Merge branch 'task_killable' of git://git.kernel.org/pub/scm/linux/kernel/git/willy...
[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 void ip6addrlbl_free_rcu(struct rcu_head *h)
110 {
111         ip6addrlbl_free(container_of(h, struct ip6addrlbl_entry, rcu));
112 }
113
114 static inline int ip6addrlbl_hold(struct ip6addrlbl_entry *p)
115 {
116         return atomic_inc_not_zero(&p->refcnt);
117 }
118
119 static inline void ip6addrlbl_put(struct ip6addrlbl_entry *p)
120 {
121         if (atomic_dec_and_test(&p->refcnt))
122                 call_rcu(&p->rcu, ip6addrlbl_free_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 static 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 static 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                                 goto out;
244                         } else if ((p->prefixlen == newp->prefixlen && !p->ifindex) ||
245                                    (p->prefixlen < newp->prefixlen)) {
246                                 hlist_add_before_rcu(&newp->list, &p->list);
247                                 goto out;
248                         }
249                 }
250                 hlist_add_after_rcu(&p->list, &newp->list);
251         }
252 out:
253         if (!ret)
254                 ip6addrlbl_table.seq++;
255         return ret;
256 }
257
258 /* add a label */
259 static int ip6addrlbl_add(const struct in6_addr *prefix, int prefixlen,
260                           int ifindex, u32 label, int replace)
261 {
262         struct ip6addrlbl_entry *newp;
263         int ret = 0;
264
265         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d, label=%u, replace=%d)\n",
266                         __FUNCTION__,
267                         NIP6(*prefix), prefixlen,
268                         ifindex,
269                         (unsigned int)label,
270                         replace);
271
272         newp = ip6addrlbl_alloc(prefix, prefixlen, ifindex, label);
273         if (IS_ERR(newp))
274                 return PTR_ERR(newp);
275         spin_lock(&ip6addrlbl_table.lock);
276         ret = __ip6addrlbl_add(newp, replace);
277         spin_unlock(&ip6addrlbl_table.lock);
278         if (ret)
279                 ip6addrlbl_free(newp);
280         return ret;
281 }
282
283 /* remove a label */
284 static int __ip6addrlbl_del(const struct in6_addr *prefix, int prefixlen,
285                             int ifindex)
286 {
287         struct ip6addrlbl_entry *p = NULL;
288         struct hlist_node *pos, *n;
289         int ret = -ESRCH;
290
291         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d)\n",
292                         __FUNCTION__,
293                         NIP6(*prefix), prefixlen,
294                         ifindex);
295
296         hlist_for_each_entry_safe(p, pos, n, &ip6addrlbl_table.head, list) {
297                 if (p->prefixlen == prefixlen &&
298                     p->ifindex == ifindex &&
299                     ipv6_addr_equal(&p->prefix, prefix)) {
300                         hlist_del_rcu(&p->list);
301                         ip6addrlbl_put(p);
302                         ret = 0;
303                         break;
304                 }
305         }
306         return ret;
307 }
308
309 static int ip6addrlbl_del(const struct in6_addr *prefix, int prefixlen,
310                           int ifindex)
311 {
312         struct in6_addr prefix_buf;
313         int ret;
314
315         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d)\n",
316                         __FUNCTION__,
317                         NIP6(*prefix), prefixlen,
318                         ifindex);
319
320         ipv6_addr_prefix(&prefix_buf, prefix, prefixlen);
321         spin_lock(&ip6addrlbl_table.lock);
322         ret = __ip6addrlbl_del(&prefix_buf, prefixlen, ifindex);
323         spin_unlock(&ip6addrlbl_table.lock);
324         return ret;
325 }
326
327 /* add default label */
328 static __init int ip6addrlbl_init(void)
329 {
