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_route.h>
34 #include <net/rtnetlink.h>
35 #include <net/route.h>
36 #include <net/addrconf.h>
37 #include <net/l3mdev.h>
39 #define RT_FL_TOS(oldflp4) \
40 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
42 #define DRV_NAME "vrf"
43 #define DRV_VERSION "1.0"
45 #define vrf_master_get_rcu(dev) \
46 ((struct net_device *)rcu_dereference(dev->rx_handler_data))
49 struct list_head list
;
50 struct net_device
*dev
;
54 struct list_head all_slaves
;
58 struct slave_queue queue
;
69 struct u64_stats_sync syncp
;
72 static struct dst_entry
*vrf_ip_check(struct dst_entry
*dst
, u32 cookie
)
77 static int vrf_ip_local_out(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
79 return ip_local_out(net
, sk
, skb
);
82 static unsigned int vrf_v4_mtu(const struct dst_entry
*dst
)
84 /* TO-DO: return max ethernet size? */
88 static void vrf_dst_destroy(struct dst_entry
*dst
)
90 /* our dst lives forever - or until the device is closed */
93 static unsigned int vrf_default_advmss(const struct dst_entry
*dst
)
98 static struct dst_ops vrf_dst_ops
= {
100 .local_out
= vrf_ip_local_out
,
101 .check
= vrf_ip_check
,
103 .destroy
= vrf_dst_destroy
,
104 .default_advmss
= vrf_default_advmss
,
107 static bool is_ip_rx_frame(struct sk_buff
*skb
)
109 switch (skb
->protocol
) {
110 case htons(ETH_P_IP
):
111 case htons(ETH_P_IPV6
):
117 static void vrf_tx_error(struct net_device
*vrf_dev
, struct sk_buff
*skb
)
119 vrf_dev
->stats
.tx_errors
++;
123 /* note: already called with rcu_read_lock */
124 static rx_handler_result_t
vrf_handle_frame(struct sk_buff
**pskb
)
126 struct sk_buff
*skb
= *pskb
;
128 if (is_ip_rx_frame(skb
)) {
129 struct net_device
*dev
= vrf_master_get_rcu(skb
->dev
);
130 struct pcpu_dstats
*dstats
= this_cpu_ptr(dev
->dstats
);
132 u64_stats_update_begin(&dstats
->syncp
);
134 dstats
->rx_bytes
+= skb
->len
;
135 u64_stats_update_end(&dstats
->syncp
);
139 return RX_HANDLER_ANOTHER
;
141 return RX_HANDLER_PASS
;
144 static struct rtnl_link_stats64
*vrf_get_stats64(struct net_device
*dev
,
145 struct rtnl_link_stats64
*stats
)
149 for_each_possible_cpu(i
) {
150 const struct pcpu_dstats
*dstats
;
151 u64 tbytes
, tpkts
, tdrops
, rbytes
, rpkts
;
154 dstats
= per_cpu_ptr(dev
->dstats
, i
);
156 start
= u64_stats_fetch_begin_irq(&dstats
->syncp
);
157 tbytes
= dstats
->tx_bytes
;
158 tpkts
= dstats
->tx_pkts
;
159 tdrops
= dstats
->tx_drps
;
160 rbytes
= dstats
->rx_bytes
;
161 rpkts
= dstats
->rx_pkts
;
162 } while (u64_stats_fetch_retry_irq(&dstats
->syncp
, start
));
163 stats
->tx_bytes
+= tbytes
;
164 stats
->tx_packets
+= tpkts
;
165 stats
->tx_dropped
+= tdrops
;
166 stats
->rx_bytes
+= rbytes
;
167 stats
->rx_packets
+= rpkts
;
172 static netdev_tx_t
vrf_process_v6_outbound(struct sk_buff
*skb
,
173 struct net_device
*dev
)
175 vrf_tx_error(dev
, skb
);
176 return NET_XMIT_DROP
;
179 static int vrf_send_v4_prep(struct sk_buff
*skb
, struct flowi4
*fl4
,
180 struct net_device
*vrf_dev
)
185 rt
= ip_route_output_flow(dev_net(vrf_dev
), fl4
, NULL
);
189 /* TO-DO: what about broadcast ? */
