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 #include <net/lwtunnel.h> 36 37 #include <net/ip6_fib.h> 38 #include <net/ip6_route.h> 39 40 #define RT6_DEBUG 2 41 42 #if RT6_DEBUG >= 3 43 #define RT6_TRACE(x...) pr_debug(x) 44 #else 45 #define RT6_TRACE(x...) do { ; } while (0) 46 #endif 47 48 static struct kmem_cache *fib6_node_kmem __read_mostly; 49 50 struct fib6_cleaner { 51 struct fib6_walker w; 52 struct net *net; 53 int (*func)(struct rt6_info *, void *arg); 54 int sernum; 55 void *arg; 56 }; 57 58 #ifdef CONFIG_IPV6_SUBTREES 59 #define FWS_INIT FWS_S 60 #else 61 #define FWS_INIT FWS_L 62 #endif 63 64 static void fib6_prune_clones(struct net *net, struct fib6_node *fn); 65 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn); 66 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn); 67 static int fib6_walk(struct net *net, struct fib6_walker *w); 68 static int fib6_walk_continue(struct fib6_walker *w); 69 70 /* 71 * A routing update causes an increase of the serial number on the 72 * affected subtree. This allows for cached routes to be asynchronously 73 * tested when modifications are made to the destination cache as a 74 * result of redirects, path MTU changes, etc. 75 */ 76 77 static void fib6_gc_timer_cb(unsigned long arg); 78 79 #define FOR_WALKERS(net, w) \ 80 list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh) 81 82 static void fib6_walker_link(struct net *net, struct fib6_walker *w) 83 { 84 write_lock_bh(&net->ipv6.fib6_walker_lock); 85 list_add(&w->lh, &net->ipv6.fib6_walkers); 86 write_unlock_bh(&net->ipv6.fib6_walker_lock); 87 } 88 89 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w) 90 { 91 write_lock_bh(&net->ipv6.fib6_walker_lock); 92 list_del(&w->lh); 93 write_unlock_bh(&net->ipv6.fib6_walker_lock); 94 } 95 96 static int fib6_new_sernum(struct net *net) 97 { 98 int new, old; 99 100 do { 101 old = atomic_read(&net->ipv6.fib6_sernum); 102 new = old < INT_MAX ? old + 1 : 1; 103 } while (atomic_cmpxchg(&net->ipv6.fib6_sernum, 104 old, new) != old); 105 return new; 106 } 107 108 enum { 109 FIB6_NO_SERNUM_CHANGE = 0, 110 }; 111 112 /* 113 * Auxiliary address test functions for the radix tree. 114 * 115 * These assume a 32bit processor (although it will work on 116 * 64bit processors) 117 */ 118 119 /* 120 * test bit 121 */ 122 #if defined(__LITTLE_ENDIAN) 123 # define BITOP_BE32_SWIZZLE (0x1F & ~7) 124 #else 125 # define BITOP_BE32_SWIZZLE 0 126 #endif 127 128 static __be32 addr_bit_set(const void *token, int fn_bit) 129 { 130 const __be32 *addr = token; 131 /* 132 * Here, 133 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f) 134 * is optimized version of 135 * htonl(1 << ((~fn_bit)&0x1F)) 136 * See include/asm-generic/bitops/le.h. 137 */ 138 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) & 139 addr[fn_bit >> 5]; 140 } 141 142 static struct fib6_node *node_alloc(void) 143 { 144 struct fib6_node *fn; 145 146 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC); 147 148 return fn; 149 } 150 151 static void node_free(struct fib6_node *fn) 152 { 153 kmem_cache_free(fib6_node_kmem, fn); 154 } 155 156 static void rt6_rcu_free(struct rt6_info *rt) 157 { 158 call_rcu(&rt->dst.rcu_head, dst_rcu_free); 159 } 160 161 static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt) 162 { 163 int cpu; 164 165 if (!non_pcpu_rt->rt6i_pcpu) 166 return; 167 168 for_each_possible_cpu(cpu) { 169 struct rt6_info **ppcpu_rt; 170 struct rt6_info *pcpu_rt; 171 172 ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu); 173 pcpu_rt = *ppcpu_rt; 174 if (pcpu_rt) { 175 rt6_rcu_free(pcpu_rt); 176 *ppcpu_rt = NULL; 177 } 178 } 179 180 non_pcpu_rt->rt6i_pcpu = NULL; 181 } 182 183 static void rt6_release(struct rt6_info *rt) 184 { 185 if (atomic_dec_and_test(&rt->rt6i_ref)) { 186 rt6_free_pcpu(rt); 187 rt6_rcu_free(rt); 188 } 189 } 190 191 static void fib6_link_table(struct net *net, struct fib6_table *tb) 192 { 193 unsigned int h; 194 195 /* 196 * Initialize table lock at a single place to give lockdep a key, 197 * tables aren't visible prior to being linked to the list. 198 */ 199 rwlock_init(&tb->tb6_lock); 200 201 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1); 202 203 /* 204 * No protection necessary, this is the only list mutatation 205 * operation, tables never disappear once they exist. 206 */ 207 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]); 208 } 209 210 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 211 212 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id) 213 { 214 struct fib6_table *table; 215 216 table = kzalloc(sizeof(*table), GFP_ATOMIC); 217 if (table) { 218 table->tb6_id = id; 219 table->tb6_root.leaf = net->ipv6.ip6_null_entry; 220 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 221 inet_peer_base_init(&table->tb6_peers); 222 } 223 224 return table; 225 } 226 227 struct fib6_table *fib6_new_table(struct net *net, u32 id) 228 { 229 struct fib6_table *tb; 230 231 if (id == 0) 232 id = RT6_TABLE_MAIN; 233 tb = fib6_get_table(net, id); 234 if (tb) 235 return tb; 236 237 tb = fib6_alloc_table(net, id); 238 if (tb) 239 fib6_link_table(net, tb); 240 241 return tb; 242 } 243 EXPORT_SYMBOL_GPL(fib6_new_table); 244 245 struct fib6_table *fib6_get_table(struct net *net, u32 id) 246 { 247 struct fib6_table *tb; 248 struct hlist_head *head; 249 unsigned int h; 250 251 if (id == 0) 252 id = RT6_TABLE_MAIN; 253 h = id & (FIB6_TABLE_HASHSZ - 1); 254 rcu_read_lock(); 255 head = &net->ipv6.fib_table_hash[h]; 256 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 257 if (tb->tb6_id == id) { 258 rcu_read_unlock(); 259 return tb; 260 } 261 } 262 rcu_read_unlock(); 263 264 return NULL; 265 } 266 EXPORT_SYMBOL_GPL(fib6_get_table); 267 268 static void __net_init fib6_tables_init(struct net *net) 269 { 270 fib6_link_table(net, net->ipv6.fib6_main_tbl); 271 fib6_link_table(net, net->ipv6.fib6_local_tbl); 272 } 273 #else 274 275 struct fib6_table *fib6_new_table(struct net *net, u32 id) 276 { 277 return fib6_get_table(net, id); 278 } 279 280 struct fib6_table *fib6_get_table(struct net *net, u32 id) 281 { 282 return net->ipv6.fib6_main_tbl; 283 } 284 285 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, 286 int flags, pol_lookup_t lookup) 287 { 288 struct rt6_info *rt; 289 290 rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags); 291 if (rt->rt6i_flags & RTF_REJECT && 292 rt->dst.error == -EAGAIN) { 293 ip6_rt_put(rt); 294 rt = net->ipv6.ip6_null_entry; 295 dst_hold(&rt->dst); 296 } 297 298 return &rt->dst; 299 } 300 301 static void __net_init fib6_tables_init(struct net *net) 302 { 303 fib6_link_table(net, net->ipv6.fib6_main_tbl); 304 } 305 306 #endif 307 308 static int fib6_dump_node(struct fib6_walker *w) 309 { 310 int res; 311 struct rt6_info *rt; 312 313 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 314 res = rt6_dump_route(rt, w->args); 315 if (res < 0) { 316 /* Frame is full, suspend walking */ 317 w->leaf = rt; 318 return 1; 319 } 320 } 321 w->leaf = NULL; 322 return 0; 323 } 324 325 static void fib6_dump_end(struct netlink_callback *cb) 326 { 327 struct net *net = sock_net(cb->skb->sk); 328 struct fib6_walker *w = (void *)cb->args[2]; 329 330 if (w) { 331 if (cb->args[4]) { 332 cb->args[4] = 0; 333 fib6_walker_unlink(net, w); 334 } 335 cb->args[2] = 0; 336 kfree(w); 337 } 338 cb->done = (void *)cb->args[3]; 339 cb->args[1] = 3; 340 } 341 342 static int fib6_dump_done(struct netlink_callback *cb) 343 { 344 fib6_dump_end(cb); 345 return cb->done ? cb->done(cb) : 0; 346 } 347 348 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 349 struct netlink_callback *cb) 350 { 351 struct net *net = sock_net(skb->sk); 352 struct fib6_walker *w; 353 int res; 354 355 w = (void *)cb->args[2]; 356 w->root = &table->tb6_root; 357 358 if (cb->args[4] == 0) { 359 w->count = 0; 360 w->skip = 0; 361 362 read_lock_bh(&table->tb6_lock); 363 res = fib6_walk(net, w); 364 read_unlock_bh(&table->tb6_lock); 365 if (res > 0) { 366 cb->args[4] = 1; 367 cb->args[5] = w->root->fn_sernum; 368 } 369 } else { 370 if (cb->args[5] != w->root->fn_sernum) { 371 /* Begin at the root if the tree changed */ 372 cb->args[5] = w->root->fn_sernum; 373 w->state = FWS_INIT; 374 w->node = w->root; 375 w->skip = w->count; 376 } else 377 w->skip = 0; 378 379 read_lock_bh(&table->tb6_lock); 380 res = fib6_walk_continue(w); 381 read_unlock_bh(&table->tb6_lock); 382 if (res <= 0) { 383 fib6_walker_unlink(net, w); 384 cb->args[4] = 0; 385 } 386 } 387 388 return res; 389 } 390 391 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 392 { 393 struct net *net = sock_net(skb->sk); 394 unsigned int h, s_h; 395 unsigned int e = 0, s_e; 396 struct rt6_rtnl_dump_arg arg; 397 struct fib6_walker *w; 398 struct fib6_table *tb; 399 struct hlist_head *head; 400 int res = 0; 401 402 s_h = cb->args[0]; 403 s_e = cb->args[1]; 404 405 w = (void *)cb->args[2]; 406 if (!w) { 407 /* New dump: 408 * 409 * 1. hook callback destructor. 410 */ 411 cb->args[3] = (long)cb->done; 412 cb->done = fib6_dump_done; 413 414 /* 415 * 2. allocate and initialize walker. 416 */ 417 w = kzalloc(sizeof(*w), GFP_ATOMIC); 418 if (!w) 419 return -ENOMEM; 420 w->func = fib6_dump_node; 421 cb->args[2] = (long)w; 422 } 423 424 arg.skb = skb; 425 arg.cb = cb; 426 arg.net = net; 427 w->args = &arg; 428 429 rcu_read_lock(); 430 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) { 431 e = 0; 432 head = &net->ipv6.fib_table_hash[h]; 433 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 434 if (e < s_e) 435 goto next; 436 res = fib6_dump_table(tb, skb, cb); 437 if (res != 0) 438 goto out; 439 next: 440 e++; 441 } 442 } 443 out: 444 rcu_read_unlock(); 445 cb->args[1] = e; 446 cb->args[0] = h; 447 448 res = res < 0 ? res : skb->len; 449 if (res <= 0) 450 fib6_dump_end(cb); 451 return res; 452 } 453 454 /* 455 * Routing Table 456 * 457 * return the appropriate node for a routing tree "add" operation 458 * by either creating and inserting or by returning an existing 459 * node. 460 */ 461 462 static struct fib6_node *fib6_add_1(struct fib6_node *root, 463 struct in6_addr *addr, int plen, 464 int offset, int allow_create, 465 int replace_required, int sernum) 466 { 467 struct fib6_node *fn, *in, *ln; 468 struct fib6_node *pn = NULL; 469 struct rt6key *key; 470 int bit; 471 __be32 dir = 0; 472 473 RT6_TRACE("fib6_add_1\n"); 474 475 /* insert node in tree */ 476 477 fn = root; 478 479 do { 480 key = (struct rt6key *)((u8 *)fn->leaf + offset); 481 482 /* 483 * Prefix match 484 */ 485 if (plen < fn->fn_bit || 486 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) { 487 if (!allow_create) { 488 if (replace_required) { 489 pr_warn("Can't replace route, no match found\n"); 490 return ERR_PTR(-ENOENT); 491 } 492 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 493 } 494 goto insert_above; 495 } 496 497 /* 498 * Exact match ? 499 */ 500 501 if (plen == fn->fn_bit) { 502 /* clean up an intermediate node */ 503 if (!(fn->fn_flags & RTN_RTINFO)) { 504 rt6_release(fn->leaf); 505 fn->leaf = NULL; 506 } 507 508 fn->fn_sernum = sernum; 509 510 return fn; 511 } 512 513 /* 514 * We have more bits to go 515 */ 516 517 /* Try to walk down on tree. */ 518 fn->fn_sernum = sernum; 519 dir = addr_bit_set(addr, fn->fn_bit); 520 pn = fn; 521 fn = dir ? fn->right : fn->left; 522 } while (fn); 523 524 if (!allow_create) { 525 /* We should not create new node because 526 * NLM_F_REPLACE was specified without NLM_F_CREATE 527 * I assume it is safe to require NLM_F_CREATE when 528 * REPLACE flag is used! Later we may want to remove the 529 * check for replace_required, because according 530 * to netlink specification, NLM_F_CREATE 531 * MUST be specified if new route is created. 532 * That would keep IPv6 consistent with IPv4 533 */ 534 if (replace_required) { 535 pr_warn("Can't replace route, no match found\n"); 536 return ERR_PTR(-ENOENT); 537 } 538 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 539 } 540 /* 541 * We walked to the bottom of tree. 542 * Create new leaf node without children. 543 */ 544 545 ln = node_alloc(); 546 547 if (!ln) 548 return ERR_PTR(-ENOMEM); 549 ln->fn_bit = plen; 550 551 ln->parent = pn; 552 ln->fn_sernum = sernum; 553 554 if (dir) 555 pn->right = ln; 556 else 557 pn->left = ln; 558 559 return ln; 560 561 562 insert_above: 563 /* 564 * split since we don't have a common prefix anymore or 565 * we have a less significant route. 566 * we've to insert an intermediate node on the list 567 * this new node will point to the one we need to create 568 * and the current 569 */ 570 571 pn = fn->parent; 572 573 /* find 1st bit in difference between the 2 addrs. 