1 /* 2 * Linux INET6 implementation 3 * Forwarding Information Database 4 * 5 * Authors: 6 * Pedro Roque <roque@di.fc.ul.pt> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 */ 13 14 /* 15 * Changes: 16 * Yuji SEKIYA @USAGI: Support default route on router node; 17 * remove ip6_null_entry from the top of 18 * routing table. 19 * Ville Nuorvala: Fixed routing subtrees. 20 */ 21 22 #define pr_fmt(fmt) "IPv6: " fmt 23 24 #include <linux/errno.h> 25 #include <linux/types.h> 26 #include <linux/net.h> 27 #include <linux/route.h> 28 #include <linux/netdevice.h> 29 #include <linux/in6.h> 30 #include <linux/init.h> 31 #include <linux/list.h> 32 #include <linux/slab.h> 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...) pr_debug(x) 45 #else 46 #define RT6_TRACE(x...) do { ; } while (0) 47 #endif 48 49 static struct kmem_cache * fib6_node_kmem __read_mostly; 50 51 enum fib_walk_state_t 52 { 53 #ifdef CONFIG_IPV6_SUBTREES 54 FWS_S, 55 #endif 56 FWS_L, 57 FWS_R, 58 FWS_C, 59 FWS_U 60 }; 61 62 struct fib6_cleaner_t 63 { 64 struct fib6_walker_t w; 65 struct net *net; 66 int (*func)(struct rt6_info *, void *arg); 67 void *arg; 68 }; 69 70 static DEFINE_RWLOCK(fib6_walker_lock); 71 72 #ifdef CONFIG_IPV6_SUBTREES 73 #define FWS_INIT FWS_S 74 #else 75 #define FWS_INIT FWS_L 76 #endif 77 78 static void fib6_prune_clones(struct net *net, struct fib6_node *fn, 79 struct rt6_info *rt); 80 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn); 81 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn); 82 static int fib6_walk(struct fib6_walker_t *w); 83 static int fib6_walk_continue(struct fib6_walker_t *w); 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 void fib6_gc_timer_cb(unsigned long arg); 95 96 static LIST_HEAD(fib6_walkers); 97 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh) 98 99 static inline void fib6_walker_link(struct fib6_walker_t *w) 100 { 101 write_lock_bh(&fib6_walker_lock); 102 list_add(&w->lh, &fib6_walkers); 103 write_unlock_bh(&fib6_walker_lock); 104 } 105 106 static inline void fib6_walker_unlink(struct fib6_walker_t *w) 107 { 108 write_lock_bh(&fib6_walker_lock); 109 list_del(&w->lh); 110 write_unlock_bh(&fib6_walker_lock); 111 } 112 static __inline__ u32 fib6_new_sernum(void) 113 { 114 u32 n = ++rt_sernum; 115 if ((__s32)n <= 0) 116 rt_sernum = n = 1; 117 return n; 118 } 119 120 /* 121 * Auxiliary address test functions for the radix tree. 122 * 123 * These assume a 32bit processor (although it will work on 124 * 64bit processors) 125 */ 126 127 /* 128 * test bit 129 */ 130 #if defined(__LITTLE_ENDIAN) 131 # define BITOP_BE32_SWIZZLE (0x1F & ~7) 132 #else 133 # define BITOP_BE32_SWIZZLE 0 134 #endif 135 136 static __inline__ __be32 addr_bit_set(const void *token, int fn_bit) 137 { 138 const __be32 *addr = token; 139 /* 140 * Here, 141 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f) 142 * is optimized version of 143 * htonl(1 << ((~fn_bit)&0x1F)) 144 * See include/asm-generic/bitops/le.h. 145 */ 146 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) & 147 addr[fn_bit >> 5]; 148 } 149 150 static __inline__ struct fib6_node * node_alloc(void) 151 { 152 struct fib6_node *fn; 153 154 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC); 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->dst); 168 } 169 170 static void fib6_link_table(struct net *net, struct fib6_table *tb) 171 { 172 unsigned int h; 173 174 /* 175 * Initialize table lock at a single place to give lockdep a key, 176 * tables aren't visible prior to being linked to the list. 177 */ 178 rwlock_init(&tb->tb6_lock); 179 180 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1); 181 182 /* 183 * No protection necessary, this is the only list mutatation 184 * operation, tables never disappear once they exist. 185 */ 186 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]); 187 } 188 189 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 190 191 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id) 192 { 193 struct fib6_table *table; 194 195 table = kzalloc(sizeof(*table), GFP_ATOMIC); 196 if (table) { 197 table->tb6_id = id; 198 table->tb6_root.leaf = net->ipv6.ip6_null_entry; 199 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 200 inet_peer_base_init(&table->tb6_peers); 201 } 202 203 return table; 204 } 205 206 struct fib6_table *fib6_new_table(struct net *net, u32 id) 207 { 208 struct fib6_table *tb; 209 210 if (id == 0) 211 id = RT6_TABLE_MAIN; 212 tb = fib6_get_table(net, id); 213 if (tb) 214 return tb; 215 216 tb = fib6_alloc_table(net, id); 217 if (tb) 218 fib6_link_table(net, tb); 219 220 return tb; 221 } 222 223 struct fib6_table *fib6_get_table(struct net *net, u32 id) 224 { 225 struct fib6_table *tb; 226 struct hlist_head *head; 227 struct hlist_node *node; 228 unsigned int h; 229 230 if (id == 0) 231 id = RT6_TABLE_MAIN; 232 h = id & (FIB6_TABLE_HASHSZ - 1); 233 rcu_read_lock(); 234 head = &net->ipv6.fib_table_hash[h]; 235 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) { 236 if (tb->tb6_id == id) { 237 rcu_read_unlock(); 238 return tb; 239 } 240 } 241 rcu_read_unlock(); 242 243 return NULL; 244 } 245 246 static void __net_init fib6_tables_init(struct net *net) 247 { 248 fib6_link_table(net, net->ipv6.fib6_main_tbl); 