blob: cfc30ce4c6e1acec8c152cfb7e292bda54b504ea [file] [log] [blame]
Thomas Gleixner2874c5f2019-05-27 08:55:01 +02001// SPDX-License-Identifier: GPL-2.0-or-later
David Ahern193125d2015-08-13 14:59:10 -06002/*
3 * vrf.c: device driver to encapsulate a VRF space
4 *
5 * Copyright (c) 2015 Cumulus Networks. All rights reserved.
6 * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com>
7 * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com>
8 *
9 * Based on dummy, team and ipvlan drivers
David Ahern193125d2015-08-13 14:59:10 -060010 */
11
Jakub Kicinskicc698372020-11-20 14:50:52 -080012#include <linux/ethtool.h>
David Ahern193125d2015-08-13 14:59:10 -060013#include <linux/module.h>
14#include <linux/kernel.h>
15#include <linux/netdevice.h>
16#include <linux/etherdevice.h>
17#include <linux/ip.h>
18#include <linux/init.h>
19#include <linux/moduleparam.h>
20#include <linux/netfilter.h>
21#include <linux/rtnetlink.h>
22#include <net/rtnetlink.h>
23#include <linux/u64_stats_sync.h>
24#include <linux/hashtable.h>
Andrea Mayerc8baec32020-06-20 00:54:44 +020025#include <linux/spinlock_types.h>
David Ahern193125d2015-08-13 14:59:10 -060026
27#include <linux/inetdevice.h>
David Ahern8f583362015-08-27 10:10:50 -070028#include <net/arp.h>
David Ahern193125d2015-08-13 14:59:10 -060029#include <net/ip.h>
30#include <net/ip_fib.h>
David Ahern35402e32015-10-12 11:47:09 -070031#include <net/ip6_fib.h>
David Ahern193125d2015-08-13 14:59:10 -060032#include <net/ip6_route.h>
David Ahern193125d2015-08-13 14:59:10 -060033#include <net/route.h>
34#include <net/addrconf.h>
David Ahernee15ee52015-09-29 20:07:12 -070035#include <net/l3mdev.h>
David Ahern1aa6c4f2016-06-08 10:55:40 -070036#include <net/fib_rules.h>
Jakub Kicinskib6459412021-12-28 16:49:13 -080037#include <net/sch_generic.h>
David Ahern097d3c92017-06-08 11:31:11 -060038#include <net/netns/generic.h>
Florian Westphal8c9c2962021-10-25 16:14:00 +020039#include <net/netfilter/nf_conntrack.h>
David Ahern193125d2015-08-13 14:59:10 -060040
41#define DRV_NAME "vrf"
Andrea Mayerc8baec32020-06-20 00:54:44 +020042#define DRV_VERSION "1.1"
David Ahern193125d2015-08-13 14:59:10 -060043
David Ahern1aa6c4f2016-06-08 10:55:40 -070044#define FIB_RULE_PREF 1000 /* default preference for FIB rules */
David Ahern097d3c92017-06-08 11:31:11 -060045
Andrea Mayerc8baec32020-06-20 00:54:44 +020046#define HT_MAP_BITS 4
47#define HASH_INITVAL ((u32)0xcafef00d)
48
49struct vrf_map {
50 DECLARE_HASHTABLE(ht, HT_MAP_BITS);
51 spinlock_t vmap_lock;
52
53 /* shared_tables:
54 * count how many distinct tables do not comply with the strict mode
55 * requirement.
56 * shared_tables value must be 0 in order to enable the strict mode.
57 *
58 * example of the evolution of shared_tables:
59 * | time
60 * add vrf0 --> table 100 shared_tables = 0 | t0
61 * add vrf1 --> table 101 shared_tables = 0 | t1
62 * add vrf2 --> table 100 shared_tables = 1 | t2
63 * add vrf3 --> table 100 shared_tables = 1 | t3
64 * add vrf4 --> table 101 shared_tables = 2 v t4
65 *
66 * shared_tables is a "step function" (or "staircase function")
67 * and it is increased by one when the second vrf is associated to a
68 * table.
69 *
70 * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1.
71 *
72 * at t3, another dev (vrf3) is bound to the same table 100 but the
73 * value of shared_tables is still 1.
74 * This means that no matter how many new vrfs will register on the
75 * table 100, the shared_tables will not increase (considering only
76 * table 100).
77 *
78 * at t4, vrf4 is bound to table 101, and shared_tables = 2.
79 *
80 * Looking at the value of shared_tables we can immediately know if
81 * the strict_mode can or cannot be enforced. Indeed, strict_mode
82 * can be enforced iff shared_tables = 0.
83 *
84 * Conversely, shared_tables is decreased when a vrf is de-associated
85 * from a table with exactly two associated vrfs.
86 */
87 u32 shared_tables;
88
89 bool strict_mode;
90};
91
92struct vrf_map_elem {
93 struct hlist_node hnode;
94 struct list_head vrf_list; /* VRFs registered to this table */
95
96 u32 table_id;
97 int users;
98 int ifindex;
99};
100
David Ahern097d3c92017-06-08 11:31:11 -0600101static unsigned int vrf_net_id;
David Ahern1aa6c4f2016-06-08 10:55:40 -0700102
Andrea Mayerc8baec32020-06-20 00:54:44 +0200103/* per netns vrf data */
104struct netns_vrf {
105 /* protected by rtnl lock */
106 bool add_fib_rules;
107
108 struct vrf_map vmap;
Andrea Mayer33306f12020-06-20 00:54:45 +0200109 struct ctl_table_header *ctl_hdr;
Andrea Mayerc8baec32020-06-20 00:54:44 +0200110};
111
David Ahernec539512015-09-29 20:07:17 -0700112struct net_vrf {
David Ahernb0e95cc2016-05-13 12:23:45 -0700113 struct rtable __rcu *rth;
114 struct rt6_info __rcu *rt6;
David Ahern43b059a2018-04-17 17:33:09 -0700115#if IS_ENABLED(CONFIG_IPV6)
116 struct fib6_table *fib6_table;
117#endif
David Ahernec539512015-09-29 20:07:17 -0700118 u32 tb_id;
Andrea Mayerc8baec32020-06-20 00:54:44 +0200119
120 struct list_head me_list; /* entry in vrf_map_elem */
121 int ifindex;
David Ahernec539512015-09-29 20:07:17 -0700122};
123
David Ahern193125d2015-08-13 14:59:10 -0600124struct pcpu_dstats {
125 u64 tx_pkts;
126 u64 tx_bytes;
127 u64 tx_drps;
128 u64 rx_pkts;
129 u64 rx_bytes;
David Ahernafe80a42016-06-06 20:50:39 -0700130 u64 rx_drps;
David Ahern193125d2015-08-13 14:59:10 -0600131 struct u64_stats_sync syncp;
132};
133
David Ahernafe80a42016-06-06 20:50:39 -0700134static void vrf_rx_stats(struct net_device *dev, int len)
135{
136 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
137
138 u64_stats_update_begin(&dstats->syncp);
139 dstats->rx_pkts++;
140 dstats->rx_bytes += len;
141 u64_stats_update_end(&dstats->syncp);
142}
143
Nikolay Aleksandrov57b8efa2015-08-19 06:12:29 +0300144static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
145{
146 vrf_dev->stats.tx_errors++;
147 kfree_skb(skb);
148}
149
stephen hemmingerbc1f4472017-01-06 19:12:52 -0800150static void vrf_get_stats64(struct net_device *dev,
151 struct rtnl_link_stats64 *stats)
David Ahern193125d2015-08-13 14:59:10 -0600152{
153 int i;
154
155 for_each_possible_cpu(i) {
156 const struct pcpu_dstats *dstats;
157 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
158 unsigned int start;
159
160 dstats = per_cpu_ptr(dev->dstats, i);
161 do {
162 start = u64_stats_fetch_begin_irq(&dstats->syncp);
163 tbytes = dstats->tx_bytes;
164 tpkts = dstats->tx_pkts;
165 tdrops = dstats->tx_drps;
166 rbytes = dstats->rx_bytes;
167 rpkts = dstats->rx_pkts;
168 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
169 stats->tx_bytes += tbytes;
170 stats->tx_packets += tpkts;
171 stats->tx_dropped += tdrops;
172 stats->rx_bytes += rbytes;
173 stats->rx_packets += rpkts;
174 }
David Ahern193125d2015-08-13 14:59:10 -0600175}
176
Andrea Mayerc8baec32020-06-20 00:54:44 +0200177static struct vrf_map *netns_vrf_map(struct net *net)
178{
179 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
180
181 return &nn_vrf->vmap;
182}
183
184static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev)
185{
186 return netns_vrf_map(dev_net(dev));
187}
188
Andrea Mayera59a8ff2020-06-20 00:54:46 +0200189static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me)
190{
191 struct list_head *me_head = &me->vrf_list;
192 struct net_vrf *vrf;
193
194 if (list_empty(me_head))
195 return -ENODEV;
196
197 vrf = list_first_entry(me_head, struct net_vrf, me_list);
198
199 return vrf->ifindex;
200}
201
Andrea Mayerc8baec32020-06-20 00:54:44 +0200202static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags)
203{
204 struct vrf_map_elem *me;
205
206 me = kmalloc(sizeof(*me), flags);
207 if (!me)
208 return NULL;
209
210 return me;
211}
212
213static void vrf_map_elem_free(struct vrf_map_elem *me)
214{
215 kfree(me);
216}
217
218static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id,
219 int ifindex, int users)
220{
221 me->table_id = table_id;
222 me->ifindex = ifindex;
223 me->users = users;
224 INIT_LIST_HEAD(&me->vrf_list);
225}
226
227static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap,
228 u32 table_id)
229{
230 struct vrf_map_elem *me;
231 u32 key;
232
233 key = jhash_1word(table_id, HASH_INITVAL);
234 hash_for_each_possible(vmap->ht, me, hnode, key) {
235 if (me->table_id == table_id)
236 return me;
237 }
238
239 return NULL;
240}
241
242static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me)
243{
244 u32 table_id = me->table_id;
245 u32 key;
246
247 key = jhash_1word(table_id, HASH_INITVAL);
248 hash_add(vmap->ht, &me->hnode, key);
249}
250
251static void vrf_map_del_elem(struct vrf_map_elem *me)
252{
253 hash_del(&me->hnode);
254}
255
256static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock)
257{
258 spin_lock(&vmap->vmap_lock);
259}
260
261static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock)
262{
263 spin_unlock(&vmap->vmap_lock);
264}
265
266/* called with rtnl lock held */
267static int
268vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack)
269{
270 struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
271 struct net_vrf *vrf = netdev_priv(dev);
272 struct vrf_map_elem *new_me, *me;
273 u32 table_id = vrf->tb_id;
274 bool free_new_me = false;
275 int users;
276 int res;
277
278 /* we pre-allocate elements used in the spin-locked section (so that we
Zheng Yongjune9a0bf62021-06-01 22:16:35 +0800279 * keep the spinlock as short as possible).
Andrea Mayerc8baec32020-06-20 00:54:44 +0200280 */
281 new_me = vrf_map_elem_alloc(GFP_KERNEL);
282 if (!new_me)
283 return -ENOMEM;
284
285 vrf_map_elem_init(new_me, table_id, dev->ifindex, 0);
286
287 vrf_map_lock(vmap);
288
289 me = vrf_map_lookup_elem(vmap, table_id);
290 if (!me) {
291 me = new_me;
292 vrf_map_add_elem(vmap, me);
293 goto link_vrf;
294 }
295
296 /* we already have an entry in the vrf_map, so it means there is (at
297 * least) a vrf registered on the specific table.
