vrf.c (9d066a252786e1a18484a6283f82614d42a9f4ac) vrf.c (0d240e7811c4ec1965760ee4643b5bbc9cfacbb3)
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

--- 21 unchanged lines hidden (view full) ---

30#include <net/arp.h>
31#include <net/ip.h>
32#include <net/ip_fib.h>
33#include <net/ip6_fib.h>
34#include <net/ip6_route.h>
35#include <net/route.h>
36#include <net/addrconf.h>
37#include <net/l3mdev.h>
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

--- 21 unchanged lines hidden (view full) ---

30#include <net/arp.h>
31#include <net/ip.h>
32#include <net/ip_fib.h>
33#include <net/ip6_fib.h>
34#include <net/ip6_route.h>
35#include <net/route.h>
36#include <net/addrconf.h>
37#include <net/l3mdev.h>
38#include <net/fib_rules.h>
38
39#define RT_FL_TOS(oldflp4) \
40 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
41
42#define DRV_NAME "vrf"
43#define DRV_VERSION "1.0"
44
39
40#define RT_FL_TOS(oldflp4) \
41 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
42
43#define DRV_NAME "vrf"
44#define DRV_VERSION "1.0"
45
46#define FIB_RULE_PREF 1000 /* default preference for FIB rules */
47static bool add_fib_rules = true;
48
45struct net_vrf {
46 struct rtable __rcu *rth;
49struct net_vrf {
50 struct rtable __rcu *rth;
51 struct rtable __rcu *rth_local;
47 struct rt6_info __rcu *rt6;
52 struct rt6_info __rcu *rt6;
53 struct rt6_info __rcu *rt6_local;
48 u32 tb_id;
49};
50
51struct pcpu_dstats {
52 u64 tx_pkts;
53 u64 tx_bytes;
54 u64 tx_drps;
55 u64 rx_pkts;
56 u64 rx_bytes;
54 u32 tb_id;
55};
56
57struct pcpu_dstats {
58 u64 tx_pkts;
59 u64 tx_bytes;
60 u64 tx_drps;
61 u64 rx_pkts;
62 u64 rx_bytes;
63 u64 rx_drps;
57 struct u64_stats_sync syncp;
58};
59
64 struct u64_stats_sync syncp;
65};
66
67static 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
60static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
61{
62 vrf_dev->stats.tx_errors++;
63 kfree_skb(skb);
64}
65
66static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
67 struct rtnl_link_stats64 *stats)

--- 18 unchanged lines hidden (view full) ---

86 stats->tx_packets += tpkts;
87 stats->tx_dropped += tdrops;
88 stats->rx_bytes += rbytes;
89 stats->rx_packets += rpkts;
90 }
91 return stats;
92}
93
77static 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
83static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
84 struct rtnl_link_stats64 *stats)

--- 18 unchanged lines hidden (view full) ---

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
111/* 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
94#if IS_ENABLED(CONFIG_IPV6)
95static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
96 struct net_device *dev)
97{
98 const struct ipv6hdr *iph = ipv6_hdr(skb);
99 struct net *net = dev_net(skb->dev);
100 struct flowi6 fl6 = {
101 /* needed to match OIF rule */

--- 10 unchanged lines hidden (view full) ---

112 struct dst_entry *dst;
113 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
114
115 dst = ip6_route_output(net, NULL, &fl6);
116 if (dst == dst_null)
117 goto err;
118
119 skb_dst_drop(skb);
139#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 */

--- 10 unchanged lines hidden (view full) ---

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);
165
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
120 skb_dst_set(skb, dst);
121
205 skb_dst_set(skb, dst);
206
207 /* strip the ethernet header added for pass through VRF device */
208 __skb_pull(skb, skb_network_offset(skb));
209
122 ret = ip6_local_out(net, skb->sk, skb);
123 if (unlikely(net_xmit_eval(ret)))
124 dev->stats.tx_errors++;
125 else
126 ret = NET_XMIT_SUCCESS;
127
128 return ret;
129err:

