2 * IPv6 Address Label subsystem
3 * for the IPv6 "Default" Source Address Selection
5 * Copyright (C)2007 USAGI/WIDE Project
9 * YOSHIFUJI Hideaki @ USAGI/WIDE Project <yoshfuji@linux-ipv6.org>
12 #include <linux/kernel.h>
13 #include <linux/list.h>
14 #include <linux/rcupdate.h>
15 #include <linux/in6.h>
16 #include <linux/slab.h>
17 #include <net/addrconf.h>
18 #include <linux/if_addrlabel.h>
19 #include <linux/netlink.h>
20 #include <linux/rtnetlink.h>
23 #define ADDRLABEL(x...) printk(x)
25 #define ADDRLABEL(x...) do { ; } while(0)
31 struct ip6addrlbl_entry
36 struct in6_addr prefix;
41 struct hlist_node list;
46 static struct ip6addrlbl_table
48 struct hlist_head head;
54 struct net *ip6addrlbl_net(const struct ip6addrlbl_entry *lbl)
56 return read_pnet(&lbl->lbl_net);
60 * Default policy table (RFC6724 + extensions)
62 * prefix addr_type label
63 * -------------------------------------------------------------------------
68 * ::ffff:0:0/96 V4MAPPED 4
69 * fc00::/7 N/A 5 ULA (RFC 4193)
70 * 2001::/32 N/A 6 Teredo (RFC 4380)
71 * 2001:10::/28 N/A 7 ORCHID (RFC 4843)
72 * fec0::/10 N/A 11 Site-local
73 * (deprecated by RFC3879)
74 * 3ffe::/16 N/A 12 6bone
76 * Note: 0xffffffff is used if we do not have any policies.
77 * Note: Labels for ULA and 6to4 are different from labels listed in RFC6724.
80 #define IPV6_ADDR_LABEL_DEFAULT 0xffffffffUL
82 static const __net_initconst struct ip6addrlbl_init_table
84 const struct in6_addr *prefix;
87 } ip6addrlbl_init_table[] = {
89 .prefix = &in6addr_any,
92 .prefix = &(struct in6_addr){{{ 0xfc }}},
96 .prefix = &(struct in6_addr){{{ 0xfe, 0xc0 }}},
100 .prefix = &(struct in6_addr){{{ 0x20, 0x02 }}},
104 .prefix = &(struct in6_addr){{{ 0x3f, 0xfe }}},
108 .prefix = &(struct in6_addr){{{ 0x20, 0x01 }}},
111 },{ /* 2001:10::/28 */
112 .prefix = &(struct in6_addr){{{ 0x20, 0x01, 0x00, 0x10 }}},
116 .prefix = &(struct in6_addr){{{ [10] = 0xff, [11] = 0xff }}},
120 .prefix = &in6addr_any,
124 .prefix = &in6addr_loopback,
130 /* Object management */
131 static inline void ip6addrlbl_free(struct ip6addrlbl_entry *p)
134 release_net(p->lbl_net);
139 static void ip6addrlbl_free_rcu(struct rcu_head *h)
141 ip6addrlbl_free(container_of(h, struct ip6addrlbl_entry, rcu));
144 static bool ip6addrlbl_hold(struct ip6addrlbl_entry *p)
146 return atomic_inc_not_zero(&p->refcnt);
149 static inline void ip6addrlbl_put(struct ip6addrlbl_entry *p)
151 if (atomic_dec_and_test(&p->refcnt))
152 call_rcu(&p->rcu, ip6addrlbl_free_rcu);
156 static bool __ip6addrlbl_match(struct net *net,
157 const struct ip6addrlbl_entry *p,
158 const struct in6_addr *addr,
159 int addrtype, int ifindex)
161 if (!net_eq(ip6addrlbl_net(p), net))
163 if (p->ifindex && p->ifindex != ifindex)
165 if (p->addrtype && p->addrtype != addrtype)
167 if (!ipv6_prefix_equal(addr, &p->prefix, p->prefixlen))
172 static struct ip6addrlbl_entry *__ipv6_addr_label(struct net *net,
