addrlabel.c revision 22c352195ee09dcce9f4f0e2d4cd5f382b90f0fb 
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 (RFC6724 + 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 * fec0::/10		N/A		11		Site-local
73 *							(deprecated by RFC3879)
74 * 3ffe::/16		N/A		12		6bone
75 *
76 * Note: 0xffffffff is used if we do not have any policies.
77 * Note: Labels for ULA and 6to4 are different from labels listed in RFC6724.
78 */
79
80#define IPV6_ADDR_LABEL_DEFAULT	0xffffffffUL
81
82static const __net_initconst 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	},{	/* fec0::/10 */
96		.prefix = &(struct in6_addr){{{ 0xfe, 0xc0 }}},
97		.prefixlen = 10,
98		.label = 11,
99	},{	/* 2002::/16 */
100		.prefix = &(struct in6_addr){{{ 0x20, 0x02 }}},
101		.prefixlen = 16,
102		.label = 2,
103	},{	/* 3ffe::/16 */
104		.prefix = &(struct in6_addr){{{ 0x3f, 0xfe }}},
105		.prefixlen = 16,
106		.label = 12,
107	},{	/* 2001::/32 */
108		.prefix = &(struct in6_addr){{{ 0x20, 0x01 }}},
109		.prefixlen = 32,
110		.label = 6,
111	},{	/* 2001:10::/28 */
112		.prefix = &(struct in6_addr){{{ 0x20, 0x01, 0x00, 0x10 }}},
113		.prefixlen = 28,
114		.label = 7,
115	},{	/* ::ffff:0:0 */
116		.prefix = &(struct in6_addr){{{ [10] = 0xff, [11] = 0xff }}},
117		.prefixlen = 96,
118		.label = 4,
119	},{	/* ::/96 */
120		.prefix = &in6addr_any,
121		.prefixlen = 96,
122		.label = 3,
123	},{	/* ::1/128 */
124		.prefix = &in6addr_loopback,
125		.prefixlen = 128,
126		.label = 0,
127	}
128};
129
130/* Object management */
131static inline void ip6addrlbl_free(struct ip6addrlbl_entry *p)
132{
133#ifdef CONFIG_NET_NS
134	release_net(p->lbl_net);
135#endif
136	kfree(p);
137}
138
139static void ip6addrlbl_free_rcu(struct rcu_head *h)
140{
141	ip6addrlbl_free(container_of(h, struct ip6addrlbl_entry, rcu));
142}
143
144static bool ip6addrlbl_hold(struct ip6addrlbl_entry *p)
145{
146	return atomic_inc_not_zero(&p->refcnt);
147}
148
149static inline void ip6addrlbl_put(struct ip6addrlbl_entry *p)
150{
151	if (atomic_dec_and_test(&p->refcnt))
152		call_rcu(&p->rcu, ip6addrlbl_free_rcu);
153}
154
155/* Find label */
156static bool __ip6addrlbl_match(struct net *net,
157			       const struct ip6addrlbl_entry *p,
158			       const struct in6_addr *addr,
159			       int addrtype, int ifindex)
160{
161	if (!net_eq(ip6addrlbl_net(p), net))
162		return false;
163	if (p->ifindex && p->ifindex != ifindex)
164		return false;
165	if (p->addrtype && p->addrtype != addrtype)
166		return false;
167	if (!ipv6_prefix_equal(addr, &p->prefix, p->prefixlen))
168		return false;
169	return true;
170}
171
172static struct ip6addrlbl_entry *__ipv6_addr_label(struct net *net,
173						  const struct in6_addr *addr,
174						  int type, int ifindex)
175{
176	struct ip6addrlbl_entry *p;
177	hlist_for_each_entry_rcu(p, &ip6addrlbl_table.head, list) {
178		if (__ip6addrlbl_match(net, p, addr, type, ifindex))
179			return p;
180	}
181	return NULL;
182}
183
184u32 ipv6_addr_label(struct net *net,
185		    const struct in6_addr *addr, int type, int ifindex)
186{
187	u32 label;
188	struct ip6addrlbl_entry *p;
189
190	type &= IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK;
191
192	rcu_read_lock();
193	p = __ipv6_addr_label(net, addr, type, ifindex);
194	label = p ? p->label : IPV6_ADDR_LABEL_DEFAULT;
195	rcu_read_unlock();
196
197	ADDRLABEL(KERN_DEBUG "%s(addr=%pI6, type=%d, ifindex=%d) => %08x\n",
198		  __func__, addr, type, ifindex, label);
199
200	return label;
201}
202
203/* allocate one entry */
204static struct ip6addrlbl_entry *ip6addrlbl_alloc(struct net *net,
