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 DRV_NAME "vrf"
41 #define DRV_VERSION "1.0"
43 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */
44 static bool add_fib_rules
= true;
47 struct rtable __rcu
*rth
;
48 struct rtable __rcu
*rth_local
;
49 struct rt6_info __rcu
*rt6
;
50 struct rt6_info __rcu
*rt6_local
;
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 struct rtnl_link_stats64
*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
;
108 /* Local traffic destined to local address. Reinsert the packet to rx
109 * path, similar to loopback handling.
111 static int vrf_local_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
112 struct dst_entry
*dst
)
118 skb_dst_set(skb
, dst
);
121 /* set pkt_type to avoid skb hitting packet taps twice -
122 * once on Tx and again in Rx processing
124 skb
->pkt_type
= PACKET_LOOPBACK
;
126 skb
->protocol
= eth_type_trans(skb
, dev
);
128 if (likely(netif_rx(skb
) == NET_RX_SUCCESS
))
129 vrf_rx_stats(dev
, len
);
131 this_cpu_inc(dev
->dstats
->rx_drps
);
136 #if IS_ENABLED(CONFIG_IPV6)
137 static int vrf_ip6_local_out(struct net
*net
, struct sock
*sk
,
142 err
= nf_hook(NFPROTO_IPV6
, NF_INET_LOCAL_OUT
, net
,
143 sk
, skb
, NULL
, skb_dst(skb
)->dev
, dst_output
);
145 if (likely(err
== 1))
146 err
= dst_output(net
, sk
, skb
);
151 static netdev_tx_t
vrf_process_v6_outbound(struct sk_buff
*skb
,
152 struct net_device
*dev
)
154 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
155 struct net
*net
= dev_net(skb
->dev
);
156 struct flowi6 fl6
= {
157 /* needed to match OIF rule */
158 .flowi6_oif
= dev
->ifindex
,
159 .flowi6_iif
= LOOPBACK_IFINDEX
,
162 .flowlabel
= ip6_flowinfo(iph
),
163 .flowi6_mark
= skb
->mark
,
164 .flowi6_proto
= iph
->nexthdr
,
165 .flowi6_flags
= FLOWI_FLAG_SKIP_NH_OIF
,
167 int ret
= NET_XMIT_DROP
;
168 struct dst_entry
*dst
;
169 struct dst_entry
*dst_null
= &net
->ipv6
.ip6_null_entry
->dst
;
171 dst
= ip6_route_output(net
, NULL
, &fl6
);
177 /* if dst.dev is loopback or the VRF device again this is locally
178 * originated traffic destined to a local address. Short circuit
179 * to Rx path using our local dst
181 if (dst
->dev
== net
->loopback_dev
|| dst
->dev
== dev
) {
182 struct net_vrf
*vrf
= netdev_priv(dev
);
183 struct rt6_info
*rt6_local
;
185 /* release looked up dst and use cached local dst */
190 rt6_local
= rcu_dereference(vrf
->rt6_local
);
191 if (unlikely(!rt6_local
)) {
196 /* Ordering issue: cached local dst is created on newlink
197 * before the IPv6 initialization. Using the local dst
198 * requires rt6i_idev to be set so make sure it is.
