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>
40 #define RT_FL_TOS(oldflp4) \
41 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
43 #define DRV_NAME "vrf"
44 #define DRV_VERSION "1.0"
46 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */
47 static bool add_fib_rules
= true;
50 struct rtable __rcu
*rth
;
51 struct rtable __rcu
*rth_local
;
52 struct rt6_info __rcu
*rt6
;
53 struct rt6_info __rcu
*rt6_local
;
64 struct u64_stats_sync syncp
;
67 static void vrf_rx_stats(struct net_device
*dev
, int len
)
69 struct pcpu_dstats
*dstats
= this_cpu_ptr(dev
->dstats
);
71 u64_stats_update_begin(&dstats
->syncp
);
73 dstats
->rx_bytes
+= len
;
74 u64_stats_update_end(&dstats
->syncp
);
77 static void vrf_tx_error(struct net_device
*vrf_dev
, struct sk_buff
*skb
)
79 vrf_dev
->stats
.tx_errors
++;
83 static struct rtnl_link_stats64
*vrf_get_stats64(struct net_device
*dev
,
84 struct rtnl_link_stats64
*stats
)
88 for_each_possible_cpu(i
) {
89 const struct pcpu_dstats
*dstats
;
90 u64 tbytes
, tpkts
, tdrops
, rbytes
, rpkts
;
93 dstats
= per_cpu_ptr(dev
->dstats
, i
);
95 start
= u64_stats_fetch_begin_irq(&dstats
->syncp
);
96 tbytes
= dstats
->tx_bytes
;
97 tpkts
= dstats
->tx_pkts
;
98 tdrops
= dstats
->tx_drps
;
99 rbytes
= dstats
->rx_bytes
;
100 rpkts
= dstats
->rx_pkts
;
101 } while (u64_stats_fetch_retry_irq(&dstats
->syncp
, start
));
102 stats
->tx_bytes
+= tbytes
;
103 stats
->tx_packets
+= tpkts
;
104 stats
->tx_dropped
+= tdrops
;
105 stats
->rx_bytes
+= rbytes
;
106 stats
->rx_packets
+= rpkts
;
111 /* Local traffic destined to local address. Reinsert the packet to rx
112 * path, similar to loopback handling.
114 static int vrf_local_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
115 struct dst_entry
*dst
)
121 skb_dst_set(skb
, dst
);
124 /* set pkt_type to avoid skb hitting packet taps twice -
125 * once on Tx and again in Rx processing
127 skb
->pkt_type
= PACKET_LOOPBACK
;
129 skb
->protocol
= eth_type_trans(skb
, dev
);
131 if (likely(netif_rx(skb
) == NET_RX_SUCCESS
))
132 vrf_rx_stats(dev
, len
);
134 this_cpu_inc(dev
->dstats
->rx_drps
);
139 #if IS_ENABLED(CONFIG_IPV6)
140 static int vrf_ip6_local_out(struct net
*net
, struct sock
*sk
,
145 err
= nf_hook(NFPROTO_IPV6
, NF_INET_LOCAL_OUT
, net
,
146 sk
, skb
, NULL
, skb_dst(skb
)->dev
, dst_output
);
148 if (likely(err
== 1))
149 err
= dst_output(net
, sk
, skb
);
154 static netdev_tx_t
vrf_process_v6_outbound(struct sk_buff
*skb
,
155 struct net_device
*dev
)
157 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
158 struct net
*net
= dev_net(skb
->dev
);
159 struct flowi6 fl6
= {
160 /* needed to match OIF rule */
161 .flowi6_oif
= dev
->ifindex
,
162 .flowi6_iif
= LOOPBACK_IFINDEX
,
165 .flowlabel
= ip6_flowinfo(iph
),
166 .flowi6_mark
= skb
->mark
,
167 .flowi6_proto
= iph
->nexthdr
,
168 .flowi6_flags
= FLOWI_FLAG_SKIP_NH_OIF
,
170 int ret
= NET_XMIT_DROP
;
171 struct dst_entry
*dst
;
172 struct dst_entry
*dst_null
= &net
->ipv6
.ip6_null_entry
->dst
;
174 dst
= ip6_route_output(net
, NULL
, &fl6
);
180 /* if dst.dev is loopback or the VRF device again this is locally
181 * originated traffic destined to a local address. Short circuit
182 * to Rx path using our local dst
184 if (dst
->dev
== net
->loopback_dev
|| dst
->dev
== dev
) {
185 struct net_vrf
*vrf
= netdev_priv(dev
);
186 struct rt6_info
*rt6_local
;
188 /* release looked up dst and use cached local dst */
193 rt6_local
= rcu_dereference(vrf
->rt6_local
);
194 if (unlikely(!rt6_local
)) {
199 /* Ordering issue: cached local dst is created on newlink
200 * before the IPv6 initialization. Using the local dst
201 * requires rt6i_idev to be set so make sure it is.
