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->rt6i_flags & RTF_REJECT && 293 rt->dst.error == -EAGAIN) { 294 ip6_rt_put(rt); 295 rt = net->ipv6.ip6_null_entry; 296 dst_hold(&rt->dst); 297 } 298 299 return &rt->dst; 300 } 301 302 static void __net_init fib6_tables_init(struct net *net) 303 { 304 fib6_link_table(net, net->ipv6.fib6_main_tbl); 305 } 306 307 #endif 308 309 static int fib6_dump_node(struct fib6_walker *w) 310 { 311 int res; 312 struct rt6_info *rt; 313 314 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 315 res = rt6_dump_route(rt, w->args); 316 if (res < 0) { 317 /* Frame is full, suspend walking */ 318 w->leaf = rt; 319 return 1; 320 } 321 } 322 w->leaf = NULL; 323 return 0; 324 } 325 326 static void fib6_dump_end(struct netlink_callback *cb) 327 { 328 struct net *net = sock_net(cb->skb->sk); 329 struct fib6_walker *w = (void *)cb->args[2]; 330 331 if (w) { 332 if (cb->args[4]) { 333 cb->args[4] = 0; 334 fib6_walker_unlink(net, w); 335 } 336 cb->args[2] = 0; 337 kfree(w); 338 } 339 cb->done = (void *)cb->args[3]; 340 cb->args[1] = 3; 341 } 342 343 static int fib6_dump_done(struct netlink_callback *cb) 344 { 345 fib6_dump_end(cb); 346 return cb->done ? cb->done(cb) : 0; 347 } 348 349 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 350 struct netlink_callback *cb) 351 { 352 struct net *net = sock_net(skb->sk); 353 struct fib6_walker *w; 354 int res; 355 356 w = (void *)cb->args[2]; 357 w->root = &table->tb6_root; 358 359 if (cb->args[4] == 0) { 360 w->count = 0; 361 w->skip = 0; 362 363 read_lock_bh(&table->tb6_lock); 364 res = fib6_walk(net, w); 365 read_unlock_bh(&table->tb6_lock); 366 if (res > 0) { 367 cb->args[4] = 1; 368 cb->args[5] = w->root->fn_sernum; 369 } 370 } else { 371 if (cb->args[5] != w->root->fn_sernum) { 372 /* Begin at the root if the tree changed */ 373 cb->args[5] = w->root->fn_sernum; 374 w->state = FWS_INIT; 375 w->node = w->root; 376 w->skip = w->count; 377 } else 378 w->skip = 0; 379 380 read_lock_bh(&table->tb6_lock); 381 res = fib6_walk_continue(w); 382 read_unlock_bh(&table->tb6_lock); 383 if (res <= 0) { 384 fib6_walker_unlink(net, w); 385 cb->args[4] = 0; 386 } 387 } 388 389 return res; 390 } 391 392 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 393 { 394 struct net *net = sock_net(skb->sk); 395 unsigned int h, s_h; 396 unsigned int e = 0, s_e; 397 struct rt6_rtnl_dump_arg arg; 398 struct fib6_walker *w; 399 struct fib6_table *tb; 400 struct hlist_head *head; 401 int res = 0; 402 403 s_h = cb->args[0]; 404 s_e = cb->args[1]; 405 406 w = (void *)cb->args[2]; 407 if (!w) { 408 /* New dump: 409 * 410 * 1. hook callback destructor. 411 */ 412 cb->args[3] = (long)cb->done; 413 cb->done = fib6_dump_done; 414 415 /* 416 * 2. allocate and initialize walker. 417 */ 418 w = kzalloc(sizeof(*w), GFP_ATOMIC); 419 if (!w) 420 return -ENOMEM; 421 w->func = fib6_dump_node; 422 cb->args[2] = (long)w; 423 } 424 425 arg.skb = skb; 426 arg.cb = cb; 427 arg.net = net; 428 w->args = &arg; 429 430 rcu_read_lock(); 431 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) { 432 e = 0; 433 head = &net->ipv6.fib_table_hash[h]; 434 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 435 if (e < s_e) 436 goto next; 437 res = fib6_dump_table(tb, skb, cb); 438 if (res != 0) 439 goto out; 440 next: 441 e++; 442 } 443 } 444 out: 445 rcu_read_unlock(); 446 cb->args[1] = e; 447 cb->args[0] = h; 448 449 res = res < 0 ? res : skb->len; 450 if (res <= 0) 451 fib6_dump_end(cb); 452 return res; 453 } 454 455 /* 456 * Routing Table 457 * 458 * return the appropriate node for a routing tree "add" operation 459 * by either creating and inserting or by returning an existing 460 * node. 461 */ 462 463 static struct fib6_node *fib6_add_1(struct fib6_node *root, 464 struct in6_addr *addr, int plen, 465 int offset, int allow_create, 466 int replace_required, int sernum) 467 { 468 struct fib6_node *fn, *in, *ln; 469 struct fib6_node *pn = NULL; 470 struct rt6key *key; 471 int bit; 472 __be32 dir = 0; 473 474 RT6_TRACE("fib6_add_1\n"); 475 476 /* insert node in tree */ 477 478 fn = root; 479 480 do { 481 key = (struct rt6key *)((u8 *)fn->leaf + offset); 482 483 /* 484 * Prefix match 485 */ 486 if (plen < fn->fn_bit || 487 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) { 488 if (!allow_create) { 489 if (replace_required) { 490 pr_warn("Can't replace route, no match found\n"); 491 return ERR_PTR(-ENOENT); 492 } 493 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 494 } 495 goto insert_above; 496 } 497 498 /* 499 * Exact match ? 500 */ 501 502 if (plen == fn->fn_bit) { 503 /* clean up an intermediate node */ 504 if (!(fn->fn_flags & RTN_RTINFO)) { 505 rt6_release(fn->leaf); 506 fn->leaf = NULL; 507 } 508 509 fn->fn_sernum = sernum; 510 511 return fn; 512 } 513 514 /* 515 * We have more bits to go 516 */ 517 518 /* Try to walk down on tree. */ 519 fn->fn_sernum = sernum; 520 dir = addr_bit_set(addr, fn->fn_bit); 521 pn = fn; 522 fn = dir ? fn->right : fn->left; 523 } while (fn); 524 525 if (!allow_create) { 526 /* We should not create new node because 527 * NLM_F_REPLACE was specified without NLM_F_CREATE 528 * I assume it is safe to require NLM_F_CREATE when 529 * REPLACE flag is used! Later we may want to remove the 530 * check for replace_required, because according 531 * to netlink specification, NLM_F_CREATE 532 * MUST be specified if new route is created. 533 * That would keep IPv6 consistent with IPv4 534 */ 535 if (replace_required) { 536 pr_warn("Can't replace route, no match found\n"); 537 return ERR_PTR(-ENOENT); 538 } 539 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 540 } 541 /* 542 * We walked to the bottom of tree. 543 * Create new leaf node without children. 