200 if (unlikely(!rt6_local
->rt6i_idev
)) {
201 rt6_local
->rt6i_idev
= in6_dev_get(dev
);
202 if (!rt6_local
->rt6i_idev
) {
208 dst
= &rt6_local
->dst
;
213 return vrf_local_xmit(skb
, dev
, &rt6_local
->dst
);
216 skb_dst_set(skb
, dst
);
218 /* strip the ethernet header added for pass through VRF device */
219 __skb_pull(skb
, skb_network_offset(skb
));
221 ret
= vrf_ip6_local_out(net
, skb
->sk
, skb
);
222 if (unlikely(net_xmit_eval(ret
)))
223 dev
->stats
.tx_errors
++;
225 ret
= NET_XMIT_SUCCESS
;
229 vrf_tx_error(dev
, skb
);
230 return NET_XMIT_DROP
;
233 static netdev_tx_t
vrf_process_v6_outbound(struct sk_buff
*skb
,
234 struct net_device
*dev
)
236 vrf_tx_error(dev
, skb
);
237 return NET_XMIT_DROP
;
241 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
242 static int vrf_ip_local_out(struct net
*net
, struct sock
*sk
,
247 err
= nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
, net
, sk
,
248 skb
, NULL
, skb_dst(skb
)->dev
, dst_output
);
249 if (likely(err
== 1))
250 err
= dst_output(net
, sk
, skb
);
255 static netdev_tx_t
vrf_process_v4_outbound(struct sk_buff
*skb
,
256 struct net_device
*vrf_dev
)
258 struct iphdr
*ip4h
= ip_hdr(skb
);
259 int ret
= NET_XMIT_DROP
;
260 struct flowi4 fl4
= {
261 /* needed to match OIF rule */
262 .flowi4_oif
= vrf_dev
->ifindex
,
263 .flowi4_iif
= LOOPBACK_IFINDEX
,
264 .flowi4_tos
= RT_TOS(ip4h
->tos
),
265 .flowi4_flags
= FLOWI_FLAG_ANYSRC
| FLOWI_FLAG_SKIP_NH_OIF
,
266 .flowi4_proto
= ip4h
->protocol
,
267 .daddr
= ip4h
->daddr
,
268 .saddr
= ip4h
->saddr
,
270 struct net
*net
= dev_net(vrf_dev
);
273 rt
= ip_route_output_flow(net
, &fl4
, NULL
);
279 /* if dst.dev is loopback or the VRF device again this is locally
280 * originated traffic destined to a local address. Short circuit
281 * to Rx path using our local dst
283 if (rt
->dst
.dev
== net
->loopback_dev
|| rt
->dst
.dev
== vrf_dev
) {
284 struct net_vrf
*vrf
= netdev_priv(vrf_dev
);
285 struct rtable
*rth_local
;
286 struct dst_entry
*dst
= NULL
;
292 rth_local
= rcu_dereference(vrf
->rth_local
);
293 if (likely(rth_local
)) {
294 dst
= &rth_local
->dst
;
303 return vrf_local_xmit(skb
, vrf_dev
, dst
);
306 skb_dst_set(skb
, &rt
->dst
);
308 /* strip the ethernet header added for pass through VRF device */
309 __skb_pull(skb
, skb_network_offset(skb
));
312 ip4h
->saddr
= inet_select_addr(skb_dst(skb
)->dev
, 0,
316 ret
= vrf_ip_local_out(dev_net(skb_dst(skb
)->dev
), skb
->sk
, skb
);
317 if (unlikely(net_xmit_eval(ret
)))
318 vrf_dev
->stats
.tx_errors
++;
320 ret
= NET_XMIT_SUCCESS
;
325 vrf_tx_error(vrf_dev
, skb
);
329 static netdev_tx_t
is_ip_tx_frame(struct sk_buff
*skb
, struct net_device
*dev
)
331 switch (skb
->protocol
) {
332 case htons(ETH_P_IP
):
333 return vrf_process_v4_outbound(skb
, dev
);
334 case htons(ETH_P_IPV6
):
335 return vrf_process_v6_outbound(skb
, dev
);
337 vrf_tx_error(dev
, skb
);
338 return NET_XMIT_DROP
;
342 static netdev_tx_t
vrf_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
345 netdev_tx_t ret
= is_ip_tx_frame(skb
, dev
);
347 if (likely(ret
== NET_XMIT_SUCCESS
|| ret
== NET_XMIT_CN
)) {
348 struct pcpu_dstats
*dstats
= this_cpu_ptr(dev
->dstats
);
350 u64_stats_update_begin(&dstats
->syncp
);
352 dstats
->tx_bytes
+= len
;
353 u64_stats_update_end(&dstats
->syncp
);
355 this_cpu_inc(dev
->dstats
->tx_drps
);
361 #if IS_ENABLED(CONFIG_IPV6)
362 /* modelled after ip6_finish_output2 */
