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 rtable __rcu
*rth_local
;
51 struct rt6_info __rcu
*rt6
;
52 struct rt6_info __rcu
*rt6_local
;
63 struct u64_stats_sync syncp
;
66 static void vrf_rx_stats(struct net_device
*dev
, int len
)
68 struct pcpu_dstats
*dstats
= this_cpu_ptr(dev
->dstats
);
70 u64_stats_update_begin(&dstats
->syncp
);
72 dstats
->rx_bytes
+= len
;
73 u64_stats_update_end(&dstats
->syncp
);
76 static void vrf_tx_error(struct net_device
*vrf_dev
, struct sk_buff
*skb
)
78 vrf_dev
->stats
.tx_errors
++;
82 static void vrf_get_stats64(struct net_device
*dev
,
83 struct rtnl_link_stats64
*stats
)
87 for_each_possible_cpu(i
) {
88 const struct pcpu_dstats
*dstats
;
89 u64 tbytes
, tpkts
, tdrops
, rbytes
, rpkts
;
92 dstats
= per_cpu_ptr(dev
->dstats
, i
);
94 start
= u64_stats_fetch_begin_irq(&dstats
->syncp
);
95 tbytes
= dstats
->tx_bytes
;
96 tpkts
= dstats
->tx_pkts
;
97 tdrops
= dstats
->tx_drps
;
98 rbytes
= dstats
->rx_bytes
;
99 rpkts
= dstats
->rx_pkts
;
100 } while (u64_stats_fetch_retry_irq(&dstats
->syncp
, start
));
101 stats
->tx_bytes
+= tbytes
;
102 stats
->tx_packets
+= tpkts
;
103 stats
->tx_dropped
+= tdrops
;
104 stats
->rx_bytes
+= rbytes
;
105 stats
->rx_packets
+= rpkts
;
109 /* by default VRF devices do not have a qdisc and are expected
110 * to be created with only a single queue.
112 static bool qdisc_tx_is_default(const struct net_device
*dev
)
114 struct netdev_queue
*txq
;
117 if (dev
->num_tx_queues
> 1)
120 txq
= netdev_get_tx_queue(dev
, 0);
121 qdisc
= rcu_access_pointer(txq
->qdisc
);
123 return !qdisc
->enqueue
;
126 /* Local traffic destined to local address. Reinsert the packet to rx
127 * path, similar to loopback handling.
129 static int vrf_local_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
130 struct dst_entry
*dst
)
136 skb_dst_set(skb
, dst
);
139 /* set pkt_type to avoid skb hitting packet taps twice -
140 * once on Tx and again in Rx processing
142 skb
->pkt_type
= PACKET_LOOPBACK
;
144 skb
->protocol
= eth_type_trans(skb
, dev
);
146 if (likely(netif_rx(skb
) == NET_RX_SUCCESS
))
147 vrf_rx_stats(dev
, len
);
149 this_cpu_inc(dev
->dstats
->rx_drps
);
154 #if IS_ENABLED(CONFIG_IPV6)
155 static int vrf_ip6_local_out(struct net
*net
, struct sock
*sk
,
160 err
= nf_hook(NFPROTO_IPV6
, NF_INET_LOCAL_OUT
, net
,
161 sk
, skb
, NULL
, skb_dst(skb
)->dev
, dst_output
);
163 if (likely(err
== 1))
164 err
= dst_output(net
, sk
, skb
);
169 static netdev_tx_t
vrf_process_v6_outbound(struct sk_buff
*skb
,
170 struct net_device
*dev
)
172 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
173 struct net
*net
= dev_net(skb
->dev
);
174 struct flowi6 fl6
= {
175 /* needed to match OIF rule */
176 .flowi6_oif
= dev
->ifindex
,
177 .flowi6_iif
= LOOPBACK_IFINDEX
,
180 .flowlabel
= ip6_flowinfo(iph
),
181 .flowi6_mark
= skb
->mark
,
182 .flowi6_proto
= iph
->nexthdr
,
183 .flowi6_flags
= FLOWI_FLAG_SKIP_NH_OIF
,
185 int ret
= NET_XMIT_DROP
;
186 struct dst_entry
*dst
;
187 struct dst_entry
*dst_null
= &net
->ipv6
.ip6_null_entry
->dst
;
189 dst
= ip6_route_output(net
, NULL
, &fl6
);
195 /* if dst.dev is loopback or the VRF device again this is locally
196 * originated traffic destined to a local address. Short circuit
197 * to Rx path using our local dst
199 if (dst
->dev
== net
->loopback_dev
|| dst
->dev
== dev
) {
200 struct net_vrf
*vrf
= netdev_priv(dev
);
201 struct rt6_info
*rt6_local
;
203 /* release looked up dst and use cached local dst */
208 rt6_local
= rcu_dereference(vrf
->rt6_local
);
209 if (unlikely(!rt6_local
)) {
214 /* Ordering issue: cached local dst is created on newlink
215 * before the IPv6 initialization. Using the local dst
216 * requires rt6i_idev to be set so make sure it is.
