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	possible_net_t lbl_net;
33	struct in6_addr prefix;
34	int prefixlen;
35	int ifindex;
36	int addrtype;
37	u32 label;
38	struct hlist_node list;
39	atomic_t refcnt;
40	struct rcu_head rcu;
41};
42
43static struct ip6addrlbl_table
44{
45	struct hlist_head head;
46	spinlock_t lock;
47	u32 seq;
48} ip6addrlbl_table;
49
50static inline
51struct net *ip6addrlbl_net(const struct ip6addrlbl_entry *lbl)
52{
53	return read_pnet(&lbl->lbl_net);
54}
55
56/*
57 * Default policy table (RFC6724 + extensions)
58 *
59 * prefix		addr_type	label
60 * -------------------------------------------------------------------------
61 * ::1/128		LOOPBACK	0
62 * ::/0			N/A		1
63 * 2002::/16		N/A		2
64 * ::/96		COMPATv4	3
65 * ::ffff:0:0/96	V4MAPPED	4
66 * fc00::/7		N/A		5		ULA (RFC 4193)
67 * 2001::/32		N/A		6		Teredo (RFC 4380)
68 * 2001:10::/28		N/A		7		ORCHID (RFC 4843)
69 * fec0::/10		N/A		11		Site-local
70 *							(deprecated by RFC3879)
71 * 3ffe::/16		N/A		12		6bone
72 *
73 * Note: 0xffffffff is used if we do not have any policies.
74 * Note: Labels for ULA and 6to4 are different from labels listed in RFC6724.
75 */
76
77#define IPV6_ADDR_LABEL_DEFAULT	0xffffffffUL
78
79static const __net_initconst struct ip6addrlbl_init_table
80{
81	const struct in6_addr *prefix;
82	int prefixlen;
83	u32 label;
84} ip6addrlbl_init_table[] = {
85	{	/* ::/0 */
86		.prefix = &in6addr_any,
87		.label = 1,
88	}, {	/* fc00::/7 */
89		.prefix = &(struct in6_addr){ { { 0xfc } } } ,
90		.prefixlen = 7,
91		.label = 5,
92	}, {	/* fec0::/10 */
93		.prefix = &(struct in6_addr){ { { 0xfe, 0xc0 } } },
94		.prefixlen = 10,
95		.label = 11,
96	}, {	/* 2002::/16 */
97		.prefix = &(struct in6_addr){ { { 0x20, 0x02 } } },
98		.prefixlen = 16,
99		.label = 2,
100	}, {	/* 3ffe::/16 */
101		.prefix = &(struct in6_addr){ { { 0x3f, 0xfe } } },
102		.prefixlen = 16,
103		.label = 12,
104	}, {	/* 2001::/32 */
105		.prefix = &(struct in6_addr){ { { 0x20, 0x01 } } },
106		.prefixlen = 32,
107		.label = 6,
108	}, {	/* 2001:10::/28 */
109		.prefix = &(struct in6_addr){ { { 0x20, 0x01, 0x00, 0x10 } } },
110		.prefixlen = 28,
111		.label = 7,
112	}, {	/* ::ffff:0:0 */
113		.prefix = &(struct in6_addr){ { { [10] = 0xff, [11] = 0xff } } },
114		.prefixlen = 96,
115		.label = 4,
116	}, {	/* ::/96 */
117		.prefix = &in6addr_any,
118		.prefixlen = 96,
119		.label = 3,
120	}, {	/* ::1/128 */
121		.prefix = &in6addr_loopback,
122		.prefixlen = 128,
123		.label = 0,
124	}
125};
126
127/* Object management */
128static inline void ip6addrlbl_free(struct ip6addrlbl_entry *p)
129{
130	kfree(p);
131}
132
133static void ip6addrlbl_free_rcu(struct rcu_head *h)
134{
135	ip6addrlbl_free(container_of(h, struct ip6addrlbl_entry, rcu));
136}
137
138static bool ip6addrlbl_hold(struct ip6addrlbl_entry *p)
139{
140	return atomic_inc_not_zero(&p->refcnt);
141}
142
143static inline void ip6addrlbl_put(struct ip6addrlbl_entry *p)
144{
145	if (atomic_dec_and_test(&p->refcnt))
146		call_rcu(&p->rcu, ip6addrlbl_free_rcu);
147}
148
