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 * Changes: 14 * Yuji SEKIYA @USAGI: Support default route on router node; 15 * remove ip6_null_entry from the top of 16 * routing table. 17 * Ville Nuorvala: Fixed routing subtrees. 18 */ 19 20 #define pr_fmt(fmt) "IPv6: " fmt 21 22 #include <linux/errno.h> 23 #include <linux/types.h> 24 #include <linux/net.h> 25 #include <linux/route.h> 26 #include <linux/netdevice.h> 27 #include <linux/in6.h> 28 #include <linux/init.h> 29 #include <linux/list.h> 30 #include <linux/slab.h> 31 32 #include <net/ipv6.h> 33 #include <net/ndisc.h> 34 #include <net/addrconf.h> 35 36 #include <net/ip6_fib.h> 37 #include <net/ip6_route.h> 38 39 #define RT6_DEBUG 2 40 41 #if RT6_DEBUG >= 3 42 #define RT6_TRACE(x...) pr_debug(x) 43 #else 44 #define RT6_TRACE(x...) do { ; } while (0) 45 #endif 46 47 static struct kmem_cache *fib6_node_kmem __read_mostly; 48 49 enum fib_walk_state_t { 50 #ifdef CONFIG_IPV6_SUBTREES 51 FWS_S, 52 #endif 53 FWS_L, 54 FWS_R, 55 FWS_C, 56 FWS_U 57 }; 58 59 struct fib6_cleaner_t { 60 struct fib6_walker_t w; 61 struct net *net; 62 int (*func)(struct rt6_info *, void *arg); 63 void *arg; 64 }; 65 66 static DEFINE_RWLOCK(fib6_walker_lock); 67 68 #ifdef CONFIG_IPV6_SUBTREES 69 #define FWS_INIT FWS_S 70 #else 71 #define FWS_INIT FWS_L 72 #endif 73 74 static void fib6_prune_clones(struct net *net, struct fib6_node *fn); 75 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn); 76 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn); 77 static int fib6_walk(struct fib6_walker_t *w); 78 static int fib6_walk_continue(struct fib6_walker_t *w); 79 80 /* 81 * A routing update causes an increase of the serial number on the 82 * affected subtree. This allows for cached routes to be asynchronously 83 * tested when modifications are made to the destination cache as a 84 * result of redirects, path MTU changes, etc. 85 */ 86 87 static __u32 rt_sernum; 88 89 static void fib6_gc_timer_cb(unsigned long arg); 90 91 static LIST_HEAD(fib6_walkers); 92 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh) 93 94 static inline void fib6_walker_link(struct fib6_walker_t *w) 95 { 96 write_lock_bh(&fib6_walker_lock); 97 list_add(&w->lh, &fib6_walkers); 98 write_unlock_bh(&fib6_walker_lock); 99 } 100 101 static inline void fib6_walker_unlink(struct fib6_walker_t *w) 102 { 103 write_lock_bh(&fib6_walker_lock); 104 list_del(&w->lh); 105 write_unlock_bh(&fib6_walker_lock); 106 } 107 static __inline__ u32 fib6_new_sernum(void) 108 { 109 u32 n = ++rt_sernum; 110 if ((__s32)n <= 0) 111 rt_sernum = n = 1; 112 return n; 113 } 114 115 /* 116 * Auxiliary address test functions for the radix tree. 117 * 118 * These assume a 32bit processor (although it will work on 119 * 64bit processors) 120 */ 121 122 /* 123 * test bit 124 */ 125 #if defined(__LITTLE_ENDIAN) 126 # define BITOP_BE32_SWIZZLE (0x1F & ~7) 127 #else 128 # define BITOP_BE32_SWIZZLE 0 129 #endif 130 131 static __inline__ __be32 addr_bit_set(const void *token, int fn_bit) 132 { 133 const __be32 *addr = token; 134 /* 135 * Here, 136 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f) 137 * is optimized version of 138 * htonl(1 << ((~fn_bit)&0x1F)) 139 * See include/asm-generic/bitops/le.h. 140 */ 141 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) & 142 addr[fn_bit >> 5]; 143 } 144 145 static __inline__ struct fib6_node *node_alloc(void) 146 { 147 struct fib6_node *fn; 148 149 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC); 150 151 return fn; 152 } 153 154 static __inline__ void node_free(struct fib6_node *fn) 155 { 156 kmem_cache_free(fib6_node_kmem, fn); 157 } 158 159 static __inline__ void rt6_release(struct rt6_info *rt) 160 { 161 if (atomic_dec_and_test(&rt->rt6i_ref)) 162 dst_free(&rt->dst); 163 } 164 165 static void fib6_link_table(struct net *net, struct fib6_table *tb) 166 { 167 unsigned int h; 168 169 /* 170 * Initialize table lock at a single place to give lockdep a key, 171 * tables aren't visible prior to being linked to the list. 172 */ 173 rwlock_init(&tb->tb6_lock); 174 175 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1); 176 177 /* 178 * No protection necessary, this is the only list mutatation 179 * operation, tables never disappear once they exist. 180 */ 181 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]); 182 } 183 184 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 185 186 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id) 187 { 188 struct fib6_table *table; 189 190 table = kzalloc(sizeof(*table), GFP_ATOMIC); 191 if (table) { 192 table->tb6_id = id; 193 table->tb6_root.leaf = net->ipv6.ip6_null_entry; 194 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 195 inet_peer_base_init(&table->tb6_peers); 196 } 197 198 return table; 199 } 200 201 struct fib6_table *fib6_new_table(struct net *net, u32 id) 202 { 203 struct fib6_table *tb; 204 205 if (id == 0) 206 id = RT6_TABLE_MAIN; 207 tb = fib6_get_table(net, id); 208 if (tb) 209 return tb; 210 211 tb = fib6_alloc_table(net, id); 212 if (tb) 213 fib6_link_table(net, tb); 214 215 return tb; 216 } 217 218 struct fib6_table *fib6_get_table(struct net *net, u32 id) 219 { 220 struct fib6_table *tb; 221 struct hlist_head *head; 222 unsigned int h; 223 224 if (id == 0) 225 id = RT6_TABLE_MAIN; 226 h = id & (FIB6_TABLE_HASHSZ - 1); 227 rcu_read_lock(); 228 head = &net->ipv6.fib_table_hash[h]; 229 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 230 if (tb->tb6_id == id) { 231 rcu_read_unlock(); 232 return tb; 233 } 234 } 235 rcu_read_unlock(); 236 237 return NULL; 238 } 239 240 static void __net_init fib6_tables_init(struct net *net) 241 { 242 fib6_link_table(net, net->ipv6.fib6_main_tbl); 243 fib6_link_table(net, net->ipv6.fib6_local_tbl); 244 } 245 #else 246 247 struct fib6_table *fib6_new_table(struct net *net, u32 id) 248 { 249 return fib6_get_table(net, id); 250 } 251 252 struct fib6_table *fib6_get_table(struct net *net, u32 id) 253 { 254 return net->ipv6.fib6_main_tbl; 255 } 256 257 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, 