218 if (unlikely(!rt6_local
->rt6i_idev
)) {
219 rt6_local
->rt6i_idev
= in6_dev_get(dev
);
220 if (!rt6_local
->rt6i_idev
) {
226 dst
= &rt6_local
->dst
;
231 return vrf_local_xmit(skb
, dev
, &rt6_local
->dst
);
234 skb_dst_set(skb
, dst
);
236 /* strip the ethernet header added for pass through VRF device */
237 __skb_pull(skb
, skb_network_offset(skb
));
239 ret
= vrf_ip6_local_out(net
, skb
->sk
, skb
);
240 if (unlikely(net_xmit_eval(ret
)))
241 dev
->stats
.tx_errors
++;
243 ret
= NET_XMIT_SUCCESS
;
247 vrf_tx_error(dev
, skb
);
248 return NET_XMIT_DROP
;
251 static netdev_tx_t
vrf_process_v6_outbound(struct sk_buff
*skb
,
252 struct net_device
*dev
)
254 vrf_tx_error(dev
, skb
);
255 return NET_XMIT_DROP
;
259 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
260 static int vrf_ip_local_out(struct net
*net
, struct sock
*sk
,
265 err
= nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
, net
, sk
,
266 skb
, NULL
, skb_dst(skb
)->dev
, dst_output
);
267 if (likely(err
== 1))
268 err
= dst_output(net
, sk
, skb
);
273 static netdev_tx_t
vrf_process_v4_outbound(struct sk_buff
*skb
,
274 struct net_device
*vrf_dev
)
276 struct iphdr
*ip4h
= ip_hdr(skb
);
277 int ret
= NET_XMIT_DROP
;
278 struct flowi4 fl4
= {
279 /* needed to match OIF rule */
280 .flowi4_oif
= vrf_dev
->ifindex
,
281 .flowi4_iif
= LOOPBACK_IFINDEX
,
282 .flowi4_tos
= RT_TOS(ip4h
->tos
),
283 .flowi4_flags
= FLOWI_FLAG_ANYSRC
| FLOWI_FLAG_SKIP_NH_OIF
,
284 .flowi4_proto
= ip4h
->protocol
,
285 .daddr
= ip4h
->daddr
,
286 .saddr
= ip4h
->saddr
,
288 struct net
*net
= dev_net(vrf_dev
);
291 rt
= ip_route_output_flow(net
, &fl4
, NULL
);
297 /* if dst.dev is loopback or the VRF device again this is locally
298 * originated traffic destined to a local address. Short circuit
299 * to Rx path using our local dst
301 if (rt
->dst
.dev
== net
->loopback_dev
|| rt
->dst
.dev
== vrf_dev
) {
302 struct net_vrf
*vrf
= netdev_priv(vrf_dev
);
303 struct rtable
*rth_local
;
304 struct dst_entry
*dst
= NULL
;
310 rth_local
= rcu_dereference(vrf
->rth_local
);
311 if (likely(rth_local
)) {
312 dst
= &rth_local
->dst
;
321 return vrf_local_xmit(skb
, vrf_dev
, dst
);
324 skb_dst_set(skb
, &rt
->dst
);
326 /* strip the ethernet header added for pass through VRF device */
327 __skb_pull(skb
, skb_network_offset(skb
));
330 ip4h
->saddr
= inet_select_addr(skb_dst(skb
)->dev
, 0,
334 ret
= vrf_ip_local_out(dev_net(skb_dst(skb
)->dev
), skb
->sk
, skb
);
335 if (unlikely(net_xmit_eval(ret
)))
336 vrf_dev
->stats
.tx_errors
++;
338 ret
= NET_XMIT_SUCCESS
;
343 vrf_tx_error(vrf_dev
, skb
);
347 static netdev_tx_t
is_ip_tx_frame(struct sk_buff
*skb
, struct net_device
*dev
)
349 switch (skb
->protocol
) {
350 case htons(ETH_P_IP
):
351 return vrf_process_v4_outbound(skb
, dev
);
352 case htons(ETH_P_IPV6
):
353 return vrf_process_v6_outbound(skb
, dev
);
355 vrf_tx_error(dev
, skb
);
356 return NET_XMIT_DROP
;
360 static netdev_tx_t
vrf_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
363 netdev_tx_t ret
= is_ip_tx_frame(skb
, dev
);
365 if (likely(ret
== NET_XMIT_SUCCESS
|| ret
== NET_XMIT_CN
)) {
366 struct pcpu_dstats
*dstats
= this_cpu_ptr(dev
->dstats
);
368 u64_stats_update_begin(&dstats
->syncp
);
370 dstats
->tx_bytes
+= len
;
371 u64_stats_update_end(&dstats
->syncp
);
373 this_cpu_inc(dev
->dstats
->tx_drps
);
379 static int vrf_finish_direct(struct net
*net
, struct sock
*sk
,
382 struct net_device
*vrf_dev
= skb
->dev
;
384 if (!list_empty(&vrf_dev
->ptype_all
) &&
385 likely(skb_headroom(skb
) >= ETH_HLEN
)) {
386 struct ethhdr
*eth
= skb_push(skb
, ETH_HLEN
);
388 ether_addr_copy(eth
->h_source
, vrf_dev
->dev_addr
);
389 eth_zero_addr(eth
->h_dest
);
390 eth
->h_proto
= skb
->protocol
;
393 dev_queue_xmit_nit(skb
, vrf_dev
);
394 rcu_read_unlock_bh();
396 skb_pull(skb
, ETH_HLEN
);
402 #if IS_ENABLED(CONFIG_IPV6)
403 /* modelled after ip6_finish_output2 */
404 static int vrf_finish_output6(struct net
*net
, struct sock
*sk
,
407 struct dst_entry
*dst
= skb_dst(skb
);
408 struct net_device
*dev
= dst
->dev
;
409 struct neighbour
*neigh
;
410 struct in6_addr
*nexthop
;
415 skb
->protocol
= htons(ETH_P_IPV6
);
419 nexthop
= rt6_nexthop((struct rt6_info
*)dst
, &ipv6_hdr(skb
)->daddr
);
420 neigh
= __ipv6_neigh_lookup_noref(dst
->dev
, nexthop
);
421 if (unlikely(!neigh
))
422 neigh
= __neigh_create(&nd_tbl
, nexthop
, dst
->dev
, false);
