Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[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 #ifdef CONFIG_NET_NS
33         struct net *lbl_net;
34 #endif
35         struct in6_addr prefix;
36         int prefixlen;
37         int ifindex;
38         int addrtype;
39         u32 label;
40         struct hlist_node list;
41         atomic_t refcnt;
42         struct rcu_head rcu;
43 };
44
45 static struct ip6addrlbl_table
46 {
47         struct hlist_head head;
48         spinlock_t lock;
49         u32 seq;
50 } ip6addrlbl_table;
51
52 static inline
53 struct net *ip6addrlbl_net(const struct ip6addrlbl_entry *lbl)
54 {
55 #ifdef CONFIG_NET_NS
56         return lbl->lbl_net;
57 #else
58         return &init_net;
59 #endif
60 }
61
62 /*
63  * Default policy table (RFC3484 + extensions)
64  *
65  * prefix               addr_type       label
66  * -------------------------------------------------------------------------
67  * ::1/128              LOOPBACK        0
68  * ::/0                 N/A             1
69  * 2002::/16            N/A             2
70  * ::/96                COMPATv4        3
71  * ::ffff:0:0/96        V4MAPPED        4
72  * fc00::/7             N/A             5               ULA (RFC 4193)
73  * 2001::/32            N/A             6               Teredo (RFC 4380)
74  * 2001:10::/28         N/A             7               ORCHID (RFC 4843)
75  *
76  * Note: 0xffffffff is used if we do not have any policies.
77  */
78
79 #define IPV6_ADDR_LABEL_DEFAULT 0xffffffffUL
80
81 static const __net_initdata struct ip6addrlbl_init_table
82 {
83         const struct in6_addr *prefix;
84         int prefixlen;
85         u32 label;
86 } ip6addrlbl_init_table[] = {
87         {       /* ::/0 */
88                 .prefix = &in6addr_any,
89                 .label = 1,
90         },{     /* fc00::/7 */
91                 .prefix = &(struct in6_addr){{{ 0xfc }}},
92                 .prefixlen = 7,
93                 .label = 5,
94         },{     /* 2002::/16 */
95                 .prefix = &(struct in6_addr){{{ 0x20, 0x02 }}},
96                 .prefixlen = 16,
97                 .label = 2,
98         },{     /* 2001::/32 */
99                 .prefix = &(struct in6_addr){{{ 0x20, 0x01 }}},
100                 .prefixlen = 32,
101                 .label = 6,
102         },{     /* 2001:10::/28 */
103                 .prefix = &(struct in6_addr){{{ 0x20, 0x01, 0x00, 0x10 }}},
104                 .prefixlen = 28,
105                 .label = 7,
106         },{     /* ::ffff:0:0 */
107                 .prefix = &(struct in6_addr){{{ [10] = 0xff, [11] = 0xff }}},
108                 .prefixlen = 96,
109                 .label = 4,
110         },{     /* ::/96 */
111                 .prefix = &in6addr_any,
112                 .prefixlen = 96,
113                 .label = 3,
114         },{     /* ::1/128 */
115                 .prefix = &in6addr_loopback,
116                 .prefixlen = 128,
117                 .label = 0,
118         }
119 };
120
121 /* Object management */
122 static inline void ip6addrlbl_free(struct ip6addrlbl_entry *p)
123 {
124 #ifdef CONFIG_NET_NS
125         release_net(p->lbl_net);
126 #endif
127         kfree(p);
128 }
129
130 static void ip6addrlbl_free_rcu(struct rcu_head *h)
131 {
132         ip6addrlbl_free(container_of(h, struct ip6addrlbl_entry, rcu));
133 }
134
135 static inline int ip6addrlbl_hold(struct ip6addrlbl_entry *p)
136 {
137         return atomic_inc_not_zero(&p->refcnt);
138 }
139
140 static inline void ip6addrlbl_put(struct ip6addrlbl_entry *p)
141 {
142         if (atomic_dec_and_test(&p->refcnt))
143                 call_rcu(&p->rcu, ip6addrlbl_free_rcu);
144 }
145
146 /* Find label */
147 static int __ip6addrlbl_match(struct net *net,
148                               struct ip6addrlbl_entry *p,
149                               const struct in6_addr *addr,
150                               int addrtype, int ifindex)
151 {
152         if (!net_eq(ip6addrlbl_net(p), net))
153                 return 0;
154         if (p->ifindex && p->ifindex != ifindex)
155                 return 0;
156         if (p->addrtype && p->addrtype != addrtype)
157                 return 0;
