xref: /linux/net/ipv6/anycast.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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(const struct net *net,
53 			     const struct in6_addr *addr)
54 {
55 	u32 val = __ipv6_addr_jhash(addr, net_hash_mix(net));
56 
57 	return hash_32(val, IN6_ADDR_HSIZE_SHIFT);
58 }
59 
60 /*
61  *	socket join an anycast group
62  */
63 
64 int ipv6_sock_ac_join(struct sock *sk, int ifindex, const struct in6_addr *addr)
65 {
66 	struct ipv6_pinfo *np = inet6_sk(sk);
67 	struct net_device *dev = NULL;
68 	struct inet6_dev *idev;
69 	struct ipv6_ac_socklist *pac;
70 	struct net *net = sock_net(sk);
71 	int	ishost = !net->ipv6.devconf_all->forwarding;
72 	int	err = 0;
73 
74 	ASSERT_RTNL();
75 
76 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
77 		return -EPERM;
78 	if (ipv6_addr_is_multicast(addr))
79 		return -EINVAL;
80 
81 	if (ifindex)
82 		dev = __dev_get_by_index(net, ifindex);
83 
84 	if (ipv6_chk_addr_and_flags(net, addr, dev, true, 0, IFA_F_TENTATIVE))
85 		return -EINVAL;
86 
87 	pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL);
88 	if (!pac)
89 		return -ENOMEM;
90 	pac->acl_next = NULL;
91 	pac->acl_addr = *addr;
92 
93 	if (ifindex == 0) {
94 		struct rt6_info *rt;
95 
96 		rt = rt6_lookup(net, addr, NULL, 0, NULL, 0);
97 		if (rt) {
98 			dev = rt->dst.dev;
99 			ip6_rt_put(rt);
100 		} else if (ishost) {
101 			err = -EADDRNOTAVAIL;
102 			goto error;
103 		} else {
104 			/* router, no matching interface: just pick one */
105 			dev = __dev_get_by_flags(net, IFF_UP,
106 						 IFF_UP | IFF_LOOPBACK);
107 		}
108 	}
109 
110 	if (!dev) {
111 		err = -ENODEV;
112 		goto error;
113 	}
114 
115 	idev = __in6_dev_get(dev);
116 	if (!idev) {
117 		if (ifindex)
118 			err = -ENODEV;
119 		else
120 			err = -EADDRNOTAVAIL;
121 		goto error;
122 	}
123 	/* reset ishost, now that we have a specific device */
124 	ishost = !idev->cnf.forwarding;
125 
126 	pac->acl_ifindex = dev->ifindex;
127 
128 	/* XXX
129 	 * For hosts, allow link-local or matching prefix anycasts.
130 	 * This obviates the need for propagating anycast routes while
131 	 * still allowing some non-router anycast participation.
132 	 */
133 	if (!ipv6_chk_prefix(addr, dev)) {
134 		if (ishost)
135 			err = -EADDRNOTAVAIL;
136 		if (err)
137 			goto error;
138 	}
139 
140 	err = __ipv6_dev_ac_inc(idev, addr);
141 	if (!err) {
142 		pac->acl_next = np->ipv6_ac_list;
143 		np->ipv6_ac_list = pac;
144 		pac = NULL;
145 	}
146 
147 error:
148 	if (pac)
149 		sock_kfree_s(sk, pac, sizeof(*pac));
150 	return err;
151 }
152 
153 /*
154  *	socket leave an anycast group
155  */
156 int ipv6_sock_ac_drop(struct sock *sk, int ifindex, const struct in6_addr *addr)
157 {
158 	struct ipv6_pinfo *np = inet6_sk(sk);
159 	struct net_device *dev;
160 	struct ipv6_ac_socklist *pac, *prev_pac;
161 	struct net *net = sock_net(sk);
162 
163 	ASSERT_RTNL();
164 
165 	prev_pac = NULL;
166 	for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) {
167 		if ((ifindex == 0 || pac->acl_ifindex == ifindex) &&
168 		     ipv6_addr_equal(&pac->acl_addr, addr))
169 			break;
170 		prev_pac = pac;
171 	}
172 	if (!pac)
173 		return -ENOENT;
174 	if (prev_pac)
175 		prev_pac->acl_next = pac->acl_next;
176 	else
177 		np->ipv6_ac_list = pac->acl_next;
178 
179 	dev = __dev_get_by_index(net, pac->acl_ifindex);
180 	if (dev)
181 		ipv6_dev_ac_dec(dev, &pac->acl_addr);
182 
183 	sock_kfree_s(sk, pac, sizeof(*pac));
184 	return 0;
