addrlabel.c revision 5a0e3ad6af8660be21ca98a971cd00f331318c05 
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#ifdef CONFIG_NET_NS
57	return lbl->lbl_net;
58#else
59	return &init_net;
60#endif
61}
62
63/*
64 * Default policy table (RFC3484 + extensions)
65 *
66 * prefix		addr_type	label
67 * -------------------------------------------------------------------------
68 * ::1/128		LOOPBACK	0
69 * ::/0			N/A		1
70 * 2002::/16		N/A		2
71 * ::/96		COMPATv4	3
72 * ::ffff:0:0/96	V4MAPPED	4
73 * fc00::/7		N/A		5		ULA (RFC 4193)
74 * 2001::/32		N/A		6		Teredo (RFC 4380)
75 * 2001:10::/28		N/A		7		ORCHID (RFC 4843)
76 *
77 * Note: 0xffffffff is used if we do not have any policies.
78 */
79
80#define IPV6_ADDR_LABEL_DEFAULT	0xffffffffUL
81
82static const __net_initdata struct ip6addrlbl_init_table
83{
84	const struct in6_addr *prefix;
85	int prefixlen;
86	u32 label;
87} ip6addrlbl_init_table[] = {
88	{	/* ::/0 */
89		.prefix = &in6addr_any,
90		.label = 1,
91	},{	/* fc00::/7 */
92		.prefix = &(struct in6_addr){{{ 0xfc }}},
93		.prefixlen = 7,
94		.label = 5,
95	},{	/* 2002::/16 */
96		.prefix = &(struct in6_addr){{{ 0x20, 0x02 }}},
97		.prefixlen = 16,
98		.label = 2,
99	},{	/* 2001::/32 */
100		.prefix = &(struct in6_addr){{{ 0x20, 0x01 }}},
101		.prefixlen = 32,
102		.label = 6,
103	},{	/* 2001:10::/28 */
104		.prefix = &(struct in6_addr){{{ 0x20, 0x01, 0x00, 0x10 }}},
105		.prefixlen = 28,
106		.label = 7,
107	},{	/* ::ffff:0:0 */
108		.prefix = &(struct in6_addr){{{ [10] = 0xff, [11] = 0xff }}},
109		.prefixlen = 96,
110		.label = 4,
111	},{	/* ::/96 */
112		.prefix = &in6addr_any,
113		.prefixlen = 96,
114		.label = 3,
115	},{	/* ::1/128 */
116		.prefix = &in6addr_loopback,
117		.prefixlen = 128,
118		.label = 0,
119	}
120};
121
122/* Object management */
123static inline void ip6addrlbl_free(struct ip6addrlbl_entry *p)
124{
125#ifdef CONFIG_NET_NS
126	release_net(p->lbl_net);
127#endif
128	kfree(p);
129}
130
131static void ip6addrlbl_free_rcu(struct rcu_head *h)
132{
133	ip6addrlbl_free(container_of(h, struct ip6addrlbl_entry, rcu));
134}
135
136static inline int ip6addrlbl_hold(struct ip6addrlbl_entry *p)
137{
138	return atomic_inc_not_zero(&p->refcnt);
139}
140
141static inline void ip6addrlbl_put(struct ip6addrlbl_entry *p)
142{
143	if (atomic_dec_and_test(&p->refcnt))
144		call_rcu(&p->rcu, ip6addrlbl_free_rcu);
145}
146
147/* Find label */
148static int __ip6addrlbl_match(struct net *net,
149			      struct ip6addrlbl_entry *p,
150			      const struct in6_addr *addr,
151			      int addrtype, int ifindex)
152{
153	if (!net_eq(ip6addrlbl_net(p), net))
154		return 0;
155	if (p->ifindex && p->ifindex != ifindex)
156		return 0;
157	if (p->addrtype && p->addrtype != addrtype)
158		return 0;
159	if (!ipv6_prefix_equal(addr, &p->prefix, p->prefixlen))
160		return 0;
161	return 1;
162}
163
164static struct ip6addrlbl_entry *__ipv6_addr_label(struct net *net,
165						  const struct in6_addr *addr,
