]>
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 DA |
139 | #if IS_ENABLED(CONFIG_IPV6) |
140 | static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, | |
141 | struct net_device *dev) | |
142 | { | |
143 | const struct ipv6hdr *iph = ipv6_hdr(skb); | |
144 | struct net *net = dev_net(skb->dev); | |
145 | struct flowi6 fl6 = { | |
146 | /* needed to match OIF rule */ | |
147 | .flowi6_oif = dev->ifindex, | |
148 | .flowi6_iif = LOOPBACK_IFINDEX, | |
149 | .daddr = iph->daddr, | |
150 | .saddr = iph->saddr, | |
151 | .flowlabel = ip6_flowinfo(iph), | |
152 | .flowi6_mark = skb->mark, | |
153 | .flowi6_proto = iph->nexthdr, | |
154 | .flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF, | |
155 | }; | |
156 | int ret = NET_XMIT_DROP; | |
157 | struct dst_entry *dst; | |
158 | struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; | |
159 | ||
160 | dst = ip6_route_output(net, NULL, &fl6); | |
161 | if (dst == dst_null) | |
162 | goto err; | |
163 | ||
164 | skb_dst_drop(skb); | |
b4869aa2 DA |
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 | ||
35402e31 DA |
205 | skb_dst_set(skb, dst); |
206 | ||
911a66fb DA |
207 | /* strip the ethernet header added for pass through VRF device */ |
208 | __skb_pull(skb, skb_network_offset(skb)); | |
209 | ||
35402e31 DA |
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; | |
217 | err: | |
218 | vrf_tx_error(dev, skb); | |
219 | return NET_XMIT_DROP; | |
220 | } | |
221 | #else | |
193125db DA |
222 | static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, |
223 | struct net_device *dev) | |
224 | { | |
57b8efa1 NA |
225 | vrf_tx_error(dev, skb); |
226 | return NET_XMIT_DROP; | |
193125db | 227 | } |
35402e31 | 228 | #endif |
193125db | 229 | |
193125db DA |
230 | static 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), | |
6e2895a8 | 240 | .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC | |
58189ca7 | 241 | FLOWI_FLAG_SKIP_NH_OIF, |
193125db DA |
242 | .daddr = ip4h->daddr, |
243 | }; | |
911a66fb DA |
244 | struct net *net = dev_net(vrf_dev); |
245 | struct rtable *rt; | |
246 | ||
247 | rt = ip_route_output_flow(net, &fl4, NULL); | |
248 | if (IS_ERR(rt)) | |
249 | goto err; | |
193125db | 250 | |
911a66fb DA |
251 | if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { |
252 | ip_rt_put(rt); | |
193125db | 253 | goto err; |
911a66fb DA |
254 | } |
255 | ||
256 | skb_dst_drop(skb); | |
afe80a49 DA |
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 | ||
911a66fb DA |
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)); | |
193125db DA |
289 | |
290 | if (!ip4h->saddr) { | |
291 | ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, | |
292 | RT_SCOPE_LINK); | |
293 | } | |
294 | ||
33224b16 | 295 | ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); |
193125db DA |
296 | if (unlikely(net_xmit_eval(ret))) |
297 | vrf_dev->stats.tx_errors++; | |
298 | else | |
299 | ret = NET_XMIT_SUCCESS; | |
300 | ||
301 | out: | |
302 | return ret; | |
303 | err: | |
57b8efa1 | 304 | vrf_tx_error(vrf_dev, skb); |
193125db DA |
305 | goto out; |
306 | } | |
307 | ||
308 | static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) | |
309 | { | |
310 | switch (skb->protocol) { | |
311 | case htons(ETH_P_IP): | |
312 | return vrf_process_v4_outbound(skb, dev); | |
313 | case htons(ETH_P_IPV6): | |
314 | return vrf_process_v6_outbound(skb, dev); | |
315 | default: | |
57b8efa1 | 316 | vrf_tx_error(dev, skb); |
193125db DA |
317 | return NET_XMIT_DROP; |
318 | } | |
319 | } | |
320 | ||
321 | static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) | |
322 | { | |
323 | netdev_tx_t ret = is_ip_tx_frame(skb, dev); | |
324 | ||
325 | if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { | |
326 | struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); | |
327 | ||
328 | u64_stats_update_begin(&dstats->syncp); | |
329 | dstats->tx_pkts++; | |
330 | dstats->tx_bytes += skb->len; | |