--- 4 unchanged lines hidden (view full) ---

134static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
135 struct net_device *dev)
136{
137 vrf_tx_error(dev, skb);
138 return NET_XMIT_DROP;
139}
140#endif
141
210 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:

--- 4 unchanged lines hidden (view full) ---

222static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
223 struct net_device *dev)
224{
225 vrf_tx_error(dev, skb);
226 return NET_XMIT_DROP;
227}
228#endif
229
142static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
143 struct net_device *vrf_dev)
144{
145 struct rtable *rt;
146 int err = 1;
147
148 rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
149 if (IS_ERR(rt))
150 goto out;
151
152 /* TO-DO: what about broadcast ? */
153 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
154 ip_rt_put(rt);
155 goto out;
156 }
157
158 skb_dst_drop(skb);
159 skb_dst_set(skb, &rt->dst);
160 err = 0;
161out:
162 return err;
163}
164
165static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
166 struct net_device *vrf_dev)
167{
168 struct iphdr *ip4h = ip_hdr(skb);
169 int ret = NET_XMIT_DROP;
170 struct flowi4 fl4 = {
171 /* needed to match OIF rule */
172 .flowi4_oif = vrf_dev->ifindex,
173 .flowi4_iif = LOOPBACK_IFINDEX,
174 .flowi4_tos = RT_TOS(ip4h->tos),
175 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
176 FLOWI_FLAG_SKIP_NH_OIF,
177 .daddr = ip4h->daddr,
178 };
230static 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),
240 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
241 FLOWI_FLAG_SKIP_NH_OIF,
242 .daddr = ip4h->daddr,
243 };
244 struct net *net = dev_net(vrf_dev);
245 struct rtable *rt;
179
246
180 if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
247 rt = ip_route_output_flow(net, &fl4, NULL);
248 if (IS_ERR(rt))
181 goto err;
182
249 goto err;
250
251 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);
257
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
285 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
183 if (!ip4h->saddr) {
184 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
185 RT_SCOPE_LINK);
186 }
187
188 ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
189 if (unlikely(net_xmit_eval(ret)))
190 vrf_dev->stats.tx_errors++;

--- 4 unchanged lines hidden (view full) ---

195 return ret;
196err:
197 vrf_tx_error(vrf_dev, skb);
198 goto out;
199}
200
201static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
202{
290 if (!ip4h->saddr) {
291 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
292 RT_SCOPE_LINK);
293 }
294
295 ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
296 if (unlikely(net_xmit_eval(ret)))
297 vrf_dev->stats.tx_errors++;

--- 4 unchanged lines hidden (view full) ---

302 return ret;
303err:
304 vrf_tx_error(vrf_dev, skb);
305 goto out;
306}
307
308static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
309{
203 /* strip the ethernet header added for pass through VRF device */
204 __skb_pull(skb, skb_network_offset(skb));
205
206 switch (skb->protocol) {
207 case htons(ETH_P_IP):
208 return vrf_process_v4_outbound(skb, dev);
209 case htons(ETH_P_IPV6):
210 return vrf_process_v6_outbound(skb, dev);
211 default:
212 vrf_tx_error(dev, skb);
213 return NET_XMIT_DROP;

--- 55 unchanged lines hidden (view full) ---

269{
270 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
271 net, sk, skb, NULL, skb_dst(skb)->dev,
272 vrf_finish_output6,
273 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
274}
275
276/* holding rtnl */
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:
316 vrf_tx_error(dev, skb);
317 return NET_XMIT_DROP;

