xref: /openbmc/linux/net/ipv6/anycast.c (revision 95298d63) 
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 	if (!np->ipv6_ac_list)
195 		return;
196 
197 	rtnl_lock();
198 	pac = np->ipv6_ac_list;
199 	np->ipv6_ac_list = NULL;
200 
201 	prev_index = 0;
202 	while (pac) {
203 		struct ipv6_ac_socklist *next = pac->acl_next;
204 
205 		if (pac->acl_ifindex != prev_index) {
206 			dev = __dev_get_by_index(net, pac->acl_ifindex);
207 			prev_index = pac->acl_ifindex;
208 		}
209 		if (dev)
210 			ipv6_dev_ac_dec(dev, &pac->acl_addr);
211 		sock_kfree_s(sk, pac, sizeof(*pac));
212 		pac = next;
213 	}
214 	rtnl_unlock();
215 }
216 
217 static void ipv6_add_acaddr_hash(struct net *net, struct ifacaddr6 *aca)
218 {
219 	unsigned int hash = inet6_acaddr_hash(net, &aca->aca_addr);
220 
221 	spin_lock(&acaddr_hash_lock);
222 	hlist_add_head_rcu(&aca->aca_addr_lst, &inet6_acaddr_lst[hash]);
223 	spin_unlock(&acaddr_hash_lock);
224 }
225 
226 static void ipv6_del_acaddr_hash(struct ifacaddr6 *aca)
227 {
228 	spin_lock(&acaddr_hash_lock);
229 	hlist_del_init_rcu(&aca->aca_addr_lst);
230 	spin_unlock(&acaddr_hash_lock);
231 }
232 
233 static void aca_get(struct ifacaddr6 *aca)
234 {
235 	refcount_inc(&aca->aca_refcnt);
236 }
237 
238 static void aca_free_rcu(struct rcu_head *h)
239 {
240 	struct ifacaddr6 *aca = container_of(h, struct ifacaddr6, rcu);
241 
242 	fib6_info_release(aca->aca_rt);
243 	kfree(aca);
244 }
245 
246 static void aca_put(struct ifacaddr6 *ac)
247 {
248 	if (refcount_dec_and_test(&ac->aca_refcnt)) {
249 		call_rcu(&ac->rcu, aca_free_rcu);
250 	}
251 }
252 
253 static struct ifacaddr6 *aca_alloc(struct fib6_info *f6i,
254 				   const struct in6_addr *addr)
255 {
256 	struct ifacaddr6 *aca;
257 
258 	aca = kzalloc(sizeof(*aca), GFP_ATOMIC);
259 	if (!aca)
260 		return NULL;
261 
262 	aca->aca_addr = *addr;
263 	fib6_info_hold(f6i);
264 	aca->aca_rt = f6i;
265 	INIT_HLIST_NODE(&aca->aca_addr_lst);
266 	aca->aca_users = 1;
267 	/* aca_tstamp should be updated upon changes */
268 	aca->aca_cstamp = aca->aca_tstamp = jiffies;
269 	refcount_set(&aca->aca_refcnt, 1);
270 
271 	return aca;
272 }
273 
274 /*
275  *	device anycast group inc (add if not found)
276  */
277 int __ipv6_dev_ac_inc(struct inet6_dev *idev, const struct in6_addr *addr)
278 {
279 	struct ifacaddr6 *aca;
280 	struct fib6_info *f6i;
281 	struct net *net;
282 	int err;
283 
284 	ASSERT_RTNL();
285 
286 	write_lock_bh(&idev->lock);
287 	if (idev->dead) {
288 		err = -ENODEV;
289 		goto out;
290 	}
291 
292 	for (aca = idev->ac_list; aca; aca = aca->aca_next) {
293 		if (ipv6_addr_equal(&aca->aca_addr, addr)) {
294 			aca->aca_users++;
295 			err = 0;
296 			goto out;
297 		}
298 	}
299 
300 	net = dev_net(idev->dev);
301 	f6i = addrconf_f6i_alloc(net, idev, addr, true, GFP_ATOMIC);
302 	if (IS_ERR(f6i)) {
303 		err = PTR_ERR(f6i);
304 		goto out;
305 	}
306 	aca = aca_alloc(f6i, addr);
307 	if (!aca) {
308 		fib6_info_release(f6i);
309 		err = -ENOMEM;
310 		goto out;
311 	}
312 
313 	aca->aca_next = idev->ac_list;
314 	idev->ac_list = aca;
315 
316 	/* Hold this for addrconf_join_solict() below before we unlock,
317 	 * it is already exposed via idev->ac_list.