249 fib6_link_table(net, net->ipv6.fib6_local_tbl); 250 } 251 #else 252 253 struct fib6_table *fib6_new_table(struct net *net, u32 id) 254 { 255 return fib6_get_table(net, id); 256 } 257 258 struct fib6_table *fib6_get_table(struct net *net, u32 id) 259 { 260 return net->ipv6.fib6_main_tbl; 261 } 262 263 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, 264 int flags, pol_lookup_t lookup) 265 { 266 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags); 267 } 268 269 static void __net_init fib6_tables_init(struct net *net) 270 { 271 fib6_link_table(net, net->ipv6.fib6_main_tbl); 272 } 273 274 #endif 275 276 static int fib6_dump_node(struct fib6_walker_t *w) 277 { 278 int res; 279 struct rt6_info *rt; 280 281 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 282 res = rt6_dump_route(rt, w->args); 283 if (res < 0) { 284 /* Frame is full, suspend walking */ 285 w->leaf = rt; 286 return 1; 287 } 288 WARN_ON(res == 0); 289 } 290 w->leaf = NULL; 291 return 0; 292 } 293 294 static void fib6_dump_end(struct netlink_callback *cb) 295 { 296 struct fib6_walker_t *w = (void*)cb->args[2]; 297 298 if (w) { 299 if (cb->args[4]) { 300 cb->args[4] = 0; 301 fib6_walker_unlink(w); 302 } 303 cb->args[2] = 0; 304 kfree(w); 305 } 306 cb->done = (void*)cb->args[3]; 307 cb->args[1] = 3; 308 } 309 310 static int fib6_dump_done(struct netlink_callback *cb) 311 { 312 fib6_dump_end(cb); 313 return cb->done ? cb->done(cb) : 0; 314 } 315 316 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 317 struct netlink_callback *cb) 318 { 319 struct fib6_walker_t *w; 320 int res; 321 322 w = (void *)cb->args[2]; 323 w->root = &table->tb6_root; 324 325 if (cb->args[4] == 0) { 326 w->count = 0; 327 w->skip = 0; 328 329 read_lock_bh(&table->tb6_lock); 330 res = fib6_walk(w); 331 read_unlock_bh(&table->tb6_lock); 332 if (res > 0) { 333 cb->args[4] = 1; 334 cb->args[5] = w->root->fn_sernum; 335 } 336 } else { 337 if (cb->args[5] != w->root->fn_sernum) { 338 /* Begin at the root if the tree changed */ 339 cb->args[5] = w->root->fn_sernum; 340 w->state = FWS_INIT; 341 w->node = w->root; 342 w->skip = w->count; 343 } else 344 w->skip = 0; 345 346 read_lock_bh(&table->tb6_lock); 347 res = fib6_walk_continue(w); 348 read_unlock_bh(&table->tb6_lock); 349 if (res <= 0) { 350 fib6_walker_unlink(w); 351 cb->args[4] = 0; 352 } 353 } 354 355 return res; 356 } 357 358 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 359 { 360 struct net *net = sock_net(skb->sk); 361 unsigned int h, s_h; 362 unsigned int e = 0, s_e; 363 struct rt6_rtnl_dump_arg arg; 364 struct fib6_walker_t *w; 365 struct fib6_table *tb; 366 struct hlist_node *node; 367 struct hlist_head *head; 368 int res = 0; 369 370 s_h = cb->args[0]; 371 s_e = cb->args[1]; 372 373 w = (void *)cb->args[2]; 374 if (!w) { 375 /* New dump: 376 * 377 * 1. hook callback destructor. 378 */ 379 cb->args[3] = (long)cb->done; 380 cb->done = fib6_dump_done; 381 382 /* 383 * 2. allocate and initialize walker. 384 */ 385 w = kzalloc(sizeof(*w), GFP_ATOMIC); 386 if (!w) 387 return -ENOMEM; 388 w->func = fib6_dump_node; 389 cb->args[2] = (long)w; 390 } 391 392 arg.skb = skb; 393 arg.cb = cb; 394 arg.net = net; 395 w->args = &arg; 396 397 rcu_read_lock(); 398 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) { 399 e = 0; 400 head = &net->ipv6.fib_table_hash[h]; 401 hlist_for_each_entry_rcu(tb, node, head, 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 rcu_read_unlock(); 413 cb->args[1] = e; 414 cb->args[0] = h; 415 416 res = res < 0 ? res : skb->len; 417 if (res <= 0) 418 fib6_dump_end(cb); 419 return res; 420 } 421 422 /* 423 * Routing Table 424 * 425 * return the appropriate node for a routing tree "add" operation 426 * by either creating and inserting or by returning an existing 427 * node. 428 */ 429 430 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr, 431 int addrlen, int plen, 432 int offset, int allow_create, 433 int replace_required) 434 { 435 struct fib6_node *fn, *in, *ln; 436 struct fib6_node *pn = NULL; 437 struct rt6key *key; 438 int bit; 439 __be32 dir = 0; 440 __u32 sernum = fib6_new_sernum(); 441 442 RT6_TRACE("fib6_add_1\n"); 443 444 /* insert node in tree */ 445 446 fn = root; 447 448 do { 449 key = (struct rt6key *)((u8 *)fn->leaf + offset); 450 451 /* 452 * Prefix match 453 */ 454 if (plen < fn->fn_bit || 455 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) { 456 if (!allow_create) { 457 if (replace_required) { 458 pr_warn("Can't replace route, no match found\n"); 459 return ERR_PTR(-ENOENT); 460 } 461 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 462 } 463 goto insert_above; 464 } 465 466 /* 467 * Exact match ? 468 */ 469 470 if (plen == fn->fn_bit) { 471 /* clean up an intermediate node */ 472 if (!(fn->fn_flags & RTN_RTINFO)) { 473 rt6_release(fn->leaf); 474 fn->leaf = NULL; 475 } 476 477 fn->fn_sernum = sernum; 478 479 return fn; 480 } 481 482 /* 483 * We have more bits to go 484 */ 485 486 /* Try to walk down on tree. */ 487 fn->fn_sernum = sernum; 488 dir = addr_bit_set(addr, fn->fn_bit); 489 pn = fn; 490 fn = dir ? fn->right: fn->left; 491 } while (fn); 492 493 if (!allow_create) { 494 /* We should not create new node because 495 * NLM_F_REPLACE was specified without NLM_F_CREATE 496 * I assume it is safe to require NLM_F_CREATE when 497 * REPLACE flag is used! Later we may want to remove the 498 * check for replace_required, because according 499 * to netlink specification, NLM_F_CREATE 500 * MUST be specified if new route is created. 