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 #include <linux/slab.h>
  33
  34 #include <net/net_namespace.h>
  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, const struct in6_addr *addr);
  48
  49 /* Big ac list lock for all the sockets */
  50 static DEFINE_SPINLOCK(ipv6_sk_ac_lock);
  51
  52
  53 /*
  54  *      socket join an anycast group
  55  */
  56
  57 int ipv6_sock_ac_join(struct sock *sk, int ifindex, const struct in6_addr *addr)
  58 {
  59         struct ipv6_pinfo *np = inet6_sk(sk);
  60         struct net_device *dev = NULL;
  61         struct inet6_dev *idev;
  62         struct ipv6_ac_socklist *pac;
  63         struct net *net = sock_net(sk);
  64         int     ishost = !net->ipv6.devconf_all->forwarding;
  65         int     err = 0;
  66
  67         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
  68                 return -EPERM;
  69         if (ipv6_addr_is_multicast(addr))
  70                 return -EINVAL;
  71         if (ipv6_chk_addr(net, addr, NULL, 0))
  72                 return -EINVAL;
  73
  74         pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL);
  75         if (pac == NULL)
  76                 return -ENOMEM;
  77         pac->acl_next = NULL;
  78         pac->acl_addr = *addr;
  79
  80         rcu_read_lock();
  81         if (ifindex == 0) {
  82                 struct rt6_info *rt;
  83
  84                 rt = rt6_lookup(net, addr, NULL, 0, 0);
  85                 if (rt) {
  86                         dev = rt->dst.dev;
  87                         ip6_rt_put(rt);
  88                 } else if (ishost) {
  89                         err = -EADDRNOTAVAIL;
  90                         goto error;
  91                 } else {
  92                         /* router, no matching interface: just pick one */
  93                         dev = dev_get_by_flags_rcu(net, IFF_UP,
  94                                                    IFF_UP | IFF_LOOPBACK);
  95                 }
  96         } else
  97                 dev = dev_get_by_index_rcu(net, ifindex);
  98
  99         if (dev == NULL) {
 100                 err = -ENODEV;
 101                 goto error;
 102         }
 103
 104         idev = __in6_dev_get(dev);
 105         if (!idev) {
 106                 if (ifindex)
 107                         err = -ENODEV;
 108                 else
 109                         err = -EADDRNOTAVAIL;
 110                 goto error;
 111         }
 112         /* reset ishost, now that we have a specific device */
 113         ishost = !idev->cnf.forwarding;
 114
 115         pac->acl_ifindex = dev->ifindex;
 116
 117         /* XXX
 118          * For hosts, allow link-local or matching prefix anycasts.
 119          * This obviates the need for propagating anycast routes while
 120          * still allowing some non-router anycast participation.
 121          */
 122         if (!ipv6_chk_prefix(addr, dev)) {
 123                 if (ishost)
 124                         err = -EADDRNOTAVAIL;
 125                 if (err)
 126                         goto error;
 127         }
 128
 129         err = ipv6_dev_ac_inc(dev, addr);
 130         if (!err) {
 131                 spin_lock_bh(&ipv6_sk_ac_lock);
 132                 pac->acl_next = np->ipv6_ac_list;
 133                 np->ipv6_ac_list = pac;
 134                 spin_unlock_bh(&ipv6_sk_ac_lock);
 135                 pac = NULL;
 136         }
 137
 138 error:
 139         rcu_read_unlock();
 140         if (pac)
 141                 sock_kfree_s(sk, pac, sizeof(*pac));
 142         return err;
 143 }
 144
 145 /*
 146  *      socket leave an anycast group