149/* Find label */
150static bool __ip6addrlbl_match(struct net *net,
151			       const struct ip6addrlbl_entry *p,
152			       const struct in6_addr *addr,
153			       int addrtype, int ifindex)
154{
155	if (!net_eq(ip6addrlbl_net(p), net))
156		return false;
157	if (p->ifindex && p->ifindex != ifindex)
158		return false;
159	if (p->addrtype && p->addrtype != addrtype)
160		return false;
161	if (!ipv6_prefix_equal(addr, &p->prefix, p->prefixlen))
162		return false;
163	return true;
164}
165
166static struct ip6addrlbl_entry *__ipv6_addr_label(struct net *net,
167						  const struct in6_addr *addr,
168						  int type, int ifindex)
169{
170	struct ip6addrlbl_entry *p;
171	hlist_for_each_entry_rcu(p, &ip6addrlbl_table.head, list) {
172		if (__ip6addrlbl_match(net, p, addr, type, ifindex))
173			return p;
174	}
175	return NULL;
176}
177
178u32 ipv6_addr_label(struct net *net,
179		    const struct in6_addr *addr, int type, int ifindex)
180{
181	u32 label;
182	struct ip6addrlbl_entry *p;
183
184	type &= IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK;
185
186	rcu_read_lock();
187	p = __ipv6_addr_label(net, addr, type, ifindex);
188	label = p ? p->label : IPV6_ADDR_LABEL_DEFAULT;
189	rcu_read_unlock();
190
191	ADDRLABEL(KERN_DEBUG "%s(addr=%pI6, type=%d, ifindex=%d) => %08x\n",
192		  __func__, addr, type, ifindex, label);
193
194	return label;
195}
196
197/* allocate one entry */
198static struct ip6addrlbl_entry *ip6addrlbl_alloc(struct net *net,
199						 const struct in6_addr *prefix,
200						 int prefixlen, int ifindex,
201						 u32 label)
202{
203	struct ip6addrlbl_entry *newp;
204	int addrtype;
205
206	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u)\n",
207		  __func__, prefix, prefixlen, ifindex, (unsigned int)label);
208
209	addrtype = ipv6_addr_type(prefix) & (IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK);
210
211	switch (addrtype) {
212	case IPV6_ADDR_MAPPED:
213		if (prefixlen > 96)
214			return ERR_PTR(-EINVAL);
215		if (prefixlen < 96)
216			addrtype = 0;
217		break;
218	case IPV6_ADDR_COMPATv4:
219		if (prefixlen != 96)
220			addrtype = 0;
221		break;
222	case IPV6_ADDR_LOOPBACK:
223		if (prefixlen != 128)
224			addrtype = 0;
225		break;
226	}
227
228	newp = kmalloc(sizeof(*newp), GFP_KERNEL);
229	if (!newp)
230		return ERR_PTR(-ENOMEM);
231
232	ipv6_addr_prefix(&newp->prefix, prefix, prefixlen);
233	newp->prefixlen = prefixlen;
234	newp->ifindex = ifindex;
235	newp->addrtype = addrtype;
236	newp->label = label;
237	INIT_HLIST_NODE(&newp->list);
238	write_pnet(&newp->lbl_net, net);
239	atomic_set(&newp->refcnt, 1);
240	return newp;
241}
242
243/* add a label */
244static int __ip6addrlbl_add(struct ip6addrlbl_entry *newp, int replace)
245{
246	struct hlist_node *n;
247	struct ip6addrlbl_entry *last = NULL, *p = NULL;
248	int ret = 0;
249
250	ADDRLABEL(KERN_DEBUG "%s(newp=%p, replace=%d)\n", __func__, newp,
251		  replace);
252
253	hlist_for_each_entry_safe(p, n,	&ip6addrlbl_table.head, list) {
254		if (p->prefixlen == newp->prefixlen &&