258 int flags, pol_lookup_t lookup) 259 { 260 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags); 261 } 262 263 static void __net_init fib6_tables_init(struct net *net) 264 { 265 fib6_link_table(net, net->ipv6.fib6_main_tbl); 266 } 267 268 #endif 269 270 static int fib6_dump_node(struct fib6_walker_t *w) 271 { 272 int res; 273 struct rt6_info *rt; 274 275 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 276 res = rt6_dump_route(rt, w->args); 277 if (res < 0) { 278 /* Frame is full, suspend walking */ 279 w->leaf = rt; 280 return 1; 281 } 282 WARN_ON(res == 0); 283 } 284 w->leaf = NULL; 285 return 0; 286 } 287 288 static void fib6_dump_end(struct netlink_callback *cb) 289 { 290 struct fib6_walker_t *w = (void *)cb->args[2]; 291 292 if (w) { 293 if (cb->args[4]) { 294 cb->args[4] = 0; 295 fib6_walker_unlink(w); 296 } 297 cb->args[2] = 0; 298 kfree(w); 299 } 300 cb->done = (void *)cb->args[3]; 301 cb->args[1] = 3; 302 } 303 304 static int fib6_dump_done(struct netlink_callback *cb) 305 { 306 fib6_dump_end(cb); 307 return cb->done ? cb->done(cb) : 0; 308 } 309 310 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 311 struct netlink_callback *cb) 312 { 313 struct fib6_walker_t *w; 314 int res; 315 316 w = (void *)cb->args[2]; 317 w->root = &table->tb6_root; 318 319 if (cb->args[4] == 0) { 320 w->count = 0; 321 w->skip = 0; 322 323 read_lock_bh(&table->tb6_lock); 324 res = fib6_walk(w); 325 read_unlock_bh(&table->tb6_lock); 326 if (res > 0) { 327 cb->args[4] = 1; 328 cb->args[5] = w->root->fn_sernum; 329 } 330 } else { 331 if (cb->args[5] != w->root->fn_sernum) { 332 /* Begin at the root if the tree changed */ 333 cb->args[5] = w->root->fn_sernum; 334 w->state = FWS_INIT; 335 w->node = w->root; 336 w->skip = w->count; 337 } else 338 w->skip = 0; 339 340 read_lock_bh(&table->tb6_lock); 341 res = fib6_walk_continue(w); 342 read_unlock_bh(&table->tb6_lock); 343 if (res <= 0) { 344 fib6_walker_unlink(w); 345 cb->args[4] = 0; 346 } 347 } 348 349 return res; 350 } 351 352 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 353 { 354 struct net *net = sock_net(skb->sk); 355 unsigned int h, s_h; 356 unsigned int e = 0, s_e; 357 struct rt6_rtnl_dump_arg arg; 358 struct fib6_walker_t *w; 359 struct fib6_table *tb; 360 struct hlist_head *head; 361 int res = 0; 362 363 s_h = cb->args[0]; 364 s_e = cb->args[1]; 365 366 w = (void *)cb->args[2]; 367 if (!w) { 368 /* New dump: 369 * 370 * 1. hook callback destructor. 371 */ 372 cb->args[3] = (long)cb->done; 373 cb->done = fib6_dump_done; 374 375 /* 376 * 2. allocate and initialize walker. 377 */ 378 w = kzalloc(sizeof(*w), GFP_ATOMIC); 379 if (!w) 380 return -ENOMEM; 381 w->func = fib6_dump_node; 382 cb->args[2] = (long)w; 383 } 384 385 arg.skb = skb; 386 arg.cb = cb; 387 arg.net = net; 388 w->args = &arg; 389 390 rcu_read_lock(); 391 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) { 392 e = 0; 393 head = &net->ipv6.fib_table_hash[h]; 394 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 395 if (e < s_e) 396 goto next; 397 res = fib6_dump_table(tb, skb, cb); 398 if (res != 0) 399 goto out; 400 next: 401 e++; 402 } 403 } 404 out: 405 rcu_read_unlock(); 406 cb->args[1] = e; 407 cb->args[0] = h; 408 409 res = res < 0 ? res : skb->len; 410 if (res <= 0) 411 fib6_dump_end(cb); 412 return res; 413 } 414 415 /* 416 * Routing Table 417 * 418 * return the appropriate node for a routing tree "add" operation 419 * by either creating and inserting or by returning an existing 420 * node. 421 */ 422 423 static struct fib6_node *fib6_add_1(struct fib6_node *root, 424 struct in6_addr *addr, int plen, 425 int offset, int allow_create, 426 int replace_required) 427 { 428 struct fib6_node *fn, *in, *ln; 429 struct fib6_node *pn = NULL; 430 struct rt6key *key; 431 int bit; 432 __be32 dir = 0; 433 __u32 sernum = fib6_new_sernum(); 434 435 RT6_TRACE("fib6_add_1\n"); 436 437 /* insert node in tree */ 438 439 fn = root; 440 441 do { 442 key = (struct rt6key *)((u8 *)fn->leaf + offset); 443 444 /* 445 * Prefix match 446 */ 447 if (plen < fn->fn_bit || 448 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) { 449 if (!allow_create) { 450 if (replace_required) { 451 pr_warn("Can't replace route, no match found\n"); 452 return ERR_PTR(-ENOENT); 453 } 454 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 455 } 456 goto insert_above; 457 } 458 459 /* 460 * Exact match ? 461 */ 462 463 if (plen == fn->fn_bit) { 464 /* clean up an intermediate node */ 465 if (!(fn->fn_flags & RTN_RTINFO)) { 466 rt6_release(fn->leaf); 467 fn->leaf = NULL; 468 } 469 470 fn->fn_sernum = sernum; 471 472 return fn; 473 } 474 475 /* 476 * We have more bits to go 477 */ 478 479 /* Try to walk down on tree. */ 480 fn->fn_sernum = sernum; 481 dir = addr_bit_set(addr, fn->fn_bit); 482 pn = fn; 483 fn = dir ? fn->right : fn->left; 484 } while (fn); 485 486 if (!allow_create) { 487 /* We should not create new node because 488 * NLM_F_REPLACE was specified without NLM_F_CREATE 489 * I assume it is safe to require NLM_F_CREATE when 490 * REPLACE flag is used! Later we may want to remove the 491 * check for replace_required, because according 492 * to netlink specification, NLM_F_CREATE 493 * MUST be specified if new route is created. 494 * That would keep IPv6 consistent with IPv4 495 */ 496 if (replace_required) { 497 pr_warn("Can't replace route, no match found\n"); 498 return ERR_PTR(-ENOENT); 499 } 500 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 501 } 502 /* 503 * We walked to the bottom of tree. 504 * Create new leaf node without children. 505 */ 506 507 ln = node_alloc(); 508 509 if (!ln) 510 return ERR_PTR(-ENOMEM); 511 ln->fn_bit = plen; 512 513 ln->parent = pn; 514 ln->fn_sernum = sernum; 515 516 if (dir) 517 pn->right = ln; 518 else 519 pn->left = ln; 520 521 return ln; 522 523 524 insert_above: 525 /* 526 * split since we don't have a common prefix anymore or 527 * we have a less significant route. 