1 /* 2 * Linux INET6 implementation 3 * Forwarding Information Database 4 * 5 * Authors: 6 * Pedro Roque <roque@di.fc.ul.pt> 7 * 8 * $Id: ip6_fib.c,v 1.25 2001/10/31 21:55:55 davem Exp $ 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 /* 17 * Changes: 18 * Yuji SEKIYA @USAGI: Support default route on router node; 19 * remove ip6_null_entry from the top of 20 * routing table. 21 * Ville Nuorvala: Fixed routing subtrees. 22 */ 23 #include <linux/errno.h> 24 #include <linux/types.h> 25 #include <linux/net.h> 26 #include <linux/route.h> 27 #include <linux/netdevice.h> 28 #include <linux/in6.h> 29 #include <linux/init.h> 30 #include <linux/list.h> 31 32 #ifdef CONFIG_PROC_FS 33 #include <linux/proc_fs.h> 34 #endif 35 36 #include <net/ipv6.h> 37 #include <net/ndisc.h> 38 #include <net/addrconf.h> 39 40 #include <net/ip6_fib.h> 41 #include <net/ip6_route.h> 42 43 #define RT6_DEBUG 2 44 45 #if RT6_DEBUG >= 3 46 #define RT6_TRACE(x...) printk(KERN_DEBUG x) 47 #else 48 #define RT6_TRACE(x...) do { ; } while (0) 49 #endif 50 51 struct rt6_statistics rt6_stats; 52 53 static struct kmem_cache * fib6_node_kmem __read_mostly; 54 55 enum fib_walk_state_t 56 { 57 #ifdef CONFIG_IPV6_SUBTREES 58 FWS_S, 59 #endif 60 FWS_L, 61 FWS_R, 62 FWS_C, 63 FWS_U 64 }; 65 66 struct fib6_cleaner_t 67 { 68 struct fib6_walker_t w; 69 int (*func)(struct rt6_info *, void *arg); 70 void *arg; 71 }; 72 73 static DEFINE_RWLOCK(fib6_walker_lock); 74 75 #ifdef CONFIG_IPV6_SUBTREES 76 #define FWS_INIT FWS_S 77 #else 78 #define FWS_INIT FWS_L 79 #endif 80 81 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt); 82 static struct rt6_info * fib6_find_prefix(struct fib6_node *fn); 83 static struct fib6_node * fib6_repair_tree(struct fib6_node *fn); 84 static int fib6_walk(struct fib6_walker_t *w); 85 static int fib6_walk_continue(struct fib6_walker_t *w); 86 87 /* 88 * A routing update causes an increase of the serial number on the 89 * affected subtree. This allows for cached routes to be asynchronously 90 * tested when modifications are made to the destination cache as a 91 * result of redirects, path MTU changes, etc. 92 */ 93 94 static __u32 rt_sernum; 95 96 static DEFINE_TIMER(ip6_fib_timer, fib6_run_gc, 0, 0); 97 98 static struct fib6_walker_t fib6_walker_list = { 99 .prev = &fib6_walker_list, 100 .next = &fib6_walker_list, 101 }; 102 103 #define FOR_WALKERS(w) for ((w)=fib6_walker_list.next; (w) != &fib6_walker_list; (w)=(w)->next) 104 105 static inline void fib6_walker_link(struct fib6_walker_t *w) 106 { 107 write_lock_bh(&fib6_walker_lock); 108 w->next = fib6_walker_list.next; 109 w->prev = &fib6_walker_list; 110 w->next->prev = w; 111 w->prev->next = w; 112 write_unlock_bh(&fib6_walker_lock); 113 } 114 115 static inline void fib6_walker_unlink(struct fib6_walker_t *w) 116 { 117 write_lock_bh(&fib6_walker_lock); 118 w->next->prev = w->prev; 119 w->prev->next = w->next; 120 w->prev = w->next = w; 121 write_unlock_bh(&fib6_walker_lock); 122 } 123 static __inline__ u32 fib6_new_sernum(void) 124 { 125 u32 n = ++rt_sernum; 126 if ((__s32)n <= 0) 127 rt_sernum = n = 1; 128 return n; 129 } 130 131 /* 132 * Auxiliary address test functions for the radix tree. 133 * 134 * These assume a 32bit processor (although it will work on 135 * 64bit processors) 136 */ 137 138 /* 139 * test bit 140 */ 141 142 static __inline__ __be32 addr_bit_set(void *token, int fn_bit) 143 { 144 __be32 *addr = token; 145 146 return htonl(1 << ((~fn_bit)&0x1F)) & addr[fn_bit>>5]; 147 } 148 149 static __inline__ struct fib6_node * node_alloc(void) 150 { 151 struct fib6_node *fn; 152 153 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC); 154 155 return fn; 156 } 157 158 static __inline__ void node_free(struct fib6_node * fn) 159 { 160 kmem_cache_free(fib6_node_kmem, fn); 161 } 162 163 static __inline__ void rt6_release(struct rt6_info *rt) 164 { 165 if (atomic_dec_and_test(&rt->rt6i_ref)) 166 dst_free(&rt->u.dst); 167 } 168 169 static struct fib6_table fib6_main_tbl = { 170 .tb6_id = RT6_TABLE_MAIN, 171 .tb6_root = { 172 .leaf = &ip6_null_entry, 173 .fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO, 174 }, 175 }; 176 177 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 178 #define FIB_TABLE_HASHSZ 256 179 #else 180 #define FIB_TABLE_HASHSZ 1 181 #endif 182 static struct hlist_head fib_table_hash[FIB_TABLE_HASHSZ]; 183 184 static void fib6_link_table(struct fib6_table *tb) 185 { 186 unsigned int h; 187 188 /* 189 * Initialize table lock at a single place to give lockdep a key, 190 * tables aren't visible prior to being linked to the list. 191 */ 192 rwlock_init(&tb->tb6_lock); 193 194 h = tb->tb6_id & (FIB_TABLE_HASHSZ - 1); 195 196 /* 197 * No protection necessary, this is the only list mutatation 198 * operation, tables never disappear once they exist. 