net/ipv6/anycast.c

   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/config.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/sched.h>
  25 #include <linux/net.h>
  26 #include <linux/in6.h>
  27 #include <linux/netdevice.h>
  28 #include <linux/if_arp.h>
  29 #include <linux/route.h>
  30 #include <linux/init.h>
  31 #include <linux/proc_fs.h>
  32 #include <linux/seq_file.h>
  33
  34 #include <net/sock.h>
  35 #include <net/snmp.h>
  36
  37 #include <net/ipv6.h>
  38 #include <net/protocol.h>
  39 #include <net/if_inet6.h>
  40 #include <net/ndisc.h>
  41 #include <net/addrconf.h>
  42 #include <net/ip6_route.h>
  43
  44 #include <net/checksum.h>
  45
  46 /* Big ac list lock for all the sockets */
  47 static rwlock_t ipv6_sk_ac_lock = RW_LOCK_UNLOCKED;
  48
  49 /* XXX ip6_addr_match() and ip6_onlink() really belong in net/core.c */
  50
  51 static int
  52 ip6_addr_match(struct in6_addr *addr1, struct in6_addr *addr2, int prefix)
  53 {
  54         __u32   mask;
  55         int     i;
  56
  57         if (prefix > 128 || prefix < 0)
  58                 return 0;
  59         if (prefix == 0)
  60                 return 1;
  61         for (i=0; i<4; ++i) {
  62                 if (prefix >= 32)
  63                         mask = ~0;
  64                 else
  65                         mask = htonl(~0 << (32 - prefix));
  66                 if ((addr1->s6_addr32[i] ^ addr2->s6_addr32[i]) & mask)
  67                         return 0;
  68                 prefix -= 32;
  69                 if (prefix <= 0)
  70                         break;
  71         }
  72         return 1;
  73 }
  74
  75 static int
  76 ip6_onlink(struct in6_addr *addr, struct net_device *dev)
  77 {
  78         struct inet6_dev        *idev;
  79         struct inet6_ifaddr     *ifa;
  80         int     onlink;
  81
  82         onlink = 0;
  83         read_lock(&addrconf_lock);
  84         idev = __in6_dev_get(dev);
  85         if (idev) {
  86                 read_lock_bh(&idev->lock);
  87                 for (ifa=idev->addr_list; ifa; ifa=ifa->if_next) {
  88                         onlink = ip6_addr_match(addr, &ifa->addr,
  89                                         ifa->prefix_len);
  90                         if (onlink)
  91                                 break;
  92                 }
  93                 read_unlock_bh(&idev->lock);
  94         }
  95         read_unlock(&addrconf_lock);
  96         return onlink;
  97 }
  98
  99 /*
 100  *      socket join an anycast group
 101  */
 102
 103 int ipv6_sock_ac_join(struct sock *sk, int ifindex, struct in6_addr *addr)
 104 {
 105         struct ipv6_pinfo *np = inet6_sk(sk);
 106         struct net_device *dev = NULL;
 107         struct inet6_dev *idev;
 108         struct ipv6_ac_socklist *pac;
 109         int     ishost = !ipv6_devconf.forwarding;
 110         int     err = 0;
 111
 112         if (!capable(CAP_NET_ADMIN))
 113                 return -EPERM;
 114         if (ipv6_addr_is_multicast(addr))
 115                 return -EINVAL;
 116         if (ipv6_chk_addr(addr, NULL, 0))
 117                 return -EINVAL;
 118
 119         pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL);
 120         if (pac == NULL)
 121                 return -ENOMEM;
 122         pac->acl_next = NULL;
 123         ipv6_addr_copy(&pac->acl_addr, addr);
 124
 125         if (ifindex == 0) {
 126                 struct rt6_info *rt;
 127
 128                 rt = rt6_lookup(addr, NULL, 0, 0);
 129                 if (rt) {
 130                         dev = rt->rt6i_dev;
 131                         dev_hold(dev);
 132                         dst_release(&rt->u.dst);
 133                 } else if (ishost) {
 134                         err = -EADDRNOTAVAIL;
 135                         goto out_free_pac;
 136                 } else {
 137                         /* router, no matching interface: just pick one */
 138
 139                         dev = dev_get_by_flags(IFF_UP, IFF_UP|IFF_LOOPBACK);
 140                 }
 141         } else
 142                 dev = dev_get_by_index(ifindex);
 143
 144         if (dev == NULL) {
 145                 err = -ENODEV;
 146                 goto out_free_pac;
 147         }
 148
 149         idev = in6_dev_get(dev);
 150         if (!idev) {
 151                 if (ifindex)
 152                         err = -ENODEV;
 153                 else
 154                         err = -EADDRNOTAVAIL;
 155                 goto out_dev_put;
 156         }
 157         /* reset ishost, now that we have a specific device */
 158         ishost = !idev->cnf.forwarding;
 159         in6_dev_put(idev);
 160
 161         pac->acl_ifindex = dev->ifindex;
 162
 163         /* XXX
 164          * For hosts, allow link-local or matching prefix anycasts.
