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 #include <linux/refcount.h>
  22
  23 #if 0
  24 #define ADDRLABEL(x...) printk(x)
  25 #else
  26 #define ADDRLABEL(x...) do { ; } while (0)
  27 #endif
  28
  29 /*
  30  * Policy Table
  31  */
  32 struct ip6addrlbl_entry {
  33         possible_net_t lbl_net;
  34         struct in6_addr prefix;
  35         int prefixlen;
  36         int ifindex;
  37         int addrtype;
  38         u32 label;
  39         struct hlist_node list;
  40         refcount_t refcnt;
  41         struct rcu_head rcu;
  42 };
  43
  44 static struct ip6addrlbl_table
  45 {
  46         struct hlist_head head;
  47         spinlock_t lock;
  48         u32 seq;
  49 } ip6addrlbl_table;
  50
  51 static inline
  52 struct net *ip6addrlbl_net(const struct ip6addrlbl_entry *lbl)
  53 {
  54         return read_pnet(&lbl->lbl_net);
  55 }
  56
  57 /*
  58  * Default policy table (RFC6724 + extensions)
  59  *
  60  * prefix               addr_type       label
  61  * -------------------------------------------------------------------------
  62  * ::1/128              LOOPBACK        0
  63  * ::/0                 N/A             1
  64  * 2002::/16            N/A             2
  65  * ::/96                COMPATv4        3
  66  * ::ffff:0:0/96        V4MAPPED        4
  67  * fc00::/7             N/A             5               ULA (RFC 4193)
  68  * 2001::/32            N/A             6               Teredo (RFC 4380)
  69  * 2001:10::/28         N/A             7               ORCHID (RFC 4843)
  70  * fec0::/10            N/A             11              Site-local
  71  *                                                      (deprecated by RFC3879)
  72  * 3ffe::/16            N/A             12              6bone
  73  *
  74  * Note: 0xffffffff is used if we do not have any policies.
  75  * Note: Labels for ULA and 6to4 are different from labels listed in RFC6724.
  76  */
  77
  78 #define IPV6_ADDR_LABEL_DEFAULT 0xffffffffUL
  79
  80 static const __net_initconst struct ip6addrlbl_init_table
  81 {
  82         const struct in6_addr *prefix;
  83         int prefixlen;
  84         u32 label;
  85 } ip6addrlbl_init_table[] = {
  86         {       /* ::/0 */
  87                 .prefix = &in6addr_any,
  88                 .label = 1,
  89         }, {    /* fc00::/7 */
  90                 .prefix = &(struct in6_addr){ { { 0xfc } } } ,
  91                 .prefixlen = 7,
  92                 .label = 5,
  93         }, {    /* fec0::/10 */
  94                 .prefix = &(struct in6_addr){ { { 0xfe, 0xc0 } } },
  95                 .prefixlen = 10,
  96                 .label = 11,
  97         }, {    /* 2002::/16 */
  98                 .prefix = &(struct in6_addr){ { { 0x20, 0x02 } } },
  99                 .prefixlen = 16,
 100                 .label = 2,
 101         }, {    /* 3ffe::/16 */
 102                 .prefix = &(struct in6_addr){ { { 0x3f, 0xfe } } },
 103                 .prefixlen = 16,
 104                 .label = 12,
 105         }, {    /* 2001::/32 */
 106                 .prefix = &(struct in6_addr){ { { 0x20, 0x01 } } },
 107                 .prefixlen = 32,
 108                 .label = 6,
 109         }, {    /* 2001:10::/28 */
 110                 .prefix = &(struct in6_addr){ { { 0x20, 0x01, 0x00, 0x10 } } },
 111                 .prefixlen = 28,
 112                 .label = 7,
 113         }, {    /* ::ffff:0:0 */
 114                 .prefix = &(struct in6_addr){ { { [10] = 0xff, [11] = 0xff } } },
 115                 .prefixlen = 96,
 116                 .label = 4,
 117         }, {    /* ::/96 */
