blob: b3762822b6533ea8cae30c16580408d92d657773 [file] [log] [blame]
David Ahern193125d2015-08-13 14:59:10 -06001/*
2 * vrf.c: device driver to encapsulate a VRF space
3 *
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>
7 *
8 * Based on dummy, team and ipvlan drivers
9 *
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.
14 */
15
16#include <linux/module.h>
17#include <linux/kernel.h>
18#include <linux/netdevice.h>
19#include <linux/etherdevice.h>
20#include <linux/ip.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>
28
29#include <linux/inetdevice.h>
David Ahern8f583362015-08-27 10:10:50 -070030#include <net/arp.h>
David Ahern193125d2015-08-13 14:59:10 -060031#include <net/ip.h>
32#include <net/ip_fib.h>
David Ahern35402e32015-10-12 11:47:09 -070033#include <net/ip6_fib.h>
David Ahern193125d2015-08-13 14:59:10 -060034#include <net/ip6_route.h>
David Ahern193125d2015-08-13 14:59:10 -060035#include <net/route.h>
36#include <net/addrconf.h>
David Ahernee15ee52015-09-29 20:07:12 -070037#include <net/l3mdev.h>
David Ahern1aa6c4f2016-06-08 10:55:40 -070038#include <net/fib_rules.h>
David Ahern193125d2015-08-13 14:59:10 -060039
David Ahern8cbb512c2015-10-05 08:51:26 -070040#define RT_FL_TOS(oldflp4) \
41 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
42
David Ahern193125d2015-08-13 14:59:10 -060043#define DRV_NAME "vrf"
44#define DRV_VERSION "1.0"
45
David Ahern1aa6c4f2016-06-08 10:55:40 -070046#define FIB_RULE_PREF 1000 /* default preference for FIB rules */
47static bool add_fib_rules = true;
48
David Ahernec539512015-09-29 20:07:17 -070049struct net_vrf {
David Ahernb0e95cc2016-05-13 12:23:45 -070050 struct rtable __rcu *rth;
David Ahernafe80a42016-06-06 20:50:39 -070051 struct rtable __rcu *rth_local;
David Ahernb0e95cc2016-05-13 12:23:45 -070052 struct rt6_info __rcu *rt6;
David Ahernb4869aa2016-06-06 20:50:40 -070053 struct rt6_info __rcu *rt6_local;
David Ahernec539512015-09-29 20:07:17 -070054 u32 tb_id;
55};
56
David Ahern193125d2015-08-13 14:59:10 -060057struct pcpu_dstats {
58 u64 tx_pkts;
59 u64 tx_bytes;
60 u64 tx_drps;
61 u64 rx_pkts;
62 u64 rx_bytes;
David Ahernafe80a42016-06-06 20:50:39 -070063 u64 rx_drps;
David Ahern193125d2015-08-13 14:59:10 -060064 struct u64_stats_sync syncp;
65};
66
David Ahernafe80a42016-06-06 20:50:39 -070067static void vrf_rx_stats(struct net_device *dev, int len)
68{
69 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
70
71 u64_stats_update_begin(&dstats->syncp);
72 dstats->rx_pkts++;
73 dstats->rx_bytes += len;
74 u64_stats_update_end(&dstats->syncp);
75}
76
Nikolay Aleksandrov57b8efa2015-08-19 06:12:29 +030077static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
78{
79 vrf_dev->stats.tx_errors++;
80 kfree_skb(skb);
81}
82
David Ahern193125d2015-08-13 14:59:10 -060083static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
84 struct rtnl_link_stats64 *stats)
85{
86 int i;
87
88 for_each_possible_cpu(i) {
89 const struct pcpu_dstats *dstats;
90 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
91 unsigned int start;
92
93 dstats = per_cpu_ptr(dev->dstats, i);
94 do {
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;
107 }
108 return stats;
109}
110
David Ahernafe80a42016-06-06 20:50:39 -0700111/* Local traffic destined to local address. Reinsert the packet to rx
112 * path, similar to loopback handling.
113 */
114static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
115 struct dst_entry *dst)
116{
117 int len = skb->len;
118
119 skb_orphan(skb);
120
121 skb_dst_set(skb, dst);
122 skb_dst_force(skb);
123
124 /* set pkt_type to avoid skb hitting packet taps twice -
125 * once on Tx and again in Rx processing
126 */
127 skb->pkt_type = PACKET_LOOPBACK;
128
129 skb->protocol = eth_type_trans(skb, dev);
130
131 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
132 vrf_rx_stats(dev, len);
133 else
134 this_cpu_inc(dev->dstats->rx_drps);
135
136 return NETDEV_TX_OK;
137}
138
David Ahern35402e32015-10-12 11:47:09 -0700139#if IS_ENABLED(CONFIG_IPV6)
140static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
141 struct net_device *dev)
142{
143 const struct ipv6hdr *iph = ipv6_hdr(skb);
144 struct net *net = dev_net(skb->dev);
145 struct flowi6 fl6 = {
146 /* needed to match OIF rule */
147 .flowi6_oif = dev->ifindex,
148 .flowi6_iif = LOOPBACK_IFINDEX,
149 .daddr = iph->daddr,
150 .saddr = iph->saddr,
151 .flowlabel = ip6_flowinfo(iph),
152 .flowi6_mark = skb->mark,
153 .flowi6_proto = iph->nexthdr,
154 .flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF,
155 };
156 int ret = NET_XMIT_DROP;
157 struct dst_entry *dst;
158 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
159
160 dst = ip6_route_output(net, NULL, &fl6);
161 if (dst == dst_null)
162 goto err;
163
164 skb_dst_drop(skb);
David Ahernb4869aa2016-06-06 20:50:40 -0700165
166 /* if dst.dev is loopback or the VRF device again this is locally
167 * originated traffic destined to a local address. Short circuit
168 * to Rx path using our local dst
169 */
170 if (dst->dev == net->loopback_dev || dst->dev == dev) {
171 struct net_vrf *vrf = netdev_priv(dev);
172 struct rt6_info *rt6_local;
173
174 /* release looked up dst and use cached local dst */
175 dst_release(dst);
176
177 rcu_read_lock();
178
179 rt6_local = rcu_dereference(vrf->rt6_local);
180 if (unlikely(!rt6_local)) {
181 rcu_read_unlock();
182 goto err;
183 }
184
185 /* Ordering issue: cached local dst is created on newlink
186 * before the IPv6 initialization. Using the local dst
187 * requires rt6i_idev to be set so make sure it is.