501 * That would keep IPv6 consistent with IPv4 502 */ 503 if (replace_required) { 504 pr_warn("Can't replace route, no match found\n"); 505 return ERR_PTR(-ENOENT); 506 } 507 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 508 } 509 /* 510 * We walked to the bottom of tree. 511 * Create new leaf node without children. 512 */ 513 514 ln = node_alloc(); 515 516 if (!ln) 517 return ERR_PTR(-ENOMEM); 518 ln->fn_bit = plen; 519 520 ln->parent = pn; 521 ln->fn_sernum = sernum; 522 523 if (dir) 524 pn->right = ln; 525 else 526 pn->left = ln; 527 528 return ln; 529 530 531 insert_above: 532 /* 533 * split since we don't have a common prefix anymore or 534 * we have a less significant route. 535 * we've to insert an intermediate node on the list 536 * this new node will point to the one we need to create 537 * and the current 538 */ 539 540 pn = fn->parent; 541 542 /* find 1st bit in difference between the 2 addrs. 543 544 See comment in __ipv6_addr_diff: bit may be an invalid value, 545 but if it is >= plen, the value is ignored in any case. 546 */ 547 548 bit = __ipv6_addr_diff(addr, &key->addr, addrlen); 549 550 /* 551 * (intermediate)[in] 552 * / \ 553 * (new leaf node)[ln] (old node)[fn] 554 */ 555 if (plen > bit) { 556 in = node_alloc(); 557 ln = node_alloc(); 558 559 if (!in || !ln) { 560 if (in) 561 node_free(in); 562 if (ln) 563 node_free(ln); 564 return ERR_PTR(-ENOMEM); 565 } 566 567 /* 568 * new intermediate node. 569 * RTN_RTINFO will 570 * be off since that an address that chooses one of 571 * the branches would not match less specific routes 572 * in the other branch 573 */ 574 575 in->fn_bit = bit; 576 577 in->parent = pn; 578 in->leaf = fn->leaf; 579 atomic_inc(&in->leaf->rt6i_ref); 580 581 in->fn_sernum = sernum; 582 583 /* update parent pointer */ 584 if (dir) 585 pn->right = in; 586 else 587 pn->left = in; 588 589 ln->fn_bit = plen; 590 591 ln->parent = in; 592 fn->parent = in; 593 594 ln->fn_sernum = sernum; 595 596 if (addr_bit_set(addr, bit)) { 597 in->right = ln; 598 in->left = fn; 599 } else { 600 in->left = ln; 601 in->right = fn; 602 } 603 } else { /* plen <= bit */ 604 605 /* 606 * (new leaf node)[ln] 607 * / \ 608 * (old node)[fn] NULL 609 */ 610 611 ln = node_alloc(); 612 613 if (!ln) 614 return ERR_PTR(-ENOMEM); 615 616 ln->fn_bit = plen; 617 618 ln->parent = pn; 619 620 ln->fn_sernum = sernum; 621 622 if (dir) 623 pn->right = ln; 624 else 625 pn->left = ln; 626 627 if (addr_bit_set(&key->addr, plen)) 628 ln->right = fn; 629 else 630 ln->left = fn; 631 632 fn->parent = ln; 633 } 634 return ln; 635 } 636 637 /* 638 * Insert routing information in a node. 639 */ 640 641 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, 642 struct nl_info *info) 643 { 644 struct rt6_info *iter = NULL; 645 struct rt6_info **ins; 646 int replace = (info->nlh && 647 (info->nlh->nlmsg_flags & NLM_F_REPLACE)); 648 int add = (!info->nlh || 649 (info->nlh->nlmsg_flags & NLM_F_CREATE)); 650 int found = 0; 651 652 ins = &fn->leaf; 653 654 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) { 655 /* 656 * Search for duplicates 657 */ 658 659 if (iter->rt6i_metric == rt->rt6i_metric) { 660 /* 661 * Same priority level 662 */ 663 if (info->nlh && 664 (info->nlh->nlmsg_flags & NLM_F_EXCL)) 665 return -EEXIST; 666 if (replace) { 667 found++; 668 break; 669 } 670 671 if (iter->dst.dev == rt->dst.dev && 672 iter->rt6i_idev == rt->rt6i_idev && 673 ipv6_addr_equal(&iter->rt6i_gateway, 674 &rt->rt6i_gateway)) { 675 if (rt->rt6i_nsiblings) 676 rt->rt6i_nsiblings = 0; 677 if (!(iter->rt6i_flags & RTF_EXPIRES)) 678 return -EEXIST; 679 if (!(rt->rt6i_flags & RTF_EXPIRES)) 680 rt6_clean_expires(iter); 681 else 682 rt6_set_expires(iter, rt->dst.expires); 683 return -EEXIST; 684 } 685 /* If we have the same destination and the same metric, 686 * but not the same gateway, then the route we try to 687 * add is sibling to this route, increment our counter 688 * of siblings, and later we will add our route to the 689 * list. 690 * Only static routes (which don't have flag 691 * RTF_EXPIRES) are used for ECMPv6. 692 * 693 * To avoid long list, we only had siblings if the 694 * route have a gateway. 695 */ 696 if (rt->rt6i_flags & RTF_GATEWAY && 697 !(rt->rt6i_flags & RTF_EXPIRES) && 698 !(iter->rt6i_flags & RTF_EXPIRES)) 699 rt->rt6i_nsiblings++; 700 } 701 702 if (iter->rt6i_metric > rt->rt6i_metric) 703 break; 704 705 ins = &iter->dst.rt6_next; 706 } 707 708 /* Reset round-robin state, if necessary */ 709 if (ins == &fn->leaf) 710 fn->rr_ptr = NULL; 711 712 /* Link this route to others same route. */ 713 if (rt->rt6i_nsiblings) { 714 unsigned int rt6i_nsiblings; 715 struct rt6_info *sibling, *temp_sibling; 716 717 /* Find the first route that have the same metric */ 718 sibling = fn->leaf; 719 while (sibling) { 720 if (sibling->rt6i_metric == rt->rt6i_metric) { 721 list_add_tail(&rt->rt6i_siblings, 722 &sibling->rt6i_siblings); 723 break; 724 } 725 sibling = sibling->dst.rt6_next; 726 } 727 /* For each sibling in the list, increment the counter of 728 * siblings. BUG() if counters does not match, list of siblings 729 * is broken! 