Pull regset into release branch
[linux-2.6] / net / ipv6 / anycast.c
1 /*
2  *      Anycast support for IPv6
3  *      Linux INET6 implementation
4  *
5  *      Authors:
6  *      David L Stevens (dlstevens@us.ibm.com)
7  *
8  *      based heavily on net/ipv6/mcast.c
9  *
10  *      This program is free software; you can redistribute it and/or
11  *      modify it under the terms of the GNU General Public License
12  *      as published by the Free Software Foundation; either version
13  *      2 of the License, or (at your option) any later version.
14  */
15
16 #include <linux/capability.h>
17 #include <linux/module.h>
18 #include <linux/errno.h>
19 #include <linux/types.h>
20 #include <linux/random.h>
21 #include <linux/string.h>
22 #include <linux/socket.h>
23 #include <linux/sockios.h>
24 #include <linux/net.h>
25 #include <linux/in6.h>
26 #include <linux/netdevice.h>
27 #include <linux/if_arp.h>
28 #include <linux/route.h>
29 #include <linux/init.h>
30 #include <linux/proc_fs.h>
31 #include <linux/seq_file.h>
32
33 #include <net/net_namespace.h>
34 #include <net/sock.h>
35 #include <net/snmp.h>
36
37 #include <net/ipv6.h>
38 #include <net/protocol.h>
39 #include <net/if_inet6.h>
40 #include <net/ndisc.h>
41 #include <net/addrconf.h>
42 #include <net/ip6_route.h>
43
44 #include <net/checksum.h>
45
46 static int ipv6_dev_ac_dec(struct net_device *dev, struct in6_addr *addr);
47
48 /* Big ac list lock for all the sockets */
49 static DEFINE_RWLOCK(ipv6_sk_ac_lock);
50
51 static int
52 ip6_onlink(struct in6_addr *addr, struct net_device *dev)
53 {
54         struct inet6_dev        *idev;
55         struct inet6_ifaddr     *ifa;
56         int     onlink;
57
58         onlink = 0;
59         rcu_read_lock();
60         idev = __in6_dev_get(dev);
61         if (idev) {
62                 read_lock_bh(&idev->lock);
63                 for (ifa=idev->addr_list; ifa; ifa=ifa->if_next) {
64                         onlink = ipv6_prefix_equal(addr, &ifa->addr,
65                                                    ifa->prefix_len);
66                         if (onlink)
67                                 break;
68                 }
69                 read_unlock_bh(&idev->lock);
70         }
71         rcu_read_unlock();
72         return onlink;
73 }
74
75 /*
76  *      socket join an anycast group
77  */
78
79 int ipv6_sock_ac_join(struct sock *sk, int ifindex, struct in6_addr *addr)
80 {
81         struct ipv6_pinfo *np = inet6_sk(sk);
82         struct net_device *dev = NULL;
83         struct inet6_dev *idev;
84         struct ipv6_ac_socklist *pac;
85         int     ishost = !ipv6_devconf.forwarding;
86         int     err = 0;
87
88         if (!capable(CAP_NET_ADMIN))
89                 return -EPERM;
90         if (ipv6_addr_is_multicast(addr))
91                 return -EINVAL;
92         if (ipv6_chk_addr(&init_net, addr, NULL, 0))
93                 return -EINVAL;
94
95         pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL);
96         if (pac == NULL)
97                 return -ENOMEM;
98         pac->acl_next = NULL;
99         ipv6_addr_copy(&pac->acl_addr, addr);
100
101         if (ifindex == 0) {
102                 struct rt6_info *rt;
103
104                 rt = rt6_lookup(addr, NULL, 0, 0);
105                 if (rt) {
106                         dev = rt->rt6i_dev;
107                         dev_hold(dev);
108                         dst_release(&rt->u.dst);
109                 } else if (ishost) {
110                         err = -EADDRNOTAVAIL;
111                         goto out_free_pac;
112                 } else {
113                         /* router, no matching interface: just pick one */
114
115                         dev = dev_get_by_flags(&init_net, IFF_UP, IFF_UP|IFF_LOOPBACK);
116                 }
117         } else
118                 dev = dev_get_by_index(&init_net, ifindex);
119
120         if (dev == NULL) {
121                 err = -ENODEV;
122                 goto out_free_pac;
123         }
124
125         idev = in6_dev_get(dev);
126         if (!idev) {
127                 if (ifindex)
128                         err = -ENODEV;
129                 else
130                         err = -EADDRNOTAVAIL;
131                 goto out_dev_put;
132         }
133         /* reset ishost, now that we have a specific device */
134         ishost = !idev->cnf.forwarding;
135         in6_dev_put(idev);
136
137         pac->acl_ifindex = dev->ifindex;
138
139         /* XXX
140          * For hosts, allow link-local or matching prefix anycasts.