 165          * This obviates the need for propagating anycast routes while
 166          * still allowing some non-router anycast participation.
 167          */
 168         if (!ip6_onlink(addr, dev)) {
 169                 if (ishost)
 170                         err = -EADDRNOTAVAIL;
 171                 if (err)
 172                         goto out_dev_put;
 173         }
 174
 175         err = ipv6_dev_ac_inc(dev, addr);
 176         if (err)
 177                 goto out_dev_put;
 178
 179         write_lock_bh(&ipv6_sk_ac_lock);
 180         pac->acl_next = np->ipv6_ac_list;
 181         np->ipv6_ac_list = pac;
 182         write_unlock_bh(&ipv6_sk_ac_lock);
 183
 184         dev_put(dev);
 185
 186         return 0;
 187
 188 out_dev_put:
 189         dev_put(dev);
 190 out_free_pac:
 191         sock_kfree_s(sk, pac, sizeof(*pac));
 192         return err;
 193 }
 194
 195 /*
 196  *      socket leave an anycast group
 197  */
 198 int ipv6_sock_ac_drop(struct sock *sk, int ifindex, struct in6_addr *addr)
 199 {
 200         struct ipv6_pinfo *np = inet6_sk(sk);
 201         struct net_device *dev;
 202         struct ipv6_ac_socklist *pac, *prev_pac;
 203
 204         write_lock_bh(&ipv6_sk_ac_lock);
 205         prev_pac = 0;
 206         for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) {
 207                 if ((ifindex == 0 || pac->acl_ifindex == ifindex) &&
 208                      ipv6_addr_cmp(&pac->acl_addr, addr) == 0)
 209                         break;
 210                 prev_pac = pac;
 211         }
 212         if (!pac) {
 213                 write_unlock_bh(&ipv6_sk_ac_lock);
 214                 return -ENOENT;
 215         }
 216         if (prev_pac)
 217                 prev_pac->acl_next = pac->acl_next;
 218         else
 219                 np->ipv6_ac_list = pac->acl_next;
 220
 221         write_unlock_bh(&ipv6_sk_ac_lock);
 222
 223         dev = dev_get_by_index(pac->acl_ifindex);
 224         if (dev) {
 225                 ipv6_dev_ac_dec(dev, &pac->acl_addr);
 226                 dev_put(dev);
 227         }
 228         sock_kfree_s(sk, pac, sizeof(*pac));
 229         return 0;
 230 }
 231
 232 void ipv6_sock_ac_close(struct sock *sk)
 233 {
 234         struct ipv6_pinfo *np = inet6_sk(sk);
 235         struct net_device *dev = 0;
 236         struct ipv6_ac_socklist *pac;
 237         int     prev_index;
 238
 239         write_lock_bh(&ipv6_sk_ac_lock);
 240         pac = np->ipv6_ac_list;
 241         np->ipv6_ac_list = 0;
 242         write_unlock_bh(&ipv6_sk_ac_lock);
 243
 244         prev_index = 0;
 245         while (pac) {
 246                 struct ipv6_ac_socklist *next = pac->acl_next;
 247
 248                 if (pac->acl_ifindex != prev_index) {
 249                         if (dev)
 250                                 dev_put(dev);
 251                         dev = dev_get_by_index(pac->acl_ifindex);
 252                         prev_index = pac->acl_ifindex;
 253                 }
 254                 if (dev)
 255                         ipv6_dev_ac_dec(dev, &pac->acl_addr);
 256                 sock_kfree_s(sk, pac, sizeof(*pac));
 257                 pac = next;
 258         }
 259         if (dev)
 260                 dev_put(dev);
 261 }
 262
 263 #if 0
 264 /* The function is not used, which is funny. Apparently, author
 265  * supposed to use it to filter out datagrams inside udp/raw but forgot.
