1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Anycast support for IPv6 4 * Linux INET6 implementation 5 * 6 * Authors: 7 * David L Stevens (dlstevens@us.ibm.com) 8 * 9 * based heavily on net/ipv6/mcast.c 10 */ 11 12 #include <linux/capability.h> 13 #include <linux/module.h> 14 #include <linux/errno.h> 15 #include <linux/types.h> 16 #include <linux/random.h> 17 #include <linux/string.h> 18 #include <linux/socket.h> 19 #include <linux/sockios.h> 20 #include <linux/net.h> 21 #include <linux/in6.h> 22 #include <linux/netdevice.h> 23 #include <linux/if_arp.h> 24 #include <linux/route.h> 25 #include <linux/init.h> 26 #include <linux/proc_fs.h> 27 #include <linux/seq_file.h> 28 #include <linux/slab.h> 29 30 #include <net/net_namespace.h> 31 #include <net/sock.h> 32 #include <net/snmp.h> 33 34 #include <net/ipv6.h> 35 #include <net/protocol.h> 36 #include <net/if_inet6.h> 37 #include <net/ndisc.h> 38 #include <net/addrconf.h> 39 #include <net/ip6_route.h> 40 41 #include <net/checksum.h> 42 43 #define IN6_ADDR_HSIZE_SHIFT 8 44 #define IN6_ADDR_HSIZE BIT(IN6_ADDR_HSIZE_SHIFT) 45 /* anycast address hash table 46 */ 47 static struct hlist_head inet6_acaddr_lst[IN6_ADDR_HSIZE]; 48 static DEFINE_SPINLOCK(acaddr_hash_lock); 49 50 static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr); 51 52 static u32 inet6_acaddr_hash(const struct net *net, 53 const struct in6_addr *addr) 54 { 55 u32 val = __ipv6_addr_jhash(addr, net_hash_mix(net)); 56 57 return hash_32(val, IN6_ADDR_HSIZE_SHIFT); 58 } 59 60 /* 61 * socket join an anycast group 62 */ 63 64 int ipv6_sock_ac_join(struct sock *sk, int ifindex, const struct in6_addr *addr) 65 { 66 struct ipv6_pinfo *np = inet6_sk(sk); 67 struct net_device *dev = NULL; 68 struct inet6_dev *idev; 69 struct ipv6_ac_socklist *pac; 70 struct net *net = sock_net(sk); 71 int ishost = !net->ipv6.devconf_all->forwarding; 72 int err = 0; 73 74 ASSERT_RTNL(); 75 76 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 77 return -EPERM; 78 if (ipv6_addr_is_multicast(addr)) 79 return -EINVAL; 80 81 if (ifindex) 82 dev = __dev_get_by_index(net, ifindex); 83 84 if (ipv6_chk_addr_and_flags(net, addr, dev, true, 0, IFA_F_TENTATIVE)) 85 return -EINVAL; 86 87 pac = sock_kmalloc(sk, sizeof(struct ipv6_ac_socklist), GFP_KERNEL); 88 if (!pac) 89 return -ENOMEM; 90 pac->acl_next = NULL; 91 pac->acl_addr = *addr; 92 93 if (ifindex == 0) { 94 struct rt6_info *rt; 95 96 rt = rt6_lookup(net, addr, NULL, 0, NULL, 0); 97 if (rt) { 98 dev = rt->dst.dev; 99 ip6_rt_put(rt); 100 } else if (ishost) { 101 err = -EADDRNOTAVAIL; 102 goto error; 103 } else { 104 /* router, no matching interface: just pick one */ 105 dev = __dev_get_by_flags(net, IFF_UP, 106 IFF_UP | IFF_LOOPBACK); 107 } 108 } 109 110 if (!dev) { 111 err = -ENODEV; 112 goto error; 113 } 114 115 idev = __in6_dev_get(dev); 116 if (!idev) { 117 if (ifindex) 118 err = -ENODEV; 119 else 120 err = -EADDRNOTAVAIL; 121 goto error; 122 } 123 /* reset ishost, now that we have a specific device */ 124 ishost = !idev->cnf.forwarding; 125 126 pac->acl_ifindex = dev->ifindex; 127 128 /* XXX 129 * For hosts, allow link-local or matching prefix anycasts. 