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 free_percpu(non_pcpu_rt->rt6i_pcpu); 181 non_pcpu_rt->rt6i_pcpu = NULL; 182 } 183 184 static void rt6_release(struct rt6_info *rt) 185 { 186 if (atomic_dec_and_test(&rt->rt6i_ref)) { 187 rt6_free_pcpu(rt); 188 rt6_rcu_free(rt); 189 } 190 } 191 192 static void fib6_link_table(struct net *net, struct fib6_table *tb) 193 { 194 unsigned int h; 195 196 /* 197 * Initialize table lock at a single place to give lockdep a key, 198 * tables aren't visible prior to being linked to the list. 199 */ 200 rwlock_init(&tb->tb6_lock); 201 202 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1); 203 204 /* 205 * No protection necessary, this is the only list mutatation 206 * operation, tables never disappear once they exist. 207 */ 208 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]); 209 } 210 211 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 212 213 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id) 214 { 215 struct fib6_table *table; 216 217 table = kzalloc(sizeof(*table), GFP_ATOMIC); 218 if (table) { 219 table->tb6_id = id; 220 table->tb6_root.leaf = net->ipv6.ip6_null_entry; 221 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 222 inet_peer_base_init(&table->tb6_peers); 223 } 224 225 return table; 226 } 227 228 struct fib6_table *fib6_new_table(struct net *net, u32 id) 229 { 230 struct fib6_table *tb; 231 232 if (id == 0) 233 id = RT6_TABLE_MAIN; 234 tb = fib6_get_table(net, id); 235 if (tb) 236 return tb; 237 238 tb = fib6_alloc_table(net, id); 239 if (tb) 240 fib6_link_table(net, tb); 241 242 return tb; 243 } 244 EXPORT_SYMBOL_GPL(fib6_new_table); 245 246 struct fib6_table *fib6_get_table(struct net *net, u32 id) 247 { 248 struct fib6_table *tb; 249 struct hlist_head *head; 250 unsigned int h; 251 252 if (id == 0) 253 id = RT6_TABLE_MAIN; 254 h = id & (FIB6_TABLE_HASHSZ - 1); 255 rcu_read_lock(); 256 head = &net->ipv6.fib_table_hash[h]; 257 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 258 if (tb->tb6_id == id) { 259 rcu_read_unlock(); 260 return tb; 261 } 262 } 263 rcu_read_unlock(); 264 265 return NULL; 266 } 267 EXPORT_SYMBOL_GPL(fib6_get_table); 268 269 static void __net_init fib6_tables_init(struct net *net) 270 { 271 fib6_link_table(net, net->ipv6.fib6_main_tbl); 272 fib6_link_table(net, net->ipv6.fib6_local_tbl); 273 } 274 #else 275 276 struct fib6_table *fib6_new_table(struct net *net, u32 id) 277 { 278 return fib6_get_table(net, id); 279 } 280 281 struct fib6_table *fib6_get_table(struct net *net, u32 id) 282 { 283 return net->ipv6.fib6_main_tbl; 284 } 285 286 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, 287 int flags, pol_lookup_t lookup) 288 { 289 struct rt6_info *rt; 290 291 rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags); 292 if (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 /* Multipath routes are dumped in one route with the 322 * RTA_MULTIPATH attribute. Jump 'rt' to point to the 323 * last sibling of this route (no need to dump the 324 * sibling routes again) 325 */ 326 if (rt->rt6i_nsiblings) 327 rt = list_last_entry(&rt->rt6i_siblings, 328 struct rt6_info, 329 rt6i_siblings); 330 } 331 w->leaf = NULL; 332 return 0; 333 } 334 335 static void fib6_dump_end(struct netlink_callback *cb) 336 { 337 struct net *net = sock_net(cb->skb->sk); 338 struct fib6_walker *w = (void *)cb->args[2]; 339 340 if (w) { 341 if (cb->args[4]) { 342 cb->args[4] = 0; 343 fib6_walker_unlink(net, w); 344 } 345 cb->args[2] = 0; 346 kfree(w); 347 } 348 cb->done = (void *)cb->args[3]; 349 cb->args[1] = 3; 350 } 351 352 static int fib6_dump_done(struct netlink_callback *cb) 353 { 354 fib6_dump_end(cb); 355 return cb->done ? cb->done(cb) : 0; 356 } 357 358 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 359 struct netlink_callback *cb) 360 { 361 struct net *net = sock_net(skb->sk); 362 struct fib6_walker *w; 363 int res; 364 365 w = (void *)cb->args[2]; 366 w->root = &table->tb6_root; 367 368 if (cb->args[4] == 0) { 369 w->count = 0; 370 w->skip = 0; 371 372 read_lock_bh(&table->tb6_lock); 373 res = fib6_walk(net, w); 374 read_unlock_bh(&table->tb6_lock); 375 if (res > 0) { 376 cb->args[4] = 1; 377 cb->args[5] = w->root->fn_sernum; 378 } 379 } else { 380 if (cb->args[5] != w->root->fn_sernum) { 381 /* Begin at the root if the tree changed */ 382 cb->args[5] = w->root->fn_sernum; 383 w->state = FWS_INIT; 384 w->node = w->root; 385 w->skip = w->count; 386 } else 387 w->skip = 0; 388 389 read_lock_bh(&table->tb6_lock); 390 res = fib6_walk_continue(w); 391 read_unlock_bh(&table->tb6_lock); 392 if (res <= 0) { 393 fib6_walker_unlink(net, w); 394 cb->args[4] = 0; 395 } 396 } 397 398 return res; 399 } 400 401 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 402 { 403 struct net *net = sock_net(skb->sk); 404 unsigned int h, s_h; 405 unsigned int e = 0, s_e; 406 struct rt6_rtnl_dump_arg arg; 407 struct fib6_walker *w; 408 struct fib6_table *tb; 409 struct hlist_head *head; 410 int res = 0; 411 412 s_h = cb->args[0]; 413 s_e = cb->args[1]; 414 415 w = (void *)cb->args[2]; 416 if (!w) { 417 /* New dump: 418 * 419 * 1. hook callback destructor. 420 */ 421 cb->args[3] = (long)cb->done; 422 cb->done = fib6_dump_done; 423 424 /* 425 * 2. allocate and initialize walker. 