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/capability.h>
  17 #include <linux/config.h>
  18 #include <linux/module.h>
  19 #include <linux/errno.h>
  20 #include <linux/types.h>
  21 #include <linux/random.h>
  22 #include <linux/string.h>
  23 #include <linux/socket.h>
  24 #include <linux/sockios.h>
  25 #include <linux/sched.h>
  26 #include <linux/net.h>
  27 #include <linux/in6.h>
  28 #include <linux/netdevice.h>
  29 #include <linux/if_arp.h>
  30 #include <linux/route.h>
  31 #include <linux/init.h>
  32 #include <linux/proc_fs.h>
  33 #include <linux/seq_file.h>
  34
  35 #include <net/sock.h>
  36 #include <net/snmp.h>
  37
  38 #include <net/ipv6.h>
  39 #include <net/protocol.h>
  40 #include <net/if_inet6.h>
  41 #include <net/ndisc.h>
  42 #include <net/addrconf.h>
  43 #include <net/ip6_route.h>
  44
  45 #include <net/checksum.h>
  46
  47 static int ipv6_dev_ac_dec(struct net_device *dev, struct in6_addr *addr);
  48
  49 /* Big ac list lock for all the sockets */
  50 static DEFINE_RWLOCK(ipv6_sk_ac_lock);
  51
  52 static int
  53 ip6_onlink(struct in6_addr *addr, struct net_device *dev)
  54 {
  55         struct inet6_dev        *idev;
  56         struct inet6_ifaddr     *ifa;
  57         int     onlink;
  58
  59         onlink = 0;
  60         read_lock(&addrconf_lock);
  61         idev = __in6_dev_get(dev);
  62         if (idev) {
  63                 read_lock_bh(&idev->lock);
  64                 for (ifa=idev->addr_list; ifa; ifa=ifa->if_next) {
  65                         onlink = ipv6_prefix_equal(addr, &ifa->addr,
  66                                                    ifa->prefix_len);
  67                         if (onlink)
  68                                 break;
  69                 }
  70                 read_unlock_bh(&idev->lock);
  71         }
  72         read_unlock(&addrconf_lock);
  73         return onlink;
  74 }
  75
  76 /*
  77  *      socket join an anycast group
  78  */
  79
  80 int ipv6_sock_ac_join(struct sock *sk, int ifindex, struct in6_addr *addr)
  81 {
  82         struct ipv6_pinfo *np = inet6_sk(sk);
  83         struct net_device *dev = NULL;
  84         struct inet6_dev *idev;
  85         struct ipv6_ac_socklist *pac;
  86         int     ishost = !ipv6_devconf.forwarding;
  87         int     err = 0;
  88
  89         if (!capable(CAP_NET_ADMIN))
  90                 return -EPERM;
  91         if (ipv6_addr_is_multicast(addr))
  92                 return -EINVAL;
  93         if (ipv6_chk_addr(addr, NULL, 0))
  94                 return -EINVAL;
  95
  96         pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL);
  97         if (pac == NULL)
  98                 return -ENOMEM;
  99         pac->acl_next = NULL;
 100         ipv6_addr_copy(&pac->acl_addr, addr);
 101
 102         if (ifindex == 0) {
 103                 struct rt6_info *rt;
 104
 105                 rt = rt6_lookup(addr, NULL, 0, 0);
 106                 if (rt) {
 107                         dev = rt->rt6i_dev;
 108                         dev_hold(dev);
 109                         dst_release(&rt->u.dst);
 110                 } else if (ishost) {
 111                         err = -EADDRNOTAVAIL;
 112                         goto out_free_pac;
 113                 } else {
 114                         /* router, no matching interface: just pick one */
 115
 116                         dev = dev_get_by_flags(IFF_UP, IFF_UP|IFF_LOOPBACK);
 117                 }
 118         } else
 119                 dev = dev_get_by_index(ifindex);
 120
 121         if (dev == NULL) {
 122                 err = -ENODEV;
 123                 goto out_free_pac;
 124         }
 125
 126         idev = in6_dev_get(dev);
 127         if (!idev) {
 128                 if (ifindex)
 129                         err = -ENODEV;
 130                 else
 131                         err = -EADDRNOTAVAIL;
 132                 goto out_dev_put;
 133         }
 134         /* reset ishost, now that we have a specific device */
 135         ishost = !idev->cnf.forwarding;
 136         in6_dev_put(idev);
 137
 138         pac->acl_ifindex = dev->ifindex;
 139
 140         /* XXX
 141          * For hosts, allow link-local or matching prefix anycasts.
