addrlabel.c revision c2d9ba9bce8d7323ca96f239e1f505c14d6244fb 
1/*
2 * IPv6 Address Label subsystem
3 * for the IPv6 "Default" Source Address Selection
4 *
5 * Copyright (C)2007 USAGI/WIDE Project
6 */
7/*
8 * Author:
9 * 	YOSHIFUJI Hideaki @ USAGI/WIDE Project <yoshfuji@linux-ipv6.org>
10 */
11
12#include <linux/kernel.h>
13#include <linux/list.h>
14#include <linux/rcupdate.h>
15#include <linux/in6.h>
16#include <linux/slab.h>
17#include <net/addrconf.h>
18#include <linux/if_addrlabel.h>
19#include <linux/netlink.h>
20#include <linux/rtnetlink.h>
21
22#if 0
23#define ADDRLABEL(x...) printk(x)
24#else
25#define ADDRLABEL(x...) do { ; } while(0)
26#endif
27
28/*
29 * Policy Table
30 */
31struct ip6addrlbl_entry
32{
33#ifdef CONFIG_NET_NS
34	struct net *lbl_net;
35#endif
36	struct in6_addr prefix;
37	int prefixlen;
38	int ifindex;
39	int addrtype;
40	u32 label;
41	struct hlist_node list;
42	atomic_t refcnt;
43	struct rcu_head rcu;
44};
45
46static struct ip6addrlbl_table
47{
48	struct hlist_head head;
49	spinlock_t lock;
50	u32 seq;
51} ip6addrlbl_table;
52
53static inline
54struct net *ip6addrlbl_net(const struct ip6addrlbl_entry *lbl)
55{
56	return read_pnet(&lbl->lbl_net);
57}
58
59/*
60 * Default policy table (RFC3484 + extensions)
61 *
62 * prefix		addr_type	label
63 * -------------------------------------------------------------------------
64 * ::1/128		LOOPBACK	0
65 * ::/0			N/A		1
66 * 2002::/16		N/A		2
67 * ::/96		COMPATv4	3
68 * ::ffff:0:0/96	V4MAPPED	4
69 * fc00::/7		N/A		5		ULA (RFC 4193)
70 * 2001::/32		N/A		6		Teredo (RFC 4380)
71 * 2001:10::/28		N/A		7		ORCHID (RFC 4843)
72 *
73 * Note: 0xffffffff is used if we do not have any policies.
74 */
75
76#define IPV6_ADDR_LABEL_DEFAULT	0xffffffffUL
77
78static const __net_initdata struct ip6addrlbl_init_table
79{
80	const struct in6_addr *prefix;
81	int prefixlen;
82	u32 label;
83} ip6addrlbl_init_table[] = {
84	{	/* ::/0 */
85		.prefix = &in6addr_any,
86		.label = 1,
87	},{	/* fc00::/7 */
88		.prefix = &(struct in6_addr){{{ 0xfc }}},
89		.prefixlen = 7,
90		.label = 5,
91	},{	/* 2002::/16 */
92		.prefix = &(struct in6_addr){{{ 0x20, 0x02 }}},
93		.prefixlen = 16,
94		.label = 2,
95	},{	/* 2001::/32 */
96		.prefix = &(struct in6_addr){{{ 0x20, 0x01 }}},
97		.prefixlen = 32,
98		.label = 6,
99	},{	/* 2001:10::/28 */
100		.prefix = &(struct in6_addr){{{ 0x20, 0x01, 0x00, 0x10 }}},
101		.prefixlen = 28,
102		.label = 7,
103	},{	/* ::ffff:0:0 */
104		.prefix = &(struct in6_addr){{{ [10] = 0xff, [11] = 0xff }}},
105		.prefixlen = 96,
106		.label = 4,
107	},{	/* ::/96 */
108		.prefix = &in6addr_any,
109		.prefixlen = 96,
110		.label = 3,
111	},{	/* ::1/128 */
112		.prefix = &in6addr_loopback,
113		.prefixlen = 128,
114		.label = 0,
115	}
116};
117
118/* Object management */
119static inline void ip6addrlbl_free(struct ip6addrlbl_entry *p)
120{
121#ifdef CONFIG_NET_NS
122	release_net(p->lbl_net);
123#endif
124	kfree(p);
125}
126
127static void ip6addrlbl_free_rcu(struct rcu_head *h)
128{
129	ip6addrlbl_free(container_of(h, struct ip6addrlbl_entry, rcu));
130}
131
132static inline int ip6addrlbl_hold(struct ip6addrlbl_entry *p)
