Merge git://git.infradead.org/mtd-2.6
[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/sock.h>
34 #include <net/snmp.h>
35
36 #include <net/ipv6.h>
37 #include <net/protocol.h>
38 #include <net/if_inet6.h>
39 #include <net/ndisc.h>
40 #include <net/addrconf.h>
41 #include <net/ip6_route.h>
42
43 #include <net/checksum.h>
44
45 static int ipv6_dev_ac_dec(struct net_device *dev, struct in6_addr *addr);
46
47 /* Big ac list lock for all the sockets */
48 static DEFINE_RWLOCK(ipv6_sk_ac_lock);
49
50 static int
51 ip6_onlink(struct in6_addr *addr, struct net_device *dev)
52 {
53         struct inet6_dev        *idev;
54         struct inet6_ifaddr     *ifa;
55         int     onlink;
56
57         onlink = 0;
58         rcu_read_lock();
59         idev = __in6_dev_get(dev);
60         if (idev) {
61                 read_lock_bh(&idev->lock);
62                 for (ifa=idev->addr_list; ifa; ifa=ifa->if_next) {
63                         onlink = ipv6_prefix_equal(addr, &ifa->addr,
64                                                    ifa->prefix_len);
65                         if (onlink)
66                                 break;
67                 }
68                 read_unlock_bh(&idev->lock);
69         }
70         rcu_read_unlock();
71         return onlink;
72 }
73
74 /*
75  *      socket join an anycast group
76  */
77
78 int ipv6_sock_ac_join(struct sock *sk, int ifindex, struct in6_addr *addr)
79 {
80         struct ipv6_pinfo *np = inet6_sk(sk);
81         struct net_device *dev = NULL;
82         struct inet6_dev *idev;
83         struct ipv6_ac_socklist *pac;
84         int     ishost = !ipv6_devconf.forwarding;
85         int     err = 0;
86
87         if (!capable(CAP_NET_ADMIN))
88                 return -EPERM;
89         if (ipv6_addr_is_multicast(addr))
90                 return -EINVAL;
91         if (ipv6_chk_addr(addr, NULL, 0))
92                 return -EINVAL;
93
94         pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL);
95         if (pac == NULL)
96                 return -ENOMEM;
97         pac->acl_next = NULL;
98         ipv6_addr_copy(&pac->acl_addr, addr);
99
100         if (ifindex == 0) {
101                 struct rt6_info *rt;
102
103                 rt = rt6_lookup(addr, NULL, 0, 0);
104                 if (rt) {
105                         dev = rt->rt6i_dev;
106                         dev_hold(dev);
107                         dst_release(&rt->u.dst);
108                 } else if (ishost) {
109                         err = -EADDRNOTAVAIL;
110                         goto out_free_pac;
111                 } else {
112                         /* router, no matching interface: just pick one */
113
114                         dev = dev_get_by_flags(IFF_UP, IFF_UP|IFF_LOOPBACK);
115                 }
116         } else
117                 dev = dev_get_by_index(ifindex);
118
119         if (dev == NULL) {
120                 err = -ENODEV;
121                 goto out_free_pac;
122         }
123
124         idev = in6_dev_get(dev);
125         if (!idev) {
126                 if (ifindex)
127                         err = -ENODEV;
128                 else
129                         err = -EADDRNOTAVAIL;
130                 goto out_dev_put;
131         }
132         /* reset ishost, now that we have a specific device */
133         ishost = !idev->cnf.forwarding;
134         in6_dev_put(idev);
135
136         pac->acl_ifindex = dev->ifindex;
137
138         /* XXX
139          * For hosts, allow link-local or matching prefix anycasts.
140          * This obviates the need for propagating anycast routes while
141          * still allowing some non-router anycast participation.