199 */ 200 hlist_add_head_rcu(&tb->tb6_hlist, &fib_table_hash[h]); 201 } 202 203 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 204 static struct fib6_table fib6_local_tbl = { 205 .tb6_id = RT6_TABLE_LOCAL, 206 .tb6_root = { 207 .leaf = &ip6_null_entry, 208 .fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO, 209 }, 210 }; 211 212 static struct fib6_table *fib6_alloc_table(u32 id) 213 { 214 struct fib6_table *table; 215 216 table = kzalloc(sizeof(*table), GFP_ATOMIC); 217 if (table != NULL) { 218 table->tb6_id = id; 219 table->tb6_root.leaf = &ip6_null_entry; 220 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 221 } 222 223 return table; 224 } 225 226 struct fib6_table *fib6_new_table(u32 id) 227 { 228 struct fib6_table *tb; 229 230 if (id == 0) 231 id = RT6_TABLE_MAIN; 232 tb = fib6_get_table(id); 233 if (tb) 234 return tb; 235 236 tb = fib6_alloc_table(id); 237 if (tb != NULL) 238 fib6_link_table(tb); 239 240 return tb; 241 } 242 243 struct fib6_table *fib6_get_table(u32 id) 244 { 245 struct fib6_table *tb; 246 struct hlist_node *node; 247 unsigned int h; 248 249 if (id == 0) 250 id = RT6_TABLE_MAIN; 251 h = id & (FIB_TABLE_HASHSZ - 1); 252 rcu_read_lock(); 253 hlist_for_each_entry_rcu(tb, node, &fib_table_hash[h], tb6_hlist) { 254 if (tb->tb6_id == id) { 255 rcu_read_unlock(); 256 return tb; 257 } 258 } 259 rcu_read_unlock(); 260 261 return NULL; 262 } 263 264 static void __init fib6_tables_init(void) 265 { 266 fib6_link_table(&fib6_main_tbl); 267 fib6_link_table(&fib6_local_tbl); 268 } 269 270 #else 271 272 struct fib6_table *fib6_new_table(u32 id) 273 { 274 return fib6_get_table(id); 275 } 276 277 struct fib6_table *fib6_get_table(u32 id) 278 { 279 return &fib6_main_tbl; 280 } 281 282 struct dst_entry *fib6_rule_lookup(struct flowi *fl, int flags, 283 pol_lookup_t lookup) 284 { 285 return (struct dst_entry *) lookup(&fib6_main_tbl, fl, flags); 286 } 287 288 static void __init fib6_tables_init(void) 289 { 290 fib6_link_table(&fib6_main_tbl); 291 } 292 293 #endif 294 295 static int fib6_dump_node(struct fib6_walker_t *w) 296 { 297 int res; 298 struct rt6_info *rt; 299 300 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) { 301 res = rt6_dump_route(rt, w->args); 302 if (res < 0) { 303 /* Frame is full, suspend walking */ 304 w->leaf = rt; 305 return 1; 306 } 307 BUG_TRAP(res!=0); 308 } 309 w->leaf = NULL; 310 return 0; 311 } 312 313 static void fib6_dump_end(struct netlink_callback *cb) 314 { 315 struct fib6_walker_t *w = (void*)cb->args[2]; 316 317 if (w) { 318 cb->args[2] = 0; 319 kfree(w); 320 } 321 cb->done = (void*)cb->args[3]; 322 cb->args[1] = 3; 323 } 324 325 static int fib6_dump_done(struct netlink_callback *cb) 326 { 327 fib6_dump_end(cb); 328 return cb->done ? cb->done(cb) : 0; 329 } 330 331 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 332 struct netlink_callback *cb) 333 { 334 struct fib6_walker_t *w; 335 int res; 336 337 w = (void *)cb->args[2]; 338 w->root = &table->tb6_root; 339 340 if (cb->args[4] == 0) { 341 read_lock_bh(&table->tb6_lock); 342 res = fib6_walk(w); 343 read_unlock_bh(&table->tb6_lock); 344 if (res > 0) 345 cb->args[4] = 1; 346 } else { 347 read_lock_bh(&table->tb6_lock); 348 res = fib6_walk_continue(w); 349 read_unlock_bh(&table->tb6_lock); 350 if (res != 0) { 351 if (res < 0) 352 fib6_walker_unlink(w); 353 goto end; 354 } 355 fib6_walker_unlink(w); 356 cb->args[4] = 0; 357 } 358 end: 359 return res; 360 } 361 362 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 363 { 364 struct net *net = skb->sk->sk_net; 365 unsigned int h, s_h; 366 unsigned int e = 0, s_e; 367 struct rt6_rtnl_dump_arg arg; 368 struct fib6_walker_t *w; 369 struct fib6_table *tb; 370 struct hlist_node *node; 371 int res = 0; 372 373 if (net != &init_net) 374 return 0; 375 376 s_h = cb->args[0]; 377 s_e = cb->args[1]; 378 379 w = (void *)cb->args[2]; 380 if (w == NULL) { 381 /* New dump: 382 * 383 * 1. hook callback destructor. 