1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * vrf.c: device driver to encapsulate a VRF space
5 * Copyright (c) 2015 Cumulus Networks. All rights reserved.
6 * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com>
7 * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com>
9 * Based on dummy, team and ipvlan drivers
12 #include <linux/module.h>
13 #include <linux/kernel.h>
14 #include <linux/netdevice.h>
15 #include <linux/etherdevice.h>
17 #include <linux/init.h>
18 #include <linux/moduleparam.h>
19 #include <linux/netfilter.h>
20 #include <linux/rtnetlink.h>
21 #include <net/rtnetlink.h>
22 #include <linux/u64_stats_sync.h>
23 #include <linux/hashtable.h>
25 #include <linux/inetdevice.h>
28 #include <net/ip_fib.h>
29 #include <net/ip6_fib.h>
30 #include <net/ip6_route.h>
31 #include <net/route.h>
32 #include <net/addrconf.h>
33 #include <net/l3mdev.h>
34 #include <net/fib_rules.h>
35 #include <net/netns/generic.h>
37 #define DRV_NAME "vrf"
38 #define DRV_VERSION "1.0"
40 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */
42 static unsigned int vrf_net_id;
45 struct rtable __rcu *rth;
46 struct rt6_info __rcu *rt6;
47 #if IS_ENABLED(CONFIG_IPV6)
48 struct fib6_table *fib6_table;
60 struct u64_stats_sync syncp;
63 static void vrf_rx_stats(struct net_device *dev, int len)
65 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
67 u64_stats_update_begin(&dstats->syncp);
69 dstats->rx_bytes += len;
70 u64_stats_update_end(&dstats->syncp);
73 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
75 vrf_dev->stats.tx_errors++;
79 static void vrf_get_stats64(struct net_device *dev,
80 struct rtnl_link_stats64 *stats)
84 for_each_possible_cpu(i) {
85 const struct pcpu_dstats *dstats;
86 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
89 dstats = per_cpu_ptr(dev->dstats, i);
91 start = u64_stats_fetch_begin_irq(&dstats->syncp);
92 tbytes = dstats->tx_bytes;
93 tpkts = dstats->tx_pkts;
94 tdrops = dstats->tx_drps;
95 rbytes = dstats->rx_bytes;
96 rpkts = dstats->rx_pkts;
97 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
98 stats->tx_bytes += tbytes;
99 stats->tx_packets += tpkts;
100 stats->tx_dropped += tdrops;
101 stats->rx_bytes += rbytes;
102 stats->rx_packets += rpkts;
106 /* by default VRF devices do not have a qdisc and are expected
107 * to be created with only a single queue.
109 static bool qdisc_tx_is_default(const struct net_device *dev)
111 struct netdev_queue *txq;
114 if (dev->num_tx_queues > 1)
117 txq = netdev_get_tx_queue(dev, 0);
118 qdisc = rcu_access_pointer(txq->qdisc);
120 return !qdisc->enqueue;
123 /* Local traffic destined to local address. Reinsert the packet to rx
124 * path, similar to loopback handling.