331 | u64_stats_update_end(&dstats->syncp); | |
332 | } else { | |
333 | this_cpu_inc(dev->dstats->tx_drps); | |
334 | } | |
335 | ||
336 | return ret; | |
337 | } | |
338 | ||
35402e31 | 339 | #if IS_ENABLED(CONFIG_IPV6) |
35402e31 DA |
340 | /* modelled after ip6_finish_output2 */ |
341 | static int vrf_finish_output6(struct net *net, struct sock *sk, | |
342 | struct sk_buff *skb) | |
343 | { | |
344 | struct dst_entry *dst = skb_dst(skb); | |
345 | struct net_device *dev = dst->dev; | |
346 | struct neighbour *neigh; | |
347 | struct in6_addr *nexthop; | |
348 | int ret; | |
349 | ||
350 | skb->protocol = htons(ETH_P_IPV6); | |
351 | skb->dev = dev; | |
352 | ||
353 | rcu_read_lock_bh(); | |
354 | nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); | |
355 | neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); | |
356 | if (unlikely(!neigh)) | |
357 | neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); | |
358 | if (!IS_ERR(neigh)) { | |
359 | ret = dst_neigh_output(dst, neigh, skb); | |
360 | rcu_read_unlock_bh(); | |
361 | return ret; | |
362 | } | |
363 | rcu_read_unlock_bh(); | |
364 | ||
365 | IP6_INC_STATS(dev_net(dst->dev), | |
366 | ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); | |
367 | kfree_skb(skb); | |
368 | return -EINVAL; | |
369 | } | |
370 | ||
371 | /* modelled after ip6_output */ | |
372 | static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) | |
373 | { | |
374 | return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, | |
375 | net, sk, skb, NULL, skb_dst(skb)->dev, | |
376 | vrf_finish_output6, | |
377 | !(IP6CB(skb)->flags & IP6SKB_REROUTED)); | |
378 | } | |
379 | ||
b0e95ccd | 380 | /* holding rtnl */ |
9ab179d8 | 381 | static void vrf_rt6_release(struct net_vrf *vrf) |
35402e31 | 382 | { |
b0e95ccd | 383 | struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); |
b4869aa2 | 384 | struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local); |
b0e95ccd | 385 | |
b4869aa2 DA |
386 | RCU_INIT_POINTER(vrf->rt6, NULL); |
387 | RCU_INIT_POINTER(vrf->rt6_local, NULL); | |
388 | synchronize_rcu(); | |
b0e95ccd DA |
389 | |
390 | if (rt6) | |
391 | dst_release(&rt6->dst); | |
b4869aa2 DA |
392 | |
393 | if (rt6_local) { | |
394 | if (rt6_local->rt6i_idev) | |
395 | in6_dev_put(rt6_local->rt6i_idev); | |
396 | ||
397 | dst_release(&rt6_local->dst); | |
398 | } | |
35402e31 DA |
399 | } |
400 | ||
401 | static int vrf_rt6_create(struct net_device *dev) | |
402 | { | |
b4869aa2 | 403 | int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE; |
35402e31 | 404 | struct net_vrf *vrf = netdev_priv(dev); |
9ab179d8 | 405 | struct net *net = dev_net(dev); |
b3b4663c | 406 | struct fib6_table *rt6i_table; |
b4869aa2 | 407 | struct rt6_info *rt6, *rt6_local; |
35402e31 DA |
408 | int rc = -ENOMEM; |
409 | ||
b3b4663c DA |
410 | rt6i_table = fib6_new_table(net, vrf->tb_id); |
411 | if (!rt6i_table) | |
412 | goto out; | |
413 | ||
b4869aa2 DA |
414 | /* create a dst for routing packets out a VRF device */ |
415 | rt6 = ip6_dst_alloc(net, dev, flags); | |
35402e31 DA |
416 | if (!rt6) |
417 | goto out; | |
418 | ||
9ab179d8 | 419 | dst_hold(&rt6->dst); |
b3b4663c DA |
420 | |
421 | rt6->rt6i_table = rt6i_table; | |
422 | rt6->dst.output = vrf_output6; | |
b4869aa2 DA |
423 | |
424 | /* create a dst for local routing - packets sent locally | |
425 | * to local address via the VRF device as a loopback | |
426 | */ | |
427 | rt6_local = ip6_dst_alloc(net, dev, flags); | |
428 | if (!rt6_local) { | |
429 | dst_release(&rt6->dst); | |
430 | goto out; | |
431 | } | |
432 | ||
433 | dst_hold(&rt6_local->dst); | |
434 | ||
435 | rt6_local->rt6i_idev = in6_dev_get(dev); | |
436 | rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL; | |
437 | rt6_local->rt6i_table = rt6i_table; | |
438 | rt6_local->dst.input = ip6_input; | |
439 | ||
b0e95ccd | 440 | rcu_assign_pointer(vrf->rt6, rt6); |
