1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Anycast support for IPv6
4 * Linux INET6 implementation
5 *
6 * Authors:
7 * David L Stevens (dlstevens@us.ibm.com)
8 *
9 * based heavily on net/ipv6/mcast.c
10 */
11
12 #include <linux/capability.h>
13 #include <linux/module.h>
14 #include <linux/errno.h>
15 #include <linux/types.h>
16 #include <linux/random.h>
17 #include <linux/string.h>
18 #include <linux/socket.h>
19 #include <linux/sockios.h>
20 #include <linux/net.h>
21 #include <linux/in6.h>
22 #include <linux/netdevice.h>
23 #include <linux/if_arp.h>
24 #include <linux/route.h>
25 #include <linux/init.h>
26 #include <linux/proc_fs.h>
27 #include <linux/seq_file.h>
28 #include <linux/slab.h>
29
30 #include <net/net_namespace.h>
31 #include <net/sock.h>
32 #include <net/snmp.h>
33
34 #include <net/ipv6.h>
35 #include <net/protocol.h>
36 #include <net/if_inet6.h>
37 #include <net/ndisc.h>
38 #include <net/addrconf.h>
39 #include <net/ip6_route.h>
40
41 #include <net/checksum.h>
42
43 #define IN6_ADDR_HSIZE_SHIFT 8
44 #define IN6_ADDR_HSIZE BIT(IN6_ADDR_HSIZE_SHIFT)
45 /* anycast address hash table
46 */
47 static struct hlist_head inet6_acaddr_lst[IN6_ADDR_HSIZE];
48 static DEFINE_SPINLOCK(acaddr_hash_lock);
49
50 static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr);
51
52 static u32 inet6_acaddr_hash(struct net *net, const struct in6_addr *addr)
53 {
54 u32 val = ipv6_addr_hash(addr) ^ net_hash_mix(net);
55
56 return hash_32(val, IN6_ADDR_HSIZE_SHIFT);
57 }
58
59 /*
60 * socket join an anycast group
61 */
62
63 int ipv6_sock_ac_join(struct sock *sk, int ifindex, const struct in6_addr *addr)
64 {
65 struct ipv6_pinfo *np = inet6_sk(sk);
66 struct net_device *dev = NULL;
67 struct inet6_dev *idev;
68 struct ipv6_ac_socklist *pac;
69 struct net *net = sock_net(sk);
70 int ishost = !net->ipv6.devconf_all->forwarding;
71 int err = 0;
72
73 ASSERT_RTNL();
74
75 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
76 return -EPERM;
77 if (ipv6_addr_is_multicast(addr))
78 return -EINVAL;
79
80 if (ifindex)
81 dev = __dev_get_by_index(net, ifindex);
82
83 if (ipv6_chk_addr_and_flags(net, addr, dev, true, 0, IFA_F_TENTATIVE))
84 return -EINVAL;
85
86 pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL);
87 if (!pac)
88 return -ENOMEM;
89 pac->acl_next = NULL;
90 pac->acl_addr = *addr;
91
92 if (ifindex == 0) {
93 struct rt6_info *rt;
94
95 rt = rt6_lookup(net, addr, NULL, 0, NULL, 0);
96 if (rt) {
97 dev = rt->dst.dev;
98 ip6_rt_put(rt);
99 } else if (ishost) {
100 err = -EADDRNOTAVAIL;
101 goto error;
102 } else {
103 /* router, no matching interface: just pick one */
104 dev = __dev_get_by_flags(net, IFF_UP,
105 IFF_UP | IFF_LOOPBACK);
106 }
107 }
108
109 if (!dev) {
110 err = -ENODEV;
111 goto error;
112 }
113
114 idev = __in6_dev_get(dev);
115 if (!idev) {
116 if (ifindex)
117 err = -ENODEV;
118 else
119 err = -EADDRNOTAVAIL;
120 goto error;
121 }
122 /* reset ishost, now that we have a specific device */
123 ishost = !idev->cnf.forwarding;
124
125 pac->acl_ifindex = dev->ifindex;
126
127 /* XXX
128 * For hosts, allow link-local or matching prefix anycasts.
