Commit | Line | Data |
---|---|---|
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> |
193125db | 39 | |
8cbb512c DA |
40 | #define RT_FL_TOS(oldflp4) \ |
41 | ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)) | |
42 | ||
193125db DA |
43 | #define DRV_NAME "vrf" |
44 | #define DRV_VERSION "1.0" | |
45 | ||
1aa6c4f6 DA |
46 | #define FIB_RULE_PREF 1000 /* default preference for FIB rules */ |
47 | static bool add_fib_rules = true; | |
48 | ||
ec539514 | 49 | struct net_vrf { |
b0e95ccd | 50 | struct rtable __rcu *rth; |
afe80a49 | 51 | struct rtable __rcu *rth_local; |
b0e95ccd | 52 | struct rt6_info __rcu *rt6; |
b4869aa2 | 53 | struct rt6_info __rcu *rt6_local; |
ec539514 DA |
54 | u32 tb_id; |
55 | }; | |
56 | ||
193125db DA |
57 | struct pcpu_dstats { |
58 | u64 tx_pkts; | |
59 | u64 tx_bytes; | |
60 | u64 tx_drps; | |
61 | u64 rx_pkts; | |
62 | u64 rx_bytes; | |
afe80a49 | 63 | u64 rx_drps; |
193125db DA |
64 | struct u64_stats_sync syncp; |
65 | }; | |
66 | ||
afe80a49 DA |
67 | static 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 | ||
57b8efa1 NA |
77 | static 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 | ||
193125db DA |
83 | static 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 | ||
afe80a49 DA |
111 | /* Local traffic destined to local address. Reinsert the packet to rx |
112 | * path, similar to loopback handling. | |
113 | */ | |
114 | static 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 | ||
35402e31 | 139 | #if IS_ENABLED(CONFIG_IPV6) |
4c1feac5 DA |
140 | static int vrf_ip6_local_out(struct net *net, struct sock *sk, |
141 | struct sk_buff *skb) | |
142 | { | |
143 | int err; | |
144 | ||
145 | err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, | |
146 | sk, skb, NULL, skb_dst(skb)->dev, dst_output); | |
147 | ||
148 | if (likely(err == 1)) | |
149 | err = dst_output(net, sk, skb); | |
150 | ||
151 | return err; | |
152 | } | |
153 | ||
35402e31 DA |
154 | static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, |
155 | struct net_device *dev) | |
156 | { | |
157 | const struct ipv6hdr *iph = ipv6_hdr(skb); | |
158 | struct net *net = dev_net(skb->dev); | |
159 | struct flowi6 fl6 = { | |
160 | /* needed to match OIF rule */ | |
161 | .flowi6_oif = dev->ifindex, | |
162 | .flowi6_iif = LOOPBACK_IFINDEX, | |
163 | .daddr = iph->daddr, | |
164 | .saddr = iph->saddr, | |
165 | .flowlabel = ip6_flowinfo(iph), | |
166 | .flowi6_mark = skb->mark, | |
167 | .flowi6_proto = iph->nexthdr, | |
c71ad3d4 | 168 | .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF, |
35402e31 DA |
169 | }; |
170 | int ret = NET_XMIT_DROP; | |
171 | struct dst_entry *dst; | |
172 | struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; | |
173 | ||
174 | dst = ip6_route_output(net, NULL, &fl6); | |
175 | if (dst == dst_null) | |
176 | goto err; | |
177 | ||
178 | skb_dst_drop(skb); | |
b4869aa2 DA |
179 | |
180 | /* if dst.dev is loopback or the VRF device again this is locally | |
181 | * originated traffic destined to a local address. Short circuit | |
182 | * to Rx path using our local dst | |
183 | */ | |
184 | if (dst->dev == net->loopback_dev || dst->dev == dev) { | |
185 | struct net_vrf *vrf = netdev_priv(dev); | |
186 | struct rt6_info *rt6_local; | |
187 | ||
188 | /* release looked up dst and use cached local dst */ | |
189 | dst_release(dst); | |
190 | ||
191 | rcu_read_lock(); | |
192 | ||
193 | rt6_local = rcu_dereference(vrf->rt6_local); | |
194 | if (unlikely(!rt6_local)) { | |
195 | rcu_read_unlock(); | |
196 | goto err; | |
197 | } | |
198 | ||
199 | /* Ordering issue: cached local dst is created on newlink | |
200 | * before the IPv6 initialization. Using the local dst | |
201 | * requires rt6i_idev to be set so make sure it is. | |
202 | */ | |
203 | if (unlikely(!rt6_local->rt6i_idev)) { | |
204 | rt6_local->rt6i_idev = in6_dev_get(dev); | |
205 | if (!rt6_local->rt6i_idev) { | |
206 | rcu_read_unlock(); | |
207 | goto err; | |
208 | } | |
209 | } | |
210 | ||
211 | dst = &rt6_local->dst; | |
212 | dst_hold(dst); | |
213 | ||
214 | rcu_read_unlock(); | |
215 | ||
216 | return vrf_local_xmit(skb, dev, &rt6_local->dst); | |
217 | } | |
218 | ||
35402e31 DA |
219 | skb_dst_set(skb, dst); |
220 | ||
911a66fb DA |
221 | /* strip the ethernet header added for pass through VRF device */ |
222 | __skb_pull(skb, skb_network_offset(skb)); | |
223 | ||
4c1feac5 | 224 | ret = vrf_ip6_local_out(net, skb->sk, skb); |
35402e31 DA |
225 | if (unlikely(net_xmit_eval(ret))) |
226 | dev->stats.tx_errors++; | |
227 | else | |
228 | ret = NET_XMIT_SUCCESS; | |
229 | ||
230 | return ret; | |
231 | err: | |
232 | vrf_tx_error(dev, skb); | |
233 | return NET_XMIT_DROP; | |
234 | } | |
235 | #else | |
193125db DA |
236 | static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, |
237 | struct net_device *dev) | |
238 | { | |
57b8efa1 NA |
239 | vrf_tx_error(dev, skb); |
240 | return NET_XMIT_DROP; | |
193125db | 241 | } |
35402e31 | 242 | #endif |
193125db | 243 | |
ebfc102c DA |
244 | /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ |
245 | static int vrf_ip_local_out(struct net *net, struct sock *sk, | |
246 | struct sk_buff *skb) | |
247 | { | |
248 | int err; | |
249 | ||
250 | err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, | |