166						  int type, int ifindex)
167{
168	struct hlist_node *pos;
169	struct ip6addrlbl_entry *p;
170	hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
171		if (__ip6addrlbl_match(net, p, addr, type, ifindex))
172			return p;
173	}
174	return NULL;
175}
176
177u32 ipv6_addr_label(struct net *net,
178		    const struct in6_addr *addr, int type, int ifindex)
179{
180	u32 label;
181	struct ip6addrlbl_entry *p;
182
183	type &= IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK;
184
185	rcu_read_lock();
186	p = __ipv6_addr_label(net, addr, type, ifindex);
187	label = p ? p->label : IPV6_ADDR_LABEL_DEFAULT;
188	rcu_read_unlock();
189
190	ADDRLABEL(KERN_DEBUG "%s(addr=%pI6, type=%d, ifindex=%d) => %08x\n",
191		  __func__, addr, type, ifindex, label);
192
193	return label;
194}
195
196/* allocate one entry */
197static struct ip6addrlbl_entry *ip6addrlbl_alloc(struct net *net,
198						 const struct in6_addr *prefix,
199						 int prefixlen, int ifindex,
200						 u32 label)
201{
202	struct ip6addrlbl_entry *newp;
203	int addrtype;
204
205	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u)\n",
206		  __func__, prefix, prefixlen, ifindex, (unsigned int)label);
207
208	addrtype = ipv6_addr_type(prefix) & (IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK);
209
210	switch (addrtype) {
211	case IPV6_ADDR_MAPPED:
212		if (prefixlen > 96)
213			return ERR_PTR(-EINVAL);
214		if (prefixlen < 96)
215			addrtype = 0;
216		break;
217	case IPV6_ADDR_COMPATv4:
218		if (prefixlen != 96)
219			addrtype = 0;
220		break;
221	case IPV6_ADDR_LOOPBACK:
222		if (prefixlen != 128)
223			addrtype = 0;
224		break;
225	}
226
227	newp = kmalloc(sizeof(*newp), GFP_KERNEL);
228	if (!newp)
229		return ERR_PTR(-ENOMEM);
230
231	ipv6_addr_prefix(&newp->prefix, prefix, prefixlen);
232	newp->prefixlen = prefixlen;
233	newp->ifindex = ifindex;
234	newp->addrtype = addrtype;
235	newp->label = label;
236	INIT_HLIST_NODE(&newp->list);
237#ifdef CONFIG_NET_NS
238	newp->lbl_net = hold_net(net);
239#endif
240	atomic_set(&newp->refcnt, 1);
241	return newp;
242}
243
244/* add a label */
245static int __ip6addrlbl_add(struct ip6addrlbl_entry *newp, int replace)
246{
247	int ret = 0;
248
249	ADDRLABEL(KERN_DEBUG "%s(newp=%p, replace=%d)\n",
250			__func__,
251			newp, replace);
252
253	if (hlist_empty(&ip6addrlbl_table.head)) {
254		hlist_add_head_rcu(&newp->list, &ip6addrlbl_table.head);
255	} else {
256		struct hlist_node *pos, *n;
257		struct ip6addrlbl_entry *p = NULL;
258		hlist_for_each_entry_safe(p, pos, n,
259					  &ip6addrlbl_table.head, list) {
260			if (p->prefixlen == newp->prefixlen &&
261			    net_eq(ip6addrlbl_net(p), ip6addrlbl_net(newp)) &&
262			    p->ifindex == newp->ifindex &&
263			    ipv6_addr_equal(&p->prefix, &newp->prefix)) {
264				if (!replace) {
265					ret = -EEXIST;
266					goto out;
267				}
268				hlist_replace_rcu(&p->list, &newp->list);
269				ip6addrlbl_put(p);
