xref: /openbmc/linux/net/ipv6/anycast.c (revision af958a38)
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 #include <linux/slab.h>
33 
34 #include <net/net_namespace.h>
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 
47 static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr);
48 
49 /* Big ac list lock for all the sockets */
50 static DEFINE_SPINLOCK(ipv6_sk_ac_lock);
51 
52 
53 /*
54  *	socket join an anycast group
55  */
56 
57 int ipv6_sock_ac_join(struct sock *sk, int ifindex, const struct in6_addr *addr)
58 {
59 	struct ipv6_pinfo *np = inet6_sk(sk);
60 	struct net_device *dev = NULL;
61 	struct inet6_dev *idev;
62 	struct ipv6_ac_socklist *pac;
63 	struct net *net = sock_net(sk);
64 	int	ishost = !net->ipv6.devconf_all->forwarding;
65 	int	err = 0;
66 
67 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
68 		return -EPERM;
69 	if (ipv6_addr_is_multicast(addr))
70 		return -EINVAL;
71 	if (ipv6_chk_addr(net, addr, NULL, 0))
72 		return -EINVAL;
73 
74 	pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL);
75 	if (pac == NULL)
76 		return -ENOMEM;
77 	pac->acl_next = NULL;
78 	pac->acl_addr = *addr;
79 
80 	rtnl_lock();
81 	rcu_read_lock();
82 	if (ifindex == 0) {
83 		struct rt6_info *rt;
84 
85 		rt = rt6_lookup(net, addr, NULL, 0, 0);
86 		if (rt) {
87 			dev = rt->dst.dev;
88 			ip6_rt_put(rt);
89 		} else if (ishost) {
90 			err = -EADDRNOTAVAIL;
91 			goto error;
92 		} else {
93 			/* router, no matching interface: just pick one */
94 			dev = dev_get_by_flags_rcu(net, IFF_UP,
95 						   IFF_UP | IFF_LOOPBACK);
96 		}
97 	} else
98 		dev = dev_get_by_index_rcu(net, ifindex);
99 
100 	if (dev == NULL) {
101 		err = -ENODEV;
102 		goto error;
103 	}
104 
105 	idev = __in6_dev_get(dev);
106 	if (!idev) {
107 		if (ifindex)
108 			err = -ENODEV;
109 		else
110 			err = -EADDRNOTAVAIL;
111 		goto error;
112 	}
113 	/* reset ishost, now that we have a specific device */
114 	ishost = !idev->cnf.forwarding;
115 
116 	pac->acl_ifindex = dev->ifindex;
117 
118 	/* XXX
119 	 * For hosts, allow link-local or matching prefix anycasts.
120 	 * This obviates the need for propagating anycast routes while
121 	 * still allowing some non-router anycast participation.
122 	 */
123 	if (!ipv6_chk_prefix(addr, dev)) {
124 		if (ishost)
125 			err = -EADDRNOTAVAIL;
126 		if (err)
127 			goto error;
128 	}
129 
130 	err = ipv6_dev_ac_inc(dev, addr);
131 	if (!err) {
132 		spin_lock_bh(&ipv6_sk_ac_lock);
133 		pac->acl_next = np->ipv6_ac_list;
134 		np->ipv6_ac_list = pac;
135 		spin_unlock_bh(&ipv6_sk_ac_lock);
136 		pac = NULL;
137 	}
138 
139 error:
140 	rcu_read_unlock();
141 	rtnl_unlock();
142 	if (pac)
143 		sock_kfree_s(sk, pac, sizeof(*pac));
144 	return err;
145 }
146 
147 /*
148  *	socket leave an anycast group
149  */