185 }
186 
187 void __ipv6_sock_ac_close(struct sock *sk)
188 {
189 	struct ipv6_pinfo *np = inet6_sk(sk);
190 	struct net_device *dev = NULL;
191 	struct ipv6_ac_socklist *pac;
192 	struct net *net = sock_net(sk);
193 	int	prev_index;
194 
195 	ASSERT_RTNL();
196 	pac = np->ipv6_ac_list;
197 	np->ipv6_ac_list = NULL;
198 
199 	prev_index = 0;
200 	while (pac) {
201 		struct ipv6_ac_socklist *next = pac->acl_next;
202 
203 		if (pac->acl_ifindex != prev_index) {
204 			dev = __dev_get_by_index(net, pac->acl_ifindex);
205 			prev_index = pac->acl_ifindex;
206 		}
207 		if (dev)
208 			ipv6_dev_ac_dec(dev, &pac->acl_addr);
209 		sock_kfree_s(sk, pac, sizeof(*pac));
210 		pac = next;
211 	}
212 }
213 
214 void ipv6_sock_ac_close(struct sock *sk)
215 {
216 	struct ipv6_pinfo *np = inet6_sk(sk);
217 
218 	if (!np->ipv6_ac_list)
219 		return;
220 	rtnl_lock();
221 	__ipv6_sock_ac_close(sk);
222 	rtnl_unlock();
223 }
224 
225 static void ipv6_add_acaddr_hash(struct net *net, struct ifacaddr6 *aca)
226 {
227 	unsigned int hash = inet6_acaddr_hash(net, &aca->aca_addr);
228 
229 	spin_lock(&acaddr_hash_lock);
230 	hlist_add_head_rcu(&aca->aca_addr_lst, &inet6_acaddr_lst[hash]);
231 	spin_unlock(&acaddr_hash_lock);
232 }
233 
234 static void ipv6_del_acaddr_hash(struct ifacaddr6 *aca)
235 {
236 	spin_lock(&acaddr_hash_lock);
237 	hlist_del_init_rcu(&aca->aca_addr_lst);
238 	spin_unlock(&acaddr_hash_lock);
239 }
240 
241 static void aca_get(struct ifacaddr6 *aca)
242 {
243 	refcount_inc(&aca->aca_refcnt);
244 }
245 
246 static void aca_free_rcu(struct rcu_head *h)
247 {
248 	struct ifacaddr6 *aca = container_of(h, struct ifacaddr6, rcu);
249 
250 	fib6_info_release(aca->aca_rt);
251 	kfree(aca);
252 }
253 
254 static void aca_put(struct ifacaddr6 *ac)
255 {
256 	if (refcount_dec_and_test(&ac->aca_refcnt))
257 		call_rcu_hurry(&ac->rcu, aca_free_rcu);
258 }
259 
260 static struct ifacaddr6 *aca_alloc(struct fib6_info *f6i,
261 				   const struct in6_addr *addr)
262 {
263 	struct ifacaddr6 *aca;
264 
265 	aca = kzalloc(sizeof(*aca), GFP_ATOMIC);
266 	if (!aca)
267 		return NULL;
268 
269 	aca->aca_addr = *addr;
270 	fib6_info_hold(f6i);
271 	aca->aca_rt = f6i;
272 	INIT_HLIST_NODE(&aca->aca_addr_lst);
273 	aca->aca_users = 1;
274 	/* aca_tstamp should be updated upon changes */
275 	aca->aca_cstamp = aca->aca_tstamp = jiffies;
276 	refcount_set(&aca->aca_refcnt, 1);
277 
278 	return aca;
279 }
280 
281 /*
282  *	device anycast group inc (add if not found)
283  */
284 int __ipv6_dev_ac_inc(struct inet6_dev *idev, const struct in6_addr *addr)
285 {
286 	struct ifacaddr6 *aca;
287 	struct fib6_info *f6i;
288 	struct net *net;
289 	int err;
290 
291 	ASSERT_RTNL();
292 
293 	write_lock_bh(&idev->lock);
294 	if (idev->dead) {
295 		err = -ENODEV;
296 		goto out;
297 	}
298 
299 	for (aca = rtnl_dereference(idev->ac_list); aca;
300 	     aca = rtnl_dereference(aca->aca_next)) {
301 		if (ipv6_addr_equal(&aca->aca_addr, addr)) {
302 			aca->aca_users++;
303 			err = 0;
304 			goto out;
305 		}
306 	}
307 
308 	net = dev_net(idev->dev);
309 	f6i = addrconf_f6i_alloc(net, idev, addr, true, GFP_ATOMIC, NULL);
310 	if (IS_ERR(f6i)) {
311 		err = PTR_ERR(f6i);
312 		goto out;
313 	}
314 	aca = aca_alloc(f6i, addr);
315 	if (!aca) {
316 		fib6_info_release(f6i);
317 		err = -ENOMEM;
318 		goto out;
319 	}
320 
321 	/* Hold this for addrconf_join_solict() below before we unlock,
322 	 * it is already exposed via idev->ac_list.