129 * This obviates the need for propagating anycast routes while
130 * still allowing some non-router anycast participation.
131 */
132 if (!ipv6_chk_prefix(addr, dev)) {
133 if (ishost)
134 err = -EADDRNOTAVAIL;
135 if (err)
136 goto error;
137 }
138
139 err = __ipv6_dev_ac_inc(idev, addr);
140 if (!err) {
141 pac->acl_next = np->ipv6_ac_list;
142 np->ipv6_ac_list = pac;
143 pac = NULL;
144 }
145
146 error:
147 if (pac)
148 sock_kfree_s(sk, pac, sizeof(*pac));
149 return err;
150 }
151
152 /*
153 * socket leave an anycast group
154 */
155 int ipv6_sock_ac_drop(struct sock *sk, int ifindex, const struct in6_addr *addr)
156 {
157 struct ipv6_pinfo *np = inet6_sk(sk);
158 struct net_device *dev;
159 struct ipv6_ac_socklist *pac, *prev_pac;
160 struct net *net = sock_net(sk);
161
162 ASSERT_RTNL();
163
164 prev_pac = NULL;
165 for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) {
166 if ((ifindex == 0 || pac->acl_ifindex == ifindex) &&
167 ipv6_addr_equal(&pac->acl_addr, addr))
168 break;
169 prev_pac = pac;
170 }
171 if (!pac)
172 return -ENOENT;
173 if (prev_pac)
174 prev_pac->acl_next = pac->acl_next;
175 else
176 np->ipv6_ac_list = pac->acl_next;
177
178 dev = __dev_get_by_index(net, pac->acl_ifindex);
179 if (dev)
180 ipv6_dev_ac_dec(dev, &pac->acl_addr);
181
182 sock_kfree_s(sk, pac, sizeof(*pac));
183 return 0;
184 }
185
186 void __ipv6_sock_ac_close(struct sock *sk)
187 {
188 struct ipv6_pinfo *np = inet6_sk(sk);
189 struct net_device *dev = NULL;
190 struct ipv6_ac_socklist *pac;
191 struct net *net = sock_net(sk);
192 int prev_index;
193
194 ASSERT_RTNL();
195 pac = np->ipv6_ac_list;
196 np->ipv6_ac_list = NULL;
197
198 prev_index = 0;
199 while (pac) {
200 struct ipv6_ac_socklist *next = pac->acl_next;
201
202 if (pac->acl_ifindex != prev_index) {
203 dev = __dev_get_by_index(net, pac->acl_ifindex);
204 prev_index = pac->acl_ifindex;
205 }
206 if (dev)
207 ipv6_dev_ac_dec(dev, &pac->acl_addr);
208 sock_kfree_s(sk, pac, sizeof(*pac));
209 pac = next;
210 }
211 }
212
213 void ipv6_sock_ac_close(struct sock *sk)
214 {
215 struct ipv6_pinfo *np = inet6_sk(sk);
216
217 if (!np->ipv6_ac_list)
218 return;
219 rtnl_lock();
220 __ipv6_sock_ac_close(sk);
221 rtnl_unlock();
222 }
223
224 static void ipv6_add_acaddr_hash(struct net *net, struct ifacaddr6 *aca)
225 {
