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/config.h> 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 31 #ifdef CONFIG_PROC_FS 32 #include <linux/proc_fs.h> 33 #endif 34 35 #include <net/ipv6.h> 36 #include <net/ndisc.h> 37 #include <net/addrconf.h> 38 39 #include <net/ip6_fib.h> 40 #include <net/ip6_route.h> 41 42 #define RT6_DEBUG 2 43 44 #if RT6_DEBUG >= 3 45 #define RT6_TRACE(x...) printk(KERN_DEBUG x) 46 #else 47 #define RT6_TRACE(x...) do { ; } while (0) 48 #endif 49 50 struct rt6_statistics rt6_stats; 51 52 static kmem_cache_t * fib6_node_kmem __read_mostly; 53 54 enum fib_walk_state_t 55 { 56 #ifdef CONFIG_IPV6_SUBTREES 57 FWS_S, 58 #endif 59 FWS_L, 60 FWS_R, 61 FWS_C, 62 FWS_U 63 }; 64 65 struct fib6_cleaner_t 66 { 67 struct fib6_walker_t w; 68 int (*func)(struct rt6_info *, void *arg); 69 void *arg; 70 }; 71 72 DEFINE_RWLOCK(fib6_walker_lock); 73 74 75 #ifdef CONFIG_IPV6_SUBTREES 76 #define FWS_INIT FWS_S 77 #define SUBTREE(fn) ((fn)->subtree) 78 #else 79 #define FWS_INIT FWS_L 80 #define SUBTREE(fn) NULL 81 #endif 82 83 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt); 84 static struct fib6_node * fib6_repair_tree(struct fib6_node *fn); 85 86 /* 87 * A routing update causes an increase of the serial number on the 88 * affected subtree. This allows for cached routes to be asynchronously 89 * tested when modifications are made to the destination cache as a 90 * result of redirects, path MTU changes, etc. 91 */ 92 93 static __u32 rt_sernum; 94 95 static struct timer_list ip6_fib_timer = TIMER_INITIALIZER(fib6_run_gc, 0, 0); 96 97 struct fib6_walker_t fib6_walker_list = { 98 .prev = &fib6_walker_list, 99 .next = &fib6_walker_list, 100 }; 101 102 #define FOR_WALKERS(w) for ((w)=fib6_walker_list.next; (w) != &fib6_walker_list; (w)=(w)->next) 103 104 static __inline__ u32 fib6_new_sernum(void) 105 { 106 u32 n = ++rt_sernum; 107 if ((__s32)n <= 0) 108 rt_sernum = n = 1; 109 return n; 110 } 111 112 /* 113 * Auxiliary address test functions for the radix tree. 114 * 115 * These assume a 32bit processor (although it will work on 116 * 64bit processors) 117 */ 118 119 /* 120 * test bit 121 */ 122 123 static __inline__ int addr_bit_set(void *token, int fn_bit) 124 { 125 __u32 *addr = token; 126 127 return htonl(1 << ((~fn_bit)&0x1F)) & addr[fn_bit>>5]; 128 } 129 130 /* 131 * find the first different bit between two addresses 132 * length of address must be a multiple of 32bits 133 */ 134 135 static __inline__ int addr_diff(void *token1, void *token2, int addrlen) 136 { 137 __u32 *a1 = token1; 138 __u32 *a2 = token2; 139 int i; 140 141 addrlen >>= 2; 142 143 for (i = 0; i < addrlen; i++) { 144 __u32 xb; 145 146 xb = a1[i] ^ a2[i]; 147 148 if (xb) { 149 int j = 31; 150 151 xb = ntohl(xb); 152 153 while ((xb & (1 << j)) == 0) 154 j--; 155 156 return (i * 32 + 31 - j); 157 } 158 } 159 160 /* 161 * we should *never* get to this point since that 162 * would mean the addrs are equal 163 * 164 * However, we do get to it 8) And exacly, when 165 * addresses are equal 8) 166 * 167 * ip route add 1111::/128 via ... 168 * ip route add 1111::/64 via ... 169 * and we are here. 170 * 171 * Ideally, this function should stop comparison 172 * at prefix length. It does not, but it is still OK, 173 * if returned value is greater than prefix length. 