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 unsigned int h; 228 229 if (id == 0) 230 id = RT6_TABLE_MAIN; 231 h = id & (FIB6_TABLE_HASHSZ - 1); 232 rcu_read_lock(); 233 head = &net->ipv6.fib_table_hash[h]; 234 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 235 if (tb->tb6_id == id) { 236 rcu_read_unlock(); 237 return tb; 238 } 239 } 240 rcu_read_unlock(); 241 242 return NULL; 243 } 244 245 static void __net_init fib6_tables_init(struct net *net) 246 { 247 fib6_link_table(net, net->ipv6.fib6_main_tbl); 248 fib6_link_table(net, net->ipv6.fib6_local_tbl); 249 } 250 #else 251 252 struct fib6_table *fib6_new_table(struct net *net, u32 id) 253 { 254 return fib6_get_table(net, id); 255 } 256 257 struct fib6_table *fib6_get_table(struct net *net, u32 id) 258 { 259 return net->ipv6.fib6_main_tbl; 260 } 261 262 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, 263 int flags, pol_lookup_t lookup) 264 { 265 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags); 266 } 267 268 static void __net_init fib6_tables_init(struct net *net) 269 { 270 fib6_link_table(net, net->ipv6.fib6_main_tbl); 271 } 272 273 #endif 274 275 static int fib6_dump_node(struct fib6_walker_t *w) 276 { 277 int res; 278 struct rt6_info *rt; 279 280 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 281 res = rt6_dump_route(rt, w->args); 282 if (res < 0) { 283 /* Frame is full, suspend walking */ 284 w->leaf = rt; 285 return 1; 286 } 287 WARN_ON(res == 0); 288 } 289 w->leaf = NULL; 290 return 0; 291 } 292 293 static void fib6_dump_end(struct netlink_callback *cb) 294 { 295 struct fib6_walker_t *w = (void*)cb->args[2]; 296 297 if (w) { 298 if (cb->args[4]) { 299 cb->args[4] = 0; 300 fib6_walker_unlink(w); 301 } 302 cb->args[2] = 0; 303 kfree(w); 304 } 305 cb->done = (void*)cb->args[3]; 306 cb->args[1] = 3; 307 } 308 309 static int fib6_dump_done(struct netlink_callback *cb) 310 { 311 fib6_dump_end(cb); 312 return cb->done ? cb->done(cb) : 0; 313 } 314 315 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 316 struct netlink_callback *cb) 317 { 318 struct fib6_walker_t *w; 319 int res; 320 321 w = (void *)cb->args[2]; 322 w->root = &table->tb6_root; 323 324 if (cb->args[4] == 0) { 325 w->count = 0; 326 w->skip = 0; 327 328 read_lock_bh(&table->tb6_lock); 329 res = fib6_walk(w); 330 read_unlock_bh(&table->tb6_lock); 331 if (res > 0) { 332 cb->args[4] = 1; 333 cb->args[5] = w->root->fn_sernum; 334 } 335 } else { 336 if (cb->args[5] != w->root->fn_sernum) { 337 /* Begin at the root if the tree changed */ 338 cb->args[5] = w->root->fn_sernum; 339 w->state = FWS_INIT; 340 w->node = w->root; 341 w->skip = w->count; 342 } else 343 w->skip = 0; 344 345 read_lock_bh(&table->tb6_lock); 346 res = fib6_walk_continue(w); 347 read_unlock_bh(&table->tb6_lock); 348 if (res <= 0) { 349 fib6_walker_unlink(w); 350 cb->args[4] = 0; 351 } 352 } 353 354 return res; 355 } 356 357 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 358 { 359 struct net *net = sock_net(skb->sk); 360 unsigned int h, s_h; 361 unsigned int e = 0, s_e; 362 struct rt6_rtnl_dump_arg arg; 363 struct fib6_walker_t *w; 364 struct fib6_table *tb; 365 struct hlist_head *head; 366 int res = 0; 367 368 s_h = cb->args[0]; 369 s_e = cb->args[1]; 370 371 w = (void *)cb->args[2]; 372 if (!w) { 373 /* New dump: 374 * 375 * 1. hook callback destructor. 376 */ 377 cb->args[3] = (long)cb->done; 378 cb->done = fib6_dump_done; 379 380 /* 381 * 2. allocate and initialize walker. 382 */ 383 w = kzalloc(sizeof(*w), GFP_ATOMIC); 384 if (!w) 385 return -ENOMEM; 386 w->func = fib6_dump_node; 387 cb->args[2] = (long)w; 388 } 389 390 arg.skb = skb; 391 arg.cb = cb; 392 arg.net = net; 393 w->args = &arg; 394 395 rcu_read_lock(); 396 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) { 397 e = 0; 398 head = &net->ipv6.fib_table_hash[h]; 399 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 400 if (e < s_e) 401 goto next; 402 res = fib6_dump_table(tb, skb, cb); 403 if (res != 0) 404 goto out; 405 next: 406 e++; 407 } 408 } 409 out: 410 rcu_read_unlock(); 411 cb->args[1] = e; 412 cb->args[0] = h; 413 414 res = res < 0 ? res : skb->len; 415 if (res <= 0) 416 fib6_dump_end(cb); 417 return res; 418 } 419 420 /* 421 * Routing Table 422 * 423 * return the appropriate node for a routing tree "add" operation 424 * by either creating and inserting or by returning an existing 425 * node. 426 */ 427 428 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr, 429 int addrlen, int plen, 430 int offset, int allow_create, 431 int replace_required) 432 { 433 struct fib6_node *fn, *in, *ln; 434 struct fib6_node *pn = NULL; 435 struct rt6key *key; 436 int bit; 437 __be32 dir = 0; 438 __u32 sernum = fib6_new_sernum(); 439 440 RT6_TRACE("fib6_add_1\n"); 441 442 /* insert node in tree */ 443 444 fn = root; 445 446 do { 447 key = (struct rt6key *)((u8 *)fn->leaf + offset); 448 449 /* 450 * Prefix match 451 */ 452 if (plen < fn->fn_bit || 453 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) { 454 if (!allow_create) { 455 if (replace_required) { 456 pr_warn("Can't replace route, no match found\n"); 457 return ERR_PTR(-ENOENT); 458 } 459 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 460 } 461 goto insert_above; 462 } 463 464 /* 465 * Exact match ? 