270				goto out;
271			} else if ((p->prefixlen == newp->prefixlen && !p->ifindex) ||
272				   (p->prefixlen < newp->prefixlen)) {
273				hlist_add_before_rcu(&newp->list, &p->list);
274				goto out;
275			}
276		}
277		hlist_add_after_rcu(&p->list, &newp->list);
278	}
279out:
280	if (!ret)
281		ip6addrlbl_table.seq++;
282	return ret;
283}
284
285/* add a label */
286static int ip6addrlbl_add(struct net *net,
287			  const struct in6_addr *prefix, int prefixlen,
288			  int ifindex, u32 label, int replace)
289{
290	struct ip6addrlbl_entry *newp;
291	int ret = 0;
292
293	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u, replace=%d)\n",
294		  __func__, prefix, prefixlen, ifindex, (unsigned int)label,
295		  replace);
296
297	newp = ip6addrlbl_alloc(net, prefix, prefixlen, ifindex, label);
298	if (IS_ERR(newp))
299		return PTR_ERR(newp);
300	spin_lock(&ip6addrlbl_table.lock);
301	ret = __ip6addrlbl_add(newp, replace);
302	spin_unlock(&ip6addrlbl_table.lock);
303	if (ret)
304		ip6addrlbl_free(newp);
305	return ret;
306}
307
308/* remove a label */
309static int __ip6addrlbl_del(struct net *net,
310			    const struct in6_addr *prefix, int prefixlen,
311			    int ifindex)
312{
313	struct ip6addrlbl_entry *p = NULL;
314	struct hlist_node *pos, *n;
315	int ret = -ESRCH;
316
317	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
318		  __func__, prefix, prefixlen, ifindex);
319
320	hlist_for_each_entry_safe(p, pos, n, &ip6addrlbl_table.head, list) {
321		if (p->prefixlen == prefixlen &&
322		    net_eq(ip6addrlbl_net(p), net) &&
323		    p->ifindex == ifindex &&
324		    ipv6_addr_equal(&p->prefix, prefix)) {
325			hlist_del_rcu(&p->list);
326			ip6addrlbl_put(p);
327			ret = 0;
328			break;
329		}
330	}
331	return ret;
332}
333
334static int ip6addrlbl_del(struct net *net,
335			  const struct in6_addr *prefix, int prefixlen,
336			  int ifindex)
337{
338	struct in6_addr prefix_buf;
339	int ret;
340
341	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
342		  __func__, prefix, prefixlen, ifindex);
343
344	ipv6_addr_prefix(&prefix_buf, prefix, prefixlen);
345	spin_lock(&ip6addrlbl_table.lock);
346	ret = __ip6addrlbl_del(net, &prefix_buf, prefixlen, ifindex);
347	spin_unlock(&ip6addrlbl_table.lock);
348	return ret;
349}
350
351/* add default label */
352static int __net_init ip6addrlbl_net_init(struct net *net)
353{
354	int err = 0;
355	int i;
356
357	ADDRLABEL(KERN_DEBUG "%s()\n", __func__);
358
359	for (i = 0; i < ARRAY_SIZE(ip6addrlbl_init_table); i++) {
360		int ret = ip6addrlbl_add(net,
361					 ip6addrlbl_init_table[i].prefix,
362					 ip6addrlbl_init_table[i].prefixlen,
363					 0,
364					 ip6addrlbl_init_table[i].label, 0);
365		/* XXX: should we free all rules when we catch an error? */
366		if (ret && (!err || err != -ENOMEM))
367			err = ret;
368	}
369	return err;
370}
371
372static void __net_exit ip6addrlbl_net_exit(struct net *net)
373{