574 575 See comment in __ipv6_addr_diff: bit may be an invalid value, 576 but if it is >= plen, the value is ignored in any case. 577 */ 578 579 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr)); 580 581 /* 582 * (intermediate)[in] 583 * / \ 584 * (new leaf node)[ln] (old node)[fn] 585 */ 586 if (plen > bit) { 587 in = node_alloc(); 588 ln = node_alloc(); 589 590 if (!in || !ln) { 591 if (in) 592 node_free(in); 593 if (ln) 594 node_free(ln); 595 return ERR_PTR(-ENOMEM); 596 } 597 598 /* 599 * new intermediate node. 600 * RTN_RTINFO will 601 * be off since that an address that chooses one of 602 * the branches would not match less specific routes 603 * in the other branch 604 */ 605 606 in->fn_bit = bit; 607 608 in->parent = pn; 609 in->leaf = fn->leaf; 610 atomic_inc(&in->leaf->rt6i_ref); 611 612 in->fn_sernum = sernum; 613 614 /* update parent pointer */ 615 if (dir) 616 pn->right = in; 617 else 618 pn->left = in; 619 620 ln->fn_bit = plen; 621 622 ln->parent = in; 623 fn->parent = in; 624 625 ln->fn_sernum = sernum; 626 627 if (addr_bit_set(addr, bit)) { 628 in->right = ln; 629 in->left = fn; 630 } else { 631 in->left = ln; 632 in->right = fn; 633 } 634 } else { /* plen <= bit */ 635 636 /* 637 * (new leaf node)[ln] 638 * / \ 639 * (old node)[fn] NULL 640 */ 641 642 ln = node_alloc(); 643 644 if (!ln) 645 return ERR_PTR(-ENOMEM); 646 647 ln->fn_bit = plen; 648 649 ln->parent = pn; 650 651 ln->fn_sernum = sernum; 652 653 if (dir) 654 pn->right = ln; 655 else 656 pn->left = ln; 657 658 if (addr_bit_set(&key->addr, plen)) 659 ln->right = fn; 660 else 661 ln->left = fn; 662 663 fn->parent = ln; 664 } 665 return ln; 666 } 667 668 static bool rt6_qualify_for_ecmp(struct rt6_info *rt) 669 { 670 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) == 671 RTF_GATEWAY; 672 } 673 674 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc) 675 { 676 int i; 677 678 for (i = 0; i < RTAX_MAX; i++) { 679 if (test_bit(i, mxc->mx_valid)) 680 mp[i] = mxc->mx[i]; 681 } 682 } 683 684 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc) 685 { 686 if (!mxc->mx) 687 return 0; 688 689 if (dst->flags & DST_HOST) { 690 u32 *mp = dst_metrics_write_ptr(dst); 691 692 if (unlikely(!mp)) 693 return -ENOMEM; 694 695 fib6_copy_metrics(mp, mxc); 696 } else { 697 dst_init_metrics(dst, mxc->mx, false); 698 699 /* We've stolen mx now. */ 700 mxc->mx = NULL; 701 } 702 703 return 0; 704 } 705 706 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn, 707 struct net *net) 708 { 709 if (atomic_read(&rt->rt6i_ref) != 1) { 710 /* This route is used as dummy address holder in some split 711 * nodes. It is not leaked, but it still holds other resources, 712 * which must be released in time. So, scan ascendant nodes 713 * and replace dummy references to this route with references 714 * to still alive ones. 715 */ 716 while (fn) { 717 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) { 718 fn->leaf = fib6_find_prefix(net, fn); 719 atomic_inc(&fn->leaf->rt6i_ref); 720 rt6_release(rt); 721 } 722 fn = fn->parent; 723 } 724 /* No more references are possible at this point. */ 725 BUG_ON(atomic_read(&rt->rt6i_ref) != 1); 726 } 727 } 728 729 /* 730 * Insert routing information in a node. 731 */ 732 733 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, 734 struct nl_info *info, struct mx6_config *mxc) 735 { 736 struct rt6_info *iter = NULL; 737 struct rt6_info **ins; 738 struct rt6_info **fallback_ins = NULL; 739 int replace = (info->nlh && 740 (info->nlh->nlmsg_flags & NLM_F_REPLACE)); 741 int add = (!info->nlh || 742 (info->nlh->nlmsg_flags & NLM_F_CREATE)); 743 int found = 0; 744 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt); 745 int err; 746 747 ins = &fn->leaf; 748 749 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) { 750 /* 751 * Search for duplicates 752 */ 753 754 if (iter->rt6i_metric == rt->rt6i_metric) { 755 /* 756 * Same priority level 757 */ 758 if (info->nlh && 759 (info->nlh->nlmsg_flags & NLM_F_EXCL)) 760 return -EEXIST; 761 if (replace) { 762 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) { 763 found++; 764 break; 765 } 766 if (rt_can_ecmp) 767 fallback_ins = fallback_ins ?: ins; 768 goto next_iter; 769 } 770 771 if (iter->dst.dev == rt->dst.dev && 772 iter->rt6i_idev == rt->rt6i_idev && 773 ipv6_addr_equal(&iter->rt6i_gateway, 774 &rt->rt6i_gateway)) { 775 if (rt->rt6i_nsiblings) 776 rt->rt6i_nsiblings = 0; 777 if (!(iter->rt6i_flags & RTF_EXPIRES)) 778 return -EEXIST; 779 if (!(rt->rt6i_flags & RTF_EXPIRES)) 780 rt6_clean_expires(iter); 781 else 782 rt6_set_expires(iter, rt->dst.expires); 783 iter->rt6i_pmtu = rt->rt6i_pmtu; 784 return -EEXIST; 785 } 786 /* If we have the same destination and the same metric, 787 * but not the same gateway, then the route we try to 788 * add is sibling to this route, increment our counter 789 * of siblings, and later we will add our route to the 790 * list. 791 * Only static routes (which don't have flag 792 * RTF_EXPIRES) are used for ECMPv6. 793 * 794 * To avoid long list, we only had siblings if the 795 * route have a gateway. 796 */ 797 if (rt_can_ecmp && 798 rt6_qualify_for_ecmp(iter)) 799 rt->rt6i_nsiblings++; 800 } 801 802 if (iter->rt6i_metric > rt->rt6i_metric) 803 break; 804 805 next_iter: 806 ins = &iter->dst.rt6_next; 807 } 808 809 if (fallback_ins && !found) { 810 /* No ECMP-able route found, replace first non-ECMP one */ 811 ins = fallback_ins; 812 iter = *ins; 813 found++; 814 } 815 816 /* Reset round-robin state, if necessary */ 817 if (ins == &fn->leaf) 818 fn->rr_ptr = NULL; 819 820 /* Link this route to others same route. */ 821 if (rt->rt6i_nsiblings) { 822 unsigned int rt6i_nsiblings; 823 struct rt6_info *sibling, *temp_sibling; 824 825 /* Find the first route that have the same metric */ 826 sibling = fn->leaf; 827 while (sibling) { 828 if (sibling->rt6i_metric == rt->rt6i_metric && 829 rt6_qualify_for_ecmp(sibling)) { 830 list_add_tail(&rt->rt6i_siblings, 831 &sibling->rt6i_siblings); 832 break; 833 } 834 sibling = sibling->dst.rt6_next; 835 } 836 /* For each sibling in the list, increment the counter of 837 * siblings. BUG() if counters does not match, list of siblings 838 * is broken! 