]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - drivers/net/vrf.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'iio-for-4.13b' of git://git.kernel.org/pub/scm/linux/kernel/git/jic23...
[mirror_ubuntu-artful-kernel.git] / drivers / net / vrf.c
CommitLineData
193125db
DA		 1/*
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>
8f58336d 30#include <net/arp.h>
193125db
DA		 31#include <net/ip.h>
32#include <net/ip_fib.h>
35402e31 33#include <net/ip6_fib.h>
193125db 34#include <net/ip6_route.h>
193125db
DA		 35#include <net/route.h>
36#include <net/addrconf.h>
ee15ee5d 37#include <net/l3mdev.h>
1aa6c4f6 38#include <net/fib_rules.h>
097d3c95 39#include <net/netns/generic.h>
193125db
DA		 40
41#define DRV_NAME "vrf"
42#define DRV_VERSION "1.0"
43
1aa6c4f6 44#define FIB_RULE_PREF 1000 /* default preference for FIB rules */
097d3c95
DA		 45
46static unsigned int vrf_net_id;
1aa6c4f6 47
ec539514 48struct net_vrf {
b0e95ccd 49 struct rtable __rcu *rth;
afe80a49 50 struct rtable __rcu *rth_local;
b0e95ccd 51 struct rt6_info __rcu *rt6;
b4869aa2 52 struct rt6_info __rcu *rt6_local;
ec539514
DA		 53 u32 tb_id;
54};
55
193125db
DA		 56struct pcpu_dstats {
57 u64 tx_pkts;
58 u64 tx_bytes;
59 u64 tx_drps;
60 u64 rx_pkts;
61 u64 rx_bytes;
afe80a49 62 u64 rx_drps;
193125db
DA		 63 struct u64_stats_sync syncp;
64};
65
afe80a49
DA		 66static void vrf_rx_stats(struct net_device *dev, int len)
67{
68 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
69
70 u64_stats_update_begin(&dstats->syncp);
71 dstats->rx_pkts++;
72 dstats->rx_bytes += len;
73 u64_stats_update_end(&dstats->syncp);
74}
75
57b8efa1
NA		 76static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
77{
78 vrf_dev->stats.tx_errors++;
79 kfree_skb(skb);
80}
81
bc1f4470 82static void vrf_get_stats64(struct net_device *dev,
83 struct rtnl_link_stats64 *stats)
193125db
DA		 84{
85 int i;
86
87 for_each_possible_cpu(i) {
88 const struct pcpu_dstats *dstats;
89 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
90 unsigned int start;
91
92 dstats = per_cpu_ptr(dev->dstats, i);
93 do {
94 start = u64_stats_fetch_begin_irq(&dstats->syncp);
95 tbytes = dstats->tx_bytes;
96 tpkts = dstats->tx_pkts;
97 tdrops = dstats->tx_drps;
98 rbytes = dstats->rx_bytes;
99 rpkts = dstats->rx_pkts;
100 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
101 stats->tx_bytes += tbytes;
102 stats->tx_packets += tpkts;
103 stats->tx_dropped += tdrops;
104 stats->rx_bytes += rbytes;
105 stats->rx_packets += rpkts;
106 }
193125db
DA		 107}
108
dcdd43c4
DA		 109/* by default VRF devices do not have a qdisc and are expected
110 * to be created with only a single queue.
111 */
112static bool qdisc_tx_is_default(const struct net_device *dev)
113{
114 struct netdev_queue *txq;
115 struct Qdisc *qdisc;
116
117 if (dev->num_tx_queues > 1)
118 return false;
119
120 txq = netdev_get_tx_queue(dev, 0);
121 qdisc = rcu_access_pointer(txq->qdisc);
122
123 return !qdisc->enqueue;
124}
125
afe80a49
DA		 126/* Local traffic destined to local address. Reinsert the packet to rx
127 * path, similar to loopback handling.
128 */
129static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
130 struct dst_entry *dst)
131{
132 int len = skb->len;
133
134 skb_orphan(skb);
135
136 skb_dst_set(skb, dst);
137 skb_dst_force(skb);
138
139 /* set pkt_type to avoid skb hitting packet taps twice -
140 * once on Tx and again in Rx processing
141 */
142 skb->pkt_type = PACKET_LOOPBACK;
143
144 skb->protocol = eth_type_trans(skb, dev);
145
146 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
147 vrf_rx_stats(dev, len);
148 else
149 this_cpu_inc(dev->dstats->rx_drps);
150
151 return NETDEV_TX_OK;
152}
153
35402e31 154#if IS_ENABLED(CONFIG_IPV6)
4c1feac5
DA		 155static int vrf_ip6_local_out(struct net *net, struct sock *sk,
156 struct sk_buff *skb)
157{
158 int err;
159
160 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
161 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
162
163 if (likely(err == 1))
164 err = dst_output(net, sk, skb);
165
166 return err;
167}
168
35402e31
DA		 169static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
170 struct net_device *dev)
171{
172 const struct ipv6hdr *iph = ipv6_hdr(skb);
173 struct net *net = dev_net(skb->dev);
174 struct flowi6 fl6 = {
175 /* needed to match OIF rule */
176 .flowi6_oif = dev->ifindex,
177 .flowi6_iif = LOOPBACK_IFINDEX,
178 .daddr = iph->daddr,
179 .saddr = iph->saddr,
180 .flowlabel = ip6_flowinfo(iph),
181 .flowi6_mark = skb->mark,
182 .flowi6_proto = iph->nexthdr,
c71ad3d4 183 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
35402e31
DA		 184 };
185 int ret = NET_XMIT_DROP;
186 struct dst_entry *dst;
187 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
188
189 dst = ip6_route_output(net, NULL, &fl6);
190 if (dst == dst_null)
191 goto err;
192
193 skb_dst_drop(skb);
b4869aa2
DA		 194
195 /* if dst.dev is loopback or the VRF device again this is locally
196 * originated traffic destined to a local address. Short circuit
197 * to Rx path using our local dst
198 */
199 if (dst->dev == net->loopback_dev || dst->dev == dev) {
200 struct net_vrf *vrf = netdev_priv(dev);
201 struct rt6_info *rt6_local;
202
203 /* release looked up dst and use cached local dst */
204 dst_release(dst);
205
206 rcu_read_lock();
207
208 rt6_local = rcu_dereference(vrf->rt6_local);
209 if (unlikely(!rt6_local)) {
210 rcu_read_unlock();
211 goto err;
212 }
213
214 /* Ordering issue: cached local dst is created on newlink
215 * before the IPv6 initialization. Using the local dst
216 * requires rt6i_idev to be set so make sure it is.
