addrlabel.c revision dfc47ef8639facd77210e74be831943c2fdd9c74 
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 hlist_node *pos;
177	struct ip6addrlbl_entry *p;
178	hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
179		if (__ip6addrlbl_match(net, p, addr, type, ifindex))
180			return p;
181	}
182	return NULL;
183}
184
185u32 ipv6_addr_label(struct net *net,
186		    const struct in6_addr *addr, int type, int ifindex)
187{
188	u32 label;
189	struct ip6addrlbl_entry *p;
190
191	type &= IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK;
192
193	rcu_read_lock();
194	p = __ipv6_addr_label(net, addr, type, ifindex);
195	label = p ? p->label : IPV6_ADDR_LABEL_DEFAULT;
196	rcu_read_unlock();
197
198	ADDRLABEL(KERN_DEBUG "%s(addr=%pI6, type=%d, ifindex=%d) => %08x\n",
199		  __func__, addr, type, ifindex, label);
200
201	return label;
202}
203
204/* allocate one entry */
205static struct ip6addrlbl_entry *ip6addrlbl_alloc(struct net *net,
206						 const struct in6_addr *prefix,
207						 int prefixlen, int ifindex,
208						 u32 label)
209{
210	struct ip6addrlbl_entry *newp;
211	int addrtype;
212
213	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u)\n",
214		  __func__, prefix, prefixlen, ifindex, (unsigned int)label);
215
216	addrtype = ipv6_addr_type(prefix) & (IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK);
217
218	switch (addrtype) {
219	case IPV6_ADDR_MAPPED:
220		if (prefixlen > 96)
221			return ERR_PTR(-EINVAL);
222		if (prefixlen < 96)
223			addrtype = 0;
224		break;
225	case IPV6_ADDR_COMPATv4:
226		if (prefixlen != 96)
227			addrtype = 0;
228		break;
229	case IPV6_ADDR_LOOPBACK:
230		if (prefixlen != 128)
231			addrtype = 0;
232		break;
233	}
234
235	newp = kmalloc(sizeof(*newp), GFP_KERNEL);
236	if (!newp)
237		return ERR_PTR(-ENOMEM);
238
239	ipv6_addr_prefix(&newp->prefix, prefix, prefixlen);
240	newp->prefixlen = prefixlen;
241	newp->ifindex = ifindex;
242	newp->addrtype = addrtype;
243	newp->label = label;
244	INIT_HLIST_NODE(&newp->list);
245#ifdef CONFIG_NET_NS
246	newp->lbl_net = hold_net(net);
247#endif
248	atomic_set(&newp->refcnt, 1);
249	return newp;
250}
251
252/* add a label */
253static int __ip6addrlbl_add(struct ip6addrlbl_entry *newp, int replace)
254{
255	int ret = 0;
256
257	ADDRLABEL(KERN_DEBUG "%s(newp=%p, replace=%d)\n",
258			__func__,
259			newp, replace);
260
261	if (hlist_empty(&ip6addrlbl_table.head)) {
262		hlist_add_head_rcu(&newp->list, &ip6addrlbl_table.head);
263	} else {
264		struct hlist_node *pos, *n;
265		struct ip6addrlbl_entry *p = NULL;
266		hlist_for_each_entry_safe(p, pos, n,
267					  &ip6addrlbl_table.head, list) {
268			if (p->prefixlen == newp->prefixlen &&
269			    net_eq(ip6addrlbl_net(p), ip6addrlbl_net(newp)) &&
270			    p->ifindex == newp->ifindex &&
271			    ipv6_addr_equal(&p->prefix, &newp->prefix)) {
272				if (!replace) {
273					ret = -EEXIST;
274					goto out;
275				}
276				hlist_replace_rcu(&p->list, &newp->list);
277				ip6addrlbl_put(p);