544 */ 545 546 ln = node_alloc(); 547 548 if (!ln) 549 return ERR_PTR(-ENOMEM); 550 ln->fn_bit = plen; 551 552 ln->parent = pn; 553 ln->fn_sernum = sernum; 554 555 if (dir) 556 pn->right = ln; 557 else 558 pn->left = ln; 559 560 return ln; 561 562 563 insert_above: 564 /* 565 * split since we don't have a common prefix anymore or 566 * we have a less significant route. 567 * we've to insert an intermediate node on the list 568 * this new node will point to the one we need to create 569 * and the current 570 */ 571 572 pn = fn->parent; 573 574 /* find 1st bit in difference between the 2 addrs. 575 576 See comment in __ipv6_addr_diff: bit may be an invalid value, 577 but if it is >= plen, the value is ignored in any case. 578 */ 579 580 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr)); 581 582 /* 583 * (intermediate)[in] 584 * / \ 585 * (new leaf node)[ln] (old node)[fn] 586 */ 587 if (plen > bit) { 588 in = node_alloc(); 589 ln = node_alloc(); 590 591 if (!in || !ln) { 592 if (in) 593 node_free(in); 594 if (ln) 595 node_free(ln); 596 return ERR_PTR(-ENOMEM); 597 } 598 599 /* 600 * new intermediate node. 601 * RTN_RTINFO will 602 * be off since that an address that chooses one of 603 * the branches would not match less specific routes 604 * in the other branch 605 */ 606 607 in->fn_bit = bit; 608 609 in->parent = pn; 610 in->leaf = fn->leaf; 611 atomic_inc(&in->leaf->rt6i_ref); 612 613 in->fn_sernum = sernum; 614 615 /* update parent pointer */ 616 if (dir) 617 pn->right = in; 618 else 619 pn->left = in; 620 621 ln->fn_bit = plen; 622 623 ln->parent = in; 624 fn->parent = in; 625 626 ln->fn_sernum = sernum; 627 628 if (addr_bit_set(addr, bit)) { 629 in->right = ln; 630 in->left = fn; 631 } else { 632 in->left = ln; 633 in->right = fn; 634 } 635 } else { /* plen <= bit */ 636 637 /* 638 * (new leaf node)[ln] 639 * / \ 640 * (old node)[fn] NULL 641 */ 642 643 ln = node_alloc(); 644 645 if (!ln) 646 return ERR_PTR(-ENOMEM); 647 648 ln->fn_bit = plen; 649 650 ln->parent = pn; 651 652 ln->fn_sernum = sernum; 653 654 if (dir) 655 pn->right = ln; 656 else 657 pn->left = ln; 658 659 if (addr_bit_set(&key->addr, plen)) 660 ln->right = fn; 661 else 662 ln->left = fn; 663 664 fn->parent = ln; 665 } 666 return ln; 667 } 668 669 static bool rt6_qualify_for_ecmp(struct rt6_info *rt) 670 { 671 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) == 672 RTF_GATEWAY; 673 } 674 675 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc) 676 { 677 int i; 678 679 for (i = 0; i < RTAX_MAX; i++) { 680 if (test_bit(i, mxc->mx_valid)) 681 mp[i] = mxc->mx[i]; 682 } 683 } 684 685 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc) 686 { 687 if (!mxc->mx) 688 return 0; 689 690 if (dst->flags & DST_HOST) { 691 u32 *mp = dst_metrics_write_ptr(dst); 692 693 if (unlikely(!mp)) 694 return -ENOMEM; 695 696 fib6_copy_metrics(mp, mxc); 697 } else { 698 dst_init_metrics(dst, mxc->mx, false); 699 700 /* We've stolen mx now. */ 701 mxc->mx = NULL; 702 } 703 704 return 0; 705 } 706 707 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn, 708 struct net *net) 709 { 710 if (atomic_read(&rt->rt6i_ref) != 1) { 711 /* This route is used as dummy address holder in some split 712 * nodes. It is not leaked, but it still holds other resources, 713 * which must be released in time. So, scan ascendant nodes 714 * and replace dummy references to this route with references 715 * to still alive ones. 716 */ 717 while (fn) { 718 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) { 719 fn->leaf = fib6_find_prefix(net, fn); 720 atomic_inc(&fn->leaf->rt6i_ref); 721 rt6_release(rt); 722 } 723 fn = fn->parent; 724 } 725 /* No more references are possible at this point. */ 726 BUG_ON(atomic_read(&rt->rt6i_ref) != 1); 727 } 728 } 729 730 /* 731 * Insert routing information in a node. 732 */ 733 734 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, 735 struct nl_info *info, struct mx6_config *mxc) 736 { 737 struct rt6_info *iter = NULL; 738 struct rt6_info **ins; 739 struct rt6_info **fallback_ins = NULL; 740 int replace = (info->nlh && 741 (info->nlh->nlmsg_flags & NLM_F_REPLACE)); 742 int add = (!info->nlh || 743 (info->nlh->nlmsg_flags & NLM_F_CREATE)); 744 int found = 0; 745 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt); 746 u16 nlflags = NLM_F_EXCL; 747 int err; 748 749 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND)) 750 nlflags |= NLM_F_APPEND; 751 752 ins = &fn->leaf; 753 754 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) { 755 /* 756 * Search for duplicates 757 */ 758 759 if (iter->rt6i_metric == rt->rt6i_metric) { 760 /* 761 * Same priority level 762 */ 763 if (info->nlh && 764 (info->nlh->nlmsg_flags & NLM_F_EXCL)) 765 return -EEXIST; 766 767 nlflags &= ~NLM_F_EXCL; 768 if (replace) { 769 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) { 770 found++; 771 break; 772 } 773 if (rt_can_ecmp) 774 fallback_ins = fallback_ins ?: ins; 775 goto next_iter; 776 } 777 778 if (iter->dst.dev == rt->dst.dev && 779 iter->rt6i_idev == rt->rt6i_idev && 780 ipv6_addr_equal(&iter->rt6i_gateway, 781 &rt->rt6i_gateway)) { 782 if (rt->rt6i_nsiblings) 783 rt->rt6i_nsiblings = 0; 784 if (!(iter->rt6i_flags & RTF_EXPIRES)) 785 return -EEXIST; 786 if (!