 118                 .prefix = &in6addr_any,
 119                 .prefixlen = 96,
 120                 .label = 3,
 121         }, {    /* ::1/128 */
 122                 .prefix = &in6addr_loopback,
 123                 .prefixlen = 128,
 124                 .label = 0,
 125         }
 126 };
 127
 128 /* Object management */
 129 static inline void ip6addrlbl_free(struct ip6addrlbl_entry *p)
 130 {
 131         kfree(p);
 132 }
 133
 134 static void ip6addrlbl_free_rcu(struct rcu_head *h)
 135 {
 136         ip6addrlbl_free(container_of(h, struct ip6addrlbl_entry, rcu));
 137 }
 138
 139 static bool ip6addrlbl_hold(struct ip6addrlbl_entry *p)
 140 {
 141         return refcount_inc_not_zero(&p->refcnt);
 142 }
 143
 144 static inline void ip6addrlbl_put(struct ip6addrlbl_entry *p)
 145 {
 146         if (refcount_dec_and_test(&p->refcnt))
 147                 call_rcu(&p->rcu, ip6addrlbl_free_rcu);
 148 }
 149
 150 /* Find label */
 151 static bool __ip6addrlbl_match(struct net *net,
 152                                const struct ip6addrlbl_entry *p,
 153                                const struct in6_addr *addr,
 154                                int addrtype, int ifindex)
 155 {
 156         if (!net_eq(ip6addrlbl_net(p), net))
 157                 return false;
 158         if (p->ifindex && p->ifindex != ifindex)
 159                 return false;
 160         if (p->addrtype && p->addrtype != addrtype)
 161                 return false;
 162         if (!ipv6_prefix_equal(addr, &p->prefix, p->prefixlen))
 163                 return false;
 164         return true;
 165 }
 166
 167 static struct ip6addrlbl_entry *__ipv6_addr_label(struct net *net,
 168                                                   const struct in6_addr *addr,
 169                                                   int type, int ifindex)
 170 {
 171         struct ip6addrlbl_entry *p;
 172         hlist_for_each_entry_rcu(p, &ip6addrlbl_table.head, list) {
 173                 if (__ip6addrlbl_match(net, p, addr, type, ifindex))
 174                         return p;
 175         }
 176         return NULL;
 177 }
 178
 179 u32 ipv6_addr_label(struct net *net,
 180                     const struct in6_addr *addr, int type, int ifindex)
 181 {
 182         u32 label;
 183         struct ip6addrlbl_entry *p;
 184
 185         type &= IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK;
 186
 187         rcu_read_lock();
 188         p = __ipv6_addr_label(net, addr, type, ifindex);
 189         label = p ? p->label : IPV6_ADDR_LABEL_DEFAULT;
 190         rcu_read_unlock();
 191
 192         ADDRLABEL(KERN_DEBUG "%s(addr=%pI6, type=%d, ifindex=%d) => %08x\n",
 193                   __func__, addr, type, ifindex, label);
 194
 195         return label;
 196 }
 197
 198 /* allocate one entry */
 199 static struct ip6addrlbl_entry *ip6addrlbl_alloc(struct net *net,
 200                                                  const struct in6_addr *prefix,
 201                                                  int prefixlen, int ifindex,
 202                                                  u32 label)
 203 {
 204         struct ip6addrlbl_entry *newp;
 205         int addrtype;
 206
 207         ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u)\n",
 208                   __func__, prefix, prefixlen, ifindex, (unsigned int)label);
 209
 210         addrtype = ipv6_addr_type(prefix) & (IPV6_ADDR_MAPPED | IPV6_ADDR_COMPATv4 | IPV6_ADDR_LOOPBACK);
 211
 212         switch (addrtype) {
 213         case IPV6_ADDR_MAPPED:
 214                 if (prefixlen > 96)