278				goto out;
279			} else if ((p->prefixlen == newp->prefixlen && !p->ifindex) ||
280				   (p->prefixlen < newp->prefixlen)) {
281				hlist_add_before_rcu(&newp->list, &p->list);
282				goto out;
283			}
284		}
285		hlist_add_after_rcu(&p->list, &newp->list);
286	}
287out:
288	if (!ret)
289		ip6addrlbl_table.seq++;
290	return ret;
291}
292
293/* add a label */
294static int ip6addrlbl_add(struct net *net,
295			  const struct in6_addr *prefix, int prefixlen,
296			  int ifindex, u32 label, int replace)
297{
298	struct ip6addrlbl_entry *newp;
299	int ret = 0;
300
301	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u, replace=%d)\n",
302		  __func__, prefix, prefixlen, ifindex, (unsigned int)label,
303		  replace);
304
305	newp = ip6addrlbl_alloc(net, prefix, prefixlen, ifindex, label);
306	if (IS_ERR(newp))
307		return PTR_ERR(newp);
308	spin_lock(&ip6addrlbl_table.lock);
309	ret = __ip6addrlbl_add(newp, replace);
310	spin_unlock(&ip6addrlbl_table.lock);
311	if (ret)
312		ip6addrlbl_free(newp);
313	return ret;
314}
315
316/* remove a label */
317static int __ip6addrlbl_del(struct net *net,
318			    const struct in6_addr *prefix, int prefixlen,
319			    int ifindex)
320{
321	struct ip6addrlbl_entry *p = NULL;
322	struct hlist_node *pos, *n;
323	int ret = -ESRCH;
324
325	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
326		  __func__, prefix, prefixlen, ifindex);
327
328	hlist_for_each_entry_safe(p, pos, n, &ip6addrlbl_table.head, list) {
329		if (p->prefixlen == prefixlen &&
330		    net_eq(ip6addrlbl_net(p), net) &&
331		    p->ifindex == ifindex &&
332		    ipv6_addr_equal(&p->prefix, prefix)) {
333			hlist_del_rcu(&p->list);
334			ip6addrlbl_put(p);
335			ret = 0;
336			break;
337		}
338	}
339	return ret;
340}
341
342static int ip6addrlbl_del(struct net *net,
343			  const struct in6_addr *prefix, int prefixlen,
344			  int ifindex)
345{
346	struct in6_addr prefix_buf;
347	int ret;
348
349	ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
350		  __func__, prefix, prefixlen, ifindex);
351
352	ipv6_addr_prefix(&prefix_buf, prefix, prefixlen);
353	spin_lock(&ip6addrlbl_table.lock);
354	ret = __ip6addrlbl_del(net, &prefix_buf, prefixlen, ifindex);
355	spin_unlock(&ip6addrlbl_table.lock);
356	return ret;
357}
358
359/* add default label */
360static int __net_init ip6addrlbl_net_init(struct net *net)
361{
362	int err = 0;
363	int i;
364
365	ADDRLABEL(KERN_DEBUG "%s\n", __func__);
366
367	for (i = 0; i < ARRAY_SIZE(ip6addrlbl_init_table); i++) {
368		int ret = ip6addrlbl_add(net,
369					 ip6addrlbl_init_table[i].prefix,
370					 ip6addrlbl_init_table[i].prefixlen,
371					 0,
372					 ip6addrlbl_init_table[i].label, 0);
373		/* XXX: should we free all rules when we catch an error? */
374		if (ret && (!err || err != -ENOMEM))
375			err = ret;
376	}
377	return err;
378}
379
380static void __net_exit ip6addrlbl_net_exit(struct net *net)
381{
382	struct ip6addrlbl_entry *p = NULL;
383	struct hlist_node *pos, *n;
384