251 | skb, NULL, skb_dst(skb)->dev, dst_output); | |
252 | if (likely(err == 1)) | |
253 | err = dst_output(net, sk, skb); | |
254 | ||
255 | return err; | |
256 | } | |
257 | ||
193125db DA |
258 | static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, |
259 | struct net_device *vrf_dev) | |
260 | { | |
261 | struct iphdr *ip4h = ip_hdr(skb); | |
262 | int ret = NET_XMIT_DROP; | |
263 | struct flowi4 fl4 = { | |
264 | /* needed to match OIF rule */ | |
265 | .flowi4_oif = vrf_dev->ifindex, | |
266 | .flowi4_iif = LOOPBACK_IFINDEX, | |
267 | .flowi4_tos = RT_TOS(ip4h->tos), | |
c71ad3d4 | 268 | .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF, |
193125db DA |
269 | .daddr = ip4h->daddr, |
270 | }; | |
911a66fb DA |
271 | struct net *net = dev_net(vrf_dev); |
272 | struct rtable *rt; | |
273 | ||
274 | rt = ip_route_output_flow(net, &fl4, NULL); | |
275 | if (IS_ERR(rt)) | |
276 | goto err; | |
193125db | 277 | |
911a66fb DA |
278 | if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { |
279 | ip_rt_put(rt); | |
193125db | 280 | goto err; |
911a66fb DA |
281 | } |
282 | ||
283 | skb_dst_drop(skb); | |
afe80a49 DA |
284 | |
285 | /* if dst.dev is loopback or the VRF device again this is locally | |
286 | * originated traffic destined to a local address. Short circuit | |
287 | * to Rx path using our local dst | |
288 | */ | |
289 | if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) { | |
290 | struct net_vrf *vrf = netdev_priv(vrf_dev); | |
291 | struct rtable *rth_local; | |
292 | struct dst_entry *dst = NULL; | |
293 | ||
294 | ip_rt_put(rt); | |
295 | ||
296 | rcu_read_lock(); | |
297 | ||
298 | rth_local = rcu_dereference(vrf->rth_local); | |
299 | if (likely(rth_local)) { | |
300 | dst = &rth_local->dst; | |
301 | dst_hold(dst); | |
302 | } | |
303 | ||
304 | rcu_read_unlock(); | |
305 | ||
306 | if (unlikely(!dst)) | |
307 | goto err; | |
308 | ||
309 | return vrf_local_xmit(skb, vrf_dev, dst); | |
310 | } | |
311 | ||
911a66fb DA |
312 | skb_dst_set(skb, &rt->dst); |
313 | ||
314 | /* strip the ethernet header added for pass through VRF device */ | |
315 | __skb_pull(skb, skb_network_offset(skb)); | |
193125db DA |
316 | |
317 | if (!ip4h->saddr) { | |
318 | ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, | |
319 | RT_SCOPE_LINK); | |
320 | } | |
321 | ||
ebfc102c | 322 | ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); |
193125db DA |
323 | if (unlikely(net_xmit_eval(ret))) |
324 | vrf_dev->stats.tx_errors++; | |
325 | else | |
326 | ret = NET_XMIT_SUCCESS; | |
327 | ||
328 | out: | |
329 | return ret; | |
330 | err: | |
57b8efa1 | 331 | vrf_tx_error(vrf_dev, skb); |
193125db DA |
332 | goto out; |
333 | } | |
334 | ||
335 | static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) | |
336 | { | |
337 | switch (skb->protocol) { | |
338 | case htons(ETH_P_IP): | |
339 | return vrf_process_v4_outbound(skb, dev); | |
340 | case htons(ETH_P_IPV6): | |
341 | return vrf_process_v6_outbound(skb, dev); | |
342 | default: | |
57b8efa1 | 343 | vrf_tx_error(dev, skb); |
193125db DA |
344 | return NET_XMIT_DROP; |
345 | } | |
346 | } | |
347 | ||
348 | static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) | |
349 | { | |
350 | netdev_tx_t ret = is_ip_tx_frame(skb, dev); | |
351 | ||
352 | if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { | |
353 | struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); | |
354 | ||
355 | u64_stats_update_begin(&dstats->syncp); | |
356 | dstats->tx_pkts++; | |
357 | dstats->tx_bytes += skb->len; | |
358 | u64_stats_update_end(&dstats->syncp); | |
359 | } else { | |
360 | this_cpu_inc(dev->dstats->tx_drps); | |
361 | } | |
362 | ||
363 | return ret; | |
364 | } | |
365 | ||
35402e31 | 366 | #if IS_ENABLED(CONFIG_IPV6) |
35402e31 DA |
367 | /* modelled after ip6_finish_output2 */ |
368 | static int vrf_finish_output6(struct net *net, struct sock *sk, | |
369 | struct sk_buff *skb) | |
370 | { | |
371 | struct dst_entry *dst = skb_dst(skb); | |
372 | struct net_device *dev = dst->dev; | |
373 | struct neighbour *neigh; | |
374 | struct in6_addr *nexthop; | |
375 | int ret; | |
376 | ||
377 | skb->protocol = htons(ETH_P_IPV6); | |
378 | skb->dev = dev; | |
379 | ||
380 | rcu_read_lock_bh(); | |
381 | nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); | |
382 | neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); | |
383 | if (unlikely(!neigh)) | |
384 | neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); | |
385 | if (!IS_ERR(neigh)) { | |
386 | ret = dst_neigh_output(dst, neigh, skb); | |
387 | rcu_read_unlock_bh(); | |
388 | return ret; | |
389 | } | |
390 | rcu_read_unlock_bh(); | |
391 | ||
392 | IP6_INC_STATS(dev_net(dst->dev), | |
393 | ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); | |
394 | kfree_skb(skb); | |
395 | return -EINVAL; | |
396 | } | |
397 | ||
398 | /* modelled after ip6_output */ | |
399 | static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) | |
400 | { | |
401 | return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, | |
402 | net, sk, skb, NULL, skb_dst(skb)->dev, | |
403 | vrf_finish_output6, | |
404 | !(IP6CB(skb)->flags & IP6SKB_REROUTED)); | |
405 | } | |
406 | ||
4c1feac5 DA |
407 | /* set dst on skb to send packet to us via dev_xmit path. Allows |