 215                         return ERR_PTR(-EINVAL);
 216                 if (prefixlen < 96)
 217                         addrtype = 0;
 218                 break;
 219         case IPV6_ADDR_COMPATv4:
 220                 if (prefixlen != 96)
 221                         addrtype = 0;
 222                 break;
 223         case IPV6_ADDR_LOOPBACK:
 224                 if (prefixlen != 128)
 225                         addrtype = 0;
 226                 break;
 227         }
 228
 229         newp = kmalloc(sizeof(*newp), GFP_KERNEL);
 230         if (!newp)
 231                 return ERR_PTR(-ENOMEM);
 232
 233         ipv6_addr_prefix(&newp->prefix, prefix, prefixlen);
 234         newp->prefixlen = prefixlen;
 235         newp->ifindex = ifindex;
 236         newp->addrtype = addrtype;
 237         newp->label = label;
 238         INIT_HLIST_NODE(&newp->list);
 239         write_pnet(&newp->lbl_net, net);
 240         refcount_set(&newp->refcnt, 1);
 241         return newp;
 242 }
 243
 244 /* add a label */
 245 static int __ip6addrlbl_add(struct ip6addrlbl_entry *newp, int replace)
 246 {
 247         struct hlist_node *n;
 248         struct ip6addrlbl_entry *last = NULL, *p = NULL;
 249         int ret = 0;
 250
 251         ADDRLABEL(KERN_DEBUG "%s(newp=%p, replace=%d)\n", __func__, newp,
 252                   replace);
 253
 254         hlist_for_each_entry_safe(p, n, &ip6addrlbl_table.head, list) {
 255                 if (p->prefixlen == newp->prefixlen &&
 256                     net_eq(ip6addrlbl_net(p), ip6addrlbl_net(newp)) &&
 257                     p->ifindex == newp->ifindex &&
 258                     ipv6_addr_equal(&p->prefix, &newp->prefix)) {
 259                         if (!replace) {
 260                                 ret = -EEXIST;
 261                                 goto out;
 262                         }
 263                         hlist_replace_rcu(&p->list, &newp->list);
 264                         ip6addrlbl_put(p);
 265                         goto out;
 266                 } else if ((p->prefixlen == newp->prefixlen && !p->ifindex) ||
 267                            (p->prefixlen < newp->prefixlen)) {
 268                         hlist_add_before_rcu(&newp->list, &p->list);
 269                         goto out;
 270                 }
 271                 last = p;
 272         }
 273         if (last)
 274                 hlist_add_behind_rcu(&newp->list, &last->list);
 275         else
 276                 hlist_add_head_rcu(&newp->list, &ip6addrlbl_table.head);
 277 out:
 278         if (!ret)
 279                 ip6addrlbl_table.seq++;
 280         return ret;
 281 }
 282
 283 /* add a label */
 284 static int ip6addrlbl_add(struct net *net,
 285                           const struct in6_addr *prefix, int prefixlen,
 286                           int ifindex, u32 label, int replace)
 287 {
 288         struct ip6addrlbl_entry *newp;
 289         int ret = 0;
 290
 291         ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d, label=%u, replace=%d)\n",
 292                   __func__, prefix, prefixlen, ifindex, (unsigned int)label,
 293                   replace);
 294
 295         newp = ip6addrlbl_alloc(net, prefix, prefixlen, ifindex, label);
 296         if (IS_ERR(newp))
 297                 return PTR_ERR(newp);
 298         spin_lock(&ip6addrlbl_table.lock);
 299         ret = __ip6addrlbl_add(newp, replace);
 300         spin_unlock(&ip6addrlbl_table.lock);
 301         if (ret)
 302                 ip6addrlbl_free(newp);
 303         return ret;
 304 }
 305
 306 /* remove a label */
 307 static int __ip6addrlbl_del(struct net *net,