374	struct ip6addrlbl_entry *p = NULL;
375	struct hlist_node *pos, *n;
376
377	/* Remove all labels belonging to the exiting net */
378	spin_lock(&ip6addrlbl_table.lock);
379	hlist_for_each_entry_safe(p, pos, n, &ip6addrlbl_table.head, list) {
380		if (net_eq(ip6addrlbl_net(p), net)) {
381			hlist_del_rcu(&p->list);
382			ip6addrlbl_put(p);
383		}
384	}
385	spin_unlock(&ip6addrlbl_table.lock);
386}
387
388static struct pernet_operations ipv6_addr_label_ops = {
389	.init = ip6addrlbl_net_init,
390	.exit = ip6addrlbl_net_exit,
391};
392
393int __init ipv6_addr_label_init(void)
394{
395	spin_lock_init(&ip6addrlbl_table.lock);
396
397	return register_pernet_subsys(&ipv6_addr_label_ops);
398}
399
400static const struct nla_policy ifal_policy[IFAL_MAX+1] = {
401	[IFAL_ADDRESS]		= { .len = sizeof(struct in6_addr), },
402	[IFAL_LABEL]		= { .len = sizeof(u32), },
403};
404
405static int ip6addrlbl_newdel(struct sk_buff *skb, struct nlmsghdr *nlh,
406			     void *arg)
407{
408	struct net *net = sock_net(skb->sk);
409	struct ifaddrlblmsg *ifal;
410	struct nlattr *tb[IFAL_MAX+1];
411	struct in6_addr *pfx;
412	u32 label;
413	int err = 0;
414
415	err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
416	if (err < 0)
417		return err;
418
419	ifal = nlmsg_data(nlh);
420
421	if (ifal->ifal_family != AF_INET6 ||
422	    ifal->ifal_prefixlen > 128)
423		return -EINVAL;
424
425	if (ifal->ifal_index &&
426	    !__dev_get_by_index(net, ifal->ifal_index))
427		return -EINVAL;
428
429	if (!tb[IFAL_ADDRESS])
430		return -EINVAL;
431
432	pfx = nla_data(tb[IFAL_ADDRESS]);
433	if (!pfx)
434		return -EINVAL;
435
436	if (!tb[IFAL_LABEL])
437		return -EINVAL;
438	label = nla_get_u32(tb[IFAL_LABEL]);
439	if (label == IPV6_ADDR_LABEL_DEFAULT)
440		return -EINVAL;
441
442	switch(nlh->nlmsg_type) {
443	case RTM_NEWADDRLABEL:
444		err = ip6addrlbl_add(net, pfx, ifal->ifal_prefixlen,
445				     ifal->ifal_index, label,
446				     nlh->nlmsg_flags & NLM_F_REPLACE);
447		break;
448	case RTM_DELADDRLABEL:
449		err = ip6addrlbl_del(net, pfx, ifal->ifal_prefixlen,
450				     ifal->ifal_index);
451		break;
452	default:
453		err = -EOPNOTSUPP;
454	}
455	return err;
456}
457
458static inline void ip6addrlbl_putmsg(struct nlmsghdr *nlh,
459				     int prefixlen, int ifindex, u32 lseq)
460{
461	struct ifaddrlblmsg *ifal = nlmsg_data(nlh);
462	ifal->ifal_family = AF_INET6;
463	ifal->ifal_prefixlen = prefixlen;
464	ifal->ifal_flags = 0;
465	ifal->ifal_index = ifindex;
466	ifal->ifal_seq = lseq;
467};
468
469static int ip6addrlbl_fill(struct sk_buff *skb,
470			   struct ip6addrlbl_entry *p,
471			   u32 lseq,
472			   u32 pid, u32 seq, int event,
473			   unsigned int flags)
474{
475	struct nlmsghdr *nlh = nlmsg_put(skb, pid, seq, event,
476					 sizeof(struct ifaddrlblmsg), flags);
477	if (!nlh)
478		return -EMSGSIZE;
479
480	ip6addrlbl_putmsg(nlh, p->prefixlen, p->ifindex, lseq);
481
482	if (nla_put(skb, IFAL_ADDRESS, 16, &p->prefix) < 0 ||
