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 /* Multipath routes are dumped in one route with the 323 * RTA_MULTIPATH attribute. Jump 'rt' to point to the 324 * last sibling of this route (no need to dump the 325 * sibling routes again) 326 */ 327 if (rt->rt6i_nsiblings) 328 rt = list_last_entry(&rt->rt6i_siblings, 329 struct rt6_info, 330 rt6i_siblings); 331 } 332 w->leaf = NULL; 333 return 0; 334 } 335 336 static void fib6_dump_end(struct netlink_callback *cb) 337 { 338 struct net *net = sock_net(cb->skb->sk); 339 struct fib6_walker *w = (void *)cb->args[2]; 340 341 if (w) { 342 if (cb->args[4]) { 343 cb->args[4] = 0; 344 fib6_walker_unlink(net, w); 345 } 346 cb->args[2] = 0; 347 kfree(w); 348 } 349 cb->done = (void *)cb->args[3]; 350 cb->args[1] = 3; 351 } 352 353 static int fib6_dump_done(struct netlink_callback *cb) 354 { 355 fib6_dump_end(cb); 356 return cb->done ? cb->done(cb) : 0; 357 } 358 359 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 360 struct netlink_callback *cb) 361 { 362 struct net *net = sock_net(skb->sk); 363 struct fib6_walker *w; 364 int res; 365 366 w = (void *)cb->args[2]; 367 w->root = &table->tb6_root; 368 369 if (cb->args[4] == 0) { 370 w->count = 0; 371 w->skip = 0; 372 373 read_lock_bh(&table->tb6_lock); 374 res = fib6_walk(net, w); 375 read_unlock_bh(&table->tb6_lock); 376 if (res > 0) { 377 cb->args[4] = 1; 378 cb->args[5] = w->root->fn_sernum; 379 } 380 } else { 381 if (cb->args[5] != w->root->fn_sernum) { 382 /* Begin at the root if the tree changed */ 383 cb->args[5] = w->root->fn_sernum; 384 w->state = FWS_INIT; 385 w->node = w->root; 386 w->skip = w->count; 387 } else 388 w->skip = 0; 389 390 read_lock_bh(&table->tb6_lock); 391 res = fib6_walk_continue(w); 392 read_unlock_bh(&table->tb6_lock); 393 if (res <= 0) { 394 fib6_walker_unlink(net, w); 395 cb->args[4] = 0; 396 } 397 } 398 399 return res; 400 } 401 402 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 403 { 404 struct net *net = sock_net(skb->sk); 405 unsigned int h, s_h; 406 unsigned int e = 0, s_e; 407 struct rt6_rtnl_dump_arg arg; 408 struct fib6_walker *w; 409 struct fib6_table *tb; 410 struct hlist_head *head; 411 int res = 0; 412 413 s_h = cb->args[0]; 414 s_e = cb->args[1]; 415 416 w = (void *)cb->args[2]; 417 if (!w) { 418 /* New dump: 419 * 420 * 1. hook callback destructor. 421 */ 422 cb->args[3] = (long)cb->done; 423 cb->done = fib6_dump_done; 424 425 /* 426 * 2. allocate and initialize walker. 427 */ 428 w = kzalloc(sizeof(*w), GFP_ATOMIC); 429 if (!w) 430 return -ENOMEM; 431 w->func = fib6_dump_node; 432 cb->args[2] = (long)w; 433 } 434 435 arg.skb = skb; 436 arg.cb = cb; 437 arg.net = net; 438 w->args = &arg; 439 440 rcu_read_lock(); 441 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) { 442 e = 0; 443 head = &net->ipv6.fib_table_hash[h]; 444 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 445 if (e < s_e) 446 goto next; 447 res = fib6_dump_table(tb, skb, cb); 448 if (res != 0) 449 goto out; 450 next: 451 e++; 452 } 453 } 454 out: 455 rcu_read_unlock(); 456 cb->args[1] = e; 457 cb->args[0] = h; 458 459 res = res < 0 ? res : skb->len; 460 if (res <= 0) 461 fib6_dump_end(cb); 462 return res; 463 } 464 465 /* 466 * Routing Table 467 * 468 * return the appropriate node for a routing tree "add" operation 469 * by either creating and inserting or by returning an existing 470 * node. 471 */ 472 473 static struct fib6_node *fib6_add_1(struct fib6_node *root, 474 struct in6_addr *addr, int plen, 475 int offset, int allow_create, 476 int replace_required, int sernum) 477 { 478 struct fib6_node *fn, *in, *ln; 479 struct fib6_node *pn = NULL; 480 struct rt6key *key; 481 int bit; 482 __be32 dir = 0; 483 484 RT6_TRACE("fib6_add_1\n"); 485 486 /* insert node in tree */ 487 488 fn = root; 489 490 do { 491 key = (struct rt6key *)((u8 *)fn->leaf + offset); 492 493 /* 494 * Prefix match 495 */ 496 if (plen < fn->fn_bit || 497 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) { 498 if (!allow_create) { 499 if (replace_required) { 500 pr_warn("Can't replace route, no match found\n"); 501 return ERR_PTR(-ENOENT); 502 } 503 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 504 } 505 goto insert_above; 506 } 507 508 /* 509 * Exact match ? 510 */ 511 512 if (plen == fn->fn_bit) { 513 /* clean up an intermediate node */ 514 if (!(fn->fn_flags & RTN_RTINFO)) { 515 rt6_release(fn->leaf); 516 fn->leaf = NULL; 517 } 518 519 fn->fn_sernum = sernum; 520 521 return fn; 522 } 523 524 /* 525 * We have more bits to go 526 */ 527 528 /* Try to walk down on tree. */ 529 fn->fn_sernum = sernum; 530 dir = addr_bit_set(addr, fn->fn_bit); 531 pn = fn; 532 fn = dir ? fn->right : fn->left; 533 } while (fn); 534 535 if (!allow_create) { 536 /* We should not create new node because 537 * NLM_F_REPLACE was specified without NLM_F_CREATE 538 * I assume it is safe to require NLM_F_CREATE when 539 * REPLACE flag is used! Later we may want to remove the 540 * check for replace_required, because according 541 * to netlink specification, NLM_F_CREATE 542 * MUST be specified if new route is created. 