363 static int vrf_finish_output6(struct net
*net
, struct sock
*sk
,
366 struct dst_entry
*dst
= skb_dst(skb
);
367 struct net_device
*dev
= dst
->dev
;
368 struct neighbour
*neigh
;
369 struct in6_addr
*nexthop
;
374 skb
->protocol
= htons(ETH_P_IPV6
);
378 nexthop
= rt6_nexthop((struct rt6_info
*)dst
, &ipv6_hdr(skb
)->daddr
);
379 neigh
= __ipv6_neigh_lookup_noref(dst
->dev
, nexthop
);
380 if (unlikely(!neigh
))
381 neigh
= __neigh_create(&nd_tbl
, nexthop
, dst
->dev
, false);
382 if (!IS_ERR(neigh
)) {
383 sock_confirm_neigh(skb
, neigh
);
384 ret
= dst_neigh_output(dst
, neigh
, skb
);
385 rcu_read_unlock_bh();
388 rcu_read_unlock_bh();
390 IP6_INC_STATS(dev_net(dst
->dev
),
391 ip6_dst_idev(dst
), IPSTATS_MIB_OUTNOROUTES
);
396 /* modelled after ip6_output */
397 static int vrf_output6(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
399 return NF_HOOK_COND(NFPROTO_IPV6
, NF_INET_POST_ROUTING
,
400 net
, sk
, skb
, NULL
, skb_dst(skb
)->dev
,
402 !(IP6CB(skb
)->flags
& IP6SKB_REROUTED
));
405 /* set dst on skb to send packet to us via dev_xmit path. Allows
406 * packet to go through device based features such as qdisc, netfilter
407 * hooks and packet sockets with skb->dev set to vrf device.
409 static struct sk_buff
*vrf_ip6_out(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
;
417 /* don't divert link scope packets */
418 if (rt6_need_strict(&ipv6_hdr(skb
)->daddr
))
423 rt6
= rcu_dereference(vrf
->rt6
);
431 if (unlikely(!dst
)) {
432 vrf_tx_error(vrf_dev
, skb
);
437 skb_dst_set(skb
, dst
);
443 static void vrf_rt6_release(struct net_device
*dev
, struct net_vrf
*vrf
)
445 struct rt6_info
*rt6
= rtnl_dereference(vrf
->rt6
);
446 struct rt6_info
*rt6_local
= rtnl_dereference(vrf
->rt6_local
);
447 struct net
*net
= dev_net(dev
);
448 struct dst_entry
*dst
;
450 RCU_INIT_POINTER(vrf
->rt6
, NULL
);
451 RCU_INIT_POINTER(vrf
->rt6_local
, NULL
);
454 /* move dev in dst's to loopback so this VRF device can be deleted
455 * - based on dst_ifdown
460 dst
->dev
= net
->loopback_dev
;
466 if (rt6_local
->rt6i_idev
) {
467 in6_dev_put(rt6_local
->rt6i_idev
);
468 rt6_local
->rt6i_idev
= NULL
;
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
;
560 /* Be paranoid, rather than too clever. */
561 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
562 struct sk_buff
*skb2
;
564 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
570 skb_set_owner_w(skb2
, skb
->sk
);
578 nexthop
= (__force u32
)rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
579 neigh
= __ipv4_neigh_lookup_noref(dev
, nexthop
);
580 if (unlikely(!neigh
))
581 neigh
= __neigh_create(&arp_tbl
, &nexthop
, dev
, false);
582 if (!IS_ERR(neigh
)) {
583 sock_confirm_neigh(skb
, neigh
);
584 ret
= dst_neigh_output(dst
, neigh
, skb
);
587 rcu_read_unlock_bh();
589 if (unlikely(ret
< 0))
590 vrf_tx_error(skb
->dev
, skb
);
594 static int vrf_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
596 struct net_device
*dev
= skb_dst(skb
)->dev
;
598 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUT
, skb
->len
);
601 skb
->protocol
= htons(ETH_P_IP
);
603 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
604 net
, sk
, skb
, NULL
, dev
,
606 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
609 /* set dst on skb to send packet to us via dev_xmit path. Allows
610 * packet to go through device based features such as qdisc, netfilter
611 * hooks and packet sockets with skb->dev set to vrf device.
