anycast.c revision f79d52c254e4e2cef3da64dc02ade3bc8f10c539 
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
47static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr);
48
49/* Big ac list lock for all the sockets */
50static DEFINE_RWLOCK(ipv6_sk_ac_lock);
51
52
53/*
54 *	socket join an anycast group
55 */
56
57int ipv6_sock_ac_join(struct sock *sk, int ifindex, const struct in6_addr *addr)
58{
59	struct ipv6_pinfo *np = inet6_sk(sk);
60	struct net_device *dev = NULL;
61	struct inet6_dev *idev;
62	struct ipv6_ac_socklist *pac;
63	struct net *net = sock_net(sk);
64	int	ishost = !net->ipv6.devconf_all->forwarding;
65	int	err = 0;
66
67	if (!capable(CAP_NET_ADMIN))
68		return -EPERM;
69	if (ipv6_addr_is_multicast(addr))
70		return -EINVAL;
71	if (ipv6_chk_addr(net, addr, NULL, 0))
72		return -EINVAL;
73
74	pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL);
75	if (pac == NULL)
76		return -ENOMEM;
77	pac->acl_next = NULL;
78	pac->acl_addr = *addr;
79
80	rcu_read_lock();
81	if (ifindex == 0) {
82		struct rt6_info *rt;
83
84		rt = rt6_lookup(net, addr, NULL, 0, 0);
85		if (rt) {
86			dev = rt->dst.dev;
87			dst_release(&rt->dst);
88		} else if (ishost) {
89			err = -EADDRNOTAVAIL;
90			goto error;
91		} else {
92			/* router, no matching interface: just pick one */
93			dev = dev_get_by_flags_rcu(net, IFF_UP,
94						   IFF_UP | IFF_LOOPBACK);
95		}
96	} else
97		dev = dev_get_by_index_rcu(net, ifindex);
98
99	if (dev == NULL) {
100		err = -ENODEV;
101		goto error;
102	}
103
104	idev = __in6_dev_get(dev);
105	if (!idev) {
106		if (ifindex)
107			err = -ENODEV;
108		else
109			err = -EADDRNOTAVAIL;
110		goto error;
111	}
112	/* reset ishost, now that we have a specific device */
113	ishost = !idev->cnf.forwarding;
114
115	pac->acl_ifindex = dev->ifindex;
116
117	/* XXX
118	 * For hosts, allow link-local or matching prefix anycasts.
119	 * This obviates the need for propagating anycast routes while
120	 * still allowing some non-router anycast participation.
121	 */
122	if (!ipv6_chk_prefix(addr, dev)) {
123		if (ishost)
124			err = -EADDRNOTAVAIL;
125		if (err)
126			goto error;
127	}
128
129	err = ipv6_dev_ac_inc(dev, addr);
130	if (!err) {
131		write_lock_bh(&ipv6_sk_ac_lock);
132		pac->acl_next = np->ipv6_ac_list;
133		np->ipv6_ac_list = pac;
134		write_unlock_bh(&ipv6_sk_ac_lock);
135		pac = NULL;
136	}
137
138error:
139	rcu_read_unlock();
140	if (pac)
141		sock_kfree_s(sk, pac, sizeof(*pac));
142	return err;
143}
144
145/*
146 *	socket leave an anycast group
147 */
148int ipv6_sock_ac_drop(struct sock *sk, int ifindex, const struct in6_addr *addr)
149{
150	struct ipv6_pinfo *np = inet6_sk(sk);
151	struct net_device *dev;
152	struct ipv6_ac_socklist *pac, *prev_pac;
153	struct net *net = sock_net(sk);
154
155	write_lock_bh(&ipv6_sk_ac_lock);
156	prev_pac = NULL;
157	for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) {
158		if ((ifindex == 0 || pac->acl_ifindex == ifindex) &&
159		     ipv6_addr_equal(&pac->acl_addr, addr))
160			break;
161		prev_pac = pac;
162	}
163	if (!pac) {
164		write_unlock_bh(&ipv6_sk_ac_lock);
165		return -ENOENT;
166	}
167	if (prev_pac)
168		prev_pac->acl_next = pac->acl_next;
169	else
170		np->ipv6_ac_list = pac->acl_next;
171
172	write_unlock_bh(&ipv6_sk_ac_lock);
173
174	rcu_read_lock();
175	dev = dev_get_by_index_rcu(net, pac->acl_ifindex);
176	if (dev)
177		ipv6_dev_ac_dec(dev, &pac->acl_addr);
178	rcu_read_unlock();
179
180	sock_kfree_s(sk, pac, sizeof(*pac));
181	return 0;
182}
183
184void ipv6_sock_ac_close(struct sock *sk)
185{
186	struct ipv6_pinfo *np = inet6_sk(sk);
187	struct net_device *dev = NULL;
188	struct ipv6_ac_socklist *pac;
189	struct net *net = sock_net(sk);
190	int	prev_index;
191
