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