613 static struct sk_buff
*vrf_ip_out(struct net_device
*vrf_dev
,
617 struct net_vrf
*vrf
= netdev_priv(vrf_dev
);
618 struct dst_entry
*dst
= NULL
;
621 /* don't divert multicast */
622 if (ipv4_is_multicast(ip_hdr(skb
)->daddr
))
627 rth
= rcu_dereference(vrf
->rth
);
635 if (unlikely(!dst
)) {
636 vrf_tx_error(vrf_dev
, skb
);
641 skb_dst_set(skb
, dst
);
646 /* called with rcu lock held */
647 static struct sk_buff
*vrf_l3_out(struct net_device
*vrf_dev
,
654 return vrf_ip_out(vrf_dev
, sk
, skb
);
656 return vrf_ip6_out(vrf_dev
, sk
, skb
);
663 static void vrf_rtable_release(struct net_device
*dev
, struct net_vrf
*vrf
)
665 struct rtable
*rth
= rtnl_dereference(vrf
->rth
);
666 struct rtable
*rth_local
= rtnl_dereference(vrf
->rth_local
);
667 struct net
*net
= dev_net(dev
);
668 struct dst_entry
*dst
;
670 RCU_INIT_POINTER(vrf
->rth
, NULL
);
671 RCU_INIT_POINTER(vrf
->rth_local
, NULL
);
674 /* move dev in dst's to loopback so this VRF device can be deleted
675 * - based on dst_ifdown
680 dst
->dev
= net
->loopback_dev
;
686 dst
= &rth_local
->dst
;
688 dst
->dev
= net
->loopback_dev
;
694 static int vrf_rtable_create(struct net_device
*dev
)
696 struct net_vrf
*vrf
= netdev_priv(dev
);
697 struct rtable
*rth
, *rth_local
;
699 if (!fib_new_table(dev_net(dev
), vrf
->tb_id
))
702 /* create a dst for routing packets out through a VRF device */
703 rth
= rt_dst_alloc(dev
, 0, RTN_UNICAST
, 1, 1, 0);
707 /* create a dst for local ingress routing - packets sent locally
708 * to local address via the VRF device as a loopback
710 rth_local
= rt_dst_alloc(dev
, RTCF_LOCAL
, RTN_LOCAL
, 1, 1, 0);
712 dst_release(&rth
->dst
);
716 rth
->dst
.output
= vrf_output
;
717 rth
->rt_table_id
= vrf
->tb_id
;
719 rth_local
->rt_table_id
= vrf
->tb_id
;
721 rcu_assign_pointer(vrf
->rth
, rth
);
722 rcu_assign_pointer(vrf
->rth_local
, rth_local
);
727 /**************************** device handling ********************/
729 /* cycle interface to flush neighbor cache and move routes across tables */
730 static void cycle_netdev(struct net_device
*dev
)
732 unsigned int flags
= dev
->flags
;
735 if (!netif_running(dev
))
738 ret
= dev_change_flags(dev
, flags
& ~IFF_UP
);
740 ret
= dev_change_flags(dev
, flags
);
744 "Failed to cycle device %s; route tables might be wrong!\n",
749 static int do_vrf_add_slave(struct net_device
*dev
, struct net_device
*port_dev
)
753 ret
= netdev_master_upper_dev_link(port_dev
, dev
, NULL
, NULL
);
757 port_dev
->priv_flags
|= IFF_L3MDEV_SLAVE
;
758 cycle_netdev(port_dev
);
763 static int vrf_add_slave(struct net_device
*dev
, struct net_device
*port_dev
)
765 if (netif_is_l3_master(port_dev
) || netif_is_l3_slave(port_dev
))
768 return do_vrf_add_slave(dev
, port_dev
);
771 /* inverse of do_vrf_add_slave */
772 static int do_vrf_del_slave(struct net_device
*dev
, struct net_device
*port_dev
)
774 netdev_upper_dev_unlink(port_dev
, dev
);
775 port_dev
->priv_flags
&= ~IFF_L3MDEV_SLAVE
;
777 cycle_netdev(port_dev
);
782 static int vrf_del_slave(struct net_device
*dev
, struct net_device
*port_dev
)
784 return do_vrf_del_slave(dev
, port_dev
);
787 static void vrf_dev_uninit(struct net_device
*dev
)
789 struct net_vrf
*vrf
= netdev_priv(dev
);
790 struct net_device
*port_dev
;
791 struct list_head
*iter
;
793 vrf_rtable_release(dev
, vrf
);
794 vrf_rt6_release(dev
, vrf
);