126 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
127 struct dst_entry *dst)
133 skb_dst_set(skb, dst);
135 /* set pkt_type to avoid skb hitting packet taps twice -
136 * once on Tx and again in Rx processing
138 skb->pkt_type = PACKET_LOOPBACK;
140 skb->protocol = eth_type_trans(skb, dev);
142 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
143 vrf_rx_stats(dev, len);
145 this_cpu_inc(dev->dstats->rx_drps);
150 #if IS_ENABLED(CONFIG_IPV6)
151 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
156 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
157 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
159 if (likely(err == 1))
160 err = dst_output(net, sk, skb);
165 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
166 struct net_device *dev)
168 const struct ipv6hdr *iph;
169 struct net *net = dev_net(skb->dev);
171 int ret = NET_XMIT_DROP;
172 struct dst_entry *dst;
173 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
175 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
180 memset(&fl6, 0, sizeof(fl6));
181 /* needed to match OIF rule */
182 fl6.flowi6_oif = dev->ifindex;
183 fl6.flowi6_iif = LOOPBACK_IFINDEX;
184 fl6.daddr = iph->daddr;
185 fl6.saddr = iph->saddr;
186 fl6.flowlabel = ip6_flowinfo(iph);
187 fl6.flowi6_mark = skb->mark;
188 fl6.flowi6_proto = iph->nexthdr;
189 fl6.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF;
191 dst = ip6_route_output(net, NULL, &fl6);
197 /* if dst.dev is loopback or the VRF device again this is locally
198 * originated traffic destined to a local address. Short circuit
202 return vrf_local_xmit(skb, dev, dst);
204 skb_dst_set(skb, dst);
206 /* strip the ethernet header added for pass through VRF device */
207 __skb_pull(skb, skb_network_offset(skb));
209 ret = vrf_ip6_local_out(net, skb->sk, skb);
210 if (unlikely(net_xmit_eval(ret)))
211 dev->stats.tx_errors++;
213 ret = NET_XMIT_SUCCESS;
217 vrf_tx_error(dev, skb);
218 return NET_XMIT_DROP;
221 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
222 struct net_device *dev)
224 vrf_tx_error(dev, skb);
225 return NET_XMIT_DROP;
229 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
230 static int vrf_ip_local_out(struct net *net, struct sock *sk,
235 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
236 skb, NULL, skb_dst(skb)->dev, dst_output);
237 if (likely(err == 1))
238 err = dst_output(net, sk, skb);
243 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
244 struct net_device *vrf_dev)
247 int ret = NET_XMIT_DROP;
249 struct net *net = dev_net(vrf_dev);
252 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
257 memset(&fl4, 0, sizeof(fl4));
258 /* needed to match OIF rule */
259 fl4.flowi4_oif = vrf_dev->ifindex;
260 fl4.flowi4_iif = LOOPBACK_IFINDEX;
261 fl4.flowi4_tos = RT_TOS(ip4h->tos);
262 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF;
263 fl4.flowi4_proto = ip4h->protocol;
264 fl4.daddr = ip4h->daddr;
265 fl4.saddr = ip4h->saddr;
267 rt = ip_route_output_flow(net, &fl4, NULL);
273 /* if dst.dev is loopback or the VRF device again this is locally
274 * originated traffic destined to a local address. Short circuit
277 if (rt->dst.dev == vrf_dev)
278 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
280 skb_dst_set(skb, &rt->dst);
282 /* strip the ethernet header added for pass through VRF device */
283 __skb_pull(skb, skb_network_offset(skb));
286 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
290 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
291 if (unlikely(net_xmit_eval(ret)))
292 vrf_dev->stats.tx_errors++;
294 ret = NET_XMIT_SUCCESS;
299 vrf_tx_error(vrf_dev, skb);
303 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
305 switch (skb->protocol) {
306 case htons(ETH_P_IP):
307 return vrf_process_v4_outbound(skb, dev);
308 case htons(ETH_P_IPV6):
309 return vrf_process_v6_outbound(skb, dev);
311 vrf_tx_error(dev, skb);
312 return NET_XMIT_DROP;
316 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
319 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
321 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
322 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
324 u64_stats_update_begin(&dstats->syncp);
326 dstats->tx_bytes += len;
327 u64_stats_update_end(&dstats->syncp);
329 this_cpu_inc(dev->dstats->tx_drps);
335 static int vrf_finish_direct(struct net *net, struct sock *sk,
338 struct net_device *vrf_dev = skb->dev;
340 if (!list_empty(&vrf_dev->ptype_all) &&
341 likely(skb_headroom(skb) >= ETH_HLEN)) {
342 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
344 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
345 eth_zero_addr(eth->h_dest);
346 eth->h_proto = skb->protocol;
349 dev_queue_xmit_nit(skb, vrf_dev);
350 rcu_read_unlock_bh();
352 skb_pull(skb, ETH_HLEN);
358 #if IS_ENABLED(CONFIG_IPV6)
359 /* modelled after ip6_finish_output2 */
360 static int vrf_finish_output6(struct net *net, struct sock *sk,
363 struct dst_entry *dst = skb_dst(skb);
364 struct net_device *dev = dst->dev;
365 const struct in6_addr *nexthop;
366 struct neighbour *neigh;
371 skb->protocol = htons(ETH_P_IPV6);
375 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
376 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
377 if (unlikely(!neigh))
378 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
379 if (!IS_ERR(neigh)) {
380 sock_confirm_neigh(skb, neigh);
381 ret = neigh_output(neigh, skb, false);
382 rcu_read_unlock_bh();
385 rcu_read_unlock_bh();
387 IP6_INC_STATS(dev_net(dst->dev),
388 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
393 /* modelled after ip6_output */
394 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
396 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
397 net, sk, skb, NULL, skb_dst(skb)->dev,
399 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
402 /* set dst on skb to send packet to us via dev_xmit path. Allows
403 * packet to go through device based features such as qdisc, netfilter
404 * hooks and packet sockets with skb->dev set to vrf device.