255		    net_eq(ip6addrlbl_net(p), ip6addrlbl_net(newp)) &&
256		    p->ifindex == newp->ifindex &&
257		    ipv6_addr_equal(&p->prefix, &newp->prefix)) {
258			if (!replace) {
259				ret = -EEXIST;
260				goto out;
261			}
262			hlist_replace_rcu(&p->list, &newp->list);
263			ip6addrlbl_put(p);
264			goto out;
265		} else if ((p->prefixlen == newp->prefixlen && !p->ifindex) ||
266			   (p->prefixlen < newp->prefixlen)) {
267			hlist_add_before_rcu(&newp->list, &p->list);
268			goto out;
269		}
270		last = p;
271	}
272	if (last)
273		hlist_add_behind_rcu(&newp->list, &last->list);
274	else
275		hlist_add_head_rcu(&newp->list, &ip6addrlbl_table.head);
276out:
277	if (!ret)
278		ip6addrlbl_table.seq++;
279	return ret;
280}
281
282/* add a label */
283static int ip6addrlbl_add(struct net *net,
284			  const struct in6_addr *prefix, int prefixlen,
285			  int ifindex, u32 label, int replace)
286{
287	struct ip6addrlbl_entry *newp;
288	int ret = 0;
289
290	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u, replace=%d)\n",
291		  __func__, prefix, prefixlen, ifindex, (unsigned int)label,
292		  replace);
293
294	newp = ip6addrlbl_alloc(net, prefix, prefixlen, ifindex, label);
295	if (IS_ERR(newp))
296		return PTR_ERR(newp);
297	spin_lock(&ip6addrlbl_table.lock);
298	ret = __ip6addrlbl_add(newp, replace);
299	spin_unlock(&ip6addrlbl_table.lock);
300	if (ret)
301		ip6addrlbl_free(newp);
302	return ret;
303}
304
305/* remove a label */
306static int __ip6addrlbl_del(struct net *net,
307			    const struct in6_addr *prefix, int prefixlen,
308			    int ifindex)
309{
310	struct ip6addrlbl_entry *p = NULL;
311	struct hlist_node *n;
312	int ret = -ESRCH;
313
314	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
315		  __func__, prefix, prefixlen, ifindex);
316
317	hlist_for_each_entry_safe(p, n, &ip6addrlbl_table.head, list) {
318		if (p->prefixlen == prefixlen &&
319		    net_eq(ip6addrlbl_net(p), net) &&
320		    p->ifindex == ifindex &&
321		    ipv6_addr_equal(&p->prefix, prefix)) {
322			hlist_del_rcu(&p->list);
323			ip6addrlbl_put(p);
324			ret = 0;
325			break;
326		}
327	}
328	return ret;
329}
330
331static int ip6addrlbl_del(struct net *net,
332			  const struct in6_addr *prefix, int prefixlen,
333			  int ifindex)
334{
335	struct in6_addr prefix_buf;
336	int ret;
337
338	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
339		  __func__, prefix, prefixlen, ifindex);
340
341	ipv6_addr_prefix(&prefix_buf, prefix, prefixlen);
342	spin_lock(&ip6addrlbl_table.lock);
343	ret = __ip6addrlbl_del(net, &prefix_buf, prefixlen, ifindex);
344	spin_unlock(&ip6addrlbl_table.lock);
345	return ret;
346}
347
348/* add default label */
349static int __net_init ip6addrlbl_net_init(struct net *net)
350{
351	int err = 0;
352	int i;
353
354	ADDRLABEL(KERN_DEBUG "%s\n", __func__);
355
356	for (i = 0; i < ARRAY_SIZE(ip6addrlbl_init_table); i++) {
357		int ret = ip6addrlbl_add(net,
358					 ip6addrlbl_init_table[i].prefix,
359					 ip6addrlbl_init_table[i].prefixlen,
360					 0,
361					 ip6addrlbl_init_table[i].label, 0);