 266  *
 267  * It is OK, anycasts are not special comparing to delivery to unicasts.
 268  */
 269
 270 int inet6_ac_check(struct sock *sk, struct in6_addr *addr, int ifindex)
 271 {
 272         struct ipv6_ac_socklist *pac;
 273         struct ipv6_pinfo *np = inet6_sk(sk);
 274         int     found;
 275
 276         found = 0;
 277         read_lock(&ipv6_sk_ac_lock);
 278         for (pac=np->ipv6_ac_list; pac; pac=pac->acl_next) {
 279                 if (ifindex && pac->acl_ifindex != ifindex)
 280                         continue;
 281                 found = ipv6_addr_cmp(&pac->acl_addr, addr) == 0;
 282                 if (found)
 283                         break;
 284         }
 285         read_unlock(&ipv6_sk_ac_lock);
 286
 287         return found;
 288 }
 289
 290 #endif
 291
 292 static void aca_put(struct ifacaddr6 *ac)
 293 {
 294         if (atomic_dec_and_test(&ac->aca_refcnt)) {
 295                 in6_dev_put(ac->aca_idev);
 296                 kfree(ac);
 297         }
 298 }
 299
 300 /*
 301  *      device anycast group inc (add if not found)
 302  */
 303 int ipv6_dev_ac_inc(struct net_device *dev, struct in6_addr *addr)
 304 {
 305         struct ifacaddr6 *aca;
 306         struct inet6_dev *idev;
 307
 308         idev = in6_dev_get(dev);
 309
 310         if (idev == NULL)
 311                 return -EINVAL;
 312
 313         write_lock_bh(&idev->lock);
 314         if (idev->dead) {
 315                 write_unlock_bh(&idev->lock);
 316                 in6_dev_put(idev);
 317                 return -ENODEV;
 318         }
 319
 320         for (aca = idev->ac_list; aca; aca = aca->aca_next) {
 321                 if (ipv6_addr_cmp(&aca->aca_addr, addr) == 0) {
 322                         aca->aca_users++;
 323                         write_unlock_bh(&idev->lock);
 324                         in6_dev_put(idev);
 325                         return 0;
 326                 }
 327         }
 328
 329         /*
 330          *      not found: create a new one.
