[IPV6] ADDRCONF: Support RFC3484 configurable address selection policy table.
[linux-2.6] / net / ipv6 / addrlabel.c
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 <net/addrconf.h>
17 #include <linux/if_addrlabel.h>
18 #include <linux/netlink.h>
19 #include <linux/rtnetlink.h>
20
21 #if 0
22 #define ADDRLABEL(x...) printk(x)
23 #else
24 #define ADDRLABEL(x...) do { ; } while(0)
25 #endif
26
27 /*
28  * Policy Table
29  */
30 struct ip6addrlbl_entry
31 {
32         struct in6_addr prefix;
33         int prefixlen;
34         int ifindex;
35         int addrtype;
36         u32 label;
37         struct hlist_node list;
38         atomic_t refcnt;
39         struct rcu_head rcu;
40 };
41
42 static struct ip6addrlbl_table
43 {
44         struct hlist_head head;
45         spinlock_t lock;
46         u32 seq;
47 } ip6addrlbl_table;
48
49 /*
50  * Default policy table (RFC3484 + extensions)
51  *
52  * prefix               addr_type       label
53  * -------------------------------------------------------------------------
54  * ::1/128              LOOPBACK        0
55  * ::/0                 N/A             1
56  * 2002::/16            N/A             2
57  * ::/96                COMPATv4        3
58  * ::ffff:0:0/96        V4MAPPED        4
59  * fc00::/7             N/A             5               ULA (RFC 4193)
60  * 2001::/32            N/A             6               Teredo (RFC 4380)
61  *
62  * Note: 0xffffffff is used if we do not have any policies.
63  */
64
65 #define IPV6_ADDR_LABEL_DEFAULT 0xffffffffUL
66
67 static const __initdata struct ip6addrlbl_init_table
68 {
69         const struct in6_addr *prefix;
70         int prefixlen;
71         u32 label;
72 } ip6addrlbl_init_table[] = {
73         {       /* ::/0 */
74                 .prefix = &in6addr_any,
75                 .label = 1,
76         },{     /* fc00::/7 */
77                 .prefix = &(struct in6_addr){{{ 0xfc }}},
78                 .prefixlen = 7,
79                 .label = 5,
80         },{     /* 2002::/16 */
81                 .prefix = &(struct in6_addr){{{ 0x20, 0x02 }}},
82                 .prefixlen = 16,
83                 .label = 2,
84         },{     /* 2001::/32 */
85                 .prefix = &(struct in6_addr){{{ 0x20, 0x01 }}},
86                 .prefixlen = 32,
87                 .label = 6,
88         },{     /* ::ffff:0:0 */
89                 .prefix = &(struct in6_addr){{{ [10] = 0xff, [11] = 0xff }}},
90                 .prefixlen = 96,
91                 .label = 4,
92         },{     /* ::/96 */
93                 .prefix = &in6addr_any,
94                 .prefixlen = 96,
95                 .label = 3,
96         },{     /* ::1/128 */
97                 .prefix = &in6addr_loopback,
98                 .prefixlen = 128,
99                 .label = 0,
100         }
101 };
102
103 /* Object management */
104 static inline void ip6addrlbl_free(struct ip6addrlbl_entry *p)
105 {
106         kfree(p);
107 }
108
109 static inline int ip6addrlbl_hold(struct ip6addrlbl_entry *p)
110 {
111         return atomic_inc_not_zero(&p->refcnt);
112 }
113
114 static inline void ip6addrlbl_put(struct ip6addrlbl_entry *p)
115 {
116         if (atomic_dec_and_test(&p->refcnt))
117                 ip6addrlbl_free(p);
118 }
119
120 static void ip6addrlbl_free_rcu(struct rcu_head *h)
121 {
122         ip6addrlbl_free(container_of(h, struct ip6addrlbl_entry, rcu));
123 }
124
125 /* Find label */
126 static int __ip6addrlbl_match(struct ip6addrlbl_entry *p,
127                               const struct in6_addr *addr,
128                               int addrtype, int ifindex)
129 {
130         if (p->ifindex && p->ifindex != ifindex)
131                 return 0;
132         if (p->addrtype && p->addrtype != addrtype)
133                 return 0;