217 */
218 if (unlikely(!rt6_local->rt6i_idev)) {
219 rt6_local->rt6i_idev = in6_dev_get(dev);
220 if (!rt6_local->rt6i_idev) {
221 rcu_read_unlock();
222 goto err;
223 }
224 }
225
226 dst = &rt6_local->dst;
227 dst_hold(dst);
228
229 rcu_read_unlock();
230
231 return vrf_local_xmit(skb, dev, &rt6_local->dst);
232 }
233
35402e31
DA		 234 skb_dst_set(skb, dst);
235
911a66fb
DA		 236 /* strip the ethernet header added for pass through VRF device */
237 __skb_pull(skb, skb_network_offset(skb));
238
4c1feac5 239 ret = vrf_ip6_local_out(net, skb->sk, skb);
35402e31
DA		 240 if (unlikely(net_xmit_eval(ret)))
241 dev->stats.tx_errors++;
242 else
243 ret = NET_XMIT_SUCCESS;
244
245 return ret;
246err:
247 vrf_tx_error(dev, skb);
248 return NET_XMIT_DROP;
249}
250#else
193125db
DA		 251static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
252 struct net_device *dev)
253{
57b8efa1
NA		 254 vrf_tx_error(dev, skb);
255 return NET_XMIT_DROP;
193125db 256}
35402e31 257#endif
193125db 258
ebfc102c
DA		 259/* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
260static int vrf_ip_local_out(struct net *net, struct sock *sk,
261 struct sk_buff *skb)
262{
263 int err;
264
265 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
266 skb, NULL, skb_dst(skb)->dev, dst_output);
267 if (likely(err == 1))
268 err = dst_output(net, sk, skb);
269
270 return err;
271}
272
193125db
DA		 273static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
274 struct net_device *vrf_dev)
275{
276 struct iphdr *ip4h = ip_hdr(skb);
277 int ret = NET_XMIT_DROP;
278 struct flowi4 fl4 = {
279 /* needed to match OIF rule */
280 .flowi4_oif = vrf_dev->ifindex,
281 .flowi4_iif = LOOPBACK_IFINDEX,
282 .flowi4_tos = RT_TOS(ip4h->tos),
c71ad3d4 283 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
7a18c5b9 284 .flowi4_proto = ip4h->protocol,
193125db 285 .daddr = ip4h->daddr,
7a18c5b9 286 .saddr = ip4h->saddr,
193125db 287 };
911a66fb
DA		 288 struct net *net = dev_net(vrf_dev);
289 struct rtable *rt;
290
291 rt = ip_route_output_flow(net, &fl4, NULL);
292 if (IS_ERR(rt))
293 goto err;
193125db 294
911a66fb 295 skb_dst_drop(skb);
afe80a49
DA		 296
297 /* if dst.dev is loopback or the VRF device again this is locally
298 * originated traffic destined to a local address. Short circuit
299 * to Rx path using our local dst
300 */
301 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) {
302 struct net_vrf *vrf = netdev_priv(vrf_dev);
303 struct rtable *rth_local;
304 struct dst_entry *dst = NULL;
305
306 ip_rt_put(rt);
307
308 rcu_read_lock();
309
310 rth_local = rcu_dereference(vrf->rth_local);
311 if (likely(rth_local)) {
312 dst = &rth_local->dst;
313 dst_hold(dst);
314 }
315
316 rcu_read_unlock();
317
318 if (unlikely(!dst))
319 goto err;
320
321 return vrf_local_xmit(skb, vrf_dev, dst);
322 }
323
911a66fb
DA		 324 skb_dst_set(skb, &rt->dst);
325
326 /* strip the ethernet header added for pass through VRF device */
327 __skb_pull(skb, skb_network_offset(skb));
193125db
DA		 328
329 if (!ip4h->saddr) {
330 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
331 RT_SCOPE_LINK);
332 }
333
ebfc102c 334 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
193125db
DA		 335 if (unlikely(net_xmit_eval(ret)))
336 vrf_dev->stats.tx_errors++;
337 else
338 ret = NET_XMIT_SUCCESS;
339
340out:
341 return ret;
342err:
57b8efa1 343 vrf_tx_error(vrf_dev, skb);
193125db
DA		 344 goto out;
345}
346
347static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
348{
349 switch (skb->protocol) {
350 case htons(ETH_P_IP):
351 return vrf_process_v4_outbound(skb, dev);
352 case htons(ETH_P_IPV6):
353 return vrf_process_v6_outbound(skb, dev);
354 default:
57b8efa1 355 vrf_tx_error(dev, skb);
193125db
DA		 356 return NET_XMIT_DROP;
357 }
358}
359
360static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
361{
f7887d40 362 int len = skb->len;
193125db
DA		 363 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
364
365 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
366 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
367
368 u64_stats_update_begin(&dstats->syncp);
369 dstats->tx_pkts++;
f7887d40 370 dstats->tx_bytes += len;
193125db
DA		 371 u64_stats_update_end(&dstats->syncp);
372 } else {
373 this_cpu_inc(dev->dstats->tx_drps);
374 }
375
376 return ret;
377}
378
dcdd43c4
DA		 379static int vrf_finish_direct(struct net *net, struct sock *sk,
380 struct sk_buff *skb)
381{
382 struct net_device *vrf_dev = skb->dev;
383
384 if (!list_empty(&vrf_dev->ptype_all) &&
385 likely(skb_headroom(skb) >= ETH_HLEN)) {
386 struct ethhdr *eth = (struct ethhdr *)skb_push(skb, ETH_HLEN);
387
388 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
389 eth_zero_addr(eth->h_dest);
390 eth->h_proto = skb->protocol;
391
392 rcu_read_lock_bh();
393 dev_queue_xmit_nit(skb, vrf_dev);
394 rcu_read_unlock_bh();
395
396 skb_pull(skb, ETH_HLEN);
397 }
398
399 return 1;
400}
401
35402e31 402#if IS_ENABLED(CONFIG_IPV6)
35402e31
DA		 403/* modelled after ip6_finish_output2 */
404static int vrf_finish_output6(struct net *net, struct sock *sk,
405 struct sk_buff *skb)
406{
407 struct dst_entry *dst = skb_dst(skb);
408 struct net_device *dev = dst->dev;
409 struct neighbour *neigh;
410 struct in6_addr *nexthop;
411 int ret;
412
eb63ecc1
DA		 413 nf_reset(skb);
414
35402e31
DA		 415 skb->protocol = htons(ETH_P_IPV6);
416 skb->dev = dev;
417
418 rcu_read_lock_bh();
419 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
420 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
421 if (unlikely(!neigh))
422 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
423 if (!IS_ERR(neigh)) {
4ff06203 424 sock_confirm_neigh(skb, neigh);
c16ec185 425 ret = neigh_output(neigh, skb);
35402e31
DA		 426 rcu_read_unlock_bh();
427 return ret;
428 }
429 rcu_read_unlock_bh();
430
431 IP6_INC_STATS(dev_net(dst->dev),
432 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
433 kfree_skb(skb);
434 return -EINVAL;
435}
436
437/* modelled after ip6_output */
438static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
439{
440 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
441 net, sk, skb, NULL, skb_dst(skb)->dev,
442 vrf_finish_output6,
443 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
444}
445
4c1feac5
DA		 446/* set dst on skb to send packet to us via dev_xmit path. Allows
447 * packet to go through device based features such as qdisc, netfilter
448 * hooks and packet sockets with skb->dev set to vrf device.