426 */ 427 w = kzalloc(sizeof(*w), GFP_ATOMIC); 428 if (!w) 429 return -ENOMEM; 430 w->func = fib6_dump_node; 431 cb->args[2] = (long)w; 432 } 433 434 arg.skb = skb; 435 arg.cb = cb; 436 arg.net = net; 437 w->args = &arg; 438 439 rcu_read_lock(); 440 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) { 441 e = 0; 442 head = &net->ipv6.fib_table_hash[h]; 443 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 444 if (e < s_e) 445 goto next; 446 res = fib6_dump_table(tb, skb, cb); 447 if (res != 0) 448 goto out; 449 next: 450 e++; 451 } 452 } 453 out: 454 rcu_read_unlock(); 455 cb->args[1] = e; 456 cb->args[0] = h; 457 458 res = res < 0 ? res : skb->len; 459 if (res <= 0) 460 fib6_dump_end(cb); 461 return res; 462 } 463 464 /* 465 * Routing Table 466 * 467 * return the appropriate node for a routing tree "add" operation 468 * by either creating and inserting or by returning an existing 469 * node. 470 */ 471 472 static struct fib6_node *fib6_add_1(struct fib6_node *root, 473 struct in6_addr *addr, int plen, 474 int offset, int allow_create, 475 int replace_required, int sernum) 476 { 477 struct fib6_node *fn, *in, *ln; 478 struct fib6_node *pn = NULL; 479 struct rt6key *key; 480 int bit; 481 __be32 dir = 0; 482 483 RT6_TRACE("fib6_add_1\n"); 484 485 /* insert node in tree */ 486 487 fn = root; 488 489 do { 490 key = (struct rt6key *)((u8 *)fn->leaf + offset); 491 492 /* 493 * Prefix match 494 */ 495 if (plen < fn->fn_bit || 496 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) { 497 if (!allow_create) { 498 if (replace_required) { 499 pr_warn("Can't replace route, no match found\n"); 500 return ERR_PTR(-ENOENT); 501 } 502 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 503 } 504 goto insert_above; 505 } 506 507 /* 508 * Exact match ? 509 */ 510 511 if (plen == fn->fn_bit) { 512 /* clean up an intermediate node */ 513 if (!(fn->fn_flags & RTN_RTINFO)) { 514 rt6_release(fn->leaf); 515 fn->leaf = NULL; 516 } 517 518 fn->fn_sernum = sernum; 519 520 return fn; 521 } 522 523 /* 524 * We have more bits to go 525 */ 526 527 /* Try to walk down on tree. */ 528 fn->fn_sernum = sernum; 529 dir = addr_bit_set(addr, fn->fn_bit); 530 pn = fn; 531 fn = dir ? fn->right : fn->left; 532 } while (fn); 533 534 if (!allow_create) { 535 /* We should not create new node because 536 * NLM_F_REPLACE was specified without NLM_F_CREATE 537 * I assume it is safe to require NLM_F_CREATE when 538 * REPLACE flag is used! Later we may want to remove the 539 * check for replace_required, because according 540 * to netlink specification, NLM_F_CREATE 541 * MUST be specified if new route is created. 542 * That would keep IPv6 consistent with IPv4 543 */ 544 if (replace_required) { 545 pr_warn("Can't replace route, no match found\n"); 546 return ERR_PTR(-ENOENT); 547 } 548 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 549 } 550 /* 551 * We walked to the bottom of tree. 552 * Create new leaf node without children. 553 */ 554 555 ln = node_alloc(); 556 557 if (!ln) 558 return ERR_PTR(-ENOMEM); 559 ln->fn_bit = plen; 560 561 ln->parent = pn; 562 ln->fn_sernum = sernum; 563 564 if (dir) 565 pn->right = ln; 566 else 567 pn->left = ln; 568 569 return ln; 570 571 572 insert_above: 573 /* 574 * split since we don't have a common prefix anymore or 575 * we have a less significant route. 576 * we've to insert an intermediate node on the list 577 * this new node will point to the one we need to create 578 * and the current 579 */ 580 581 pn = fn->parent; 582 583 /* find 1st bit in difference between the 2 addrs. 584 585 See comment in __ipv6_addr_diff: bit may be an invalid value, 586 but if it is >= plen, the value is ignored in any case. 587 */ 588 589 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr)); 590 591 /* 592 * (intermediate)[in] 593 * / \ 594 * (new leaf node)[ln] (old node)[fn] 595 */ 596 if (plen > bit) { 597 in = node_alloc(); 598 ln = node_alloc(); 599 600 if (!in || !ln) { 601 if (in) 602 node_free(in); 603 if (ln) 604 node_free(ln); 605 return ERR_PTR(-ENOMEM); 606 } 607 608 /* 609 * new intermediate node. 610 * RTN_RTINFO will 611 * be off since that an address that chooses one of 612 * the branches would not match less specific routes 613 * in the other branch 614 */ 615 616 in->fn_bit = bit; 617 618 in->parent = pn; 619 in->leaf = fn->leaf; 620 atomic_inc(&in->leaf->rt6i_ref); 621 622 in->fn_sernum = sernum; 623 624 /* update parent pointer */ 625 if (dir) 626 pn->right = in; 627 else 628 pn->left = in; 629 630 ln->fn_bit = plen; 631 632 ln->parent = in; 633 fn->parent = in; 634 635 ln->fn_sernum = sernum; 636 637 if (addr_bit_set(addr, bit)) { 638 in->right = ln; 639 in->left = fn; 640 } else { 641 in->left = ln; 642 in->right = fn; 643 } 644 } else { /* plen <= bit */ 645 646 /* 647 * (new leaf node)[ln] 648 * / \ 649 * (old node)[fn] NULL 650 */ 651 652 ln = node_alloc(); 653 654 if (!ln) 655 return ERR_PTR(-ENOMEM); 656 657 ln->fn_bit = plen; 658 659 ln->parent = pn; 660 661 ln->fn_sernum = sernum; 662 663 if (dir) 664 pn->right = ln; 665 else 666 pn->left = ln; 667 668 if (addr_bit_set(&key->addr, plen)) 669 ln->right = fn; 670 else 671 ln->left = fn; 672 673 fn->parent = ln; 674 } 675 return ln; 676 } 677 678 static bool rt6_qualify_for_ecmp(struct rt6_info *rt) 679 { 680 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) == 681 RTF_GATEWAY; 682 } 683 684 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc) 685 { 686 int i; 687 688 for (i = 0; i < RTAX_MAX; i++) { 689 if (test_bit(i, mxc->mx_valid)) 690 mp[i] = mxc->mx[i]; 691 } 692 } 693 694 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc) 695 { 696 if (!mxc->mx) 697 return 0; 698 699 if (dst->flags & DST_HOST) { 700 u32 *mp = dst_metrics_write_ptr(dst); 701 702 if (unlikely(!mp)) 703 return -ENOMEM; 704 705 fib6_copy_metrics(mp, mxc); 706 } else { 707 dst_init_metrics(dst, mxc->mx, false); 708 709 /* We've stolen mx now. */ 710 mxc->mx = NULL; 711 } 712 713 return 0; 714 } 715 716 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn, 717 struct net *net) 718 { 719 if (atomic_read(&rt->rt6i_ref) != 1) { 720 /* This route is used as dummy address holder in some split 721 * nodes. It is not leaked, but it still holds other resources, 722 * which must be released in time. So, scan ascendant nodes 723 * and replace dummy references to this route with references 724 * to still alive ones. 725 */ 726 while (fn) { 727 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) { 728 fn->leaf = fib6_find_prefix(net, fn); 729 atomic_inc(&fn->leaf->rt6i_ref); 730 rt6_release(rt); 731 } 732 fn = fn->parent; 733 } 734 /* No more references are possible at this point. */ 735 BUG_ON(atomic_read(&rt->rt6i_ref) != 1); 736 } 737 } 738 739 /* 740 * Insert routing information in a node. 741 */ 742 743 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, 744 struct nl_info *info, struct mx6_config *mxc) 745 { 746 struct rt6_info *iter = NULL; 747 struct rt6_info **ins; 748 struct rt6_info **fallback_ins = NULL; 749 int replace = (info->nlh && 750 (info->nlh->nlmsg_flags & NLM_F_REPLACE)); 751 int add = (!info->nlh || 752 (info->nlh->nlmsg_flags & NLM_F_CREATE)); 753 int found = 0; 754 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt); 755 u16 nlflags = NLM_F_EXCL; 756 int err; 757 758 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND)) 759 nlflags |= NLM_F_APPEND; 760 761 ins = &fn->leaf; 762 763 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) { 764 /* 765 * Search for duplicates 766 */ 767 768 if (iter->rt6i_metric == rt->rt6i_metric) { 769 /* 770 * Same priority level 771 */ 772 if (info->nlh && 773 (info->nlh->nlmsg_flags & NLM_F_EXCL)) 774 return -EEXIST; 775 776 nlflags &= ~NLM_F_EXCL; 777 if (replace) { 778 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) { 779 found++; 780 break; 781 } 782 if (rt_can_ecmp) 783 fallback_ins = fallback_ins ?: ins; 784 goto next_iter; 785 } 786 787 if (iter->dst.dev == rt->dst.dev && 788 iter->rt6i_idev == rt->rt6i_idev && 789 ipv6_addr_equal(&iter->rt6i_gateway, 790 &rt->rt6i_gateway)) { 791 if (rt->rt6i_nsiblings) 792 rt->rt6i_nsiblings = 0; 793 if (!(iter->rt6i_flags & RTF_EXPIRES)) 794 return -EEXIST; 795 if (!(rt->rt6i_flags & RTF_EXPIRES)) 796 rt6_clean_expires(iter); 797 else 798 rt6_set_expires(iter, rt->dst.expires); 799 iter->rt6i_pmtu = rt->rt6i_pmtu; 800 return -EEXIST; 801 } 802 /* If we have the same destination and the same metric, 803 * but not the same gateway, then the route we try to 804 * add is sibling to this route, increment our counter 805 * of siblings, and later we will add our route to the 806 * list. 807 * Only static routes (which don't have flag 808 * RTF_EXPIRES) are used for ECMPv6. 809 * 810 * To avoid long list, we only had siblings if the 811 * route have a gateway. 812 */ 813 if (rt_can_ecmp && 814 rt6_qualify_for_ecmp(iter)) 815 rt->rt6i_nsiblings++; 816 } 817 818 if (iter->rt6i_metric > rt->rt6i_metric) 819 break; 820 821 next_iter: 822 ins = &iter->dst.rt6_next; 823 } 824 825 if (fallback_ins && !found) { 826 /* No ECMP-able route found, replace first non-ECMP one */ 827 ins = fallback_ins; 828 iter = *ins; 829 found++; 830 } 831 832 /* Reset round-robin state, if necessary */ 833 if (ins == &fn->leaf) 834 fn->rr_ptr = NULL; 835 836 /* Link this route to others same route. */ 837 if (rt->rt6i_nsiblings) { 838 unsigned int rt6i_nsiblings; 839 struct rt6_info *sibling, *temp_sibling; 840 841 /* Find the first route that have the same metric */ 842 sibling = fn->leaf; 843 while (sibling) { 844 if (sibling->rt6i_metric == rt->rt6i_metric && 845 rt6_qualify_for_ecmp(sibling)) { 846 list_add_tail(&rt->rt6i_siblings, 847 &sibling->rt6i_siblings); 848 break; 849 } 850 sibling = sibling->dst.rt6_next; 851 } 852 /* For each sibling in the list, increment the counter of 853 * siblings. BUG() if counters does not match, list of siblings 854 * is broken! 855 */ 856 rt6i_nsiblings = 0; 857 list_for_each_entry_safe(sibling, temp_sibling, 858 &rt->rt6i_siblings, rt6i_siblings) { 859 sibling->rt6i_nsiblings++; 860 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings); 861 rt6i_nsiblings++; 862 } 863 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings); 864 } 865 866 /* 867 * insert node 868 */ 869 if (!replace) { 870 if (!add) 871 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 872 873 add: 874 nlflags |= NLM_F_CREATE; 875 err = fib6_commit_metrics(&rt->dst, mxc); 876 if (err) 877 return err; 878 879 rt->dst.rt6_next = iter; 880 *ins = rt; 881 rt->rt6i_node = fn; 882 atomic_inc(&rt->rt6i_ref); 883 if (!info->skip_notify) 884 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags); 885 info->nl_net->ipv6.rt6_stats->fib_rt_entries++; 886 887 if (!(fn->fn_flags & RTN_RTINFO)) { 888 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 889 fn->fn_flags |= RTN_RTINFO; 890 } 891 892 } else { 893 int nsiblings; 894 895 if (!found) { 896 if (add) 897 goto add; 898 pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); 899 return -ENOENT; 900 } 901 902 err = fib6_commit_metrics(&rt->dst, mxc); 903 if (err) 904 return err; 905 906 *ins = rt; 907 rt->rt6i_node = fn; 908 rt->dst.rt6_next = iter->dst.rt6_next; 909 atomic_inc(&rt->rt6i_ref); 910 if (!info->skip_notify) 911 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE); 912 if (!(fn->fn_flags & RTN_RTINFO)) { 913 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 914 fn->fn_flags |= RTN_RTINFO; 915 } 916 nsiblings = iter->rt6i_nsiblings; 917 fib6_purge_rt(iter, fn, info->nl_net); 918 rt6_release(iter); 919 920 if (nsiblings) { 921 /* Replacing an ECMP route, remove all siblings */ 922 ins = &rt->dst.rt6_next; 923 iter = *ins; 924 while (iter) { 925 if (iter->rt6i_metric > rt->rt6i_metric) 926 break; 927 if (rt6_qualify_for_ecmp(iter)) { 928 *ins = iter->dst.rt6_next; 929 fib6_purge_rt(iter, fn, info->nl_net); 930 rt6_release(iter); 931 nsiblings--; 932 } else { 933 ins = &iter->dst.rt6_next; 934 } 935 iter = *ins; 936 } 937 WARN_ON(nsiblings != 0); 938 } 939 } 940 941 return 0; 942 } 943 944 static void fib6_start_gc(struct net *net, struct rt6_info *rt) 945 { 946 if (!timer_pending(&net->ipv6.ip6_fib_timer) && 947 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE))) 948 mod_timer(&net->ipv6.ip6_fib_timer, 949 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 950 } 951 952 void fib6_force_start_gc(struct net *net) 953 { 954 if (!timer_pending(&net->ipv6.ip6_fib_timer)) 955 mod_timer(&net->ipv6.ip6_fib_timer, 956 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 957 } 958 959 /* 960 * Add routing information to the routing tree. 961 * <destination addr>/<source addr> 962 * with source addr info in sub-trees 963 */ 964 965 int fib6_add(struct fib6_node *root, struct rt6_info *rt, 966 struct nl_info *info, struct mx6_config *mxc) 967 { 968 struct fib6_node *fn, *pn = NULL; 969 int err = -ENOMEM; 970 int allow_create = 1; 971 int replace_required = 0; 972 int sernum = fib6_new_sernum(info->nl_net); 973 974 if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) && 975 !atomic_read(&rt->dst.