141          * This obviates the need for propagating anycast routes while
142          * still allowing some non-router anycast participation.
143          */
144         if (!ip6_onlink(addr, dev)) {
145                 if (ishost)
146                         err = -EADDRNOTAVAIL;
147                 if (err)
148                         goto out_dev_put;
149         }
150
151         err = ipv6_dev_ac_inc(dev, addr);
152         if (err)
153                 goto out_dev_put;
154
155         write_lock_bh(&ipv6_sk_ac_lock);
156         pac->acl_next = np->ipv6_ac_list;
157         np->ipv6_ac_list = pac;
158         write_unlock_bh(&ipv6_sk_ac_lock);
159
160         dev_put(dev);
161
162         return 0;
163
164 out_dev_put:
165         dev_put(dev);
166 out_free_pac:
167         sock_kfree_s(sk, pac, sizeof(*pac));
168         return err;
169 }
170
171 /*
172  *      socket leave an anycast group
173  */
174 int ipv6_sock_ac_drop(struct sock *sk, int ifindex, struct in6_addr *addr)
175 {
176         struct ipv6_pinfo *np = inet6_sk(sk);
177         struct net_device *dev;
178         struct ipv6_ac_socklist *pac, *prev_pac;
179
180         write_lock_bh(&ipv6_sk_ac_lock);
181         prev_pac = NULL;
182         for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) {
183                 if ((ifindex == 0 || pac->acl_ifindex == ifindex) &&
184                      ipv6_addr_equal(&pac->acl_addr, addr))
185                         break;
186                 prev_pac = pac;
187         }
188         if (!pac) {
189                 write_unlock_bh(&ipv6_sk_ac_lock);
190                 return -ENOENT;
191         }
192         if (prev_pac)
193                 prev_pac->acl_next = pac->acl_next;
194         else
195                 np->ipv6_ac_list = pac->acl_next;
196
197         write_unlock_bh(&ipv6_sk_ac_lock);
198
199         dev = dev_get_by_index(&init_net, pac->acl_ifindex);
200         if (dev) {
201                 ipv6_dev_ac_dec(dev, &pac->acl_addr);
202                 dev_put(dev);
203         }
204         sock_kfree_s(sk, pac, sizeof(*pac));
205         return 0;
206 }
207
208 void ipv6_sock_ac_close(struct sock *sk)
209 {
210         struct ipv6_pinfo *np = inet6_sk(sk);
211         struct net_device *dev = NULL;
212         struct ipv6_ac_socklist *pac;
213         int     prev_index;
214
215         write_lock_bh(&ipv6_sk_ac_lock);
216         pac = np->ipv6_ac_list;
217         np->ipv6_ac_list = NULL;
218         write_unlock_bh(&ipv6_sk_ac_lock);
219
220         prev_index = 0;
221         while (pac) {
222                 struct ipv6_ac_socklist *next = pac->acl_next;
223
224                 if (pac->acl_ifindex != prev_index) {
225                         if (dev)
226                                 dev_put(dev);
227                         dev = dev_get_by_index(&init_net, pac->acl_ifindex);
228                         prev_index = pac->acl_ifindex;
229                 }
230                 if (dev)
231                         ipv6_dev_ac_dec(dev, &pac->acl_addr);
232                 sock_kfree_s(sk, pac, sizeof(*pac));
233                 pac = next;
234         }
235         if (dev)
236                 dev_put(dev);
237 }
238
239 #if 0
240 /* The function is not used, which is funny. Apparently, author
241  * supposed to use it to filter out datagrams inside udp/raw but forgot.
242  *
243  * It is OK, anycasts are not special comparing to delivery to unicasts.
244  */
245
246 int inet6_ac_check(struct sock *sk, struct in6_addr *addr, int ifindex)
247 {
248         struct ipv6_ac_socklist *pac;
249         struct ipv6_pinfo *np = inet6_sk(sk);
250         int     found;
251
252         found = 0;
253         read_lock(&ipv6_sk_ac_lock);
254         for (pac=np->ipv6_ac_list; pac; pac=pac->acl_next) {
255                 if (ifindex && pac->acl_ifindex != ifindex)
256                         continue;
257                 found = ipv6_addr_equal(&pac->acl_addr, addr);
258                 if (found)
259                         break;
260         }
261         read_unlock(&ipv6_sk_ac_lock);
262
263         return found;
264 }
265
266 #endif
267
268 static void aca_put(struct ifacaddr6 *ac)
269 {
270         if (atomic_dec_and_test(&ac->aca_refcnt)) {
271                 in6_dev_put(ac->aca_idev);
272                 dst_release(&ac->aca_rt->u.dst);
273                 kfree(ac);
274         }
275 }
276
277 /*
278  *      device anycast group inc (add if not found)
279  */
280 int ipv6_dev_ac_inc(struct net_device *dev, struct in6_addr *addr)
281 {
282         struct ifacaddr6 *aca;
283         struct inet6_dev *idev;
284         struct rt6_info *rt;
285         int err;
286
287         idev = in6_dev_get(dev);
288
289         if (idev == NULL)
290                 return -EINVAL;
291
292         write_lock_bh(&idev->lock);
293         if (idev->dead) {
294                 err = -ENODEV;
295                 goto out;
296         }
297
298         for (aca = idev->ac_list; aca; aca = aca->aca_next) {
299                 if (ipv6_addr_equal(&aca->aca_addr, addr)) {
300                         aca->aca_users++;
301                         err = 0;
302                         goto out;
303                 }
304         }
305
306         /*
307          *      not found: create a new one.