--- 55 unchanged lines hidden (view full) ---

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
380/* holding rtnl */
277static void vrf_rt6_release(struct net_vrf *vrf)
381static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
278{
279 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
382{
383 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
384 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local);
385 struct net *net = dev_net(dev);
386 struct dst_entry *dst;
280
387
281 rcu_assign_pointer(vrf->rt6, NULL);
388 RCU_INIT_POINTER(vrf->rt6, NULL);
389 RCU_INIT_POINTER(vrf->rt6_local, NULL);
390 synchronize_rcu();
282
391
283 if (rt6)
284 dst_release(&rt6->dst);
392 /* 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 }
402
403 if (rt6_local) {
404 if (rt6_local->rt6i_idev)
405 in6_dev_put(rt6_local->rt6i_idev);
406
407 dst = &rt6_local->dst;
408 dev_put(dst->dev);
409 dst->dev = net->loopback_dev;
410 dev_hold(dst->dev);
411 dst_release(dst);
412 }
285}
286
287static int vrf_rt6_create(struct net_device *dev)
288{
413}
414
415static int vrf_rt6_create(struct net_device *dev)
416{
417 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE;
289 struct net_vrf *vrf = netdev_priv(dev);
290 struct net *net = dev_net(dev);
291 struct fib6_table *rt6i_table;
418 struct net_vrf *vrf = netdev_priv(dev);
419 struct net *net = dev_net(dev);
420 struct fib6_table *rt6i_table;
292 struct rt6_info *rt6;
421 struct rt6_info *rt6, *rt6_local;
293 int rc = -ENOMEM;
294
422 int rc = -ENOMEM;
423
424 /* IPv6 can be CONFIG enabled and then disabled runtime */
425 if (!ipv6_mod_enabled())
426 return 0;
427
295 rt6i_table = fib6_new_table(net, vrf->tb_id);
296 if (!rt6i_table)
297 goto out;
298
428 rt6i_table = fib6_new_table(net, vrf->tb_id);
429 if (!rt6i_table)
430 goto out;
431
299 rt6 = ip6_dst_alloc(net, dev,
300 DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE);
432 /* create a dst for routing packets out a VRF device */
433 rt6 = ip6_dst_alloc(net, dev, flags);
301 if (!rt6)
302 goto out;
303
304 dst_hold(&rt6->dst);
305
306 rt6->rt6i_table = rt6i_table;
307 rt6->dst.output = vrf_output6;
434 if (!rt6)
435 goto out;
436
437 dst_hold(&rt6->dst);
438
439 rt6->rt6i_table = rt6i_table;
440 rt6->dst.output = vrf_output6;
441
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
308 rcu_assign_pointer(vrf->rt6, rt6);
458 rcu_assign_pointer(vrf->rt6, rt6);
459 rcu_assign_pointer(vrf->rt6_local, rt6_local);
309
310 rc = 0;
311out:
312 return rc;
313}
314#else
460
461 rc = 0;
462out:
463 return rc;
464}
465#else
315static void vrf_rt6_release(struct net_vrf *vrf)
466static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
316{
317}
318
319static int vrf_rt6_create(struct net_device *dev)
320{
321 return 0;
322}
323#endif

--- 52 unchanged lines hidden (view full) ---

376
377 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
378 net, sk, skb, NULL, dev,
379 vrf_finish_output,
380 !(IPCB(skb)->flags & IPSKB_REROUTED));
381}
382
383/* holding rtnl */
467{
468}
469
470static int vrf_rt6_create(struct net_device *dev)
471{
472 return 0;
473}
474#endif