226 unsigned int hash = inet6_acaddr_hash(net, &aca->aca_addr);
227
228 spin_lock(&acaddr_hash_lock);
229 hlist_add_head_rcu(&aca->aca_addr_lst, &inet6_acaddr_lst[hash]);
230 spin_unlock(&acaddr_hash_lock);
231 }
232
233 static void ipv6_del_acaddr_hash(struct ifacaddr6 *aca)
234 {
235 spin_lock(&acaddr_hash_lock);
236 hlist_del_init_rcu(&aca->aca_addr_lst);
237 spin_unlock(&acaddr_hash_lock);
238 }
239
240 static void aca_get(struct ifacaddr6 *aca)
241 {
242 refcount_inc(&aca->aca_refcnt);
243 }
244
245 static void aca_free_rcu(struct rcu_head *h)
246 {
247 struct ifacaddr6 *aca = container_of(h, struct ifacaddr6, rcu);
248
249 fib6_info_release(aca->aca_rt);
250 kfree(aca);
251 }
252
253 static void aca_put(struct ifacaddr6 *ac)
254 {
255 if (refcount_dec_and_test(&ac->aca_refcnt))
256 call_rcu_hurry(&ac->rcu, aca_free_rcu);
257 }
258
259 static struct ifacaddr6 *aca_alloc(struct fib6_info *f6i,
260 const struct in6_addr *addr)
261 {
262 struct ifacaddr6 *aca;
263
264 aca = kzalloc(sizeof(*aca), GFP_ATOMIC);
265 if (!aca)
266 return NULL;
267
268 aca->aca_addr = *addr;
269 fib6_info_hold(f6i);
270 aca->aca_rt = f6i;
271 INIT_HLIST_NODE(&aca->aca_addr_lst);
272 aca->aca_users = 1;
273 /* aca_tstamp should be updated upon changes */
274 aca->aca_cstamp = aca->aca_tstamp = jiffies;
275 refcount_set(&aca->aca_refcnt, 1);
276
277 return aca;
278 }
279
280 /*
281 * device anycast group inc (add if not found)
282 */
283 int __ipv6_dev_ac_inc(struct inet6_dev *idev, const struct in6_addr *addr)
284 {
285 struct ifacaddr6 *aca;
286 struct fib6_info *f6i;
287 struct net *net;
288 int err;
289
290 ASSERT_RTNL();
291
292 write_lock_bh(&idev->lock);
293 if (idev->dead) {
294 err = -ENODEV;
295 goto out;
296 }
297
298 for (aca = rtnl_dereference(idev->ac_list); aca;
299 aca = rtnl_dereference(aca->aca_next)) {
300 if (ipv6_addr_equal(&aca->aca_addr, addr)) {
301 aca->aca_users++;
302 err = 0;
303 goto out;
304 }
305 }
306
307 net = dev_net(idev->dev);
308 f6i = addrconf_f6i_alloc(net, idev, addr, true, GFP_ATOMIC, NULL);
309 if (IS_ERR(f6i)) {
310 err = PTR_ERR(f6i);
311 goto out;
312 }
313 aca = aca_alloc(f6i, addr);
314 if (!aca) {
315 fib6_info_release(f6i);
316 err = -ENOMEM;
317 goto out;
318 }
319
320 /* Hold this for addrconf_join_solict() below before we unlock,
321 * it is already exposed via idev->ac_list.