528 * we've to insert an intermediate node on the list 529 * this new node will point to the one we need to create 530 * and the current 531 */ 532 533 pn = fn->parent; 534 535 /* find 1st bit in difference between the 2 addrs. 536 537 See comment in __ipv6_addr_diff: bit may be an invalid value, 538 but if it is >= plen, the value is ignored in any case. 539 */ 540 541 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr)); 542 543 /* 544 * (intermediate)[in] 545 * / \ 546 * (new leaf node)[ln] (old node)[fn] 547 */ 548 if (plen > bit) { 549 in = node_alloc(); 550 ln = node_alloc(); 551 552 if (!in || !ln) { 553 if (in) 554 node_free(in); 555 if (ln) 556 node_free(ln); 557 return ERR_PTR(-ENOMEM); 558 } 559 560 /* 561 * new intermediate node. 562 * RTN_RTINFO will 563 * be off since that an address that chooses one of 564 * the branches would not match less specific routes 565 * in the other branch 566 */ 567 568 in->fn_bit = bit; 569 570 in->parent = pn; 571 in->leaf = fn->leaf; 572 atomic_inc(&in->leaf->rt6i_ref); 573 574 in->fn_sernum = sernum; 575 576 /* update parent pointer */ 577 if (dir) 578 pn->right = in; 579 else 580 pn->left = in; 581 582 ln->fn_bit = plen; 583 584 ln->parent = in; 585 fn->parent = in; 586 587 ln->fn_sernum = sernum; 588 589 if (addr_bit_set(addr, bit)) { 590 in->right = ln; 591 in->left = fn; 592 } else { 593 in->left = ln; 594 in->right = fn; 595 } 596 } else { /* plen <= bit */ 597 598 /* 599 * (new leaf node)[ln] 600 * / \ 601 * (old node)[fn] NULL 602 */ 603 604 ln = node_alloc(); 605 606 if (!ln) 607 return ERR_PTR(-ENOMEM); 608 609 ln->fn_bit = plen; 610 611 ln->parent = pn; 612 613 ln->fn_sernum = sernum; 614 615 if (dir) 616 pn->right = ln; 617 else 618 pn->left = ln; 619 620 if (addr_bit_set(&key->addr, plen)) 621 ln->right = fn; 622 else 623 ln->left = fn; 624 625 fn->parent = ln; 626 } 627 return ln; 628 } 629 630 static inline bool rt6_qualify_for_ecmp(struct rt6_info *rt) 631 { 632 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) == 633 RTF_GATEWAY; 634 } 635 636 static int fib6_commit_metrics(struct dst_entry *dst, 637 struct nlattr *mx, int mx_len) 638 { 639 struct nlattr *nla; 640 int remaining; 641 u32 *mp; 642 643 if (dst->flags & DST_HOST) { 644 mp = dst_metrics_write_ptr(dst); 645 } else { 646 mp = kzalloc(sizeof(u32) * RTAX_MAX, GFP_ATOMIC); 647 if (!mp) 648 return -ENOMEM; 649 dst_init_metrics(dst, mp, 0); 650 } 651 652 nla_for_each_attr(nla, mx, mx_len, remaining) { 653 int type = nla_type(nla); 654 655 if (type) { 656 if (type > RTAX_MAX) 657 return -EINVAL; 658 659 mp[type - 1] = nla_get_u32(nla); 660 } 661 } 662 return 0; 663 } 664 665 /* 666 * Insert routing information in a node. 667 */ 668 669 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, 670 struct nl_info *info, struct nlattr *mx, int mx_len) 671 { 672 struct rt6_info *iter = NULL; 673 struct rt6_info **ins; 674 int replace = (info->nlh && 675 (info->nlh->nlmsg_flags & NLM_F_REPLACE)); 676 int add = (!info->nlh || 677 (info->nlh->nlmsg_flags & NLM_F_CREATE)); 678 int found = 0; 679 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt); 680 int err; 681 682 ins = &fn->leaf; 683 684 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) { 685 /* 686 * Search for duplicates 687 */ 688 689 if (iter->rt6i_metric == rt->rt6i_metric) { 690 /* 691 * Same priority level 692 */ 693 if (info->nlh && 694 (info->nlh->nlmsg_flags & NLM_F_EXCL)) 695 return -EEXIST; 696 if (replace) { 697 found++; 698 break; 699 } 700 701 if (iter->dst.dev == rt->dst.dev && 702 iter->rt6i_idev == rt->rt6i_idev && 703 ipv6_addr_equal(&iter->rt6i_gateway, 704 &rt->rt6i_gateway)) { 705 if (rt->rt6i_nsiblings) 706 rt->rt6i_nsiblings = 0; 707 if (!(iter->rt6i_flags & RTF_EXPIRES)) 708 return -EEXIST; 709 if (!(rt->rt6i_flags & RTF_EXPIRES)) 710 rt6_clean_expires(iter); 711 else 712 rt6_set_expires(iter, rt->dst.expires); 713 return -EEXIST; 714 } 715 /* If we have the same destination and the same metric, 716 * but not the same gateway, then the route we try to 717 * add is sibling to this route, increment our counter 718 * of siblings, and later we will add our route to the 719 * list. 720 * Only static routes (which don't have flag 721 * RTF_EXPIRES) are used for ECMPv6. 722 * 723 * To avoid long list, we only had siblings if the 724 * route have a gateway. 725 */ 726 if (rt_can_ecmp && 727 rt6_qualify_for_ecmp(iter)) 728 rt->rt6i_nsiblings++; 729 } 730 731 if (iter->rt6i_metric > rt->rt6i_metric) 732 break; 733 734 ins = &iter->dst.rt6_next; 735 } 736 737 /* Reset round-robin state, if necessary */ 738 if (ins == &fn->leaf) 739 fn->rr_ptr = NULL; 740 741 /* Link this route to others same route. */ 742 if (rt->rt6i_nsiblings) { 743 unsigned int rt6i_nsiblings; 744 struct rt6_info *sibling, *temp_sibling; 745 746 /* Find the first route that have the same metric */ 747 sibling = fn->leaf; 748 while (sibling) { 749 if (sibling->rt6i_metric == rt->rt6i_metric && 750 rt6_qualify_for_ecmp(sibling)) { 751 list_add_tail(&rt->rt6i_siblings, 752 &sibling->rt6i_siblings); 753 break; 754 } 755 sibling = sibling->dst.rt6_next; 756 } 757 /* For each sibling in the list, increment the counter of 758 * siblings. BUG() if counters does not match, list of siblings 759 * is broken! 760 */ 761 rt6i_nsiblings = 0; 762 list_for_each_entry_safe(sibling, temp_sibling, 763 &rt->rt6i_siblings, rt6i_siblings) { 764 sibling->rt6i_nsiblings++; 765 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings); 766 rt6i_nsiblings++; 767 } 768 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings); 769 } 770 771 /* 772 * insert node 773 */ 774 if (!replace) { 775 if (!add) 776 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 777 778 add: 779 if (mx) { 780 err = fib6_commit_metrics(&rt->dst, mx, mx_len); 781 if (err) 782 return err; 783 } 784 rt->dst.rt6_next = iter; 785 *ins = rt; 786 rt->rt6i_node = fn; 787 atomic_inc(&rt->rt6i_ref); 788 inet6_rt_notify(RTM_NEWROUTE, rt, info); 789 info->nl_net->ipv6.rt6_stats->fib_rt_entries++; 790 791 if (!