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 if ((fn = kmem_cache_alloc(fib6_node_kmem, GFP_ATOMIC)) != NULL) 154 memset(fn, 0, sizeof(struct fib6_node)); 155 156 return fn; 157 } 158 159 static __inline__ void node_free(struct fib6_node * fn) 160 { 161 kmem_cache_free(fib6_node_kmem, fn); 162 } 163 164 static __inline__ void rt6_release(struct rt6_info *rt) 165 { 166 if (atomic_dec_and_test(&rt->rt6i_ref)) 167 dst_free(&rt->u.dst); 168 } 169 170 static struct fib6_table fib6_main_tbl = { 171 .tb6_id = RT6_TABLE_MAIN, 172 .tb6_root = { 173 .leaf = &ip6_null_entry, 174 .fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO, 175 }, 176 }; 177 178 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 179 #define FIB_TABLE_HASHSZ 256 180 #else 181 #define FIB_TABLE_HASHSZ 1 182 #endif 183 static struct hlist_head fib_table_hash[FIB_TABLE_HASHSZ]; 184 185 static void fib6_link_table(struct fib6_table *tb) 186 { 187 unsigned int h; 188 189 /* 190 * Initialize table lock at a single place to give lockdep a key, 191 * tables aren't visible prior to being linked to the list. 192 */ 193 rwlock_init(&tb->tb6_lock); 194 195 h = tb->tb6_id & (FIB_TABLE_HASHSZ - 1); 196 197 /* 198 * No protection necessary, this is the only list mutatation 199 * operation, tables never disappear once they exist. 200 */ 201 hlist_add_head_rcu(&tb->tb6_hlist, &fib_table_hash[h]); 202 } 203 204 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 205 static struct fib6_table fib6_local_tbl = { 206 .tb6_id = RT6_TABLE_LOCAL, 207 .tb6_root = { 208 .leaf = &ip6_null_entry, 209 .fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO, 210 }, 211 }; 212 213 static struct fib6_table *fib6_alloc_table(u32 id) 214 { 215 struct fib6_table *table; 216 217 table = kzalloc(sizeof(*table), GFP_ATOMIC); 218 if (table != NULL) { 219 table->tb6_id = id; 220 table->tb6_root.leaf = &ip6_null_entry; 221 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 222 } 223 224 return table; 225 } 226 227 struct fib6_table *fib6_new_table(u32 id) 228 { 229 struct fib6_table *tb; 230 231 if (id == 0) 232 id = RT6_TABLE_MAIN; 233 tb = fib6_get_table(id); 234 if (tb) 235 return tb; 236 237 tb = fib6_alloc_table(id); 238 if (tb != NULL) 239 fib6_link_table(tb); 240 241 return tb; 242 } 243 244 struct fib6_table *fib6_get_table(u32 id) 245 { 246 struct fib6_table *tb; 247 struct hlist_node *node; 248 unsigned int h; 249 250 if (id == 0) 251 id = RT6_TABLE_MAIN; 252 h = id & (FIB_TABLE_HASHSZ - 1); 253 rcu_read_lock(); 254 hlist_for_each_entry_rcu(tb, node, &fib_table_hash[h], tb6_hlist) { 255 if (tb->tb6_id == id) { 256 rcu_read_unlock(); 257 return tb; 258 } 259 } 260 rcu_read_unlock(); 261 262 return NULL; 263 } 264 265 static void __init fib6_tables_init(void) 266 { 267 fib6_link_table(&fib6_main_tbl); 268 fib6_link_table(&fib6_local_tbl); 269 } 270 271 #else 272 273 struct fib6_table *fib6_new_table(u32 id) 274 { 275 return fib6_get_table(id); 276 } 277 278 struct fib6_table *fib6_get_table(u32 id) 279 { 280 return &fib6_main_tbl; 281 } 282 283 struct dst_entry *fib6_rule_lookup(struct flowi *fl, int flags, 284 pol_lookup_t lookup) 285 { 286 return (struct dst_entry *) lookup(&fib6_main_tbl, fl, flags); 287 } 288 289 static void __init fib6_tables_init(void) 290 { 291 fib6_link_table(&fib6_main_tbl); 292 } 293 294 #endif 295 296 static int fib6_dump_node(struct fib6_walker_t *w) 297 { 298 int res; 299 struct rt6_info *rt; 300 301 for (rt = w->leaf; rt; rt = rt->u.next) { 302 res = rt6_dump_route(rt, w->args); 303 if (res < 0) { 304 /* Frame is full, suspend walking */ 305 w->leaf = rt; 306 return 1; 307 } 308 BUG_TRAP(res!