 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
430 net, sk, skb, NULL, skb->dev,
432 !(IPCB(skb)->flags & IPSKB_REROUTED));
435 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
439 struct net *net = dev_net(vrf_dev);
444 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
445 skb, NULL, vrf_dev, vrf_output6_direct);
447 if (likely(err == 1))
448 err = vrf_output6_direct(net, sk, skb);
450 /* reset skb device */
451 if (likely(err == 1))
459 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
463 /* don't divert link scope packets */
464 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
467 if (qdisc_tx_is_default(vrf_dev))
468 return vrf_ip6_out_direct(vrf_dev, sk, skb);
470 return vrf_ip6_out_redirect(vrf_dev, skb);
474 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
476 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
477 struct net *net = dev_net(dev);
478 struct dst_entry *dst;
480 RCU_INIT_POINTER(vrf->rt6, NULL);
483 /* move dev in dst's to loopback so this VRF device can be deleted
484 * - based on dst_ifdown
489 dst->dev = net->loopback_dev;
495 static int vrf_rt6_create(struct net_device *dev)
497 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
498 struct net_vrf *vrf = netdev_priv(dev);
499 struct net *net = dev_net(dev);
500 struct fib6_table *rt6i_table;
501 struct rt6_info *rt6;
504 /* IPv6 can be CONFIG enabled and then disabled runtime */
505 if (!ipv6_mod_enabled())
508 rt6i_table = fib6_new_table(net, vrf->tb_id);
512 /* create a dst for routing packets out a VRF device */
513 rt6 = ip6_dst_alloc(net, dev, flags);
517 rt6->rt6i_table = rt6i_table;
518 rt6->dst.output = vrf_output6;
520 rcu_assign_pointer(vrf->rt6, rt6);
527 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
534 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
538 static int vrf_rt6_create(struct net_device *dev)
544 /* modelled after ip_finish_output2 */
545 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
547 struct dst_entry *dst = skb_dst(skb);
548 struct rtable *rt = (struct rtable *)dst;
549 struct net_device *dev = dst->dev;
550 unsigned int hh_len = LL_RESERVED_SPACE(dev);
551 struct neighbour *neigh;
557 /* Be paranoid, rather than too clever. */
558 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
559 struct sk_buff *skb2;
561 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
567 skb_set_owner_w(skb2, skb->sk);
575 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
576 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
577 if (unlikely(!neigh))
578 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
579 if (!IS_ERR(neigh)) {
580 sock_confirm_neigh(skb, neigh);
581 ret = neigh_output(neigh, skb);
582 rcu_read_unlock_bh();
586 rcu_read_unlock_bh();
588 vrf_tx_error(skb->dev, skb);
592 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
594 struct net_device *dev = skb_dst(skb)->dev;
596 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
599 skb->protocol = htons(ETH_P_IP);
601 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
602 net, sk, skb, NULL, dev,
604 !(IPCB(skb)->flags & IPSKB_REROUTED));
607 /* set dst on skb to send packet to us via dev_xmit path. Allows
608 * packet to go through device based features such as qdisc, netfilter
609 * hooks and packet sockets with skb->dev set to vrf device.
611 static struct sk_buff *vrf_ip_out_redirect(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 static int vrf_output_direct(struct net *net, struct sock *sk,
642 skb->protocol = htons(ETH_P_IP);
644 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
645 net, sk, skb, NULL, skb->dev,
647 !(IPCB(skb)->flags & IPSKB_REROUTED));
650 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
654 struct net *net = dev_net(vrf_dev);
659 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
660 skb, NULL, vrf_dev, vrf_output_direct);
662 if (likely(err == 1))
663 err = vrf_output_direct(net, sk, skb);
665 /* reset skb device */
666 if (likely(err == 1))
674 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
678 /* don't divert multicast or local broadcast */
679 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
680 ipv4_is_lbcast(ip_hdr(skb)->daddr))
683 if (qdisc_tx_is_default(vrf_dev))
684 return vrf_ip_out_direct(vrf_dev, sk, skb);
686 return vrf_ip_out_redirect(vrf_dev, skb);
689 /* called with rcu lock held */
690 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
697 return vrf_ip_out(vrf_dev, sk, skb);
699 return vrf_ip6_out(vrf_dev, sk, skb);
706 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
708 struct rtable *rth = rtnl_dereference(vrf->rth);
709 struct net *net = dev_net(dev);
710 struct dst_entry *dst;
712 RCU_INIT_POINTER(vrf->rth, NULL);
715 /* move dev in dst's to loopback so this VRF device can be deleted
716 * - based on dst_ifdown
721 dst->dev = net->loopback_dev;
727 static int vrf_rtable_create(struct net_device *dev)
729 struct net_vrf *vrf = netdev_priv(dev);
732 if (!fib_new_table(dev_net(dev), vrf->tb_id))
735 /* create a dst for routing packets out through a VRF device */
736 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
740 rth->dst.output = vrf_output;
741 rth->rt_table_id = vrf->tb_id;
743 rcu_assign_pointer(vrf->rth, rth);
748 /**************************** device handling ********************/
750 /* cycle interface to flush neighbor cache and move routes across tables */
751 static void cycle_netdev(struct net_device *dev)
753 unsigned int flags = dev->flags;
756 if (!netif_running(dev))
759 ret = dev_change_flags(dev, flags & ~IFF_UP);
761 ret = dev_change_flags(dev, flags);
765 "Failed to cycle device %s; route tables might be wrong!\n",
770 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
774 /* do not allow loopback device to be enslaved to a VRF.