384 */ 385 cb->args[3] = (long)cb->done; 386 cb->done = fib6_dump_done; 387 388 /* 389 * 2. allocate and initialize walker. 390 */ 391 w = kzalloc(sizeof(*w), GFP_ATOMIC); 392 if (w == NULL) 393 return -ENOMEM; 394 w->func = fib6_dump_node; 395 cb->args[2] = (long)w; 396 } 397 398 arg.skb = skb; 399 arg.cb = cb; 400 w->args = &arg; 401 402 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 403 e = 0; 404 hlist_for_each_entry(tb, node, &fib_table_hash[h], tb6_hlist) { 405 if (e < s_e) 406 goto next; 407 res = fib6_dump_table(tb, skb, cb); 408 if (res != 0) 409 goto out; 410 next: 411 e++; 412 } 413 } 414 out: 415 cb->args[1] = e; 416 cb->args[0] = h; 417 418 res = res < 0 ? res : skb->len; 419 if (res <= 0) 420 fib6_dump_end(cb); 421 return res; 422 } 423 424 /* 425 * Routing Table 426 * 427 * return the appropriate node for a routing tree "add" operation 428 * by either creating and inserting or by returning an existing 429 * node. 430 */ 431 432 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr, 433 int addrlen, int plen, 434 int offset) 435 { 436 struct fib6_node *fn, *in, *ln; 437 struct fib6_node *pn = NULL; 438 struct rt6key *key; 439 int bit; 440 __be32 dir = 0; 441 __u32 sernum = fib6_new_sernum(); 442 443 RT6_TRACE("fib6_add_1\n"); 444 445 /* insert node in tree */ 446 447 fn = root; 448 449 do { 450 key = (struct rt6key *)((u8 *)fn->leaf + offset); 451 452 /* 453 * Prefix match 454 */ 455 if (plen < fn->fn_bit || 456 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 457 goto insert_above; 458 459 /* 460 * Exact match ? 461 */ 462 463 if (plen == fn->fn_bit) { 464 /* clean up an intermediate node */ 465 if ((fn->fn_flags & RTN_RTINFO) == 0) { 466 rt6_release(fn->leaf); 467 fn->leaf = NULL; 468 } 469 470 fn->fn_sernum = sernum; 471 472 return fn; 473 } 474 475 /* 476 * We have more bits to go 477 */ 478 479 /* Try to walk down on tree. */ 480 fn->fn_sernum = sernum; 481 dir = addr_bit_set(addr, fn->fn_bit); 482 pn = fn; 483 fn = dir ? fn->right: fn->left; 484 } while (fn); 485 486 /* 487 * We walked to the bottom of tree. 488 * Create new leaf node without children. 489 */ 490 491 ln = node_alloc(); 492 493 if (ln == NULL) 494 return NULL; 495 ln->fn_bit = plen; 496 497 ln->parent = pn; 498 ln->fn_sernum = sernum; 499 500 if (dir) 501 pn->right = ln; 502 else 503 pn->left = ln; 504 505 return ln; 506 507 508 insert_above: 509 /* 510 * split since we don't have a common prefix anymore or 511 * we have a less significant route. 512 * we've to insert an intermediate node on the list 513 * this new node will point to the one we need to create 514 * and the current 515 */ 516 517 pn = fn->parent; 518 519 /* find 1st bit in difference between the 2 addrs. 520 521 See comment in __ipv6_addr_diff: bit may be an invalid value, 522 but if it is >= plen, the value is ignored in any case. 523 */ 524 525 bit = __ipv6_addr_diff(addr, &key->addr, addrlen); 526 527 /* 528 * (intermediate)[in] 529 * / \ 530 * (new leaf node)[ln] (old node)[fn] 531 */ 532 if (plen > bit) { 533 in = node_alloc(); 534 ln = node_alloc(); 535 536 if (in == NULL || ln == NULL) { 537 if (in) 538 node_free(in); 539 if (ln) 540 node_free(ln); 541 return NULL; 542 } 543 544 /* 545 * new intermediate node. 546 * RTN_RTINFO will 547 * be off since that an address that chooses one of 548 * the branches would not match less specific routes 549 * in the other branch 550 */ 551 552 in->fn_bit = bit; 553 554 in->parent = pn; 555 in->leaf = fn->leaf; 556 atomic_inc(&in->leaf->rt6i_ref); 557 558 in->fn_sernum = sernum; 559 560 /* update parent pointer */ 561 if (dir) 562 pn->right = in; 563 else 564 pn->left = in; 565 566 ln->fn_bit = plen; 567 568 ln->parent = in; 569 fn->parent = in; 570 571 ln->fn_sernum = sernum; 572 573 if (addr_bit_set(addr, bit)) { 574 in->right = ln; 575 in->left = fn; 576 } else { 577 in->left = ln; 578 in->right = fn; 579 } 580 } else { /* plen <= bit */ 581 582 /* 583 * (new leaf node)[ln] 584 * / \ 585 * (old node)[fn] NULL 586 */ 587 588 ln = node_alloc(); 589 590 if (ln == NULL) 591 return NULL; 592 593 ln->fn_bit = plen; 594 595 ln->parent = pn; 596 597 ln->fn_sernum = sernum; 598 599 if (dir) 600 pn->right = ln; 601 else 602 pn->left = ln; 603 604 if (addr_bit_set(&key->addr, plen)) 605 ln->right = fn; 606 else 607 ln->left = fn; 608 609 fn->parent = ln; 610 } 611 return ln; 612 } 613 614 /* 615 * Insert routing information in a node. 616 */ 617 618 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, 619 struct nl_info *info) 620 { 621 struct rt6_info *iter = NULL; 622 struct rt6_info **ins; 623 624 ins = &fn->leaf; 625 626 for (iter = fn->leaf; iter; iter=iter->u.dst.rt6_next) { 627 /* 628 * Search for duplicates 629 */ 630 631 if (iter->rt6i_metric == rt->rt6i_metric) { 632 /* 633 * Same priority level 634 */ 635 636 if (iter->rt6i_dev == rt->rt6i_dev && 637 iter->rt6i_idev == rt->rt6i_idev && 638 ipv6_addr_equal(&iter->rt6i_gateway, 639 &rt->rt6i_gateway)) { 640 if (!(iter->rt6i_flags&RTF_EXPIRES)) 641 return -EEXIST; 642 iter->rt6i_expires = rt->rt6i_expires; 643 if (!(rt->rt6i_flags&RTF_EXPIRES)) { 644 iter->rt6i_flags &= ~RTF_EXPIRES; 645 iter->rt6i_expires = 0; 646 } 647 return -EEXIST; 648 } 649 } 650 651 if (iter->rt6i_metric > rt->rt6i_metric) 652 break; 653 654 ins = &iter->u.dst.rt6_next; 655 } 656 657 /* Reset round-robin state, if necessary */ 658 if (ins == &fn->leaf) 659 fn->rr_ptr = NULL; 660 661 /* 662 * insert node 663 */ 664 665 rt->u.dst.rt6_next = iter; 666 *ins = rt; 667 rt->rt6i_node = fn; 668 atomic_inc(&rt->rt6i_ref); 669 inet6_rt_notify(RTM_NEWROUTE, rt, info); 670 rt6_stats.fib_rt_entries++; 671 672 if ((fn->fn_flags & RTN_RTINFO) == 0) { 673 rt6_stats.fib_route_nodes++; 674 fn->fn_flags |= RTN_RTINFO; 675 } 676 677 return 0; 678 } 679 680 static __inline__ void fib6_start_gc(struct rt6_info *rt) 681 { 682 if (ip6_fib_timer.expires == 0 && 683 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE))) 684 mod_timer(&ip6_fib_timer, jiffies + 685 init_net.ipv6.sysctl.ip6_rt_gc_interval); 686 } 687 688 void fib6_force_start_gc(void) 689 { 690 if (ip6_fib_timer.expires == 0) 691 mod_timer(&ip6_fib_timer, jiffies + 692 init_net.ipv6.sysctl.ip6_rt_gc_interval); 693 } 694 695 /* 696 * Add routing information to the routing tree. 697 * <destination addr>/<source addr> 698 * with source addr info in sub-trees 699 */ 700 701 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info) 702 { 703 struct fib6_node *fn, *pn = NULL; 704 int err = -ENOMEM; 705 706 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr), 707 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst)); 708 709 if (fn == NULL) 710 goto out; 711 712 pn = fn; 713 714 #ifdef CONFIG_IPV6_SUBTREES 715 if (rt->rt6i_src.plen) { 716 struct fib6_node *sn; 717 718 if (fn->subtree == NULL) { 719 struct fib6_node *sfn; 720 721 /* 722 * Create subtree. 723 * 724 * fn[main tree] 725 * | 726 * sfn[subtree root] 727 * \ 728 * sn[new leaf node] 729 */ 730 731 /* Create subtree root node */ 732 sfn = node_alloc(); 733 if (sfn == NULL) 734 goto st_failure; 735 736 sfn->leaf = &ip6_null_entry; 737 atomic_inc(&ip6_null_entry.rt6i_ref); 738 sfn->fn_flags = RTN_ROOT; 739 sfn->fn_sernum = fib6_new_sernum(); 740 741 /* Now add the first leaf node to new subtree */ 742 743 sn = fib6_add_1(sfn, &rt->rt6i_src.addr, 744 sizeof(struct in6_addr), rt->rt6i_src.plen, 745 offsetof(struct rt6_info, rt6i_src)); 746 747 if (sn == NULL) { 748 /* If it is failed, discard just allocated 749 root, and then (in st_failure) stale node 750 in main tree. 751 */ 752 node_free(sfn); 753 goto st_failure; 754 } 755 756 /* Now link new subtree to main tree */ 757 sfn->parent = fn; 758 fn->subtree = sfn; 759 } else { 760 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, 761 sizeof(struct in6_addr), rt->rt6i_src.plen, 762 offsetof(struct rt6_info, rt6i_src)); 763 764 if (sn == NULL) 765 goto st_failure; 766 } 767 768 if (fn->leaf == NULL) { 769 fn->leaf = rt; 770 atomic_inc(&rt->rt6i_ref); 771 } 772 fn = sn; 773 } 774 #endif 775 776 err = fib6_add_rt2node(fn, rt, info); 777 778 if (err == 0) { 779 fib6_start_gc(rt); 780 if (!