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/rtnetlink.h>
36 #include <net/route.h>
37 #include <net/addrconf.h>
38 #include <net/l3mdev.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 vrf_master_get_rcu(dev) \
47 ((struct net_device *)rcu_dereference(dev->rx_handler_data))
61 struct u64_stats_sync syncp
;
64 static struct dst_entry
*vrf_ip_check(struct dst_entry
*dst
, u32 cookie
)
69 static int vrf_ip_local_out(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
71 return ip_local_out(net
, sk
, skb
);
74 static unsigned int vrf_v4_mtu(const struct dst_entry
*dst
)
76 /* TO-DO: return max ethernet size? */
80 static void vrf_dst_destroy(struct dst_entry
*dst
)
82 /* our dst lives forever - or until the device is closed */
85 static unsigned int vrf_default_advmss(const struct dst_entry
*dst
)
90 static struct dst_ops vrf_dst_ops
= {
92 .local_out
= vrf_ip_local_out
,
93 .check
= vrf_ip_check
,
95 .destroy
= vrf_dst_destroy
,
96 .default_advmss
= vrf_default_advmss
,
99 /* neighbor handling is done with actual device; do not want
100 * to flip skb->dev for those ndisc packets. This really fails
101 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
104 #if IS_ENABLED(CONFIG_IPV6)
105 static bool check_ipv6_frame(const struct sk_buff
*skb
)
107 const struct ipv6hdr
*ipv6h
;
108 struct ipv6hdr _ipv6h
;
111 ipv6h
= skb_header_pointer(skb
, 0, sizeof(_ipv6h
), &_ipv6h
);
115 if (ipv6h
->nexthdr
== NEXTHDR_ICMP
) {
116 const struct icmp6hdr
*icmph
;
117 struct icmp6hdr _icmph
;
119 icmph
= skb_header_pointer(skb
, sizeof(_ipv6h
),
120 sizeof(_icmph
), &_icmph
);
124 switch (icmph
->icmp6_type
) {
125 case NDISC_ROUTER_SOLICITATION
:
126 case NDISC_ROUTER_ADVERTISEMENT
:
127 case NDISC_NEIGHBOUR_SOLICITATION
:
128 case NDISC_NEIGHBOUR_ADVERTISEMENT
:
139 static bool check_ipv6_frame(const struct sk_buff
*skb
)
145 static bool is_ip_rx_frame(struct sk_buff
*skb
)
147 switch (skb
->protocol
) {
148 case htons(ETH_P_IP
):
150 case htons(ETH_P_IPV6
):
151 return check_ipv6_frame(skb
);
156 static void vrf_tx_error(struct net_device
*vrf_dev
, struct sk_buff
*skb
)
158 vrf_dev
->stats
.tx_errors
++;
162 /* note: already called with rcu_read_lock */
163 static rx_handler_result_t
vrf_handle_frame(struct sk_buff
**pskb
)
165 struct sk_buff
*skb
= *pskb
;
167 if (is_ip_rx_frame(skb
)) {
168 struct net_device
*dev
= vrf_master_get_rcu(skb
->dev
);
169 struct pcpu_dstats
*dstats
= this_cpu_ptr(dev
->dstats
);
171 u64_stats_update_begin(&dstats
->syncp
);
173 dstats
->rx_bytes
+= skb
->len
;
174 u64_stats_update_end(&dstats
->syncp
);
178 return RX_HANDLER_ANOTHER
;
180 return RX_HANDLER_PASS
;
183 static struct rtnl_link_stats64
*vrf_get_stats64(struct net_device
*dev
,
184 struct rtnl_link_stats64
*stats
)
188 for_each_possible_cpu(i
) {
189 const struct pcpu_dstats
*dstats
;
190 u64 tbytes
, tpkts
, tdrops
, rbytes
, rpkts
;
193 dstats
= per_cpu_ptr(dev
->dstats
, i
);
195 start
= u64_stats_fetch_begin_irq(&dstats
->syncp
);
196 tbytes
= dstats
->tx_bytes
;
197 tpkts
= dstats
->tx_pkts
;
198 tdrops
= dstats
->tx_drps
;
199 rbytes
= dstats
->rx_bytes
;
200 rpkts
= dstats
->rx_pkts
;
201 } while (u64_stats_fetch_retry_irq(&dstats
