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 */ 22 #include <linux/errno.h> 23 #include <linux/types.h> 24 #include <linux/net.h> 25 #include <linux/route.h> 26 #include <linux/netdevice.h> 27 #include <linux/in6.h> 28 #include <linux/init.h> 29 30 #ifdef CONFIG_PROC_FS 31 #include <linux/proc_fs.h> 32 #endif 33 34 #include <net/ipv6.h> 35 #include <net/ndisc.h> 36 #include <net/addrconf.h> 37 38 #include <net/ip6_fib.h> 39 #include <net/ip6_route.h> 40 41 #define RT6_DEBUG 2 42 43 #if RT6_DEBUG >= 3 44 #define RT6_TRACE(x...) printk(KERN_DEBUG x) 45 #else 46 #define RT6_TRACE(x...) do { ; } while (0) 47 #endif 48 49 struct rt6_statistics rt6_stats; 50 51 static kmem_cache_t * fib6_node_kmem __read_mostly; 52 53 enum fib_walk_state_t 54 { 55 #ifdef CONFIG_IPV6_SUBTREES 56 FWS_S, 57 #endif 58 FWS_L, 59 FWS_R, 60 FWS_C, 61 FWS_U 62 }; 63 64 struct fib6_cleaner_t 65 { 66 struct fib6_walker_t w; 67 int (*func)(struct rt6_info *, void *arg); 68 void *arg; 69 }; 70 71 DEFINE_RWLOCK(fib6_walker_lock); 72 73 74 #ifdef CONFIG_IPV6_SUBTREES 75 #define FWS_INIT FWS_S 76 #define SUBTREE(fn) ((fn)->subtree) 77 #else 78 #define FWS_INIT FWS_L 79 #define SUBTREE(fn) NULL 80 #endif 81 82 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt); 83 static struct fib6_node * fib6_repair_tree(struct fib6_node *fn); 84 85 /* 86 * A routing update causes an increase of the serial number on the 87 * affected subtree. This allows for cached routes to be asynchronously 88 * tested when modifications are made to the destination cache as a 89 * result of redirects, path MTU changes, etc. 90 */ 91 92 static __u32 rt_sernum; 93 94 static DEFINE_TIMER(ip6_fib_timer, fib6_run_gc, 0, 0); 95 96 struct fib6_walker_t fib6_walker_list = { 97 .prev = &fib6_walker_list, 98 .next = &fib6_walker_list, 99 }; 100 101 #define FOR_WALKERS(w) for ((w)=fib6_walker_list.next; (w) != &fib6_walker_list; (w)=(w)->next) 102 103 static __inline__ u32 fib6_new_sernum(void) 104 { 105 u32 n = ++rt_sernum; 106 if ((__s32)n <= 0) 107 rt_sernum = n = 1; 108 return n; 109 } 110 111 /* 112 * Auxiliary address test functions for the radix tree. 113 * 114 * These assume a 32bit processor (although it will work on 115 * 64bit processors) 116 */ 117 118 /* 119 * test bit 120 */ 121 122 static __inline__ int addr_bit_set(void *token, int fn_bit) 123 { 124 __u32 *addr = token; 125 126 return htonl(1 << ((~fn_bit)&0x1F)) & addr[fn_bit>>5]; 127 } 128 129 static __inline__ struct fib6_node * node_alloc(void) 130 { 131 struct fib6_node *fn; 132 133 if ((fn = kmem_cache_alloc(fib6_node_kmem, SLAB_ATOMIC)) != NULL) 134 memset(fn, 0, sizeof(struct fib6_node)); 135 136 return fn; 137 } 138 139 static __inline__ void node_free(struct fib6_node * fn) 140 { 141 kmem_cache_free(fib6_node_kmem, fn); 142 } 143 144 static __inline__ void rt6_release(struct rt6_info *rt) 145 { 146 if (atomic_dec_and_test(&rt->rt6i_ref)) 147 dst_free(&rt->u.dst); 148 } 149 150 151 /* 152 * Routing Table 153 * 154 * return the appropriate node for a routing tree "add" operation 155 * by either creating and inserting or by returning an existing 156 * node. 157 */ 158 159 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr, 160 int addrlen, int