130 * This obviates the need for propagating anycast routes while 131 * still allowing some non-router anycast participation. 132 */ 133 if (!ipv6_chk_prefix(addr, dev)) { 134 if (ishost) 135 err = -EADDRNOTAVAIL; 136 if (err) 137 goto error; 138 } 139 140 err = __ipv6_dev_ac_inc(idev, addr); 141 if (!err) { 142 pac->acl_next = np->ipv6_ac_list; 143 np->ipv6_ac_list = pac; 144 pac = NULL; 145 } 146 147 error: 148 if (pac) 149 sock_kfree_s(sk, pac, sizeof(*pac)); 150 return err; 151 } 152 153 /* 154 * socket leave an anycast group 155 */ 156 int ipv6_sock_ac_drop(struct sock *sk, int ifindex, const struct in6_addr *addr) 157 { 158 struct ipv6_pinfo *np = inet6_sk(sk); 159 struct net_device *dev; 160 struct ipv6_ac_socklist *pac, *prev_pac; 161 struct net *net = sock_net(sk); 162 163 ASSERT_RTNL(); 164 165 prev_pac = NULL; 166 for (pac = np->ipv6_ac_list; pac; pac = pac->acl_next) { 167 if ((ifindex == 0 || pac->acl_ifindex == ifindex) && 168 ipv6_addr_equal(&pac->acl_addr, addr)) 169 break; 170 prev_pac = pac; 171 } 172 if (!pac) 173 return -ENOENT; 174 if (prev_pac) 175 prev_pac->acl_next = pac->acl_next; 176 else 177 np->ipv6_ac_list = pac->acl_next; 178 179 dev = __dev_get_by_index(net, pac->acl_ifindex); 180 if (dev) 181 ipv6_dev_ac_dec(dev, &pac->acl_addr); 182 183 sock_kfree_s(sk, pac, sizeof(*pac)); 184 return 0; 185 } 186 187 void __ipv6_sock_ac_close(struct sock *sk) 188 { 189 struct ipv6_pinfo *np = inet6_sk(sk); 190 struct net_device *dev = NULL; 191 struct ipv6_ac_socklist *pac; 192 struct net *net = sock_net(sk); 193 int prev_index; 194 195 ASSERT_RTNL(); 196 pac = np->ipv6_ac_list; 197 np->ipv6_ac_list = NULL; 198 199 prev_index = 0; 200 while (pac) { 201 struct ipv6_ac_socklist *next = pac->acl_next; 202 203 if (pac->acl_ifindex != prev_index) { 204 dev = __dev_get_by_index(net, pac->acl_ifindex); 205 prev_index = pac->acl_ifindex; 206 } 207 if (dev) 208 ipv6_dev_ac_dec(dev, &pac->acl_addr); 209 sock_kfree_s(sk, pac, sizeof(*pac)); 210 pac = next; 211 } 212 } 213 214 void ipv6_sock_ac_close(struct sock *sk) 215 { 216 struct ipv6_pinfo *np = inet6_sk(sk); 217 218 if (!np->ipv6_ac_list) 219 return; 220 rtnl_lock(); 221 __ipv6_sock_ac_close(sk); 222 rtnl_unlock(); 223 } 224 225 static void ipv6_add_acaddr_hash(struct net *net, struct ifacaddr6 *aca) 226 { 227 unsigned int hash = inet6_acaddr_hash(net, &aca->aca_addr); 228 229 spin_lock(&acaddr_hash_lock); 230 hlist_add_head_rcu(&aca->aca_addr_lst, &inet6_acaddr_lst[hash]); 231 spin_unlock(&acaddr_hash_lock); 232 } 233 234 static void ipv6_del_acaddr_hash(struct ifacaddr6 *aca) 235 { 236 spin_lock(&acaddr_hash_lock); 237 hlist_del_init_rcu(&aca->aca_addr_lst); 238 spin_unlock(&acaddr_hash_lock); 239 } 240 241 static void aca_get(struct ifacaddr6 *aca) 242 { 243 refcount_inc(&aca->aca_refcnt); 244 } 245 246 static void aca_free_rcu(struct rcu_head *h) 247 { 248 struct ifacaddr6 *aca = container_of(h, struct ifacaddr6, rcu); 249 250 fib6_info_release(aca->aca_rt); 251 kfree(aca); 252 } 253 254 static void aca_put(struct ifacaddr6 *ac) 255 { 256 if (refcount_dec_and_test(&ac->aca_refcnt)) 257 call_rcu_hurry(&ac->rcu, aca_free_rcu); 258 } 259 260 static struct ifacaddr6 *aca_alloc(struct fib6_info *f6i, 261 const struct in6_addr *addr) 262 { 263 struct ifacaddr6 *aca; 264 265 aca = kzalloc(sizeof(*aca), GFP_ATOMIC); 266 if (!aca) 267 return NULL; 268 269 aca->aca_addr = *addr; 270 fib6_info_hold(f6i); 271 aca->aca_rt = f6i; 272 INIT_HLIST_NODE(&aca->aca_addr_lst); 273 aca->aca_users = 1; 274 /* aca_tstamp should be updated upon changes */ 275 aca->aca_cstamp = aca->aca_tstamp = jiffies; 276 refcount_set(&aca->aca_refcnt, 1); 277 278 return aca; 279 } 280 281 static void inet6_ifacaddr_notify(struct net_device *dev, 282 const struct ifacaddr6 *ifaca, int event) 283 { 284 struct inet6_fill_args fillargs = { 285 .event = event, 286 .netnsid = -1, 287 }; 288 struct net *net = dev_net(dev); 289 struct sk_buff *skb; 290 int err = -ENOMEM; 291 292 skb = nlmsg_new(NLMSG_ALIGN(sizeof(struct ifaddrmsg)) + 293 nla_total_size(sizeof(struct in6_addr)) + 294 nla_total_size(sizeof(struct ifa_cacheinfo)), 295 GFP_KERNEL); 296 if (!skb) 297 goto error; 298 299 err = inet6_fill_ifacaddr(skb, ifaca, &fillargs); 300 if (err < 0) { 301 pr_err("Failed to fill in anycast addresses (err %d)\n", err); 302 nlmsg_free(skb); 303 goto error; 304 } 305 306 rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ACADDR, NULL, GFP_KERNEL); 307 return; 308 error: 309 rtnl_set_sk_err(net, RTNLGRP_IPV6_ACADDR, err); 310 } 311 312 /* 313 * device anycast group inc (add if not found) 314 */ 315 int __ipv6_dev_ac_inc(struct inet6_dev *idev, const struct in6_addr *addr) 316 { 317 struct ifacaddr6 *aca; 318 struct fib6_info *f6i; 319 struct net *net; 320 int err; 321 322 ASSERT_RTNL(); 323 324 write_lock_bh(&idev->lock); 325 if (idev->dead) { 326 err = -ENODEV; 327 goto out; 328 } 329 330 for (aca = rtnl_dereference(idev->ac_list); aca; 331 aca = rtnl_dereference(aca->aca_next)) { 332 if (ipv6_addr_equal(&aca->aca_addr, addr)) { 333 aca->aca_users++; 334 err = 0; 335 goto out; 336 } 337 } 338 339 net = dev_net(idev->dev); 340 f6i = addrconf_f6i_alloc(net, idev, addr, true, GFP_ATOMIC, NULL); 341 if (IS_ERR(f6i)) { 342 err = PTR_ERR(f6i); 343 goto out; 344 } 345 aca = aca_alloc(f6i, addr); 346 if (!aca) { 347 fib6_info_release(f6i); 348 err = -ENOMEM; 349 goto out; 350 } 351 352 /* Hold this for addrconf_join_solict() below before we unlock, 353 * it is already exposed via idev->ac_list. 354 */ 355 aca_get(aca); 356 aca->aca_next = idev->ac_list; 357 rcu_assign_pointer(idev->ac_list, aca); 358 359 write_unlock_bh(&idev->lock); 360 361 ipv6_add_acaddr_hash(net, aca); 362 363 ip6_ins_rt(net, f6i); 364 365 addrconf_join_solict(idev->dev, &aca->aca_addr); 366 367 inet6_ifacaddr_notify(idev->dev, aca, RTM_NEWANYCAST); 368 369 aca_put(aca); 370 return 0; 371 out: 372 write_unlock_bh(&idev->lock); 373 return err; 374 } 375 376 /* 377 * device anycast group decrement 378 */ 379 int __ipv6_dev_ac_dec(struct inet6_dev *idev, const struct in6_addr *addr) 380 { 381 struct ifacaddr6 *aca, *prev_aca; 382 383 ASSERT_RTNL(); 384 385 write_lock_bh(&idev->lock); 386 prev_aca = NULL; 387 for (aca = rtnl_dereference(idev->ac_list); aca; 388 aca = rtnl_dereference(aca->aca_next)) { 389 if (ipv6_addr_equal(&aca->aca_addr, addr)) 390 break; 391 prev_aca = aca; 392 } 393 if (!aca) { 394 write_unlock_bh(&idev->lock); 395 return -ENOENT; 396 } 397 if (--aca->aca_users > 0) { 398 write_unlock_bh(&idev->lock); 399 return 0; 400 } 401 if (prev_aca) 402 rcu_assign_pointer(prev_aca->aca_next, aca->aca_next); 403 else 404 rcu_assign_pointer(idev->ac_list, aca->aca_next); 405 write_unlock_bh(&idev->lock); 406 ipv6_del_acaddr_hash(aca); 407 addrconf_leave_solict(idev, &aca->aca_addr); 408 409 ip6_del_rt(dev_net(idev->dev), aca->aca_rt, false); 410 411 inet6_ifacaddr_notify(idev->dev, aca, RTM_DELANYCAST); 412 413 aca_put(aca); 414 return 0; 415 } 416 417 /* called with rtnl_lock() */ 418 static int ipv6_dev_ac_dec(struct net_device *dev, const struct in6_addr *addr) 419 { 420 struct inet6_dev *idev = __in6_dev_get(dev); 421 422 if (!idev) 423 return -ENODEV; 424 return __ipv6_dev_ac_dec(idev, addr); 425 } 426 427 void ipv6_ac_destroy_dev(struct inet6_dev *idev) 428 { 429 struct ifacaddr6 *aca; 430 431 write_lock_bh(&idev->lock); 432 while ((aca = rtnl_dereference(idev->ac_list)) != NULL) { 433 rcu_assign_pointer(idev->ac_list, aca->aca_next); 434 write_unlock_bh(&idev->lock); 435 436 ipv6_del_acaddr_hash(aca); 437 438 addrconf_leave_solict(idev, &aca->aca_addr); 439 440 ip6_del_rt(dev_net(idev->dev), aca->aca_rt, false); 441 442 aca_put(aca); 443 444 write_lock_bh(&idev->lock); 445 } 446 write_unlock_bh(&idev->lock); 447 } 448 449 /* 450 * check if the interface has this anycast address 451 * called with rcu_read_lock() 452 */ 453 static bool ipv6_chk_acast_dev(struct net_device *dev, const struct in6_addr *addr) 454 { 455 struct inet6_dev *idev; 456 struct ifacaddr6 *aca; 457 458 idev = __in6_dev_get(dev); 459 if (idev) { 460 for (aca = rcu_dereference(idev->ac_list); aca; 461 aca = rcu_dereference(aca->aca_next)) 462 if (ipv6_addr_equal(&aca->aca_addr, addr)) 463 break; 464 return aca != NULL; 465 } 466 return false; 467 } 468 469 /* 470 * check if given interface (or any, if dev==0) has this anycast address 471 */ 472 bool ipv6_chk_acast_addr(struct net *net, struct net_device *dev, 473 const struct in6_addr *addr) 474 { 475 struct net_device *nh_dev; 476 struct ifacaddr6 *aca; 477 bool found = false; 478 479 rcu_read_lock(); 480 if (dev) 481 found = ipv6_chk_acast_dev(dev, addr); 482 else { 483 unsigned int hash = inet6_acaddr_hash(net, addr); 484 485 hlist_for_each_entry_rcu(aca, &inet6_acaddr_lst[hash], 486 aca_addr_lst) { 487 nh_dev = fib6_info_nh_dev(aca->aca_rt); 488 if (!nh_dev || !net_eq(dev_net(nh_dev), net)) 489 continue; 490 if (ipv6_addr_equal(&aca->aca_addr, addr)) { 491 found = true; 492 break; 493 } 494 } 495 } 496 rcu_read_unlock(); 497 return found; 498 } 499 500 /* check if this anycast address is link-local on given interface or 501 * is global 502 */ 503 bool ipv6_chk_acast_addr_src(struct net *net, struct net_device *dev, 504 const struct in6_addr *addr) 505 { 506 return ipv6_chk_acast_addr(net, 507 (ipv6_addr_type(addr) & IPV6_ADDR_LINKLOCAL ? 