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;
171 struct net *net = dev_net(skb->dev);
173 int ret = NET_XMIT_DROP;
174 struct dst_entry *dst;
175 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
177 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
182 memset(&fl6, 0, sizeof(fl6));
183 /* needed to match OIF rule */
184 fl6.flowi6_oif = dev->ifindex;
185 fl6.flowi6_iif = LOOPBACK_IFINDEX;
186 fl6.daddr = iph->daddr;
187 fl6.saddr = iph->saddr;
188 fl6.flowlabel = ip6_flowinfo(iph);
189 fl6.flowi6_mark = skb->mark;
190 fl6.flowi6_proto = iph->nexthdr;
191 fl6.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF;
193 dst = ip6_route_output(net, NULL, &fl6);
199 /* if dst.dev is loopback or the VRF device again this is locally
200 * originated traffic destined to a local address. Short circuit
204 return vrf_local_xmit(skb, dev, dst);
206 skb_dst_set(skb, dst);
208 /* strip the ethernet header added for pass through VRF device */
209 __skb_pull(skb, skb_network_offset(skb));
211 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
212 ret = vrf_ip6_local_out(net, skb->sk, skb);
213 if (unlikely(net_xmit_eval(ret)))
214 dev->stats.tx_errors++;
216 ret = NET_XMIT_SUCCESS;
220 vrf_tx_error(dev, skb);
221 return NET_XMIT_DROP;
224 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
225 struct net_device *dev)
227 vrf_tx_error(dev, skb);
228 return NET_XMIT_DROP;
232 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
233 static int vrf_ip_local_out(struct net *net, struct sock *sk,
238 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
239 skb, NULL, skb_dst(skb)->dev, dst_output);
240 if (likely(err == 1))
241 err = dst_output(net, sk, skb);
246 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
247 struct net_device *vrf_dev)
250 int ret = NET_XMIT_DROP;
252 struct net *net = dev_net(vrf_dev);
255 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
260 memset(&fl4, 0, sizeof(fl4));
261 /* needed to match OIF rule */
262 fl4.flowi4_oif = vrf_dev->ifindex;
263 fl4.flowi4_iif = LOOPBACK_IFINDEX;
264 fl4.flowi4_tos = RT_TOS(ip4h->tos);
265 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF;
266 fl4.flowi4_proto = ip4h->protocol;
267 fl4.daddr = ip4h->daddr;
268 fl4.saddr = ip4h->saddr;
270 rt = ip_route_output_flow(net, &fl4, NULL);
276 /* if dst.dev is loopback or the VRF device again this is locally
277 * originated traffic destined to a local address. Short circuit
280 if (rt->dst.dev == vrf_dev)
281 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
283 skb_dst_set(skb, &rt->dst);
285 /* strip the ethernet header added for pass through VRF device */
286 __skb_pull(skb, skb_network_offset(skb));
289 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
293 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
294 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
295 if (unlikely(net_xmit_eval(ret)))
296 vrf_dev->stats.tx_errors++;
298 ret = NET_XMIT_SUCCESS;
303 vrf_tx_error(vrf_dev, skb);
307 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
309 switch (skb->protocol) {
310 case htons(ETH_P_IP):
311 return vrf_process_v4_outbound(skb, dev);
312 case htons(ETH_P_IPV6):
313 return vrf_process_v6_outbound(skb, dev);
315 vrf_tx_error(dev, skb);
316 return NET_XMIT_DROP;
320 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
323 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
325 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
326 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
328 u64_stats_update_begin(&dstats->syncp);
330 dstats->tx_bytes += len;
331 u64_stats_update_end(&dstats->syncp);
333 this_cpu_inc(dev->dstats->tx_drps);
339 static void vrf_finish_direct(struct sk_buff *skb)
341 struct net_device *vrf_dev = skb->dev;
343 if (!list_empty(&vrf_dev->ptype_all) &&
344 likely(skb_headroom(skb) >= ETH_HLEN)) {
345 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
347 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
348 eth_zero_addr(eth->h_dest);
349 eth->h_proto = skb->protocol;
352 dev_queue_xmit_nit(skb, vrf_dev);
353 rcu_read_unlock_bh();
355 skb_pull(skb, ETH_HLEN);
358 /* reset skb device */
362 #if IS_ENABLED(CONFIG_IPV6)
363 /* modelled after ip6_finish_output2 */
364 static int vrf_finish_output6(struct net *net, struct sock *sk,
367 struct dst_entry *dst = skb_dst(skb);
368 struct net_device *dev = dst->dev;
369 struct neighbour *neigh;
370 struct in6_addr *nexthop;
375 skb->protocol = htons(ETH_P_IPV6);
379 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
380 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
381 if (unlikely(!neigh))
382 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
383 if (!IS_ERR(neigh)) {
384 sock_confirm_neigh(skb, neigh);
385 ret = neigh_output(neigh, skb);
386 rcu_read_unlock_bh();
389 rcu_read_unlock_bh();
391 IP6_INC_STATS(dev_net(dst->dev),
392 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
397 /* modelled after ip6_output */
398 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
400 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
401 net, sk, skb, NULL, skb_dst(skb)->dev,
403 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
406 /* set dst on skb to send packet to us via dev_xmit path. Allows
407 * packet to go through device based features such as qdisc, netfilter
408 * hooks and packet sockets with skb->dev set to vrf device.