plen, 161 int offset) 162 { 163 struct fib6_node *fn, *in, *ln; 164 struct fib6_node *pn = NULL; 165 struct rt6key *key; 166 int bit; 167 int dir = 0; 168 __u32 sernum = fib6_new_sernum(); 169 170 RT6_TRACE("fib6_add_1\n"); 171 172 /* insert node in tree */ 173 174 fn = root; 175 176 do { 177 key = (struct rt6key *)((u8 *)fn->leaf + offset); 178 179 /* 180 * Prefix match 181 */ 182 if (plen < fn->fn_bit || 183 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 184 goto insert_above; 185 186 /* 187 * Exact match ? 188 */ 189 190 if (plen == fn->fn_bit) { 191 /* clean up an intermediate node */ 192 if ((fn->fn_flags & RTN_RTINFO) == 0) { 193 rt6_release(fn->leaf); 194 fn->leaf = NULL; 195 } 196 197 fn->fn_sernum = sernum; 198 199 return fn; 200 } 201 202 /* 203 * We have more bits to go 204 */ 205 206 /* Try to walk down on tree. */ 207 fn->fn_sernum = sernum; 208 dir = addr_bit_set(addr, fn->fn_bit); 209 pn = fn; 210 fn = dir ? fn->right: fn->left; 211 } while (fn); 212 213 /* 214 * We walked to the bottom of tree. 215 * Create new leaf node without children. 216 */ 217 218 ln = node_alloc(); 219 220 if (ln == NULL) 221 return NULL; 222 ln->fn_bit = plen; 223 224 ln->parent = pn; 225 ln->fn_sernum = sernum; 226 227 if (dir) 228 pn->right = ln; 229 else 230 pn->left = ln; 231 232 return ln; 233 234 235 insert_above: 236 /* 237 * split since we don't have a common prefix anymore or 238 * we have a less significant route. 239 * we've to insert an intermediate node on the list 240 * this new node will point to the one we need to create 241 * and the current 242 */ 243 244 pn = fn->parent; 245 246 /* find 1st bit in difference between the 2 addrs. 247 248 See comment in __ipv6_addr_diff: bit may be an invalid value, 249 but if it is >= plen, the value is ignored in any case. 250 */ 251 252 bit = __ipv6_addr_diff(addr, &key->addr, addrlen); 253 254 /* 255 * (intermediate)[in] 256 * / \ 257 * (new leaf node)[ln] (old node)[fn] 258 */ 259 if (plen > bit) { 260 in = node_alloc(); 261 ln = node_alloc(); 262 263 if (in == NULL || ln == NULL) { 264 if (in) 265 node_free(in); 266 if (ln) 267 node_free(ln); 268 return NULL; 269 } 270 271 /* 272 * new intermediate node. 273 * RTN_RTINFO will 274 * be off since that an address that chooses one of 275 * the branches would not match less specific routes 276 * in the other branch 277 */ 278 279 in->fn_bit = bit; 280 281 in->parent = pn; 282 in->leaf = fn->leaf; 283 atomic_inc(&in->leaf->rt6i_ref); 284 285 in->fn_sernum = sernum; 286 287 /* update parent pointer */ 288 if (dir) 289 pn->right = in; 290 else 291 pn->left = in; 292 293 ln->fn_bit = plen; 294 295 ln->parent = in; 296 fn->parent = in; 297 298 ln->fn_sernum = sernum; 299 300 if (addr_bit_set(addr, bit)) { 301 in->right = ln; 302 in->left = fn; 303 } else { 304 in->left = ln; 305 in->right = fn; 306 } 307 } else { /* plen <= bit */ 308 309 /* 310 * (new leaf node)[ln] 311 * / \ 312 * (old node)[fn] NULL 313 */ 314 315 ln = node_alloc(); 316 317 if (ln == NULL) 318 return NULL; 319 320 ln->fn_bit = plen; 321 322 ln->parent = pn; 323 324 ln->fn_sernum = sernum; 325 326 if (dir) 327 pn->right = ln; 328 else 329 pn->left = ln; 330 331 if (addr_bit_set(&key->addr, plen)) 332 ln->right = fn; 333 else 334 ln->left = fn; 335 336 fn->parent = ln; 337 } 338 return ln; 339 } 340 341 /* 342 * Insert routing information in a node. 343 */ 344 345 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, 346 struct nlmsghdr *nlh, struct netlink_skb_parms *req) 347 { 348 struct rt6_info *iter = NULL; 349 struct rt6_info **ins; 350 351 ins = &fn->leaf; 352 353 if (fn->fn_flags&RTN_TL_ROOT && 354 fn->leaf == &ip6_null_entry && 355 !