xref: /openbmc/linux/net/ipv6/anycast.c (revision ae213c44) 
1 /*
2  *	Anycast support for IPv6
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	David L Stevens (dlstevens@us.ibm.com)
7  *
8  *	based heavily on net/ipv6/mcast.c
9  *
10  *	This program is free software; you can redistribute it and/or
11  *      modify it under the terms of the GNU General Public License
12  *      as published by the Free Software Foundation; either version
13  *      2 of the License, or (at your option) any later version.
14  */
15 
16 #include <linux/capability.h>
17 #include <linux/module.h>
18 #include <linux/errno.h>
19 #include <linux/types.h>
20 #include <linux/random.h>
21 #include <linux/string.h>
22 #include <linux/socket.h>
23 #include <linux/sockios.h>
24 #include <linux/net.h>
25 #include <linux/in6.h>
26 #include <linux/netdevice.h>
27 #include <linux/if_arp.h>
28 #include <linux/route.h>
29 #include <linux/init.h>
30 #include <linux/proc_fs.h>
31 #include <linux/seq_file.h>
32 #include <linux/slab.h>
33 
34 #include <net/net_namespace.h>
35 #include <net/sock.h>
36 #include <net/snmp.h>
37 
38 #include <net/ipv6.h>
39 #include <net/protocol.h>
40 #include <net/if_inet6.h>
41 #include <net/ndisc.h>
42 #include <net/addrconf.h>
43 #include <net/ip6_route.h>
44 
45 #include <net/checksum.h>
46 
47 #define IN6_ADDR_HSIZE_SHIFT	8
48 #define IN6_ADDR_HSIZE		BIT(IN6_ADDR_HSIZE_SHIFT)
49 /*	anycast address hash table
50  */
51 static struct hlist_head inet6_acaddr_lst[IN6_ADDR_HSIZE];
52 static DEFINE_SPINLOCK(acaddr_hash_lock);
53 
54 static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr);
55 
56 static u32 inet6_acaddr_hash(struct net *net, const struct in6_addr *addr)
57 {
58 	u32 val = ipv6_addr_hash(addr) ^ net_hash_mix(net);
59 
60 	return hash_32(val, IN6_ADDR_HSIZE_SHIFT);
61 }
62 
63 /*
64  *	socket join an anycast group
65  */
66 
67 int ipv6_sock_ac_join(struct sock *sk, int ifindex, const struct in6_addr *addr)
68 {
69 	struct ipv6_pinfo *np = inet6_sk(sk);
70 	struct net_device *dev = NULL;
71 	struct inet6_dev *idev;
72 	struct ipv6_ac_socklist *pac;
73 	struct net *net = sock_net(sk);
74 	int	ishost = !net->ipv6.devconf_all->forwarding;
75 	int	err = 0;
76 
77 	ASSERT_RTNL();
78 
79 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
80 		return -EPERM;
81 	if (ipv6_addr_is_multicast(addr))
82 		return -EINVAL;
83 
84 	if (ifindex)
85 		dev = __dev_get_by_index(net, ifindex);
86 
87 	if (ipv6_chk_addr_and_flags(net, addr, dev, true, 0, IFA_F_TENTATIVE))
88 		return -EINVAL;
89 
90 	pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL);
91 	if (!pac)
92 		return -ENOMEM;
93 	pac->acl_next = NULL;
94 	pac->acl_addr = *addr;
95 
96 	if (ifindex == 0) {
97 		struct rt6_info *rt;
98 
99 		rt = rt6_lookup(net, addr, NULL, 0, NULL, 0);
100 		if (rt) {
101 			dev = rt->dst.dev;
102 			ip6_rt_put(rt);
103 		} else if (ishost) {
104 			err = -EADDRNOTAVAIL;
105 			goto error;
106 		} else {
107 			/* router, no matching interface: just pick one */
108 			dev = __dev_get_by_flags(net, IFF_UP,
109 						 IFF_UP | IFF_LOOPBACK);
110 		}
111 	}
112 
113 	if (!dev) {
114 		err = -ENODEV;
115 		goto error;
116 	}
117 
118 	idev = __in6_dev_get(dev);
119 	if (!idev) {
120 		if (ifindex)
121 			err = -ENODEV;
122 		else
123 			err = -EADDRNOTAVAIL;
124 		goto error;
125 	}
126 	/* reset ishost, now that we have a specific device */
127 	ishost = !idev->cnf.forwarding;
128 
129 	pac->acl_ifindex = dev->ifindex;
130 
131 	/* XXX
132 	 * For hosts, allow link-local or matching prefix anycasts.