410 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
413 struct net_vrf *vrf = netdev_priv(vrf_dev);
414 struct dst_entry *dst = NULL;
415 struct rt6_info *rt6;
419 rt6 = rcu_dereference(vrf->rt6);
427 if (unlikely(!dst)) {
428 vrf_tx_error(vrf_dev, skb);
433 skb_dst_set(skb, dst);
438 static int vrf_output6_direct_finish(struct net *net, struct sock *sk,
441 vrf_finish_direct(skb);
443 return vrf_ip6_local_out(net, sk, skb);
446 static int vrf_output6_direct(struct net *net, struct sock *sk,
451 skb->protocol = htons(ETH_P_IPV6);
453 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
454 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
455 NULL, skb->dev, vrf_output6_direct_finish);
457 if (likely(err == 1))
458 vrf_finish_direct(skb);
463 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk,
468 err = vrf_output6_direct(net, sk, skb);
469 if (likely(err == 1))
470 err = vrf_ip6_local_out(net, sk, skb);
475 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
479 struct net *net = dev_net(vrf_dev);
484 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
485 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish);
487 if (likely(err == 1))
488 err = vrf_output6_direct(net, sk, skb);
490 if (likely(err == 1))
496 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
500 /* don't divert link scope packets */
501 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
504 if (qdisc_tx_is_default(vrf_dev) ||
505 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
506 return vrf_ip6_out_direct(vrf_dev, sk, skb);
508 return vrf_ip6_out_redirect(vrf_dev, skb);
512 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
514 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
515 struct net *net = dev_net(dev);
516 struct dst_entry *dst;
518 RCU_INIT_POINTER(vrf->rt6, NULL);
521 /* move dev in dst's to loopback so this VRF device can be deleted
522 * - based on dst_ifdown
527 dst->dev = net->loopback_dev;
533 static int vrf_rt6_create(struct net_device *dev)
535 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
536 struct net_vrf *vrf = netdev_priv(dev);
537 struct net *net = dev_net(dev);
538 struct fib6_table *rt6i_table;
539 struct rt6_info *rt6;
542 /* IPv6 can be CONFIG enabled and then disabled runtime */
543 if (!ipv6_mod_enabled())
546 rt6i_table = fib6_new_table(net, vrf->tb_id);
550 /* create a dst for routing packets out a VRF device */
551 rt6 = ip6_dst_alloc(net, dev, flags);
555 rt6->rt6i_table = rt6i_table;
556 rt6->dst.output = vrf_output6;
558 rcu_assign_pointer(vrf->rt6, rt6);
565 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
572 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
576 static int vrf_rt6_create(struct net_device *dev)
582 /* modelled after ip_finish_output2 */
583 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
585 struct dst_entry *dst = skb_dst(skb);
586 struct rtable *rt = (struct rtable *)dst;
587 struct net_device *dev = dst->dev;
588 unsigned int hh_len = LL_RESERVED_SPACE(dev);
589 struct neighbour *neigh;
595 /* Be paranoid, rather than too clever. */
596 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
597 struct sk_buff *skb2;
599 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
605 skb_set_owner_w(skb2, skb->sk);
613 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
614 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
615 if (unlikely(!neigh))
616 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
617 if (!IS_ERR(neigh)) {
618 sock_confirm_neigh(skb, neigh);
619 ret = neigh_output(neigh, skb);
620 rcu_read_unlock_bh();
624 rcu_read_unlock_bh();
626 vrf_tx_error(skb->dev, skb);
630 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
632 struct net_device *dev = skb_dst(skb)->dev;
634 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
637 skb->protocol = htons(ETH_P_IP);
639 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
640 net, sk, skb, NULL, dev,
642 !(IPCB(skb)->flags & IPSKB_REROUTED));
645 /* set dst on skb to send packet to us via dev_xmit path. Allows
646 * packet to go through device based features such as qdisc, netfilter
647 * hooks and packet sockets with skb->dev set to vrf device.