150 int ipv6_sock_ac_drop(struct sock *sk, int ifindex, const struct in6_addr *addr)
151 {
152 	struct ipv6_pinfo *np = inet6_sk(sk);
153 	struct net_device *dev;
154 	struct ipv6_ac_socklist *pac, *prev_pac;
155 	struct net *net = sock_net(sk);
156 
157 	spin_lock_bh(&ipv6_sk_ac_lock);
158 	prev_pac = NULL;
159 	for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) {
160 		if ((ifindex == 0 || pac->acl_ifindex == ifindex) &&
161 		     ipv6_addr_equal(&pac->acl_addr, addr))
162 			break;
163 		prev_pac = pac;
164 	}
165 	if (!pac) {
166 		spin_unlock_bh(&ipv6_sk_ac_lock);
167 		return -ENOENT;
168 	}
169 	if (prev_pac)
170 		prev_pac->acl_next = pac->acl_next;
171 	else
172 		np->ipv6_ac_list = pac->acl_next;
173 
174 	spin_unlock_bh(&ipv6_sk_ac_lock);
175 
176 	rtnl_lock();
177 	rcu_read_lock();
178 	dev = dev_get_by_index_rcu(net, pac->acl_ifindex);
179 	if (dev)
180 		ipv6_dev_ac_dec(dev, &pac->acl_addr);
181 	rcu_read_unlock();
182 	rtnl_unlock();
183 
184 	sock_kfree_s(sk, pac, sizeof(*pac));
185 	return 0;
186 }
187 
188 void ipv6_sock_ac_close(struct sock *sk)
189 {
190 	struct ipv6_pinfo *np = inet6_sk(sk);
191 	struct net_device *dev = NULL;
192 	struct ipv6_ac_socklist *pac;
193 	struct net *net = sock_net(sk);
194 	int	prev_index;
195 
196 	if (!np->ipv6_ac_list)
197 		return;
198 
199 	spin_lock_bh(&ipv6_sk_ac_lock);
200 	pac = np->ipv6_ac_list;
201 	np->ipv6_ac_list = NULL;
202 	spin_unlock_bh(&ipv6_sk_ac_lock);
203 
204 	prev_index = 0;
205 	rtnl_lock();
206 	rcu_read_lock();
207 	while (pac) {
208 		struct ipv6_ac_socklist *next = pac->acl_next;
209 
210 		if (pac->acl_ifindex != prev_index) {
211 			dev = dev_get_by_index_rcu(net, pac->acl_ifindex);
212 			prev_index = pac->acl_ifindex;
213 		}
214 		if (dev)
215 			ipv6_dev_ac_dec(dev, &pac->acl_addr);
216 		sock_kfree_s(sk, pac, sizeof(*pac));
217 		pac = next;
218 	}
219 	rcu_read_unlock();
220 	rtnl_unlock();
221 }
222 
223 static void aca_put(struct ifacaddr6 *ac)
224 {
225 	if (atomic_dec_and_test(&ac->aca_refcnt)) {
226 		in6_dev_put(ac->aca_idev);
227 		dst_release(&ac->aca_rt->dst);
228 		kfree(ac);
229 	}
230 }
231 
232 /*
233  *	device anycast group inc (add if not found)
234  */
235 int ipv6_dev_ac_inc(struct net_device *dev, const struct in6_addr *addr)
236 {
237 	struct ifacaddr6 *aca;
238 	struct inet6_dev *idev;
239 	struct rt6_info *rt;
240 	int err;
241 
242 	ASSERT_RTNL();
243 
244 	idev = in6_dev_get(dev);
245 
246 	if (idev == NULL)
247 		return -EINVAL;
248 
249 	write_lock_bh(&idev->lock);
250 	if (idev->dead) {
251 		err = -ENODEV;
252 		goto out;
253 	}
254 
255 	for (aca = idev->ac_list; aca; aca = aca->aca_next) {
256 		if (ipv6_addr_equal(&aca->aca_addr, addr)) {
257 			aca->aca_users++;
258 			err = 0;
259 			goto out;
260 		}
261 	}
262 
263 	/*
264 	 *	not found: create a new one.