(rt->rt6i_flags & RTF_EXPIRES)) 787 rt6_clean_expires(iter); 788 else 789 rt6_set_expires(iter, rt->dst.expires); 790 iter->rt6i_pmtu = rt->rt6i_pmtu; 791 return -EEXIST; 792 } 793 /* If we have the same destination and the same metric, 794 * but not the same gateway, then the route we try to 795 * add is sibling to this route, increment our counter 796 * of siblings, and later we will add our route to the 797 * list. 798 * Only static routes (which don't have flag 799 * RTF_EXPIRES) are used for ECMPv6. 800 * 801 * To avoid long list, we only had siblings if the 802 * route have a gateway. 803 */ 804 if (rt_can_ecmp && 805 rt6_qualify_for_ecmp(iter)) 806 rt->rt6i_nsiblings++; 807 } 808 809 if (iter->rt6i_metric > rt->rt6i_metric) 810 break; 811 812 next_iter: 813 ins = &iter->dst.rt6_next; 814 } 815 816 if (fallback_ins && !found) { 817 /* No ECMP-able route found, replace first non-ECMP one */ 818 ins = fallback_ins; 819 iter = *ins; 820 found++; 821 } 822 823 /* Reset round-robin state, if necessary */ 824 if (ins == &fn->leaf) 825 fn->rr_ptr = NULL; 826 827 /* Link this route to others same route. */ 828 if (rt->rt6i_nsiblings) { 829 unsigned int rt6i_nsiblings; 830 struct rt6_info *sibling, *temp_sibling; 831 832 /* Find the first route that have the same metric */ 833 sibling = fn->leaf; 834 while (sibling) { 835 if (sibling->rt6i_metric == rt->rt6i_metric && 836 rt6_qualify_for_ecmp(sibling)) { 837 list_add_tail(&rt->rt6i_siblings, 838 &sibling->rt6i_siblings); 839 break; 840 } 841 sibling = sibling->dst.rt6_next; 842 } 843 /* For each sibling in the list, increment the counter of 844 * siblings. BUG() if counters does not match, list of siblings 845 * is broken! 846 */ 847 rt6i_nsiblings = 0; 848 list_for_each_entry_safe(sibling, temp_sibling, 849 &rt->rt6i_siblings, rt6i_siblings) { 850 sibling->rt6i_nsiblings++; 851 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings); 852 rt6i_nsiblings++; 853 } 854 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings); 855 } 856 857 /* 858 * insert node 859 */ 860 if (!replace) { 861 if (!add) 862 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 863 864 add: 865 nlflags |= NLM_F_CREATE; 866 err = fib6_commit_metrics(&rt->dst, mxc); 867 if (err) 868 return err; 869 870 rt->dst.rt6_next = iter; 871 *ins = rt; 872 rt->rt6i_node = fn; 873 atomic_inc(&rt->rt6i_ref); 874 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags); 875 info->nl_net->ipv6.rt6_stats->fib_rt_entries++; 876 877 if (!(fn->fn_flags & RTN_RTINFO)) { 878 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 879 fn->fn_flags |= RTN_RTINFO; 880 } 881 882 } else { 883 int nsiblings; 884 885 if (!found) { 886 if (add) 887 goto add; 888 pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); 889 return -ENOENT; 890 } 891 892 err = fib6_commit_metrics(&rt->dst, mxc); 893 if (err) 894 return err; 895 896 *ins = rt; 897 rt->rt6i_node = fn; 898 rt->dst.rt6_next = iter->dst.rt6_next; 899 atomic_inc(&rt->rt6i_ref); 900 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE); 901 if (!(fn->fn_flags & RTN_RTINFO)) { 902 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 903 fn->fn_flags |= RTN_RTINFO; 904 } 905 nsiblings = iter->rt6i_nsiblings; 906 fib6_purge_rt(iter, fn, info->nl_net); 907 rt6_release(iter); 908 909 if (nsiblings) { 910 /* Replacing an ECMP route, remove all siblings */ 911 ins = &rt->dst.rt6_next; 912 iter = *ins; 913 while (iter) { 914 if (rt6_qualify_for_ecmp(iter)) { 915 *ins = iter->dst.rt6_next; 916 fib6_purge_rt(iter, fn, info->nl_net); 917 rt6_release(iter); 918 nsiblings--; 919 } else { 920 ins = &iter->dst.rt6_next; 921 } 922 iter = *ins; 923 } 924 WARN_ON(nsiblings != 0); 925 } 926 } 927 928 return 0; 929 } 930 931 static void fib6_start_gc(struct net *net, struct rt6_info *rt) 932 { 933 if (!timer_pending(&net->ipv6.ip6_fib_timer) && 934 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE))) 935 mod_timer(&net->ipv6.ip6_fib_timer, 936 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 937 } 938 939 void fib6_force_start_gc(struct net *net) 940 { 941 if (!timer_pending(&net->ipv6.ip6_fib_timer)) 942 mod_timer(&net->ipv6.ip6_fib_timer, 943 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 944 } 945 946 /* 947 * Add routing information to the routing tree. 948 * <destination addr>/<source addr> 949 * with source addr info in sub-trees 950 */ 951 952 int fib6_add(struct fib6_node *root, struct rt6_info *rt, 953 struct nl_info *info, struct mx6_config *mxc) 954 { 955 struct fib6_node *fn, *pn = NULL; 956 int err = -ENOMEM; 957 int allow_create = 1; 958 int replace_required = 0; 959 int sernum = fib6_new_sernum(info->nl_net); 960 961 if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) && 962 !atomic_read(&rt->dst.__refcnt))) 963 return -EINVAL; 964 965 if (info->nlh) { 966 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) 967 allow_create = 0; 968 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) 969 replace_required = 1; 970 } 971 if (!allow_create && !replace_required) 972 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); 973 974 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen, 975 offsetof(struct rt6_info, rt6i_dst), allow_create, 976 replace_required, sernum); 977 if (IS_ERR(fn)) { 978 err = PTR_ERR(fn); 979 fn = NULL; 980 goto out; 981 } 982 983 pn = fn; 984 985 #ifdef CONFIG_IPV6_SUBTREES 986 if (rt->rt6i_src.plen) { 987 struct fib6_node *sn; 988 989 if (!fn->subtree) { 990 struct fib6_node *sfn; 991 992 /* 993 * Create subtree. 