__refcnt))) 976 return -EINVAL; 977 978 if (info->nlh) { 979 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) 980 allow_create = 0; 981 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) 982 replace_required = 1; 983 } 984 if (!allow_create && !replace_required) 985 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); 986 987 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen, 988 offsetof(struct rt6_info, rt6i_dst), allow_create, 989 replace_required, sernum); 990 if (IS_ERR(fn)) { 991 err = PTR_ERR(fn); 992 fn = NULL; 993 goto out; 994 } 995 996 pn = fn; 997 998 #ifdef CONFIG_IPV6_SUBTREES 999 if (rt->rt6i_src.plen) { 1000 struct fib6_node *sn; 1001 1002 if (!fn->subtree) { 1003 struct fib6_node *sfn; 1004 1005 /* 1006 * Create subtree. 1007 * 1008 * fn[main tree] 1009 * | 1010 * sfn[subtree root] 1011 * \ 1012 * sn[new leaf node] 1013 */ 1014 1015 /* Create subtree root node */ 1016 sfn = node_alloc(); 1017 if (!sfn) 1018 goto st_failure; 1019 1020 sfn->leaf = info->nl_net->ipv6.ip6_null_entry; 1021 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref); 1022 sfn->fn_flags = RTN_ROOT; 1023 sfn->fn_sernum = sernum; 1024 1025 /* Now add the first leaf node to new subtree */ 1026 1027 sn = fib6_add_1(sfn, &rt->rt6i_src.addr, 1028 rt->rt6i_src.plen, 1029 offsetof(struct rt6_info, rt6i_src), 1030 allow_create, replace_required, sernum); 1031 1032 if (IS_ERR(sn)) { 1033 /* If it is failed, discard just allocated 1034 root, and then (in st_failure) stale node 1035 in main tree. 1036 */ 1037 node_free(sfn); 1038 err = PTR_ERR(sn); 1039 goto st_failure; 1040 } 1041 1042 /* Now link new subtree to main tree */ 1043 sfn->parent = fn; 1044 fn->subtree = sfn; 1045 } else { 1046 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, 1047 rt->rt6i_src.plen, 1048 offsetof(struct rt6_info, rt6i_src), 1049 allow_create, replace_required, sernum); 1050 1051 if (IS_ERR(sn)) { 1052 err = PTR_ERR(sn); 1053 goto st_failure; 1054 } 1055 } 1056 1057 if (!fn->leaf) { 1058 fn->leaf = rt; 1059 atomic_inc(&rt->rt6i_ref); 1060 } 1061 fn = sn; 1062 } 1063 #endif 1064 1065 err = fib6_add_rt2node(fn, rt, info, mxc); 1066 if (!err) { 1067 fib6_start_gc(info->nl_net, rt); 1068 if (!(rt->rt6i_flags & RTF_CACHE)) 1069 fib6_prune_clones(info->nl_net, pn); 1070 rt->dst.flags &= ~DST_NOCACHE; 1071 } 1072 1073 out: 1074 if (err) { 1075 #ifdef CONFIG_IPV6_SUBTREES 1076 /* 1077 * If fib6_add_1 has cleared the old leaf pointer in the 1078 * super-tree leaf node we have to find a new one for it. 1079 */ 1080 if (pn != fn && pn->leaf == rt) { 1081 pn->leaf = NULL; 1082 atomic_dec(&rt->rt6i_ref); 1083 } 1084 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) { 1085 pn->leaf = fib6_find_prefix(info->nl_net, pn); 1086 #if RT6_DEBUG >= 2 1087 if (!pn->leaf) { 1088 WARN_ON(pn->leaf == NULL); 1089 pn->leaf = info->nl_net->ipv6.ip6_null_entry; 1090 } 1091 #endif 1092 atomic_inc(&pn->leaf->rt6i_ref); 1093 } 1094 #endif 1095 if (!(rt->dst.flags & DST_NOCACHE)) 1096 dst_free(&rt->dst); 1097 } 1098 return err; 1099 1100 #ifdef CONFIG_IPV6_SUBTREES 1101 /* Subtree creation failed, probably main tree node 1102 is orphan. If it is, shoot it. 1103 */ 1104 st_failure: 1105 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) 1106 fib6_repair_tree(info->nl_net, fn); 1107 if (!(rt->dst.flags & DST_NOCACHE)) 1108 dst_free(&rt->dst); 1109 return err; 1110 #endif 1111 } 1112 1113 /* 1114 * Routing tree lookup 1115 * 1116 */ 1117 1118 struct lookup_args { 1119 int offset; /* key offset on rt6_info */ 1120 const struct in6_addr *addr; /* search key */ 1121 }; 1122 1123 static struct fib6_node *fib6_lookup_1(struct fib6_node *root, 1124 struct lookup_args *args) 1125 { 1126 struct fib6_node *fn; 1127 __be32 dir; 1128 1129 if (unlikely(args->offset == 0)) 1130 return NULL; 1131 1132 /* 1133 * Descend on a tree 1134 */ 1135 1136 fn = root; 1137 1138 for (;;) { 1139 struct fib6_node *next; 1140 1141 dir = addr_bit_set(args->addr, fn->fn_bit); 1142 1143 next = dir ? fn->right : fn->left; 1144 1145 if (next) { 1146 fn = next; 1147 continue; 1148 } 1149 break; 1150 } 1151 1152 while (fn) { 1153 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) { 1154 struct rt6key *key; 1155 1156 key = (struct rt6key *) ((u8 *) fn->leaf + 1157 args->offset); 1158 1159 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 1160 #ifdef CONFIG_IPV6_SUBTREES 1161 if (fn->subtree) { 1162 struct fib6_node *sfn; 1163 sfn = fib6_lookup_1(fn->subtree, 1164 args + 1); 1165 if (!sfn) 1166 goto backtrack; 1167 fn = sfn; 1168 } 1169 #endif 1170 if (fn->fn_flags & RTN_RTINFO) 1171 return fn; 1172 } 1173 } 1174 #ifdef CONFIG_IPV6_SUBTREES 1175 backtrack: 1176 #endif 1177 if (fn->fn_flags & RTN_ROOT) 1178 break; 1179 1180 fn = fn->parent; 1181 } 1182 1183 return NULL; 1184 } 1185 1186 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr, 1187 const struct in6_addr *saddr) 1188 { 1189 struct fib6_node *fn; 1190 struct lookup_args args[] = { 1191 { 1192 .offset = offsetof(struct rt6_info, rt6i_dst), 1193 .addr = daddr, 1194 }, 1195 #ifdef CONFIG_IPV6_SUBTREES 1196 { 1197 .offset = offsetof(struct rt6_info, rt6i_src), 1198 .addr = saddr, 1199 }, 1200 #endif 1201 { 1202 .offset = 0, /* sentinel */ 1203 } 1204 }; 1205 1206 fn = fib6_lookup_1(root, daddr ? args : args + 1); 1207 if (!fn || fn->fn_flags & RTN_TL_ROOT) 1208 fn = root; 1209 