142          */
143         if (!ip6_onlink(addr, dev)) {
144                 if (ishost)
145                         err = -EADDRNOTAVAIL;
146                 if (err)
147                         goto out_dev_put;
148         }
149
150         err = ipv6_dev_ac_inc(dev, addr);
151         if (err)
152                 goto out_dev_put;
153
154         write_lock_bh(&ipv6_sk_ac_lock);
155         pac->acl_next = np->ipv6_ac_list;
156         np->ipv6_ac_list = pac;
157         write_unlock_bh(&ipv6_sk_ac_lock);
158
159         dev_put(dev);
160
161         return 0;
162
163 out_dev_put:
164         dev_put(dev);
165 out_free_pac:
166         sock_kfree_s(sk, pac, sizeof(*pac));
167         return err;
168 }
169
170 /*
171  *      socket leave an anycast group
172  */
173 int ipv6_sock_ac_drop(struct sock *sk, int ifindex, struct in6_addr *addr)
174 {
175         struct ipv6_pinfo *np = inet6_sk(sk);
176         struct net_device *dev;
177         struct ipv6_ac_socklist *pac, *prev_pac;
178
179         write_lock_bh(&ipv6_sk_ac_lock);
180         prev_pac = NULL;
181         for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) {
182                 if ((ifindex == 0 || pac->acl_ifindex == ifindex) &&
183                      ipv6_addr_equal(&pac->acl_addr, addr))
184                         break;
185                 prev_pac = pac;
186         }
187         if (!pac) {
188                 write_unlock_bh(&ipv6_sk_ac_lock);
189                 return -ENOENT;
190         }
191         if (prev_pac)
192                 prev_pac->acl_next = pac->acl_next;
193         else
194                 np->ipv6_ac_list = pac->acl_next;
195
196         write_unlock_bh(&ipv6_sk_ac_lock);
197
198         dev = dev_get_by_index(pac->acl_ifindex);
199         if (dev) {
200                 ipv6_dev_ac_dec(dev, &pac->acl_addr);
201                 dev_put(dev);
202         }
203         sock_kfree_s(sk, pac, sizeof(*pac));
204         return 0;
205 }
206
207 void ipv6_sock_ac_close(struct sock *sk)
208 {
209         struct ipv6_pinfo *np = inet6_sk(sk);
210         struct net_device *dev = NULL;
211         struct ipv6_ac_socklist *pac;
212         int     prev_index;
213
214         write_lock_bh(&ipv6_sk_ac_lock);
215         pac = np->ipv6_ac_list;
216         np->ipv6_ac_list = NULL;
217         write_unlock_bh(&ipv6_sk_ac_lock);
218
219         prev_index = 0;
220         while (pac) {
221                 struct ipv6_ac_socklist *next = pac->acl_next;
222
223                 if (pac->acl_ifindex != prev_index) {
224                         if (dev)
225                                 dev_put(dev);
226                         dev = dev_get_by_index(pac->acl_ifindex);
227                         prev_index = pac->acl_ifindex;
228                 }
229                 if (dev)
230                         ipv6_dev_ac_dec(dev, &pac->acl_addr);
231                 sock_kfree_s(sk, pac, sizeof(*pac));
232                 pac = next;
233         }
234         if (dev)
235                 dev_put(dev);
236 }
237
238 #if 0
239 /* The function is not used, which is funny. Apparently, author
240  * supposed to use it to filter out datagrams inside udp/raw but forgot.
241  *
242  * It is OK, anycasts are not special comparing to delivery to unicasts.
243  */
244
245 int inet6_ac_check(struct sock *sk, struct in6_addr *addr, int ifindex)
246 {
247         struct ipv6_ac_socklist *pac;
248         struct ipv6_pinfo *np = inet6_sk(sk);
249         int     found;
250
251         found = 0;
252         read_lock(&ipv6_sk_ac_lock);
253         for (pac=np->ipv6_ac_list; pac; pac=pac->acl_next) {
254                 if (ifindex && pac->acl_ifindex != ifindex)
255                         continue;
256                 found = ipv6_addr_equal(&pac->acl_addr, addr);
257                 if (found)
258                         break;
259         }
260         read_unlock(&ipv6_sk_ac_lock);
261
262         return found;
263 }
264
265 #endif
266
267 static void aca_put(struct ifacaddr6 *ac)
268 {
269         if (atomic_dec_and_test(&ac->aca_refcnt)) {
270                 in6_dev_put(ac->aca_idev);
271                 dst_release(&ac->aca_rt->u.dst);
272                 kfree(ac);
273         }
274 }
275
276 /*
277  *      device anycast group inc (add if not found)
278  */
279 int ipv6_dev_ac_inc(struct net_device *dev, struct in6_addr *addr)
280 {
281         struct ifacaddr6 *aca;
282         struct inet6_dev *idev;
283         struct rt6_info *rt;
284         int err;
285
286         idev = in6_dev_get(dev);
287
288         if (idev == NULL)
289                 return -EINVAL;
290
291         write_lock_bh(&idev->lock);
292         if (idev->dead) {
293                 err = -ENODEV;
294                 goto out;
295         }
296
297         for (aca = idev->ac_list; aca; aca = aca->aca_next) {
298                 if (ipv6_addr_equal(&aca->aca_addr, addr)) {
299                         aca->aca_users++;
300                         err = 0;
301                         goto out;
302                 }
303         }
304
305         /*
306          *      not found: create a new one.