->syncp
, start
));
202 stats
->tx_bytes
+= tbytes
;
203 stats
->tx_packets
+= tpkts
;
204 stats
->tx_dropped
+= tdrops
;
205 stats
->rx_bytes
+= rbytes
;
206 stats
->rx_packets
+= rpkts
;
211 #if IS_ENABLED(CONFIG_IPV6)
212 static netdev_tx_t
vrf_process_v6_outbound(struct sk_buff
*skb
,
213 struct net_device
*dev
)
215 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
216 struct net
*net
= dev_net(skb
->dev
);
217 struct flowi6 fl6
= {
218 /* needed to match OIF rule */
219 .flowi6_oif
= dev
->ifindex
,
220 .flowi6_iif
= LOOPBACK_IFINDEX
,
223 .flowlabel
= ip6_flowinfo(iph
),
224 .flowi6_mark
= skb
->mark
,
225 .flowi6_proto
= iph
->nexthdr
,
226 .flowi6_flags
= FLOWI_FLAG_L3MDEV_SRC
| FLOWI_FLAG_SKIP_NH_OIF
,
228 int ret
= NET_XMIT_DROP
;
229 struct dst_entry
*dst
;
230 struct dst_entry
*dst_null
= &net
->ipv6
.ip6_null_entry
->dst
;
232 dst
= ip6_route_output(net
, NULL
, &fl6
);
237 skb_dst_set(skb
, dst
);
239 ret
= 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 static int vrf_send_v4_prep(struct sk_buff
*skb
, struct flowi4
*fl4
,
260 struct net_device
*vrf_dev
)
265 rt
= ip_route_output_flow(dev_net(vrf_dev
), fl4
, NULL
);
269 /* TO-DO: what about broadcast ? */
270 if (rt
->rt_type
!= RTN_UNICAST
&& rt
->rt_type
!= RTN_LOCAL
) {
276 skb_dst_set(skb
, &rt
->dst
);
282 static netdev_tx_t
vrf_process_v4_outbound(struct sk_buff
*skb
,
283 struct net_device
*vrf_dev
)
285 struct iphdr
*ip4h
= ip_hdr(skb
);
286 int ret
= NET_XMIT_DROP
;
287 struct flowi4 fl4
= {
288 /* needed to match OIF rule */
289 .flowi4_oif
= vrf_dev
->ifindex
,
290 .flowi4_iif
= LOOPBACK_IFINDEX
,
291 .flowi4_tos
= RT_TOS(ip4h
->tos
),
292 .flowi4_flags
= FLOWI_FLAG_ANYSRC
| FLOWI_FLAG_L3MDEV_SRC
|
293 FLOWI_FLAG_SKIP_NH_OIF
,
294 .daddr
= ip4h
->daddr
,
297 if (vrf_send_v4_prep(skb
, &fl4
, vrf_dev
))
301 ip4h
->saddr
= inet_select_addr(skb_dst(skb
)->dev
, 0,
305 ret
= ip_local_out(dev_net(skb_dst(skb
)->dev
), skb
->sk
, skb
);
306 if (unlikely(net_xmit_eval(ret
)))
307 vrf_dev
->stats
.tx_errors
++;
309 ret
= NET_XMIT_SUCCESS
;
314 vrf_tx_error(vrf_dev
, skb
);
318 static netdev_tx_t
is_ip_tx_frame(struct sk_buff
*skb
, struct net_device
*dev
)
320 /* strip the ethernet header added for pass through VRF device */
321 __skb_pull(skb
, skb_network_offset(skb
));
323 switch (skb
->protocol
) {
324 case htons(ETH_P_IP
):
325 return vrf_process_v4_outbound(skb
, dev
);
326 case htons(ETH_P_IPV6
):
327 return vrf_process_v6_outbound(skb
, dev
);
329 vrf_tx_error(dev
, skb
);
330 return NET_XMIT_DROP
;
334 static netdev_tx_t
vrf_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
336 netdev_tx_t ret
= is_ip_tx_frame(skb
, dev
);
338 if (likely(ret
== NET_XMIT_SUCCESS
|| ret
== NET_XMIT_CN
)) {
339 struct pcpu_dstats
*dstats
= this_cpu_ptr(dev
->dstats
);
341 u64_stats_update_begin(&dstats
->syncp
);
343 dstats
->tx_bytes
+= skb
->len
;
344 u64_stats_update_end(&dstats
->syncp
);
346 this_cpu_inc(dev
->dstats
->tx_drps
);
352 #if IS_ENABLED(CONFIG_IPV6)
353 static struct dst_entry
*vrf_ip6_check(struct dst_entry
*dst
, u32 cookie
)
358 static struct dst_ops vrf_dst_ops6
= {
360 .local_out
= ip6_local_out
,
361 .check
= vrf_ip6_check
,