--- 52 unchanged lines hidden (view full) ---

527
528 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
529 net, sk, skb, NULL, dev,
530 vrf_finish_output,
531 !(IPCB(skb)->flags & IPSKB_REROUTED));
532}
533
534/* holding rtnl */
384static void vrf_rtable_release(struct net_vrf *vrf)
535static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
385{
386 struct rtable *rth = rtnl_dereference(vrf->rth);
536{
537 struct rtable *rth = rtnl_dereference(vrf->rth);
538 struct rtable *rth_local = rtnl_dereference(vrf->rth_local);
539 struct net *net = dev_net(dev);
540 struct dst_entry *dst;
387
541
388 rcu_assign_pointer(vrf->rth, NULL);
542 RCU_INIT_POINTER(vrf->rth, NULL);
543 RCU_INIT_POINTER(vrf->rth_local, NULL);
544 synchronize_rcu();
389
545
390 if (rth)
391 dst_release(&rth->dst);
546 /* 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 }
556
557 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 }
392}
393
394static int vrf_rtable_create(struct net_device *dev)
395{
396 struct net_vrf *vrf = netdev_priv(dev);
564}
565
566static int vrf_rtable_create(struct net_device *dev)
567{
568 struct net_vrf *vrf = netdev_priv(dev);
397 struct rtable *rth;
569 struct rtable *rth, *rth_local;
398
399 if (!fib_new_table(dev_net(dev), vrf->tb_id))
400 return -ENOMEM;
401
570
571 if (!fib_new_table(dev_net(dev), vrf->tb_id))
572 return -ENOMEM;
573
574 /* create a dst for routing packets out through a VRF device */
402 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
403 if (!rth)
404 return -ENOMEM;
405
575 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
576 if (!rth)
577 return -ENOMEM;
578
579 /* 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
406 rth->dst.output = vrf_output;
407 rth->rt_table_id = vrf->tb_id;
408
588 rth->dst.output = vrf_output;
589 rth->rt_table_id = vrf->tb_id;
590
591 rth_local->rt_table_id = vrf->tb_id;
592
409 rcu_assign_pointer(vrf->rth, rth);
593 rcu_assign_pointer(vrf->rth, rth);
594 rcu_assign_pointer(vrf->rth_local, rth_local);
410
411 return 0;
412}
413
414/**************************** device handling ********************/
415
416/* cycle interface to flush neighbor cache and move routes across tables */
417static void cycle_netdev(struct net_device *dev)

--- 54 unchanged lines hidden (view full) ---

472}
473
474static void vrf_dev_uninit(struct net_device *dev)
475{
476 struct net_vrf *vrf = netdev_priv(dev);
477 struct net_device *port_dev;
478 struct list_head *iter;
479
595
596 return 0;
597}
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)

--- 54 unchanged lines hidden (view full) ---

657}
658
659static void vrf_dev_uninit(struct net_device *dev)
660{
661 struct net_vrf *vrf = netdev_priv(dev);
662 struct net_device *port_dev;
663 struct list_head *iter;
664
480 vrf_rtable_release(vrf);
481 vrf_rt6_release(vrf);
665 vrf_rtable_release(dev, vrf);
666 vrf_rt6_release(dev, vrf);
482
483 netdev_for_each_lower_dev(dev, port_dev, iter)
484 vrf_del_slave(dev, port_dev);
485
486 free_percpu(dev->dstats);
487 dev->dstats = NULL;
488}
489

--- 9 unchanged lines hidden (view full) ---

499 if (vrf_rtable_create(dev) != 0)
500 goto out_stats;
501
502 if (vrf_rt6_create(dev) != 0)
503 goto out_rth;
504
505 dev->flags = IFF_MASTER | IFF_NOARP;
506
667
668 netdev_for_each_lower_dev(dev, port_dev, iter)
669 vrf_del_slave(dev, port_dev);
670
671 free_percpu(dev->dstats);
672 dev->dstats = NULL;
673}
674

--- 9 unchanged lines hidden (view full) ---

684 if (vrf_rtable_create(dev) != 0)
685 goto out_stats;
686
687 if (vrf_rt6_create(dev) != 0)
688 goto out_rth;
689
690 dev->flags = IFF_MASTER | IFF_NOARP;
691
692 /* 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;
697 netdev_lockdep_set_classes(dev);
507 return 0;
508
509out_rth:
698 return 0;
699
700out_rth:
510 vrf_rtable_release(vrf);
701 vrf_rtable_release(dev, vrf);
511out_stats:
512 free_percpu(dev->dstats);
513 dev->dstats = NULL;
514out_nomem:
515 return -ENOMEM;
516}
517
518static const struct net_device_ops vrf_netdev_ops = {