173 const struct in6_addr *addr,
174 int type, int ifindex)
176 struct ip6addrlbl_entry *p;
177 hlist_for_each_entry_rcu(p, &ip6addrlbl_table.head, list) {
178 if (__ip6addrlbl_match(net, p, addr, type, ifindex))
184 u32 ipv6_addr_label(struct net *net,
185 const struct in6_addr *addr, int type, int ifindex)
188 struct ip6addrlbl_entry *p;
190 type &= IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK;
193 p = __ipv6_addr_label(net, addr, type, ifindex);
194 label = p ? p->label : IPV6_ADDR_LABEL_DEFAULT;
197 ADDRLABEL(KERN_DEBUG "%s(addr=%pI6, type=%d, ifindex=%d) => %08x\n",
198 __func__, addr, type, ifindex, label);
203 /* allocate one entry */
204 static struct ip6addrlbl_entry *ip6addrlbl_alloc(struct net *net,
205 const struct in6_addr *prefix,
206 int prefixlen, int ifindex,
209 struct ip6addrlbl_entry *newp;
212 ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u)\n",
213 __func__, prefix, prefixlen, ifindex, (unsigned int)label);
215 addrtype = ipv6_addr_type(prefix) & (IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK);
218 case IPV6_ADDR_MAPPED:
220 return ERR_PTR(-EINVAL);
224 case IPV6_ADDR_COMPATv4:
228 case IPV6_ADDR_LOOPBACK:
229 if (prefixlen != 128)
234 newp = kmalloc(sizeof(*newp), GFP_KERNEL);
236 return ERR_PTR(-ENOMEM);
238 ipv6_addr_prefix(&newp->prefix, prefix, prefixlen);
239 newp->prefixlen = prefixlen;
240 newp->ifindex = ifindex;
241 newp->addrtype = addrtype;
243 INIT_HLIST_NODE(&newp->list);
245 newp->lbl_net = hold_net(net);
247 atomic_set(&newp->refcnt, 1);
252 static int __ip6addrlbl_add(struct ip6addrlbl_entry *newp, int replace)
256 ADDRLABEL(KERN_DEBUG "%s(newp=%p, replace=%d)\n",
260 if (hlist_empty(&ip6addrlbl_table.head)) {
261 hlist_add_head_rcu(&newp->list, &ip6addrlbl_table.head);
263 struct hlist_node *n;
264 struct ip6addrlbl_entry *p = NULL;
265 hlist_for_each_entry_safe(p, n,
266 &ip6addrlbl_table.head, list) {
267 if (p->prefixlen == newp->prefixlen &&
268 net_eq(ip6addrlbl_net(p), ip6addrlbl_net(newp)) &&
269 p->ifindex == newp->ifindex &&
270 ipv6_addr_equal(&p->prefix, &newp->prefix)) {
275 hlist_replace_rcu(&p->list, &newp->list);
278 } else if ((p->prefixlen == newp->prefixlen && !p->ifindex) ||
279 (p->prefixlen < newp->prefixlen)) {
280 hlist_add_before_rcu(&newp->list, &p->list);
284 hlist_add_after_rcu(&p->list, &newp->list);
288 ip6addrlbl_table.seq++;
293 static int ip6addrlbl_add(struct net *net,
294 const struct in6_addr *prefix, int prefixlen,
295 int ifindex, u32 label, int replace)
297 struct ip6addrlbl_entry *newp;
300 ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u, replace=%d)\n",
301 __func__, prefix, prefixlen, ifindex, (unsigned int)label,
304 newp = ip6addrlbl_alloc(net, prefix, prefixlen, ifindex, label);
306 return PTR_ERR(newp);
307 spin_lock(&ip6addrlbl_table.lock);
308 ret = __ip6addrlbl_add(newp, replace);
309 spin_unlock(&ip6addrlbl_table.lock);