363 .destroy
= vrf_dst_destroy
,
364 .default_advmss
= vrf_default_advmss
,
367 static int init_dst_ops6_kmem_cachep(void)
369 vrf_dst_ops6
.kmem_cachep
= kmem_cache_create("vrf_ip6_dst_cache",
370 sizeof(struct rt6_info
),
375 if (!vrf_dst_ops6
.kmem_cachep
)
381 static void free_dst_ops6_kmem_cachep(void)
383 kmem_cache_destroy(vrf_dst_ops6
.kmem_cachep
);
386 static int vrf_input6(struct sk_buff
*skb
)
388 skb
->dev
->stats
.rx_errors
++;
393 /* modelled after ip6_finish_output2 */
394 static int vrf_finish_output6(struct net
*net
, struct sock
*sk
,
397 struct dst_entry
*dst
= skb_dst(skb
);
398 struct net_device
*dev
= dst
->dev
;
399 struct neighbour
*neigh
;
400 struct in6_addr
*nexthop
;
403 skb
->protocol
= htons(ETH_P_IPV6
);
407 nexthop
= rt6_nexthop((struct rt6_info
*)dst
, &ipv6_hdr(skb
)->daddr
);
408 neigh
= __ipv6_neigh_lookup_noref(dst
->dev
, nexthop
);
409 if (unlikely(!neigh
))
410 neigh
= __neigh_create(&nd_tbl
, nexthop
, dst
->dev
, false);
411 if (!IS_ERR(neigh
)) {
412 ret
= dst_neigh_output(dst
, neigh
, skb
);
413 rcu_read_unlock_bh();
416 rcu_read_unlock_bh();
418 IP6_INC_STATS(dev_net(dst
->dev
),
419 ip6_dst_idev(dst
), IPSTATS_MIB_OUTNOROUTES
);
424 /* modelled after ip6_output */
425 static int vrf_output6(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
427 return NF_HOOK_COND(NFPROTO_IPV6
, NF_INET_POST_ROUTING
,
428 net
, sk
, skb
, NULL
, skb_dst(skb
)->dev
,
430 !(IP6CB(skb
)->flags
& IP6SKB_REROUTED
));
433 static void vrf_rt6_destroy(struct net_vrf
*vrf
)
435 dst_destroy(&vrf
->rt6
->dst
);
436 free_percpu(vrf
->rt6
->rt6i_pcpu
);
440 static int vrf_rt6_create(struct net_device
*dev
)
442 struct net_vrf
*vrf
= netdev_priv(dev
);
443 struct dst_entry
*dst
;
444 struct rt6_info
*rt6
;
448 rt6
= dst_alloc(&vrf_dst_ops6
, dev
, 0,
450 (DST_HOST
| DST_NOPOLICY
| DST_NOXFRM
));
456 rt6
->rt6i_pcpu
= alloc_percpu_gfp(struct rt6_info
*, GFP_KERNEL
);
457 if (!rt6
->rt6i_pcpu
) {
461 for_each_possible_cpu(cpu
) {
462 struct rt6_info
**p
= per_cpu_ptr(rt6
->rt6i_pcpu
, cpu
);
466 memset(dst
+ 1, 0, sizeof(*rt6
) - sizeof(*dst
));
468 INIT_LIST_HEAD(&rt6
->rt6i_siblings
);
469 INIT_LIST_HEAD(&rt6
->rt6i_uncached
);
471 rt6
->dst
.input
= vrf_input6
;
472 rt6
->dst
.output
= vrf_output6
;
474 rt6
->rt6i_table
= fib6_get_table(dev_net(dev
), vrf
->tb_id
);
476 atomic_set(&rt6
->dst
.__refcnt
, 2);
484 static int init_dst_ops6_kmem_cachep(void)
489 static void free_dst_ops6_kmem_cachep(void)
493 static void vrf_rt6_destroy(struct net_vrf
*vrf
)
497 static int vrf_rt6_create(struct net_device
*dev
)
503 /* modelled after ip_finish_output2 */
504 static int vrf_finish_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
506 struct dst_entry
*dst
= skb_dst(skb
);
507 struct rtable
*rt
= (struct rtable
*)dst
;
508 struct net_device
*dev
= dst
->dev
;
509 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
510 struct neighbour
*neigh
;
514 /* Be paranoid, rather than too clever. */
515 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
516 struct sk_buff
*skb2
;
518 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
524 skb_set_owner_w(skb2
, skb
->sk
);
532 nexthop
= (__force u32
)rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
533 neigh
= __ipv4_neigh_lookup_noref(dev
, nexthop
);
534 if (unlikely(!neigh
))
535 neigh
= __neigh_create(&arp_tbl
, &nexthop
, dev
, false);
537 ret
= dst_neigh_output(dst
, neigh
, skb
);
539 rcu_read_unlock_bh();
541 if (unlikely(ret
< 0))
542 vrf_tx_error(skb
->dev
, skb
);
546 static int vrf_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
548 struct net_device
*dev
= skb_dst(skb
)->dev
;
550 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUT
, skb
->len
);
553 skb
->protocol
= htons(ETH_P_IP
);
555 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
556 net
, sk
, skb
, NULL
, dev
,
558 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
561 static void vrf_rtable_destroy(struct net_vrf
*vrf
)
563 struct dst_entry
*dst
= (struct dst_entry
*)vrf
->rth
;
569 static struct rtable
*vrf_rtable_create(struct net_device
*dev
)
571 struct net_vrf
*vrf
= netdev_priv(dev
);
574 rth
= dst_alloc(&vrf_dst_ops
, dev
, 2,
576 (DST_HOST
| DST_NOPOLICY
| DST_NOXFRM
));
578 rth
->dst
.output
= vrf_output
;
579 rth
->rt_genid
= rt_genid_ipv4(dev_net(dev
));
581 rth
->rt_type
= RTN_UNICAST
;
582 rth
->rt_is_input
= 0;
586 rth
->rt_uses_gateway
= 0;
587 rth
->rt_table_id
= vrf
->tb_id
;
588 INIT_LIST_HEAD(&rth
->rt_uncached
);
589 rth
->rt_uncached_list
= NULL
;
595 /**************************** device handling ********************/
597 /* cycle interface to flush neighbor cache and move routes across tables */
598 static void cycle_netdev(struct net_device
*dev
)
600 unsigned int flags
= dev
->flags
;
603 if (!netif_running(dev
))
606 ret
= dev_change_flags(dev
, flags
& ~IFF_UP
);
608 ret
= dev_change_flags(dev
, flags
);
612 "Failed to cycle device %s; route tables might be wrong!\n",
617 static int do_vrf_add_slave(struct net_device
*dev
, struct net_device
*port_dev
)
621 /* register the packet handler for slave ports */
622 ret
= netdev_rx_handler_register(port_dev
, vrf_handle_frame
, dev
);
625 "Device %s failed to register rx_handler\n",
630 ret
= netdev_master_upper_dev_link(port_dev
, dev
, NULL
, NULL
);
634 port_dev
->priv_flags
|= IFF_L3MDEV_SLAVE
;
635 cycle_netdev(port_dev
);
640 netdev_rx_handler_unregister(port_dev
);
645 static int vrf_add_slave(struct net_device
*dev
, struct net_device
*port_dev
)
647 if (netif_is_l3_master(port_dev
) || netif_is_l3_slave(port_dev
))
650 return do_vrf_add_slave(dev
, port_dev
);
653 /* inverse of do_vrf_add_slave */
654 static int do_vrf_del_slave(struct net_device
*dev
, struct net_device
*port_dev
)
656 netdev_upper_dev_unlink(port_dev
, dev
);
657 port_dev
->priv_flags
&= ~IFF_L3MDEV_SLAVE
;
659 netdev_rx_handler_unregister(port_dev
);
661 cycle_netdev(port_dev
);
666 static int vrf_del_slave(struct net_device
*dev
, struct net_device
*port_dev
)
668 return do_vrf_del_slave(dev
, port_dev
);
671 static void vrf_dev_uninit(struct net_device
*dev
)
673 struct net_vrf
*vrf
= netdev_priv(dev
);
674 struct net_device
*port_dev
;
675 struct list_head
*iter
;
677 vrf_rtable_destroy(vrf
);
678 vrf_rt6_destroy(vrf
);
680 netdev_for_each_lower_dev(dev
, port_dev
, iter
)
681 vrf_del_slave(dev
, port_dev
);
683 free_percpu(dev
->dstats
);