1210 return fn; 1211 } 1212 1213 /* 1214 * Get node with specified destination prefix (and source prefix, 1215 * if subtrees are used) 1216 */ 1217 1218 1219 static struct fib6_node *fib6_locate_1(struct fib6_node *root, 1220 const struct in6_addr *addr, 1221 int plen, int offset) 1222 { 1223 struct fib6_node *fn; 1224 1225 for (fn = root; fn ; ) { 1226 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); 1227 1228 /* 1229 * Prefix match 1230 */ 1231 if (plen < fn->fn_bit || 1232 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 1233 return NULL; 1234 1235 if (plen == fn->fn_bit) 1236 return fn; 1237 1238 /* 1239 * We have more bits to go 1240 */ 1241 if (addr_bit_set(addr, fn->fn_bit)) 1242 fn = fn->right; 1243 else 1244 fn = fn->left; 1245 } 1246 return NULL; 1247 } 1248 1249 struct fib6_node *fib6_locate(struct fib6_node *root, 1250 const struct in6_addr *daddr, int dst_len, 1251 const struct in6_addr *saddr, int src_len) 1252 { 1253 struct fib6_node *fn; 1254 1255 fn = fib6_locate_1(root, daddr, dst_len, 1256 offsetof(struct rt6_info, rt6i_dst)); 1257 1258 #ifdef CONFIG_IPV6_SUBTREES 1259 if (src_len) { 1260 WARN_ON(saddr == NULL); 1261 if (fn && fn->subtree) 1262 fn = fib6_locate_1(fn->subtree, saddr, src_len, 1263 offsetof(struct rt6_info, rt6i_src)); 1264 } 1265 #endif 1266 1267 if (fn && fn->fn_flags & RTN_RTINFO) 1268 return fn; 1269 1270 return NULL; 1271 } 1272 1273 1274 /* 1275 * Deletion 1276 * 1277 */ 1278 1279 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn) 1280 { 1281 if (fn->fn_flags & RTN_ROOT) 1282 return net->ipv6.ip6_null_entry; 1283 1284 while (fn) { 1285 if (fn->left) 1286 return fn->left->leaf; 1287 if (fn->right) 1288 return fn->right->leaf; 1289 1290 fn = FIB6_SUBTREE(fn); 1291 } 1292 return NULL; 1293 } 1294 1295 /* 1296 * Called to trim the tree of intermediate nodes when possible. "fn" 1297 * is the node we want to try and remove. 1298 */ 1299 1300 static struct fib6_node *fib6_repair_tree(struct net *net, 1301 struct fib6_node *fn) 1302 { 1303 int children; 1304 int nstate; 1305 struct fib6_node *child, *pn; 1306 struct fib6_walker *w; 1307 int iter = 0; 1308 1309 for (;;) { 1310 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1311 iter++; 1312 1313 WARN_ON(fn->fn_flags & RTN_RTINFO); 1314 WARN_ON(fn->fn_flags & RTN_TL_ROOT); 1315 WARN_ON(fn->leaf); 1316 1317 children = 0; 1318 child = NULL; 1319 if (fn->right) 1320 child = fn->right, children |= 1; 1321 if (fn->left) 1322 child = fn->left, children |= 2; 1323 1324 if (children == 3 || FIB6_SUBTREE(fn) 1325 #ifdef CONFIG_IPV6_SUBTREES 1326 /* Subtree root (i.e. fn) may have one child */ 1327 || (children && fn->fn_flags & RTN_ROOT) 1328 #endif 1329 ) { 1330 fn->leaf = fib6_find_prefix(net, fn); 1331 #if RT6_DEBUG >= 2 1332 if (!fn->leaf) { 1333 WARN_ON(!fn->leaf); 1334 fn->leaf = net->ipv6.ip6_null_entry; 1335 } 1336 #endif 1337 atomic_inc(&fn->leaf->rt6i_ref); 1338 return fn->parent; 1339 } 1340 1341 pn = fn->parent; 1342 #ifdef CONFIG_IPV6_SUBTREES 1343 if (FIB6_SUBTREE(pn) == fn) { 1344 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1345 FIB6_SUBTREE(pn) = NULL; 1346 nstate = FWS_L; 1347 } else { 1348 WARN_ON(fn->fn_flags & RTN_ROOT); 1349 #endif 1350 if (pn->right == fn) 1351 pn->right = child; 1352 else if (pn->left == fn) 1353 pn->left = child; 1354 #if RT6_DEBUG >= 2 1355 else 1356 WARN_ON(1); 1357 #endif 1358 if (child) 1359 child->parent = pn; 1360 nstate = FWS_R; 1361 #ifdef CONFIG_IPV6_SUBTREES 1362 } 1363 #endif 1364 1365 read_lock(&net->ipv6.fib6_walker_lock); 1366 FOR_WALKERS(net, w) { 1367 if (!child) { 1368 if (w->root == fn) { 1369 w->root = w->node = NULL; 1370 RT6_TRACE("W %p adjusted by delroot 1\n", w); 1371 } else if (w->node == fn) { 1372 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1373 w->node = pn; 1374 w->state = nstate; 1375 } 1376 } else { 1377 if (w->root == fn) { 1378 w->root = child; 1379 RT6_TRACE("W %p adjusted by delroot 2\n", w); 1380 } 1381 if (w->node == fn) { 1382 w->node = child; 1383 if (children&2) { 1384 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1385 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT; 1386 } else { 1387 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1388 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT; 1389 } 1390 } 1391 } 1392 } 1393 read_unlock(&net->ipv6.fib6_walker_lock); 1394 1395 node_free(fn); 1396 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) 1397 return pn; 1398 1399 rt6_release(pn->leaf); 1400 pn->leaf = NULL; 1401 fn = pn; 1402 } 1403 } 1404 1405 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, 1406 struct nl_info *info) 1407 { 1408 struct fib6_walker *w; 1409 struct rt6_info *rt = *rtp; 1410 struct net *net = info->nl_net; 1411 1412 RT6_TRACE("fib6_del_route\n"); 1413 1414 /* Unlink it */ 1415 *rtp = rt->dst.rt6_next; 1416 rt->rt6i_node = NULL; 1417 net->ipv6.rt6_stats->fib_rt_entries--; 1418 net->ipv6.rt6_stats->fib_discarded_routes++; 1419 1420 /* Reset round-robin state, if necessary */ 1421 if (fn->rr_ptr == rt) 1422 fn->rr_ptr = NULL; 1423 1424 /* Remove this entry from other siblings */ 1425 if (rt->rt6i_nsiblings) { 1426 struct rt6_info *sibling, *next_sibling; 1427 1428 list_for_each_entry_safe(sibling, next_sibling, 1429 &rt->rt6i_siblings, rt6i_siblings) 1430 sibling->rt6i_nsiblings--; 1431 rt->rt6i_nsiblings = 0; 1432 list_del_init(&rt->rt6i_siblings); 1433 } 1434 1435 /* Adjust walkers */ 1436 read_lock(&net->ipv6.fib6_walker_lock); 1437 FOR_WALKERS(net, w) { 1438 if (w->state == FWS_C && w->leaf == rt) { 1439 RT6_TRACE("walker %p adjusted by delroute\n", w); 1440 w->leaf = rt->dst.rt6_next; 1441 if (!w->leaf) 1442 w->state = FWS_U; 1443 } 1444 } 1445 read_unlock(&net->ipv6.fib6_walker_lock); 1446 1447 rt->dst.rt6_next = NULL; 1448 1449 /* If it was last route, expunge its radix tree node */ 1450 if (!fn->leaf) { 1451 fn->fn_flags &= ~RTN_RTINFO; 1452 net->ipv6.rt6_stats->fib_route_nodes--; 1453 fn = fib6_repair_tree(net, fn); 1454 } 1455 1456 fib6_purge_rt(rt, fn, net); 1457 1458 if (!info->skip_notify) 1459 inet6_rt_notify(RTM_DELROUTE, rt, info, 0); 1460 rt6_release(rt); 1461 } 1462 1463 int fib6_del(struct rt6_info *rt, struct nl_info *info) 1464 { 1465 struct net *net = info->nl_net; 1466 struct fib6_node *fn = rt->rt6i_node; 1467 struct rt6_info **rtp; 1468 1469 #if RT6_DEBUG >= 2 1470 if (rt->dst.obsolete > 0) { 1471 WARN_ON(fn); 1472 return -ENOENT; 1473 } 1474 #endif 1475 if (!fn || rt == net->ipv6.ip6_null_entry) 1476 return -ENOENT; 1477 1478 WARN_ON(!