423 if (!IS_ERR(neigh
)) {
424 sock_confirm_neigh(skb
, neigh
);
425 ret
= neigh_output(neigh
, skb
);
426 rcu_read_unlock_bh();
429 rcu_read_unlock_bh();
431 IP6_INC_STATS(dev_net(dst
->dev
),
432 ip6_dst_idev(dst
), IPSTATS_MIB_OUTNOROUTES
);
437 /* modelled after ip6_output */
438 static int vrf_output6(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
440 return NF_HOOK_COND(NFPROTO_IPV6
, NF_INET_POST_ROUTING
,
441 net
, sk
, skb
, NULL
, skb_dst(skb
)->dev
,
443 !(IP6CB(skb
)->flags
& IP6SKB_REROUTED
));
446 /* set dst on skb to send packet to us via dev_xmit path. Allows
447 * packet to go through device based features such as qdisc, netfilter
448 * hooks and packet sockets with skb->dev set to vrf device.
450 static struct sk_buff
*vrf_ip6_out_redirect(struct net_device
*vrf_dev
,
453 struct net_vrf
*vrf
= netdev_priv(vrf_dev
);
454 struct dst_entry
*dst
= NULL
;
455 struct rt6_info
*rt6
;
459 rt6
= rcu_dereference(vrf
->rt6
);
467 if (unlikely(!dst
)) {
468 vrf_tx_error(vrf_dev
, skb
);
473 skb_dst_set(skb
, dst
);
478 static int vrf_output6_direct(struct net
*net
, struct sock
*sk
,
481 skb
->protocol
= htons(ETH_P_IPV6
);
483 return NF_HOOK_COND(NFPROTO_IPV6
, NF_INET_POST_ROUTING
,
484 net
, sk
, skb
, NULL
, skb
->dev
,
486 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
489 static struct sk_buff
*vrf_ip6_out_direct(struct net_device
*vrf_dev
,
493 struct net
*net
= dev_net(vrf_dev
);
498 err
= nf_hook(NFPROTO_IPV6
, NF_INET_LOCAL_OUT
, net
, sk
,
499 skb
, NULL
, vrf_dev
, vrf_output6_direct
);
501 if (likely(err
== 1))
502 err
= vrf_output6_direct(net
, sk
, skb
);
504 /* reset skb device */
505 if (likely(err
== 1))
513 static struct sk_buff
*vrf_ip6_out(struct net_device
*vrf_dev
,
517 /* don't divert link scope packets */
518 if (rt6_need_strict(&ipv6_hdr(skb
)->daddr
))
521 if (qdisc_tx_is_default(vrf_dev
))
522 return vrf_ip6_out_direct(vrf_dev
, sk
, skb
);
524 return vrf_ip6_out_redirect(vrf_dev
, skb
);
528 static void vrf_rt6_release(struct net_device
*dev
, struct net_vrf
*vrf
)
530 struct rt6_info
*rt6
= rtnl_dereference(vrf
->rt6
);
531 struct rt6_info
*rt6_local
= rtnl_dereference(vrf
->rt6_local
);
532 struct net
*net
= dev_net(dev
);
533 struct dst_entry
*dst
;
535 RCU_INIT_POINTER(vrf
->rt6
, NULL
);
536 RCU_INIT_POINTER(vrf
->rt6_local
, NULL
);
539 /* move dev in dst's to loopback so this VRF device can be deleted
540 * - based on dst_ifdown
545 dst
->dev
= net
->loopback_dev
;
551 if (rt6_local
->rt6i_idev
) {
552 in6_dev_put(rt6_local
->rt6i_idev
);
553 rt6_local
->rt6i_idev
= NULL
;
556 dst
= &rt6_local
->dst
;
558 dst
->dev
= net
->loopback_dev
;
564 static int vrf_rt6_create(struct net_device
*dev
)
566 int flags
= DST_HOST
| DST_NOPOLICY
| DST_NOXFRM
;
567 struct net_vrf
*vrf
= netdev_priv(dev
);
568 struct net
*net
= dev_net(dev
);
569 struct fib6_table
*rt6i_table
;
570 struct rt6_info
*rt6
, *rt6_local
;
573 /* IPv6 can be CONFIG enabled and then disabled runtime */
574 if (!ipv6_mod_enabled())
577 rt6i_table
= fib6_new_table(net
, vrf
->tb_id
);
581 /* create a dst for routing packets out a VRF device */
582 rt6
= ip6_dst_alloc(net
, dev
, flags
);
586 rt6
->rt6i_table
= rt6i_table
;
587 rt6
->dst
.output
= vrf_output6
;
589 /* create a dst for local routing - packets sent locally
590 * to local address via the VRF device as a loopback
592 rt6_local
= ip6_dst_alloc(net
, dev
, flags
);
594 dst_release(&rt6
->dst
);
598 rt6_local
->rt6i_idev
= in6_dev_get(dev
);
599 rt6_local
->rt6i_flags
= RTF_UP
| RTF_NONEXTHOP
| RTF_LOCAL
;
600 rt6_local
->rt6i_table
= rt6i_table
;
601 rt6_local
->dst
.input
= ip6_input
;
603 rcu_assign_pointer(vrf
->rt6
, rt6
);
604 rcu_assign_pointer(vrf
->rt6_local
, rt6_local
);
611 static struct sk_buff
*vrf_ip6_out(struct net_device
*vrf_dev
,
618 static void vrf_rt6_release(struct net_device
*dev
, struct net_vrf
*vrf
)
622 static int vrf_rt6_create(struct net_device
*dev
)
628 /* modelled after ip_finish_output2 */
629 static int vrf_finish_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
631 struct dst_entry
*dst
= skb_dst(skb
);
632 struct rtable
*rt
= (struct rtable
*)dst
;
633 struct net_device
*dev
= dst
->dev
;
634 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
635 struct neighbour
*neigh
;
641 /* Be paranoid, rather than too clever. */
642 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
643 struct sk_buff
*skb2
;