174 * --ANK (980803) 175 */ 176 177 return addrlen<<5; 178 } 179 180 static __inline__ struct fib6_node * node_alloc(void) 181 { 182 struct fib6_node *fn; 183 184 if ((fn = kmem_cache_alloc(fib6_node_kmem, SLAB_ATOMIC)) != NULL) 185 memset(fn, 0, sizeof(struct fib6_node)); 186 187 return fn; 188 } 189 190 static __inline__ void node_free(struct fib6_node * fn) 191 { 192 kmem_cache_free(fib6_node_kmem, fn); 193 } 194 195 static __inline__ void rt6_release(struct rt6_info *rt) 196 { 197 if (atomic_dec_and_test(&rt->rt6i_ref)) 198 dst_free(&rt->u.dst); 199 } 200 201 202 /* 203 * Routing Table 204 * 205 * return the appropriate node for a routing tree "add" operation 206 * by either creating and inserting or by returning an existing 207 * node. 208 */ 209 210 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr, 211 int addrlen, int plen, 212 int offset) 213 { 214 struct fib6_node *fn, *in, *ln; 215 struct fib6_node *pn = NULL; 216 struct rt6key *key; 217 int bit; 218 int dir = 0; 219 __u32 sernum = fib6_new_sernum(); 220 221 RT6_TRACE("fib6_add_1\n"); 222 223 /* insert node in tree */ 224 225 fn = root; 226 227 do { 228 key = (struct rt6key *)((u8 *)fn->leaf + offset); 229 230 /* 231 * Prefix match 232 */ 233 if (plen < fn->fn_bit || 234 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 235 goto insert_above; 236 237 /* 238 * Exact match ? 239 */ 240 241 if (plen == fn->fn_bit) { 242 /* clean up an intermediate node */ 243 if ((fn->fn_flags & RTN_RTINFO) == 0) { 244 rt6_release(fn->leaf); 245 fn->leaf = NULL; 246 } 247 248 fn->fn_sernum = sernum; 249 250 return fn; 251 } 252 253 /* 254 * We have more bits to go 255 */ 256 257 /* Try to walk down on tree. */ 258 fn->fn_sernum = sernum; 259 dir = addr_bit_set(addr, fn->fn_bit); 260 pn = fn; 261 fn = dir ? fn->right: fn->left; 262 } while (fn); 263 264 /* 265 * We walked to the bottom of tree. 266 * Create new leaf node without children. 267 */ 268 269 ln = node_alloc(); 270 271 if (ln == NULL) 272 return NULL; 273 ln->fn_bit = plen; 274 275 ln->parent = pn; 276 ln->fn_sernum = sernum; 277 278 if (dir) 279 pn->right = ln; 280 else 281 pn->left = ln; 282 283 return ln; 284 285 286 insert_above: 287 /* 288 * split since we don't have a common prefix anymore or 289 * we have a less significant route. 290 * we've to insert an intermediate node on the list 291 * this new node will point to the one we need to create 292 * and the current 293 */ 294 295 pn = fn->parent; 296 297 /* find 1st bit in difference between the 2 addrs. 298 299 See comment in addr_diff: bit may be an invalid value, 300 but if it is >= plen, the value is ignored in any case. 301 */ 302 303 bit = addr_diff(addr, &key->addr, addrlen); 304 305 /* 306 * (intermediate)[in] 307 * / \ 308 * (new leaf node)[ln] (old node)[fn] 309 */ 310 if (plen > bit) { 311 in = node_alloc(); 312 ln = node_alloc(); 313 314 if (in == NULL || ln == NULL) { 315 if (in) 316 node_free(in); 317 if (ln) 318 node_free(ln); 319 return NULL; 320 } 321 322 /* 323 * new intermediate node. 324 * RTN_RTINFO will 325 * be off since that an address that chooses one of 326 * the branches would not match less specific routes 327 * in the other branch 328 */ 329 330 in->fn_bit = bit; 331 332 in->parent = pn; 333 in->leaf = fn->leaf; 334 atomic_inc(&in->leaf->rt6i_ref); 335 336 in->fn_sernum = sernum; 337 338 /* update parent pointer */ 339 if (dir) 340 pn->right = in; 341 else 342 pn->left = in; 343 344 ln->fn_bit = plen; 345 346 ln->parent = in; 347 fn->parent = in; 348 349 ln->fn_sernum = sernum; 350 351 if (addr_bit_set(addr, bit)) { 352 in->right = ln; 353 in->left = fn; 354 } else { 355 in->left = ln; 356 in->right = fn; 357 } 358 } else { /* plen <= bit */ 359 360 /* 361 * (new leaf node)[ln] 362 * / \ 363 * (old node)[fn] NULL 364 */ 365 366 ln = node_alloc(); 367 368 if (ln == NULL) 369 return NULL; 370 371 ln->fn_bit = plen; 372 373 ln->parent = pn; 374 375 ln->fn_sernum = sernum; 376 377 if (dir) 378 pn->right = ln; 379 else 380 pn->left = ln; 381 382 if (addr_bit_set(&key->addr, plen)) 383 ln->right = fn; 384 else 385 ln->left = fn; 386 387 fn->parent = ln; 388 } 389 return ln; 390 } 391 392 /* 393 * Insert routing information in a node. 394 */ 395 396 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, 397 struct nlmsghdr *nlh, struct netlink_skb_parms *req) 398 { 399 struct rt6_info *iter = NULL; 400 struct rt6_info **ins; 401 402 ins = &fn->leaf; 403 404 if (fn->fn_flags&RTN_TL_ROOT && 405 fn->leaf == &ip6_null_entry && 406 !(rt->rt6i_flags & (RTF_DEFAULT | RTF_ADDRCONF)) ){ 407 fn->leaf = rt; 408 rt->u.next = NULL; 409 goto out; 410 } 411 412 for (iter = fn->leaf; iter; iter=iter->u.next) { 413 /* 414 * Search for duplicates 415 */ 416 417 if (iter->rt6i_metric == rt->rt6i_metric) { 418 /* 419 * Same priority level 420 */ 421 422 if (iter->rt6i_dev == rt->rt6i_dev && 423 iter->rt6i_idev == rt->rt6i_idev && 424 ipv6_addr_equal(&iter->rt6i_gateway, 425 &rt->rt6i_gateway)) { 426 if (!(iter->rt6i_flags&RTF_EXPIRES)) 427 return -EEXIST; 428 iter->rt6i_expires = rt->rt6i_expires; 429 if (!(rt->rt6i_flags&RTF_EXPIRES)) { 430 iter->rt6i_flags &= ~RTF_EXPIRES; 431 iter->rt6i_expires = 0; 432 } 433 return -EEXIST; 434 } 435 } 436 437 if (iter->rt6i_metric > rt->rt6i_metric) 438 break; 439 440 ins = &iter->u.next; 441 } 442 443 /* 444 * insert node 445 */ 446 447 out: 448 rt->u.next = iter; 449 *ins = rt; 450 rt->rt6i_node = fn; 451 atomic_inc(&rt->rt6i_ref); 452 inet6_rt_notify(RTM_NEWROUTE, rt, nlh, req); 453 rt6_stats.fib_rt_entries++; 454 455 if ((fn->fn_flags & RTN_RTINFO) == 0) { 456 rt6_stats.fib_route_nodes++; 457 fn->fn_flags |= RTN_RTINFO; 458 } 459 460 return 0; 461 } 462 463 static __inline__ void fib6_start_gc(struct rt6_info *rt) 464 { 465 if (ip6_fib_timer.expires == 0 && 466 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE))) 467 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval); 468 } 469 470 void fib6_force_start_gc(void) 471 { 472 if (ip6_fib_timer.expires == 0) 473 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval); 474 } 475 476 /* 477 * Add routing information to the routing tree. 478 * <destination addr>/<source addr> 479 * with source addr info in sub-trees 480 */ 481 482 int fib6_add(struct fib6_node *root, struct rt6_info *rt, 483 struct nlmsghdr *nlh, void *_rtattr, struct netlink_skb_parms *req) 484 { 485 struct fib6_node *fn; 486 int err = -ENOMEM; 487 488 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr), 489 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst)); 490 491 if (fn == NULL) 492 goto out; 493 494 #ifdef CONFIG_IPV6_SUBTREES 495 if (rt->rt6i_src.plen) { 496 struct fib6_node *sn; 497 498 if (fn->subtree == NULL) { 499 struct fib6_node *sfn; 500 501 /* 502 * Create subtree. 