=0); 309 } 310 w->leaf = NULL; 311 return 0; 312 } 313 314 static void fib6_dump_end(struct netlink_callback *cb) 315 { 316 struct fib6_walker_t *w = (void*)cb->args[2]; 317 318 if (w) { 319 cb->args[2] = 0; 320 kfree(w); 321 } 322 cb->done = (void*)cb->args[3]; 323 cb->args[1] = 3; 324 } 325 326 static int fib6_dump_done(struct netlink_callback *cb) 327 { 328 fib6_dump_end(cb); 329 return cb->done ? cb->done(cb) : 0; 330 } 331 332 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 333 struct netlink_callback *cb) 334 { 335 struct fib6_walker_t *w; 336 int res; 337 338 w = (void *)cb->args[2]; 339 w->root = &table->tb6_root; 340 341 if (cb->args[4] == 0) { 342 read_lock_bh(&table->tb6_lock); 343 res = fib6_walk(w); 344 read_unlock_bh(&table->tb6_lock); 345 if (res > 0) 346 cb->args[4] = 1; 347 } else { 348 read_lock_bh(&table->tb6_lock); 349 res = fib6_walk_continue(w); 350 read_unlock_bh(&table->tb6_lock); 351 if (res != 0) { 352 if (res < 0) 353 fib6_walker_unlink(w); 354 goto end; 355 } 356 fib6_walker_unlink(w); 357 cb->args[4] = 0; 358 } 359 end: 360 return res; 361 } 362 363 int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 364 { 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 s_h = cb->args[0]; 374 s_e = cb->args[1]; 375 376 w = (void *)cb->args[2]; 377 if (w == NULL) { 378 /* New dump: 379 * 380 * 1. hook callback destructor. 381 */ 382 cb->args[3] = (long)cb->done; 383 cb->done = fib6_dump_done; 384 385 /* 386 * 2. allocate and initialize walker. 387 */ 388 w = kzalloc(sizeof(*w), GFP_ATOMIC); 389 if (w == NULL) 390 return -ENOMEM; 391 w->func = fib6_dump_node; 392 cb->args[2] = (long)w; 393 } 394 395 arg.skb = skb; 396 arg.cb = cb; 397 w->args = &arg; 398 399 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 400 e = 0; 401 hlist_for_each_entry(tb, node, &fib_table_hash[h], tb6_hlist) { 402 if (e < s_e) 403 goto next; 404 res = fib6_dump_table(tb, skb, cb); 405 if (res != 0) 406 goto out; 407 next: 408 e++; 409 } 410 } 411 out: 412 cb->args[1] = e; 413 cb->args[0] = h; 414 415 res = res < 0 ? res : skb->len; 416 if (res <= 0) 417 fib6_dump_end(cb); 418 return res; 419 } 420 421 /* 422 * Routing Table 423 * 424 * return the appropriate node for a routing tree "add" operation 425 * by either creating and inserting or by returning an existing 426 * node. 427 */ 428 429 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr, 430 int addrlen, int plen, 431 int offset) 432 { 433 struct fib6_node *fn, *in, *ln; 434 struct fib6_node *pn = NULL; 435 struct rt6key *key; 436 int bit; 437 __be32 dir = 0; 438 __u32 sernum = fib6_new_sernum(); 439 440 RT6_TRACE("fib6_add_1\n"); 441 442 /* insert node in tree */ 443 444 fn = root; 445 446 do { 447 key = (struct rt6key *)((u8 *)fn->leaf + offset); 448 449 /* 450 * Prefix match 451 */ 452 if (plen < fn->fn_bit || 453 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 454 goto insert_above; 455 456 /* 457 * Exact match ? 458 */ 459 460 if (plen == fn->fn_bit) { 461 /* clean up an intermediate node */ 462 if ((fn->fn_flags & RTN_RTINFO) == 0) { 463 rt6_release(fn->leaf); 464 fn->leaf = NULL; 465 } 466 467 fn->fn_sernum = sernum; 468 469 return fn; 470 } 471 472 /* 473 * We have more bits to go 474 */ 475 476 /* Try to walk down on tree. */ 477 fn->fn_sernum = sernum; 478 dir = addr_bit_set(addr, fn->fn_bit); 479 pn = fn; 480 fn = dir ? fn->right: fn->left; 481 } while (fn); 482 483 /* 484 * We walked to the bottom of tree. 485 * Create new leaf node without children. 