]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Anycast support for IPv6 | |
1ab1457c | 3 | * Linux INET6 implementation |
1da177e4 LT |
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 | ||
4fc268d2 | 16 | #include <linux/capability.h> |
1da177e4 LT |
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> | |
1da177e4 LT |
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> | |
5a0e3ad6 | 32 | #include <linux/slab.h> |
1da177e4 | 33 | |
457c4cbc | 34 | #include <net/net_namespace.h> |
1da177e4 LT |
35 | #include <net/sock.h> |
36 | #include <net/snmp.h> | |
37 | ||
38 | #include <net/ipv6.h> | |
39 | #include <net/protocol.h> | |
40 | #include <net/if_inet6.h> | |
41 | #include <net/ndisc.h> | |
42 | #include <net/addrconf.h> | |
43 | #include <net/ip6_route.h> | |
44 | ||
45 | #include <net/checksum.h> | |
46 | ||
b71d1d42 | 47 | static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr); |
1da177e4 | 48 | |
1da177e4 LT |
49 | /* |
50 | * socket join an anycast group | |
51 | */ | |
52 | ||
b71d1d42 | 53 | int ipv6_sock_ac_join(struct sock *sk, int ifindex, const struct in6_addr *addr) |
1da177e4 LT |
54 | { |
55 | struct ipv6_pinfo *np = inet6_sk(sk); | |
56 | struct net_device *dev = NULL; | |
57 | struct inet6_dev *idev; | |
58 | struct ipv6_ac_socklist *pac; | |
6ab57e7e | 59 | struct net *net = sock_net(sk); |
53b7997f | 60 | int ishost = !net->ipv6.devconf_all->forwarding; |
1da177e4 LT |
61 | int err = 0; |
62 | ||
af31f412 | 63 | if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) |
1da177e4 LT |
64 | return -EPERM; |
65 | if (ipv6_addr_is_multicast(addr)) | |
66 | return -EINVAL; | |
6ab57e7e | 67 | if (ipv6_chk_addr(net, addr, NULL, 0)) |
1da177e4 LT |
68 | return -EINVAL; |
69 | ||
70 | pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL); | |
71 | if (pac == NULL) | |
72 | return -ENOMEM; | |
73 | pac->acl_next = NULL; | |
4e3fd7a0 | 74 | pac->acl_addr = *addr; |
1da177e4 | 75 | |
a9ed4a29 | 76 | rtnl_lock(); |
1da177e4 LT |
77 | if (ifindex == 0) { |
78 | struct rt6_info *rt; | |
79 | ||
6ab57e7e | 80 | rt = rt6_lookup(net, addr, NULL, 0, 0); |
1da177e4 | 81 | if (rt) { |
d1918542 | 82 | dev = rt->dst.dev; |
94e187c0 | 83 | ip6_rt_put(rt); |
1da177e4 LT |
84 | } else if (ishost) { |
85 | err = -EADDRNOTAVAIL; | |
bb69ae04 | 86 | goto error; |
1da177e4 LT |
87 | } else { |
88 | /* router, no matching interface: just pick one */ | |
6c555490 WC |
89 | dev = __dev_get_by_flags(net, IFF_UP, |
90 | IFF_UP | IFF_LOOPBACK); | |
1da177e4 LT |
91 | } |
92 | } else | |
6c555490 | 93 | dev = __dev_get_by_index(net, ifindex); |
1da177e4 LT |
94 | |
95 | if (dev == NULL) { | |
96 | err = -ENODEV; | |
bb69ae04 | 97 | goto error; |
1da177e4 LT |
98 | } |
99 | ||
bb69ae04 | 100 | idev = __in6_dev_get(dev); |
1da177e4 LT |
101 | if (!idev) { |
102 | if (ifindex) | |
103 | err = -ENODEV; | |
104 | else | |
105 | err = -EADDRNOTAVAIL; | |
bb69ae04 | 106 | goto error; |
1da177e4 LT |
107 | } |
108 | /* reset ishost, now that we have a specific device */ | |
109 | ishost = !idev->cnf.forwarding; | |
1da177e4 LT |
110 | |
111 | pac->acl_ifindex = dev->ifindex; | |
112 | ||
113 | /* XXX | |
114 | * For hosts, allow link-local or matching prefix anycasts. | |
115 | * This obviates the need for propagating anycast routes while | |