322 */
323 aca_get(aca);
324 aca->aca_next = idev->ac_list;
325 rcu_assign_pointer(idev->ac_list, aca);
326
327 write_unlock_bh(&idev->lock);
328
329 ipv6_add_acaddr_hash(net, aca);
330
331 ip6_ins_rt(net, f6i);
332
333 addrconf_join_solict(idev->dev, &aca->aca_addr);
334
335 aca_put(aca);
336 return 0;
337 out:
338 write_unlock_bh(&idev->lock);
339 return err;
340 }
341
342 /*
343 * device anycast group decrement
344 */
345 int __ipv6_dev_ac_dec(struct inet6_dev *idev, const struct in6_addr *addr)
346 {
347 struct ifacaddr6 *aca, *prev_aca;
348
349 ASSERT_RTNL();
350
351 write_lock_bh(&idev->lock);
352 prev_aca = NULL;
353 for (aca = rtnl_dereference(idev->ac_list); aca;
354 aca = rtnl_dereference(aca->aca_next)) {
355 if (ipv6_addr_equal(&aca->aca_addr, addr))
356 break;
357 prev_aca = aca;
358 }
359 if (!aca) {
360 write_unlock_bh(&idev->lock);
361 return -ENOENT;
362 }
363 if (--aca->aca_users > 0) {
364 write_unlock_bh(&idev->lock);
365 return 0;
366 }
367 if (prev_aca)
368 rcu_assign_pointer(prev_aca->aca_next, aca->aca_next);
369 else
370 rcu_assign_pointer(idev->ac_list, aca->aca_next);
371 write_unlock_bh(&idev->lock);
372 ipv6_del_acaddr_hash(aca);
373 addrconf_leave_solict(idev, &aca->aca_addr);
374
375 ip6_del_rt(dev_net(idev->dev), aca->aca_rt, false);
376
377 aca_put(aca);
378 return 0;
379 }
380
381 /* called with rtnl_lock() */
382 static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr)
383 {
384 struct inet6_dev *idev = __in6_dev_get(dev);
385
386 if (!idev)
387 return -ENODEV;
388 return __ipv6_dev_ac_dec(idev, addr);
389 }
390
391 void ipv6_ac_destroy_dev(struct inet6_dev *idev)
392 {
393 struct ifacaddr6 *aca;
394
395 write_lock_bh(&idev->lock);
396 while ((aca = rtnl_dereference(idev->ac_list)) != NULL) {
397 rcu_assign_pointer(idev->ac_list, aca->aca_next);
398 write_unlock_bh(&idev->lock);
399
400 ipv6_del_acaddr_hash(aca);
401
402 addrconf_leave_solict(idev, &aca->aca_addr);
403
404 ip6_del_rt(dev_net(idev->dev), aca->aca_rt, false);
405
406 aca_put(aca);
407
408 write_lock_bh(&idev->lock);
409 }
410 write_unlock_bh(&idev->lock);
411 }
412
413 /*
414 * check if the interface has this anycast address
415 * called with rcu_read_lock()
416 */
417 static bool ipv6_chk_acast_dev(struct net_device *dev, const struct in6_addr *addr)
418 {
419 struct inet6_dev *idev;
420 struct ifacaddr6 *aca;
421
422 idev = __in6_dev_get(dev);
423 if (idev) {
424 for (aca = rcu_dereference(idev->ac_list); aca;
425 aca = rcu_dereference(aca->aca_next))
426 if (ipv6_addr_equal(&aca->aca_addr, addr))
427 break;
428 return aca != NULL;
429 }
430 return false;
431 }
432
433 /*
434 * check if given interface (or any, if dev==0) has this anycast address
435 */
436 bool ipv6_chk_acast_addr(struct net *net, struct net_device *dev,
437 const struct in6_addr *addr)
438 {
439 struct net_device *nh_dev;
440 struct ifacaddr6 *aca;
441 bool found = false;
442
443 rcu_read_lock();
444 if (dev)
445 found = ipv6_chk_acast_dev(dev, addr);
446 else {
447 unsigned int hash = inet6_acaddr_hash(net, addr);
448
449 hlist_for_each_entry_rcu(aca, &inet6_acaddr_lst[hash],
450 aca_addr_lst) {
451 nh_dev = fib6_info_nh_dev(aca->aca_rt);
452 if (!nh_dev || !net_eq(dev_net(nh_dev), net))
453 continue;
454 if (ipv6_addr_equal(&aca->aca_addr, addr)) {
455 found = true;
456 break;
457 }
458 }
459 }
460 rcu_read_unlock();
461 return found;
462 }
463
464 /* check if this anycast address is link-local on given interface or
465 * is global
466 */
467 bool ipv6_chk_acast_addr_src(struct net *net, struct net_device *dev,
468 const struct in6_addr *addr)
469 {
470 return ipv6_chk_acast_addr(net,
471 (ipv6_addr_type(addr) & IPV6_ADDR_LINKLOCAL ?