486 */ 487 488 ln = node_alloc(); 489 490 if (ln == NULL) 491 return NULL; 492 ln->fn_bit = plen; 493 494 ln->parent = pn; 495 ln->fn_sernum = sernum; 496 497 if (dir) 498 pn->right = ln; 499 else 500 pn->left = ln; 501 502 return ln; 503 504 505 insert_above: 506 /* 507 * split since we don't have a common prefix anymore or 508 * we have a less significant route. 509 * we've to insert an intermediate node on the list 510 * this new node will point to the one we need to create 511 * and the current 512 */ 513 514 pn = fn->parent; 515 516 /* find 1st bit in difference between the 2 addrs. 517 518 See comment in __ipv6_addr_diff: bit may be an invalid value, 519 but if it is >= plen, the value is ignored in any case. 520 */ 521 522 bit = __ipv6_addr_diff(addr, &key->addr, addrlen); 523 524 /* 525 * (intermediate)[in] 526 * / \ 527 * (new leaf node)[ln] (old node)[fn] 528 */ 529 if (plen > bit) { 530 in = node_alloc(); 531 ln = node_alloc(); 532 533 if (in == NULL || ln == NULL) { 534 if (in) 535 node_free(in); 536 if (ln) 537 node_free(ln); 538 return NULL; 539 } 540 541 /* 542 * new intermediate node. 543 * RTN_RTINFO will 544 * be off since that an address that chooses one of 545 * the branches would not match less specific routes 546 * in the other branch 547 */ 548 549 in->fn_bit = bit; 550 551 in->parent = pn; 552 in->leaf = fn->leaf; 553 atomic_inc(&in->leaf->rt6i_ref); 554 555 in->fn_sernum = sernum; 556 557 /* update parent pointer */ 558 if (dir) 559 pn->right = in; 560 else 561 pn->left = in; 562 563 ln->fn_bit = plen; 564 565 ln->parent = in; 566 fn->parent = in; 567 568 ln->fn_sernum = sernum; 569 570 if (addr_bit_set(addr, bit)) { 571 in->right = ln; 572 in->left = fn; 573 } else { 574 in->left = ln; 575 in->right = fn; 576 } 577 } else { /* plen <= bit */ 578 579 /* 580 * (new leaf node)[ln] 581 * / \ 582 * (old node)[fn] NULL 583 */ 584 585 ln = node_alloc(); 586 587 if (ln == NULL) 588 return NULL; 589 590 ln->fn_bit = plen; 591 592 ln->parent = pn; 593 594 ln->fn_sernum = sernum; 595 596 if (dir) 597 pn->right = ln; 598 else 599 pn->left = ln; 600 601 if (addr_bit_set(&key->addr, plen)) 602 ln->right = fn; 603 else 604 ln->left = fn; 605 606 fn->parent = ln; 607 } 608 return ln; 609 } 610 611 /* 612 * Insert routing information in a node. 613 */ 614 615 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, 616 struct nl_info *info) 617 { 618 struct rt6_info *iter = NULL; 619 struct rt6_info **ins; 620 621 ins = &fn->leaf; 622 623 if (fn->fn_flags&RTN_TL_ROOT && 624 fn->leaf == &ip6_null_entry && 625 !(rt->rt6i_flags & (RTF_DEFAULT | RTF_ADDRCONF)) ){ 626 fn->leaf = rt; 627 rt->u.next = NULL; 628 goto out; 629 } 630 631 for (iter = fn->leaf; iter; iter=iter->u.next) { 632 /* 633 * Search for duplicates 634 */ 635 636 if (iter->rt6i_metric == rt->rt6i_metric) { 637 /* 638 * Same priority level 639 */ 640 641 if (iter->rt6i_dev == rt->rt6i_dev && 642 iter->rt6i_idev == rt->rt6i_idev && 643 ipv6_addr_equal(&iter->rt6i_gateway, 644 &rt->rt6i_gateway)) { 645 if (!(iter->rt6i_flags&RTF_EXPIRES)) 646 return -EEXIST; 647 iter->rt6i_expires = rt->rt6i_expires; 648 if (!