(rt->rt6i_flags & (RTF_DEFAULT | RTF_ADDRCONF)) ){ 356 fn->leaf = rt; 357 rt->u.next = NULL; 358 goto out; 359 } 360 361 for (iter = fn->leaf; iter; iter=iter->u.next) { 362 /* 363 * Search for duplicates 364 */ 365 366 if (iter->rt6i_metric == rt->rt6i_metric) { 367 /* 368 * Same priority level 369 */ 370 371 if (iter->rt6i_dev == rt->rt6i_dev && 372 iter->rt6i_idev == rt->rt6i_idev && 373 ipv6_addr_equal(&iter->rt6i_gateway, 374 &rt->rt6i_gateway)) { 375 if (!(iter->rt6i_flags&RTF_EXPIRES)) 376 return -EEXIST; 377 iter->rt6i_expires = rt->rt6i_expires; 378 if (!(rt->rt6i_flags&RTF_EXPIRES)) { 379 iter->rt6i_flags &= ~RTF_EXPIRES; 380 iter->rt6i_expires = 0; 381 } 382 return -EEXIST; 383 } 384 } 385 386 if (iter->rt6i_metric > rt->rt6i_metric) 387 break; 388 389 ins = &iter->u.next; 390 } 391 392 /* 393 * insert node 394 */ 395 396 out: 397 rt->u.next = iter; 398 *ins = rt; 399 rt->rt6i_node = fn; 400 atomic_inc(&rt->rt6i_ref); 401 inet6_rt_notify(RTM_NEWROUTE, rt, nlh, req); 402 rt6_stats.fib_rt_entries++; 403 404 if ((fn->fn_flags & RTN_RTINFO) == 0) { 405 rt6_stats.fib_route_nodes++; 406 fn->fn_flags |= RTN_RTINFO; 407 } 408 409 return 0; 410 } 411 412 static __inline__ void fib6_start_gc(struct rt6_info *rt) 413 { 414 if (ip6_fib_timer.expires == 0 && 415 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE))) 416 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval); 417 } 418 419 void fib6_force_start_gc(void) 420 { 421 if (ip6_fib_timer.expires == 0) 422 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval); 423 } 424 425 /* 426 * Add routing information to the routing tree. 427 * <destination addr>/<source addr> 428 * with source addr info in sub-trees 429 */ 430 431 int fib6_add(struct fib6_node *root, struct rt6_info *rt, 432 struct nlmsghdr *nlh, void *_rtattr, struct netlink_skb_parms *req) 433 { 434 struct fib6_node *fn; 435 int err = -ENOMEM; 436 437 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr), 438 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst)); 439 440 if (fn == NULL) 441 goto out; 442 443 #ifdef CONFIG_IPV6_SUBTREES 444 if (rt->rt6i_src.plen) { 445 struct fib6_node *sn; 446 447 if (fn->subtree == NULL) { 448 struct fib6_node *sfn; 449 450 /* 451 * Create subtree. 452 * 453 * fn[main tree] 454 * | 455 * sfn[subtree root] 456 * \ 457 * sn[new leaf node] 458 */ 459 460 /* Create subtree root node */ 461 sfn = node_alloc(); 462 if (sfn == NULL) 463 goto st_failure; 464 465 sfn->leaf = &ip6_null_entry; 466 atomic_inc(&ip6_null_entry.rt6i_ref); 467 sfn->fn_flags = RTN_ROOT; 468 sfn->fn_sernum = fib6_new_sernum(); 469 470 /* Now add the first leaf node to new subtree */ 471 472 sn = fib6_add_1(sfn, &rt->rt6i_src.addr, 473 sizeof(struct in6_addr), rt->rt6i_src.plen, 