311 ip6addrlbl_free(newp);
316 static int __ip6addrlbl_del(struct net *net,
317 const struct in6_addr *prefix, int prefixlen,
320 struct ip6addrlbl_entry *p = NULL;
321 struct hlist_node *n;
324 ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
325 __func__, prefix, prefixlen, ifindex);
327 hlist_for_each_entry_safe(p, n, &ip6addrlbl_table.head, list) {
328 if (p->prefixlen == prefixlen &&
329 net_eq(ip6addrlbl_net(p), net) &&
330 p->ifindex == ifindex &&
331 ipv6_addr_equal(&p->prefix, prefix)) {
332 hlist_del_rcu(&p->list);
341 static int ip6addrlbl_del(struct net *net,
342 const struct in6_addr *prefix, int prefixlen,
345 struct in6_addr prefix_buf;
348 ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
349 __func__, prefix, prefixlen, ifindex);
351 ipv6_addr_prefix(&prefix_buf, prefix, prefixlen);
352 spin_lock(&ip6addrlbl_table.lock);
353 ret = __ip6addrlbl_del(net, &prefix_buf, prefixlen, ifindex);
354 spin_unlock(&ip6addrlbl_table.lock);
358 /* add default label */
359 static int __net_init ip6addrlbl_net_init(struct net *net)
364 ADDRLABEL(KERN_DEBUG "%s\n", __func__);
366 for (i = 0; i < ARRAY_SIZE(ip6addrlbl_init_table); i++) {
367 int ret = ip6addrlbl_add(net,
368 ip6addrlbl_init_table[i].prefix,
369 ip6addrlbl_init_table[i].prefixlen,
371 ip6addrlbl_init_table[i].label, 0);
372 /* XXX: should we free all rules when we catch an error? */
373 if (ret && (!err || err != -ENOMEM))
379 static void __net_exit ip6addrlbl_net_exit(struct net *net)
381 struct ip6addrlbl_entry *p = NULL;
382 struct hlist_node *n;
384 /* Remove all labels belonging to the exiting net */
385 spin_lock(&ip6addrlbl_table.lock);
386 hlist_for_each_entry_safe(p, n, &ip6addrlbl_table.head, list) {
387 if (net_eq(ip6addrlbl_net(p), net)) {
388 hlist_del_rcu(&p->list);
392 spin_unlock(&ip6addrlbl_table.lock);
395 static struct pernet_operations ipv6_addr_label_ops = {
396 .init = ip6addrlbl_net_init,
397 .exit = ip6addrlbl_net_exit,
400 int __init ipv6_addr_label_init(void)
402 spin_lock_init(&ip6addrlbl_table.lock);
404 return register_pernet_subsys(&ipv6_addr_label_ops);
407 void ipv6_addr_label_cleanup(void)
409 unregister_pernet_subsys(&ipv6_addr_label_ops);
412 static const struct nla_policy ifal_policy[IFAL_MAX+1] = {
413 [IFAL_ADDRESS] = { .len = sizeof(struct in6_addr), },
414 [IFAL_LABEL] = { .len = sizeof(u32), },
417 static int ip6addrlbl_newdel(struct sk_buff *skb, struct nlmsghdr *nlh)
419 struct net *net = sock_net(skb->sk);
420 struct ifaddrlblmsg *ifal;
421 struct nlattr *tb[IFAL_MAX+1];
422 struct in6_addr *pfx;
426 err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
430 ifal = nlmsg_data(nlh);
432 if (ifal->ifal_family != AF_INET6 ||
433 ifal->ifal_prefixlen > 128)
436 if (!tb[IFAL_ADDRESS])
438 pfx = nla_data(tb[IFAL_ADDRESS]);
442 label = nla_get_u32(tb[IFAL_LABEL]);
443 if (label == IPV6_ADDR_LABEL_DEFAULT)
446 switch(nlh->nlmsg_type) {