362		/* XXX: should we free all rules when we catch an error? */
363		if (ret && (!err || err != -ENOMEM))
364			err = ret;
365	}
366	return err;
367}
368
369static void __net_exit ip6addrlbl_net_exit(struct net *net)
370{
371	struct ip6addrlbl_entry *p = NULL;
372	struct hlist_node *n;
373
374	/* Remove all labels belonging to the exiting net */
375	spin_lock(&ip6addrlbl_table.lock);
376	hlist_for_each_entry_safe(p, n, &ip6addrlbl_table.head, list) {
377		if (net_eq(ip6addrlbl_net(p), net)) {
378			hlist_del_rcu(&p->list);
379			ip6addrlbl_put(p);
380		}
381	}
382	spin_unlock(&ip6addrlbl_table.lock);
383}
384
385static struct pernet_operations ipv6_addr_label_ops = {
386	.init = ip6addrlbl_net_init,
387	.exit = ip6addrlbl_net_exit,
388};
389
390int __init ipv6_addr_label_init(void)
391{
392	spin_lock_init(&ip6addrlbl_table.lock);
393
394	return register_pernet_subsys(&ipv6_addr_label_ops);
395}
396
397void ipv6_addr_label_cleanup(void)
398{
399	unregister_pernet_subsys(&ipv6_addr_label_ops);
400}
401
402static const struct nla_policy ifal_policy[IFAL_MAX+1] = {
403	[IFAL_ADDRESS]		= { .len = sizeof(struct in6_addr), },
404	[IFAL_LABEL]		= { .len = sizeof(u32), },
405};
406
407static int ip6addrlbl_newdel(struct sk_buff *skb, struct nlmsghdr *nlh)
408{
409	struct net *net = sock_net(skb->sk);
410	struct ifaddrlblmsg *ifal;
411	struct nlattr *tb[IFAL_MAX+1];
412	struct in6_addr *pfx;
413	u32 label;
414	int err = 0;
415
416	err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
417	if (err < 0)
418		return err;
419
420	ifal = nlmsg_data(nlh);
421
422	if (ifal->ifal_family != AF_INET6 ||
423	    ifal->ifal_prefixlen > 128)
424		return -EINVAL;
425
426	if (!tb[IFAL_ADDRESS])
427		return -EINVAL;
428	pfx = nla_data(tb[IFAL_ADDRESS]);
429
430	if (!tb[IFAL_LABEL])
431		return -EINVAL;
432	label = nla_get_u32(tb[IFAL_LABEL]);
433	if (label == IPV6_ADDR_LABEL_DEFAULT)
434		return -EINVAL;
435
436	switch (nlh->nlmsg_type) {
437	case RTM_NEWADDRLABEL:
438		if (ifal->ifal_index &&
439		    !__dev_get_by_index(net, ifal->ifal_index))
440			return -EINVAL;
441
442		err = ip6addrlbl_add(net, pfx, ifal->ifal_prefixlen,
443				     ifal->ifal_index, label,
444				     nlh->nlmsg_flags & NLM_F_REPLACE);
445		break;
446	case RTM_DELADDRLABEL:
447		err = ip6addrlbl_del(net, pfx, ifal->ifal_prefixlen,
448				     ifal->ifal_index);
449		break;
450	default:
451		err = -EOPNOTSUPP;
452	}
453	return err;
454}
455
456static void ip6addrlbl_putmsg(struct nlmsghdr *nlh,
457			      int prefixlen, int ifindex, u32 lseq)
458{
459	struct ifaddrlblmsg *ifal = nlmsg_data(nlh);
460	ifal->ifal_family = AF_INET6;
461	ifal->ifal_prefixlen = prefixlen;
462	ifal->ifal_flags = 0;
463	ifal->ifal_index = ifindex;
464	ifal->ifal_seq = lseq;
465};
466
467static int ip6addrlbl_fill(struct sk_buff *skb,
468			   struct ip6addrlbl_entry *p,
469			   u32 lseq,
470			   u32 portid, u32 seq, int event,
471			   unsigned int flags)
472{
473	struct nlmsghdr *nlh = nlmsg_put(skb, portid, seq, event,
474					 sizeof(struct ifaddrlblmsg), flags);
475	if (!nlh)