385	/* Remove all labels belonging to the exiting net */
386	spin_lock(&ip6addrlbl_table.lock);
387	hlist_for_each_entry_safe(p, pos, n, &ip6addrlbl_table.head, list) {
388		if (net_eq(ip6addrlbl_net(p), net)) {
389			hlist_del_rcu(&p->list);
390			ip6addrlbl_put(p);
391		}
392	}
393	spin_unlock(&ip6addrlbl_table.lock);
394}
395
396static struct pernet_operations ipv6_addr_label_ops = {
397	.init = ip6addrlbl_net_init,
398	.exit = ip6addrlbl_net_exit,
399};
400
401int __init ipv6_addr_label_init(void)
402{
403	spin_lock_init(&ip6addrlbl_table.lock);
404
405	return register_pernet_subsys(&ipv6_addr_label_ops);
406}
407
408void ipv6_addr_label_cleanup(void)
409{
410	unregister_pernet_subsys(&ipv6_addr_label_ops);
411}
412
413static const struct nla_policy ifal_policy[IFAL_MAX+1] = {
414	[IFAL_ADDRESS]		= { .len = sizeof(struct in6_addr), },
415	[IFAL_LABEL]		= { .len = sizeof(u32), },
416};
417
418static int ip6addrlbl_newdel(struct sk_buff *skb, struct nlmsghdr *nlh,
419			     void *arg)
420{
421	struct net *net = sock_net(skb->sk);
422	struct ifaddrlblmsg *ifal;
423	struct nlattr *tb[IFAL_MAX+1];
424	struct in6_addr *pfx;
425	u32 label;
426	int err = 0;
427
428	if (!capable(CAP_NET_ADMIN))
429		return -EPERM;
430
431	err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
432	if (err < 0)
433		return err;
434
435	ifal = nlmsg_data(nlh);
436
437	if (ifal->ifal_family != AF_INET6 ||
438	    ifal->ifal_prefixlen > 128)
439		return -EINVAL;
440
441	if (!tb[IFAL_ADDRESS])
442		return -EINVAL;
443
444	pfx = nla_data(tb[IFAL_ADDRESS]);
445	if (!pfx)
446		return -EINVAL;
447
448	if (!tb[IFAL_LABEL])
449		return -EINVAL;
450	label = nla_get_u32(tb[IFAL_LABEL]);
451	if (label == IPV6_ADDR_LABEL_DEFAULT)
452		return -EINVAL;
453
454	switch(nlh->nlmsg_type) {
455	case RTM_NEWADDRLABEL:
456		if (ifal->ifal_index &&
457		    !__dev_get_by_index(net, ifal->ifal_index))
458			return -EINVAL;
459
460		err = ip6addrlbl_add(net, pfx, ifal->ifal_prefixlen,
461				     ifal->ifal_index, label,
462				     nlh->nlmsg_flags & NLM_F_REPLACE);
463		break;
464	case RTM_DELADDRLABEL:
465		err = ip6addrlbl_del(net, pfx, ifal->ifal_prefixlen,
466				     ifal->ifal_index);
467		break;
468	default:
469		err = -EOPNOTSUPP;
470	}
471	return err;
472}
473
474static void ip6addrlbl_putmsg(struct nlmsghdr *nlh,
475			      int prefixlen, int ifindex, u32 lseq)
476{
477	struct ifaddrlblmsg *ifal = nlmsg_data(nlh);
478	ifal->ifal_family = AF_INET6;
479	ifal->ifal_prefixlen = prefixlen;
480	ifal->ifal_flags = 0;
481	ifal->ifal_index = ifindex;
482	ifal->ifal_seq = lseq;
483};
484
485static int ip6addrlbl_fill(struct sk_buff *skb,
486			   struct ip6addrlbl_entry *p,
487			   u32 lseq,
488			   u32 portid, u32 seq, int event,
489			   unsigned int flags)
490{
491	struct nlmsghdr *nlh = nlmsg_put(skb, portid, seq, event,
492					 sizeof(struct ifaddrlblmsg), flags);
493	if (!nlh)
494		return -EMSGSIZE;
495
496	ip6addrlbl_putmsg(nlh, p->prefixlen, p->ifindex, lseq);
497