(rt->rt6i_flags&RTF_CACHE)) 781 fib6_prune_clones(pn, rt); 782 } 783 784 out: 785 if (err) { 786 #ifdef CONFIG_IPV6_SUBTREES 787 /* 788 * If fib6_add_1 has cleared the old leaf pointer in the 789 * super-tree leaf node we have to find a new one for it. 790 */ 791 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) { 792 pn->leaf = fib6_find_prefix(pn); 793 #if RT6_DEBUG >= 2 794 if (!pn->leaf) { 795 BUG_TRAP(pn->leaf != NULL); 796 pn->leaf = &ip6_null_entry; 797 } 798 #endif 799 atomic_inc(&pn->leaf->rt6i_ref); 800 } 801 #endif 802 dst_free(&rt->u.dst); 803 } 804 return err; 805 806 #ifdef CONFIG_IPV6_SUBTREES 807 /* Subtree creation failed, probably main tree node 808 is orphan. If it is, shoot it. 809 */ 810 st_failure: 811 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) 812 fib6_repair_tree(fn); 813 dst_free(&rt->u.dst); 814 return err; 815 #endif 816 } 817 818 /* 819 * Routing tree lookup 820 * 821 */ 822 823 struct lookup_args { 824 int offset; /* key offset on rt6_info */ 825 struct in6_addr *addr; /* search key */ 826 }; 827 828 static struct fib6_node * fib6_lookup_1(struct fib6_node *root, 829 struct lookup_args *args) 830 { 831 struct fib6_node *fn; 832 __be32 dir; 833 834 if (unlikely(args->offset == 0)) 835 return NULL; 836 837 /* 838 * Descend on a tree 839 */ 840 841 fn = root; 842 843 for (;;) { 844 struct fib6_node *next; 845 846 dir = addr_bit_set(args->addr, fn->fn_bit); 847 848 next = dir ? fn->right : fn->left; 849 850 if (next) { 851 fn = next; 852 continue; 853 } 854 855 break; 856 } 857 858 while(fn) { 859 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) { 860 struct rt6key *key; 861 862 key = (struct rt6key *) ((u8 *) fn->leaf + 863 args->offset); 864 865 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 866 #ifdef CONFIG_IPV6_SUBTREES 867 if (fn->subtree) 868 fn = fib6_lookup_1(fn->subtree, args + 1); 869 #endif 870 if (!fn || fn->fn_flags & RTN_RTINFO) 871 return fn; 872 } 873 } 874 875 if (fn->fn_flags & RTN_ROOT) 876 break; 877 878 fn = fn->parent; 879 } 880 881 return NULL; 882 } 883 884 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr, 885 struct in6_addr *saddr) 886 { 887 struct fib6_node *fn; 888 struct lookup_args args[] = { 889 { 890 .offset = offsetof(struct rt6_info, rt6i_dst), 891 .addr = daddr, 892 }, 893 #ifdef CONFIG_IPV6_SUBTREES 894 { 895 .offset = offsetof(struct rt6_info, rt6i_src), 896 .addr = saddr, 897 }, 898 #endif 899 { 900 .offset = 0, /* sentinel */ 901 } 902 }; 903 904 fn = fib6_lookup_1(root, daddr ? args : args + 1); 905 906 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT) 907 fn = root; 908 909 return fn; 910 } 911 912 /* 913 * Get node with specified destination prefix (and source prefix, 914 * if subtrees are used) 915 */ 916 917 918 static struct fib6_node * fib6_locate_1(struct fib6_node *root, 919 struct in6_addr *addr, 920 int plen, int offset) 921 { 922 struct fib6_node *fn; 923 924 for (fn = root; fn ; ) { 925 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); 926 927 /* 928 * Prefix match 929 */ 930 if (plen < fn->fn_bit || 931 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 932 return NULL; 933 934 if (plen == fn->fn_bit) 935 return fn; 936 937 /* 938 * We have more bits to go 939 */ 940 if (addr_bit_set(addr, fn->fn_bit)) 941 fn = fn->right; 942 else 943 fn = fn->left; 944 } 945 return NULL; 946 } 947 948 struct fib6_node * fib6_locate(struct fib6_node *root, 949 struct in6_addr *daddr, int dst_len, 950 struct in6_addr *saddr, int src_len) 951 { 952 struct fib6_node *fn; 953 954 fn = fib6_locate_1(root, daddr, dst_len, 955 offsetof(struct rt6_info, rt6i_dst)); 956 957 #ifdef CONFIG_IPV6_SUBTREES 958 if (src_len) { 959 BUG_TRAP(saddr!=NULL); 960 if (fn && fn->subtree) 961 fn = fib6_locate_1(fn->subtree, saddr, src_len, 962 offsetof(struct rt6_info, rt6i_src)); 963 } 964 #endif 965 966 if (fn && fn->fn_flags&RTN_RTINFO) 967 return fn; 968 969 return NULL; 970 } 971 972 973 /* 974 * Deletion 975 * 976 */ 977 978 static struct rt6_info * fib6_find_prefix(struct fib6_node *fn) 979 { 980 if (fn->fn_flags&RTN_ROOT) 981 return &ip6_null_entry; 982 983 while(fn) { 984 if(fn->left) 985 return fn->left->leaf; 986 987 if(fn->right) 988 return fn->right->leaf; 989 990 fn = FIB6_SUBTREE(fn); 991 } 992 return NULL; 993 } 994 995 /* 996 * Called to trim the tree of intermediate nodes when possible. "fn" 997 * is the node we want to try and remove. 