(rt->rt6i_flags&RTF_EXPIRES)) { 649 iter->rt6i_flags &= ~RTF_EXPIRES; 650 iter->rt6i_expires = 0; 651 } 652 return -EEXIST; 653 } 654 } 655 656 if (iter->rt6i_metric > rt->rt6i_metric) 657 break; 658 659 ins = &iter->u.next; 660 } 661 662 /* 663 * insert node 664 */ 665 666 out: 667 rt->u.next = iter; 668 *ins = rt; 669 rt->rt6i_node = fn; 670 atomic_inc(&rt->rt6i_ref); 671 inet6_rt_notify(RTM_NEWROUTE, rt, info); 672 rt6_stats.fib_rt_entries++; 673 674 if ((fn->fn_flags & RTN_RTINFO) == 0) { 675 rt6_stats.fib_route_nodes++; 676 fn->fn_flags |= RTN_RTINFO; 677 } 678 679 return 0; 680 } 681 682 static __inline__ void fib6_start_gc(struct rt6_info *rt) 683 { 684 if (ip6_fib_timer.expires == 0 && 685 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE))) 686 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval); 687 } 688 689 void fib6_force_start_gc(void) 690 { 691 if (ip6_fib_timer.expires == 0) 692 mod_timer(&ip6_fib_timer, jiffies + 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.next; 1109 rt->rt6i_node = NULL; 1110 rt6_stats.fib_rt_entries--; 1111 rt6_stats.fib_discarded_routes++; 1112 1113 /* Adjust walkers */ 1114 read_lock(&fib6_walker_lock); 1115 FOR_WALKERS(w) { 1116 if (w->state == FWS_C && w->leaf == rt) { 1117 RT6_TRACE("walker %p adjusted by delroute\n", w); 1118 w->leaf = rt->u.next; 1119 if (w->leaf == NULL) 1120 w->state = FWS_U; 1121 } 1122 } 1123 read_unlock(&fib6_walker_lock); 1124 1125 rt->u.next = NULL; 1126 1127 if (fn->leaf == NULL && fn->fn_flags&RTN_TL_ROOT) 1128 fn->leaf = &ip6_null_entry; 1129 1130 /* If it was last route, expunge its radix tree node */ 1131 if (fn->leaf == NULL) { 1132 fn->fn_flags &= ~RTN_RTINFO; 1133 rt6_stats.fib_route_nodes--; 1134 fn = fib6_repair_tree(fn); 1135 } 1136 1137 if (atomic_read(&rt->rt6i_ref) != 1) { 1138 /* This route is used as dummy address holder in some split 1139 * nodes. It is not leaked, but it still holds other resources, 1140 * which must be released in time. So, scan ascendant nodes 1141 * and replace dummy references to this route with references 1142 * to still alive ones. 1143 */ 1144 while (fn) { 1145 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) { 1146 fn->leaf = fib6_find_prefix(fn); 1147 atomic_inc(&fn->leaf->rt6i_ref); 1148 rt6_release(rt); 1149 } 1150 fn = fn->parent; 1151 } 1152 /* No more references are possible at this point. */ 1153 if (atomic_read(&rt->rt6i_ref) != 1) BUG(); 1154 } 1155 1156 inet6_rt_notify(RTM_DELROUTE, rt, info); 1157 rt6_release(rt); 1158 } 1159 1160 int fib6_del(struct rt6_info *rt, struct nl_info *info) 1161 { 1162 struct fib6_node *fn = rt->rt6i_node; 1163 struct rt6_info **rtp; 1164 1165 #if RT6_DEBUG >= 2 1166 if (rt->u.dst.obsolete>0) { 1167 BUG_TRAP(fn==NULL); 1168 return -ENOENT; 1169 } 1170 #endif 1171 if (fn == NULL || rt == &ip6_null_entry) 1172 return -ENOENT; 1173 1174 BUG_TRAP(fn->fn_flags&RTN_RTINFO); 1175 1176 if (!(rt->rt6i_flags&RTF_CACHE)) { 1177 struct fib6_node *pn = fn; 1178 #ifdef CONFIG_IPV6_SUBTREES 1179 /* clones of this route might be in another subtree */ 1180 if (rt->rt6i_src.plen) { 1181 while (!(pn->fn_flags&RTN_ROOT)) 1182 pn = pn->parent; 1183 pn = pn->parent; 1184 } 1185 #endif 1186 fib6_prune_clones(pn, rt); 1187 } 1188 1189 /* 1190 * Walk the leaf entries looking for ourself 1191 */ 1192 1193 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.next) { 1194 if (*rtp == rt) { 1195 fib6_del_route(fn, rtp, info); 1196 return 0; 1197 } 1198 } 1199 return -ENOENT; 1200 } 1201 1202 /* 1203 * Tree traversal function. 