308          */
309
310         aca = kzalloc(sizeof(struct ifacaddr6), GFP_ATOMIC);
311
312         if (aca == NULL) {
313                 err = -ENOMEM;
314                 goto out;
315         }
316
317         rt = addrconf_dst_alloc(idev, addr, 1);
318         if (IS_ERR(rt)) {
319                 kfree(aca);
320                 err = PTR_ERR(rt);
321                 goto out;
322         }
323
324         ipv6_addr_copy(&aca->aca_addr, addr);
325         aca->aca_idev = idev;
326         aca->aca_rt = rt;
327         aca->aca_users = 1;
328         /* aca_tstamp should be updated upon changes */
329         aca->aca_cstamp = aca->aca_tstamp = jiffies;
330         atomic_set(&aca->aca_refcnt, 2);
331         spin_lock_init(&aca->aca_lock);
332
333         aca->aca_next = idev->ac_list;
334         idev->ac_list = aca;
335         write_unlock_bh(&idev->lock);
336
337         ip6_ins_rt(rt);
338
339         addrconf_join_solict(dev, &aca->aca_addr);
340
341         aca_put(aca);
342         return 0;
343 out:
344         write_unlock_bh(&idev->lock);
345         in6_dev_put(idev);
346         return err;
347 }
348
349 /*
350  *      device anycast group decrement
351  */
352 int __ipv6_dev_ac_dec(struct inet6_dev *idev, struct in6_addr *addr)
353 {
354         struct ifacaddr6 *aca, *prev_aca;
355
356         write_lock_bh(&idev->lock);
357         prev_aca = NULL;
358         for (aca = idev->ac_list; aca; aca = aca->aca_next) {
359                 if (ipv6_addr_equal(&aca->aca_addr, addr))
360                         break;
361                 prev_aca = aca;
362         }
363         if (!aca) {
364                 write_unlock_bh(&idev->lock);
365                 return -ENOENT;
366         }
367         if (--aca->aca_users > 0) {
368                 write_unlock_bh(&idev->lock);
369                 return 0;
370         }
371         if (prev_aca)
372                 prev_aca->aca_next = aca->aca_next;
373         else
374                 idev->ac_list = aca->aca_next;
375         write_unlock_bh(&idev->lock);
376         addrconf_leave_solict(idev, &aca->aca_addr);
377
378         dst_hold(&aca->aca_rt->u.dst);
379         ip6_del_rt(aca->aca_rt);
380
381         aca_put(aca);
382         return 0;
383 }
384
385 static int ipv6_dev_ac_dec(struct net_device *dev, struct in6_addr *addr)
386 {
387         int ret;
388         struct inet6_dev *idev = in6_dev_get(dev);
389         if (idev == NULL)
390                 return -ENODEV;
391         ret = __ipv6_dev_ac_dec(idev, addr);
392         in6_dev_put(idev);
393         return ret;
394 }
395
396 /*
397  *      check if the interface has this anycast address
398  */
399 static int ipv6_chk_acast_dev(struct net_device *dev, struct in6_addr *addr)
400 {
401         struct inet6_dev *idev;
402         struct ifacaddr6 *aca;
403
404         idev = in6_dev_get(dev);
405         if (idev) {
406                 read_lock_bh(&idev->lock);
407                 for (aca = idev->ac_list; aca; aca = aca->aca_next)
408                         if (ipv6_addr_equal(&aca->aca_addr, addr))
409                                 break;
410                 read_unlock_bh(&idev->lock);
411                 in6_dev_put(idev);
412                 return aca != NULL;
413         }
414         return 0;
415 }
416
417 /*
418  *      check if given interface (or any, if dev==0) has this anycast address
419  */
420 int ipv6_chk_acast_addr(struct net_device *dev, struct in6_addr *addr)
421 {
422         int found = 0;
423
424         if (dev)
425                 return ipv6_chk_acast_dev(dev, addr);
426         read_lock(&dev_base_lock);
427         for_each_netdev(&init_net, dev)
428                 if (ipv6_chk_acast_dev(dev, addr)) {
429                         found = 1;
430                         break;
431                 }
432         read_unlock(&dev_base_lock);
433         return found;
434 }
435
436
437 #ifdef CONFIG_PROC_FS
438 struct ac6_iter_state {
439         struct net_device *dev;
440         struct inet6_dev *idev;
441 };
442
443 #define ac6_seq_private(seq)    ((struct ac6_iter_state *)(seq)->private)
444
445 static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq)