188 */
189 if (unlikely(!rt6_local->rt6i_idev)) {
190 rt6_local->rt6i_idev = in6_dev_get(dev);
191 if (!rt6_local->rt6i_idev) {
192 rcu_read_unlock();
193 goto err;
194 }
195 }
196
197 dst = &rt6_local->dst;
198 dst_hold(dst);
199
200 rcu_read_unlock();
201
202 return vrf_local_xmit(skb, dev, &rt6_local->dst);
203 }
204
David Ahern35402e32015-10-12 11:47:09 -0700205 skb_dst_set(skb, dst);
206
David Ahern911a66f2016-06-06 20:50:38 -0700207 /* strip the ethernet header added for pass through VRF device */
208 __skb_pull(skb, skb_network_offset(skb));
209
David Ahern35402e32015-10-12 11:47:09 -0700210 ret = ip6_local_out(net, skb->sk, skb);
211 if (unlikely(net_xmit_eval(ret)))
212 dev->stats.tx_errors++;
213 else
214 ret = NET_XMIT_SUCCESS;
215
216 return ret;
217err:
218 vrf_tx_error(dev, skb);
219 return NET_XMIT_DROP;
220}
221#else
David Ahern193125d2015-08-13 14:59:10 -0600222static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
223 struct net_device *dev)
224{
Nikolay Aleksandrov57b8efa2015-08-19 06:12:29 +0300225 vrf_tx_error(dev, skb);
226 return NET_XMIT_DROP;
David Ahern193125d2015-08-13 14:59:10 -0600227}
David Ahern35402e32015-10-12 11:47:09 -0700228#endif
David Ahern193125d2015-08-13 14:59:10 -0600229
David Ahern193125d2015-08-13 14:59:10 -0600230static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
231 struct net_device *vrf_dev)
232{
233 struct iphdr *ip4h = ip_hdr(skb);
234 int ret = NET_XMIT_DROP;
235 struct flowi4 fl4 = {
236 /* needed to match OIF rule */
237 .flowi4_oif = vrf_dev->ifindex,
238 .flowi4_iif = LOOPBACK_IFINDEX,
239 .flowi4_tos = RT_TOS(ip4h->tos),
David Ahern6e2895a2015-10-05 08:51:23 -0700240 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
David Ahern58189ca2015-09-15 15:10:50 -0700241 FLOWI_FLAG_SKIP_NH_OIF,
David Ahern193125d2015-08-13 14:59:10 -0600242 .daddr = ip4h->daddr,
243 };
David Ahern911a66f2016-06-06 20:50:38 -0700244 struct net *net = dev_net(vrf_dev);
245 struct rtable *rt;
David Ahern193125d2015-08-13 14:59:10 -0600246
David Ahern911a66f2016-06-06 20:50:38 -0700247 rt = ip_route_output_flow(net, &fl4, NULL);
248 if (IS_ERR(rt))
David Ahern193125d2015-08-13 14:59:10 -0600249 goto err;
250
David Ahern911a66f2016-06-06 20:50:38 -0700251 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
252 ip_rt_put(rt);
253 goto err;
254 }
255
256 skb_dst_drop(skb);
David Ahernafe80a42016-06-06 20:50:39 -0700257
258 /* if dst.dev is loopback or the VRF device again this is locally
259 * originated traffic destined to a local address. Short circuit
260 * to Rx path using our local dst
261 */
262 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) {
263 struct net_vrf *vrf = netdev_priv(vrf_dev);
264 struct rtable *rth_local;
265 struct dst_entry *dst = NULL;
266
267 ip_rt_put(rt);
268
269 rcu_read_lock();
270
271 rth_local = rcu_dereference(vrf->rth_local);
272 if (likely(rth_local)) {
273 dst = &rth_local->dst;
274 dst_hold(dst);
275 }
276
277 rcu_read_unlock();
278
279 if (unlikely(!dst))
280 goto err;
281
282 return vrf_local_xmit(skb, vrf_dev, dst);
283 }
284
David Ahern911a66f2016-06-06 20:50:38 -0700285 skb_dst_set(skb, &rt->dst);
286
287 /* strip the ethernet header added for pass through VRF device */
288 __skb_pull(skb, skb_network_offset(skb));
289
David Ahern193125d2015-08-13 14:59:10 -0600290 if (!ip4h->saddr) {
291 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
292 RT_SCOPE_LINK);
293 }
294
Eric W. Biederman33224b12015-10-07 16:48:46 -0500295 ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
David Ahern193125d2015-08-13 14:59:10 -0600296 if (unlikely(net_xmit_eval(ret)))
297 vrf_dev->stats.tx_errors++;
298 else
299 ret = NET_XMIT_SUCCESS;
300
301out:
302 return ret;
303err:
Nikolay Aleksandrov57b8efa2015-08-19 06:12:29 +0300304 vrf_tx_error(vrf_dev, skb);
David Ahern193125d2015-08-13 14:59:10 -0600305 goto out;
306}
307
308static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
309{
310 switch (skb->protocol) {
311 case htons(ETH_P_IP):
312 return vrf_process_v4_outbound(skb, dev);
313 case htons(ETH_P_IPV6):
314 return vrf_process_v6_outbound(skb, dev);
315 default:
Nikolay Aleksandrov57b8efa2015-08-19 06:12:29 +0300316 vrf_tx_error(dev, skb);
David Ahern193125d2015-08-13 14:59:10 -0600317 return NET_XMIT_DROP;
318 }
319}
320
321static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
322{
323 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
324
325 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
326 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
327
328 u64_stats_update_begin(&dstats->syncp);
329 dstats->tx_pkts++;
330 dstats->tx_bytes += skb->len;
331 u64_stats_update_end(&dstats->syncp);
332 } else {
333 this_cpu_inc(dev->dstats->tx_drps);
334 }
335
336 return ret;
337}
338
David Ahern35402e32015-10-12 11:47:09 -0700339#if IS_ENABLED(CONFIG_IPV6)
David Ahern35402e32015-10-12 11:47:09 -0700340/* modelled after ip6_finish_output2 */
341static int vrf_finish_output6(struct net *net, struct sock *sk,
342 struct sk_buff *skb)
343{
344 struct dst_entry *dst = skb_dst(skb);
345 struct net_device *dev = dst->dev;
346 struct neighbour *neigh;
347 struct in6_addr *nexthop;
348 int ret;
349
350 skb->protocol = htons(ETH_P_IPV6);
351 skb->dev = dev;
352
353 rcu_read_lock_bh();
354 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
355 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
356 if (unlikely(!neigh))
357 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
358 if (!IS_ERR(neigh)) {
359 ret = dst_neigh_output(dst, neigh, skb);
360 rcu_read_unlock_bh();
361 return ret;