1204 * 1205 * Certainly, it is not interrupt safe. 1206 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 1207 * It means, that we can modify tree during walking 1208 * and use this function for garbage collection, clone pruning, 1209 * cleaning tree when a device goes down etc. etc. 1210 * 1211 * It guarantees that every node will be traversed, 1212 * and that it will be traversed only once. 1213 * 1214 * Callback function w->func may return: 1215 * 0 -> continue walking. 1216 * positive value -> walking is suspended (used by tree dumps, 1217 * and probably by gc, if it will be split to several slices) 1218 * negative value -> terminate walking. 1219 * 1220 * The function itself returns: 1221 * 0 -> walk is complete. 1222 * >0 -> walk is incomplete (i.e. suspended) 1223 * <0 -> walk is terminated by an error. 1224 */ 1225 1226 static int fib6_walk_continue(struct fib6_walker_t *w) 1227 { 1228 struct fib6_node *fn, *pn; 1229 1230 for (;;) { 1231 fn = w->node; 1232 if (fn == NULL) 1233 return 0; 1234 1235 if (w->prune && fn != w->root && 1236 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) { 1237 w->state = FWS_C; 1238 w->leaf = fn->leaf; 1239 } 1240 switch (w->state) { 1241 #ifdef CONFIG_IPV6_SUBTREES 1242 case FWS_S: 1243 if (FIB6_SUBTREE(fn)) { 1244 w->node = FIB6_SUBTREE(fn); 1245 continue; 1246 } 1247 w->state = FWS_L; 1248 #endif 1249 case FWS_L: 1250 if (fn->left) { 1251 w->node = fn->left; 1252 w->state = FWS_INIT; 1253 continue; 1254 } 1255 w->state = FWS_R; 1256 case FWS_R: 1257 if (fn->right) { 1258 w->node = fn->right; 1259 w->state = FWS_INIT; 1260 continue; 1261 } 1262 w->state = FWS_C; 1263 w->leaf = fn->leaf; 1264 case FWS_C: 1265 if (w->leaf && fn->fn_flags&RTN_RTINFO) { 1266 int err = w->func(w); 1267 if (err) 1268 return err; 1269 continue; 1270 } 1271 w->state = FWS_U; 1272 case FWS_U: 1273 if (fn == w->root) 1274 return 0; 1275 pn = fn->parent; 1276 w->node = pn; 1277 #ifdef CONFIG_IPV6_SUBTREES 1278 if (FIB6_SUBTREE(pn) == fn) { 1279 BUG_TRAP(fn->fn_flags&RTN_ROOT); 1280 w->state = FWS_L; 1281 continue; 1282 } 1283 #endif 1284 if (pn->left == fn) { 1285 w->state = FWS_R; 1286 continue; 1287 } 1288 if (pn->right == fn) { 1289 w->state = FWS_C; 1290 w->leaf = w->node->leaf; 1291 continue; 1292 } 1293 #if RT6_DEBUG >= 2 1294 BUG_TRAP(0); 1295 #endif 1296 } 1297 } 1298 } 1299 1300 static int fib6_walk(struct fib6_walker_t *w) 1301 { 1302 int res; 1303 1304 w->state = FWS_INIT; 1305 w->node = w->root; 1306 1307 fib6_walker_link(w); 1308 res = fib6_walk_continue(w); 1309 if (res <= 0) 1310 fib6_walker_unlink(w); 1311 return res; 1312 } 1313 1314 static int fib6_clean_node(struct fib6_walker_t *w) 1315 { 1316 int res; 1317 struct rt6_info *rt; 1318 struct fib6_cleaner_t *c = (struct fib6_cleaner_t*)w; 1319 1320 for (rt = w->leaf; rt; rt = rt->u.next) { 1321 res = c->func(rt, c->arg); 1322 if (res < 0) { 1323 w->leaf = rt; 1324 res = fib6_del(rt, NULL); 1325 if (res) { 1326 #if RT6_DEBUG >= 2 1327 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res); 1328 #endif 1329 continue; 1330 } 1331 return 0; 1332 } 1333 BUG_TRAP(res==0); 1334 } 1335 w->leaf = rt; 1336 return 0; 1337 } 1338 1339 /* 1340 * Convenient frontend to tree walker. 1341 * 1342 * func is called on each route. 1343 * It may return -1 -> delete this route. 1344 * 0 -> continue walking 1345 * 1346 * prune==1 -> only immediate children of node (certainly, 1347 * ignoring pure split nodes) will be scanned. 