(fn->fn_flags & RTN_RTINFO)) { 792 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 793 fn->fn_flags |= RTN_RTINFO; 794 } 795 796 } else { 797 if (!found) { 798 if (add) 799 goto add; 800 pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); 801 return -ENOENT; 802 } 803 if (mx) { 804 err = fib6_commit_metrics(&rt->dst, mx, mx_len); 805 if (err) 806 return err; 807 } 808 *ins = rt; 809 rt->rt6i_node = fn; 810 rt->dst.rt6_next = iter->dst.rt6_next; 811 atomic_inc(&rt->rt6i_ref); 812 inet6_rt_notify(RTM_NEWROUTE, rt, info); 813 rt6_release(iter); 814 if (!(fn->fn_flags & RTN_RTINFO)) { 815 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 816 fn->fn_flags |= RTN_RTINFO; 817 } 818 } 819 820 return 0; 821 } 822 823 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt) 824 { 825 if (!timer_pending(&net->ipv6.ip6_fib_timer) && 826 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE))) 827 mod_timer(&net->ipv6.ip6_fib_timer, 828 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 829 } 830 831 void fib6_force_start_gc(struct net *net) 832 { 833 if (!timer_pending(&net->ipv6.ip6_fib_timer)) 834 mod_timer(&net->ipv6.ip6_fib_timer, 835 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 836 } 837 838 /* 839 * Add routing information to the routing tree. 840 * <destination addr>/<source addr> 841 * with source addr info in sub-trees 842 */ 843 844 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info, 845 struct nlattr *mx, int mx_len) 846 { 847 struct fib6_node *fn, *pn = NULL; 848 int err = -ENOMEM; 849 int allow_create = 1; 850 int replace_required = 0; 851 852 if (info->nlh) { 853 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) 854 allow_create = 0; 855 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) 856 replace_required = 1; 857 } 858 if (!allow_create && !replace_required) 859 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); 860 861 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen, 862 offsetof(struct rt6_info, rt6i_dst), allow_create, 863 replace_required); 864 if (IS_ERR(fn)) { 865 err = PTR_ERR(fn); 866 fn = NULL; 867 goto out; 868 } 869 870 pn = fn; 871 872 #ifdef CONFIG_IPV6_SUBTREES 873 if (rt->rt6i_src.plen) { 874 struct fib6_node *sn; 875 876 if (!fn->subtree) { 877 struct fib6_node *sfn; 878 879 /* 880 * Create subtree. 881 * 882 * fn[main tree] 883 * | 884 * sfn[subtree root] 885 * \ 886 * sn[new leaf node] 887 */ 888 889 /* Create subtree root node */ 890 sfn = node_alloc(); 891 if (!sfn) 892 goto st_failure; 893 894 sfn->leaf = info->nl_net->ipv6.ip6_null_entry; 895 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref); 896 sfn->fn_flags = RTN_ROOT; 897 sfn->fn_sernum = fib6_new_sernum(); 898 899 /* Now add the first leaf node to new subtree */ 900 901 sn = fib6_add_1(sfn, &rt->rt6i_src.addr, 902 rt->rt6i_src.plen, 903 offsetof(struct rt6_info, rt6i_src), 904 allow_create, replace_required); 905 906 if (IS_ERR(sn)) { 907 /* If it is failed, discard just allocated 908 root, and then (in st_failure) stale node 909 in main tree. 910 */ 911 node_free(sfn); 912 err = PTR_ERR(sn); 913 goto st_failure; 914 } 915 916 /* Now link new subtree to main tree */ 917 sfn->parent = fn; 918 fn->subtree = sfn; 919 } else { 920 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, 921 rt->rt6i_src.plen, 922 offsetof(struct rt6_info, rt6i_src), 923 allow_create, replace_required); 924 925 if (IS_ERR(sn)) { 926 err = PTR_ERR(sn); 927 goto st_failure; 928 } 929 } 930 931 if (!fn->leaf) { 932 fn->leaf = rt; 933 atomic_inc(&rt->rt6i_ref); 934 } 935 fn = sn; 936 } 937 #endif 938 939 err = fib6_add_rt2node(fn, rt, info, mx, mx_len); 940 if (!err) { 941 fib6_start_gc(info->nl_net, rt); 942 if (!(rt->rt6i_flags & RTF_CACHE)) 943 fib6_prune_clones(info->nl_net, pn); 944 } 945 946 out: 947 if (err) { 948 #ifdef CONFIG_IPV6_SUBTREES 949 /* 950 * If fib6_add_1 has cleared the old leaf pointer in the 951 * super-tree leaf node we have to find a new one for it. 952 */ 953 if (pn != fn && pn->leaf == rt) { 954 pn->leaf = NULL; 955 atomic_dec(&rt->rt6i_ref); 956 } 957 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) { 958 pn->leaf = fib6_find_prefix(info->nl_net, pn); 959 #if RT6_DEBUG >= 2 960 if (!pn->leaf) { 961 WARN_ON(pn->leaf == NULL); 962 pn->leaf = info->nl_net->ipv6.ip6_null_entry; 963 } 964 #endif 965 atomic_inc(&pn->leaf->rt6i_ref); 966 } 967 #endif 968 dst_free(&rt->dst); 969 } 970 return err; 971 972 #ifdef CONFIG_IPV6_SUBTREES 973 /* Subtree creation failed, probably main tree node 974 is orphan. If it is, shoot it. 975 */ 976 st_failure: 977 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) 978 fib6_repair_tree(info->nl_net, fn); 979 dst_free(&rt->dst); 980 return err; 981 #endif 982 } 983 984 /* 985 * Routing tree lookup 986 * 987 */ 988 989 struct lookup_args { 990 int offset; /* key offset on rt6_info */ 991 const struct in6_addr *addr; /* search key */ 992 }; 993 994 static struct fib6_node *fib6_lookup_1(struct fib6_node *root, 995 struct lookup_args *args) 996 { 997 struct fib6_node *fn; 998 __be32 dir; 999 1000 if (unlikely(args->offset == 0)) 1001 return NULL; 1002 1003 /* 1004 * Descend on a tree 1005 */ 1006 1007 fn = root; 1008 1009 for (;;) { 1010 struct fib6_node *next; 1011 1012 dir = addr_bit_set(args->addr, fn->fn_bit); 1013 1014 next = dir ? fn->right : fn->left; 1015 1016 if (next) { 1017 fn = next; 1018 continue; 1019 } 1020 break; 1021 } 1022 1023 while (fn) { 1024 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) { 1025 struct rt6key *key; 1026 1027 key = (struct rt6key *) ((u8 *) fn->leaf + 1028 args->offset); 1029 1030 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 1031 #ifdef CONFIG_IPV6_SUBTREES 1032 if (fn->subtree) { 1033 struct fib6_node *sfn; 1034 sfn = fib6_lookup_1(fn->subtree, 1035 args + 1); 1036 if (!sfn) 1037 goto backtrack; 1038 fn = sfn; 1039 } 1040 #endif 1041 if (fn->fn_flags & RTN_RTINFO) 1042 return fn; 1043 } 1044 } 1045 #ifdef CONFIG_IPV6_SUBTREES 1046 backtrack: 1047 #endif 1048 if (fn->fn_flags & RTN_ROOT) 1049 break; 1050 1051 fn = fn->parent; 1052 } 1053 1054 return NULL; 1055 } 1056 1057 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr, 