net/ipv6/addrlabel.c

   1 /*
   2  * IPv6 Address Label subsystem
   3  * for the IPv6 "Default" Source Address Selection
   4  *
   5  * Copyright (C)2007 USAGI/WIDE Project
   6  */
   7 /*
   8  * Author:
   9  *      YOSHIFUJI Hideaki @ USAGI/WIDE Project <yoshfuji@linux-ipv6.org>
  10  */
  11
  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>
  21
  22 #if 0
  23 #define ADDRLABEL(x...) printk(x)
  24 #else
  25 #define ADDRLABEL(x...) do { ; } while (0)
  26 #endif
  27
  28 /*
  29  * Policy Table
  30  */
  31 struct ip6addrlbl_entry {
  32 #ifdef CONFIG_NET_NS
  33         struct net *lbl_net;
  34 #endif
  35         struct in6_addr prefix;
  36         int prefixlen;
  37         int ifindex;
  38         int addrtype;
  39         u32 label;
  40         struct hlist_node list;
  41         atomic_t refcnt;
  42         struct rcu_head rcu;
  43 };
  44
  45 static struct ip6addrlbl_table
  46 {
  47         struct hlist_head head;
  48         spinlock_t lock;
  49         u32 seq;
  50 } ip6addrlbl_table;
  51
  52 static inline
  53 struct net *ip6addrlbl_net(const struct ip6addrlbl_entry *lbl)
  54 {
  55         return read_pnet(&lbl->lbl_net);
  56 }
  57
  58 /*
  59  * Default policy table (RFC6724 + extensions)
  60  *
  61  * prefix               addr_type       label
  62  * -------------------------------------------------------------------------
  63  * ::1/128              LOOPBACK        0
  64  * ::/0                 N/A             1
  65  * 2002::/16            N/A             2
  66  * ::/96                COMPATv4        3
  67  * ::ffff:0:0/96        V4MAPPED        4
  68  * fc00::/7             N/A             5               ULA (RFC 4193)
  69  * 2001::/32            N/A             6               Teredo (RFC 4380)
  70  * 2001:10::/28         N/A             7               ORCHID (RFC 4843)
  71  * fec0::/10            N/A             11              Site-local
  72  *                                                      (deprecated by RFC3879)
  73  * 3ffe::/16            N/A             12              6bone
  74  *
  75  * Note: 0xffffffff is used if we do not have any policies.
  76  * Note: Labels for ULA and 6to4 are different from labels listed in RFC6724.
  77  */
  78
  79 #define IPV6_ADDR_LABEL_DEFAULT 0xffffffffUL
  80
  81 static const __net_initconst struct ip6addrlbl_init_table
  82 {
  83         const struct in6_addr *prefix;
  84         int prefixlen;
  85         u32 label;
  86 } ip6addrlbl_init_table[] = {
  87         {       /* ::/0 */
  88                 .prefix = &in6addr_any,
  89                 .label = 1,
  90         }, {    /* fc00::/7 */
  91                 .prefix = &(struct in6_addr){ { { 0xfc } } } ,
  92                 .prefixlen = 7,
  93                 .label = 5,
  94         }, {    /* fec0::/10 */
  95                 .prefix = &(struct in6_addr){ { { 0xfe, 0xc0 } } },
  96                 .prefixlen = 10,
  97                 .label = 11,
  98         }, {    /* 2002::/16 */
  99                 .prefix = &(struct in6_addr){ { { 0x20, 0x02 } } },
 100                 .prefixlen = 16,
 101                 .label = 2,
 102         }, {    /* 3ffe::/16 */
 103                 .prefix = &(struct in6_addr){ { { 0x3f, 0xfe } } },
 104                 .prefixlen = 16,
 105                 .label = 12,
 106         }, {    /* 2001::/32 */
 107                 .prefix = &(struct in6_addr){ { { 0x20, 0x01 } } },
 108                 .prefixlen = 32,
 109                 .label = 6,
 110         }, {    /* 2001:10::/28 */
 111                 .prefix = &(struct in6_addr){ { { 0x20, 0x01, 0x00, 0x10 } } },