323 	 */
324 	aca_get(aca);
325 	aca->aca_next = idev->ac_list;
326 	rcu_assign_pointer(idev->ac_list, aca);
327 
328 	write_unlock_bh(&idev->lock);
329 
330 	ipv6_add_acaddr_hash(net, aca);
331 
332 	ip6_ins_rt(net, f6i);
333 
334 	addrconf_join_solict(idev->dev, &aca->aca_addr);
335 
336 	aca_put(aca);
337 	return 0;
338 out:
339 	write_unlock_bh(&idev->lock);
340 	return err;
341 }
342 
343 /*
344  *	device anycast group decrement
345  */
346 int __ipv6_dev_ac_dec(struct inet6_dev *idev, const struct in6_addr *addr)
347 {
348 	struct ifacaddr6 *aca, *prev_aca;
349 
350 	ASSERT_RTNL();
351 
352 	write_lock_bh(&idev->lock);
353 	prev_aca = NULL;
354 	for (aca = rtnl_dereference(idev->ac_list); aca;
355 	     aca = rtnl_dereference(aca->aca_next)) {
356 		if (ipv6_addr_equal(&aca->aca_addr, addr))
357 			break;
358 		prev_aca = aca;
359 	}
360 	if (!aca) {
361 		write_unlock_bh(&idev->lock);
362 		return -ENOENT;
363 	}
364 	if (--aca->aca_users > 0) {
365 		write_unlock_bh(&idev->lock);
366 		return 0;
367 	}
368 	if (prev_aca)
369 		rcu_assign_pointer(prev_aca->aca_next, aca->aca_next);
370 	else
371 		rcu_assign_pointer(idev->ac_list, aca->aca_next);
372 	write_unlock_bh(&idev->lock);
373 	ipv6_del_acaddr_hash(aca);
374 	addrconf_leave_solict(idev, &aca->aca_addr);
375 
376 	ip6_del_rt(dev_net(idev->dev), aca->aca_rt, false);
377 
378 	aca_put(aca);
379 	return 0;
380 }
381 
382 /* called with rtnl_lock() */
383 static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr)
384 {
385 	struct inet6_dev *idev = __in6_dev_get(dev);
386 
387 	if (!idev)
388 		return -ENODEV;
389 	return __ipv6_dev_ac_dec(idev, addr);
390 }
391 
392 void ipv6_ac_destroy_dev(struct inet6_dev *idev)
393 {
394 	struct ifacaddr6 *aca;
395 
396 	write_lock_bh(&idev->lock);
397 	while ((aca = rtnl_dereference(idev->ac_list)) != NULL) {
398 		rcu_assign_pointer(idev->ac_list, aca->aca_next);
399 		write_unlock_bh(&idev->lock);
400 
401 		ipv6_del_acaddr_hash(aca);
402 
403 		addrconf_leave_solict(idev, &aca->aca_addr);
404 
405 		ip6_del_rt(dev_net(idev->dev), aca->aca_rt, false);
406 
407 		aca_put(aca);
408 
409 		write_lock_bh(&idev->lock);
410 	}
411 	write_unlock_bh(&idev->lock);
412 }
413 
414 /*
415  *	check if the interface has this anycast address
416  *	called with rcu_read_lock()
417  */
418 static bool ipv6_chk_acast_dev(struct net_device *dev, const struct in6_addr *addr)
419 {
420 	struct inet6_dev *idev;
421 	struct ifacaddr6 *aca;
422 
423 	idev = __in6_dev_get(dev);
424 	if (idev) {
425 		for (aca = rcu_dereference(idev->ac_list); aca;
426 		     aca = rcu_dereference(aca->aca_next))
427 			if (ipv6_addr_equal(&aca->aca_addr, addr))
428 				break;
429 		return aca != NULL;
430 	}
431 	return false;
432 }
433 
434 /*
435  *	check if given interface (or any, if dev==0) has this anycast address
436  */
437 bool ipv6_chk_acast_addr(struct net *net, struct net_device *dev,
438 			 const struct in6_addr *addr)
439 {
440 	struct net_device *nh_dev;
441 	struct ifacaddr6 *aca;
442 	bool found = false;
443 
444 	rcu_read_lock();
445 	if (dev)
446 		found = ipv6_chk_acast_dev(dev, addr);
447 	else {
448 		unsigned int hash = inet6_acaddr_hash(net, addr);
449 
450 		hlist_for_each_entry_rcu(aca, &inet6_acaddr_lst[hash],
451 					 aca_addr_lst) {
452 			nh_dev = fib6_info_nh_dev(aca->aca_rt);
453 			if (!nh_dev || !net_eq(dev_net(nh_dev), net))
454 				continue;
455 			if (ipv6_addr_equal(&aca->aca_addr, addr)) {
456 				found = true;
457 				break;
458 			}
459 		}
460 	}
461 	rcu_read_unlock();
462 	return found;
463 }
464 
465 /*	check if this anycast address is link-local on given interface or
466  *	is global
467  */
468 bool ipv6_chk_acast_addr_src(struct net *net, struct net_device *dev,
469 			     const struct in6_addr *addr)
470 {
471 	return ipv6_chk_acast_addr(net,
472 				   (ipv6_addr_type(addr) & IPV6_ADDR_LINKLOCAL ?