994 * 995 * fn[main tree] 996 * | 997 * sfn[subtree root] 998 * \ 999 * sn[new leaf node] 1000 */ 1001 1002 /* Create subtree root node */ 1003 sfn = node_alloc(); 1004 if (!sfn) 1005 goto st_failure; 1006 1007 sfn->leaf = info->nl_net->ipv6.ip6_null_entry; 1008 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref); 1009 sfn->fn_flags = RTN_ROOT; 1010 sfn->fn_sernum = sernum; 1011 1012 /* Now add the first leaf node to new subtree */ 1013 1014 sn = fib6_add_1(sfn, &rt->rt6i_src.addr, 1015 rt->rt6i_src.plen, 1016 offsetof(struct rt6_info, rt6i_src), 1017 allow_create, replace_required, sernum); 1018 1019 if (IS_ERR(sn)) { 1020 /* If it is failed, discard just allocated 1021 root, and then (in st_failure) stale node 1022 in main tree. 1023 */ 1024 node_free(sfn); 1025 err = PTR_ERR(sn); 1026 goto st_failure; 1027 } 1028 1029 /* Now link new subtree to main tree */ 1030 sfn->parent = fn; 1031 fn->subtree = sfn; 1032 } else { 1033 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, 1034 rt->rt6i_src.plen, 1035 offsetof(struct rt6_info, rt6i_src), 1036 allow_create, replace_required, sernum); 1037 1038 if (IS_ERR(sn)) { 1039 err = PTR_ERR(sn); 1040 goto st_failure; 1041 } 1042 } 1043 1044 if (!fn->leaf) { 1045 fn->leaf = rt; 1046 atomic_inc(&rt->rt6i_ref); 1047 } 1048 fn = sn; 1049 } 1050 #endif 1051 1052 err = fib6_add_rt2node(fn, rt, info, mxc); 1053 if (!err) { 1054 fib6_start_gc(info->nl_net, rt); 1055 if (!(rt->rt6i_flags & RTF_CACHE)) 1056 fib6_prune_clones(info->nl_net, pn); 1057 rt->dst.flags &= ~DST_NOCACHE; 1058 } 1059 1060 out: 1061 if (err) { 1062 #ifdef CONFIG_IPV6_SUBTREES 1063 /* 1064 * If fib6_add_1 has cleared the old leaf pointer in the 1065 * super-tree leaf node we have to find a new one for it. 1066 */ 1067 if (pn != fn && pn->leaf == rt) { 1068 pn->leaf = NULL; 1069 atomic_dec(&rt->rt6i_ref); 1070 } 1071 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) { 1072 pn->leaf = fib6_find_prefix(info->nl_net, pn); 1073 #if RT6_DEBUG >= 2 1074 if (!pn->leaf) { 1075 WARN_ON(pn->leaf == NULL); 1076 pn->leaf = info->nl_net->ipv6.ip6_null_entry; 1077 } 1078 #endif 1079 atomic_inc(&pn->leaf->rt6i_ref); 1080 } 1081 #endif 1082 if (!(rt->dst.flags & DST_NOCACHE)) 1083 dst_free(&rt->dst); 1084 } 1085 return err; 1086 1087 #ifdef CONFIG_IPV6_SUBTREES 1088 /* Subtree creation failed, probably main tree node 1089 is orphan. If it is, shoot it. 1090 */ 1091 st_failure: 1092 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) 1093 fib6_repair_tree(info->nl_net, fn); 1094 if (!(rt->dst.flags & DST_NOCACHE)) 1095 dst_free(&rt->dst); 1096 return err; 1097 #endif 1098 } 1099 1100 /* 1101 * Routing tree lookup 1102 * 1103 */ 1104 1105 struct lookup_args { 1106 int offset; /* key offset on rt6_info */ 1107 const struct in6_addr *addr; /* search key */ 1108 }; 1109 1110 static struct fib6_node *fib6_lookup_1(struct fib6_node *root, 1111 struct lookup_args *args) 1112 { 1113 struct fib6_node *fn; 1114 __be32 dir; 1115 1116 if (unlikely(args->offset == 0)) 1117 return NULL; 1118 1119 /* 1120 * Descend on a tree 1121 */ 1122 1123 fn = root; 1124 1125 for (;;) { 1126 struct fib6_node *next; 1127 1128 dir = addr_bit_set(args->addr, fn->fn_bit); 1129 1130 next = dir ? fn->right : fn->left; 1131 1132 if (next) { 1133 fn = next; 1134 continue; 1135 } 1136 break; 1137 } 1138 1139 while (fn) { 1140 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) { 1141 struct rt6key *key; 1142 1143 key = (struct rt6key *) ((u8 *) fn->leaf + 1144 args->offset); 1145 1146 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 1147 #ifdef CONFIG_IPV6_SUBTREES 1148 if (fn->subtree) { 1149 struct fib6_node *sfn; 1150 sfn = fib6_lookup_1(fn->subtree, 1151 args + 1); 1152 if (!sfn) 1153 goto backtrack; 1154 fn = sfn; 1155 } 1156 #endif 1157 if (fn->fn_flags & RTN_RTINFO) 1158 return fn; 1159 } 1160 } 1161 #ifdef CONFIG_IPV6_SUBTREES 1162 backtrack: 1163 #endif 1164 if (fn->fn_flags & RTN_ROOT) 1165 break; 1166 1167 fn = fn->parent; 1168 } 1169 1170 return NULL; 1171 } 1172 1173 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr, 1174 const struct in6_addr *saddr) 1175 { 1176 struct fib6_node *fn; 1177 struct lookup_args args[] = { 1178 { 1179 .offset = offsetof(struct rt6_info, rt6i_dst), 1180 .addr = daddr, 1181 }, 1182 #ifdef CONFIG_IPV6_SUBTREES 1183 { 1184 .offset = offsetof(struct rt6_info, rt6i_src), 1185 .addr = saddr, 1186 }, 1187 #endif 1188 { 1189 .offset = 0, /* sentinel */ 1190 } 1191 }; 1192 1193 fn = fib6_lookup_1(root, daddr ? args : args + 1); 1194 if (!fn || fn->fn_flags & RTN_TL_ROOT) 1195 fn = root; 1196 1197 return fn; 1198 } 1199 1200 /* 1201 * Get node with specified destination prefix (and source prefix, 1202 * if subtrees are used) 1203 */ 1204 1205 1206 static struct fib6_node *fib6_locate_1(struct fib6_node *root, 1207 const struct in6_addr *addr, 1208 int plen, int offset) 1209 { 1210 struct fib6_node *fn; 1211 1212 for (fn = root; fn ; ) { 1213 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); 1214 1215 /* 1216 * Prefix match 1217 */ 1218 if (plen < fn->fn_bit || 1219 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 1220 return NULL; 1221 1222 if (plen == fn->fn_bit) 1223 return fn; 1224 1225 /* 1226 * We have more bits to go 1227 */ 1228 if (addr_bit_set(addr, fn->fn_bit)) 1229 fn = fn->right; 1230 else 1231 fn = fn->left; 1232 } 1233 return NULL; 1234 } 1235 1236 struct fib6_node *fib6_locate(struct fib6_node *root, 1237 const struct in6_addr *daddr, int dst_len, 1238 const struct in6_addr *saddr, int src_len) 1239 { 1240 struct fib6_node *fn; 1241 1242 fn = fib6_locate_1(root, daddr, dst_len, 1243 offsetof(struct rt6_info, rt6i_dst)); 1244 1245 #ifdef CONFIG_IPV6_SUBTREES 1246 if (src_len) { 1247 WARN_ON(saddr == NULL); 1248 if (fn && fn->subtree) 1249 fn = fib6_locate_1(fn->subtree, saddr, src_len, 1250 offsetof(struct rt6_info, rt6i_src)); 1251 } 1252 #endif 1253 1254 if (fn && fn->fn_flags & RTN_RTINFO) 1255 return fn; 1256 1257 return NULL; 1258 } 1259 1260 1261 /* 1262 * Deletion 1263 * 1264 */ 1265 1266 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn) 1267 { 1268 if (fn->fn_flags & RTN_ROOT) 1269 return net->ipv6.ip6_null_entry; 1270 1271 while (fn) { 1272 if (fn->left) 1273 return fn->left->leaf; 1274 if (fn->right) 1275 return fn->right->leaf; 1276 1277 fn = FIB6_SUBTREE(fn); 1278 } 1279 return NULL; 1280 } 1281 1282 /* 1283 * Called to trim the tree of intermediate nodes when possible. "fn" 1284 * is the node we want to try and remove. 