134         if (!ipv6_prefix_equal(addr, &p->prefix, p->prefixlen))
135                 return 0;
136         return 1;
137 }
138
139 static struct ip6addrlbl_entry *__ipv6_addr_label(const struct in6_addr *addr,
140                                                   int type, int ifindex)
141 {
142         struct hlist_node *pos;
143         struct ip6addrlbl_entry *p;
144         hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
145                 if (__ip6addrlbl_match(p, addr, type, ifindex))
146                         return p;
147         }
148         return NULL;
149 }
150
151 u32 ipv6_addr_label(const struct in6_addr *addr, int type, int ifindex)
152 {
153         u32 label;
154         struct ip6addrlbl_entry *p;
155
156         type &= IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK;
157
158         rcu_read_lock();
159         p = __ipv6_addr_label(addr, type, ifindex);
160         label = p ? p->label : IPV6_ADDR_LABEL_DEFAULT;
161         rcu_read_unlock();
162
163         ADDRLABEL(KERN_DEBUG "%s(addr=" NIP6_FMT ", type=%d, ifindex=%d) => %08x\n",
164                         __FUNCTION__,
165                         NIP6(*addr), type, ifindex,
166                         label);
167
168         return label;
169 }
170
171 /* allocate one entry */
172 struct ip6addrlbl_entry *ip6addrlbl_alloc(const struct in6_addr *prefix,
173                                           int prefixlen, int ifindex,
174                                           u32 label)
175 {
176         struct ip6addrlbl_entry *newp;
177         int addrtype;
178
179         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d, label=%u)\n",
180                         __FUNCTION__,
181                         NIP6(*prefix), prefixlen,
182                         ifindex,
183                         (unsigned int)label);
184
185         addrtype = ipv6_addr_type(prefix) & (IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK);
186
187         switch (addrtype) {
188         case IPV6_ADDR_MAPPED:
189                 if (prefixlen > 96)
190                         return ERR_PTR(-EINVAL);
191                 if (prefixlen < 96)
192                         addrtype = 0;
193                 break;
194         case IPV6_ADDR_COMPATv4:
195                 if (prefixlen != 96)
196                         addrtype = 0;
197                 break;
198         case IPV6_ADDR_LOOPBACK:
199                 if (prefixlen != 128)
200                         addrtype = 0;
201                 break;
202         }
203
204         newp = kmalloc(sizeof(*newp), GFP_KERNEL);
205         if (!newp)
206                 return ERR_PTR(-ENOMEM);
207
208         ipv6_addr_prefix(&newp->prefix, prefix, prefixlen);
209         newp->prefixlen = prefixlen;
210         newp->ifindex = ifindex;
211         newp->addrtype = addrtype;
212         newp->label = label;
213         INIT_HLIST_NODE(&newp->list);
214         atomic_set(&newp->refcnt, 1);
215         return newp;
216 }
217
218 /* add a label */
219 int __ip6addrlbl_add(struct ip6addrlbl_entry *newp, int replace)
220 {
221         int ret = 0;
222
223         ADDRLABEL(KERN_DEBUG "%s(newp=%p, replace=%d)\n",
224                         __FUNCTION__,
225                         newp, replace);
226
227         if (hlist_empty(&ip6addrlbl_table.head)) {
228                 hlist_add_head_rcu(&newp->list, &ip6addrlbl_table.head);
229         } else {
230                 struct hlist_node *pos, *n;
231                 struct ip6addrlbl_entry *p = NULL;
232                 hlist_for_each_entry_safe(p, pos, n,
233                                           &ip6addrlbl_table.head, list) {
234                         if (p->prefixlen == newp->prefixlen &&
235                             p->ifindex == newp->ifindex &&