466 */ 467 468 if (plen == fn->fn_bit) { 469 /* clean up an intermediate node */ 470 if (!(fn->fn_flags & RTN_RTINFO)) { 471 rt6_release(fn->leaf); 472 fn->leaf = NULL; 473 } 474 475 fn->fn_sernum = sernum; 476 477 return fn; 478 } 479 480 /* 481 * We have more bits to go 482 */ 483 484 /* Try to walk down on tree. */ 485 fn->fn_sernum = sernum; 486 dir = addr_bit_set(addr, fn->fn_bit); 487 pn = fn; 488 fn = dir ? fn->right: fn->left; 489 } while (fn); 490 491 if (!allow_create) { 492 /* We should not create new node because 493 * NLM_F_REPLACE was specified without NLM_F_CREATE 494 * I assume it is safe to require NLM_F_CREATE when 495 * REPLACE flag is used! Later we may want to remove the 496 * check for replace_required, because according 497 * to netlink specification, NLM_F_CREATE 498 * MUST be specified if new route is created. 499 * That would keep IPv6 consistent with IPv4 500 */ 501 if (replace_required) { 502 pr_warn("Can't replace route, no match found\n"); 503 return ERR_PTR(-ENOENT); 504 } 505 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 506 } 507 /* 508 * We walked to the bottom of tree. 509 * Create new leaf node without children. 510 */ 511 512 ln = node_alloc(); 513 514 if (!ln) 515 return ERR_PTR(-ENOMEM); 516 ln->fn_bit = plen; 517 518 ln->parent = pn; 519 ln->fn_sernum = sernum; 520 521 if (dir) 522 pn->right = ln; 523 else 524 pn->left = ln; 525 526 return ln; 527 528 529 insert_above: 530 /* 531 * split since we don't have a common prefix anymore or 532 * we have a less significant route. 533 * we've to insert an intermediate node on the list 534 * this new node will point to the one we need to create 535 * and the current 536 */ 537 538 pn = fn->parent; 539 540 /* find 1st bit in difference between the 2 addrs. 541 542 See comment in __ipv6_addr_diff: bit may be an invalid value, 543 but if it is >= plen, the value is ignored in any case. 544 */ 545 546 bit = __ipv6_addr_diff(addr, &key->addr, addrlen); 547 548 /* 549 * (intermediate)[in] 550 * / \ 551 * (new leaf node)[ln] (old node)[fn] 552 */ 553 if (plen > bit) { 554 in = node_alloc(); 555 ln = node_alloc(); 556 557 if (!in || !ln) { 558 if (in) 559 node_free(in); 560 if (ln) 561 node_free(ln); 562 return ERR_PTR(-ENOMEM); 563 } 564 565 /* 566 * new intermediate node. 567 * RTN_RTINFO will 568 * be off since that an address that chooses one of 569 * the branches would not match less specific routes 570 * in the other branch 571 */ 572 573 in->fn_bit = bit; 574 575 in->parent = pn; 576 in->leaf = fn->leaf; 577 atomic_inc(&in->leaf->rt6i_ref); 578 579 in->fn_sernum = sernum; 580 581 /* update parent pointer */ 582 if (dir) 583 pn->right = in; 584 else 585 pn->left = in; 586 587 ln->fn_bit = plen; 588 589 ln->parent = in; 590 fn->parent = in; 591 592 ln->fn_sernum = sernum; 593 594 if (addr_bit_set(addr, bit)) { 595 in->right = ln; 596 in->left = fn; 597 } else { 598 in->left = ln; 599 in->right = fn; 600 } 601 } else { /* plen <= bit */ 602 603 /* 604 * (new leaf node)[ln] 605 * / \ 606 * (old node)[fn] NULL 607 */ 608 609 ln = node_alloc(); 610 611 if (!ln) 612 return ERR_PTR(-ENOMEM); 613 614 ln->fn_bit = plen; 615 616 ln->parent = pn; 617 618 ln->fn_sernum = sernum; 619 620 if (dir) 621 pn->right = ln; 622 else 623 pn->left = ln; 624 625 if (addr_bit_set(&key->addr, plen)) 626 ln->right = fn; 627 else 628 ln->left = fn; 629 630 fn->parent = ln; 631 } 632 return ln; 633 } 634 635 /* 636 * Insert routing information in a node. 637 */ 638 639 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, 640 struct nl_info *info) 641 { 642 struct rt6_info *iter = NULL; 643 struct rt6_info **ins; 644 int replace = (info->nlh && 645 (info->nlh->nlmsg_flags & NLM_F_REPLACE)); 646 int add = (!info->nlh || 647 (info->nlh->nlmsg_flags & NLM_F_CREATE)); 648 int found = 0; 649 650 ins = &fn->leaf; 651 652 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) { 653 /* 654 * Search for duplicates 655 */ 656 657 if (iter->rt6i_metric == rt->rt6i_metric) { 658 /* 659 * Same priority level 660 */ 661 if (info->nlh && 662 (info->nlh->nlmsg_flags & NLM_F_EXCL)) 663 return -EEXIST; 664 if (replace) { 665 found++; 666 break; 667 } 668 669 if (iter->dst.dev == rt->dst.dev && 670 iter->rt6i_idev == rt->rt6i_idev && 671 ipv6_addr_equal(&iter->rt6i_gateway, 672 &rt->rt6i_gateway)) { 673 if (rt->rt6i_nsiblings) 674 rt->rt6i_nsiblings = 0; 675 if (!