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/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/net.h>
  25 #include <linux/in6.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/if_arp.h>
  28 #include <linux/route.h>
  29 #include <linux/init.h>
  30 #include <linux/proc_fs.h>
  31 #include <linux/seq_file.h>
  32
  33 #include <net/net_namespace.h>
  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 static int ipv6_dev_ac_dec(struct net_device *dev, struct in6_addr *addr);
  47
  48 /* Big ac list lock for all the sockets */
  49 static DEFINE_RWLOCK(ipv6_sk_ac_lock);
  50
  51 static int
  52 ip6_onlink(struct in6_addr *addr, struct net_device *dev)
  53 {
  54         struct inet6_dev        *idev;
  55         struct inet6_ifaddr     *ifa;
  56         int     onlink;
  57
  58         onlink = 0;
  59         rcu_read_lock();
  60         idev = __in6_dev_get(dev);
  61         if (idev) {
  62                 read_lock_bh(&idev->lock);
  63                 for (ifa=idev->addr_list; ifa; ifa=ifa->if_next) {
  64                         onlink = ipv6_prefix_equal(addr, &ifa->addr,
  65                                                    ifa->prefix_len);
  66                         if (onlink)
  67                                 break;
  68                 }
  69                 read_unlock_bh(&idev->lock);
  70         }
  71         rcu_read_unlock();
  72         return onlink;
  73 }
  74
  75 /*
  76  *      socket join an anycast group
  77  */
  78
  79 int ipv6_sock_ac_join(struct sock *sk, int ifindex, struct in6_addr *addr)
  80 {
  81         struct ipv6_pinfo *np = inet6_sk(sk);
  82         struct net_device *dev = NULL;
  83         struct inet6_dev *idev;
  84         struct ipv6_ac_socklist *pac;
  85         struct net *net = sock_net(sk);
  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(net, 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(net, 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(net, IFF_UP, IFF_UP|IFF_LOOPBACK);
 117                 }
 118         } else
 119                 dev = dev_get_by_index(net, 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         struct net *net = sock_net(sk);
 181
 182         write_lock_bh(&ipv6_sk_ac_lock);
 183         prev_pac = NULL;
 184         for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) {
 185                 if ((ifindex == 0 || pac->acl_ifindex == ifindex) &&
 186                      ipv6_addr_equal(&pac->acl_addr, addr))
 187                         break;
 188                 prev_pac = pac;
 189         }
 190         if (!pac) {
 191                 write_unlock_bh(&ipv6_sk_ac_lock);
 192                 return -ENOENT;
 193         }
 194         if (prev_pac)
 195                 prev_pac->acl_next = pac->acl_next;
 196         else
 197                 np->ipv6_ac_list = pac->acl_next;
 198
 199         write_unlock_bh(&ipv6_sk_ac_lock);
 200
 201         dev = dev_get_by_index(net, pac->acl_ifindex);
 202         if (dev) {
 203                 ipv6_dev_ac_dec(dev, &pac->acl_addr);
 204                 dev_put(dev);
 205         }
 206         sock_kfree_s(sk, pac, sizeof(*pac));
 207         return 0;
 208 }
 209
 210 void ipv6_sock_ac_close(struct sock *sk)
 211 {
 212         struct ipv6_pinfo *np = inet6_sk(sk);
 213         struct net_device *dev = NULL;
 214         struct ipv6_ac_socklist *pac;
 215         struct net *net = sock_net(sk);
 216         int     prev_index;
 217
 218         write_lock_bh(&ipv6_sk_ac_lock);
 219         pac = np->ipv6_ac_list;
 220         np->ipv6_ac_list = NULL;
 221         write_unlock_bh(&ipv6_sk_ac_lock);
 222
 223         prev_index = 0;
 224         while (pac) {
 225                 struct ipv6_ac_socklist *next = pac->acl_next;
 226
 227                 if (pac->acl_ifindex != prev_index) {
 228                         if (dev)
 229                                 dev_put(dev);
 230                         dev = dev_get_by_index(net, pac->acl_ifindex);
 231                         prev_index = pac->acl_ifindex;
 232                 }
 233                 if (dev)
 234                         ipv6_dev_ac_dec(dev, &pac->acl_addr);
 235                 sock_kfree_s(sk, pac, sizeof(*pac));
 236                 pac = next;
 237         }
 238         if (dev)
 239                 dev_put(dev);
 240 }
 241
 242 #if 0
 243 /* The function is not used, which is funny. Apparently, author
 244  * supposed to use it to filter out datagrams inside udp/raw but forgot.