b4869aa2 | 441 | rcu_assign_pointer(vrf->rt6_local, rt6_local); |
b0e95ccd | 442 | |
35402e31 DA |
443 | rc = 0; |
444 | out: | |
445 | return rc; | |
446 | } | |
447 | #else | |
9ab179d8 | 448 | static void vrf_rt6_release(struct net_vrf *vrf) |
35402e31 DA |
449 | { |
450 | } | |
451 | ||
452 | static int vrf_rt6_create(struct net_device *dev) | |
453 | { | |
454 | return 0; | |
455 | } | |
456 | #endif | |
457 | ||
8f58336d | 458 | /* modelled after ip_finish_output2 */ |
0c4b51f0 | 459 | static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
193125db | 460 | { |
8f58336d DA |
461 | struct dst_entry *dst = skb_dst(skb); |
462 | struct rtable *rt = (struct rtable *)dst; | |
463 | struct net_device *dev = dst->dev; | |
464 | unsigned int hh_len = LL_RESERVED_SPACE(dev); | |
465 | struct neighbour *neigh; | |
466 | u32 nexthop; | |
467 | int ret = -EINVAL; | |
468 | ||
469 | /* Be paranoid, rather than too clever. */ | |
470 | if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { | |
471 | struct sk_buff *skb2; | |
472 | ||
473 | skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); | |
474 | if (!skb2) { | |
475 | ret = -ENOMEM; | |
476 | goto err; | |
477 | } | |
478 | if (skb->sk) | |
479 | skb_set_owner_w(skb2, skb->sk); | |
480 | ||
481 | consume_skb(skb); | |
482 | skb = skb2; | |
483 | } | |
484 | ||
485 | rcu_read_lock_bh(); | |
486 | ||
487 | nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); | |
488 | neigh = __ipv4_neigh_lookup_noref(dev, nexthop); | |
489 | if (unlikely(!neigh)) | |
490 | neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); | |
491 | if (!IS_ERR(neigh)) | |
492 | ret = dst_neigh_output(dst, neigh, skb); | |
493 | ||
494 | rcu_read_unlock_bh(); | |
495 | err: | |
496 | if (unlikely(ret < 0)) | |
497 | vrf_tx_error(skb->dev, skb); | |
498 | return ret; | |
193125db DA |
499 | } |
500 | ||
ede2059d | 501 | static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
193125db DA |
502 | { |
503 | struct net_device *dev = skb_dst(skb)->dev; | |
504 | ||
29a26a56 | 505 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); |
193125db DA |
506 | |
507 | skb->dev = dev; | |
508 | skb->protocol = htons(ETH_P_IP); | |
509 | ||
29a26a56 EB |
510 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
511 | net, sk, skb, NULL, dev, | |
8f58336d | 512 | vrf_finish_output, |
193125db DA |
513 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
514 | } | |
515 | ||
b0e95ccd | 516 | /* holding rtnl */ |
9ab179d8 | 517 | static void vrf_rtable_release(struct net_vrf *vrf) |
193125db | 518 | { |
b0e95ccd | 519 | struct rtable *rth = rtnl_dereference(vrf->rth); |
afe80a49 | 520 | struct rtable *rth_local = rtnl_dereference(vrf->rth_local); |
b0e95ccd | 521 | |
afe80a49 DA |
522 | RCU_INIT_POINTER(vrf->rth, NULL); |
523 | RCU_INIT_POINTER(vrf->rth_local, NULL); | |
524 | synchronize_rcu(); | |
193125db | 525 | |
b0e95ccd DA |
526 | if (rth) |
527 | dst_release(&rth->dst); | |
afe80a49 DA |
528 | |
529 | if (rth_local) | |
530 | dst_release(&rth_local->dst); | |
193125db DA |
531 | } |
532 | ||
b0e95ccd | 533 | static int vrf_rtable_create(struct net_device *dev) |
193125db | 534 | { |
b7503e0c | 535 | struct net_vrf *vrf = netdev_priv(dev); |
afe80a49 | 536 | struct rtable *rth, *rth_local; |
193125db | 537 | |
b3b4663c | 538 | if (!fib_new_table(dev_net(dev), vrf->tb_id)) |
b0e95ccd | 539 | return -ENOMEM; |
b3b4663c | 540 | |
afe80a49 | 541 | /* create a dst for routing packets out through a VRF device */ |
9ab179d8 | 542 | rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0); |
b0e95ccd DA |
543 | if (!rth) |
544 | return -ENOMEM; | |
193125db | 545 | |
afe80a49 DA |
546 | /* create a dst for local ingress routing - packets sent locally |
547 | * to local address via the VRF device as a loopback | |
548 | */ | |
549 | rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0); | |
550 | if (!rth_local) { | |
551 | dst_release(&rth->dst); | |