649 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
652 struct net_vrf *vrf = netdev_priv(vrf_dev);
653 struct dst_entry *dst = NULL;
658 rth = rcu_dereference(vrf->rth);
666 if (unlikely(!dst)) {
667 vrf_tx_error(vrf_dev, skb);
672 skb_dst_set(skb, dst);
677 static int vrf_output_direct_finish(struct net *net, struct sock *sk,
680 vrf_finish_direct(skb);
682 return vrf_ip_local_out(net, sk, skb);
685 static int vrf_output_direct(struct net *net, struct sock *sk,
690 skb->protocol = htons(ETH_P_IP);
692 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
693 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
694 NULL, skb->dev, vrf_output_direct_finish);
696 if (likely(err == 1))
697 vrf_finish_direct(skb);
702 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk,
707 err = vrf_output_direct(net, sk, skb);
708 if (likely(err == 1))
709 err = vrf_ip_local_out(net, sk, skb);
714 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
718 struct net *net = dev_net(vrf_dev);
723 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
724 skb, NULL, vrf_dev, vrf_ip_out_direct_finish);
726 if (likely(err == 1))
727 err = vrf_output_direct(net, sk, skb);
729 if (likely(err == 1))
735 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
739 /* don't divert multicast or local broadcast */
740 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
741 ipv4_is_lbcast(ip_hdr(skb)->daddr))
744 if (qdisc_tx_is_default(vrf_dev) ||
745 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
746 return vrf_ip_out_direct(vrf_dev, sk, skb);
748 return vrf_ip_out_redirect(vrf_dev, skb);
751 /* called with rcu lock held */
752 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
759 return vrf_ip_out(vrf_dev, sk, skb);
761 return vrf_ip6_out(vrf_dev, sk, skb);
768 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
770 struct rtable *rth = rtnl_dereference(vrf->rth);
771 struct net *net = dev_net(dev);
772 struct dst_entry *dst;
774 RCU_INIT_POINTER(vrf->rth, NULL);
777 /* move dev in dst's to loopback so this VRF device can be deleted
778 * - based on dst_ifdown
783 dst->dev = net->loopback_dev;
789 static int vrf_rtable_create(struct net_device *dev)
791 struct net_vrf *vrf = netdev_priv(dev);
794 if (!fib_new_table(dev_net(dev), vrf->tb_id))
797 /* create a dst for routing packets out through a VRF device */
798 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
802 rth->dst.output = vrf_output;
803 rth->rt_table_id = vrf->tb_id;
805 rcu_assign_pointer(vrf->rth, rth);
810 /**************************** device handling ********************/
812 /* cycle interface to flush neighbor cache and move routes across tables */
813 static void cycle_netdev(struct net_device *dev)
815 unsigned int flags = dev->flags;
818 if (!netif_running(dev))
821 ret = dev_change_flags(dev, flags & ~IFF_UP);
823 ret = dev_change_flags(dev, flags);
827 "Failed to cycle device %s; route tables might be wrong!\n",
832 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
836 /* do not allow loopback device to be enslaved to a VRF.
837 * The vrf device acts as the loopback for the vrf.