730 */ 731 rt6i_nsiblings = 0; 732 list_for_each_entry_safe(sibling, temp_sibling, 733 &rt->rt6i_siblings, rt6i_siblings) { 734 sibling->rt6i_nsiblings++; 735 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings); 736 rt6i_nsiblings++; 737 } 738 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings); 739 } 740 741 /* 742 * insert node 743 */ 744 if (!replace) { 745 if (!add) 746 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 747 748 add: 749 rt->dst.rt6_next = iter; 750 *ins = rt; 751 rt->rt6i_node = fn; 752 atomic_inc(&rt->rt6i_ref); 753 inet6_rt_notify(RTM_NEWROUTE, rt, info); 754 info->nl_net->ipv6.rt6_stats->fib_rt_entries++; 755 756 if (!(fn->fn_flags & RTN_RTINFO)) { 757 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 758 fn->fn_flags |= RTN_RTINFO; 759 } 760 761 } else { 762 if (!found) { 763 if (add) 764 goto add; 765 pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); 766 return -ENOENT; 767 } 768 *ins = rt; 769 rt->rt6i_node = fn; 770 rt->dst.rt6_next = iter->dst.rt6_next; 771 atomic_inc(&rt->rt6i_ref); 772 inet6_rt_notify(RTM_NEWROUTE, rt, info); 773 rt6_release(iter); 774 if (!(fn->fn_flags & RTN_RTINFO)) { 775 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 776 fn->fn_flags |= RTN_RTINFO; 777 } 778 } 779 780 return 0; 781 } 782 783 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt) 784 { 785 if (!timer_pending(&net->ipv6.ip6_fib_timer) && 786 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE))) 787 mod_timer(&net->ipv6.ip6_fib_timer, 788 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 789 } 790 791 void fib6_force_start_gc(struct net *net) 792 { 793 if (!timer_pending(&net->ipv6.ip6_fib_timer)) 794 mod_timer(&net->ipv6.ip6_fib_timer, 795 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 796 } 797 798 /* 799 * Add routing information to the routing tree. 800 * <destination addr>/<source addr> 801 * with source addr info in sub-trees 802 */ 803 804 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info) 805 { 806 struct fib6_node *fn, *pn = NULL; 807 int err = -ENOMEM; 808 int allow_create = 1; 809 int replace_required = 0; 810 811 if (info->nlh) { 812 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) 813 allow_create = 0; 814 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) 815 replace_required = 1; 816 } 817 if (!allow_create && !replace_required) 818 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); 819 820 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr), 821 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst), 822 allow_create, replace_required); 823 824 if (IS_ERR(fn)) { 825 err = PTR_ERR(fn); 826 goto out; 827 } 828 829 pn = fn; 830 831 #ifdef CONFIG_IPV6_SUBTREES 832 if (rt->rt6i_src.plen) { 833 struct fib6_node *sn; 834 835 if (!fn->subtree) { 836 struct fib6_node *sfn; 837 838 /* 839 * Create subtree. 840 * 841 * fn[main tree] 842 * | 843 * sfn[subtree root] 844 * \ 845 * sn[new leaf node] 846 */ 847 848 /* Create subtree root node */ 849 sfn = node_alloc(); 850 if (!sfn) 851 goto st_failure; 852 853 sfn->leaf = info->nl_net->ipv6.ip6_null_entry; 854 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref); 855 sfn->fn_flags = RTN_ROOT; 856 sfn->fn_sernum = fib6_new_sernum(); 857 858 /* Now add the first leaf node to new subtree */ 859 860 sn = fib6_add_1(sfn, &rt->rt6i_src.addr, 861 sizeof(struct in6_addr), rt->rt6i_src.plen, 862 offsetof(struct rt6_info, rt6i_src), 863 allow_create, replace_required); 864 865 if (IS_ERR(sn)) { 866 /* If it is failed, discard just allocated 867 root, and then (in st_failure) stale node 868 in main tree. 869 */ 870 node_free(sfn); 871 err = PTR_ERR(sn); 872 goto st_failure; 873 } 874 875 /* Now link new subtree to main tree */ 876 sfn->parent = fn; 877 fn->subtree = sfn; 878 } else { 879 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, 880 sizeof(struct in6_addr), rt->rt6i_src.plen, 881 offsetof(struct rt6_info, rt6i_src), 882 allow_create, replace_required); 883 884 if (IS_ERR(sn)) { 885 err = PTR_ERR(sn); 886 goto st_failure; 887 } 888 } 889 890 if (!fn->leaf) { 891 fn->leaf = rt; 892 atomic_inc(&rt->rt6i_ref); 893 } 894 fn = sn; 895 } 896 #endif 897 898 err = fib6_add_rt2node(fn, rt, info); 899 if (!err) { 900 fib6_start_gc(info->nl_net, rt); 901 if (!(rt->rt6i_flags & RTF_CACHE)) 902 fib6_prune_clones(info->nl_net, pn, rt); 903 } 904 905 out: 906 if (err) { 907 #ifdef CONFIG_IPV6_SUBTREES 908 /* 909 * If fib6_add_1 has cleared the old leaf pointer in the 910 * super-tree leaf node we have to find a new one for it. 911 */ 912 if (pn != fn && pn->leaf == rt) { 913 pn->leaf = NULL; 914 atomic_dec(&rt->rt6i_ref); 915 } 916 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) { 917 pn->leaf = fib6_find_prefix(info->nl_net, pn); 918 #if RT6_DEBUG >= 2 919 if (!pn->leaf) { 920 WARN_ON(pn->leaf == NULL); 921 pn->leaf = info->nl_net->ipv6.ip6_null_entry; 922 } 923 #endif 924 atomic_inc(&pn->leaf->rt6i_ref); 925 } 926 #endif 927 dst_free(&rt->dst); 928 } 929 return err; 930 931 #ifdef CONFIG_IPV6_SUBTREES 932 /* Subtree creation failed, probably main