474 offsetof(struct rt6_info, rt6i_src)); 475 476 if (sn == NULL) { 477 /* If it is failed, discard just allocated 478 root, and then (in st_failure) stale node 479 in main tree. 480 */ 481 node_free(sfn); 482 goto st_failure; 483 } 484 485 /* Now link new subtree to main tree */ 486 sfn->parent = fn; 487 fn->subtree = sfn; 488 if (fn->leaf == NULL) { 489 fn->leaf = rt; 490 atomic_inc(&rt->rt6i_ref); 491 } 492 } else { 493 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, 494 sizeof(struct in6_addr), rt->rt6i_src.plen, 495 offsetof(struct rt6_info, rt6i_src)); 496 497 if (sn == NULL) 498 goto st_failure; 499 } 500 501 fn = sn; 502 } 503 #endif 504 505 err = fib6_add_rt2node(fn, rt, nlh, req); 506 507 if (err == 0) { 508 fib6_start_gc(rt); 509 if (!(rt->rt6i_flags&RTF_CACHE)) 510 fib6_prune_clones(fn, rt); 511 } 512 513 out: 514 if (err) 515 dst_free(&rt->u.dst); 516 return err; 517 518 #ifdef CONFIG_IPV6_SUBTREES 519 /* Subtree creation failed, probably main tree node 520 is orphan. If it is, shoot it. 521 */ 522 st_failure: 523 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) 524 fib6_repair_tree(fn); 525 dst_free(&rt->u.dst); 526 return err; 527 #endif 528 } 529 530 /* 531 * Routing tree lookup 532 * 533 */ 534 535 struct lookup_args { 536 int offset; /* key offset on rt6_info */ 537 struct in6_addr *addr; /* search key */ 538 }; 539 540 static struct fib6_node * fib6_lookup_1(struct fib6_node *root, 541 struct lookup_args *args) 542 { 543 struct fib6_node *fn; 544 int dir; 545 546 /* 547 * Descend on a tree 548 */ 549 550 fn = root; 551 552 for (;;) { 553 struct fib6_node *next; 554 555 dir = addr_bit_set(args->addr, fn->fn_bit); 556 557 next = dir ? fn->right : fn->left; 558 559 if (next) { 560 fn = next; 561 continue; 562 } 563 564 break; 565 } 566 567 while ((fn->fn_flags & RTN_ROOT) == 0) { 568 #ifdef CONFIG_IPV6_SUBTREES 569 if (fn->subtree) { 570 struct fib6_node *st; 571 struct lookup_args *narg; 572 573 narg = args + 1; 574 575 if (narg->addr) { 576 st = fib6_lookup_1(fn->subtree, narg); 577 578 if (st && !(st->fn_flags & RTN_ROOT)) 579 return st; 580 } 581 } 582 #endif 583 584 if (fn->fn_flags & RTN_RTINFO) { 585 struct rt6key *key; 586 587 key = (struct rt6key *) ((u8 *) fn->leaf + 588 args->offset); 589 590 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) 591 return fn; 592 } 593 594 fn = fn->parent; 595 } 596 597 return NULL; 598 } 599 600 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr, 601 struct in6_addr *saddr) 602 { 603 struct lookup_args args[2]; 604 struct fib6_node *fn; 605 606 args[0].offset = offsetof(struct rt6_info, rt6i_dst); 607 args[0].addr = daddr; 608 609 #ifdef CONFIG_IPV6_SUBTREES 610 args[1].offset = offsetof(struct rt6_info, rt6i_src); 611 args[1].addr = saddr; 612 #endif 613 614 fn = fib6_lookup_1(root, args); 615 616 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT) 617 fn = root; 618 619 return fn; 620 } 621 622 /* 623 * Get node with specified destination prefix (and source prefix, 624 * if subtrees are used) 625 */ 626 627 628 static struct fib6_node * fib6_locate_1(struct fib6_node *root, 629 struct in6_addr *addr, 630 int plen, int offset) 631 { 632 struct fib6_node *fn; 633 