362 }
363 rcu_read_unlock_bh();
364
365 IP6_INC_STATS(dev_net(dst->dev),
366 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
367 kfree_skb(skb);
368 return -EINVAL;
369}
370
371/* modelled after ip6_output */
372static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
373{
374 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
375 net, sk, skb, NULL, skb_dst(skb)->dev,
376 vrf_finish_output6,
377 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
378}
379
David Ahernb0e95cc2016-05-13 12:23:45 -0700380/* holding rtnl */
David Ahern810e5302016-06-14 11:37:21 -0700381static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
David Ahern35402e32015-10-12 11:47:09 -0700382{
David Ahernb0e95cc2016-05-13 12:23:45 -0700383 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
David Ahernb4869aa2016-06-06 20:50:40 -0700384 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local);
David Ahern810e5302016-06-14 11:37:21 -0700385 struct net *net = dev_net(dev);
386 struct dst_entry *dst;
David Ahernb0e95cc2016-05-13 12:23:45 -0700387
David Ahernb4869aa2016-06-06 20:50:40 -0700388 RCU_INIT_POINTER(vrf->rt6, NULL);
389 RCU_INIT_POINTER(vrf->rt6_local, NULL);
390 synchronize_rcu();
David Ahernb0e95cc2016-05-13 12:23:45 -0700391
David Ahern810e5302016-06-14 11:37:21 -0700392 /* move dev in dst's to loopback so this VRF device can be deleted
393 * - based on dst_ifdown
394 */
395 if (rt6) {
396 dst = &rt6->dst;
397 dev_put(dst->dev);
398 dst->dev = net->loopback_dev;
399 dev_hold(dst->dev);
400 dst_release(dst);
401 }
David Ahernb4869aa2016-06-06 20:50:40 -0700402
403 if (rt6_local) {
404 if (rt6_local->rt6i_idev)
405 in6_dev_put(rt6_local->rt6i_idev);
406
David Ahern810e5302016-06-14 11:37:21 -0700407 dst = &rt6_local->dst;
408 dev_put(dst->dev);
409 dst->dev = net->loopback_dev;
410 dev_hold(dst->dev);
411 dst_release(dst);
David Ahernb4869aa2016-06-06 20:50:40 -0700412 }
David Ahern35402e32015-10-12 11:47:09 -0700413}
414
415static int vrf_rt6_create(struct net_device *dev)
416{
David Ahernb4869aa2016-06-06 20:50:40 -0700417 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE;
David Ahern35402e32015-10-12 11:47:09 -0700418 struct net_vrf *vrf = netdev_priv(dev);
David Ahern9ab179d2016-04-07 11:10:06 -0700419 struct net *net = dev_net(dev);
David Ahernb3b46632016-05-04 21:46:12 -0700420 struct fib6_table *rt6i_table;
David Ahernb4869aa2016-06-06 20:50:40 -0700421 struct rt6_info *rt6, *rt6_local;
David Ahern35402e32015-10-12 11:47:09 -0700422 int rc = -ENOMEM;
423
David Aherne4348632016-06-09 10:21:00 -0700424 /* IPv6 can be CONFIG enabled and then disabled runtime */
425 if (!ipv6_mod_enabled())
426 return 0;
427
David Ahernb3b46632016-05-04 21:46:12 -0700428 rt6i_table = fib6_new_table(net, vrf->tb_id);
429 if (!rt6i_table)
430 goto out;
431
David Ahernb4869aa2016-06-06 20:50:40 -0700432 /* create a dst for routing packets out a VRF device */
433 rt6 = ip6_dst_alloc(net, dev, flags);
David Ahern35402e32015-10-12 11:47:09 -0700434 if (!rt6)
435 goto out;
436
David Ahern9ab179d2016-04-07 11:10:06 -0700437 dst_hold(&rt6->dst);
David Ahernb3b46632016-05-04 21:46:12 -0700438
439 rt6->rt6i_table = rt6i_table;
440 rt6->dst.output = vrf_output6;
David Ahernb4869aa2016-06-06 20:50:40 -0700441
442 /* create a dst for local routing - packets sent locally
443 * to local address via the VRF device as a loopback
444 */
445 rt6_local = ip6_dst_alloc(net, dev, flags);
446 if (!rt6_local) {
447 dst_release(&rt6->dst);
448 goto out;
449 }
450
451 dst_hold(&rt6_local->dst);
452
453 rt6_local->rt6i_idev = in6_dev_get(dev);
454 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL;
455 rt6_local->rt6i_table = rt6i_table;
456 rt6_local->dst.input = ip6_input;
457
David Ahernb0e95cc2016-05-13 12:23:45 -0700458 rcu_assign_pointer(vrf->rt6, rt6);
David Ahernb4869aa2016-06-06 20:50:40 -0700459 rcu_assign_pointer(vrf->rt6_local, rt6_local);
David Ahernb0e95cc2016-05-13 12:23:45 -0700460
David Ahern35402e32015-10-12 11:47:09 -0700461 rc = 0;
462out:
463 return rc;
464}
465#else
David Ahern810e5302016-06-14 11:37:21 -0700466static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
David Ahern35402e32015-10-12 11:47:09 -0700467{
468}
469
470static int vrf_rt6_create(struct net_device *dev)
471{
472 return 0;
473}
474#endif
475
David Ahern8f583362015-08-27 10:10:50 -0700476/* modelled after ip_finish_output2 */
Eric W. Biederman0c4b51f2015-09-15 20:04:18 -0500477static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
David Ahern193125d2015-08-13 14:59:10 -0600478{
David Ahern8f583362015-08-27 10:10:50 -0700479 struct dst_entry *dst = skb_dst(skb);
480 struct rtable *rt = (struct rtable *)dst;
481 struct net_device *dev = dst->dev;
482 unsigned int hh_len = LL_RESERVED_SPACE(dev);
483 struct neighbour *neigh;
484 u32 nexthop;
485 int ret = -EINVAL;
486
487 /* Be paranoid, rather than too clever. */
488 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
489 struct sk_buff *skb2;
490
491 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
492 if (!skb2) {
493 ret = -ENOMEM;
494 goto err;
495 }
496 if (skb->sk)
497 skb_set_owner_w(skb2, skb->sk);
498
499 consume_skb(skb);
500 skb = skb2;
501 }
502
503 rcu_read_lock_bh();
504
505 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
506 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
507 if (unlikely(!neigh))
508 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
509 if (!IS_ERR(neigh))
510 ret = dst_neigh_output(dst, neigh, skb);
511
512 rcu_read_unlock_bh();
513err:
514 if (unlikely(ret < 0))
515 vrf_tx_error(skb->dev, skb);
516 return ret;
