Merge branch 'x86/irq' into x86/devel
[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  * 2001:10::/28         N/A             7               ORCHID (RFC 4843)
62  *
63  * Note: 0xffffffff is used if we do not have any policies.
64  */
65
66 #define IPV6_ADDR_LABEL_DEFAULT 0xffffffffUL
67
68 static const __initdata struct ip6addrlbl_init_table
69 {
70         const struct in6_addr *prefix;
71         int prefixlen;
72         u32 label;
73 } ip6addrlbl_init_table[] = {
74         {       /* ::/0 */
75                 .prefix = &in6addr_any,
76                 .label = 1,
77         },{     /* fc00::/7 */
78                 .prefix = &(struct in6_addr){{{ 0xfc }}},
79                 .prefixlen = 7,
80                 .label = 5,
81         },{     /* 2002::/16 */
82                 .prefix = &(struct in6_addr){{{ 0x20, 0x02 }}},
83                 .prefixlen = 16,
84                 .label = 2,
85         },{     /* 2001::/32 */
86                 .prefix = &(struct in6_addr){{{ 0x20, 0x01 }}},
87                 .prefixlen = 32,
88                 .label = 6,
89         },{     /* 2001:10::/28 */
90                 .prefix = &(struct in6_addr){{{ 0x20, 0x01, 0x00, 0x10 }}},
91                 .prefixlen = 28,
92                 .label = 7,
93         },{     /* ::ffff:0:0 */
94                 .prefix = &(struct in6_addr){{{ [10] = 0xff, [11] = 0xff }}},
95                 .prefixlen = 96,
96                 .label = 4,
97         },{     /* ::/96 */
98                 .prefix = &in6addr_any,
99                 .prefixlen = 96,
100                 .label = 3,
101         },{     /* ::1/128 */
102                 .prefix = &in6addr_loopback,
103                 .prefixlen = 128,
104                 .label = 0,
105         }
106 };
107
108 /* Object management */
109 static inline void ip6addrlbl_free(struct ip6addrlbl_entry *p)
110 {
111         kfree(p);
112 }
113
114 static void ip6addrlbl_free_rcu(struct rcu_head *h)
115 {
116         ip6addrlbl_free(container_of(h, struct ip6addrlbl_entry, rcu));
117 }
118
119 static inline int ip6addrlbl_hold(struct ip6addrlbl_entry *p)
120 {
121         return atomic_inc_not_zero(&p->refcnt);
122 }
123
124 static inline void ip6addrlbl_put(struct ip6addrlbl_entry *p)
125 {
126         if (atomic_dec_and_test(&p->refcnt))
127                 call_rcu(&p->rcu, ip6addrlbl_free_rcu);
128 }
129
130 /* Find label */
131 static int __ip6addrlbl_match(struct ip6addrlbl_entry *p,
132                               const struct in6_addr *addr,
133                               int addrtype, int ifindex)
134 {
135         if (p->ifindex && p->ifindex != ifindex)
136                 return 0;
137         if (p->addrtype && p->addrtype != addrtype)
138                 return 0;
139         if (!ipv6_prefix_equal(addr, &p->prefix, p->prefixlen))
140                 return 0;
141         return 1;
142 }
143
144 static struct ip6addrlbl_entry *__ipv6_addr_label(const struct in6_addr *addr,
145                                                   int type, int ifindex)
146 {
147         struct hlist_node *pos;
148         struct ip6addrlbl_entry *p;
149         hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
150                 if (__ip6addrlbl_match(p, addr, type, ifindex))
151                         return p;
152         }
153         return NULL;
154 }
155
156 u32 ipv6_addr_label(const struct in6_addr *addr, int type, int ifindex)
157 {
158         u32 label;
159         struct ip6addrlbl_entry *p;
160
161         type &= IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK;
162
163         rcu_read_lock();
164         p = __ipv6_addr_label(addr, type, ifindex);
165         label = p ? p->label : IPV6_ADDR_LABEL_DEFAULT;
166         rcu_read_unlock();
167
168         ADDRLABEL(KERN_DEBUG "%s(addr=" NIP6_FMT ", type=%d, ifindex=%d) => %08x\n",
169                         __func__,
170                         NIP6(*addr), type, ifindex,
171                         label);
172
173         return label;
174 }
175
176 /* allocate one entry */
177 static struct ip6addrlbl_entry *ip6addrlbl_alloc(const struct in6_addr *prefix,
178                                                  int prefixlen, int ifindex,
179                                                  u32 label)
180 {
181         struct ip6addrlbl_entry *newp;
182         int addrtype;
183
184         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d, label=%u)\n",
185                         __func__,
186                         NIP6(*prefix), prefixlen,
187                         ifindex,