298 */
299 free_new_me = true;
300 if (vmap->strict_mode) {
301 /* vrfs cannot share the same table */
302 NL_SET_ERR_MSG(extack, "Table is used by another VRF");
303 res = -EBUSY;
304 goto unlock;
305 }
306
307link_vrf:
308 users = ++me->users;
309 if (users == 2)
310 ++vmap->shared_tables;
311
312 list_add(&vrf->me_list, &me->vrf_list);
313
314 res = 0;
315
316unlock:
317 vrf_map_unlock(vmap);
318
319 /* clean-up, if needed */
320 if (free_new_me)
321 vrf_map_elem_free(new_me);
322
323 return res;
324}
325
326/* called with rtnl lock held */
327static void vrf_map_unregister_dev(struct net_device *dev)
328{
329 struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
330 struct net_vrf *vrf = netdev_priv(dev);
331 u32 table_id = vrf->tb_id;
332 struct vrf_map_elem *me;
333 int users;
334
335 vrf_map_lock(vmap);
336
337 me = vrf_map_lookup_elem(vmap, table_id);
338 if (!me)
339 goto unlock;
340
341 list_del(&vrf->me_list);
342
343 users = --me->users;
344 if (users == 1) {
345 --vmap->shared_tables;
346 } else if (users == 0) {
347 vrf_map_del_elem(me);
348
349 /* no one will refer to this element anymore */
350 vrf_map_elem_free(me);
351 }
352
353unlock:
354 vrf_map_unlock(vmap);
355}
356
Andrea Mayera59a8ff2020-06-20 00:54:46 +0200357/* return the vrf device index associated with the table_id */
358static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id)
359{
360 struct vrf_map *vmap = netns_vrf_map(net);
361 struct vrf_map_elem *me;
362 int ifindex;
363
364 vrf_map_lock(vmap);
365
366 if (!vmap->strict_mode) {
367 ifindex = -EPERM;
368 goto unlock;
369 }
370
371 me = vrf_map_lookup_elem(vmap, table_id);
372 if (!me) {
373 ifindex = -ENODEV;
374 goto unlock;
375 }
376
377 ifindex = vrf_map_elem_get_vrf_ifindex(me);
378
379unlock:
380 vrf_map_unlock(vmap);
381
382 return ifindex;
383}
384
David Aherndcdd43c2017-03-20 11:19:44 -0700385/* by default VRF devices do not have a qdisc and are expected
386 * to be created with only a single queue.
387 */
388static bool qdisc_tx_is_default(const struct net_device *dev)
389{
390 struct netdev_queue *txq;
391 struct Qdisc *qdisc;
392
393 if (dev->num_tx_queues > 1)
394 return false;
395
396 txq = netdev_get_tx_queue(dev, 0);
397 qdisc = rcu_access_pointer(txq->qdisc);
398
399 return !qdisc->enqueue;
400}
401
David Ahernafe80a42016-06-06 20:50:39 -0700402/* Local traffic destined to local address. Reinsert the packet to rx
403 * path, similar to loopback handling.
404 */
405static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
406 struct dst_entry *dst)
407{
408 int len = skb->len;
409
410 skb_orphan(skb);
411
412 skb_dst_set(skb, dst);
David Ahernafe80a42016-06-06 20:50:39 -0700413
414 /* set pkt_type to avoid skb hitting packet taps twice -
415 * once on Tx and again in Rx processing
416 */
417 skb->pkt_type = PACKET_LOOPBACK;
418
419 skb->protocol = eth_type_trans(skb, dev);
420
Sebastian Andrzej Siewiorbaebdf482022-02-12 00:38:38 +0100421 if (likely(__netif_rx(skb) == NET_RX_SUCCESS))
David Ahernafe80a42016-06-06 20:50:39 -0700422 vrf_rx_stats(dev, len);
423 else
424 this_cpu_inc(dev->dstats->rx_drps);
425
426 return NETDEV_TX_OK;
427}
428
Florian Westphal8c9c2962021-10-25 16:14:00 +0200429static void vrf_nf_set_untracked(struct sk_buff *skb)
430{
431 if (skb_get_nfct(skb) == 0)
432 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
433}
434
435static void vrf_nf_reset_ct(struct sk_buff *skb)
436{
437 if (skb_get_nfct(skb) == IP_CT_UNTRACKED)
438 nf_reset_ct(skb);
439}
440
David Ahern35402e32015-10-12 11:47:09 -0700441#if IS_ENABLED(CONFIG_IPV6)
David Ahern4c1feac2016-09-10 12:09:56 -0700442static int vrf_ip6_local_out(struct net *net, struct sock *sk,
443 struct sk_buff *skb)
444{
445 int err;
446
Florian Westphal8c9c2962021-10-25 16:14:00 +0200447 vrf_nf_reset_ct(skb);
448
David Ahern4c1feac2016-09-10 12:09:56 -0700449 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
450 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
451
452 if (likely(err == 1))
453 err = dst_output(net, sk, skb);
454
455 return err;
456}
457
David Ahern35402e32015-10-12 11:47:09 -0700458static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
459 struct net_device *dev)
460{
Peter Kosyh107e47c2019-07-19 11:11:47 +0300461 const struct ipv6hdr *iph;
David Ahern35402e32015-10-12 11:47:09 -0700462 struct net *net = dev_net(skb->dev);
Peter Kosyh107e47c2019-07-19 11:11:47 +0300463 struct flowi6 fl6;
David Ahern35402e32015-10-12 11:47:09 -0700464 int ret = NET_XMIT_DROP;
465 struct dst_entry *dst;
466 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
467
Peter Kosyh107e47c2019-07-19 11:11:47 +0300468 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
469 goto err;
470
471 iph = ipv6_hdr(skb);
472
473 memset(&fl6, 0, sizeof(fl6));
474 /* needed to match OIF rule */
David Ahern40867d72022-03-14 14:45:51 -0600475 fl6.flowi6_l3mdev = dev->ifindex;
Peter Kosyh107e47c2019-07-19 11:11:47 +0300476 fl6.flowi6_iif = LOOPBACK_IFINDEX;
477 fl6.daddr = iph->daddr;
478 fl6.saddr = iph->saddr;
479 fl6.flowlabel = ip6_flowinfo(iph);
480 fl6.flowi6_mark = skb->mark;
481 fl6.flowi6_proto = iph->nexthdr;
Peter Kosyh107e47c2019-07-19 11:11:47 +0300482
David Aherna53c1022020-04-21 17:48:27 -0600483 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL);
484 if (IS_ERR(dst) || dst == dst_null)
David Ahern35402e32015-10-12 11:47:09 -0700485 goto err;
486
487 skb_dst_drop(skb);
David Ahernb4869aa2016-06-06 20:50:40 -0700488
Nicolas Dichtel2e1534f2021-04-14 12:03:25 +0200489 /* if dst.dev is the VRF device again this is locally originated traffic
490 * destined to a local address. Short circuit to Rx path.
David Ahernb4869aa2016-06-06 20:50:40 -0700491 */
David Ahern4f042562017-08-11 17:11:14 -0700492 if (dst->dev == dev)
493 return vrf_local_xmit(skb, dev, dst);
David Ahernb4869aa2016-06-06 20:50:40 -0700494
David Ahern35402e32015-10-12 11:47:09 -0700495 skb_dst_set(skb, dst);
496
David Ahern911a66f2016-06-06 20:50:38 -0700497 /* strip the ethernet header added for pass through VRF device */
498 __skb_pull(skb, skb_network_offset(skb));
499
Stephen Suryaputraee201012021-11-30 11:26:37 -0500500 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
David Ahern4c1feac2016-09-10 12:09:56 -0700501 ret = vrf_ip6_local_out(net, skb->sk, skb);
David Ahern35402e32015-10-12 11:47:09 -0700502 if (unlikely(net_xmit_eval(ret)))
503 dev->stats.tx_errors++;
504 else
505 ret = NET_XMIT_SUCCESS;
506
507 return ret;
508err:
509 vrf_tx_error(dev, skb);
510 return NET_XMIT_DROP;
511}
512#else
David Ahern193125d2015-08-13 14:59:10 -0600513static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
514 struct net_device *dev)
515{
Nikolay Aleksandrov57b8efa2015-08-19 06:12:29 +0300516 vrf_tx_error(dev, skb);
517 return NET_XMIT_DROP;
David Ahern193125d2015-08-13 14:59:10 -0600518}
David Ahern35402e32015-10-12 11:47:09 -0700519#endif
David Ahern193125d2015-08-13 14:59:10 -0600520
David Ahernebfc1022016-09-10 12:09:55 -0700521/* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
522static int vrf_ip_local_out(struct net *net, struct sock *sk,
523 struct sk_buff *skb)
524{
525 int err;
526
Florian Westphal8c9c2962021-10-25 16:14:00 +0200527 vrf_nf_reset_ct(skb);
528
David Ahernebfc1022016-09-10 12:09:55 -0700529 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
530 skb, NULL, skb_dst(skb)->dev, dst_output);
531 if (likely(err == 1))
532 err = dst_output(net, sk, skb);
533
534 return err;
535}
536
David Ahern193125d2015-08-13 14:59:10 -0600537static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
538 struct net_device *vrf_dev)
539{
Peter Kosyh107e47c2019-07-19 11:11:47 +0300540 struct iphdr *ip4h;
David Ahern193125d2015-08-13 14:59:10 -0600541 int ret = NET_XMIT_DROP;
Peter Kosyh107e47c2019-07-19 11:11:47 +0300542 struct flowi4 fl4;
David Ahern911a66f2016-06-06 20:50:38 -0700543 struct net *net = dev_net(vrf_dev);
544 struct rtable *rt;
David Ahern193125d2015-08-13 14:59:10 -0600545
Peter Kosyh107e47c2019-07-19 11:11:47 +0300546 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
547 goto err;
548
549 ip4h = ip_hdr(skb);
550
551 memset(&fl4, 0, sizeof(fl4));
552 /* needed to match OIF rule */
David Ahern40867d72022-03-14 14:45:51 -0600553 fl4.flowi4_l3mdev = vrf_dev->ifindex;
Peter Kosyh107e47c2019-07-19 11:11:47 +0300554 fl4.flowi4_iif = LOOPBACK_IFINDEX;
555 fl4.flowi4_tos = RT_TOS(ip4h->tos);
David Ahern40867d72022-03-14 14:45:51 -0600556 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC;
Peter Kosyh107e47c2019-07-19 11:11:47 +0300557 fl4.flowi4_proto = ip4h->protocol;
558 fl4.daddr = ip4h->daddr;
559 fl4.saddr = ip4h->saddr;
560
David Ahern911a66f2016-06-06 20:50:38 -0700561 rt = ip_route_output_flow(net, &fl4, NULL);
562 if (IS_ERR(rt))
David Ahern193125d2015-08-13 14:59:10 -0600563 goto err;
564
David Ahern911a66f2016-06-06 20:50:38 -0700565 skb_dst_drop(skb);
David Ahernafe80a42016-06-06 20:50:39 -0700566
Nicolas Dichtel2e1534f2021-04-14 12:03:25 +0200567 /* if dst.dev is the VRF device again this is locally originated traffic
568 * destined to a local address. Short circuit to Rx path.