408 | * packet to go through device based features such as qdisc, netfilter | |
409 | * hooks and packet sockets with skb->dev set to vrf device. | |
410 | */ | |
411 | static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, | |
412 | struct sock *sk, | |
413 | struct sk_buff *skb) | |
414 | { | |
415 | struct net_vrf *vrf = netdev_priv(vrf_dev); | |
416 | struct dst_entry *dst = NULL; | |
417 | struct rt6_info *rt6; | |
418 | ||
419 | /* don't divert link scope packets */ | |
420 | if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) | |
421 | return skb; | |
422 | ||
423 | rcu_read_lock(); | |
424 | ||
425 | rt6 = rcu_dereference(vrf->rt6); | |
426 | if (likely(rt6)) { | |
427 | dst = &rt6->dst; | |
428 | dst_hold(dst); | |
429 | } | |
430 | ||
431 | rcu_read_unlock(); | |
432 | ||
433 | if (unlikely(!dst)) { | |
434 | vrf_tx_error(vrf_dev, skb); | |
435 | return NULL; | |
436 | } | |
437 | ||
438 | skb_dst_drop(skb); | |
439 | skb_dst_set(skb, dst); | |
440 | ||
441 | return skb; | |
442 | } | |
443 | ||
b0e95ccd | 444 | /* holding rtnl */ |
810e530b | 445 | static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) |
35402e31 | 446 | { |
b0e95ccd | 447 | struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); |
b4869aa2 | 448 | struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local); |
810e530b DA |
449 | struct net *net = dev_net(dev); |
450 | struct dst_entry *dst; | |
b0e95ccd | 451 | |
b4869aa2 DA |
452 | RCU_INIT_POINTER(vrf->rt6, NULL); |
453 | RCU_INIT_POINTER(vrf->rt6_local, NULL); | |
454 | synchronize_rcu(); | |
b0e95ccd | 455 | |
810e530b DA |
456 | /* move dev in dst's to loopback so this VRF device can be deleted |
457 | * - based on dst_ifdown | |
458 | */ | |
459 | if (rt6) { | |
460 | dst = &rt6->dst; | |
461 | dev_put(dst->dev); | |
462 | dst->dev = net->loopback_dev; | |
463 | dev_hold(dst->dev); | |
464 | dst_release(dst); | |
465 | } | |
b4869aa2 DA |
466 | |
467 | if (rt6_local) { | |
468 | if (rt6_local->rt6i_idev) | |
469 | in6_dev_put(rt6_local->rt6i_idev); | |
470 | ||
810e530b DA |
471 | dst = &rt6_local->dst; |
472 | dev_put(dst->dev); | |
473 | dst->dev = net->loopback_dev; | |
474 | dev_hold(dst->dev); | |
475 | dst_release(dst); | |
b4869aa2 | 476 | } |
35402e31 DA |
477 | } |
478 | ||
479 | static int vrf_rt6_create(struct net_device *dev) | |
480 | { | |
b4869aa2 | 481 | int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE; |
35402e31 | 482 | struct net_vrf *vrf = netdev_priv(dev); |
9ab179d8 | 483 | struct net *net = dev_net(dev); |
b3b4663c | 484 | struct fib6_table *rt6i_table; |
b4869aa2 | 485 | struct rt6_info *rt6, *rt6_local; |
35402e31 DA |
486 | int rc = -ENOMEM; |
487 | ||
e4348637 DA |
488 | /* IPv6 can be CONFIG enabled and then disabled runtime */ |
489 | if (!ipv6_mod_enabled()) | |
490 | return 0; | |
491 | ||
b3b4663c DA |
492 | rt6i_table = fib6_new_table(net, vrf->tb_id); |
493 | if (!rt6i_table) | |
494 | goto out; | |
495 | ||
b4869aa2 DA |
496 | /* create a dst for routing packets out a VRF device */ |
497 | rt6 = ip6_dst_alloc(net, dev, flags); | |
35402e31 DA |
498 | if (!rt6) |
499 | goto out; | |
500 | ||
9ab179d8 | 501 | dst_hold(&rt6->dst); |
b3b4663c DA |
502 | |
503 | rt6->rt6i_table = rt6i_table; | |
504 | rt6->dst.output = vrf_output6; | |
b4869aa2 DA |
505 | |
506 | /* create a dst for local routing - packets sent locally | |
507 | * to local address via the VRF device as a loopback | |
508 | */ | |
509 | rt6_local = ip6_dst_alloc(net, dev, flags); | |
510 | if (!rt6_local) { | |
511 | dst_release(&rt6->dst); | |
512 | goto out; | |
513 | } | |
514 | ||
515 | dst_hold(&rt6_local->dst); | |
516 | ||
517 | rt6_local->rt6i_idev = in6_dev_get(dev); | |
518 | rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL; | |
519 | rt6_local->rt6i_table = rt6i_table; | |
520 | rt6_local->dst.input = ip6_input; | |
521 | ||
b0e95ccd | 522 | rcu_assign_pointer(vrf->rt6, rt6); |
b4869aa2 | 523 | rcu_assign_pointer(vrf->rt6_local, rt6_local); |
b0e95ccd | 524 | |
35402e31 DA |
525 | rc = 0; |
526 | out: | |
527 | return rc; | |
528 | } | |
529 | #else | |
4c1feac5 DA |
530 | static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, |
531 | struct sock *sk, | |
532 | struct sk_buff *skb) | |
533 | { | |
534 | return skb; | |
535 | } | |
536 | ||
810e530b | 537 | static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) |
35402e31 DA |
538 | { |
539 | } | |
540 | ||
541 | static int vrf_rt6_create(struct net_device *dev) | |
542 | { | |
543 | return 0; | |
544 | } | |
545 | #endif | |
546 | ||
8f58336d | 547 | /* modelled after ip_finish_output2 */ |
0c4b51f0 | 548 | static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
193125db | 549 | { |
8f58336d DA |
550 | struct dst_entry *dst = skb_dst(skb); |
551 | struct rtable *rt = (struct rtable *)dst; | |
552 | struct net_device *dev = dst->dev; | |
553 | unsigned int hh_len = LL_RESERVED_SPACE(dev); | |
554 | struct neighbour *neigh; | |
555 | u32 nexthop; | |
556 | int ret = -EINVAL; | |
557 | ||
558 | /* Be paranoid, rather than too clever. */ | |
559 | if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { | |
560 | struct sk_buff *skb2; | |
561 | ||
562 | skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); | |
563 | if (!skb2) { | |