1058 const struct in6_addr *saddr) 1059 { 1060 struct fib6_node *fn; 1061 struct lookup_args args[] = { 1062 { 1063 .offset = offsetof(struct rt6_info, rt6i_dst), 1064 .addr = daddr, 1065 }, 1066 #ifdef CONFIG_IPV6_SUBTREES 1067 { 1068 .offset = offsetof(struct rt6_info, rt6i_src), 1069 .addr = saddr, 1070 }, 1071 #endif 1072 { 1073 .offset = 0, /* sentinel */ 1074 } 1075 }; 1076 1077 fn = fib6_lookup_1(root, daddr ? args : args + 1); 1078 if (!fn || fn->fn_flags & RTN_TL_ROOT) 1079 fn = root; 1080 1081 return fn; 1082 } 1083 1084 /* 1085 * Get node with specified destination prefix (and source prefix, 1086 * if subtrees are used) 1087 */ 1088 1089 1090 static struct fib6_node *fib6_locate_1(struct fib6_node *root, 1091 const struct in6_addr *addr, 1092 int plen, int offset) 1093 { 1094 struct fib6_node *fn; 1095 1096 for (fn = root; fn ; ) { 1097 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); 1098 1099 /* 1100 * Prefix match 1101 */ 1102 if (plen < fn->fn_bit || 1103 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 1104 return NULL; 1105 1106 if (plen == fn->fn_bit) 1107 return fn; 1108 1109 /* 1110 * We have more bits to go 1111 */ 1112 if (addr_bit_set(addr, fn->fn_bit)) 1113 fn = fn->right; 1114 else 1115 fn = fn->left; 1116 } 1117 return NULL; 1118 } 1119 1120 struct fib6_node *fib6_locate(struct fib6_node *root, 1121 const struct in6_addr *daddr, int dst_len, 1122 const struct in6_addr *saddr, int src_len) 1123 { 1124 struct fib6_node *fn; 1125 1126 fn = fib6_locate_1(root, daddr, dst_len, 1127 offsetof(struct rt6_info, rt6i_dst)); 1128 1129 #ifdef CONFIG_IPV6_SUBTREES 1130 if (src_len) { 1131 WARN_ON(saddr == NULL); 1132 if (fn && fn->subtree) 1133 fn = fib6_locate_1(fn->subtree, saddr, src_len, 1134 offsetof(struct rt6_info, rt6i_src)); 1135 } 1136 #endif 1137 1138 if (fn && fn->fn_flags & RTN_RTINFO) 1139 return fn; 1140 1141 return NULL; 1142 } 1143 1144 1145 /* 1146 * Deletion 1147 * 1148 */ 1149 1150 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn) 1151 { 1152 if (fn->fn_flags & RTN_ROOT) 1153 return net->ipv6.ip6_null_entry; 1154 1155 while (fn) { 1156 if (fn->left) 1157 return fn->left->leaf; 1158 if (fn->right) 1159 return fn->right->leaf; 1160 1161 fn = FIB6_SUBTREE(fn); 1162 } 1163 return NULL; 1164 } 1165 1166 /* 1167 * Called to trim the tree of intermediate nodes when possible. "fn" 1168 * is the node we want to try and remove. 1169 */ 1170 1171 static struct fib6_node *fib6_repair_tree(struct net *net, 1172 struct fib6_node *fn) 1173 { 1174 int children; 1175 int nstate; 1176 struct fib6_node *child, *pn; 1177 struct fib6_walker_t *w; 1178 int iter = 0; 1179 1180 for (;;) { 1181 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1182 iter++; 1183 1184 WARN_ON(fn->fn_flags & RTN_RTINFO); 1185 WARN_ON(fn->fn_flags & RTN_TL_ROOT); 1186 WARN_ON(fn->leaf != NULL); 1187 1188 children = 0; 1189 child = NULL; 1190 if (fn->right) 1191 child = fn->right, children |= 1; 1192 if (fn->left) 1193 child = fn->left, children |= 2; 1194 1195 if (children == 3 || FIB6_SUBTREE(fn) 1196 #ifdef CONFIG_IPV6_SUBTREES 1197 /* Subtree root (i.e. fn) may have one child */ 1198 || (children && fn->fn_flags & RTN_ROOT) 1199 #endif 1200 ) { 1201 fn->leaf = fib6_find_prefix(net, fn); 1202 #if RT6_DEBUG >= 2 1203 if (!fn->leaf) { 1204 WARN_ON(!fn->leaf); 1205 fn->leaf = net->ipv6.ip6_null_entry; 1206 } 1207 #endif 1208 atomic_inc(&fn->leaf->rt6i_ref); 1209 return fn->parent; 1210 } 1211 1212 pn = fn->parent; 1213 #ifdef CONFIG_IPV6_SUBTREES 1214 if (FIB6_SUBTREE(pn) == fn) { 1215 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1216 FIB6_SUBTREE(pn) = NULL; 1217 nstate = FWS_L; 1218 } else { 1219 WARN_ON(fn->fn_flags & RTN_ROOT); 1220 #endif 1221 if (pn->right == fn) 1222 pn->right = child; 1223 else if (pn->left == fn) 1224 pn->left = child; 1225 #if RT6_DEBUG >= 2 1226 else 1227 WARN_ON(1); 1228 #endif 1229 if (child) 1230 child->parent = pn; 1231 nstate = FWS_R; 1232 #ifdef CONFIG_IPV6_SUBTREES 1233 } 1234 #endif 1235 1236 read_lock(&fib6_walker_lock); 1237 FOR_WALKERS(w) { 1238 if (!child) { 1239 if (w->root == fn) { 1240 w->root = w->node = NULL; 1241 RT6_TRACE("W %p adjusted by delroot 1\n", w); 1242 } else if (w->node == fn) { 1243 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1244 w->node = pn; 1245 w->state = nstate; 1246 } 1247 } else { 1248 if (w->root == fn) { 1249 w->root = child; 1250 RT6_TRACE("W %p adjusted by delroot 2\n", w); 1251 } 1252 if (w->node == fn) { 1253 w->node = child; 1254 if (children&2) { 1255 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1256 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT; 1257 } else { 1258 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1259 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT; 1260 } 1261 } 1262 } 1263 } 1264 read_unlock(&fib6_walker_lock); 1265 1266 node_free(fn); 1267 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) 1268 return pn; 1269 1270 rt6_release(pn->leaf); 1271 pn->leaf = NULL; 1272 fn = pn; 1273 } 1274 } 1275 1276 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, 1277 struct nl_info *info) 1278 { 1279 struct fib6_walker_t *w; 1280 struct rt6_info *rt = *rtp; 1281 struct net *net = info->nl_net; 1282 1283 RT6_TRACE("fib6_del_route\n"); 1284 1285 /* Unlink it */ 1286 *rtp = rt->dst.rt6_next; 1287 rt->rt6i_node = NULL; 1288 net->ipv6.rt6_stats->fib_rt_entries--; 1289 net->ipv6.rt6_stats->fib_discarded_routes++; 1290 1291 /* Reset round-robin state, if necessary */ 1292 if (fn->rr_ptr == rt) 1293 fn->rr_ptr = NULL; 1294 1295 /* Remove this entry from other siblings */ 1296 if (rt->rt6i_nsiblings) { 1297 struct rt6_info *sibling, *next_sibling; 1298 1299 list_for_each_entry_safe(sibling, next_sibling, 1300 &rt->rt6i_siblings, rt6i_siblings) 1301 sibling->rt6i_nsiblings--; 1302 rt->rt6i_nsiblings = 0; 1303 list_del_init(&rt->rt6i_siblings); 1304 } 1305 1306 /* Adjust walkers */ 1307 read_lock(&fib6_walker_lock); 1308 FOR_WALKERS(w) { 1309 if (w->state == FWS_C && w->leaf == rt) { 1310 RT6_TRACE("walker %p adjusted by delroute\n", w); 1311 w->leaf = rt->dst.rt6_next; 1312 if (!w->leaf) 1313 w->state = FWS_U; 1314 } 1315 } 1316 read_unlock(&fib6_walker_lock); 1317 1318 rt->dst.rt6_next = NULL; 1319 1320 /* If it was last route, expunge its radix tree node */ 1321 if (!fn->leaf) { 1322 fn->fn_flags &= ~RTN_RTINFO; 1323 net->ipv6.rt6_stats->fib_route_nodes--; 1324 fn = fib6_repair_tree(net, fn); 1325 } 1326 1327 if (atomic_read(&rt->rt6i_ref) != 1) { 1328 /* This route is used as dummy address holder in some split 1329 * nodes. It is not leaked, but it still holds other resources, 1330 * which must be released in time. So, scan ascendant nodes 1331 * and replace dummy references to this route with references 1332 * to still alive ones. 1333 */ 1334 while (fn) { 1335 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) { 1336 fn->leaf = fib6_find_prefix(net, fn); 1337 atomic_inc(&fn->leaf->rt6i_ref); 1338 rt6_release(rt); 1339 } 1340 fn = fn->parent; 1341 } 1342 /* No more references are possible at this point. */ 1343 BUG_ON(atomic_read(&rt->rt6i_ref) != 1); 1344 } 1345 1346 inet6_rt_notify(RTM_DELROUTE, rt, info); 1347 rt6_release(rt); 1348 } 1349 1350 int fib6_del(struct rt6_info *rt, struct nl_info *info) 1351 { 1352 struct net *net = info->nl_net; 1353 struct fib6_node *fn = rt->rt6i_node; 1354 struct rt6_info **rtp; 1355 1356 #if RT6_DEBUG >= 2 1357 if (rt->dst.obsolete > 0) { 1358 WARN_ON(fn != NULL); 1359 return -ENOENT; 1360 } 1361 #endif 1362 if (!fn || rt == net->ipv6.ip6_null_entry) 1363 return -ENOENT; 1364 1365 WARN_ON(!(fn->fn_flags & RTN_RTINFO)); 1366 1367 if (!(rt->rt6i_flags & RTF_CACHE)) { 1368 struct fib6_node *pn = fn; 1369 #ifdef CONFIG_IPV6_SUBTREES 1370 /* clones of this route might be in another subtree */ 1371 if (rt->rt6i_src.plen) { 1372 while (!(pn->fn_flags & RTN_ROOT)) 1373 pn = pn->parent; 1374 pn = pn->parent; 1375 } 1376 #endif 1377 fib6_prune_clones(info->nl_net, pn); 1378 } 1379 1380 /* 1381 * Walk the leaf entries looking for ourself 1382 */ 1383 1384 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) { 1385 if (*rtp == rt) { 1386 fib6_del_route(fn, rtp, info); 1387 return 0; 1388 } 1389 } 1390 return -ENOENT; 1391 } 1392 1393 /* 1394 * Tree traversal function. 1395 * 1396 * Certainly, it is not interrupt safe. 1397 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 1398 * It means, that we can modify tree during walking 1399 * and use this function for garbage collection, clone pruning, 1400 * cleaning tree when a device goes down etc. etc. 1401 * 1402 * It guarantees that every node will be traversed, 1403 * and that it will be traversed only once. 1404 * 1405 * Callback function w->func may return: 1406 * 0 -> continue walking. 1407 * positive value -> walking is suspended (used by tree dumps, 1408 * and probably by gc, if it will be split to several slices) 1409 * negative value -> terminate walking. 1410 * 1411 * The function itself returns: 1412 * 0 -> walk is complete. 1413 * >0 -> walk is incomplete (i.e. suspended) 1414 * <0 -> walk is terminated by an error. 1415 */ 1416 1417 static int fib6_walk_continue(struct fib6_walker_t *w) 1418 { 1419 struct fib6_node *fn, *pn; 1420 1421 for (;;) { 1422 fn = w->node; 1423 if (!fn) 1424 return 0; 1425 1426 if (w->prune && fn != w->root && 1427 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) { 1428 w->state = FWS_C; 1429 w->leaf = fn->leaf; 1430 } 1431 switch (w->state) { 1432 #ifdef CONFIG_IPV6_SUBTREES 1433 case FWS_S: 1434 if (FIB6_SUBTREE(fn)) { 1435 w->node = FIB6_SUBTREE(fn); 1436 continue; 1437 } 1438 w->state = FWS_L; 1439 #endif 1440 case FWS_L: 1441 if (fn->left) { 1442 w->node = fn->left; 1443 w->state = FWS_INIT; 1444 continue; 1445 } 1446 w->state = FWS_R; 1447 case FWS_R: 1448 if (fn->right) { 1449 w->node = fn->right; 1450 w->state = FWS_INIT; 1451 continue; 1452 } 1453 w->state = FWS_C; 1454 w->leaf = fn->leaf; 1455 case FWS_C: 1456 if (w->leaf && fn->fn_flags & RTN_RTINFO) { 1457 int err; 1458 1459 if (w->skip) { 1460 w->skip--; 1461 goto skip; 1462 } 1463 1464 err = w->func(w); 1465 if (err) 1466 return err; 1467 1468 w->count++; 1469 continue; 1470 } 1471 skip: 1472 w->state = FWS_U; 1473 case FWS_U: 1474 if (fn == w->root) 1475 return 0; 1476 pn = fn->parent; 1477 w->node = pn; 1478 #ifdef CONFIG_IPV6_SUBTREES 1479 if (FIB6_SUBTREE(pn) == fn) { 1480 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1481 w->state = FWS_L; 1482 continue; 1483 } 1484 #endif 1485 if (pn->left == fn) { 1486 w->state = FWS_R; 1487 continue; 1488 } 1489 if (pn->right == fn) { 1490 w->state = FWS_C; 1491 w->leaf = w->node->leaf; 1492 continue; 1493 } 1494 #if RT6_DEBUG >= 2 1495 WARN_ON(1); 1496 #endif 1497 } 1498 } 1499 } 1500 1501 static int fib6_walk(struct fib6_walker_t *w) 1502 { 1503 int res; 1504 1505 w->state = FWS_INIT; 1506 w->node = w->root; 1507 1508 fib6_walker_link(w); 1509 res = fib6_walk_continue(w); 1510 if (res <= 0) 1511 fib6_walker_unlink(w); 1512 return res; 1513 } 1514 1515 static int fib6_clean_node(struct fib6_walker_t *w) 1516 { 1517 int res; 1518 struct rt6_info *rt; 1519 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w); 1520 struct nl_info info = { 1521 .nl_net = c->net, 1522 }; 1523 1524 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 1525 res = c->func(rt, c->arg); 1526 if (res < 0) { 1527 w->leaf = rt; 1528 res = fib6_del(rt, &info); 1529 if (res) { 1530 #if RT6_DEBUG >= 2 1531 pr_debug("%s: del failed: rt=%p@%p err=%d\n", 1532 __func__, rt, rt->rt6i_node, res); 1533 #endif 1534 continue; 1535 } 1536 return 0; 1537 } 1538 WARN_ON(res != 0); 1539 } 1540 w->leaf = rt; 1541 return 0; 1542 } 1543 1544 /* 1545 * Convenient frontend to tree walker. 1546 * 1547 * func is called on each route. 1548 * It may return -1 -> delete this route. 1549 * 0 -> continue walking 1550 * 1551 * prune==1 -> only immediate children of node (certainly, 1552 * ignoring pure split nodes) will be scanned. 