645 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
651 skb_set_owner_w(skb2
, skb
->sk
);
659 nexthop
= (__force u32
)rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
660 neigh
= __ipv4_neigh_lookup_noref(dev
, nexthop
);
661 if (unlikely(!neigh
))
662 neigh
= __neigh_create(&arp_tbl
, &nexthop
, dev
, false);
663 if (!IS_ERR(neigh
)) {
664 sock_confirm_neigh(skb
, neigh
);
665 ret
= neigh_output(neigh
, skb
);
668 rcu_read_unlock_bh();
670 if (unlikely(ret
< 0))
671 vrf_tx_error(skb
->dev
, skb
);
675 static int vrf_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
677 struct net_device
*dev
= skb_dst(skb
)->dev
;
679 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUT
, skb
->len
);
682 skb
->protocol
= htons(ETH_P_IP
);
684 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
685 net
, sk
, skb
, NULL
, dev
,
687 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
690 /* set dst on skb to send packet to us via dev_xmit path. Allows
691 * packet to go through device based features such as qdisc, netfilter
692 * hooks and packet sockets with skb->dev set to vrf device.
694 static struct sk_buff
*vrf_ip_out_redirect(struct net_device
*vrf_dev
,
697 struct net_vrf
*vrf
= netdev_priv(vrf_dev
);
698 struct dst_entry
*dst
= NULL
;
703 rth
= rcu_dereference(vrf
->rth
);
711 if (unlikely(!dst
)) {
712 vrf_tx_error(vrf_dev
, skb
);
717 skb_dst_set(skb
, dst
);
722 static int vrf_output_direct(struct net
*net
, struct sock
*sk
,
725 skb
->protocol
= htons(ETH_P_IP
);
727 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
728 net
, sk
, skb
, NULL
, skb
->dev
,
730 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
733 static struct sk_buff
*vrf_ip_out_direct(struct net_device
*vrf_dev
,
737 struct net
*net
= dev_net(vrf_dev
);
742 err
= nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
, net
, sk
,
743 skb
, NULL
, vrf_dev
, vrf_output_direct
);
745 if (likely(err
== 1))
746 err
= vrf_output_direct(net
, sk
, skb
);
748 /* reset skb device */
749 if (likely(err
== 1))
757 static struct sk_buff
*vrf_ip_out(struct net_device
*vrf_dev
,
761 /* don't divert multicast */
762 if (ipv4_is_multicast(ip_hdr(skb
)->daddr
))
765 if (qdisc_tx_is_default(vrf_dev
))
766 return vrf_ip_out_direct(vrf_dev
, sk
, skb
);
768 return vrf_ip_out_redirect(vrf_dev
, skb
);
771 /* called with rcu lock held */
772 static struct sk_buff
*vrf_l3_out(struct net_device
*vrf_dev
,
779 return vrf_ip_out(vrf_dev
, sk
, skb
);
781 return vrf_ip6_out(vrf_dev
, sk
, skb
);
788 static void vrf_rtable_release(struct net_device
*dev
, struct net_vrf
*vrf
)
790 struct rtable
*rth
= rtnl_dereference(vrf
->rth
);
791 struct rtable
*rth_local
= rtnl_dereference(vrf
->rth_local
);
792 struct net
*net
= dev_net(dev
);
793 struct dst_entry
*dst
;
795 RCU_INIT_POINTER(vrf
->rth
, NULL
);
796 RCU_INIT_POINTER(vrf
->rth_local
, NULL
);
799 /* move dev in dst's to loopback so this VRF device can be deleted
800 * - based on dst_ifdown
805 dst
->dev
= net
->loopback_dev
;
811 dst
= &rth_local
->dst
;
813 dst
->dev
= net
->loopback_dev
;
819 static int vrf_rtable_create(struct net_device
*dev
)
821 struct net_vrf
*vrf
= netdev_priv(dev
);
822 struct rtable
*rth
, *rth_local
;
824 if (!fib_new_table(dev_net(dev
), vrf
->tb_id
))
827 /* create a dst for routing packets out through a VRF device */
828 rth
= rt_dst_alloc(dev
, 0, RTN_UNICAST
, 1, 1, 0);
832 /* create a dst for local ingress routing - packets sent locally
833 * to local address via the VRF device as a loopback
835 rth_local
= rt_dst_alloc(dev
, RTCF_LOCAL
, RTN_LOCAL
, 1, 1, 0);
837 dst_release(&rth
->dst
);
841 rth
->dst
.output
= vrf_output
;
842 rth
->rt_table_id
= vrf
->tb_id
;
844 rth_local
->rt_table_id
= vrf
->tb_id
;
846 rcu_assign_pointer(vrf
->rth
, rth
);
847 rcu_assign_pointer(vrf
->rth_local
, rth_local
);
852 /**************************** device handling ********************/
854 /* cycle interface to flush neighbor cache and move routes across tables */
855 static void cycle_netdev(struct net_device
*dev
)
857 unsigned int flags
= dev
->flags
;
860 if (!netif_running(dev
))
863 ret
= dev_change_flags(dev
, flags
& ~IFF_UP
);
865 ret
= dev_change_flags(dev
, flags
);
869 "Failed to cycle device %s; route tables might be wrong!\n",
874 static int do_vrf_add_slave(struct net_device
*dev
, struct net_device
*port_dev
)
878 /* do not allow loopback device to be enslaved to a VRF.