564 | ret = -ENOMEM; | |
565 | goto err; | |
566 | } | |
567 | if (skb->sk) | |
568 | skb_set_owner_w(skb2, skb->sk); | |
569 | ||
570 | consume_skb(skb); | |
571 | skb = skb2; | |
572 | } | |
573 | ||
574 | rcu_read_lock_bh(); | |
575 | ||
576 | nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); | |
577 | neigh = __ipv4_neigh_lookup_noref(dev, nexthop); | |
578 | if (unlikely(!neigh)) | |
579 | neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); | |
580 | if (!IS_ERR(neigh)) | |
581 | ret = dst_neigh_output(dst, neigh, skb); | |
582 | ||
583 | rcu_read_unlock_bh(); | |
584 | err: | |
585 | if (unlikely(ret < 0)) | |
586 | vrf_tx_error(skb->dev, skb); | |
587 | return ret; | |
193125db DA |
588 | } |
589 | ||
ede2059d | 590 | static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
193125db DA |
591 | { |
592 | struct net_device *dev = skb_dst(skb)->dev; | |
593 | ||
29a26a56 | 594 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); |
193125db DA |
595 | |
596 | skb->dev = dev; | |
597 | skb->protocol = htons(ETH_P_IP); | |
598 | ||
29a26a56 EB |
599 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
600 | net, sk, skb, NULL, dev, | |
8f58336d | 601 | vrf_finish_output, |
193125db DA |
602 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
603 | } | |
604 | ||
ebfc102c DA |
605 | /* set dst on skb to send packet to us via dev_xmit path. Allows |
606 | * packet to go through device based features such as qdisc, netfilter | |
607 | * hooks and packet sockets with skb->dev set to vrf device. | |
608 | */ | |
609 | static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, | |
610 | struct sock *sk, | |
611 | struct sk_buff *skb) | |
612 | { | |
613 | struct net_vrf *vrf = netdev_priv(vrf_dev); | |
614 | struct dst_entry *dst = NULL; | |
615 | struct rtable *rth; | |
616 | ||
617 | rcu_read_lock(); | |
618 | ||
619 | rth = rcu_dereference(vrf->rth); | |
620 | if (likely(rth)) { | |
621 | dst = &rth->dst; | |
622 | dst_hold(dst); | |
623 | } | |
624 | ||
625 | rcu_read_unlock(); | |
626 | ||
627 | if (unlikely(!dst)) { | |
628 | vrf_tx_error(vrf_dev, skb); | |
629 | return NULL; | |
630 | } | |
631 | ||
632 | skb_dst_drop(skb); | |
633 | skb_dst_set(skb, dst); | |
634 | ||
635 | return skb; | |
636 | } | |
637 | ||
638 | /* called with rcu lock held */ | |
639 | static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, | |
640 | struct sock *sk, | |
641 | struct sk_buff *skb, | |
642 | u16 proto) | |
643 | { | |
644 | switch (proto) { | |
645 | case AF_INET: | |
646 | return vrf_ip_out(vrf_dev, sk, skb); | |
4c1feac5 DA |
647 | case AF_INET6: |
648 | return vrf_ip6_out(vrf_dev, sk, skb); | |
ebfc102c DA |
649 | } |
650 | ||
651 | return skb; | |
652 | } | |
653 | ||
b0e95ccd | 654 | /* holding rtnl */ |
810e530b | 655 | static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) |
193125db | 656 | { |
b0e95ccd | 657 | struct rtable *rth = rtnl_dereference(vrf->rth); |
afe80a49 | 658 | struct rtable *rth_local = rtnl_dereference(vrf->rth_local); |
810e530b DA |
659 | struct net *net = dev_net(dev); |
660 | struct dst_entry *dst; | |
b0e95ccd | 661 | |
afe80a49 DA |
662 | RCU_INIT_POINTER(vrf->rth, NULL); |
663 | RCU_INIT_POINTER(vrf->rth_local, NULL); | |
664 | synchronize_rcu(); | |
193125db | 665 | |
810e530b DA |
666 | /* move dev in dst's to loopback so this VRF device can be deleted |
667 | * - based on dst_ifdown | |
668 | */ | |
669 | if (rth) { | |
670 | dst = &rth->dst; | |
671 | dev_put(dst->dev); | |
672 | dst->dev = net->loopback_dev; | |
673 | dev_hold(dst->dev); | |
674 | dst_release(dst); | |
675 | } | |
afe80a49 | 676 | |
810e530b DA |
677 | if (rth_local) { |
678 | dst = &rth_local->dst; | |
679 | dev_put(dst->dev); | |
680 | dst->dev = net->loopback_dev; | |
681 | dev_hold(dst->dev); | |
682 | dst_release(dst); | |
683 | } | |
193125db DA |
684 | } |
685 | ||
b0e95ccd | 686 | static int vrf_rtable_create(struct net_device *dev) |
193125db | 687 | { |
b7503e0c | 688 | struct net_vrf *vrf = netdev_priv(dev); |
afe80a49 | 689 | struct rtable *rth, *rth_local; |
193125db | 690 | |
b3b4663c | 691 | if (!fib_new_table(dev_net(dev), vrf->tb_id)) |
b0e95ccd | 692 | return -ENOMEM; |
b3b4663c | 693 | |
afe80a49 | 694 | /* create a dst for routing packets out through a VRF device */ |
9ab179d8 | 695 | rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0); |
b0e95ccd DA |
696 | if (!rth) |
697 | return -ENOMEM; | |
193125db | 698 | |
afe80a49 DA |
699 | /* create a dst for local ingress routing - packets sent locally |
700 | * to local address via the VRF device as a loopback | |
701 | */ | |
702 | rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0); | |
703 | if (!rth_local) { | |
704 | dst_release(&rth->dst); | |
705 | return -ENOMEM; | |
706 | } | |
707 | ||
b0e95ccd DA |
708 | rth->dst.output = vrf_output; |
709 | rth->rt_table_id = vrf->tb_id; | |
710 | ||
afe80a49 DA |
711 | rth_local->rt_table_id = vrf->tb_id; |
712 | ||
b0e95ccd | 713 | rcu_assign_pointer(vrf->rth, rth); |
afe80a49 | 714 | rcu_assign_pointer(vrf->rth_local, rth_local); |
b0e95ccd DA |
715 | |
716 | return 0; | |
193125db DA |
717 | } |
718 | ||
719 | /**************************** device handling ********************/ | |
720 | ||
721 | /* cycle interface to flush neighbor cache and move routes across tables */ | |