330         int err = 0;
331         int i;
332
333         ADDRLABEL(KERN_DEBUG "%s()\n", __FUNCTION__);
334
335         for (i = 0; i < ARRAY_SIZE(ip6addrlbl_init_table); i++) {
336                 int ret = ip6addrlbl_add(ip6addrlbl_init_table[i].prefix,
337                                          ip6addrlbl_init_table[i].prefixlen,
338                                          0,
339                                          ip6addrlbl_init_table[i].label, 0);
340                 /* XXX: should we free all rules when we catch an error? */
341                 if (ret && (!err || err != -ENOMEM))
342                         err = ret;
343         }
344         return err;
345 }
346
347 int __init ipv6_addr_label_init(void)
348 {
349         spin_lock_init(&ip6addrlbl_table.lock);
350
351         return ip6addrlbl_init();
352 }
353
354 static const struct nla_policy ifal_policy[IFAL_MAX+1] = {
355         [IFAL_ADDRESS]          = { .len = sizeof(struct in6_addr), },
356         [IFAL_LABEL]            = { .len = sizeof(u32), },
357 };
358
359 static int ip6addrlbl_newdel(struct sk_buff *skb, struct nlmsghdr *nlh,
360                              void *arg)
361 {
362         struct net *net = skb->sk->sk_net;
363         struct ifaddrlblmsg *ifal;
364         struct nlattr *tb[IFAL_MAX+1];
365         struct in6_addr *pfx;
366         u32 label;
367         int err = 0;
368
369         if (net != &init_net)
370                 return 0;
371
372         err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
373         if (err < 0)
374                 return err;
375
376         ifal = nlmsg_data(nlh);
377
378         if (ifal->ifal_family != AF_INET6 ||
379             ifal->ifal_prefixlen > 128)
380                 return -EINVAL;
381
382         if (ifal->ifal_index &&
383             !__dev_get_by_index(&init_net, ifal->ifal_index))
384                 return -EINVAL;
385
386         if (!tb[IFAL_ADDRESS])
387                 return -EINVAL;
388
389         pfx = nla_data(tb[IFAL_ADDRESS]);
390         if (!pfx)
391                 return -EINVAL;
392
393         if (!tb[IFAL_LABEL])
394                 return -EINVAL;
395         label = nla_get_u32(tb[IFAL_LABEL]);
396         if (label == IPV6_ADDR_LABEL_DEFAULT)
397                 return -EINVAL;
398
399         switch(nlh->nlmsg_type) {
400         case RTM_NEWADDRLABEL:
401                 err = ip6addrlbl_add(pfx, ifal->ifal_prefixlen,
402                                      ifal->ifal_index, label,
403                                      nlh->nlmsg_flags & NLM_F_REPLACE);
404                 break;
405         case RTM_DELADDRLABEL:
406                 err = ip6addrlbl_del(pfx, ifal->ifal_prefixlen,
407                                      ifal->ifal_index);
408                 break;
409         default:
410                 err = -EOPNOTSUPP;
411         }
412         return err;
413 }
414
415 static inline void ip6addrlbl_putmsg(struct nlmsghdr *nlh,
416                                      int prefixlen, int ifindex, u32 lseq)
417 {
418         struct ifaddrlblmsg *ifal = nlmsg_data(nlh);
419         ifal->ifal_family = AF_INET6;
420         ifal->ifal_prefixlen = prefixlen;
421         ifal->ifal_flags = 0;
422         ifal->ifal_index = ifindex;
423         ifal->ifal_seq = lseq;
424 };
425
426 static int ip6addrlbl_fill(struct sk_buff *skb,
427                            struct ip6addrlbl_entry *p,
428                            u32 lseq,
429                            u32 pid, u32 seq, int event,
430                            unsigned int flags)
431 {
432         struct nlmsghdr *nlh = nlmsg_put(skb, pid, seq, event,
433                                          sizeof(struct ifaddrlblmsg), flags);
434         if (!nlh)
435                 return -EMSGSIZE;
436
437         ip6addrlbl_putmsg(nlh, p->prefixlen, p->ifindex, lseq);
438
439         if (nla_put(skb, IFAL_ADDRESS, 16, &p->prefix) < 0 ||
440             nla_put_u32(skb, IFAL_LABEL, p->label) < 0) {