 142          * This obviates the need for propagating anycast routes while
 143          * still allowing some non-router anycast participation.
 144          */
 145         if (!ip6_onlink(addr, dev)) {
 146                 if (ishost)
 147                         err = -EADDRNOTAVAIL;
 148                 if (err)
 149                         goto out_dev_put;
 150         }
 151
 152         err = ipv6_dev_ac_inc(dev, addr);
 153         if (err)
 154                 goto out_dev_put;
 155
 156         write_lock_bh(&ipv6_sk_ac_lock);
 157         pac->acl_next = np->ipv6_ac_list;
 158         np->ipv6_ac_list = pac;
 159         write_unlock_bh(&ipv6_sk_ac_lock);
 160
 161         dev_put(dev);
 162
 163         return 0;
 164
 165 out_dev_put:
 166         dev_put(dev);
 167 out_free_pac:
 168         sock_kfree_s(sk, pac, sizeof(*pac));
 169         return err;
 170 }
 171
 172 /*
 173  *      socket leave an anycast group
 174  */
 175 int ipv6_sock_ac_drop(struct sock *sk, int ifindex, struct in6_addr *addr)
 176 {
 177         struct ipv6_pinfo *np = inet6_sk(sk);
 178         struct net_device *dev;
 179         struct ipv6_ac_socklist *pac, *prev_pac;
 180
 181         write_lock_bh(&ipv6_sk_ac_lock);
 182         prev_pac = NULL;
 183         for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) {
 184                 if ((ifindex == 0 || pac->acl_ifindex == ifindex) &&
 185                      ipv6_addr_equal(&pac->acl_addr, addr))
 186                         break;
 187                 prev_pac = pac;
 188         }
 189         if (!pac) {
 190                 write_unlock_bh(&ipv6_sk_ac_lock);
 191                 return -ENOENT;
 192         }
 193         if (prev_pac)
 194                 prev_pac->acl_next = pac->acl_next;
 195         else
 196                 np->ipv6_ac_list = pac->acl_next;
 197
 198         write_unlock_bh(&ipv6_sk_ac_lock);
 199
 200         dev = dev_get_by_index(pac->acl_ifindex);
 201         if (dev) {
 202                 ipv6_dev_ac_dec(dev, &pac->acl_addr);
 203                 dev_put(dev);
 204         }
 205         sock_kfree_s(sk, pac, sizeof(*pac));
 206         return 0;
 207 }
 208
 209 void ipv6_sock_ac_close(struct sock *sk)
 210 {
 211         struct ipv6_pinfo *np = inet6_sk(sk);
 212         struct net_device *dev = NULL;
 213         struct ipv6_ac_socklist *pac;
 214         int     prev_index;
 215
 216         write_lock_bh(&ipv6_sk_ac_lock);
 217         pac = np->ipv6_ac_list;
 218         np->ipv6_ac_list = NULL;
 219         write_unlock_bh(&ipv6_sk_ac_lock);
 220
 221         prev_index = 0;
 222         while (pac) {
 223                 struct ipv6_ac_socklist *next = pac->acl_next;
 224
 225                 if (pac->acl_ifindex != prev_index) {
 226                         if (dev)
 227                                 dev_put(dev);
 228                         dev = dev_get_by_index(pac->acl_ifindex);
 229                         prev_index = pac->acl_ifindex;
 230                 }
 231                 if (dev)
 232                         ipv6_dev_ac_dec(dev, &pac->acl_addr);
 233                 sock_kfree_s(sk, pac, sizeof(*pac));
 234                 pac = next;
 235         }
 236         if (dev)
 237                 dev_put(dev);
 238 }
 239
 240 #if 0
 241 /* The function is not used, which is funny. Apparently, author
 242  * supposed to use it to filter out datagrams inside udp/raw but forgot.