449 */
a9ec54d1
DA		 450static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
451 struct sk_buff *skb)
4c1feac5
DA		 452{
453 struct net_vrf *vrf = netdev_priv(vrf_dev);
454 struct dst_entry *dst = NULL;
455 struct rt6_info *rt6;
456
4c1feac5
DA		 457 rcu_read_lock();
458
459 rt6 = rcu_dereference(vrf->rt6);
460 if (likely(rt6)) {
461 dst = &rt6->dst;
462 dst_hold(dst);
463 }
464
465 rcu_read_unlock();
466
467 if (unlikely(!dst)) {
468 vrf_tx_error(vrf_dev, skb);
469 return NULL;
470 }
471
472 skb_dst_drop(skb);
473 skb_dst_set(skb, dst);
474
475 return skb;
476}
477
a9ec54d1
DA		 478static int vrf_output6_direct(struct net *net, struct sock *sk,
479 struct sk_buff *skb)
480{
481 skb->protocol = htons(ETH_P_IPV6);
482
483 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
484 net, sk, skb, NULL, skb->dev,
485 vrf_finish_direct,
486 !(IPCB(skb)->flags & IPSKB_REROUTED));
487}
488
489static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
490 struct sock *sk,
491 struct sk_buff *skb)
492{
493 struct net *net = dev_net(vrf_dev);
494 int err;
495
496 skb->dev = vrf_dev;
497
498 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
499 skb, NULL, vrf_dev, vrf_output6_direct);
500
501 if (likely(err == 1))
502 err = vrf_output6_direct(net, sk, skb);
503
504 /* reset skb device */
505 if (likely(err == 1))
506 nf_reset(skb);
507 else
508 skb = NULL;
509
510 return skb;
511}
512
513static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
514 struct sock *sk,
515 struct sk_buff *skb)
516{
517 /* don't divert link scope packets */
518 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
519 return skb;
520
521 if (qdisc_tx_is_default(vrf_dev))
522 return vrf_ip6_out_direct(vrf_dev, sk, skb);
523
524 return vrf_ip6_out_redirect(vrf_dev, skb);
525}
526
b0e95ccd 527/* holding rtnl */
810e530b 528static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
35402e31 529{
b0e95ccd 530 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
b4869aa2 531 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local);
810e530b
DA		 532 struct net *net = dev_net(dev);
533 struct dst_entry *dst;
b0e95ccd 534
b4869aa2
DA		 535 RCU_INIT_POINTER(vrf->rt6, NULL);
536 RCU_INIT_POINTER(vrf->rt6_local, NULL);
537 synchronize_rcu();
b0e95ccd 538
810e530b
DA		 539 /* move dev in dst's to loopback so this VRF device can be deleted
540 * - based on dst_ifdown
541 */
542 if (rt6) {
543 dst = &rt6->dst;
544 dev_put(dst->dev);
545 dst->dev = net->loopback_dev;
546 dev_hold(dst->dev);
547 dst_release(dst);
548 }
b4869aa2
DA		 549
550 if (rt6_local) {
3dc857f0 551 if (rt6_local->rt6i_idev) {
b4869aa2 552 in6_dev_put(rt6_local->rt6i_idev);
3dc857f0
DA		 553 rt6_local->rt6i_idev = NULL;
554 }
b4869aa2 555
810e530b
DA		 556 dst = &rt6_local->dst;
557 dev_put(dst->dev);
558 dst->dev = net->loopback_dev;
559 dev_hold(dst->dev);
560 dst_release(dst);
b4869aa2 561 }
35402e31
DA		 562}
563
564static int vrf_rt6_create(struct net_device *dev)
565{
b4869aa2 566 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE;
35402e31 567 struct net_vrf *vrf = netdev_priv(dev);
9ab179d8 568 struct net *net = dev_net(dev);
b3b4663c 569 struct fib6_table *rt6i_table;
b4869aa2 570 struct rt6_info *rt6, *rt6_local;
35402e31
DA		 571 int rc = -ENOMEM;
572
e4348637
DA		 573 /* IPv6 can be CONFIG enabled and then disabled runtime */
574 if (!ipv6_mod_enabled())
575 return 0;
576
b3b4663c
DA		 577 rt6i_table = fib6_new_table(net, vrf->tb_id);
578 if (!rt6i_table)
579 goto out;
580
b4869aa2
DA		 581 /* create a dst for routing packets out a VRF device */
582 rt6 = ip6_dst_alloc(net, dev, flags);
35402e31
DA		 583 if (!rt6)
584 goto out;
585
9ab179d8 586 dst_hold(&rt6->dst);
b3b4663c
DA		 587
588 rt6->rt6i_table = rt6i_table;
589 rt6->dst.output = vrf_output6;
b4869aa2
DA		 590
591 /* create a dst for local routing - packets sent locally
592 * to local address via the VRF device as a loopback
593 */
594 rt6_local = ip6_dst_alloc(net, dev, flags);
595 if (!rt6_local) {
596 dst_release(&rt6->dst);
597 goto out;
598 }
599
600 dst_hold(&rt6_local->dst);
601
602 rt6_local->rt6i_idev = in6_dev_get(dev);
603 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL;
604 rt6_local->rt6i_table = rt6i_table;
605 rt6_local->dst.input = ip6_input;
606
b0e95ccd 607 rcu_assign_pointer(vrf->rt6, rt6);
b4869aa2 608 rcu_assign_pointer(vrf->rt6_local, rt6_local);
b0e95ccd 609
35402e31
DA		 610 rc = 0;
611out:
612 return rc;
613}
614#else
4c1feac5
DA		 615static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
616 struct sock *sk,
617 struct sk_buff *skb)
618{
619 return skb;
620}
621
810e530b 622static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
35402e31
DA		 623{
624}
625
626static int vrf_rt6_create(struct net_device *dev)
627{
628 return 0;
629}
630#endif
631
8f58336d 632/* modelled after ip_finish_output2 */
0c4b51f0 633static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
193125db 634{
8f58336d
DA		 635 struct dst_entry *dst = skb_dst(skb);
636 struct rtable *rt = (struct rtable *)dst;
637 struct net_device *dev = dst->dev;
638 unsigned int hh_len = LL_RESERVED_SPACE(dev);
639 struct neighbour *neigh;
640 u32 nexthop;
641 int ret = -EINVAL;