552 | return -ENOMEM; | |
553 | } | |
554 | ||
b0e95ccd DA |
555 | rth->dst.output = vrf_output; |
556 | rth->rt_table_id = vrf->tb_id; | |
557 | ||
afe80a49 DA |
558 | rth_local->rt_table_id = vrf->tb_id; |
559 | ||
b0e95ccd | 560 | rcu_assign_pointer(vrf->rth, rth); |
afe80a49 | 561 | rcu_assign_pointer(vrf->rth_local, rth_local); |
b0e95ccd DA |
562 | |
563 | return 0; | |
193125db DA |
564 | } |
565 | ||
566 | /**************************** device handling ********************/ | |
567 | ||
568 | /* cycle interface to flush neighbor cache and move routes across tables */ | |
569 | static void cycle_netdev(struct net_device *dev) | |
570 | { | |
571 | unsigned int flags = dev->flags; | |
572 | int ret; | |
573 | ||
574 | if (!netif_running(dev)) | |
575 | return; | |
576 | ||
577 | ret = dev_change_flags(dev, flags & ~IFF_UP); | |
578 | if (ret >= 0) | |
579 | ret = dev_change_flags(dev, flags); | |
580 | ||
581 | if (ret < 0) { | |
582 | netdev_err(dev, | |
583 | "Failed to cycle device %s; route tables might be wrong!\n", | |
584 | dev->name); | |
585 | } | |
586 | } | |
587 | ||
193125db DA |
588 | static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) |
589 | { | |
bad53162 | 590 | int ret; |
193125db | 591 | |
29bf24af | 592 | ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL); |
193125db | 593 | if (ret < 0) |
74b20582 | 594 | return ret; |
193125db | 595 | |
fee6d4c7 | 596 | port_dev->priv_flags |= IFF_L3MDEV_SLAVE; |
193125db DA |
597 | cycle_netdev(port_dev); |
598 | ||
599 | return 0; | |
193125db DA |
600 | } |
601 | ||
602 | static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) | |
603 | { | |
fee6d4c7 | 604 | if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev)) |
193125db DA |
605 | return -EINVAL; |
606 | ||
607 | return do_vrf_add_slave(dev, port_dev); | |
608 | } | |
609 | ||
610 | /* inverse of do_vrf_add_slave */ | |
611 | static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) | |
612 | { | |
193125db | 613 | netdev_upper_dev_unlink(port_dev, dev); |
fee6d4c7 | 614 | port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; |
193125db | 615 | |
193125db DA |
616 | cycle_netdev(port_dev); |
617 | ||
193125db DA |
618 | return 0; |
619 | } | |
620 | ||
621 | static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) | |
622 | { | |
193125db DA |
623 | return do_vrf_del_slave(dev, port_dev); |
624 | } | |
625 | ||
626 | static void vrf_dev_uninit(struct net_device *dev) | |
627 | { | |
628 | struct net_vrf *vrf = netdev_priv(dev); | |
bad53162 NA |
629 | struct net_device *port_dev; |
630 | struct list_head *iter; | |
193125db | 631 | |
9ab179d8 DA |
632 | vrf_rtable_release(vrf); |
633 | vrf_rt6_release(vrf); | |
193125db | 634 | |
bad53162 NA |
635 | netdev_for_each_lower_dev(dev, port_dev, iter) |
636 | vrf_del_slave(dev, port_dev); | |
193125db | 637 | |
3a4a27d3 | 638 | free_percpu(dev->dstats); |
193125db DA |
639 | dev->dstats = NULL; |
640 | } | |
641 | ||
642 | static int vrf_dev_init(struct net_device *dev) | |
643 | { | |
644 | struct net_vrf *vrf = netdev_priv(dev); | |
645 | ||
193125db DA |
646 | dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); |
647 | if (!dev->dstats) | |
648 | goto out_nomem; | |
649 | ||
650 | /* create the default dst which points back to us */ | |
b0e95ccd | 651 | if (vrf_rtable_create(dev) != 0) |
193125db DA |
652 | goto out_stats; |
653 | ||
35402e31 DA |
654 | if (vrf_rt6_create(dev) != 0) |
655 | goto out_rth; | |
656 | ||
193125db DA |
657 | dev->flags = IFF_MASTER | IFF_NOARP; |
658 | ||
b87ab6b8 DA |
659 | /* MTU is irrelevant for VRF device; set to 64k similar to lo */ |
660 | dev->mtu = 64 * 1024; | |
661 | ||
662 | /* similarly, oper state is irrelevant; set to up to avoid confusion */ | |
663 | dev->operstate = IF_OPER_UP; | |
78e7a2ae | 664 | netdev_lockdep_set_classes(dev); |
193125db DA |
665 | return 0; |
666 | ||
35402e31 | 667 | out_rth: |