796 netdev_for_each_lower_dev(dev
, port_dev
, iter
)
797 vrf_del_slave(dev
, port_dev
);
799 free_percpu(dev
->dstats
);
803 static int vrf_dev_init(struct net_device
*dev
)
805 struct net_vrf
*vrf
= netdev_priv(dev
);
807 dev
->dstats
= netdev_alloc_pcpu_stats(struct pcpu_dstats
);
811 /* create the default dst which points back to us */
812 if (vrf_rtable_create(dev
) != 0)
815 if (vrf_rt6_create(dev
) != 0)
818 dev
->flags
= IFF_MASTER
| IFF_NOARP
;
820 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
821 dev
->mtu
= 64 * 1024;
823 /* similarly, oper state is irrelevant; set to up to avoid confusion */
824 dev
->operstate
= IF_OPER_UP
;
825 netdev_lockdep_set_classes(dev
);
829 vrf_rtable_release(dev
, vrf
);
831 free_percpu(dev
->dstats
);
837 static const struct net_device_ops vrf_netdev_ops
= {
838 .ndo_init
= vrf_dev_init
,
839 .ndo_uninit
= vrf_dev_uninit
,
840 .ndo_start_xmit
= vrf_xmit
,
841 .ndo_get_stats64
= vrf_get_stats64
,
842 .ndo_add_slave
= vrf_add_slave
,
843 .ndo_del_slave
= vrf_del_slave
,
846 static u32
vrf_fib_table(const struct net_device
*dev
)
848 struct net_vrf
*vrf
= netdev_priv(dev
);
853 static int vrf_rcv_finish(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
858 static struct sk_buff
*vrf_rcv_nfhook(u8 pf
, unsigned int hook
,
860 struct net_device
*dev
)
862 struct net
*net
= dev_net(dev
);
864 if (NF_HOOK(pf
, hook
, net
, NULL
, skb
, dev
, NULL
, vrf_rcv_finish
) < 0)
865 skb
= NULL
; /* kfree_skb(skb) handled by nf code */
870 #if IS_ENABLED(CONFIG_IPV6)
871 /* neighbor handling is done with actual device; do not want
872 * to flip skb->dev for those ndisc packets. This really fails
873 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
876 static bool ipv6_ndisc_frame(const struct sk_buff
*skb
)
878 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
881 if (iph
->nexthdr
== NEXTHDR_ICMP
) {
882 const struct icmp6hdr
*icmph
;
883 struct icmp6hdr _icmph
;
885 icmph
= skb_header_pointer(skb
, sizeof(*iph
),
886 sizeof(_icmph
), &_icmph
);
890 switch (icmph
->icmp6_type
) {
891 case NDISC_ROUTER_SOLICITATION
:
892 case NDISC_ROUTER_ADVERTISEMENT
:
893 case NDISC_NEIGHBOUR_SOLICITATION
:
894 case NDISC_NEIGHBOUR_ADVERTISEMENT
:
905 static struct rt6_info
*vrf_ip6_route_lookup(struct net
*net
,
906 const struct net_device
*dev
,
911 struct net_vrf
*vrf
= netdev_priv(dev
);
912 struct fib6_table
*table
= NULL
;
913 struct rt6_info
*rt6
;
917 /* fib6_table does not have a refcnt and can not be freed */
918 rt6
= rcu_dereference(vrf
->rt6
);
920 table
= rt6
->rt6i_table
;
927 return ip6_pol_route(net
, table
, ifindex
, fl6
, flags
);
930 static void vrf_ip6_input_dst(struct sk_buff
*skb
, struct net_device
*vrf_dev
,
933 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
934 struct flowi6 fl6
= {
937 .flowlabel
= ip6_flowinfo(iph
),
938 .flowi6_mark
= skb
->mark
,
939 .flowi6_proto
= iph
->nexthdr
,
940 .flowi6_iif
= ifindex
,
942 struct net
*net
= dev_net(vrf_dev
);
943 struct rt6_info
*rt6
;
945 rt6
= vrf_ip6_route_lookup(net
, vrf_dev
, &fl6
, ifindex
,
946 RT6_LOOKUP_F_HAS_SADDR
| RT6_LOOKUP_F_IFACE
);
950 if (unlikely(&rt6
->dst
== &net
->ipv6
.ip6_null_entry
->dst
))
953 skb_dst_set(skb
, &rt6
->dst
);
956 static struct sk_buff
*vrf_ip6_rcv(struct net_device
*vrf_dev
,
959 int orig_iif
= skb
->skb_iif
;
962 /* loopback traffic; do not push through packet taps again.