503 * 504 * fn[main tree] 505 * | 506 * sfn[subtree root] 507 * \ 508 * sn[new leaf node] 509 */ 510 511 /* Create subtree root node */ 512 sfn = node_alloc(); 513 if (sfn == NULL) 514 goto st_failure; 515 516 sfn->leaf = &ip6_null_entry; 517 atomic_inc(&ip6_null_entry.rt6i_ref); 518 sfn->fn_flags = RTN_ROOT; 519 sfn->fn_sernum = fib6_new_sernum(); 520 521 /* Now add the first leaf node to new subtree */ 522 523 sn = fib6_add_1(sfn, &rt->rt6i_src.addr, 524 sizeof(struct in6_addr), rt->rt6i_src.plen, 525 offsetof(struct rt6_info, rt6i_src)); 526 527 if (sn == NULL) { 528 /* If it is failed, discard just allocated 529 root, and then (in st_failure) stale node 530 in main tree. 531 */ 532 node_free(sfn); 533 goto st_failure; 534 } 535 536 /* Now link new subtree to main tree */ 537 sfn->parent = fn; 538 fn->subtree = sfn; 539 if (fn->leaf == NULL) { 540 fn->leaf = rt; 541 atomic_inc(&rt->rt6i_ref); 542 } 543 } else { 544 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, 545 sizeof(struct in6_addr), rt->rt6i_src.plen, 546 offsetof(struct rt6_info, rt6i_src)); 547 548 if (sn == NULL) 549 goto st_failure; 550 } 551 552 fn = sn; 553 } 554 #endif 555 556 err = fib6_add_rt2node(fn, rt, nlh, req); 557 558 if (err == 0) { 559 fib6_start_gc(rt); 560 if (!(rt->rt6i_flags&RTF_CACHE)) 561 fib6_prune_clones(fn, rt); 562 } 563 564 out: 565 if (err) 566 dst_free(&rt->u.dst); 567 return err; 568 569 #ifdef CONFIG_IPV6_SUBTREES 570 /* Subtree creation failed, probably main tree node 571 is orphan. If it is, shoot it. 572 */ 573 st_failure: 574 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) 575 fib6_repair_tree(fn); 576 dst_free(&rt->u.dst); 577 return err; 578 #endif 579 } 580 581 /* 582 * Routing tree lookup 583 * 584 */ 585 586 struct lookup_args { 587 int offset; /* key offset on rt6_info */ 588 struct in6_addr *addr; /* search key */ 589 }; 590 591 static struct fib6_node * fib6_lookup_1(struct fib6_node *root, 592 struct lookup_args *args) 593 { 594 struct fib6_node *fn; 595 int dir; 596 597 /* 598 * Descend on a tree 599 */ 600 601 fn = root; 602 603 for (;;) { 604 struct fib6_node *next; 605 606 dir = addr_bit_set(args->addr, fn->fn_bit); 607 608 next = dir ? fn->right : fn->left; 609 610 if (next) { 611 fn = next; 612 continue; 613 } 614 615 break; 616 } 617 618 while ((fn->fn_flags & RTN_ROOT) == 0) { 619 #ifdef CONFIG_IPV6_SUBTREES 620 if (fn->subtree) { 621 struct fib6_node *st; 622 struct lookup_args *narg; 623 624 narg = args + 1; 625 626 if (narg->addr) { 627 st = fib6_lookup_1(fn->subtree, narg); 628 629 if (st && !(st->fn_flags & RTN_ROOT)) 630 return st; 631 } 632 } 633 #endif 634 635 if (fn->fn_flags & RTN_RTINFO) { 636 struct rt6key *key; 637 638 key = (struct rt6key *) ((u8 *) fn->leaf + 639 args->offset); 640 641 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) 642 return fn; 643 } 644 645 fn = fn->parent; 646 } 647 648 return NULL; 649 } 650 651 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr, 652 struct in6_addr *saddr) 653 { 654 struct lookup_args args[2]; 655 struct fib6_node *fn; 656 657 args[0].offset = offsetof(struct rt6_info, rt6i_dst); 658 args[0].addr = daddr; 659 660 #ifdef CONFIG_IPV6_SUBTREES 661 args[1].offset = offsetof(struct rt6_info, rt6i_src); 662 args[1].addr = saddr; 663 #endif 664 