203 if (unlikely(!rt6_local
->rt6i_idev
)) {
204 rt6_local
->rt6i_idev
= in6_dev_get(dev
);
205 if (!rt6_local
->rt6i_idev
) {
211 dst
= &rt6_local
->dst
;
216 return vrf_local_xmit(skb
, dev
, &rt6_local
->dst
);
219 skb_dst_set(skb
, dst
);
221 /* strip the ethernet header added for pass through VRF device */
222 __skb_pull(skb
, skb_network_offset(skb
));
224 ret
= vrf_ip6_local_out(net
, skb
->sk
, skb
);
225 if (unlikely(net_xmit_eval(ret
)))
226 dev
->stats
.tx_errors
++;
228 ret
= NET_XMIT_SUCCESS
;
232 vrf_tx_error(dev
, skb
);
233 return NET_XMIT_DROP
;
236 static netdev_tx_t
vrf_process_v6_outbound(struct sk_buff
*skb
,
237 struct net_device
*dev
)
239 vrf_tx_error(dev
, skb
);
240 return NET_XMIT_DROP
;
244 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
245 static int vrf_ip_local_out(struct net
*net
, struct sock
*sk
,
250 err
= nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
, net
, sk
,
251 skb
, NULL
, skb_dst(skb
)->dev
, dst_output
);
252 if (likely(err
== 1))
253 err
= dst_output(net
, sk
, skb
);
258 static netdev_tx_t
vrf_process_v4_outbound(struct sk_buff
*skb
,
259 struct net_device
*vrf_dev
)
261 struct iphdr
*ip4h
= ip_hdr(skb
);
262 int ret
= NET_XMIT_DROP
;
263 struct flowi4 fl4
= {
264 /* needed to match OIF rule */
265 .flowi4_oif
= vrf_dev
->ifindex
,
266 .flowi4_iif
= LOOPBACK_IFINDEX
,
267 .flowi4_tos
= RT_TOS(ip4h
->tos
),
268 .flowi4_flags
= FLOWI_FLAG_ANYSRC
| FLOWI_FLAG_SKIP_NH_OIF
,
269 .daddr
= ip4h
->daddr
,
271 struct net
*net
= dev_net(vrf_dev
);
274 rt
= ip_route_output_flow(net
, &fl4
, NULL
);
278 if (rt
->rt_type
!= RTN_UNICAST
&& rt
->rt_type
!= RTN_LOCAL
) {
285 /* if dst.dev is loopback or the VRF device again this is locally
286 * originated traffic destined to a local address. Short circuit
287 * to Rx path using our local dst
289 if (rt
->dst
.dev
== net
->loopback_dev
|| rt
->dst
.dev
== vrf_dev
) {
290 struct net_vrf
*vrf
= netdev_priv(vrf_dev
);
291 struct rtable
*rth_local
;
292 struct dst_entry
*dst
= NULL
;
298 rth_local
= rcu_dereference(vrf
->rth_local
);
299 if (likely(rth_local
)) {
300 dst
= &rth_local
->dst
;
309 return vrf_local_xmit(skb
, vrf_dev
, dst
);
312 skb_dst_set(skb
, &rt
->dst
);
314 /* strip the ethernet header added for pass through VRF device */
315 __skb_pull(skb
, skb_network_offset(skb
));
318 ip4h
->saddr
= inet_select_addr(skb_dst(skb
)->dev
, 0,
322 ret
= vrf_ip_local_out(dev_net(skb_dst(skb
)->dev
), skb
->sk
, skb
);
323 if (unlikely(net_xmit_eval(ret
)))
324 vrf_dev
->stats
.tx_errors
++;
326 ret
= NET_XMIT_SUCCESS
;
331 vrf_tx_error(vrf_dev
, skb
);
335 static netdev_tx_t
is_ip_tx_frame(struct sk_buff
*skb
, struct net_device
*dev
)
337 switch (skb
->protocol
) {
338 case htons(ETH_P_IP
):
339 return vrf_process_v4_outbound(skb
, dev
);
340 case htons(ETH_P_IPV6
):
341 return vrf_process_v6_outbound(skb
, dev
);
343 vrf_tx_error(dev
, skb
);
344 return NET_XMIT_DROP
;
348 static netdev_tx_t
vrf_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
350 netdev_tx_t ret
= is_ip_tx_frame(skb
, dev
);
352 if (likely(ret
== NET_XMIT_SUCCESS
|| ret
== NET_XMIT_CN
)) {
353 struct pcpu_dstats
*dstats
= this_cpu_ptr(dev
->dstats
);
355 u64_stats_update_begin(&dstats
->syncp