839 */ 840 rt6i_nsiblings = 0; 841 list_for_each_entry_safe(sibling, temp_sibling, 842 &rt->rt6i_siblings, rt6i_siblings) { 843 sibling->rt6i_nsiblings++; 844 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings); 845 rt6i_nsiblings++; 846 } 847 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings); 848 } 849 850 /* 851 * insert node 852 */ 853 if (!replace) { 854 if (!add) 855 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 856 857 add: 858 err = fib6_commit_metrics(&rt->dst, mxc); 859 if (err) 860 return err; 861 862 rt->dst.rt6_next = iter; 863 *ins = rt; 864 rt->rt6i_node = fn; 865 atomic_inc(&rt->rt6i_ref); 866 inet6_rt_notify(RTM_NEWROUTE, rt, info, 0); 867 info->nl_net->ipv6.rt6_stats->fib_rt_entries++; 868 869 if (!(fn->fn_flags & RTN_RTINFO)) { 870 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 871 fn->fn_flags |= RTN_RTINFO; 872 } 873 874 } else { 875 int nsiblings; 876 877 if (!found) { 878 if (add) 879 goto add; 880 pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); 881 return -ENOENT; 882 } 883 884 err = fib6_commit_metrics(&rt->dst, mxc); 885 if (err) 886 return err; 887 888 *ins = rt; 889 rt->rt6i_node = fn; 890 rt->dst.rt6_next = iter->dst.rt6_next; 891 atomic_inc(&rt->rt6i_ref); 892 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE); 893 if (!(fn->fn_flags & RTN_RTINFO)) { 894 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 895 fn->fn_flags |= RTN_RTINFO; 896 } 897 nsiblings = iter->rt6i_nsiblings; 898 fib6_purge_rt(iter, fn, info->nl_net); 899 rt6_release(iter); 900 901 if (nsiblings) { 902 /* Replacing an ECMP route, remove all siblings */ 903 ins = &rt->dst.rt6_next; 904 iter = *ins; 905 while (iter) { 906 if (rt6_qualify_for_ecmp(iter)) { 907 *ins = iter->dst.rt6_next; 908 fib6_purge_rt(iter, fn, info->nl_net); 909 rt6_release(iter); 910 nsiblings--; 911 } else { 912 ins = &iter->dst.rt6_next; 913 } 914 iter = *ins; 915 } 916 WARN_ON(nsiblings != 0); 917 } 918 } 919 920 return 0; 921 } 922 923 static void fib6_start_gc(struct net *net, struct rt6_info *rt) 924 { 925 if (!timer_pending(&net->ipv6.ip6_fib_timer) && 926 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE))) 927 mod_timer(&net->ipv6.ip6_fib_timer, 928 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 929 } 930 931 void fib6_force_start_gc(struct net *net) 932 { 933 if (!timer_pending(&net->ipv6.ip6_fib_timer)) 934 mod_timer(&net->ipv6.ip6_fib_timer, 935 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 936 } 937 938 /* 939 * Add routing information to the routing tree. 940 * <destination addr>/<source addr> 941 * with source addr info in sub-trees 942 */ 943 944 int fib6_add(struct fib6_node *root, struct rt6_info *rt, 945 struct nl_info *info, struct mx6_config *mxc) 946 { 947 struct fib6_node *fn, *pn = NULL; 948 int err = -ENOMEM; 949 int allow_create = 1; 950 int replace_required = 0; 951 int sernum = fib6_new_sernum(info->nl_net); 952 953 if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) && 954 !atomic_read(&rt->dst.__refcnt))) 955 return -EINVAL; 956 957 if (info->nlh) { 958 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) 959 allow_create = 0; 960 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) 961 replace_required = 1; 962 } 963 if (!allow_create && !replace_required) 964 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); 965 966 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen, 967 offsetof(struct rt6_info, rt6i_dst), allow_create, 968 replace_required, sernum); 969 if (IS_ERR(fn)) { 970 err = PTR_ERR(fn); 971 fn = NULL; 972 goto out; 973 } 974 975 pn = fn; 976 977 #ifdef CONFIG_IPV6_SUBTREES 978 if (rt->rt6i_src.plen) { 979 struct fib6_node *sn; 980 981 if (!fn->subtree) { 982 struct fib6_node *sfn; 983 984 /* 985 * Create subtree. 986 * 987 * fn[main tree] 988 * | 989 * sfn[subtree root] 990 * \ 991 * sn[new leaf node] 992 */ 993 994 /* Create subtree root node */ 995 sfn = node_alloc(); 996 if (!sfn) 997 goto st_failure; 998 999 sfn->leaf = info->nl_net->ipv6.ip6_null_entry; 1000 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref); 1001 sfn->fn_flags = RTN_ROOT; 1002 sfn->fn_sernum = sernum; 1003 1004 /* Now add the first leaf node to new subtree */ 1005 1006 sn = fib6_add_1(sfn, &rt->rt6i_src.addr, 1007 rt->rt6i_src.plen, 1008 offsetof(struct rt6_info, rt6i_src), 1009 allow_create, replace_required, sernum); 1010 1011 if (IS_ERR(sn)) { 1012 /* If it is failed, discard just allocated 1013 root, and then (in st_failure) stale node 1014 in main tree. 1015 */ 1016 node_free(sfn); 1017 err = PTR_ERR(sn); 1018 goto st_failure; 1019 } 1020 1021 /* Now link new subtree to main tree */ 1022 sfn->parent = fn; 1023 fn->subtree = sfn; 1024 } else { 1025 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, 1026 rt->rt6i_src.plen, 1027 offsetof(struct rt6_info, rt6i_src), 1028 allow_create, replace_required, sernum); 1029 1030 if (IS_ERR(sn)) { 1031 err = PTR_ERR(sn); 1032 goto st_failure; 1033 } 1034 } 1035 1036 if (!fn->leaf) { 1037 fn->leaf = rt; 1038 atomic_inc(&rt->rt6i_ref); 1039 } 1040 fn = sn; 1041 } 1042 #endif 1043 1044 err = fib6_add_rt2node(fn, rt, info, mxc); 1045 if (!err) { 1046 fib6_start_gc(info->nl_net, rt); 1047 if (!(rt->rt6i_flags & RTF_CACHE)) 1048 fib6_prune_clones(info->nl_net, pn); 1049 rt->dst.flags &= ~DST_NOCACHE; 1050 } 1051 1052 out: 1053 if (err) { 1054 #ifdef CONFIG_IPV6_SUBTREES 1055 /* 1056 * If fib6_add_1 has cleared the old leaf pointer in the 1057 * super-tree leaf node we have to find a new one for it. 1058 */ 1059 if (pn != fn && pn->leaf == rt) { 1060 pn->leaf = NULL; 1061 atomic_dec(&rt->rt6i_ref); 1062 } 1063 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) { 1064 pn->leaf = fib6_find_prefix(info->nl_net, pn); 1065 #if RT6_DEBUG >= 2 1066 if (!pn->leaf) { 1067 WARN_ON(pn->leaf == NULL); 1068 pn->leaf = info->nl_net->ipv6.ip6_null_entry; 1069 } 1070 #endif 1071 atomic_inc(&pn->leaf->rt6i_ref); 1072 } 1073 #endif 1074 if (!(rt->dst.flags & DST_NOCACHE)) 1075 dst_free(&rt->dst); 1076 } 1077 return err; 1078 1079 #ifdef CONFIG_IPV6_SUBTREES 1080 /* Subtree creation failed, probably main tree node 1081 is orphan. If it is, shoot it. 1082 */ 1083 st_failure: 1084 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) 1085 fib6_repair_tree(info->nl_net, fn); 1086 if (!