498	if (nla_put(skb, IFAL_ADDRESS, 16, &p->prefix) < 0 ||
499	    nla_put_u32(skb, IFAL_LABEL, p->label) < 0) {
500		nlmsg_cancel(skb, nlh);
501		return -EMSGSIZE;
502	}
503
504	return nlmsg_end(skb, nlh);
505}
506
507static int ip6addrlbl_dump(struct sk_buff *skb, struct netlink_callback *cb)
508{
509	struct net *net = sock_net(skb->sk);
510	struct ip6addrlbl_entry *p;
511	struct hlist_node *pos;
512	int idx = 0, s_idx = cb->args[0];
513	int err;
514
515	rcu_read_lock();
516	hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
517		if (idx >= s_idx &&
518		    net_eq(ip6addrlbl_net(p), net)) {
519			if ((err = ip6addrlbl_fill(skb, p,
520						   ip6addrlbl_table.seq,
521						   NETLINK_CB(cb->skb).portid,
522						   cb->nlh->nlmsg_seq,
523						   RTM_NEWADDRLABEL,
524						   NLM_F_MULTI)) <= 0)
525				break;
526		}
527		idx++;
528	}
529	rcu_read_unlock();
530	cb->args[0] = idx;
531	return skb->len;
532}
533
534static inline int ip6addrlbl_msgsize(void)
535{
536	return NLMSG_ALIGN(sizeof(struct ifaddrlblmsg))
537		+ nla_total_size(16)	/* IFAL_ADDRESS */
538		+ nla_total_size(4);	/* IFAL_LABEL */
539}
540
541static int ip6addrlbl_get(struct sk_buff *in_skb, struct nlmsghdr* nlh,
542			  void *arg)
543{
544	struct net *net = sock_net(in_skb->sk);
545	struct ifaddrlblmsg *ifal;
546	struct nlattr *tb[IFAL_MAX+1];
547	struct in6_addr *addr;
548	u32 lseq;
549	int err = 0;
550	struct ip6addrlbl_entry *p;
551	struct sk_buff *skb;
552
553	err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
554	if (err < 0)
555		return err;
556
557	ifal = nlmsg_data(nlh);
558
559	if (ifal->ifal_family != AF_INET6 ||
560	    ifal->ifal_prefixlen != 128)
561		return -EINVAL;
562
563	if (ifal->ifal_index &&
564	    !__dev_get_by_index(net, ifal->ifal_index))
565		return -EINVAL;
566
567	if (!tb[IFAL_ADDRESS])
568		return -EINVAL;
569
570	addr = nla_data(tb[IFAL_ADDRESS]);
571	if (!addr)
572		return -EINVAL;
573
574	rcu_read_lock();
575	p = __ipv6_addr_label(net, addr, ipv6_addr_type(addr), ifal->ifal_index);
576	if (p && ip6addrlbl_hold(p))
577		p = NULL;
578	lseq = ip6addrlbl_table.seq;
579	rcu_read_unlock();
580
581	if (!p) {
582		err = -ESRCH;
583		goto out;
584	}
585
586	if (!(skb = nlmsg_new(ip6addrlbl_msgsize(), GFP_KERNEL))) {
587		ip6addrlbl_put(p);
588		return -ENOBUFS;
589	}
590
591	err = ip6addrlbl_fill(skb, p, lseq,
592			      NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
593			      RTM_NEWADDRLABEL, 0);
594
595	ip6addrlbl_put(p);
596
597	if (err < 0) {
598		WARN_ON(err == -EMSGSIZE);
599		kfree_skb(skb);
600		goto out;
601	}
602
603	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
604out:
605	return err;
606}
607
608void __init ipv6_addr_label_rtnl_register(void)
609{
610	__rtnl_register(PF_INET6, RTM_NEWADDRLABEL, ip6addrlbl_newdel,
611			NULL, NULL);
612	__rtnl_register(PF_INET6, RTM_DELADDRLABEL, ip6addrlbl_newdel,
613			NULL, NULL);
614	__rtnl_register(PF_INET6, RTM_GETADDRLABEL, ip6addrlbl_get,
615			ip6addrlbl_dump, NULL);
616}
617
618