406 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
409 struct net_vrf *vrf = netdev_priv(vrf_dev);
410 struct dst_entry *dst = NULL;
411 struct rt6_info *rt6;
415 rt6 = rcu_dereference(vrf->rt6);
423 if (unlikely(!dst)) {
424 vrf_tx_error(vrf_dev, skb);
429 skb_dst_set(skb, dst);
434 static int vrf_output6_direct(struct net *net, struct sock *sk,
437 skb->protocol = htons(ETH_P_IPV6);
439 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
440 net, sk, skb, NULL, skb->dev,
442 !(IPCB(skb)->flags & IPSKB_REROUTED));
445 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
449 struct net *net = dev_net(vrf_dev);
454 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
455 skb, NULL, vrf_dev, vrf_output6_direct);
457 if (likely(err == 1))
458 err = vrf_output6_direct(net, sk, skb);
460 /* reset skb device */
461 if (likely(err == 1))
469 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
473 /* don't divert link scope packets */
474 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
477 if (qdisc_tx_is_default(vrf_dev) ||
478 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
479 return vrf_ip6_out_direct(vrf_dev, sk, skb);
481 return vrf_ip6_out_redirect(vrf_dev, skb);
485 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
487 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
488 struct net *net = dev_net(dev);
489 struct dst_entry *dst;
491 RCU_INIT_POINTER(vrf->rt6, NULL);
494 /* move dev in dst's to loopback so this VRF device can be deleted
495 * - based on dst_ifdown
500 dst->dev = net->loopback_dev;
506 static int vrf_rt6_create(struct net_device *dev)
508 int flags = DST_NOPOLICY | DST_NOXFRM;
509 struct net_vrf *vrf = netdev_priv(dev);
510 struct net *net = dev_net(dev);
511 struct rt6_info *rt6;
514 /* IPv6 can be CONFIG enabled and then disabled runtime */
515 if (!ipv6_mod_enabled())
518 vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
519 if (!vrf->fib6_table)
522 /* create a dst for routing packets out a VRF device */
523 rt6 = ip6_dst_alloc(net, dev, flags);
527 rt6->dst.output = vrf_output6;
529 rcu_assign_pointer(vrf->rt6, rt6);
536 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
543 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
547 static int vrf_rt6_create(struct net_device *dev)
553 /* modelled after ip_finish_output2 */
554 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
556 struct dst_entry *dst = skb_dst(skb);
557 struct rtable *rt = (struct rtable *)dst;
558 struct net_device *dev = dst->dev;
559 unsigned int hh_len = LL_RESERVED_SPACE(dev);
560 struct neighbour *neigh;
561 bool is_v6gw = false;
566 /* Be paranoid, rather than too clever. */
567 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
568 struct sk_buff *skb2;
570 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
576 skb_set_owner_w(skb2, skb->sk);
584 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
585 if (!IS_ERR(neigh)) {
586 sock_confirm_neigh(skb, neigh);
587 /* if crossing protocols, can not use the cached header */
588 ret = neigh_output(neigh, skb, is_v6gw);
589 rcu_read_unlock_bh();
593 rcu_read_unlock_bh();
595 vrf_tx_error(skb->dev, skb);
599 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
601 struct net_device *dev = skb_dst(skb)->dev;
603 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
606 skb->protocol = htons(ETH_P_IP);
608 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
609 net, sk, skb, NULL, dev,
611 !(IPCB(skb)->flags & IPSKB_REROUTED));
614 /* set dst on skb to send packet to us via dev_xmit path. Allows
615 * packet to go through device based features such as qdisc, netfilter
616 * hooks and packet sockets with skb->dev set to vrf device.