David Ahern193125d2015-08-13 14:59:10 -0600517}
518
Eric W. Biedermanede20592015-10-07 16:48:47 -0500519static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
David Ahern193125d2015-08-13 14:59:10 -0600520{
521 struct net_device *dev = skb_dst(skb)->dev;
522
Eric W. Biederman29a26a52015-09-15 20:04:16 -0500523 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
David Ahern193125d2015-08-13 14:59:10 -0600524
525 skb->dev = dev;
526 skb->protocol = htons(ETH_P_IP);
527
Eric W. Biederman29a26a52015-09-15 20:04:16 -0500528 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
529 net, sk, skb, NULL, dev,
David Ahern8f583362015-08-27 10:10:50 -0700530 vrf_finish_output,
David Ahern193125d2015-08-13 14:59:10 -0600531 !(IPCB(skb)->flags & IPSKB_REROUTED));
532}
533
David Ahernb0e95cc2016-05-13 12:23:45 -0700534/* holding rtnl */
David Ahern810e5302016-06-14 11:37:21 -0700535static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
David Ahern193125d2015-08-13 14:59:10 -0600536{
David Ahernb0e95cc2016-05-13 12:23:45 -0700537 struct rtable *rth = rtnl_dereference(vrf->rth);
David Ahernafe80a42016-06-06 20:50:39 -0700538 struct rtable *rth_local = rtnl_dereference(vrf->rth_local);
David Ahern810e5302016-06-14 11:37:21 -0700539 struct net *net = dev_net(dev);
540 struct dst_entry *dst;
David Ahern193125d2015-08-13 14:59:10 -0600541
David Ahernafe80a42016-06-06 20:50:39 -0700542 RCU_INIT_POINTER(vrf->rth, NULL);
543 RCU_INIT_POINTER(vrf->rth_local, NULL);
544 synchronize_rcu();
David Ahernb0e95cc2016-05-13 12:23:45 -0700545
David Ahern810e5302016-06-14 11:37:21 -0700546 /* move dev in dst's to loopback so this VRF device can be deleted
547 * - based on dst_ifdown
548 */
549 if (rth) {
550 dst = &rth->dst;
551 dev_put(dst->dev);
552 dst->dev = net->loopback_dev;
553 dev_hold(dst->dev);
554 dst_release(dst);
555 }
David Ahernafe80a42016-06-06 20:50:39 -0700556
David Ahern810e5302016-06-14 11:37:21 -0700557 if (rth_local) {
558 dst = &rth_local->dst;
559 dev_put(dst->dev);
560 dst->dev = net->loopback_dev;
561 dev_hold(dst->dev);
562 dst_release(dst);
563 }
David Ahern193125d2015-08-13 14:59:10 -0600564}
565
David Ahernb0e95cc2016-05-13 12:23:45 -0700566static int vrf_rtable_create(struct net_device *dev)
David Ahern193125d2015-08-13 14:59:10 -0600567{
David Ahernb7503e02015-09-02 13:58:35 -0700568 struct net_vrf *vrf = netdev_priv(dev);
David Ahernafe80a42016-06-06 20:50:39 -0700569 struct rtable *rth, *rth_local;
David Ahern193125d2015-08-13 14:59:10 -0600570
David Ahernb3b46632016-05-04 21:46:12 -0700571 if (!fib_new_table(dev_net(dev), vrf->tb_id))
David Ahernb0e95cc2016-05-13 12:23:45 -0700572 return -ENOMEM;
David Ahernb3b46632016-05-04 21:46:12 -0700573
David Ahernafe80a42016-06-06 20:50:39 -0700574 /* create a dst for routing packets out through a VRF device */
David Ahern9ab179d2016-04-07 11:10:06 -0700575 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
David Ahernb0e95cc2016-05-13 12:23:45 -0700576 if (!rth)
577 return -ENOMEM;
David Ahern193125d2015-08-13 14:59:10 -0600578
David Ahernafe80a42016-06-06 20:50:39 -0700579 /* create a dst for local ingress routing - packets sent locally
580 * to local address via the VRF device as a loopback
581 */
582 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0);
583 if (!rth_local) {
584 dst_release(&rth->dst);
585 return -ENOMEM;
586 }
587
David Ahernb0e95cc2016-05-13 12:23:45 -0700588 rth->dst.output = vrf_output;
589 rth->rt_table_id = vrf->tb_id;
590
David Ahernafe80a42016-06-06 20:50:39 -0700591 rth_local->rt_table_id = vrf->tb_id;
592
David Ahernb0e95cc2016-05-13 12:23:45 -0700593 rcu_assign_pointer(vrf->rth, rth);
David Ahernafe80a42016-06-06 20:50:39 -0700594 rcu_assign_pointer(vrf->rth_local, rth_local);
David Ahernb0e95cc2016-05-13 12:23:45 -0700595
596 return 0;
David Ahern193125d2015-08-13 14:59:10 -0600597}
598
599/**************************** device handling ********************/
600
601/* cycle interface to flush neighbor cache and move routes across tables */
602static void cycle_netdev(struct net_device *dev)
603{
604 unsigned int flags = dev->flags;
605 int ret;
606
607 if (!netif_running(dev))
608 return;
609
610 ret = dev_change_flags(dev, flags & ~IFF_UP);
611 if (ret >= 0)
612 ret = dev_change_flags(dev, flags);
613
614 if (ret < 0) {
615 netdev_err(dev,
616 "Failed to cycle device %s; route tables might be wrong!\n",
617 dev->name);
618 }
619}
620
David Ahern193125d2015-08-13 14:59:10 -0600621static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
622{
Nikolay Aleksandrovbad53162015-11-24 14:29:16 +0100623 int ret;
David Ahern193125d2015-08-13 14:59:10 -0600624
Jiri Pirko29bf24a2015-12-03 12:12:11 +0100625 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
David Ahern193125d2015-08-13 14:59:10 -0600626 if (ret < 0)
David Ahern74b20582016-05-10 11:19:50 -0700627 return ret;
David Ahern193125d2015-08-13 14:59:10 -0600628
David Ahernfee6d4c2015-10-05 08:51:24 -0700629 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
David Ahern193125d2015-08-13 14:59:10 -0600630 cycle_netdev(port_dev);
631
632 return 0;
David Ahern193125d2015-08-13 14:59:10 -0600633}
634
635static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
636{
David Ahernfee6d4c2015-10-05 08:51:24 -0700637 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
David Ahern193125d2015-08-13 14:59:10 -0600638 return -EINVAL;
639
640 return do_vrf_add_slave(dev, port_dev);
641}
642
643/* inverse of do_vrf_add_slave */
644static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
645{
David Ahern193125d2015-08-13 14:59:10 -0600646 netdev_upper_dev_unlink(port_dev, dev);