(iter->rt6i_flags & RTF_EXPIRES)) 676 return -EEXIST; 677 if (!(rt->rt6i_flags & RTF_EXPIRES)) 678 rt6_clean_expires(iter); 679 else 680 rt6_set_expires(iter, rt->dst.expires); 681 return -EEXIST; 682 } 683 /* If we have the same destination and the same metric, 684 * but not the same gateway, then the route we try to 685 * add is sibling to this route, increment our counter 686 * of siblings, and later we will add our route to the 687 * list. 688 * Only static routes (which don't have flag 689 * RTF_EXPIRES) are used for ECMPv6. 690 * 691 * To avoid long list, we only had siblings if the 692 * route have a gateway. 693 */ 694 if (rt->rt6i_flags & RTF_GATEWAY && 695 !(rt->rt6i_flags & RTF_EXPIRES) && 696 !(iter->rt6i_flags & RTF_EXPIRES)) 697 rt->rt6i_nsiblings++; 698 } 699 700 if (iter->rt6i_metric > rt->rt6i_metric) 701 break; 702 703 ins = &iter->dst.rt6_next; 704 } 705 706 /* Reset round-robin state, if necessary */ 707 if (ins == &fn->leaf) 708 fn->rr_ptr = NULL; 709 710 /* Link this route to others same route. */ 711 if (rt->rt6i_nsiblings) { 712 unsigned int rt6i_nsiblings; 713 struct rt6_info *sibling, *temp_sibling; 714 715 /* Find the first route that have the same metric */ 716 sibling = fn->leaf; 717 while (sibling) { 718 if (sibling->rt6i_metric == rt->rt6i_metric) { 719 list_add_tail(&rt->rt6i_siblings, 720 &sibling->rt6i_siblings); 721 break; 722 } 723 sibling = sibling->dst.rt6_next; 724 } 725 /* For each sibling in the list, increment the counter of 726 * siblings. BUG() if counters does not match, list of siblings 727 * is broken! 728 */ 729 rt6i_nsiblings = 0; 730 list_for_each_entry_safe(sibling, temp_sibling, 731 &rt->rt6i_siblings, rt6i_siblings) { 732 sibling->rt6i_nsiblings++; 733 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings); 734 rt6i_nsiblings++; 735 } 736 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings); 737 } 738 739 /* 740 * insert node 741 */ 742 if (!replace) { 743 if (!add) 744 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 745 746 add: 747 rt->dst.rt6_next = iter; 748 *ins = rt; 749 rt->rt6i_node = fn; 750 atomic_inc(&rt->rt6i_ref); 751 inet6_rt_notify(RTM_NEWROUTE, rt, info); 752 info->nl_net->ipv6.rt6_stats->fib_rt_entries++; 753 754 if (!(fn->fn_flags & RTN_RTINFO)) { 755 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 756 fn->fn_flags |= RTN_RTINFO; 757 } 758 759 } else { 760 if (!found) { 761 if (add) 762 goto add; 763 pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); 764 return -ENOENT; 765 } 766 *ins = rt; 767 rt->rt6i_node = fn; 768 rt->dst.rt6_next = iter->dst.rt6_next; 769 atomic_inc(&rt->rt6i_ref); 770 inet6_rt_notify(RTM_NEWROUTE, rt, info); 771 rt6_release(iter); 772 if (!(fn->fn_flags & RTN_RTINFO)) { 773 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 774 fn->fn_flags |= RTN_RTINFO; 775 } 776 } 777 778 return 0; 779 } 780 781 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt) 782 { 783 if (!timer_pending(&net->ipv6.ip6_fib_timer) && 784 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE))) 785 mod_timer(&net->ipv6.ip6_fib_timer, 786 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 787 } 788 789 void fib6_force_start_gc(struct net *net) 790 { 791 if (!timer_pending(&net->ipv6.ip6_fib_timer)) 792 mod_timer(&net->ipv6.ip6_fib_timer, 793 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 794 } 795 796 /* 797 * Add routing information to the routing tree. 798 * <destination addr>/<source addr> 799 * with source addr info in sub-trees 800 */ 801 802 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info) 803 { 804 struct fib6_node *fn, *pn = NULL; 805 int err = -ENOMEM; 806 int allow_create = 1; 807 int replace_required = 0; 808 809 if (info->nlh) { 810 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) 811 allow_create = 0; 812 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) 813 replace_required = 1; 814 } 815 if (!allow_create && !replace_required) 816 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); 817 818 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr), 819 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst), 820 allow_create, replace_required); 821 822 if (IS_ERR(fn)) { 823 err = PTR_ERR(fn); 824 goto out; 825 } 826 827 pn = fn; 828 829 #ifdef CONFIG_IPV6_SUBTREES 830 if (rt->rt6i_src.plen) { 831 struct fib6_node *sn; 832 833 if (!fn->subtree) { 834 struct fib6_node *sfn; 835 836 /* 837 * Create subtree. 