133{
134	return atomic_inc_not_zero(&p->refcnt);
135}
136
137static inline void ip6addrlbl_put(struct ip6addrlbl_entry *p)
138{
139	if (atomic_dec_and_test(&p->refcnt))
140		call_rcu(&p->rcu, ip6addrlbl_free_rcu);
141}
142
143/* Find label */
144static int __ip6addrlbl_match(struct net *net,
145			      struct ip6addrlbl_entry *p,
146			      const struct in6_addr *addr,
147			      int addrtype, int ifindex)
148{
149	if (!net_eq(ip6addrlbl_net(p), net))
150		return 0;
151	if (p->ifindex && p->ifindex != ifindex)
152		return 0;
153	if (p->addrtype && p->addrtype != addrtype)
154		return 0;
155	if (!ipv6_prefix_equal(addr, &p->prefix, p->prefixlen))
156		return 0;
157	return 1;
158}
159
160static struct ip6addrlbl_entry *__ipv6_addr_label(struct net *net,
161						  const struct in6_addr *addr,
162						  int type, int ifindex)
163{
164	struct hlist_node *pos;
165	struct ip6addrlbl_entry *p;
166	hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
167		if (__ip6addrlbl_match(net, p, addr, type, ifindex))
168			return p;
169	}
170	return NULL;
171}
172
173u32 ipv6_addr_label(struct net *net,
174		    const struct in6_addr *addr, int type, int ifindex)
175{
176	u32 label;
177	struct ip6addrlbl_entry *p;
178
179	type &= IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK;
180
181	rcu_read_lock();
182	p = __ipv6_addr_label(net, addr, type, ifindex);
183	label = p ? p->label : IPV6_ADDR_LABEL_DEFAULT;
184	rcu_read_unlock();
185
186	ADDRLABEL(KERN_DEBUG "%s(addr=%pI6, type=%d, ifindex=%d) => %08x\n",
187		  __func__, addr, type, ifindex, label);
188
189	return label;
190}
191
192/* allocate one entry */
193static struct ip6addrlbl_entry *ip6addrlbl_alloc(struct net *net,
194						 const struct in6_addr *prefix,
195						 int prefixlen, int ifindex,
196						 u32 label)
197{
198	struct ip6addrlbl_entry *newp;
199	int addrtype;
200
201	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u)\n",
202		  __func__, prefix, prefixlen, ifindex, (unsigned int)label);
203
204	addrtype = ipv6_addr_type(prefix) & (IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK);
205
206	switch (addrtype) {
207	case IPV6_ADDR_MAPPED:
208		if (prefixlen > 96)
209			return ERR_PTR(-EINVAL);
210		if (prefixlen < 96)
211			addrtype = 0;
212		break;
213	case IPV6_ADDR_COMPATv4:
214		if (prefixlen != 96)
215			addrtype = 0;
216		break;
217	case IPV6_ADDR_LOOPBACK:
218		if (prefixlen != 128)
219			addrtype = 0;
220		break;
221	}
222
223	newp = kmalloc(sizeof(*newp), GFP_KERNEL);
224	if (!newp)
225		return ERR_PTR(-ENOMEM);
226
227	ipv6_addr_prefix(&newp->prefix, prefix, prefixlen);
228	newp->prefixlen = prefixlen;
229	newp->ifindex = ifindex;
230	newp->addrtype = addrtype;
231	newp->label = label;
232	INIT_HLIST_NODE(&newp->list);
233#ifdef CONFIG_NET_NS
234	newp->lbl_net = hold_net(net);
235#endif
236	atomic_set(&newp->refcnt, 1);
237	return newp;
238}
239
240/* add a label */
241static int __ip6addrlbl_add(struct ip6addrlbl_entry *newp, int replace)
242{
243	int ret = 0;
244
245	ADDRLABEL(KERN_DEBUG "%s(newp=%p, replace=%d)\n",
246			__func__,
247			newp, replace);
248
249	if (hlist_empty(&ip6addrlbl_table.head)) {