);
357 dstats
->tx_bytes
+= skb
->len
;
358 u64_stats_update_end(&dstats
->syncp
);
360 this_cpu_inc(dev
->dstats
->tx_drps
);
366 #if IS_ENABLED(CONFIG_IPV6)
367 /* modelled after ip6_finish_output2 */
368 static int vrf_finish_output6(struct net
*net
, struct sock
*sk
,
371 struct dst_entry
*dst
= skb_dst(skb
);
372 struct net_device
*dev
= dst
->dev
;
373 struct neighbour
*neigh
;
374 struct in6_addr
*nexthop
;
377 skb
->protocol
= htons(ETH_P_IPV6
);
381 nexthop
= rt6_nexthop((struct rt6_info
*)dst
, &ipv6_hdr(skb
)->daddr
);
382 neigh
= __ipv6_neigh_lookup_noref(dst
->dev
, nexthop
);
383 if (unlikely(!neigh
))
384 neigh
= __neigh_create(&nd_tbl
, nexthop
, dst
->dev
, false);
385 if (!IS_ERR(neigh
)) {
386 ret
= dst_neigh_output(dst
, neigh
, skb
);
387 rcu_read_unlock_bh();
390 rcu_read_unlock_bh();
392 IP6_INC_STATS(dev_net(dst
->dev
),
393 ip6_dst_idev(dst
), IPSTATS_MIB_OUTNOROUTES
);
398 /* modelled after ip6_output */
399 static int vrf_output6(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
401 return NF_HOOK_COND(NFPROTO_IPV6
, NF_INET_POST_ROUTING
,
402 net
, sk
, skb
, NULL
, skb_dst(skb
)->dev
,
404 !(IP6CB(skb
)->flags
& IP6SKB_REROUTED
));
407 /* set dst on skb to send packet to us via dev_xmit path. Allows
408 * packet to go through device based features such as qdisc, netfilter
409 * hooks and packet sockets with skb->dev set to vrf device.
411 static struct sk_buff
*vrf_ip6_out(struct net_device
*vrf_dev
,
415 struct net_vrf
*vrf
= netdev_priv(vrf_dev
);
416 struct dst_entry
*dst
= NULL
;
417 struct rt6_info
*rt6
;
419 /* don't divert link scope packets */
420 if (rt6_need_strict(&ipv6_hdr(skb
)->daddr
))
425 rt6
= rcu_dereference(vrf
->rt6
);
433 if (unlikely(!dst
)) {
434 vrf_tx_error(vrf_dev
, skb
);
439 skb_dst_set(skb
, dst
);
445 static void vrf_rt6_release(struct net_device
*dev
, struct net_vrf
*vrf
)
447 struct rt6_info
*rt6
= rtnl_dereference(vrf
->rt6
);
448 struct rt6_info
*rt6_local
= rtnl_dereference(vrf
->rt6_local
);
449 struct net
*net
= dev_net(dev
);
450 struct dst_entry
*dst
;
452 RCU_INIT_POINTER(vrf
->rt6
, NULL
);
453 RCU_INIT_POINTER(vrf
->rt6_local
, NULL
);
456 /* move dev in dst's to loopback so this VRF device can be deleted
457 * - based on dst_ifdown
462 dst
->dev
= net
->loopback_dev
;
468 if (rt6_local
->rt6i_idev
)
469 in6_dev_put(rt6_local
->rt6i_idev
);
471 dst
= &rt6_local
->dst
;
473 dst
->dev
= net
->loopback_dev
;
479 static int vrf_rt6_create(struct net_device
*dev
)
481 int flags
= DST_HOST
| DST_NOPOLICY
| DST_NOXFRM
| DST_NOCACHE
;
482 struct net_vrf
*vrf
= netdev_priv(dev
);
483 struct net
*net
= dev_net(dev
);
484 struct fib6_table
*rt6i_table
;
485 struct rt6_info
*rt6
, *rt6_local
;
488 /* IPv6 can be CONFIG enabled and then disabled runtime */
489 if (!ipv6_mod_enabled())
492 rt6i_table
= fib6_new_table(net
, vrf
->tb_id
);
496 /* create a dst for routing packets out a VRF device */
497 rt6
= ip6_dst_alloc(net
, dev
, flags
);
503 rt6
->rt6i_table
= rt6i_table
;
504 rt6
->dst
.output
= vrf_output6
;
506 /* create a dst for local routing - packets sent locally
507 * to local address via the VRF device as a loopback
509 rt6_local
= ip6_dst_alloc(net
, dev
, flags
);
511 dst_release(&rt6
->dst