133 	 * This obviates the need for propagating anycast routes while
134 	 * still allowing some non-router anycast participation.
135 	 */
136 	if (!ipv6_chk_prefix(addr, dev)) {
137 		if (ishost)
138 			err = -EADDRNOTAVAIL;
139 		if (err)
140 			goto error;
141 	}
142 
143 	err = __ipv6_dev_ac_inc(idev, addr);
144 	if (!err) {
145 		pac->acl_next = np->ipv6_ac_list;
146 		np->ipv6_ac_list = pac;
147 		pac = NULL;
148 	}
149 
150 error:
151 	if (pac)
152 		sock_kfree_s(sk, pac, sizeof(*pac));
153 	return err;
154 }
155 
156 /*
157  *	socket leave an anycast group
158  */
159 int ipv6_sock_ac_drop(struct sock *sk, int ifindex, const struct in6_addr *addr)
160 {
161 	struct ipv6_pinfo *np = inet6_sk(sk);
162 	struct net_device *dev;
163 	struct ipv6_ac_socklist *pac, *prev_pac;
164 	struct net *net = sock_net(sk);
165 
166 	ASSERT_RTNL();
167 
168 	prev_pac = NULL;
169 	for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) {
170 		if ((ifindex == 0 || pac->acl_ifindex == ifindex) &&
171 		     ipv6_addr_equal(&pac->acl_addr, addr))
172 			break;
173 		prev_pac = pac;
174 	}
175 	if (!pac)
176 		return -ENOENT;
177 	if (prev_pac)
178 		prev_pac->acl_next = pac->acl_next;
179 	else
180 		np->ipv6_ac_list = pac->acl_next;
181 
182 	dev = __dev_get_by_index(net, pac->acl_ifindex);
183 	if (dev)
184 		ipv6_dev_ac_dec(dev, &pac->acl_addr);
185 
186 	sock_kfree_s(sk, pac, sizeof(*pac));
187 	return 0;
188 }
189 
190 void ipv6_sock_ac_close(struct sock *sk)
191 {
192 	struct ipv6_pinfo *np = inet6_sk(sk);
193 	struct net_device *dev = NULL;
194 	struct ipv6_ac_socklist *pac;
195 	struct net *net = sock_net(sk);
196 	int	prev_index;
197 
198 	if (!np->ipv6_ac_list)
199 		return;
200 
201 	rtnl_lock();
202 	pac = np->ipv6_ac_list;
203 	np->ipv6_ac_list = NULL;
204 
205 	prev_index = 0;
206 	while (pac) {
207 		struct ipv6_ac_socklist *next = pac->acl_next;
208 
209 		if (pac->acl_ifindex != prev_index) {
210 			dev = __dev_get_by_index(net, pac->acl_ifindex);
211 			prev_index = pac->acl_ifindex;
212 		}
213 		if (dev)
214 			ipv6_dev_ac_dec(dev, &pac->acl_addr);
215 		sock_kfree_s(sk, pac, sizeof(*pac));
216 		pac = next;
217 	}
218 	rtnl_unlock();
219 }
220 
221 static void ipv6_add_acaddr_hash(struct net *net, struct ifacaddr6 *aca)
222 {
223 	unsigned int hash = inet6_acaddr_hash(net, &aca->aca_addr);
224 
225 	spin_lock(&acaddr_hash_lock);
226 	hlist_add_head_rcu(&aca->aca_addr_lst, &inet6_acaddr_lst[hash]);
227 	spin_unlock(&acaddr_hash_lock);
228 }
229 
230 static void ipv6_del_acaddr_hash(struct ifacaddr6 *aca)
231 {
232 	spin_lock(&acaddr_hash_lock);
233 	hlist_del_init_rcu(&aca->aca_addr_lst);
234 	spin_unlock(&acaddr_hash_lock);
235 }
236 
237 static void aca_get(struct ifacaddr6 *aca)
238 {
239 	refcount_inc(&aca->aca_refcnt);