250		hlist_add_head_rcu(&newp->list, &ip6addrlbl_table.head);
251	} else {
252		struct hlist_node *pos, *n;
253		struct ip6addrlbl_entry *p = NULL;
254		hlist_for_each_entry_safe(p, pos, n,
255					  &ip6addrlbl_table.head, list) {
256			if (p->prefixlen == newp->prefixlen &&
257			    net_eq(ip6addrlbl_net(p), ip6addrlbl_net(newp)) &&
258			    p->ifindex == newp->ifindex &&
259			    ipv6_addr_equal(&p->prefix, &newp->prefix)) {
260				if (!replace) {
261					ret = -EEXIST;
262					goto out;
263				}
264				hlist_replace_rcu(&p->list, &newp->list);
265				ip6addrlbl_put(p);
266				goto out;
267			} else if ((p->prefixlen == newp->prefixlen && !p->ifindex) ||
268				   (p->prefixlen < newp->prefixlen)) {
269				hlist_add_before_rcu(&newp->list, &p->list);
270				goto out;
271			}
272		}
273		hlist_add_after_rcu(&p->list, &newp->list);
274	}
275out:
276	if (!ret)
277		ip6addrlbl_table.seq++;
278	return ret;
279}
280
281/* add a label */
282static int ip6addrlbl_add(struct net *net,
283			  const struct in6_addr *prefix, int prefixlen,
284			  int ifindex, u32 label, int replace)
285{
286	struct ip6addrlbl_entry *newp;
287	int ret = 0;
288
289	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u, replace=%d)\n",
290		  __func__, prefix, prefixlen, ifindex, (unsigned int)label,
291		  replace);
292
293	newp = ip6addrlbl_alloc(net, prefix, prefixlen, ifindex, label);
294	if (IS_ERR(newp))
295		return PTR_ERR(newp);
296	spin_lock(&ip6addrlbl_table.lock);
297	ret = __ip6addrlbl_add(newp, replace);
298	spin_unlock(&ip6addrlbl_table.lock);
299	if (ret)
300		ip6addrlbl_free(newp);
301	return ret;
302}
303
304/* remove a label */
305static int __ip6addrlbl_del(struct net *net,
306			    const struct in6_addr *prefix, int prefixlen,
307			    int ifindex)
308{
309	struct ip6addrlbl_entry *p = NULL;
310	struct hlist_node *pos, *n;
311	int ret = -ESRCH;
312
313	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
314		  __func__, prefix, prefixlen, ifindex);
315
316	hlist_for_each_entry_safe(p, pos, n, &ip6addrlbl_table.head, list) {
317		if (p->prefixlen == prefixlen &&
318		    net_eq(ip6addrlbl_net(p), net) &&
319		    p->ifindex == ifindex &&
320		    ipv6_addr_equal(&p->prefix, prefix)) {
321			hlist_del_rcu(&p->list);
322			ip6addrlbl_put(p);
323			ret = 0;
324			break;
325		}
326	}
327	return ret;
328}
329
330static int ip6addrlbl_del(struct net *net,
331			  const struct in6_addr *prefix, int prefixlen,
332			  int ifindex)
333{
334	struct in6_addr prefix_buf;
335	int ret;
336
337	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
338		  __func__, prefix, prefixlen, ifindex);
339
340	ipv6_addr_prefix(&prefix_buf, prefix, prefixlen);
341	spin_lock(&ip6addrlbl_table.lock);
342	ret = __ip6addrlbl_del(net, &prefix_buf, prefixlen, ifindex);
343	spin_unlock(&ip6addrlbl_table.lock);
344	return ret;
345}
346
347/* add default label */
348static int __net_init ip6addrlbl_net_init(struct net *net)
349{
350	int err = 0;
351	int i;
352
353	ADDRLABEL(KERN_DEBUG "%s()\n", __func__);
354
355	for (i = 0; i < ARRAY_SIZE(ip6addrlbl_init_table); i++) {