 147  */
 148 int ipv6_sock_ac_drop(struct sock *sk, int ifindex, const struct in6_addr *addr)
 149 {
 150         struct ipv6_pinfo *np = inet6_sk(sk);
 151         struct net_device *dev;
 152         struct ipv6_ac_socklist *pac, *prev_pac;
 153         struct net *net = sock_net(sk);
 154
 155         spin_lock_bh(&ipv6_sk_ac_lock);
 156         prev_pac = NULL;
 157         for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) {
 158                 if ((ifindex == 0 || pac->acl_ifindex == ifindex) &&
 159                      ipv6_addr_equal(&pac->acl_addr, addr))
 160                         break;
 161                 prev_pac = pac;
 162         }
 163         if (!pac) {
 164                 spin_unlock_bh(&ipv6_sk_ac_lock);
 165                 return -ENOENT;
 166         }
 167         if (prev_pac)
 168                 prev_pac->acl_next = pac->acl_next;
 169         else
 170                 np->ipv6_ac_list = pac->acl_next;
 171
 172         spin_unlock_bh(&ipv6_sk_ac_lock);
 173
 174         rcu_read_lock();
 175         dev = dev_get_by_index_rcu(net, pac->acl_ifindex);
 176         if (dev)
 177                 ipv6_dev_ac_dec(dev, &pac->acl_addr);
 178         rcu_read_unlock();
 179
 180         sock_kfree_s(sk, pac, sizeof(*pac));
 181         return 0;
 182 }
 183
 184 void ipv6_sock_ac_close(struct sock *sk)
 185 {
 186         struct ipv6_pinfo *np = inet6_sk(sk);
 187         struct net_device *dev = NULL;
 188         struct ipv6_ac_socklist *pac;
 189         struct net *net = sock_net(sk);
 190         int     prev_index;
 191
 192         if (!np->ipv6_ac_list)
 193                 return;
 194
 195         spin_lock_bh(&ipv6_sk_ac_lock);
 196         pac = np->ipv6_ac_list;
 197         np->ipv6_ac_list = NULL;
 198         spin_unlock_bh(&ipv6_sk_ac_lock);
 199
 200         prev_index = 0;
 201         rcu_read_lock();
 202         while (pac) {
 203                 struct ipv6_ac_socklist *next = pac->acl_next;
 204
 205                 if (pac->acl_ifindex != prev_index) {
 206                         dev = dev_get_by_index_rcu(net, pac->acl_ifindex);
 207                         prev_index = pac->acl_ifindex;
 208                 }
 209                 if (dev)
 210                         ipv6_dev_ac_dec(dev, &pac->acl_addr);
 211                 sock_kfree_s(sk, pac, sizeof(*pac));
 212                 pac = next;
 213         }
 214         rcu_read_unlock();
 215 }
 216
 217 static void aca_put(struct ifacaddr6 *ac)
 218 {
 219         if (atomic_dec_and_test(&ac->aca_refcnt)) {
 220                 in6_dev_put(ac->aca_idev);
 221                 dst_release(&ac->aca_rt->dst);
 222                 kfree(ac);
 223         }
 224 }
 225
 226 /*
 227  *      device anycast group inc (add if not found)
 228  */
 229 int ipv6_dev_ac_inc(struct net_device *dev, const struct in6_addr *addr)
 230 {
 231         struct ifacaddr6 *aca;
 232         struct inet6_dev *idev;
 233         struct rt6_info *rt;
 234         int err;
 235
 236         idev = in6_dev_get(dev);
 237
 238         if (idev == NULL)
 239                 return -EINVAL;
 240
 241         write_lock_bh(&idev->lock);
 242         if (idev->dead) {
 243                 err = -ENODEV;
 244                 goto out;
 245         }
 246
 247         for (aca = idev->ac_list; aca; aca = aca->aca_next) {
 248                 if (ipv6_addr_equal(&aca->aca_addr, addr)) {
 249                         aca->aca_users++;
 250                         err = 0;
 251                         goto out;
 252                 }
 253         }
 254
 255         /*
 256          *      not found: create a new one.