David Ahernafe80a42016-06-06 20:50:39 -0700569 */
David Ahern4f042562017-08-11 17:11:14 -0700570 if (rt->dst.dev == vrf_dev)
571 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
David Ahernafe80a42016-06-06 20:50:39 -0700572
David Ahern911a66f2016-06-06 20:50:38 -0700573 skb_dst_set(skb, &rt->dst);
574
575 /* strip the ethernet header added for pass through VRF device */
576 __skb_pull(skb, skb_network_offset(skb));
577
David Ahern193125d2015-08-13 14:59:10 -0600578 if (!ip4h->saddr) {
579 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
580 RT_SCOPE_LINK);
581 }
582
Stephen Suryaputraee201012021-11-30 11:26:37 -0500583 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
David Ahernebfc1022016-09-10 12:09:55 -0700584 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
David Ahern193125d2015-08-13 14:59:10 -0600585 if (unlikely(net_xmit_eval(ret)))
586 vrf_dev->stats.tx_errors++;
587 else
588 ret = NET_XMIT_SUCCESS;
589
590out:
591 return ret;
592err:
Nikolay Aleksandrov57b8efa2015-08-19 06:12:29 +0300593 vrf_tx_error(vrf_dev, skb);
David Ahern193125d2015-08-13 14:59:10 -0600594 goto out;
595}
596
597static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
598{
599 switch (skb->protocol) {
600 case htons(ETH_P_IP):
601 return vrf_process_v4_outbound(skb, dev);
602 case htons(ETH_P_IPV6):
603 return vrf_process_v6_outbound(skb, dev);
604 default:
Nikolay Aleksandrov57b8efa2015-08-19 06:12:29 +0300605 vrf_tx_error(dev, skb);
David Ahern193125d2015-08-13 14:59:10 -0600606 return NET_XMIT_DROP;
607 }
608}
609
610static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
611{
David Ahernf7887d42017-03-06 08:53:04 -0800612 int len = skb->len;
David Ahern193125d2015-08-13 14:59:10 -0600613 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
614
615 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
616 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
617
618 u64_stats_update_begin(&dstats->syncp);
619 dstats->tx_pkts++;
David Ahernf7887d42017-03-06 08:53:04 -0800620 dstats->tx_bytes += len;
David Ahern193125d2015-08-13 14:59:10 -0600621 u64_stats_update_end(&dstats->syncp);
622 } else {
623 this_cpu_inc(dev->dstats->tx_drps);
624 }
625
626 return ret;
627}
628
Martin Willi9e2b7fa2020-11-06 08:30:30 +0100629static void vrf_finish_direct(struct sk_buff *skb)
David Aherndcdd43c2017-03-20 11:19:44 -0700630{
631 struct net_device *vrf_dev = skb->dev;
632
633 if (!list_empty(&vrf_dev->ptype_all) &&
634 likely(skb_headroom(skb) >= ETH_HLEN)) {
Johannes Bergd58ff352017-06-16 14:29:23 +0200635 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
David Aherndcdd43c2017-03-20 11:19:44 -0700636
637 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
638 eth_zero_addr(eth->h_dest);
639 eth->h_proto = skb->protocol;
640
641 rcu_read_lock_bh();
642 dev_queue_xmit_nit(skb, vrf_dev);
643 rcu_read_unlock_bh();
644
645 skb_pull(skb, ETH_HLEN);
646 }
647
Florian Westphal8c9c2962021-10-25 16:14:00 +0200648 vrf_nf_reset_ct(skb);
David Aherndcdd43c2017-03-20 11:19:44 -0700649}
650
David Ahern35402e32015-10-12 11:47:09 -0700651#if IS_ENABLED(CONFIG_IPV6)
David Ahern35402e32015-10-12 11:47:09 -0700652/* modelled after ip6_finish_output2 */
653static int vrf_finish_output6(struct net *net, struct sock *sk,
654 struct sk_buff *skb)
655{
656 struct dst_entry *dst = skb_dst(skb);
657 struct net_device *dev = dst->dev;
Nicolas Dichtel9b1c1ef2019-06-24 16:01:08 +0200658 const struct in6_addr *nexthop;
David Ahern35402e32015-10-12 11:47:09 -0700659 struct neighbour *neigh;
David Ahern35402e32015-10-12 11:47:09 -0700660 int ret;
661
Florian Westphal8c9c2962021-10-25 16:14:00 +0200662 vrf_nf_reset_ct(skb);
David Aherneb63ecc2016-12-14 14:31:11 -0800663
David Ahern35402e32015-10-12 11:47:09 -0700664 skb->protocol = htons(ETH_P_IPV6);
665 skb->dev = dev;
666
667 rcu_read_lock_bh();
668 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
669 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
670 if (unlikely(!neigh))
671 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
672 if (!IS_ERR(neigh)) {
Julian Anastasov4ff06202017-02-06 23:14:12 +0200673 sock_confirm_neigh(skb, neigh);
David Ahern0353f282019-04-05 16:30:33 -0700674 ret = neigh_output(neigh, skb, false);
David Ahern35402e32015-10-12 11:47:09 -0700675 rcu_read_unlock_bh();
676 return ret;
677 }
678 rcu_read_unlock_bh();
679
680 IP6_INC_STATS(dev_net(dst->dev),
681 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
682 kfree_skb(skb);
683 return -EINVAL;
684}
685
686/* modelled after ip6_output */
687static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
688{
689 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
690 net, sk, skb, NULL, skb_dst(skb)->dev,
691 vrf_finish_output6,
692 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
693}
694
David Ahern4c1feac2016-09-10 12:09:56 -0700695/* set dst on skb to send packet to us via dev_xmit path. Allows
696 * packet to go through device based features such as qdisc, netfilter
697 * hooks and packet sockets with skb->dev set to vrf device.
698 */
David Aherna9ec54d2017-03-20 11:19:45 -0700699static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
700 struct sk_buff *skb)
David Ahern4c1feac2016-09-10 12:09:56 -0700701{
702 struct net_vrf *vrf = netdev_priv(vrf_dev);
703 struct dst_entry *dst = NULL;
704 struct rt6_info *rt6;
705
David Ahern4c1feac2016-09-10 12:09:56 -0700706 rcu_read_lock();
707
708 rt6 = rcu_dereference(vrf->rt6);
709 if (likely(rt6)) {
710 dst = &rt6->dst;
711 dst_hold(dst);
712 }
713
714 rcu_read_unlock();
715
716 if (unlikely(!dst)) {
717 vrf_tx_error(vrf_dev, skb);
718 return NULL;
719 }
720
721 skb_dst_drop(skb);
722 skb_dst_set(skb, dst);
723
724 return skb;
725}
726
Martin Willi9e2b7fa2020-11-06 08:30:30 +0100727static int vrf_output6_direct_finish(struct net *net, struct sock *sk,
728 struct sk_buff *skb)
729{
730 vrf_finish_direct(skb);
731
732 return vrf_ip6_local_out(net, sk, skb);
733}
734
David Aherna9ec54d2017-03-20 11:19:45 -0700735static int vrf_output6_direct(struct net *net, struct sock *sk,
736 struct sk_buff *skb)
737{
Martin Willi9e2b7fa2020-11-06 08:30:30 +0100738 int err = 1;
739
David Aherna9ec54d2017-03-20 11:19:45 -0700740 skb->protocol = htons(ETH_P_IPV6);
741
Martin Willi9e2b7fa2020-11-06 08:30:30 +0100742 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
743 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
744 NULL, skb->dev, vrf_output6_direct_finish);
745
746 if (likely(err == 1))
747 vrf_finish_direct(skb);
748
749 return err;
750}
751
752static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk,
753 struct sk_buff *skb)
754{
755 int err;
756
757 err = vrf_output6_direct(net, sk, skb);
758 if (likely(err == 1))
759 err = vrf_ip6_local_out(net, sk, skb);
760
761 return err;
David Aherna9ec54d2017-03-20 11:19:45 -0700762}
763
764static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
765 struct sock *sk,
766 struct sk_buff *skb)
767{
768 struct net *net = dev_net(vrf_dev);
769 int err;
770
771 skb->dev = vrf_dev;
772
773 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
Martin Willi9e2b7fa2020-11-06 08:30:30 +0100774 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish);
David Aherna9ec54d2017-03-20 11:19:45 -0700775
776 if (likely(err == 1))
777 err = vrf_output6_direct(net, sk, skb);
778
David Aherna9ec54d2017-03-20 11:19:45 -0700779 if (likely(err == 1))
Martin Willi9e2b7fa2020-11-06 08:30:30 +0100780 return skb;
David Aherna9ec54d2017-03-20 11:19:45 -0700781
Martin Willi9e2b7fa2020-11-06 08:30:30 +0100782 return NULL;
David Aherna9ec54d2017-03-20 11:19:45 -0700783}
784
785static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
786 struct sock *sk,
787 struct sk_buff *skb)
788{
789 /* don't divert link scope packets */
790 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
791 return skb;
792
Nicolas Dichteld43b75f2021-11-26 15:36:12 +0100793 vrf_nf_set_untracked(skb);
794
David Ahern16b9db12020-04-20 17:13:52 -0600795 if (qdisc_tx_is_default(vrf_dev) ||
796 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
David Aherna9ec54d2017-03-20 11:19:45 -0700797 return vrf_ip6_out_direct(vrf_dev, sk, skb);
798
799 return vrf_ip6_out_redirect(vrf_dev, skb);
800}
801
David Ahernb0e95cc2016-05-13 12:23:45 -0700802/* holding rtnl */
David Ahern810e5302016-06-14 11:37:21 -0700803static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
David Ahern35402e32015-10-12 11:47:09 -0700804{
David Ahernb0e95cc2016-05-13 12:23:45 -0700805 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
David Ahern810e5302016-06-14 11:37:21 -0700806 struct net *net = dev_net(dev);
807 struct dst_entry *dst;
David Ahernb0e95cc2016-05-13 12:23:45 -0700808
David Ahernb4869aa2016-06-06 20:50:40 -0700809 RCU_INIT_POINTER(vrf->rt6, NULL);
David Ahernb4869aa2016-06-06 20:50:40 -0700810 synchronize_rcu();
David Ahernb0e95cc2016-05-13 12:23:45 -0700811
David Ahern810e5302016-06-14 11:37:21 -0700812 /* move dev in dst's to loopback so this VRF device can be deleted
813 * - based on dst_ifdown
814 */
815 if (rt6) {
816 dst = &rt6->dst;
Eric Dumazetc0e5e112021-12-06 21:56:03 -0800817 dev_replace_track(dst->dev, net->loopback_dev,
818 &dst->dev_tracker, GFP_KERNEL);
David Ahern810e5302016-06-14 11:37:21 -0700819 dst->dev = net->loopback_dev;
David Ahern810e5302016-06-14 11:37:21 -0700820 dst_release(dst);
821 }
David Ahern35402e32015-10-12 11:47:09 -0700822}
823
824static int vrf_rt6_create(struct net_device *dev)
825{
David Laightaf13b3c2020-03-23 14:31:19 +0000826 int flags = DST_NOPOLICY | DST_NOXFRM;
David Ahern35402e32015-10-12 11:47:09 -0700827 struct net_vrf *vrf = netdev_priv(dev);
David Ahern9ab179d2016-04-07 11:10:06 -0700828 struct net *net = dev_net(dev);
David Ahern4f042562017-08-11 17:11:14 -0700829 struct rt6_info *rt6;
David Ahern35402e32015-10-12 11:47:09 -0700830 int rc = -ENOMEM;
831
David Aherne4348632016-06-09 10:21:00 -0700832 /* IPv6 can be CONFIG enabled and then disabled runtime */
833 if (!ipv6_mod_enabled())
834 return 0;
835
David Ahern43b059a2018-04-17 17:33:09 -0700836 vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
837 if (!vrf->fib6_table)
David Ahernb3b46632016-05-04 21:46:12 -0700838 goto out;
839
David Ahernb4869aa2016-06-06 20:50:40 -0700840 /* create a dst for routing packets out a VRF device */
841 rt6 = ip6_dst_alloc(net, dev, flags);
David Ahern35402e32015-10-12 11:47:09 -0700842 if (!rt6)
843 goto out;
844
David Ahernb3b46632016-05-04 21:46:12 -0700845 rt6->dst.output = vrf_output6;
David Ahernb4869aa2016-06-06 20:50:40 -0700846
David Ahernb0e95cc2016-05-13 12:23:45 -0700847 rcu_assign_pointer(vrf->rt6, rt6);
848
David Ahern35402e32015-10-12 11:47:09 -0700849 rc = 0;
850out:
851 return rc;
852}
853#else
David Ahern4c1feac2016-09-10 12:09:56 -0700854static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