447 case RTM_NEWADDRLABEL:
448 if (ifal->ifal_index &&
449 !__dev_get_by_index(net, ifal->ifal_index))
452 err = ip6addrlbl_add(net, pfx, ifal->ifal_prefixlen,
453 ifal->ifal_index, label,
454 nlh->nlmsg_flags & NLM_F_REPLACE);
456 case RTM_DELADDRLABEL:
457 err = ip6addrlbl_del(net, pfx, ifal->ifal_prefixlen,
466 static void ip6addrlbl_putmsg(struct nlmsghdr *nlh,
467 int prefixlen, int ifindex, u32 lseq)
469 struct ifaddrlblmsg *ifal = nlmsg_data(nlh);
470 ifal->ifal_family = AF_INET6;
471 ifal->ifal_prefixlen = prefixlen;
472 ifal->ifal_flags = 0;
473 ifal->ifal_index = ifindex;
474 ifal->ifal_seq = lseq;
477 static int ip6addrlbl_fill(struct sk_buff *skb,
478 struct ip6addrlbl_entry *p,
480 u32 portid, u32 seq, int event,
483 struct nlmsghdr *nlh = nlmsg_put(skb, portid, seq, event,
484 sizeof(struct ifaddrlblmsg), flags);
488 ip6addrlbl_putmsg(nlh, p->prefixlen, p->ifindex, lseq);
490 if (nla_put(skb, IFAL_ADDRESS, 16, &p->prefix) < 0 ||
491 nla_put_u32(skb, IFAL_LABEL, p->label) < 0) {
492 nlmsg_cancel(skb, nlh);
496 return nlmsg_end(skb, nlh);
499 static int ip6addrlbl_dump(struct sk_buff *skb, struct netlink_callback *cb)
501 struct net *net = sock_net(skb->sk);
502 struct ip6addrlbl_entry *p;
503 int idx = 0, s_idx = cb->args[0];
507 hlist_for_each_entry_rcu(p, &ip6addrlbl_table.head, list) {
509 net_eq(ip6addrlbl_net(p), net)) {
510 if ((err = ip6addrlbl_fill(skb, p,
511 ip6addrlbl_table.seq,
512 NETLINK_CB(cb->skb).portid,
525 static inline int ip6addrlbl_msgsize(void)
527 return NLMSG_ALIGN(sizeof(struct ifaddrlblmsg))
528 + nla_total_size(16) /* IFAL_ADDRESS */
529 + nla_total_size(4); /* IFAL_LABEL */
532 static int ip6addrlbl_get(struct sk_buff *in_skb, struct nlmsghdr* nlh)
534 struct net *net = sock_net(in_skb->sk);
535 struct ifaddrlblmsg *ifal;
536 struct nlattr *tb[IFAL_MAX+1];
537 struct in6_addr *addr;
540 struct ip6addrlbl_entry *p;
543 err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
547 ifal = nlmsg_data(nlh);
549 if (ifal->ifal_family != AF_INET6 ||
550 ifal->ifal_prefixlen != 128)
553 if (ifal->ifal_index &&
554 !__dev_get_by_index(net, ifal->ifal_index))
557 if (!tb[IFAL_ADDRESS])
559 addr = nla_data(tb[IFAL_ADDRESS]);
562 p = __ipv6_addr_label(net, addr, ipv6_addr_type(addr), ifal->ifal_index);
563 if (p && ip6addrlbl_hold(p))
565 lseq = ip6addrlbl_table.seq;
573 if (!(skb = nlmsg_new(ip6addrlbl_msgsize(), GFP_KERNEL))) {
578 err = ip6addrlbl_fill(skb, p, lseq,
579 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
580 RTM_NEWADDRLABEL, 0);
585 WARN_ON(err == -EMSGSIZE);
590 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
595 void __init ipv6_addr_label_rtnl_register(void)
597 __rtnl_register(PF_INET6, RTM_NEWADDRLABEL, ip6addrlbl_newdel,
599 __rtnl_register(PF_INET6, RTM_DELADDRLABEL, ip6addrlbl_newdel,
601 __rtnl_register(PF_INET6, RTM_GETADDRLABEL, ip6addrlbl_get,
602 ip6addrlbl_dump, NULL);