963 * Reset pkt_type for upper layers to process skb
965 if (skb
->pkt_type
== PACKET_LOOPBACK
) {
967 skb
->skb_iif
= vrf_dev
->ifindex
;
968 IP6CB(skb
)->flags
|= IP6SKB_L3SLAVE
;
969 skb
->pkt_type
= PACKET_HOST
;
973 /* if packet is NDISC or addressed to multicast or link-local
974 * then keep the ingress interface
976 need_strict
= rt6_need_strict(&ipv6_hdr(skb
)->daddr
);
977 if (!ipv6_ndisc_frame(skb
) && !need_strict
) {
978 vrf_rx_stats(vrf_dev
, skb
->len
);
980 skb
->skb_iif
= vrf_dev
->ifindex
;
982 skb_push(skb
, skb
->mac_len
);
983 dev_queue_xmit_nit(skb
, vrf_dev
);
984 skb_pull(skb
, skb
->mac_len
);
986 IP6CB(skb
)->flags
|= IP6SKB_L3SLAVE
;
990 vrf_ip6_input_dst(skb
, vrf_dev
, orig_iif
);
992 skb
= vrf_rcv_nfhook(NFPROTO_IPV6
, NF_INET_PRE_ROUTING
, skb
, vrf_dev
);
998 static struct sk_buff
*vrf_ip6_rcv(struct net_device
*vrf_dev
,
1005 static struct sk_buff
*vrf_ip_rcv(struct net_device
*vrf_dev
,
1006 struct sk_buff
*skb
)
1009 skb
->skb_iif
= vrf_dev
->ifindex
;
1010 IPCB(skb
)->flags
|= IPSKB_L3SLAVE
;
1012 if (ipv4_is_multicast(ip_hdr(skb
)->daddr
))
1015 /* loopback traffic; do not push through packet taps again.
1016 * Reset pkt_type for upper layers to process skb
1018 if (skb
->pkt_type
== PACKET_LOOPBACK
) {
1019 skb
->pkt_type
= PACKET_HOST
;
1023 vrf_rx_stats(vrf_dev
, skb
->len
);
1025 skb_push(skb
, skb
->mac_len
);
1026 dev_queue_xmit_nit(skb
, vrf_dev
);
1027 skb_pull(skb
, skb
->mac_len
);
1029 skb
= vrf_rcv_nfhook(NFPROTO_IPV4
, NF_INET_PRE_ROUTING
, skb
, vrf_dev
);
1034 /* called with rcu lock held */
1035 static struct sk_buff
*vrf_l3_rcv(struct net_device
*vrf_dev
,
1036 struct sk_buff
*skb
,
1041 return vrf_ip_rcv(vrf_dev
, skb
);
1043 return vrf_ip6_rcv(vrf_dev
, skb
);
1049 #if IS_ENABLED(CONFIG_IPV6)
1050 /* send to link-local or multicast address via interface enslaved to
1051 * VRF device. Force lookup to VRF table without changing flow struct
1053 static struct dst_entry
*vrf_link_scope_lookup(const struct net_device
*dev
,
1056 struct net
*net
= dev_net(dev
);
1057 int flags
= RT6_LOOKUP_F_IFACE
;
1058 struct dst_entry
*dst
= NULL
;
1059 struct rt6_info
*rt
;
1061 /* VRF device does not have a link-local address and
1062 * sending packets to link-local or mcast addresses over
1063 * a VRF device does not make sense
1065 if (fl6
->flowi6_oif
== dev
->ifindex
) {
1066 dst
= &net
->ipv6
.ip6_null_entry
->dst
;
1071 if (!ipv6_addr_any(&fl6
->saddr
))
1072 flags
|= RT6_LOOKUP_F_HAS_SADDR
;
1074 rt
= vrf_ip6_route_lookup(net
, dev
, fl6
, fl6
->flowi6_oif
, flags
);
1082 static const struct l3mdev_ops vrf_l3mdev_ops
= {
1083 .l3mdev_fib_table
= vrf_fib_table
,
1084 .l3mdev_l3_rcv
= vrf_l3_rcv
,
1085 .l3mdev_l3_out
= vrf_l3_out
,
1086 #if IS_ENABLED(CONFIG_IPV6)
1087 .l3mdev_link_scope_lookup
= vrf_link_scope_lookup
,
1091 static void vrf_get_drvinfo(struct net_device
*dev
,
1092 struct ethtool_drvinfo
*info
)
1094 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
1095 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
