addrlabel.c revision b51642f6d77b131dc85d1d71029c3cbb5b07c262 
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	err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
429	if (err < 0)
430		return err;
431
432	ifal = nlmsg_data(nlh);
433
434	if (ifal->ifal_family != AF_INET6 ||
435	    ifal->ifal_prefixlen > 128)
436		return -EINVAL;
437
438	if (!tb[IFAL_ADDRESS])
439		return -EINVAL;
440
441	pfx = nla_data(tb[IFAL_ADDRESS]);
442	if (!pfx)
443		return -EINVAL;
444
445	if (!tb[IFAL_LABEL])
446		return -EINVAL;
447	label = nla_get_u32(tb[IFAL_LABEL]);
448	if (label == IPV6_ADDR_LABEL_DEFAULT)
449		return -EINVAL;
450
451	switch(nlh->nlmsg_type) {
452	case RTM_NEWADDRLABEL:
453		if (ifal->ifal_index &&
454		    !__dev_get_by_index(net, ifal->ifal_index))
455			return -EINVAL;
456
457		err = ip6addrlbl_add(net, pfx, ifal->ifal_prefixlen,
458				     ifal->ifal_index, label,
459				     nlh->nlmsg_flags & NLM_F_REPLACE);
460		break;
461	case RTM_DELADDRLABEL:
462		err = ip6addrlbl_del(net, pfx, ifal->ifal_prefixlen,
463				     ifal->ifal_index);
464		break;
465	default:
466		err = -EOPNOTSUPP;
467	}
468	return err;
469}
470
471static void ip6addrlbl_putmsg(struct nlmsghdr *nlh,
472			      int prefixlen, int ifindex, u32 lseq)
473{
474	struct ifaddrlblmsg *ifal = nlmsg_data(nlh);
475	ifal->ifal_family = AF_INET6;
476	ifal->ifal_prefixlen = prefixlen;
477	ifal->ifal_flags = 0;
478	ifal->ifal_index = ifindex;
479	ifal->ifal_seq = lseq;
480};
481
482static int ip6addrlbl_fill(struct sk_buff *skb,
483			   struct ip6addrlbl_entry *p,
484			   u32 lseq,
485			   u32 portid, u32 seq, int event,
486			   unsigned int flags)
487{
488	struct nlmsghdr *nlh = nlmsg_put(skb, portid, seq, event,
489					 sizeof(struct ifaddrlblmsg), flags);
490	if (!nlh)
491		return -EMSGSIZE;
492
493	ip6addrlbl_putmsg(nlh, p->prefixlen, p->ifindex, lseq);
494
495	if (nla_put(skb, IFAL_ADDRESS, 16, &p->prefix) < 0 ||
496	    nla_put_u32(skb, IFAL_LABEL, p->label) < 0) {
497		nlmsg_cancel(skb, nlh);
498		return -EMSGSIZE;
499	}
500
501	return nlmsg_end(skb, nlh);
502}
503
504static int ip6addrlbl_dump(struct sk_buff *skb, struct netlink_callback *cb)
505{
506	struct net *net = sock_net(skb->sk);
507	struct ip6addrlbl_entry *p;
508	struct hlist_node *pos;
509	int idx = 0, s_idx = cb->args[0];
510	int err;
511
512	rcu_read_lock();
513	hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
514		if (idx >= s_idx &&
515		    net_eq(ip6addrlbl_net(p), net)) {
516			if ((err = ip6addrlbl_fill(skb, p,
517						   ip6addrlbl_table.seq,
518						   NETLINK_CB(cb->skb).portid,
519						   cb->nlh->nlmsg_seq,
520						   RTM_NEWADDRLABEL,
521						   NLM_F_MULTI)) <= 0)
522				break;
523		}
524		idx++;
525	}
526	rcu_read_unlock();
527	cb->args[0] = idx;
528	return skb->len;
529}
530
531static inline int ip6addrlbl_msgsize(void)
532{
533	return NLMSG_ALIGN(sizeof(struct ifaddrlblmsg))
534		+ nla_total_size(16)	/* IFAL_ADDRESS */
535		+ nla_total_size(4);	/* IFAL_LABEL */
536}
537
538static int ip6addrlbl_get(struct sk_buff *in_skb, struct nlmsghdr* nlh,
539			  void *arg)
540{
541	struct net *net = sock_net(in_skb->sk);
542	struct ifaddrlblmsg *ifal;
543	struct nlattr *tb[IFAL_MAX+1];
544	struct in6_addr *addr;
545	u32 lseq;
546	int err = 0;
547	struct ip6addrlbl_entry *p;
548	struct sk_buff *skb;
549
550	err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
551	if (err < 0)
552		return err;
553
554	ifal = nlmsg_data(nlh);
555
556	if (ifal->ifal_family != AF_INET6 ||
557	    ifal->ifal_prefixlen != 128)
558		return -EINVAL;
559
560	if (ifal->ifal_index &&
561	    !__dev_get_by_index(net, ifal->ifal_index))
562		return -EINVAL;
563
564	if (!tb[IFAL_ADDRESS])
565		return -EINVAL;
566
567	addr = nla_data(tb[IFAL_ADDRESS]);
568	if (!addr)
569		return -EINVAL;
570
571	rcu_read_lock();
572	p = __ipv6_addr_label(net, addr, ipv6_addr_type(addr), ifal->ifal_index);
573	if (p && ip6addrlbl_hold(p))
574		p = NULL;
575	lseq = ip6addrlbl_table.seq;
576	rcu_read_unlock();
577
578	if (!p) {
579		err = -ESRCH;
580		goto out;
581	}
582
583	if (!(skb = nlmsg_new(ip6addrlbl_msgsize(), GFP_KERNEL))) {
584		ip6addrlbl_put(p);
585		return -ENOBUFS;
586	}
587
588	err = ip6addrlbl_fill(skb, p, lseq,
589			      NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
590			      RTM_NEWADDRLABEL, 0);
591
592	ip6addrlbl_put(p);
593
594	if (err < 0) {
595		WARN_ON(err == -EMSGSIZE);
596		kfree_skb(skb);
597		goto out;
598	}
599
600	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
601out:
602	return err;
603}
604
605void __init ipv6_addr_label_rtnl_register(void)
606{
607	__rtnl_register(PF_INET6, RTM_NEWADDRLABEL, ip6addrlbl_newdel,
608			NULL, NULL);
609	__rtnl_register(PF_INET6, RTM_DELADDRLABEL, ip6addrlbl_newdel,
610			NULL, NULL);
611	__rtnl_register(PF_INET6, RTM_GETADDRLABEL, ip6addrlbl_get,
612			ip6addrlbl_dump, NULL);
613}
614
615