(rt->dst.flags & DST_NOCACHE)) 1087 dst_free(&rt->dst); 1088 return err; 1089 #endif 1090 } 1091 1092 /* 1093 * Routing tree lookup 1094 * 1095 */ 1096 1097 struct lookup_args { 1098 int offset; /* key offset on rt6_info */ 1099 const struct in6_addr *addr; /* search key */ 1100 }; 1101 1102 static struct fib6_node *fib6_lookup_1(struct fib6_node *root, 1103 struct lookup_args *args) 1104 { 1105 struct fib6_node *fn; 1106 __be32 dir; 1107 1108 if (unlikely(args->offset == 0)) 1109 return NULL; 1110 1111 /* 1112 * Descend on a tree 1113 */ 1114 1115 fn = root; 1116 1117 for (;;) { 1118 struct fib6_node *next; 1119 1120 dir = addr_bit_set(args->addr, fn->fn_bit); 1121 1122 next = dir ? fn->right : fn->left; 1123 1124 if (next) { 1125 fn = next; 1126 continue; 1127 } 1128 break; 1129 } 1130 1131 while (fn) { 1132 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) { 1133 struct rt6key *key; 1134 1135 key = (struct rt6key *) ((u8 *) fn->leaf + 1136 args->offset); 1137 1138 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 1139 #ifdef CONFIG_IPV6_SUBTREES 1140 if (fn->subtree) { 1141 struct fib6_node *sfn; 1142 sfn = fib6_lookup_1(fn->subtree, 1143 args + 1); 1144 if (!sfn) 1145 goto backtrack; 1146 fn = sfn; 1147 } 1148 #endif 1149 if (fn->fn_flags & RTN_RTINFO) 1150 return fn; 1151 } 1152 } 1153 #ifdef CONFIG_IPV6_SUBTREES 1154 backtrack: 1155 #endif 1156 if (fn->fn_flags & RTN_ROOT) 1157 break; 1158 1159 fn = fn->parent; 1160 } 1161 1162 return NULL; 1163 } 1164 1165 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr, 1166 const struct in6_addr *saddr) 1167 { 1168 struct fib6_node *fn; 1169 struct lookup_args args[] = { 1170 { 1171 .offset = offsetof(struct rt6_info, rt6i_dst), 1172 .addr = daddr, 1173 }, 1174 #ifdef CONFIG_IPV6_SUBTREES 1175 { 1176 .offset = offsetof(struct rt6_info, rt6i_src), 1177 .addr = saddr, 1178 }, 1179 #endif 1180 { 1181 .offset = 0, /* sentinel */ 1182 } 1183 }; 1184 1185 fn = fib6_lookup_1(root, daddr ? args : args + 1); 1186 if (!fn || fn->fn_flags & RTN_TL_ROOT) 1187 fn = root; 1188 1189 return fn; 1190 } 1191 1192 /* 1193 * Get node with specified destination prefix (and source prefix, 1194 * if subtrees are used) 1195 */ 1196 1197 1198 static struct fib6_node *fib6_locate_1(struct fib6_node *root, 1199 const struct in6_addr *addr, 1200 int plen, int offset) 1201 { 1202 struct fib6_node *fn; 1203 1204 for (fn = root; fn ; ) { 1205 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); 1206 1207 /* 1208 * Prefix match 1209 */ 1210 if (plen < fn->fn_bit || 1211 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 1212 return NULL; 1213 1214 if (plen == fn->fn_bit) 1215 return fn; 1216 1217 /* 1218 * We have more bits to go 1219 */ 1220 if (addr_bit_set(addr, fn->fn_bit)) 1221 fn = fn->right; 1222 else 1223 fn = fn->left; 1224 } 1225 return NULL; 1226 } 1227 1228 struct fib6_node *fib6_locate(struct fib6_node *root, 1229 const struct in6_addr *daddr, int dst_len, 1230 const struct in6_addr *saddr, int src_len) 1231 { 1232 struct fib6_node *fn; 1233 1234 fn = fib6_locate_1(root, daddr, dst_len, 1235 offsetof(struct rt6_info, rt6i_dst)); 1236 1237 #ifdef CONFIG_IPV6_SUBTREES 1238 if (src_len) { 1239 WARN_ON(saddr == NULL); 1240 if (fn && fn->subtree) 1241 fn = fib6_locate_1(fn->subtree, saddr, src_len, 1242 offsetof(struct rt6_info, rt6i_src)); 1243 } 1244 #endif 1245 1246 if (fn && fn->fn_flags & RTN_RTINFO) 1247 return fn; 1248 1249 return NULL; 1250 } 1251 1252 1253 /* 1254 * Deletion 1255 * 1256 */ 1257 1258 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn) 1259 { 1260 if (fn->fn_flags & RTN_ROOT) 1261 return net->ipv6.ip6_null_entry; 1262 1263 while (fn) { 1264 if (fn->left) 1265 return fn->left->leaf; 1266 if (fn->right) 1267 return fn->right->leaf; 1268 1269 fn = FIB6_SUBTREE(fn); 1270 } 1271 return NULL; 1272 } 1273 1274 /* 1275 * Called to trim the tree of intermediate nodes when possible. "fn" 1276 * is the node we want to try and remove. 1277 */ 1278 1279 static struct fib6_node *fib6_repair_tree(struct net *net, 1280 struct fib6_node *fn) 1281 { 1282 int children; 1283 int nstate; 1284 struct fib6_node *child, *pn; 1285 struct fib6_walker *w; 1286 int iter = 0; 1287 1288 for (;;) { 1289 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1290 iter++; 1291 1292 WARN_ON(fn->fn_flags & RTN_RTINFO); 1293 WARN_ON(fn->fn_flags & RTN_TL_ROOT); 1294 WARN_ON(fn->leaf); 1295 1296 children = 0; 1297 child = NULL; 1298 if (fn->right) 1299 child = fn->right, children |= 1; 1300 if (fn->left) 1301 child = fn->left, children |= 2; 1302 1303 if (children == 3 || FIB6_SUBTREE(fn) 1304 #ifdef CONFIG_IPV6_SUBTREES 1305 /* Subtree root (i.e. fn) may have one child */ 1306 || (children && fn->fn_flags & RTN_ROOT) 1307 #endif 1308 ) { 1309 fn->leaf = fib6_find_prefix(net, fn); 1310 #if RT6_DEBUG >= 2 1311 if (!fn->leaf) { 1312 WARN_ON(!fn->leaf); 1313 fn->leaf = net->ipv6.ip6_null_entry; 1314 } 1315 #endif 1316 atomic_inc(&fn->leaf->rt6i_ref); 1317 return fn->parent; 1318 } 1319 1320 pn = fn->parent; 1321 #ifdef CONFIG_IPV6_SUBTREES 1322 if (FIB6_SUBTREE(pn) == fn) { 1323 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1324 FIB6_SUBTREE(pn) = NULL; 1325 nstate = FWS_L; 1326 } else { 1327 WARN_ON(fn->fn_flags & RTN_ROOT); 1328 #endif 1329 if (pn->right == fn) 1330 pn->right = child; 1331 else if (pn->left == fn) 1332 pn->left = child; 1333 #if RT6_DEBUG >= 2 1334 else 1335 WARN_ON(1); 1336 #endif 1337 if (child) 1338 child->parent = pn; 1339 nstate = FWS_R; 1340 #ifdef CONFIG_IPV6_SUBTREES 1341 } 1342 #endif 1343 1344 read_lock(&net->ipv6.fib6_walker_lock); 1345 FOR_WALKERS(net, w) { 1346 if (!child) { 1347 if (w->root == fn) { 1348 w->root = w->node = NULL; 1349 RT6_TRACE("W %p adjusted by delroot 1\n", w); 1350 } else if (w->node == fn) { 1351 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1352 w->node = pn; 1353 w->state = nstate; 1354 } 1355 } else { 1356 if (w->root == fn) { 1357 w->root = child; 1358 RT6_TRACE("W %p adjusted by delroot 2\n", w); 1359 } 1360 if (w->node == fn) { 1361 w->node = child; 1362 if (children&2) { 1363 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1364 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT; 1365 } else { 1366 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1367 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT; 1368 } 1369 } 1370 } 1371 } 1372 read_unlock(&net->ipv6.fib6_walker_lock); 1373 1374 node_free(fn); 1375 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) 1376 return pn; 1377 1378 rt6_release(pn->leaf); 1379 pn->leaf = NULL; 1380 fn = pn; 1381 } 1382 } 1383 1384 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, 