1098 static const struct ethtool_ops vrf_ethtool_ops
= {
1099 .get_drvinfo
= vrf_get_drvinfo
,
1102 static inline size_t vrf_fib_rule_nl_size(void)
1106 sz
= NLMSG_ALIGN(sizeof(struct fib_rule_hdr
));
1107 sz
+= nla_total_size(sizeof(u8
)); /* FRA_L3MDEV */
1108 sz
+= nla_total_size(sizeof(u32
)); /* FRA_PRIORITY */
1113 static int vrf_fib_rule(const struct net_device
*dev
, __u8 family
, bool add_it
)
1115 struct fib_rule_hdr
*frh
;
1116 struct nlmsghdr
*nlh
;
1117 struct sk_buff
*skb
;
1120 if (family
== AF_INET6
&& !ipv6_mod_enabled())
1123 skb
= nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL
);
1127 nlh
= nlmsg_put(skb
, 0, 0, 0, sizeof(*frh
), 0);
1129 goto nla_put_failure
;
1131 /* rule only needs to appear once */
1132 nlh
->nlmsg_flags
|= NLM_F_EXCL
;
1134 frh
= nlmsg_data(nlh
);
1135 memset(frh
, 0, sizeof(*frh
));
1136 frh
->family
= family
;
1137 frh
->action
= FR_ACT_TO_TBL
;
1139 if (nla_put_u32(skb
, FRA_L3MDEV
, 1))
1140 goto nla_put_failure
;
1142 if (nla_put_u32(skb
, FRA_PRIORITY
, FIB_RULE_PREF
))
1143 goto nla_put_failure
;
1145 nlmsg_end(skb
, nlh
);
1147 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1148 skb
->sk
= dev_net(dev
)->rtnl
;
1150 err
= fib_nl_newrule(skb
, nlh
);
1154 err
= fib_nl_delrule(skb
, nlh
);
1168 static int vrf_add_fib_rules(const struct net_device
*dev
)
1172 err
= vrf_fib_rule(dev
, AF_INET
, true);
1176 err
= vrf_fib_rule(dev
, AF_INET6
, true);
1180 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1181 err
= vrf_fib_rule(dev
, RTNL_FAMILY_IPMR
, true);
1188 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1190 vrf_fib_rule(dev
, AF_INET6
, false);
1194 vrf_fib_rule(dev
, AF_INET
, false);
1197 netdev_err(dev
, "Failed to add FIB rules.\n");
1201 static void vrf_setup(struct net_device
*dev
)
1205 /* Initialize the device structure. */
1206 dev
->netdev_ops
= &vrf_netdev_ops
;
1207 dev
->l3mdev_ops
= &vrf_l3mdev_ops
;
1208 dev
->ethtool_ops
= &vrf_ethtool_ops
;
1209 dev
->destructor
= free_netdev
;
1211 /* Fill in device structure with ethernet-generic values. */
1212 eth_hw_addr_random(dev
);
1214 /* don't acquire vrf device's netif_tx_lock when transmitting */
1215 dev
->features
|= NETIF_F_LLTX
;
1217 /* don't allow vrf devices to change network namespaces. */
1218 dev
->features
|= NETIF_F_NETNS_LOCAL
;
1220 /* does not make sense for a VLAN to be added to a vrf device */
1221 dev
->features
|= NETIF_F_VLAN_CHALLENGED
;
1223 /* enable offload features */
1224 dev
->features
|= NETIF_F_GSO_SOFTWARE
;
1225 dev
->features
|= NETIF_F_RXCSUM
| NETIF_F_HW_CSUM
;
1226 dev
->features
|= NETIF_F_SG
| NETIF_F_FRAGLIST
| NETIF_F_HIGHDMA
;
1228 dev
->hw_features
= dev
->features
;
1229 dev
->hw_enc_features
= dev
->features
;
1231 /* default to no qdisc; user can add if desired */
1232 dev
->priv_flags
|= IFF_NO_QUEUE
;
1235 static int vrf_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
1237 if (tb
[IFLA_ADDRESS
]) {
1238 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