 112                 .prefixlen = 28,
 113                 .label = 7,
 114         }, {    /* ::ffff:0:0 */
 115                 .prefix = &(struct in6_addr){ { { [10] = 0xff, [11] = 0xff } } },
 116                 .prefixlen = 96,
 117                 .label = 4,
 118         }, {    /* ::/96 */
 119                 .prefix = &in6addr_any,
 120                 .prefixlen = 96,
 121                 .label = 3,
 122         }, {    /* ::1/128 */
 123                 .prefix = &in6addr_loopback,
 124                 .prefixlen = 128,
 125                 .label = 0,
 126         }
 127 };
 128
 129 /* Object management */
 130 static inline void ip6addrlbl_free(struct ip6addrlbl_entry *p)
 131 {
 132 #ifdef CONFIG_NET_NS
 133         release_net(p->lbl_net);
 134 #endif
 135         kfree(p);
 136 }
 137
 138 static void ip6addrlbl_free_rcu(struct rcu_head *h)
 139 {
 140         ip6addrlbl_free(container_of(h, struct ip6addrlbl_entry, rcu));
 141 }
 142
 143 static bool ip6addrlbl_hold(struct ip6addrlbl_entry *p)
 144 {
 145         return atomic_inc_not_zero(&p->refcnt);
 146 }
 147
 148 static inline void ip6addrlbl_put(struct ip6addrlbl_entry *p)
 149 {
 150         if (atomic_dec_and_test(&p->refcnt))
 151                 call_rcu(&p->rcu, ip6addrlbl_free_rcu);
 152 }
 153
 154 /* Find label */
 155 static bool __ip6addrlbl_match(struct net *net,
 156                                const struct ip6addrlbl_entry *p,
 157                                const struct in6_addr *addr,
 158                                int addrtype, int ifindex)
 159 {
 160         if (!net_eq(ip6addrlbl_net(p), net))
 161                 return false;
 162         if (p->ifindex && p->ifindex != ifindex)
 163                 return false;
 164         if (p->addrtype && p->addrtype != addrtype)
 165                 return false;
 166         if (!ipv6_prefix_equal(addr, &p->prefix, p->prefixlen))
 167                 return false;
 168         return true;
 169 }
 170
 171 static struct ip6addrlbl_entry *__ipv6_addr_label(struct net *net,
 172                                                   const struct in6_addr *addr,
 173                                                   int type, int ifindex)
 174 {
 175         struct ip6addrlbl_entry *p;
 176         hlist_for_each_entry_rcu(p, &ip6addrlbl_table.head, list) {
 177                 if (__ip6addrlbl_match(net, p, addr, type, ifindex))
 178                         return p;
 179         }
 180         return NULL;
 181 }
 182
 183 u32 ipv6_addr_label(struct net *net,
 184                     const struct in6_addr *addr, int type, int ifindex)
 185 {
 186         u32 label;
 187         struct ip6addrlbl_entry *p;
 188
 189         type &= IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK;
 190
 191         rcu_read_lock();
 192         p = __ipv6_addr_label(net, addr, type, ifindex);
 193         label = p ? p->label : IPV6_ADDR_LABEL_DEFAULT;
 194         rcu_read_unlock();
 195
 196         ADDRLABEL(KERN_DEBUG "%s(addr=%pI6, type=%d, ifindex=%d) => %08x\n",
 197                   __func__, addr, type, ifindex, label);
 198
 199         return label;
 200 }
 201
 202 /* allocate one entry */
 203 static struct ip6addrlbl_entry *ip6addrlbl_alloc(struct net *net,
 204                                                  const struct in6_addr *prefix,
 205                                                  int prefixlen, int ifindex,
 206                                                  u32 label)
 207 {
 208         struct ip6addrlbl_entry *newp;
 209         int addrtype;
 210
 211         ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u)\n",