158         if (!ipv6_prefix_equal(addr, &p->prefix, p->prefixlen))
159                 return 0;
160         return 1;
161 }
162
163 static struct ip6addrlbl_entry *__ipv6_addr_label(struct net *net,
164                                                   const struct in6_addr *addr,
165                                                   int type, int ifindex)
166 {
167         struct hlist_node *pos;
168         struct ip6addrlbl_entry *p;
169         hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
170                 if (__ip6addrlbl_match(net, p, addr, type, ifindex))
171                         return p;
172         }
173         return NULL;
174 }
175
176 u32 ipv6_addr_label(struct net *net,
177                     const struct in6_addr *addr, int type, int ifindex)
178 {
179         u32 label;
180         struct ip6addrlbl_entry *p;
181
182         type &= IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK;
183
184         rcu_read_lock();
185         p = __ipv6_addr_label(net, addr, type, ifindex);
186         label = p ? p->label : IPV6_ADDR_LABEL_DEFAULT;
187         rcu_read_unlock();
188
189         ADDRLABEL(KERN_DEBUG "%s(addr=" NIP6_FMT ", type=%d, ifindex=%d) => %08x\n",
190                         __func__,
191                         NIP6(*addr), type, ifindex,
192                         label);
193
194         return label;
195 }
196
197 /* allocate one entry */
198 static struct ip6addrlbl_entry *ip6addrlbl_alloc(struct net *net,
199                                                  const struct in6_addr *prefix,
200                                                  int prefixlen, int ifindex,
201                                                  u32 label)
202 {
203         struct ip6addrlbl_entry *newp;
204         int addrtype;
205
206         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d, label=%u)\n",
207                         __func__,
208                         NIP6(*prefix), prefixlen,
209                         ifindex,
210                         (unsigned int)label);
211
212         addrtype = ipv6_addr_type(prefix) & (IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK);
213
214         switch (addrtype) {
215         case IPV6_ADDR_MAPPED:
216                 if (prefixlen > 96)
217                         return ERR_PTR(-EINVAL);
218                 if (prefixlen < 96)
219                         addrtype = 0;
220                 break;
221         case IPV6_ADDR_COMPATv4:
222                 if (prefixlen != 96)
223                         addrtype = 0;
224                 break;
225         case IPV6_ADDR_LOOPBACK:
226                 if (prefixlen != 128)
227                         addrtype = 0;
228                 break;
229         }
230
231         newp = kmalloc(sizeof(*newp), GFP_KERNEL);
232         if (!newp)
233                 return ERR_PTR(-ENOMEM);
234
235         ipv6_addr_prefix(&newp->prefix, prefix, prefixlen);
236         newp->prefixlen = prefixlen;
237         newp->ifindex = ifindex;
238         newp->addrtype = addrtype;
239         newp->label = label;
240         INIT_HLIST_NODE(&newp->list);
241 #ifdef CONFIG_NET_NS
242         newp->lbl_net = hold_net(net);
243 #endif
244         atomic_set(&newp->refcnt, 1);
245         return newp;
246 }
247
248 /* add a label */
249 static int __ip6addrlbl_add(struct ip6addrlbl_entry *newp, int replace)
250 {
251         int ret = 0;
252
253         ADDRLABEL(KERN_DEBUG "%s(newp=%p, replace=%d)\n",
254                         __func__,
255                         newp, replace);
256
257         if (hlist_empty(&ip6addrlbl_table.head)) {
258                 hlist_add_head_rcu(&newp->list, &ip6addrlbl_table.head);
259         } else {
260                 struct hlist_node *pos, *n;
261                 struct ip6addrlbl_entry *p = NULL;
262                 hlist_for_each_entry_safe(p, pos, n,
263                                           &ip6addrlbl_table.head, list) {
264                         if (p->prefixlen == newp->prefixlen &&
265                             net_eq(ip6addrlbl_net(p), ip6addrlbl_net(newp)) &&
266                             p->ifindex == newp->ifindex &&