9ab179d8 | 668 | vrf_rtable_release(vrf); |
193125db DA |
669 | out_stats: |
670 | free_percpu(dev->dstats); | |
671 | dev->dstats = NULL; | |
672 | out_nomem: | |
673 | return -ENOMEM; | |
674 | } | |
675 | ||
676 | static const struct net_device_ops vrf_netdev_ops = { | |
677 | .ndo_init = vrf_dev_init, | |
678 | .ndo_uninit = vrf_dev_uninit, | |
679 | .ndo_start_xmit = vrf_xmit, | |
680 | .ndo_get_stats64 = vrf_get_stats64, | |
681 | .ndo_add_slave = vrf_add_slave, | |
682 | .ndo_del_slave = vrf_del_slave, | |
683 | }; | |
684 | ||
ee15ee5d DA |
685 | static u32 vrf_fib_table(const struct net_device *dev) |
686 | { | |
687 | struct net_vrf *vrf = netdev_priv(dev); | |
688 | ||
689 | return vrf->tb_id; | |
690 | } | |
691 | ||
692 | static struct rtable *vrf_get_rtable(const struct net_device *dev, | |
693 | const struct flowi4 *fl4) | |
694 | { | |
695 | struct rtable *rth = NULL; | |
696 | ||
6e2895a8 | 697 | if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) { |
ee15ee5d DA |
698 | struct net_vrf *vrf = netdev_priv(dev); |
699 | ||
b0e95ccd DA |
700 | rcu_read_lock(); |
701 | ||
702 | rth = rcu_dereference(vrf->rth); | |
703 | if (likely(rth)) | |
704 | dst_hold(&rth->dst); | |
705 | ||
706 | rcu_read_unlock(); | |
ee15ee5d DA |
707 | } |
708 | ||
709 | return rth; | |
710 | } | |
711 | ||
8cbb512c | 712 | /* called under rcu_read_lock */ |
b5bdacf3 | 713 | static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4) |
8cbb512c DA |
714 | { |
715 | struct fib_result res = { .tclassid = 0 }; | |
716 | struct net *net = dev_net(dev); | |
717 | u32 orig_tos = fl4->flowi4_tos; | |
718 | u8 flags = fl4->flowi4_flags; | |
719 | u8 scope = fl4->flowi4_scope; | |
720 | u8 tos = RT_FL_TOS(fl4); | |
b5bdacf3 | 721 | int rc; |
8cbb512c DA |
722 | |
723 | if (unlikely(!fl4->daddr)) | |
b5bdacf3 | 724 | return 0; |
8cbb512c DA |
725 | |
726 | fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF; | |
727 | fl4->flowi4_iif = LOOPBACK_IFINDEX; | |
1ff23bee DA |
728 | /* make sure oif is set to VRF device for lookup */ |
729 | fl4->flowi4_oif = dev->ifindex; | |
8cbb512c DA |
730 | fl4->flowi4_tos = tos & IPTOS_RT_MASK; |
731 | fl4->flowi4_scope = ((tos & RTO_ONLINK) ? | |
732 | RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); | |
733 | ||
b5bdacf3 DA |
734 | rc = fib_lookup(net, fl4, &res, 0); |
735 | if (!rc) { | |
8cbb512c DA |
736 | if (res.type == RTN_LOCAL) |
737 | fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr; | |
738 | else | |
739 | fib_select_path(net, &res, fl4, -1); | |
740 | } | |
741 | ||
742 | fl4->flowi4_flags = flags; | |
743 | fl4->flowi4_tos = orig_tos; | |
744 | fl4->flowi4_scope = scope; | |
b5bdacf3 DA |
745 | |
746 | return rc; | |
8cbb512c DA |
747 | } |
748 | ||
74b20582 DA |
749 | #if IS_ENABLED(CONFIG_IPV6) |
750 | /* neighbor handling is done with actual device; do not want | |
751 | * to flip skb->dev for those ndisc packets. This really fails | |
752 | * for multiple next protocols (e.g., NEXTHDR_HOP). But it is | |
753 | * a start. | |
754 | */ | |
755 | static bool ipv6_ndisc_frame(const struct sk_buff *skb) | |
756 | { | |
757 | const struct ipv6hdr *iph = ipv6_hdr(skb); | |
758 | bool rc = false; | |
759 | ||
760 | if (iph->nexthdr == NEXTHDR_ICMP) { | |
761 | const struct icmp6hdr *icmph; | |
762 | struct icmp6hdr _icmph; | |
763 | ||
764 | icmph = skb_header_pointer(skb, sizeof(*iph), | |
765 | sizeof(_icmph), &_icmph); | |
766 | if (!icmph) | |
767 | goto out; | |
768 | ||
769 | switch (icmph->icmp6_type) { | |
770 | case NDISC_ROUTER_SOLICITATION: | |
771 | case NDISC_ROUTER_ADVERTISEMENT: | |
772 | case NDISC_NEIGHBOUR_SOLICITATION: | |
773 | case NDISC_NEIGHBOUR_ADVERTISEMENT: | |
774 | case NDISC_REDIRECT: | |
775 | rc = true; | |
776 | break; | |
777 | } | |
778 | } | |
779 | ||
780 | out: | |
781 | return rc; | |
782 | } | |
783 | ||
784 | static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, | |