 212                   __func__, prefix, prefixlen, ifindex, (unsigned int)label);
 213
 214         addrtype = ipv6_addr_type(prefix) & (IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK);
 215
 216         switch (addrtype) {
 217         case IPV6_ADDR_MAPPED:
 218                 if (prefixlen > 96)
 219                         return ERR_PTR(-EINVAL);
 220                 if (prefixlen < 96)
 221                         addrtype = 0;
 222                 break;
 223         case IPV6_ADDR_COMPATv4:
 224                 if (prefixlen != 96)
 225                         addrtype = 0;
 226                 break;
 227         case IPV6_ADDR_LOOPBACK:
 228                 if (prefixlen != 128)
 229                         addrtype = 0;
 230                 break;
 231         }
 232
 233         newp = kmalloc(sizeof(*newp), GFP_KERNEL);
 234         if (!newp)
 235                 return ERR_PTR(-ENOMEM);
 236
 237         ipv6_addr_prefix(&newp->prefix, prefix, prefixlen);
 238         newp->prefixlen = prefixlen;
 239         newp->ifindex = ifindex;
 240         newp->addrtype = addrtype;
 241         newp->label = label;
 242         INIT_HLIST_NODE(&newp->list);
 243 #ifdef CONFIG_NET_NS
 244         newp->lbl_net = hold_net(net);
 245 #endif
 246         atomic_set(&newp->refcnt, 1);
 247         return newp;
 248 }
 249
 250 /* add a label */
 251 static int __ip6addrlbl_add(struct ip6addrlbl_entry *newp, int replace)
 252 {
 253         struct hlist_node *n;
 254         struct ip6addrlbl_entry *last = NULL, *p = NULL;
 255         int ret = 0;
 256
 257         ADDRLABEL(KERN_DEBUG "%s(newp=%p, replace=%d)\n", __func__, newp,
 258                   replace);
 259
 260         hlist_for_each_entry_safe(p, n, &ip6addrlbl_table.head, list) {
 261                 if (p->prefixlen == newp->prefixlen &&
 262                     net_eq(ip6addrlbl_net(p), ip6addrlbl_net(newp)) &&
 263                     p->ifindex == newp->ifindex &&
 264                     ipv6_addr_equal(&p->prefix, &newp->prefix)) {
 265                         if (!replace) {
 266                                 ret = -EEXIST;
 267                                 goto out;
 268                         }
 269                         hlist_replace_rcu(&p->list, &newp->list);
 270                         ip6addrlbl_put(p);
 271                         goto out;
 272                 } else if ((p->prefixlen == newp->prefixlen && !p->ifindex) ||
 273                            (p->prefixlen < newp->prefixlen)) {
 274                         hlist_add_before_rcu(&newp->list, &p->list);
 275                         goto out;
 276                 }
 277                 last = p;
 278         }
 279         if (last)
 280                 hlist_add_behind_rcu(&newp->list, &last->list);
 281         else
 282                 hlist_add_head_rcu(&newp->list, &ip6addrlbl_table.head);
 283 out:
 284         if (!ret)
 285                 ip6addrlbl_table.seq++;
 286         return ret;
 287 }
 288
 289 /* add a label */
 290 static int ip6addrlbl_add(struct net *net,
 291                           const struct in6_addr *prefix, int prefixlen,
 292                           int ifindex, u32 label, int replace)
 293 {
 294         struct ip6addrlbl_entry *newp;
 295         int ret = 0;
 296
 297         ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u, replace=%d)\n",
 298                   __func__, prefix, prefixlen, ifindex, (unsigned int)label,
 299                   replace);
 300
 301         newp = ip6addrlbl_alloc(net, prefix, prefixlen, ifindex, label);
 302         if (IS_ERR(newp))
 303                 return PTR_ERR(newp);