855 struct sock *sk,
856 struct sk_buff *skb)
857{
858 return skb;
859}
860
David Ahern810e5302016-06-14 11:37:21 -0700861static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
David Ahern35402e32015-10-12 11:47:09 -0700862{
863}
864
865static int vrf_rt6_create(struct net_device *dev)
866{
867 return 0;
868}
869#endif
870
David Ahern8f583362015-08-27 10:10:50 -0700871/* modelled after ip_finish_output2 */
Eric W. Biederman0c4b51f2015-09-15 20:04:18 -0500872static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
David Ahern193125d2015-08-13 14:59:10 -0600873{
David Ahern8f583362015-08-27 10:10:50 -0700874 struct dst_entry *dst = skb_dst(skb);
875 struct rtable *rt = (struct rtable *)dst;
876 struct net_device *dev = dst->dev;
877 unsigned int hh_len = LL_RESERVED_SPACE(dev);
878 struct neighbour *neigh;
David Ahern5c9f7c12019-04-05 16:30:34 -0700879 bool is_v6gw = false;
David Ahern8f583362015-08-27 10:10:50 -0700880
Florian Westphal8c9c2962021-10-25 16:14:00 +0200881 vrf_nf_reset_ct(skb);
David Aherneb63ecc2016-12-14 14:31:11 -0800882
David Ahern8f583362015-08-27 10:10:50 -0700883 /* Be paranoid, rather than too clever. */
884 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
Vasily Averin14ee70c2021-08-02 11:52:40 +0300885 skb = skb_expand_head(skb, hh_len);
886 if (!skb) {
Dan Carpenter06669e62021-08-06 18:04:35 +0300887 dev->stats.tx_errors++;
Vasily Averin14ee70c2021-08-02 11:52:40 +0300888 return -ENOMEM;
David Ahern8f583362015-08-27 10:10:50 -0700889 }
David Ahern8f583362015-08-27 10:10:50 -0700890 }
891
892 rcu_read_lock_bh();
893
David Ahern5c9f7c12019-04-05 16:30:34 -0700894 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
Julian Anastasov4ff06202017-02-06 23:14:12 +0200895 if (!IS_ERR(neigh)) {
Vasily Averin14ee70c2021-08-02 11:52:40 +0300896 int ret;
897
Julian Anastasov4ff06202017-02-06 23:14:12 +0200898 sock_confirm_neigh(skb, neigh);
David Ahern5c9f7c12019-04-05 16:30:34 -0700899 /* if crossing protocols, can not use the cached header */
900 ret = neigh_output(neigh, skb, is_v6gw);
David Ahern82dd0d22018-03-29 12:49:52 -0700901 rcu_read_unlock_bh();
902 return ret;
Julian Anastasov4ff06202017-02-06 23:14:12 +0200903 }
David Ahern8f583362015-08-27 10:10:50 -0700904
905 rcu_read_unlock_bh();
David Ahern82dd0d22018-03-29 12:49:52 -0700906 vrf_tx_error(skb->dev, skb);
Vasily Averin14ee70c2021-08-02 11:52:40 +0300907 return -EINVAL;
David Ahern193125d2015-08-13 14:59:10 -0600908}
909
Eric W. Biedermanede20592015-10-07 16:48:47 -0500910static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
David Ahern193125d2015-08-13 14:59:10 -0600911{
912 struct net_device *dev = skb_dst(skb)->dev;
913
Eric W. Biederman29a26a52015-09-15 20:04:16 -0500914 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
David Ahern193125d2015-08-13 14:59:10 -0600915
916 skb->dev = dev;
917 skb->protocol = htons(ETH_P_IP);
918
Eric W. Biederman29a26a52015-09-15 20:04:16 -0500919 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
920 net, sk, skb, NULL, dev,
David Ahern8f583362015-08-27 10:10:50 -0700921 vrf_finish_output,
David Ahern193125d2015-08-13 14:59:10 -0600922 !(IPCB(skb)->flags & IPSKB_REROUTED));
923}
924
David Ahernebfc1022016-09-10 12:09:55 -0700925/* set dst on skb to send packet to us via dev_xmit path. Allows
926 * packet to go through device based features such as qdisc, netfilter
927 * hooks and packet sockets with skb->dev set to vrf device.
928 */
David Aherndcdd43c2017-03-20 11:19:44 -0700929static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
930 struct sk_buff *skb)
David Ahernebfc1022016-09-10 12:09:55 -0700931{
932 struct net_vrf *vrf = netdev_priv(vrf_dev);
933 struct dst_entry *dst = NULL;
934 struct rtable *rth;
935
936 rcu_read_lock();
937
938 rth = rcu_dereference(vrf->rth);
939 if (likely(rth)) {
940 dst = &rth->dst;
941 dst_hold(dst);
942 }
943
944 rcu_read_unlock();
945
946 if (unlikely(!dst)) {
947 vrf_tx_error(vrf_dev, skb);
948 return NULL;
949 }
950
951 skb_dst_drop(skb);
952 skb_dst_set(skb, dst);
953
954 return skb;
955}
956
Martin Willi9e2b7fa2020-11-06 08:30:30 +0100957static int vrf_output_direct_finish(struct net *net, struct sock *sk,
958 struct sk_buff *skb)
959{
960 vrf_finish_direct(skb);
961
962 return vrf_ip_local_out(net, sk, skb);
963}
964
David Aherndcdd43c2017-03-20 11:19:44 -0700965static int vrf_output_direct(struct net *net, struct sock *sk,
966 struct sk_buff *skb)
967{
Martin Willi9e2b7fa2020-11-06 08:30:30 +0100968 int err = 1;
969
David Aherndcdd43c2017-03-20 11:19:44 -0700970 skb->protocol = htons(ETH_P_IP);
971
Martin Willi9e2b7fa2020-11-06 08:30:30 +0100972 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
973 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
974 NULL, skb->dev, vrf_output_direct_finish);
975
976 if (likely(err == 1))
977 vrf_finish_direct(skb);
978
979 return err;
980}
981
982static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk,
983 struct sk_buff *skb)
984{
985 int err;
986
987 err = vrf_output_direct(net, sk, skb);
988 if (likely(err == 1))
989 err = vrf_ip_local_out(net, sk, skb);
990
991 return err;
David Aherndcdd43c2017-03-20 11:19:44 -0700992}
993
994static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
995 struct sock *sk,
996 struct sk_buff *skb)
997{
998 struct net *net = dev_net(vrf_dev);
999 int err;
1000
1001 skb->dev = vrf_dev;
1002
1003 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
Martin Willi9e2b7fa2020-11-06 08:30:30 +01001004 skb, NULL, vrf_dev, vrf_ip_out_direct_finish);
David Aherndcdd43c2017-03-20 11:19:44 -07001005
1006 if (likely(err == 1))
1007 err = vrf_output_direct(net, sk, skb);
1008
David Aherndcdd43c2017-03-20 11:19:44 -07001009 if (likely(err == 1))
Martin Willi9e2b7fa2020-11-06 08:30:30 +01001010 return skb;
David Aherndcdd43c2017-03-20 11:19:44 -07001011
Martin Willi9e2b7fa2020-11-06 08:30:30 +01001012 return NULL;
David Aherndcdd43c2017-03-20 11:19:44 -07001013}
1014
1015static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
1016 struct sock *sk,
1017 struct sk_buff *skb)
1018{
David Ahern1e19c4d2018-01-24 19:37:37 -08001019 /* don't divert multicast or local broadcast */
1020 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
1021 ipv4_is_lbcast(ip_hdr(skb)->daddr))
David Aherndcdd43c2017-03-20 11:19:44 -07001022 return skb;
1023
Nicolas Dichteld43b75f2021-11-26 15:36:12 +01001024 vrf_nf_set_untracked(skb);
1025
David Ahern16b9db12020-04-20 17:13:52 -06001026 if (qdisc_tx_is_default(vrf_dev) ||
1027 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
David Aherndcdd43c2017-03-20 11:19:44 -07001028 return vrf_ip_out_direct(vrf_dev, sk, skb);
1029
1030 return vrf_ip_out_redirect(vrf_dev, skb);
1031}
1032
David Ahernebfc1022016-09-10 12:09:55 -07001033/* called with rcu lock held */
1034static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
1035 struct sock *sk,
1036 struct sk_buff *skb,
1037 u16 proto)
1038{
1039 switch (proto) {
1040 case AF_INET:
1041 return vrf_ip_out(vrf_dev, sk, skb);
David Ahern4c1feac2016-09-10 12:09:56 -07001042 case AF_INET6:
1043 return vrf_ip6_out(vrf_dev, sk, skb);
David Ahernebfc1022016-09-10 12:09:55 -07001044 }
1045
1046 return skb;
1047}
1048
David Ahernb0e95cc2016-05-13 12:23:45 -07001049/* holding rtnl */
David Ahern810e5302016-06-14 11:37:21 -07001050static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
David Ahern193125d2015-08-13 14:59:10 -06001051{
David Ahernb0e95cc2016-05-13 12:23:45 -07001052 struct rtable *rth = rtnl_dereference(vrf->rth);
David Ahern810e5302016-06-14 11:37:21 -07001053 struct net *net = dev_net(dev);
1054 struct dst_entry *dst;
David Ahern193125d2015-08-13 14:59:10 -06001055
David Ahernafe80a42016-06-06 20:50:39 -07001056 RCU_INIT_POINTER(vrf->rth, NULL);
David Ahernafe80a42016-06-06 20:50:39 -07001057 synchronize_rcu();
David Ahernb0e95cc2016-05-13 12:23:45 -07001058
David Ahern810e5302016-06-14 11:37:21 -07001059 /* move dev in dst's to loopback so this VRF device can be deleted
1060 * - based on dst_ifdown
1061 */
1062 if (rth) {
1063 dst = &rth->dst;
Eric Dumazetc0e5e112021-12-06 21:56:03 -08001064 dev_replace_track(dst->dev, net->loopback_dev,
1065 &dst->dev_tracker, GFP_KERNEL);
David Ahern810e5302016-06-14 11:37:21 -07001066 dst->dev = net->loopback_dev;
David Ahern810e5302016-06-14 11:37:21 -07001067 dst_release(dst);
1068 }
David Ahern193125d2015-08-13 14:59:10 -06001069}
1070
David Ahernb0e95cc2016-05-13 12:23:45 -07001071static int vrf_rtable_create(struct net_device *dev)
David Ahern193125d2015-08-13 14:59:10 -06001072{
David Ahernb7503e02015-09-02 13:58:35 -07001073 struct net_vrf *vrf = netdev_priv(dev);
David Ahern4f042562017-08-11 17:11:14 -07001074 struct rtable *rth;
David Ahern193125d2015-08-13 14:59:10 -06001075
David Ahernb3b46632016-05-04 21:46:12 -07001076 if (!fib_new_table(dev_net(dev), vrf->tb_id))
David Ahernb0e95cc2016-05-13 12:23:45 -07001077 return -ENOMEM;
David Ahernb3b46632016-05-04 21:46:12 -07001078
David Ahernafe80a42016-06-06 20:50:39 -07001079 /* create a dst for routing packets out through a VRF device */
David Laightaf13b3c2020-03-23 14:31:19 +00001080 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1);
David Ahernb0e95cc2016-05-13 12:23:45 -07001081 if (!rth)
1082 return -ENOMEM;
David Ahern193125d2015-08-13 14:59:10 -06001083
David Ahernb0e95cc2016-05-13 12:23:45 -07001084 rth->dst.output = vrf_output;
David Ahernb0e95cc2016-05-13 12:23:45 -07001085
1086 rcu_assign_pointer(vrf->rth, rth);
1087
1088 return 0;
David Ahern193125d2015-08-13 14:59:10 -06001089}
1090
1091/**************************** device handling ********************/
1092
1093/* cycle interface to flush neighbor cache and move routes across tables */
Petr Machatadc1aea12018-12-06 17:05:38 +00001094static void cycle_netdev(struct net_device *dev,
1095 struct netlink_ext_ack *extack)
David Ahern193125d2015-08-13 14:59:10 -06001096{
1097 unsigned int flags = dev->flags;
1098 int ret;
1099
1100 if (!netif_running(dev))
1101 return;
1102
Petr Machata567c5e12018-12-06 17:05:42 +00001103 ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
David Ahern193125d2015-08-13 14:59:10 -06001104 if (ret >= 0)
Petr Machata567c5e12018-12-06 17:05:42 +00001105 ret = dev_change_flags(dev, flags, extack);
David Ahern193125d2015-08-13 14:59:10 -06001106
1107 if (ret < 0) {
1108 netdev_err(dev,
1109 "Failed to cycle device %s; route tables might be wrong!\n",
1110 dev->name);
1111 }
1112}
1113
David Ahern42ab19e2017-10-04 17:48:47 -07001114static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1115 struct netlink_ext_ack *extack)
David Ahern193125d2015-08-13 14:59:10 -06001116{
Nikolay Aleksandrovbad53162015-11-24 14:29:16 +01001117 int ret;
David Ahern193125d2015-08-13 14:59:10 -06001118
David Ahern26d31ac2017-04-26 07:58:22 -07001119 /* do not allow loopback device to be enslaved to a VRF.