1385 struct nl_info *info) 1386 { 1387 struct fib6_walker *w; 1388 struct rt6_info *rt = *rtp; 1389 struct net *net = info->nl_net; 1390 1391 RT6_TRACE("fib6_del_route\n"); 1392 1393 /* Unlink it */ 1394 *rtp = rt->dst.rt6_next; 1395 rt->rt6i_node = NULL; 1396 net->ipv6.rt6_stats->fib_rt_entries--; 1397 net->ipv6.rt6_stats->fib_discarded_routes++; 1398 1399 /* Reset round-robin state, if necessary */ 1400 if (fn->rr_ptr == rt) 1401 fn->rr_ptr = NULL; 1402 1403 /* Remove this entry from other siblings */ 1404 if (rt->rt6i_nsiblings) { 1405 struct rt6_info *sibling, *next_sibling; 1406 1407 list_for_each_entry_safe(sibling, next_sibling, 1408 &rt->rt6i_siblings, rt6i_siblings) 1409 sibling->rt6i_nsiblings--; 1410 rt->rt6i_nsiblings = 0; 1411 list_del_init(&rt->rt6i_siblings); 1412 } 1413 1414 /* Adjust walkers */ 1415 read_lock(&net->ipv6.fib6_walker_lock); 1416 FOR_WALKERS(net, w) { 1417 if (w->state == FWS_C && w->leaf == rt) { 1418 RT6_TRACE("walker %p adjusted by delroute\n", w); 1419 w->leaf = rt->dst.rt6_next; 1420 if (!w->leaf) 1421 w->state = FWS_U; 1422 } 1423 } 1424 read_unlock(&net->ipv6.fib6_walker_lock); 1425 1426 rt->dst.rt6_next = NULL; 1427 1428 /* If it was last route, expunge its radix tree node */ 1429 if (!fn->leaf) { 1430 fn->fn_flags &= ~RTN_RTINFO; 1431 net->ipv6.rt6_stats->fib_route_nodes--; 1432 fn = fib6_repair_tree(net, fn); 1433 } 1434 1435 fib6_purge_rt(rt, fn, net); 1436 1437 inet6_rt_notify(RTM_DELROUTE, rt, info, 0); 1438 rt6_release(rt); 1439 } 1440 1441 int fib6_del(struct rt6_info *rt, struct nl_info *info) 1442 { 1443 struct net *net = info->nl_net; 1444 struct fib6_node *fn = rt->rt6i_node; 1445 struct rt6_info **rtp; 1446 1447 #if RT6_DEBUG >= 2 1448 if (rt->dst.obsolete > 0) { 1449 WARN_ON(fn); 1450 return -ENOENT; 1451 } 1452 #endif 1453 if (!fn || rt == net->ipv6.ip6_null_entry) 1454 return -ENOENT; 1455 1456 WARN_ON(!(fn->fn_flags & RTN_RTINFO)); 1457 1458 if (!(rt->rt6i_flags & RTF_CACHE)) { 1459 struct fib6_node *pn = fn; 1460 #ifdef CONFIG_IPV6_SUBTREES 1461 /* clones of this route might be in another subtree */ 1462 if (rt->rt6i_src.plen) { 1463 while (!(pn->fn_flags & RTN_ROOT)) 1464 pn = pn->parent; 1465 pn = pn->parent; 1466 } 1467 #endif 1468 fib6_prune_clones(info->nl_net, pn); 1469 } 1470 1471 /* 1472 * Walk the leaf entries looking for ourself 1473 */ 1474 1475 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) { 1476 if (*rtp == rt) { 1477 fib6_del_route(fn, rtp, info); 1478 return 0; 1479 } 1480 } 1481 return -ENOENT; 1482 } 1483 1484 /* 1485 * Tree traversal function. 1486 * 1487 * Certainly, it is not interrupt safe. 1488 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 1489 * It means, that we can modify tree during walking 1490 * and use this function for garbage collection, clone pruning, 1491 * cleaning tree when a device goes down etc. etc. 1492 * 1493 * It guarantees that every node will be traversed, 1494 * and that it will be traversed only once. 1495 * 1496 * Callback function w->func may return: 1497 * 0 -> continue walking. 1498 * positive value -> walking is suspended (used by tree dumps, 1499 * and probably by gc, if it will be split to several slices) 1500 * negative value -> terminate walking. 1501 * 1502 * The function itself returns: 1503 * 0 -> walk is complete. 1504 * >0 -> walk is incomplete (i.e. suspended) 1505 * <0 -> walk is terminated by an error. 1506 */ 1507 1508 static int fib6_walk_continue(struct fib6_walker *w) 1509 { 1510 struct fib6_node *fn, *pn; 1511 1512 for (;;) { 1513 fn = w->node; 1514 if (!fn) 1515 return 0; 1516 1517 if (w->prune && fn != w->root && 1518 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) { 1519 w->state = FWS_C; 1520 w->leaf = fn->leaf; 1521 } 1522 switch (w->state) { 1523 #ifdef CONFIG_IPV6_SUBTREES 1524 case FWS_S: 1525 if (FIB6_SUBTREE(fn)) { 1526 w->node = FIB6_SUBTREE(fn); 1527 continue; 1528 } 1529 w->state = FWS_L; 1530 #endif 1531 case FWS_L: 1532 if (fn->left) { 1533 w->node = fn->left; 1534 w->state = FWS_INIT; 1535 continue; 1536 } 1537 w->state = FWS_R; 1538 case FWS_R: 1539 if (fn->right) { 1540 w->node = fn->right; 1541 w->state = FWS_INIT; 1542 continue; 1543 } 1544 w->state = FWS_C; 1545 w->leaf = fn->leaf; 1546 case FWS_C: 1547 if (w->leaf && fn->fn_flags & RTN_RTINFO) { 1548 int err; 1549 1550 if (w->skip) { 1551 w->skip--; 1552 goto skip; 1553 } 1554 1555 err = w->func(w); 1556 if (err) 1557 return err; 1558 1559 w->count++; 1560 continue; 1561 } 1562 skip: 1563 w->state = FWS_U; 1564 case FWS_U: 1565 if (fn == w->root) 1566 return 0; 1567 pn = fn->parent; 1568 w->node = pn; 1569 #ifdef CONFIG_IPV6_SUBTREES 1570 if (FIB6_SUBTREE(pn) == fn) { 1571 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1572 w->state = FWS_L; 1573 continue; 1574 } 1575 #endif 1576 if (pn->left == fn) { 1577 w->state = FWS_R; 1578 continue; 1579 } 1580 if (pn->right == fn) { 1581 w->state = FWS_C; 1582 w->leaf = w->node->leaf; 1583 continue; 1584 } 1585 #if RT6_DEBUG >= 2 1586 WARN_ON(1); 1587 #endif 1588 } 1589 } 1590 } 1591 1592 static int fib6_walk(struct net *net, struct fib6_walker *w) 1593 { 1594 int res; 1595 1596 w->state = FWS_INIT; 1597 w->node = w->root; 1598 1599 fib6_walker_link(net, w); 1600 res = fib6_walk_continue(w); 1601 if (res <= 0) 1602 fib6_walker_unlink(net, w); 1603 return res; 1604 } 1605 1606 static int fib6_clean_node(struct fib6_walker *w) 1607 { 1608 int res; 1609 struct rt6_info *rt; 1610 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w); 1611 struct nl_info info = { 1612 .nl_net = c->net, 1613 }; 1614 1615 if (c->sernum != FIB6_NO_SERNUM_CHANGE && 1616 w->node->fn_sernum != c->sernum) 1617 w->node->fn_sernum = c->sernum; 1618 1619 if (!c->func) { 1620 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE); 1621 w->leaf = NULL; 1622 return 0; 1623 } 1624 1625 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 1626 res = c->func(rt, c->arg); 1627 if (res < 0) { 1628 w->leaf = rt; 1629 res = fib6_del(rt, &info); 1630 if (res) { 1631 #if RT6_DEBUG >= 2 1632 pr_debug("%s: del failed: rt=%p@%p err=%d\n", 1633 __func__, rt, rt->rt6i_node, res); 1634 #endif 1635 continue; 1636 } 1637 return 0; 1638 } 1639 WARN_ON(res != 0); 1640 } 1641 w->leaf = rt; 1642 return 0; 1643 } 1644 1645 /* 1646 * Convenient frontend to tree walker. 1647 * 1648 * func is called on each route. 