 304         spin_lock(&ip6addrlbl_table.lock);
 305         ret = __ip6addrlbl_add(newp, replace);
 306         spin_unlock(&ip6addrlbl_table.lock);
 307         if (ret)
 308                 ip6addrlbl_free(newp);
 309         return ret;
 310 }
 311
 312 /* remove a label */
 313 static int __ip6addrlbl_del(struct net *net,
 314                             const struct in6_addr *prefix, int prefixlen,
 315                             int ifindex)
 316 {
 317         struct ip6addrlbl_entry *p = NULL;
 318         struct hlist_node *n;
 319         int ret = -ESRCH;
 320
 321         ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
 322                   __func__, prefix, prefixlen, ifindex);
 323
 324         hlist_for_each_entry_safe(p, n, &ip6addrlbl_table.head, list) {
 325                 if (p->prefixlen == prefixlen &&
 326                     net_eq(ip6addrlbl_net(p), net) &&
 327                     p->ifindex == ifindex &&
 328                     ipv6_addr_equal(&p->prefix, prefix)) {
 329                         hlist_del_rcu(&p->list);
 330                         ip6addrlbl_put(p);
 331                         ret = 0;
 332                         break;
 333                 }
 334         }
 335         return ret;
 336 }
 337
 338 static int ip6addrlbl_del(struct net *net,
 339                           const struct in6_addr *prefix, int prefixlen,
 340                           int ifindex)
 341 {
 342         struct in6_addr prefix_buf;
 343         int ret;
 344
 345         ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
 346                   __func__, prefix, prefixlen, ifindex);
 347
 348         ipv6_addr_prefix(&prefix_buf, prefix, prefixlen);
 349         spin_lock(&ip6addrlbl_table.lock);
 350         ret = __ip6addrlbl_del(net, &prefix_buf, prefixlen, ifindex);
 351         spin_unlock(&ip6addrlbl_table.lock);
 352         return ret;
 353 }
 354
 355 /* add default label */
 356 static int __net_init ip6addrlbl_net_init(struct net *net)
 357 {
 358         int err = 0;
 359         int i;
 360
 361         ADDRLABEL(KERN_DEBUG "%s\n", __func__);
 362
 363         for (i = 0; i < ARRAY_SIZE(ip6addrlbl_init_table); i++) {
 364                 int ret = ip6addrlbl_add(net,
 365                                          ip6addrlbl_init_table[i].prefix,
 366                                          ip6addrlbl_init_table[i].prefixlen,
 367                                          0,
 368                                          ip6addrlbl_init_table[i].label, 0);
 369                 /* XXX: should we free all rules when we catch an error? */
 370                 if (ret && (!err || err != -ENOMEM))
 371                         err = ret;
 372         }
 373         return err;
 374 }
 375
 376 static void __net_exit ip6addrlbl_net_exit(struct net *net)
 377 {
 378         struct ip6addrlbl_entry *p = NULL;
 379         struct hlist_node *n;
 380
 381         /* Remove all labels belonging to the exiting net */
 382         spin_lock(&ip6addrlbl_table.lock);
 383         hlist_for_each_entry_safe(p, n, &ip6addrlbl_table.head, list) {
 384                 if (net_eq(ip6addrlbl_net(p), net)) {
 385                         hlist_del_rcu(&p->list);
 386                         ip6addrlbl_put(p);
 387                 }
 388         }
 389         spin_unlock(&ip6addrlbl_table.lock);
 390 }
 391
 392 static struct pernet_operations ipv6_addr_label_ops = {
 393         .init = ip6addrlbl_net_init,
 394         .exit = ip6addrlbl_net_exit,
 395 };
 396
 397 int __init ipv6_addr_label_init(void)
 398 {
 399         spin_lock_init(&ip6addrlbl_table.lock);
 400