240 }
241 
242 static void aca_free_rcu(struct rcu_head *h)
243 {
244 	struct ifacaddr6 *aca = container_of(h, struct ifacaddr6, rcu);
245 
246 	fib6_info_release(aca->aca_rt);
247 	kfree(aca);
248 }
249 
250 static void aca_put(struct ifacaddr6 *ac)
251 {
252 	if (refcount_dec_and_test(&ac->aca_refcnt)) {
253 		call_rcu(&ac->rcu, aca_free_rcu);
254 	}
255 }
256 
257 static struct ifacaddr6 *aca_alloc(struct fib6_info *f6i,
258 				   const struct in6_addr *addr)
259 {
260 	struct ifacaddr6 *aca;
261 
262 	aca = kzalloc(sizeof(*aca), GFP_ATOMIC);
263 	if (!aca)
264 		return NULL;
265 
266 	aca->aca_addr = *addr;
267 	fib6_info_hold(f6i);
268 	aca->aca_rt = f6i;
269 	INIT_HLIST_NODE(&aca->aca_addr_lst);
270 	aca->aca_users = 1;
271 	/* aca_tstamp should be updated upon changes */
272 	aca->aca_cstamp = aca->aca_tstamp = jiffies;
273 	refcount_set(&aca->aca_refcnt, 1);
274 
275 	return aca;
276 }
277 
278 /*
279  *	device anycast group inc (add if not found)
280  */
281 int __ipv6_dev_ac_inc(struct inet6_dev *idev, const struct in6_addr *addr)
282 {
283 	struct ifacaddr6 *aca;
284 	struct fib6_info *f6i;
285 	struct net *net;
286 	int err;
287 
288 	ASSERT_RTNL();
289 
290 	write_lock_bh(&idev->lock);
291 	if (idev->dead) {
292 		err = -ENODEV;
293 		goto out;
294 	}
295 
296 	for (aca = idev->ac_list; aca; aca = aca->aca_next) {
297 		if (ipv6_addr_equal(&aca->aca_addr, addr)) {
298 			aca->aca_users++;
299 			err = 0;
300 			goto out;
301 		}
302 	}
303 
304 	net = dev_net(idev->dev);
305 	f6i = addrconf_f6i_alloc(net, idev, addr, true, GFP_ATOMIC);
306 	if (IS_ERR(f6i)) {
307 		err = PTR_ERR(f6i);
308 		goto out;
309 	}
310 	aca = aca_alloc(f6i, addr);
311 	if (!aca) {
312 		fib6_info_release(f6i);
313 		err = -ENOMEM;
314 		goto out;
315 	}
316 
317 	aca->aca_next = idev->ac_list;
318 	idev->ac_list = aca;
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 	write_unlock_bh(&idev->lock);
325 
326 	ipv6_add_acaddr_hash(net, aca);
327 
328 	ip6_ins_rt(net, f6i);
329 
330 	addrconf_join_solict(idev->dev, &aca->aca_addr);
331 
332 	aca_put(aca);
333 	return 0;
334 out:
335 	write_unlock_bh(&idev->lock);
336 	return err;
337 }
338 
339 /*
340  *	device anycast group decrement
341  */
342 int __ipv6_dev_ac_dec(struct inet6_dev *idev, const struct in6_addr *addr)
343 {
344 	struct ifacaddr6 *aca, *prev_aca;
345 
346 	ASSERT_RTNL();
347 
348 	write_lock_bh(&idev->lock);
349 	prev_aca = NULL;
350 	for (aca = idev->ac_list; aca; aca = aca->aca_next) {
351 		if (ipv6_addr_equal(&aca->aca_addr, addr))
352 			break;
353 		prev_aca = aca;
354 	}
355 	if (!aca) {
356 		write_unlock_bh(&idev->lock);
357 		return -ENOENT;
358 	}
359 	if (--aca->aca_users > 0) {
360 		write_unlock_bh(&idev->lock);
361 		return 0;
362 	}
363 	if (prev_aca)