205						 const struct in6_addr *prefix,
206						 int prefixlen, int ifindex,
207						 u32 label)
208{
209	struct ip6addrlbl_entry *newp;
210	int addrtype;
211
212	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u)\n",
213		  __func__, prefix, prefixlen, ifindex, (unsigned int)label);
214
215	addrtype = ipv6_addr_type(prefix) & (IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK);
216
217	switch (addrtype) {
218	case IPV6_ADDR_MAPPED:
219		if (prefixlen > 96)
220			return ERR_PTR(-EINVAL);
221		if (prefixlen < 96)
222			addrtype = 0;
223		break;
224	case IPV6_ADDR_COMPATv4:
225		if (prefixlen != 96)
226			addrtype = 0;
227		break;
228	case IPV6_ADDR_LOOPBACK:
229		if (prefixlen != 128)
230			addrtype = 0;
231		break;
232	}
233
234	newp = kmalloc(sizeof(*newp), GFP_KERNEL);
235	if (!newp)
236		return ERR_PTR(-ENOMEM);
237
238	ipv6_addr_prefix(&newp->prefix, prefix, prefixlen);
239	newp->prefixlen = prefixlen;
240	newp->ifindex = ifindex;
241	newp->addrtype = addrtype;
242	newp->label = label;
243	INIT_HLIST_NODE(&newp->list);
244#ifdef CONFIG_NET_NS
245	newp->lbl_net = hold_net(net);
246#endif
247	atomic_set(&newp->refcnt, 1);
248	return newp;
249}
250
251/* add a label */
252static int __ip6addrlbl_add(struct ip6addrlbl_entry *newp, int replace)
253{
254	int ret = 0;
255
256	ADDRLABEL(KERN_DEBUG "%s(newp=%p, replace=%d)\n",
257			__func__,
258			newp, replace);
259
260	if (hlist_empty(&ip6addrlbl_table.head)) {
261		hlist_add_head_rcu(&newp->list, &ip6addrlbl_table.head);
262	} else {
263		struct hlist_node *n;
264		struct ip6addrlbl_entry *p = NULL;
265		hlist_for_each_entry_safe(p, n,
266					  &ip6addrlbl_table.head, list) {
267			if (p->prefixlen == newp->prefixlen &&
268			    net_eq(ip6addrlbl_net(p), ip6addrlbl_net(newp)) &&
269			    p->ifindex == newp->ifindex &&
270			    ipv6_addr_equal(&p->prefix, &newp->prefix)) {
271				if (!replace) {
272					ret = -EEXIST;
273					goto out;
274				}
275				hlist_replace_rcu(&p->list, &newp->list);
276				ip6addrlbl_put(p);
277				goto out;
278			} else if ((p->prefixlen == newp->prefixlen && !p->ifindex) ||
279				   (p->prefixlen < newp->prefixlen)) {
280				hlist_add_before_rcu(&newp->list, &p->list);
281				goto out;
282			}
283		}
284		hlist_add_after_rcu(&p->list, &newp->list);
285	}
286out:
287	if (!ret)
288		ip6addrlbl_table.seq++;
289	return ret;
290}
291
292/* add a label */
293static int ip6addrlbl_add(struct net *net,
294			  const struct in6_addr *prefix, int prefixlen,
295			  int ifindex, u32 label, int replace)
296{
297	struct ip6addrlbl_entry *newp;
298	int ret = 0;
299
300	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u, replace=%d)\n",
301		  __func__, prefix, prefixlen, ifindex, (unsigned int)label,
302		  replace);
303
304	newp = ip6addrlbl_alloc(net, prefix, prefixlen, ifindex, label);
305	if (IS_ERR(newp))
306		return PTR_ERR(newp);
307	spin_lock(&ip6addrlbl_table.lock);
308	ret = __ip6addrlbl_add(newp, replace);
309	spin_unlock(&ip6addrlbl_table.lock);
310	if (ret)
311		ip6addrlbl_free(newp);
312	return ret;
313}
314
315/* remove a label */
316static int __ip6addrlbl_del(struct net *net,
317			    const struct in6_addr *prefix, int prefixlen,
318			    int ifindex)
319{
320	struct ip6addrlbl_entry *p = NULL;
321	struct hlist_node *n;
322	int ret = -ESRCH;
323
324	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
325		  __func__, prefix, prefixlen, ifindex);
326
327	hlist_for_each_entry_safe(p, n, &ip6addrlbl_table.head, list) {
328		if (p->prefixlen == prefixlen &&
329		    net_eq(ip6addrlbl_net(p), net) &&
330		    p->ifindex == ifindex &&