543 * That would keep IPv6 consistent with IPv4 544 */ 545 if (replace_required) { 546 pr_warn("Can't replace route, no match found\n"); 547 return ERR_PTR(-ENOENT); 548 } 549 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 550 } 551 /* 552 * We walked to the bottom of tree. 553 * Create new leaf node without children. 554 */ 555 556 ln = node_alloc(); 557 558 if (!ln) 559 return ERR_PTR(-ENOMEM); 560 ln->fn_bit = plen; 561 562 ln->parent = pn; 563 ln->fn_sernum = sernum; 564 565 if (dir) 566 pn->right = ln; 567 else 568 pn->left = ln; 569 570 return ln; 571 572 573 insert_above: 574 /* 575 * split since we don't have a common prefix anymore or 576 * we have a less significant route. 577 * we've to insert an intermediate node on the list 578 * this new node will point to the one we need to create 579 * and the current 580 */ 581 582 pn = fn->parent; 583 584 /* find 1st bit in difference between the 2 addrs. 585 586 See comment in __ipv6_addr_diff: bit may be an invalid value, 587 but if it is >= plen, the value is ignored in any case. 588 */ 589 590 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr)); 591 592 /* 593 * (intermediate)[in] 594 * / \ 595 * (new leaf node)[ln] (old node)[fn] 596 */ 597 if (plen > bit) { 598 in = node_alloc(); 599 ln = node_alloc(); 600 601 if (!in || !ln) { 602 if (in) 603 node_free(in); 604 if (ln) 605 node_free(ln); 606 return ERR_PTR(-ENOMEM); 607 } 608 609 /* 610 * new intermediate node. 611 * RTN_RTINFO will 612 * be off since that an address that chooses one of 613 * the branches would not match less specific routes 614 * in the other branch 615 */ 616 617 in->fn_bit = bit; 618 619 in->parent = pn; 620 in->leaf = fn->leaf; 621 atomic_inc(&in->leaf->rt6i_ref); 622 623 in->fn_sernum = sernum; 624 625 /* update parent pointer */ 626 if (dir) 627 pn->right = in; 628 else 629 pn->left = in; 630 631 ln->fn_bit = plen; 632 633 ln->parent = in; 634 fn->parent = in; 635 636 ln->fn_sernum = sernum; 637 638 if (addr_bit_set(addr, bit)) { 639 in->right = ln; 640 in->left = fn; 641 } else { 642 in->left = ln; 643 in->right = fn; 644 } 645 } else { /* plen <= bit */ 646 647 /* 648 * (new leaf node)[ln] 649 * / \ 650 * (old node)[fn] NULL 651 */ 652 653 ln = node_alloc(); 654 655 if (!ln) 656 return ERR_PTR(-ENOMEM); 657 658 ln->fn_bit = plen; 659 660 ln->parent = pn; 661 662 ln->fn_sernum = sernum; 663 664 if (dir) 665 pn->right = ln; 666 else 667 pn->left = ln; 668 669 if (addr_bit_set(&key->addr, plen)) 670 ln->right = fn; 671 else 672 ln->left = fn; 673 674 fn->parent = ln; 675 } 676 return ln; 677 } 678 679 static bool rt6_qualify_for_ecmp(struct rt6_info *rt) 680 { 681 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) == 682 RTF_GATEWAY; 683 } 684 685 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc) 686 { 687 int i; 688 689 for (i = 0; i < RTAX_MAX; i++) { 690 if (test_bit(i, mxc->mx_valid)) 691 mp[i] = mxc->mx[i]; 692 } 693 } 694 695 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc) 696 { 697 if (!mxc->mx) 698 return 0; 699 700 if (dst->flags & DST_HOST) { 701 u32 *mp = dst_metrics_write_ptr(dst); 702 703 if (unlikely(!mp)) 704 return -ENOMEM; 705 706 fib6_copy_metrics(mp, mxc); 707 } else { 708 dst_init_metrics(dst, mxc->mx, false); 709 710 /* We've stolen mx now. */ 711 mxc->mx = NULL; 712 } 713 714 return 0; 715 } 716 717 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn, 718 struct net *net) 719 { 720 if (atomic_read(&rt->rt6i_ref) != 1) { 721 /* This route is used as dummy address holder in some split 722 * nodes. It is not leaked, but it still holds other resources, 723 * which must be released in time. So, scan ascendant nodes 724 * and replace dummy references to this route with references 725 * to still alive ones. 726 */ 727 while (fn) { 728 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) { 729 fn->leaf = fib6_find_prefix(net, fn); 730 atomic_inc(&fn->leaf->rt6i_ref); 731 rt6_release(rt); 732 } 733 fn = fn->parent; 734 } 735 /* No more references are possible at this point. */ 736 BUG_ON(atomic_read(&rt->rt6i_ref) != 1); 737 } 738 } 739 740 /* 741 * Insert routing information in a node. 742 */ 743 744 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, 745 struct nl_info *info, struct mx6_config *mxc) 746 { 747 struct rt6_info *iter = NULL; 748 struct rt6_info **ins; 749 struct rt6_info **fallback_ins = NULL; 750 int replace = (info->nlh && 751 (info->nlh->nlmsg_flags & NLM_F_REPLACE)); 752 int add = (!info->nlh || 753 (info->nlh->nlmsg_flags & NLM_F_CREATE)); 754 int found = 0; 755 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt); 756 u16 nlflags = NLM_F_EXCL; 757 int err; 758 759 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND)) 760 nlflags |= NLM_F_APPEND; 761 762 ins = &fn->leaf; 763 764 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) { 765 /* 766 * Search for duplicates 767 */ 768 769 if (iter->rt6i_metric == rt->rt6i_metric) { 770 /* 771 * Same priority level 772 */ 773 if (info->nlh && 774 (info->nlh->nlmsg_flags & NLM_F_EXCL)) 775 return -EEXIST; 776 777 nlflags &= ~NLM_F_EXCL; 778 if (replace) { 779 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) { 780 found++; 781 break; 782 } 783 if (rt_can_ecmp) 784 fallback_ins = fallback_ins ?: ins; 785 goto next_iter; 786 } 787 788 if (iter->dst.dev == rt->dst.dev && 789 iter->rt6i_idev == rt->rt6i_idev && 790 ipv6_addr_equal(&iter->rt6i_gateway, 791 &rt->rt6i_gateway)) { 792 if (rt->rt6i_nsiblings) 793 rt->rt6i_nsiblings = 0; 794 if (!