 331          */
 332
 333         aca = kmalloc(sizeof(struct ifacaddr6), GFP_ATOMIC);
 334
 335         if (aca == NULL) {
 336                 write_unlock_bh(&idev->lock);
 337                 in6_dev_put(idev);
 338                 return -ENOMEM;
 339         }
 340
 341         memset(aca, 0, sizeof(struct ifacaddr6));
 342
 343         ipv6_addr_copy(&aca->aca_addr, addr);
 344         aca->aca_idev = idev;
 345         aca->aca_users = 1;
 346         /* aca_tstamp should be updated upon changes */
 347         aca->aca_cstamp = aca->aca_tstamp = jiffies;
 348         atomic_set(&aca->aca_refcnt, 2);
 349         aca->aca_lock = SPIN_LOCK_UNLOCKED;
 350
 351         aca->aca_next = idev->ac_list;
 352         idev->ac_list = aca;
 353         write_unlock_bh(&idev->lock);
 354
 355         ip6_rt_addr_add(&aca->aca_addr, dev, 1);
 356
 357         addrconf_join_solict(dev, &aca->aca_addr);
 358
 359         aca_put(aca);
 360         return 0;
 361 }
 362
 363 /*
 364  *      device anycast group decrement
 365  */
 366 int ipv6_dev_ac_dec(struct net_device *dev, struct in6_addr *addr)
 367 {
 368         struct inet6_dev *idev;
 369         struct ifacaddr6 *aca, *prev_aca;
 370
 371         idev = in6_dev_get(dev);
 372         if (idev == NULL)
 373                 return -ENODEV;
 374
 375         write_lock_bh(&idev->lock);
 376         prev_aca = 0;
 377         for (aca = idev->ac_list; aca; aca = aca->aca_next) {
 378                 if (ipv6_addr_cmp(&aca->aca_addr, addr) == 0)
 379                         break;
 380                 prev_aca = aca;
 381         }
 382         if (!aca) {
 383                 write_unlock_bh(&idev->lock);
 384                 in6_dev_put(idev);
 385                 return -ENOENT;
 386         }
 387         if (--aca->aca_users > 0) {
 388                 write_unlock_bh(&idev->lock);
 389                 in6_dev_put(idev);
 390                 return 0;
 391         }
 392         if (prev_aca)
 393                 prev_aca->aca_next = aca->aca_next;
 394         else
 395                 idev->ac_list = aca->aca_next;
 396         write_unlock_bh(&idev->lock);
 397         addrconf_leave_solict(dev, &aca->aca_addr);
 398
 399         ip6_rt_addr_del(&aca->aca_addr, dev);
 400
 401         aca_put(aca);
 402         in6_dev_put(idev);
 403         return 0;
 404 }
 405
 406 /*
 407  *      check if the interface has this anycast address
 408  */
 409 static int ipv6_chk_acast_dev(struct net_device *dev, struct in6_addr *addr)
 410 {
 411         struct inet6_dev *idev;
 412         struct ifacaddr6 *aca;
 413
 414         idev = in6_dev_get(dev);
 415         if (idev) {
 416                 read_lock_bh(&idev->lock);
 417                 for (aca = idev->ac_list; aca; aca = aca->aca_next)
 418                         if (ipv6_addr_cmp(&aca->aca_addr, addr) == 0)
 419                                 break;
 420                 read_unlock_bh(&idev->lock);
 421                 in6_dev_put(idev);
 422                 return aca != 0;
 423         }
 424         return 0;
 425 }
 426
 427 /*
 428  *      check if given interface (or any, if dev==0) has this anycast address
 429  */
 430 int ipv6_chk_acast_addr(struct net_device *dev, struct in6_addr *addr)
 431 {
 432         if (dev)
 433                 return ipv6_chk_acast_dev(dev, addr);
 434         read_lock(&dev_base_lock);
 435         for (dev=dev_base; dev; dev=dev->next)
 436                 if (ipv6_chk_acast_dev(dev, addr))
 437                         break;
 438         read_unlock(&dev_base_lock);
 439         return dev != 0;
 440 }
 441
 442
 443 #ifdef CONFIG_PROC_FS
 444 struct ac6_iter_state {
 445         struct net_device *dev;
 446         struct inet6_dev *idev;
 447 };
 448
 449 #define ac6_seq_private(seq)    ((struct ac6_iter_state *)(seq)->private)
 450
 451 static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq)
 452 {
 453         struct ifacaddr6 *im = NULL;
 454         struct ac6_iter_state *state = ac6_seq_private(seq);
 455
 456         for (state->dev = dev_base, state->idev = NULL;
 457              state->dev;
 458              state->dev = state->dev->next) {
 459                 struct inet6_dev *idev;
 460                 idev = in6_dev_get(state->dev);
 461                 if (!idev)