775 * The vrf device acts as the loopback for the vrf.
777 if (port_dev == dev_net(dev)->loopback_dev)
780 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
781 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
785 cycle_netdev(port_dev);
790 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
794 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
796 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
799 return do_vrf_add_slave(dev, port_dev);
802 /* inverse of do_vrf_add_slave */
803 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
805 netdev_upper_dev_unlink(port_dev, dev);
806 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
808 cycle_netdev(port_dev);
813 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
815 return do_vrf_del_slave(dev, port_dev);
818 static void vrf_dev_uninit(struct net_device *dev)
820 struct net_vrf *vrf = netdev_priv(dev);
822 vrf_rtable_release(dev, vrf);
823 vrf_rt6_release(dev, vrf);
825 free_percpu(dev->dstats);
829 static int vrf_dev_init(struct net_device *dev)
831 struct net_vrf *vrf = netdev_priv(dev);
833 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
837 /* create the default dst which points back to us */
838 if (vrf_rtable_create(dev) != 0)
841 if (vrf_rt6_create(dev) != 0)
844 dev->flags = IFF_MASTER | IFF_NOARP;
846 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
847 dev->mtu = 64 * 1024;
849 /* similarly, oper state is irrelevant; set to up to avoid confusion */
850 dev->operstate = IF_OPER_UP;
851 netdev_lockdep_set_classes(dev);
855 vrf_rtable_release(dev, vrf);
857 free_percpu(dev->dstats);
863 static const struct net_device_ops vrf_netdev_ops = {
864 .ndo_init = vrf_dev_init,
865 .ndo_uninit = vrf_dev_uninit,
866 .ndo_start_xmit = vrf_xmit,
867 .ndo_get_stats64 = vrf_get_stats64,
868 .ndo_add_slave = vrf_add_slave,
869 .ndo_del_slave = vrf_del_slave,
872 static u32 vrf_fib_table(const struct net_device *dev)
874 struct net_vrf *vrf = netdev_priv(dev);
879 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
885 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
887 struct net_device *dev)
889 struct net *net = dev_net(dev);
891 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
892 skb = NULL; /* kfree_skb(skb) handled by nf code */
897 #if IS_ENABLED(CONFIG_IPV6)
898 /* neighbor handling is done with actual device; do not want
899 * to flip skb->dev for those ndisc packets. This really fails
900 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
903 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
905 const struct ipv6hdr *iph = ipv6_hdr(skb);
908 if (iph->nexthdr == NEXTHDR_ICMP) {
909 const struct icmp6hdr *icmph;
910 struct icmp6hdr _icmph;
912 icmph = skb_header_pointer(skb, sizeof(*iph),
913 sizeof(_icmph), &_icmph);
917 switch (icmph->icmp6_type) {
918 case NDISC_ROUTER_SOLICITATION:
919 case NDISC_ROUTER_ADVERTISEMENT:
920 case NDISC_NEIGHBOUR_SOLICITATION:
921 case NDISC_NEIGHBOUR_ADVERTISEMENT:
932 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
933 const struct net_device *dev,
938 struct net_vrf *vrf = netdev_priv(dev);
939 struct fib6_table *table = NULL;
940 struct rt6_info *rt6;
944 /* fib6_table does not have a refcnt and can not be freed */
945 rt6 = rcu_dereference(vrf->rt6);
947 table = rt6->rt6i_table;
954 return ip6_pol_route(net, table, ifindex, fl6, flags);
957 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
960 const struct ipv6hdr *iph = ipv6_hdr(skb);
961 struct flowi6 fl6 = {
962 .flowi6_iif = ifindex,
963 .flowi6_mark = skb->mark,
964 .flowi6_proto = iph->nexthdr,
967 .flowlabel = ip6_flowinfo(iph),
969 struct net *net = dev_net(vrf_dev);
970 struct rt6_info *rt6;
972 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
973 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
977 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
980 skb_dst_set(skb, &rt6->dst);
983 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
986 int orig_iif = skb->skb_iif;
989 /* loopback traffic; do not push through packet taps again.