473 				    dev : NULL),
474 				   addr);
475 }
476 
477 #ifdef CONFIG_PROC_FS
478 struct ac6_iter_state {
479 	struct seq_net_private p;
480 	struct net_device *dev;
481 };
482 
483 #define ac6_seq_private(seq)	((struct ac6_iter_state *)(seq)->private)
484 
485 static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq)
486 {
487 	struct ac6_iter_state *state = ac6_seq_private(seq);
488 	struct net *net = seq_file_net(seq);
489 	struct ifacaddr6 *im = NULL;
490 
491 	for_each_netdev_rcu(net, state->dev) {
492 		struct inet6_dev *idev;
493 
494 		idev = __in6_dev_get(state->dev);
495 		if (!idev)
496 			continue;
497 		im = rcu_dereference(idev->ac_list);
498 		if (im)
499 			break;
500 	}
501 	return im;
502 }
503 
504 static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im)
505 {
506 	struct ac6_iter_state *state = ac6_seq_private(seq);
507 	struct inet6_dev *idev;
508 
509 	im = rcu_dereference(im->aca_next);
510 	while (!im) {
511 		state->dev = next_net_device_rcu(state->dev);
512 		if (!state->dev)
513 			break;
514 		idev = __in6_dev_get(state->dev);
515 		if (!idev)
516 			continue;
517 		im = rcu_dereference(idev->ac_list);
518 	}
519 	return im;
520 }
521 
522 static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos)
523 {
524 	struct ifacaddr6 *im = ac6_get_first(seq);
525 	if (im)
526 		while (pos && (im = ac6_get_next(seq, im)) != NULL)
527 			--pos;
528 	return pos ? NULL : im;
529 }
530 
531 static void *ac6_seq_start(struct seq_file *seq, loff_t *pos)
532 	__acquires(RCU)
533 {
534 	rcu_read_lock();
535 	return ac6_get_idx(seq, *pos);
536 }
537 
538 static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
539 {
540 	struct ifacaddr6 *im = ac6_get_next(seq, v);
541 
542 	++*pos;
543 	return im;
544 }
545 
546 static void ac6_seq_stop(struct seq_file *seq, void *v)
547 	__releases(RCU)
548 {
549 	rcu_read_unlock();
550 }
551 
552 static int ac6_seq_show(struct seq_file *seq, void *v)
553 {
554 	struct ifacaddr6 *im = (struct ifacaddr6 *)v;
555 	struct ac6_iter_state *state = ac6_seq_private(seq);
556 
557 	seq_printf(seq, "%-4d %-15s %pi6 %5d\n",
558 		   state->dev->ifindex, state->dev->name,
559 		   &im->aca_addr, im->aca_users);
560 	return 0;
561 }
562 
563 static const struct seq_operations ac6_seq_ops = {
564 	.start	=	ac6_seq_start,
565 	.next	=	ac6_seq_next,
566 	.stop	=	ac6_seq_stop,
567 	.show	=	ac6_seq_show,
568 };
569 
570 int __net_init ac6_proc_init(struct net *net)
571 {
572 	if (!proc_create_net("anycast6", 0444, net->proc_net, &ac6_seq_ops,
573 			sizeof(struct ac6_iter_state)))
574 		return -ENOMEM;
575 
576 	return 0;
577 }
578 
579 void ac6_proc_exit(struct net *net)
580 {
581 	remove_proc_entry("anycast6", net->proc_net);
582 }
583 #endif
584 
585 /*	Init / cleanup code
586  */
587 int __init ipv6_anycast_init(void)
588 {
589 	int i;
590 
591 	for (i = 0; i < IN6_ADDR_HSIZE; i++)
592 		INIT_HLIST_HEAD(&inet6_acaddr_lst[i]);
593 	return 0;
594 }
595 
596 void ipv6_anycast_cleanup(void)
597 {
598 	int i;
599 
600 	spin_lock(&acaddr_hash_lock);
601 	for (i = 0; i < IN6_ADDR_HSIZE; i++)
602 		WARN_ON(!hlist_empty(&inet6_acaddr_lst[i]));
603 	spin_unlock(&acaddr_hash_lock);
604 }
605