190 if (rt
->rt_type
!= RTN_UNICAST
&& rt
->rt_type
!= RTN_LOCAL
) {
196 skb_dst_set(skb
, &rt
->dst
);
202 static netdev_tx_t
vrf_process_v4_outbound(struct sk_buff
*skb
,
203 struct net_device
*vrf_dev
)
205 struct iphdr
*ip4h
= ip_hdr(skb
);
206 int ret
= NET_XMIT_DROP
;
207 struct flowi4 fl4
= {
208 /* needed to match OIF rule */
209 .flowi4_oif
= vrf_dev
->ifindex
,
210 .flowi4_iif
= LOOPBACK_IFINDEX
,
211 .flowi4_tos
= RT_TOS(ip4h
->tos
),
212 .flowi4_flags
= FLOWI_FLAG_ANYSRC
| FLOWI_FLAG_L3MDEV_SRC
|
213 FLOWI_FLAG_SKIP_NH_OIF
,
214 .daddr
= ip4h
->daddr
,
217 if (vrf_send_v4_prep(skb
, &fl4
, vrf_dev
))
221 ip4h
->saddr
= inet_select_addr(skb_dst(skb
)->dev
, 0,
225 ret
= ip_local_out(dev_net(skb_dst(skb
)->dev
), skb
->sk
, skb
);
226 if (unlikely(net_xmit_eval(ret
)))
227 vrf_dev
->stats
.tx_errors
++;
229 ret
= NET_XMIT_SUCCESS
;
234 vrf_tx_error(vrf_dev
, skb
);
238 static netdev_tx_t
is_ip_tx_frame(struct sk_buff
*skb
, struct net_device
*dev
)
240 /* strip the ethernet header added for pass through VRF device */
241 __skb_pull(skb
, skb_network_offset(skb
));
243 switch (skb
->protocol
) {
244 case htons(ETH_P_IP
):
245 return vrf_process_v4_outbound(skb
, dev
);
246 case htons(ETH_P_IPV6
):
247 return vrf_process_v6_outbound(skb
, dev
);
249 vrf_tx_error(dev
, skb
);
250 return NET_XMIT_DROP
;
254 static netdev_tx_t
vrf_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
256 netdev_tx_t ret
= is_ip_tx_frame(skb
, dev
);
258 if (likely(ret
== NET_XMIT_SUCCESS
|| ret
== NET_XMIT_CN
)) {
259 struct pcpu_dstats
*dstats
= this_cpu_ptr(dev
->dstats
);
261 u64_stats_update_begin(&dstats
->syncp
);
263 dstats
->tx_bytes
+= skb
->len
;
264 u64_stats_update_end(&dstats
->syncp
);
266 this_cpu_inc(dev
->dstats
->tx_drps
);
272 /* modelled after ip_finish_output2 */
273 static int vrf_finish_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
275 struct dst_entry
*dst
= skb_dst(skb
);
276 struct rtable
*rt
= (struct rtable
*)dst
;
277 struct net_device
*dev
= dst
->dev
;
278 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
279 struct neighbour
*neigh
;
283 /* Be paranoid, rather than too clever. */
284 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
285 struct sk_buff
*skb2
;
287 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
293 skb_set_owner_w(skb2
, skb
->sk
);
301 nexthop
= (__force u32
)rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
302 neigh
= __ipv4_neigh_lookup_noref(dev
, nexthop
);
303 if (unlikely(!neigh
))
304 neigh
= __neigh_create(&arp_tbl
, &nexthop
, dev
, false);
306 ret
= dst_neigh_output(dst
, neigh
, skb
);
308 rcu_read_unlock_bh();
310 if (unlikely(ret
< 0))
311 vrf_tx_error(skb
->dev
, skb
);
315 static int vrf_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
317 struct net_device
*dev
= skb_dst(skb
)->dev
;
319 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUT
, skb
->len
);
322 skb
->protocol
= htons(ETH_P_IP
);
324 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
325 net
, sk
, skb
, NULL
, dev
,
327 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