839 if (port_dev == dev_net(dev)->loopback_dev)
842 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
843 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
847 cycle_netdev(port_dev);
852 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
856 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
858 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
861 return do_vrf_add_slave(dev, port_dev);
864 /* inverse of do_vrf_add_slave */
865 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
867 netdev_upper_dev_unlink(port_dev, dev);
868 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
870 cycle_netdev(port_dev);
875 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
877 return do_vrf_del_slave(dev, port_dev);
880 static void vrf_dev_uninit(struct net_device *dev)
882 struct net_vrf *vrf = netdev_priv(dev);
884 vrf_rtable_release(dev, vrf);
885 vrf_rt6_release(dev, vrf);
887 free_percpu(dev->dstats);
891 static int vrf_dev_init(struct net_device *dev)
893 struct net_vrf *vrf = netdev_priv(dev);
895 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
899 /* create the default dst which points back to us */
900 if (vrf_rtable_create(dev) != 0)
903 if (vrf_rt6_create(dev) != 0)
906 dev->flags = IFF_MASTER | IFF_NOARP;
908 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
909 dev->mtu = 64 * 1024;
911 /* similarly, oper state is irrelevant; set to up to avoid confusion */
912 dev->operstate = IF_OPER_UP;
913 netdev_lockdep_set_classes(dev);
917 vrf_rtable_release(dev, vrf);
919 free_percpu(dev->dstats);
925 static const struct net_device_ops vrf_netdev_ops = {
926 .ndo_init = vrf_dev_init,
927 .ndo_uninit = vrf_dev_uninit,
928 .ndo_start_xmit = vrf_xmit,
929 .ndo_get_stats64 = vrf_get_stats64,
930 .ndo_add_slave = vrf_add_slave,
931 .ndo_del_slave = vrf_del_slave,
934 static u32 vrf_fib_table(const struct net_device *dev)
936 struct net_vrf *vrf = netdev_priv(dev);
941 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
947 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
949 struct net_device *dev)
951 struct net *net = dev_net(dev);
953 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
954 skb = NULL; /* kfree_skb(skb) handled by nf code */
959 #if IS_ENABLED(CONFIG_IPV6)
960 /* neighbor handling is done with actual device; do not want
961 * to flip skb->dev for those ndisc packets. This really fails
962 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
965 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
967 const struct ipv6hdr *iph = ipv6_hdr(skb);
970 if (iph->nexthdr == NEXTHDR_ICMP) {
971 const struct icmp6hdr *icmph;
972 struct icmp6hdr _icmph;
974 icmph = skb_header_pointer(skb, sizeof(*iph),
975 sizeof(_icmph), &_icmph);
979 switch (icmph->icmp6_type) {
980 case NDISC_ROUTER_SOLICITATION:
981 case NDISC_ROUTER_ADVERTISEMENT:
982 case NDISC_NEIGHBOUR_SOLICITATION:
983 case NDISC_NEIGHBOUR_ADVERTISEMENT:
994 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
995 const struct net_device *dev,
1000 struct net_vrf *vrf = netdev_priv(dev);
1001 struct fib6_table *table = NULL;
1002 struct rt6_info *rt6;
1006 /* fib6_table does not have a refcnt and can not be freed */
1007 rt6 = rcu_dereference(vrf->rt6);
1009 table = rt6->rt6i_table;
1016 return ip6_pol_route(net, table, ifindex, fl6, flags);
1019 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1022 const struct ipv6hdr *iph = ipv6_hdr(skb);
1023 struct flowi6 fl6 = {
1024 .flowi6_iif = ifindex,
1025 .flowi6_mark = skb->mark,
1026 .flowi6_proto = iph->nexthdr,
1027 .daddr = iph->daddr,
1028 .saddr = iph->saddr,
1029 .flowlabel = ip6_flowinfo(iph),
1031 struct net *net = dev_net(vrf_dev);
1032 struct rt6_info *rt6;
1034 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
1035 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1039 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1042 skb_dst_set(skb, &rt6->dst);
1045 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1046 struct sk_buff *skb)
1048 int orig_iif = skb->skb_iif;
1051 /* loopback traffic; do not push through packet taps again.
1052 * Reset pkt_type for upper layers to process skb
1054 if (skb->pkt_type == PACKET_LOOPBACK) {
1056 skb->skb_iif = vrf_dev->ifindex;
1057 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1058 skb->pkt_type = PACKET_HOST;
1062 /* if packet is NDISC or addressed to multicast or link-local
1063 * then keep the ingress interface
1065 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1066 if (!ipv6_ndisc_frame(skb) && !need_strict) {
1067 vrf_rx_stats(vrf_dev, skb->len);
1069 skb->skb_iif = vrf_dev->ifindex;
1071 if (!list_empty(&vrf_dev->ptype_all)) {
1072 skb_push(skb, skb->mac_len);
1073 dev_queue_xmit_nit(skb, vrf_dev);
1074 skb_pull(skb, skb->mac_len);
1077 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1081 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1083 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1089 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1090 struct sk_buff *skb)
1096 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1097 struct sk_buff *skb)
1100 skb->skb_iif = vrf_dev->ifindex;
1101 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1103 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1106 /* loopback traffic; do not push through packet taps again.