318 	 */
319 	aca_get(aca);
320 	write_unlock_bh(&idev->lock);
321 
322 	ipv6_add_acaddr_hash(net, aca);
323 
324 	ip6_ins_rt(net, f6i);
325 
326 	addrconf_join_solict(idev->dev, &aca->aca_addr);
327 
328 	aca_put(aca);
329 	return 0;
330 out:
331 	write_unlock_bh(&idev->lock);
332 	return err;
333 }
334 
335 /*
336  *	device anycast group decrement
337  */
338 int __ipv6_dev_ac_dec(struct inet6_dev *idev, const struct in6_addr *addr)
339 {
340 	struct ifacaddr6 *aca, *prev_aca;
341 
342 	ASSERT_RTNL();
343 
344 	write_lock_bh(&idev->lock);
345 	prev_aca = NULL;
346 	for (aca = idev->ac_list; aca; aca = aca->aca_next) {
347 		if (ipv6_addr_equal(&aca->aca_addr, addr))
348 			break;
349 		prev_aca = aca;
350 	}
351 	if (!aca) {
352 		write_unlock_bh(&idev->lock);
353 		return -ENOENT;
354 	}
355 	if (--aca->aca_users > 0) {
356 		write_unlock_bh(&idev->lock);
357 		return 0;
358 	}
359 	if (prev_aca)
360 		prev_aca->aca_next = aca->aca_next;
361 	else
362 		idev->ac_list = aca->aca_next;
363 	write_unlock_bh(&idev->lock);
364 	ipv6_del_acaddr_hash(aca);
365 	addrconf_leave_solict(idev, &aca->aca_addr);
366 
367 	ip6_del_rt(dev_net(idev->dev), aca->aca_rt, false);
368 
369 	aca_put(aca);
370 	return 0;
371 }
372 
373 /* called with rtnl_lock() */
374 static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr)
375 {
376 	struct inet6_dev *idev = __in6_dev_get(dev);
377 
378 	if (!idev)
379 		return -ENODEV;
380 	return __ipv6_dev_ac_dec(idev, addr);
381 }
382 
383 void ipv6_ac_destroy_dev(struct inet6_dev *idev)
384 {
385 	struct ifacaddr6 *aca;
386 
387 	write_lock_bh(&idev->lock);
388 	while ((aca = idev->ac_list) != NULL) {
389 		idev->ac_list = aca->aca_next;
390 		write_unlock_bh(&idev->lock);
391 
392 		ipv6_del_acaddr_hash(aca);
393 
394 		addrconf_leave_solict(idev, &aca->aca_addr);
395 
396 		ip6_del_rt(dev_net(idev->dev), aca->aca_rt, false);
397 
398 		aca_put(aca);
399 
400 		write_lock_bh(&idev->lock);
401 	}
402 	write_unlock_bh(&idev->lock);
403 }
404 
405 /*
406  *	check if the interface has this anycast address
407  *	called with rcu_read_lock()
408  */
409 static bool ipv6_chk_acast_dev(struct net_device *dev, const struct in6_addr *addr)
410 {
411 	struct inet6_dev *idev;
412 	struct ifacaddr6 *aca;
413 
414 	idev = __in6_dev_get(dev);
415 	if (idev) {
416 		read_lock_bh(&idev->lock);
417 		for (aca = idev->ac_list; aca; aca = aca->aca_next)
418 			if (ipv6_addr_equal(&aca->aca_addr, addr))
419 				break;
420 		read_unlock_bh(&idev->lock);
421 		return aca != NULL;
422 	}
423 	return false;
424 }
425 
426 /*
427  *	check if given interface (or any, if dev==0) has this anycast address
428  */
429 bool ipv6_chk_acast_addr(struct net *net, struct net_device *dev,
430 			 const struct in6_addr *addr)
431 {
432 	struct net_device *nh_dev;
433 	struct ifacaddr6 *aca;
434 	bool found = false;
435 
436 	rcu_read_lock();
437 	if (dev)
438 		found = ipv6_chk_acast_dev(dev, addr);
439 	else {
440 		unsigned int hash = inet6_acaddr_hash(net, addr);
441 
442 		hlist_for_each_entry_rcu(aca, &inet6_acaddr_lst[hash],
443 					 aca_addr_lst) {
444 			nh_dev = fib6_info_nh_dev(aca->aca_rt);
445 			if (!nh_dev || !net_eq(dev_net(nh_dev), net))
446 				continue;
447 			if (ipv6_addr_equal(&aca->aca_addr, addr)) {
448 				found = true;
449 				break;
450 			}
451 		}
452 	}
453 	rcu_read_unlock();
454 	return found;
455 }
456 
457 /*	check if this anycast address is link-local on given interface or
458  *	is global
459  */
460 bool ipv6_chk_acast_addr_src(struct net *net, struct net_device *dev,
461 			     const struct in6_addr *addr)
462 {
463 	return ipv6_chk_acast_addr(net,
464 				   (ipv6_addr_type(addr) & IPV6_ADDR_LINKLOCAL ?