990 * Reset pkt_type for upper layers to process skb
992 if (skb->pkt_type == PACKET_LOOPBACK) {
994 skb->skb_iif = vrf_dev->ifindex;
995 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
996 skb->pkt_type = PACKET_HOST;
1000 /* if packet is NDISC or addressed to multicast or link-local
1001 * then keep the ingress interface
1003 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1004 if (!ipv6_ndisc_frame(skb) && !need_strict) {
1005 vrf_rx_stats(vrf_dev, skb->len);
1007 skb->skb_iif = vrf_dev->ifindex;
1009 if (!list_empty(&vrf_dev->ptype_all)) {
1010 skb_push(skb, skb->mac_len);
1011 dev_queue_xmit_nit(skb, vrf_dev);
1012 skb_pull(skb, skb->mac_len);
1015 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1019 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1021 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1027 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1028 struct sk_buff *skb)
1034 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1035 struct sk_buff *skb)
1038 skb->skb_iif = vrf_dev->ifindex;
1039 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1041 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1044 /* loopback traffic; do not push through packet taps again.
1045 * Reset pkt_type for upper layers to process skb
1047 if (skb->pkt_type == PACKET_LOOPBACK) {
1048 skb->pkt_type = PACKET_HOST;
1052 vrf_rx_stats(vrf_dev, skb->len);
1054 if (!list_empty(&vrf_dev->ptype_all)) {
1055 skb_push(skb, skb->mac_len);
1056 dev_queue_xmit_nit(skb, vrf_dev);
1057 skb_pull(skb, skb->mac_len);
1060 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1065 /* called with rcu lock held */
1066 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1067 struct sk_buff *skb,
1072 return vrf_ip_rcv(vrf_dev, skb);
1074 return vrf_ip6_rcv(vrf_dev, skb);
1080 #if IS_ENABLED(CONFIG_IPV6)
1081 /* send to link-local or multicast address via interface enslaved to
1082 * VRF device. Force lookup to VRF table without changing flow struct
1084 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1087 struct net *net = dev_net(dev);
1088 int flags = RT6_LOOKUP_F_IFACE;
1089 struct dst_entry *dst = NULL;
1090 struct rt6_info *rt;
1092 /* VRF device does not have a link-local address and
1093 * sending packets to link-local or mcast addresses over
1094 * a VRF device does not make sense
1096 if (fl6->flowi6_oif == dev->ifindex) {
1097 dst = &net->ipv6.ip6_null_entry->dst;
1102 if (!ipv6_addr_any(&fl6->saddr))
1103 flags |= RT6_LOOKUP_F_HAS_SADDR;
1105 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1113 static const struct l3mdev_ops vrf_l3mdev_ops = {
1114 .l3mdev_fib_table = vrf_fib_table,
1115 .l3mdev_l3_rcv = vrf_l3_rcv,
1116 .l3mdev_l3_out = vrf_l3_out,
1117 #if IS_ENABLED(CONFIG_IPV6)
1118 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1122 static void vrf_get_drvinfo(struct net_device *dev,
1123 struct ethtool_drvinfo *info)
1125 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1126 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1129 static const struct ethtool_ops vrf_ethtool_ops = {
1130 .get_drvinfo = vrf_get_drvinfo,
1133 static inline size_t vrf_fib_rule_nl_size(void)
1137 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1138 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1139 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1144 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1146 struct fib_rule_hdr *frh;
1147 struct nlmsghdr *nlh;
1148 struct sk_buff *skb;
1151 if (family == AF_INET6 && !ipv6_mod_enabled())
1154 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1158 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1160 goto nla_put_failure;
1162 /* rule only needs to appear once */
1163 nlh->nlmsg_flags |= NLM_F_EXCL;
1165 frh = nlmsg_data(nlh);
1166 memset(frh, 0, sizeof(*frh));
1167 frh->family = family;
1168 frh->action = FR_ACT_TO_TBL;
1170 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1171 goto nla_put_failure;
1173 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1174 goto nla_put_failure;
1176 nlmsg_end(skb, nlh);
1178 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1179 skb->sk = dev_net(dev)->rtnl;
1181 err = fib_nl_newrule(skb, nlh, NULL);
1185 err = fib_nl_delrule(skb, nlh, NULL);
1199 static int vrf_add_fib_rules(const struct net_device *dev)
1203 err = vrf_fib_rule(dev, AF_INET, true);
1207 err = vrf_fib_rule(dev, AF_INET6, true);
1211 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1212 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1219 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1221 vrf_fib_rule(dev, AF_INET6, false);
1225 vrf_fib_rule(dev, AF_INET, false);
1228 netdev_err(dev, "Failed to add FIB rules.\n");
1232 static void vrf_setup(struct net_device *dev)
1236 /* Initialize the device structure. */
1237 dev->netdev_ops = &vrf_netdev_ops;
1238 dev->l3mdev_ops = &vrf_l3mdev_ops;