1553 */ 1554 1555 static void fib6_clean_tree(struct net *net, struct fib6_node *root, 1556 int (*func)(struct rt6_info *, void *arg), 1557 int prune, void *arg) 1558 { 1559 struct fib6_cleaner_t c; 1560 1561 c.w.root = root; 1562 c.w.func = fib6_clean_node; 1563 c.w.prune = prune; 1564 c.w.count = 0; 1565 c.w.skip = 0; 1566 c.func = func; 1567 c.arg = arg; 1568 c.net = net; 1569 1570 fib6_walk(&c.w); 1571 } 1572 1573 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg), 1574 void *arg) 1575 { 1576 struct fib6_table *table; 1577 struct hlist_head *head; 1578 unsigned int h; 1579 1580 rcu_read_lock(); 1581 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 1582 head = &net->ipv6.fib_table_hash[h]; 1583 hlist_for_each_entry_rcu(table, head, tb6_hlist) { 1584 write_lock_bh(&table->tb6_lock); 1585 fib6_clean_tree(net, &table->tb6_root, 1586 func, 0, arg); 1587 write_unlock_bh(&table->tb6_lock); 1588 } 1589 } 1590 rcu_read_unlock(); 1591 } 1592 1593 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1594 { 1595 if (rt->rt6i_flags & RTF_CACHE) { 1596 RT6_TRACE("pruning clone %p\n", rt); 1597 return -1; 1598 } 1599 1600 return 0; 1601 } 1602 1603 static void fib6_prune_clones(struct net *net, struct fib6_node *fn) 1604 { 1605 fib6_clean_tree(net, fn, fib6_prune_clone, 1, NULL); 1606 } 1607 1608 static int fib6_update_sernum(struct rt6_info *rt, void *arg) 1609 { 1610 __u32 sernum = *(__u32 *)arg; 1611 1612 if (rt->rt6i_node && 1613 rt->rt6i_node->fn_sernum != sernum) 1614 rt->rt6i_node->fn_sernum = sernum; 1615 1616 return 0; 1617 } 1618 1619 static void fib6_flush_trees(struct net *net) 1620 { 1621 __u32 new_sernum = fib6_new_sernum(); 1622 1623 fib6_clean_all(net, fib6_update_sernum, &new_sernum); 1624 } 1625 1626 /* 1627 * Garbage collection 1628 */ 1629 1630 static struct fib6_gc_args 1631 { 1632 int timeout; 1633 int more; 1634 } gc_args; 1635 1636 static int fib6_age(struct rt6_info *rt, void *arg) 1637 { 1638 unsigned long now = jiffies; 1639 1640 /* 1641 * check addrconf expiration here. 1642 * Routes are expired even if they are in use. 1643 * 1644 * Also age clones. Note, that clones are aged out 1645 * only if they are not in use now. 1646 */ 1647 1648 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) { 1649 if (time_after(now, rt->dst.expires)) { 1650 RT6_TRACE("expiring %p\n", rt); 1651 return -1; 1652 } 1653 gc_args.more++; 1654 } else if (rt->rt6i_flags & RTF_CACHE) { 1655 if (atomic_read(&rt->dst.__refcnt) == 0 && 1656 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) { 1657 RT6_TRACE("aging clone %p\n", rt); 1658 return -1; 1659 } else if (rt->rt6i_flags & RTF_GATEWAY) { 1660 struct neighbour *neigh; 1661 __u8 neigh_flags = 0; 1662 1663 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway); 1664 if (neigh) { 1665 neigh_flags = neigh->flags; 1666 neigh_release(neigh); 1667 } 1668 if (!(neigh_flags & NTF_ROUTER)) { 1669 RT6_TRACE("purging route %p via non-router but gateway\n", 1670 rt); 1671 return -1; 1672 } 1673 } 1674 gc_args.more++; 1675 } 1676 1677 return 0; 1678 } 1679 1680 static DEFINE_SPINLOCK(fib6_gc_lock); 1681 1682 void fib6_run_gc(unsigned long expires, struct net *net, bool force) 1683 { 1684 unsigned long now; 1685 1686 if (force) { 1687 spin_lock_bh(&fib6_gc_lock); 1688 } else if (!spin_trylock_bh(&fib6_gc_lock)) { 1689 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); 1690 return; 1691 } 1692 gc_args.timeout = expires ? (int)expires : 1693 net->ipv6.sysctl.ip6_rt_gc_interval; 1694 1695 gc_args.more = icmp6_dst_gc(); 1696 1697 fib6_clean_all(net, fib6_age, NULL); 1698 now = jiffies; 1699 net->ipv6.ip6_rt_last_gc = now; 1700 1701 if (gc_args.more) 1702 mod_timer(&net->ipv6.ip6_fib_timer, 1703 round_jiffies(now 1704 + net->ipv6.sysctl.ip6_rt_gc_interval)); 1705 else 1706 del_timer(&net->ipv6.ip6_fib_timer); 1707 spin_unlock_bh(&fib6_gc_lock); 1708 } 1709 1710 static void fib6_gc_timer_cb(unsigned long arg) 1711 { 1712 fib6_run_gc(0, (struct net *)arg, true); 1713 } 1714 1715 static int __net_init fib6_net_init(struct net *net) 1716 { 1717 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; 1718 1719 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net); 1720 1721 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); 1722 if (!net->ipv6.rt6_stats) 1723 goto out_timer; 1724 1725 /* Avoid false sharing : Use at least a full cache line */ 1726 size = max_t(size_t, size, L1_CACHE_BYTES); 1727 1728 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); 1729 if (!net->ipv6.fib_table_hash) 1730 goto out_rt6_stats; 1731 1732 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), 1733 GFP_KERNEL); 1734 if (!net->ipv6.fib6_main_tbl) 1735 goto out_fib_table_hash; 1736 1737 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; 1738 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1739 net->ipv6.fib6_main_tbl->tb6_root.fn_flags = 1740 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1741 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); 1742 1743 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1744 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), 1745 GFP_KERNEL); 1746 if (!net->ipv6.fib6_local_tbl) 1747 goto out_fib6_main_tbl; 1748 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; 1749 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1750 net->ipv6.fib6_local_tbl->tb6_root.fn_flags = 1751 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1752 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); 1753 #endif 1754 fib6_tables_init(net); 1755 1756 return 0; 1757 1758 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1759 out_fib6_main_tbl: 1760 kfree(net->ipv6.fib6_main_tbl); 1761 #endif 1762 out_fib_table_hash: 1763 kfree(net->ipv6.fib_table_hash); 1764 out_rt6_stats: 1765 kfree(net->ipv6.rt6_stats); 1766 out_timer: 1767 return -ENOMEM; 1768 } 1769 1770 static void fib6_net_exit(struct net *net) 1771 { 1772 rt6_ifdown(net, NULL); 1773 del_timer_sync(&net->ipv6.ip6_fib_timer); 1774 1775 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1776 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers); 1777 kfree(net->ipv6.fib6_local_tbl); 1778 #endif 