(fn->fn_flags & RTN_RTINFO)); 1479 1480 if (!(rt->rt6i_flags & RTF_CACHE)) { 1481 struct fib6_node *pn = fn; 1482 #ifdef CONFIG_IPV6_SUBTREES 1483 /* clones of this route might be in another subtree */ 1484 if (rt->rt6i_src.plen) { 1485 while (!(pn->fn_flags & RTN_ROOT)) 1486 pn = pn->parent; 1487 pn = pn->parent; 1488 } 1489 #endif 1490 fib6_prune_clones(info->nl_net, pn); 1491 } 1492 1493 /* 1494 * Walk the leaf entries looking for ourself 1495 */ 1496 1497 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) { 1498 if (*rtp == rt) { 1499 fib6_del_route(fn, rtp, info); 1500 return 0; 1501 } 1502 } 1503 return -ENOENT; 1504 } 1505 1506 /* 1507 * Tree traversal function. 1508 * 1509 * Certainly, it is not interrupt safe. 1510 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 1511 * It means, that we can modify tree during walking 1512 * and use this function for garbage collection, clone pruning, 1513 * cleaning tree when a device goes down etc. etc. 1514 * 1515 * It guarantees that every node will be traversed, 1516 * and that it will be traversed only once. 1517 * 1518 * Callback function w->func may return: 1519 * 0 -> continue walking. 1520 * positive value -> walking is suspended (used by tree dumps, 1521 * and probably by gc, if it will be split to several slices) 1522 * negative value -> terminate walking. 1523 * 1524 * The function itself returns: 1525 * 0 -> walk is complete. 1526 * >0 -> walk is incomplete (i.e. suspended) 1527 * <0 -> walk is terminated by an error. 1528 */ 1529 1530 static int fib6_walk_continue(struct fib6_walker *w) 1531 { 1532 struct fib6_node *fn, *pn; 1533 1534 for (;;) { 1535 fn = w->node; 1536 if (!fn) 1537 return 0; 1538 1539 if (w->prune && fn != w->root && 1540 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) { 1541 w->state = FWS_C; 1542 w->leaf = fn->leaf; 1543 } 1544 switch (w->state) { 1545 #ifdef CONFIG_IPV6_SUBTREES 1546 case FWS_S: 1547 if (FIB6_SUBTREE(fn)) { 1548 w->node = FIB6_SUBTREE(fn); 1549 continue; 1550 } 1551 w->state = FWS_L; 1552 #endif 1553 case FWS_L: 1554 if (fn->left) { 1555 w->node = fn->left; 1556 w->state = FWS_INIT; 1557 continue; 1558 } 1559 w->state = FWS_R; 1560 case FWS_R: 1561 if (fn->right) { 1562 w->node = fn->right; 1563 w->state = FWS_INIT; 1564 continue; 1565 } 1566 w->state = FWS_C; 1567 w->leaf = fn->leaf; 1568 case FWS_C: 1569 if (w->leaf && fn->fn_flags & RTN_RTINFO) { 1570 int err; 1571 1572 if (w->skip) { 1573 w->skip--; 1574 goto skip; 1575 } 1576 1577 err = w->func(w); 1578 if (err) 1579 return err; 1580 1581 w->count++; 1582 continue; 1583 } 1584 skip: 1585 w->state = FWS_U; 1586 case FWS_U: 1587 if (fn == w->root) 1588 return 0; 1589 pn = fn->parent; 1590 w->node = pn; 1591 #ifdef CONFIG_IPV6_SUBTREES 1592 if (FIB6_SUBTREE(pn) == fn) { 1593 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1594 w->state = FWS_L; 1595 continue; 1596 } 1597 #endif 1598 if (pn->left == fn) { 1599 w->state = FWS_R; 1600 continue; 1601 } 1602 if (pn->right == fn) { 1603 w->state = FWS_C; 1604 w->leaf = w->node->leaf; 1605 continue; 1606 } 1607 #if RT6_DEBUG >= 2 1608 WARN_ON(1); 1609 #endif 1610 } 1611 } 1612 } 1613 1614 static int fib6_walk(struct net *net, struct fib6_walker *w) 1615 { 1616 int res; 1617 1618 w->state = FWS_INIT; 1619 w->node = w->root; 1620 1621 fib6_walker_link(net, w); 1622 res = fib6_walk_continue(w); 1623 if (res <= 0) 1624 fib6_walker_unlink(net, w); 1625 return res; 1626 } 1627 1628 static int fib6_clean_node(struct fib6_walker *w) 1629 { 1630 int res; 1631 struct rt6_info *rt; 1632 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w); 1633 struct nl_info info = { 1634 .nl_net = c->net, 1635 }; 1636 1637 if (c->sernum != FIB6_NO_SERNUM_CHANGE && 1638 w->node->fn_sernum != c->sernum) 1639 w->node->fn_sernum = c->sernum; 1640 1641 if (!c->func) { 1642 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE); 1643 w->leaf = NULL; 1644 return 0; 1645 } 1646 1647 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 1648 res = c->func(rt, c->arg); 1649 if (res < 0) { 1650 w->leaf = rt; 1651 res = fib6_del(rt, &info); 1652 if (res) { 1653 #if RT6_DEBUG >= 2 1654 pr_debug("%s: del failed: rt=%p@%p err=%d\n", 1655 __func__, rt, rt->rt6i_node, res); 1656 #endif 1657 continue; 1658 } 1659 return 0; 1660 } 1661 WARN_ON(res != 0); 1662 } 1663 w->leaf = rt; 1664 return 0; 1665 } 1666 1667 /* 1668 * Convenient frontend to tree walker. 1669 * 1670 * func is called on each route. 1671 * It may return -1 -> delete this route. 1672 * 0 -> continue walking 1673 * 1674 * prune==1 -> only immediate children of node (certainly, 1675 * ignoring pure split nodes) will be scanned. 