236                             ipv6_addr_equal(&p->prefix, &newp->prefix)) {
237                                 if (!replace) {
238                                         ret = -EEXIST;
239                                         goto out;
240                                 }
241                                 hlist_replace_rcu(&p->list, &newp->list);
242                                 ip6addrlbl_put(p);
243                                 call_rcu(&p->rcu, ip6addrlbl_free_rcu);
244                                 goto out;
245                         } else if ((p->prefixlen == newp->prefixlen && !p->ifindex) ||
246                                    (p->prefixlen < newp->prefixlen)) {
247                                 hlist_add_before_rcu(&newp->list, &p->list);
248                                 goto out;
249                         }
250                 }
251                 hlist_add_after_rcu(&p->list, &newp->list);
252         }
253 out:
254         if (!ret)
255                 ip6addrlbl_table.seq++;
256         return ret;
257 }
258
259 /* add a label */
260 int ip6addrlbl_add(const struct in6_addr *prefix, int prefixlen,
261                        int ifindex, u32 label, int replace)
262 {
263         struct ip6addrlbl_entry *newp;
264         int ret = 0;
265
266         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d, label=%u, replace=%d)\n",
267                         __FUNCTION__,
268                         NIP6(*prefix), prefixlen,
269                         ifindex,
270                         (unsigned int)label,
271                         replace);
272
273         newp = ip6addrlbl_alloc(prefix, prefixlen, ifindex, label);
274         if (IS_ERR(newp))
275                 return PTR_ERR(newp);
276         spin_lock(&ip6addrlbl_table.lock);
277         ret = __ip6addrlbl_add(newp, replace);
278         spin_unlock(&ip6addrlbl_table.lock);
279         if (ret)
280                 ip6addrlbl_free(newp);
281         return ret;
282 }
283
284 /* remove a label */
285 int __ip6addrlbl_del(const struct in6_addr *prefix, int prefixlen,
286                           int ifindex)
287 {
288         struct ip6addrlbl_entry *p = NULL;
289         struct hlist_node *pos, *n;
290         int ret = -ESRCH;
291
292         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d)\n",
293                         __FUNCTION__,
294                         NIP6(*prefix), prefixlen,
295                         ifindex);
296
297         hlist_for_each_entry_safe(p, pos, n, &ip6addrlbl_table.head, list) {
298                 if (p->prefixlen == prefixlen &&
299                     p->ifindex == ifindex &&
300                     ipv6_addr_equal(&p->prefix, prefix)) {
301                         hlist_del_rcu(&p->list);
302                         ip6addrlbl_put(p);
303                         call_rcu(&p->rcu, ip6addrlbl_free_rcu);
304                         ret = 0;
305                         break;
306                 }
307         }
308         return ret;
309 }
310
311 int ip6addrlbl_del(const struct in6_addr *prefix, int prefixlen,
312                        int ifindex)
313 {
314         struct in6_addr prefix_buf;
315         int ret;
316
317         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d)\n",
318                         __FUNCTION__,
319                         NIP6(*prefix), prefixlen,
320                         ifindex);
321
322         ipv6_addr_prefix(&prefix_buf, prefix, prefixlen);
323         spin_lock(&ip6addrlbl_table.lock);
324         ret = __ip6addrlbl_del(&prefix_buf, prefixlen, ifindex);
325         spin_unlock(&ip6addrlbl_table.lock);
326         return ret;
327 }
328
329 /* add default label */
330 static __init int ip6addrlbl_init(void)
331 {
332         int err = 0;
333         int i;
334