);
515 dst_hold(&rt6_local
->dst
);
517 rt6_local
->rt6i_idev
= in6_dev_get(dev
);
518 rt6_local
->rt6i_flags
= RTF_UP
| RTF_NONEXTHOP
| RTF_LOCAL
;
519 rt6_local
->rt6i_table
= rt6i_table
;
520 rt6_local
->dst
.input
= ip6_input
;
522 rcu_assign_pointer(vrf
->rt6
, rt6
);
523 rcu_assign_pointer(vrf
->rt6_local
, rt6_local
);
530 static struct sk_buff
*vrf_ip6_out(struct net_device
*vrf_dev
,
537 static void vrf_rt6_release(struct net_device
*dev
, struct net_vrf
*vrf
)
541 static int vrf_rt6_create(struct net_device
*dev
)
547 /* modelled after ip_finish_output2 */
548 static int vrf_finish_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
550 struct dst_entry
*dst
= skb_dst(skb
);
551 struct rtable
*rt
= (struct rtable
*)dst
;
552 struct net_device
*dev
= dst
->dev
;
553 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
554 struct neighbour
*neigh
;
558 /* Be paranoid, rather than too clever. */
559 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
560 struct sk_buff
*skb2
;
562 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
568 skb_set_owner_w(skb2
, skb
->sk
);
576 nexthop
= (__force u32
)rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
577 neigh
= __ipv4_neigh_lookup_noref(dev
, nexthop
);
578 if (unlikely(!neigh
))
579 neigh
= __neigh_create(&arp_tbl
, &nexthop
, dev
, false);
581 ret
= dst_neigh_output(dst
, neigh
, skb
);
583 rcu_read_unlock_bh();
585 if (unlikely(ret
< 0))
586 vrf_tx_error(skb
->dev
, skb
);
590 static int vrf_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
592 struct net_device
*dev
= skb_dst(skb
)->dev
;
594 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUT
, skb
->len
);
597 skb
->protocol
= htons(ETH_P_IP
);
599 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
600 net
, sk
, skb
, NULL
, dev
,
602 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
605 /* set dst on skb to send packet to us via dev_xmit path. Allows
606 * packet to go through device based features such as qdisc, netfilter
607 * hooks and packet sockets with skb->dev set to vrf device.
609 static struct sk_buff
*vrf_ip_out(struct net_device
*vrf_dev
,
613 struct net_vrf
*vrf
= netdev_priv(vrf_dev
);
614 struct dst_entry
*dst
= NULL
;
619 rth
= rcu_dereference(vrf
->rth
);
627 if (unlikely(!dst
)) {
628 vrf_tx_error(vrf_dev
, skb
);
633 skb_dst_set(skb
, dst
);
638 /* called with rcu lock held */
639 static struct sk_buff
*vrf_l3_out(struct net_device
*vrf_dev
,
646 return vrf_ip_out(vrf_dev
, sk
, skb
);
648 return vrf_ip6_out(vrf_dev
, sk
, skb
);
655 static void vrf_rtable_release(struct net_device
*dev
, struct net_vrf
*vrf
)
657 struct rtable
*rth
= rtnl_dereference(vrf
->rth
);
658 struct rtable
*rth_local
= rtnl_dereference(vrf
->rth_local
);
659 struct net
*net
= dev_net(dev
);
660 struct dst_entry
*dst
;
662 RCU_INIT_POINTER(vrf
->rth
, NULL
);
663 RCU_INIT_POINTER(vrf
->rth_local
, NULL
);
666 /* move dev in dst's to loopback so this VRF device can be deleted
667 * - based on dst_ifdown
672 dst
->dev
= net
->loopback_dev
;
678 dst
= &rth_local
->dst
;
680 dst
->dev
= net
->loopback_dev
;
686 static int vrf_rtable_create(struct net_device
*dev
)
688 struct net_vrf
*vrf
= netdev_priv(dev
);
689 struct rtable
*rth
, *rth_local
