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