1676 */ 1677 1678 static void fib6_clean_tree(struct net *net, struct fib6_node *root, 1679 int (*func)(struct rt6_info *, void *arg), 1680 bool prune, int sernum, void *arg) 1681 { 1682 struct fib6_cleaner c; 1683 1684 c.w.root = root; 1685 c.w.func = fib6_clean_node; 1686 c.w.prune = prune; 1687 c.w.count = 0; 1688 c.w.skip = 0; 1689 c.func = func; 1690 c.sernum = sernum; 1691 c.arg = arg; 1692 c.net = net; 1693 1694 fib6_walk(net, &c.w); 1695 } 1696 1697 static void __fib6_clean_all(struct net *net, 1698 int (*func)(struct rt6_info *, void *), 1699 int sernum, void *arg) 1700 { 1701 struct fib6_table *table; 1702 struct hlist_head *head; 1703 unsigned int h; 1704 1705 rcu_read_lock(); 1706 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 1707 head = &net->ipv6.fib_table_hash[h]; 1708 hlist_for_each_entry_rcu(table, head, tb6_hlist) { 1709 write_lock_bh(&table->tb6_lock); 1710 fib6_clean_tree(net, &table->tb6_root, 1711 func, false, sernum, arg); 1712 write_unlock_bh(&table->tb6_lock); 1713 } 1714 } 1715 rcu_read_unlock(); 1716 } 1717 1718 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *), 1719 void *arg) 1720 { 1721 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg); 1722 } 1723 1724 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1725 { 1726 if (rt->rt6i_flags & RTF_CACHE) { 1727 RT6_TRACE("pruning clone %p\n", rt); 1728 return -1; 1729 } 1730 1731 return 0; 1732 } 1733 1734 static void fib6_prune_clones(struct net *net, struct fib6_node *fn) 1735 { 1736 fib6_clean_tree(net, fn, fib6_prune_clone, true, 1737 FIB6_NO_SERNUM_CHANGE, NULL); 1738 } 1739 1740 static void fib6_flush_trees(struct net *net) 1741 { 1742 int new_sernum = fib6_new_sernum(net); 1743 1744 __fib6_clean_all(net, NULL, new_sernum, NULL); 1745 } 1746 1747 /* 1748 * Garbage collection 1749 */ 1750 1751 struct fib6_gc_args 1752 { 1753 int timeout; 1754 int more; 1755 }; 1756 1757 static int fib6_age(struct rt6_info *rt, void *arg) 1758 { 1759 struct fib6_gc_args *gc_args = arg; 1760 unsigned long now = jiffies; 1761 1762 /* 1763 * check addrconf expiration here. 1764 * Routes are expired even if they are in use. 1765 * 1766 * Also age clones. Note, that clones are aged out 1767 * only if they are not in use now. 1768 */ 1769 1770 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) { 1771 if (time_after(now, rt->dst.expires)) { 1772 RT6_TRACE("expiring %p\n", rt); 1773 return -1; 1774 } 1775 gc_args->more++; 1776 } else if (rt->rt6i_flags & RTF_CACHE) { 1777 if (atomic_read(&rt->dst.__refcnt) == 0 && 1778 time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) { 1779 RT6_TRACE("aging clone %p\n", rt); 1780 return -1; 1781 } else if (rt->rt6i_flags & RTF_GATEWAY) { 1782 struct neighbour *neigh; 1783 __u8 neigh_flags = 0; 1784 1785 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway); 1786 if (neigh) { 1787 neigh_flags = neigh->flags; 1788 neigh_release(neigh); 1789 } 1790 if (!(neigh_flags & NTF_ROUTER)) { 1791 RT6_TRACE("purging route %p via non-router but gateway\n", 1792 rt); 1793 return -1; 1794 } 1795 } 1796 gc_args->more++; 1797 } 1798 1799 return 0; 1800 } 1801 1802 void fib6_run_gc(unsigned long expires, struct net *net, bool force) 1803 { 1804 struct fib6_gc_args gc_args; 1805 unsigned long now; 1806 1807 if (force) { 1808 spin_lock_bh(&net->ipv6.fib6_gc_lock); 1809 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) { 1810 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); 1811 return; 1812 } 1813 gc_args.timeout = expires ? (int)expires : 1814 net->ipv6.sysctl.ip6_rt_gc_interval; 1815 1816 gc_args.more = icmp6_dst_gc(); 1817 1818 fib6_clean_all(net, fib6_age, &gc_args); 1819 now = jiffies; 1820 net->ipv6.ip6_rt_last_gc = now; 1821 1822 if (gc_args.more) 1823 mod_timer(&net->ipv6.ip6_fib_timer, 1824 round_jiffies(now 1825 + net->ipv6.sysctl.ip6_rt_gc_interval)); 1826 else 1827 del_timer(&net->ipv6.ip6_fib_timer); 1828 spin_unlock_bh(&net->ipv6.fib6_gc_lock); 1829 } 1830 1831 static void fib6_gc_timer_cb(unsigned long arg) 1832 { 1833 fib6_run_gc(0, (struct net *)arg, true); 1834 } 1835 1836 static int __net_init fib6_net_init(struct net *net) 1837 { 1838 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; 1839 1840 spin_lock_init(&net->ipv6.fib6_gc_lock); 1841 rwlock_init(&net->ipv6.fib6_walker_lock); 1842 INIT_LIST_HEAD(&net->ipv6.fib6_walkers); 1843 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net); 1844 1845 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); 1846 if (!net->ipv6.rt6_stats) 1847 goto out_timer; 1848 1849 /* Avoid false sharing : Use at least a full cache line */ 1850 size = max_t(size_t, size, L1_CACHE_BYTES); 1851 1852 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); 1853 if (!net->ipv6.fib_table_hash) 1854 goto out_rt6_stats; 1855 1856 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), 1857 GFP_KERNEL); 1858 if (!net->ipv6.fib6_main_tbl) 1859 goto out_fib_table_hash; 1860 1861 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; 1862 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1863 net->ipv6.fib6_main_tbl->tb6_root.fn_flags = 1864 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1865 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); 1866 1867 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1868 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), 1869 GFP_KERNEL); 1870 if (!net->ipv6.fib6_local_tbl) 1871 goto out_fib6_main_tbl; 1872 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; 1873 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1874 net->ipv6.fib6_local_tbl->tb6_root.fn_flags = 1875 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1876 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); 1877 #endif 1878 fib6_tables_init(net); 1879 1880 return 0; 1881 1882 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1883 out_fib6_main_tbl: 1884 kfree(net->ipv6.fib6_main_tbl); 1885 #endif 1886 out_fib_table_hash: 1887 kfree(net->ipv6.fib_table_hash); 1888 out_rt6_stats: 1889 kfree(net->ipv6.rt6_stats); 1890 out_timer: 1891 return -ENOMEM; 1892 } 1893 1894 static void fib6_net_exit(struct net *net) 1895 { 1896 rt6_ifdown(net, NULL); 1897 del_timer_sync(&net->ipv6.ip6_fib_timer); 1898 1899 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1900 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers); 1901 kfree(net->ipv6.fib6_local_tbl); 1902 #endif 1903 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers); 1904 kfree(net->ipv6.fib6_main_tbl); 1905 kfree(net->ipv6.fib_table_hash); 1906 kfree(net->ipv6.rt6_stats); 1907 } 1908 1909 static struct