116 | * still allowing some non-router anycast participation. | |
117 | */ | |
52eeeb84 | 118 | if (!ipv6_chk_prefix(addr, dev)) { |
1da177e4 LT |
119 | if (ishost) |
120 | err = -EADDRNOTAVAIL; | |
121 | if (err) | |
bb69ae04 | 122 | goto error; |
1da177e4 LT |
123 | } |
124 | ||
125 | err = ipv6_dev_ac_inc(dev, addr); | |
bb69ae04 | 126 | if (!err) { |
bb69ae04 ED |
127 | pac->acl_next = np->ipv6_ac_list; |
128 | np->ipv6_ac_list = pac; | |
bb69ae04 ED |
129 | pac = NULL; |
130 | } | |
1da177e4 | 131 | |
bb69ae04 | 132 | error: |
a9ed4a29 | 133 | rtnl_unlock(); |
bb69ae04 ED |
134 | if (pac) |
135 | sock_kfree_s(sk, pac, sizeof(*pac)); | |
1da177e4 LT |
136 | return err; |
137 | } | |
138 | ||
139 | /* | |
140 | * socket leave an anycast group | |
141 | */ | |
b71d1d42 | 142 | int ipv6_sock_ac_drop(struct sock *sk, int ifindex, const struct in6_addr *addr) |
1da177e4 LT |
143 | { |
144 | struct ipv6_pinfo *np = inet6_sk(sk); | |
145 | struct net_device *dev; | |
146 | struct ipv6_ac_socklist *pac, *prev_pac; | |
6ab57e7e | 147 | struct net *net = sock_net(sk); |
1da177e4 | 148 | |
b03a9c04 | 149 | rtnl_lock(); |
1da177e4 LT |
150 | prev_pac = NULL; |
151 | for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) { | |
152 | if ((ifindex == 0 || pac->acl_ifindex == ifindex) && | |
153 | ipv6_addr_equal(&pac->acl_addr, addr)) | |
154 | break; | |
155 | prev_pac = pac; | |
156 | } | |
157 | if (!pac) { | |
b03a9c04 | 158 | rtnl_unlock(); |
1da177e4 LT |
159 | return -ENOENT; |
160 | } | |
161 | if (prev_pac) | |
162 | prev_pac->acl_next = pac->acl_next; | |
163 | else | |
164 | np->ipv6_ac_list = pac->acl_next; | |
165 | ||
6c555490 | 166 | dev = __dev_get_by_index(net, pac->acl_ifindex); |
bb69ae04 | 167 | if (dev) |
1da177e4 | 168 | ipv6_dev_ac_dec(dev, &pac->acl_addr); |
a9ed4a29 | 169 | rtnl_unlock(); |
bb69ae04 | 170 | |
1da177e4 LT |
171 | sock_kfree_s(sk, pac, sizeof(*pac)); |
172 | return 0; | |
173 | } | |
174 | ||
175 | void ipv6_sock_ac_close(struct sock *sk) | |
176 | { | |
177 | struct ipv6_pinfo *np = inet6_sk(sk); | |
178 | struct net_device *dev = NULL; | |
179 | struct ipv6_ac_socklist *pac; | |
6ab57e7e | 180 | struct net *net = sock_net(sk); |
1da177e4 LT |
181 | int prev_index; |
182 | ||
0e1efe9d ED |
183 | if (!np->ipv6_ac_list) |
184 | return; | |
185 | ||
b03a9c04 | 186 | rtnl_lock(); |
1da177e4 LT |
187 | pac = np->ipv6_ac_list; |
188 | np->ipv6_ac_list = NULL; | |
1da177e4 LT |
189 | |
190 | prev_index = 0; | |
191 | while (pac) { | |
192 | struct ipv6_ac_socklist *next = pac->acl_next; | |
193 | ||
194 | if (pac->acl_ifindex != prev_index) { | |
6c555490 | 195 | dev = __dev_get_by_index(net, pac->acl_ifindex); |
1da177e4 LT |
196 | prev_index = pac->acl_ifindex; |
197 | } | |
198 | if (dev) | |
199 | ipv6_dev_ac_dec(dev, &pac->acl_addr); | |
200 | sock_kfree_s(sk, pac, sizeof(*pac)); | |
201 | pac = next; | |
202 | } | |
a9ed4a29 | 203 | rtnl_unlock(); |
1da177e4 LT |
204 | } |
205 | ||
1da177e4 LT |
206 | static void aca_put(struct ifacaddr6 *ac) |
207 | { | |
208 | if (atomic_dec_and_test(&ac->aca_refcnt)) { | |
209 | in6_dev_put(ac->aca_idev); | |
d8d1f30b | 210 | dst_release(&ac->aca_rt->dst); |
1da177e4 LT |
211 | kfree(ac); |
212 | } | |
213 | } | |
214 | ||
215 | /* | |
216 | * device anycast group inc (add if not found) | |
217 | */ | |