642
eb63ecc1
DA		 643 nf_reset(skb);
644
8f58336d
DA		 645 /* Be paranoid, rather than too clever. */
646 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
647 struct sk_buff *skb2;
648
649 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
650 if (!skb2) {
651 ret = -ENOMEM;
652 goto err;
653 }
654 if (skb->sk)
655 skb_set_owner_w(skb2, skb->sk);
656
657 consume_skb(skb);
658 skb = skb2;
659 }
660
661 rcu_read_lock_bh();
662
663 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
664 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
665 if (unlikely(!neigh))
666 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
4ff06203
JA		 667 if (!IS_ERR(neigh)) {
668 sock_confirm_neigh(skb, neigh);
c16ec185 669 ret = neigh_output(neigh, skb);
4ff06203 670 }
8f58336d
DA		 671
672 rcu_read_unlock_bh();
673err:
674 if (unlikely(ret < 0))
675 vrf_tx_error(skb->dev, skb);
676 return ret;
193125db
DA		 677}
678
ede2059d 679static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
193125db
DA		 680{
681 struct net_device *dev = skb_dst(skb)->dev;
682
29a26a56 683 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
193125db
DA		 684
685 skb->dev = dev;
686 skb->protocol = htons(ETH_P_IP);
687
29a26a56
EB		 688 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
689 net, sk, skb, NULL, dev,
8f58336d 690 vrf_finish_output,
193125db
DA		 691 !(IPCB(skb)->flags & IPSKB_REROUTED));
692}
693
ebfc102c
DA		 694/* set dst on skb to send packet to us via dev_xmit path. Allows
695 * packet to go through device based features such as qdisc, netfilter
696 * hooks and packet sockets with skb->dev set to vrf device.
697 */
dcdd43c4
DA		 698static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
699 struct sk_buff *skb)
ebfc102c
DA		 700{
701 struct net_vrf *vrf = netdev_priv(vrf_dev);
702 struct dst_entry *dst = NULL;
703 struct rtable *rth;
704
705 rcu_read_lock();
706
707 rth = rcu_dereference(vrf->rth);
708 if (likely(rth)) {
709 dst = &rth->dst;
710 dst_hold(dst);
711 }
712
713 rcu_read_unlock();
714
715 if (unlikely(!dst)) {
716 vrf_tx_error(vrf_dev, skb);
717 return NULL;
718 }
719
720 skb_dst_drop(skb);
721 skb_dst_set(skb, dst);
722
723 return skb;
724}
725
dcdd43c4
DA		 726static int vrf_output_direct(struct net *net, struct sock *sk,
727 struct sk_buff *skb)
728{
729 skb->protocol = htons(ETH_P_IP);
730
731 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
732 net, sk, skb, NULL, skb->dev,
733 vrf_finish_direct,
734 !(IPCB(skb)->flags & IPSKB_REROUTED));
735}
736
737static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
738 struct sock *sk,
739 struct sk_buff *skb)
740{
741 struct net *net = dev_net(vrf_dev);
742 int err;
743
744 skb->dev = vrf_dev;
745
746 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
747 skb, NULL, vrf_dev, vrf_output_direct);
748
749 if (likely(err == 1))
750 err = vrf_output_direct(net, sk, skb);
751
752 /* reset skb device */
753 if (likely(err == 1))
754 nf_reset(skb);
755 else
756 skb = NULL;
757
758 return skb;
759}
760
761static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
762 struct sock *sk,
763 struct sk_buff *skb)
764{
765 /* don't divert multicast */
766 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
767 return skb;
768
769 if (qdisc_tx_is_default(vrf_dev))
770 return vrf_ip_out_direct(vrf_dev, sk, skb);
771
772 return vrf_ip_out_redirect(vrf_dev, skb);
773}
774
ebfc102c
DA		 775/* called with rcu lock held */
776static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
777 struct sock *sk,
778 struct sk_buff *skb,
779 u16 proto)
780{
781 switch (proto) {
782 case AF_INET:
783 return vrf_ip_out(vrf_dev, sk, skb);
4c1feac5
DA		 784 case AF_INET6:
785 return vrf_ip6_out(vrf_dev, sk, skb);
ebfc102c
DA		 786 }
787
788 return skb;
789}
790
b0e95ccd 791/* holding rtnl */
810e530b 792static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
193125db 793{
b0e95ccd 794 struct rtable *rth = rtnl_dereference(vrf->rth);
afe80a49 795 struct rtable *rth_local = rtnl_dereference(vrf->rth_local);
810e530b
DA		 796 struct net *net = dev_net(dev);
797 struct dst_entry *dst;
b0e95ccd 798
afe80a49
DA		 799 RCU_INIT_POINTER(vrf->rth, NULL);
800 RCU_INIT_POINTER(vrf->rth_local, NULL);
801 synchronize_rcu();
193125db 802
810e530b
DA		 803 /* move dev in dst's to loopback so this VRF device can be deleted
804 * - based on dst_ifdown
805 */
806 if (rth) {
807 dst = &rth->dst;
808 dev_put(dst->dev);
809 dst->dev = net->loopback_dev;
810 dev_hold(dst->dev);
811 dst_release(dst);
812 }
afe80a49 813
810e530b
DA		 814 if (rth_local) {
815 dst = &rth_local->dst;
816 dev_put(dst->dev);
817 dst->dev = net->loopback_dev;
818 dev_hold(dst->dev);
819 dst_release(dst);
820 }
193125db
DA		 821}
822
b0e95ccd 823static int vrf_rtable_create(struct net_device *dev)
193125db 824{
b7503e0c 825 struct net_vrf *vrf = netdev_priv(dev);
afe80a49 826 struct rtable *rth, *rth_local;
193125db 827
b3b4663c 828 if (!fib_new_table(dev_net(dev), vrf->tb_id))
b0e95ccd 829 return -ENOMEM;
b3b4663c 830
afe80a49 831 /* create a dst for routing packets out through a VRF device */
9ab179d8 832 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
b0e95ccd
DA		 833 if (!rth)