188                         (unsigned int)label);
189
190         addrtype = ipv6_addr_type(prefix) & (IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK);
191
192         switch (addrtype) {
193         case IPV6_ADDR_MAPPED:
194                 if (prefixlen > 96)
195                         return ERR_PTR(-EINVAL);
196                 if (prefixlen < 96)
197                         addrtype = 0;
198                 break;
199         case IPV6_ADDR_COMPATv4:
200                 if (prefixlen != 96)
201                         addrtype = 0;
202                 break;
203         case IPV6_ADDR_LOOPBACK:
204                 if (prefixlen != 128)
205                         addrtype = 0;
206                 break;
207         }
208
209         newp = kmalloc(sizeof(*newp), GFP_KERNEL);
210         if (!newp)
211                 return ERR_PTR(-ENOMEM);
212
213         ipv6_addr_prefix(&newp->prefix, prefix, prefixlen);
214         newp->prefixlen = prefixlen;
215         newp->ifindex = ifindex;
216         newp->addrtype = addrtype;
217         newp->label = label;
218         INIT_HLIST_NODE(&newp->list);
219         atomic_set(&newp->refcnt, 1);
220         return newp;
221 }
222
223 /* add a label */
224 static int __ip6addrlbl_add(struct ip6addrlbl_entry *newp, int replace)
225 {
226         int ret = 0;
227
228         ADDRLABEL(KERN_DEBUG "%s(newp=%p, replace=%d)\n",
229                         __func__,
230                         newp, replace);
231
232         if (hlist_empty(&ip6addrlbl_table.head)) {
233                 hlist_add_head_rcu(&newp->list, &ip6addrlbl_table.head);
234         } else {
235                 struct hlist_node *pos, *n;
236                 struct ip6addrlbl_entry *p = NULL;
237                 hlist_for_each_entry_safe(p, pos, n,
238                                           &ip6addrlbl_table.head, list) {
239                         if (p->prefixlen == newp->prefixlen &&
240                             p->ifindex == newp->ifindex &&
241                             ipv6_addr_equal(&p->prefix, &newp->prefix)) {
242                                 if (!replace) {
243                                         ret = -EEXIST;
244                                         goto out;
245                                 }
246                                 hlist_replace_rcu(&p->list, &newp->list);
247                                 ip6addrlbl_put(p);
248                                 goto out;
249                         } else if ((p->prefixlen == newp->prefixlen && !p->ifindex) ||
250                                    (p->prefixlen < newp->prefixlen)) {
251                                 hlist_add_before_rcu(&newp->list, &p->list);
252                                 goto out;
253                         }
254                 }
255                 hlist_add_after_rcu(&p->list, &newp->list);
256         }
257 out:
258         if (!ret)
259                 ip6addrlbl_table.seq++;
260         return ret;
261 }
262
263 /* add a label */
264 static int ip6addrlbl_add(const struct in6_addr *prefix, int prefixlen,
265                           int ifindex, u32 label, int replace)
266 {
267         struct ip6addrlbl_entry *newp;
268         int ret = 0;
269
270         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d, label=%u, replace=%d)\n",
271                         __func__,
272                         NIP6(*prefix), prefixlen,
273                         ifindex,
274                         (unsigned int)label,
275                         replace);
276
277         newp = ip6addrlbl_alloc(prefix, prefixlen, ifindex, label);
278         if (IS_ERR(newp))
279                 return PTR_ERR(newp);
280         spin_lock(&ip6addrlbl_table.lock);
281         ret = __ip6addrlbl_add(newp, replace);
282         spin_unlock(&ip6addrlbl_table.lock);
283         if (ret)
284                 ip6addrlbl_free(newp);
285         return ret;
286 }
287
288 /* remove a label */
289 static int __ip6addrlbl_del(const struct in6_addr *prefix, int prefixlen,
290                             int ifindex)
291 {
292         struct ip6addrlbl_entry *p = NULL;
293         struct hlist_node *pos, *n;
294         int ret = -ESRCH;
295
296         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d)\n",
297                         __func__,
298                         NIP6(*prefix), prefixlen,
299                         ifindex);
300
301         hlist_for_each_entry_safe(p, pos, n, &ip6addrlbl_table.head, list) {
302                 if (p->prefixlen == prefixlen &&