508 dev : NULL), 509 addr); 510 } 511 512 #ifdef CONFIG_PROC_FS 513 struct ac6_iter_state { 514 struct seq_net_private p; 515 struct net_device *dev; 516 }; 517 518 #define ac6_seq_private(seq) ((struct ac6_iter_state *)(seq)->private) 519 520 static inline struct ifacaddr6 *ac6_get_first(struct seq_file *seq) 521 { 522 struct ac6_iter_state *state = ac6_seq_private(seq); 523 struct net *net = seq_file_net(seq); 524 struct ifacaddr6 *im = NULL; 525 526 for_each_netdev_rcu(net, state->dev) { 527 struct inet6_dev *idev; 528 529 idev = __in6_dev_get(state->dev); 530 if (!idev) 531 continue; 532 im = rcu_dereference(idev->ac_list); 533 if (im) 534 break; 535 } 536 return im; 537 } 538 539 static struct ifacaddr6 *ac6_get_next(struct seq_file *seq, struct ifacaddr6 *im) 540 { 541 struct ac6_iter_state *state = ac6_seq_private(seq); 542 struct inet6_dev *idev; 543 544 im = rcu_dereference(im->aca_next); 545 while (!im) { 546 state->dev = next_net_device_rcu(state->dev); 547 if (!state->dev) 548 break; 549 idev = __in6_dev_get(state->dev); 550 if (!idev) 551 continue; 552 im = rcu_dereference(idev->ac_list); 553 } 554 return im; 555 } 556 557 static struct ifacaddr6 *ac6_get_idx(struct seq_file *seq, loff_t pos) 558 { 559 struct ifacaddr6 *im = ac6_get_first(seq); 560 if (im) 561 while (pos && (im = ac6_get_next(seq, im)) != NULL) 562 --pos; 563 return pos ? NULL : im; 564 } 565 566 static void *ac6_seq_start(struct seq_file *seq, loff_t *pos) 567 __acquires(RCU) 568 { 569 rcu_read_lock(); 570 return ac6_get_idx(seq, *pos); 571 } 572 573 static void *ac6_seq_next(struct seq_file *seq, void *v, loff_t *pos) 574 { 575 struct ifacaddr6 *im = ac6_get_next(seq, v); 576 577 ++*pos; 578 return im; 579 } 580 581 static void ac6_seq_stop(struct seq_file *seq, void *v) 582 __releases(RCU) 583 { 584 rcu_read_unlock(); 585 } 586 587 static int ac6_seq_show(struct seq_file *seq, void *v) 588 { 589 struct ifacaddr6 *im = (struct ifacaddr6 *)v; 590 struct ac6_iter_state *state = ac6_seq_private(seq); 591 592 seq_printf(seq, "%-4d %-15s %pi6 %5d\n", 593 state->dev->ifindex, state->dev->name, 594 &im->aca_addr, im->aca_users); 595 return 0; 596 } 597 598 static const struct seq_operations ac6_seq_ops = { 599 .start = ac6_seq_start, 600 .next = ac6_seq_next, 601 .stop = ac6_seq_stop, 602 .show = ac6_seq_show, 603 }; 604 605 int __net_init ac6_proc_init(struct net *net) 606 { 607 if (!proc_create_net("anycast6", 0444, net->proc_net, &ac6_seq_ops, 608 sizeof(struct ac6_iter_state))) 609 return -ENOMEM; 610 611 return 0; 612 } 613 614 void ac6_proc_exit(struct net *net) 615 { 616 remove_proc_entry("anycast6", net->proc_net); 617 } 618 #endif 619 620 /* Init / cleanup code 621 */ 622 int __init ipv6_anycast_init(void) 623 { 624 int i; 625 626 for (i = 0; i < IN6_ADDR_HSIZE; i++) 627 INIT_HLIST_HEAD(&inet6_acaddr_lst[i]); 628 return 0; 629 } 630 631 void ipv6_anycast_cleanup(void) 632 { 633 int i; 634 635 spin_lock(&acaddr_hash_lock); 636 for (i = 0; i < IN6_ADDR_HSIZE; i++) 637 WARN_ON(!hlist_empty(&inet6_acaddr_lst[i])); 638 spin_unlock(&acaddr_hash_lock); 639 } 640