267                             ipv6_addr_equal(&p->prefix, &newp->prefix)) {
268                                 if (!replace) {
269                                         ret = -EEXIST;
270                                         goto out;
271                                 }
272                                 hlist_replace_rcu(&p->list, &newp->list);
273                                 ip6addrlbl_put(p);
274                                 goto out;
275                         } else if ((p->prefixlen == newp->prefixlen && !p->ifindex) ||
276                                    (p->prefixlen < newp->prefixlen)) {
277                                 hlist_add_before_rcu(&newp->list, &p->list);
278                                 goto out;
279                         }
280                 }
281                 hlist_add_after_rcu(&p->list, &newp->list);
282         }
283 out:
284         if (!ret)
285                 ip6addrlbl_table.seq++;
286         return ret;
287 }
288
289 /* add a label */
290 static int ip6addrlbl_add(struct net *net,
291                           const struct in6_addr *prefix, int prefixlen,
292                           int ifindex, u32 label, int replace)
293 {
294         struct ip6addrlbl_entry *newp;
295         int ret = 0;
296
297         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d, label=%u, replace=%d)\n",
298                         __func__,
299                         NIP6(*prefix), prefixlen,
300                         ifindex,
301                         (unsigned int)label,
302                         replace);
303
304         newp = ip6addrlbl_alloc(net, prefix, prefixlen, ifindex, label);
305         if (IS_ERR(newp))
306                 return PTR_ERR(newp);
307         spin_lock(&ip6addrlbl_table.lock);
308         ret = __ip6addrlbl_add(newp, replace);
309         spin_unlock(&ip6addrlbl_table.lock);
310         if (ret)
311                 ip6addrlbl_free(newp);
312         return ret;
313 }
314
315 /* remove a label */
316 static int __ip6addrlbl_del(struct net *net,
317                             const struct in6_addr *prefix, int prefixlen,
318                             int ifindex)
319 {
320         struct ip6addrlbl_entry *p = NULL;
321         struct hlist_node *pos, *n;
322         int ret = -ESRCH;
323
324         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d)\n",
325                         __func__,
326                         NIP6(*prefix), prefixlen,
327                         ifindex);
328
329         hlist_for_each_entry_safe(p, pos, n, &ip6addrlbl_table.head, list) {
330                 if (p->prefixlen == prefixlen &&
331                     net_eq(ip6addrlbl_net(p), net) &&
332                     p->ifindex == ifindex &&
333                     ipv6_addr_equal(&p->prefix, prefix)) {
334                         hlist_del_rcu(&p->list);
335                         ip6addrlbl_put(p);
336                         ret = 0;
337                         break;
338                 }
339         }
340         return ret;
341 }
342
343 static int ip6addrlbl_del(struct net *net,
344                           const struct in6_addr *prefix, int prefixlen,
345                           int ifindex)
346 {
347         struct in6_addr prefix_buf;
348         int ret;
349
350         ADDRLABEL(KERN_DEBUG "%s(prefix=" NIP6_FMT ", prefixlen=%d, ifindex=%d)\n",
351                         __func__,
352                         NIP6(*prefix), prefixlen,
353                         ifindex);
354
355         ipv6_addr_prefix(&prefix_buf, prefix, prefixlen);
356         spin_lock(&ip6addrlbl_table.lock);
357         ret = __ip6addrlbl_del(net, &prefix_buf, prefixlen, ifindex);
358         spin_unlock(&ip6addrlbl_table.lock);
359         return ret;
360 }
361
362 /* add default label */
363 static int __net_init ip6addrlbl_net_init(struct net *net)
364 {
365         int err = 0;
366         int i;
367
368         ADDRLABEL(KERN_DEBUG "%s()\n", __func__);
369
370         for (i = 0; i < ARRAY_SIZE(ip6addrlbl_init_table); i++) {
371                 int ret = ip6addrlbl_add(net,
372                                          ip6addrlbl_init_table[i].prefix,