618 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
621 struct net_vrf *vrf = netdev_priv(vrf_dev);
622 struct dst_entry *dst = NULL;
627 rth = rcu_dereference(vrf->rth);
635 if (unlikely(!dst)) {
636 vrf_tx_error(vrf_dev, skb);
641 skb_dst_set(skb, dst);
646 static int vrf_output_direct(struct net *net, struct sock *sk,
649 skb->protocol = htons(ETH_P_IP);
651 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
652 net, sk, skb, NULL, skb->dev,
654 !(IPCB(skb)->flags & IPSKB_REROUTED));
657 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
661 struct net *net = dev_net(vrf_dev);
666 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
667 skb, NULL, vrf_dev, vrf_output_direct);
669 if (likely(err == 1))
670 err = vrf_output_direct(net, sk, skb);
672 /* reset skb device */
673 if (likely(err == 1))
681 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
685 /* don't divert multicast or local broadcast */
686 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
687 ipv4_is_lbcast(ip_hdr(skb)->daddr))
690 if (qdisc_tx_is_default(vrf_dev) ||
691 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
692 return vrf_ip_out_direct(vrf_dev, sk, skb);
694 return vrf_ip_out_redirect(vrf_dev, skb);
697 /* called with rcu lock held */
698 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
705 return vrf_ip_out(vrf_dev, sk, skb);
707 return vrf_ip6_out(vrf_dev, sk, skb);
714 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
716 struct rtable *rth = rtnl_dereference(vrf->rth);
717 struct net *net = dev_net(dev);
718 struct dst_entry *dst;
720 RCU_INIT_POINTER(vrf->rth, NULL);
723 /* move dev in dst's to loopback so this VRF device can be deleted
724 * - based on dst_ifdown
729 dst->dev = net->loopback_dev;
735 static int vrf_rtable_create(struct net_device *dev)
737 struct net_vrf *vrf = netdev_priv(dev);
740 if (!fib_new_table(dev_net(dev), vrf->tb_id))
743 /* create a dst for routing packets out through a VRF device */
744 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1);
748 rth->dst.output = vrf_output;
750 rcu_assign_pointer(vrf->rth, rth);
755 /**************************** device handling ********************/
757 /* cycle interface to flush neighbor cache and move routes across tables */
758 static void cycle_netdev(struct net_device *dev,
759 struct netlink_ext_ack *extack)
761 unsigned int flags = dev->flags;
764 if (!netif_running(dev))
767 ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
769 ret = dev_change_flags(dev, flags, extack);
773 "Failed to cycle device %s; route tables might be wrong!\n",
778 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
779 struct netlink_ext_ack *extack)
783 /* do not allow loopback device to be enslaved to a VRF.
784 * The vrf device acts as the loopback for the vrf.
786 if (port_dev == dev_net(dev)->loopback_dev) {
787 NL_SET_ERR_MSG(extack,
788 "Can not enslave loopback device to a VRF");
792 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
793 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
797 cycle_netdev(port_dev, extack);
802 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
806 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
807 struct netlink_ext_ack *extack)
809 if (netif_is_l3_master(port_dev)) {
810 NL_SET_ERR_MSG(extack,
811 "Can not enslave an L3 master device to a VRF");
815 if (netif_is_l3_slave(port_dev))
818 return do_vrf_add_slave(dev, port_dev, extack);
821 /* inverse of do_vrf_add_slave */
822 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
824 netdev_upper_dev_unlink(port_dev, dev);
825 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
827 cycle_netdev(port_dev, NULL);
832 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
834 return do_vrf_del_slave(dev, port_dev);
837 static void vrf_dev_uninit(struct net_device *dev)
839 struct net_vrf *vrf = netdev_priv(dev);
841 vrf_rtable_release(dev, vrf);
842 vrf_rt6_release(dev, vrf);
844 free_percpu(dev->dstats);
848 static int vrf_dev_init(struct net_device *dev)
850 struct net_vrf *vrf = netdev_priv(dev);
852 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
856 /* create the default dst which points back to us */
857 if (vrf_rtable_create(dev) != 0)
860 if (vrf_rt6_create(dev) != 0)
863 dev->flags = IFF_MASTER | IFF_NOARP;
865 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
866 dev->mtu = 64 * 1024;
868 /* similarly, oper state is irrelevant; set to up to avoid confusion */
869 dev->operstate = IF_OPER_UP;
873 vrf_rtable_release(dev, vrf);