838 * 839 * fn[main tree] 840 * | 841 * sfn[subtree root] 842 * \ 843 * sn[new leaf node] 844 */ 845 846 /* Create subtree root node */ 847 sfn = node_alloc(); 848 if (!sfn) 849 goto st_failure; 850 851 sfn->leaf = info->nl_net->ipv6.ip6_null_entry; 852 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref); 853 sfn->fn_flags = RTN_ROOT; 854 sfn->fn_sernum = fib6_new_sernum(); 855 856 /* Now add the first leaf node to new subtree */ 857 858 sn = fib6_add_1(sfn, &rt->rt6i_src.addr, 859 sizeof(struct in6_addr), rt->rt6i_src.plen, 860 offsetof(struct rt6_info, rt6i_src), 861 allow_create, replace_required); 862 863 if (IS_ERR(sn)) { 864 /* If it is failed, discard just allocated 865 root, and then (in st_failure) stale node 866 in main tree. 867 */ 868 node_free(sfn); 869 err = PTR_ERR(sn); 870 goto st_failure; 871 } 872 873 /* Now link new subtree to main tree */ 874 sfn->parent = fn; 875 fn->subtree = sfn; 876 } else { 877 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, 878 sizeof(struct in6_addr), rt->rt6i_src.plen, 879 offsetof(struct rt6_info, rt6i_src), 880 allow_create, replace_required); 881 882 if (IS_ERR(sn)) { 883 err = PTR_ERR(sn); 884 goto st_failure; 885 } 886 } 887 888 if (!fn->leaf) { 889 fn->leaf = rt; 890 atomic_inc(&rt->rt6i_ref); 891 } 892 fn = sn; 893 } 894 #endif 895 896 err = fib6_add_rt2node(fn, rt, info); 897 if (!err) { 898 fib6_start_gc(info->nl_net, rt); 899 if (!(rt->rt6i_flags & RTF_CACHE)) 900 fib6_prune_clones(info->nl_net, pn, rt); 901 } 902 903 out: 904 if (err) { 905 #ifdef CONFIG_IPV6_SUBTREES 906 /* 907 * If fib6_add_1 has cleared the old leaf pointer in the 908 * super-tree leaf node we have to find a new one for it. 909 */ 910 if (pn != fn && pn->leaf == rt) { 911 pn->leaf = NULL; 912 atomic_dec(&rt->rt6i_ref); 913 } 914 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) { 915 pn->leaf = fib6_find_prefix(info->nl_net, pn); 916 #if RT6_DEBUG >= 2 917 if (!pn->leaf) { 918 WARN_ON(pn->leaf == NULL); 919 pn->leaf = info->nl_net->ipv6.ip6_null_entry; 920 } 921 #endif 922 atomic_inc(&pn->leaf->rt6i_ref); 923 } 924 #endif 925 dst_free(&rt->dst); 926 } 927 return err; 928 929 #ifdef CONFIG_IPV6_SUBTREES 930 /* Subtree creation failed, probably main tree node 931 is orphan. If it is, shoot it. 932 */ 933 st_failure: 934 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) 935 fib6_repair_tree(info->nl_net, fn); 936 dst_free(&rt->dst); 937 return err; 938 #endif 939 } 940 941 /* 942 * Routing tree lookup 943 * 944 */ 945 946 struct lookup_args { 947 int offset; /* key offset on rt6_info */ 948 const struct in6_addr *addr; /* search key */ 949 }; 950 951 static struct fib6_node * fib6_lookup_1(struct fib6_node *root, 952 struct lookup_args *args) 953 { 954 struct fib6_node *fn; 955 __be32 dir; 956 957 if (unlikely(args->offset == 0)) 958 return NULL; 959 960 /* 961 * Descend on a tree 962 */ 963 964 fn = root; 965 966 for (;;) { 967 struct fib6_node *next; 968 969 dir = addr_bit_set(args->addr, fn->fn_bit); 970 971 next = dir ? fn->right : fn->left; 972 973 if (next) { 974 fn = next; 975 continue; 976 } 977 break; 978 } 979 980 while (fn) { 981 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) { 982 struct rt6key *key; 983 984 key = (struct rt6key *) ((u8 *) fn->leaf + 985 args->offset); 986 987 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 988 #ifdef CONFIG_IPV6_SUBTREES 989 if (fn->subtree) 990 fn = fib6_lookup_1(fn->subtree, args + 1); 991 #endif 992 if (!fn || fn->fn_flags & RTN_RTINFO) 993 return fn; 994 } 995 } 996 997 if (fn->fn_flags & RTN_ROOT) 998 break; 999 1000 fn = fn->parent; 1001 } 1002 1003 return NULL; 1004 } 1005 1006 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr, 1007 const struct in6_addr *saddr) 1008 { 1009 struct fib6_node *fn; 1010 struct lookup_args args[] = { 1011 { 1012 .offset = offsetof(struct rt6_info, rt6i_dst), 1013 .addr = daddr, 1014 }, 1015 #ifdef CONFIG_IPV6_SUBTREES 1016 { 1017 .offset = offsetof(struct rt6_info, rt6i_src), 1018 .addr = saddr, 1019 }, 1020 #endif 1021 { 1022 .offset = 0, /* sentinel */ 1023 } 1024 }; 1025 1026 fn = fib6_lookup_1(root, daddr ? args : args + 1); 1027 if (!fn || fn->fn_flags & RTN_TL_ROOT) 1028 fn = root; 1029 1030 return fn; 1031 } 1032 1033 /* 1034 * Get node with specified destination prefix (and source prefix, 1035 * if subtrees are used) 1036 */ 1037 1038 1039 static struct fib6_node * fib6_locate_1(struct fib6_node *root, 1040 const struct in6_addr *addr, 1041 int plen, int offset) 1042 { 1043 struct