 257          */
 258
 259         aca = kzalloc(sizeof(struct ifacaddr6), GFP_ATOMIC);
 260
 261         if (aca == NULL) {
 262                 err = -ENOMEM;
 263                 goto out;
 264         }
 265
 266         rt = addrconf_dst_alloc(idev, addr, true);
 267         if (IS_ERR(rt)) {
 268                 kfree(aca);
 269                 err = PTR_ERR(rt);
 270                 goto out;
 271         }
 272
 273         aca->aca_addr = *addr;
 274         aca->aca_idev = idev;
 275         aca->aca_rt = rt;
 276         aca->aca_users = 1;
 277         /* aca_tstamp should be updated upon changes */
 278         aca->aca_cstamp = aca->aca_tstamp = jiffies;
 279         atomic_set(&aca->aca_refcnt, 2);
 280         spin_lock_init(&aca->aca_lock);
 281
 282         aca->aca_next = idev->ac_list;
 283         idev->ac_list = aca;
 284         write_unlock_bh(&idev->lock);
 285
 286         ip6_ins_rt(rt);
 287
 288         addrconf_join_solict(dev, &aca->aca_addr);
 289
 290         aca_put(aca);
 291         return 0;
 292 out:
 293         write_unlock_bh(&idev->lock);
 294         in6_dev_put(idev);
 295         return err;
 296 }
 297
 298 /*
 299  *      device anycast group decrement
 300  */
 301 int __ipv6_dev_ac_dec(struct inet6_dev *idev, const struct in6_addr *addr)
 302 {
 303         struct ifacaddr6 *aca, *prev_aca;
 304
 305         write_lock_bh(&idev->lock);
 306         prev_aca = NULL;
 307         for (aca = idev->ac_list; aca; aca = aca->aca_next) {
 308                 if (ipv6_addr_equal(&aca->aca_addr, addr))
 309                         break;
 310                 prev_aca = aca;
 311         }
 312         if (!aca) {
 313                 write_unlock_bh(&idev->lock);
 314                 return -ENOENT;
 315         }
 316         if (--aca->aca_users > 0) {
 317                 write_unlock_bh(&idev->lock);
 318                 return 0;
 319         }
 320         if (prev_aca)
 321                 prev_aca->aca_next = aca->aca_next;
 322         else
 323                 idev->ac_list = aca->aca_next;
 324         write_unlock_bh(&idev->lock);
 325         addrconf_leave_solict(idev, &aca->aca_addr);
 326
 327         dst_hold(&aca->aca_rt->dst);
 328         ip6_del_rt(aca->aca_rt);
 329
 330         aca_put(aca);
 331         return 0;
 332 }
 333
 334 /* called with rcu_read_lock() */
 335 static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr)
 336 {
 337         struct inet6_dev *idev = __in6_dev_get(dev);
 338
 339         if (idev == NULL)
 340                 return -ENODEV;
 341         return __ipv6_dev_ac_dec(idev, addr);
 342 }
 343
 344 void ipv6_ac_destroy_dev(struct inet6_dev *idev)
 345 {
 346         struct ifacaddr6 *aca;
 347
 348         write_lock_bh(&idev->lock);
 349         while ((aca = idev->ac_list) != NULL) {
 350                 idev->ac_list = aca->aca_next;
 351                 write_unlock_bh(&idev->lock);
 352
 353                 addrconf_leave_solict(idev, &aca->aca_addr);
 354
 355                 dst_hold(&aca->aca_rt->dst);
 356                 ip6_del_rt(aca->aca_rt);
 357
 358                 aca_put(aca);
 359
 360                 write_lock_bh(&idev->lock);
 361         }
 362         write_unlock_bh(&idev->lock);
 363 }
 364
 365 /*
 366  *      check if the interface has this anycast address
 367  *      called with rcu_read_lock()
 368  */
 369 static bool ipv6_chk_acast_dev(struct net_device *dev, const struct in6_addr *addr)
 370 {
 371         struct inet6_dev *idev;
 372         struct ifacaddr6 *aca;
 373
 374         idev = __in6_dev_get(dev);
 375         if (idev) {
 376                 read_lock_bh(&idev->lock);
 377                 for (aca = idev->ac_list; aca; aca = aca->aca_next)
 378                         if (ipv6_addr_equal(&aca->aca_addr, addr))
 379                                 break;
 380                 read_unlock_bh(&idev->lock);
 381                 return aca != NULL;
 382         }
 383         return false;
 384 }
 385
 386 /*
 387  *      check if given interface (or any, if dev==0) has this anycast address
 388  */
 389 bool ipv6_chk_acast_addr(struct net *net, struct net_device *dev,
 390                          const struct in6_addr *addr)
 391 {
 392         bool found = false;
 393
 394         rcu_read_lock();
 395         if (dev)
 396                 found = ipv6_chk_acast_dev(dev, addr);
 397         else
 398                 for_each_netdev_rcu(net, dev)
 399                         if (ipv6_chk_acast_dev(dev, addr)) {
 400                                 found = true;
 401                                 break;