 245  *
 246  * It is OK, anycasts are not special comparing to delivery to unicasts.
 247  */
 248
 249 int inet6_ac_check(struct sock *sk, struct in6_addr *addr, int ifindex)
 250 {
 251         struct ipv6_ac_socklist *pac;
 252         struct ipv6_pinfo *np = inet6_sk(sk);
 253         int     found;
 254
 255         found = 0;
 256         read_lock(&ipv6_sk_ac_lock);
 257         for (pac=np->ipv6_ac_list; pac; pac=pac->acl_next) {
 258                 if (ifindex && pac->acl_ifindex != ifindex)
 259                         continue;
 260                 found = ipv6_addr_equal(&pac->acl_addr, addr);
 261                 if (found)
 262                         break;
 263         }
 264         read_unlock(&ipv6_sk_ac_lock);
 265
 266         return found;
 267 }
 268
 269 #endif
 270
 271 static void aca_put(struct ifacaddr6 *ac)
 272 {
 273         if (atomic_dec_and_test(&ac->aca_refcnt)) {
 274                 in6_dev_put(ac->aca_idev);
 275                 dst_release(&ac->aca_rt->u.dst);
 276                 kfree(ac);
 277         }
 278 }
 279
 280 /*
 281  *      device anycast group inc (add if not found)
 282  */
 283 int ipv6_dev_ac_inc(struct net_device *dev, struct in6_addr *addr)
 284 {
 285         struct ifacaddr6 *aca;
 286         struct inet6_dev *idev;
 287         struct rt6_info *rt;
 288         int err;
 289
 290         idev = in6_dev_get(dev);
 291
 292         if (idev == NULL)
 293                 return -EINVAL;
 294
 295         write_lock_bh(&idev->lock);
 296         if (idev->dead) {
 297                 err = -ENODEV;
 298                 goto out;
 299         }
 300
 301         for (aca = idev->ac_list; aca; aca = aca->aca_next) {
 302                 if (ipv6_addr_equal(&aca->aca_addr, addr)) {
 303                         aca->aca_users++;
 304                         err = 0;
 305                         goto out;
 306                 }
 307         }
 308
 309         /*
 310          *      not found: create a new one.
 311          */
 312
 313         aca = kzalloc(sizeof(struct ifacaddr6), GFP_ATOMIC);
 314
 315         if (aca == NULL) {
 316                 err = -ENOMEM;
 317                 goto out;
 318         }
 319
 320         rt = addrconf_dst_alloc(idev, addr, 1);
 321         if (IS_ERR(rt)) {
 322                 kfree(aca);
 323                 err = PTR_ERR(rt);
 324                 goto out;
 325         }
 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))
 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))
 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 != NULL;
 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 *net, struct net_device *dev,
 429                         struct in6_addr *addr)
 430 {
 431         int found = 0;
 432
 433         if (dev)
 434                 return ipv6_chk_acast_dev(dev, addr);
 435         read_lock(&dev_base_lock);
 436         for_each_netdev(net, dev)
 437                 if (ipv6_chk_acast_dev(dev, addr)) {
 438                         found = 1;
 439                         break;
 440                 }
 441         read_unlock(&dev_base_lock);
 442         return found;
 443 }
 444
 445
 446 #ifdef CONFIG_PROC_FS
 447 struct ac6_iter_state {
 448         struct seq_net_private p;
 449         struct net_device *dev;
 450         struct inet6_dev *idev;
 451 };
 452
 453 #define ac6_seq_private(seq)    ((struct ac6_iter_state *)(seq)->private)
 454
 455 static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq)
 456 {
 457         struct ifacaddr6 *im = NULL;
 458         struct ac6_iter_state *state = ac6_seq_private(seq);
 459         struct net *net = seq_file_net(seq);
 460
 461         state->idev = NULL;
 462         for_each_netdev(net, state->dev) {
 463                 struct inet6_dev *idev;
 464                 idev = in6_dev_get(state->dev);
 465                 if (!idev)
 466                         continue;