 308                             const struct in6_addr *prefix, int prefixlen,
 309                             int ifindex)
 310 {
 311         struct ip6addrlbl_entry *p = NULL;
 312         struct hlist_node *n;
 313         int ret = -ESRCH;
 314
 315         ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
 316                   __func__, prefix, prefixlen, ifindex);
 317
 318         hlist_for_each_entry_safe(p, n, &ip6addrlbl_table.head, list) {
 319                 if (p->prefixlen == prefixlen &&
 320                     net_eq(ip6addrlbl_net(p), net) &&
 321                     p->ifindex == ifindex &&
 322                     ipv6_addr_equal(&p->prefix, prefix)) {
 323                         hlist_del_rcu(&p->list);
 324                         ip6addrlbl_put(p);
 325                         ret = 0;
 326                         break;
 327                 }
 328         }
 329         return ret;
 330 }
 331
 332 static int ip6addrlbl_del(struct net *net,
 333                           const struct in6_addr *prefix, int prefixlen,
 334                           int ifindex)
 335 {
 336         struct in6_addr prefix_buf;
 337         int ret;
 338
 339         ADDRLABEL(KERN_DEBUG "%s(prefix=%pI6, prefixlen=%d, ifindex=%d)\n",
 340                   __func__, prefix, prefixlen, ifindex);
 341
 342         ipv6_addr_prefix(&prefix_buf, prefix, prefixlen);
 343         spin_lock(&ip6addrlbl_table.lock);
 344         ret = __ip6addrlbl_del(net, &prefix_buf, prefixlen, ifindex);
 345         spin_unlock(&ip6addrlbl_table.lock);
 346         return ret;
 347 }
 348
 349 /* add default label */
 350 static int __net_init ip6addrlbl_net_init(struct net *net)
 351 {
 352         int err = 0;
 353         int i;
 354
 355         ADDRLABEL(KERN_DEBUG "%s\n", __func__);
 356
 357         for (i = 0; i < ARRAY_SIZE(ip6addrlbl_init_table); i++) {
 358                 int ret = ip6addrlbl_add(net,
 359                                          ip6addrlbl_init_table[i].prefix,
 360                                          ip6addrlbl_init_table[i].prefixlen,
 361                                          0,
 362                                          ip6addrlbl_init_table[i].label, 0);
 363                 /* XXX: should we free all rules when we catch an error? */
 364                 if (ret && (!err || err != -ENOMEM))
 365                         err = ret;
 366         }
 367         return err;
 368 }
 369
 370 static void __net_exit ip6addrlbl_net_exit(struct net *net)
 371 {
 372         struct ip6addrlbl_entry *p = NULL;
 373         struct hlist_node *n;
 374
 375         /* Remove all labels belonging to the exiting net */
 376         spin_lock(&ip6addrlbl_table.lock);
 377         hlist_for_each_entry_safe(p, n, &ip6addrlbl_table.head, list) {
 378                 if (net_eq(ip6addrlbl_net(p), net)) {
 379                         hlist_del_rcu(&p->list);
 380                         ip6addrlbl_put(p);
 381                 }
 382         }
 383         spin_unlock(&ip6addrlbl_table.lock);
 384 }
 385
 386 static struct pernet_operations ipv6_addr_label_ops = {
 387         .init = ip6addrlbl_net_init,
 388         .exit = ip6addrlbl_net_exit,
 389 };
 390
 391 int __init ipv6_addr_label_init(void)
 392 {
 393         spin_lock_init(&ip6addrlbl_table.lock);
 394
 395         return register_pernet_subsys(&ipv6_addr_label_ops);
 396 }
 397
 398 void ipv6_addr_label_cleanup(void)
 399 {
 400         unregister_pernet_subsys(&ipv6_addr_label_ops);
 401 }
 402
 403 static const struct nla_policy ifal_policy[IFAL_MAX+1] = {