307          */
308
309         aca = kzalloc(sizeof(struct ifacaddr6), GFP_ATOMIC);
310
311         if (aca == NULL) {
312                 err = -ENOMEM;
313                 goto out;
314         }
315
316         rt = addrconf_dst_alloc(idev, addr, 1);
317         if (IS_ERR(rt)) {
318                 kfree(aca);
319                 err = PTR_ERR(rt);
320                 goto out;
321         }
322
323         ipv6_addr_copy(&aca->aca_addr, addr);
324         aca->aca_idev = idev;
325         aca->aca_rt = rt;
326         aca->aca_users = 1;
327         /* aca_tstamp should be updated upon changes */
328         aca->aca_cstamp = aca->aca_tstamp = jiffies;
329         atomic_set(&aca->aca_refcnt, 2);
330         spin_lock_init(&aca->aca_lock);
331
332         aca->aca_next = idev->ac_list;
333         idev->ac_list = aca;
334         write_unlock_bh(&idev->lock);
335
336         dst_hold(&rt->u.dst);
337         if (ip6_ins_rt(rt))
338                 dst_release(&rt->u.dst);
339
340         addrconf_join_solict(dev, &aca->aca_addr);
341
342         aca_put(aca);
343         return 0;
344 out:
345         write_unlock_bh(&idev->lock);
346         in6_dev_put(idev);
347         return err;
348 }
349
350 /*
351  *      device anycast group decrement
352  */
353 int __ipv6_dev_ac_dec(struct inet6_dev *idev, struct in6_addr *addr)
354 {
355         struct ifacaddr6 *aca, *prev_aca;
356
357         write_lock_bh(&idev->lock);
358         prev_aca = NULL;
359         for (aca = idev->ac_list; aca; aca = aca->aca_next) {
360                 if (ipv6_addr_equal(&aca->aca_addr, addr))
361                         break;
362                 prev_aca = aca;
363         }
364         if (!aca) {
365                 write_unlock_bh(&idev->lock);
366                 return -ENOENT;
367         }
368         if (--aca->aca_users > 0) {
369                 write_unlock_bh(&idev->lock);
370                 return 0;
371         }
372         if (prev_aca)
373                 prev_aca->aca_next = aca->aca_next;
374         else
375                 idev->ac_list = aca->aca_next;
376         write_unlock_bh(&idev->lock);
377         addrconf_leave_solict(idev, &aca->aca_addr);
378
379         dst_hold(&aca->aca_rt->u.dst);
380         if (ip6_del_rt(aca->aca_rt))
381                 dst_free(&aca->aca_rt->u.dst);
382         else
383                 dst_release(&aca->aca_rt->u.dst);
384
385         aca_put(aca);
386         return 0;
387 }
388
389 static int ipv6_dev_ac_dec(struct net_device *dev, struct in6_addr *addr)
390 {
391         int ret;
392         struct inet6_dev *idev = in6_dev_get(dev);
393         if (idev == NULL)
394                 return -ENODEV;
395         ret = __ipv6_dev_ac_dec(idev, addr);
396         in6_dev_put(idev);
397         return ret;
398 }
399
400 /*
401  *      check if the interface has this anycast address
402  */
403 static int ipv6_chk_acast_dev(struct net_device *dev, struct in6_addr *addr)
404 {
405         struct inet6_dev *idev;
406         struct ifacaddr6 *aca;
407
408         idev = in6_dev_get(dev);
409         if (idev) {
410                 read_lock_bh(&idev->lock);
411                 for (aca = idev->ac_list; aca; aca = aca->aca_next)
412                         if (ipv6_addr_equal(&aca->aca_addr, addr))
413                                 break;
414                 read_unlock_bh(&idev->lock);
415                 in6_dev_put(idev);
416                 return aca != 0;
417         }
418         return 0;
419 }
420
421 /*
422  *      check if given interface (or any, if dev==0) has this anycast address
423  */
424 int ipv6_chk_acast_addr(struct net_device *dev, struct in6_addr *addr)
425 {
426         int found = 0;
427
428         if (dev)
429                 return ipv6_chk_acast_dev(dev, addr);
430         read_lock(&dev_base_lock);
431         for_each_netdev(dev)
432                 if (ipv6_chk_acast_dev(dev, addr)) {
433                         found = 1;
434                         break;
435                 }
436         read_unlock(&dev_base_lock);
437         return found;
438 }
439
440
441 #ifdef CONFIG_PROC_FS
442 struct ac6_iter_state {
443         struct net_device *dev;
444         struct inet6_dev *idev;
445 };
446
447 #define ac6_seq_private(seq)    ((struct ac6_iter_state *)(seq)->private)
448
449 static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq)
450 {