1120 * The vrf device acts as the loopback for the vrf.
1121 */
David Ahernde3baa32017-10-04 17:48:48 -07001122 if (port_dev == dev_net(dev)->loopback_dev) {
1123 NL_SET_ERR_MSG(extack,
1124 "Can not enslave loopback device to a VRF");
David Ahern26d31ac2017-04-26 07:58:22 -07001125 return -EOPNOTSUPP;
David Ahernde3baa32017-10-04 17:48:48 -07001126 }
David Ahern26d31ac2017-04-26 07:58:22 -07001127
Ido Schimmelfdeea7b2017-03-16 09:08:15 +01001128 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
David Ahern42ab19e2017-10-04 17:48:47 -07001129 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
David Ahern193125d2015-08-13 14:59:10 -06001130 if (ret < 0)
Ido Schimmelfdeea7b2017-03-16 09:08:15 +01001131 goto err;
David Ahern193125d2015-08-13 14:59:10 -06001132
Petr Machatadc1aea12018-12-06 17:05:38 +00001133 cycle_netdev(port_dev, extack);
David Ahern193125d2015-08-13 14:59:10 -06001134
1135 return 0;
Ido Schimmelfdeea7b2017-03-16 09:08:15 +01001136
1137err:
1138 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1139 return ret;
David Ahern193125d2015-08-13 14:59:10 -06001140}
1141
David Ahern33eaf2a2017-10-04 17:48:46 -07001142static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1143 struct netlink_ext_ack *extack)
David Ahern193125d2015-08-13 14:59:10 -06001144{
David Ahernde3baa32017-10-04 17:48:48 -07001145 if (netif_is_l3_master(port_dev)) {
1146 NL_SET_ERR_MSG(extack,
1147 "Can not enslave an L3 master device to a VRF");
1148 return -EINVAL;
1149 }
1150
1151 if (netif_is_l3_slave(port_dev))
David Ahern193125d2015-08-13 14:59:10 -06001152 return -EINVAL;
1153
David Ahern42ab19e2017-10-04 17:48:47 -07001154 return do_vrf_add_slave(dev, port_dev, extack);
David Ahern193125d2015-08-13 14:59:10 -06001155}
1156
1157/* inverse of do_vrf_add_slave */
1158static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1159{
David Ahern193125d2015-08-13 14:59:10 -06001160 netdev_upper_dev_unlink(port_dev, dev);
David Ahernfee6d4c2015-10-05 08:51:24 -07001161 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
David Ahern193125d2015-08-13 14:59:10 -06001162
Petr Machatadc1aea12018-12-06 17:05:38 +00001163 cycle_netdev(port_dev, NULL);
David Ahern193125d2015-08-13 14:59:10 -06001164
David Ahern193125d2015-08-13 14:59:10 -06001165 return 0;
1166}
1167
1168static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1169{
David Ahern193125d2015-08-13 14:59:10 -06001170 return do_vrf_del_slave(dev, port_dev);
1171}
1172
1173static void vrf_dev_uninit(struct net_device *dev)
1174{
1175 struct net_vrf *vrf = netdev_priv(dev);
David Ahern193125d2015-08-13 14:59:10 -06001176
David Ahern810e5302016-06-14 11:37:21 -07001177 vrf_rtable_release(dev, vrf);
1178 vrf_rt6_release(dev, vrf);
David Ahern193125d2015-08-13 14:59:10 -06001179
Nikolay Aleksandrov3a4a27d2015-08-18 20:28:03 +03001180 free_percpu(dev->dstats);
David Ahern193125d2015-08-13 14:59:10 -06001181 dev->dstats = NULL;
1182}
1183
1184static int vrf_dev_init(struct net_device *dev)
1185{
1186 struct net_vrf *vrf = netdev_priv(dev);
1187
David Ahern193125d2015-08-13 14:59:10 -06001188 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
1189 if (!dev->dstats)
1190 goto out_nomem;
1191
1192 /* create the default dst which points back to us */
David Ahernb0e95cc2016-05-13 12:23:45 -07001193 if (vrf_rtable_create(dev) != 0)
David Ahern193125d2015-08-13 14:59:10 -06001194 goto out_stats;
1195
David Ahern35402e32015-10-12 11:47:09 -07001196 if (vrf_rt6_create(dev) != 0)
1197 goto out_rth;
1198
David Ahern193125d2015-08-13 14:59:10 -06001199 dev->flags = IFF_MASTER | IFF_NOARP;
1200
David Ahernb87ab6b2016-06-01 21:16:39 -07001201 /* similarly, oper state is irrelevant; set to up to avoid confusion */
1202 dev->operstate = IF_OPER_UP;
Cong Wang1a33e102020-05-02 22:22:19 -07001203 netdev_lockdep_set_classes(dev);
David Ahern193125d2015-08-13 14:59:10 -06001204 return 0;
1205
David Ahern35402e32015-10-12 11:47:09 -07001206out_rth:
David Ahern810e5302016-06-14 11:37:21 -07001207 vrf_rtable_release(dev, vrf);
David Ahern193125d2015-08-13 14:59:10 -06001208out_stats:
1209 free_percpu(dev->dstats);
1210 dev->dstats = NULL;
1211out_nomem:
1212 return -ENOMEM;
1213}
1214
1215static const struct net_device_ops vrf_netdev_ops = {
1216 .ndo_init = vrf_dev_init,
1217 .ndo_uninit = vrf_dev_uninit,
1218 .ndo_start_xmit = vrf_xmit,
Miaohe Lin6819e3f2019-04-20 12:09:39 +08001219 .ndo_set_mac_address = eth_mac_addr,
David Ahern193125d2015-08-13 14:59:10 -06001220 .ndo_get_stats64 = vrf_get_stats64,
1221 .ndo_add_slave = vrf_add_slave,
1222 .ndo_del_slave = vrf_del_slave,
1223};
1224
David Ahernee15ee52015-09-29 20:07:12 -07001225static u32 vrf_fib_table(const struct net_device *dev)
1226{
1227 struct net_vrf *vrf = netdev_priv(dev);
1228
1229 return vrf->tb_id;
1230}
1231
David Ahern73e20b72016-07-04 18:47:41 -07001232static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
1233{
Gao Feng1a4a5bf2017-05-09 18:27:33 +08001234 kfree_skb(skb);
David Ahern73e20b72016-07-04 18:47:41 -07001235 return 0;
1236}
1237
1238static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
1239 struct sk_buff *skb,
1240 struct net_device *dev)
1241{
1242 struct net *net = dev_net(dev);
1243
Gao Feng1a4a5bf2017-05-09 18:27:33 +08001244 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
David Ahern73e20b72016-07-04 18:47:41 -07001245 skb = NULL; /* kfree_skb(skb) handled by nf code */
1246
1247 return skb;
1248}
1249
Andrea Mayer9125abe2020-12-08 18:52:10 +01001250static int vrf_prepare_mac_header(struct sk_buff *skb,
1251 struct net_device *vrf_dev, u16 proto)
1252{
1253 struct ethhdr *eth;
1254 int err;
1255
1256 /* in general, we do not know if there is enough space in the head of
1257 * the packet for hosting the mac header.
1258 */
1259 err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev));
1260 if (unlikely(err))
1261 /* no space in the skb head */
1262 return -ENOBUFS;
1263
1264 __skb_push(skb, ETH_HLEN);
1265 eth = (struct ethhdr *)skb->data;
1266
1267 skb_reset_mac_header(skb);
Eyal Birger012d69f2022-03-31 10:26:43 +03001268 skb_reset_mac_len(skb);
Andrea Mayer9125abe2020-12-08 18:52:10 +01001269
1270 /* we set the ethernet destination and the source addresses to the
1271 * address of the VRF device.
1272 */
1273 ether_addr_copy(eth->h_dest, vrf_dev->dev_addr);
1274 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
1275 eth->h_proto = htons(proto);
1276
1277 /* the destination address of the Ethernet frame corresponds to the
1278 * address set on the VRF interface; therefore, the packet is intended
1279 * to be processed locally.
1280 */
1281 skb->protocol = eth->h_proto;
1282 skb->pkt_type = PACKET_HOST;
1283
1284 skb_postpush_rcsum(skb, skb->data, ETH_HLEN);
1285
1286 skb_pull_inline(skb, ETH_HLEN);
1287
1288 return 0;
1289}
1290
1291/* prepare and add the mac header to the packet if it was not set previously.
1292 * In this way, packet sniffers such as tcpdump can parse the packet correctly.
1293 * If the mac header was already set, the original mac header is left
1294 * untouched and the function returns immediately.
1295 */
1296static int vrf_add_mac_header_if_unset(struct sk_buff *skb,
1297 struct net_device *vrf_dev,
Eyal Birger012d69f2022-03-31 10:26:43 +03001298 u16 proto, struct net_device *orig_dev)
Andrea Mayer9125abe2020-12-08 18:52:10 +01001299{
Eyal Birger012d69f2022-03-31 10:26:43 +03001300 if (skb_mac_header_was_set(skb) && dev_has_header(orig_dev))
Andrea Mayer9125abe2020-12-08 18:52:10 +01001301 return 0;
1302
1303 return vrf_prepare_mac_header(skb, vrf_dev, proto);
1304}
1305
David Ahern35402e32015-10-12 11:47:09 -07001306#if IS_ENABLED(CONFIG_IPV6)
David Ahern74b20582016-05-10 11:19:50 -07001307/* neighbor handling is done with actual device; do not want
1308 * to flip skb->dev for those ndisc packets. This really fails
1309 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1310 * a start.