834 return -ENOMEM;
193125db 835
afe80a49
DA		 836 /* create a dst for local ingress routing - packets sent locally
837 * to local address via the VRF device as a loopback
838 */
839 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0);
840 if (!rth_local) {
841 dst_release(&rth->dst);
842 return -ENOMEM;
843 }
844
b0e95ccd
DA		 845 rth->dst.output = vrf_output;
846 rth->rt_table_id = vrf->tb_id;
847
afe80a49
DA		 848 rth_local->rt_table_id = vrf->tb_id;
849
b0e95ccd 850 rcu_assign_pointer(vrf->rth, rth);
afe80a49 851 rcu_assign_pointer(vrf->rth_local, rth_local);
b0e95ccd
DA		 852
853 return 0;
193125db
DA		 854}
855
856/**************************** device handling ********************/
857
858/* cycle interface to flush neighbor cache and move routes across tables */
859static void cycle_netdev(struct net_device *dev)
860{
861 unsigned int flags = dev->flags;
862 int ret;
863
864 if (!netif_running(dev))
865 return;
866
867 ret = dev_change_flags(dev, flags & ~IFF_UP);
868 if (ret >= 0)
869 ret = dev_change_flags(dev, flags);
870
871 if (ret < 0) {
872 netdev_err(dev,
873 "Failed to cycle device %s; route tables might be wrong!\n",
874 dev->name);
875 }
876}
877
193125db
DA		 878static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
879{
bad53162 880 int ret;
193125db 881
26d31ac1
DA		 882 /* do not allow loopback device to be enslaved to a VRF.
883 * The vrf device acts as the loopback for the vrf.
884 */
885 if (port_dev == dev_net(dev)->loopback_dev)
886 return -EOPNOTSUPP;
887
fdeea7be 888 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
29bf24af 889 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
193125db 890 if (ret < 0)
fdeea7be 891 goto err;
193125db 892
193125db
DA		 893 cycle_netdev(port_dev);
894
895 return 0;
fdeea7be
IS		 896
897err:
898 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
899 return ret;
193125db
DA		 900}
901
902static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
903{
fee6d4c7 904 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
193125db
DA		 905 return -EINVAL;
906
907 return do_vrf_add_slave(dev, port_dev);
908}
909
910/* inverse of do_vrf_add_slave */
911static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
912{
193125db 913 netdev_upper_dev_unlink(port_dev, dev);
fee6d4c7 914 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
193125db 915
193125db
DA		 916 cycle_netdev(port_dev);
917
193125db
DA		 918 return 0;
919}
920
921static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
922{
193125db
DA		 923 return do_vrf_del_slave(dev, port_dev);
924}
925
926static void vrf_dev_uninit(struct net_device *dev)
927{
928 struct net_vrf *vrf = netdev_priv(dev);
bad53162
NA		 929 struct net_device *port_dev;
930 struct list_head *iter;
193125db 931
810e530b
DA		 932 vrf_rtable_release(dev, vrf);
933 vrf_rt6_release(dev, vrf);
193125db 934
bad53162
NA		 935 netdev_for_each_lower_dev(dev, port_dev, iter)
936 vrf_del_slave(dev, port_dev);
193125db 937
3a4a27d3 938 free_percpu(dev->dstats);
193125db
DA		 939 dev->dstats = NULL;
940}
941
942static int vrf_dev_init(struct net_device *dev)
943{
944 struct net_vrf *vrf = netdev_priv(dev);
945
193125db
DA		 946 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
947 if (!dev->dstats)
948 goto out_nomem;
949
950 /* create the default dst which points back to us */
b0e95ccd 951 if (vrf_rtable_create(dev) != 0)
193125db
DA		 952 goto out_stats;
953
35402e31
DA		 954 if (vrf_rt6_create(dev) != 0)
955 goto out_rth;
956
193125db
DA		 957 dev->flags = IFF_MASTER | IFF_NOARP;
958
b87ab6b8
DA		 959 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
960 dev->mtu = 64 * 1024;
961
962 /* similarly, oper state is irrelevant; set to up to avoid confusion */
963 dev->operstate = IF_OPER_UP;
78e7a2ae 964 netdev_lockdep_set_classes(dev);
193125db
DA		 965 return 0;
966
35402e31 967out_rth:
810e530b 968 vrf_rtable_release(dev, vrf);
193125db
DA		 969out_stats:
970 free_percpu(dev->dstats);
971 dev->dstats = NULL;
972out_nomem:
973 return -ENOMEM;
974}
975
976static const struct net_device_ops vrf_netdev_ops = {
977 .ndo_init = vrf_dev_init,
978 .ndo_uninit = vrf_dev_uninit,
979 .ndo_start_xmit = vrf_xmit,
980 .ndo_get_stats64 = vrf_get_stats64,
981 .ndo_add_slave = vrf_add_slave,
982 .ndo_del_slave = vrf_del_slave,
983};
984
ee15ee5d
DA		 985static u32 vrf_fib_table(const struct net_device *dev)
986{
987 struct net_vrf *vrf = netdev_priv(dev);
988
989 return vrf->tb_id;
990}
991
73e20b76
DA		 992static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
993{
1a4a5bf5 994 kfree_skb(skb);
73e20b76
DA		 995 return 0;
996}
997
998static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
999 struct sk_buff *skb,
1000 struct net_device *dev)
1001{
1002 struct net *net = dev_net(dev);
1003
1a4a5bf5 1004 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
73e20b76
DA		 1005 skb = NULL; /* kfree_skb(skb) handled by nf code */
1006
1007 return skb;
1008}
1009
74b20582
DA		 1010#if IS_ENABLED(CONFIG_IPV6)
1011/* neighbor handling is done with actual device; do not want
1012 * to flip skb->dev for those ndisc packets. This really fails
1013 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1014 * a start.