(iter->rt6i_flags & RTF_EXPIRES)) 795 return -EEXIST; 796 if (!(rt->rt6i_flags & RTF_EXPIRES)) 797 rt6_clean_expires(iter); 798 else 799 rt6_set_expires(iter, rt->dst.expires); 800 iter->rt6i_pmtu = rt->rt6i_pmtu; 801 return -EEXIST; 802 } 803 /* If we have the same destination and the same metric, 804 * but not the same gateway, then the route we try to 805 * add is sibling to this route, increment our counter 806 * of siblings, and later we will add our route to the 807 * list. 808 * Only static routes (which don't have flag 809 * RTF_EXPIRES) are used for ECMPv6. 810 * 811 * To avoid long list, we only had siblings if the 812 * route have a gateway. 813 */ 814 if (rt_can_ecmp && 815 rt6_qualify_for_ecmp(iter)) 816 rt->rt6i_nsiblings++; 817 } 818 819 if (iter->rt6i_metric > rt->rt6i_metric) 820 break; 821 822 next_iter: 823 ins = &iter->dst.rt6_next; 824 } 825 826 if (fallback_ins && !found) { 827 /* No ECMP-able route found, replace first non-ECMP one */ 828 ins = fallback_ins; 829 iter = *ins; 830 found++; 831 } 832 833 /* Reset round-robin state, if necessary */ 834 if (ins == &fn->leaf) 835 fn->rr_ptr = NULL; 836 837 /* Link this route to others same route. */ 838 if (rt->rt6i_nsiblings) { 839 unsigned int rt6i_nsiblings; 840 struct rt6_info *sibling, *temp_sibling; 841 842 /* Find the first route that have the same metric */ 843 sibling = fn->leaf; 844 while (sibling) { 845 if (sibling->rt6i_metric == rt->rt6i_metric && 846 rt6_qualify_for_ecmp(sibling)) { 847 list_add_tail(&rt->rt6i_siblings, 848 &sibling->rt6i_siblings); 849 break; 850 } 851 sibling = sibling->dst.rt6_next; 852 } 853 /* For each sibling in the list, increment the counter of 854 * siblings. BUG() if counters does not match, list of siblings 855 * is broken! 856 */ 857 rt6i_nsiblings = 0; 858 list_for_each_entry_safe(sibling, temp_sibling, 859 &rt->rt6i_siblings, rt6i_siblings) { 860 sibling->rt6i_nsiblings++; 861 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings); 862 rt6i_nsiblings++; 863 } 864 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings); 865 } 866 867 /* 868 * insert node 869 */ 870 if (!replace) { 871 if (!add) 872 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 873 874 add: 875 nlflags |= NLM_F_CREATE; 876 err = fib6_commit_metrics(&rt->dst, mxc); 877 if (err) 878 return err; 879 880 rt->dst.rt6_next = iter; 881 *ins = rt; 882 rt->rt6i_node = fn; 883 atomic_inc(&rt->rt6i_ref); 884 if (!info->skip_notify) 885 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags); 886 info->nl_net->ipv6.rt6_stats->fib_rt_entries++; 887 888 if (!(fn->fn_flags & RTN_RTINFO)) { 889 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 890 fn->fn_flags |= RTN_RTINFO; 891 } 892 893 } else { 894 int nsiblings; 895 896 if (!found) { 897 if (add) 898 goto add; 899 pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); 900 return -ENOENT; 901 } 902 903 err = fib6_commit_metrics(&rt->dst, mxc); 904 if (err) 905 return err; 906 907 *ins = rt; 908 rt->rt6i_node = fn; 909 rt->dst.rt6_next = iter->dst.rt6_next; 910 atomic_inc(&rt->rt6i_ref); 911 if (!info->skip_notify) 912 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE); 913 if (!(fn->fn_flags & RTN_RTINFO)) { 914 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 915 fn->fn_flags |= RTN_RTINFO; 916 } 917 nsiblings = iter->rt6i_nsiblings; 918 fib6_purge_rt(iter, fn, info->nl_net); 919 rt6_release(iter); 920 921 if (nsiblings) { 922 /* Replacing an ECMP route, remove all siblings */ 923 ins = &rt->dst.rt6_next; 924 iter = *ins; 925 while (iter) { 926 if (iter->rt6i_metric > rt->rt6i_metric) 927 break; 928 if (rt6_qualify_for_ecmp(iter)) { 929 *ins = iter->dst.rt6_next; 930 fib6_purge_rt(iter, fn, info->nl_net); 931 rt6_release(iter); 932 nsiblings--; 933 } else { 934 ins = &iter->dst.rt6_next; 935 } 936 iter = *ins; 937 } 938 WARN_ON(nsiblings != 0); 939 } 940 } 941 942 return 0; 943 } 944 945 static void fib6_start_gc(struct net *net, struct rt6_info *rt) 946 { 947 if (!timer_pending(&net->ipv6.ip6_fib_timer) && 948 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE))) 949 mod_timer(&net->ipv6.ip6_fib_timer, 950 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 951 } 952 953 void fib6_force_start_gc(struct net *net) 954 { 955 if (!timer_pending(&net->ipv6.ip6_fib_timer)) 956 mod_timer(&net->ipv6.ip6_fib_timer, 957 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 958 } 959 960 /* 961 * Add routing information to the routing tree. 962 * <destination addr>/<source addr> 963 * with source addr info in sub-trees 964 */ 965 966 int fib6_add(struct fib6_node *root, struct rt6_info *rt, 967 struct nl_info *info, struct mx6_config *mxc) 968 { 969 struct fib6_node *fn, *pn = NULL; 970 int err = -ENOMEM; 971 int allow_create = 1; 972 int replace_required = 0; 973 int sernum = fib6_new_sernum(info->nl_net); 974 975 if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) && 976 !atomic_read(&rt->dst.