fib6_node *fn; 1044 1045 for (fn = root; fn ; ) { 1046 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); 1047 1048 /* 1049 * Prefix match 1050 */ 1051 if (plen < fn->fn_bit || 1052 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 1053 return NULL; 1054 1055 if (plen == fn->fn_bit) 1056 return fn; 1057 1058 /* 1059 * We have more bits to go 1060 */ 1061 if (addr_bit_set(addr, fn->fn_bit)) 1062 fn = fn->right; 1063 else 1064 fn = fn->left; 1065 } 1066 return NULL; 1067 } 1068 1069 struct fib6_node * fib6_locate(struct fib6_node *root, 1070 const struct in6_addr *daddr, int dst_len, 1071 const struct in6_addr *saddr, int src_len) 1072 { 1073 struct fib6_node *fn; 1074 1075 fn = fib6_locate_1(root, daddr, dst_len, 1076 offsetof(struct rt6_info, rt6i_dst)); 1077 1078 #ifdef CONFIG_IPV6_SUBTREES 1079 if (src_len) { 1080 WARN_ON(saddr == NULL); 1081 if (fn && fn->subtree) 1082 fn = fib6_locate_1(fn->subtree, saddr, src_len, 1083 offsetof(struct rt6_info, rt6i_src)); 1084 } 1085 #endif 1086 1087 if (fn && fn->fn_flags & RTN_RTINFO) 1088 return fn; 1089 1090 return NULL; 1091 } 1092 1093 1094 /* 1095 * Deletion 1096 * 1097 */ 1098 1099 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn) 1100 { 1101 if (fn->fn_flags & RTN_ROOT) 1102 return net->ipv6.ip6_null_entry; 1103 1104 while (fn) { 1105 if (fn->left) 1106 return fn->left->leaf; 1107 if (fn->right) 1108 return fn->right->leaf; 1109 1110 fn = FIB6_SUBTREE(fn); 1111 } 1112 return NULL; 1113 } 1114 1115 /* 1116 * Called to trim the tree of intermediate nodes when possible. "fn" 1117 * is the node we want to try and remove. 1118 */ 1119 1120 static struct fib6_node *fib6_repair_tree(struct net *net, 1121 struct fib6_node *fn) 1122 { 1123 int children; 1124 int nstate; 1125 struct fib6_node *child, *pn; 1126 struct fib6_walker_t *w; 1127 int iter = 0; 1128 1129 for (;;) { 1130 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1131 iter++; 1132 1133 WARN_ON(fn->fn_flags & RTN_RTINFO); 1134 WARN_ON(fn->fn_flags & RTN_TL_ROOT); 1135 WARN_ON(fn->leaf != NULL); 1136 1137 children = 0; 1138 child = NULL; 1139 if (fn->right) child = fn->right, children |= 1; 1140 if (fn->left) child = fn->left, children |= 2; 1141 1142 if (children == 3 || FIB6_SUBTREE(fn) 1143 #ifdef CONFIG_IPV6_SUBTREES 1144 /* Subtree root (i.e. fn) may have one child */ 1145 || (children && fn->fn_flags & RTN_ROOT) 1146 #endif 1147 ) { 1148 fn->leaf = fib6_find_prefix(net, fn); 1149 #if RT6_DEBUG >= 2 1150 if (!fn->leaf) { 1151 WARN_ON(!fn->leaf); 1152 fn->leaf = net->ipv6.ip6_null_entry; 1153 } 1154 #endif 1155 atomic_inc(&fn->leaf->rt6i_ref); 1156 return fn->parent; 1157 } 1158 1159 pn = fn->parent; 1160 #ifdef CONFIG_IPV6_SUBTREES 1161 if (FIB6_SUBTREE(pn) == fn) { 1162 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1163 FIB6_SUBTREE(pn) = NULL; 1164 nstate = FWS_L; 1165 } else { 1166 WARN_ON(fn->fn_flags & RTN_ROOT); 1167 #endif 1168 if (pn->right == fn) pn->right = child; 1169 else if (pn->left == fn) pn->left = child; 1170 #if RT6_DEBUG >= 2 1171 else 1172 WARN_ON(1); 1173 #endif 1174 if (child) 1175 child->parent = pn; 1176 nstate = FWS_R; 1177 #ifdef CONFIG_IPV6_SUBTREES 1178 } 1179 #endif 1180 1181 read_lock(&fib6_walker_lock); 1182 FOR_WALKERS(w) { 1183 if (!child) { 1184 if (w->root == fn) { 1185 w->root = w->node = NULL; 1186 RT6_TRACE("W %p adjusted by delroot 1\n", w); 1187 } else if (w->node == fn) { 1188 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1189 w->node = pn; 1190 w->state = nstate; 1191 } 1192 } else { 1193 if (w->root == fn) { 1194 w->root = child; 1195 RT6_TRACE("W %p adjusted by delroot 2\n", w); 1196 } 1197 if (w->node == fn) { 1198 w->node = child; 1199 if (children&2) { 1200 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1201 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT; 1202 } else { 1203 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1204 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT; 1205 } 1206 } 1207 } 1208 } 1209 read_unlock(&fib6_walker_lock); 1210 1211 node_free(fn); 1212 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) 1213 return pn; 1214 1215 rt6_release(pn->leaf); 1216 pn->leaf = NULL; 1217 fn = pn; 1218 } 1219 } 1220 1221 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, 1222 struct nl_info *info) 1223 { 1224 struct fib6_walker_t *w; 1225 struct rt6_info *rt = *rtp; 1226 struct net *net = info->nl_net; 1227 1228 RT6_TRACE("fib6_del_route\n"); 1229 1230 /* Unlink it */ 1231 *rtp = rt->dst.rt6_next; 1232 rt->rt6i_node = NULL; 1233 net->ipv6.rt6_stats->fib_rt_entries--; 1234 net->ipv6.rt6_stats->fib_discarded_routes++; 1235 1236 /* Reset