 401         return register_pernet_subsys(&ipv6_addr_label_ops);
 402 }
 403
 404 void ipv6_addr_label_cleanup(void)
 405 {
 406         unregister_pernet_subsys(&ipv6_addr_label_ops);
 407 }
 408
 409 static const struct nla_policy ifal_policy[IFAL_MAX+1] = {
 410         [IFAL_ADDRESS]          = { .len = sizeof(struct in6_addr), },
 411         [IFAL_LABEL]            = { .len = sizeof(u32), },
 412 };
 413
 414 static int ip6addrlbl_newdel(struct sk_buff *skb, struct nlmsghdr *nlh)
 415 {
 416         struct net *net = sock_net(skb->sk);
 417         struct ifaddrlblmsg *ifal;
 418         struct nlattr *tb[IFAL_MAX+1];
 419         struct in6_addr *pfx;
 420         u32 label;
 421         int err = 0;
 422
 423         err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
 424         if (err < 0)
 425                 return err;
 426
 427         ifal = nlmsg_data(nlh);
 428
 429         if (ifal->ifal_family != AF_INET6 ||
 430             ifal->ifal_prefixlen > 128)
 431                 return -EINVAL;
 432
 433         if (!tb[IFAL_ADDRESS])
 434                 return -EINVAL;
 435         pfx = nla_data(tb[IFAL_ADDRESS]);
 436
 437         if (!tb[IFAL_LABEL])
 438                 return -EINVAL;
 439         label = nla_get_u32(tb[IFAL_LABEL]);
 440         if (label == IPV6_ADDR_LABEL_DEFAULT)
 441                 return -EINVAL;
 442
 443         switch (nlh->nlmsg_type) {
 444         case RTM_NEWADDRLABEL:
 445                 if (ifal->ifal_index &&
 446                     !__dev_get_by_index(net, ifal->ifal_index))
 447                         return -EINVAL;
 448
 449                 err = ip6addrlbl_add(net, pfx, ifal->ifal_prefixlen,
 450                                      ifal->ifal_index, label,
 451                                      nlh->nlmsg_flags & NLM_F_REPLACE);
 452                 break;
 453         case RTM_DELADDRLABEL:
 454                 err = ip6addrlbl_del(net, pfx, ifal->ifal_prefixlen,
 455                                      ifal->ifal_index);
 456                 break;
 457         default:
 458                 err = -EOPNOTSUPP;
 459         }
 460         return err;
 461 }
 462
 463 static void ip6addrlbl_putmsg(struct nlmsghdr *nlh,
 464                               int prefixlen, int ifindex, u32 lseq)
 465 {
 466         struct ifaddrlblmsg *ifal = nlmsg_data(nlh);
 467         ifal->ifal_family = AF_INET6;
 468         ifal->ifal_prefixlen = prefixlen;
 469         ifal->ifal_flags = 0;
 470         ifal->ifal_index = ifindex;
 471         ifal->ifal_seq = lseq;
 472 };
 473
 474 static int ip6addrlbl_fill(struct sk_buff *skb,
 475                            struct ip6addrlbl_entry *p,
 476                            u32 lseq,
 477                            u32 portid, u32 seq, int event,
 478                            unsigned int flags)
 479 {
 480         struct nlmsghdr *nlh = nlmsg_put(skb, portid, seq, event,
 481                                          sizeof(struct ifaddrlblmsg), flags);
 482         if (!nlh)
 483                 return -EMSGSIZE;
 484
 485         ip6addrlbl_putmsg(nlh, p->prefixlen, p->ifindex, lseq);
 486
 487         if (nla_put(skb, IFAL_ADDRESS, 16, &p->prefix) < 0 ||
 488             nla_put_u32(skb, IFAL_LABEL, p->label) < 0) {
 489                 nlmsg_cancel(skb, nlh);
 490                 return -EMSGSIZE;
 491         }
 492
 493         nlmsg_end(skb, nlh);
 494         return 0;
 495 }
 496
 497 static int ip6addrlbl_dump(struct sk_buff *skb, struct netlink_callback *cb)
 498 {
 499         struct net *net = sock_net(skb->sk);
 500         struct ip6addrlbl_entry *p;
 501         int idx = 0, s_idx = cb->args[0];