1649 * It may return -1 -> delete this route. 1650 * 0 -> continue walking 1651 * 1652 * prune==1 -> only immediate children of node (certainly, 1653 * ignoring pure split nodes) will be scanned. 1654 */ 1655 1656 static void fib6_clean_tree(struct net *net, struct fib6_node *root, 1657 int (*func)(struct rt6_info *, void *arg), 1658 bool prune, int sernum, void *arg) 1659 { 1660 struct fib6_cleaner c; 1661 1662 c.w.root = root; 1663 c.w.func = fib6_clean_node; 1664 c.w.prune = prune; 1665 c.w.count = 0; 1666 c.w.skip = 0; 1667 c.func = func; 1668 c.sernum = sernum; 1669 c.arg = arg; 1670 c.net = net; 1671 1672 fib6_walk(net, &c.w); 1673 } 1674 1675 static void __fib6_clean_all(struct net *net, 1676 int (*func)(struct rt6_info *, void *), 1677 int sernum, void *arg) 1678 { 1679 struct fib6_table *table; 1680 struct hlist_head *head; 1681 unsigned int h; 1682 1683 rcu_read_lock(); 1684 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 1685 head = &net->ipv6.fib_table_hash[h]; 1686 hlist_for_each_entry_rcu(table, head, tb6_hlist) { 1687 write_lock_bh(&table->tb6_lock); 1688 fib6_clean_tree(net, &table->tb6_root, 1689 func, false, sernum, arg); 1690 write_unlock_bh(&table->tb6_lock); 1691 } 1692 } 1693 rcu_read_unlock(); 1694 } 1695 1696 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *), 1697 void *arg) 1698 { 1699 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg); 1700 } 1701 1702 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1703 { 1704 if (rt->rt6i_flags & RTF_CACHE) { 1705 RT6_TRACE("pruning clone %p\n", rt); 1706 return -1; 1707 } 1708 1709 return 0; 1710 } 1711 1712 static void fib6_prune_clones(struct net *net, struct fib6_node *fn) 1713 { 1714 fib6_clean_tree(net, fn, fib6_prune_clone, true, 1715 FIB6_NO_SERNUM_CHANGE, NULL); 1716 } 1717 1718 static void fib6_flush_trees(struct net *net) 1719 { 1720 int new_sernum = fib6_new_sernum(net); 1721 1722 __fib6_clean_all(net, NULL, new_sernum, NULL); 1723 } 1724 1725 /* 1726 * Garbage collection 1727 */ 1728 1729 struct fib6_gc_args 1730 { 1731 int timeout; 1732 int more; 1733 }; 1734 1735 static int fib6_age(struct rt6_info *rt, void *arg) 1736 { 1737 struct fib6_gc_args *gc_args = arg; 1738 unsigned long now = jiffies; 1739 1740 /* 1741 * check addrconf expiration here. 1742 * Routes are expired even if they are in use. 1743 * 1744 * Also age clones. Note, that clones are aged out 1745 * only if they are not in use now. 1746 */ 1747 1748 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) { 1749 if (time_after(now, rt->dst.expires)) { 1750 RT6_TRACE("expiring %p\n", rt); 1751 return -1; 1752 } 1753 gc_args->more++; 1754 } else if (rt->rt6i_flags & RTF_CACHE) { 1755 if (atomic_read(&rt->dst.__refcnt) == 0 && 1756 time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) { 1757 RT6_TRACE("aging clone %p\n", rt); 1758 return -1; 1759 } else if (rt->rt6i_flags & RTF_GATEWAY) { 1760 struct neighbour *neigh; 1761 __u8 neigh_flags = 0; 1762 1763 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway); 1764 if (neigh) { 1765 neigh_flags = neigh->flags; 1766 neigh_release(neigh); 1767 } 1768 if (!(neigh_flags & NTF_ROUTER)) { 1769 RT6_TRACE("purging route %p via non-router but gateway\n", 1770 rt); 1771 return -1; 1772 } 1773 } 1774 gc_args->more++; 1775 } 1776 1777 return 0; 1778 } 1779 1780 void fib6_run_gc(unsigned long expires, struct net *net, bool force) 1781 { 1782 struct fib6_gc_args gc_args; 1783 unsigned long now; 1784 1785 if (force) { 1786 spin_lock_bh(&net->ipv6.fib6_gc_lock); 1787 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) { 1788 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); 1789 return; 1790 } 1791 gc_args.timeout = expires ? (int)expires : 1792 net->ipv6.sysctl.ip6_rt_gc_interval; 1793 1794 gc_args.more = icmp6_dst_gc(); 1795 1796 fib6_clean_all(net, fib6_age, &gc_args); 1797 now = jiffies; 1798 net->ipv6.ip6_rt_last_gc = now; 1799 1800 if (gc_args.more) 1801 mod_timer(&net->ipv6.ip6_fib_timer, 1802 round_jiffies(now 1803 + net->ipv6.sysctl.ip6_rt_gc_interval)); 1804 else 1805 del_timer(&net->ipv6.ip6_fib_timer); 1806 spin_unlock_bh(&net->ipv6.fib6_gc_lock); 1807 } 1808 1809 static void fib6_gc_timer_cb(unsigned long arg) 1810 { 1811 fib6_run_gc(0, (struct net *)arg, true); 1812 } 1813 1814 static int __net_init fib6_net_init(struct net *net) 1815 { 1816 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; 1817 1818 spin_lock_init(&net->ipv6.fib6_gc_lock); 1819 rwlock_init(&net->ipv6.fib6_walker_lock); 1820 INIT_LIST_HEAD(&net->ipv6.fib6_walkers); 1821 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net); 1822 1823 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); 1824 if (!net->ipv6.rt6_stats) 1825 goto out_timer; 1826 1827 /* Avoid false sharing : Use at least a full cache line */ 1828 size = max_t(size_t, size, L1_CACHE_BYTES); 1829 1830 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); 1831 if (!net->ipv6.fib_table_hash) 1832 goto out_rt6_stats; 1833 1834 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), 1835 GFP_KERNEL); 1836 if (!net->ipv6.fib6_main_tbl) 1837 goto out_fib_table_hash; 1838 1839 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; 1840 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1841 net->ipv6.fib6_main_tbl->tb6_root.fn_flags = 1842 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1843 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); 1844 1845 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1846 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), 1847 GFP_KERNEL); 1848 if (!net->ipv6.fib6_local_tbl) 1849 goto out_fib6_main_tbl; 1850 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; 1851 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1852 net->ipv6.fib6_local_tbl->tb6_root.fn_flags = 1853 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1854 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); 1855 #endif 1856 fib6_tables_init(net); 1857 1858 return 0; 1859 1860 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1861 out_fib6_main_tbl: 1862 kfree(net->ipv6.fib6_main_tbl); 1863 #endif 1864 out_fib_table_hash: 1865 kfree(net->ipv6.fib_table_hash); 1866 out_rt6_stats: 1867 kfree(net->ipv6.rt6_stats); 1868 out_timer: 1869 return -ENOMEM; 1870 } 1871 1872 static void fib6_net_exit(struct net *net) 1873 { 1874 rt6_ifdown(net, NULL); 1875 