round-robin state, if necessary */ 1237 if (fn->rr_ptr == rt) 1238 fn->rr_ptr = NULL; 1239 1240 /* Remove this entry from other siblings */ 1241 if (rt->rt6i_nsiblings) { 1242 struct rt6_info *sibling, *next_sibling; 1243 1244 list_for_each_entry_safe(sibling, next_sibling, 1245 &rt->rt6i_siblings, rt6i_siblings) 1246 sibling->rt6i_nsiblings--; 1247 rt->rt6i_nsiblings = 0; 1248 list_del_init(&rt->rt6i_siblings); 1249 } 1250 1251 /* Adjust walkers */ 1252 read_lock(&fib6_walker_lock); 1253 FOR_WALKERS(w) { 1254 if (w->state == FWS_C && w->leaf == rt) { 1255 RT6_TRACE("walker %p adjusted by delroute\n", w); 1256 w->leaf = rt->dst.rt6_next; 1257 if (!w->leaf) 1258 w->state = FWS_U; 1259 } 1260 } 1261 read_unlock(&fib6_walker_lock); 1262 1263 rt->dst.rt6_next = NULL; 1264 1265 /* If it was last route, expunge its radix tree node */ 1266 if (!fn->leaf) { 1267 fn->fn_flags &= ~RTN_RTINFO; 1268 net->ipv6.rt6_stats->fib_route_nodes--; 1269 fn = fib6_repair_tree(net, fn); 1270 } 1271 1272 if (atomic_read(&rt->rt6i_ref) != 1) { 1273 /* This route is used as dummy address holder in some split 1274 * nodes. It is not leaked, but it still holds other resources, 1275 * which must be released in time. So, scan ascendant nodes 1276 * and replace dummy references to this route with references 1277 * to still alive ones. 1278 */ 1279 while (fn) { 1280 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) { 1281 fn->leaf = fib6_find_prefix(net, fn); 1282 atomic_inc(&fn->leaf->rt6i_ref); 1283 rt6_release(rt); 1284 } 1285 fn = fn->parent; 1286 } 1287 /* No more references are possible at this point. */ 1288 BUG_ON(atomic_read(&rt->rt6i_ref) != 1); 1289 } 1290 1291 inet6_rt_notify(RTM_DELROUTE, rt, info); 1292 rt6_release(rt); 1293 } 1294 1295 int fib6_del(struct rt6_info *rt, struct nl_info *info) 1296 { 1297 struct net *net = info->nl_net; 1298 struct fib6_node *fn = rt->rt6i_node; 1299 struct rt6_info **rtp; 1300 1301 #if RT6_DEBUG >= 2 1302 if (rt->dst.obsolete>0) { 1303 WARN_ON(fn != NULL); 1304 return -ENOENT; 1305 } 1306 #endif 1307 if (!fn || rt == net->ipv6.ip6_null_entry) 1308 return -ENOENT; 1309 1310 WARN_ON(!(fn->fn_flags & RTN_RTINFO)); 1311 1312 if (!(rt->rt6i_flags & RTF_CACHE)) { 1313 struct fib6_node *pn = fn; 1314 #ifdef CONFIG_IPV6_SUBTREES 1315 /* clones of this route might be in another subtree */ 1316 if (rt->rt6i_src.plen) { 1317 while (!(pn->fn_flags & RTN_ROOT)) 1318 pn = pn->parent; 1319 pn = pn->parent; 1320 } 1321 #endif 1322 fib6_prune_clones(info->nl_net, pn, rt); 1323 } 1324 1325 /* 1326 * Walk the leaf entries looking for ourself 1327 */ 1328 1329 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) { 1330 if (*rtp == rt) { 1331 fib6_del_route(fn, rtp, info); 1332 return 0; 1333 } 1334 } 1335 return -ENOENT; 1336 } 1337 1338 /* 1339 * Tree traversal function. 1340 * 1341 * Certainly, it is not interrupt safe. 1342 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 1343 * It means, that we can modify tree during walking 1344 * and use this function for garbage collection, clone pruning, 1345 * cleaning tree when a device goes down etc. etc. 1346 * 1347 * It guarantees that every node will be traversed, 1348 * and that it will be traversed only once. 1349 * 1350 * Callback function w->func may return: 1351 * 0 -> continue walking. 1352 * positive value -> walking is suspended (used by tree dumps, 1353 * and probably by gc, if it will be split to several slices) 1354 * negative value -> terminate walking. 1355 * 1356 * The function itself returns: 1357 * 0 -> walk is complete. 1358 * >0 -> walk is incomplete (i.e. suspended) 1359 * <0 -> walk is terminated by an error. 1360 */ 1361 1362 static int fib6_walk_continue(struct fib6_walker_t *w) 1363 { 1364 struct fib6_node *fn, *pn; 1365 1366 for (;;) { 1367 fn = w->node; 1368 if (!fn) 1369 return 0; 1370 1371 if (w->prune && fn != w->root && 1372 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) { 1373 w->state = FWS_C; 1374 w->leaf = fn->leaf; 1375 } 1376 switch (w->state) { 1377 #ifdef CONFIG_IPV6_SUBTREES 1378 case FWS_S: 1379 if (FIB6_SUBTREE(fn)) { 1380 w->node = FIB6_SUBTREE(fn); 1381 continue; 1382 } 1383 w->state = FWS_L; 1384 #endif 1385 case FWS_L: 1386 if (fn->left) { 1387 w->node = fn->left; 1388 w->state = FWS_INIT; 1389 continue; 1390 } 1391 w->state = FWS_R; 1392 case FWS_R: 1393 if (fn->right) { 1394 w->node = fn->right; 1395 w->state = FWS_INIT; 1396 continue; 1397 } 1398 w->state = FWS_C; 1399 w->leaf = fn->leaf; 1400 case FWS_C: 1401 if (w->leaf && fn->fn_flags & RTN_RTINFO) { 1402 int err; 1403 1404 if (w->skip) { 1405 w->skip--; 1406 continue; 1407 } 1408 1409 err = w->func(w); 