 404         [IFAL_ADDRESS]          = { .len = sizeof(struct in6_addr), },
 405         [IFAL_LABEL]            = { .len = sizeof(u32), },
 406 };
 407
 408 static int ip6addrlbl_newdel(struct sk_buff *skb, struct nlmsghdr *nlh,
 409                              struct netlink_ext_ack *extack)
 410 {
 411         struct net *net = sock_net(skb->sk);
 412         struct ifaddrlblmsg *ifal;
 413         struct nlattr *tb[IFAL_MAX+1];
 414         struct in6_addr *pfx;
 415         u32 label;
 416         int err = 0;
 417
 418         err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy,
 419                           extack);
 420         if (err < 0)
 421                 return err;
 422
 423         ifal = nlmsg_data(nlh);
 424
 425         if (ifal->ifal_family != AF_INET6 ||
 426             ifal->ifal_prefixlen > 128)
 427                 return -EINVAL;
 428
 429         if (!tb[IFAL_ADDRESS])
 430                 return -EINVAL;
 431         pfx = nla_data(tb[IFAL_ADDRESS]);
 432
 433         if (!tb[IFAL_LABEL])
 434                 return -EINVAL;
 435         label = nla_get_u32(tb[IFAL_LABEL]);
 436         if (label == IPV6_ADDR_LABEL_DEFAULT)
 437                 return -EINVAL;
 438
 439         switch (nlh->nlmsg_type) {
 440         case RTM_NEWADDRLABEL:
 441                 if (ifal->ifal_index &&
 442                     !__dev_get_by_index(net, ifal->ifal_index))
 443                         return -EINVAL;
 444
 445                 err = ip6addrlbl_add(net, pfx, ifal->ifal_prefixlen,
 446                                      ifal->ifal_index, label,
 447                                      nlh->nlmsg_flags & NLM_F_REPLACE);
 448                 break;
 449         case RTM_DELADDRLABEL:
 450                 err = ip6addrlbl_del(net, pfx, ifal->ifal_prefixlen,
 451                                      ifal->ifal_index);
 452                 break;
 453         default:
 454                 err = -EOPNOTSUPP;
 455         }
 456         return err;
 457 }
 458
 459 static void ip6addrlbl_putmsg(struct nlmsghdr *nlh,
 460                               int prefixlen, int ifindex, u32 lseq)
 461 {
 462         struct ifaddrlblmsg *ifal = nlmsg_data(nlh);
 463         ifal->ifal_family = AF_INET6;
 464         ifal->ifal_prefixlen = prefixlen;
 465         ifal->ifal_flags = 0;
 466         ifal->ifal_index = ifindex;
 467         ifal->ifal_seq = lseq;
 468 };
 469
 470 static int ip6addrlbl_fill(struct sk_buff *skb,
 471                            struct ip6addrlbl_entry *p,
 472                            u32 lseq,
 473                            u32 portid, u32 seq, int event,
 474                            unsigned int flags)
 475 {
 476         struct nlmsghdr *nlh = nlmsg_put(skb, portid, seq, event,
 477                                          sizeof(struct ifaddrlblmsg), flags);
 478         if (!nlh)
 479                 return -EMSGSIZE;
 480
 481         ip6addrlbl_putmsg(nlh, p->prefixlen, p->ifindex, lseq);
 482
 483         if (nla_put_in6_addr(skb, IFAL_ADDRESS, &p->prefix) < 0 ||
 484             nla_put_u32(skb, IFAL_LABEL, p->label) < 0) {
 485                 nlmsg_cancel(skb, nlh);
 486                 return -EMSGSIZE;
 487         }
 488
 489         nlmsg_end(skb, nlh);
 490         return 0;
 491 }
 492
 493 static int ip6addrlbl_dump(struct sk_buff *skb, struct netlink_callback *cb)
 494 {
 495         struct net *net = sock_net(skb->sk);
 496         struct ip6addrlbl_entry *p;
 497         int idx = 0, s_idx = cb->args[0];
 498         int err;
 499
 500         rcu_read_lock();
 501         hlist_for_each_entry_rcu(p, &ip6addrlbl_table.head, list) {