722 | static void cycle_netdev(struct net_device *dev) | |
723 | { | |
724 | unsigned int flags = dev->flags; | |
725 | int ret; | |
726 | ||
727 | if (!netif_running(dev)) | |
728 | return; | |
729 | ||
730 | ret = dev_change_flags(dev, flags & ~IFF_UP); | |
731 | if (ret >= 0) | |
732 | ret = dev_change_flags(dev, flags); | |
733 | ||
734 | if (ret < 0) { | |
735 | netdev_err(dev, | |
736 | "Failed to cycle device %s; route tables might be wrong!\n", | |
737 | dev->name); | |
738 | } | |
739 | } | |
740 | ||
193125db DA |
741 | static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) |
742 | { | |
bad53162 | 743 | int ret; |
193125db | 744 | |
29bf24af | 745 | ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL); |
193125db | 746 | if (ret < 0) |
74b20582 | 747 | return ret; |
193125db | 748 | |
fee6d4c7 | 749 | port_dev->priv_flags |= IFF_L3MDEV_SLAVE; |
193125db DA |
750 | cycle_netdev(port_dev); |
751 | ||
752 | return 0; | |
193125db DA |
753 | } |
754 | ||
755 | static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) | |
756 | { | |
fee6d4c7 | 757 | if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev)) |
193125db DA |
758 | return -EINVAL; |
759 | ||
760 | return do_vrf_add_slave(dev, port_dev); | |
761 | } | |
762 | ||
763 | /* inverse of do_vrf_add_slave */ | |
764 | static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) | |
765 | { | |
193125db | 766 | netdev_upper_dev_unlink(port_dev, dev); |
fee6d4c7 | 767 | port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; |
193125db | 768 | |
193125db DA |
769 | cycle_netdev(port_dev); |
770 | ||
193125db DA |
771 | return 0; |
772 | } | |
773 | ||
774 | static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) | |
775 | { | |
193125db DA |
776 | return do_vrf_del_slave(dev, port_dev); |
777 | } | |
778 | ||
779 | static void vrf_dev_uninit(struct net_device *dev) | |
780 | { | |
781 | struct net_vrf *vrf = netdev_priv(dev); | |
bad53162 NA |
782 | struct net_device *port_dev; |
783 | struct list_head *iter; | |
193125db | 784 | |
810e530b DA |
785 | vrf_rtable_release(dev, vrf); |
786 | vrf_rt6_release(dev, vrf); | |
193125db | 787 | |
bad53162 NA |
788 | netdev_for_each_lower_dev(dev, port_dev, iter) |
789 | vrf_del_slave(dev, port_dev); | |
193125db | 790 | |
3a4a27d3 | 791 | free_percpu(dev->dstats); |
193125db DA |
792 | dev->dstats = NULL; |
793 | } | |
794 | ||
795 | static int vrf_dev_init(struct net_device *dev) | |
796 | { | |
797 | struct net_vrf *vrf = netdev_priv(dev); | |
798 | ||
193125db DA |
799 | dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); |
800 | if (!dev->dstats) | |
801 | goto out_nomem; | |
802 | ||
803 | /* create the default dst which points back to us */ | |
b0e95ccd | 804 | if (vrf_rtable_create(dev) != 0) |
193125db DA |
805 | goto out_stats; |
806 | ||
35402e31 DA |
807 | if (vrf_rt6_create(dev) != 0) |
808 | goto out_rth; | |
809 | ||
193125db DA |
810 | dev->flags = IFF_MASTER | IFF_NOARP; |
811 | ||
b87ab6b8 DA |
812 | /* MTU is irrelevant for VRF device; set to 64k similar to lo */ |
813 | dev->mtu = 64 * 1024; | |
814 | ||
815 | /* similarly, oper state is irrelevant; set to up to avoid confusion */ | |
816 | dev->operstate = IF_OPER_UP; | |
78e7a2ae | 817 | netdev_lockdep_set_classes(dev); |
193125db DA |
818 | return 0; |
819 | ||
35402e31 | 820 | out_rth: |
810e530b | 821 | vrf_rtable_release(dev, vrf); |
193125db DA |
822 | out_stats: |
823 | free_percpu(dev->dstats); | |
824 | dev->dstats = NULL; | |
825 | out_nomem: | |
826 | return -ENOMEM; | |
827 | } | |
828 | ||
829 | static const struct net_device_ops vrf_netdev_ops = { | |
830 | .ndo_init = vrf_dev_init, | |
831 | .ndo_uninit = vrf_dev_uninit, | |
832 | .ndo_start_xmit = vrf_xmit, | |
833 | .ndo_get_stats64 = vrf_get_stats64, | |
834 | .ndo_add_slave = vrf_add_slave, | |
835 | .ndo_del_slave = vrf_del_slave, | |
836 | }; | |
837 | ||
ee15ee5d DA |
838 | static u32 vrf_fib_table(const struct net_device *dev) |
839 | { | |
840 | struct net_vrf *vrf = netdev_priv(dev); | |
841 | ||
842 | return vrf->tb_id; | |
843 | } | |
844 | ||
73e20b76 DA |
845 | static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) |
846 | { | |
847 | return 0; | |
848 | } | |
849 | ||
850 | static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, | |
851 | struct sk_buff *skb, | |
852 | struct net_device *dev) | |
853 | { | |
854 | struct net *net = dev_net(dev); | |
855 | ||
856 | nf_reset(skb); | |
857 | ||
858 | if (NF_HOOK(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) < 0) | |
859 | skb = NULL; /* kfree_skb(skb) handled by nf code */ | |
860 | ||
861 | return skb; | |
862 | } | |
863 | ||
74b20582 DA |
864 | #if IS_ENABLED(CONFIG_IPV6) |
865 | /* neighbor handling is done with actual device; do not want | |
866 | * to flip skb->dev for those ndisc packets. This really fails | |
867 | * for multiple next protocols (e.g., NEXTHDR_HOP). But it is | |
868 | * a start. | |
869 | */ | |
870 | static bool ipv6_ndisc_frame(const struct sk_buff *skb) | |
871 | { | |
872 | const struct ipv6hdr *iph = ipv6_hdr(skb); | |
873 | bool rc = false; | |
874 | ||
875 | if (iph->nexthdr == NEXTHDR_ICMP) { | |
876 | const struct icmp6hdr *icmph; | |
877 | struct icmp6hdr _icmph; | |
878 | ||