1285 */ 1286 1287 static struct fib6_node *fib6_repair_tree(struct net *net, 1288 struct fib6_node *fn) 1289 { 1290 int children; 1291 int nstate; 1292 struct fib6_node *child, *pn; 1293 struct fib6_walker *w; 1294 int iter = 0; 1295 1296 for (;;) { 1297 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1298 iter++; 1299 1300 WARN_ON(fn->fn_flags & RTN_RTINFO); 1301 WARN_ON(fn->fn_flags & RTN_TL_ROOT); 1302 WARN_ON(fn->leaf); 1303 1304 children = 0; 1305 child = NULL; 1306 if (fn->right) 1307 child = fn->right, children |= 1; 1308 if (fn->left) 1309 child = fn->left, children |= 2; 1310 1311 if (children == 3 || FIB6_SUBTREE(fn) 1312 #ifdef CONFIG_IPV6_SUBTREES 1313 /* Subtree root (i.e. fn) may have one child */ 1314 || (children && fn->fn_flags & RTN_ROOT) 1315 #endif 1316 ) { 1317 fn->leaf = fib6_find_prefix(net, fn); 1318 #if RT6_DEBUG >= 2 1319 if (!fn->leaf) { 1320 WARN_ON(!fn->leaf); 1321 fn->leaf = net->ipv6.ip6_null_entry; 1322 } 1323 #endif 1324 atomic_inc(&fn->leaf->rt6i_ref); 1325 return fn->parent; 1326 } 1327 1328 pn = fn->parent; 1329 #ifdef CONFIG_IPV6_SUBTREES 1330 if (FIB6_SUBTREE(pn) == fn) { 1331 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1332 FIB6_SUBTREE(pn) = NULL; 1333 nstate = FWS_L; 1334 } else { 1335 WARN_ON(fn->fn_flags & RTN_ROOT); 1336 #endif 1337 if (pn->right == fn) 1338 pn->right = child; 1339 else if (pn->left == fn) 1340 pn->left = child; 1341 #if RT6_DEBUG >= 2 1342 else 1343 WARN_ON(1); 1344 #endif 1345 if (child) 1346 child->parent = pn; 1347 nstate = FWS_R; 1348 #ifdef CONFIG_IPV6_SUBTREES 1349 } 1350 #endif 1351 1352 read_lock(&net->ipv6.fib6_walker_lock); 1353 FOR_WALKERS(net, w) { 1354 if (!child) { 1355 if (w->root == fn) { 1356 w->root = w->node = NULL; 1357 RT6_TRACE("W %p adjusted by delroot 1\n", w); 1358 } else if (w->node == fn) { 1359 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1360 w->node = pn; 1361 w->state = nstate; 1362 } 1363 } else { 1364 if (w->root == fn) { 1365 w->root = child; 1366 RT6_TRACE("W %p adjusted by delroot 2\n", w); 1367 } 1368 if (w->node == fn) { 1369 w->node = child; 1370 if (children&2) { 1371 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1372 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT; 1373 } else { 1374 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1375 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT; 1376 } 1377 } 1378 } 1379 } 1380 read_unlock(&net->ipv6.fib6_walker_lock); 1381 1382 node_free(fn); 1383 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) 1384 return pn; 1385 1386 rt6_release(pn->leaf); 1387 pn->leaf = NULL; 1388 fn = pn; 1389 } 1390 } 1391 1392 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, 1393 struct nl_info *info) 1394 { 1395 struct fib6_walker *w; 1396 struct rt6_info *rt = *rtp; 1397 struct net *net = info->nl_net; 1398 1399 RT6_TRACE("fib6_del_route\n"); 1400 1401 /* Unlink it */ 1402 *rtp = rt->dst.rt6_next; 1403 rt->rt6i_node = NULL; 1404 net->ipv6.rt6_stats->fib_rt_entries--; 1405 net->ipv6.rt6_stats->fib_discarded_routes++; 1406 1407 /* Reset round-robin state, if necessary */ 1408 if (fn->rr_ptr == rt) 1409 fn->rr_ptr = NULL; 1410 1411 /* Remove this entry from other siblings */ 1412 if (rt->rt6i_nsiblings) { 1413 struct rt6_info *sibling, *next_sibling; 1414 1415 list_for_each_entry_safe(sibling, next_sibling, 1416 &rt->rt6i_siblings, rt6i_siblings) 1417 sibling->rt6i_nsiblings--; 1418 rt->rt6i_nsiblings = 0; 1419 list_del_init(&rt->rt6i_siblings); 1420 } 1421 1422 /* Adjust walkers */ 1423 read_lock(&net->ipv6.fib6_walker_lock); 1424 FOR_WALKERS(net, w) { 1425 if (w->state == FWS_C && w->leaf == rt) { 1426 RT6_TRACE("walker %p adjusted by delroute\n", w); 1427 w->leaf = rt->dst.rt6_next; 1428 if (!w->leaf) 1429 w->state = FWS_U; 1430 } 1431 } 1432 read_unlock(&net->ipv6.fib6_walker_lock); 1433 1434 rt->dst.rt6_next = NULL; 1435 1436 /* If it was last route, expunge its radix tree node */ 1437 if (!fn->leaf) { 1438 fn->fn_flags &= ~RTN_RTINFO; 1439 net->ipv6.rt6_stats->fib_route_nodes--; 1440 fn = fib6_repair_tree(net, fn); 1441 } 1442 1443 fib6_purge_rt(rt, fn, net); 1444 1445 inet6_rt_notify(RTM_DELROUTE, rt, info, 0); 1446 rt6_release(rt); 1447 } 1448 1449 int fib6_del(struct rt6_info *rt, struct nl_info *info) 1450 { 1451 struct net *net = info->nl_net; 1452 struct fib6_node *fn = rt->rt6i_node; 1453 struct rt6_info **rtp; 1454 1455 #if RT6_DEBUG >= 2 1456 if (rt->dst.obsolete > 0) { 1457 WARN_ON(fn); 1458 return -ENOENT; 1459 } 1460 #endif 1461 if (!fn || rt == net->ipv6.ip6_null_entry) 1462 return -ENOENT; 1463 1464 WARN_ON(!(fn->fn_flags & RTN_RTINFO)); 1465 1466 if (!(rt->rt6i_flags & RTF_CACHE)) { 1467 struct fib6_node *pn = fn; 1468 #ifdef CONFIG_IPV6_SUBTREES 1469 /* clones of this route might be in another subtree */ 1470 if (rt->rt6i_src.plen) { 1471 while (!