__refcnt))) 977 return -EINVAL; 978 979 if (info->nlh) { 980 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) 981 allow_create = 0; 982 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) 983 replace_required = 1; 984 } 985 if (!allow_create && !replace_required) 986 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); 987 988 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen, 989 offsetof(struct rt6_info, rt6i_dst), allow_create, 990 replace_required, sernum); 991 if (IS_ERR(fn)) { 992 err = PTR_ERR(fn); 993 fn = NULL; 994 goto out; 995 } 996 997 pn = fn; 998 999 #ifdef CONFIG_IPV6_SUBTREES 1000 if (rt->rt6i_src.plen) { 1001 struct fib6_node *sn; 1002 1003 if (!fn->subtree) { 1004 struct fib6_node *sfn; 1005 1006 /* 1007 * Create subtree. 1008 * 1009 * fn[main tree] 1010 * | 1011 * sfn[subtree root] 1012 * \ 1013 * sn[new leaf node] 1014 */ 1015 1016 /* Create subtree root node */ 1017 sfn = node_alloc(); 1018 if (!sfn) 1019 goto st_failure; 1020 1021 sfn->leaf = info->nl_net->ipv6.ip6_null_entry; 1022 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref); 1023 sfn->fn_flags = RTN_ROOT; 1024 sfn->fn_sernum = sernum; 1025 1026 /* Now add the first leaf node to new subtree */ 1027 1028 sn = fib6_add_1(sfn, &rt->rt6i_src.addr, 1029 rt->rt6i_src.plen, 1030 offsetof(struct rt6_info, rt6i_src), 1031 allow_create, replace_required, sernum); 1032 1033 if (IS_ERR(sn)) { 1034 /* If it is failed, discard just allocated 1035 root, and then (in st_failure) stale node 1036 in main tree. 1037 */ 1038 node_free(sfn); 1039 err = PTR_ERR(sn); 1040 goto st_failure; 1041 } 1042 1043 /* Now link new subtree to main tree */ 1044 sfn->parent = fn; 1045 fn->subtree = sfn; 1046 } else { 1047 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, 1048 rt->rt6i_src.plen, 1049 offsetof(struct rt6_info, rt6i_src), 1050 allow_create, replace_required, sernum); 1051 1052 if (IS_ERR(sn)) { 1053 err = PTR_ERR(sn); 1054 goto st_failure; 1055 } 1056 } 1057 1058 if (!fn->leaf) { 1059 fn->leaf = rt; 1060 atomic_inc(&rt->rt6i_ref); 1061 } 1062 fn = sn; 1063 } 1064 #endif 1065 1066 err = fib6_add_rt2node(fn, rt, info, mxc); 1067 if (!err) { 1068 fib6_start_gc(info->nl_net, rt); 1069 if (!(rt->rt6i_flags & RTF_CACHE)) 1070 fib6_prune_clones(info->nl_net, pn); 1071 rt->dst.flags &= ~DST_NOCACHE; 1072 } 1073 1074 out: 1075 if (err) { 1076 #ifdef CONFIG_IPV6_SUBTREES 1077 /* 1078 * If fib6_add_1 has cleared the old leaf pointer in the 1079 * super-tree leaf node we have to find a new one for it. 1080 */ 1081 if (pn != fn && pn->leaf == rt) { 1082 pn->leaf = NULL; 1083 atomic_dec(&rt->rt6i_ref); 1084 } 1085 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) { 1086 pn->leaf = fib6_find_prefix(info->nl_net, pn); 1087 #if RT6_DEBUG >= 2 1088 if (!pn->leaf) { 1089 WARN_ON(pn->leaf == NULL); 1090 pn->leaf = info->nl_net->ipv6.ip6_null_entry; 1091 } 1092 #endif 1093 atomic_inc(&pn->leaf->rt6i_ref); 1094 } 1095 #endif 1096 if (!(rt->dst.flags & DST_NOCACHE)) 1097 dst_free(&rt->dst); 1098 } 1099 return err; 1100 1101 #ifdef CONFIG_IPV6_SUBTREES 1102 /* Subtree creation failed, probably main tree node 1103 is orphan. If it is, shoot it. 1104 */ 1105 st_failure: 1106 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) 1107 fib6_repair_tree(info->nl_net, fn); 1108 if (!(rt->dst.flags & DST_NOCACHE)) 1109 dst_free(&rt->dst); 1110 return err; 1111 #endif 1112 } 1113 1114 /* 1115 * Routing tree lookup 1116 * 1117 */ 1118 1119 struct lookup_args { 1120 int offset; /* key offset on rt6_info */ 1121 const struct in6_addr *addr; /* search key */ 1122 }; 1123 1124 static struct fib6_node *fib6_lookup_1(struct fib6_node *root, 1125 struct lookup_args *args) 1126 { 1127 struct fib6_node *fn; 1128 __be32 dir; 1129 1130 if (unlikely(args->offset == 0)) 1131 return NULL; 1132 1133 /* 1134 * Descend on a tree 1135 */ 1136 1137 fn = root; 1138 1139 for (;;) { 1140 struct fib6_node *next; 1141 1142 dir = addr_bit_set(args->addr, fn->fn_bit); 1143 1144 next = dir ? fn->right : fn->left; 1145 1146 if (next) { 1147 fn = next; 1148 continue; 1149 } 1150 break; 1151 } 1152 1153 while (fn) { 1154 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) { 1155 struct rt6key *key; 1156 1157 key = (struct rt6key *) ((u8 *) fn->leaf + 1158 args->offset); 1159 1160 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 1161 #ifdef CONFIG_IPV6_SUBTREES 1162 if (fn->subtree) { 1163 struct fib6_node *sfn; 1164 sfn = fib6_lookup_1(fn->subtree, 1165 args + 1); 1166 if (!sfn) 1167 goto backtrack; 1168 fn = sfn; 1169 } 1170 #endif 1171 if (fn->fn_flags & RTN_RTINFO) 1172 return fn; 1173 } 1174 } 1175 #ifdef CONFIG_IPV6_SUBTREES 1176 backtrack: 1177 #endif 1178 if (fn->fn_flags & RTN_ROOT) 1179 break; 1180 1181 fn = fn->parent; 1182 } 1183 1184 return NULL; 1185 } 1186 1187 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr, 1188 const struct in6_addr *saddr) 1189 { 1190 struct fib6_node *fn; 1191 struct lookup_args args[] = { 1192 { 1193 .offset = offsetof(struct