785 | struct sk_buff *skb) | |
786 | { | |
b4869aa2 DA |
787 | /* loopback traffic; do not push through packet taps again. |
788 | * Reset pkt_type for upper layers to process skb | |
789 | */ | |
790 | if (skb->pkt_type == PACKET_LOOPBACK) { | |
791 | skb->dev = vrf_dev; | |
792 | skb->skb_iif = vrf_dev->ifindex; | |
793 | skb->pkt_type = PACKET_HOST; | |
794 | goto out; | |
795 | } | |
796 | ||
74b20582 DA |
797 | /* if packet is NDISC keep the ingress interface */ |
798 | if (!ipv6_ndisc_frame(skb)) { | |
799 | skb->dev = vrf_dev; | |
800 | skb->skb_iif = vrf_dev->ifindex; | |
801 | ||
802 | skb_push(skb, skb->mac_len); | |
803 | dev_queue_xmit_nit(skb, vrf_dev); | |
804 | skb_pull(skb, skb->mac_len); | |
805 | ||
806 | IP6CB(skb)->flags |= IP6SKB_L3SLAVE; | |
807 | } | |
808 | ||
b4869aa2 | 809 | out: |
74b20582 DA |
810 | return skb; |
811 | } | |
812 | ||
813 | #else | |
814 | static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, | |
815 | struct sk_buff *skb) | |
816 | { | |
817 | return skb; | |
818 | } | |
819 | #endif | |
820 | ||
821 | static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, | |
822 | struct sk_buff *skb) | |
823 | { | |
824 | skb->dev = vrf_dev; | |
825 | skb->skb_iif = vrf_dev->ifindex; | |
826 | ||
afe80a49 DA |
827 | /* loopback traffic; do not push through packet taps again. |
828 | * Reset pkt_type for upper layers to process skb | |
829 | */ | |
830 | if (skb->pkt_type == PACKET_LOOPBACK) { | |
831 | skb->pkt_type = PACKET_HOST; | |
832 | goto out; | |
833 | } | |
834 | ||
74b20582 DA |
835 | skb_push(skb, skb->mac_len); |
836 | dev_queue_xmit_nit(skb, vrf_dev); | |
837 | skb_pull(skb, skb->mac_len); | |
838 | ||
afe80a49 | 839 | out: |
74b20582 DA |
840 | return skb; |
841 | } | |
842 | ||
843 | /* called with rcu lock held */ | |
844 | static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, | |
845 | struct sk_buff *skb, | |
846 | u16 proto) | |
847 | { | |
848 | switch (proto) { | |
849 | case AF_INET: | |
850 | return vrf_ip_rcv(vrf_dev, skb); | |
851 | case AF_INET6: | |
852 | return vrf_ip6_rcv(vrf_dev, skb); | |
853 | } | |
854 | ||
855 | return skb; | |
856 | } | |
857 | ||
35402e31 DA |
858 | #if IS_ENABLED(CONFIG_IPV6) |
859 | static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev, | |
860 | const struct flowi6 *fl6) | |
861 | { | |
b0e95ccd | 862 | struct dst_entry *dst = NULL; |
35402e31 DA |
863 | |
864 | if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) { | |
865 | struct net_vrf *vrf = netdev_priv(dev); | |
b0e95ccd DA |
866 | struct rt6_info *rt; |
867 | ||
868 | rcu_read_lock(); | |
869 | ||
870 | rt = rcu_dereference(vrf->rt6); | |
871 | if (likely(rt)) { | |
872 | dst = &rt->dst; | |
873 | dst_hold(dst); | |
874 | } | |
35402e31 | 875 | |
b0e95ccd | 876 | rcu_read_unlock(); |
35402e31 DA |
877 | } |
878 | ||
b0e95ccd | 879 | return dst; |
35402e31 DA |
880 | } |
881 | #endif | |
882 | ||
ee15ee5d DA |
883 | static const struct l3mdev_ops vrf_l3mdev_ops = { |
884 | .l3mdev_fib_table = vrf_fib_table, | |
885 | .l3mdev_get_rtable = vrf_get_rtable, | |
8cbb512c | 886 | .l3mdev_get_saddr = vrf_get_saddr, |
74b20582 | 887 | .l3mdev_l3_rcv = vrf_l3_rcv, |
35402e31 DA |
888 | #if IS_ENABLED(CONFIG_IPV6) |
889 | .l3mdev_get_rt6_dst = vrf_get_rt6_dst, | |
890 | #endif | |
ee15ee5d DA |
891 | }; |
892 | ||
193125db DA |
893 | static void vrf_get_drvinfo(struct net_device *dev, |
894 | struct ethtool_drvinfo *info) | |
895 | { | |
896 | strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); | |
897 | strlcpy(info->version, DRV_VERSION, sizeof(info->version)); | |
898 | } | |
899 | ||
900 | static const struct ethtool_ops vrf_ethtool_ops = { | |
901 | .get_drvinfo = vrf_get_drvinfo, | |
902 | }; | |
903 | ||
1aa6c4f6 DA |
904 | static inline size_t vrf_fib_rule_nl_size(void) |
905 | { | |
906 | size_t sz; | |
907 | ||
908 | sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); | |