687 static int vrf_dev_init(struct net_device
*dev
)
689 struct net_vrf
*vrf
= netdev_priv(dev
);
691 dev
->dstats
= netdev_alloc_pcpu_stats(struct pcpu_dstats
);
695 /* create the default dst which points back to us */
696 vrf
->rth
= vrf_rtable_create(dev
);
700 if (vrf_rt6_create(dev
) != 0)
703 dev
->flags
= IFF_MASTER
| IFF_NOARP
;
708 vrf_rtable_destroy(vrf
);
710 free_percpu(dev
->dstats
);
716 static const struct net_device_ops vrf_netdev_ops
= {
717 .ndo_init
= vrf_dev_init
,
718 .ndo_uninit
= vrf_dev_uninit
,
719 .ndo_start_xmit
= vrf_xmit
,
720 .ndo_get_stats64
= vrf_get_stats64
,
721 .ndo_add_slave
= vrf_add_slave
,
722 .ndo_del_slave
= vrf_del_slave
,
725 static u32
vrf_fib_table(const struct net_device
*dev
)
727 struct net_vrf
*vrf
= netdev_priv(dev
);
732 static struct rtable
*vrf_get_rtable(const struct net_device
*dev
,
733 const struct flowi4
*fl4
)
735 struct rtable
*rth
= NULL
;
737 if (!(fl4
->flowi4_flags
& FLOWI_FLAG_L3MDEV_SRC
)) {
738 struct net_vrf
*vrf
= netdev_priv(dev
);
741 atomic_inc(&rth
->dst
.__refcnt
);
747 /* called under rcu_read_lock */
748 static int vrf_get_saddr(struct net_device
*dev
, struct flowi4
*fl4
)
750 struct fib_result res
= { .tclassid
= 0 };
751 struct net
*net
= dev_net(dev
);
752 u32 orig_tos
= fl4
->flowi4_tos
;
753 u8 flags
= fl4
->flowi4_flags
;
754 u8 scope
= fl4
->flowi4_scope
;
755 u8 tos
= RT_FL_TOS(fl4
);
758 if (unlikely(!fl4
->daddr
))
761 fl4
->flowi4_flags
|= FLOWI_FLAG_SKIP_NH_OIF
;
762 fl4
->flowi4_iif
= LOOPBACK_IFINDEX
;
763 fl4
->flowi4_tos
= tos
& IPTOS_RT_MASK
;
764 fl4
->flowi4_scope
= ((tos
& RTO_ONLINK
) ?
765 RT_SCOPE_LINK
: RT_SCOPE_UNIVERSE
);
767 rc
= fib_lookup(net
, fl4
, &res
, 0);
769 if (res
.type
== RTN_LOCAL
)
770 fl4
->saddr
= res
.fi
->fib_prefsrc
? : fl4
->daddr
;
772 fib_select_path(net
, &res
, fl4
, -1);
775 fl4
->flowi4_flags
= flags
;
776 fl4
->flowi4_tos
= orig_tos
;
777 fl4
->flowi4_scope
= scope
;
782 #if IS_ENABLED(CONFIG_IPV6)
783 static struct dst_entry
*vrf_get_rt6_dst(const struct net_device
*dev
,
784 const struct flowi6
*fl6
)
786 struct rt6_info
*rt
= NULL
;
788 if (!(fl6
->flowi6_flags
& FLOWI_FLAG_L3MDEV_SRC
)) {
789 struct net_vrf
*vrf
= netdev_priv(dev
);
792 atomic_inc(&rt
->dst
.__refcnt
);
795 return (struct dst_entry
*)rt
;
799 static const struct l3mdev_ops vrf_l3mdev_ops
= {
800 .l3mdev_fib_table
= vrf_fib_table
,
801 .l3mdev_get_rtable
= vrf_get_rtable
,
802 .l3mdev_get_saddr
= vrf_get_saddr
,
803 #if IS_ENABLED(CONFIG_IPV6)
804 .l3mdev_get_rt6_dst
= vrf_get_rt6_dst
,
808 static void vrf_get_drvinfo(struct net_device
*dev
,
809 struct ethtool_drvinfo
*info
)
811 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
812 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
815 static const struct ethtool_ops vrf_ethtool_ops
= {
816 .get_drvinfo
= vrf_get_drvinfo
,
819 static void vrf_setup(struct net_device
*dev
)
823 /* Initialize the device structure. */
824 dev
->netdev_ops
= &vrf_netdev_ops
;
825 dev
->l3mdev_ops
= &vrf_l3mdev_ops
;
826 dev
->ethtool_ops
= &vrf_ethtool_ops
;
827 dev
->destructor
= free_netdev
;
829 /* Fill in device structure with ethernet-generic values. */
830 eth_hw_addr_random(dev
);