441                 nlmsg_cancel(skb, nlh);
442                 return -EMSGSIZE;
443         }
444
445         return nlmsg_end(skb, nlh);
446 }
447
448 static int ip6addrlbl_dump(struct sk_buff *skb, struct netlink_callback *cb)
449 {
450         struct net *net = skb->sk->sk_net;
451         struct ip6addrlbl_entry *p;
452         struct hlist_node *pos;
453         int idx = 0, s_idx = cb->args[0];
454         int err;
455
456         if (net != &init_net)
457                 return 0;
458
459         rcu_read_lock();
460         hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
461                 if (idx >= s_idx) {
462                         if ((err = ip6addrlbl_fill(skb, p,
463                                                    ip6addrlbl_table.seq,
464                                                    NETLINK_CB(cb->skb).pid,
465                                                    cb->nlh->nlmsg_seq,
466                                                    RTM_NEWADDRLABEL,
467                                                    NLM_F_MULTI)) <= 0)
468                                 break;
469                 }
470                 idx++;
471         }
472         rcu_read_unlock();
473         cb->args[0] = idx;
474         return skb->len;
475 }
476
477 static inline int ip6addrlbl_msgsize(void)
478 {
479         return (NLMSG_ALIGN(sizeof(struct ifaddrlblmsg))
480                 + nla_total_size(16)    /* IFAL_ADDRESS */
481                 + nla_total_size(4)     /* IFAL_LABEL */
482         );
483 }
484
485 static int ip6addrlbl_get(struct sk_buff *in_skb, struct nlmsghdr* nlh,
486                           void *arg)
487 {
488         struct net *net = in_skb->sk->sk_net;
489         struct ifaddrlblmsg *ifal;
490         struct nlattr *tb[IFAL_MAX+1];
491         struct in6_addr *addr;
492         u32 lseq;
493         int err = 0;
494         struct ip6addrlbl_entry *p;
495         struct sk_buff *skb;
496
497         if (net != &init_net)
498                 return 0;
499
500         err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
501         if (err < 0)
502                 return err;
503
504         ifal = nlmsg_data(nlh);
505
506         if (ifal->ifal_family != AF_INET6 ||
507             ifal->ifal_prefixlen != 128)
508                 return -EINVAL;
509
510         if (ifal->ifal_index &&
511             !__dev_get_by_index(&init_net, ifal->ifal_index))
512                 return -EINVAL;
513
514         if (!tb[IFAL_ADDRESS])
515                 return -EINVAL;
516
517         addr = nla_data(tb[IFAL_ADDRESS]);
518         if (!addr)
519                 return -EINVAL;
520
521         rcu_read_lock();
522         p = __ipv6_addr_label(addr, ipv6_addr_type(addr), ifal->ifal_index);
523         if (p && ip6addrlbl_hold(p))
524                 p = NULL;
525         lseq = ip6addrlbl_table.seq;
526         rcu_read_unlock();
527
528         if (!p) {
529                 err = -ESRCH;
530                 goto out;
531         }
532
533         if (!(skb = nlmsg_new(ip6addrlbl_msgsize(), GFP_KERNEL))) {
534                 ip6addrlbl_put(p);
535                 return -ENOBUFS;
536         }
537
538         err = ip6addrlbl_fill(skb, p, lseq,
539                               NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
540                               RTM_NEWADDRLABEL, 0);
541
542         ip6addrlbl_put(p);
543
544         if (err < 0) {
545                 WARN_ON(err == -EMSGSIZE);
546                 kfree_skb(skb);
547                 goto out;
548         }
549
550         err = rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
551 out:
552         return err;
553 }
554
555 void __init ipv6_addr_label_rtnl_register(void)
556 {
557         __rtnl_register(PF_INET6, RTM_NEWADDRLABEL, ip6addrlbl_newdel, NULL);
558         __rtnl_register(PF_INET6, RTM_DELADDRLABEL, ip6addrlbl_newdel, NULL);
559         __rtnl_register(PF_INET6, RTM_GETADDRLABEL, ip6addrlbl_get, ip6addrlbl_dump);
560 }
561