--- 99 unchanged lines hidden (view full) ---

618 break;
619 }
620 }
621
622out:
623 return rc;
624}
625
702out_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 = {

--- 99 unchanged lines hidden (view full) ---

809 break;
810 }
811 }
812
813out:
814 return rc;
815}
816
817static 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
626static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
627 struct sk_buff *skb)
628{
868static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
869 struct sk_buff *skb)
870{
629 /* if packet is NDISC keep the ingress interface */
630 if (!ipv6_ndisc_frame(skb)) {
871 int orig_iif = skb->skb_iif;
872 bool need_strict;
873
874 /* 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) {
631 skb->dev = vrf_dev;
632 skb->skb_iif = vrf_dev->ifindex;
878 skb->dev = vrf_dev;
879 skb->skb_iif = vrf_dev->ifindex;
880 skb->pkt_type = PACKET_HOST;
881 goto out;
882 }
633
883
884 /* 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) {
889 skb->dev = vrf_dev;
890 skb->skb_iif = vrf_dev->ifindex;
891
634 skb_push(skb, skb->mac_len);
635 dev_queue_xmit_nit(skb, vrf_dev);
636 skb_pull(skb, skb->mac_len);
637
638 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
639 }
640
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
899 if (need_strict)
900 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
901
902out:
641 return skb;
642}
643
644#else
645static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
646 struct sk_buff *skb)
647{
648 return skb;
649}
650#endif
651
652static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
653 struct sk_buff *skb)
654{
655 skb->dev = vrf_dev;
656 skb->skb_iif = vrf_dev->ifindex;
657
903 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
920 /* 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
658 skb_push(skb, skb->mac_len);
659 dev_queue_xmit_nit(skb, vrf_dev);
660 skb_pull(skb, skb->mac_len);
661
928 skb_push(skb, skb->mac_len);
929 dev_queue_xmit_nit(skb, vrf_dev);
930 skb_pull(skb, skb->mac_len);
931
932out:
662 return skb;
663}
664
665/* called with rcu lock held */
666static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
667 struct sk_buff *skb,
668 u16 proto)
669{

--- 4 unchanged lines hidden (view full) ---

674 return vrf_ip6_rcv(vrf_dev, skb);
675 }
676
677 return skb;
678}
679
680#if IS_ENABLED(CONFIG_IPV6)
681static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
933 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{