832 /* don't acquire vrf device's netif_tx_lock when transmitting */
833 dev
->features
|= NETIF_F_LLTX
;
835 /* don't allow vrf devices to change network namespaces. */
836 dev
->features
|= NETIF_F_NETNS_LOCAL
;
839 static int vrf_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
841 if (tb
[IFLA_ADDRESS
]) {
842 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
844 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
845 return -EADDRNOTAVAIL
;
850 static void vrf_dellink(struct net_device
*dev
, struct list_head
*head
)
852 unregister_netdevice_queue(dev
, head
);
855 static int vrf_newlink(struct net
*src_net
, struct net_device
*dev
,
856 struct nlattr
*tb
[], struct nlattr
*data
[])
858 struct net_vrf
*vrf
= netdev_priv(dev
);
860 if (!data
|| !data
[IFLA_VRF_TABLE
])
863 vrf
->tb_id
= nla_get_u32(data
[IFLA_VRF_TABLE
]);
865 dev
->priv_flags
|= IFF_L3MDEV_MASTER
;
867 return register_netdevice(dev
);
870 static size_t vrf_nl_getsize(const struct net_device
*dev
)
872 return nla_total_size(sizeof(u32
)); /* IFLA_VRF_TABLE */
875 static int vrf_fillinfo(struct sk_buff
*skb
,
876 const struct net_device
*dev
)
878 struct net_vrf
*vrf
= netdev_priv(dev
);
880 return nla_put_u32(skb
, IFLA_VRF_TABLE
, vrf
->tb_id
);
883 static const struct nla_policy vrf_nl_policy
[IFLA_VRF_MAX
+ 1] = {
884 [IFLA_VRF_TABLE
] = { .type
= NLA_U32
},
887 static struct rtnl_link_ops vrf_link_ops __read_mostly
= {
889 .priv_size
= sizeof(struct net_vrf
),
891 .get_size
= vrf_nl_getsize
,
892 .policy
= vrf_nl_policy
,
893 .validate
= vrf_validate
,
894 .fill_info
= vrf_fillinfo
,
896 .newlink
= vrf_newlink
,
897 .dellink
= vrf_dellink
,
899 .maxtype
= IFLA_VRF_MAX
,
902 static int vrf_device_event(struct notifier_block
*unused
,
903 unsigned long event
, void *ptr
)
905 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
907 /* only care about unregister events to drop slave references */
908 if (event
== NETDEV_UNREGISTER
) {
909 struct net_device
*vrf_dev
;
911 if (!netif_is_l3_slave(dev
))
914 vrf_dev
= netdev_master_upper_dev_get(dev
);
915 vrf_del_slave(vrf_dev
, dev
);
921 static struct notifier_block vrf_notifier_block __read_mostly
= {
922 .notifier_call
= vrf_device_event
,
925 static int __init
vrf_init_module(void)
929 vrf_dst_ops
.kmem_cachep
=
930 kmem_cache_create("vrf_ip_dst_cache",
931 sizeof(struct rtable
), 0,
935 if (!vrf_dst_ops
.kmem_cachep
)
938 rc
= init_dst_ops6_kmem_cachep();
942 register_netdevice_notifier(&vrf_notifier_block
);
944 rc
= rtnl_link_register(&vrf_link_ops
);
951 unregister_netdevice_notifier(&vrf_notifier_block
);
952 free_dst_ops6_kmem_cachep();
954 kmem_cache_destroy(vrf_dst_ops
.kmem_cachep
);
958 static void __exit
vrf_cleanup_module(void)
960 rtnl_link_unregister(&vrf_link_ops
);
961 unregister_netdevice_notifier(&vrf_notifier_block
);
962 kmem_cache_destroy(vrf_dst_ops
.kmem_cachep
);
963 free_dst_ops6_kmem_cachep();
966 module_init(vrf_init_module
);
967 module_exit(vrf_cleanup_module
);
968 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
969 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
970 MODULE_LICENSE("GPL");
971 MODULE_ALIAS_RTNL_LINK(DRV_NAME
);
972 MODULE_VERSION(DRV_VERSION
);