665 fn = fib6_lookup_1(root, args); 666 667 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT) 668 fn = root; 669 670 return fn; 671 } 672 673 /* 674 * Get node with specified destination prefix (and source prefix, 675 * if subtrees are used) 676 */ 677 678 679 static struct fib6_node * fib6_locate_1(struct fib6_node *root, 680 struct in6_addr *addr, 681 int plen, int offset) 682 { 683 struct fib6_node *fn; 684 685 for (fn = root; fn ; ) { 686 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); 687 688 /* 689 * Prefix match 690 */ 691 if (plen < fn->fn_bit || 692 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 693 return NULL; 694 695 if (plen == fn->fn_bit) 696 return fn; 697 698 /* 699 * We have more bits to go 700 */ 701 if (addr_bit_set(addr, fn->fn_bit)) 702 fn = fn->right; 703 else 704 fn = fn->left; 705 } 706 return NULL; 707 } 708 709 struct fib6_node * fib6_locate(struct fib6_node *root, 710 struct in6_addr *daddr, int dst_len, 711 struct in6_addr *saddr, int src_len) 712 { 713 struct fib6_node *fn; 714 715 fn = fib6_locate_1(root, daddr, dst_len, 716 offsetof(struct rt6_info, rt6i_dst)); 717 718 #ifdef CONFIG_IPV6_SUBTREES 719 if (src_len) { 720 BUG_TRAP(saddr!=NULL); 721 if (fn == NULL) 722 fn = fn->subtree; 723 if (fn) 724 fn = fib6_locate_1(fn, saddr, src_len, 725 offsetof(struct rt6_info, rt6i_src)); 726 } 727 #endif 728 729 if (fn && fn->fn_flags&RTN_RTINFO) 730 return fn; 731 732 return NULL; 733 } 734 735 736 /* 737 * Deletion 738 * 739 */ 740 741 static struct rt6_info * fib6_find_prefix(struct fib6_node *fn) 742 { 743 if (fn->fn_flags&RTN_ROOT) 744 return &ip6_null_entry; 745 746 while(fn) { 747 if(fn->left) 748 return fn->left->leaf; 749 750 if(fn->right) 751 return fn->right->leaf; 752 753 fn = SUBTREE(fn); 754 } 755 return NULL; 756 } 757 758 /* 759 * Called to trim the tree of intermediate nodes when possible. "fn" 760 * is the node we want to try and remove. 761 */ 762 763 static struct fib6_node * fib6_repair_tree(struct fib6_node *fn) 764 { 765 int children; 766 int nstate; 767 struct fib6_node *child, *pn; 768 struct fib6_walker_t *w; 769 int iter = 0; 770 771 for (;;) { 772 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 773 iter++; 774 775 BUG_TRAP(!(fn->fn_flags&RTN_RTINFO)); 776 BUG_TRAP(!(fn->fn_flags&RTN_TL_ROOT)); 777 BUG_TRAP(fn->leaf==NULL); 778 779 children = 0; 780 child = NULL; 781 if (fn->right) child = fn->right, children |= 1; 782 if (fn->left) child = fn->left, children |= 2; 783 784 if (children == 3 || SUBTREE(fn) 785 #ifdef CONFIG_IPV6_SUBTREES 786 /* Subtree root (i.e. fn) may have one child */ 787 || (children && fn->fn_flags&RTN_ROOT) 788 #endif 789 ) { 790 fn->leaf = fib6_find_prefix(fn); 791 #if RT6_DEBUG >= 2 792 if (fn->leaf==NULL) { 793 BUG_TRAP(fn->leaf); 794 fn->leaf = &ip6_null_entry; 795 } 796 #endif 797 atomic_inc(&fn->leaf->rt6i_ref); 798 return fn->parent; 799 } 800 801 pn = fn->parent; 802 #ifdef CONFIG_IPV6_SUBTREES 803 if (SUBTREE(pn) == fn) { 804 BUG_TRAP(fn->fn_flags&RTN_ROOT); 805 SUBTREE(pn) = NULL; 806 nstate = FWS_L; 807 } else { 808 BUG_TRAP(!