998 */ 999 1000 static struct fib6_node * fib6_repair_tree(struct fib6_node *fn) 1001 { 1002 int children; 1003 int nstate; 1004 struct fib6_node *child, *pn; 1005 struct fib6_walker_t *w; 1006 int iter = 0; 1007 1008 for (;;) { 1009 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1010 iter++; 1011 1012 BUG_TRAP(!(fn->fn_flags&RTN_RTINFO)); 1013 BUG_TRAP(!(fn->fn_flags&RTN_TL_ROOT)); 1014 BUG_TRAP(fn->leaf==NULL); 1015 1016 children = 0; 1017 child = NULL; 1018 if (fn->right) child = fn->right, children |= 1; 1019 if (fn->left) child = fn->left, children |= 2; 1020 1021 if (children == 3 || FIB6_SUBTREE(fn) 1022 #ifdef CONFIG_IPV6_SUBTREES 1023 /* Subtree root (i.e. fn) may have one child */ 1024 || (children && fn->fn_flags&RTN_ROOT) 1025 #endif 1026 ) { 1027 fn->leaf = fib6_find_prefix(fn); 1028 #if RT6_DEBUG >= 2 1029 if (fn->leaf==NULL) { 1030 BUG_TRAP(fn->leaf); 1031 fn->leaf = &ip6_null_entry; 1032 } 1033 #endif 1034 atomic_inc(&fn->leaf->rt6i_ref); 1035 return fn->parent; 1036 } 1037 1038 pn = fn->parent; 1039 #ifdef CONFIG_IPV6_SUBTREES 1040 if (FIB6_SUBTREE(pn) == fn) { 1041 BUG_TRAP(fn->fn_flags&RTN_ROOT); 1042 FIB6_SUBTREE(pn) = NULL; 1043 nstate = FWS_L; 1044 } else { 1045 BUG_TRAP(!(fn->fn_flags&RTN_ROOT)); 1046 #endif 1047 if (pn->right == fn) pn->right = child; 1048 else if (pn->left == fn) pn->left = child; 1049 #if RT6_DEBUG >= 2 1050 else BUG_TRAP(0); 1051 #endif 1052 if (child) 1053 child->parent = pn; 1054 nstate = FWS_R; 1055 #ifdef CONFIG_IPV6_SUBTREES 1056 } 1057 #endif 1058 1059 read_lock(&fib6_walker_lock); 1060 FOR_WALKERS(w) { 1061 if (child == NULL) { 1062 if (w->root == fn) { 1063 w->root = w->node = NULL; 1064 RT6_TRACE("W %p adjusted by delroot 1\n", w); 1065 } else if (w->node == fn) { 1066 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1067 w->node = pn; 1068 w->state = nstate; 1069 } 1070 } else { 1071 if (w->root == fn) { 1072 w->root = child; 1073 RT6_TRACE("W %p adjusted by delroot 2\n", w); 1074 } 1075 if (w->node == fn) { 1076 w->node = child; 1077 if (children&2) { 1078 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1079 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT; 1080 } else { 1081 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1082 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT; 1083 } 1084 } 1085 } 1086 } 1087 read_unlock(&fib6_walker_lock); 1088 1089 node_free(fn); 1090 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn)) 1091 return pn; 1092 1093 rt6_release(pn->leaf); 1094 pn->leaf = NULL; 1095 fn = pn; 1096 } 1097 } 1098 1099 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, 1100 struct nl_info *info) 1101 { 1102 struct fib6_walker_t *w; 1103 struct rt6_info *rt = *rtp; 1104 1105 RT6_TRACE("fib6_del_route\n"); 1106 1107 /* Unlink it */ 1108 *rtp = rt->u.dst.rt6_next; 1109 rt->rt6i_node = NULL; 1110 rt6_stats.fib_rt_entries--; 1111 rt6_stats.fib_discarded_routes++; 1112 1113 /* Reset round-robin state, if necessary */ 1114 if (fn->rr_ptr == rt) 1115 fn->rr_ptr = NULL; 1116 1117 /* Adjust walkers */ 1118 read_lock(&fib6_walker_lock); 1119 FOR_WALKERS(w) { 1120 if (w->state == FWS_C && w->leaf == rt) { 1121 RT6_TRACE("walker %p adjusted by delroute\n", w); 1122 w->leaf = rt->u.dst.rt6_next; 1123 if (w->leaf == NULL) 1124 w->state = FWS_U; 1125 } 1126 } 1127 read_unlock(&fib6_walker_lock); 1128 1129 rt->u.dst.rt6_next = NULL; 1130 1131 /* If it was last route, expunge its radix tree node */ 1132 if (fn->leaf == NULL) { 1133 fn->fn_flags &= ~RTN_RTINFO; 1134 rt6_stats.fib_route_nodes--; 1135 fn = fib6_repair_tree(fn); 1136 } 1137 1138 if (atomic_read(&rt->rt6i_ref) != 1) { 1139 /* This route is used as dummy address holder in some split 1140 * nodes. It is not leaked, but it still holds other resources, 1141 * which must be released in time. So, scan ascendant nodes 1142 * and replace dummy references to this route with references 1143 * to still alive ones. 1144 */ 1145 while (fn) { 1146 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) { 1147 fn->leaf = fib6_find_prefix(fn); 1148 atomic_inc(&fn->leaf->rt6i_ref); 1149 rt6_release(rt); 1150 } 1151 fn = fn->parent; 1152 } 1153 /* No more references are possible at this point. */ 1154 BUG_ON(atomic_read(&rt->rt6i_ref) != 1); 1155 } 1156 1157 inet6_rt_notify(RTM_DELROUTE, rt, info); 1158 rt6_release(rt); 1159 } 1160 1161 int fib6_del(struct rt6_info *rt, struct nl_info *info) 1162 { 1163 struct fib6_node *fn = rt->rt6i_node; 1164 struct rt6_info **rtp; 1165 1166 #if RT6_DEBUG >= 2 1167 if (rt->u.dst.obsolete>0) { 1168 BUG_TRAP(fn==NULL); 1169 return -ENOENT; 1170 } 1171 #endif 1172 if (fn == NULL || rt == &ip6_null_entry) 1173 return -ENOENT; 1174 1175 BUG_TRAP(fn->fn_flags&RTN_RTINFO); 1176 1177 if (!