rt6_info, rt6i_dst), 1194 .addr = daddr, 1195 }, 1196 #ifdef CONFIG_IPV6_SUBTREES 1197 { 1198 .offset = offsetof(struct rt6_info, rt6i_src), 1199 .addr = saddr, 1200 }, 1201 #endif 1202 { 1203 .offset = 0, /* sentinel */ 1204 } 1205 }; 1206 1207 fn = fib6_lookup_1(root, daddr ? args : args + 1); 1208 if (!fn || fn->fn_flags & RTN_TL_ROOT) 1209 fn = root; 1210 1211 return fn; 1212 } 1213 1214 /* 1215 * Get node with specified destination prefix (and source prefix, 1216 * if subtrees are used) 1217 */ 1218 1219 1220 static struct fib6_node *fib6_locate_1(struct fib6_node *root, 1221 const struct in6_addr *addr, 1222 int plen, int offset) 1223 { 1224 struct fib6_node *fn; 1225 1226 for (fn = root; fn ; ) { 1227 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); 1228 1229 /* 1230 * Prefix match 1231 */ 1232 if (plen < fn->fn_bit || 1233 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 1234 return NULL; 1235 1236 if (plen == fn->fn_bit) 1237 return fn; 1238 1239 /* 1240 * We have more bits to go 1241 */ 1242 if (addr_bit_set(addr, fn->fn_bit)) 1243 fn = fn->right; 1244 else 1245 fn = fn->left; 1246 } 1247 return NULL; 1248 } 1249 1250 struct fib6_node *fib6_locate(struct fib6_node *root, 1251 const struct in6_addr *daddr, int dst_len, 1252 const struct in6_addr *saddr, int src_len) 1253 { 1254 struct fib6_node *fn; 1255 1256 fn = fib6_locate_1(root, daddr, dst_len, 1257 offsetof(struct rt6_info, rt6i_dst)); 1258 1259 #ifdef CONFIG_IPV6_SUBTREES 1260 if (src_len) { 1261 WARN_ON(saddr == NULL); 1262 if (fn && fn->subtree) 1263 fn = fib6_locate_1(fn->subtree, saddr, src_len, 1264 offsetof(struct rt6_info, rt6i_src)); 1265 } 1266 #endif 1267 1268 if (fn && fn->fn_flags & RTN_RTINFO) 1269 return fn; 1270 1271 return NULL; 1272 } 1273 1274 1275 /* 1276 * Deletion 1277 * 1278 */ 1279 1280 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn) 1281 { 1282 if (fn->fn_flags & RTN_ROOT) 1283 return net->ipv6.ip6_null_entry; 1284 1285 while (fn) { 1286 if (fn->left) 1287 return fn->left->leaf; 1288 if (fn->right) 1289 return fn->right->leaf; 1290 1291 fn = FIB6_SUBTREE(fn); 1292 } 1293 return NULL; 1294 } 1295 1296 /* 1297 * Called to trim the tree of intermediate nodes when possible. "fn" 1298 * is the node we want to try and remove. 1299 */ 1300 1301 static struct fib6_node *fib6_repair_tree(struct net *net, 1302 struct fib6_node *fn) 1303 { 1304 int children; 1305 int nstate; 1306 struct fib6_node *child, *pn; 1307 struct fib6_walker *w; 1308 int iter = 0; 1309 1310 for (;;) { 1311 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1312 iter++; 1313 1314 WARN_ON(fn->fn_flags & RTN_RTINFO); 1315 WARN_ON(fn->fn_flags & RTN_TL_ROOT); 1316 WARN_ON(fn->leaf); 1317 1318 children = 0; 1319 child = NULL; 1320 if (fn->right) 1321 child = fn->right, children |= 1; 1322 if (fn->left) 1323 child = fn->left, children |= 2; 1324 1325 if (children == 3 || FIB6_SUBTREE(fn) 1326 #ifdef CONFIG_IPV6_SUBTREES 1327 /* Subtree root (i.e. fn) may have one child */ 1328 || (children && fn->fn_flags & RTN_ROOT) 1329 #endif 1330 ) { 1331 fn->leaf = fib6_find_prefix(net, fn); 1332 #if RT6_DEBUG >= 2 1333 if (!fn->leaf) { 1334 WARN_ON(!fn->leaf); 1335 fn->leaf = net->ipv6.ip6_null_entry; 1336 } 1337 #endif 1338 atomic_inc(&fn->leaf->rt6i_ref); 1339 return fn->parent; 1340 } 1341 1342 pn = fn->parent; 1343 #ifdef CONFIG_IPV6_SUBTREES 1344 if (FIB6_SUBTREE(pn) == fn) { 1345 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1346 FIB6_SUBTREE(pn) = NULL; 1347 nstate = FWS_L; 1348 } else { 1349 WARN_ON(fn->fn_flags & RTN_ROOT); 1350 #endif 1351 if (pn->right == fn) 1352 pn->right = child; 1353 else if (pn->left == fn) 1354 pn->left = child; 1355 #if RT6_DEBUG >= 2 1356 else 1357 WARN_ON(1); 1358 #endif 1359 if (child) 1360 child->parent = pn; 1361 nstate = FWS_R; 1362 #ifdef CONFIG_IPV6_SUBTREES 1363 } 1364 #endif 1365 1366 read_lock(&net->ipv6.fib6_walker_lock); 1367 FOR_WALKERS(net, w) { 1368 if (!child) { 1369 if (w->root == fn) { 1370 w->root = w->node = NULL; 1371 RT6_TRACE("W %p adjusted by delroot 1\n", w); 1372 } else if (w->node == fn) { 1373 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1374 w->node = pn; 1375 w->state = nstate; 1376 } 1377 } else { 1378 if (w->root == fn) { 1379 w->root = child; 1380 RT6_TRACE("W %p adjusted by delroot 2\n", w); 1381 } 1382 if (w->node == fn) { 1383 w->node = child; 1384 if (children&2) { 1385 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1386 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT; 1387 } else { 1388 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1389 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT; 1390 } 1391 } 1392 } 1393 } 1394 read_unlock(&net->ipv6.fib6_walker_lock); 1395 1396 node_free(fn); 1397 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) 1398 return pn; 1399 1400 rt6_release(pn->leaf); 1401 pn->leaf = NULL; 1402 fn = pn; 1403 } 1404 } 1405 1406 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, 1407 struct nl_info *info) 1408 { 1409 struct fib6_walker *w; 1410 struct rt6_info *rt = *rtp; 1411 struct net *net = info->nl_net; 1412 1413 RT6_TRACE("fib6_del_route\n"); 