335         ADDRLABEL(KERN_DEBUG "%s()\n", __FUNCTION__);
336
337         for (i = 0; i < ARRAY_SIZE(ip6addrlbl_init_table); i++) {
338                 int ret = ip6addrlbl_add(ip6addrlbl_init_table[i].prefix,
339                                          ip6addrlbl_init_table[i].prefixlen,
340                                          0,
341                                          ip6addrlbl_init_table[i].label, 0);
342                 /* XXX: should we free all rules when we catch an error? */
343                 if (ret && (!err || err != -ENOMEM))
344                         err = ret;
345         }
346         return err;
347 }
348
349 int __init ipv6_addr_label_init(void)
350 {
351         spin_lock_init(&ip6addrlbl_table.lock);
352
353         return ip6addrlbl_init();
354 }
355
356 static const struct nla_policy ifal_policy[IFAL_MAX+1] = {
357         [IFAL_ADDRESS]          = { .len = sizeof(struct in6_addr), },
358         [IFAL_LABEL]            = { .len = sizeof(u32), },
359 };
360
361 static int ip6addrlbl_newdel(struct sk_buff *skb, struct nlmsghdr *nlh,
362                              void *arg)
363 {
364         struct ifaddrlblmsg *ifal;
365         struct nlattr *tb[IFAL_MAX+1];
366         struct in6_addr *pfx;
367         u32 label;
368         int err = 0;
369
370         err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
371         if (err < 0)
372                 return err;
373
374         ifal = nlmsg_data(nlh);
375
376         if (ifal->ifal_family != AF_INET6 ||
377             ifal->ifal_prefixlen > 128)
378                 return -EINVAL;
379
380         if (ifal->ifal_index &&
381             !__dev_get_by_index(&init_net, ifal->ifal_index))
382                 return -EINVAL;
383
384         if (!tb[IFAL_ADDRESS])
385                 return -EINVAL;
386
387         pfx = nla_data(tb[IFAL_ADDRESS]);
388         if (!pfx)
389                 return -EINVAL;
390
391         if (!tb[IFAL_LABEL])
392                 return -EINVAL;
393         label = nla_get_u32(tb[IFAL_LABEL]);
394         if (label == IPV6_ADDR_LABEL_DEFAULT)
395                 return -EINVAL;
396
397         switch(nlh->nlmsg_type) {
398         case RTM_NEWADDRLABEL:
399                 err = ip6addrlbl_add(pfx, ifal->ifal_prefixlen,
400                                      ifal->ifal_index, label,
401                                      nlh->nlmsg_flags & NLM_F_REPLACE);
402                 break;
403         case RTM_DELADDRLABEL:
404                 err = ip6addrlbl_del(pfx, ifal->ifal_prefixlen,
405                                      ifal->ifal_index);
406                 break;
407         default:
408                 err = -EOPNOTSUPP;
409         }
410         return err;
411 }
412
413 static inline void ip6addrlbl_putmsg(struct nlmsghdr *nlh,
414                                      int prefixlen, int ifindex, u32 lseq)
415 {
416         struct ifaddrlblmsg *ifal = nlmsg_data(nlh);
417         ifal->ifal_family = AF_INET6;
418         ifal->ifal_prefixlen = prefixlen;
419         ifal->ifal_flags = 0;
420         ifal->ifal_index = ifindex;
421         ifal->ifal_seq = lseq;
422 };
423
424 static int ip6addrlbl_fill(struct sk_buff *skb,
425                            struct ip6addrlbl_entry *p,
426                            u32 lseq,
427                            u32 pid, u32 seq, int event,
428                            unsigned int flags)
429 {
430         struct nlmsghdr *nlh = nlmsg_put(skb, pid, seq, event,
431                                          sizeof(struct ifaddrlblmsg), flags);
432         if (!nlh)
433                 return -EMSGSIZE;
434
435         ip6addrlbl_putmsg(nlh, p->prefixlen, p->ifindex, lseq);
436
437         if (nla_put(skb, IFAL_ADDRESS, 16, &p->prefix) < 0 ||