David Ahernfee6d4c2015-10-05 08:51:24 -0700647 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
David Ahern193125d2015-08-13 14:59:10 -0600648
David Ahern193125d2015-08-13 14:59:10 -0600649 cycle_netdev(port_dev);
650
David Ahern193125d2015-08-13 14:59:10 -0600651 return 0;
652}
653
654static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
655{
David Ahern193125d2015-08-13 14:59:10 -0600656 return do_vrf_del_slave(dev, port_dev);
657}
658
659static void vrf_dev_uninit(struct net_device *dev)
660{
661 struct net_vrf *vrf = netdev_priv(dev);
Nikolay Aleksandrovbad53162015-11-24 14:29:16 +0100662 struct net_device *port_dev;
663 struct list_head *iter;
David Ahern193125d2015-08-13 14:59:10 -0600664
David Ahern810e5302016-06-14 11:37:21 -0700665 vrf_rtable_release(dev, vrf);
666 vrf_rt6_release(dev, vrf);
David Ahern193125d2015-08-13 14:59:10 -0600667
Nikolay Aleksandrovbad53162015-11-24 14:29:16 +0100668 netdev_for_each_lower_dev(dev, port_dev, iter)
669 vrf_del_slave(dev, port_dev);
David Ahern193125d2015-08-13 14:59:10 -0600670
Nikolay Aleksandrov3a4a27d2015-08-18 20:28:03 +0300671 free_percpu(dev->dstats);
David Ahern193125d2015-08-13 14:59:10 -0600672 dev->dstats = NULL;
673}
674
675static int vrf_dev_init(struct net_device *dev)
676{
677 struct net_vrf *vrf = netdev_priv(dev);
678
David Ahern193125d2015-08-13 14:59:10 -0600679 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
680 if (!dev->dstats)
681 goto out_nomem;
682
683 /* create the default dst which points back to us */
David Ahernb0e95cc2016-05-13 12:23:45 -0700684 if (vrf_rtable_create(dev) != 0)
David Ahern193125d2015-08-13 14:59:10 -0600685 goto out_stats;
686
David Ahern35402e32015-10-12 11:47:09 -0700687 if (vrf_rt6_create(dev) != 0)
688 goto out_rth;
689
David Ahern193125d2015-08-13 14:59:10 -0600690 dev->flags = IFF_MASTER | IFF_NOARP;
691
David Ahernb87ab6b2016-06-01 21:16:39 -0700692 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
693 dev->mtu = 64 * 1024;
694
695 /* similarly, oper state is irrelevant; set to up to avoid confusion */
696 dev->operstate = IF_OPER_UP;
Eric Dumazet78e7a2a2016-06-09 07:45:13 -0700697 netdev_lockdep_set_classes(dev);
David Ahern193125d2015-08-13 14:59:10 -0600698 return 0;
699
David Ahern35402e32015-10-12 11:47:09 -0700700out_rth:
David Ahern810e5302016-06-14 11:37:21 -0700701 vrf_rtable_release(dev, vrf);
David Ahern193125d2015-08-13 14:59:10 -0600702out_stats:
703 free_percpu(dev->dstats);
704 dev->dstats = NULL;
705out_nomem:
706 return -ENOMEM;
707}
708
709static const struct net_device_ops vrf_netdev_ops = {
710 .ndo_init = vrf_dev_init,
711 .ndo_uninit = vrf_dev_uninit,
712 .ndo_start_xmit = vrf_xmit,
713 .ndo_get_stats64 = vrf_get_stats64,
714 .ndo_add_slave = vrf_add_slave,
715 .ndo_del_slave = vrf_del_slave,
716};
717
David Ahernee15ee52015-09-29 20:07:12 -0700718static u32 vrf_fib_table(const struct net_device *dev)
719{
720 struct net_vrf *vrf = netdev_priv(dev);
721
722 return vrf->tb_id;
723}
724
725static struct rtable *vrf_get_rtable(const struct net_device *dev,
726 const struct flowi4 *fl4)
727{
728 struct rtable *rth = NULL;
729
David Ahern6e2895a2015-10-05 08:51:23 -0700730 if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
David Ahernee15ee52015-09-29 20:07:12 -0700731 struct net_vrf *vrf = netdev_priv(dev);
732
David Ahernb0e95cc2016-05-13 12:23:45 -0700733 rcu_read_lock();
734
735 rth = rcu_dereference(vrf->rth);
736 if (likely(rth))
737 dst_hold(&rth->dst);
738
739 rcu_read_unlock();
David Ahernee15ee52015-09-29 20:07:12 -0700740 }
741
742 return rth;
743}
744
David Ahern8cbb512c2015-10-05 08:51:26 -0700745/* called under rcu_read_lock */
David Ahernb5bdacf2016-01-04 09:09:27 -0800746static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4)
David Ahern8cbb512c2015-10-05 08:51:26 -0700747{
748 struct fib_result res = { .tclassid = 0 };
749 struct net *net = dev_net(dev);
750 u32 orig_tos = fl4->flowi4_tos;
751 u8 flags = fl4->flowi4_flags;
752 u8 scope = fl4->flowi4_scope;
753 u8 tos = RT_FL_TOS(fl4);
David Ahernb5bdacf2016-01-04 09:09:27 -0800754 int rc;
David Ahern8cbb512c2015-10-05 08:51:26 -0700755
756 if (unlikely(!fl4->daddr))
David Ahernb5bdacf2016-01-04 09:09:27 -0800757 return 0;
David Ahern8cbb512c2015-10-05 08:51:26 -0700758
759 fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF;
760 fl4->flowi4_iif = LOOPBACK_IFINDEX;
David Ahern1ff23be2016-05-07 16:49:00 -0700761 /* make sure oif is set to VRF device for lookup */
762 fl4->flowi4_oif = dev->ifindex;
David Ahern8cbb512c2015-10-05 08:51:26 -0700763 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
764 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
765 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
766
David Ahernb5bdacf2016-01-04 09:09:27 -0800767 rc = fib_lookup(net, fl4, &res, 0);
768 if (!rc) {
David Ahern8cbb512c2015-10-05 08:51:26 -0700769 if (res.type == RTN_LOCAL)
770 fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr;
771 else
772 fib_select_path(net, &res, fl4, -1);
773 }
774
775 fl4->flowi4_flags = flags;
776 fl4->flowi4_tos = orig_tos;
777 fl4->flowi4_scope = scope;
David Ahernb5bdacf2016-01-04 09:09:27 -0800778
779 return rc;
David Ahern8cbb512c2015-10-05 08:51:26 -0700780}
781
David Ahern35402e32015-10-12 11:47:09 -0700782#if IS_ENABLED(CONFIG_IPV6)
David Ahern74b20582016-05-10 11:19:50 -0700783/* neighbor handling is done with actual device; do not want
784 * to flip skb->dev for those ndisc packets. This really fails
785 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
786 * a start.