 243  *
 244  * It is OK, anycasts are not special comparing to delivery to unicasts.
 245  */
 246
 247 int inet6_ac_check(struct sock *sk, struct in6_addr *addr, int ifindex)
 248 {
 249         struct ipv6_ac_socklist *pac;
 250         struct ipv6_pinfo *np = inet6_sk(sk);
 251         int     found;
 252
 253         found = 0;
 254         read_lock(&ipv6_sk_ac_lock);
 255         for (pac=np->ipv6_ac_list; pac; pac=pac->acl_next) {
 256                 if (ifindex && pac->acl_ifindex != ifindex)
 257                         continue;
 258                 found = ipv6_addr_equal(&pac->acl_addr, addr);
 259                 if (found)
 260                         break;
 261         }
 262         read_unlock(&ipv6_sk_ac_lock);
 263
 264         return found;
 265 }
 266
 267 #endif
 268
 269 static void aca_put(struct ifacaddr6 *ac)
 270 {
 271         if (atomic_dec_and_test(&ac->aca_refcnt)) {
 272                 in6_dev_put(ac->aca_idev);
 273                 dst_release(&ac->aca_rt->u.dst);
 274                 kfree(ac);
 275         }
 276 }
 277
 278 /*
 279  *      device anycast group inc (add if not found)
 280  */
 281 int ipv6_dev_ac_inc(struct net_device *dev, struct in6_addr *addr)
 282 {
 283         struct ifacaddr6 *aca;
 284         struct inet6_dev *idev;
 285         struct rt6_info *rt;
 286         int err;
 287
 288         idev = in6_dev_get(dev);
 289
 290         if (idev == NULL)
 291                 return -EINVAL;
 292
 293         write_lock_bh(&idev->lock);
 294         if (idev->dead) {
 295                 err = -ENODEV;
 296                 goto out;
 297         }
 298
 299         for (aca = idev->ac_list; aca; aca = aca->aca_next) {
 300                 if (ipv6_addr_equal(&aca->aca_addr, addr)) {
 301                         aca->aca_users++;
 302                         err = 0;
 303                         goto out;
 304                 }
 305         }
 306
 307         /*
 308          *      not found: create a new one.
 309          */
 310
 311         aca = kmalloc(sizeof(struct ifacaddr6), GFP_ATOMIC);
 312
 313         if (aca == NULL) {
 314                 err = -ENOMEM;
 315                 goto out;
 316         }
 317
 318         rt = addrconf_dst_alloc(idev, addr, 1);
 319         if (IS_ERR(rt)) {
 320                 kfree(aca);
 321                 err = PTR_ERR(rt);
 322                 goto out;
 323         }
 324
 325         memset(aca, 0, sizeof(struct ifacaddr6));
 326
 327         ipv6_addr_copy(&aca->aca_addr, addr);
 328         aca->aca_idev = idev;
 329         aca->aca_rt = rt;
 330         aca->aca_users = 1;
 331         /* aca_tstamp should be updated upon changes */
 332         aca->aca_cstamp = aca->aca_tstamp = jiffies;
 333         atomic_set(&aca->aca_refcnt, 2);
 334         spin_lock_init(&aca->aca_lock);
 335
 336         aca->aca_next = idev->ac_list;
 337         idev->ac_list = aca;
 338         write_unlock_bh(&idev->lock);
 339
 340         dst_hold(&rt->u.dst);
 341         if (ip6_ins_rt(rt, NULL, NULL, NULL))
 342                 dst_release(&rt->u.dst);
 343
 344         addrconf_join_solict(dev, &aca->aca_addr);
 345
 346         aca_put(aca);
 347         return 0;
 348 out:
 349         write_unlock_bh(&idev->lock);
 350         in6_dev_put(idev);
 351         return err;
 352 }
 353
 354 /*
 355  *      device anycast group decrement
 356  */
 357 int __ipv6_dev_ac_dec(struct inet6_dev *idev, struct in6_addr *addr)
 358 {
 359         struct ifacaddr6 *aca, *prev_aca;
 360
 361         write_lock_bh(&idev->lock);
 362         prev_aca = NULL;
 363         for (aca = idev->ac_list; aca; aca = aca->aca_next) {
 364                 if (ipv6_addr_equal(&aca->aca_addr, addr))