438             nla_put_u32(skb, IFAL_LABEL, p->label) < 0) {
439                 nlmsg_cancel(skb, nlh);
440                 return -EMSGSIZE;
441         }
442
443         return nlmsg_end(skb, nlh);
444 }
445
446 static int ip6addrlbl_dump(struct sk_buff *skb, struct netlink_callback *cb)
447 {
448         struct ip6addrlbl_entry *p;
449         struct hlist_node *pos;
450         int idx = 0, s_idx = cb->args[0];
451         int err;
452
453         rcu_read_lock();
454         hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
455                 if (idx >= s_idx) {
456                         if ((err = ip6addrlbl_fill(skb, p,
457                                                    ip6addrlbl_table.seq,
458                                                    NETLINK_CB(cb->skb).pid,
459                                                    cb->nlh->nlmsg_seq,
460                                                    RTM_NEWADDRLABEL,
461                                                    NLM_F_MULTI)) <= 0)
462                                 break;
463                 }
464                 idx++;
465         }
466         rcu_read_unlock();
467         cb->args[0] = idx;
468         return skb->len;
469 }
470
471 static inline int ip6addrlbl_msgsize(void)
472 {
473         return (NLMSG_ALIGN(sizeof(struct ifaddrlblmsg))
474                 + nla_total_size(16)    /* IFAL_ADDRESS */
475                 + nla_total_size(4)     /* IFAL_LABEL */
476         );
477 }
478
479 static int ip6addrlbl_get(struct sk_buff *in_skb, struct nlmsghdr* nlh,
480                           void *arg)
481 {
482         struct ifaddrlblmsg *ifal;
483         struct nlattr *tb[IFAL_MAX+1];
484         struct in6_addr *addr;
485         u32 lseq;
486         int err = 0;
487         struct ip6addrlbl_entry *p;
488         struct sk_buff *skb;
489
490         err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
491         if (err < 0)
492                 return err;
493
494         ifal = nlmsg_data(nlh);
495
496         if (ifal->ifal_family != AF_INET6 ||
497             ifal->ifal_prefixlen != 128)
498                 return -EINVAL;
499
500         if (ifal->ifal_index &&
501             !__dev_get_by_index(&init_net, ifal->ifal_index))
502                 return -EINVAL;
503
504         if (!tb[IFAL_ADDRESS])
505                 return -EINVAL;
506
507         addr = nla_data(tb[IFAL_ADDRESS]);
508         if (!addr)
509                 return -EINVAL;
510
511         rcu_read_lock();
512         p = __ipv6_addr_label(addr, ipv6_addr_type(addr), ifal->ifal_index);
513         if (p && ip6addrlbl_hold(p))
514                 p = NULL;
515         lseq = ip6addrlbl_table.seq;
516         rcu_read_unlock();
517
518         if (!p) {
519                 err = -ESRCH;
520                 goto out;
521         }
522
523         if (!(skb = nlmsg_new(ip6addrlbl_msgsize(), GFP_KERNEL))) {
524                 ip6addrlbl_put(p);
525                 return -ENOBUFS;
526         }
527
528         err = ip6addrlbl_fill(skb, p, lseq,
529                               NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
530                               RTM_NEWADDRLABEL, 0);
531
532         ip6addrlbl_put(p);
533
534         if (err < 0) {
535                 WARN_ON(err == -EMSGSIZE);
536                 kfree_skb(skb);
537                 goto out;
538         }
539
540         err = rtnl_unicast(skb, NETLINK_CB(in_skb).pid);
541 out:
542         return err;
543 }
544
545 void __init ipv6_addr_label_rtnl_register(void)
546 {
547         __rtnl_register(PF_INET6, RTM_NEWADDRLABEL, ip6addrlbl_newdel, NULL);
548         __rtnl_register(PF_INET6, RTM_DELADDRLABEL, ip6addrlbl_newdel, NULL);
549         __rtnl_register(PF_INET6, RTM_GETADDRLABEL, ip6addrlbl_get, ip6addrlbl_dump);
550 }
551