356		int ret = ip6addrlbl_add(net,
357					 ip6addrlbl_init_table[i].prefix,
358					 ip6addrlbl_init_table[i].prefixlen,
359					 0,
360					 ip6addrlbl_init_table[i].label, 0);
361		/* XXX: should we free all rules when we catch an error? */
362		if (ret && (!err || err != -ENOMEM))
363			err = ret;
364	}
365	return err;
366}
367
368static void __net_exit ip6addrlbl_net_exit(struct net *net)
369{
370	struct ip6addrlbl_entry *p = NULL;
371	struct hlist_node *pos, *n;
372
373	/* Remove all labels belonging to the exiting net */
374	spin_lock(&ip6addrlbl_table.lock);
375	hlist_for_each_entry_safe(p, pos, n, &ip6addrlbl_table.head, list) {
376		if (net_eq(ip6addrlbl_net(p), net)) {
377			hlist_del_rcu(&p->list);
378			ip6addrlbl_put(p);
379		}
380	}
381	spin_unlock(&ip6addrlbl_table.lock);
382}
383
384static struct pernet_operations ipv6_addr_label_ops = {
385	.init = ip6addrlbl_net_init,
386	.exit = ip6addrlbl_net_exit,
387};
388
389int __init ipv6_addr_label_init(void)
390{
391	spin_lock_init(&ip6addrlbl_table.lock);
392
393	return register_pernet_subsys(&ipv6_addr_label_ops);
394}
395
396static const struct nla_policy ifal_policy[IFAL_MAX+1] = {
397	[IFAL_ADDRESS]		= { .len = sizeof(struct in6_addr), },
398	[IFAL_LABEL]		= { .len = sizeof(u32), },
399};
400
401static int ip6addrlbl_newdel(struct sk_buff *skb, struct nlmsghdr *nlh,
402			     void *arg)
403{
404	struct net *net = sock_net(skb->sk);
405	struct ifaddrlblmsg *ifal;
406	struct nlattr *tb[IFAL_MAX+1];
407	struct in6_addr *pfx;
408	u32 label;
409	int err = 0;
410
411	err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
412	if (err < 0)
413		return err;
414
415	ifal = nlmsg_data(nlh);
416
417	if (ifal->ifal_family != AF_INET6 ||
418	    ifal->ifal_prefixlen > 128)
419		return -EINVAL;
420
421	if (!tb[IFAL_ADDRESS])
422		return -EINVAL;
423
424	pfx = nla_data(tb[IFAL_ADDRESS]);
425	if (!pfx)
426		return -EINVAL;
427
428	if (!tb[IFAL_LABEL])
429		return -EINVAL;
430	label = nla_get_u32(tb[IFAL_LABEL]);
431	if (label == IPV6_ADDR_LABEL_DEFAULT)
432		return -EINVAL;
433
434	switch(nlh->nlmsg_type) {
435	case RTM_NEWADDRLABEL:
436		if (ifal->ifal_index &&
437		    !__dev_get_by_index(net, ifal->ifal_index))
438			return -EINVAL;
439
440		err = ip6addrlbl_add(net, pfx, ifal->ifal_prefixlen,
441				     ifal->ifal_index, label,
442				     nlh->nlmsg_flags & NLM_F_REPLACE);
443		break;
444	case RTM_DELADDRLABEL:
445		err = ip6addrlbl_del(net, pfx, ifal->ifal_prefixlen,
446				     ifal->ifal_index);
447		break;
448	default:
449		err = -EOPNOTSUPP;
450	}
451	return err;
452}
453
454static inline void ip6addrlbl_putmsg(struct nlmsghdr *nlh,
455				     int prefixlen, int ifindex, u32 lseq)
456{
457	struct ifaddrlblmsg *ifal = nlmsg_data(nlh);
458	ifal->ifal_family = AF_INET6;
459	ifal->ifal_prefixlen = prefixlen;
460	ifal->ifal_flags = 0;
461	ifal->ifal_index = ifindex;
462	ifal->ifal_seq = lseq;
463};
464
465static int ip6addrlbl_fill(struct sk_buff *skb,
466			   struct ip6addrlbl_entry *p,
467			   u32 lseq,
468			   u32 pid, u32 seq, int event,
469			   unsigned int flags)
470{
471	struct nlmsghdr *nlh = nlmsg_put(skb, pid, seq, event,