 402                         }
 403         rcu_read_unlock();
 404         return found;
 405 }
 406
 407
 408 #ifdef CONFIG_PROC_FS
 409 struct ac6_iter_state {
 410         struct seq_net_private p;
 411         struct net_device *dev;
 412         struct inet6_dev *idev;
 413 };
 414
 415 #define ac6_seq_private(seq)    ((struct ac6_iter_state *)(seq)->private)
 416
 417 static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq)
 418 {
 419         struct ifacaddr6 *im = NULL;
 420         struct ac6_iter_state *state = ac6_seq_private(seq);
 421         struct net *net = seq_file_net(seq);
 422
 423         state->idev = NULL;
 424         for_each_netdev_rcu(net, state->dev) {
 425                 struct inet6_dev *idev;
 426                 idev = __in6_dev_get(state->dev);
 427                 if (!idev)
 428                         continue;
 429                 read_lock_bh(&idev->lock);
 430                 im = idev->ac_list;
 431                 if (im) {
 432                         state->idev = idev;
 433                         break;
 434                 }
 435                 read_unlock_bh(&idev->lock);
 436         }
 437         return im;
 438 }
 439
 440 static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im)
 441 {
 442         struct ac6_iter_state *state = ac6_seq_private(seq);
 443
 444         im = im->aca_next;
 445         while (!im) {
 446                 if (likely(state->idev != NULL))
 447                         read_unlock_bh(&state->idev->lock);
 448
 449                 state->dev = next_net_device_rcu(state->dev);
 450                 if (!state->dev) {
 451                         state->idev = NULL;
 452                         break;
 453                 }
 454                 state->idev = __in6_dev_get(state->dev);
 455                 if (!state->idev)
 456                         continue;
 457                 read_lock_bh(&state->idev->lock);
 458                 im = state->idev->ac_list;
 459         }
 460         return im;
 461 }
 462
 463 static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos)
 464 {
 465         struct ifacaddr6 *im = ac6_get_first(seq);
 466         if (im)
 467                 while (pos && (im = ac6_get_next(seq, im)) != NULL)
 468                         --pos;
 469         return pos ? NULL : im;
 470 }
 471
 472 static void *ac6_seq_start(struct seq_file *seq, loff_t *pos)
 473         __acquires(RCU)
 474 {
 475         rcu_read_lock();
 476         return ac6_get_idx(seq, *pos);
 477 }
 478
 479 static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 480 {
 481         struct ifacaddr6 *im = ac6_get_next(seq, v);
 482
 483         ++*pos;
 484         return im;
 485 }
 486
 487 static void ac6_seq_stop(struct seq_file *seq, void *v)
 488         __releases(RCU)
 489 {
 490         struct ac6_iter_state *state = ac6_seq_private(seq);
 491
 492         if (likely(state->idev != NULL)) {
 493                 read_unlock_bh(&state->idev->lock);
 494                 state->idev = NULL;
 495         }
 496         rcu_read_unlock();
 497 }
 498
 499 static int ac6_seq_show(struct seq_file *seq, void *v)
 500 {
 501         struct ifacaddr6 *im = (struct ifacaddr6 *)v;
 502         struct ac6_iter_state *state = ac6_seq_private(seq);
 503
 504         seq_printf(seq, "%-4d %-15s %pi6 %5d\n",
 505                    state->dev->ifindex, state->dev->name,
 506                    &im->aca_addr, im->aca_users);
 507         return 0;
 508 }
 509
 510 static const struct seq_operations ac6_seq_ops = {
 511         .start  =       ac6_seq_start,
 512         .next   =       ac6_seq_next,
 513         .stop   =       ac6_seq_stop,
 514         .show   =       ac6_seq_show,
 515 };
 516
 517 static int ac6_seq_open(struct inode *inode, struct file *file)
 518 {
 519         return seq_open_net(inode, file, &ac6_seq_ops,
 520                             sizeof(struct ac6_iter_state));
 521 }
 522
 523 static const struct file_operations ac6_seq_fops = {
 524         .owner          =       THIS_MODULE,
 525         .open           =       ac6_seq_open,
 526         .read           =       seq_read,
 527         .llseek         =       seq_lseek,
 528         .release        =       seq_release_net,
 529 };
 530
 531 int __net_init ac6_proc_init(struct net *net)
 532 {
 533         if (!proc_create("anycast6", S_IRUGO, net->proc_net, &ac6_seq_fops))
 534                 return -ENOMEM;
 535
 536         return 0;
 537 }
 538
 539 void ac6_proc_exit(struct net *net)
 540 {
 541         remove_proc_entry("anycast6", net->proc_net);
 542 }
 543 #endif
 544