 365                         break;
 366                 prev_aca = aca;
 367         }
 368         if (!aca) {
 369                 write_unlock_bh(&idev->lock);
 370                 return -ENOENT;
 371         }
 372         if (--aca->aca_users > 0) {
 373                 write_unlock_bh(&idev->lock);
 374                 return 0;
 375         }
 376         if (prev_aca)
 377                 prev_aca->aca_next = aca->aca_next;
 378         else
 379                 idev->ac_list = aca->aca_next;
 380         write_unlock_bh(&idev->lock);
 381         addrconf_leave_solict(idev, &aca->aca_addr);
 382
 383         dst_hold(&aca->aca_rt->u.dst);
 384         if (ip6_del_rt(aca->aca_rt, NULL, NULL, NULL))
 385                 dst_free(&aca->aca_rt->u.dst);
 386         else
 387                 dst_release(&aca->aca_rt->u.dst);
 388
 389         aca_put(aca);
 390         return 0;
 391 }
 392
 393 static int ipv6_dev_ac_dec(struct net_device *dev, struct in6_addr *addr)
 394 {
 395         int ret;
 396         struct inet6_dev *idev = in6_dev_get(dev);
 397         if (idev == NULL)
 398                 return -ENODEV;
 399         ret = __ipv6_dev_ac_dec(idev, addr);
 400         in6_dev_put(idev);
 401         return ret;
 402 }
 403
 404 /*
 405  *      check if the interface has this anycast address
 406  */
 407 static int ipv6_chk_acast_dev(struct net_device *dev, struct in6_addr *addr)
 408 {
 409         struct inet6_dev *idev;
 410         struct ifacaddr6 *aca;
 411
 412         idev = in6_dev_get(dev);
 413         if (idev) {
 414                 read_lock_bh(&idev->lock);
 415                 for (aca = idev->ac_list; aca; aca = aca->aca_next)
 416                         if (ipv6_addr_equal(&aca->aca_addr, addr))
 417                                 break;
 418                 read_unlock_bh(&idev->lock);
 419                 in6_dev_put(idev);
 420                 return aca != 0;
 421         }
 422         return 0;
 423 }
 424
 425 /*
 426  *      check if given interface (or any, if dev==0) has this anycast address
 427  */
 428 int ipv6_chk_acast_addr(struct net_device *dev, struct in6_addr *addr)
 429 {
 430         if (dev)
 431                 return ipv6_chk_acast_dev(dev, addr);
 432         read_lock(&dev_base_lock);
 433         for (dev=dev_base; dev; dev=dev->next)
 434                 if (ipv6_chk_acast_dev(dev, addr))
 435                         break;
 436         read_unlock(&dev_base_lock);
 437         return dev != 0;
 438 }
 439
 440
 441 #ifdef CONFIG_PROC_FS
 442 struct ac6_iter_state {
 443         struct net_device *dev;
 444         struct inet6_dev *idev;
 445 };
 446
 447 #define ac6_seq_private(seq)    ((struct ac6_iter_state *)(seq)->private)
 448
 449 static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq)
 450 {
 451         struct ifacaddr6 *im = NULL;
 452         struct ac6_iter_state *state = ac6_seq_private(seq);
 453
 454         for (state->dev = dev_base, state->idev = NULL;
 455              state->dev;
 456              state->dev = state->dev->next) {
 457                 struct inet6_dev *idev;
 458                 idev = in6_dev_get(state->dev);
 459                 if (!idev)
 460                         continue;
 461                 read_lock_bh(&idev->lock);
 462                 im = idev->ac_list;
 463                 if (im) {
 464                         state->idev = idev;
 465                         break;
 466                 }
 467                 read_unlock_bh(&idev->lock);
 468         }
 469         return im;
 470 }
 471
 472 static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im)
 473 {
 474         struct ac6_iter_state *state = ac6_seq_private(seq);