330 static void vrf_rtable_destroy(struct net_vrf
*vrf
)
332 struct dst_entry
*dst
= (struct dst_entry
*)vrf
->rth
;
338 static struct rtable
*vrf_rtable_create(struct net_device
*dev
)
340 struct net_vrf
*vrf
= netdev_priv(dev
);
343 rth
= dst_alloc(&vrf_dst_ops
, dev
, 2,
345 (DST_HOST
| DST_NOPOLICY
| DST_NOXFRM
));
347 rth
->dst
.output
= vrf_output
;
348 rth
->rt_genid
= rt_genid_ipv4(dev_net(dev
));
350 rth
->rt_type
= RTN_UNICAST
;
351 rth
->rt_is_input
= 0;
355 rth
->rt_uses_gateway
= 0;
356 rth
->rt_table_id
= vrf
->tb_id
;
357 INIT_LIST_HEAD(&rth
->rt_uncached
);
358 rth
->rt_uncached_list
= NULL
;
364 /**************************** device handling ********************/
366 /* cycle interface to flush neighbor cache and move routes across tables */
367 static void cycle_netdev(struct net_device
*dev
)
369 unsigned int flags
= dev
->flags
;
372 if (!netif_running(dev
))
375 ret
= dev_change_flags(dev
, flags
& ~IFF_UP
);
377 ret
= dev_change_flags(dev
, flags
);
381 "Failed to cycle device %s; route tables might be wrong!\n",
386 static struct slave
*__vrf_find_slave_dev(struct slave_queue
*queue
,
387 struct net_device
*dev
)
389 struct list_head
*head
= &queue
->all_slaves
;
392 list_for_each_entry(slave
, head
, list
) {
393 if (slave
->dev
== dev
)
400 /* inverse of __vrf_insert_slave */
401 static void __vrf_remove_slave(struct slave_queue
*queue
, struct slave
*slave
)
403 list_del(&slave
->list
);
406 static void __vrf_insert_slave(struct slave_queue
*queue
, struct slave
*slave
)
408 list_add(&slave
->list
, &queue
->all_slaves
);
411 static int do_vrf_add_slave(struct net_device
*dev
, struct net_device
*port_dev
)
413 struct slave
*slave
= kzalloc(sizeof(*slave
), GFP_KERNEL
);
414 struct net_vrf
*vrf
= netdev_priv(dev
);
415 struct slave_queue
*queue
= &vrf
->queue
;
421 slave
->dev
= port_dev
;
423 /* register the packet handler for slave ports */
424 ret
= netdev_rx_handler_register(port_dev
, vrf_handle_frame
, dev
);
427 "Device %s failed to register rx_handler\n",
432 ret
= netdev_master_upper_dev_link(port_dev
, dev
);
436 port_dev
->priv_flags
|= IFF_L3MDEV_SLAVE
;
437 __vrf_insert_slave(queue
, slave
);
438 cycle_netdev(port_dev
);
443 netdev_rx_handler_unregister(port_dev
);
449 static int vrf_add_slave(struct net_device
*dev
, struct net_device
*port_dev
)
451 if (netif_is_l3_master(port_dev
) || netif_is_l3_slave(port_dev
))
454 return do_vrf_add_slave(dev
, port_dev
);
457 /* inverse of do_vrf_add_slave */
458 static int do_vrf_del_slave(struct net_device
*dev
, struct net_device
*port_dev
)
460 struct net_vrf
*vrf
= netdev_priv(dev
);
461 struct slave_queue
*queue
= &vrf
->queue
;
464 netdev_upper_dev_unlink(port_dev
, dev
);
465 port_dev
->priv_flags
&= ~IFF_L3MDEV_SLAVE
;
467 netdev_rx_handler_unregister(port_dev
);
469 cycle_netdev(port_dev
);
471 slave
= __vrf_find_slave_dev(queue
, port_dev
);
473 __vrf_remove_slave(queue
, slave
);
480 static int vrf_del_slave(struct net_device
*dev
, struct net_device
*port_dev
)
482 return do_vrf_del_slave(dev
, port_dev
);
485 static void vrf_dev_uninit(struct net_device