192	write_lock_bh(&ipv6_sk_ac_lock);
193	pac = np->ipv6_ac_list;
194	np->ipv6_ac_list = NULL;
195	write_unlock_bh(&ipv6_sk_ac_lock);
196
197	prev_index = 0;
198	rcu_read_lock();
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_rcu(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	rcu_read_unlock();
212}
213
214static void aca_put(struct ifacaddr6 *ac)
215{
216	if (atomic_dec_and_test(&ac->aca_refcnt)) {
217		in6_dev_put(ac->aca_idev);
218		dst_release(&ac->aca_rt->dst);
219		kfree(ac);
220	}
221}
222
223/*
224 *	device anycast group inc (add if not found)
225 */
226int ipv6_dev_ac_inc(struct net_device *dev, const struct in6_addr *addr)
227{
228	struct ifacaddr6 *aca;
229	struct inet6_dev *idev;
230	struct rt6_info *rt;
231	int err;
232
233	idev = in6_dev_get(dev);
234
235	if (idev == NULL)
236		return -EINVAL;
237
238	write_lock_bh(&idev->lock);
239	if (idev->dead) {
240		err = -ENODEV;
241		goto out;
242	}
243
244	for (aca = idev->ac_list; aca; aca = aca->aca_next) {
245		if (ipv6_addr_equal(&aca->aca_addr, addr)) {
246			aca->aca_users++;
247			err = 0;
248			goto out;
249		}
250	}
251
252	/*
253	 *	not found: create a new one.
254	 */
255
256	aca = kzalloc(sizeof(struct ifacaddr6), GFP_ATOMIC);
257
258	if (aca == NULL) {
259		err = -ENOMEM;
260		goto out;
261	}
262
263	rt = addrconf_dst_alloc(idev, addr, true);
264	if (IS_ERR(rt)) {
265		kfree(aca);
266		err = PTR_ERR(rt);
267		goto out;
268	}
269
270	aca->aca_addr = *addr;
271	aca->aca_idev = idev;
272	aca->aca_rt = rt;
273	aca->aca_users = 1;
274	/* aca_tstamp should be updated upon changes */
275	aca->aca_cstamp = aca->aca_tstamp = jiffies;
276	atomic_set(&aca->aca_refcnt, 2);
277	spin_lock_init(&aca->aca_lock);
278
279	aca->aca_next = idev->ac_list;
280	idev->ac_list = aca;
281	write_unlock_bh(&idev->lock);
282
283	ip6_ins_rt(rt);
284
285	addrconf_join_solict(dev, &aca->aca_addr);
286
287	aca_put(aca);
288	return 0;
289out:
290	write_unlock_bh(&idev->lock);
291	in6_dev_put(idev);
292	return err;
293}
294
295/*
296 *	device anycast group decrement
297 */
298int __ipv6_dev_ac_dec(struct inet6_dev *idev, const struct in6_addr *addr)
299{
300	struct ifacaddr6 *aca, *prev_aca;
301
302	write_lock_bh(&idev->lock);
303	prev_aca = NULL;
304	for (aca = idev->ac_list; aca; aca = aca->aca_next) {
305		if (ipv6_addr_equal(&aca->aca_addr, addr))
306			break;
307		prev_aca = aca;
308	}
309	if (!aca) {
310		write_unlock_bh(&idev->lock);
311		return -ENOENT;
312	}
313	if (--aca->aca_users > 0) {
314		write_unlock_bh(&idev->lock);
315		return 0;
316	}
317	if (prev_aca)
318		prev_aca->aca_next = aca->aca_next;
319	else
320		idev->ac_list = aca->aca_next;
321	write_unlock_bh(&idev->lock);
322	addrconf_leave_solict(idev, &aca->aca_addr);
323
324	dst_hold(&aca->aca_rt->dst);
325	ip6_del_rt(aca->aca_rt);
326
327	aca_put(aca);
328	return 0;
329}
330
331/* called with rcu_read_lock() */
332static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr)
333{
334	struct inet6_dev *idev = __in6_dev_get(dev);
335
336	if (idev == NULL)
337		return -ENODEV;
338	return __ipv6_dev_ac_dec(idev, addr);
339}
340
341/*
342 *	check if the interface has this anycast address
343 *	called with rcu_read_lock()
344 */
345static int ipv6_chk_acast_dev(struct net_device *dev, const struct in6_addr *addr)
346{
347	struct inet6_dev *idev;
348	struct ifacaddr6 *aca;
349
350	idev = __in6_dev_get(dev);
351	if (idev) {
352		read_lock_bh(&idev->lock);
353		for (aca = idev->ac_list; aca; aca = aca->aca_next)
354			if (ipv6_addr_equal(&aca->aca_addr, addr))
355				break;
356		read_unlock_bh(&idev->lock);
357		return aca != NULL;
358	}
359	return 0;
360}
361
362/*
363 *	check if given interface (or any, if dev==0) has this anycast address
364 */