(rt->rt6i_flags&RTF_CACHE)) { 1178 struct fib6_node *pn = fn; 1179 #ifdef CONFIG_IPV6_SUBTREES 1180 /* clones of this route might be in another subtree */ 1181 if (rt->rt6i_src.plen) { 1182 while (!(pn->fn_flags&RTN_ROOT)) 1183 pn = pn->parent; 1184 pn = pn->parent; 1185 } 1186 #endif 1187 fib6_prune_clones(pn, rt); 1188 } 1189 1190 /* 1191 * Walk the leaf entries looking for ourself 1192 */ 1193 1194 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.dst.rt6_next) { 1195 if (*rtp == rt) { 1196 fib6_del_route(fn, rtp, info); 1197 return 0; 1198 } 1199 } 1200 return -ENOENT; 1201 } 1202 1203 /* 1204 * Tree traversal function. 1205 * 1206 * Certainly, it is not interrupt safe. 1207 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 1208 * It means, that we can modify tree during walking 1209 * and use this function for garbage collection, clone pruning, 1210 * cleaning tree when a device goes down etc. etc. 1211 * 1212 * It guarantees that every node will be traversed, 1213 * and that it will be traversed only once. 1214 * 1215 * Callback function w->func may return: 1216 * 0 -> continue walking. 1217 * positive value -> walking is suspended (used by tree dumps, 1218 * and probably by gc, if it will be split to several slices) 1219 * negative value -> terminate walking. 1220 * 1221 * The function itself returns: 1222 * 0 -> walk is complete. 1223 * >0 -> walk is incomplete (i.e. suspended) 1224 * <0 -> walk is terminated by an error. 1225 */ 1226 1227 static int fib6_walk_continue(struct fib6_walker_t *w) 1228 { 1229 struct fib6_node *fn, *pn; 1230 1231 for (;;) { 1232 fn = w->node; 1233 if (fn == NULL) 1234 return 0; 1235 1236 if (w->prune && fn != w->root && 1237 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) { 1238 w->state = FWS_C; 1239 w->leaf = fn->leaf; 1240 } 1241 switch (w->state) { 1242 #ifdef CONFIG_IPV6_SUBTREES 1243 case FWS_S: 1244 if (FIB6_SUBTREE(fn)) { 1245 w->node = FIB6_SUBTREE(fn); 1246 continue; 1247 } 1248 w->state = FWS_L; 1249 #endif 1250 case FWS_L: 1251 if (fn->left) { 1252 w->node = fn->left; 1253 w->state = FWS_INIT; 1254 continue; 1255 } 1256 w->state = FWS_R; 1257 case FWS_R: 1258 if (fn->right) { 1259 w->node = fn->right; 1260 w->state = FWS_INIT; 1261 continue; 1262 } 1263 w->state = FWS_C; 1264 w->leaf = fn->leaf; 1265 case FWS_C: 1266 if (w->leaf && fn->fn_flags&RTN_RTINFO) { 1267 int err = w->func(w); 1268 if (err) 1269 return err; 1270 continue; 1271 } 1272 w->state = FWS_U; 1273 case FWS_U: 1274 if (fn == w->root) 1275 return 0; 1276 pn = fn->parent; 1277 w->node = pn; 1278 #ifdef CONFIG_IPV6_SUBTREES 1279 if (FIB6_SUBTREE(pn) == fn) { 1280 BUG_TRAP(fn->fn_flags&RTN_ROOT); 1281 w->state = FWS_L; 1282 continue; 1283 } 1284 #endif 1285 if (pn->left == fn) { 1286 w->state = FWS_R; 1287 continue; 1288 } 1289 if (pn->right == fn) { 1290 w->state = FWS_C; 1291 w->leaf = w->node->leaf; 1292 continue; 1293 } 1294 #if RT6_DEBUG >= 2 1295 BUG_TRAP(0); 1296 #endif 1297 } 1298 } 1299 } 1300 1301 static int fib6_walk(struct fib6_walker_t *w) 1302 { 1303 int res; 1304 1305 w->state = FWS_INIT; 1306 w->node = w->root; 1307 1308 fib6_walker_link(w); 1309 res = fib6_walk_continue(w); 1310 if (res <= 0) 1311 fib6_walker_unlink(w); 1312 return res; 1313 } 1314 1315 static int fib6_clean_node(struct fib6_walker_t *w) 1316 { 1317 struct nl_info info = { 1318 .nl_net = &init_net, 1319 }; 1320 int res; 1321 struct rt6_info *rt; 1322 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w); 1323 1324 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) { 1325 res = c->func(rt, c->arg); 1326 if (res < 0) { 1327 w->leaf = rt; 1328 res = fib6_del(rt, &info); 1329 if (res) { 1330 #if RT6_DEBUG >= 2 1331 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res); 1332 #endif 1333 continue; 1334 } 1335 return 0; 1336 } 1337 BUG_TRAP(res==0); 1338 } 1339 w->leaf = rt; 1340 return 0; 1341 } 1342 1343 /* 1344 * Convenient frontend to tree walker. 