1311 */
1312static bool ipv6_ndisc_frame(const struct sk_buff *skb)
1313{
1314 const struct ipv6hdr *iph = ipv6_hdr(skb);
1315 bool rc = false;
1316
1317 if (iph->nexthdr == NEXTHDR_ICMP) {
1318 const struct icmp6hdr *icmph;
1319 struct icmp6hdr _icmph;
1320
1321 icmph = skb_header_pointer(skb, sizeof(*iph),
1322 sizeof(_icmph), &_icmph);
1323 if (!icmph)
1324 goto out;
1325
1326 switch (icmph->icmp6_type) {
1327 case NDISC_ROUTER_SOLICITATION:
1328 case NDISC_ROUTER_ADVERTISEMENT:
1329 case NDISC_NEIGHBOUR_SOLICITATION:
1330 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1331 case NDISC_REDIRECT:
1332 rc = true;
1333 break;
1334 }
1335 }
1336
1337out:
1338 return rc;
1339}
1340
David Ahern9ff74382016-06-13 13:44:19 -07001341static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
1342 const struct net_device *dev,
1343 struct flowi6 *fl6,
1344 int ifindex,
David Ahernb75cc8f2018-03-02 08:32:17 -08001345 const struct sk_buff *skb,
David Ahern9ff74382016-06-13 13:44:19 -07001346 int flags)
1347{
1348 struct net_vrf *vrf = netdev_priv(dev);
David Ahern9ff74382016-06-13 13:44:19 -07001349
David Ahern43b059a2018-04-17 17:33:09 -07001350 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
David Ahern9ff74382016-06-13 13:44:19 -07001351}
1352
1353static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1354 int ifindex)
1355{
1356 const struct ipv6hdr *iph = ipv6_hdr(skb);
1357 struct flowi6 fl6 = {
Arnd Bergmannecf09112017-09-12 22:10:53 +02001358 .flowi6_iif = ifindex,
1359 .flowi6_mark = skb->mark,
1360 .flowi6_proto = iph->nexthdr,
David Ahern9ff74382016-06-13 13:44:19 -07001361 .daddr = iph->daddr,
1362 .saddr = iph->saddr,
1363 .flowlabel = ip6_flowinfo(iph),
David Ahern9ff74382016-06-13 13:44:19 -07001364 };
1365 struct net *net = dev_net(vrf_dev);
1366 struct rt6_info *rt6;
1367
David Ahernb75cc8f2018-03-02 08:32:17 -08001368 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
David Ahern9ff74382016-06-13 13:44:19 -07001369 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1370 if (unlikely(!rt6))
1371 return;
1372
1373 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1374 return;
1375
1376 skb_dst_set(skb, &rt6->dst);
1377}
1378
David Ahern74b20582016-05-10 11:19:50 -07001379static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1380 struct sk_buff *skb)
1381{
David Ahern9ff74382016-06-13 13:44:19 -07001382 int orig_iif = skb->skb_iif;
Mike Manning6f12fa72018-11-07 15:36:07 +00001383 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1384 bool is_ndisc = ipv6_ndisc_frame(skb);
David Ahern9ff74382016-06-13 13:44:19 -07001385
Mike Manning6f12fa72018-11-07 15:36:07 +00001386 /* loopback, multicast & non-ND link-local traffic; do not push through
Stephen Suryaputra205704c2020-12-03 22:06:04 -05001387 * packet taps again. Reset pkt_type for upper layers to process skb.
Antoine Tenart603113c52021-06-18 17:15:53 +02001388 * For strict packets with a source LLA, determine the dst using the
1389 * original ifindex.
David Ahernb4869aa2016-06-06 20:50:40 -07001390 */
Antoine Tenart603113c52021-06-18 17:15:53 +02001391 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
David Ahernb4869aa2016-06-06 20:50:40 -07001392 skb->dev = vrf_dev;
1393 skb->skb_iif = vrf_dev->ifindex;
David Aherna04a4802016-10-16 20:02:52 -07001394 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
Antoine Tenart603113c52021-06-18 17:15:53 +02001395
Mike Manning6f12fa72018-11-07 15:36:07 +00001396 if (skb->pkt_type == PACKET_LOOPBACK)
1397 skb->pkt_type = PACKET_HOST;
Antoine Tenart603113c52021-06-18 17:15:53 +02001398 else if (ipv6_addr_type(&ipv6_hdr(skb)->saddr) & IPV6_ADDR_LINKLOCAL)
1399 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1400
David Ahernb4869aa2016-06-06 20:50:40 -07001401 goto out;
1402 }
1403
Mike Manning6f12fa72018-11-07 15:36:07 +00001404 /* if packet is NDISC then keep the ingress interface */
1405 if (!is_ndisc) {
Eyal Birger012d69f2022-03-31 10:26:43 +03001406 struct net_device *orig_dev = skb->dev;
1407
David Ahern926d93a2017-01-03 09:37:55 -08001408 vrf_rx_stats(vrf_dev, skb->len);
David Ahern74b20582016-05-10 11:19:50 -07001409 skb->dev = vrf_dev;
1410 skb->skb_iif = vrf_dev->ifindex;
1411
David Aherna9ec54d2017-03-20 11:19:45 -07001412 if (!list_empty(&vrf_dev->ptype_all)) {
Andrea Mayer04893902020-12-02 14:05:10 +01001413 int err;
1414
1415 err = vrf_add_mac_header_if_unset(skb, vrf_dev,
Eyal Birger012d69f2022-03-31 10:26:43 +03001416 ETH_P_IPV6,
1417 orig_dev);
Andrea Mayer04893902020-12-02 14:05:10 +01001418 if (likely(!err)) {
1419 skb_push(skb, skb->mac_len);
1420 dev_queue_xmit_nit(skb, vrf_dev);
1421 skb_pull(skb, skb->mac_len);
1422 }
David Aherna9ec54d2017-03-20 11:19:45 -07001423 }
David Ahern74b20582016-05-10 11:19:50 -07001424
1425 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1426 }
1427
David Ahern9ff74382016-06-13 13:44:19 -07001428 if (need_strict)
1429 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1430
David Ahern73e20b72016-07-04 18:47:41 -07001431 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
David Ahernb4869aa2016-06-06 20:50:40 -07001432out:
David Ahern74b20582016-05-10 11:19:50 -07001433 return skb;
1434}
1435
1436#else
1437static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1438 struct sk_buff *skb)
1439{
1440 return skb;
1441}
1442#endif
1443
1444static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1445 struct sk_buff *skb)
1446{
Eyal Birger012d69f2022-03-31 10:26:43 +03001447 struct net_device *orig_dev = skb->dev;
1448
David Ahern74b20582016-05-10 11:19:50 -07001449 skb->dev = vrf_dev;
1450 skb->skb_iif = vrf_dev->ifindex;
David Aherna04a4802016-10-16 20:02:52 -07001451 IPCB(skb)->flags |= IPSKB_L3SLAVE;
David Ahern74b20582016-05-10 11:19:50 -07001452
David Aherne58e4152016-10-31 15:54:00 -07001453 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1454 goto out;
1455
David Ahernafe80a42016-06-06 20:50:39 -07001456 /* loopback traffic; do not push through packet taps again.
1457 * Reset pkt_type for upper layers to process skb
1458 */
1459 if (skb->pkt_type == PACKET_LOOPBACK) {
1460 skb->pkt_type = PACKET_HOST;
1461 goto out;
1462 }
1463
David Ahern926d93a2017-01-03 09:37:55 -08001464 vrf_rx_stats(vrf_dev, skb->len);
1465
David Aherndcdd43c2017-03-20 11:19:44 -07001466 if (!list_empty(&vrf_dev->ptype_all)) {
Andrea Mayer04893902020-12-02 14:05:10 +01001467 int err;
1468
Eyal Birger012d69f2022-03-31 10:26:43 +03001469 err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP,
1470 orig_dev);
Andrea Mayer04893902020-12-02 14:05:10 +01001471 if (likely(!err)) {
1472 skb_push(skb, skb->mac_len);
1473 dev_queue_xmit_nit(skb, vrf_dev);
1474 skb_pull(skb, skb->mac_len);
1475 }
David Aherndcdd43c2017-03-20 11:19:44 -07001476 }
David Ahern74b20582016-05-10 11:19:50 -07001477
David Ahern73e20b72016-07-04 18:47:41 -07001478 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
David Ahernafe80a42016-06-06 20:50:39 -07001479out:
David Ahern74b20582016-05-10 11:19:50 -07001480 return skb;
1481}
1482
1483/* called with rcu lock held */
1484static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1485 struct sk_buff *skb,
1486 u16 proto)
1487{
1488 switch (proto) {
1489 case AF_INET:
1490 return vrf_ip_rcv(vrf_dev, skb);
1491 case AF_INET6:
1492 return vrf_ip6_rcv(vrf_dev, skb);
1493 }
1494
1495 return skb;
1496}
1497
1498#if IS_ENABLED(CONFIG_IPV6)
David Ahern4c1feac2016-09-10 12:09:56 -07001499/* send to link-local or multicast address via interface enslaved to
1500 * VRF device. Force lookup to VRF table without changing flow struct
Wei Wang7d9e5f42019-06-20 17:36:41 -07001501 * Note: Caller to this function must hold rcu_read_lock() and no refcnt
1502 * is taken on the dst by this function.