(fn->fn_flags&RTN_ROOT)); 809 #endif 810 if (pn->right == fn) pn->right = child; 811 else if (pn->left == fn) pn->left = child; 812 #if RT6_DEBUG >= 2 813 else BUG_TRAP(0); 814 #endif 815 if (child) 816 child->parent = pn; 817 nstate = FWS_R; 818 #ifdef CONFIG_IPV6_SUBTREES 819 } 820 #endif 821 822 read_lock(&fib6_walker_lock); 823 FOR_WALKERS(w) { 824 if (child == NULL) { 825 if (w->root == fn) { 826 w->root = w->node = NULL; 827 RT6_TRACE("W %p adjusted by delroot 1\n", w); 828 } else if (w->node == fn) { 829 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 830 w->node = pn; 831 w->state = nstate; 832 } 833 } else { 834 if (w->root == fn) { 835 w->root = child; 836 RT6_TRACE("W %p adjusted by delroot 2\n", w); 837 } 838 if (w->node == fn) { 839 w->node = child; 840 if (children&2) { 841 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 842 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT; 843 } else { 844 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 845 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT; 846 } 847 } 848 } 849 } 850 read_unlock(&fib6_walker_lock); 851 852 node_free(fn); 853 if (pn->fn_flags&RTN_RTINFO || SUBTREE(pn)) 854 return pn; 855 856 rt6_release(pn->leaf); 857 pn->leaf = NULL; 858 fn = pn; 859 } 860 } 861 862 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, 863 struct nlmsghdr *nlh, void *_rtattr, struct netlink_skb_parms *req) 864 { 865 struct fib6_walker_t *w; 866 struct rt6_info *rt = *rtp; 867 868 RT6_TRACE("fib6_del_route\n"); 869 870 /* Unlink it */ 871 *rtp = rt->u.next; 872 rt->rt6i_node = NULL; 873 rt6_stats.fib_rt_entries--; 874 rt6_stats.fib_discarded_routes++; 875 876 /* Adjust walkers */ 877 read_lock(&fib6_walker_lock); 878 FOR_WALKERS(w) { 879 if (w->state == FWS_C && w->leaf == rt) { 880 RT6_TRACE("walker %p adjusted by delroute\n", w); 881 w->leaf = rt->u.next; 882 if (w->leaf == NULL) 883 w->state = FWS_U; 884 } 885 } 886 read_unlock(&fib6_walker_lock); 887 888 rt->u.next = NULL; 889 890 if (fn->leaf == NULL && fn->fn_flags&RTN_TL_ROOT) 891 fn->leaf = &ip6_null_entry; 892 893 /* If it was last route, expunge its radix tree node */ 894 if (fn->leaf == NULL) { 895 fn->fn_flags &= ~RTN_RTINFO; 896 rt6_stats.fib_route_nodes--; 897 fn = fib6_repair_tree(fn); 898 } 899 900 if (atomic_read(&rt->rt6i_ref) != 1) { 901 /* This route is used as dummy address holder in some split 902 * nodes. It is not leaked, but it still holds other resources, 903 * which must be released in time. So, scan ascendant nodes 904 * and replace dummy references to this route with references 905 * to still alive ones. 906 */ 907 while (fn) { 908 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) { 909 fn->leaf = fib6_find_prefix(fn); 910 atomic_inc(&fn->leaf->rt6i_ref); 911 rt6_release(rt); 912 } 913 fn = fn->parent; 914 } 915 /* No more references are possible at this point. */ 916 if (atomic_read(&rt->rt6i_ref) != 1) BUG(); 917 } 918 919 inet6_rt_notify(RTM_DELROUTE, rt, nlh, req); 920 rt6_release(rt); 921 } 922 923 int fib6_del(struct rt6_info *rt, struct nlmsghdr *nlh, void *_rtattr, struct netlink_skb_parms *req) 924 { 925 struct fib6_node *fn = rt->rt6i_node; 926 struct rt6_info **rtp; 927 928 #if RT6_DEBUG >= 2 929 if (rt->u.dst.obsolete>0) { 930 BUG_TRAP(fn==NULL); 931 return -ENOENT; 932 } 933 #endif 934 if (fn == NULL || rt == &ip6_null_entry) 935 return -ENOENT; 936 937 BUG_TRAP(fn->fn_flags&RTN_RTINFO); 938 939 if (!