1015 */
1016static bool ipv6_ndisc_frame(const struct sk_buff *skb)
1017{
1018 const struct ipv6hdr *iph = ipv6_hdr(skb);
1019 bool rc = false;
1020
1021 if (iph->nexthdr == NEXTHDR_ICMP) {
1022 const struct icmp6hdr *icmph;
1023 struct icmp6hdr _icmph;
1024
1025 icmph = skb_header_pointer(skb, sizeof(*iph),
1026 sizeof(_icmph), &_icmph);
1027 if (!icmph)
1028 goto out;
1029
1030 switch (icmph->icmp6_type) {
1031 case NDISC_ROUTER_SOLICITATION:
1032 case NDISC_ROUTER_ADVERTISEMENT:
1033 case NDISC_NEIGHBOUR_SOLICITATION:
1034 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1035 case NDISC_REDIRECT:
1036 rc = true;
1037 break;
1038 }
1039 }
1040
1041out:
1042 return rc;
1043}
1044
9ff74384
DA		 1045static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
1046 const struct net_device *dev,
1047 struct flowi6 *fl6,
1048 int ifindex,
1049 int flags)
1050{
1051 struct net_vrf *vrf = netdev_priv(dev);
1052 struct fib6_table *table = NULL;
1053 struct rt6_info *rt6;
1054
1055 rcu_read_lock();
1056
1057 /* fib6_table does not have a refcnt and can not be freed */
1058 rt6 = rcu_dereference(vrf->rt6);
1059 if (likely(rt6))
1060 table = rt6->rt6i_table;
1061
1062 rcu_read_unlock();
1063
1064 if (!table)
1065 return NULL;
1066
1067 return ip6_pol_route(net, table, ifindex, fl6, flags);
1068}
1069
1070static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1071 int ifindex)
1072{
1073 const struct ipv6hdr *iph = ipv6_hdr(skb);
1074 struct flowi6 fl6 = {
1075 .daddr = iph->daddr,
1076 .saddr = iph->saddr,
1077 .flowlabel = ip6_flowinfo(iph),
1078 .flowi6_mark = skb->mark,
1079 .flowi6_proto = iph->nexthdr,
1080 .flowi6_iif = ifindex,
1081 };
1082 struct net *net = dev_net(vrf_dev);
1083 struct rt6_info *rt6;
1084
1085 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
1086 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1087 if (unlikely(!rt6))
1088 return;
1089
1090 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1091 return;
1092
1093 skb_dst_set(skb, &rt6->dst);
1094}
1095
74b20582
DA		 1096static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1097 struct sk_buff *skb)
1098{
9ff74384
DA		 1099 int orig_iif = skb->skb_iif;
1100 bool need_strict;
1101
b4869aa2
DA		 1102 /* loopback traffic; do not push through packet taps again.
1103 * Reset pkt_type for upper layers to process skb
1104 */
1105 if (skb->pkt_type == PACKET_LOOPBACK) {
1106 skb->dev = vrf_dev;
1107 skb->skb_iif = vrf_dev->ifindex;
a04a480d 1108 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
b4869aa2
DA		 1109 skb->pkt_type = PACKET_HOST;
1110 goto out;
1111 }
1112
9ff74384
DA		 1113 /* if packet is NDISC or addressed to multicast or link-local
1114 * then keep the ingress interface
1115 */
1116 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1117 if (!ipv6_ndisc_frame(skb) && !need_strict) {
926d93a3 1118 vrf_rx_stats(vrf_dev, skb->len);
74b20582
DA		 1119 skb->dev = vrf_dev;
1120 skb->skb_iif = vrf_dev->ifindex;
1121
a9ec54d1
DA		 1122 if (!list_empty(&vrf_dev->ptype_all)) {
1123 skb_push(skb, skb->mac_len);
1124 dev_queue_xmit_nit(skb, vrf_dev);
1125 skb_pull(skb, skb->mac_len);
1126 }
74b20582
DA		 1127
1128 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1129 }
1130
9ff74384
DA		 1131 if (need_strict)
1132 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1133
73e20b76 1134 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
b4869aa2 1135out:
74b20582
DA		 1136 return skb;
1137}
1138
1139#else
1140static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1141 struct sk_buff *skb)
1142{
1143 return skb;
1144}
1145#endif
1146
1147static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1148 struct sk_buff *skb)
1149{
1150 skb->dev = vrf_dev;
1151 skb->skb_iif = vrf_dev->ifindex;
a04a480d 1152 IPCB(skb)->flags |= IPSKB_L3SLAVE;
74b20582 1153
e58e4159
DA		 1154 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1155 goto out;
1156
afe80a49
DA		 1157 /* loopback traffic; do not push through packet taps again.