b71d1d42 | 218 | int ipv6_dev_ac_inc(struct net_device *dev, const struct in6_addr *addr) |
1da177e4 LT |
219 | { |
220 | struct ifacaddr6 *aca; | |
221 | struct inet6_dev *idev; | |
222 | struct rt6_info *rt; | |
223 | int err; | |
224 | ||
a9ed4a29 SD |
225 | ASSERT_RTNL(); |
226 | ||
1da177e4 LT |
227 | idev = in6_dev_get(dev); |
228 | ||
229 | if (idev == NULL) | |
230 | return -EINVAL; | |
231 | ||
232 | write_lock_bh(&idev->lock); | |
233 | if (idev->dead) { | |
234 | err = -ENODEV; | |
235 | goto out; | |
236 | } | |
237 | ||
238 | for (aca = idev->ac_list; aca; aca = aca->aca_next) { | |
239 | if (ipv6_addr_equal(&aca->aca_addr, addr)) { | |
240 | aca->aca_users++; | |
241 | err = 0; | |
242 | goto out; | |
243 | } | |
244 | } | |
245 | ||
246 | /* | |
247 | * not found: create a new one. | |
248 | */ | |
249 | ||
0c600eda | 250 | aca = kzalloc(sizeof(struct ifacaddr6), GFP_ATOMIC); |
1da177e4 LT |
251 | |
252 | if (aca == NULL) { | |
253 | err = -ENOMEM; | |
254 | goto out; | |
255 | } | |
256 | ||
8f031519 | 257 | rt = addrconf_dst_alloc(idev, addr, true); |
1da177e4 LT |
258 | if (IS_ERR(rt)) { |
259 | kfree(aca); | |
260 | err = PTR_ERR(rt); | |
261 | goto out; | |
262 | } | |
263 | ||
4e3fd7a0 | 264 | aca->aca_addr = *addr; |
1da177e4 LT |
265 | aca->aca_idev = idev; |
266 | aca->aca_rt = rt; | |
267 | aca->aca_users = 1; | |
268 | /* aca_tstamp should be updated upon changes */ | |
269 | aca->aca_cstamp = aca->aca_tstamp = jiffies; | |
270 | atomic_set(&aca->aca_refcnt, 2); | |
271 | spin_lock_init(&aca->aca_lock); | |
272 | ||
273 | aca->aca_next = idev->ac_list; | |
274 | idev->ac_list = aca; | |
275 | write_unlock_bh(&idev->lock); | |
276 | ||
84f59370 | 277 | ip6_ins_rt(rt); |
1da177e4 LT |
278 | |
279 | addrconf_join_solict(dev, &aca->aca_addr); | |
280 | ||
281 | aca_put(aca); | |
282 | return 0; | |
283 | out: | |
284 | write_unlock_bh(&idev->lock); | |
285 | in6_dev_put(idev); | |
286 | return err; | |
287 | } | |
288 | ||
289 | /* | |
290 | * device anycast group decrement | |
291 | */ | |
b71d1d42 | 292 | int __ipv6_dev_ac_dec(struct inet6_dev *idev, const struct in6_addr *addr) |
1da177e4 LT |
293 | { |
294 | struct ifacaddr6 *aca, *prev_aca; | |
295 | ||
a9ed4a29 SD |
296 | ASSERT_RTNL(); |
297 | ||
1da177e4 LT |
298 | write_lock_bh(&idev->lock); |
299 | prev_aca = NULL; | |
300 | for (aca = idev->ac_list; aca; aca = aca->aca_next) { | |
301 | if (ipv6_addr_equal(&aca->aca_addr, addr)) | |
302 | break; | |
303 | prev_aca = aca; | |
304 | } | |
305 | if (!aca) { | |
306 | write_unlock_bh(&idev->lock); | |
307 | return -ENOENT; | |
308 | } | |
309 | if (--aca->aca_users > 0) { | |
310 | write_unlock_bh(&idev->lock); | |
311 | return 0; | |
312 | } | |
313 | if (prev_aca) | |
314 | prev_aca->aca_next = aca->aca_next; | |
315 | else | |
316 | idev->ac_list = aca->aca_next; | |
317 | write_unlock_bh(&idev->lock); | |
318 | addrconf_leave_solict(idev, &aca->aca_addr); | |
319 | ||
d8d1f30b | 320 | dst_hold(&aca->aca_rt->dst); |
84f59370 | 321 | ip6_del_rt(aca->aca_rt); |
1da177e4 LT |
322 | |
323 | aca_put(aca); | |
324 | return 0; | |
325 | } | |
326 | ||
6c555490 | 327 | /* called with rtnl_lock() */ |
b71d1d42 | 328 | static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr) |
1da177e4 | 329 | { |
bb69ae04 ED |
330 | struct inet6_dev *idev = __in6_dev_get(dev); |
331 | ||
1da177e4 LT |