446 {
447         struct ifacaddr6 *im = NULL;
448         struct ac6_iter_state *state = ac6_seq_private(seq);
449
450         state->idev = NULL;
451         for_each_netdev(&init_net, state->dev) {
452                 struct inet6_dev *idev;
453                 idev = in6_dev_get(state->dev);
454                 if (!idev)
455                         continue;
456                 read_lock_bh(&idev->lock);
457                 im = idev->ac_list;
458                 if (im) {
459                         state->idev = idev;
460                         break;
461                 }
462                 read_unlock_bh(&idev->lock);
463                 in6_dev_put(idev);
464         }
465         return im;
466 }
467
468 static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im)
469 {
470         struct ac6_iter_state *state = ac6_seq_private(seq);
471
472         im = im->aca_next;
473         while (!im) {
474                 if (likely(state->idev != NULL)) {
475                         read_unlock_bh(&state->idev->lock);
476                         in6_dev_put(state->idev);
477                 }
478                 state->dev = next_net_device(state->dev);
479                 if (!state->dev) {
480                         state->idev = NULL;
481                         break;
482                 }
483                 state->idev = in6_dev_get(state->dev);
484                 if (!state->idev)
485                         continue;
486                 read_lock_bh(&state->idev->lock);
487                 im = state->idev->ac_list;
488         }
489         return im;
490 }
491
492 static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos)
493 {
494         struct ifacaddr6 *im = ac6_get_first(seq);
495         if (im)
496                 while (pos && (im = ac6_get_next(seq, im)) != NULL)
497                         --pos;
498         return pos ? NULL : im;
499 }
500
501 static void *ac6_seq_start(struct seq_file *seq, loff_t *pos)
502         __acquires(dev_base_lock)
503 {
504         read_lock(&dev_base_lock);
505         return ac6_get_idx(seq, *pos);
506 }
507
508 static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
509 {
510         struct ifacaddr6 *im;
511         im = ac6_get_next(seq, v);
512         ++*pos;
513         return im;
514 }
515
516 static void ac6_seq_stop(struct seq_file *seq, void *v)
517         __releases(dev_base_lock)
518 {
519         struct ac6_iter_state *state = ac6_seq_private(seq);
520         if (likely(state->idev != NULL)) {
521                 read_unlock_bh(&state->idev->lock);
522                 in6_dev_put(state->idev);
523         }
524         read_unlock(&dev_base_lock);
525 }
526
527 static int ac6_seq_show(struct seq_file *seq, void *v)
528 {
529         struct ifacaddr6 *im = (struct ifacaddr6 *)v;
530         struct ac6_iter_state *state = ac6_seq_private(seq);
531
532         seq_printf(seq,
533                    "%-4d %-15s " NIP6_SEQFMT " %5d\n",
534                    state->dev->ifindex, state->dev->name,
535                    NIP6(im->aca_addr),
536                    im->aca_users);
537         return 0;
538 }
539
540 static const struct seq_operations ac6_seq_ops = {
541         .start  =       ac6_seq_start,
542         .next   =       ac6_seq_next,
543         .stop   =       ac6_seq_stop,
544         .show   =       ac6_seq_show,
545 };
546
547 static int ac6_seq_open(struct inode *inode, struct file *file)
548 {
549         return seq_open_private(file, &ac6_seq_ops,
550                         sizeof(struct ac6_iter_state));
551 }
552
553 static const struct file_operations ac6_seq_fops = {
554         .owner          =       THIS_MODULE,
555         .open           =       ac6_seq_open,
556         .read           =       seq_read,
557         .llseek         =       seq_lseek,
558         .release        =       seq_release_private,
559 };
560
561 int __init ac6_proc_init(void)
562 {
563         if (!proc_net_fops_create(&init_net, "anycast6", S_IRUGO, &ac6_seq_fops))
564                 return -ENOMEM;
565
566         return 0;
567 }
568
569 void ac6_proc_exit(void)
570 {
571         proc_net_remove(&init_net, "anycast6");
572 }
573 #endif
574