451         struct ifacaddr6 *im = NULL;
452         struct ac6_iter_state *state = ac6_seq_private(seq);
453
454         state->idev = NULL;
455         for_each_netdev(state->dev) {
456                 struct inet6_dev *idev;
457                 idev = in6_dev_get(state->dev);
458                 if (!idev)
459                         continue;
460                 read_lock_bh(&idev->lock);
461                 im = idev->ac_list;
462                 if (im) {
463                         state->idev = idev;
464                         break;
465                 }
466                 read_unlock_bh(&idev->lock);
467                 in6_dev_put(idev);
468         }
469         return im;
470 }
471
472 static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im)
473 {
474         struct ac6_iter_state *state = ac6_seq_private(seq);
475
476         im = im->aca_next;
477         while (!im) {
478                 if (likely(state->idev != NULL)) {
479                         read_unlock_bh(&state->idev->lock);
480                         in6_dev_put(state->idev);
481                 }
482                 state->dev = next_net_device(state->dev);
483                 if (!state->dev) {
484                         state->idev = NULL;
485                         break;
486                 }
487                 state->idev = in6_dev_get(state->dev);
488                 if (!state->idev)
489                         continue;
490                 read_lock_bh(&state->idev->lock);
491                 im = state->idev->ac_list;
492         }
493         return im;
494 }
495
496 static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos)
497 {
498         struct ifacaddr6 *im = ac6_get_first(seq);
499         if (im)
500                 while (pos && (im = ac6_get_next(seq, im)) != NULL)
501                         --pos;
502         return pos ? NULL : im;
503 }
504
505 static void *ac6_seq_start(struct seq_file *seq, loff_t *pos)
506 {
507         read_lock(&dev_base_lock);
508         return ac6_get_idx(seq, *pos);
509 }
510
511 static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
512 {
513         struct ifacaddr6 *im;
514         im = ac6_get_next(seq, v);
515         ++*pos;
516         return im;
517 }
518
519 static void ac6_seq_stop(struct seq_file *seq, void *v)
520 {
521         struct ac6_iter_state *state = ac6_seq_private(seq);
522         if (likely(state->idev != NULL)) {
523                 read_unlock_bh(&state->idev->lock);
524                 in6_dev_put(state->idev);
525         }
526         read_unlock(&dev_base_lock);
527 }
528
529 static int ac6_seq_show(struct seq_file *seq, void *v)
530 {
531         struct ifacaddr6 *im = (struct ifacaddr6 *)v;
532         struct ac6_iter_state *state = ac6_seq_private(seq);
533
534         seq_printf(seq,
535                    "%-4d %-15s " NIP6_SEQFMT " %5d\n",
536                    state->dev->ifindex, state->dev->name,
537                    NIP6(im->aca_addr),
538                    im->aca_users);
539         return 0;
540 }
541
542 static struct seq_operations ac6_seq_ops = {
543         .start  =       ac6_seq_start,
544         .next   =       ac6_seq_next,
545         .stop   =       ac6_seq_stop,
546         .show   =       ac6_seq_show,
547 };
548
549 static int ac6_seq_open(struct inode *inode, struct file *file)
550 {
551         struct seq_file *seq;
552         int rc = -ENOMEM;
553         struct ac6_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
554
555         if (!s)
556                 goto out;
557
558         rc = seq_open(file, &ac6_seq_ops);
559         if (rc)
560                 goto out_kfree;
561
562         seq = file->private_data;
563         seq->private = s;
564 out:
565         return rc;
566 out_kfree:
567         kfree(s);
568         goto out;
569 }
570
571 static const struct file_operations ac6_seq_fops = {
572         .owner          =       THIS_MODULE,
573         .open           =       ac6_seq_open,
574         .read           =       seq_read,
575         .llseek         =       seq_lseek,
576         .release        =       seq_release_private,
577 };
578
579 int __init ac6_proc_init(void)
580 {
581         if (!proc_net_fops_create("anycast6", S_IRUGO, &ac6_seq_fops))
582                 return -ENOMEM;
583
584         return 0;
585 }
586
587 void ac6_proc_exit(void)
588 {
589         proc_net_remove("anycast6");
590 }
591 #endif
592