del_timer_sync(&net->ipv6.ip6_fib_timer); 1876 1877 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1878 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers); 1879 kfree(net->ipv6.fib6_local_tbl); 1880 #endif 1881 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers); 1882 kfree(net->ipv6.fib6_main_tbl); 1883 kfree(net->ipv6.fib_table_hash); 1884 kfree(net->ipv6.rt6_stats); 1885 } 1886 1887 static struct pernet_operations fib6_net_ops = { 1888 .init = fib6_net_init, 1889 .exit = fib6_net_exit, 1890 }; 1891 1892 int __init fib6_init(void) 1893 { 1894 int ret = -ENOMEM; 1895 1896 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1897 sizeof(struct fib6_node), 1898 0, SLAB_HWCACHE_ALIGN, 1899 NULL); 1900 if (!fib6_node_kmem) 1901 goto out; 1902 1903 ret = register_pernet_subsys(&fib6_net_ops); 1904 if (ret) 1905 goto out_kmem_cache_create; 1906 1907 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib, 1908 NULL); 1909 if (ret) 1910 goto out_unregister_subsys; 1911 1912 __fib6_flush_trees = fib6_flush_trees; 1913 out: 1914 return ret; 1915 1916 out_unregister_subsys: 1917 unregister_pernet_subsys(&fib6_net_ops); 1918 out_kmem_cache_create: 1919 kmem_cache_destroy(fib6_node_kmem); 1920 goto out; 1921 } 1922 1923 void fib6_gc_cleanup(void) 1924 { 1925 unregister_pernet_subsys(&fib6_net_ops); 1926 kmem_cache_destroy(fib6_node_kmem); 1927 } 1928 1929 #ifdef CONFIG_PROC_FS 1930 1931 struct ipv6_route_iter { 1932 struct seq_net_private p; 1933 struct fib6_walker w; 1934 loff_t skip; 1935 struct fib6_table *tbl; 1936 int sernum; 1937 }; 1938 1939 static int ipv6_route_seq_show(struct seq_file *seq, void *v) 1940 { 1941 struct rt6_info *rt = v; 1942 struct ipv6_route_iter *iter = seq->private; 1943 1944 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen); 1945 1946 #ifdef CONFIG_IPV6_SUBTREES 1947 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen); 1948 #else 1949 seq_puts(seq, "00000000000000000000000000000000 00 "); 1950 #endif 1951 if (rt->rt6i_flags & RTF_GATEWAY) 1952 seq_printf(seq, "%pi6", &rt->rt6i_gateway); 1953 else 1954 seq_puts(seq, "00000000000000000000000000000000"); 1955 1956 seq_printf(seq, " %08x %08x %08x %08x %8s\n", 1957 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt), 1958 rt->dst.__use, rt->rt6i_flags, 1959 rt->dst.dev ? rt->dst.dev->name : ""); 1960 iter->w.leaf = NULL; 1961 return 0; 1962 } 1963 1964 static int ipv6_route_yield(struct fib6_walker *w) 1965 { 1966 struct ipv6_route_iter *iter = w->args; 1967 1968 if (!iter->skip) 1969 return 1; 1970 1971 do { 1972 iter->w.leaf = iter->w.leaf->dst.rt6_next; 1973 iter->skip--; 1974 if (!iter->skip && iter->w.leaf) 1975 return 1; 1976 } while (iter->w.leaf); 1977 1978 return 0; 1979 } 1980 1981 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter, 1982 struct net *net) 1983 { 1984 memset(&iter->w, 0, sizeof(iter->w)); 1985 iter->w.func = ipv6_route_yield; 1986 iter->w.root = &iter->tbl->tb6_root; 1987 iter->w.state = FWS_INIT; 1988 iter->w.node = iter->w.root; 1989 iter->w.args = iter; 1990 iter->sernum = iter->w.root->fn_sernum; 1991 INIT_LIST_HEAD(&iter->w.lh); 1992 fib6_walker_link(net, &iter->w); 1993 } 1994 1995 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl, 1996 struct net *net) 1997 { 1998 unsigned int h; 1999 struct hlist_node *node; 2000 2001 if (tbl) { 2002 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1; 2003 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist)); 2004 } else { 2005 h = 0; 2006 node = NULL; 2007 } 2008 2009 while (!node && h < FIB6_TABLE_HASHSZ) { 2010 node = rcu_dereference_bh( 2011 hlist_first_rcu(&net->ipv6.fib_table_hash[h++])); 2012 } 2013 return hlist_entry_safe(node, struct fib6_table, tb6_hlist); 2014 } 2015 2016 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter) 2017 { 2018 if (iter->sernum != iter->w.root->fn_sernum) { 2019 iter->sernum = iter->w.root->fn_sernum; 2020 iter->w.state = FWS_INIT; 2021 iter->w.node = iter->w.root; 2022 WARN_ON(iter->w.skip); 2023 iter->w.skip = iter->w.count; 2024 } 2025 } 2026 2027 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2028 { 2029 int r; 2030 struct rt6_info *n; 2031 struct net *net = seq_file_net(seq); 2032 struct ipv6_route_iter *iter = seq->private; 2033 2034 if (!v) 2035 goto iter_table; 2036 2037 n = ((struct rt6_info *)v)->dst.rt6_next; 2038 if (n) { 2039 ++*pos; 2040 return n; 2041 } 2042 2043 iter_table: 2044 ipv6_route_check_sernum(iter); 2045 read_lock(&iter->tbl->tb6_lock); 2046 r = fib6_walk_continue(&iter->w); 2047 read_unlock(&iter->tbl->tb6_lock); 2048 if (r > 0) { 2049 if (v) 2050 ++*pos; 2051 return iter->w.leaf; 2052 } else if (r < 0) { 2053 fib6_walker_unlink(net, &iter->w); 2054 return NULL; 2055 } 2056 fib6_walker_unlink(net, &iter->w); 2057 2058 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net); 2059 if (!iter->tbl) 2060 return NULL; 2061 2062 ipv6_route_seq_setup_walk(iter, net); 2063 goto iter_table; 2064 } 2065 2066 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos) 2067 __acquires(RCU_BH) 2068 { 2069 struct net *net = seq_file_net(seq); 2070 struct ipv6_route_iter *iter = seq->private; 2071 2072 rcu_read_lock_bh(); 2073 iter->tbl = ipv6_route_seq_next_table(NULL, net); 2074 iter->skip = *pos; 2075 2076 if (iter->tbl) { 2077 ipv6_route_seq_setup_walk(iter, net); 2078 return ipv6_route_seq_next(seq, NULL, pos); 2079 } else { 2080 return NULL; 2081 } 2082 } 2083 2084 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter) 2085 { 2086 struct fib6_walker *w = &iter->w; 2087 return w->node && !(w->state == FWS_U && w->node == w->root); 2088 } 2089 2090 static void ipv6_route_seq_stop(struct seq_file *seq, void *v) 2091 __releases(RCU_BH) 2092 { 2093 struct net *net = seq_file_net(seq); 2094 struct ipv6_route_iter *iter = seq->private; 2095 2096 if (ipv6_route_iter_active(iter)) 2097 fib6_walker_unlink(net, &iter->w); 2098 2099 rcu_read_unlock_bh(); 2100 } 2101 2102 static const struct seq_operations ipv6_route_seq_ops = { 2103 .start = ipv6_route_seq_start, 2104 .next = ipv6_route_seq_next, 2105 .stop = ipv6_route_seq_stop, 2106 .show = ipv6_route_seq_show 2107 }; 2108 2109 int ipv6_route_open(struct inode *inode, struct file *file) 2110 { 2111 return seq_open_net(inode, file, &ipv6_route_seq_ops, 2112 sizeof(struct ipv6_route_iter)); 2113 } 2114 2115 #endif /* CONFIG_PROC_FS */ 2116