634 for (fn = root; fn ; ) { 635 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); 636 637 /* 638 * Prefix match 639 */ 640 if (plen < fn->fn_bit || 641 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 642 return NULL; 643 644 if (plen == fn->fn_bit) 645 return fn; 646 647 /* 648 * We have more bits to go 649 */ 650 if (addr_bit_set(addr, fn->fn_bit)) 651 fn = fn->right; 652 else 653 fn = fn->left; 654 } 655 return NULL; 656 } 657 658 struct fib6_node * fib6_locate(struct fib6_node *root, 659 struct in6_addr *daddr, int dst_len, 660 struct in6_addr *saddr, int src_len) 661 { 662 struct fib6_node *fn; 663 664 fn = fib6_locate_1(root, daddr, dst_len, 665 offsetof(struct rt6_info, rt6i_dst)); 666 667 #ifdef CONFIG_IPV6_SUBTREES 668 if (src_len) { 669 BUG_TRAP(saddr!=NULL); 670 if (fn == NULL) 671 fn = fn->subtree; 672 if (fn) 673 fn = fib6_locate_1(fn, saddr, src_len, 674 offsetof(struct rt6_info, rt6i_src)); 675 } 676 #endif 677 678 if (fn && fn->fn_flags&RTN_RTINFO) 679 return fn; 680 681 return NULL; 682 } 683 684 685 /* 686 * Deletion 687 * 688 */ 689 690 static struct rt6_info * fib6_find_prefix(struct fib6_node *fn) 691 { 692 if (fn->fn_flags&RTN_ROOT) 693 return &ip6_null_entry; 694 695 while(fn) { 696 if(fn->left) 697 return fn->left->leaf; 698 699 if(fn->right) 700 return fn->right->leaf; 701 702 fn = SUBTREE(fn); 703 } 704 return NULL; 705 } 706 707 /* 708 * Called to trim the tree of intermediate nodes when possible. "fn" 709 * is the node we want to try and remove. 710 */ 711 712 static struct fib6_node * fib6_repair_tree(struct fib6_node *fn) 713 { 714 int children; 715 int nstate; 716 struct fib6_node *child, *pn; 717 struct fib6_walker_t *w; 718 int iter = 0; 719 720 for (;;) { 721 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 722 iter++; 723 724 BUG_TRAP(!(fn->fn_flags&RTN_RTINFO)); 725 BUG_TRAP(!(fn->fn_flags&RTN_TL_ROOT)); 726 BUG_TRAP(fn->leaf==NULL); 727 728 children = 0; 729 child = NULL; 730 if (fn->right) child = fn->right, children |= 1; 731 if (fn->left) child = fn->left, children |= 2; 732 733 if (children == 3 || SUBTREE(fn) 734 #ifdef CONFIG_IPV6_SUBTREES 735 /* Subtree root (i.e. fn) may have one child */ 736 || (children && fn->fn_flags&RTN_ROOT) 737 #endif 738 ) { 739 fn->leaf = fib6_find_prefix(fn); 740 #if RT6_DEBUG >= 2 741 if (fn->leaf==NULL) { 742 BUG_TRAP(fn->leaf); 743 fn->leaf = &ip6_null_entry; 744 } 745 #endif 746 atomic_inc(&fn->leaf->rt6i_ref); 747 return fn->parent; 748 } 749 750 pn = fn->parent; 751 #ifdef CONFIG_IPV6_SUBTREES 752 if (SUBTREE(pn) == fn) { 753 BUG_TRAP(fn->fn_flags&RTN_ROOT); 754 SUBTREE(pn) = NULL; 755 nstate = FWS_L; 756 } else { 757 BUG_TRAP(!(fn->fn_flags&RTN_ROOT)); 758 #endif 759 if (pn->right == fn) pn->right = child; 760 else if (pn->left == fn) pn->left = child; 761 #if RT6_DEBUG >= 2 762 else BUG_TRAP(0); 763 #endif 764 if (child) 765 child->parent = pn; 766 nstate = FWS_R; 767 #ifdef CONFIG_IPV6_SUBTREES 768 } 769 #endif 770 771 read_lock(&fib6_walker_lock); 772 FOR_WALKERS(w) { 773 if (child == NULL) { 774 if (w->root == fn) { 775 w->root = w->node = NULL; 776 RT6_TRACE("W %p adjusted by delroot 1\n", w); 777 } else if (w->node == fn) { 778 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 