472					 sizeof(struct ifaddrlblmsg), flags);
473	if (!nlh)
474		return -EMSGSIZE;
475
476	ip6addrlbl_putmsg(nlh, p->prefixlen, p->ifindex, lseq);
477
478	if (nla_put(skb, IFAL_ADDRESS, 16, &p->prefix) < 0 ||
479	    nla_put_u32(skb, IFAL_LABEL, p->label) < 0) {
480		nlmsg_cancel(skb, nlh);
481		return -EMSGSIZE;
482	}
483
484	return nlmsg_end(skb, nlh);
485}
486
487static int ip6addrlbl_dump(struct sk_buff *skb, struct netlink_callback *cb)
488{
489	struct net *net = sock_net(skb->sk);
490	struct ip6addrlbl_entry *p;
491	struct hlist_node *pos;
492	int idx = 0, s_idx = cb->args[0];
493	int err;
494
495	rcu_read_lock();
496	hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
497		if (idx >= s_idx &&
498		    net_eq(ip6addrlbl_net(p), net)) {
499			if ((err = ip6addrlbl_fill(skb, p,
500						   ip6addrlbl_table.seq,
501						   NETLINK_CB(cb->skb).pid,
502						   cb->nlh->nlmsg_seq,
503						   RTM_NEWADDRLABEL,
504						   NLM_F_MULTI)) <= 0)
505				break;
506		}
507		idx++;
508	}
509	rcu_read_unlock();
510	cb->args[0] = idx;
511	return skb->len;
512}
513
514static inline int ip6addrlbl_msgsize(void)
515{
516	return (NLMSG_ALIGN(sizeof(struct ifaddrlblmsg))
517		+ nla_total_size(16)	/* IFAL_ADDRESS */
518		+ nla_total_size(4)	/* IFAL_LABEL */
519	);
520}
521
522static int ip6addrlbl_get(struct sk_buff *in_skb, struct nlmsghdr* nlh,
523			  void *arg)
524{
525	struct net *net = sock_net(in_skb->sk);
526	struct ifaddrlblmsg *ifal;
527	struct nlattr *tb[IFAL_MAX+1];
528	struct in6_addr *addr;
529	u32 lseq;
530	int err = 0;
531	struct ip6addrlbl_entry *p;
532	struct sk_buff *skb;
533
534	err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
535	if (err < 0)
536		return err;
537
538	ifal = nlmsg_data(nlh);
539
540	if (ifal->ifal_family != AF_INET6 ||
541	    ifal->ifal_prefixlen != 128)
542		return -EINVAL;
543
544	if (ifal->ifal_index &&
545	    !__dev_get_by_index(net, ifal->ifal_index))
546		return -EINVAL;
547
548	if (!tb[IFAL_ADDRESS])
549		return -EINVAL;
550
551	addr = nla_data(tb[IFAL_ADDRESS]);
552	if (!addr)
553		return -EINVAL;
554
555	rcu_read_lock();
556	p = __ipv6_addr_label(net, addr, ipv6_addr_type(addr), ifal->ifal_index);
557	if (p && ip6addrlbl_hold(p))
558		p = NULL;
559	lseq = ip6addrlbl_table.seq;
560	rcu_read_unlock();
561
562	if (!p) {
563		err = -ESRCH;
564		goto out;
565	}
566
567	if (!(skb = nlmsg_new(ip6addrlbl_msgsize(), GFP_KERNEL))) {
568		ip6addrlbl_put(p);
569		return -ENOBUFS;
570	}
571
572	err = ip6addrlbl_fill(skb, p, lseq,
573			      NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
574			      RTM_NEWADDRLABEL, 0);
575
576	ip6addrlbl_put(p);
577
578	if (err < 0) {
579		WARN_ON(err == -EMSGSIZE);
580		kfree_skb(skb);
581		goto out;
582	}
583
584	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
585out:
586	return err;
587}
588
589void __init ipv6_addr_label_rtnl_register(void)
590{
591	__rtnl_register(PF_INET6, RTM_NEWADDRLABEL, ip6addrlbl_newdel, NULL);
592	__rtnl_register(PF_INET6, RTM_DELADDRLABEL, ip6addrlbl_newdel, NULL);
593	__rtnl_register(PF_INET6, RTM_GETADDRLABEL, ip6addrlbl_get, ip6addrlbl_dump);
594}
595
596