879 * The vrf device acts as the loopback for the vrf.
881 if (port_dev
== dev_net(dev
)->loopback_dev
)
884 port_dev
->priv_flags
|= IFF_L3MDEV_SLAVE
;
885 ret
= netdev_master_upper_dev_link(port_dev
, dev
, NULL
, NULL
);
889 cycle_netdev(port_dev
);
894 port_dev
->priv_flags
&= ~IFF_L3MDEV_SLAVE
;
898 static int vrf_add_slave(struct net_device
*dev
, struct net_device
*port_dev
)
900 if (netif_is_l3_master(port_dev
) || netif_is_l3_slave(port_dev
))
903 return do_vrf_add_slave(dev
, port_dev
);
906 /* inverse of do_vrf_add_slave */
907 static int do_vrf_del_slave(struct net_device
*dev
, struct net_device
*port_dev
)
909 netdev_upper_dev_unlink(port_dev
, dev
);
910 port_dev
->priv_flags
&= ~IFF_L3MDEV_SLAVE
;
912 cycle_netdev(port_dev
);
917 static int vrf_del_slave(struct net_device
*dev
, struct net_device
*port_dev
)
919 return do_vrf_del_slave(dev
, port_dev
);
922 static void vrf_dev_uninit(struct net_device
*dev
)
924 struct net_vrf
*vrf
= netdev_priv(dev
);
925 struct net_device
*port_dev
;
926 struct list_head
*iter
;
928 vrf_rtable_release(dev
, vrf
);
929 vrf_rt6_release(dev
, vrf
);
931 netdev_for_each_lower_dev(dev
, port_dev
, iter
)
932 vrf_del_slave(dev
, port_dev
);
934 free_percpu(dev
->dstats
);
938 static int vrf_dev_init(struct net_device
*dev
)
940 struct net_vrf
*vrf
= netdev_priv(dev
);
942 dev
->dstats
= netdev_alloc_pcpu_stats(struct pcpu_dstats
);
946 /* create the default dst which points back to us */
947 if (vrf_rtable_create(dev
) != 0)
950 if (vrf_rt6_create(dev
) != 0)
953 dev
->flags
= IFF_MASTER
| IFF_NOARP
;
955 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
956 dev
->mtu
= 64 * 1024;
958 /* similarly, oper state is irrelevant; set to up to avoid confusion */
959 dev
->operstate
= IF_OPER_UP
;
960 netdev_lockdep_set_classes(dev
);
964 vrf_rtable_release(dev
, vrf
);
966 free_percpu(dev
->dstats
);
972 static const struct net_device_ops vrf_netdev_ops
= {
973 .ndo_init
= vrf_dev_init
,
974 .ndo_uninit
= vrf_dev_uninit
,
975 .ndo_start_xmit
= vrf_xmit
,
976 .ndo_get_stats64
= vrf_get_stats64
,
977 .ndo_add_slave
= vrf_add_slave
,
978 .ndo_del_slave
= vrf_del_slave
,
981 static u32
vrf_fib_table(const struct net_device
*dev
)
983 struct net_vrf
*vrf
= netdev_priv(dev
);
988 static int vrf_rcv_finish(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
994 static struct sk_buff
*vrf_rcv_nfhook(u8 pf
, unsigned int hook
,
996 struct net_device
*dev
)
998 struct net
*net
= dev_net(dev
);
1000 if (nf_hook(pf
, hook
, net
, NULL
, skb
, dev
, NULL
, vrf_rcv_finish
) != 1)
1001 skb
= NULL
; /* kfree_skb(skb) handled by nf code */
1006 #if IS_ENABLED(CONFIG_IPV6)
1007 /* neighbor handling is done with actual device; do not want
1008 * to flip skb->dev for those ndisc packets. This really fails
1009 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1012 static bool ipv6_ndisc_frame(const struct sk_buff
*skb
)
1014 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