 502         int err;
 503
 504         rcu_read_lock();
 505         hlist_for_each_entry_rcu(p, &ip6addrlbl_table.head, list) {
 506                 if (idx >= s_idx &&
 507                     net_eq(ip6addrlbl_net(p), net)) {
 508                         err = ip6addrlbl_fill(skb, p,
 509                                               ip6addrlbl_table.seq,
 510                                               NETLINK_CB(cb->skb).portid,
 511                                               cb->nlh->nlmsg_seq,
 512                                               RTM_NEWADDRLABEL,
 513                                               NLM_F_MULTI);
 514                         if (err < 0)
 515                                 break;
 516                 }
 517                 idx++;
 518         }
 519         rcu_read_unlock();
 520         cb->args[0] = idx;
 521         return skb->len;
 522 }
 523
 524 static inline int ip6addrlbl_msgsize(void)
 525 {
 526         return NLMSG_ALIGN(sizeof(struct ifaddrlblmsg))
 527                 + nla_total_size(16)    /* IFAL_ADDRESS */
 528                 + nla_total_size(4);    /* IFAL_LABEL */
 529 }
 530
 531 static int ip6addrlbl_get(struct sk_buff *in_skb, struct nlmsghdr *nlh)
 532 {
 533         struct net *net = sock_net(in_skb->sk);
 534         struct ifaddrlblmsg *ifal;
 535         struct nlattr *tb[IFAL_MAX+1];
 536         struct in6_addr *addr;
 537         u32 lseq;
 538         int err = 0;
 539         struct ip6addrlbl_entry *p;
 540         struct sk_buff *skb;
 541
 542         err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy);
 543         if (err < 0)
 544                 return err;
 545
 546         ifal = nlmsg_data(nlh);
 547
 548         if (ifal->ifal_family != AF_INET6 ||
 549             ifal->ifal_prefixlen != 128)
 550                 return -EINVAL;
 551
 552         if (ifal->ifal_index &&
 553             !__dev_get_by_index(net, ifal->ifal_index))
 554                 return -EINVAL;
 555
 556         if (!tb[IFAL_ADDRESS])
 557                 return -EINVAL;
 558         addr = nla_data(tb[IFAL_ADDRESS]);
 559
 560         rcu_read_lock();
 561         p = __ipv6_addr_label(net, addr, ipv6_addr_type(addr), ifal->ifal_index);
 562         if (p && ip6addrlbl_hold(p))
 563                 p = NULL;
 564         lseq = ip6addrlbl_table.seq;
 565         rcu_read_unlock();
 566
 567         if (!p) {
 568                 err = -ESRCH;
 569                 goto out;
 570         }
 571
 572         skb = nlmsg_new(ip6addrlbl_msgsize(), GFP_KERNEL);
 573         if (!skb) {
 574                 ip6addrlbl_put(p);
 575                 return -ENOBUFS;
 576         }
 577
 578         err = ip6addrlbl_fill(skb, p, lseq,
 579                               NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
 580                               RTM_NEWADDRLABEL, 0);
 581
 582         ip6addrlbl_put(p);
 583
 584         if (err < 0) {
 585                 WARN_ON(err == -EMSGSIZE);
 586                 kfree_skb(skb);
 587                 goto out;
 588         }
 589
 590         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
 591 out:
 592         return err;
 593 }
 594
 595 void __init ipv6_addr_label_rtnl_register(void)
 596 {
 597         __rtnl_register(PF_INET6, RTM_NEWADDRLABEL, ip6addrlbl_newdel,
 598                         NULL, NULL);
 599         __rtnl_register(PF_INET6, RTM_DELADDRLABEL, ip6addrlbl_newdel,
 600                         NULL, NULL);
 601         __rtnl_register(PF_INET6, RTM_GETADDRLABEL, ip6addrlbl_get,
 602                         ip6addrlbl_dump, NULL);
 603 }
 604