465 				    dev : NULL),
466 				   addr);
467 }
468 
469 #ifdef CONFIG_PROC_FS
470 struct ac6_iter_state {
471 	struct seq_net_private p;
472 	struct net_device *dev;
473 	struct inet6_dev *idev;
474 };
475 
476 #define ac6_seq_private(seq)	((struct ac6_iter_state *)(seq)->private)
477 
478 static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq)
479 {
480 	struct ifacaddr6 *im = NULL;
481 	struct ac6_iter_state *state = ac6_seq_private(seq);
482 	struct net *net = seq_file_net(seq);
483 
484 	state->idev = NULL;
485 	for_each_netdev_rcu(net, state->dev) {
486 		struct inet6_dev *idev;
487 		idev = __in6_dev_get(state->dev);
488 		if (!idev)
489 			continue;
490 		read_lock_bh(&idev->lock);
491 		im = idev->ac_list;
492 		if (im) {
493 			state->idev = idev;
494 			break;
495 		}
496 		read_unlock_bh(&idev->lock);
497 	}
498 	return im;
499 }
500 
501 static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im)
502 {
503 	struct ac6_iter_state *state = ac6_seq_private(seq);
504 
505 	im = im->aca_next;
506 	while (!im) {
507 		if (likely(state->idev != NULL))
508 			read_unlock_bh(&state->idev->lock);
509 
510 		state->dev = next_net_device_rcu(state->dev);
511 		if (!state->dev) {
512 			state->idev = NULL;
513 			break;
514 		}
515 		state->idev = __in6_dev_get(state->dev);
516 		if (!state->idev)
517 			continue;
518 		read_lock_bh(&state->idev->lock);
519 		im = state->idev->ac_list;
520 	}
521 	return im;
522 }
523 
524 static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos)
525 {
526 	struct ifacaddr6 *im = ac6_get_first(seq);
527 	if (im)
528 		while (pos && (im = ac6_get_next(seq, im)) != NULL)
529 			--pos;
530 	return pos ? NULL : im;
531 }
532 
533 static void *ac6_seq_start(struct seq_file *seq, loff_t *pos)
534 	__acquires(RCU)
535 {
536 	rcu_read_lock();
537 	return ac6_get_idx(seq, *pos);
538 }
539 
540 static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
541 {
542 	struct ifacaddr6 *im = ac6_get_next(seq, v);
543 
544 	++*pos;
545 	return im;
546 }
547 
548 static void ac6_seq_stop(struct seq_file *seq, void *v)
549 	__releases(RCU)
550 {
551 	struct ac6_iter_state *state = ac6_seq_private(seq);
552 
553 	if (likely(state->idev != NULL)) {
554 		read_unlock_bh(&state->idev->lock);
555 		state->idev = NULL;
556 	}
557 	rcu_read_unlock();
558 }
559 
560 static int ac6_seq_show(struct seq_file *seq, void *v)
561 {
562 	struct ifacaddr6 *im = (struct ifacaddr6 *)v;
563 	struct ac6_iter_state *state = ac6_seq_private(seq);
564 
565 	seq_printf(seq, "%-4d %-15s %pi6 %5d\n",
566 		   state->dev->ifindex, state->dev->name,
567 		   &im->aca_addr, im->aca_users);
568 	return 0;
569 }
570 
571 static const struct seq_operations ac6_seq_ops = {
572 	.start	=	ac6_seq_start,
573 	.next	=	ac6_seq_next,
574 	.stop	=	ac6_seq_stop,
575 	.show	=	ac6_seq_show,
576 };
577 
578 int __net_init ac6_proc_init(struct net *net)
579 {
580 	if (!proc_create_net("anycast6", 0444, net->proc_net, &ac6_seq_ops,
581 			sizeof(struct ac6_iter_state)))
582 		return -ENOMEM;
583 
584 	return 0;
585 }
586 
587 void ac6_proc_exit(struct net *net)
588 {
589 	remove_proc_entry("anycast6", net->proc_net);
590 }
591 #endif
592 
593 /*	Init / cleanup code
594  */
595 int __init ipv6_anycast_init(void)
596 {
597 	int i;
598 
599 	for (i = 0; i < IN6_ADDR_HSIZE; i++)
600 		INIT_HLIST_HEAD(&inet6_acaddr_lst[i]);
601 	return 0;
602 }
603 
604 void ipv6_anycast_cleanup(void)
605 {
606 	int i;
607 
608 	spin_lock(&acaddr_hash_lock);
609 	for (i = 0; i < IN6_ADDR_HSIZE; i++)
610 		WARN_ON(!hlist_empty(&inet6_acaddr_lst[i]));
611 	spin_unlock(&acaddr_hash_lock);
612 }
613