365int ipv6_chk_acast_addr(struct net *net, struct net_device *dev,
366			const struct in6_addr *addr)
367{
368	int found = 0;
369
370	rcu_read_lock();
371	if (dev)
372		found = ipv6_chk_acast_dev(dev, addr);
373	else
374		for_each_netdev_rcu(net, dev)
375			if (ipv6_chk_acast_dev(dev, addr)) {
376				found = 1;
377				break;
378			}
379	rcu_read_unlock();
380	return found;
381}
382
383
384#ifdef CONFIG_PROC_FS
385struct ac6_iter_state {
386	struct seq_net_private p;
387	struct net_device *dev;
388	struct inet6_dev *idev;
389};
390
391#define ac6_seq_private(seq)	((struct ac6_iter_state *)(seq)->private)
392
393static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq)
394{
395	struct ifacaddr6 *im = NULL;
396	struct ac6_iter_state *state = ac6_seq_private(seq);
397	struct net *net = seq_file_net(seq);
398
399	state->idev = NULL;
400	for_each_netdev_rcu(net, state->dev) {
401		struct inet6_dev *idev;
402		idev = __in6_dev_get(state->dev);
403		if (!idev)
404			continue;
405		read_lock_bh(&idev->lock);
406		im = idev->ac_list;
407		if (im) {
408			state->idev = idev;
409			break;
410		}
411		read_unlock_bh(&idev->lock);
412	}
413	return im;
414}
415
416static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im)
417{
418	struct ac6_iter_state *state = ac6_seq_private(seq);
419
420	im = im->aca_next;
421	while (!im) {
422		if (likely(state->idev != NULL))
423			read_unlock_bh(&state->idev->lock);
424
425		state->dev = next_net_device_rcu(state->dev);
426		if (!state->dev) {
427			state->idev = NULL;
428			break;
429		}
430		state->idev = __in6_dev_get(state->dev);
431		if (!state->idev)
432			continue;
433		read_lock_bh(&state->idev->lock);
434		im = state->idev->ac_list;
435	}
436	return im;
437}
438
439static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos)
440{
441	struct ifacaddr6 *im = ac6_get_first(seq);
442	if (im)
443		while (pos && (im = ac6_get_next(seq, im)) != NULL)
444			--pos;
445	return pos ? NULL : im;
446}
447
448static void *ac6_seq_start(struct seq_file *seq, loff_t *pos)
449	__acquires(RCU)
450{
451	rcu_read_lock();
452	return ac6_get_idx(seq, *pos);
453}
454
455static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
456{
457	struct ifacaddr6 *im = ac6_get_next(seq, v);
458
459	++*pos;
460	return im;
461}
462
463static void ac6_seq_stop(struct seq_file *seq, void *v)
464	__releases(RCU)
465{
466	struct ac6_iter_state *state = ac6_seq_private(seq);
467
468	if (likely(state->idev != NULL)) {
469		read_unlock_bh(&state->idev->lock);
470		state->idev = NULL;
471	}
472	rcu_read_unlock();
473}
474
475static int ac6_seq_show(struct seq_file *seq, void *v)
476{
477	struct ifacaddr6 *im = (struct ifacaddr6 *)v;
478	struct ac6_iter_state *state = ac6_seq_private(seq);
479
480	seq_printf(seq, "%-4d %-15s %pi6 %5d\n",
481		   state->dev->ifindex, state->dev->name,
482		   &im->aca_addr, im->aca_users);
483	return 0;
484}
485
486static const struct seq_operations ac6_seq_ops = {
487	.start	=	ac6_seq_start,
488	.next	=	ac6_seq_next,
489	.stop	=	ac6_seq_stop,
490	.show	=	ac6_seq_show,
491};
492
493static int ac6_seq_open(struct inode *inode, struct file *file)
494{
495	return seq_open_net(inode, file, &ac6_seq_ops,
496			    sizeof(struct ac6_iter_state));
497}
498
499static const struct file_operations ac6_seq_fops = {
500	.owner		=	THIS_MODULE,
501	.open		=	ac6_seq_open,
502	.read		=	seq_read,
503	.llseek		=	seq_lseek,
504	.release	=	seq_release_net,
505};
506
507int __net_init ac6_proc_init(struct net *net)
508{
509	if (!proc_net_fops_create(net, "anycast6", S_IRUGO, &ac6_seq_fops))
510		return -ENOMEM;
511
512	return 0;
513}
514
515void ac6_proc_exit(struct net *net)
516{
517	proc_net_remove(net, "anycast6");
518}
519#endif
520
521