--- 4 unchanged lines hidden (view full) ---

945 return vrf_ip6_rcv(vrf_dev, skb);
946 }
947
948 return skb;
949}
950
951#if IS_ENABLED(CONFIG_IPV6)
952static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
682 const struct flowi6 *fl6)
953 struct flowi6 *fl6)
683{
954{
955 bool need_strict = rt6_need_strict(&fl6->daddr);
956 struct net_vrf *vrf = netdev_priv(dev);
957 struct net *net = dev_net(dev);
684 struct dst_entry *dst = NULL;
958 struct dst_entry *dst = NULL;
959 struct rt6_info *rt;
685
960
686 if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) {
687 struct net_vrf *vrf = netdev_priv(dev);
688 struct rt6_info *rt;
961 /* send to link-local or multicast address */
962 if (need_strict) {
963 int flags = RT6_LOOKUP_F_IFACE;
689
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)) {
984
690 rcu_read_lock();
691
692 rt = rcu_dereference(vrf->rt6);
693 if (likely(rt)) {
694 dst = &rt->dst;
695 dst_hold(dst);
696 }
697
698 rcu_read_unlock();
699 }
700
985 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();
994 }
995
996 /* make sure oif is set to VRF device for lookup */
997 if (!need_strict)
998 fl6->flowi6_oif = dev->ifindex;
999
701 return dst;
702}
1000 return dst;
1001}
1002
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}
703#endif
704
705static const struct l3mdev_ops vrf_l3mdev_ops = {
706 .l3mdev_fib_table = vrf_fib_table,
707 .l3mdev_get_rtable = vrf_get_rtable,
708 .l3mdev_get_saddr = vrf_get_saddr,
709 .l3mdev_l3_rcv = vrf_l3_rcv,
710#if IS_ENABLED(CONFIG_IPV6)
711 .l3mdev_get_rt6_dst = vrf_get_rt6_dst,
1042#endif
1043
1044static const struct l3mdev_ops vrf_l3mdev_ops = {
1045 .l3mdev_fib_table = vrf_fib_table,
1046 .l3mdev_get_rtable = vrf_get_rtable,
1047 .l3mdev_get_saddr = vrf_get_saddr,
1048 .l3mdev_l3_rcv = vrf_l3_rcv,
1049#if IS_ENABLED(CONFIG_IPV6)
1050 .l3mdev_get_rt6_dst = vrf_get_rt6_dst,
1051 .l3mdev_get_saddr6 = vrf_get_saddr6,
712#endif
713};
714
715static void vrf_get_drvinfo(struct net_device *dev,
716 struct ethtool_drvinfo *info)
717{
718 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
719 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
720}
721
722static const struct ethtool_ops vrf_ethtool_ops = {
723 .get_drvinfo = vrf_get_drvinfo,
724};
725
1052#endif
1053};
1054
1055static 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
1066static 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
1084 if (family == AF_INET6 && !ipv6_mod_enabled())
1085 return 0;
1086
1087 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
726static void vrf_setup(struct net_device *dev)
727{
728 ether_setup(dev);
729
730 /* Initialize the device structure. */
731 dev->netdev_ops = &vrf_netdev_ops;
732 dev->l3mdev_ops = &vrf_l3mdev_ops;
733 dev->ethtool_ops = &vrf_ethtool_ops;
734 dev->destructor = free_netdev;
735
736 /* Fill in device structure with ethernet-generic values. */
737 eth_hw_addr_random(dev);
738
739 /* don't acquire vrf device's netif_tx_lock when transmitting */
740 dev->features |= NETIF_F_LLTX;
741
742 /* don't allow vrf devices to change network namespaces. */
743 dev->features |= NETIF_F_NETNS_LOCAL;
1154static 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;
1160 dev->l3mdev_ops = &vrf_l3mdev_ops;
1161 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;
1172
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;
744}
745
746static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
747{
748 if (tb[IFLA_ADDRESS]) {
749 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
750 return -EINVAL;
751 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))

--- 6 unchanged lines hidden (view full) ---

758{
759 unregister_netdevice_queue(dev, head);
760}
761
762static int vrf_newlink(struct net *src_net, struct net_device *dev,
763 struct nlattr *tb[], struct nlattr *data[])
764{
765 struct net_vrf *vrf = netdev_priv(dev);
1186}
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])))

--- 6 unchanged lines hidden (view full) ---

1200{
1201 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);
1208 int err;
766
767 if (!data || !data[IFLA_VRF_TABLE])
768 return -EINVAL;
769
770 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
771
772 dev->priv_flags |= IFF_L3MDEV_MASTER;
773
1209
1210 if (!data || !data[IFLA_VRF_TABLE])
1211 return -EINVAL;
1212
1213 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1214
1215 dev->priv_flags |= IFF_L3MDEV_MASTER;
1216
774 return register_netdevice(dev);
1217 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;
775}
776
777static size_t vrf_nl_getsize(const struct net_device *dev)
778{
779 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
780}
781
782static int vrf_fillinfo(struct sk_buff *skb,

--- 93 unchanged lines hidden --- 		1232}
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,

--- 93 unchanged lines hidden ---