476		return -EMSGSIZE;
477
478	ip6addrlbl_putmsg(nlh, p->prefixlen, p->ifindex, lseq);
479
480	if (nla_put_in6_addr(skb, IFAL_ADDRESS, &p->prefix) < 0 ||
481	    nla_put_u32(skb, IFAL_LABEL, p->label) < 0) {
482		nlmsg_cancel(skb, nlh);
483		return -EMSGSIZE;
484	}
485
486	nlmsg_end(skb, nlh);
487	return 0;
488}
489
490static int ip6addrlbl_dump(struct sk_buff *skb, struct netlink_callback *cb)
491{
492	struct net *net = sock_net(skb->sk);
493	struct ip6addrlbl_entry *p;
494	int idx = 0, s_idx = cb->args[0];
495	int err;
496
497	rcu_read_lock();
498	hlist_for_each_entry_rcu(p, &ip6addrlbl_table.head, list) {
499		if (idx >= s_idx &&
500		    net_eq(ip6addrlbl_net(p), net)) {
501			err = ip6addrlbl_fill(skb, p,
502					      ip6addrlbl_table.seq,
503					      NETLINK_CB(cb->skb).portid,
504					      cb->nlh->nlmsg_seq,
505					      RTM_NEWADDRLABEL,
506					      NLM_F_MULTI);
507			if (err < 0)
508				break;
509		}
510		idx++;
511	}
512	rcu_read_unlock();
513	cb->args[0] = idx;
514	return skb->len;
515}
516
517static inline int ip6addrlbl_msgsize(void)
518{
519	return NLMSG_ALIGN(sizeof(struct ifaddrlblmsg))
520		+ nla_total_size(16)	/* IFAL_ADDRESS */
521		+ nla_total_size(4);	/* IFAL_LABEL */
522}
523
524static int ip6addrlbl_get(struct sk_buff *in_skb, struct nlmsghdr *nlh)
525{
526	struct net *net = sock_net(in_skb->sk);
527	struct ifaddrlblmsg *ifal;
528	struct nlattr *tb[IFAL_MAX+1];
529	struct in6_addr *addr;
530	u32 lseq;
531	int err = 0;
532	struct ip6addrlbl_entry *p;
533	struct sk_buff *skb;
534
535	err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
536	if (err < 0)
537		return err;
538
539	ifal = nlmsg_data(nlh);
540
541	if (ifal->ifal_family != AF_INET6 ||
542	    ifal->ifal_prefixlen != 128)
543		return -EINVAL;
544
545	if (ifal->ifal_index &&
546	    !__dev_get_by_index(net, ifal->ifal_index))
547		return -EINVAL;
548
549	if (!tb[IFAL_ADDRESS])
550		return -EINVAL;
551	addr = nla_data(tb[IFAL_ADDRESS]);
552
553	rcu_read_lock();
554	p = __ipv6_addr_label(net, addr, ipv6_addr_type(addr), ifal->ifal_index);
555	if (p && !ip6addrlbl_hold(p))
556		p = NULL;
557	lseq = ip6addrlbl_table.seq;
558	rcu_read_unlock();
559
560	if (!p) {
561		err = -ESRCH;
562		goto out;
563	}
564
565	skb = nlmsg_new(ip6addrlbl_msgsize(), GFP_KERNEL);
566	if (!skb) {
567		ip6addrlbl_put(p);
568		return -ENOBUFS;
569	}
570
571	err = ip6addrlbl_fill(skb, p, lseq,
572			      NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
573			      RTM_NEWADDRLABEL, 0);
574
575	ip6addrlbl_put(p);
576
577	if (err < 0) {
578		WARN_ON(err == -EMSGSIZE);
579		kfree_skb(skb);
580		goto out;
581	}
582
583	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
584out:
585	return err;
586}
587
588void __init ipv6_addr_label_rtnl_register(void)
589{
590	__rtnl_register(PF_INET6, RTM_NEWADDRLABEL, ip6addrlbl_newdel,
591			NULL, NULL);
592	__rtnl_register(PF_INET6, RTM_DELADDRLABEL, ip6addrlbl_newdel,
593			NULL, NULL);
594	__rtnl_register(PF_INET6, RTM_GETADDRLABEL, ip6addrlbl_get,
595			ip6addrlbl_dump, NULL);
596}
597
598