;
691 if (!fib_new_table(dev_net(dev
), vrf
->tb_id
))
694 /* create a dst for routing packets out through a VRF device */
695 rth
= rt_dst_alloc(dev
, 0, RTN_UNICAST
, 1, 1, 0);
699 /* create a dst for local ingress routing - packets sent locally
700 * to local address via the VRF device as a loopback
702 rth_local
= rt_dst_alloc(dev
, RTCF_LOCAL
, RTN_LOCAL
, 1, 1, 0);
704 dst_release(&rth
->dst
);
708 rth
->dst
.output
= vrf_output
;
709 rth
->rt_table_id
= vrf
->tb_id
;
711 rth_local
->rt_table_id
= vrf
->tb_id
;
713 rcu_assign_pointer(vrf
->rth
, rth
);
714 rcu_assign_pointer(vrf
->rth_local
, rth_local
);
719 /**************************** device handling ********************/
721 /* cycle interface to flush neighbor cache and move routes across tables */
722 static void cycle_netdev(struct net_device
*dev
)
724 unsigned int flags
= dev
->flags
;
727 if (!netif_running(dev
))
730 ret
= dev_change_flags(dev
, flags
& ~IFF_UP
);
732 ret
= dev_change_flags(dev
, flags
);
736 "Failed to cycle device %s; route tables might be wrong!\n",
741 static int do_vrf_add_slave(struct net_device
*dev
, struct net_device
*port_dev
)
745 ret
= netdev_master_upper_dev_link(port_dev
, dev
, NULL
, NULL
);
749 port_dev
->priv_flags
|= IFF_L3MDEV_SLAVE
;
750 cycle_netdev(port_dev
);
755 static int vrf_add_slave(struct net_device
*dev
, struct net_device
*port_dev
)
757 if (netif_is_l3_master(port_dev
) || netif_is_l3_slave(port_dev
))
760 return do_vrf_add_slave(dev
, port_dev
);
763 /* inverse of do_vrf_add_slave */
764 static int do_vrf_del_slave(struct net_device
*dev
, struct net_device
*port_dev
)
766 netdev_upper_dev_unlink(port_dev
, dev
);
767 port_dev
->priv_flags
&= ~IFF_L3MDEV_SLAVE
;
769 cycle_netdev(port_dev
);
774 static int vrf_del_slave(struct net_device
*dev
, struct net_device
*port_dev
)
776 return do_vrf_del_slave(dev
, port_dev
);
779 static void vrf_dev_uninit(struct net_device
*dev
)
781 struct net_vrf
*vrf
= netdev_priv(dev
);
782 struct net_device
*port_dev
;
783 struct list_head
*iter
;
785 vrf_rtable_release(dev
, vrf
);
786 vrf_rt6_release(dev
, vrf
);
788 netdev_for_each_lower_dev(dev
, port_dev
, iter
)
789 vrf_del_slave(dev
, port_dev
);
791 free_percpu(dev
->dstats
);
795 static int vrf_dev_init(struct net_device
*dev
)
797 struct net_vrf
*vrf
= netdev_priv(dev
);
799 dev
->dstats
= netdev_alloc_pcpu_stats(struct pcpu_dstats
);
803 /* create the default dst which points back to us */
804 if (vrf_rtable_create(dev
) != 0)
807 if (vrf_rt6_create(dev
) != 0)
810 dev
->flags
= IFF_MASTER
| IFF_NOARP
;
812 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
813 dev
->mtu
= 64 * 1024;
815 /* similarly, oper state is irrelevant; set to up to avoid confusion */
816 dev
->operstate
= IF_OPER_UP
;
817 netdev_lockdep_set_classes(dev
);
821 vrf_rtable_release(dev
, vrf
);
823 free_percpu(dev
->dstats
);
829 static const struct net_device_ops vrf_netdev_ops
= {
830 .ndo_init
= vrf_dev_init
,
831 .ndo_uninit
= vrf_dev_uninit
,
832 .ndo_start_xmit
= vrf_xmit
,
833 .ndo_get_stats64
= vrf_get_stats64
,
834 .ndo_add_slave
= vrf_add_slave
,
835 .ndo_del_slave
= vrf_del_slave
,
838 static u32
vrf_fib_table(const struct net_device
*dev
)
840 struct net_vrf
*vrf