1779 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers); 1780 kfree(net->ipv6.fib6_main_tbl); 1781 kfree(net->ipv6.fib_table_hash); 1782 kfree(net->ipv6.rt6_stats); 1783 } 1784 1785 static struct pernet_operations fib6_net_ops = { 1786 .init = fib6_net_init, 1787 .exit = fib6_net_exit, 1788 }; 1789 1790 int __init fib6_init(void) 1791 { 1792 int ret = -ENOMEM; 1793 1794 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1795 sizeof(struct fib6_node), 1796 0, SLAB_HWCACHE_ALIGN, 1797 NULL); 1798 if (!fib6_node_kmem) 1799 goto out; 1800 1801 ret = register_pernet_subsys(&fib6_net_ops); 1802 if (ret) 1803 goto out_kmem_cache_create; 1804 1805 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib, 1806 NULL); 1807 if (ret) 1808 goto out_unregister_subsys; 1809 1810 __fib6_flush_trees = fib6_flush_trees; 1811 out: 1812 return ret; 1813 1814 out_unregister_subsys: 1815 unregister_pernet_subsys(&fib6_net_ops); 1816 out_kmem_cache_create: 1817 kmem_cache_destroy(fib6_node_kmem); 1818 goto out; 1819 } 1820 1821 void fib6_gc_cleanup(void) 1822 { 1823 unregister_pernet_subsys(&fib6_net_ops); 1824 kmem_cache_destroy(fib6_node_kmem); 1825 } 1826 1827 #ifdef CONFIG_PROC_FS 1828 1829 struct ipv6_route_iter { 1830 struct seq_net_private p; 1831 struct fib6_walker_t w; 1832 loff_t skip; 1833 struct fib6_table *tbl; 1834 __u32 sernum; 1835 }; 1836 1837 static int ipv6_route_seq_show(struct seq_file *seq, void *v) 1838 { 1839 struct rt6_info *rt = v; 1840 struct ipv6_route_iter *iter = seq->private; 1841 1842 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen); 1843 1844 #ifdef CONFIG_IPV6_SUBTREES 1845 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen); 1846 #else 1847 seq_puts(seq, "00000000000000000000000000000000 00 "); 1848 #endif 1849 if (rt->rt6i_flags & RTF_GATEWAY) 1850 seq_printf(seq, "%pi6", &rt->rt6i_gateway); 1851 else 1852 seq_puts(seq, "00000000000000000000000000000000"); 1853 1854 seq_printf(seq, " %08x %08x %08x %08x %8s\n", 1855 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt), 1856 rt->dst.__use, rt->rt6i_flags, 1857 rt->dst.dev ? rt->dst.dev->name : ""); 1858 iter->w.leaf = NULL; 1859 return 0; 1860 } 1861 1862 static int ipv6_route_yield(struct fib6_walker_t *w) 1863 { 1864 struct ipv6_route_iter *iter = w->args; 1865 1866 if (!iter->skip) 1867 return 1; 1868 1869 do { 1870 iter->w.leaf = iter->w.leaf->dst.rt6_next; 1871 iter->skip--; 1872 if (!iter->skip && iter->w.leaf) 1873 return 1; 1874 } while (iter->w.leaf); 1875 1876 return 0; 1877 } 1878 1879 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter) 1880 { 1881 memset(&iter->w, 0, sizeof(iter->w)); 1882 iter->w.func = ipv6_route_yield; 1883 iter->w.root = &iter->tbl->tb6_root; 1884 iter->w.state = FWS_INIT; 1885 iter->w.node = iter->w.root; 1886 iter->w.args = iter; 1887 iter->sernum = iter->w.root->fn_sernum; 1888 INIT_LIST_HEAD(&iter->w.lh); 1889 fib6_walker_link(&iter->w); 1890 } 1891 1892 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl, 1893 struct net *net) 1894 { 1895 unsigned int h; 1896 struct hlist_node *node; 1897 1898 if (tbl) { 1899 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1; 1900 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist)); 1901 } else { 1902 h = 0; 1903 node = NULL; 1904 } 1905 1906 while (!node && h < FIB6_TABLE_HASHSZ) { 1907 node = rcu_dereference_bh( 1908 hlist_first_rcu(&net->ipv6.fib_table_hash[h++])); 1909 } 1910 return hlist_entry_safe(node, struct fib6_table, tb6_hlist); 1911 } 1912 1913 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter) 1914 { 1915 if (iter->sernum != iter->w.root->fn_sernum) { 1916 iter->sernum = iter->w.root->fn_sernum; 1917 iter->w.state = FWS_INIT; 1918 iter->w.node = iter->w.root; 1919 WARN_ON(iter->w.skip); 1920 iter->w.skip = iter->w.count; 1921 } 1922 } 1923 1924 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1925 { 1926 int r; 1927 struct rt6_info *n; 1928 struct net *net = seq_file_net(seq); 1929 struct ipv6_route_iter *iter = seq->private; 1930 1931 if (!v) 1932 goto iter_table; 1933 1934 n = ((struct rt6_info *)v)->dst.rt6_next; 1935 if (n) { 1936 ++*pos; 1937 return n; 1938 } 1939 1940 iter_table: 1941 ipv6_route_check_sernum(iter); 1942 read_lock(&iter->tbl->tb6_lock); 1943 r = fib6_walk_continue(&iter->w); 1944 read_unlock(&iter->tbl->tb6_lock); 1945 if (r > 0) { 1946 if (v) 1947 ++*pos; 1948 return iter->w.leaf; 1949 } else if (r < 0) { 1950 fib6_walker_unlink(&iter->w); 1951 return NULL; 1952 } 1953 fib6_walker_unlink(&iter->w); 1954 1955 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net); 1956 if (!iter->tbl) 1957 return NULL; 1958 1959 ipv6_route_seq_setup_walk(iter); 1960 goto iter_table; 1961 } 1962 1963 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos) 1964 __acquires(RCU_BH) 1965 { 1966 struct net *net = seq_file_net(seq); 1967 struct ipv6_route_iter *iter = seq->private; 1968 1969 rcu_read_lock_bh(); 1970 iter->tbl = ipv6_route_seq_next_table(NULL, net); 1971 iter->skip = *pos; 1972 1973 if (iter->tbl) { 1974 ipv6_route_seq_setup_walk(iter); 1975 return ipv6_route_seq_next(seq, NULL, pos); 1976 } else { 1977 return NULL; 1978 } 1979 } 1980 1981 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter) 1982 { 1983 struct fib6_walker_t *w = &iter->w; 1984 return w->node && !(w->state == FWS_U && w->node == w->root); 1985 } 1986 1987 static void ipv6_route_seq_stop(struct seq_file *seq, void *v) 1988 __releases(RCU_BH) 1989 { 1990 struct ipv6_route_iter *iter = seq->private; 1991 1992 if (ipv6_route_iter_active(iter)) 1993 fib6_walker_unlink(&iter->w); 1994 1995 rcu_read_unlock_bh(); 1996 } 1997 1998 static const struct seq_operations ipv6_route_seq_ops = { 1999 .start = ipv6_route_seq_start, 2000 .next = ipv6_route_seq_next, 2001 .stop = ipv6_route_seq_stop, 2002 .show = ipv6_route_seq_show 2003 }; 2004 2005 int ipv6_route_open(struct inode *inode, struct file *file) 2006 { 2007 return seq_open_net(inode, file, &ipv6_route_seq_ops, 2008 sizeof(struct ipv6_route_iter)); 2009 } 2010 2011 #endif /* CONFIG_PROC_FS */ 2012