875 free_percpu(dev->dstats);
881 static const struct net_device_ops vrf_netdev_ops = {
882 .ndo_init = vrf_dev_init,
883 .ndo_uninit = vrf_dev_uninit,
884 .ndo_start_xmit = vrf_xmit,
885 .ndo_set_mac_address = eth_mac_addr,
886 .ndo_get_stats64 = vrf_get_stats64,
887 .ndo_add_slave = vrf_add_slave,
888 .ndo_del_slave = vrf_del_slave,
891 static u32 vrf_fib_table(const struct net_device *dev)
893 struct net_vrf *vrf = netdev_priv(dev);
898 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
904 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
906 struct net_device *dev)
908 struct net *net = dev_net(dev);
910 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
911 skb = NULL; /* kfree_skb(skb) handled by nf code */
916 #if IS_ENABLED(CONFIG_IPV6)
917 /* neighbor handling is done with actual device; do not want
918 * to flip skb->dev for those ndisc packets. This really fails
919 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
922 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
924 const struct ipv6hdr *iph = ipv6_hdr(skb);
927 if (iph->nexthdr == NEXTHDR_ICMP) {
928 const struct icmp6hdr *icmph;
929 struct icmp6hdr _icmph;
931 icmph = skb_header_pointer(skb, sizeof(*iph),
932 sizeof(_icmph), &_icmph);
936 switch (icmph->icmp6_type) {
937 case NDISC_ROUTER_SOLICITATION:
938 case NDISC_ROUTER_ADVERTISEMENT:
939 case NDISC_NEIGHBOUR_SOLICITATION:
940 case NDISC_NEIGHBOUR_ADVERTISEMENT:
951 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
952 const struct net_device *dev,
955 const struct sk_buff *skb,
958 struct net_vrf *vrf = netdev_priv(dev);
960 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
963 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
966 const struct ipv6hdr *iph = ipv6_hdr(skb);
967 struct flowi6 fl6 = {
968 .flowi6_iif = ifindex,
969 .flowi6_mark = skb->mark,
970 .flowi6_proto = iph->nexthdr,
973 .flowlabel = ip6_flowinfo(iph),
975 struct net *net = dev_net(vrf_dev);
976 struct rt6_info *rt6;
978 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
979 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
983 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
986 skb_dst_set(skb, &rt6->dst);
989 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
992 int orig_iif = skb->skb_iif;
993 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
994 bool is_ndisc = ipv6_ndisc_frame(skb);
996 /* loopback, multicast & non-ND link-local traffic; do not push through
997 * packet taps again. Reset pkt_type for upper layers to process skb
999 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
1001 skb->skb_iif = vrf_dev->ifindex;
1002 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1003 if (skb->pkt_type == PACKET_LOOPBACK)
1004 skb->pkt_type = PACKET_HOST;
1008 /* if packet is NDISC then keep the ingress interface */
1010 vrf_rx_stats(vrf_dev, skb->len);
1012 skb->skb_iif = vrf_dev->ifindex;
1014 if (!list_empty(&vrf_dev->ptype_all)) {
1015 skb_push(skb, skb->mac_len);
1016 dev_queue_xmit_nit(skb, vrf_dev);
1017 skb_pull(skb, skb->mac_len);
1020 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1024 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1026 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1032 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1033 struct sk_buff *skb)
1039 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1040 struct sk_buff *skb)
1043 skb->skb_iif = vrf_dev->ifindex;
1044 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1046 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1049 /* loopback traffic; do not push through packet taps again.
1050 * Reset pkt_type for upper layers to process skb
1052 if (skb->pkt_type == PACKET_LOOPBACK) {
1053 skb->pkt_type = PACKET_HOST;
1057 vrf_rx_stats(vrf_dev, skb->len);
1059 if (!list_empty(&vrf_dev->ptype_all)) {
1060 skb_push(skb, skb->mac_len);
1061 dev_queue_xmit_nit(skb, vrf_dev);
1062 skb_pull(skb, skb->mac_len);
1065 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1070 /* called with rcu lock held */
1071 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1072 struct sk_buff *skb,
1077 return vrf_ip_rcv(vrf_dev, skb);
1079 return vrf_ip6_rcv(vrf_dev, skb);
1085 #if IS_ENABLED(CONFIG_IPV6)
1086 /* send to link-local or multicast address via interface enslaved to
1087 * VRF device. Force lookup to VRF table without changing flow struct
1088 * Note: Caller to this function must hold rcu_read_lock() and no refcnt
1089 * is taken on the dst by this function.