 462                         continue;
 463                 read_lock_bh(&idev->lock);
 464                 im = idev->ac_list;
 465                 if (im) {
 466                         state->idev = idev;
 467                         break;
 468                 }
 469                 read_unlock_bh(&idev->lock);
 470         }
 471         return im;
 472 }
 473
 474 static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im)
 475 {
 476         struct ac6_iter_state *state = ac6_seq_private(seq);
 477
 478         im = im->aca_next;
 479         while (!im) {
 480                 if (likely(state->idev != NULL)) {
 481                         read_unlock_bh(&state->idev->lock);
 482                         in6_dev_put(state->idev);
 483                 }
 484                 state->dev = state->dev->next;
 485                 if (!state->dev) {
 486                         state->idev = NULL;
 487                         break;
 488                 }
 489                 state->idev = in6_dev_get(state->dev);
 490                 if (!state->idev)
 491                         continue;
 492                 read_lock_bh(&state->idev->lock);
 493                 im = state->idev->ac_list;
 494         }
 495         return im;
 496 }
 497
 498 static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos)
 499 {
 500         struct ifacaddr6 *im = ac6_get_first(seq);
 501         if (im)
 502                 while (pos && (im = ac6_get_next(seq, im)) != NULL)
 503                         --pos;
 504         return pos ? NULL : im;
 505 }
 506
 507 static void *ac6_seq_start(struct seq_file *seq, loff_t *pos)
 508 {
 509         read_lock(&dev_base_lock);
 510         return ac6_get_idx(seq, *pos);
 511 }
 512
 513 static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 514 {
 515         struct ifacaddr6 *im;
 516         im = ac6_get_next(seq, v);
 517         ++*pos;
 518         return im;
 519 }
 520
 521 static void ac6_seq_stop(struct seq_file *seq, void *v)
 522 {
 523         struct ac6_iter_state *state = ac6_seq_private(seq);
 524         if (likely(state->idev != NULL)) {
 525                 read_unlock_bh(&state->idev->lock);
 526                 in6_dev_put(state->idev);
 527         }
 528         read_unlock(&dev_base_lock);
 529 }
 530
 531 static int ac6_seq_show(struct seq_file *seq, void *v)
 532 {
 533         struct ifacaddr6 *im = (struct ifacaddr6 *)v;
 534         struct ac6_iter_state *state = ac6_seq_private(seq);
 535
 536         seq_printf(seq,
 537                    "%-4d %-15s "
 538                    "%04x%04x%04x%04x%04x%04x%04x%04x "
 539                    "%5d\n",
 540                    state->dev->ifindex, state->dev->name,
 541                    NIP6(im->aca_addr),
 542                    im->aca_users);
 543         return 0;
 544 }
 545
 546 static struct seq_operations ac6_seq_ops = {
 547         .start  =       ac6_seq_start,
 548         .next   =       ac6_seq_next,
 549         .stop   =       ac6_seq_stop,
 550         .show   =       ac6_seq_show,
 551 };
 552
 553 static int ac6_seq_open(struct inode *inode, struct file *file)
 554 {
 555         struct seq_file *seq;
 556         int rc = -ENOMEM;
 557         struct ac6_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
 558
 559         if (!s)
 560                 goto out;
 561
 562         rc = seq_open(file, &ac6_seq_ops);
 563         if (rc)
 564                 goto out_kfree;
 565
 566         seq = file->private_data;
 567         seq->private = s;
 568         memset(s, 0, sizeof(*s));
 569 out:
 570         return rc;
 571 out_kfree:
 572         kfree(s);
 573         goto out;
 574 }
 575
 576 static struct file_operations ac6_seq_fops = {
 577         .owner          =       THIS_MODULE,
 578         .open           =       ac6_seq_open,
 579         .read           =       seq_read,
 580         .llseek         =       seq_lseek,
 581         .release        =       seq_release_private,
 582 };
 583
 584 int __init ac6_proc_init(void)
 585 {
 586         if (!proc_net_fops_create("anycast6", S_IRUGO, &ac6_seq_fops))
 587                 return -ENOMEM;
 588
 589         return 0;
 590 }
 591
 592 void ac6_proc_exit(void)
 593 {
 594         proc_net_remove("anycast6");
 595 }
 596 #endif
 597