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, struct in6_addr *addr); 48 49 /* Big ac list lock for all the sockets */ 50 static DEFINE_RWLOCK(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, 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 (!capable(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 ipv6_addr_copy(&pac->acl_addr, addr); 79 80 if (ifindex == 0) { 81 struct rt6_info *rt; 82 83 rt = rt6_lookup(net, addr, NULL, 0, 0); 84 if (rt) { 85 dev = rt->rt6i_dev; 86 dev_hold(dev); 87 dst_release(&rt->u.dst); 88 } else if (ishost) { 89 err = -EADDRNOTAVAIL; 90 goto out_free_pac; 91 } else { 92 /* router, no matching interface: just pick one */ 93 94 dev = dev_get_by_flags(net, IFF_UP, IFF_UP|IFF_LOOPBACK); 95 } 96 } else 97 dev = dev_get_by_index(net, ifindex); 98 99 if (dev == NULL) { 100 err = -ENODEV; 101 goto out_free_pac; 102 } 103 104 idev = in6_dev_get(dev); 105 if (!idev) { 106 if (ifindex) 107 err = -ENODEV; 108 else 109 err = -EADDRNOTAVAIL; 110 goto out_dev_put; 111 } 112 /* reset ishost, now that we have a specific device */ 113 ishost = !idev->cnf.forwarding; 114 in6_dev_put(idev); 115 116 pac->acl_ifindex = dev->ifindex; 117 118 /* XXX 119 * For hosts, allow link-local or matching prefix anycasts. 120 * This obviates the need for propagating anycast routes while 121 * still allowing some non-router anycast participation. 122 */ 123 if (!ipv6_chk_prefix(addr, dev)) { 124 if (ishost) 125 err = -EADDRNOTAVAIL; 126 if (err) 127 goto out_dev_put; 128 } 129 130 err = ipv6_dev_ac_inc(dev, addr); 131 if (err) 132 goto out_dev_put; 133 134 write_lock_bh(&ipv6_sk_ac_lock); 135 pac->acl_next = np->ipv6_ac_list; 136 np->ipv6_ac_list = pac; 137 write_unlock_bh(&ipv6_sk_ac_lock); 138 139 dev_put(dev); 140 141 return 0; 142 143 out_dev_put: 144 dev_put(dev); 145 out_free_pac: 146 sock_kfree_s(sk, pac, sizeof(*pac)); 147 return err; 148 } 149 150 /* 151 * socket leave an anycast group 152 */ 153 int ipv6_sock_ac_drop(struct sock *sk, int ifindex, struct in6_addr *addr) 154 { 155 struct ipv6_pinfo *np = inet6_sk(sk); 156 struct net_device *dev; 157 struct ipv6_ac_socklist *pac, *prev_pac; 158 struct net *net = sock_net(sk); 159 160 write_lock_bh(&ipv6_sk_ac_lock); 161 prev_pac = NULL; 162 for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) { 163 if ((ifindex == 0 || pac->acl_ifindex == ifindex) && 164 ipv6_addr_equal(&pac->acl_addr, addr)) 165 break; 166 prev_pac = pac; 167 } 168 if (!pac) { 169 write_unlock_bh(&ipv6_sk_ac_lock); 170 return -ENOENT; 171 } 172 if (prev_pac) 173 prev_pac->acl_next = pac->acl_next; 174 else 175 np->ipv6_ac_list = pac->acl_next; 176 177 write_unlock_bh(&ipv6_sk_ac_lock); 178 179 dev = dev_get_by_index(net, pac->acl_ifindex); 180 if (dev) { 181 ipv6_dev_ac_dec(dev, &pac->acl_addr); 182 dev_put(dev); 183 } 184 sock_kfree_s(sk, pac, sizeof(*pac)); 185 return 0; 186 } 187 188 void ipv6_sock_ac_close(struct sock *sk) 189 { 190 struct ipv6_pinfo *np = inet6_sk(sk); 191 struct net_device *dev = NULL; 192 struct ipv6_ac_socklist *pac; 193 struct net *net = sock_net(sk); 194 int prev_index; 195 196 write_lock_bh(&ipv6_sk_ac_lock); 197 pac = np->ipv6_ac_list; 198 np->ipv6_ac_list = NULL; 199 write_unlock_bh(&ipv6_sk_ac_lock); 200 201 prev_index = 0; 202 while (pac) { 203 struct ipv6_ac_socklist *next = pac->acl_next; 204 205 if (pac->acl_ifindex != prev_index) { 206 if (dev) 207 dev_put(dev); 208 dev = dev_get_by_index(net, pac->acl_ifindex); 209 prev_index = pac->acl_ifindex; 210 } 211 if (dev) 212 ipv6_dev_ac_dec(dev, &pac->acl_addr); 213 sock_kfree_s(sk, pac, sizeof(*pac)); 214 pac = next; 215 } 216 if (dev) 217 dev_put(dev); 218 } 219 220 #if 0 221 /* The function is not used, which is funny. Apparently, author 222 * supposed to use it to filter out datagrams inside udp/raw but forgot. 