472 dev : NULL),
473 addr);
474 }
475
476 #ifdef CONFIG_PROC_FS
477 struct ac6_iter_state {
478 struct seq_net_private p;
479 struct net_device *dev;
480 };
481
482 #define ac6_seq_private(seq) ((struct ac6_iter_state *)(seq)->private)
483
484 static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq)
485 {
486 struct ac6_iter_state *state = ac6_seq_private(seq);
487 struct net *net = seq_file_net(seq);
488 struct ifacaddr6 *im = NULL;
489
490 for_each_netdev_rcu(net, state->dev) {
491 struct inet6_dev *idev;
492
493 idev = __in6_dev_get(state->dev);
494 if (!idev)
495 continue;
496 im = rcu_dereference(idev->ac_list);
497 if (im)
498 break;
499 }
500 return im;
501 }
502
503 static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im)
504 {
505 struct ac6_iter_state *state = ac6_seq_private(seq);
506 struct inet6_dev *idev;
507
508 im = rcu_dereference(im->aca_next);
509 while (!im) {
510 state->dev = next_net_device_rcu(state->dev);
511 if (!state->dev)
512 break;
513 idev = __in6_dev_get(state->dev);
514 if (!idev)
515 continue;
516 im = rcu_dereference(idev->ac_list);
517 }
518 return im;
519 }
520
521 static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos)
522 {
523 struct ifacaddr6 *im = ac6_get_first(seq);
524 if (im)
525 while (pos && (im = ac6_get_next(seq, im)) != NULL)
526 --pos;
527 return pos ? NULL : im;
528 }
529
530 static void *ac6_seq_start(struct seq_file *seq, loff_t *pos)
531 __acquires(RCU)
532 {
533 rcu_read_lock();
534 return ac6_get_idx(seq, *pos);
535 }
536
537 static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
538 {
539 struct ifacaddr6 *im = ac6_get_next(seq, v);
540
541 ++*pos;
542 return im;
543 }
544
545 static void ac6_seq_stop(struct seq_file *seq, void *v)
546 __releases(RCU)
547 {
548 rcu_read_unlock();
549 }
550
551 static int ac6_seq_show(struct seq_file *seq, void *v)
552 {
553 struct ifacaddr6 *im = (struct ifacaddr6 *)v;
554 struct ac6_iter_state *state = ac6_seq_private(seq);
555
556 seq_printf(seq, "%-4d %-15s %pi6 %5d\n",
557 state->dev->ifindex, state->dev->name,
558 &im->aca_addr, im->aca_users);
559 return 0;
560 }
561
562 static const struct seq_operations ac6_seq_ops = {
563 .start = ac6_seq_start,
564 .next = ac6_seq_next,
565 .stop = ac6_seq_stop,
566 .show = ac6_seq_show,
567 };
568
569 int __net_init ac6_proc_init(struct net *net)
570 {
571 if (!proc_create_net("anycast6", 0444, net->proc_net, &ac6_seq_ops,
572 sizeof(struct ac6_iter_state)))
573 return -ENOMEM;
574
575 return 0;
576 }
577
578 void ac6_proc_exit(struct net *net)
579 {
580 remove_proc_entry("anycast6", net->proc_net);
581 }
582 #endif
583
584 /* Init / cleanup code
585 */
586 int __init ipv6_anycast_init(void)
587 {
588 int i;
589
590 for (i = 0; i < IN6_ADDR_HSIZE; i++)
591 INIT_HLIST_HEAD(&inet6_acaddr_lst[i]);
592 return 0;
593 }
594
595 void ipv6_anycast_cleanup(void)
596 {
597 int i;
598
599 spin_lock(&acaddr_hash_lock);
600 for (i = 0; i < IN6_ADDR_HSIZE; i++)
601 WARN_ON(!hlist_empty(&inet6_acaddr_lst[i]));
602 spin_unlock(&acaddr_hash_lock);
603 }
604
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.