265 	 */
266 
267 	aca = kzalloc(sizeof(struct ifacaddr6), GFP_ATOMIC);
268 
269 	if (aca == NULL) {
270 		err = -ENOMEM;
271 		goto out;
272 	}
273 
274 	rt = addrconf_dst_alloc(idev, addr, true);
275 	if (IS_ERR(rt)) {
276 		kfree(aca);
277 		err = PTR_ERR(rt);
278 		goto out;
279 	}
280 
281 	aca->aca_addr = *addr;
282 	aca->aca_idev = idev;
283 	aca->aca_rt = rt;
284 	aca->aca_users = 1;
285 	/* aca_tstamp should be updated upon changes */
286 	aca->aca_cstamp = aca->aca_tstamp = jiffies;
287 	atomic_set(&aca->aca_refcnt, 2);
288 	spin_lock_init(&aca->aca_lock);
289 
290 	aca->aca_next = idev->ac_list;
291 	idev->ac_list = aca;
292 	write_unlock_bh(&idev->lock);
293 
294 	ip6_ins_rt(rt);
295 
296 	addrconf_join_solict(dev, &aca->aca_addr);
297 
298 	aca_put(aca);
299 	return 0;
300 out:
301 	write_unlock_bh(&idev->lock);
302 	in6_dev_put(idev);
303 	return err;
304 }
305 
306 /*
307  *	device anycast group decrement
308  */
309 int __ipv6_dev_ac_dec(struct inet6_dev *idev, const struct in6_addr *addr)
310 {
311 	struct ifacaddr6 *aca, *prev_aca;
312 
313 	ASSERT_RTNL();
314 
315 	write_lock_bh(&idev->lock);
316 	prev_aca = NULL;
317 	for (aca = idev->ac_list; aca; aca = aca->aca_next) {
318 		if (ipv6_addr_equal(&aca->aca_addr, addr))
319 			break;
320 		prev_aca = aca;
321 	}
322 	if (!aca) {
323 		write_unlock_bh(&idev->lock);
324 		return -ENOENT;
325 	}
326 	if (--aca->aca_users > 0) {
327 		write_unlock_bh(&idev->lock);
328 		return 0;
329 	}
330 	if (prev_aca)
331 		prev_aca->aca_next = aca->aca_next;
332 	else
333 		idev->ac_list = aca->aca_next;
334 	write_unlock_bh(&idev->lock);
335 	addrconf_leave_solict(idev, &aca->aca_addr);
336 
337 	dst_hold(&aca->aca_rt->dst);
338 	ip6_del_rt(aca->aca_rt);
339 
340 	aca_put(aca);
341 	return 0;
342 }
343 
344 /* called with rcu_read_lock() */
345 static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr)
346 {
347 	struct inet6_dev *idev = __in6_dev_get(dev);
348 
349 	if (idev == NULL)
350 		return -ENODEV;
351 	return __ipv6_dev_ac_dec(idev, addr);
352 }
353 
354 /*
355  *	check if the interface has this anycast address
356  *	called with rcu_read_lock()
357  */
358 static bool ipv6_chk_acast_dev(struct net_device *dev, const struct in6_addr *addr)
359 {
360 	struct inet6_dev *idev;
361 	struct ifacaddr6 *aca;
362 
363 	idev = __in6_dev_get(dev);
364 	if (idev) {
365 		read_lock_bh(&idev->lock);
366 		for (aca = idev->ac_list; aca; aca = aca->aca_next)
367 			if (ipv6_addr_equal(&aca->aca_addr, addr))
368 				break;
369 		read_unlock_bh(&idev->lock);
370 		return aca != NULL;
371 	}
372 	return false;
373 }
374 
375 /*
376  *	check if given interface (or any, if dev==0) has this anycast address
377  */
378 bool ipv6_chk_acast_addr(struct net *net, struct net_device *dev,
379 			 const struct in6_addr *addr)
380 {
381 	bool found = false;
382 
383 	rcu_read_lock();
384 	if (dev)
385 		found = ipv6_chk_acast_dev(dev, addr);
386 	else
387 		for_each_netdev_rcu(net, dev)
388 			if (ipv6_chk_acast_dev(dev, addr)) {
389 				found = true;
390 				break;
391 			}
392 	rcu_read_unlock();
393 	return found;
394 }
395 
396 /*	check if this anycast address is link-local on given interface or
397  *	is global
398  */
399 bool ipv6_chk_acast_addr_src(struct net *net, struct net_device *dev,
400 			     const struct in6_addr *addr)
401 {
402 	return ipv6_chk_acast_addr(net,
403 				   (ipv6_addr_type(addr) & IPV6_ADDR_LINKLOCAL ?