223 * 224 * It is OK, anycasts are not special comparing to delivery to unicasts. 225 */ 226 227 int inet6_ac_check(struct sock *sk, struct in6_addr *addr, int ifindex) 228 { 229 struct ipv6_ac_socklist *pac; 230 struct ipv6_pinfo *np = inet6_sk(sk); 231 int found; 232 233 found = 0; 234 read_lock(&ipv6_sk_ac_lock); 235 for (pac=np->ipv6_ac_list; pac; pac=pac->acl_next) { 236 if (ifindex && pac->acl_ifindex != ifindex) 237 continue; 238 found = ipv6_addr_equal(&pac->acl_addr, addr); 239 if (found) 240 break; 241 } 242 read_unlock(&ipv6_sk_ac_lock); 243 244 return found; 245 } 246 247 #endif 248 249 static void aca_put(struct ifacaddr6 *ac) 250 { 251 if (atomic_dec_and_test(&ac->aca_refcnt)) { 252 in6_dev_put(ac->aca_idev); 253 dst_release(&ac->aca_rt->u.dst); 254 kfree(ac); 255 } 256 } 257 258 /* 259 * device anycast group inc (add if not found) 260 */ 261 int ipv6_dev_ac_inc(struct net_device *dev, struct in6_addr *addr) 262 { 263 struct ifacaddr6 *aca; 264 struct inet6_dev *idev; 265 struct rt6_info *rt; 266 int err; 267 268 idev = in6_dev_get(dev); 269 270 if (idev == NULL) 271 return -EINVAL; 272 273 write_lock_bh(&idev->lock); 274 if (idev->dead) { 275 err = -ENODEV; 276 goto out; 277 } 278 279 for (aca = idev->ac_list; aca; aca = aca->aca_next) { 280 if (ipv6_addr_equal(&aca->aca_addr, addr)) { 281 aca->aca_users++; 282 err = 0; 283 goto out; 284 } 285 } 286 287 /* 288 * not found: create a new one. 289 */ 290 291 aca = kzalloc(sizeof(struct ifacaddr6), GFP_ATOMIC); 292 293 if (aca == NULL) { 294 err = -ENOMEM; 295 goto out; 296 } 297 298 rt = addrconf_dst_alloc(idev, addr, 1); 299 if (IS_ERR(rt)) { 300 kfree(aca); 301 err = PTR_ERR(rt); 302 goto out; 303 } 304 305 ipv6_addr_copy(&aca->aca_addr, addr); 306 aca->aca_idev = idev; 307 aca->aca_rt = rt; 308 aca->aca_users = 1; 309 /* aca_tstamp should be updated upon changes */ 310 aca->aca_cstamp = aca->aca_tstamp = jiffies; 311 atomic_set(&aca->aca_refcnt, 2); 312 spin_lock_init(&aca->aca_lock); 313 314 aca->aca_next = idev->ac_list; 315 idev->ac_list = aca; 316 write_unlock_bh(&idev->lock); 317 318 ip6_ins_rt(rt); 319 320 addrconf_join_solict(dev, &aca->aca_addr); 321 322 aca_put(aca); 323 return 0; 324 out: 325 write_unlock_bh(&idev->lock); 326 in6_dev_put(idev); 327 return err; 328 } 329 330 /* 331 * device anycast group decrement 332 */ 333 int __ipv6_dev_ac_dec(struct inet6_dev *idev, struct in6_addr *addr) 334 { 335 struct ifacaddr6 *aca, *prev_aca; 336 337 write_lock_bh(&idev->lock); 338 prev_aca = NULL; 339 for (aca = idev->ac_list; aca; aca = aca->aca_next) { 340 if (ipv6_addr_equal(&aca->aca_addr, addr)) 341 break; 342 prev_aca = aca; 343 } 344 if (!aca) { 345 write_unlock_bh(&idev->lock); 346 return -ENOENT; 347 } 348 if (--aca->aca_users > 0) { 349 write_unlock_bh(&idev->lock); 350 return 0; 351 } 352 if (prev_aca) 353 prev_aca->aca_next = aca->aca_next; 354 else 355 idev->ac_list = aca->aca_next; 356 write_unlock_bh(&idev->lock); 357 addrconf_leave_solict(idev, &aca->aca_addr); 358 359 dst_hold(&aca->aca_rt->u.dst); 360 ip6_del_rt(aca->aca_rt); 361 362 aca_put(aca); 363 return 0; 364 } 365 366 static int ipv6_dev_ac_dec(struct net_device *dev, struct in6_addr *addr) 367 { 368 int ret; 369 struct inet6_dev *idev = in6_dev_get(dev); 370 if (idev == NULL) 371 return -ENODEV; 372 ret = __ipv6_dev_ac_dec(idev, addr); 373 in6_dev_put(idev); 374 return ret; 375 } 376 377 /* 378 * check if the interface has this anycast address 379 */ 380 static int ipv6_chk_acast_dev(struct net_device *dev, struct in6_addr *addr) 381 { 382 struct inet6_dev *idev; 383 struct ifacaddr6 *aca; 384 385 idev = in6_dev_get(dev); 386 if (idev) { 387 read_lock_bh(&idev->lock); 388 for (aca = idev->ac_list; aca; aca = aca->aca_next) 389 if (ipv6_addr_equal(&aca->aca_addr, addr)) 390 break; 391 read_unlock_bh(&idev->lock); 392 in6_dev_put(idev); 