(pn->fn_flags & RTN_ROOT)) 1472 pn = pn->parent; 1473 pn = pn->parent; 1474 } 1475 #endif 1476 fib6_prune_clones(info->nl_net, pn); 1477 } 1478 1479 /* 1480 * Walk the leaf entries looking for ourself 1481 */ 1482 1483 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) { 1484 if (*rtp == rt) { 1485 fib6_del_route(fn, rtp, info); 1486 return 0; 1487 } 1488 } 1489 return -ENOENT; 1490 } 1491 1492 /* 1493 * Tree traversal function. 1494 * 1495 * Certainly, it is not interrupt safe. 1496 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 1497 * It means, that we can modify tree during walking 1498 * and use this function for garbage collection, clone pruning, 1499 * cleaning tree when a device goes down etc. etc. 1500 * 1501 * It guarantees that every node will be traversed, 1502 * and that it will be traversed only once. 1503 * 1504 * Callback function w->func may return: 1505 * 0 -> continue walking. 1506 * positive value -> walking is suspended (used by tree dumps, 1507 * and probably by gc, if it will be split to several slices) 1508 * negative value -> terminate walking. 1509 * 1510 * The function itself returns: 1511 * 0 -> walk is complete. 1512 * >0 -> walk is incomplete (i.e. suspended) 1513 * <0 -> walk is terminated by an error. 1514 */ 1515 1516 static int fib6_walk_continue(struct fib6_walker *w) 1517 { 1518 struct fib6_node *fn, *pn; 1519 1520 for (;;) { 1521 fn = w->node; 1522 if (!fn) 1523 return 0; 1524 1525 if (w->prune && fn != w->root && 1526 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) { 1527 w->state = FWS_C; 1528 w->leaf = fn->leaf; 1529 } 1530 switch (w->state) { 1531 #ifdef CONFIG_IPV6_SUBTREES 1532 case FWS_S: 1533 if (FIB6_SUBTREE(fn)) { 1534 w->node = FIB6_SUBTREE(fn); 1535 continue; 1536 } 1537 w->state = FWS_L; 1538 #endif 1539 case FWS_L: 1540 if (fn->left) { 1541 w->node = fn->left; 1542 w->state = FWS_INIT; 1543 continue; 1544 } 1545 w->state = FWS_R; 1546 case FWS_R: 1547 if (fn->right) { 1548 w->node = fn->right; 1549 w->state = FWS_INIT; 1550 continue; 1551 } 1552 w->state = FWS_C; 1553 w->leaf = fn->leaf; 1554 case FWS_C: 1555 if (w->leaf && fn->fn_flags & RTN_RTINFO) { 1556 int err; 1557 1558 if (w->skip) { 1559 w->skip--; 1560 goto skip; 1561 } 1562 1563 err = w->func(w); 1564 if (err) 1565 return err; 1566 1567 w->count++; 1568 continue; 1569 } 1570 skip: 1571 w->state = FWS_U; 1572 case FWS_U: 1573 if (fn == w->root) 1574 return 0; 1575 pn = fn->parent; 1576 w->node = pn; 1577 #ifdef CONFIG_IPV6_SUBTREES 1578 if (FIB6_SUBTREE(pn) == fn) { 1579 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1580 w->state = FWS_L; 1581 continue; 1582 } 1583 #endif 1584 if (pn->left == fn) { 1585 w->state = FWS_R; 1586 continue; 1587 } 1588 if (pn->right == fn) { 1589 w->state = FWS_C; 1590 w->leaf = w->node->leaf; 1591 continue; 1592 } 1593 #if RT6_DEBUG >= 2 1594 WARN_ON(1); 1595 #endif 1596 } 1597 } 1598 } 1599 1600 static int fib6_walk(struct net *net, struct fib6_walker *w) 1601 { 1602 int res; 1603 1604 w->state = FWS_INIT; 1605 w->node = w->root; 1606 1607 fib6_walker_link(net, w); 1608 res = fib6_walk_continue(w); 1609 if (res <= 0) 1610 fib6_walker_unlink(net, w); 1611 return res; 1612 } 1613 1614 static int fib6_clean_node(struct fib6_walker *w) 1615 { 1616 int res; 1617 struct rt6_info *rt; 1618 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w); 1619 struct nl_info info = { 1620 .nl_net = c->net, 1621 }; 1622 1623 if (c->sernum != FIB6_NO_SERNUM_CHANGE && 1624 w->node->fn_sernum != c->sernum) 1625 w->node->fn_sernum = c->sernum; 1626 1627 if (!c->func) { 1628 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE); 1629 w->leaf = NULL; 1630 return 0; 1631 } 1632 1633 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 1634 res = c->func(rt, c->arg); 1635 if (res < 0) { 1636 w->leaf = rt; 1637 res = fib6_del(rt, &info); 1638 if (res) { 1639 #if RT6_DEBUG >= 2 1640 pr_debug("%s: del failed: rt=%p@%p err=%d\n", 1641 __func__, rt, rt->rt6i_node, res); 1642 #endif 1643 continue; 1644 } 1645 return 0; 1646 } 1647 WARN_ON(res != 0); 1648 } 1649 w->leaf = rt; 1650 return 0; 1651 } 1652 1653 /* 1654 * Convenient frontend to tree walker. 1655 * 1656 * func is called on each route. 1657 * It may return -1 -> delete this route. 1658 * 0 -> continue walking 1659 * 1660 * prune==1 -> only immediate children of node (certainly, 1661 * ignoring pure split nodes) will be scanned. 1662 */ 1663 1664 static void fib6_clean_tree(struct net *net, struct fib6_node *root, 1665 int (*func)(struct rt6_info *, void *arg), 1666 bool prune, int sernum, void *arg) 1667 { 1668 struct fib6_cleaner c; 1669 1670 c.w.root = root; 1671 c.w.func = fib6_clean_node; 1672 c.w.prune = prune; 1673 c.w.count = 0; 1674 c.w.skip = 0; 1675 c.func = func; 1676 c.sernum = sernum; 1677 c.arg = arg; 1678 c.net = net; 1679 1680 fib6_walk(net, &c.w); 1681 } 1682 1683 static void __fib6_clean_all(struct net *net, 1684 int (*func)(struct rt6_info *, void *), 1685 int sernum, void *arg) 1686 { 1687 struct fib6_table *table; 1688 struct hlist_head *head; 1689 unsigned int h; 1690 1691 rcu_read_lock(); 1692 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 1693 head = &net->ipv6.fib_table_hash[h]; 1694 hlist_for_each_entry_rcu(table, head, tb6_hlist) { 1695 write_lock_bh(&table->tb6_lock); 1696 fib6_clean_tree(net, &table->tb6_root, 1697 func, false, sernum, arg); 1698 write_unlock_bh(&table->tb6_lock); 1699 } 1700 } 1701 rcu_read_unlock(); 1702 } 1703 1704 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *), 1705 void *arg) 1706 { 1707 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg); 1708 } 1709 1710 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1711 { 1712 if (rt->rt6i_flags & RTF_CACHE) { 1713 