1414 1415 /* Unlink it */ 1416 *rtp = rt->dst.rt6_next; 1417 rt->rt6i_node = NULL; 1418 net->ipv6.rt6_stats->fib_rt_entries--; 1419 net->ipv6.rt6_stats->fib_discarded_routes++; 1420 1421 /* Reset round-robin state, if necessary */ 1422 if (fn->rr_ptr == rt) 1423 fn->rr_ptr = NULL; 1424 1425 /* Remove this entry from other siblings */ 1426 if (rt->rt6i_nsiblings) { 1427 struct rt6_info *sibling, *next_sibling; 1428 1429 list_for_each_entry_safe(sibling, next_sibling, 1430 &rt->rt6i_siblings, rt6i_siblings) 1431 sibling->rt6i_nsiblings--; 1432 rt->rt6i_nsiblings = 0; 1433 list_del_init(&rt->rt6i_siblings); 1434 } 1435 1436 /* Adjust walkers */ 1437 read_lock(&net->ipv6.fib6_walker_lock); 1438 FOR_WALKERS(net, w) { 1439 if (w->state == FWS_C && w->leaf == rt) { 1440 RT6_TRACE("walker %p adjusted by delroute\n", w); 1441 w->leaf = rt->dst.rt6_next; 1442 if (!w->leaf) 1443 w->state = FWS_U; 1444 } 1445 } 1446 read_unlock(&net->ipv6.fib6_walker_lock); 1447 1448 rt->dst.rt6_next = NULL; 1449 1450 /* If it was last route, expunge its radix tree node */ 1451 if (!fn->leaf) { 1452 fn->fn_flags &= ~RTN_RTINFO; 1453 net->ipv6.rt6_stats->fib_route_nodes--; 1454 fn = fib6_repair_tree(net, fn); 1455 } 1456 1457 fib6_purge_rt(rt, fn, net); 1458 1459 if (!info->skip_notify) 1460 inet6_rt_notify(RTM_DELROUTE, rt, info, 0); 1461 rt6_release(rt); 1462 } 1463 1464 int fib6_del(struct rt6_info *rt, struct nl_info *info) 1465 { 1466 struct net *net = info->nl_net; 1467 struct fib6_node *fn = rt->rt6i_node; 1468 struct rt6_info **rtp; 1469 1470 #if RT6_DEBUG >= 2 1471 if (rt->dst.obsolete > 0) { 1472 WARN_ON(fn); 1473 return -ENOENT; 1474 } 1475 #endif 1476 if (!fn || rt == net->ipv6.ip6_null_entry) 1477 return -ENOENT; 1478 1479 WARN_ON(!(fn->fn_flags & RTN_RTINFO)); 1480 1481 if (!(rt->rt6i_flags & RTF_CACHE)) { 1482 struct fib6_node *pn = fn; 1483 #ifdef CONFIG_IPV6_SUBTREES 1484 /* clones of this route might be in another subtree */ 1485 if (rt->rt6i_src.plen) { 1486 while (!(pn->fn_flags & RTN_ROOT)) 1487 pn = pn->parent; 1488 pn = pn->parent; 1489 } 1490 #endif 1491 fib6_prune_clones(info->nl_net, pn); 1492 } 1493 1494 /* 1495 * Walk the leaf entries looking for ourself 1496 */ 1497 1498 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) { 1499 if (*rtp == rt) { 1500 fib6_del_route(fn, rtp, info); 1501 return 0; 1502 } 1503 } 1504 return -ENOENT; 1505 } 1506 1507 /* 1508 * Tree traversal function. 1509 * 1510 * Certainly, it is not interrupt safe. 1511 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 1512 * It means, that we can modify tree during walking 1513 * and use this function for garbage collection, clone pruning, 1514 * cleaning tree when a device goes down etc. etc. 1515 * 1516 * It guarantees that every node will be traversed, 1517 * and that it will be traversed only once. 1518 * 1519 * Callback function w->func may return: 1520 * 0 -> continue walking. 1521 * positive value -> walking is suspended (used by tree dumps, 1522 * and probably by gc, if it will be split to several slices) 1523 * negative value -> terminate walking. 1524 * 1525 * The function itself returns: 1526 * 0 -> walk is complete. 1527 * >0 -> walk is incomplete (i.e. suspended) 1528 * <0 -> walk is terminated by an error. 1529 */ 1530 1531 static int fib6_walk_continue(struct fib6_walker *w) 1532 { 1533 struct fib6_node *fn, *pn; 1534 1535 for (;;) { 1536 fn = w->node; 1537 if (!fn) 1538 return 0; 1539 1540 if (w->prune && fn != w->root && 1541 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) { 1542 w->state = FWS_C; 1543 w->leaf = fn->leaf; 1544 } 1545 switch (w->state) { 1546 #ifdef CONFIG_IPV6_SUBTREES 1547 case FWS_S: 1548 if (FIB6_SUBTREE(fn)) { 1549 w->node = FIB6_SUBTREE(fn); 1550 continue; 1551 } 1552 w->state = FWS_L; 1553 #endif 1554 case FWS_L: 1555 if (fn->left) { 1556 w->node = fn->left; 1557 w->state = FWS_INIT; 1558 continue; 1559 } 1560 w->state = FWS_R; 1561 case FWS_R: 1562 if (fn->right) { 1563 w->node = fn->right; 1564 w->state = FWS_INIT; 1565 continue; 1566 } 1567 w->state = FWS_C; 1568 w->leaf = fn->leaf; 1569 case FWS_C: 1570 if (w->leaf && fn->fn_flags & RTN_RTINFO) { 1571 int err; 1572 1573 if (w->skip) { 1574 w->skip--; 1575 goto skip; 1576 } 1577 1578 err = w->func(w); 1579 if (err) 1580 return err; 1581 1582 w->count++; 1583 continue; 1584 } 1585 skip: 1586 w->state = FWS_U; 1587 case FWS_U: 1588 if (fn == w->root) 1589 return 0; 1590 pn = fn->parent; 1591 w->node = pn; 1592 #ifdef CONFIG_IPV6_SUBTREES 1593 if (FIB6_SUBTREE(pn) == fn) { 1594 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1595 w->state = FWS_L; 1596 continue; 1597 } 1598 #endif 1599 if (pn->left == fn) { 1600 w->state = FWS_R; 1601 continue; 1602 } 1603 if (pn->right == fn) { 1604 w->state = FWS_C; 1605 w->leaf = w->node->leaf; 1606 continue; 1607 } 1608 #if RT6_DEBUG >= 2 1609 WARN_ON(1); 1610 #endif 1611 } 1612 } 1613 } 1614 1615 static int fib6_walk(struct net *net, struct fib6_walker *w) 1616 { 1617 int res; 1618 1619 w->state = FWS_INIT; 1620 w->node = w->root; 1621 1622 fib6_walker_link(net, w); 1623 res = fib6_walk_continue(w); 1624 if (res <= 0) 