787 */
788static bool ipv6_ndisc_frame(const struct sk_buff *skb)
789{
790 const struct ipv6hdr *iph = ipv6_hdr(skb);
791 bool rc = false;
792
793 if (iph->nexthdr == NEXTHDR_ICMP) {
794 const struct icmp6hdr *icmph;
795 struct icmp6hdr _icmph;
796
797 icmph = skb_header_pointer(skb, sizeof(*iph),
798 sizeof(_icmph), &_icmph);
799 if (!icmph)
800 goto out;
801
802 switch (icmph->icmp6_type) {
803 case NDISC_ROUTER_SOLICITATION:
804 case NDISC_ROUTER_ADVERTISEMENT:
805 case NDISC_NEIGHBOUR_SOLICITATION:
806 case NDISC_NEIGHBOUR_ADVERTISEMENT:
807 case NDISC_REDIRECT:
808 rc = true;
809 break;
810 }
811 }
812
813out:
814 return rc;
815}
816
David Ahern9ff74382016-06-13 13:44:19 -0700817static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
818 const struct net_device *dev,
819 struct flowi6 *fl6,
820 int ifindex,
821 int flags)
822{
823 struct net_vrf *vrf = netdev_priv(dev);
824 struct fib6_table *table = NULL;
825 struct rt6_info *rt6;
826
827 rcu_read_lock();
828
829 /* fib6_table does not have a refcnt and can not be freed */
830 rt6 = rcu_dereference(vrf->rt6);
831 if (likely(rt6))
832 table = rt6->rt6i_table;
833
834 rcu_read_unlock();
835
836 if (!table)
837 return NULL;
838
839 return ip6_pol_route(net, table, ifindex, fl6, flags);
840}
841
842static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
843 int ifindex)
844{
845 const struct ipv6hdr *iph = ipv6_hdr(skb);
846 struct flowi6 fl6 = {
847 .daddr = iph->daddr,
848 .saddr = iph->saddr,
849 .flowlabel = ip6_flowinfo(iph),
850 .flowi6_mark = skb->mark,
851 .flowi6_proto = iph->nexthdr,
852 .flowi6_iif = ifindex,
853 };
854 struct net *net = dev_net(vrf_dev);
855 struct rt6_info *rt6;
856
857 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
858 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
859 if (unlikely(!rt6))
860 return;
861
862 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
863 return;
864
865 skb_dst_set(skb, &rt6->dst);
866}
867
David Ahern74b20582016-05-10 11:19:50 -0700868static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
869 struct sk_buff *skb)
870{
David Ahern9ff74382016-06-13 13:44:19 -0700871 int orig_iif = skb->skb_iif;
872 bool need_strict;
873
David Ahernb4869aa2016-06-06 20:50:40 -0700874 /* loopback traffic; do not push through packet taps again.
875 * Reset pkt_type for upper layers to process skb
876 */
877 if (skb->pkt_type == PACKET_LOOPBACK) {
878 skb->dev = vrf_dev;
879 skb->skb_iif = vrf_dev->ifindex;
880 skb->pkt_type = PACKET_HOST;
881 goto out;
882 }
883
David Ahern9ff74382016-06-13 13:44:19 -0700884 /* if packet is NDISC or addressed to multicast or link-local
885 * then keep the ingress interface
886 */
887 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
888 if (!ipv6_ndisc_frame(skb) && !need_strict) {
David Ahern74b20582016-05-10 11:19:50 -0700889 skb->dev = vrf_dev;
890 skb->skb_iif = vrf_dev->ifindex;
891
892 skb_push(skb, skb->mac_len);
893 dev_queue_xmit_nit(skb, vrf_dev);
894 skb_pull(skb, skb->mac_len);
895
896 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
897 }
898
David Ahern9ff74382016-06-13 13:44:19 -0700899 if (need_strict)
900 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
901
David Ahernb4869aa2016-06-06 20:50:40 -0700902out:
David Ahern74b20582016-05-10 11:19:50 -0700903 return skb;
904}
905
906#else
907static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
908 struct sk_buff *skb)
909{
910 return skb;
911}
912#endif
913
914static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
915 struct sk_buff *skb)
916{
917 skb->dev = vrf_dev;
918 skb->skb_iif = vrf_dev->ifindex;
919
David Ahernafe80a42016-06-06 20:50:39 -0700920 /* loopback traffic; do not push through packet taps again.
921 * Reset pkt_type for upper layers to process skb
922 */
923 if (skb->pkt_type == PACKET_LOOPBACK) {
924 skb->pkt_type = PACKET_HOST;
925 goto out;
926 }
927
David Ahern74b20582016-05-10 11:19:50 -0700928 skb_push(skb, skb->mac_len);
929 dev_queue_xmit_nit(skb, vrf_dev);
930 skb_pull(skb, skb->mac_len);
931
David Ahernafe80a42016-06-06 20:50:39 -0700932out:
David Ahern74b20582016-05-10 11:19:50 -0700933 return skb;
934}
935
936/* called with rcu lock held */
937static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
938 struct sk_buff *skb,
939 u16 proto)
940{
941 switch (proto) {
942 case AF_INET:
943 return vrf_ip_rcv(vrf_dev, skb);
944 case AF_INET6:
945 return vrf_ip6_rcv(vrf_dev, skb);
946 }
947
948 return skb;
949}
950
951#if IS_ENABLED(CONFIG_IPV6)
David Ahern35402e32015-10-12 11:47:09 -0700952static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
David Aherncd2a9e62016-06-13 13:44:17 -0700953 struct flowi6 *fl6)
David Ahern35402e32015-10-12 11:47:09 -0700954{
David Ahern9ff74382016-06-13 13:44:19 -0700955 bool need_strict = rt6_need_strict(&fl6->daddr);
956 struct net_vrf *vrf = netdev_priv(dev);
957 struct net *net = dev_net(dev);
David Ahernb0e95cc2016-05-13 12:23:45 -0700958 struct dst_entry *dst = NULL;
David Ahern9ff74382016-06-13 13:44:19 -0700959 struct rt6_info *rt;
David Ahern35402e32015-10-12 11:47:09 -0700960
David Ahern9ff74382016-06-13 13:44:19 -0700961 /* send to link-local or multicast address */
962 if (need_strict) {
963 int flags = RT6_LOOKUP_F_IFACE;
964
965 /* VRF device does not have a link-local address and
966 * sending packets to link-local or mcast addresses over
967 * a VRF device does not make sense
968 */
969 if (fl6->flowi6_oif == dev->ifindex) {
970 struct dst_entry *dst = &net->ipv6.ip6_null_entry->dst;
971