1017 if (iph
->nexthdr
== NEXTHDR_ICMP
) {
1018 const struct icmp6hdr
*icmph
;
1019 struct icmp6hdr _icmph
;
1021 icmph
= skb_header_pointer(skb
, sizeof(*iph
),
1022 sizeof(_icmph
), &_icmph
);
1026 switch (icmph
->icmp6_type
) {
1027 case NDISC_ROUTER_SOLICITATION
:
1028 case NDISC_ROUTER_ADVERTISEMENT
:
1029 case NDISC_NEIGHBOUR_SOLICITATION
:
1030 case NDISC_NEIGHBOUR_ADVERTISEMENT
:
1031 case NDISC_REDIRECT
:
1041 static struct rt6_info
*vrf_ip6_route_lookup(struct net
*net
,
1042 const struct net_device
*dev
,
1047 struct net_vrf
*vrf
= netdev_priv(dev
);
1048 struct fib6_table
*table
= NULL
;
1049 struct rt6_info
*rt6
;
1053 /* fib6_table does not have a refcnt and can not be freed */
1054 rt6
= rcu_dereference(vrf
->rt6
);
1056 table
= rt6
->rt6i_table
;
1063 return ip6_pol_route(net
, table
, ifindex
, fl6
, flags
);
1066 static void vrf_ip6_input_dst(struct sk_buff
*skb
, struct net_device
*vrf_dev
,
1069 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
1070 struct flowi6 fl6
= {
1071 .daddr
= iph
->daddr
,
1072 .saddr
= iph
->saddr
,
1073 .flowlabel
= ip6_flowinfo(iph
),
1074 .flowi6_mark
= skb
->mark
,
1075 .flowi6_proto
= iph
->nexthdr
,
1076 .flowi6_iif
= ifindex
,
1078 struct net
*net
= dev_net(vrf_dev
);
1079 struct rt6_info
*rt6
;
1081 rt6
= vrf_ip6_route_lookup(net
, vrf_dev
, &fl6
, ifindex
,
1082 RT6_LOOKUP_F_HAS_SADDR
| RT6_LOOKUP_F_IFACE
);
1086 if (unlikely(&rt6
->dst
== &net
->ipv6
.ip6_null_entry
->dst
))
1089 skb_dst_set(skb
, &rt6
->dst
);
1092 static struct sk_buff
*vrf_ip6_rcv(struct net_device
*vrf_dev
,
1093 struct sk_buff
*skb
)
1095 int orig_iif
= skb
->skb_iif
;
1098 /* loopback traffic; do not push through packet taps again.
1099 * Reset pkt_type for upper layers to process skb
1101 if (skb
->pkt_type
== PACKET_LOOPBACK
) {
1103 skb
->skb_iif
= vrf_dev
->ifindex
;
1104 IP6CB(skb
)->flags
|= IP6SKB_L3SLAVE
;
1105 skb
->pkt_type
= PACKET_HOST
;
1109 /* if packet is NDISC or addressed to multicast or link-local
1110 * then keep the ingress interface
1112 need_strict
= rt6_need_strict(&ipv6_hdr(skb
)->daddr
);
1113 if (!ipv6_ndisc_frame(skb
) && !need_strict
) {
1114 vrf_rx_stats(vrf_dev
, skb
->len
);
1116 skb
->skb_iif
= vrf_dev
->ifindex
;
1118 if (!list_empty(&vrf_dev
->ptype_all
)) {
1119 skb_push(skb
, skb
->mac_len
);
1120 dev_queue_xmit_nit(skb
, vrf_dev
);
1121 skb_pull(skb
, skb
->mac_len
);
1124 IP6CB(skb
)->flags
|= IP6SKB_L3SLAVE
;
1128 vrf_ip6_input_dst(skb
, vrf_dev
, orig_iif
);
1130 skb
= vrf_rcv_nfhook(NFPROTO_IPV6
, NF_INET_PRE_ROUTING
, skb
, vrf_dev
);
1136 static struct sk_buff
*vrf_ip6_rcv(struct net_device
*vrf_dev
,
1137 struct sk_buff
*skb
)
1143 static struct sk_buff
*vrf_ip_rcv(struct net_device
*vrf_dev
,
1144 struct sk_buff
*skb
)
1147 skb
->skb_iif
= vrf_dev
->ifindex
;
1148 IPCB(skb
)->flags
|= IPSKB_L3SLAVE
;
1150 if (ipv4_is_multicast(ip_hdr(skb
)->daddr
))
1153 /* loopback traffic; do not push through packet taps again.