David Ahern4c1feac2016-09-10 12:09:56 -07001503 */
1504static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1505 struct flowi6 *fl6)
David Ahern35402e32015-10-12 11:47:09 -07001506{
David Ahern9ff74382016-06-13 13:44:19 -07001507 struct net *net = dev_net(dev);
Wei Wang7d9e5f42019-06-20 17:36:41 -07001508 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;
David Ahernb0e95cc2016-05-13 12:23:45 -07001509 struct dst_entry *dst = NULL;
David Ahern9ff74382016-06-13 13:44:19 -07001510 struct rt6_info *rt;
David Ahern35402e32015-10-12 11:47:09 -07001511
David Ahern4c1feac2016-09-10 12:09:56 -07001512 /* VRF device does not have a link-local address and
1513 * sending packets to link-local or mcast addresses over
1514 * a VRF device does not make sense
1515 */
1516 if (fl6->flowi6_oif == dev->ifindex) {
1517 dst = &net->ipv6.ip6_null_entry->dst;
David Ahern4c1feac2016-09-10 12:09:56 -07001518 return dst;
David Ahern35402e32015-10-12 11:47:09 -07001519 }
1520
David Ahern4c1feac2016-09-10 12:09:56 -07001521 if (!ipv6_addr_any(&fl6->saddr))
1522 flags |= RT6_LOOKUP_F_HAS_SADDR;
1523
David Ahernb75cc8f2018-03-02 08:32:17 -08001524 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
David Ahern4c1feac2016-09-10 12:09:56 -07001525 if (rt)
1526 dst = &rt->dst;
David Ahern9ff74382016-06-13 13:44:19 -07001527
David Ahernb0e95cc2016-05-13 12:23:45 -07001528 return dst;
David Ahern35402e32015-10-12 11:47:09 -07001529}
1530#endif
1531
David Ahernee15ee52015-09-29 20:07:12 -07001532static const struct l3mdev_ops vrf_l3mdev_ops = {
1533 .l3mdev_fib_table = vrf_fib_table,
David Ahern74b20582016-05-10 11:19:50 -07001534 .l3mdev_l3_rcv = vrf_l3_rcv,
David Ahernebfc1022016-09-10 12:09:55 -07001535 .l3mdev_l3_out = vrf_l3_out,
David Ahern35402e32015-10-12 11:47:09 -07001536#if IS_ENABLED(CONFIG_IPV6)
David Ahern4c1feac2016-09-10 12:09:56 -07001537 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
David Ahern35402e32015-10-12 11:47:09 -07001538#endif
David Ahernee15ee52015-09-29 20:07:12 -07001539};
1540
David Ahern193125d2015-08-13 14:59:10 -06001541static void vrf_get_drvinfo(struct net_device *dev,
1542 struct ethtool_drvinfo *info)
1543{
1544 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1545 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1546}
1547
1548static const struct ethtool_ops vrf_ethtool_ops = {
1549 .get_drvinfo = vrf_get_drvinfo,
1550};
1551
David Ahern1aa6c4f2016-06-08 10:55:40 -07001552static inline size_t vrf_fib_rule_nl_size(void)
1553{
1554 size_t sz;
1555
1556 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1557 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1558 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
Donald Sharp1b71af62018-02-23 14:01:52 -05001559 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */
David Ahern1aa6c4f2016-06-08 10:55:40 -07001560
1561 return sz;
1562}
1563
1564static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1565{
1566 struct fib_rule_hdr *frh;
1567 struct nlmsghdr *nlh;
1568 struct sk_buff *skb;
1569 int err;
1570
David Aherndac91172019-09-25 07:53:19 -07001571 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&
1572 !ipv6_mod_enabled())
David Aherne4348632016-06-09 10:21:00 -07001573 return 0;
1574
David Ahern1aa6c4f2016-06-08 10:55:40 -07001575 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1576 if (!skb)
1577 return -ENOMEM;
1578
1579 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1580 if (!nlh)
1581 goto nla_put_failure;
1582
1583 /* rule only needs to appear once */
David Ahern426c87c2017-04-13 10:57:15 -06001584 nlh->nlmsg_flags |= NLM_F_EXCL;
David Ahern1aa6c4f2016-06-08 10:55:40 -07001585
1586 frh = nlmsg_data(nlh);
1587 memset(frh, 0, sizeof(*frh));
1588 frh->family = family;
1589 frh->action = FR_ACT_TO_TBL;
Donald Sharp1b71af62018-02-23 14:01:52 -05001590
1591 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1592 goto nla_put_failure;
David Ahern1aa6c4f2016-06-08 10:55:40 -07001593
Jeff Barnhill18129a22017-11-01 14:58:09 +00001594 if (nla_put_u8(skb, FRA_L3MDEV, 1))
David Ahern1aa6c4f2016-06-08 10:55:40 -07001595 goto nla_put_failure;
1596
1597 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1598 goto nla_put_failure;
1599
1600 nlmsg_end(skb, nlh);
1601
1602 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1603 skb->sk = dev_net(dev)->rtnl;
1604 if (add_it) {
David Ahernc21ef3e2017-04-16 09:48:24 -07001605 err = fib_nl_newrule(skb, nlh, NULL);
David Ahern1aa6c4f2016-06-08 10:55:40 -07001606 if (err == -EEXIST)
1607 err = 0;
1608 } else {
David Ahernc21ef3e2017-04-16 09:48:24 -07001609 err = fib_nl_delrule(skb, nlh, NULL);
David Ahern1aa6c4f2016-06-08 10:55:40 -07001610 if (err == -ENOENT)
1611 err = 0;
1612 }
1613 nlmsg_free(skb);
1614
1615 return err;
1616
1617nla_put_failure:
1618 nlmsg_free(skb);
1619
1620 return -EMSGSIZE;
1621}
1622
1623static int vrf_add_fib_rules(const struct net_device *dev)
1624{
1625 int err;
1626
1627 err = vrf_fib_rule(dev, AF_INET, true);
1628 if (err < 0)
1629 goto out_err;
1630
1631 err = vrf_fib_rule(dev, AF_INET6, true);
1632 if (err < 0)
1633 goto ipv6_err;
1634
David Aherne58e4152016-10-31 15:54:00 -07001635#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1636 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1637 if (err < 0)
1638 goto ipmr_err;
1639#endif
1640
Patrick Ruddye4a38c02018-10-01 09:41:27 +01001641#if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1642 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1643 if (err < 0)
1644 goto ip6mr_err;
1645#endif
1646
David Ahern1aa6c4f2016-06-08 10:55:40 -07001647 return 0;
1648
Patrick Ruddye4a38c02018-10-01 09:41:27 +01001649#if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1650ip6mr_err:
1651 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false);
1652#endif
1653
David Aherne58e4152016-10-31 15:54:00 -07001654#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1655ipmr_err:
1656 vrf_fib_rule(dev, AF_INET6, false);
1657#endif
1658
David Ahern1aa6c4f2016-06-08 10:55:40 -07001659ipv6_err:
1660 vrf_fib_rule(dev, AF_INET, false);
1661
1662out_err:
1663 netdev_err(dev, "Failed to add FIB rules.\n");
1664 return err;
1665}
1666
David Ahern193125d2015-08-13 14:59:10 -06001667static void vrf_setup(struct net_device *dev)
1668{
1669 ether_setup(dev);
1670
1671 /* Initialize the device structure. */
1672 dev->netdev_ops = &vrf_netdev_ops;
David Ahernee15ee52015-09-29 20:07:12 -07001673 dev->l3mdev_ops = &vrf_l3mdev_ops;
David Ahern193125d2015-08-13 14:59:10 -06001674 dev->ethtool_ops = &vrf_ethtool_ops;
David S. Millercf124db2017-05-08 12:52:56 -04001675 dev->needs_free_netdev = true;
David Ahern193125d2015-08-13 14:59:10 -06001676
1677 /* Fill in device structure with ethernet-generic values. */
1678 eth_hw_addr_random(dev);
1679
1680 /* don't acquire vrf device's netif_tx_lock when transmitting */
1681 dev->features |= NETIF_F_LLTX;
1682
1683 /* don't allow vrf devices to change network namespaces. */
1684 dev->features |= NETIF_F_NETNS_LOCAL;
David Ahern78896812016-06-13 17:14:12 -07001685
1686 /* does not make sense for a VLAN to be added to a vrf device */
1687 dev->features |= NETIF_F_VLAN_CHALLENGED;
1688
1689 /* enable offload features */
1690 dev->features |= NETIF_F_GSO_SOFTWARE;
Davide Caratticb160392018-05-24 17:49:35 +02001691 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
David Ahern78896812016-06-13 17:14:12 -07001692 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1693
1694 dev->hw_features = dev->features;
1695 dev->hw_enc_features = dev->features;
1696
1697 /* default to no qdisc; user can add if desired */
1698 dev->priv_flags |= IFF_NO_QUEUE;
Sabrina Dubroca1017e092019-03-26 18:22:16 +01001699 dev->priv_flags |= IFF_NO_RX_HANDLER;
Miaohe Lin6819e3f2019-04-20 12:09:39 +08001700 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
Hangbin Liuad49bc62019-02-18 17:14:25 +08001701
Miaohe Lin50553762019-04-08 10:04:20 +08001702 /* VRF devices do not care about MTU, but if the MTU is set
1703 * too low then the ipv4 and ipv6 protocols are disabled
1704 * which breaks networking.
1705 */
1706 dev->min_mtu = IPV6_MIN_MTU;
Nicolas Dichtel9bb392f2021-06-08 16:59:51 +02001707 dev->max_mtu = IP6_MAX_MTU;
1708 dev->mtu = dev->max_mtu;
David Ahern193125d2015-08-13 14:59:10 -06001709}
1710
Matthias Schiffera8b8a8892017-06-25 23:56:01 +02001711static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1712 struct netlink_ext_ack *extack)
David Ahern193125d2015-08-13 14:59:10 -06001713{
1714 if (tb[IFLA_ADDRESS]) {
David Ahern53b94832017-08-07 10:08:10 -07001715 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1716 NL_SET_ERR_MSG(extack, "Invalid hardware address");
David Ahern193125d2015-08-13 14:59:10 -06001717 return -EINVAL;
David Ahern53b94832017-08-07 10:08:10 -07001718 }
1719 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1720 NL_SET_ERR_MSG(extack, "Invalid hardware address");
David Ahern193125d2015-08-13 14:59:10 -06001721 return -EADDRNOTAVAIL;
David Ahern53b94832017-08-07 10:08:10 -07001722 }
David Ahern193125d2015-08-13 14:59:10 -06001723 }
1724 return 0;
1725}
1726
1727static void vrf_dellink(struct net_device *dev, struct list_head *head)
1728{
Nikolay Aleksandrovf630c382017-07-06 15:24:40 +03001729 struct net_device *port_dev;
1730 struct list_head *iter;
1731
1732 netdev_for_each_lower_dev(dev, port_dev, iter)
1733 vrf_del_slave(dev, port_dev);
1734
Andrea Mayerc8baec32020-06-20 00:54:44 +02001735 vrf_map_unregister_dev(dev);
1736
David Ahern193125d2015-08-13 14:59:10 -06001737 unregister_netdevice_queue(dev, head);
1738}
1739
1740static int vrf_newlink(struct net *src_net, struct net_device *dev,
Matthias Schiffer7a3f4a12017-06-25 23:55:59 +02001741 struct nlattr *tb[], struct nlattr *data[],
1742 struct netlink_ext_ack *extack)
David Ahern193125d2015-08-13 14:59:10 -06001743{
1744 struct net_vrf *vrf = netdev_priv(dev);
Andrea Mayerc8baec32020-06-20 00:54:44 +02001745 struct netns_vrf *nn_vrf;
David Ahern097d3c92017-06-08 11:31:11 -06001746 bool *add_fib_rules;
1747 struct net *net;
David Ahern1aa6c4f2016-06-08 10:55:40 -07001748 int err;
David Ahern193125d2015-08-13 14:59:10 -06001749
David Ahern53b94832017-08-07 10:08:10 -07001750 if (!data || !data[IFLA_VRF_TABLE]) {
1751 NL_SET_ERR_MSG(extack, "VRF table id is missing");
David Ahern193125d2015-08-13 14:59:10 -06001752 return -EINVAL;
David Ahern53b94832017-08-07 10:08:10 -07001753 }
David Ahern193125d2015-08-13 14:59:10 -06001754
1755 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
David Ahern53b94832017-08-07 10:08:10 -07001756 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1757 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1758 "Invalid VRF table id");
David Ahern24c63bb2017-01-10 15:22:25 -08001759 return -EINVAL;
David Ahern53b94832017-08-07 10:08:10 -07001760 }
David Ahern193125d2015-08-13 14:59:10 -06001761
David Ahern007979e2015-09-29 20:07:10 -07001762 dev->priv_flags |= IFF_L3MDEV_MASTER;
David Ahern193125d2015-08-13 14:59:10 -06001763
David Ahern1aa6c4f2016-06-08 10:55:40 -07001764 err = register_netdevice(dev);
1765 if (err)
1766 goto out;
1767
Andrea Mayerc8baec32020-06-20 00:54:44 +02001768 /* mapping between table_id and vrf;
1769 * note: such binding could not be done in the dev init function
1770 * because dev->ifindex id is not available yet.