331		    ipv6_addr_equal(&p->prefix, prefix)) {
332			hlist_del_rcu(&p->list);
333			ip6addrlbl_put(p);
334			ret = 0;
335			break;
336		}
337	}
338	return ret;
339}
340
341static int ip6addrlbl_del(struct net *net,
342			  const struct in6_addr *prefix, int prefixlen,
343			  int ifindex)
344{
345	struct in6_addr prefix_buf;
346	int ret;
347
348	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
349		  __func__, prefix, prefixlen, ifindex);
350
351	ipv6_addr_prefix(&prefix_buf, prefix, prefixlen);
352	spin_lock(&ip6addrlbl_table.lock);
353	ret = __ip6addrlbl_del(net, &prefix_buf, prefixlen, ifindex);
354	spin_unlock(&ip6addrlbl_table.lock);
355	return ret;
356}
357
358/* add default label */
359static int __net_init ip6addrlbl_net_init(struct net *net)
360{
361	int err = 0;
362	int i;
363
364	ADDRLABEL(KERN_DEBUG "%s\n", __func__);
365
366	for (i = 0; i < ARRAY_SIZE(ip6addrlbl_init_table); i++) {
367		int ret = ip6addrlbl_add(net,
368					 ip6addrlbl_init_table[i].prefix,
369					 ip6addrlbl_init_table[i].prefixlen,
370					 0,
371					 ip6addrlbl_init_table[i].label, 0);
372		/* XXX: should we free all rules when we catch an error? */
373		if (ret && (!err || err != -ENOMEM))
374			err = ret;
375	}
376	return err;
377}
378
379static void __net_exit ip6addrlbl_net_exit(struct net *net)
380{
381	struct ip6addrlbl_entry *p = NULL;
382	struct hlist_node *n;
383
384	/* Remove all labels belonging to the exiting net */
385	spin_lock(&ip6addrlbl_table.lock);
386	hlist_for_each_entry_safe(p, n, &ip6addrlbl_table.head, list) {
387		if (net_eq(ip6addrlbl_net(p), net)) {
388			hlist_del_rcu(&p->list);
389			ip6addrlbl_put(p);
390		}
391	}
392	spin_unlock(&ip6addrlbl_table.lock);
393}
394
395static struct pernet_operations ipv6_addr_label_ops = {
396	.init = ip6addrlbl_net_init,
397	.exit = ip6addrlbl_net_exit,
398};
399
400int __init ipv6_addr_label_init(void)
401{
402	spin_lock_init(&ip6addrlbl_table.lock);
403
404	return register_pernet_subsys(&ipv6_addr_label_ops);
405}
406
407void ipv6_addr_label_cleanup(void)
408{
409	unregister_pernet_subsys(&ipv6_addr_label_ops);
410}
411
412static const struct nla_policy ifal_policy[IFAL_MAX+1] = {
413	[IFAL_ADDRESS]		= { .len = sizeof(struct in6_addr), },
414	[IFAL_LABEL]		= { .len = sizeof(u32), },
415};
416
417static int ip6addrlbl_newdel(struct sk_buff *skb, struct nlmsghdr *nlh,
418			     void *arg)
419{
420	struct net *net = sock_net(skb->sk);
421	struct ifaddrlblmsg *ifal;
422	struct nlattr *tb[IFAL_MAX+1];
423	struct in6_addr *pfx;
424	u32 label;
425	int err = 0;
426
427	err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
428	if (err < 0)
429		return err;
430
431	ifal = nlmsg_data(nlh);
432
433	if (ifal->ifal_family != AF_INET6 ||
434	    ifal->ifal_prefixlen > 128)
435		return -EINVAL;
436
437	if (!tb[IFAL_ADDRESS])
438		return -EINVAL;
439	pfx = nla_data(tb[IFAL_ADDRESS]);
440
441	if (!tb[IFAL_LABEL])
442		return -EINVAL;
443	label = nla_get_u32(tb[IFAL_LABEL]);
444	if (label == IPV6_ADDR_LABEL_DEFAULT)
445		return -EINVAL;
446
447	switch(nlh->nlmsg_type) {
448	case RTM_NEWADDRLABEL:
449		if (ifal->ifal_index &&
450		    !__dev_get_by_index(net, ifal->ifal_index))
451			return -EINVAL;
452
453		err = ip6addrlbl_add(net, pfx, ifal->ifal_prefixlen,
454				     ifal->ifal_index, label,
455				     nlh->nlmsg_flags & NLM_F_REPLACE);
456		break;
457	case RTM_DELADDRLABEL:
458		err = ip6addrlbl_del(net, pfx, ifal->ifal_prefixlen,
459				     ifal->ifal_index);
460		break;
461	default:
462		err = -EOPNOTSUPP;
463	}
464	return err;
465}
466
467static void ip6addrlbl_putmsg(struct nlmsghdr *nlh,