879 | icmph = skb_header_pointer(skb, sizeof(*iph), | |
880 | sizeof(_icmph), &_icmph); | |
881 | if (!icmph) | |
882 | goto out; | |
883 | ||
884 | switch (icmph->icmp6_type) { | |
885 | case NDISC_ROUTER_SOLICITATION: | |
886 | case NDISC_ROUTER_ADVERTISEMENT: | |
887 | case NDISC_NEIGHBOUR_SOLICITATION: | |
888 | case NDISC_NEIGHBOUR_ADVERTISEMENT: | |
889 | case NDISC_REDIRECT: | |
890 | rc = true; | |
891 | break; | |
892 | } | |
893 | } | |
894 | ||
895 | out: | |
896 | return rc; | |
897 | } | |
898 | ||
9ff74384 DA |
899 | static struct rt6_info *vrf_ip6_route_lookup(struct net *net, |
900 | const struct net_device *dev, | |
901 | struct flowi6 *fl6, | |
902 | int ifindex, | |
903 | int flags) | |
904 | { | |
905 | struct net_vrf *vrf = netdev_priv(dev); | |
906 | struct fib6_table *table = NULL; | |
907 | struct rt6_info *rt6; | |
908 | ||
909 | rcu_read_lock(); | |
910 | ||
911 | /* fib6_table does not have a refcnt and can not be freed */ | |
912 | rt6 = rcu_dereference(vrf->rt6); | |
913 | if (likely(rt6)) | |
914 | table = rt6->rt6i_table; | |
915 | ||
916 | rcu_read_unlock(); | |
917 | ||
918 | if (!table) | |
919 | return NULL; | |
920 | ||
921 | return ip6_pol_route(net, table, ifindex, fl6, flags); | |
922 | } | |
923 | ||
924 | static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, | |
925 | int ifindex) | |
926 | { | |
927 | const struct ipv6hdr *iph = ipv6_hdr(skb); | |
928 | struct flowi6 fl6 = { | |
929 | .daddr = iph->daddr, | |
930 | .saddr = iph->saddr, | |
931 | .flowlabel = ip6_flowinfo(iph), | |
932 | .flowi6_mark = skb->mark, | |
933 | .flowi6_proto = iph->nexthdr, | |
934 | .flowi6_iif = ifindex, | |
935 | }; | |
936 | struct net *net = dev_net(vrf_dev); | |
937 | struct rt6_info *rt6; | |
938 | ||
939 | rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, | |
940 | RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); | |
941 | if (unlikely(!rt6)) | |
942 | return; | |
943 | ||
944 | if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) | |
945 | return; | |
946 | ||
947 | skb_dst_set(skb, &rt6->dst); | |
948 | } | |
949 | ||
74b20582 DA |
950 | static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, |
951 | struct sk_buff *skb) | |
952 | { | |
9ff74384 DA |
953 | int orig_iif = skb->skb_iif; |
954 | bool need_strict; | |
955 | ||
b4869aa2 DA |
956 | /* loopback traffic; do not push through packet taps again. |
957 | * Reset pkt_type for upper layers to process skb | |
958 | */ | |
959 | if (skb->pkt_type == PACKET_LOOPBACK) { | |
960 | skb->dev = vrf_dev; | |
961 | skb->skb_iif = vrf_dev->ifindex; | |
962 | skb->pkt_type = PACKET_HOST; | |
963 | goto out; | |
964 | } | |
965 | ||
9ff74384 DA |
966 | /* if packet is NDISC or addressed to multicast or link-local |
967 | * then keep the ingress interface | |
968 | */ | |
969 | need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); | |
970 | if (!ipv6_ndisc_frame(skb) && !need_strict) { | |
74b20582 DA |
971 | skb->dev = vrf_dev; |
972 | skb->skb_iif = vrf_dev->ifindex; | |
973 | ||
974 | skb_push(skb, skb->mac_len); | |
975 | dev_queue_xmit_nit(skb, vrf_dev); | |
976 | skb_pull(skb, skb->mac_len); | |
977 | ||
978 | IP6CB(skb)->flags |= IP6SKB_L3SLAVE; | |
979 | } | |
980 | ||
9ff74384 DA |
981 | if (need_strict) |
982 | vrf_ip6_input_dst(skb, vrf_dev, orig_iif); | |
983 | ||
73e20b76 | 984 | skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); |
b4869aa2 | 985 | out: |
74b20582 DA |
986 | return skb; |
987 | } | |
988 | ||
989 | #else | |
990 | static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, | |
991 | struct sk_buff *skb) | |
992 | { | |
993 | return skb; | |
994 | } | |
995 | #endif | |
996 | ||
997 | static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, | |
998 | struct sk_buff *skb) | |
999 | { | |
1000 | skb->dev = vrf_dev; | |
1001 | skb->skb_iif = vrf_dev->ifindex; | |
1002 | ||
afe80a49 DA |
1003 | /* loopback traffic; do not push through packet taps again. |
1004 | * Reset pkt_type for upper layers to process skb | |
1005 | */ | |
1006 | if (skb->pkt_type == PACKET_LOOPBACK) { | |
1007 | skb->pkt_type = PACKET_HOST; | |
1008 | goto out; | |
1009 | } | |
1010 | ||
74b20582 DA |
1011 | skb_push(skb, skb->mac_len); |
1012 | dev_queue_xmit_nit(skb, vrf_dev); | |
1013 | skb_pull(skb, skb->mac_len); | |
1014 | ||
73e20b76 | 1015 | skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); |
afe80a49 | 1016 | out: |
74b20582 DA |
1017 | return skb; |
1018 | } | |
1019 | ||
1020 | /* called with rcu lock held */ | |
1021 | static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, | |
1022 | struct sk_buff *skb, | |
1023 | u16 proto) | |
1024 | { | |
1025 | switch (proto) { | |
1026 | case AF_INET: | |
1027 | return vrf_ip_rcv(vrf_dev, skb); | |
1028 | case AF_INET6: | |
1029 | return vrf_ip6_rcv(vrf_dev, skb); | |
1030 | } | |
1031 | ||
1032 | return skb; | |
1033 | } | |
1034 | ||
35402e31 | 1035 | #if IS_ENABLED(CONFIG_IPV6) |
4c1feac5 DA |
1036 | /* send to link-local or multicast address via interface enslaved to |
1037 | * VRF device. Force lookup to VRF table without changing flow struct | |
1038 | */ | |
1039 | static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, | |
1040 | struct flowi6 *fl6) | |
35402e31 | 1041 | { |
9ff74384 | 1042 | struct net *net = dev_net(dev); |