393 return aca != NULL; 394 } 395 return 0; 396 } 397 398 /* 399 * check if given interface (or any, if dev==0) has this anycast address 400 */ 401 int ipv6_chk_acast_addr(struct net *net, struct net_device *dev, 402 struct in6_addr *addr) 403 { 404 int found = 0; 405 406 if (dev) 407 return ipv6_chk_acast_dev(dev, addr); 408 rcu_read_lock(); 409 for_each_netdev_rcu(net, dev) 410 if (ipv6_chk_acast_dev(dev, addr)) { 411 found = 1; 412 break; 413 } 414 rcu_read_unlock(); 415 return found; 416 } 417 418 419 #ifdef CONFIG_PROC_FS 420 struct ac6_iter_state { 421 struct seq_net_private p; 422 struct net_device *dev; 423 struct inet6_dev *idev; 424 }; 425 426 #define ac6_seq_private(seq) ((struct ac6_iter_state *)(seq)->private) 427 428 static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq) 429 { 430 struct ifacaddr6 *im = NULL; 431 struct ac6_iter_state *state = ac6_seq_private(seq); 432 struct net *net = seq_file_net(seq); 433 434 state->idev = NULL; 435 for_each_netdev_rcu(net, state->dev) { 436 struct inet6_dev *idev; 437 idev = __in6_dev_get(state->dev); 438 if (!idev) 439 continue; 440 read_lock_bh(&idev->lock); 441 im = idev->ac_list; 442 if (im) { 443 state->idev = idev; 444 break; 445 } 446 read_unlock_bh(&idev->lock); 447 } 448 return im; 449 } 450 451 static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im) 452 { 453 struct ac6_iter_state *state = ac6_seq_private(seq); 454 455 im = im->aca_next; 456 while (!im) { 457 if (likely(state->idev != NULL)) 458 read_unlock_bh(&state->idev->lock); 459 460 state->dev = next_net_device_rcu(state->dev); 461 if (!state->dev) { 462 state->idev = NULL; 463 break; 464 } 465 state->idev = __in6_dev_get(state->dev); 466 if (!state->idev) 467 continue; 468 read_lock_bh(&state->idev->lock); 469 im = state->idev->ac_list; 470 } 471 return im; 472 } 473 474 static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos) 475 { 476 struct ifacaddr6 *im = ac6_get_first(seq); 477 if (im) 478 while (pos && (im = ac6_get_next(seq, im)) != NULL) 479 --pos; 480 return pos ? NULL : im; 481 } 482 483 static void *ac6_seq_start(struct seq_file *seq, loff_t *pos) 484 __acquires(RCU) 485 { 486 rcu_read_lock(); 487 return ac6_get_idx(seq, *pos); 488 } 489 490 static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos) 491 { 492 struct ifacaddr6 *im = ac6_get_next(seq, v); 493 494 ++*pos; 495 return im; 496 } 497 498 static void ac6_seq_stop(struct seq_file *seq, void *v) 499 __releases(RCU) 500 { 501 struct ac6_iter_state *state = ac6_seq_private(seq); 502 503 if (likely(state->idev != NULL)) { 504 read_unlock_bh(&state->idev->lock); 505 state->idev = NULL; 506 } 507 rcu_read_unlock(); 508 } 509 510 static int ac6_seq_show(struct seq_file *seq, void *v) 511 { 512 struct ifacaddr6 *im = (struct ifacaddr6 *)v; 513 struct ac6_iter_state *state = ac6_seq_private(seq); 514 515 seq_printf(seq, "%-4d %-15s %pi6 %5d\n", 516 state->dev->ifindex, state->dev->name, 517 &im->aca_addr, im->aca_users); 518 return 0; 519 } 520 521 static const struct seq_operations ac6_seq_ops = { 522 .start = ac6_seq_start, 523 .next = ac6_seq_next, 524 .stop = ac6_seq_stop, 525 .show = ac6_seq_show, 526 }; 527 528 static int ac6_seq_open(struct inode *inode, struct file *file) 529 { 530 return seq_open_net(inode, file, &ac6_seq_ops, 531 sizeof(struct ac6_iter_state)); 532 } 533 534 static const struct file_operations ac6_seq_fops = { 535 .owner = THIS_MODULE, 536 .open = ac6_seq_open, 537 .read = seq_read, 538 .llseek = seq_lseek, 539 .release = seq_release_net, 540 }; 541 542 int __net_init ac6_proc_init(struct net *net) 543 { 544 if (!proc_net_fops_create(net, "anycast6", S_IRUGO, &ac6_seq_fops)) 545 return -ENOMEM; 546 547 return 0; 548 } 549 550 void ac6_proc_exit(struct net *net) 551 { 552 proc_net_remove(net, "anycast6"); 553 } 554 #endif 555 556