404 				    dev : NULL),
405 				   addr);
406 }
407 
408 #ifdef CONFIG_PROC_FS
409 struct ac6_iter_state {
410 	struct seq_net_private p;
411 	struct net_device *dev;
412 	struct inet6_dev *idev;
413 };
414 
415 #define ac6_seq_private(seq)	((struct ac6_iter_state *)(seq)->private)
416 
417 static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq)
418 {
419 	struct ifacaddr6 *im = NULL;
420 	struct ac6_iter_state *state = ac6_seq_private(seq);
421 	struct net *net = seq_file_net(seq);
422 
423 	state->idev = NULL;
424 	for_each_netdev_rcu(net, state->dev) {
425 		struct inet6_dev *idev;
426 		idev = __in6_dev_get(state->dev);
427 		if (!idev)
428 			continue;
429 		read_lock_bh(&idev->lock);
430 		im = idev->ac_list;
431 		if (im) {
432 			state->idev = idev;
433 			break;
434 		}
435 		read_unlock_bh(&idev->lock);
436 	}
437 	return im;
438 }
439 
440 static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im)
441 {
442 	struct ac6_iter_state *state = ac6_seq_private(seq);
443 
444 	im = im->aca_next;
445 	while (!im) {
446 		if (likely(state->idev != NULL))
447 			read_unlock_bh(&state->idev->lock);
448 
449 		state->dev = next_net_device_rcu(state->dev);
450 		if (!state->dev) {
451 			state->idev = NULL;
452 			break;
453 		}
454 		state->idev = __in6_dev_get(state->dev);
455 		if (!state->idev)
456 			continue;
457 		read_lock_bh(&state->idev->lock);
458 		im = state->idev->ac_list;
459 	}
460 	return im;
461 }
462 
463 static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos)
464 {
465 	struct ifacaddr6 *im = ac6_get_first(seq);
466 	if (im)
467 		while (pos && (im = ac6_get_next(seq, im)) != NULL)
468 			--pos;
469 	return pos ? NULL : im;
470 }
471 
472 static void *ac6_seq_start(struct seq_file *seq, loff_t *pos)
473 	__acquires(RCU)
474 {
475 	rcu_read_lock();
476 	return ac6_get_idx(seq, *pos);
477 }
478 
479 static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
480 {
481 	struct ifacaddr6 *im = ac6_get_next(seq, v);
482 
483 	++*pos;
484 	return im;
485 }
486 
487 static void ac6_seq_stop(struct seq_file *seq, void *v)
488 	__releases(RCU)
489 {
490 	struct ac6_iter_state *state = ac6_seq_private(seq);
491 
492 	if (likely(state->idev != NULL)) {
493 		read_unlock_bh(&state->idev->lock);
494 		state->idev = NULL;
495 	}
496 	rcu_read_unlock();
497 }
498 
499 static int ac6_seq_show(struct seq_file *seq, void *v)
500 {
501 	struct ifacaddr6 *im = (struct ifacaddr6 *)v;
502 	struct ac6_iter_state *state = ac6_seq_private(seq);
503 
504 	seq_printf(seq, "%-4d %-15s %pi6 %5d\n",
505 		   state->dev->ifindex, state->dev->name,
506 		   &im->aca_addr, im->aca_users);
507 	return 0;
508 }
509 
510 static const struct seq_operations ac6_seq_ops = {
511 	.start	=	ac6_seq_start,
512 	.next	=	ac6_seq_next,
513 	.stop	=	ac6_seq_stop,
514 	.show	=	ac6_seq_show,
515 };
516 
517 static int ac6_seq_open(struct inode *inode, struct file *file)
518 {
519 	return seq_open_net(inode, file, &ac6_seq_ops,
520 			    sizeof(struct ac6_iter_state));
521 }
522 
523 static const struct file_operations ac6_seq_fops = {
524 	.owner		=	THIS_MODULE,
525 	.open		=	ac6_seq_open,
526 	.read		=	seq_read,
527 	.llseek		=	seq_lseek,
528 	.release	=	seq_release_net,
529 };
530 
531 int __net_init ac6_proc_init(struct net *net)
532 {
533 	if (!proc_create("anycast6", S_IRUGO, net->proc_net, &ac6_seq_fops))
534 		return -ENOMEM;
535 
536 	return 0;
537 }
538 
539 void ac6_proc_exit(struct net *net)
540 {
541 	remove_proc_entry("anycast6", net->proc_net);
542 }
543 #endif
544 
545