364 		prev_aca->aca_next = aca->aca_next;
365 	else
366 		idev->ac_list = aca->aca_next;
367 	write_unlock_bh(&idev->lock);
368 	ipv6_del_acaddr_hash(aca);
369 	addrconf_leave_solict(idev, &aca->aca_addr);
370 
371 	ip6_del_rt(dev_net(idev->dev), aca->aca_rt);
372 
373 	aca_put(aca);
374 	return 0;
375 }
376 
377 /* called with rtnl_lock() */
378 static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr)
379 {
380 	struct inet6_dev *idev = __in6_dev_get(dev);
381 
382 	if (!idev)
383 		return -ENODEV;
384 	return __ipv6_dev_ac_dec(idev, addr);
385 }
386 
387 void ipv6_ac_destroy_dev(struct inet6_dev *idev)
388 {
389 	struct ifacaddr6 *aca;
390 
391 	write_lock_bh(&idev->lock);
392 	while ((aca = idev->ac_list) != NULL) {
393 		idev->ac_list = aca->aca_next;
394 		write_unlock_bh(&idev->lock);
395 
396 		ipv6_del_acaddr_hash(aca);
397 
398 		addrconf_leave_solict(idev, &aca->aca_addr);
399 
400 		ip6_del_rt(dev_net(idev->dev), aca->aca_rt);
401 
402 		aca_put(aca);
403 
404 		write_lock_bh(&idev->lock);
405 	}
406 	write_unlock_bh(&idev->lock);
407 }
408 
409 /*
410  *	check if the interface has this anycast address
411  *	called with rcu_read_lock()
412  */
413 static bool ipv6_chk_acast_dev(struct net_device *dev, const struct in6_addr *addr)
414 {
415 	struct inet6_dev *idev;
416 	struct ifacaddr6 *aca;
417 
418 	idev = __in6_dev_get(dev);
419 	if (idev) {
420 		read_lock_bh(&idev->lock);
421 		for (aca = idev->ac_list; aca; aca = aca->aca_next)
422 			if (ipv6_addr_equal(&aca->aca_addr, addr))
423 				break;
424 		read_unlock_bh(&idev->lock);
425 		return aca != NULL;
426 	}
427 	return false;
428 }
429 
430 /*
431  *	check if given interface (or any, if dev==0) has this anycast address
432  */
433 bool ipv6_chk_acast_addr(struct net *net, struct net_device *dev,
434 			 const struct in6_addr *addr)
435 {
436 	struct net_device *nh_dev;
437 	struct ifacaddr6 *aca;
438 	bool found = false;
439 
440 	rcu_read_lock();
441 	if (dev)
442 		found = ipv6_chk_acast_dev(dev, addr);
443 	else {
444 		unsigned int hash = inet6_acaddr_hash(net, addr);
445 
446 		hlist_for_each_entry_rcu(aca, &inet6_acaddr_lst[hash],
447 					 aca_addr_lst) {
448 			nh_dev = fib6_info_nh_dev(aca->aca_rt);
449 			if (!nh_dev || !net_eq(dev_net(nh_dev), net))
450 				continue;
451 			if (ipv6_addr_equal(&aca->aca_addr, addr)) {
452 				found = true;
453 				break;
454 			}
455 		}
456 	}
457 	rcu_read_unlock();
458 	return found;
459 }
460 
461 /*	check if this anycast address is link-local on given interface or
462  *	is global
463  */
464 bool ipv6_chk_acast_addr_src(struct net *net, struct net_device *dev,
465 			     const struct in6_addr *addr)
466 {
467 	return ipv6_chk_acast_addr(net,
468 				   (ipv6_addr_type(addr) & IPV6_ADDR_LINKLOCAL ?