4c1feac5 | 1043 | int flags = RT6_LOOKUP_F_IFACE; |
b0e95ccd | 1044 | struct dst_entry *dst = NULL; |
9ff74384 | 1045 | struct rt6_info *rt; |
35402e31 | 1046 | |
4c1feac5 DA |
1047 | /* VRF device does not have a link-local address and |
1048 | * sending packets to link-local or mcast addresses over | |
1049 | * a VRF device does not make sense | |
1050 | */ | |
1051 | if (fl6->flowi6_oif == dev->ifindex) { | |
1052 | dst = &net->ipv6.ip6_null_entry->dst; | |
1053 | dst_hold(dst); | |
1054 | return dst; | |
35402e31 DA |
1055 | } |
1056 | ||
4c1feac5 DA |
1057 | if (!ipv6_addr_any(&fl6->saddr)) |
1058 | flags |= RT6_LOOKUP_F_HAS_SADDR; | |
1059 | ||
1060 | rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags); | |
1061 | if (rt) | |
1062 | dst = &rt->dst; | |
9ff74384 | 1063 | |
b0e95ccd | 1064 | return dst; |
35402e31 DA |
1065 | } |
1066 | #endif | |
1067 | ||
ee15ee5d DA |
1068 | static const struct l3mdev_ops vrf_l3mdev_ops = { |
1069 | .l3mdev_fib_table = vrf_fib_table, | |
74b20582 | 1070 | .l3mdev_l3_rcv = vrf_l3_rcv, |
ebfc102c | 1071 | .l3mdev_l3_out = vrf_l3_out, |
35402e31 | 1072 | #if IS_ENABLED(CONFIG_IPV6) |
4c1feac5 | 1073 | .l3mdev_link_scope_lookup = vrf_link_scope_lookup, |
35402e31 | 1074 | #endif |
ee15ee5d DA |
1075 | }; |
1076 | ||
193125db DA |
1077 | static void vrf_get_drvinfo(struct net_device *dev, |
1078 | struct ethtool_drvinfo *info) | |
1079 | { | |
1080 | strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); | |
1081 | strlcpy(info->version, DRV_VERSION, sizeof(info->version)); | |
1082 | } | |
1083 | ||
1084 | static const struct ethtool_ops vrf_ethtool_ops = { | |
1085 | .get_drvinfo = vrf_get_drvinfo, | |
1086 | }; | |
1087 | ||
1aa6c4f6 DA |
1088 | static inline size_t vrf_fib_rule_nl_size(void) |
1089 | { | |
1090 | size_t sz; | |
1091 | ||
1092 | sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); | |
1093 | sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ | |
1094 | sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ | |
1095 | ||
1096 | return sz; | |
1097 | } | |
1098 | ||
1099 | static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) | |
1100 | { | |
1101 | struct fib_rule_hdr *frh; | |
1102 | struct nlmsghdr *nlh; | |
1103 | struct sk_buff *skb; | |
1104 | int err; | |
1105 | ||
e4348637 DA |
1106 | if (family == AF_INET6 && !ipv6_mod_enabled()) |
1107 | return 0; | |
1108 | ||
1aa6c4f6 DA |
1109 | skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); |
1110 | if (!skb) | |
1111 | return -ENOMEM; | |
1112 | ||
1113 | nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); | |
1114 | if (!nlh) | |
1115 | goto nla_put_failure; | |
1116 | ||
1117 | /* rule only needs to appear once */ | |
1118 | nlh->nlmsg_flags &= NLM_F_EXCL; | |
1119 | ||
1120 | frh = nlmsg_data(nlh); | |
1121 | memset(frh, 0, sizeof(*frh)); | |
1122 | frh->family = family; | |
1123 | frh->action = FR_ACT_TO_TBL; | |
1124 | ||
1125 | if (nla_put_u32(skb, FRA_L3MDEV, 1)) | |
1126 | goto nla_put_failure; | |
1127 | ||
1128 | if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) | |
1129 | goto nla_put_failure; | |
1130 | ||
1131 | nlmsg_end(skb, nlh); | |
1132 | ||
1133 | /* fib_nl_{new,del}rule handling looks for net from skb->sk */ | |
1134 | skb->sk = dev_net(dev)->rtnl; | |
1135 | if (add_it) { | |
1136 | err = fib_nl_newrule(skb, nlh); | |
1137 | if (err == -EEXIST) | |
1138 | err = 0; | |
1139 | } else { | |
1140 | err = fib_nl_delrule(skb, nlh); | |
1141 | if (err == -ENOENT) | |
1142 | err = 0; | |
1143 | } | |
1144 | nlmsg_free(skb); | |
1145 | ||
1146 | return err; | |
1147 | ||
1148 | nla_put_failure: | |
1149 | nlmsg_free(skb); | |
1150 | ||
1151 | return -EMSGSIZE; | |
1152 | } | |
1153 | ||
1154 | static int vrf_add_fib_rules(const struct net_device *dev) | |
1155 | { | |
1156 | int err; | |
1157 | ||
1158 | err = vrf_fib_rule(dev, AF_INET, true); | |
1159 | if (err < 0) | |
1160 | goto out_err; | |
1161 | ||
1162 | err = vrf_fib_rule(dev, AF_INET6, true); | |
1163 | if (err < 0) | |
1164 | goto ipv6_err; | |
1165 | ||
1166 | return 0; | |
1167 | ||
1168 | ipv6_err: | |
1169 | vrf_fib_rule(dev, AF_INET, false); | |
1170 | ||
1171 | out_err: | |
1172 | netdev_err(dev, "Failed to add FIB rules.\n"); | |
1173 | return err; | |
1174 | } | |
1175 | ||
193125db DA |
1176 | static void vrf_setup(struct net_device *dev) |
1177 | { | |
1178 | ether_setup(dev); | |
1179 | ||
1180 | /* Initialize the device structure. */ | |
1181 | dev->netdev_ops = &vrf_netdev_ops; | |
ee15ee5d | 1182 | dev->l3mdev_ops = &vrf_l3mdev_ops; |
193125db DA |
1183 | dev->ethtool_ops = &vrf_ethtool_ops; |
1184 | dev->destructor = free_netdev; | |
1185 | ||
1186 | /* Fill in device structure with ethernet-generic values. */ | |
1187 | eth_hw_addr_random(dev); | |
1188 | ||
1189 | /* don't acquire vrf device's netif_tx_lock when transmitting */ | |
1190 | dev->features |= NETIF_F_LLTX; | |
1191 | ||
1192 | /* don't allow vrf devices to change network namespaces. */ | |
1193 | dev->features |= NETIF_F_NETNS_LOCAL; | |
7889681f DA |
1194 | |
1195 | /* does not make sense for a VLAN to be added to a vrf device */ | |
1196 | dev->features |= NETIF_F_VLAN_CHALLENGED; | |
1197 | ||
1198 | /* enable offload features */ | |
1199 | dev->features |= NETIF_F_GSO_SOFTWARE; | |