779 w->node = pn; 780 w->state = nstate; 781 } 782 } else { 783 if (w->root == fn) { 784 w->root = child; 785 RT6_TRACE("W %p adjusted by delroot 2\n", w); 786 } 787 if (w->node == fn) { 788 w->node = child; 789 if (children&2) { 790 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 791 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT; 792 } else { 793 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 794 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT; 795 } 796 } 797 } 798 } 799 read_unlock(&fib6_walker_lock); 800 801 node_free(fn); 802 if (pn->fn_flags&RTN_RTINFO || SUBTREE(pn)) 803 return pn; 804 805 rt6_release(pn->leaf); 806 pn->leaf = NULL; 807 fn = pn; 808 } 809 } 810 811 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, 812 struct nlmsghdr *nlh, void *_rtattr, struct netlink_skb_parms *req) 813 { 814 struct fib6_walker_t *w; 815 struct rt6_info *rt = *rtp; 816 817 RT6_TRACE("fib6_del_route\n"); 818 819 /* Unlink it */ 820 *rtp = rt->u.next; 821 rt->rt6i_node = NULL; 822 rt6_stats.fib_rt_entries--; 823 rt6_stats.fib_discarded_routes++; 824 825 /* Adjust walkers */ 826 read_lock(&fib6_walker_lock); 827 FOR_WALKERS(w) { 828 if (w->state == FWS_C && w->leaf == rt) { 829 RT6_TRACE("walker %p adjusted by delroute\n", w); 830 w->leaf = rt->u.next; 831 if (w->leaf == NULL) 832 w->state = FWS_U; 833 } 834 } 835 read_unlock(&fib6_walker_lock); 836 837 rt->u.next = NULL; 838 839 if (fn->leaf == NULL && fn->fn_flags&RTN_TL_ROOT) 840 fn->leaf = &ip6_null_entry; 841 842 /* If it was last route, expunge its radix tree node */ 843 if (fn->leaf == NULL) { 844 fn->fn_flags &= ~RTN_RTINFO; 845 rt6_stats.fib_route_nodes--; 846 fn = fib6_repair_tree(fn); 847 } 848 849 if (atomic_read(&rt->rt6i_ref) != 1) { 850 /* This route is used as dummy address holder in some split 851 * nodes. It is not leaked, but it still holds other resources, 852 * which must be released in time. So, scan ascendant nodes 853 * and replace dummy references to this route with references 854 * to still alive ones. 855 */ 856 while (fn) { 857 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) { 858 fn->leaf = fib6_find_prefix(fn); 859 atomic_inc(&fn->leaf->rt6i_ref); 860 rt6_release(rt); 861 } 862 fn = fn->parent; 863 } 864 /* No more references are possible at this point. */ 865 if (atomic_read(&rt->rt6i_ref) != 1) BUG(); 866 } 867 868 inet6_rt_notify(RTM_DELROUTE, rt, nlh, req); 869 rt6_release(rt); 870 } 871 872 int fib6_del(struct rt6_info *rt, struct nlmsghdr *nlh, void *_rtattr, struct netlink_skb_parms *req) 873 { 874 struct fib6_node *fn = rt->rt6i_node; 875 struct rt6_info **rtp; 876 877 #if RT6_DEBUG >= 2 878 if (rt->u.dst.obsolete>0) { 879 BUG_TRAP(fn==NULL); 880 return -ENOENT; 881 } 882 #endif 883 if (fn == NULL || rt == &ip6_null_entry) 884 return -ENOENT; 885 886 BUG_TRAP(fn->fn_flags&RTN_RTINFO); 887 888 if (!(rt->rt6i_flags&RTF_CACHE)) 889 fib6_prune_clones(fn, rt); 890 891 /* 892 * Walk the leaf entries looking for ourself 893 */ 894 895 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.next) { 896 if (*rtp == rt) { 897 fib6_del_route(fn, rtp, nlh, _rtattr, req); 898 return 0; 899 } 900 } 901 return -ENOENT; 902 } 903 904 /* 905 * Tree traversal function. 