= netdev_priv(dev
);
845 static int vrf_rcv_finish(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
850 static struct sk_buff
*vrf_rcv_nfhook(u8 pf
, unsigned int hook
,
852 struct net_device
*dev
)
854 struct net
*net
= dev_net(dev
);
858 if (NF_HOOK(pf
, hook
, net
, NULL
, skb
, dev
, NULL
, vrf_rcv_finish
) < 0)
859 skb
= NULL
; /* kfree_skb(skb) handled by nf code */
864 #if IS_ENABLED(CONFIG_IPV6)
865 /* neighbor handling is done with actual device; do not want
866 * to flip skb->dev for those ndisc packets. This really fails
867 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
870 static bool ipv6_ndisc_frame(const struct sk_buff
*skb
)
872 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
875 if (iph
->nexthdr
== NEXTHDR_ICMP
) {
876 const struct icmp6hdr
*icmph
;
877 struct icmp6hdr _icmph
;
879 icmph
= skb_header_pointer(skb
, sizeof(*iph
),
880 sizeof(_icmph
), &_icmph
);
884 switch (icmph
->icmp6_type
) {
885 case NDISC_ROUTER_SOLICITATION
:
886 case NDISC_ROUTER_ADVERTISEMENT
:
887 case NDISC_NEIGHBOUR_SOLICITATION
:
888 case NDISC_NEIGHBOUR_ADVERTISEMENT
:
899 static struct rt6_info
*vrf_ip6_route_lookup(struct net
*net
,
900 const struct net_device
*dev
,
905 struct net_vrf
*vrf
= netdev_priv(dev
);
906 struct fib6_table
*table
= NULL
;
907 struct rt6_info
*rt6
;
911 /* fib6_table does not have a refcnt and can not be freed */
912 rt6
= rcu_dereference(vrf
->rt6
);
914 table
= rt6
->rt6i_table
;
921 return ip6_pol_route(net
, table
, ifindex
, fl6
, flags
);
924 static void vrf_ip6_input_dst(struct sk_buff
*skb
, struct net_device
*vrf_dev
,
927 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
928 struct flowi6 fl6
= {
931 .flowlabel
= ip6_flowinfo(iph
),
932 .flowi6_mark
= skb
->mark
,
933 .flowi6_proto
= iph
->nexthdr
,
934 .flowi6_iif
= ifindex
,
936 struct net
*net
= dev_net(vrf_dev
);
937 struct rt6_info
*rt6
;
939 rt6
= vrf_ip6_route_lookup(net
, vrf_dev
, &fl6
, ifindex
,
940 RT6_LOOKUP_F_HAS_SADDR
| RT6_LOOKUP_F_IFACE
);
944 if (unlikely(&rt6
->dst
== &net
->ipv6
.ip6_null_entry
->dst
))
947 skb_dst_set(skb
, &rt6
->dst
);
950 static struct sk_buff
*vrf_ip6_rcv(struct net_device
*vrf_dev
,
953 int orig_iif
= skb
->skb_iif
;
956 /* loopback traffic; do not push through packet taps again.
957 * Reset pkt_type for upper layers to process skb
959 if (skb
->pkt_type
== PACKET_LOOPBACK
) {
961 skb
->skb_iif
= vrf_dev
->ifindex
;
962 skb
->pkt_type
= PACKET_HOST
;
966 /* if packet is NDISC or addressed to multicast or link-local
967 * then keep the ingress interface
969 need_strict
= rt6_need_strict(&ipv6_hdr(skb
)->daddr
);
970 if (!ipv6_ndisc_frame(skb
) && !need_strict
) {
972 skb
->skb_iif
= vrf_dev
->ifindex
;
974 skb_push(skb
, skb
->mac_len
);
975 dev_queue_xmit_nit(skb
, vrf_dev
);
976 skb_pull(skb
, skb
->mac_len
);
978 IP6CB(skb
)->flags
|= IP6SKB_L3SLAVE
;
982 vrf_ip6_input_dst(skb
, vrf_dev
, orig_iif
);
984 skb
= vrf_rcv_nfhook(NFPROTO_IPV6
, NF_INET_PRE_ROUTING
, skb
, vrf_dev
);
990 static struct sk_buff
*vrf_ip6_rcv(struct net_device
*vrf_dev
,
997 static struct sk_buff
*vrf_ip_rcv(struct net_device
*vrf_dev
,
1001 skb
->skb_iif
= vrf_dev
->ifindex
;
1003 /* loopback traffic; do not push through packet taps again.