RT6_TRACE("pruning clone %p\n", rt); 1714 return -1; 1715 } 1716 1717 return 0; 1718 } 1719 1720 static void fib6_prune_clones(struct net *net, struct fib6_node *fn) 1721 { 1722 fib6_clean_tree(net, fn, fib6_prune_clone, true, 1723 FIB6_NO_SERNUM_CHANGE, NULL); 1724 } 1725 1726 static void fib6_flush_trees(struct net *net) 1727 { 1728 int new_sernum = fib6_new_sernum(net); 1729 1730 __fib6_clean_all(net, NULL, new_sernum, NULL); 1731 } 1732 1733 /* 1734 * Garbage collection 1735 */ 1736 1737 struct fib6_gc_args 1738 { 1739 int timeout; 1740 int more; 1741 }; 1742 1743 static int fib6_age(struct rt6_info *rt, void *arg) 1744 { 1745 struct fib6_gc_args *gc_args = arg; 1746 unsigned long now = jiffies; 1747 1748 /* 1749 * check addrconf expiration here. 1750 * Routes are expired even if they are in use. 1751 * 1752 * Also age clones. Note, that clones are aged out 1753 * only if they are not in use now. 1754 */ 1755 1756 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) { 1757 if (time_after(now, rt->dst.expires)) { 1758 RT6_TRACE("expiring %p\n", rt); 1759 return -1; 1760 } 1761 gc_args->more++; 1762 } else if (rt->rt6i_flags & RTF_CACHE) { 1763 if (atomic_read(&rt->dst.__refcnt) == 0 && 1764 time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) { 1765 RT6_TRACE("aging clone %p\n", rt); 1766 return -1; 1767 } else if (rt->rt6i_flags & RTF_GATEWAY) { 1768 struct neighbour *neigh; 1769 __u8 neigh_flags = 0; 1770 1771 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway); 1772 if (neigh) { 1773 neigh_flags = neigh->flags; 1774 neigh_release(neigh); 1775 } 1776 if (!(neigh_flags & NTF_ROUTER)) { 1777 RT6_TRACE("purging route %p via non-router but gateway\n", 1778 rt); 1779 return -1; 1780 } 1781 } 1782 gc_args->more++; 1783 } 1784 1785 return 0; 1786 } 1787 1788 void fib6_run_gc(unsigned long expires, struct net *net, bool force) 1789 { 1790 struct fib6_gc_args gc_args; 1791 unsigned long now; 1792 1793 if (force) { 1794 spin_lock_bh(&net->ipv6.fib6_gc_lock); 1795 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) { 1796 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); 1797 return; 1798 } 1799 gc_args.timeout = expires ? (int)expires : 1800 net->ipv6.sysctl.ip6_rt_gc_interval; 1801 1802 gc_args.more = icmp6_dst_gc(); 1803 1804 fib6_clean_all(net, fib6_age, &gc_args); 1805 now = jiffies; 1806 net->ipv6.ip6_rt_last_gc = now; 1807 1808 if (gc_args.more) 1809 mod_timer(&net->ipv6.ip6_fib_timer, 1810 round_jiffies(now 1811 + net->ipv6.sysctl.ip6_rt_gc_interval)); 1812 else 1813 del_timer(&net->ipv6.ip6_fib_timer); 1814 spin_unlock_bh(&net->ipv6.fib6_gc_lock); 1815 } 1816 1817 static void fib6_gc_timer_cb(unsigned long arg) 1818 { 1819 fib6_run_gc(0, (struct net *)arg, true); 1820 } 1821 1822 static int __net_init fib6_net_init(struct net *net) 1823 { 1824 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; 1825 1826 spin_lock_init(&net->ipv6.fib6_gc_lock); 1827 rwlock_init(&net->ipv6.fib6_walker_lock); 1828 INIT_LIST_HEAD(&net->ipv6.fib6_walkers); 1829 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net); 1830 1831 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); 1832 if (!net->ipv6.rt6_stats) 1833 goto out_timer; 1834 1835 /* Avoid false sharing : Use at least a full cache line */ 1836 size = max_t(size_t, size, L1_CACHE_BYTES); 1837 1838 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); 1839 if (!net->ipv6.fib_table_hash) 1840 goto out_rt6_stats; 1841 1842 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), 1843 GFP_KERNEL); 1844 if (!net->ipv6.fib6_main_tbl) 1845 goto out_fib_table_hash; 1846 1847 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; 1848 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1849 net->ipv6.fib6_main_tbl->tb6_root.fn_flags = 1850 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1851 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); 1852 1853 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1854 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), 1855 GFP_KERNEL); 1856 if (!net->ipv6.fib6_local_tbl) 1857 goto out_fib6_main_tbl; 1858 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; 1859 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1860 net->ipv6.fib6_local_tbl->tb6_root.fn_flags = 1861 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1862 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); 1863 #endif 1864 fib6_tables_init(net); 1865 1866 return 0; 1867 1868 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1869 out_fib6_main_tbl: 1870 kfree(net->ipv6.fib6_main_tbl); 1871 #endif 1872 out_fib_table_hash: 1873 kfree(net->ipv6.fib_table_hash); 1874 out_rt6_stats: 1875 kfree(net->ipv6.rt6_stats); 1876 out_timer: 1877 return -ENOMEM; 1878 } 1879 1880 static void fib6_net_exit(struct net *net) 1881 { 1882 rt6_ifdown(net, NULL); 1883 del_timer_sync(&net->ipv6.ip6_fib_timer); 1884 1885 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1886 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers); 1887 kfree(net->ipv6.fib6_local_tbl); 1888 #endif 1889 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers); 1890 kfree(net->ipv6.fib6_main_tbl); 1891 kfree(net->ipv6.fib_table_hash); 1892 kfree(net->ipv6.rt6_stats); 1893 } 1894 1895 static struct pernet_operations fib6_net_ops = { 1896 .init = fib6_net_init, 1897 .exit = fib6_net_exit, 1898 }; 1899 1900 int __init fib6_init(void) 1901 { 1902 int ret = -ENOMEM; 1903 1904 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1905 sizeof(struct fib6_node), 1906 0, SLAB_HWCACHE_ALIGN, 1907 NULL); 1908 if (!fib6_node_kmem) 1909 goto