906 * 907 * Certainly, it is not interrupt safe. 908 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 909 * It means, that we can modify tree during walking 910 * and use this function for garbage collection, clone pruning, 911 * cleaning tree when a device goes down etc. etc. 912 * 913 * It guarantees that every node will be traversed, 914 * and that it will be traversed only once. 915 * 916 * Callback function w->func may return: 917 * 0 -> continue walking. 918 * positive value -> walking is suspended (used by tree dumps, 919 * and probably by gc, if it will be split to several slices) 920 * negative value -> terminate walking. 921 * 922 * The function itself returns: 923 * 0 -> walk is complete. 924 * >0 -> walk is incomplete (i.e. suspended) 925 * <0 -> walk is terminated by an error. 926 */ 927 928 int fib6_walk_continue(struct fib6_walker_t *w) 929 { 930 struct fib6_node *fn, *pn; 931 932 for (;;) { 933 fn = w->node; 934 if (fn == NULL) 935 return 0; 936 937 if (w->prune && fn != w->root && 938 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) { 939 w->state = FWS_C; 940 w->leaf = fn->leaf; 941 } 942 switch (w->state) { 943 #ifdef CONFIG_IPV6_SUBTREES 944 case FWS_S: 945 if (SUBTREE(fn)) { 946 w->node = SUBTREE(fn); 947 continue; 948 } 949 w->state = FWS_L; 950 #endif 951 case FWS_L: 952 if (fn->left) { 953 w->node = fn->left; 954 w->state = FWS_INIT; 955 continue; 956 } 957 w->state = FWS_R; 958 case FWS_R: 959 if (fn->right) { 960 w->node = fn->right; 961 w->state = FWS_INIT; 962 continue; 963 } 964 w->state = FWS_C; 965 w->leaf = fn->leaf; 966 case FWS_C: 967 if (w->leaf && fn->fn_flags&RTN_RTINFO) { 968 int err = w->func(w); 969 if (err) 970 return err; 971 continue; 972 } 973 w->state = FWS_U; 974 case FWS_U: 975 if (fn == w->root) 976 return 0; 977 pn = fn->parent; 978 w->node = pn; 979 #ifdef CONFIG_IPV6_SUBTREES 980 if (SUBTREE(pn) == fn) { 981 BUG_TRAP(fn->fn_flags&RTN_ROOT); 982 w->state = FWS_L; 983 continue; 984 } 985 #endif 986 if (pn->left == fn) { 987 w->state = FWS_R; 988 continue; 989 } 990 if (pn->right == fn) { 991 w->state = FWS_C; 992 w->leaf = w->node->leaf; 993 continue; 994 } 995 #if RT6_DEBUG >= 2 996 BUG_TRAP(0); 997 #endif 998 } 999 } 1000 } 1001 1002 int fib6_walk(struct fib6_walker_t *w) 1003 { 1004 int res; 1005 1006 w->state = FWS_INIT; 1007 w->node = w->root; 1008 1009 fib6_walker_link(w); 1010 res = fib6_walk_continue(w); 1011 if (res <= 0) 1012 fib6_walker_unlink(w); 1013 return res; 1014 } 1015 1016 static int fib6_clean_node(struct fib6_walker_t *w) 1017 { 1018 int res; 1019 struct rt6_info *rt; 1020 struct fib6_cleaner_t *c = (struct fib6_cleaner_t*)w; 1021 1022 for (rt = w->leaf; rt; rt = rt->u.next) { 1023 res = c->func(rt, c->arg); 1024 if (res < 0) { 1025 w->leaf = rt; 1026 res = fib6_del(rt, NULL, NULL, NULL); 1027 if (res) { 1028 #if RT6_DEBUG >= 2 1029 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res); 1030 #endif 1031 continue; 1032 } 1033 return 0; 1034 } 1035 BUG_TRAP(res==0); 1036 } 1037 w->leaf = rt; 1038 return 0; 1039 } 1040 1041 /* 1042 * Convenient frontend to tree walker. 1043 * 1044 * func is called on each route. 1045 * It may return -1 -> delete this route. 