1410 if (err) 1411 return err; 1412 1413 w->count++; 1414 continue; 1415 } 1416 w->state = FWS_U; 1417 case FWS_U: 1418 if (fn == w->root) 1419 return 0; 1420 pn = fn->parent; 1421 w->node = pn; 1422 #ifdef CONFIG_IPV6_SUBTREES 1423 if (FIB6_SUBTREE(pn) == fn) { 1424 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1425 w->state = FWS_L; 1426 continue; 1427 } 1428 #endif 1429 if (pn->left == fn) { 1430 w->state = FWS_R; 1431 continue; 1432 } 1433 if (pn->right == fn) { 1434 w->state = FWS_C; 1435 w->leaf = w->node->leaf; 1436 continue; 1437 } 1438 #if RT6_DEBUG >= 2 1439 WARN_ON(1); 1440 #endif 1441 } 1442 } 1443 } 1444 1445 static int fib6_walk(struct fib6_walker_t *w) 1446 { 1447 int res; 1448 1449 w->state = FWS_INIT; 1450 w->node = w->root; 1451 1452 fib6_walker_link(w); 1453 res = fib6_walk_continue(w); 1454 if (res <= 0) 1455 fib6_walker_unlink(w); 1456 return res; 1457 } 1458 1459 static int fib6_clean_node(struct fib6_walker_t *w) 1460 { 1461 int res; 1462 struct rt6_info *rt; 1463 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w); 1464 struct nl_info info = { 1465 .nl_net = c->net, 1466 }; 1467 1468 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 1469 res = c->func(rt, c->arg); 1470 if (res < 0) { 1471 w->leaf = rt; 1472 res = fib6_del(rt, &info); 1473 if (res) { 1474 #if RT6_DEBUG >= 2 1475 pr_debug("%s: del failed: rt=%p@%p err=%d\n", 1476 __func__, rt, rt->rt6i_node, res); 1477 #endif 1478 continue; 1479 } 1480 return 0; 1481 } 1482 WARN_ON(res != 0); 1483 } 1484 w->leaf = rt; 1485 return 0; 1486 } 1487 1488 /* 1489 * Convenient frontend to tree walker. 1490 * 1491 * func is called on each route. 1492 * It may return -1 -> delete this route. 1493 * 0 -> continue walking 1494 * 1495 * prune==1 -> only immediate children of node (certainly, 1496 * ignoring pure split nodes) will be scanned. 1497 */ 1498 1499 static void fib6_clean_tree(struct net *net, struct fib6_node *root, 1500 int (*func)(struct rt6_info *, void *arg), 1501 int prune, void *arg) 1502 { 1503 struct fib6_cleaner_t c; 1504 1505 c.w.root = root; 1506 c.w.func = fib6_clean_node; 1507 c.w.prune = prune; 1508 c.w.count = 0; 1509 c.w.skip = 0; 1510 c.func = func; 1511 c.arg = arg; 1512 c.net = net; 1513 1514 fib6_walk(&c.w); 1515 } 1516 1517 void fib6_clean_all_ro(struct net *net, int (*func)(struct rt6_info *, void *arg), 1518 int prune, void *arg) 1519 { 1520 struct fib6_table *table; 1521 struct hlist_head *head; 1522 unsigned int h; 1523 1524 rcu_read_lock(); 1525 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 1526 head = &net->ipv6.fib_table_hash[h]; 1527 hlist_for_each_entry_rcu(table, head, tb6_hlist) { 1528 read_lock_bh(&table->tb6_lock); 1529 fib6_clean_tree(net, &table->tb6_root, 1530 func, prune, arg); 1531 read_unlock_bh(&table->tb6_lock); 1532 } 1533 } 1534 rcu_read_unlock(); 1535 } 1536 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg), 1537 int prune, void *arg) 1538 { 1539 struct fib6_table *table; 1540 struct hlist_head *head; 1541 unsigned int h; 1542 1543 rcu_read_lock(); 1544 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 1545 head = &net->ipv6.fib_table_hash[h]; 1546 hlist_for_each_entry_rcu(table, head, tb6_hlist) { 1547 write_lock_bh(&table->tb6_lock); 1548 fib6_clean_tree(net, &table->tb6_root, 1549 func, prune, arg); 1550 write_unlock_bh(&table->tb6_lock); 1551 } 1552 } 1553 rcu_read_unlock(); 1554 } 1555 1556 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1557 { 1558 if (rt->rt6i_flags & RTF_CACHE) { 1559 RT6_TRACE("pruning clone %p\n", rt); 1560 return -1; 1561 } 1562 1563 return 0; 1564 } 1565 1566 static void fib6_prune_clones(struct net *net, struct fib6_node *fn, 1567 struct rt6_info *rt) 1568 { 1569 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt); 1570 } 1571 1572 /* 1573 * Garbage collection 1574 */ 1575 1576 static struct fib6_gc_args 1577 { 1578 int timeout; 1579 int more; 1580 } gc_args; 1581 1582 static int fib6_age(struct rt6_info *rt, void *arg) 1583 { 1584 unsigned long now = jiffies; 1585 1586 /* 1587 * check addrconf expiration here. 1588 * Routes are expired even if they are in use. 1589 * 1590 * Also age clones. Note, that clones are aged out 1591 * only if they are not in use now. 1592 */ 1593 1594 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) { 1595 if (time_after(now, rt->dst.expires)) { 1596 RT6_TRACE("expiring %p\n", rt); 1597 return -1; 1598 } 1599 gc_args.more++; 1600 } else if (rt->rt6i_flags & RTF_CACHE) { 1601 if (atomic_read(&rt->dst.