483	    nla_put_u32(skb, IFAL_LABEL, p->label) < 0) {
484		nlmsg_cancel(skb, nlh);
485		return -EMSGSIZE;
486	}
487
488	return nlmsg_end(skb, nlh);
489}
490
491static int ip6addrlbl_dump(struct sk_buff *skb, struct netlink_callback *cb)
492{
493	struct net *net = sock_net(skb->sk);
494	struct ip6addrlbl_entry *p;
495	struct hlist_node *pos;
496	int idx = 0, s_idx = cb->args[0];
497	int err;
498
499	rcu_read_lock();
500	hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
501		if (idx >= s_idx &&
502		    net_eq(ip6addrlbl_net(p), net)) {
503			if ((err = ip6addrlbl_fill(skb, p,
504						   ip6addrlbl_table.seq,
505						   NETLINK_CB(cb->skb).pid,
506						   cb->nlh->nlmsg_seq,
507						   RTM_NEWADDRLABEL,
508						   NLM_F_MULTI)) <= 0)
509				break;
510		}
511		idx++;
512	}
513	rcu_read_unlock();
514	cb->args[0] = idx;
515	return skb->len;
516}
517
518static inline int ip6addrlbl_msgsize(void)
519{
520	return (NLMSG_ALIGN(sizeof(struct ifaddrlblmsg))
521		+ nla_total_size(16)	/* IFAL_ADDRESS */
522		+ nla_total_size(4)	/* IFAL_LABEL */
523	);
524}
525
526static int ip6addrlbl_get(struct sk_buff *in_skb, struct nlmsghdr* nlh,
527			  void *arg)
528{
529	struct net *net = sock_net(in_skb->sk);
530	struct ifaddrlblmsg *ifal;
531	struct nlattr *tb[IFAL_MAX+1];
532	struct in6_addr *addr;
533	u32 lseq;
534	int err = 0;
535	struct ip6addrlbl_entry *p;
536	struct sk_buff *skb;
537
538	err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
539	if (err < 0)
540		return err;
541
542	ifal = nlmsg_data(nlh);
543
544	if (ifal->ifal_family != AF_INET6 ||
545	    ifal->ifal_prefixlen != 128)
546		return -EINVAL;
547
548	if (ifal->ifal_index &&
549	    !__dev_get_by_index(net, ifal->ifal_index))
550		return -EINVAL;
551
552	if (!tb[IFAL_ADDRESS])
553		return -EINVAL;
554
555	addr = nla_data(tb[IFAL_ADDRESS]);
556	if (!addr)
557		return -EINVAL;
558
559	rcu_read_lock();
560	p = __ipv6_addr_label(net, addr, ipv6_addr_type(addr), ifal->ifal_index);
561	if (p && ip6addrlbl_hold(p))
562		p = NULL;
563	lseq = ip6addrlbl_table.seq;
564	rcu_read_unlock();
565
566	if (!p) {
567		err = -ESRCH;
568		goto out;
569	}
570
571	if (!(skb = nlmsg_new(ip6addrlbl_msgsize(), GFP_KERNEL))) {
572		ip6addrlbl_put(p);
573		return -ENOBUFS;
574	}
575
576	err = ip6addrlbl_fill(skb, p, lseq,
577			      NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
578			      RTM_NEWADDRLABEL, 0);
579
580	ip6addrlbl_put(p);
581
582	if (err < 0) {
583		WARN_ON(err == -EMSGSIZE);
584		kfree_skb(skb);
585		goto out;
586	}
587
588	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
589out:
590	return err;
591}
592
593void __init ipv6_addr_label_rtnl_register(void)
594{
595	__rtnl_register(PF_INET6, RTM_NEWADDRLABEL, ip6addrlbl_newdel, NULL);
596	__rtnl_register(PF_INET6, RTM_DELADDRLABEL, ip6addrlbl_newdel, NULL);
597	__rtnl_register(PF_INET6, RTM_GETADDRLABEL, ip6addrlbl_get, ip6addrlbl_dump);
598}
599
600