303                     p->ifindex == ifindex &&
304                     ipv6_addr_equal(&p->prefix, prefix)) {
305                         hlist_del_rcu(&p->list);
306                         ip6addrlbl_put(p);
307                         ret = 0;
308                         break;
309                 }
310         }
311         return ret;
312 }
313
314 static int ip6addrlbl_del(const struct in6_addr *prefix, int prefixlen,
315                           int ifindex)
316 {
317         struct in6_addr prefix_buf;
318         int ret;
319
320         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d)\n",
321                         __func__,
322                         NIP6(*prefix), prefixlen,
323                         ifindex);
324
325         ipv6_addr_prefix(&prefix_buf, prefix, prefixlen);
326         spin_lock(&ip6addrlbl_table.lock);
327         ret = __ip6addrlbl_del(&prefix_buf, prefixlen, ifindex);
328         spin_unlock(&ip6addrlbl_table.lock);
329         return ret;
330 }
331
332 /* add default label */
333 static __init int ip6addrlbl_init(void)
334 {
335         int err = 0;
336         int i;
337
338         ADDRLABEL(KERN_DEBUG "%s()\n", __func__);
339
340         for (i = 0; i < ARRAY_SIZE(ip6addrlbl_init_table); i++) {
341                 int ret = ip6addrlbl_add(ip6addrlbl_init_table[i].prefix,
342                                          ip6addrlbl_init_table[i].prefixlen,
343                                          0,
344                                          ip6addrlbl_init_table[i].label, 0);
345                 /* XXX: should we free all rules when we catch an error? */
346                 if (ret && (!err || err != -ENOMEM))
347                         err = ret;
348         }
349         return err;
350 }
351
352 int __init ipv6_addr_label_init(void)
353 {
354         spin_lock_init(&ip6addrlbl_table.lock);
355
356         return ip6addrlbl_init();
357 }
358
359 static const struct nla_policy ifal_policy[IFAL_MAX+1] = {
360         [IFAL_ADDRESS]          = { .len = sizeof(struct in6_addr), },
361         [IFAL_LABEL]            = { .len = sizeof(u32), },
362 };
363
364 static int ip6addrlbl_newdel(struct sk_buff *skb, struct nlmsghdr *nlh,
365                              void *arg)
366 {
367         struct net *net = sock_net(skb->sk);
368         struct ifaddrlblmsg *ifal;
369         struct nlattr *tb[IFAL_MAX+1];
370         struct in6_addr *pfx;
371         u32 label;
372         int err = 0;
373
374         if (net != &init_net)
375                 return 0;
376
377         err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
378         if (err < 0)
379                 return err;
380
381         ifal = nlmsg_data(nlh);
382
383         if (ifal->ifal_family != AF_INET6 ||
384             ifal->ifal_prefixlen > 128)
385                 return -EINVAL;
386
387         if (ifal->ifal_index &&
388             !__dev_get_by_index(&init_net, ifal->ifal_index))
389                 return -EINVAL;
390
391         if (!tb[IFAL_ADDRESS])
392                 return -EINVAL;
393
394         pfx = nla_data(tb[IFAL_ADDRESS]);
395         if (!pfx)
396                 return -EINVAL;
397
398         if (!tb[IFAL_LABEL])
399                 return -EINVAL;
400         label = nla_get_u32(tb[IFAL_LABEL]);
401         if (label == IPV6_ADDR_LABEL_DEFAULT)
402                 return -EINVAL;
403
404         switch(nlh->nlmsg_type) {
405         case RTM_NEWADDRLABEL:
406                 err = ip6addrlbl_add(pfx, ifal->ifal_prefixlen,
407                                      ifal->ifal_index, label,
408                                      nlh->nlmsg_flags & NLM_F_REPLACE);
409                 break;
410         case RTM_DELADDRLABEL:
411                 err = ip6addrlbl_del(pfx, ifal->ifal_prefixlen,
412                                      ifal->ifal_index);
413                 break;
414         default:
415                 err = -EOPNOTSUPP;
416         }
417         return err;
418 }
419
420 static inline void ip6addrlbl_putmsg(struct nlmsghdr *nlh,
421                                      int prefixlen, int ifindex, u32 lseq)
422 {
423         struct ifaddrlblmsg *ifal = nlmsg_data(nlh);
424         ifal->ifal_family = AF_INET6;
425         ifal->ifal_prefixlen = prefixlen;
426         ifal->ifal_flags = 0;
427         ifal->ifal_index = ifindex;
428         ifal->ifal_seq = lseq;
429 };
430
431 static int ip6addrlbl_fill(struct sk_buff *skb,
432                            struct ip6addrlbl_entry *p,
433                            u32 lseq,