469 				    dev : NULL),
470 				   addr);
471 }
472 
473 #ifdef CONFIG_PROC_FS
474 struct ac6_iter_state {
475 	struct seq_net_private p;
476 	struct net_device *dev;
477 	struct inet6_dev *idev;
478 };
479 
480 #define ac6_seq_private(seq)	((struct ac6_iter_state *)(seq)->private)
481 
482 static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq)
483 {
484 	struct ifacaddr6 *im = NULL;
485 	struct ac6_iter_state *state = ac6_seq_private(seq);
486 	struct net *net = seq_file_net(seq);
487 
488 	state->idev = NULL;
489 	for_each_netdev_rcu(net, state->dev) {
490 		struct inet6_dev *idev;
491 		idev = __in6_dev_get(state->dev);
492 		if (!idev)
493 			continue;
494 		read_lock_bh(&idev->lock);
495 		im = idev->ac_list;
496 		if (im) {
497 			state->idev = idev;
498 			break;
499 		}
500 		read_unlock_bh(&idev->lock);
501 	}
502 	return im;
503 }
504 
505 static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im)
506 {
507 	struct ac6_iter_state *state = ac6_seq_private(seq);
508 
509 	im = im->aca_next;
510 	while (!im) {
511 		if (likely(state->idev != NULL))
512 			read_unlock_bh(&state->idev->lock);
513 
514 		state->dev = next_net_device_rcu(state->dev);
515 		if (!state->dev) {
516 			state->idev = NULL;
517 			break;
518 		}
519 		state->idev = __in6_dev_get(state->dev);
520 		if (!state->idev)
521 			continue;
522 		read_lock_bh(&state->idev->lock);
523 		im = state->idev->ac_list;
524 	}
525 	return im;
526 }
527 
528 static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos)
529 {
530 	struct ifacaddr6 *im = ac6_get_first(seq);
531 	if (im)
532 		while (pos && (im = ac6_get_next(seq, im)) != NULL)
533 			--pos;
534 	return pos ? NULL : im;
535 }
536 
537 static void *ac6_seq_start(struct seq_file *seq, loff_t *pos)
538 	__acquires(RCU)
539 {
540 	rcu_read_lock();
541 	return ac6_get_idx(seq, *pos);
542 }
543 
544 static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
545 {
546 	struct ifacaddr6 *im = ac6_get_next(seq, v);
547 
548 	++*pos;
549 	return im;
550 }
551 
552 static void ac6_seq_stop(struct seq_file *seq, void *v)
553 	__releases(RCU)
554 {
555 	struct ac6_iter_state *state = ac6_seq_private(seq);
556 
557 	if (likely(state->idev != NULL)) {
558 		read_unlock_bh(&state->idev->lock);
559 		state->idev = NULL;
560 	}
561 	rcu_read_unlock();
562 }
563 
564 static int ac6_seq_show(struct seq_file *seq, void *v)
565 {
566 	struct ifacaddr6 *im = (struct ifacaddr6 *)v;
567 	struct ac6_iter_state *state = ac6_seq_private(seq);
568 
569 	seq_printf(seq, "%-4d %-15s %pi6 %5d\n",
570 		   state->dev->ifindex, state->dev->name,
571 		   &im->aca_addr, im->aca_users);
572 	return 0;
573 }
574 
575 static const struct seq_operations ac6_seq_ops = {
576 	.start	=	ac6_seq_start,
577 	.next	=	ac6_seq_next,
578 	.stop	=	ac6_seq_stop,
579 	.show	=	ac6_seq_show,
580 };
581 
582 int __net_init ac6_proc_init(struct net *net)
583 {
584 	if (!proc_create_net("anycast6", 0444, net->proc_net, &ac6_seq_ops,
585 			sizeof(struct ac6_iter_state)))
586 		return -ENOMEM;
587 
588 	return 0;
589 }
590 
591 void ac6_proc_exit(struct net *net)
592 {
593 	remove_proc_entry("anycast6", net->proc_net);
594 }
595 #endif
596 
597 /*	Init / cleanup code
598  */
599 int __init ipv6_anycast_init(void)
600 {
601 	int i;
602 
603 	for (i = 0; i < IN6_ADDR_HSIZE; i++)
604 		INIT_HLIST_HEAD(&inet6_acaddr_lst[i]);
605 	return 0;
606 }
607 
608 void ipv6_anycast_cleanup(void)
609 {
610 	int i;
611 
612 	spin_lock(&acaddr_hash_lock);
613 	for (i = 0; i < IN6_ADDR_HSIZE; i++)
614 		WARN_ON(!hlist_empty(&inet6_acaddr_lst[i]));
615 	spin_unlock(&acaddr_hash_lock);
616 }
617