1091 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1094 struct net *net = dev_net(dev);
1095 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;
1096 struct dst_entry *dst = NULL;
1097 struct rt6_info *rt;
1099 /* VRF device does not have a link-local address and
1100 * sending packets to link-local or mcast addresses over
1101 * a VRF device does not make sense
1103 if (fl6->flowi6_oif == dev->ifindex) {
1104 dst = &net->ipv6.ip6_null_entry->dst;
1108 if (!ipv6_addr_any(&fl6->saddr))
1109 flags |= RT6_LOOKUP_F_HAS_SADDR;
1111 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1119 static const struct l3mdev_ops vrf_l3mdev_ops = {
1120 .l3mdev_fib_table = vrf_fib_table,
1121 .l3mdev_l3_rcv = vrf_l3_rcv,
1122 .l3mdev_l3_out = vrf_l3_out,
1123 #if IS_ENABLED(CONFIG_IPV6)
1124 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1128 static void vrf_get_drvinfo(struct net_device *dev,
1129 struct ethtool_drvinfo *info)
1131 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1132 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1135 static const struct ethtool_ops vrf_ethtool_ops = {
1136 .get_drvinfo = vrf_get_drvinfo,
1139 static inline size_t vrf_fib_rule_nl_size(void)
1143 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1144 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1145 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1146 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */
1151 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1153 struct fib_rule_hdr *frh;
1154 struct nlmsghdr *nlh;
1155 struct sk_buff *skb;
1158 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&
1159 !ipv6_mod_enabled())
1162 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1166 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1168 goto nla_put_failure;
1170 /* rule only needs to appear once */
1171 nlh->nlmsg_flags |= NLM_F_EXCL;
1173 frh = nlmsg_data(nlh);
1174 memset(frh, 0, sizeof(*frh));
1175 frh->family = family;
1176 frh->action = FR_ACT_TO_TBL;
1178 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1179 goto nla_put_failure;
1181 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1182 goto nla_put_failure;
1184 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1185 goto nla_put_failure;
1187 nlmsg_end(skb, nlh);
1189 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1190 skb->sk = dev_net(dev)->rtnl;
1192 err = fib_nl_newrule(skb, nlh, NULL);
1196 err = fib_nl_delrule(skb, nlh, NULL);
1210 static int vrf_add_fib_rules(const struct net_device *dev)
1214 err = vrf_fib_rule(dev, AF_INET, true);
1218 err = vrf_fib_rule(dev, AF_INET6, true);
1222 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1223 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1228 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1229 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1236 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1238 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false);
1241 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1243 vrf_fib_rule(dev, AF_INET6, false);
1247 vrf_fib_rule(dev, AF_INET, false);
1250 netdev_err(dev, "Failed to add FIB rules.\n");
1254 static void vrf_setup(struct net_device *dev)
1258 /* Initialize the device structure. */
1259 dev->netdev_ops = &vrf_netdev_ops;
1260 dev->l3mdev_ops = &vrf_l3mdev_ops;
1261 dev->ethtool_ops = &vrf_ethtool_ops;
1262 dev->needs_free_netdev = true;
1264 /* Fill in device structure with ethernet-generic values. */
1265 eth_hw_addr_random(dev);
1267 /* don't acquire vrf device's netif_tx_lock when transmitting */
1268 dev->features |= NETIF_F_LLTX;
1270 /* don't allow vrf devices to change network namespaces. */
1271 dev->features |= NETIF_F_NETNS_LOCAL;
1273 /* does not make sense for a VLAN to be added to a vrf device */
1274 dev->features |= NETIF_F_VLAN_CHALLENGED;
1276 /* enable offload features */
1277 dev->features |= NETIF_F_GSO_SOFTWARE;
1278 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1279 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1281 dev->hw_features = dev->features;
1282 dev->hw_enc_features = dev->features;
1284 /* default to no qdisc; user can add if desired */
1285 dev->priv_flags |= IFF_NO_QUEUE;
1286 dev->priv_flags |= IFF_NO_RX_HANDLER;
1287 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1289 /* VRF devices do not care about MTU, but if the MTU is set
1290 * too low then the ipv4 and ipv6 protocols are disabled
1291 * which breaks networking.