1046 * 0 -> continue walking 1047 * 1048 * prune==1 -> only immediate children of node (certainly, 1049 * ignoring pure split nodes) will be scanned. 1050 */ 1051 1052 void fib6_clean_tree(struct fib6_node *root, 1053 int (*func)(struct rt6_info *, void *arg), 1054 int prune, void *arg) 1055 { 1056 struct fib6_cleaner_t c; 1057 1058 c.w.root = root; 1059 c.w.func = fib6_clean_node; 1060 c.w.prune = prune; 1061 c.func = func; 1062 c.arg = arg; 1063 1064 fib6_walk(&c.w); 1065 } 1066 1067 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1068 { 1069 if (rt->rt6i_flags & RTF_CACHE) { 1070 RT6_TRACE("pruning clone %p\n", rt); 1071 return -1; 1072 } 1073 1074 return 0; 1075 } 1076 1077 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt) 1078 { 1079 fib6_clean_tree(fn, fib6_prune_clone, 1, rt); 1080 } 1081 1082 /* 1083 * Garbage collection 1084 */ 1085 1086 static struct fib6_gc_args 1087 { 1088 int timeout; 1089 int more; 1090 } gc_args; 1091 1092 static int fib6_age(struct rt6_info *rt, void *arg) 1093 { 1094 unsigned long now = jiffies; 1095 1096 /* 1097 * check addrconf expiration here. 1098 * Routes are expired even if they are in use. 1099 * 1100 * Also age clones. Note, that clones are aged out 1101 * only if they are not in use now. 1102 */ 1103 1104 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) { 1105 if (time_after(now, rt->rt6i_expires)) { 1106 RT6_TRACE("expiring %p\n", rt); 1107 return -1; 1108 } 1109 gc_args.more++; 1110 } else if (rt->rt6i_flags & RTF_CACHE) { 1111 if (atomic_read(&rt->u.dst.__refcnt) == 0 && 1112 time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) { 1113 RT6_TRACE("aging clone %p\n", rt); 1114 return -1; 1115 } else if ((rt->rt6i_flags & RTF_GATEWAY) && 1116 (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) { 1117 RT6_TRACE("purging route %p via non-router but gateway\n", 1118 rt); 1119 return -1; 1120 } 1121 gc_args.more++; 1122 } 1123 1124 return 0; 1125 } 1126 1127 static DEFINE_SPINLOCK(fib6_gc_lock); 1128 1129 void fib6_run_gc(unsigned long dummy) 1130 { 1131 if (dummy != ~0UL) { 1132 spin_lock_bh(&fib6_gc_lock); 1133 gc_args.timeout = dummy ? (int)dummy : ip6_rt_gc_interval; 1134 } else { 1135 local_bh_disable(); 1136 if (!spin_trylock(&fib6_gc_lock)) { 1137 mod_timer(&ip6_fib_timer, jiffies + HZ); 1138 local_bh_enable(); 1139 return; 1140 } 1141 gc_args.timeout = ip6_rt_gc_interval; 1142 } 1143 gc_args.more = 0; 1144 1145 1146 write_lock_bh(&rt6_lock); 1147 ndisc_dst_gc(&gc_args.more); 1148 fib6_clean_tree(&ip6_routing_table, fib6_age, 0, NULL); 1149 write_unlock_bh(&rt6_lock); 1150 1151 if (gc_args.more) 1152 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval); 1153 else { 1154 del_timer(&ip6_fib_timer); 1155 ip6_fib_timer.expires = 0; 1156 } 1157 spin_unlock_bh(&fib6_gc_lock); 1158 } 1159 1160 void __init fib6_init(void) 1161 { 1162 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1163 sizeof(struct fib6_node), 1164 0, SLAB_HWCACHE_ALIGN, 1165 NULL, NULL); 1166 if (!fib6_node_kmem) 1167 panic("cannot create fib6_nodes cache"); 1168 } 1169 1170 void fib6_gc_cleanup(void) 1171 { 1172 del_timer(&ip6_fib_timer); 1173 kmem_cache_destroy(fib6_node_kmem); 1174 } 1175