1158 * Reset pkt_type for upper layers to process skb
1159 */
1160 if (skb->pkt_type == PACKET_LOOPBACK) {
1161 skb->pkt_type = PACKET_HOST;
1162 goto out;
1163 }
1164
926d93a3
DA		 1165 vrf_rx_stats(vrf_dev, skb->len);
1166
dcdd43c4
DA		 1167 if (!list_empty(&vrf_dev->ptype_all)) {
1168 skb_push(skb, skb->mac_len);
1169 dev_queue_xmit_nit(skb, vrf_dev);
1170 skb_pull(skb, skb->mac_len);
1171 }
74b20582 1172
73e20b76 1173 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
afe80a49 1174out:
74b20582
DA		 1175 return skb;
1176}
1177
1178/* called with rcu lock held */
1179static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1180 struct sk_buff *skb,
1181 u16 proto)
1182{
1183 switch (proto) {
1184 case AF_INET:
1185 return vrf_ip_rcv(vrf_dev, skb);
1186 case AF_INET6:
1187 return vrf_ip6_rcv(vrf_dev, skb);
1188 }
1189
1190 return skb;
1191}
1192
35402e31 1193#if IS_ENABLED(CONFIG_IPV6)
4c1feac5
DA		 1194/* send to link-local or multicast address via interface enslaved to
1195 * VRF device. Force lookup to VRF table without changing flow struct
1196 */
1197static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1198 struct flowi6 *fl6)
35402e31 1199{
9ff74384 1200 struct net *net = dev_net(dev);
4c1feac5 1201 int flags = RT6_LOOKUP_F_IFACE;
b0e95ccd 1202 struct dst_entry *dst = NULL;
9ff74384 1203 struct rt6_info *rt;
35402e31 1204
4c1feac5
DA		 1205 /* VRF device does not have a link-local address and
1206 * sending packets to link-local or mcast addresses over
1207 * a VRF device does not make sense
1208 */
1209 if (fl6->flowi6_oif == dev->ifindex) {
1210 dst = &net->ipv6.ip6_null_entry->dst;
1211 dst_hold(dst);
1212 return dst;
35402e31
DA		 1213 }
1214
4c1feac5
DA		 1215 if (!ipv6_addr_any(&fl6->saddr))
1216 flags |= RT6_LOOKUP_F_HAS_SADDR;
1217
1218 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1219 if (rt)
1220 dst = &rt->dst;
9ff74384 1221
b0e95ccd 1222 return dst;
35402e31
DA		 1223}
1224#endif
1225
ee15ee5d
DA		 1226static const struct l3mdev_ops vrf_l3mdev_ops = {
1227 .l3mdev_fib_table = vrf_fib_table,
74b20582 1228 .l3mdev_l3_rcv = vrf_l3_rcv,
ebfc102c 1229 .l3mdev_l3_out = vrf_l3_out,
35402e31 1230#if IS_ENABLED(CONFIG_IPV6)
4c1feac5 1231 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
35402e31 1232#endif
ee15ee5d
DA		 1233};
1234
193125db
DA		 1235static void vrf_get_drvinfo(struct net_device *dev,
1236 struct ethtool_drvinfo *info)
1237{
1238 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1239 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1240}
1241
1242static const struct ethtool_ops vrf_ethtool_ops = {
1243 .get_drvinfo = vrf_get_drvinfo,
1244};
1245
1aa6c4f6
DA		 1246static inline size_t vrf_fib_rule_nl_size(void)
1247{
1248 size_t sz;
1249
1250 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1251 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1252 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1253
1254 return sz;
1255}
1256
1257static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1258{
1259 struct fib_rule_hdr *frh;
1260 struct nlmsghdr *nlh;
1261 struct sk_buff *skb;
1262 int err;
1263
e4348637
DA		 1264 if (family == AF_INET6 && !ipv6_mod_enabled())
1265 return 0;
1266
1aa6c4f6
DA		 1267 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1268 if (!skb)
1269 return -ENOMEM;
1270
1271 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1272 if (!nlh)
1273 goto nla_put_failure;
1274
1275 /* rule only needs to appear once */
426c87ca 1276 nlh->nlmsg_flags |= NLM_F_EXCL;
1aa6c4f6
DA		 1277
1278 frh = nlmsg_data(nlh);
1279 memset(frh, 0, sizeof(*frh));
1280 frh->family = family;
1281 frh->action = FR_ACT_TO_TBL;
1282
1283 if (nla_put_u32(skb, FRA_L3MDEV, 1))
1284 goto nla_put_failure;
1285
1286 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1287 goto nla_put_failure;
1288
1289 nlmsg_end(skb, nlh);
1290
1291 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1292 skb->sk = dev_net(dev)->rtnl;
1293 if (add_it) {
c21ef3e3 1294 err = fib_nl_newrule(skb, nlh, NULL);
1aa6c4f6
DA		 1295 if (err == -EEXIST)
1296 err = 0;
1297 } else {
c21ef3e3 1298 err = fib_nl_delrule(skb, nlh, NULL);
1aa6c4f6
DA		 1299 if (err == -ENOENT)
1300 err = 0;
1301 }
1302 nlmsg_free(skb);
1303
1304 return err;
1305
1306nla_put_failure:
1307 nlmsg_free(skb);
1308
1309 return -EMSGSIZE;
1310}
1311
1312static int vrf_add_fib_rules(const struct net_device *dev)
1313{
1314 int err;
1315
1316 err = vrf_fib_rule(dev, AF_INET, true);
1317 if (err < 0)
1318 goto out_err;
1319
1320 err = vrf_fib_rule(dev, AF_INET6, true);
1321 if (err < 0)
1322 goto ipv6_err;
1323
e58e4159
DA		 1324#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1325 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1326 if (err < 0)
1327 goto ipmr_err;
1328#endif
1329
1aa6c4f6
DA		 1330 return 0;
1331
e58e4159
DA		 1332#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1333ipmr_err:
1334 vrf_fib_rule(dev, AF_INET6, false);
1335#endif
1336
1aa6c4f6
DA		 1337ipv6_err:
1338 vrf_fib_rule(dev, AF_INET, false);
1339
1340out_err:
1341 netdev_err(dev, "Failed to add FIB rules.\n");
1342 return err;
1343}
1344
193125db
DA		 1345static void vrf_setup(struct net_device *dev)
1346{
1347 ether_setup(dev);
1348
1349 /* Initialize the device structure. */
1350 dev->netdev_ops = &vrf_netdev_ops;
ee15ee5d 1351 dev->l3mdev_ops = &vrf_l3mdev_ops;
193125db 1352 dev->ethtool_ops = &vrf_ethtool_ops;
cf124db5 1353 dev->needs_free_netdev = true;
193125db
DA		 1354
1355 /* Fill in device structure with ethernet-generic values. */
1356 eth_hw_addr_random(dev);
1357
1358 /* don't acquire vrf device's netif_tx_lock when transmitting */