1107 * Reset pkt_type for upper layers to process skb
1109 if (skb->pkt_type == PACKET_LOOPBACK) {
1110 skb->pkt_type = PACKET_HOST;
1114 vrf_rx_stats(vrf_dev, skb->len);
1116 if (!list_empty(&vrf_dev->ptype_all)) {
1117 skb_push(skb, skb->mac_len);
1118 dev_queue_xmit_nit(skb, vrf_dev);
1119 skb_pull(skb, skb->mac_len);
1122 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1127 /* called with rcu lock held */
1128 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1129 struct sk_buff *skb,
1134 return vrf_ip_rcv(vrf_dev, skb);
1136 return vrf_ip6_rcv(vrf_dev, skb);
1142 #if IS_ENABLED(CONFIG_IPV6)
1143 /* send to link-local or multicast address via interface enslaved to
1144 * VRF device. Force lookup to VRF table without changing flow struct
1146 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1149 struct net *net = dev_net(dev);
1150 int flags = RT6_LOOKUP_F_IFACE;
1151 struct dst_entry *dst = NULL;
1152 struct rt6_info *rt;
1154 /* VRF device does not have a link-local address and
1155 * sending packets to link-local or mcast addresses over
1156 * a VRF device does not make sense
1158 if (fl6->flowi6_oif == dev->ifindex) {
1159 dst = &net->ipv6.ip6_null_entry->dst;
1164 if (!ipv6_addr_any(&fl6->saddr))
1165 flags |= RT6_LOOKUP_F_HAS_SADDR;
1167 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1175 static const struct l3mdev_ops vrf_l3mdev_ops = {
1176 .l3mdev_fib_table = vrf_fib_table,
1177 .l3mdev_l3_rcv = vrf_l3_rcv,
1178 .l3mdev_l3_out = vrf_l3_out,
1179 #if IS_ENABLED(CONFIG_IPV6)
1180 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1184 static void vrf_get_drvinfo(struct net_device *dev,
1185 struct ethtool_drvinfo *info)
1187 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1188 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1191 static const struct ethtool_ops vrf_ethtool_ops = {
1192 .get_drvinfo = vrf_get_drvinfo,
1195 static inline size_t vrf_fib_rule_nl_size(void)
1199 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1200 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1201 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1206 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1208 struct fib_rule_hdr *frh;
1209 struct nlmsghdr *nlh;
1210 struct sk_buff *skb;
1213 if (family == AF_INET6 && !ipv6_mod_enabled())
1216 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1220 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1222 goto nla_put_failure;
1224 /* rule only needs to appear once */
1225 nlh->nlmsg_flags |= NLM_F_EXCL;
1227 frh = nlmsg_data(nlh);
1228 memset(frh, 0, sizeof(*frh));
1229 frh->family = family;
1230 frh->action = FR_ACT_TO_TBL;
1232 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1233 goto nla_put_failure;
1235 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1236 goto nla_put_failure;
1238 nlmsg_end(skb, nlh);
1240 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1241 skb->sk = dev_net(dev)->rtnl;
1243 err = fib_nl_newrule(skb, nlh, NULL);
1247 err = fib_nl_delrule(skb, nlh, NULL);
1261 static int vrf_add_fib_rules(const struct net_device *dev)
1265 err = vrf_fib_rule(dev, AF_INET, true);
1269 err = vrf_fib_rule(dev, AF_INET6, true);
1273 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1274 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1281 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1283 vrf_fib_rule(dev, AF_INET6, false);
1287 vrf_fib_rule(dev, AF_INET, false);
1290 netdev_err(dev, "Failed to add FIB rules.\n");
1294 static void vrf_setup(struct net_device *dev)
1298 /* Initialize the device structure. */
1299 dev->netdev_ops = &vrf_netdev_ops;
1300 dev->l3mdev_ops = &vrf_l3mdev_ops;
1301 dev->ethtool_ops = &vrf_ethtool_ops;
1302 dev->needs_free_netdev = true;
1304 /* Fill in device structure with ethernet-generic values. */
1305 eth_hw_addr_random(dev);
1307 /* don't acquire vrf device's netif_tx_lock when transmitting */
1308 dev->features |= NETIF_F_LLTX;
1310 /* don't allow vrf devices to change network namespaces. */
1311 dev->features |= NETIF_F_NETNS_LOCAL;
1313 /* does not make sense for a VLAN to be added to a vrf device */
1314 dev->features |= NETIF_F_VLAN_CHALLENGED;
1316 /* enable offload features */
1317 dev->features |= NETIF_F_GSO_SOFTWARE;