 467                 read_lock_bh(&idev->lock);
 468                 im = idev->ac_list;
 469                 if (im) {
 470                         state->idev = idev;
 471                         break;
 472                 }
 473                 read_unlock_bh(&idev->lock);
 474                 in6_dev_put(idev);
 475         }
 476         return im;
 477 }
 478
 479 static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im)
 480 {
 481         struct ac6_iter_state *state = ac6_seq_private(seq);
 482
 483         im = im->aca_next;
 484         while (!im) {
 485                 if (likely(state->idev != NULL)) {
 486                         read_unlock_bh(&state->idev->lock);
 487                         in6_dev_put(state->idev);
 488                 }
 489                 state->dev = next_net_device(state->dev);
 490                 if (!state->dev) {
 491                         state->idev = NULL;
 492                         break;
 493                 }
 494                 state->idev = in6_dev_get(state->dev);
 495                 if (!state->idev)
 496                         continue;
 497                 read_lock_bh(&state->idev->lock);
 498                 im = state->idev->ac_list;
 499         }
 500         return im;
 501 }
 502
 503 static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos)
 504 {
 505         struct ifacaddr6 *im = ac6_get_first(seq);
 506         if (im)
 507                 while (pos && (im = ac6_get_next(seq, im)) != NULL)
 508                         --pos;
 509         return pos ? NULL : im;
 510 }
 511
 512 static void *ac6_seq_start(struct seq_file *seq, loff_t *pos)
 513         __acquires(dev_base_lock)
 514 {
 515         read_lock(&dev_base_lock);
 516         return ac6_get_idx(seq, *pos);
 517 }
 518
 519 static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 520 {
 521         struct ifacaddr6 *im;
 522         im = ac6_get_next(seq, v);
 523         ++*pos;
 524         return im;
 525 }
 526
 527 static void ac6_seq_stop(struct seq_file *seq, void *v)
 528         __releases(dev_base_lock)
 529 {
 530         struct ac6_iter_state *state = ac6_seq_private(seq);
 531         if (likely(state->idev != NULL)) {
 532                 read_unlock_bh(&state->idev->lock);
 533                 in6_dev_put(state->idev);
 534         }
 535         read_unlock(&dev_base_lock);
 536 }
 537
 538 static int ac6_seq_show(struct seq_file *seq, void *v)
 539 {
 540         struct ifacaddr6 *im = (struct ifacaddr6 *)v;
 541         struct ac6_iter_state *state = ac6_seq_private(seq);
 542
 543         seq_printf(seq,
 544                    "%-4d %-15s " NIP6_SEQFMT " %5d\n",
 545                    state->dev->ifindex, state->dev->name,
 546                    NIP6(im->aca_addr),
 547                    im->aca_users);
 548         return 0;
 549 }
 550
 551 static const struct seq_operations ac6_seq_ops = {
 552         .start  =       ac6_seq_start,
 553         .next   =       ac6_seq_next,
 554         .stop   =       ac6_seq_stop,
 555         .show   =       ac6_seq_show,
 556 };
 557
 558 static int ac6_seq_open(struct inode *inode, struct file *file)
 559 {
 560         return seq_open_net(inode, file, &ac6_seq_ops,
 561                             sizeof(struct ac6_iter_state));
 562 }
 563
 564 static const struct file_operations ac6_seq_fops = {
 565         .owner          =       THIS_MODULE,
 566         .open           =       ac6_seq_open,
 567         .read           =       seq_read,
 568         .llseek         =       seq_lseek,
 569         .release        =       seq_release_net,
 570 };
 571
 572 int ac6_proc_init(struct net *net)
 573 {
 574         if (!proc_net_fops_create(net, "anycast6", S_IRUGO, &ac6_seq_fops))
 575                 return -ENOMEM;
 576
 577         return 0;
 578 }
 579
 580 void ac6_proc_exit(struct net *net)
 581 {
 582         proc_net_remove(net, "anycast6");
 583 }
 584 #endif
 585