373                                          ip6addrlbl_init_table[i].prefixlen,
374                                          0,
375                                          ip6addrlbl_init_table[i].label, 0);
376                 /* XXX: should we free all rules when we catch an error? */
377                 if (ret && (!err || err != -ENOMEM))
378                         err = ret;
379         }
380         return err;
381 }
382
383 static void __net_exit ip6addrlbl_net_exit(struct net *net)
384 {
385         struct ip6addrlbl_entry *p = NULL;
386         struct hlist_node *pos, *n;
387
388         /* Remove all labels belonging to the exiting net */
389         spin_lock(&ip6addrlbl_table.lock);
390         hlist_for_each_entry_safe(p, pos, n, &ip6addrlbl_table.head, list) {
391                 if (net_eq(ip6addrlbl_net(p), net)) {
392                         hlist_del_rcu(&p->list);
393                         ip6addrlbl_put(p);
394                 }
395         }
396         spin_unlock(&ip6addrlbl_table.lock);
397 }
398
399 static struct pernet_operations ipv6_addr_label_ops = {
400         .init = ip6addrlbl_net_init,
401         .exit = ip6addrlbl_net_exit,
402 };
403
404 int __init ipv6_addr_label_init(void)
405 {
406         spin_lock_init(&ip6addrlbl_table.lock);
407
408         return register_pernet_subsys(&ipv6_addr_label_ops);
409 }
410
411 static const struct nla_policy ifal_policy[IFAL_MAX+1] = {
412         [IFAL_ADDRESS]          = { .len = sizeof(struct in6_addr), },
413         [IFAL_LABEL]            = { .len = sizeof(u32), },
414 };
415
416 static int ip6addrlbl_newdel(struct sk_buff *skb, struct nlmsghdr *nlh,
417                              void *arg)
418 {
419         struct net *net = sock_net(skb->sk);
420         struct ifaddrlblmsg *ifal;
421         struct nlattr *tb[IFAL_MAX+1];
422         struct in6_addr *pfx;
423         u32 label;
424         int err = 0;
425
426         err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
427         if (err < 0)
428                 return err;
429
430         ifal = nlmsg_data(nlh);
431
432         if (ifal->ifal_family != AF_INET6 ||
433             ifal->ifal_prefixlen > 128)
434                 return -EINVAL;
435
436         if (ifal->ifal_index &&
437             !__dev_get_by_index(net, ifal->ifal_index))
438                 return -EINVAL;
439
440         if (!tb[IFAL_ADDRESS])
441                 return -EINVAL;
442
443         pfx = nla_data(tb[IFAL_ADDRESS]);
444         if (!pfx)
445                 return -EINVAL;
446
447         if (!tb[IFAL_LABEL])
448                 return -EINVAL;
449         label = nla_get_u32(tb[IFAL_LABEL]);
450         if (label == IPV6_ADDR_LABEL_DEFAULT)
451                 return -EINVAL;
452
453         switch(nlh->nlmsg_type) {
454         case RTM_NEWADDRLABEL:
455                 err = ip6addrlbl_add(net, pfx, ifal->ifal_prefixlen,
456                                      ifal->ifal_index, label,
457                                      nlh->nlmsg_flags & NLM_F_REPLACE);
458                 break;
459         case RTM_DELADDRLABEL:
460                 err = ip6addrlbl_del(net, pfx, ifal->ifal_prefixlen,
461                                      ifal->ifal_index);
462                 break;
463         default:
464                 err = -EOPNOTSUPP;
465         }
466         return err;
467 }
468
469 static inline void ip6addrlbl_putmsg(struct nlmsghdr *nlh,
470                                      int prefixlen, int ifindex, u32 lseq)
471 {
472         struct ifaddrlblmsg *ifal = nlmsg_data(nlh);
473         ifal->ifal_family = AF_INET6;
474         ifal->ifal_prefixlen = prefixlen;
475         ifal->ifal_flags = 0;
476         ifal->ifal_index = ifindex;
477         ifal->ifal_seq = lseq;
478 };
479
480 static int ip6addrlbl_fill(struct sk_buff *skb,
481                            struct ip6addrlbl_entry *p,
482                            u32 lseq,
483                            u32 pid, u32 seq, int event,
484                            unsigned int flags)
485 {
486         struct nlmsghdr *nlh = nlmsg_put(skb, pid, seq, event,
487                                          sizeof(struct ifaddrlblmsg), flags);