1625 fib6_walker_unlink(net, w); 1626 return res; 1627 } 1628 1629 static int fib6_clean_node(struct fib6_walker *w) 1630 { 1631 int res; 1632 struct rt6_info *rt; 1633 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w); 1634 struct nl_info info = { 1635 .nl_net = c->net, 1636 }; 1637 1638 if (c->sernum != FIB6_NO_SERNUM_CHANGE && 1639 w->node->fn_sernum != c->sernum) 1640 w->node->fn_sernum = c->sernum; 1641 1642 if (!c->func) { 1643 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE); 1644 w->leaf = NULL; 1645 return 0; 1646 } 1647 1648 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 1649 res = c->func(rt, c->arg); 1650 if (res < 0) { 1651 w->leaf = rt; 1652 res = fib6_del(rt, &info); 1653 if (res) { 1654 #if RT6_DEBUG >= 2 1655 pr_debug("%s: del failed: rt=%p@%p err=%d\n", 1656 __func__, rt, rt->rt6i_node, res); 1657 #endif 1658 continue; 1659 } 1660 return 0; 1661 } 1662 WARN_ON(res != 0); 1663 } 1664 w->leaf = rt; 1665 return 0; 1666 } 1667 1668 /* 1669 * Convenient frontend to tree walker. 1670 * 1671 * func is called on each route. 1672 * It may return -1 -> delete this route. 1673 * 0 -> continue walking 1674 * 1675 * prune==1 -> only immediate children of node (certainly, 1676 * ignoring pure split nodes) will be scanned. 1677 */ 1678 1679 static void fib6_clean_tree(struct net *net, struct fib6_node *root, 1680 int (*func)(struct rt6_info *, void *arg), 1681 bool prune, int sernum, void *arg) 1682 { 1683 struct fib6_cleaner c; 1684 1685 c.w.root = root; 1686 c.w.func = fib6_clean_node; 1687 c.w.prune = prune; 1688 c.w.count = 0; 1689 c.w.skip = 0; 1690 c.func = func; 1691 c.sernum = sernum; 1692 c.arg = arg; 1693 c.net = net; 1694 1695 fib6_walk(net, &c.w); 1696 } 1697 1698 static void __fib6_clean_all(struct net *net, 1699 int (*func)(struct rt6_info *, void *), 1700 int sernum, void *arg) 1701 { 1702 struct fib6_table *table; 1703 struct hlist_head *head; 1704 unsigned int h; 1705 1706 rcu_read_lock(); 1707 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 1708 head = &net->ipv6.fib_table_hash[h]; 1709 hlist_for_each_entry_rcu(table, head, tb6_hlist) { 1710 write_lock_bh(&table->tb6_lock); 1711 fib6_clean_tree(net, &table->tb6_root, 1712 func, false, sernum, arg); 1713 write_unlock_bh(&table->tb6_lock); 1714 } 1715 } 1716 rcu_read_unlock(); 1717 } 1718 1719 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *), 1720 void *arg) 1721 { 1722 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg); 1723 } 1724 1725 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1726 { 1727 if (rt->rt6i_flags & RTF_CACHE) { 1728 RT6_TRACE("pruning clone %p\n", rt); 1729 return -1; 1730 } 1731 1732 return 0; 1733 } 1734 1735 static void fib6_prune_clones(struct net *net, struct fib6_node *fn) 1736 { 1737 fib6_clean_tree(net, fn, fib6_prune_clone, true, 1738 FIB6_NO_SERNUM_CHANGE, NULL); 1739 } 1740 1741 static void fib6_flush_trees(struct net *net) 1742 { 1743 int new_sernum = fib6_new_sernum(net); 1744 1745 __fib6_clean_all(net, NULL, new_sernum, NULL); 1746 } 1747 1748 /* 1749 * Garbage collection 1750 */ 1751 1752 struct fib6_gc_args 1753 { 1754 int timeout; 1755 int more; 1756 }; 1757 1758 static int fib6_age(struct rt6_info *rt, void *arg) 1759 { 1760 struct fib6_gc_args *gc_args = arg; 1761 unsigned long now = jiffies; 1762 1763 /* 1764 * check addrconf expiration here. 1765 * Routes are expired even if they are in use. 1766 * 1767 * Also age clones. Note, that clones are aged out 1768 * only if they are not in use now. 1769 */ 1770 1771 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) { 1772 if (time_after(now, rt->dst.expires)) { 1773 RT6_TRACE("expiring %p\n", rt); 1774 return -1; 1775 } 1776 gc_args->more++; 1777 } else if (rt->rt6i_flags & RTF_CACHE) { 1778 if (atomic_read(&rt->dst.__refcnt) == 0 && 1779 time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) { 1780 RT6_TRACE("aging clone %p\n", rt); 1781 return -1; 1782 } else if (rt->rt6i_flags & RTF_GATEWAY) { 1783 struct neighbour *neigh; 1784 __u8 neigh_flags = 0; 1785 1786 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway); 1787 if (neigh) { 1788 neigh_flags = neigh->flags; 1789 neigh_release(neigh); 1790 } 1791 if (!(neigh_flags & NTF_ROUTER)) { 1792 RT6_TRACE("purging route %p via non-router but gateway\n", 1793 rt); 1794 return -1; 1795 } 1796 } 1797 gc_args->more++; 1798 } 1799 1800 return 0; 1801 } 1802 1803 void fib6_run_gc(unsigned long expires, struct net *net, bool force) 1804 { 1805 struct fib6_gc_args gc_args; 1806 unsigned long now; 1807 1808 if (force) { 1809 spin_lock_bh(&net->ipv6.fib6_gc_lock); 1810 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) { 1811 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); 1812 return; 1813 } 1814 gc_args.timeout = expires ? (int)expires : 1815 net->ipv6.sysctl.ip6_rt_gc_interval; 1816 1817 gc_args.more = icmp6_dst_gc(); 1818 1819 fib6_clean_all(net, fib6_age, &gc_args); 1820 now = jiffies; 1821 net->ipv6.ip6_rt_last_gc = now; 1822 1823 if (gc_args.more) 1824 mod_timer(&net->ipv6.ip6_fib_timer, 1825 round_jiffies(now 1826 + net->ipv6.sysctl.ip6_rt_gc_interval)); 1827 else 1828 