*dev
)
487 struct net_vrf
*vrf
= netdev_priv(dev
);
488 struct slave_queue
*queue
= &vrf
->queue
;
489 struct list_head
*head
= &queue
->all_slaves
;
490 struct slave
*slave
, *next
;
492 vrf_rtable_destroy(vrf
);
494 list_for_each_entry_safe(slave
, next
, head
, list
)
495 vrf_del_slave(dev
, slave
->dev
);
497 free_percpu(dev
->dstats
);
501 static int vrf_dev_init(struct net_device
*dev
)
503 struct net_vrf
*vrf
= netdev_priv(dev
);
505 INIT_LIST_HEAD(&vrf
->queue
.all_slaves
);
507 dev
->dstats
= netdev_alloc_pcpu_stats(struct pcpu_dstats
);
511 /* create the default dst which points back to us */
512 vrf
->rth
= vrf_rtable_create(dev
);
516 dev
->flags
= IFF_MASTER
| IFF_NOARP
;
521 free_percpu(dev
->dstats
);
527 static const struct net_device_ops vrf_netdev_ops
= {
528 .ndo_init
= vrf_dev_init
,
529 .ndo_uninit
= vrf_dev_uninit
,
530 .ndo_start_xmit
= vrf_xmit
,
531 .ndo_get_stats64
= vrf_get_stats64
,
532 .ndo_add_slave
= vrf_add_slave
,
533 .ndo_del_slave
= vrf_del_slave
,
536 static u32
vrf_fib_table(const struct net_device
*dev
)
538 struct net_vrf
*vrf
= netdev_priv(dev
);
543 static struct rtable
*vrf_get_rtable(const struct net_device
*dev
,
544 const struct flowi4
*fl4
)
546 struct rtable
*rth
= NULL
;
548 if (!(fl4
->flowi4_flags
& FLOWI_FLAG_L3MDEV_SRC
)) {
549 struct net_vrf
*vrf
= netdev_priv(dev
);
552 atomic_inc(&rth
->dst
.__refcnt
);
558 /* called under rcu_read_lock */
559 static void vrf_get_saddr(struct net_device
*dev
, struct flowi4
*fl4
)
561 struct fib_result res
= { .tclassid
= 0 };
562 struct net
*net
= dev_net(dev
);
563 u32 orig_tos
= fl4
->flowi4_tos
;
564 u8 flags
= fl4
->flowi4_flags
;
565 u8 scope
= fl4
->flowi4_scope
;
566 u8 tos
= RT_FL_TOS(fl4
);
568 if (unlikely(!fl4
->daddr
))
571 fl4
->flowi4_flags
|= FLOWI_FLAG_SKIP_NH_OIF
;
572 fl4
->flowi4_iif
= LOOPBACK_IFINDEX
;
573 fl4
->flowi4_tos
= tos
& IPTOS_RT_MASK
;
574 fl4
->flowi4_scope
= ((tos
& RTO_ONLINK
) ?
575 RT_SCOPE_LINK
: RT_SCOPE_UNIVERSE
);
577 if (!fib_lookup(net
, fl4
, &res
, 0)) {
578 if (res
.type
== RTN_LOCAL
)
579 fl4
->saddr
= res
.fi
->fib_prefsrc
? : fl4
->daddr
;
581 fib_select_path(net
, &res
, fl4
, -1);
584 fl4
->flowi4_flags
= flags
;
585 fl4
->flowi4_tos
= orig_tos
;
586 fl4
->flowi4_scope
= scope
;
589 static const struct l3mdev_ops vrf_l3mdev_ops
= {
590 .l3mdev_fib_table
= vrf_fib_table
,
591 .l3mdev_get_rtable
= vrf_get_rtable
,
592 .l3mdev_get_saddr
= vrf_get_saddr
,
595 static void vrf_get_drvinfo(struct net_device
*dev
,
596 struct ethtool_drvinfo
*info
)
598 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
599 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
602 static const struct ethtool_ops vrf_ethtool_ops
= {
603 .get_drvinfo
= vrf_get_drvinfo
,
606 static void vrf_setup(struct net_device
*dev
)
610 /* Initialize the device structure. */
611 dev
->netdev_ops
= &vrf_netdev_ops
;
612 dev
->l3mdev_ops
= &vrf_l3mdev_ops
;
613 dev
->ethtool_ops
= &vrf_ethtool_ops
;
614 dev
->destructor
= free_netdev
;
616 /* Fill in device structure with ethernet-generic values. */
617 eth_hw_addr_random(dev