__refcnt) == 0 && 1602 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) { 1603 RT6_TRACE("aging clone %p\n", rt); 1604 return -1; 1605 } else if (rt->rt6i_flags & RTF_GATEWAY) { 1606 struct neighbour *neigh; 1607 __u8 neigh_flags = 0; 1608 1609 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway); 1610 if (neigh) { 1611 neigh_flags = neigh->flags; 1612 neigh_release(neigh); 1613 } 1614 if (!(neigh_flags & NTF_ROUTER)) { 1615 RT6_TRACE("purging route %p via non-router but gateway\n", 1616 rt); 1617 return -1; 1618 } 1619 } 1620 gc_args.more++; 1621 } 1622 1623 return 0; 1624 } 1625 1626 static DEFINE_SPINLOCK(fib6_gc_lock); 1627 1628 void fib6_run_gc(unsigned long expires, struct net *net) 1629 { 1630 if (expires != ~0UL) { 1631 spin_lock_bh(&fib6_gc_lock); 1632 gc_args.timeout = expires ? (int)expires : 1633 net->ipv6.sysctl.ip6_rt_gc_interval; 1634 } else { 1635 if (!spin_trylock_bh(&fib6_gc_lock)) { 1636 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); 1637 return; 1638 } 1639 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval; 1640 } 1641 1642 gc_args.more = icmp6_dst_gc(); 1643 1644 fib6_clean_all(net, fib6_age, 0, NULL); 1645 1646 if (gc_args.more) 1647 mod_timer(&net->ipv6.ip6_fib_timer, 1648 round_jiffies(jiffies 1649 + net->ipv6.sysctl.ip6_rt_gc_interval)); 1650 else 1651 del_timer(&net->ipv6.ip6_fib_timer); 1652 spin_unlock_bh(&fib6_gc_lock); 1653 } 1654 1655 static void fib6_gc_timer_cb(unsigned long arg) 1656 { 1657 fib6_run_gc(0, (struct net *)arg); 1658 } 1659 1660 static int __net_init fib6_net_init(struct net *net) 1661 { 1662 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; 1663 1664 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net); 1665 1666 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); 1667 if (!net->ipv6.rt6_stats) 1668 goto out_timer; 1669 1670 /* Avoid false sharing : Use at least a full cache line */ 1671 size = max_t(size_t, size, L1_CACHE_BYTES); 1672 1673 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); 1674 if (!net->ipv6.fib_table_hash) 1675 goto out_rt6_stats; 1676 1677 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), 1678 GFP_KERNEL); 1679 if (!net->ipv6.fib6_main_tbl) 1680 goto out_fib_table_hash; 1681 1682 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; 1683 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1684 net->ipv6.fib6_main_tbl->tb6_root.fn_flags = 1685 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1686 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); 1687 1688 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1689 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), 1690 GFP_KERNEL); 1691 if (!net->ipv6.fib6_local_tbl) 1692 goto out_fib6_main_tbl; 1693 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; 1694 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1695 net->ipv6.fib6_local_tbl->tb6_root.fn_flags = 1696 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1697 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); 1698 #endif 1699 fib6_tables_init(net); 1700 1701 return 0; 1702 1703 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1704 out_fib6_main_tbl: 1705 kfree(net->ipv6.fib6_main_tbl); 1706 #endif 1707 out_fib_table_hash: 1708 kfree(net->ipv6.fib_table_hash); 1709 out_rt6_stats: 1710 kfree(net->ipv6.rt6_stats); 1711 out_timer: 1712 return -ENOMEM; 1713 } 1714 1715 static void fib6_net_exit(struct net *net) 1716 { 1717 rt6_ifdown(net, NULL); 1718 del_timer_sync(&net->ipv6.ip6_fib_timer); 1719 1720 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1721 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers); 1722 kfree(net->ipv6.fib6_local_tbl); 1723 #endif 1724 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers); 1725 kfree(net->ipv6.fib6_main_tbl); 1726 kfree(net->ipv6.fib_table_hash); 1727 kfree(net->ipv6.rt6_stats); 1728 } 1729 1730 static struct pernet_operations fib6_net_ops = { 1731 .init = fib6_net_init, 1732 .exit = fib6_net_exit, 1733 }; 1734 1735 int __init fib6_init(void) 1736 { 1737 int ret = -ENOMEM; 1738 1739 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1740 sizeof(struct fib6_node), 1741 0, SLAB_HWCACHE_ALIGN, 1742 NULL); 1743 if (!fib6_node_kmem) 1744 goto out; 1745 1746 ret = register_pernet_subsys(&fib6_net_ops); 1747 if (ret) 1748 goto out_kmem_cache_create; 1749 1750 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib, 1751 NULL); 1752 if (ret) 1753 goto out_unregister_subsys; 1754 out: 1755 return ret; 1756 1757 out_unregister_subsys: 1758 unregister_pernet_subsys(&fib6_net_ops); 1759 out_kmem_cache_create: 1760 kmem_cache_destroy(fib6_node_kmem); 1761 goto out; 1762 } 1763 1764 void fib6_gc_cleanup(void) 1765 { 1766 unregister_pernet_subsys(&fib6_net_ops); 1767 kmem_cache_destroy(fib6_node_kmem); 1768 } 1769