434                            u32 pid, u32 seq, int event,
435                            unsigned int flags)
436 {
437         struct nlmsghdr *nlh = nlmsg_put(skb, pid, seq, event,
438                                          sizeof(struct ifaddrlblmsg), flags);
439         if (!nlh)
440                 return -EMSGSIZE;
441
442         ip6addrlbl_putmsg(nlh, p->prefixlen, p->ifindex, lseq);
443
444         if (nla_put(skb, IFAL_ADDRESS, 16, &p->prefix) < 0 ||
445             nla_put_u32(skb, IFAL_LABEL, p->label) < 0) {
446                 nlmsg_cancel(skb, nlh);
447                 return -EMSGSIZE;
448         }
449
450         return nlmsg_end(skb, nlh);
451 }
452
453 static int ip6addrlbl_dump(struct sk_buff *skb, struct netlink_callback *cb)
454 {
455         struct net *net = sock_net(skb->sk);
456         struct ip6addrlbl_entry *p;
457         struct hlist_node *pos;
458         int idx = 0, s_idx = cb->args[0];
459         int err;
460
461         if (net != &init_net)
462                 return 0;
463
464         rcu_read_lock();
465         hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
466                 if (idx >= s_idx) {
467                         if ((err = ip6addrlbl_fill(skb, p,
468                                                    ip6addrlbl_table.seq,
469                                                    NETLINK_CB(cb->skb).pid,
470                                                    cb->nlh->nlmsg_seq,
471                                                    RTM_NEWADDRLABEL,
472                                                    NLM_F_MULTI)) <= 0)
473                                 break;
474                 }
475                 idx++;
476         }
477         rcu_read_unlock();
478         cb->args[0] = idx;
479         return skb->len;
480 }
481
482 static inline int ip6addrlbl_msgsize(void)
483 {
484         return (NLMSG_ALIGN(sizeof(struct ifaddrlblmsg))
485                 + nla_total_size(16)    /* IFAL_ADDRESS */
486                 + nla_total_size(4)     /* IFAL_LABEL */
487         );
488 }
489
490 static int ip6addrlbl_get(struct sk_buff *in_skb, struct nlmsghdr* nlh,
491                           void *arg)
492 {
493         struct net *net = sock_net(in_skb->sk);
494         struct ifaddrlblmsg *ifal;
495         struct nlattr *tb[IFAL_MAX+1];
496         struct in6_addr *addr;
497         u32 lseq;
498         int err = 0;
499         struct ip6addrlbl_entry *p;
500         struct sk_buff *skb;
501
502         if (net != &init_net)
503                 return 0;
504
505         err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
506         if (err < 0)
507                 return err;
508
509         ifal = nlmsg_data(nlh);
510
511         if (ifal->ifal_family != AF_INET6 ||
512             ifal->ifal_prefixlen != 128)
513                 return -EINVAL;
514
515         if (ifal->ifal_index &&
516             !__dev_get_by_index(&init_net, ifal->ifal_index))
517                 return -EINVAL;
518
519         if (!tb[IFAL_ADDRESS])
520                 return -EINVAL;
521
522         addr = nla_data(tb[IFAL_ADDRESS]);
523         if (!addr)
524                 return -EINVAL;
525
526         rcu_read_lock();
527         p = __ipv6_addr_label(addr, ipv6_addr_type(addr), ifal->ifal_index);
528         if (p && ip6addrlbl_hold(p))
529                 p = NULL;
530         lseq = ip6addrlbl_table.seq;
531         rcu_read_unlock();
532
533         if (!p) {
534                 err = -ESRCH;
535                 goto out;
536         }
537
538         if (!(skb = nlmsg_new(ip6addrlbl_msgsize(), GFP_KERNEL))) {
539                 ip6addrlbl_put(p);
540                 return -ENOBUFS;
541         }
542
543         err = ip6addrlbl_fill(skb, p, lseq,
544                               NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
545                               RTM_NEWADDRLABEL, 0);
546
547         ip6addrlbl_put(p);
548
549         if (err < 0) {
550                 WARN_ON(err == -EMSGSIZE);
551                 kfree_skb(skb);
552                 goto out;
553         }
554
555         err = rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
556 out:
557         return err;
558 }
559
560 void __init ipv6_addr_label_rtnl_register(void)
561 {
562         __rtnl_register(PF_INET6, RTM_NEWADDRLABEL, ip6addrlbl_newdel, NULL);
563         __rtnl_register(PF_INET6, RTM_DELADDRLABEL, ip6addrlbl_newdel, NULL);
564         __rtnl_register(PF_INET6, RTM_GETADDRLABEL, ip6addrlbl_get, ip6addrlbl_dump);
565 }
566