tree node 933 is orphan. If it is, shoot it. 934 */ 935 st_failure: 936 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) 937 fib6_repair_tree(info->nl_net, fn); 938 dst_free(&rt->dst); 939 return err; 940 #endif 941 } 942 943 /* 944 * Routing tree lookup 945 * 946 */ 947 948 struct lookup_args { 949 int offset; /* key offset on rt6_info */ 950 const struct in6_addr *addr; /* search key */ 951 }; 952 953 static struct fib6_node * fib6_lookup_1(struct fib6_node *root, 954 struct lookup_args *args) 955 { 956 struct fib6_node *fn; 957 __be32 dir; 958 959 if (unlikely(args->offset == 0)) 960 return NULL; 961 962 /* 963 * Descend on a tree 964 */ 965 966 fn = root; 967 968 for (;;) { 969 struct fib6_node *next; 970 971 dir = addr_bit_set(args->addr, fn->fn_bit); 972 973 next = dir ? fn->right : fn->left; 974 975 if (next) { 976 fn = next; 977 continue; 978 } 979 break; 980 } 981 982 while (fn) { 983 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) { 984 struct rt6key *key; 985 986 key = (struct rt6key *) ((u8 *) fn->leaf + 987 args->offset); 988 989 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 990 #ifdef CONFIG_IPV6_SUBTREES 991 if (fn->subtree) 992 fn = fib6_lookup_1(fn->subtree, args + 1); 993 #endif 994 if (!fn || fn->fn_flags & RTN_RTINFO) 995 return fn; 996 } 997 } 998 999 if (fn->fn_flags & RTN_ROOT) 1000 break; 1001 1002 fn = fn->parent; 1003 } 1004 1005 return NULL; 1006 } 1007 1008 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr, 1009 const struct in6_addr *saddr) 1010 { 1011 struct fib6_node *fn; 1012 struct lookup_args args[] = { 1013 { 1014 .offset = offsetof(struct rt6_info, rt6i_dst), 1015 .addr = daddr, 1016 }, 1017 #ifdef CONFIG_IPV6_SUBTREES 1018 { 1019 .offset = offsetof(struct rt6_info, rt6i_src), 1020 .addr = saddr, 1021 }, 1022 #endif 1023 { 1024 .offset = 0, /* sentinel */ 1025 } 1026 }; 1027 1028 fn = fib6_lookup_1(root, daddr ? args : args + 1); 1029 if (!fn || fn->fn_flags & RTN_TL_ROOT) 1030 fn = root; 1031 1032 return fn; 1033 } 1034 1035 /* 1036 * Get node with specified destination prefix (and source prefix, 1037 * if subtrees are used) 1038 */ 1039 1040 1041 static struct fib6_node * fib6_locate_1(struct fib6_node *root, 1042 const struct in6_addr *addr, 1043 int plen, int offset) 1044 { 1045 struct fib6_node *fn; 1046 1047 for (fn = root; fn ; ) { 1048 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); 1049 1050 /* 1051 * Prefix match 1052 */ 1053 if (plen < fn->fn_bit || 1054 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 1055 return NULL; 1056 1057 if (plen == fn->fn_bit) 1058 return fn; 1059 1060 /* 1061 * We have more bits to go 1062 */ 1063 if (addr_bit_set(addr, fn->fn_bit)) 1064 fn = fn->right; 1065 else 1066 fn = fn->left; 1067 } 1068 return NULL; 1069 } 1070 1071 struct fib6_node * fib6_locate(struct fib6_node *root, 1072 const struct in6_addr *daddr, int dst_len, 1073 const struct in6_addr *saddr, int src_len) 1074 { 1075 struct fib6_node *fn; 1076 1077 fn = fib6_locate_1(root, daddr, dst_len, 1078 offsetof(struct rt6_info, rt6i_dst)); 1079 1080 #ifdef CONFIG_IPV6_SUBTREES 1081 if (src_len) { 1082 WARN_ON(saddr == NULL); 1083 if (fn && fn->subtree) 1084 fn = fib6_locate_1(fn->subtree, saddr, src_len, 1085 offsetof(struct rt6_info, rt6i_src)); 1086 } 1087 #endif 1088 1089 if (fn && fn->fn_flags & RTN_RTINFO) 1090 return fn; 1091 1092 return NULL; 1093 } 1094 1095 1096 /* 1097 * Deletion 1098 * 1099 */ 1100 1101 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn) 1102 { 1103 if (fn->fn_flags & RTN_ROOT) 1104 return net->ipv6.ip6_null_entry; 1105 1106 while (fn) { 1107 if (fn->left) 1108 return fn->left->leaf; 1109 if (fn->right) 1110 return fn->right->leaf; 1111 1112 fn = FIB6_SUBTREE(fn); 1113 } 1114 return NULL; 1115 } 1116 1117 /* 1118 * Called to trim the tree of intermediate nodes when possible. "fn" 1119 * is the node we want to try and remove. 1120 */ 1121 1122 static struct fib6_node *fib6_repair_tree(struct net *net, 1123 struct fib6_node *fn) 1124 { 1125 int children; 1126 int nstate; 1127 struct fib6_node *child, *pn; 1128 struct fib6_walker_t *w; 1129 int iter = 0; 1130 1131 for (;;) { 1132 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1133 iter++; 1134 1135 WARN_ON(fn->fn_flags & RTN_RTINFO); 1136 WARN_ON(fn->fn_flags & RTN_TL_ROOT); 1137 WARN_ON(fn->leaf != NULL); 1138 1139 children = 0; 1140 child = NULL; 1141 if (fn->right) child = fn->right, children |= 1; 1142 if (fn->left) child = fn->left, children |= 2; 1143 1144 if (children == 3 || FIB6_SUBTREE(fn) 1145 #ifdef CONFIG_IPV6_SUBTREES 1146 /* Subtree root (i.e. fn) may have one child */ 1147 || (children && fn->fn_flags & RTN_ROOT) 1148 #endif 1149 ) { 1150 fn->leaf = fib6_find_prefix(net, fn); 1151 #if RT6_DEBUG >= 2 1152 if (!fn->leaf) { 1153 WARN_ON(!fn->leaf); 1154 fn->leaf = net->ipv6.ip6_null_entry; 1155 } 1156 #endif 1157 atomic_inc(&fn->leaf->rt6i_ref); 1158 return fn->parent; 1159 } 1160 1161 pn = fn->parent; 1162 #ifdef CONFIG_IPV6_SUBTREES 1163 if (FIB6_SUBTREE(pn) == fn) { 1164 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1165 FIB6_SUBTREE(pn) = NULL; 1166 nstate = FWS_L; 1167 } else { 1168 WARN_ON(fn->fn_flags & RTN_ROOT); 1169 #endif 1170 if (pn->right == fn) pn->right = child; 1171 else if (pn->left == fn) pn->left = child; 1172 #if RT6_DEBUG >= 2 1173 else 1174 WARN_ON(1); 1175 #endif 1176 if (child) 1177 child->parent = pn; 1178 nstate = FWS_R; 1179 #ifdef CONFIG_IPV6_SUBTREES 1180 } 1181 #endif 1182 1183 read_lock(&fib6_walker_lock); 1184 FOR_WALKERS(w) { 1185 if (!child) { 1186 if (w->root == fn) { 1187 w->root = w->node = NULL; 1188 RT6_TRACE("W %p adjusted by delroot 1\n", w); 1189 } else if (w->node == fn) { 1190 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1191 w->node = pn; 1192 w->state = nstate; 1193 } 1194 } else { 1195 if (w->root == fn) { 1196 w->root = child; 1197 RT6_TRACE("W %p adjusted by delroot 2\n", w); 1198 } 1199 if (w->node == fn) { 1200 w->node = child; 1201 if (children&2) { 1202 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1203 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT; 1204 } else { 1205 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1206 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT; 1207 } 1208 } 1209 } 1210 } 1211 read_unlock(&fib6_walker_lock); 1212 1213 node_free(fn); 1214 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) 1215 return pn; 1216 1217 rt6_release(pn->leaf); 1218 pn->leaf = NULL; 1219 fn = pn; 1220 } 1221 } 1222 1223 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, 1224 struct nl_info *info) 1225 { 1226 struct fib6_walker_t *w; 1227 struct rt6_info *rt = *rtp; 1228 struct net *net = info->nl_net; 1229 1230 RT6_TRACE("fib6_del_route\n"); 1231 1232 /* Unlink it */ 1233 *rtp = rt->dst.rt6_next; 1234 rt->rt6i_node = NULL; 1235 net->ipv6.rt6_stats->fib_rt_entries--; 1236 net->ipv6.rt6_stats->fib_discarded_routes++; 1237 1238 /* Reset round-robin state, if necessary */ 1239 if (fn->rr_ptr == rt) 1240 fn->rr_ptr = NULL; 1241 1242 /* Remove this entry from other siblings */ 1243 if (rt->rt6i_nsiblings) { 1244 struct rt6_info *sibling, *next_sibling; 1245 1246 list_for_each_entry_safe(sibling, next_sibling, 1247 &rt->rt6i_siblings, rt6i_siblings) 1248 sibling->rt6i_nsiblings--; 1249 rt->rt6i_nsiblings = 0; 1250 list_del_init(&rt->rt6i_siblings); 1251 } 1252 1253 /* Adjust walkers */ 1254 read_lock(&fib6_walker_lock); 1255 FOR_WALKERS(w) { 1256 if (w->state == FWS_C && w->leaf == rt) { 1257 RT6_TRACE("walker %p adjusted by delroute\n", w); 1258 w->leaf = rt->dst.rt6_next; 1259 if (!w->leaf) 1260 w->state = FWS_U; 1261 } 1262 } 1263 read_unlock(&fib6_walker_lock); 1264 1265 rt->dst.rt6_next = NULL; 1266 1267 /* If it was last route, expunge its radix tree node */ 1268 if (!fn->leaf) { 1269 fn->fn_flags &= ~RTN_RTINFO; 1270 net->ipv6.rt6_stats->fib_route_nodes--; 1271 fn = fib6_repair_tree(net, fn); 1272 } 1273 1274 if (atomic_read(&rt->rt6i_ref) != 1) { 1275 /* This route is used as dummy address holder in some split 1276 * nodes. It is not leaked, but it still holds other resources, 1277 * which must be released in time. So, scan ascendant nodes 1278 * and replace dummy references to this route with references 1279 * to still alive ones. 1280 */ 1281 while (fn) { 1282 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) { 1283 fn->leaf = fib6_find_prefix(net, fn); 1284 atomic_inc(&fn->leaf->rt6i_ref); 1285 rt6_release(rt); 1286 } 1287 fn = fn->parent; 1288 } 1289 /* No more references are possible at this point. */ 1290 BUG_ON(atomic_read(&rt->rt6i_ref) != 1); 1291 } 1292 1293 inet6_rt_notify(RTM_DELROUTE, rt, info); 1294 rt6_release(rt); 1295 } 1296 1297 int fib6_del(struct rt6_info *rt, struct nl_info *info) 1298 { 1299 struct net *net = info->nl_net; 1300 struct fib6_node *fn = rt->rt6i_node; 1301 struct rt6_info **rtp; 1302 1303 #if RT6_DEBUG >= 2 1304 if (rt->dst.obsolete>0) { 1305 WARN_ON(fn != NULL); 1306 return -ENOENT; 1307 } 1308 #endif 1309 if (!fn || rt == net->ipv6.ip6_null_entry) 1310 return -ENOENT; 1311 1312 WARN_ON(!(fn->fn_flags & RTN_RTINFO)); 1313 1314 if (!(rt->rt6i_flags & RTF_CACHE)) { 1315 struct fib6_node *pn = fn; 1316 #ifdef CONFIG_IPV6_SUBTREES 1317 /* clones of this route might be in another subtree */ 1318 if (rt->rt6i_src.plen) { 1319 while (!(pn->fn_flags & RTN_ROOT)) 1320 pn = pn->parent; 1321 pn = pn->parent; 1322 } 1323 #endif 1324 fib6_prune_clones(info->nl_net, pn, rt); 1325 } 1326 1327 /* 1328 * Walk the leaf entries looking for ourself 1329 */ 1330 1331 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) { 1332 if (*rtp == rt) { 1333 fib6_del_route(fn, rtp, info); 1334 return 0; 1335 } 1336 } 1337 return -ENOENT; 1338 } 1339 1340 /* 1341 * Tree traversal function. 1342 * 1343 * Certainly, it is not interrupt safe. 1344 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 1345 * It means, that we can modify tree during walking 1346 * and use this function for garbage collection, clone pruning, 1347 * cleaning tree when a device goes down etc. etc. 1348 * 1349 * It guarantees that every node will be traversed, 1350 * and that it will be traversed only once. 1351 * 1352 * Callback function w->func may return: 1353 * 0 -> continue walking. 1354 * positive value -> walking is suspended (used by tree dumps, 1355 * and probably by gc, if it will be split to several slices) 1356 * negative value -> terminate walking. 1357 * 1358 * The function itself returns: 1359 * 0 -> walk is complete. 