488         if (!nlh)
489                 return -EMSGSIZE;
490
491         ip6addrlbl_putmsg(nlh, p->prefixlen, p->ifindex, lseq);
492
493         if (nla_put(skb, IFAL_ADDRESS, 16, &p->prefix) < 0 ||
494             nla_put_u32(skb, IFAL_LABEL, p->label) < 0) {
495                 nlmsg_cancel(skb, nlh);
496                 return -EMSGSIZE;
497         }
498
499         return nlmsg_end(skb, nlh);
500 }
501
502 static int ip6addrlbl_dump(struct sk_buff *skb, struct netlink_callback *cb)
503 {
504         struct net *net = sock_net(skb->sk);
505         struct ip6addrlbl_entry *p;
506         struct hlist_node *pos;
507         int idx = 0, s_idx = cb->args[0];
508         int err;
509
510         rcu_read_lock();
511         hlist_for_each_entry_rcu(p, pos, &ip6addrlbl_table.head, list) {
512                 if (idx >= s_idx &&
513                     net_eq(ip6addrlbl_net(p), net)) {
514                         if ((err = ip6addrlbl_fill(skb, p,
515                                                    ip6addrlbl_table.seq,
516                                                    NETLINK_CB(cb->skb).pid,
517                                                    cb->nlh->nlmsg_seq,
518                                                    RTM_NEWADDRLABEL,
519                                                    NLM_F_MULTI)) <= 0)
520                                 break;
521                 }
522                 idx++;
523         }
524         rcu_read_unlock();
525         cb->args[0] = idx;
526         return skb->len;
527 }
528
529 static inline int ip6addrlbl_msgsize(void)
530 {
531         return (NLMSG_ALIGN(sizeof(struct ifaddrlblmsg))
532                 + nla_total_size(16)    /* IFAL_ADDRESS */
533                 + nla_total_size(4)     /* IFAL_LABEL */
534         );
535 }
536
537 static int ip6addrlbl_get(struct sk_buff *in_skb, struct nlmsghdr* nlh,
538                           void *arg)
539 {
540         struct net *net = sock_net(in_skb->sk);
541         struct ifaddrlblmsg *ifal;
542         struct nlattr *tb[IFAL_MAX+1];
543         struct in6_addr *addr;
544         u32 lseq;
545         int err = 0;
546         struct ip6addrlbl_entry *p;
547         struct sk_buff *skb;
548
549         err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
550         if (err < 0)
551                 return err;
552
553         ifal = nlmsg_data(nlh);
554
555         if (ifal->ifal_family != AF_INET6 ||
556             ifal->ifal_prefixlen != 128)
557                 return -EINVAL;
558
559         if (ifal->ifal_index &&
560             !__dev_get_by_index(net, ifal->ifal_index))
561                 return -EINVAL;
562
563         if (!tb[IFAL_ADDRESS])
564                 return -EINVAL;
565
566         addr = nla_data(tb[IFAL_ADDRESS]);
567         if (!addr)
568                 return -EINVAL;
569
570         rcu_read_lock();
571         p = __ipv6_addr_label(net, addr, ipv6_addr_type(addr), ifal->ifal_index);
572         if (p && ip6addrlbl_hold(p))
573                 p = NULL;
574         lseq = ip6addrlbl_table.seq;
575         rcu_read_unlock();
576
577         if (!p) {
578                 err = -ESRCH;
579                 goto out;
580         }
581
582         if (!(skb = nlmsg_new(ip6addrlbl_msgsize(), GFP_KERNEL))) {
583                 ip6addrlbl_put(p);
584                 return -ENOBUFS;
585         }
586
587         err = ip6addrlbl_fill(skb, p, lseq,
588                               NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
589                               RTM_NEWADDRLABEL, 0);
590
591         ip6addrlbl_put(p);
592
593         if (err < 0) {
594                 WARN_ON(err == -EMSGSIZE);
595                 kfree_skb(skb);
596                 goto out;
597         }
598
599         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
600 out:
601         return err;
602 }
603
604 void __init ipv6_addr_label_rtnl_register(void)
605 {
606         __rtnl_register(PF_INET6, RTM_NEWADDRLABEL, ip6addrlbl_newdel, NULL);
607         __rtnl_register(PF_INET6, RTM_DELADDRLABEL, ip6addrlbl_newdel, NULL);
608         __rtnl_register(PF_INET6, RTM_GETADDRLABEL, ip6addrlbl_get, ip6addrlbl_dump);
609 }
610