332 | if (idev == NULL) |
333 | return -ENODEV; | |
bb69ae04 | 334 | return __ipv6_dev_ac_dec(idev, addr); |
1da177e4 | 335 | } |
1ab1457c | 336 | |
1da177e4 LT |
337 | /* |
338 | * check if the interface has this anycast address | |
bb69ae04 | 339 | * called with rcu_read_lock() |
1da177e4 | 340 | */ |
a50feda5 | 341 | static bool ipv6_chk_acast_dev(struct net_device *dev, const struct in6_addr *addr) |
1da177e4 LT |
342 | { |
343 | struct inet6_dev *idev; | |
344 | struct ifacaddr6 *aca; | |
345 | ||
bb69ae04 | 346 | idev = __in6_dev_get(dev); |
1da177e4 LT |
347 | if (idev) { |
348 | read_lock_bh(&idev->lock); | |
349 | for (aca = idev->ac_list; aca; aca = aca->aca_next) | |
350 | if (ipv6_addr_equal(&aca->aca_addr, addr)) | |
351 | break; | |
352 | read_unlock_bh(&idev->lock); | |
cfcabdcc | 353 | return aca != NULL; |
1da177e4 | 354 | } |
a50feda5 | 355 | return false; |
1da177e4 LT |
356 | } |
357 | ||
358 | /* | |
359 | * check if given interface (or any, if dev==0) has this anycast address | |
360 | */ | |
a50feda5 ED |
361 | bool ipv6_chk_acast_addr(struct net *net, struct net_device *dev, |
362 | const struct in6_addr *addr) | |
1da177e4 | 363 | { |
a50feda5 | 364 | bool found = false; |
7562f876 | 365 | |
c6d14c84 | 366 | rcu_read_lock(); |
bb69ae04 ED |
367 | if (dev) |
368 | found = ipv6_chk_acast_dev(dev, addr); | |
369 | else | |
370 | for_each_netdev_rcu(net, dev) | |
371 | if (ipv6_chk_acast_dev(dev, addr)) { | |
a50feda5 | 372 | found = true; |
bb69ae04 ED |
373 | break; |
374 | } | |
c6d14c84 | 375 | rcu_read_unlock(); |
7562f876 | 376 | return found; |
1da177e4 LT |
377 | } |
378 | ||
7c90cc2d FLB |
379 | /* check if this anycast address is link-local on given interface or |
380 | * is global | |
381 | */ | |
382 | bool ipv6_chk_acast_addr_src(struct net *net, struct net_device *dev, | |
383 | const struct in6_addr *addr) | |
384 | { | |
385 | return ipv6_chk_acast_addr(net, | |
386 | (ipv6_addr_type(addr) & IPV6_ADDR_LINKLOCAL ? | |
387 | dev : NULL), | |
388 | addr); | |
389 | } | |
1da177e4 LT |
390 | |
391 | #ifdef CONFIG_PROC_FS | |
392 | struct ac6_iter_state { | |
6ab57e7e | 393 | struct seq_net_private p; |
1da177e4 LT |
394 | struct net_device *dev; |
395 | struct inet6_dev *idev; | |
396 | }; | |
397 | ||
398 | #define ac6_seq_private(seq) ((struct ac6_iter_state *)(seq)->private) | |
399 | ||
400 | static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq) | |
401 | { | |
402 | struct ifacaddr6 *im = NULL; | |
403 | struct ac6_iter_state *state = ac6_seq_private(seq); | |
6ab57e7e | 404 | struct net *net = seq_file_net(seq); |
1da177e4 | 405 | |
7562f876 | 406 | state->idev = NULL; |
ce81b76a | 407 | for_each_netdev_rcu(net, state->dev) { |
1da177e4 | 408 | struct inet6_dev *idev; |
ce81b76a | 409 | idev = __in6_dev_get(state->dev); |
1da177e4 LT |
410 | if (!idev) |
411 | continue; | |
412 | read_lock_bh(&idev->lock); | |
413 | im = idev->ac_list; | |
414 | if (im) { | |
415 | state->idev = idev; | |
416 | break; | |
417 | } | |
418 | read_unlock_bh(&idev->lock); | |
419 | } | |
420 | return im; | |
421 | } | |
422 | ||
423 | static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im) | |
424 | { | |
425 | struct ac6_iter_state *state = ac6_seq_private(seq); | |
426 | ||
427 | im = im->aca_next; | |
428 | while (!im) { | |