1004 * Reset pkt_type for upper layers to process skb
1006 if (skb
->pkt_type
== PACKET_LOOPBACK
) {
1007 skb
->pkt_type
= PACKET_HOST
;
1011 skb_push(skb
, skb
->mac_len
);
1012 dev_queue_xmit_nit(skb
, vrf_dev
);
1013 skb_pull(skb
, skb
->mac_len
);
1015 skb
= vrf_rcv_nfhook(NFPROTO_IPV4
, NF_INET_PRE_ROUTING
, skb
, vrf_dev
);
1020 /* called with rcu lock held */
1021 static struct sk_buff
*vrf_l3_rcv(struct net_device
*vrf_dev
,
1022 struct sk_buff
*skb
,
1027 return vrf_ip_rcv(vrf_dev
, skb
);
1029 return vrf_ip6_rcv(vrf_dev
, skb
);
1035 #if IS_ENABLED(CONFIG_IPV6)
1036 /* send to link-local or multicast address via interface enslaved to
1037 * VRF device. Force lookup to VRF table without changing flow struct
1039 static struct dst_entry
*vrf_link_scope_lookup(const struct net_device
*dev
,
1042 struct net
*net
= dev_net(dev
);
1043 int flags
= RT6_LOOKUP_F_IFACE
;
1044 struct dst_entry
*dst
= NULL
;
1045 struct rt6_info
*rt
;
1047 /* VRF device does not have a link-local address and
1048 * sending packets to link-local or mcast addresses over
1049 * a VRF device does not make sense
1051 if (fl6
->flowi6_oif
== dev
->ifindex
) {
1052 dst
= &net
->ipv6
.ip6_null_entry
->dst
;
1057 if (!ipv6_addr_any(&fl6
->saddr
))
1058 flags
|= RT6_LOOKUP_F_HAS_SADDR
;
1060 rt
= vrf_ip6_route_lookup(net
, dev
, fl6
, fl6
->flowi6_oif
, flags
);
1068 static const struct l3mdev_ops vrf_l3mdev_ops
= {
1069 .l3mdev_fib_table
= vrf_fib_table
,
1070 .l3mdev_l3_rcv
= vrf_l3_rcv
,
1071 .l3mdev_l3_out
= vrf_l3_out
,
1072 #if IS_ENABLED(CONFIG_IPV6)
1073 .l3mdev_link_scope_lookup
= vrf_link_scope_lookup
,
1077 static void vrf_get_drvinfo(struct net_device
*dev
,
1078 struct ethtool_drvinfo
*info
)
1080 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
1081 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
1084 static const struct ethtool_ops vrf_ethtool_ops
= {
1085 .get_drvinfo
= vrf_get_drvinfo
,
1088 static inline size_t vrf_fib_rule_nl_size(void)
1092 sz
= NLMSG_ALIGN(sizeof(struct fib_rule_hdr
));
1093 sz
+= nla_total_size(sizeof(u8
)); /* FRA_L3MDEV */
1094 sz
+= nla_total_size(sizeof(u32
)); /* FRA_PRIORITY */
1099 static int vrf_fib_rule(const struct net_device
*dev
, __u8 family
, bool add_it
)
1101 struct fib_rule_hdr
*frh
;
1102 struct nlmsghdr
*nlh
;
1103 struct sk_buff
*skb
;
1106 if (family
== AF_INET6
&& !ipv6_mod_enabled())
1109 skb
= nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL
);
1113 nlh
= nlmsg_put(skb
, 0, 0, 0, sizeof(*frh
), 0);
1115 goto nla_put_failure
;
1117 /* rule only needs to appear once */
1118 nlh
->nlmsg_flags
&= NLM_F_EXCL
;
1120 frh
= nlmsg_data(nlh
);
1121 memset(frh
, 0, sizeof(*frh
));
1122 frh
->family
= family
;
1123 frh
->action
= FR_ACT_TO_TBL
;
1125 if (nla_put_u32(skb
, FRA_L3MDEV
, 1))
1126 goto nla_put_failure
;
1128 if (nla_put_u32(skb
, FRA_PRIORITY
, FIB_RULE_PREF
))
1129 goto nla_put_failure
;
1131 nlmsg_end(skb
, nlh
);
1133 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1134 skb
->sk
= dev_net(dev
)->rtnl
;
1136 err
= fib_nl_newrule(skb
, nlh
);
1140 err
= fib_nl_delrule(skb
, nlh
);
1154 static int vrf_add_fib_rules(const struct net_device
*dev
)
1158 err
= vrf_fib_rule(dev
, AF_INET
, true);
1162 err
= vrf_fib_rule(dev
, AF_INET6
, true);
1169 vrf_fib_rule(dev
, AF_INET
, false);
1172 netdev_err(dev
, "Failed to add FIB rules.\n");
1176 static void vrf_setup(struct net_device
*dev
)
1180 /* Initialize the device structure. */
1181 dev
->netdev_ops
= &vrf_netdev_ops
;
1182 dev
->l3mdev_ops
= &vrf_l3mdev_ops
;
1183 dev
->ethtool_ops
= &vrf_ethtool_ops
;
1184 dev
->destructor
= free_netdev
;
1186 /* Fill in device structure with ethernet-generic values. */
1187 eth_hw_addr_random(dev
);
1189 /* don't acquire vrf device's netif_tx_lock when transmitting */
1190 dev
->features