(rt->rt6i_flags&RTF_CACHE)) 940 fib6_prune_clones(fn, rt); 941 942 /* 943 * Walk the leaf entries looking for ourself 944 */ 945 946 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.next) { 947 if (*rtp == rt) { 948 fib6_del_route(fn, rtp, nlh, _rtattr, req); 949 return 0; 950 } 951 } 952 return -ENOENT; 953 } 954 955 /* 956 * Tree traversal function. 957 * 958 * Certainly, it is not interrupt safe. 959 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 960 * It means, that we can modify tree during walking 961 * and use this function for garbage collection, clone pruning, 962 * cleaning tree when a device goes down etc. etc. 963 * 964 * It guarantees that every node will be traversed, 965 * and that it will be traversed only once. 966 * 967 * Callback function w->func may return: 968 * 0 -> continue walking. 969 * positive value -> walking is suspended (used by tree dumps, 970 * and probably by gc, if it will be split to several slices) 971 * negative value -> terminate walking. 972 * 973 * The function itself returns: 974 * 0 -> walk is complete. 975 * >0 -> walk is incomplete (i.e. suspended) 976 * <0 -> walk is terminated by an error. 977 */ 978 979 int fib6_walk_continue(struct fib6_walker_t *w) 980 { 981 struct fib6_node *fn, *pn; 982 983 for (;;) { 984 fn = w->node; 985 if (fn == NULL) 986 return 0; 987 988 if (w->prune && fn != w->root && 989 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) { 990 w->state = FWS_C; 991 w->leaf = fn->leaf; 992 } 993 switch (w->state) { 994 #ifdef CONFIG_IPV6_SUBTREES 995 case FWS_S: 996 if (SUBTREE(fn)) { 997 w->node = SUBTREE(fn); 998 continue; 999 } 1000 w->state = FWS_L; 1001 #endif 1002 case FWS_L: 1003 if (fn->left) { 1004 w->node = fn->left; 1005 w->state = FWS_INIT; 1006 continue; 1007 } 1008 w->state = FWS_R; 1009 case FWS_R: 1010 if (fn->right) { 1011 w->node = fn->right; 1012 w->state = FWS_INIT; 1013 continue; 1014 } 1015 w->state = FWS_C; 1016 w->leaf = fn->leaf; 1017 case FWS_C: 1018 if (w->leaf && fn->fn_flags&RTN_RTINFO) { 1019 int err = w->func(w); 1020 if (err) 1021 return err; 1022 continue; 1023 } 1024 w->state = FWS_U; 1025 case FWS_U: 1026 if (fn == w->root) 1027 return 0; 1028 pn = fn->parent; 1029 w->node = pn; 1030 #ifdef CONFIG_IPV6_SUBTREES 1031 if (SUBTREE(pn) == fn) { 1032 BUG_TRAP(fn->fn_flags&RTN_ROOT); 1033 w->state = FWS_L; 1034 continue; 1035 } 1036 #endif 1037 if (pn->left == fn) { 1038 w->state = FWS_R; 1039 continue; 1040 } 1041 if (pn->right == fn) { 1042 w->state = FWS_C; 1043 w->leaf = w->node->leaf; 1044 continue; 1045 } 1046 #if RT6_DEBUG >= 2 1047 BUG_TRAP(0); 1048 #endif 1049 } 1050 } 1051 } 1052 1053 int fib6_walk(struct fib6_walker_t *w) 1054 { 1055 int res; 1056 1057 w->state = FWS_INIT; 1058 w->node = w->root; 1059 1060 fib6_walker_link(w); 1061 res = fib6_walk_continue(w); 1062 if (res <= 0) 1063 fib6_walker_unlink(w); 1064 return res; 1065 } 1066 1067 static int fib6_clean_node(struct fib6_walker_t *w) 1068 { 1069 int res; 1070 struct rt6_info *rt; 1071 struct fib6_cleaner_t *c = (struct fib6_cleaner_t*)w; 1072 1073 for (rt = w->leaf; rt; rt = rt->u.next) { 1074 res = c->func(rt, c->arg); 1075 if (res < 0) { 1076 w->leaf = rt; 1077 res = fib6_del(rt, NULL, NULL, NULL); 1078 if (res) { 1079 #if RT6_DEBUG >= 2 1080 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res); 1081 #endif 1082 continue; 1083 } 1084 return 0; 1085 } 1086 BUG_TRAP(res==0); 1087 } 1088 w->leaf = rt; 1089 return 0; 1090 } 1091 1092 /* 1093 * Convenient frontend to tree walker. 