1359 dev->features |= NETIF_F_LLTX;
1360
1361 /* don't allow vrf devices to change network namespaces. */
1362 dev->features |= NETIF_F_NETNS_LOCAL;
7889681f
DA		 1363
1364 /* does not make sense for a VLAN to be added to a vrf device */
1365 dev->features |= NETIF_F_VLAN_CHALLENGED;
1366
1367 /* enable offload features */
1368 dev->features |= NETIF_F_GSO_SOFTWARE;
1369 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1370 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1371
1372 dev->hw_features = dev->features;
1373 dev->hw_enc_features = dev->features;
1374
1375 /* default to no qdisc; user can add if desired */
1376 dev->priv_flags |= IFF_NO_QUEUE;
193125db
DA		 1377}
1378
1379static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
1380{
1381 if (tb[IFLA_ADDRESS]) {
1382 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1383 return -EINVAL;
1384 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1385 return -EADDRNOTAVAIL;
1386 }
1387 return 0;
1388}
1389
1390static void vrf_dellink(struct net_device *dev, struct list_head *head)
1391{
193125db
DA		 1392 unregister_netdevice_queue(dev, head);
1393}
1394
1395static int vrf_newlink(struct net *src_net, struct net_device *dev,
1396 struct nlattr *tb[], struct nlattr *data[])
1397{
1398 struct net_vrf *vrf = netdev_priv(dev);
097d3c95
DA		 1399 bool *add_fib_rules;
1400 struct net *net;
1aa6c4f6 1401 int err;
193125db
DA		 1402
1403 if (!data || !data[IFLA_VRF_TABLE])
1404 return -EINVAL;
1405
1406 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
24c63bbc
DA		 1407 if (vrf->tb_id == RT_TABLE_UNSPEC)
1408 return -EINVAL;
193125db 1409
007979ea 1410 dev->priv_flags |= IFF_L3MDEV_MASTER;
193125db 1411
1aa6c4f6
DA		 1412 err = register_netdevice(dev);
1413 if (err)
1414 goto out;
1415
097d3c95
DA		 1416 net = dev_net(dev);
1417 add_fib_rules = net_generic(net, vrf_net_id);
1418 if (*add_fib_rules) {
1aa6c4f6
DA		 1419 err = vrf_add_fib_rules(dev);
1420 if (err) {
1421 unregister_netdevice(dev);
1422 goto out;
1423 }
097d3c95 1424 *add_fib_rules = false;
1aa6c4f6
DA		 1425 }
1426
1427out:
1428 return err;
193125db
DA		 1429}
1430
1431static size_t vrf_nl_getsize(const struct net_device *dev)
1432{
1433 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1434}
1435
1436static int vrf_fillinfo(struct sk_buff *skb,
1437 const struct net_device *dev)
1438{
1439 struct net_vrf *vrf = netdev_priv(dev);
1440
1441 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1442}
1443
67eb0331
DA		 1444static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1445 const struct net_device *slave_dev)
1446{
1447 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1448}
1449
1450static int vrf_fill_slave_info(struct sk_buff *skb,
1451 const struct net_device *vrf_dev,
1452 const struct net_device *slave_dev)
1453{
1454 struct net_vrf *vrf = netdev_priv(vrf_dev);
1455
1456 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1457 return -EMSGSIZE;
1458
1459 return 0;
1460}
1461
193125db
DA		 1462static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1463 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1464};
1465
1466static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1467 .kind = DRV_NAME,
1468 .priv_size = sizeof(struct net_vrf),
1469
1470 .get_size = vrf_nl_getsize,
1471 .policy = vrf_nl_policy,
1472 .validate = vrf_validate,
1473 .fill_info = vrf_fillinfo,
1474
67eb0331
DA		 1475 .get_slave_size = vrf_get_slave_size,
1476 .fill_slave_info = vrf_fill_slave_info,
1477
193125db
DA		 1478 .newlink = vrf_newlink,
1479 .dellink = vrf_dellink,
1480 .setup = vrf_setup,
1481 .maxtype = IFLA_VRF_MAX,
1482};
1483
1484static int vrf_device_event(struct notifier_block *unused,
1485 unsigned long event, void *ptr)
1486{
1487 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1488
1489 /* only care about unregister events to drop slave references */
1490 if (event == NETDEV_UNREGISTER) {
193125db
DA		 1491 struct net_device *vrf_dev;
1492
fee6d4c7 1493 if (!netif_is_l3_slave(dev))
193125db
DA		 1494 goto out;
1495
58aa9087
NA		 1496 vrf_dev = netdev_master_upper_dev_get(dev);
1497 vrf_del_slave(vrf_dev, dev);
193125db
DA		 1498 }
1499out:
1500 return NOTIFY_DONE;
1501}
1502
1503static struct notifier_block vrf_notifier_block __read_mostly = {
1504 .notifier_call = vrf_device_event,
1505};
1506
097d3c95
DA		 1507/* Initialize per network namespace state */
1508static int __net_init vrf_netns_init(struct net *net)
1509{
1510 bool *add_fib_rules = net_generic(net, vrf_net_id);
1511
1512 *add_fib_rules = true;
1513
1514 return 0;
1515}
1516
1517static struct pernet_operations vrf_net_ops __net_initdata = {
1518 .init = vrf_netns_init,
1519 .id = &vrf_net_id,
1520 .size = sizeof(bool),
1521};
1522
193125db
DA		 1523static int __init vrf_init_module(void)
1524{
1525 int rc;
1526
193125db
DA		 1527 register_netdevice_notifier(&vrf_notifier_block);
1528
097d3c95 1529 rc = register_pernet_subsys(&vrf_net_ops);
193125db
DA		 1530 if (rc < 0)
1531 goto error;
1532
097d3c95
DA		 1533 rc = rtnl_link_register(&vrf_link_ops);
1534 if (rc < 0) {
1535 unregister_pernet_subsys(&vrf_net_ops);
1536 goto error;
1537 }
1538
193125db
DA		 1539 return 0;
1540
1541error:
1542 unregister_netdevice_notifier(&vrf_notifier_block);
193125db
DA		 1543 return rc;
1544}
1545
193125db 1546module_init(vrf_init_module);
193125db
DA		 1547MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1548MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1549MODULE_LICENSE("GPL");
1550MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1551MODULE_VERSION(DRV_VERSION);
Ubuntu Artful kernel mirror