909 | sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ | |
910 | sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ | |
911 | ||
912 | return sz; | |
913 | } | |
914 | ||
915 | static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) | |
916 | { | |
917 | struct fib_rule_hdr *frh; | |
918 | struct nlmsghdr *nlh; | |
919 | struct sk_buff *skb; | |
920 | int err; | |
921 | ||
922 | skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); | |
923 | if (!skb) | |
924 | return -ENOMEM; | |
925 | ||
926 | nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); | |
927 | if (!nlh) | |
928 | goto nla_put_failure; | |
929 | ||
930 | /* rule only needs to appear once */ | |
931 | nlh->nlmsg_flags &= NLM_F_EXCL; | |
932 | ||
933 | frh = nlmsg_data(nlh); | |
934 | memset(frh, 0, sizeof(*frh)); | |
935 | frh->family = family; | |
936 | frh->action = FR_ACT_TO_TBL; | |
937 | ||
938 | if (nla_put_u32(skb, FRA_L3MDEV, 1)) | |
939 | goto nla_put_failure; | |
940 | ||
941 | if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) | |
942 | goto nla_put_failure; | |
943 | ||
944 | nlmsg_end(skb, nlh); | |
945 | ||
946 | /* fib_nl_{new,del}rule handling looks for net from skb->sk */ | |
947 | skb->sk = dev_net(dev)->rtnl; | |
948 | if (add_it) { | |
949 | err = fib_nl_newrule(skb, nlh); | |
950 | if (err == -EEXIST) | |
951 | err = 0; | |
952 | } else { | |
953 | err = fib_nl_delrule(skb, nlh); | |
954 | if (err == -ENOENT) | |
955 | err = 0; | |
956 | } | |
957 | nlmsg_free(skb); | |
958 | ||
959 | return err; | |
960 | ||
961 | nla_put_failure: | |
962 | nlmsg_free(skb); | |
963 | ||
964 | return -EMSGSIZE; | |
965 | } | |
966 | ||
967 | static int vrf_add_fib_rules(const struct net_device *dev) | |
968 | { | |
969 | int err; | |
970 | ||
971 | err = vrf_fib_rule(dev, AF_INET, true); | |
972 | if (err < 0) | |
973 | goto out_err; | |
974 | ||
975 | err = vrf_fib_rule(dev, AF_INET6, true); | |
976 | if (err < 0) | |
977 | goto ipv6_err; | |
978 | ||
979 | return 0; | |
980 | ||
981 | ipv6_err: | |
982 | vrf_fib_rule(dev, AF_INET, false); | |
983 | ||
984 | out_err: | |
985 | netdev_err(dev, "Failed to add FIB rules.\n"); | |
986 | return err; | |
987 | } | |
988 | ||
193125db DA |
989 | static void vrf_setup(struct net_device *dev) |
990 | { | |
991 | ether_setup(dev); | |
992 | ||
993 | /* Initialize the device structure. */ | |
994 | dev->netdev_ops = &vrf_netdev_ops; | |
ee15ee5d | 995 | dev->l3mdev_ops = &vrf_l3mdev_ops; |
193125db DA |
996 | dev->ethtool_ops = &vrf_ethtool_ops; |
997 | dev->destructor = free_netdev; | |
998 | ||
999 | /* Fill in device structure with ethernet-generic values. */ | |
1000 | eth_hw_addr_random(dev); | |
1001 | ||
1002 | /* don't acquire vrf device's netif_tx_lock when transmitting */ | |
1003 | dev->features |= NETIF_F_LLTX; | |
1004 | ||
1005 | /* don't allow vrf devices to change network namespaces. */ | |
1006 | dev->features |= NETIF_F_NETNS_LOCAL; | |
1007 | } | |
1008 | ||
1009 | static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) | |
1010 | { | |
1011 | if (tb[IFLA_ADDRESS]) { | |
1012 | if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) | |
1013 | return -EINVAL; | |
1014 | if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) | |
1015 | return -EADDRNOTAVAIL; | |
1016 | } | |
1017 | return 0; | |
1018 | } | |
1019 | ||
1020 | static void vrf_dellink(struct net_device *dev, struct list_head *head) | |
1021 | { | |
193125db DA |
1022 | unregister_netdevice_queue(dev, head); |
1023 | } | |
1024 | ||
1025 | static int vrf_newlink(struct net *src_net, struct net_device *dev, | |
1026 | struct nlattr *tb[], struct nlattr *data[]) | |
1027 | { | |
1028 | struct net_vrf *vrf = netdev_priv(dev); | |
1aa6c4f6 | 1029 | int err; |
193125db DA |
1030 | |
1031 | if (!data || !data[IFLA_VRF_TABLE]) | |
1032 | return -EINVAL; | |
1033 | ||
1034 | vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); | |