1239 dev->ethtool_ops = &vrf_ethtool_ops;
1240 dev->needs_free_netdev = true;
1242 /* Fill in device structure with ethernet-generic values. */
1243 eth_hw_addr_random(dev);
1245 /* don't acquire vrf device's netif_tx_lock when transmitting */
1246 dev->features |= NETIF_F_LLTX;
1248 /* don't allow vrf devices to change network namespaces. */
1249 dev->features |= NETIF_F_NETNS_LOCAL;
1251 /* does not make sense for a VLAN to be added to a vrf device */
1252 dev->features |= NETIF_F_VLAN_CHALLENGED;
1254 /* enable offload features */
1255 dev->features |= NETIF_F_GSO_SOFTWARE;
1256 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1257 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1259 dev->hw_features = dev->features;
1260 dev->hw_enc_features = dev->features;
1262 /* default to no qdisc; user can add if desired */
1263 dev->priv_flags |= IFF_NO_QUEUE;
1266 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1267 struct netlink_ext_ack *extack)
1269 if (tb[IFLA_ADDRESS]) {
1270 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1271 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1274 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1275 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1276 return -EADDRNOTAVAIL;
1282 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1284 struct net_device *port_dev;
1285 struct list_head *iter;
1287 netdev_for_each_lower_dev(dev, port_dev, iter)
1288 vrf_del_slave(dev, port_dev);
1290 unregister_netdevice_queue(dev, head);
1293 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1294 struct nlattr *tb[], struct nlattr *data[],
1295 struct netlink_ext_ack *extack)
1297 struct net_vrf *vrf = netdev_priv(dev);
1298 bool *add_fib_rules;
1302 if (!data || !data[IFLA_VRF_TABLE]) {
1303 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1307 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1308 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1309 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1310 "Invalid VRF table id");
1314 dev->priv_flags |= IFF_L3MDEV_MASTER;
1316 err = register_netdevice(dev);
1321 add_fib_rules = net_generic(net, vrf_net_id);
1322 if (*add_fib_rules) {
1323 err = vrf_add_fib_rules(dev);
1325 unregister_netdevice(dev);
1328 *add_fib_rules = false;
1335 static size_t vrf_nl_getsize(const struct net_device *dev)
1337 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1340 static int vrf_fillinfo(struct sk_buff *skb,
1341 const struct net_device *dev)
1343 struct net_vrf *vrf = netdev_priv(dev);
1345 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1348 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1349 const struct net_device *slave_dev)
1351 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1354 static int vrf_fill_slave_info(struct sk_buff *skb,
1355 const struct net_device *vrf_dev,
1356 const struct net_device *slave_dev)
1358 struct net_vrf *vrf = netdev_priv(vrf_dev);
1360 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1366 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1367 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1370 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1372 .priv_size = sizeof(struct net_vrf),
1374 .get_size = vrf_nl_getsize,
1375 .policy = vrf_nl_policy,
1376 .validate = vrf_validate,
1377 .fill_info = vrf_fillinfo,
1379 .get_slave_size = vrf_get_slave_size,
1380 .fill_slave_info = vrf_fill_slave_info,
1382 .newlink = vrf_newlink,
1383 .dellink = vrf_dellink,
1385 .maxtype = IFLA_VRF_MAX,
1388 static int vrf_device_event(struct notifier_block *unused,
1389 unsigned long event, void *ptr)
1391 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1393 /* only care about unregister events to drop slave references */
1394 if (event == NETDEV_UNREGISTER) {
1395 struct net_device *vrf_dev;
1397 if (!netif_is_l3_slave(dev))
1400 vrf_dev = netdev_master_upper_dev_get(dev);
1401 vrf_del_slave(vrf_dev, dev);
1407 static struct notifier_block vrf_notifier_block __read_mostly = {
1408 .notifier_call = vrf_device_event,
1411 /* Initialize per network namespace state */
1412 static int __net_init vrf_netns_init(struct net *net)
1414 bool *add_fib_rules = net_generic(net, vrf_net_id);
1416 *add_fib_rules = true;
1421 static struct pernet_operations vrf_net_ops __net_initdata = {
1422 .init = vrf_netns_init,
1424 .size = sizeof(bool),
1427 static int __init vrf_init_module(void)
1431 register_netdevice_notifier(&vrf_notifier_block);
1433 rc = register_pernet_subsys(&vrf_net_ops);
1437 rc = rtnl_link_register(&vrf_link_ops);
1439 unregister_pernet_subsys(&vrf_net_ops);
1446 unregister_netdevice_notifier(&vrf_notifier_block);
1450 module_init(vrf_init_module);
1451 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1452 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1453 MODULE_LICENSE("GPL");
1454 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1455 MODULE_VERSION(DRV_VERSION);