1348 */ 1349 1350 static void fib6_clean_tree(struct fib6_node *root, 1351 int (*func)(struct rt6_info *, void *arg), 1352 int prune, void *arg) 1353 { 1354 struct fib6_cleaner_t c; 1355 1356 c.w.root = root; 1357 c.w.func = fib6_clean_node; 1358 c.w.prune = prune; 1359 c.func = func; 1360 c.arg = arg; 1361 1362 fib6_walk(&c.w); 1363 } 1364 1365 void fib6_clean_all(int (*func)(struct rt6_info *, void *arg), 1366 int prune, void *arg) 1367 { 1368 struct fib6_table *table; 1369 struct hlist_node *node; 1370 unsigned int h; 1371 1372 rcu_read_lock(); 1373 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 1374 hlist_for_each_entry_rcu(table, node, &fib_table_hash[h], 1375 tb6_hlist) { 1376 write_lock_bh(&table->tb6_lock); 1377 fib6_clean_tree(&table->tb6_root, func, prune, arg); 1378 write_unlock_bh(&table->tb6_lock); 1379 } 1380 } 1381 rcu_read_unlock(); 1382 } 1383 1384 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1385 { 1386 if (rt->rt6i_flags & RTF_CACHE) { 1387 RT6_TRACE("pruning clone %p\n", rt); 1388 return -1; 1389 } 1390 1391 return 0; 1392 } 1393 1394 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt) 1395 { 1396 fib6_clean_tree(fn, fib6_prune_clone, 1, rt); 1397 } 1398 1399 /* 1400 * Garbage collection 1401 */ 1402 1403 static struct fib6_gc_args 1404 { 1405 int timeout; 1406 int more; 1407 } gc_args; 1408 1409 static int fib6_age(struct rt6_info *rt, void *arg) 1410 { 1411 unsigned long now = jiffies; 1412 1413 /* 1414 * check addrconf expiration here. 1415 * Routes are expired even if they are in use. 1416 * 1417 * Also age clones. Note, that clones are aged out 1418 * only if they are not in use now. 1419 */ 1420 1421 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) { 1422 if (time_after(now, rt->rt6i_expires)) { 1423 RT6_TRACE("expiring %p\n", rt); 1424 return -1; 1425 } 1426 gc_args.more++; 1427 } else if (rt->rt6i_flags & RTF_CACHE) { 1428 if (atomic_read(&rt->u.dst.__refcnt) == 0 && 1429 time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) { 1430 RT6_TRACE("aging clone %p\n", rt); 1431 return -1; 1432 } else if ((rt->rt6i_flags & RTF_GATEWAY) && 1433 (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) { 1434 RT6_TRACE("purging route %p via non-router but gateway\n", 1435 rt); 1436 return -1; 1437 } 1438 gc_args.more++; 1439 } 1440 1441 return 0; 1442 } 1443 1444 static DEFINE_SPINLOCK(fib6_gc_lock); 1445 1446 void fib6_run_gc(unsigned long dummy) 1447 { 1448 if (dummy != ~0UL) { 1449 spin_lock_bh(&fib6_gc_lock); 1450 gc_args.timeout = dummy ? (int)dummy : ip6_rt_gc_interval; 1451 } else { 1452 local_bh_disable(); 1453 if (!spin_trylock(&fib6_gc_lock)) { 1454 mod_timer(&ip6_fib_timer, jiffies + HZ); 1455 local_bh_enable(); 1456 return; 1457 } 1458 gc_args.timeout = ip6_rt_gc_interval; 1459 } 1460 gc_args.more = 0; 1461 1462 ndisc_dst_gc(&gc_args.more); 1463 fib6_clean_all(fib6_age, 0, NULL); 1464 1465 if (gc_args.more) 1466 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval); 1467 else { 1468 del_timer(&ip6_fib_timer); 1469 ip6_fib_timer.expires = 0; 1470 } 1471 spin_unlock_bh(&fib6_gc_lock); 1472 } 1473 1474 void __init fib6_init(void) 1475 { 1476 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1477 sizeof(struct fib6_node), 1478 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, 1479 NULL, NULL); 1480 1481 fib6_tables_init(); 1482 } 1483 1484 void fib6_gc_cleanup(void) 1485 { 1486 del_timer(&ip6_fib_timer); 1487 kmem_cache_destroy(fib6_node_kmem); 1488 } 1489