1035 | ||
007979ea | 1036 | dev->priv_flags |= IFF_L3MDEV_MASTER; |
193125db | 1037 | |
1aa6c4f6 DA |
1038 | err = register_netdevice(dev); |
1039 | if (err) | |
1040 | goto out; | |
1041 | ||
1042 | if (add_fib_rules) { | |
1043 | err = vrf_add_fib_rules(dev); | |
1044 | if (err) { | |
1045 | unregister_netdevice(dev); | |
1046 | goto out; | |
1047 | } | |
1048 | add_fib_rules = false; | |
1049 | } | |
1050 | ||
1051 | out: | |
1052 | return err; | |
193125db DA |
1053 | } |
1054 | ||
1055 | static size_t vrf_nl_getsize(const struct net_device *dev) | |
1056 | { | |
1057 | return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ | |
1058 | } | |
1059 | ||
1060 | static int vrf_fillinfo(struct sk_buff *skb, | |
1061 | const struct net_device *dev) | |
1062 | { | |
1063 | struct net_vrf *vrf = netdev_priv(dev); | |
1064 | ||
1065 | return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); | |
1066 | } | |
1067 | ||
67eb0331 DA |
1068 | static size_t vrf_get_slave_size(const struct net_device *bond_dev, |
1069 | const struct net_device *slave_dev) | |
1070 | { | |
1071 | return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ | |
1072 | } | |
1073 | ||
1074 | static int vrf_fill_slave_info(struct sk_buff *skb, | |
1075 | const struct net_device *vrf_dev, | |
1076 | const struct net_device *slave_dev) | |
1077 | { | |
1078 | struct net_vrf *vrf = netdev_priv(vrf_dev); | |
1079 | ||
1080 | if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) | |
1081 | return -EMSGSIZE; | |
1082 | ||
1083 | return 0; | |
1084 | } | |
1085 | ||
193125db DA |
1086 | static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { |
1087 | [IFLA_VRF_TABLE] = { .type = NLA_U32 }, | |
1088 | }; | |
1089 | ||
1090 | static struct rtnl_link_ops vrf_link_ops __read_mostly = { | |
1091 | .kind = DRV_NAME, | |
1092 | .priv_size = sizeof(struct net_vrf), | |
1093 | ||
1094 | .get_size = vrf_nl_getsize, | |
1095 | .policy = vrf_nl_policy, | |
1096 | .validate = vrf_validate, | |
1097 | .fill_info = vrf_fillinfo, | |
1098 | ||
67eb0331 DA |
1099 | .get_slave_size = vrf_get_slave_size, |
1100 | .fill_slave_info = vrf_fill_slave_info, | |
1101 | ||
193125db DA |
1102 | .newlink = vrf_newlink, |
1103 | .dellink = vrf_dellink, | |
1104 | .setup = vrf_setup, | |
1105 | .maxtype = IFLA_VRF_MAX, | |
1106 | }; | |
1107 | ||
1108 | static int vrf_device_event(struct notifier_block *unused, | |
1109 | unsigned long event, void *ptr) | |
1110 | { | |
1111 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); | |
1112 | ||
1113 | /* only care about unregister events to drop slave references */ | |
1114 | if (event == NETDEV_UNREGISTER) { | |
193125db DA |
1115 | struct net_device *vrf_dev; |
1116 | ||
fee6d4c7 | 1117 | if (!netif_is_l3_slave(dev)) |
193125db DA |
1118 | goto out; |
1119 | ||
58aa9087 NA |
1120 | vrf_dev = netdev_master_upper_dev_get(dev); |
1121 | vrf_del_slave(vrf_dev, dev); | |
193125db DA |
1122 | } |
1123 | out: | |
1124 | return NOTIFY_DONE; | |
1125 | } | |
1126 | ||
1127 | static struct notifier_block vrf_notifier_block __read_mostly = { | |
1128 | .notifier_call = vrf_device_event, | |
1129 | }; | |
1130 | ||
1131 | static int __init vrf_init_module(void) | |
1132 | { | |
1133 | int rc; | |
1134 | ||
193125db DA |
1135 | register_netdevice_notifier(&vrf_notifier_block); |
1136 | ||
1137 | rc = rtnl_link_register(&vrf_link_ops); | |
1138 | if (rc < 0) | |
1139 | goto error; | |
1140 | ||
1141 | return 0; | |
1142 | ||
1143 | error: | |
1144 | unregister_netdevice_notifier(&vrf_notifier_block); | |
193125db DA |
1145 | return rc; |
1146 | } | |
1147 | ||
193125db | 1148 | module_init(vrf_init_module); |
193125db DA |
1149 | MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); |
1150 | MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); | |
1151 | MODULE_LICENSE("GPL"); | |
1152 | MODULE_ALIAS_RTNL_LINK(DRV_NAME); | |
1153 | MODULE_VERSION(DRV_VERSION); |