|= NETIF_F_LLTX
;
1192 /* don't allow vrf devices to change network namespaces. */
1193 dev
->features
|= NETIF_F_NETNS_LOCAL
;
1195 /* does not make sense for a VLAN to be added to a vrf device */
1196 dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1198 /* enable offload features */
1199 dev
->features
|= NETIF_F_GSO_SOFTWARE
;
1200 dev
->features
|= NETIF_F_RXCSUM
| NETIF_F_HW_CSUM
;
1201 dev
->features
|= NETIF_F_SG
| NETIF_F_FRAGLIST
| NETIF_F_HIGHDMA
;
1203 dev
->hw_features
= dev
->features
;
1204 dev
->hw_enc_features
= dev
->features
;
1206 /* default to no qdisc; user can add if desired */
1207 dev
->priv_flags
|= IFF_NO_QUEUE
;
1210 static int vrf_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
1212 if (tb
[IFLA_ADDRESS
]) {
1213 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
1215 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
1216 return -EADDRNOTAVAIL
;
1221 static void vrf_dellink(struct net_device
*dev
, struct list_head
*head
)
1223 unregister_netdevice_queue(dev
, head
);
1226 static int vrf_newlink(struct net
*src_net
, struct net_device
*dev
,
1227 struct nlattr
*tb
[], struct nlattr
*data
[])
1229 struct net_vrf
*vrf
= netdev_priv(dev
);
1232 if (!data
|| !data
[IFLA_VRF_TABLE
])
1235 vrf
->tb_id
= nla_get_u32(data
[IFLA_VRF_TABLE
]);
1237 dev
->priv_flags
|= IFF_L3MDEV_MASTER
;
1239 err
= register_netdevice(dev
);
1243 if (add_fib_rules
) {
1244 err
= vrf_add_fib_rules(dev
);
1246 unregister_netdevice(dev
);
1249 add_fib_rules
= false;
1256 static size_t vrf_nl_getsize(const struct net_device
*dev
)
1258 return nla_total_size(sizeof(u32
)); /* IFLA_VRF_TABLE */
1261 static int vrf_fillinfo(struct sk_buff
*skb
,
1262 const struct net_device
*dev
)
1264 struct net_vrf
*vrf
= netdev_priv(dev
);
1266 return nla_put_u32(skb
, IFLA_VRF_TABLE
, vrf
->tb_id
);
1269 static size_t vrf_get_slave_size(const struct net_device
*bond_dev
,
1270 const struct net_device
*slave_dev
)
1272 return nla_total_size(sizeof(u32
)); /* IFLA_VRF_PORT_TABLE */
1275 static int vrf_fill_slave_info(struct sk_buff
*skb
,
1276 const struct net_device
*vrf_dev
,
1277 const struct net_device
*slave_dev
)
1279 struct net_vrf
*vrf
= netdev_priv(vrf_dev
);
1281 if (nla_put_u32(skb
, IFLA_VRF_PORT_TABLE
, vrf
->tb_id
))
1287 static const struct nla_policy vrf_nl_policy
[IFLA_VRF_MAX
+ 1] = {
1288 [IFLA_VRF_TABLE
] = { .type
= NLA_U32
},
1291 static struct rtnl_link_ops vrf_link_ops __read_mostly
= {
1293 .priv_size
= sizeof(struct net_vrf
),
1295 .get_size
= vrf_nl_getsize
,
1296 .policy
= vrf_nl_policy
,
1297 .validate
= vrf_validate
,
1298 .fill_info
= vrf_fillinfo
,
1300 .get_slave_size
= vrf_get_slave_size
,
1301 .fill_slave_info
= vrf_fill_slave_info
,
1303 .newlink
= vrf_newlink
,
1304 .dellink
= vrf_dellink
,
1306 .maxtype
= IFLA_VRF_MAX
,
1309 static int vrf_device_event(struct notifier_block
*unused
,
1310 unsigned long event
, void *ptr
)
1312 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
1314 /* only care about unregister events to drop slave references */
1315 if (event
== NETDEV_UNREGISTER
) {
1316 struct net_device
*vrf_dev
;
1318 if (!netif_is_l3_slave(dev
))
1321 vrf_dev
= netdev_master_upper_dev_get(dev
);
1322 vrf_del_slave(vrf_dev
, dev
);
1328 static struct notifier_block vrf_notifier_block __read_mostly
= {
1329 .notifier_call
= vrf_device_event
,
1332 static int __init
vrf_init_module(void)
1336 register_netdevice_notifier(&vrf_notifier_block
);
1338 rc
= rtnl_link_register(&vrf_link_ops
);
1345 unregister_netdevice_notifier(&vrf_notifier_block
);
1349 module_init(vrf_init_module
);
1350 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1351 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1352 MODULE_LICENSE("GPL");
1353 MODULE_ALIAS_RTNL_LINK(DRV_NAME
);
1354 MODULE_VERSION(DRV_VERSION
);