1318 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1319 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1321 dev->hw_features = dev->features;
1322 dev->hw_enc_features = dev->features;
1324 /* default to no qdisc; user can add if desired */
1325 dev->priv_flags |= IFF_NO_QUEUE;
1328 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1329 struct netlink_ext_ack *extack)
1331 if (tb[IFLA_ADDRESS]) {
1332 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1333 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1336 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1337 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1338 return -EADDRNOTAVAIL;
1344 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1346 struct net_device *port_dev;
1347 struct list_head *iter;
1349 netdev_for_each_lower_dev(dev, port_dev, iter)
1350 vrf_del_slave(dev, port_dev);
1352 unregister_netdevice_queue(dev, head);
1355 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1356 struct nlattr *tb[], struct nlattr *data[],
1357 struct netlink_ext_ack *extack)
1359 struct net_vrf *vrf = netdev_priv(dev);
1360 bool *add_fib_rules;
1364 if (!data || !data[IFLA_VRF_TABLE]) {
1365 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1369 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1370 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1371 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1372 "Invalid VRF table id");
1376 dev->priv_flags |= IFF_L3MDEV_MASTER;
1378 err = register_netdevice(dev);
1383 add_fib_rules = net_generic(net, vrf_net_id);
1384 if (*add_fib_rules) {
1385 err = vrf_add_fib_rules(dev);
1387 unregister_netdevice(dev);
1390 *add_fib_rules = false;
1397 static size_t vrf_nl_getsize(const struct net_device *dev)
1399 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1402 static int vrf_fillinfo(struct sk_buff *skb,
1403 const struct net_device *dev)
1405 struct net_vrf *vrf = netdev_priv(dev);
1407 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1410 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1411 const struct net_device *slave_dev)
1413 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1416 static int vrf_fill_slave_info(struct sk_buff *skb,
1417 const struct net_device *vrf_dev,
1418 const struct net_device *slave_dev)
1420 struct net_vrf *vrf = netdev_priv(vrf_dev);
1422 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1428 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1429 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1432 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1434 .priv_size = sizeof(struct net_vrf),
1436 .get_size = vrf_nl_getsize,
1437 .policy = vrf_nl_policy,
1438 .validate = vrf_validate,
1439 .fill_info = vrf_fillinfo,
1441 .get_slave_size = vrf_get_slave_size,
1442 .fill_slave_info = vrf_fill_slave_info,
1444 .newlink = vrf_newlink,
1445 .dellink = vrf_dellink,
1447 .maxtype = IFLA_VRF_MAX,
1450 static int vrf_device_event(struct notifier_block *unused,
1451 unsigned long event, void *ptr)
1453 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1455 /* only care about unregister events to drop slave references */
1456 if (event == NETDEV_UNREGISTER) {
1457 struct net_device *vrf_dev;
1459 if (!netif_is_l3_slave(dev))
1462 vrf_dev = netdev_master_upper_dev_get(dev);
1463 vrf_del_slave(vrf_dev, dev);
1469 static struct notifier_block vrf_notifier_block __read_mostly = {
1470 .notifier_call = vrf_device_event,
1473 /* Initialize per network namespace state */
1474 static int __net_init vrf_netns_init(struct net *net)
1476 bool *add_fib_rules = net_generic(net, vrf_net_id);
1478 *add_fib_rules = true;
1483 static struct pernet_operations vrf_net_ops __net_initdata = {
1484 .init = vrf_netns_init,
1486 .size = sizeof(bool),
1489 static int __init vrf_init_module(void)
1493 register_netdevice_notifier(&vrf_notifier_block);
1495 rc = register_pernet_subsys(&vrf_net_ops);
1499 rc = rtnl_link_register(&vrf_link_ops);
1501 unregister_pernet_subsys(&vrf_net_ops);
1508 unregister_netdevice_notifier(&vrf_notifier_block);
1512 module_init(vrf_init_module);
1513 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1514 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1515 MODULE_LICENSE("GPL");
1516 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1517 MODULE_VERSION(DRV_VERSION);