1200 | dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM; | |
1201 | dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; | |
1202 | ||
1203 | dev->hw_features = dev->features; | |
1204 | dev->hw_enc_features = dev->features; | |
1205 | ||
1206 | /* default to no qdisc; user can add if desired */ | |
1207 | dev->priv_flags |= IFF_NO_QUEUE; | |
193125db DA |
1208 | } |
1209 | ||
1210 | static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) | |
1211 | { | |
1212 | if (tb[IFLA_ADDRESS]) { | |
1213 | if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) | |
1214 | return -EINVAL; | |
1215 | if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) | |
1216 | return -EADDRNOTAVAIL; | |
1217 | } | |
1218 | return 0; | |
1219 | } | |
1220 | ||
1221 | static void vrf_dellink(struct net_device *dev, struct list_head *head) | |
1222 | { | |
193125db DA |
1223 | unregister_netdevice_queue(dev, head); |
1224 | } | |
1225 | ||
1226 | static int vrf_newlink(struct net *src_net, struct net_device *dev, | |
1227 | struct nlattr *tb[], struct nlattr *data[]) | |
1228 | { | |
1229 | struct net_vrf *vrf = netdev_priv(dev); | |
1aa6c4f6 | 1230 | int err; |
193125db DA |
1231 | |
1232 | if (!data || !data[IFLA_VRF_TABLE]) | |
1233 | return -EINVAL; | |
1234 | ||
1235 | vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); | |
1236 | ||
007979ea | 1237 | dev->priv_flags |= IFF_L3MDEV_MASTER; |
193125db | 1238 | |
1aa6c4f6 DA |
1239 | err = register_netdevice(dev); |
1240 | if (err) | |
1241 | goto out; | |
1242 | ||
1243 | if (add_fib_rules) { | |
1244 | err = vrf_add_fib_rules(dev); | |
1245 | if (err) { | |
1246 | unregister_netdevice(dev); | |
1247 | goto out; | |
1248 | } | |
1249 | add_fib_rules = false; | |
1250 | } | |
1251 | ||
1252 | out: | |
1253 | return err; | |
193125db DA |
1254 | } |
1255 | ||
1256 | static size_t vrf_nl_getsize(const struct net_device *dev) | |
1257 | { | |
1258 | return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ | |
1259 | } | |
1260 | ||
1261 | static int vrf_fillinfo(struct sk_buff *skb, | |
1262 | const struct net_device *dev) | |
1263 | { | |
1264 | struct net_vrf *vrf = netdev_priv(dev); | |
1265 | ||
1266 | return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); | |
1267 | } | |
1268 | ||
67eb0331 DA |
1269 | static size_t vrf_get_slave_size(const struct net_device *bond_dev, |
1270 | const struct net_device *slave_dev) | |
1271 | { | |
1272 | return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ | |
1273 | } | |
1274 | ||
1275 | static int vrf_fill_slave_info(struct sk_buff *skb, | |
1276 | const struct net_device *vrf_dev, | |
1277 | const struct net_device *slave_dev) | |
1278 | { | |
1279 | struct net_vrf *vrf = netdev_priv(vrf_dev); | |
1280 | ||
1281 | if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) | |
1282 | return -EMSGSIZE; | |
1283 | ||
1284 | return 0; | |
1285 | } | |
1286 | ||
193125db DA |
1287 | static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { |
1288 | [IFLA_VRF_TABLE] = { .type = NLA_U32 }, | |
1289 | }; | |
1290 | ||
1291 | static struct rtnl_link_ops vrf_link_ops __read_mostly = { | |
1292 | .kind = DRV_NAME, | |
1293 | .priv_size = sizeof(struct net_vrf), | |
1294 | ||
1295 | .get_size = vrf_nl_getsize, | |
1296 | .policy = vrf_nl_policy, | |
1297 | .validate = vrf_validate, | |
1298 | .fill_info = vrf_fillinfo, | |
1299 | ||
67eb0331 DA |
1300 | .get_slave_size = vrf_get_slave_size, |
1301 | .fill_slave_info = vrf_fill_slave_info, | |
1302 | ||
193125db DA |
1303 | .newlink = vrf_newlink, |
1304 | .dellink = vrf_dellink, | |
1305 | .setup = vrf_setup, | |
1306 | .maxtype = IFLA_VRF_MAX, | |
1307 | }; | |
1308 | ||
1309 | static int vrf_device_event(struct notifier_block *unused, | |
1310 | unsigned long event, void *ptr) | |
1311 | { | |
1312 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); | |
1313 | ||
1314 | /* only care about unregister events to drop slave references */ | |
1315 | if (event == NETDEV_UNREGISTER) { | |
193125db DA |
1316 | struct net_device *vrf_dev; |
1317 | ||
fee6d4c7 | 1318 | if (!netif_is_l3_slave(dev)) |
193125db DA |
1319 | goto out; |
1320 | ||
58aa9087 NA |
1321 | vrf_dev = netdev_master_upper_dev_get(dev); |
1322 | vrf_del_slave(vrf_dev, dev); | |
193125db DA |
1323 | } |
1324 | out: | |
1325 | return NOTIFY_DONE; | |
1326 | } | |
1327 | ||
1328 | static struct notifier_block vrf_notifier_block __read_mostly = { | |
1329 | .notifier_call = vrf_device_event, | |
1330 | }; | |
1331 | ||
1332 | static int __init vrf_init_module(void) | |
1333 | { | |
1334 | int rc; | |
1335 | ||
193125db DA |
1336 | register_netdevice_notifier(&vrf_notifier_block); |
1337 | ||
1338 | rc = rtnl_link_register(&vrf_link_ops); | |
1339 | if (rc < 0) | |
1340 | goto error; | |
1341 | ||
1342 | return 0; | |
1343 | ||
1344 | error: | |
1345 | unregister_netdevice_notifier(&vrf_notifier_block); | |
193125db DA |
1346 | return rc; |
1347 | } | |
1348 | ||
193125db | 1349 | module_init(vrf_init_module); |
193125db DA |
1350 | MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); |
1351 | MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); | |
1352 | MODULE_LICENSE("GPL"); | |
1353 | MODULE_ALIAS_RTNL_LINK(DRV_NAME); | |
1354 | MODULE_VERSION(DRV_VERSION); |