);
619 /* don't acquire vrf device's netif_tx_lock when transmitting */
620 dev
->features
|= NETIF_F_LLTX
;
622 /* don't allow vrf devices to change network namespaces. */
623 dev
->features
|= NETIF_F_NETNS_LOCAL
;
626 static int vrf_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
628 if (tb
[IFLA_ADDRESS
]) {
629 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
631 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
632 return -EADDRNOTAVAIL
;
637 static void vrf_dellink(struct net_device
*dev
, struct list_head
*head
)
639 unregister_netdevice_queue(dev
, head
);
642 static int vrf_newlink(struct net
*src_net
, struct net_device
*dev
,
643 struct nlattr
*tb
[], struct nlattr
*data
[])
645 struct net_vrf
*vrf
= netdev_priv(dev
);
648 if (!data
|| !data
[IFLA_VRF_TABLE
])
651 vrf
->tb_id
= nla_get_u32(data
[IFLA_VRF_TABLE
]);
653 dev
->priv_flags
|= IFF_L3MDEV_MASTER
;
655 err
= register_netdevice(dev
);
666 static size_t vrf_nl_getsize(const struct net_device
*dev
)
668 return nla_total_size(sizeof(u32
)); /* IFLA_VRF_TABLE */
671 static int vrf_fillinfo(struct sk_buff
*skb
,
672 const struct net_device
*dev
)
674 struct net_vrf
*vrf
= netdev_priv(dev
);
676 return nla_put_u32(skb
, IFLA_VRF_TABLE
, vrf
->tb_id
);
679 static const struct nla_policy vrf_nl_policy
[IFLA_VRF_MAX
+ 1] = {
680 [IFLA_VRF_TABLE
] = { .type
= NLA_U32
},
683 static struct rtnl_link_ops vrf_link_ops __read_mostly
= {
685 .priv_size
= sizeof(struct net_vrf
),
687 .get_size
= vrf_nl_getsize
,
688 .policy
= vrf_nl_policy
,
689 .validate
= vrf_validate
,
690 .fill_info
= vrf_fillinfo
,
692 .newlink
= vrf_newlink
,
693 .dellink
= vrf_dellink
,
695 .maxtype
= IFLA_VRF_MAX
,
698 static int vrf_device_event(struct notifier_block
*unused
,
699 unsigned long event
, void *ptr
)
701 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
703 /* only care about unregister events to drop slave references */
704 if (event
== NETDEV_UNREGISTER
) {
705 struct net_device
*vrf_dev
;
707 if (!netif_is_l3_slave(dev
))
710 vrf_dev
= netdev_master_upper_dev_get(dev
);
711 vrf_del_slave(vrf_dev
, dev
);
717 static struct notifier_block vrf_notifier_block __read_mostly
= {
718 .notifier_call
= vrf_device_event
,
721 static int __init
vrf_init_module(void)
725 vrf_dst_ops
.kmem_cachep
=
726 kmem_cache_create("vrf_ip_dst_cache",
727 sizeof(struct rtable
), 0,
731 if (!vrf_dst_ops
.kmem_cachep
)
734 register_netdevice_notifier(&vrf_notifier_block
);
736 rc
= rtnl_link_register(&vrf_link_ops
);
743 unregister_netdevice_notifier(&vrf_notifier_block
);
744 kmem_cache_destroy(vrf_dst_ops
.kmem_cachep
);
748 static void __exit
vrf_cleanup_module(void)
750 rtnl_link_unregister(&vrf_link_ops
);
751 unregister_netdevice_notifier(&vrf_notifier_block
);
752 kmem_cache_destroy(vrf_dst_ops
.kmem_cachep
);
755 module_init(vrf_init_module
);
756 module_exit(vrf_cleanup_module
);
757 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
758 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
759 MODULE_LICENSE("GPL");
760 MODULE_ALIAS_RTNL_LINK(DRV_NAME
);
761 MODULE_VERSION(DRV_VERSION
);