972 dst_hold(dst);
973 return dst;
974 }
975
976 if (!ipv6_addr_any(&fl6->saddr))
977 flags |= RT6_LOOKUP_F_HAS_SADDR;
978
979 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
980 if (rt)
981 dst = &rt->dst;
982
983 } else if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) {
David Ahern35402e32015-10-12 11:47:09 -0700984
David Ahernb0e95cc2016-05-13 12:23:45 -0700985 rcu_read_lock();
986
987 rt = rcu_dereference(vrf->rt6);
988 if (likely(rt)) {
989 dst = &rt->dst;
990 dst_hold(dst);
991 }
992
993 rcu_read_unlock();
David Ahern35402e32015-10-12 11:47:09 -0700994 }
995
David Ahern9ff74382016-06-13 13:44:19 -0700996 /* make sure oif is set to VRF device for lookup */
997 if (!need_strict)
998 fl6->flowi6_oif = dev->ifindex;
999
David Ahernb0e95cc2016-05-13 12:23:45 -07001000 return dst;
David Ahern35402e32015-10-12 11:47:09 -07001001}
David Ahern0d240e72016-06-16 16:24:25 -07001002
1003/* called under rcu_read_lock */
1004static int vrf_get_saddr6(struct net_device *dev, const struct sock *sk,
1005 struct flowi6 *fl6)
1006{
1007 struct net *net = dev_net(dev);
1008 struct dst_entry *dst;
1009 struct rt6_info *rt;
1010 int err;
1011
1012 if (rt6_need_strict(&fl6->daddr)) {
1013 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif,
1014 RT6_LOOKUP_F_IFACE);
1015 if (unlikely(!rt))
1016 return 0;
1017
1018 dst = &rt->dst;
1019 } else {
1020 __u8 flags = fl6->flowi6_flags;
1021
1022 fl6->flowi6_flags |= FLOWI_FLAG_L3MDEV_SRC;
1023 fl6->flowi6_flags |= FLOWI_FLAG_SKIP_NH_OIF;
1024
1025 dst = ip6_route_output(net, sk, fl6);
1026 rt = (struct rt6_info *)dst;
1027
1028 fl6->flowi6_flags = flags;
1029 }
1030
1031 err = dst->error;
1032 if (!err) {
1033 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
1034 sk ? inet6_sk(sk)->srcprefs : 0,
1035 &fl6->saddr);
1036 }
1037
1038 dst_release(dst);
1039
1040 return err;
1041}
David Ahern35402e32015-10-12 11:47:09 -07001042#endif
1043
David Ahernee15ee52015-09-29 20:07:12 -07001044static const struct l3mdev_ops vrf_l3mdev_ops = {
1045 .l3mdev_fib_table = vrf_fib_table,
1046 .l3mdev_get_rtable = vrf_get_rtable,
David Ahern8cbb512c2015-10-05 08:51:26 -07001047 .l3mdev_get_saddr = vrf_get_saddr,
David Ahern74b20582016-05-10 11:19:50 -07001048 .l3mdev_l3_rcv = vrf_l3_rcv,
David Ahern35402e32015-10-12 11:47:09 -07001049#if IS_ENABLED(CONFIG_IPV6)
1050 .l3mdev_get_rt6_dst = vrf_get_rt6_dst,
David Ahern0d240e72016-06-16 16:24:25 -07001051 .l3mdev_get_saddr6 = vrf_get_saddr6,
David Ahern35402e32015-10-12 11:47:09 -07001052#endif
David Ahernee15ee52015-09-29 20:07:12 -07001053};
1054
David Ahern193125d2015-08-13 14:59:10 -06001055static void vrf_get_drvinfo(struct net_device *dev,
1056 struct ethtool_drvinfo *info)
1057{
1058 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1059 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1060}
1061
1062static const struct ethtool_ops vrf_ethtool_ops = {
1063 .get_drvinfo = vrf_get_drvinfo,
1064};
1065
David Ahern1aa6c4f2016-06-08 10:55:40 -07001066static inline size_t vrf_fib_rule_nl_size(void)
1067{
1068 size_t sz;
1069
1070 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1071 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1072 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1073
1074 return sz;
1075}
1076
1077static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1078{
1079 struct fib_rule_hdr *frh;
1080 struct nlmsghdr *nlh;
1081 struct sk_buff *skb;
1082 int err;
1083
David Aherne4348632016-06-09 10:21:00 -07001084 if (family == AF_INET6 && !ipv6_mod_enabled())
1085 return 0;
1086
David Ahern1aa6c4f2016-06-08 10:55:40 -07001087 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1088 if (!skb)
1089 return -ENOMEM;
1090
1091 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1092 if (!nlh)
1093 goto nla_put_failure;
1094
1095 /* rule only needs to appear once */
1096 nlh->nlmsg_flags &= NLM_F_EXCL;
1097
1098 frh = nlmsg_data(nlh);
1099 memset(frh, 0, sizeof(*frh));
1100 frh->family = family;
1101 frh->action = FR_ACT_TO_TBL;
1102
1103 if (nla_put_u32(skb, FRA_L3MDEV, 1))
1104 goto nla_put_failure;
1105
1106 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1107 goto nla_put_failure;
1108
1109 nlmsg_end(skb, nlh);
1110
1111 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1112 skb->sk = dev_net(dev)->rtnl;
1113 if (add_it) {
1114 err = fib_nl_newrule(skb, nlh);
1115 if (err == -EEXIST)
1116 err = 0;
1117 } else {
1118 err = fib_nl_delrule(skb, nlh);
1119 if (err == -ENOENT)
1120 err = 0;
1121 }
1122 nlmsg_free(skb);
1123
1124 return err;
1125
1126nla_put_failure:
1127 nlmsg_free(skb);
1128
1129 return -EMSGSIZE;
1130}
1131
1132static int vrf_add_fib_rules(const struct net_device *dev)
1133{
1134 int err;
1135
1136 err = vrf_fib_rule(dev, AF_INET, true);
1137 if (err < 0)
1138 goto out_err;
1139
1140 err = vrf_fib_rule(dev, AF_INET6, true);
1141 if (err < 0)
1142 goto ipv6_err;
1143
1144 return 0;
1145
1146ipv6_err:
1147 vrf_fib_rule(dev, AF_INET, false);
1148
1149out_err:
1150 netdev_err(dev, "Failed to add FIB rules.\n");
1151 return err;
1152}
1153
David Ahern193125d2015-08-13 14:59:10 -06001154static void vrf_setup(struct net_device *dev)
1155{
1156 ether_setup(dev);
1157
1158 /* Initialize the device structure. */
1159 dev->netdev_ops = &vrf_netdev_ops;
David Ahernee15ee52015-09-29 20:07:12 -07001160 dev->l3mdev_ops = &vrf_l3mdev_ops;
David Ahern193125d2015-08-13 14:59:10 -06001161 dev->ethtool_ops = &vrf_ethtool_ops;
1162 dev->destructor = free_netdev;
1163
1164 /* Fill in device structure with ethernet-generic values. */