1360 * >0 -> walk is incomplete (i.e. suspended) 1361 * <0 -> walk is terminated by an error. 1362 */ 1363 1364 static int fib6_walk_continue(struct fib6_walker_t *w) 1365 { 1366 struct fib6_node *fn, *pn; 1367 1368 for (;;) { 1369 fn = w->node; 1370 if (!fn) 1371 return 0; 1372 1373 if (w->prune && fn != w->root && 1374 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) { 1375 w->state = FWS_C; 1376 w->leaf = fn->leaf; 1377 } 1378 switch (w->state) { 1379 #ifdef CONFIG_IPV6_SUBTREES 1380 case FWS_S: 1381 if (FIB6_SUBTREE(fn)) { 1382 w->node = FIB6_SUBTREE(fn); 1383 continue; 1384 } 1385 w->state = FWS_L; 1386 #endif 1387 case FWS_L: 1388 if (fn->left) { 1389 w->node = fn->left; 1390 w->state = FWS_INIT; 1391 continue; 1392 } 1393 w->state = FWS_R; 1394 case FWS_R: 1395 if (fn->right) { 1396 w->node = fn->right; 1397 w->state = FWS_INIT; 1398 continue; 1399 } 1400 w->state = FWS_C; 1401 w->leaf = fn->leaf; 1402 case FWS_C: 1403 if (w->leaf && fn->fn_flags & RTN_RTINFO) { 1404 int err; 1405 1406 if (w->skip) { 1407 w->skip--; 1408 continue; 1409 } 1410 1411 err = w->func(w); 1412 if (err) 1413 return err; 1414 1415 w->count++; 1416 continue; 1417 } 1418 w->state = FWS_U; 1419 case FWS_U: 1420 if (fn == w->root) 1421 return 0; 1422 pn = fn->parent; 1423 w->node = pn; 1424 #ifdef CONFIG_IPV6_SUBTREES 1425 if (FIB6_SUBTREE(pn) == fn) { 1426 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1427 w->state = FWS_L; 1428 continue; 1429 } 1430 #endif 1431 if (pn->left == fn) { 1432 w->state = FWS_R; 1433 continue; 1434 } 1435 if (pn->right == fn) { 1436 w->state = FWS_C; 1437 w->leaf = w->node->leaf; 1438 continue; 1439 } 1440 #if RT6_DEBUG >= 2 1441 WARN_ON(1); 1442 #endif 1443 } 1444 } 1445 } 1446 1447 static int fib6_walk(struct fib6_walker_t *w) 1448 { 1449 int res; 1450 1451 w->state = FWS_INIT; 1452 w->node = w->root; 1453 1454 fib6_walker_link(w); 1455 res = fib6_walk_continue(w); 1456 if (res <= 0) 1457 fib6_walker_unlink(w); 1458 return res; 1459 } 1460 1461 static int fib6_clean_node(struct fib6_walker_t *w) 1462 { 1463 int res; 1464 struct rt6_info *rt; 1465 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w); 1466 struct nl_info info = { 1467 .nl_net = c->net, 1468 }; 1469 1470 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 1471 res = c->func(rt, c->arg); 1472 if (res < 0) { 1473 w->leaf = rt; 1474 res = fib6_del(rt, &info); 1475 if (res) { 1476 #if RT6_DEBUG >= 2 1477 pr_debug("%s: del failed: rt=%p@%p err=%d\n", 1478 __func__, rt, rt->rt6i_node, res); 1479 #endif 1480 continue; 1481 } 1482 return 0; 1483 } 1484 WARN_ON(res != 0); 1485 } 1486 w->leaf = rt; 1487 return 0; 1488 } 1489 1490 /* 1491 * Convenient frontend to tree walker. 1492 * 1493 * func is called on each route. 1494 * It may return -1 -> delete this route. 1495 * 0 -> continue walking 1496 * 1497 * prune==1 -> only immediate children of node (certainly, 1498 * ignoring pure split nodes) will be scanned. 1499 */ 1500 1501 static void fib6_clean_tree(struct net *net, struct fib6_node *root, 1502 int (*func)(struct rt6_info *, void *arg), 1503 int prune, void *arg) 1504 { 1505 struct fib6_cleaner_t c; 1506 1507 c.w.root = root; 1508 c.w.func = fib6_clean_node; 1509 c.w.prune = prune; 1510 c.w.count = 0; 1511 c.w.skip = 0; 1512 c.func = func; 1513 c.arg = arg; 1514 c.net = net; 1515 1516 fib6_walk(&c.w); 1517 } 1518 1519 void fib6_clean_all_ro(struct net *net, int (*func)(struct rt6_info *, void *arg), 1520 int prune, void *arg) 1521 { 1522 struct fib6_table *table; 1523 struct hlist_node *node; 1524 struct hlist_head *head; 1525 unsigned int h; 1526 1527 rcu_read_lock(); 1528 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 1529 head = &net->ipv6.fib_table_hash[h]; 1530 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) { 1531 read_lock_bh(&table->tb6_lock); 1532 fib6_clean_tree(net, &table->tb6_root, 1533 func, prune, arg); 1534 read_unlock_bh(&table->tb6_lock); 1535 } 1536 } 1537 rcu_read_unlock(); 1538 } 1539 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg), 1540 int prune, void *arg) 1541 { 1542 struct fib6_table *table; 1543 struct hlist_node *node; 1544 struct hlist_head *head; 1545 unsigned int h; 1546 1547 rcu_read_lock(); 1548 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 1549 head = &net->ipv6.fib_table_hash[h]; 1550 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) { 1551 write_lock_bh(&table->tb6_lock); 1552 fib6_clean_tree(net, &table->tb6_root, 1553 func, prune, arg); 1554 write_unlock_bh(&table->tb6_lock); 1555 } 1556 } 1557 rcu_read_unlock(); 1558 } 1559 1560 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1561 { 1562 if (rt->rt6i_flags & RTF_CACHE) { 1563 RT6_TRACE("pruning clone %p\n", rt); 1564 return -1; 1565 } 1566 1567 return 0; 1568 } 1569 1570 static void fib6_prune_clones(struct net *net, struct fib6_node *fn, 1571 struct rt6_info *rt) 1572 { 1573 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt); 1574 } 1575 1576 /* 1577 * Garbage collection 1578 */ 1579 1580 static struct fib6_gc_args 1581 { 1582 int timeout; 1583 int more; 1584 } gc_args; 1585 1586 static int fib6_age(struct rt6_info *rt, void *arg) 1587 { 1588 unsigned long now = jiffies; 1589 1590 /* 1591 * check addrconf expiration here. 1592 * Routes are expired even if they are in use. 1593 * 1594 * Also age clones. Note, that clones are aged out 1595 * only if they are not in use now. 1596 */ 1597 1598 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) { 1599 if (time_after(now, rt->dst.expires)) { 1600 RT6_TRACE("expiring %p\n", rt); 1601 return -1; 1602 } 1603 gc_args.more++; 1604 } else if (rt->rt6i_flags & RTF_CACHE) { 1605 if (atomic_read(&rt->dst.