1154 * Reset pkt_type for upper layers to process skb
1156 if (skb
->pkt_type
== PACKET_LOOPBACK
) {
1157 skb
->pkt_type
= PACKET_HOST
;
1161 vrf_rx_stats(vrf_dev
, skb
->len
);
1163 if (!list_empty(&vrf_dev
->ptype_all
)) {
1164 skb_push(skb
, skb
->mac_len
);
1165 dev_queue_xmit_nit(skb
, vrf_dev
);
1166 skb_pull(skb
, skb
->mac_len
);
1169 skb
= vrf_rcv_nfhook(NFPROTO_IPV4
, NF_INET_PRE_ROUTING
, skb
, vrf_dev
);
1174 /* called with rcu lock held */
1175 static struct sk_buff
*vrf_l3_rcv(struct net_device
*vrf_dev
,
1176 struct sk_buff
*skb
,
1181 return vrf_ip_rcv(vrf_dev
, skb
);
1183 return vrf_ip6_rcv(vrf_dev
, skb
);
1189 #if IS_ENABLED(CONFIG_IPV6)
1190 /* send to link-local or multicast address via interface enslaved to
1191 * VRF device. Force lookup to VRF table without changing flow struct
1193 static struct dst_entry
*vrf_link_scope_lookup(const struct net_device
*dev
,
1196 struct net
*net
= dev_net(dev
);
1197 int flags
= RT6_LOOKUP_F_IFACE
;
1198 struct dst_entry
*dst
= NULL
;
1199 struct rt6_info
*rt
;
1201 /* VRF device does not have a link-local address and
1202 * sending packets to link-local or mcast addresses over
1203 * a VRF device does not make sense
1205 if (fl6
->flowi6_oif
== dev
->ifindex
) {
1206 dst
= &net
->ipv6
.ip6_null_entry
->dst
;
1211 if (!ipv6_addr_any(&fl6
->saddr
))
1212 flags
|= RT6_LOOKUP_F_HAS_SADDR
;
1214 rt
= vrf_ip6_route_lookup(net
, dev
, fl6
, fl6
->flowi6_oif
, flags
);
1222 static const struct l3mdev_ops vrf_l3mdev_ops
= {
1223 .l3mdev_fib_table
= vrf_fib_table
,
1224 .l3mdev_l3_rcv
= vrf_l3_rcv
,
1225 .l3mdev_l3_out
= vrf_l3_out
,
1226 #if IS_ENABLED(CONFIG_IPV6)
1227 .l3mdev_link_scope_lookup
= vrf_link_scope_lookup
,
1231 static void vrf_get_drvinfo(struct net_device
*dev
,
1232 struct ethtool_drvinfo
*info
)
1234 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
1235 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
1238 static const struct ethtool_ops vrf_ethtool_ops
= {
1239 .get_drvinfo
= vrf_get_drvinfo
,
1242 static inline size_t vrf_fib_rule_nl_size(void)
1246 sz
= NLMSG_ALIGN(sizeof(struct fib_rule_hdr
));
1247 sz
+= nla_total_size(sizeof(u8
)); /* FRA_L3MDEV */
1248 sz
+= nla_total_size(sizeof(u32
)); /* FRA_PRIORITY */
1253 static int vrf_fib_rule(const struct net_device
*dev
, __u8 family
, bool add_it
)
1255 struct fib_rule_hdr
*frh
;
1256 struct nlmsghdr
*nlh
;
1257 struct sk_buff
*skb
;
1260 if (family
== AF_INET6
&& !ipv6_mod_enabled())
1263 skb
= nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL
);
1267 nlh
= nlmsg_put(skb
, 0, 0, 0, sizeof(*frh
), 0);
1269 goto nla_put_failure
;
1271 /* rule only needs to appear once */
1272 nlh
->nlmsg_flags
|= NLM_F_EXCL
;
1274 frh
= nlmsg_data(nlh
);
1275 memset(frh
, 0, sizeof(*frh
));
1276 frh
->family
= family
;
1277 frh
->action
= FR_ACT_TO_TBL
;
1279 if (nla_put_u32(skb
, FRA_L3MDEV
, 1))
1280 goto nla_put_failure
;
1282 if (nla_put_u32(skb
, FRA_PRIORITY
, FIB_RULE_PREF
))
1283 goto nla_put_failure
;
1285 nlmsg_end(skb
, nlh
);
1287 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1288 skb
->sk
= dev_net(dev
)->rtnl
;
1290 err
= fib_nl_newrule(skb
, nlh
, NULL
);
1294 err
= fib_nl_delrule(skb
, nlh
, NULL
);
1308 static int vrf_add_fib_rules(const struct net_device
*dev
)
1312 err
= vrf_fib_rule(dev
, AF_INET
, true);
1316 err
= vrf_fib_rule(dev
, AF_INET6
, true);
1320 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1321 err
= vrf_fib_rule(dev
, RTNL_FAMILY_IPMR
, true);
1328 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1330 vrf_fib_rule(dev
, AF_INET6
, false);
1334 vrf_fib_rule(dev
, AF_INET
, false);
1337 netdev_err(dev
, "Failed to add FIB rules.\n");
1341 static void vrf_setup(struct net_device
*dev
)
1345 /* Initialize the device structure. */
1346 dev
->netdev_ops
= &vrf_netdev_ops
;
1347 dev
->l3mdev_ops
= &vrf_l3mdev_ops
;
1348 dev
->ethtool_ops
= &vrf_ethtool_ops
;
1349 dev
->needs_free_netdev
= true;
1351 /* Fill in device structure with ethernet-generic values. */
1352 eth_hw_addr_random(dev
);
1354 /* don't acquire vrf device's netif_tx_lock when transmitting */
1355 dev
->features
|= NETIF_F_LLTX
;
1357 /* don't allow vrf devices to change network namespaces. */
1358 dev
->features
|= NETIF_F_NETNS_LOCAL
;