1240 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
1241 return -EADDRNOTAVAIL
;
1246 static void vrf_dellink(struct net_device
*dev
, struct list_head
*head
)
1248 unregister_netdevice_queue(dev
, head
);
1251 static int vrf_newlink(struct net
*src_net
, struct net_device
*dev
,
1252 struct nlattr
*tb
[], struct nlattr
*data
[])
1254 struct net_vrf
*vrf
= netdev_priv(dev
);
1257 if (!data
|| !data
[IFLA_VRF_TABLE
])
1260 vrf
->tb_id
= nla_get_u32(data
[IFLA_VRF_TABLE
]);
1261 if (vrf
->tb_id
== RT_TABLE_UNSPEC
)
1264 dev
->priv_flags
|= IFF_L3MDEV_MASTER
;
1266 err
= register_netdevice(dev
);
1270 if (add_fib_rules
) {
1271 err
= vrf_add_fib_rules(dev
);
1273 unregister_netdevice(dev
);
1276 add_fib_rules
= false;
1283 static size_t vrf_nl_getsize(const struct net_device
*dev
)
1285 return nla_total_size(sizeof(u32
)); /* IFLA_VRF_TABLE */
1288 static int vrf_fillinfo(struct sk_buff
*skb
,
1289 const struct net_device
*dev
)
1291 struct net_vrf
*vrf
= netdev_priv(dev
);
1293 return nla_put_u32(skb
, IFLA_VRF_TABLE
, vrf
->tb_id
);
1296 static size_t vrf_get_slave_size(const struct net_device
*bond_dev
,
1297 const struct net_device
*slave_dev
)
1299 return nla_total_size(sizeof(u32
)); /* IFLA_VRF_PORT_TABLE */
1302 static int vrf_fill_slave_info(struct sk_buff
*skb
,
1303 const struct net_device
*vrf_dev
,
1304 const struct net_device
*slave_dev
)
1306 struct net_vrf
*vrf
= netdev_priv(vrf_dev
);
1308 if (nla_put_u32(skb
, IFLA_VRF_PORT_TABLE
, vrf
->tb_id
))
1314 static const struct nla_policy vrf_nl_policy
[IFLA_VRF_MAX
+ 1] = {
1315 [IFLA_VRF_TABLE
] = { .type
= NLA_U32
},
1318 static struct rtnl_link_ops vrf_link_ops __read_mostly
= {
1320 .priv_size
= sizeof(struct net_vrf
),
1322 .get_size
= vrf_nl_getsize
,
1323 .policy
= vrf_nl_policy
,
1324 .validate
= vrf_validate
,
1325 .fill_info
= vrf_fillinfo
,
1327 .get_slave_size
= vrf_get_slave_size
,
1328 .fill_slave_info
= vrf_fill_slave_info
,
1330 .newlink
= vrf_newlink
,
1331 .dellink
= vrf_dellink
,
1333 .maxtype
= IFLA_VRF_MAX
,
1336 static int vrf_device_event(struct notifier_block
*unused
,
1337 unsigned long event
, void *ptr
)
1339 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
1341 /* only care about unregister events to drop slave references */
1342 if (event
== NETDEV_UNREGISTER
) {
1343 struct net_device
*vrf_dev
;
1345 if (!netif_is_l3_slave(dev
))
1348 vrf_dev
= netdev_master_upper_dev_get(dev
);
1349 vrf_del_slave(vrf_dev
, dev
);
1355 static struct notifier_block vrf_notifier_block __read_mostly
= {
1356 .notifier_call
= vrf_device_event
,
1359 static int __init
vrf_init_module(void)
1363 register_netdevice_notifier(&vrf_notifier_block
);
1365 rc
= rtnl_link_register(&vrf_link_ops
);
1372 unregister_netdevice_notifier(&vrf_notifier_block
);
1376 module_init(vrf_init_module
);
1377 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1378 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1379 MODULE_LICENSE("GPL");
1380 MODULE_ALIAS_RTNL_LINK(DRV_NAME
);
1381 MODULE_VERSION(DRV_VERSION
);