pernet_operations fib6_net_ops = { 1910 .init = fib6_net_init, 1911 .exit = fib6_net_exit, 1912 }; 1913 1914 int __init fib6_init(void) 1915 { 1916 int ret = -ENOMEM; 1917 1918 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1919 sizeof(struct fib6_node), 1920 0, SLAB_HWCACHE_ALIGN, 1921 NULL); 1922 if (!fib6_node_kmem) 1923 goto out; 1924 1925 ret = register_pernet_subsys(&fib6_net_ops); 1926 if (ret) 1927 goto out_kmem_cache_create; 1928 1929 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib, 1930 NULL); 1931 if (ret) 1932 goto out_unregister_subsys; 1933 1934 __fib6_flush_trees = fib6_flush_trees; 1935 out: 1936 return ret; 1937 1938 out_unregister_subsys: 1939 unregister_pernet_subsys(&fib6_net_ops); 1940 out_kmem_cache_create: 1941 kmem_cache_destroy(fib6_node_kmem); 1942 goto out; 1943 } 1944 1945 void fib6_gc_cleanup(void) 1946 { 1947 unregister_pernet_subsys(&fib6_net_ops); 1948 kmem_cache_destroy(fib6_node_kmem); 1949 } 1950 1951 #ifdef CONFIG_PROC_FS 1952 1953 struct ipv6_route_iter { 1954 struct seq_net_private p; 1955 struct fib6_walker w; 1956 loff_t skip; 1957 struct fib6_table *tbl; 1958 int sernum; 1959 }; 1960 1961 static int ipv6_route_seq_show(struct seq_file *seq, void *v) 1962 { 1963 struct rt6_info *rt = v; 1964 struct ipv6_route_iter *iter = seq->private; 1965 1966 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen); 1967 1968 #ifdef CONFIG_IPV6_SUBTREES 1969 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen); 1970 #else 1971 seq_puts(seq, "00000000000000000000000000000000 00 "); 1972 #endif 1973 if (rt->rt6i_flags & RTF_GATEWAY) 1974 seq_printf(seq, "%pi6", &rt->rt6i_gateway); 1975 else 1976 seq_puts(seq, "00000000000000000000000000000000"); 1977 1978 seq_printf(seq, " %08x %08x %08x %08x %8s\n", 1979 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt), 1980 rt->dst.__use, rt->rt6i_flags, 1981 rt->dst.dev ? rt->dst.dev->name : ""); 1982 iter->w.leaf = NULL; 1983 return 0; 1984 } 1985 1986 static int ipv6_route_yield(struct fib6_walker *w) 1987 { 1988 struct ipv6_route_iter *iter = w->args; 1989 1990 if (!iter->skip) 1991 return 1; 1992 1993 do { 1994 iter->w.leaf = iter->w.leaf->dst.rt6_next; 1995 iter->skip--; 1996 if (!iter->skip && iter->w.leaf) 1997 return 1; 1998 } while (iter->w.leaf); 1999 2000 return 0; 2001 } 2002 2003 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter, 2004 struct net *net) 2005 { 2006 memset(&iter->w, 0, sizeof(iter->w)); 2007 iter->w.func = ipv6_route_yield; 2008 iter->w.root = &iter->tbl->tb6_root; 2009 iter->w.state = FWS_INIT; 2010 iter->w.node = iter->w.root; 2011 iter->w.args = iter; 2012 iter->sernum = iter->w.root->fn_sernum; 2013 INIT_LIST_HEAD(&iter->w.lh); 2014 fib6_walker_link(net, &iter->w); 2015 } 2016 2017 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl, 2018 struct net *net) 2019 { 2020 unsigned int h; 2021 struct hlist_node *node; 2022 2023 if (tbl) { 2024 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1; 2025 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist)); 2026 } else { 2027 h = 0; 2028 node = NULL; 2029 } 2030 2031 while (!node && h < FIB6_TABLE_HASHSZ) { 2032 node = rcu_dereference_bh( 2033 hlist_first_rcu(&net->ipv6.fib_table_hash[h++])); 2034 } 2035 return hlist_entry_safe(node, struct fib6_table, tb6_hlist); 2036 } 2037 2038 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter) 2039 { 2040 if (iter->sernum != iter->w.root->fn_sernum) { 2041 iter->sernum = iter->w.root->fn_sernum; 2042 iter->w.state = FWS_INIT; 2043 iter->w.node = iter->w.root; 2044 WARN_ON(iter->w.skip); 2045 iter->w.skip = iter->w.count; 2046 } 2047 } 2048 2049 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2050 { 2051 int r; 2052 struct rt6_info *n; 2053 struct net *net = seq_file_net(seq); 2054 struct ipv6_route_iter *iter = seq->private; 2055 2056 if (!v) 2057 goto iter_table; 2058 2059 n = ((struct rt6_info *)v)->dst.rt6_next; 2060 if (n) { 2061 ++*pos; 2062 return n; 2063 } 2064 2065 iter_table: 2066 ipv6_route_check_sernum(iter); 2067 read_lock(&iter->tbl->tb6_lock); 2068 r = fib6_walk_continue(&iter->w); 2069 read_unlock(&iter->tbl->tb6_lock); 2070 if (r > 0) { 2071 if (v) 2072 ++*pos; 2073 return iter->w.leaf; 2074 } else if (r < 0) { 2075 fib6_walker_unlink(net, &iter->w); 2076 return NULL; 2077 } 2078 fib6_walker_unlink(net, &iter->w); 2079 2080 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net); 2081 if (!iter->tbl) 2082 return NULL; 2083 2084 ipv6_route_seq_setup_walk(iter, net); 2085 goto iter_table; 2086 } 2087 2088 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos) 2089 __acquires(RCU_BH) 2090 { 2091 struct net *net = seq_file_net(seq); 2092 struct ipv6_route_iter *iter = seq->private; 2093 2094 rcu_read_lock_bh(); 2095 iter->tbl = ipv6_route_seq_next_table(NULL, net); 2096 iter->skip = *pos; 2097 2098 if (iter->tbl) { 2099 ipv6_route_seq_setup_walk(iter, net); 2100 return ipv6_route_seq_next(seq, NULL, pos); 2101 } else { 2102 return NULL; 2103 } 2104 } 2105 2106 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter) 2107 { 2108 struct fib6_walker *w = &iter->w; 2109 return w->node && !(w->state == FWS_U && w->node == w->root); 2110 } 2111 2112 static void ipv6_route_seq_stop(struct seq_file *seq, void *v) 2113 __releases(RCU_BH) 2114 { 2115 struct net *net = seq_file_net(seq); 2116 struct ipv6_route_iter *iter = seq->private; 2117 2118 if (ipv6_route_iter_active(iter)) 2119 fib6_walker_unlink(net, &iter->w); 2120 2121 rcu_read_unlock_bh(); 2122 } 2123 2124 static const struct seq_operations ipv6_route_seq_ops = { 2125 .start = ipv6_route_seq_start, 2126 .next = ipv6_route_seq_next, 2127 .stop = ipv6_route_seq_stop, 2128 .show = ipv6_route_seq_show 2129 }; 2130 2131 int ipv6_route_open(struct inode *inode, struct file *file) 2132 { 2133 return seq_open_net(inode, file, &ipv6_route_seq_ops, 2134 sizeof(struct ipv6_route_iter)); 2135 } 2136 2137 #endif /* CONFIG_PROC_FS */ 2138