1094 * 1095 * func is called on each route. 1096 * It may return -1 -> delete this route. 1097 * 0 -> continue walking 1098 * 1099 * prune==1 -> only immediate children of node (certainly, 1100 * ignoring pure split nodes) will be scanned. 1101 */ 1102 1103 void fib6_clean_tree(struct fib6_node *root, 1104 int (*func)(struct rt6_info *, void *arg), 1105 int prune, void *arg) 1106 { 1107 struct fib6_cleaner_t c; 1108 1109 c.w.root = root; 1110 c.w.func = fib6_clean_node; 1111 c.w.prune = prune; 1112 c.func = func; 1113 c.arg = arg; 1114 1115 fib6_walk(&c.w); 1116 } 1117 1118 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1119 { 1120 if (rt->rt6i_flags & RTF_CACHE) { 1121 RT6_TRACE("pruning clone %p\n", rt); 1122 return -1; 1123 } 1124 1125 return 0; 1126 } 1127 1128 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt) 1129 { 1130 fib6_clean_tree(fn, fib6_prune_clone, 1, rt); 1131 } 1132 1133 /* 1134 * Garbage collection 1135 */ 1136 1137 static struct fib6_gc_args 1138 { 1139 int timeout; 1140 int more; 1141 } gc_args; 1142 1143 static int fib6_age(struct rt6_info *rt, void *arg) 1144 { 1145 unsigned long now = jiffies; 1146 1147 /* 1148 * check addrconf expiration here. 1149 * Routes are expired even if they are in use. 1150 * 1151 * Also age clones. Note, that clones are aged out 1152 * only if they are not in use now. 1153 */ 1154 1155 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) { 1156 if (time_after(now, rt->rt6i_expires)) { 1157 RT6_TRACE("expiring %p\n", rt); 1158 rt6_reset_dflt_pointer(rt); 1159 return -1; 1160 } 1161 gc_args.more++; 1162 } else if (rt->rt6i_flags & RTF_CACHE) { 1163 if (atomic_read(&rt->u.dst.__refcnt) == 0 && 1164 time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) { 1165 RT6_TRACE("aging clone %p\n", rt); 1166 return -1; 1167 } else if ((rt->rt6i_flags & RTF_GATEWAY) && 1168 (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) { 1169 RT6_TRACE("purging route %p via non-router but gateway\n", 1170 rt); 1171 return -1; 1172 } 1173 gc_args.more++; 1174 } 1175 1176 return 0; 1177 } 1178 1179 static DEFINE_SPINLOCK(fib6_gc_lock); 1180 1181 void fib6_run_gc(unsigned long dummy) 1182 { 1183 if (dummy != ~0UL) { 1184 spin_lock_bh(&fib6_gc_lock); 1185 gc_args.timeout = dummy ? (int)dummy : ip6_rt_gc_interval; 1186 } else { 1187 local_bh_disable(); 1188 if (!spin_trylock(&fib6_gc_lock)) { 1189 mod_timer(&ip6_fib_timer, jiffies + HZ); 1190 local_bh_enable(); 1191 return; 1192 } 1193 gc_args.timeout = ip6_rt_gc_interval; 1194 } 1195 gc_args.more = 0; 1196 1197 1198 write_lock_bh(&rt6_lock); 1199 ndisc_dst_gc(&gc_args.more); 1200 fib6_clean_tree(&ip6_routing_table, fib6_age, 0, NULL); 1201 write_unlock_bh(&rt6_lock); 1202 1203 if (gc_args.more) 1204 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval); 1205 else { 1206 del_timer(&ip6_fib_timer); 1207 ip6_fib_timer.expires = 0; 1208 } 1209 spin_unlock_bh(&fib6_gc_lock); 1210 } 1211 1212 void __init fib6_init(void) 1213 { 1214 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1215 sizeof(struct fib6_node), 1216 0, SLAB_HWCACHE_ALIGN, 1217 NULL, NULL); 1218 if (!fib6_node_kmem) 1219 panic("cannot create fib6_nodes cache"); 1220 } 1221 1222 void fib6_gc_cleanup(void) 1223 { 1224 del_timer(&ip6_fib_timer); 1225 kmem_cache_destroy(fib6_node_kmem); 1226 } 1227