1771 */
1772 vrf->ifindex = dev->ifindex;
1773
1774 err = vrf_map_register_dev(dev, extack);
1775 if (err) {
1776 unregister_netdevice(dev);
1777 goto out;
1778 }
1779
David Ahern097d3c92017-06-08 11:31:11 -06001780 net = dev_net(dev);
Andrea Mayerc8baec32020-06-20 00:54:44 +02001781 nn_vrf = net_generic(net, vrf_net_id);
1782
1783 add_fib_rules = &nn_vrf->add_fib_rules;
David Ahern097d3c92017-06-08 11:31:11 -06001784 if (*add_fib_rules) {
David Ahern1aa6c4f2016-06-08 10:55:40 -07001785 err = vrf_add_fib_rules(dev);
1786 if (err) {
Andrea Mayerc8baec32020-06-20 00:54:44 +02001787 vrf_map_unregister_dev(dev);
David Ahern1aa6c4f2016-06-08 10:55:40 -07001788 unregister_netdevice(dev);
1789 goto out;
1790 }
David Ahern097d3c92017-06-08 11:31:11 -06001791 *add_fib_rules = false;
David Ahern1aa6c4f2016-06-08 10:55:40 -07001792 }
1793
1794out:
1795 return err;
David Ahern193125d2015-08-13 14:59:10 -06001796}
1797
1798static size_t vrf_nl_getsize(const struct net_device *dev)
1799{
1800 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1801}
1802
1803static int vrf_fillinfo(struct sk_buff *skb,
1804 const struct net_device *dev)
1805{
1806 struct net_vrf *vrf = netdev_priv(dev);
1807
1808 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1809}
1810
David Ahern67eb0332016-02-02 07:43:45 -08001811static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1812 const struct net_device *slave_dev)
1813{
1814 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1815}
1816
1817static int vrf_fill_slave_info(struct sk_buff *skb,
1818 const struct net_device *vrf_dev,
1819 const struct net_device *slave_dev)
1820{
1821 struct net_vrf *vrf = netdev_priv(vrf_dev);
1822
1823 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1824 return -EMSGSIZE;
1825
1826 return 0;
1827}
1828
David Ahern193125d2015-08-13 14:59:10 -06001829static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1830 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1831};
1832
1833static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1834 .kind = DRV_NAME,
1835 .priv_size = sizeof(struct net_vrf),
1836
1837 .get_size = vrf_nl_getsize,
1838 .policy = vrf_nl_policy,
1839 .validate = vrf_validate,
1840 .fill_info = vrf_fillinfo,
1841
David Ahern67eb0332016-02-02 07:43:45 -08001842 .get_slave_size = vrf_get_slave_size,
1843 .fill_slave_info = vrf_fill_slave_info,
1844
David Ahern193125d2015-08-13 14:59:10 -06001845 .newlink = vrf_newlink,
1846 .dellink = vrf_dellink,
1847 .setup = vrf_setup,
1848 .maxtype = IFLA_VRF_MAX,
1849};
1850
1851static int vrf_device_event(struct notifier_block *unused,
1852 unsigned long event, void *ptr)
1853{
1854 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1855
1856 /* only care about unregister events to drop slave references */
1857 if (event == NETDEV_UNREGISTER) {
David Ahern193125d2015-08-13 14:59:10 -06001858 struct net_device *vrf_dev;
1859
David Ahernfee6d4c2015-10-05 08:51:24 -07001860 if (!netif_is_l3_slave(dev))
David Ahern193125d2015-08-13 14:59:10 -06001861 goto out;
1862
Nikolay Aleksandrov58aa9082015-08-18 20:28:04 +03001863 vrf_dev = netdev_master_upper_dev_get(dev);
1864 vrf_del_slave(vrf_dev, dev);
David Ahern193125d2015-08-13 14:59:10 -06001865 }
1866out:
1867 return NOTIFY_DONE;
1868}
1869
1870static struct notifier_block vrf_notifier_block __read_mostly = {
1871 .notifier_call = vrf_device_event,
1872};
1873
Andrea Mayerc8baec32020-06-20 00:54:44 +02001874static int vrf_map_init(struct vrf_map *vmap)
1875{
1876 spin_lock_init(&vmap->vmap_lock);
1877 hash_init(vmap->ht);
1878
1879 vmap->strict_mode = false;
1880
1881 return 0;
1882}
1883
David Ahern1b6687e2020-07-23 17:23:09 -06001884#ifdef CONFIG_SYSCTL
1885static bool vrf_strict_mode(struct vrf_map *vmap)
1886{
1887 bool strict_mode;
1888
1889 vrf_map_lock(vmap);
1890 strict_mode = vmap->strict_mode;
1891 vrf_map_unlock(vmap);
1892
1893 return strict_mode;
1894}
1895
1896static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode)
1897{
1898 bool *cur_mode;
1899 int res = 0;
1900
1901 vrf_map_lock(vmap);
1902
1903 cur_mode = &vmap->strict_mode;
1904 if (*cur_mode == new_mode)
1905 goto unlock;
1906
1907 if (*cur_mode) {
1908 /* disable strict mode */
1909 *cur_mode = false;
1910 } else {
1911 if (vmap->shared_tables) {
1912 /* we cannot allow strict_mode because there are some
1913 * vrfs that share one or more tables.
1914 */
1915 res = -EBUSY;
1916 goto unlock;
1917 }
1918
1919 /* no tables are shared among vrfs, so we can go back
1920 * to 1:1 association between a vrf with its table.
1921 */
1922 *cur_mode = true;
1923 }
1924
1925unlock:
1926 vrf_map_unlock(vmap);
1927
1928 return res;
1929}
1930
Andrea Mayer33306f12020-06-20 00:54:45 +02001931static int vrf_shared_table_handler(struct ctl_table *table, int write,
1932 void *buffer, size_t *lenp, loff_t *ppos)
1933{
1934 struct net *net = (struct net *)table->extra1;
1935 struct vrf_map *vmap = netns_vrf_map(net);
1936 int proc_strict_mode = 0;
1937 struct ctl_table tmp = {
1938 .procname = table->procname,
1939 .data = &proc_strict_mode,
1940 .maxlen = sizeof(int),
1941 .mode = table->mode,
1942 .extra1 = SYSCTL_ZERO,
1943 .extra2 = SYSCTL_ONE,
1944 };
1945 int ret;
1946
1947 if (!write)
1948 proc_strict_mode = vrf_strict_mode(vmap);
1949
1950 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1951
1952 if (write && ret == 0)
1953 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode);
1954
1955 return ret;
1956}
1957
1958static const struct ctl_table vrf_table[] = {
1959 {
1960 .procname = "strict_mode",
1961 .data = NULL,
1962 .maxlen = sizeof(int),
1963 .mode = 0644,
1964 .proc_handler = vrf_shared_table_handler,
1965 /* set by the vrf_netns_init */
1966 .extra1 = NULL,
1967 },
1968 { },
1969};
1970
David Ahern1b6687e2020-07-23 17:23:09 -06001971static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
David Ahern097d3c92017-06-08 11:31:11 -06001972{
Andrea Mayer33306f12020-06-20 00:54:45 +02001973 struct ctl_table *table;
David Ahern097d3c92017-06-08 11:31:11 -06001974
Andrea Mayer33306f12020-06-20 00:54:45 +02001975 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL);
1976 if (!table)
1977 return -ENOMEM;
1978
1979 /* init the extra1 parameter with the reference to current netns */
1980 table[0].extra1 = net;
1981
1982 nn_vrf->ctl_hdr = register_net_sysctl(net, "net/vrf", table);
1983 if (!nn_vrf->ctl_hdr) {
David Ahern1b6687e2020-07-23 17:23:09 -06001984 kfree(table);
1985 return -ENOMEM;
Andrea Mayer33306f12020-06-20 00:54:45 +02001986 }
1987
David Ahern097d3c92017-06-08 11:31:11 -06001988 return 0;
Andrea Mayer33306f12020-06-20 00:54:45 +02001989}
1990
David Ahern1b6687e2020-07-23 17:23:09 -06001991static void vrf_netns_exit_sysctl(struct net *net)
Andrea Mayer33306f12020-06-20 00:54:45 +02001992{
1993 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1994 struct ctl_table *table;
1995
1996 table = nn_vrf->ctl_hdr->ctl_table_arg;
1997 unregister_net_sysctl_table(nn_vrf->ctl_hdr);
1998 kfree(table);
David Ahern097d3c92017-06-08 11:31:11 -06001999}
David Ahern1b6687e2020-07-23 17:23:09 -06002000#else
2001static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
2002{
2003 return 0;
2004}
2005
2006static void vrf_netns_exit_sysctl(struct net *net)
2007{
2008}
2009#endif
2010
2011/* Initialize per network namespace state */
2012static int __net_init vrf_netns_init(struct net *net)
2013{
2014 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
2015
2016 nn_vrf->add_fib_rules = true;
2017 vrf_map_init(&nn_vrf->vmap);
2018
2019 return vrf_netns_init_sysctl(net, nn_vrf);
2020}
2021
2022static void __net_exit vrf_netns_exit(struct net *net)
2023{
2024 vrf_netns_exit_sysctl(net);
2025}
David Ahern097d3c92017-06-08 11:31:11 -06002026
2027static struct pernet_operations vrf_net_ops __net_initdata = {
2028 .init = vrf_netns_init,
Andrea Mayer33306f12020-06-20 00:54:45 +02002029 .exit = vrf_netns_exit,
David Ahern097d3c92017-06-08 11:31:11 -06002030 .id = &vrf_net_id,
Andrea Mayerc8baec32020-06-20 00:54:44 +02002031 .size = sizeof(struct netns_vrf),
David Ahern097d3c92017-06-08 11:31:11 -06002032};
2033
David Ahern193125d2015-08-13 14:59:10 -06002034static int __init vrf_init_module(void)
2035{
2036 int rc;
2037
David Ahern193125d2015-08-13 14:59:10 -06002038 register_netdevice_notifier(&vrf_notifier_block);
2039
David Ahern097d3c92017-06-08 11:31:11 -06002040 rc = register_pernet_subsys(&vrf_net_ops);
David Ahern193125d2015-08-13 14:59:10 -06002041 if (rc < 0)
2042 goto error;
2043
Andrea Mayera59a8ff2020-06-20 00:54:46 +02002044 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF,
2045 vrf_ifindex_lookup_by_table_id);
2046 if (rc < 0)
2047 goto unreg_pernet;
2048
David Ahern097d3c92017-06-08 11:31:11 -06002049 rc = rtnl_link_register(&vrf_link_ops);
Andrea Mayera59a8ff2020-06-20 00:54:46 +02002050 if (rc < 0)
2051 goto table_lookup_unreg;
David Ahern097d3c92017-06-08 11:31:11 -06002052
David Ahern193125d2015-08-13 14:59:10 -06002053 return 0;
2054
Andrea Mayera59a8ff2020-06-20 00:54:46 +02002055table_lookup_unreg:
2056 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF,
2057 vrf_ifindex_lookup_by_table_id);
2058
2059unreg_pernet:
2060 unregister_pernet_subsys(&vrf_net_ops);
2061
David Ahern193125d2015-08-13 14:59:10 -06002062error:
2063 unregister_netdevice_notifier(&vrf_notifier_block);
David Ahern193125d2015-08-13 14:59:10 -06002064 return rc;
2065}
2066
David Ahern193125d2015-08-13 14:59:10 -06002067module_init(vrf_init_module);
David Ahern193125d2015-08-13 14:59:10 -06002068MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
2069MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
2070MODULE_LICENSE("GPL");
2071MODULE_ALIAS_RTNL_LINK(DRV_NAME);
2072MODULE_VERSION(DRV_VERSION);