1165 eth_hw_addr_random(dev);
1166
1167 /* don't acquire vrf device's netif_tx_lock when transmitting */
1168 dev->features |= NETIF_F_LLTX;
1169
1170 /* don't allow vrf devices to change network namespaces. */
1171 dev->features |= NETIF_F_NETNS_LOCAL;
David Ahern78896812016-06-13 17:14:12 -07001172
1173 /* does not make sense for a VLAN to be added to a vrf device */
1174 dev->features |= NETIF_F_VLAN_CHALLENGED;
1175
1176 /* enable offload features */
1177 dev->features |= NETIF_F_GSO_SOFTWARE;
1178 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1179 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1180
1181 dev->hw_features = dev->features;
1182 dev->hw_enc_features = dev->features;
1183
1184 /* default to no qdisc; user can add if desired */
1185 dev->priv_flags |= IFF_NO_QUEUE;
David Ahern193125d2015-08-13 14:59:10 -06001186}
1187
1188static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
1189{
1190 if (tb[IFLA_ADDRESS]) {
1191 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1192 return -EINVAL;
1193 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1194 return -EADDRNOTAVAIL;
1195 }
1196 return 0;
1197}
1198
1199static void vrf_dellink(struct net_device *dev, struct list_head *head)
1200{
David Ahern193125d2015-08-13 14:59:10 -06001201 unregister_netdevice_queue(dev, head);
1202}
1203
1204static int vrf_newlink(struct net *src_net, struct net_device *dev,
1205 struct nlattr *tb[], struct nlattr *data[])
1206{
1207 struct net_vrf *vrf = netdev_priv(dev);
David Ahern1aa6c4f2016-06-08 10:55:40 -07001208 int err;
David Ahern193125d2015-08-13 14:59:10 -06001209
1210 if (!data || !data[IFLA_VRF_TABLE])
1211 return -EINVAL;
1212
1213 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1214
David Ahern007979e2015-09-29 20:07:10 -07001215 dev->priv_flags |= IFF_L3MDEV_MASTER;
David Ahern193125d2015-08-13 14:59:10 -06001216
David Ahern1aa6c4f2016-06-08 10:55:40 -07001217 err = register_netdevice(dev);
1218 if (err)
1219 goto out;
1220
1221 if (add_fib_rules) {
1222 err = vrf_add_fib_rules(dev);
1223 if (err) {
1224 unregister_netdevice(dev);
1225 goto out;
1226 }
1227 add_fib_rules = false;
1228 }
1229
1230out:
1231 return err;
David Ahern193125d2015-08-13 14:59:10 -06001232}
1233
1234static size_t vrf_nl_getsize(const struct net_device *dev)
1235{
1236 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1237}
1238
1239static int vrf_fillinfo(struct sk_buff *skb,
1240 const struct net_device *dev)
1241{
1242 struct net_vrf *vrf = netdev_priv(dev);
1243
1244 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1245}
1246
David Ahern67eb0332016-02-02 07:43:45 -08001247static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1248 const struct net_device *slave_dev)
1249{
1250 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1251}
1252
1253static int vrf_fill_slave_info(struct sk_buff *skb,
1254 const struct net_device *vrf_dev,
1255 const struct net_device *slave_dev)
1256{
1257 struct net_vrf *vrf = netdev_priv(vrf_dev);
1258
1259 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1260 return -EMSGSIZE;
1261
1262 return 0;
1263}
1264
David Ahern193125d2015-08-13 14:59:10 -06001265static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1266 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1267};
1268
1269static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1270 .kind = DRV_NAME,
1271 .priv_size = sizeof(struct net_vrf),
1272
1273 .get_size = vrf_nl_getsize,
1274 .policy = vrf_nl_policy,
1275 .validate = vrf_validate,
1276 .fill_info = vrf_fillinfo,
1277
David Ahern67eb0332016-02-02 07:43:45 -08001278 .get_slave_size = vrf_get_slave_size,
1279 .fill_slave_info = vrf_fill_slave_info,
1280
David Ahern193125d2015-08-13 14:59:10 -06001281 .newlink = vrf_newlink,
1282 .dellink = vrf_dellink,
1283 .setup = vrf_setup,
1284 .maxtype = IFLA_VRF_MAX,
1285};
1286
1287static int vrf_device_event(struct notifier_block *unused,
1288 unsigned long event, void *ptr)
1289{
1290 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1291
1292 /* only care about unregister events to drop slave references */
1293 if (event == NETDEV_UNREGISTER) {
David Ahern193125d2015-08-13 14:59:10 -06001294 struct net_device *vrf_dev;
1295
David Ahernfee6d4c2015-10-05 08:51:24 -07001296 if (!netif_is_l3_slave(dev))
David Ahern193125d2015-08-13 14:59:10 -06001297 goto out;
1298
Nikolay Aleksandrov58aa9082015-08-18 20:28:04 +03001299 vrf_dev = netdev_master_upper_dev_get(dev);
1300 vrf_del_slave(vrf_dev, dev);
David Ahern193125d2015-08-13 14:59:10 -06001301 }
1302out:
1303 return NOTIFY_DONE;
1304}
1305
1306static struct notifier_block vrf_notifier_block __read_mostly = {
1307 .notifier_call = vrf_device_event,
1308};
1309
1310static int __init vrf_init_module(void)
1311{
1312 int rc;
1313
David Ahern193125d2015-08-13 14:59:10 -06001314 register_netdevice_notifier(&vrf_notifier_block);
1315
1316 rc = rtnl_link_register(&vrf_link_ops);
1317 if (rc < 0)
1318 goto error;
1319
1320 return 0;
1321
1322error:
1323 unregister_netdevice_notifier(&vrf_notifier_block);
David Ahern193125d2015-08-13 14:59:10 -06001324 return rc;
1325}
1326
David Ahern193125d2015-08-13 14:59:10 -06001327module_init(vrf_init_module);
David Ahern193125d2015-08-13 14:59:10 -06001328MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1329MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1330MODULE_LICENSE("GPL");
1331MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1332MODULE_VERSION(DRV_VERSION);