468			      int prefixlen, int ifindex, u32 lseq)
469{
470	struct ifaddrlblmsg *ifal = nlmsg_data(nlh);
471	ifal->ifal_family = AF_INET6;
472	ifal->ifal_prefixlen = prefixlen;
473	ifal->ifal_flags = 0;
474	ifal->ifal_index = ifindex;
475	ifal->ifal_seq = lseq;
476};
477
478static int ip6addrlbl_fill(struct sk_buff *skb,
479			   struct ip6addrlbl_entry *p,
480			   u32 lseq,
481			   u32 portid, u32 seq, int event,
482			   unsigned int flags)
483{
484	struct nlmsghdr *nlh = nlmsg_put(skb, portid, seq, event,
485					 sizeof(struct ifaddrlblmsg), flags);
486	if (!nlh)
487		return -EMSGSIZE;
488
489	ip6addrlbl_putmsg(nlh, p->prefixlen, p->ifindex, lseq);
490
491	if (nla_put(skb, IFAL_ADDRESS, 16, &p->prefix) < 0 ||
492	    nla_put_u32(skb, IFAL_LABEL, p->label) < 0) {
493		nlmsg_cancel(skb, nlh);
494		return -EMSGSIZE;
495	}
496
497	return nlmsg_end(skb, nlh);
498}
499
500static int ip6addrlbl_dump(struct sk_buff *skb, struct netlink_callback *cb)
501{
502	struct net *net = sock_net(skb->sk);
503	struct ip6addrlbl_entry *p;
504	int idx = 0, s_idx = cb->args[0];
505	int err;
506
507	rcu_read_lock();
508	hlist_for_each_entry_rcu(p, &ip6addrlbl_table.head, list) {
509		if (idx >= s_idx &&
510		    net_eq(ip6addrlbl_net(p), net)) {
511			if ((err = ip6addrlbl_fill(skb, p,
512						   ip6addrlbl_table.seq,
513						   NETLINK_CB(cb->skb).portid,
514						   cb->nlh->nlmsg_seq,
515						   RTM_NEWADDRLABEL,
516						   NLM_F_MULTI)) <= 0)
517				break;
518		}
519		idx++;
520	}
521	rcu_read_unlock();
522	cb->args[0] = idx;
523	return skb->len;
524}
525
526static inline int ip6addrlbl_msgsize(void)
527{
528	return NLMSG_ALIGN(sizeof(struct ifaddrlblmsg))
529		+ nla_total_size(16)	/* IFAL_ADDRESS */
530		+ nla_total_size(4);	/* IFAL_LABEL */
531}
532
533static int ip6addrlbl_get(struct sk_buff *in_skb, struct nlmsghdr* nlh,
534			  void *arg)
535{
536	struct net *net = sock_net(in_skb->sk);
537	struct ifaddrlblmsg *ifal;
538	struct nlattr *tb[IFAL_MAX+1];
539	struct in6_addr *addr;
540	u32 lseq;
541	int err = 0;
542	struct ip6addrlbl_entry *p;
543	struct sk_buff *skb;
544
545	err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
546	if (err < 0)
547		return err;
548
549	ifal = nlmsg_data(nlh);
550
551	if (ifal->ifal_family != AF_INET6 ||
552	    ifal->ifal_prefixlen != 128)
553		return -EINVAL;
554
555	if (ifal->ifal_index &&
556	    !__dev_get_by_index(net, ifal->ifal_index))
557		return -EINVAL;
558
559	if (!tb[IFAL_ADDRESS])
560		return -EINVAL;
561	addr = nla_data(tb[IFAL_ADDRESS]);
562
563	rcu_read_lock();
564	p = __ipv6_addr_label(net, addr, ipv6_addr_type(addr), ifal->ifal_index);
565	if (p && ip6addrlbl_hold(p))
566		p = NULL;
567	lseq = ip6addrlbl_table.seq;
568	rcu_read_unlock();
569
570	if (!p) {
571		err = -ESRCH;
572		goto out;
573	}
574
575	if (!(skb = nlmsg_new(ip6addrlbl_msgsize(), GFP_KERNEL))) {
576		ip6addrlbl_put(p);
577		return -ENOBUFS;
578	}
579
580	err = ip6addrlbl_fill(skb, p, lseq,
581			      NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
582			      RTM_NEWADDRLABEL, 0);
583
584	ip6addrlbl_put(p);
585
586	if (err < 0) {
587		WARN_ON(err == -EMSGSIZE);
588		kfree_skb(skb);
589		goto out;
590	}
591
592	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
593out:
594	return err;
595}
596
597void __init ipv6_addr_label_rtnl_register(void)
598{
599	__rtnl_register(PF_INET6, RTM_NEWADDRLABEL, ip6addrlbl_newdel,
600			NULL, NULL);
601	__rtnl_register(PF_INET6, RTM_DELADDRLABEL, ip6addrlbl_newdel,
602			NULL, NULL);
603	__rtnl_register(PF_INET6, RTM_GETADDRLABEL, ip6addrlbl_get,
604			ip6addrlbl_dump, NULL);
605}
606
607