1360 /* does not make sense for a VLAN to be added to a vrf device */
1361 dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1363 /* enable offload features */
1364 dev
->features
|= NETIF_F_GSO_SOFTWARE
;
1365 dev
->features
|= NETIF_F_RXCSUM
| NETIF_F_HW_CSUM
;
1366 dev
->features
|= NETIF_F_SG
| NETIF_F_FRAGLIST
| NETIF_F_HIGHDMA
;
1368 dev
->hw_features
= dev
->features
;
1369 dev
->hw_enc_features
= dev
->features
;
1371 /* default to no qdisc; user can add if desired */
1372 dev
->priv_flags
|= IFF_NO_QUEUE
;
1375 static int vrf_validate(struct nlattr
*tb
[], struct nlattr
*data
[],
1376 struct netlink_ext_ack
*extack
)
1378 if (tb
[IFLA_ADDRESS
]) {
1379 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
1381 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
1382 return -EADDRNOTAVAIL
;
1387 static void vrf_dellink(struct net_device
*dev
, struct list_head
*head
)
1389 unregister_netdevice_queue(dev
, head
);
1392 static int vrf_newlink(struct net
*src_net
, struct net_device
*dev
,
1393 struct nlattr
*tb
[], struct nlattr
*data
[],
1394 struct netlink_ext_ack
*extack
)
1396 struct net_vrf
*vrf
= netdev_priv(dev
);
1397 bool *add_fib_rules
;
1401 if (!data
|| !data
[IFLA_VRF_TABLE
])
1404 vrf
->tb_id
= nla_get_u32(data
[IFLA_VRF_TABLE
]);
1405 if (vrf
->tb_id
== RT_TABLE_UNSPEC
)
1408 dev
->priv_flags
|= IFF_L3MDEV_MASTER
;
1410 err
= register_netdevice(dev
);
1415 add_fib_rules
= net_generic(net
, vrf_net_id
);
1416 if (*add_fib_rules
) {
1417 err
= vrf_add_fib_rules(dev
);
1419 unregister_netdevice(dev
);
1422 *add_fib_rules
= false;
1429 static size_t vrf_nl_getsize(const struct net_device
*dev
)
1431 return nla_total_size(sizeof(u32
)); /* IFLA_VRF_TABLE */
1434 static int vrf_fillinfo(struct sk_buff
*skb
,
1435 const struct net_device
*dev
)
1437 struct net_vrf
*vrf
= netdev_priv(dev
);
1439 return nla_put_u32(skb
, IFLA_VRF_TABLE
, vrf
->tb_id
);
1442 static size_t vrf_get_slave_size(const struct net_device
*bond_dev
,
1443 const struct net_device
*slave_dev
)
1445 return nla_total_size(sizeof(u32
)); /* IFLA_VRF_PORT_TABLE */
1448 static int vrf_fill_slave_info(struct sk_buff
*skb
,
1449 const struct net_device
*vrf_dev
,
1450 const struct net_device
*slave_dev
)
1452 struct net_vrf
*vrf
= netdev_priv(vrf_dev
);
1454 if (nla_put_u32(skb
, IFLA_VRF_PORT_TABLE
, vrf
->tb_id
))
1460 static const struct nla_policy vrf_nl_policy
[IFLA_VRF_MAX
+ 1] = {
1461 [IFLA_VRF_TABLE
] = { .type
= NLA_U32
},
1464 static struct rtnl_link_ops vrf_link_ops __read_mostly
= {
1466 .priv_size
= sizeof(struct net_vrf
),
1468 .get_size
= vrf_nl_getsize
,
1469 .policy
= vrf_nl_policy
,
1470 .validate
= vrf_validate
,
1471 .fill_info
= vrf_fillinfo
,
1473 .get_slave_size
= vrf_get_slave_size
,
1474 .fill_slave_info
= vrf_fill_slave_info
,
1476 .newlink
= vrf_newlink
,
1477 .dellink
= vrf_dellink
,
1479 .maxtype
= IFLA_VRF_MAX
,
1482 static int vrf_device_event(struct notifier_block
*unused
,
1483 unsigned long event
, void *ptr
)
1485 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
1487 /* only care about unregister events to drop slave references */
1488 if (event
== NETDEV_UNREGISTER
) {
1489 struct net_device
*vrf_dev
;
1491 if (!netif_is_l3_slave(dev
))
1494 vrf_dev
= netdev_master_upper_dev_get(dev
);
1495 vrf_del_slave(vrf_dev
, dev
);
1501 static struct notifier_block vrf_notifier_block __read_mostly
= {
1502 .notifier_call
= vrf_device_event
,
1505 /* Initialize per network namespace state */
1506 static int __net_init
vrf_netns_init(struct net
*net
)
1508 bool *add_fib_rules
= net_generic(net
, vrf_net_id
);
1510 *add_fib_rules
= true;
1515 static struct pernet_operations vrf_net_ops __net_initdata
= {
1516 .init
= vrf_netns_init
,
1518 .size
= sizeof(bool),
1521 static int __init
vrf_init_module(void)
1525 register_netdevice_notifier(&vrf_notifier_block
);
1527 rc
= register_pernet_subsys(&vrf_net_ops
);
1531 rc
= rtnl_link_register(&vrf_link_ops
);
1533 unregister_pernet_subsys(&vrf_net_ops
);
1540 unregister_netdevice_notifier(&vrf_notifier_block
);
1544 module_init(vrf_init_module
);
1545 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1546 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1547 MODULE_LICENSE("GPL");
1548 MODULE_ALIAS_RTNL_LINK(DRV_NAME
);
1549 MODULE_VERSION(DRV_VERSION
);