1293 dev->min_mtu = IPV6_MIN_MTU;
1294 dev->max_mtu = ETH_MAX_MTU;
1297 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1298 struct netlink_ext_ack *extack)
1300 if (tb[IFLA_ADDRESS]) {
1301 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1302 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1305 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1306 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1307 return -EADDRNOTAVAIL;
1313 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1315 struct net_device *port_dev;
1316 struct list_head *iter;
1318 netdev_for_each_lower_dev(dev, port_dev, iter)
1319 vrf_del_slave(dev, port_dev);
1321 unregister_netdevice_queue(dev, head);
1324 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1325 struct nlattr *tb[], struct nlattr *data[],
1326 struct netlink_ext_ack *extack)
1328 struct net_vrf *vrf = netdev_priv(dev);
1329 bool *add_fib_rules;
1333 if (!data || !data[IFLA_VRF_TABLE]) {
1334 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1338 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1339 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1340 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1341 "Invalid VRF table id");
1345 dev->priv_flags |= IFF_L3MDEV_MASTER;
1347 err = register_netdevice(dev);
1352 add_fib_rules = net_generic(net, vrf_net_id);
1353 if (*add_fib_rules) {
1354 err = vrf_add_fib_rules(dev);
1356 unregister_netdevice(dev);
1359 *add_fib_rules = false;
1366 static size_t vrf_nl_getsize(const struct net_device *dev)
1368 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1371 static int vrf_fillinfo(struct sk_buff *skb,
1372 const struct net_device *dev)
1374 struct net_vrf *vrf = netdev_priv(dev);
1376 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1379 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1380 const struct net_device *slave_dev)
1382 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1385 static int vrf_fill_slave_info(struct sk_buff *skb,
1386 const struct net_device *vrf_dev,
1387 const struct net_device *slave_dev)
1389 struct net_vrf *vrf = netdev_priv(vrf_dev);
1391 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1397 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1398 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1401 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1403 .priv_size = sizeof(struct net_vrf),
1405 .get_size = vrf_nl_getsize,
1406 .policy = vrf_nl_policy,
1407 .validate = vrf_validate,
1408 .fill_info = vrf_fillinfo,
1410 .get_slave_size = vrf_get_slave_size,
1411 .fill_slave_info = vrf_fill_slave_info,
1413 .newlink = vrf_newlink,
1414 .dellink = vrf_dellink,
1416 .maxtype = IFLA_VRF_MAX,
1419 static int vrf_device_event(struct notifier_block *unused,
1420 unsigned long event, void *ptr)
1422 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1424 /* only care about unregister events to drop slave references */
1425 if (event == NETDEV_UNREGISTER) {
1426 struct net_device *vrf_dev;
1428 if (!netif_is_l3_slave(dev))
1431 vrf_dev = netdev_master_upper_dev_get(dev);
1432 vrf_del_slave(vrf_dev, dev);
1438 static struct notifier_block vrf_notifier_block __read_mostly = {
1439 .notifier_call = vrf_device_event,
1442 /* Initialize per network namespace state */
1443 static int __net_init vrf_netns_init(struct net *net)
1445 bool *add_fib_rules = net_generic(net, vrf_net_id);
1447 *add_fib_rules = true;
1452 static struct pernet_operations vrf_net_ops __net_initdata = {
1453 .init = vrf_netns_init,
1455 .size = sizeof(bool),
1458 static int __init vrf_init_module(void)
1462 register_netdevice_notifier(&vrf_notifier_block);
1464 rc = register_pernet_subsys(&vrf_net_ops);
1468 rc = rtnl_link_register(&vrf_link_ops);
1470 unregister_pernet_subsys(&vrf_net_ops);
1477 unregister_netdevice_notifier(&vrf_notifier_block);
1481 module_init(vrf_init_module);
1482 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1483 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1484 MODULE_LICENSE("GPL");
1485 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1486 MODULE_VERSION(DRV_VERSION);