ce81b76a | 429 | if (likely(state->idev != NULL)) |
1da177e4 | 430 | read_unlock_bh(&state->idev->lock); |
ce81b76a ED |
431 | |
432 | state->dev = next_net_device_rcu(state->dev); | |
1da177e4 LT |
433 | if (!state->dev) { |
434 | state->idev = NULL; | |
435 | break; | |
436 | } | |
ce81b76a | 437 | state->idev = __in6_dev_get(state->dev); |
1da177e4 LT |
438 | if (!state->idev) |
439 | continue; | |
440 | read_lock_bh(&state->idev->lock); | |
441 | im = state->idev->ac_list; | |
442 | } | |
443 | return im; | |
444 | } | |
445 | ||
446 | static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos) | |
447 | { | |
448 | struct ifacaddr6 *im = ac6_get_first(seq); | |
449 | if (im) | |
450 | while (pos && (im = ac6_get_next(seq, im)) != NULL) | |
451 | --pos; | |
452 | return pos ? NULL : im; | |
453 | } | |
454 | ||
455 | static void *ac6_seq_start(struct seq_file *seq, loff_t *pos) | |
ce81b76a | 456 | __acquires(RCU) |
1da177e4 | 457 | { |
ce81b76a | 458 | rcu_read_lock(); |
1da177e4 LT |
459 | return ac6_get_idx(seq, *pos); |
460 | } | |
461 | ||
462 | static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
463 | { | |
ce81b76a ED |
464 | struct ifacaddr6 *im = ac6_get_next(seq, v); |
465 | ||
1da177e4 LT |
466 | ++*pos; |
467 | return im; | |
468 | } | |
469 | ||
470 | static void ac6_seq_stop(struct seq_file *seq, void *v) | |
ce81b76a | 471 | __releases(RCU) |
1da177e4 LT |
472 | { |
473 | struct ac6_iter_state *state = ac6_seq_private(seq); | |
ce81b76a | 474 | |
1da177e4 LT |
475 | if (likely(state->idev != NULL)) { |
476 | read_unlock_bh(&state->idev->lock); | |
ce81b76a | 477 | state->idev = NULL; |
1da177e4 | 478 | } |
ce81b76a | 479 | rcu_read_unlock(); |
1da177e4 LT |
480 | } |
481 | ||
482 | static int ac6_seq_show(struct seq_file *seq, void *v) | |
483 | { | |
484 | struct ifacaddr6 *im = (struct ifacaddr6 *)v; | |
485 | struct ac6_iter_state *state = ac6_seq_private(seq); | |
486 | ||
4b7a4274 | 487 | seq_printf(seq, "%-4d %-15s %pi6 %5d\n", |
1da177e4 | 488 | state->dev->ifindex, state->dev->name, |
b071195d | 489 | &im->aca_addr, im->aca_users); |
1da177e4 LT |
490 | return 0; |
491 | } | |
492 | ||
56b3d975 | 493 | static const struct seq_operations ac6_seq_ops = { |
1da177e4 LT |
494 | .start = ac6_seq_start, |
495 | .next = ac6_seq_next, | |
496 | .stop = ac6_seq_stop, | |
497 | .show = ac6_seq_show, | |
498 | }; | |
499 | ||
500 | static int ac6_seq_open(struct inode *inode, struct file *file) | |
501 | { | |
6ab57e7e DL |
502 | return seq_open_net(inode, file, &ac6_seq_ops, |
503 | sizeof(struct ac6_iter_state)); | |
1da177e4 LT |
504 | } |
505 | ||
9a32144e | 506 | static const struct file_operations ac6_seq_fops = { |
1da177e4 LT |
507 | .owner = THIS_MODULE, |
508 | .open = ac6_seq_open, | |
509 | .read = seq_read, | |
510 | .llseek = seq_lseek, | |
6ab57e7e | 511 | .release = seq_release_net, |
1da177e4 LT |
512 | }; |
513 | ||
2c8c1e72 | 514 | int __net_init ac6_proc_init(struct net *net) |
1da177e4 | 515 | { |
d4beaa66 | 516 | if (!proc_create("anycast6", S_IRUGO, net->proc_net, &ac6_seq_fops)) |
1da177e4 LT |
517 | return -ENOMEM; |
518 | ||
519 | return 0; | |
520 | } | |
521 | ||
6ab57e7e | 522 | void ac6_proc_exit(struct net *net) |
1da177e4 | 523 | { |
ece31ffd | 524 | remove_proc_entry("anycast6", net->proc_net); |
1da177e4 LT |
525 | } |
526 | #endif | |
527 |