__refcnt) == 0 && 1606 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) { 1607 RT6_TRACE("aging clone %p\n", rt); 1608 return -1; 1609 } else if (rt->rt6i_flags & RTF_GATEWAY) { 1610 struct neighbour *neigh; 1611 __u8 neigh_flags = 0; 1612 1613 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway); 1614 if (neigh) { 1615 neigh_flags = neigh->flags; 1616 neigh_release(neigh); 1617 } 1618 if (!(neigh_flags & NTF_ROUTER)) { 1619 RT6_TRACE("purging route %p via non-router but gateway\n", 1620 rt); 1621 return -1; 1622 } 1623 } 1624 gc_args.more++; 1625 } 1626 1627 return 0; 1628 } 1629 1630 static DEFINE_SPINLOCK(fib6_gc_lock); 1631 1632 void fib6_run_gc(unsigned long expires, struct net *net) 1633 { 1634 if (expires != ~0UL) { 1635 spin_lock_bh(&fib6_gc_lock); 1636 gc_args.timeout = expires ? (int)expires : 1637 net->ipv6.sysctl.ip6_rt_gc_interval; 1638 } else { 1639 if (!spin_trylock_bh(&fib6_gc_lock)) { 1640 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); 1641 return; 1642 } 1643 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval; 1644 } 1645 1646 gc_args.more = icmp6_dst_gc(); 1647 1648 fib6_clean_all(net, fib6_age, 0, NULL); 1649 1650 if (gc_args.more) 1651 mod_timer(&net->ipv6.ip6_fib_timer, 1652 round_jiffies(jiffies 1653 + net->ipv6.sysctl.ip6_rt_gc_interval)); 1654 else 1655 del_timer(&net->ipv6.ip6_fib_timer); 1656 spin_unlock_bh(&fib6_gc_lock); 1657 } 1658 1659 static void fib6_gc_timer_cb(unsigned long arg) 1660 { 1661 fib6_run_gc(0, (struct net *)arg); 1662 } 1663 1664 static int __net_init fib6_net_init(struct net *net) 1665 { 1666 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; 1667 1668 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net); 1669 1670 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); 1671 if (!net->ipv6.rt6_stats) 1672 goto out_timer; 1673 1674 /* Avoid false sharing : Use at least a full cache line */ 1675 size = max_t(size_t, size, L1_CACHE_BYTES); 1676 1677 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); 1678 if (!net->ipv6.fib_table_hash) 1679 goto out_rt6_stats; 1680 1681 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), 1682 GFP_KERNEL); 1683 if (!net->ipv6.fib6_main_tbl) 1684 goto out_fib_table_hash; 1685 1686 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; 1687 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1688 net->ipv6.fib6_main_tbl->tb6_root.fn_flags = 1689 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1690 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); 1691 1692 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1693 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), 1694 GFP_KERNEL); 1695 if (!net->ipv6.fib6_local_tbl) 1696 goto out_fib6_main_tbl; 1697 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; 1698 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1699 net->ipv6.fib6_local_tbl->tb6_root.fn_flags = 1700 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1701 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); 1702 #endif 1703 fib6_tables_init(net); 1704 1705 return 0; 1706 1707 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1708 out_fib6_main_tbl: 1709 kfree(net->ipv6.fib6_main_tbl); 1710 #endif 1711 out_fib_table_hash: 1712 kfree(net->ipv6.fib_table_hash); 1713 out_rt6_stats: 1714 kfree(net->ipv6.rt6_stats); 1715 out_timer: 1716 return -ENOMEM; 1717 } 1718 1719 static void fib6_net_exit(struct net *net) 1720 { 1721 rt6_ifdown(net, NULL); 1722 del_timer_sync(&net->ipv6.ip6_fib_timer); 1723 1724 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1725 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers); 1726 kfree(net->ipv6.fib6_local_tbl); 1727 #endif 1728 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers); 1729 kfree(net->ipv6.fib6_main_tbl); 1730 kfree(net->ipv6.fib_table_hash); 1731 kfree(net->ipv6.rt6_stats); 1732 } 1733 1734 static struct pernet_operations fib6_net_ops = { 1735 .init = fib6_net_init, 1736 .exit = fib6_net_exit, 1737 }; 1738 1739 int __init fib6_init(void) 1740 { 1741 int ret = -ENOMEM; 1742 1743 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1744 sizeof(struct fib6_node), 1745 0, SLAB_HWCACHE_ALIGN, 1746 NULL); 1747 if (!fib6_node_kmem) 1748 goto out; 1749 1750 ret = register_pernet_subsys(&fib6_net_ops); 1751 if (ret) 1752 goto out_kmem_cache_create; 1753 1754 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib, 1755 NULL); 1756 if (ret) 1757 goto out_unregister_subsys; 1758 out: 1759 return ret; 1760 1761 out_unregister_subsys: 1762 unregister_pernet_subsys(&fib6_net_ops); 1763 out_kmem_cache_create: 1764 kmem_cache_destroy(fib6_node_kmem); 1765 goto out; 1766 } 1767 1768 void fib6_gc_cleanup(void) 1769 { 1770 unregister_pernet_subsys(&fib6_net_ops); 1771 kmem_cache_destroy(fib6_node_kmem); 1772 } 1773