 475
 476         im = im->aca_next;
 477         while (!im) {
 478                 if (likely(state->idev != NULL)) {
 479                         read_unlock_bh(&state->idev->lock);
 480                         in6_dev_put(state->idev);
 481                 }
 482                 state->dev = state->dev->next;
 483                 if (!state->dev) {
 484                         state->idev = NULL;
 485                         break;
 486                 }
 487                 state->idev = in6_dev_get(state->dev);
 488                 if (!state->idev)
 489                         continue;
 490                 read_lock_bh(&state->idev->lock);
 491                 im = state->idev->ac_list;
 492         }
 493         return im;
 494 }
 495
 496 static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos)
 497 {
 498         struct ifacaddr6 *im = ac6_get_first(seq);
 499         if (im)
 500                 while (pos && (im = ac6_get_next(seq, im)) != NULL)
 501                         --pos;
 502         return pos ? NULL : im;
 503 }
 504
 505 static void *ac6_seq_start(struct seq_file *seq, loff_t *pos)
 506 {
 507         read_lock(&dev_base_lock);
 508         return ac6_get_idx(seq, *pos);
 509 }
 510
 511 static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 512 {
 513         struct ifacaddr6 *im;
 514         im = ac6_get_next(seq, v);
 515         ++*pos;
 516         return im;
 517 }
 518
 519 static void ac6_seq_stop(struct seq_file *seq, void *v)
 520 {
 521         struct ac6_iter_state *state = ac6_seq_private(seq);
 522         if (likely(state->idev != NULL)) {
 523                 read_unlock_bh(&state->idev->lock);
 524                 in6_dev_put(state->idev);
 525         }
 526         read_unlock(&dev_base_lock);
 527 }
 528
 529 static int ac6_seq_show(struct seq_file *seq, void *v)
 530 {
 531         struct ifacaddr6 *im = (struct ifacaddr6 *)v;
 532         struct ac6_iter_state *state = ac6_seq_private(seq);
 533
 534         seq_printf(seq,
 535                    "%-4d %-15s " NIP6_SEQFMT " %5d\n",
 536                    state->dev->ifindex, state->dev->name,
 537                    NIP6(im->aca_addr),
 538                    im->aca_users);
 539         return 0;
 540 }
 541
 542 static struct seq_operations ac6_seq_ops = {
 543         .start  =       ac6_seq_start,
 544         .next   =       ac6_seq_next,
 545         .stop   =       ac6_seq_stop,
 546         .show   =       ac6_seq_show,
 547 };
 548
 549 static int ac6_seq_open(struct inode *inode, struct file *file)
 550 {
 551         struct seq_file *seq;
 552         int rc = -ENOMEM;
 553         struct ac6_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
 554
 555         if (!s)
 556                 goto out;
 557
 558         rc = seq_open(file, &ac6_seq_ops);
 559         if (rc)
 560                 goto out_kfree;
 561
 562         seq = file->private_data;
 563         seq->private = s;
 564         memset(s, 0, sizeof(*s));
 565 out:
 566         return rc;
 567 out_kfree:
 568         kfree(s);
 569         goto out;
 570 }
 571
 572 static struct file_operations ac6_seq_fops = {
 573         .owner          =       THIS_MODULE,
 574         .open           =       ac6_seq_open,
 575         .read           =       seq_read,
 576         .llseek         =       seq_lseek,
 577         .release        =       seq_release_private,
 578 };
 579
 580 int __init ac6_proc_init(void)
 581 {
 582         if (!proc_net_fops_create("anycast6", S_IRUGO, &ac6_seq_fops))
 583                 return -ENOMEM;
 584
 585         return 0;
 586 }
 587
 588 void ac6_proc_exit(void)
 589 {
 590         proc_net_remove("anycast6");
 591 }
 592 #endif
 593