1345 * 1346 * func is called on each route. 1347 * It may return -1 -> delete this route. 1348 * 0 -> continue walking 1349 * 1350 * prune==1 -> only immediate children of node (certainly, 1351 * ignoring pure split nodes) will be scanned. 1352 */ 1353 1354 static void fib6_clean_tree(struct fib6_node *root, 1355 int (*func)(struct rt6_info *, void *arg), 1356 int prune, void *arg) 1357 { 1358 struct fib6_cleaner_t c; 1359 1360 c.w.root = root; 1361 c.w.func = fib6_clean_node; 1362 c.w.prune = prune; 1363 c.func = func; 1364 c.arg = arg; 1365 1366 fib6_walk(&c.w); 1367 } 1368 1369 void fib6_clean_all(int (*func)(struct rt6_info *, void *arg), 1370 int prune, void *arg) 1371 { 1372 struct fib6_table *table; 1373 struct hlist_node *node; 1374 unsigned int h; 1375 1376 rcu_read_lock(); 1377 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 1378 hlist_for_each_entry_rcu(table, node, &fib_table_hash[h], 1379 tb6_hlist) { 1380 write_lock_bh(&table->tb6_lock); 1381 fib6_clean_tree(&table->tb6_root, func, prune, arg); 1382 write_unlock_bh(&table->tb6_lock); 1383 } 1384 } 1385 rcu_read_unlock(); 1386 } 1387 1388 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1389 { 1390 if (rt->rt6i_flags & RTF_CACHE) { 1391 RT6_TRACE("pruning clone %p\n", rt); 1392 return -1; 1393 } 1394 1395 return 0; 1396 } 1397 1398 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt) 1399 { 1400 fib6_clean_tree(fn, fib6_prune_clone, 1, rt); 1401 } 1402 1403 /* 1404 * Garbage collection 1405 */ 1406 1407 static struct fib6_gc_args 1408 { 1409 int timeout; 1410 int more; 1411 } gc_args; 1412 1413 static int fib6_age(struct rt6_info *rt, void *arg) 1414 { 1415 unsigned long now = jiffies; 1416 1417 /* 1418 * check addrconf expiration here. 1419 * Routes are expired even if they are in use. 1420 * 1421 * Also age clones. Note, that clones are aged out 1422 * only if they are not in use now. 1423 */ 1424 1425 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) { 1426 if (time_after(now, rt->rt6i_expires)) { 1427 RT6_TRACE("expiring %p\n", rt); 1428 return -1; 1429 } 1430 gc_args.more++; 1431 } else if (rt->rt6i_flags & RTF_CACHE) { 1432 if (atomic_read(&rt->u.dst.__refcnt) == 0 && 1433 time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) { 1434 RT6_TRACE("aging clone %p\n", rt); 1435 return -1; 1436 } else if ((rt->rt6i_flags & RTF_GATEWAY) && 1437 (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) { 1438 RT6_TRACE("purging route %p via non-router but gateway\n", 1439 rt); 1440 return -1; 1441 } 1442 gc_args.more++; 1443 } 1444 1445 return 0; 1446 } 1447 1448 static DEFINE_SPINLOCK(fib6_gc_lock); 1449 1450 void fib6_run_gc(unsigned long dummy) 1451 { 1452 if (dummy != ~0UL) { 1453 spin_lock_bh(&fib6_gc_lock); 1454 gc_args.timeout = dummy ? (int)dummy : 1455 init_net.ipv6.sysctl.ip6_rt_gc_interval; 1456 } else { 1457 local_bh_disable(); 1458 if (!spin_trylock(&fib6_gc_lock)) { 1459 mod_timer(&ip6_fib_timer, jiffies + HZ); 1460 local_bh_enable(); 1461 return; 1462 } 1463 gc_args.timeout = init_net.ipv6.sysctl.ip6_rt_gc_interval; 1464 } 1465 gc_args.more = 0; 1466 1467 ndisc_dst_gc(&gc_args.more); 1468 fib6_clean_all(fib6_age, 0, NULL); 1469 1470 if (gc_args.more) 1471 mod_timer(&ip6_fib_timer, jiffies + 1472 init_net.ipv6.sysctl.ip6_rt_gc_interval); 1473 else { 1474 del_timer(&ip6_fib_timer); 1475 ip6_fib_timer.expires = 0; 1476 } 1477 spin_unlock_bh(&fib6_gc_lock); 1478 } 1479 1480 int __init fib6_init(void) 1481 { 1482 int ret; 1483 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1484 sizeof(struct fib6_node), 1485 0, SLAB_HWCACHE_ALIGN, 1486 NULL); 1487 if (!fib6_node_kmem) 1488 return -ENOMEM; 1489 1490 fib6_tables_init(); 1491 1492 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib); 1493 if (ret) 1494 goto out_kmem_cache_create; 1495 out: 1496 return ret; 1497 1498 out_kmem_cache_create: 1499 kmem_cache_destroy(fib6_node_kmem); 1500 goto out; 1501 } 1502 1503 void fib6_gc_cleanup(void) 1504 { 1505 del_timer(&ip6_fib_timer); 1506 kmem_cache_destroy(fib6_node_kmem); 1507 } 1508