out; 1910 1911 ret = register_pernet_subsys(&fib6_net_ops); 1912 if (ret) 1913 goto out_kmem_cache_create; 1914 1915 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib, 1916 NULL); 1917 if (ret) 1918 goto out_unregister_subsys; 1919 1920 __fib6_flush_trees = fib6_flush_trees; 1921 out: 1922 return ret; 1923 1924 out_unregister_subsys: 1925 unregister_pernet_subsys(&fib6_net_ops); 1926 out_kmem_cache_create: 1927 kmem_cache_destroy(fib6_node_kmem); 1928 goto out; 1929 } 1930 1931 void fib6_gc_cleanup(void) 1932 { 1933 unregister_pernet_subsys(&fib6_net_ops); 1934 kmem_cache_destroy(fib6_node_kmem); 1935 } 1936 1937 #ifdef CONFIG_PROC_FS 1938 1939 struct ipv6_route_iter { 1940 struct seq_net_private p; 1941 struct fib6_walker w; 1942 loff_t skip; 1943 struct fib6_table *tbl; 1944 int sernum; 1945 }; 1946 1947 static int ipv6_route_seq_show(struct seq_file *seq, void *v) 1948 { 1949 struct rt6_info *rt = v; 1950 struct ipv6_route_iter *iter = seq->private; 1951 1952 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen); 1953 1954 #ifdef CONFIG_IPV6_SUBTREES 1955 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen); 1956 #else 1957 seq_puts(seq, "00000000000000000000000000000000 00 "); 1958 #endif 1959 if (rt->rt6i_flags & RTF_GATEWAY) 1960 seq_printf(seq, "%pi6", &rt->rt6i_gateway); 1961 else 1962 seq_puts(seq, "00000000000000000000000000000000"); 1963 1964 seq_printf(seq, " %08x %08x %08x %08x %8s\n", 1965 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt), 1966 rt->dst.__use, rt->rt6i_flags, 1967 rt->dst.dev ? rt->dst.dev->name : ""); 1968 iter->w.leaf = NULL; 1969 return 0; 1970 } 1971 1972 static int ipv6_route_yield(struct fib6_walker *w) 1973 { 1974 struct ipv6_route_iter *iter = w->args; 1975 1976 if (!iter->skip) 1977 return 1; 1978 1979 do { 1980 iter->w.leaf = iter->w.leaf->dst.rt6_next; 1981 iter->skip--; 1982 if (!iter->skip && iter->w.leaf) 1983 return 1; 1984 } while (iter->w.leaf); 1985 1986 return 0; 1987 } 1988 1989 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter, 1990 struct net *net) 1991 { 1992 memset(&iter->w, 0, sizeof(iter->w)); 1993 iter->w.func = ipv6_route_yield; 1994 iter->w.root = &iter->tbl->tb6_root; 1995 iter->w.state = FWS_INIT; 1996 iter->w.node = iter->w.root; 1997 iter->w.args = iter; 1998 iter->sernum = iter->w.root->fn_sernum; 1999 INIT_LIST_HEAD(&iter->w.lh); 2000 fib6_walker_link(net, &iter->w); 2001 } 2002 2003 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl, 2004 struct net *net) 2005 { 2006 unsigned int h; 2007 struct hlist_node *node; 2008 2009 if (tbl) { 2010 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1; 2011 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist)); 2012 } else { 2013 h = 0; 2014 node = NULL; 2015 } 2016 2017 while (!node && h < FIB6_TABLE_HASHSZ) { 2018 node = rcu_dereference_bh( 2019 hlist_first_rcu(&net->ipv6.fib_table_hash[h++])); 2020 } 2021 return hlist_entry_safe(node, struct fib6_table, tb6_hlist); 2022 } 2023 2024 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter) 2025 { 2026 if (iter->sernum != iter->w.root->fn_sernum) { 2027 iter->sernum = iter->w.root->fn_sernum; 2028 iter->w.state = FWS_INIT; 2029 iter->w.node = iter->w.root; 2030 WARN_ON(iter->w.skip); 2031 iter->w.skip = iter->w.count; 2032 } 2033 } 2034 2035 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2036 { 2037 int r; 2038 struct rt6_info *n; 2039 struct net *net = seq_file_net(seq); 2040 struct ipv6_route_iter *iter = seq->private; 2041 2042 if (!v) 2043 goto iter_table; 2044 2045 n = ((struct rt6_info *)v)->dst.rt6_next; 2046 if (n) { 2047 ++*pos; 2048 return n; 2049 } 2050 2051 iter_table: 2052 ipv6_route_check_sernum(iter); 2053 read_lock(&iter->tbl->tb6_lock); 2054 r = fib6_walk_continue(&iter->w); 2055 read_unlock(&iter->tbl->tb6_lock); 2056 if (r > 0) { 2057 if (v) 2058 ++*pos; 2059 return iter->w.leaf; 2060 } else if (r < 0) { 2061 fib6_walker_unlink(net, &iter->w); 2062 return NULL; 2063 } 2064 fib6_walker_unlink(net, &iter->w); 2065 2066 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net); 2067 if (!iter->tbl) 2068 return NULL; 2069 2070 ipv6_route_seq_setup_walk(iter, net); 2071 goto iter_table; 2072 } 2073 2074 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos) 2075 __acquires(RCU_BH) 2076 { 2077 struct net *net = seq_file_net(seq); 2078 struct ipv6_route_iter *iter = seq->private; 2079 2080 rcu_read_lock_bh(); 2081 iter->tbl = ipv6_route_seq_next_table(NULL, net); 2082 iter->skip = *pos; 2083 2084 if (iter->tbl) { 2085 ipv6_route_seq_setup_walk(iter, net); 2086 return ipv6_route_seq_next(seq, NULL, pos); 2087 } else { 2088 return NULL; 2089 } 2090 } 2091 2092 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter) 2093 { 2094 struct fib6_walker *w = &iter->w; 2095 return w->node && !(w->state == FWS_U && w->node == w->root); 2096 } 2097 2098 static void ipv6_route_seq_stop(struct seq_file *seq, void *v) 2099 __releases(RCU_BH) 2100 { 2101 struct net *net = seq_file_net(seq); 2102 struct ipv6_route_iter *iter = seq->private; 2103 2104 if (ipv6_route_iter_active(iter)) 2105 fib6_walker_unlink(net, &iter->w); 2106 2107 rcu_read_unlock_bh(); 2108 } 2109 2110 static const struct seq_operations ipv6_route_seq_ops = { 2111 .start = ipv6_route_seq_start, 2112 .next = ipv6_route_seq_next, 2113 .stop = ipv6_route_seq_stop, 2114 .show = ipv6_route_seq_show 2115 }; 2116 2117 int ipv6_route_open(struct inode *inode, struct file *file) 2118 { 2119 return seq_open_net(inode, file, &ipv6_route_seq_ops, 2120 sizeof(struct ipv6_route_iter)); 2121 } 2122 2123 #endif /* CONFIG_PROC_FS */ 2124