 502                 if (idx >= s_idx &&
 503                     net_eq(ip6addrlbl_net(p), net)) {
 504                         err = ip6addrlbl_fill(skb, p,
 505                                               ip6addrlbl_table.seq,
 506                                               NETLINK_CB(cb->skb).portid,
 507                                               cb->nlh->nlmsg_seq,
 508                                               RTM_NEWADDRLABEL,
 509                                               NLM_F_MULTI);
 510                         if (err < 0)
 511                                 break;
 512                 }
 513                 idx++;
 514         }
 515         rcu_read_unlock();
 516         cb->args[0] = idx;
 517         return skb->len;
 518 }
 519
 520 static inline int ip6addrlbl_msgsize(void)
 521 {
 522         return NLMSG_ALIGN(sizeof(struct ifaddrlblmsg))
 523                 + nla_total_size(16)    /* IFAL_ADDRESS */
 524                 + nla_total_size(4);    /* IFAL_LABEL */
 525 }
 526
 527 static int ip6addrlbl_get(struct sk_buff *in_skb, struct nlmsghdr *nlh,
 528                           struct netlink_ext_ack *extack)
 529 {
 530         struct net *net = sock_net(in_skb->sk);
 531         struct ifaddrlblmsg *ifal;
 532         struct nlattr *tb[IFAL_MAX+1];
 533         struct in6_addr *addr;
 534         u32 lseq;
 535         int err = 0;
 536         struct ip6addrlbl_entry *p;
 537         struct sk_buff *skb;
 538
 539         err = nlmsg_parse(nlh, sizeof(*ifal), tb, IFAL_MAX, ifal_policy,
 540                           extack);
 541         if (err < 0)
 542                 return err;
 543
 544         ifal = nlmsg_data(nlh);
 545
 546         if (ifal->ifal_family != AF_INET6 ||
 547             ifal->ifal_prefixlen != 128)
 548                 return -EINVAL;
 549
 550         if (ifal->ifal_index &&
 551             !__dev_get_by_index(net, ifal->ifal_index))
 552                 return -EINVAL;
 553
 554         if (!tb[IFAL_ADDRESS])
 555                 return -EINVAL;
 556         addr = nla_data(tb[IFAL_ADDRESS]);
 557
 558         rcu_read_lock();
 559         p = __ipv6_addr_label(net, addr, ipv6_addr_type(addr), ifal->ifal_index);
 560         if (p && !ip6addrlbl_hold(p))
 561                 p = NULL;
 562         lseq = ip6addrlbl_table.seq;
 563         rcu_read_unlock();
 564
 565         if (!p) {
 566                 err = -ESRCH;
 567                 goto out;
 568         }
 569
 570         skb = nlmsg_new(ip6addrlbl_msgsize(), GFP_KERNEL);
 571         if (!skb) {
 572                 ip6addrlbl_put(p);
 573                 return -ENOBUFS;
 574         }
 575
 576         err = ip6addrlbl_fill(skb, p, lseq,
 577                               NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
 578                               RTM_NEWADDRLABEL, 0);
 579
 580         ip6addrlbl_put(p);
 581
 582         if (err < 0) {
 583                 WARN_ON(err == -EMSGSIZE);
 584                 kfree_skb(skb);
 585                 goto out;
 586         }
 587
 588         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
 589 out:
 590         return err;
 591 }
 592
 593 void __init ipv6_addr_label_rtnl_register(void)
 594 {
 595         __rtnl_register(PF_INET6, RTM_NEWADDRLABEL, ip6addrlbl_newdel,
 596                         NULL, NULL);
 597         __rtnl_register(PF_INET6, RTM_DELADDRLABEL, ip6addrlbl_newdel,
 598                         NULL, NULL);
 599         __rtnl_register(PF_INET6, RTM_GETADDRLABEL, ip6addrlbl_get,
 600                         ip6addrlbl_dump, NULL);
 601 }
 602