del_timer(&net->ipv6.ip6_fib_timer); 1829 spin_unlock_bh(&net->ipv6.fib6_gc_lock); 1830 } 1831 1832 static void fib6_gc_timer_cb(unsigned long arg) 1833 { 1834 fib6_run_gc(0, (struct net *)arg, true); 1835 } 1836 1837 static int __net_init fib6_net_init(struct net *net) 1838 { 1839 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; 1840 1841 spin_lock_init(&net->ipv6.fib6_gc_lock); 1842 rwlock_init(&net->ipv6.fib6_walker_lock); 1843 INIT_LIST_HEAD(&net->ipv6.fib6_walkers); 1844 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net); 1845 1846 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); 1847 if (!net->ipv6.rt6_stats) 1848 goto out_timer; 1849 1850 /* Avoid false sharing : Use at least a full cache line */ 1851 size = max_t(size_t, size, L1_CACHE_BYTES); 1852 1853 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); 1854 if (!net->ipv6.fib_table_hash) 1855 goto out_rt6_stats; 1856 1857 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), 1858 GFP_KERNEL); 1859 if (!net->ipv6.fib6_main_tbl) 1860 goto out_fib_table_hash; 1861 1862 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; 1863 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1864 net->ipv6.fib6_main_tbl->tb6_root.fn_flags = 1865 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1866 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); 1867 1868 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1869 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), 1870 GFP_KERNEL); 1871 if (!net->ipv6.fib6_local_tbl) 1872 goto out_fib6_main_tbl; 1873 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; 1874 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1875 net->ipv6.fib6_local_tbl->tb6_root.fn_flags = 1876 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1877 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); 1878 #endif 1879 fib6_tables_init(net); 1880 1881 return 0; 1882 1883 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1884 out_fib6_main_tbl: 1885 kfree(net->ipv6.fib6_main_tbl); 1886 #endif 1887 out_fib_table_hash: 1888 kfree(net->ipv6.fib_table_hash); 1889 out_rt6_stats: 1890 kfree(net->ipv6.rt6_stats); 1891 out_timer: 1892 return -ENOMEM; 1893 } 1894 1895 static void fib6_net_exit(struct net *net) 1896 { 1897 rt6_ifdown(net, NULL); 1898 del_timer_sync(&net->ipv6.ip6_fib_timer); 1899 1900 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1901 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers); 1902 kfree(net->ipv6.fib6_local_tbl); 1903 #endif 1904 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers); 1905 kfree(net->ipv6.fib6_main_tbl); 1906 kfree(net->ipv6.fib_table_hash); 1907 kfree(net->ipv6.rt6_stats); 1908 } 1909 1910 static struct pernet_operations fib6_net_ops = { 1911 .init = fib6_net_init, 1912 .exit = fib6_net_exit, 1913 }; 1914 1915 int __init fib6_init(void) 1916 { 1917 int ret = -ENOMEM; 1918 1919 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1920 sizeof(struct fib6_node), 1921 0, SLAB_HWCACHE_ALIGN, 1922 NULL); 1923 if (!fib6_node_kmem) 1924 goto out; 1925 1926 ret = register_pernet_subsys(&fib6_net_ops); 1927 if (ret) 1928 goto out_kmem_cache_create; 1929 1930 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib, 1931 NULL); 1932 if (ret) 1933 goto out_unregister_subsys; 1934 1935 __fib6_flush_trees = fib6_flush_trees; 1936 out: 1937 return ret; 1938 1939 out_unregister_subsys: 1940 unregister_pernet_subsys(&fib6_net_ops); 1941 out_kmem_cache_create: 1942 kmem_cache_destroy(fib6_node_kmem); 1943 goto out; 1944 } 1945 1946 void fib6_gc_cleanup(void) 1947 { 1948 unregister_pernet_subsys(&fib6_net_ops); 1949 kmem_cache_destroy(fib6_node_kmem); 1950 } 1951 1952 #ifdef CONFIG_PROC_FS 1953 1954 struct ipv6_route_iter { 1955 struct seq_net_private p; 1956 struct fib6_walker w; 1957 loff_t skip; 1958 struct fib6_table *tbl; 1959 int sernum; 1960 }; 1961 1962 static int ipv6_route_seq_show(struct seq_file *seq, void *v) 1963 { 1964 struct rt6_info *rt = v; 1965 struct ipv6_route_iter *iter = seq->private; 1966 1967 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen); 1968 1969 #ifdef CONFIG_IPV6_SUBTREES 1970 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen); 1971 #else 1972 seq_puts(seq, "00000000000000000000000000000000 00 "); 1973 #endif 1974 if (rt->rt6i_flags & RTF_GATEWAY) 1975 seq_printf(seq, "%pi6", &rt->rt6i_gateway); 1976 else 1977 seq_puts(seq, "00000000000000000000000000000000"); 1978 1979 seq_printf(seq, " %08x %08x %08x %08x %8s\n", 1980 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt), 1981 rt->dst.__use, rt->rt6i_flags, 1982 rt->dst.dev ? rt->dst.dev->name : ""); 1983 iter->w.leaf = NULL; 1984 return 0; 1985 } 1986 1987 static int ipv6_route_yield(struct fib6_walker *w) 1988 { 1989 struct ipv6_route_iter *iter = w->args; 1990 1991 if (!iter->skip) 1992 return 1; 1993 1994 do { 1995 iter->w.leaf = iter->w.leaf->dst.rt6_next; 1996 iter->skip--; 1997 if (!iter->skip && iter->w.leaf) 1998 return 1; 1999 } while (iter->w.leaf); 2000 2001 return 0; 2002 } 2003 2004 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter, 2005 struct net *net) 2006 { 2007 memset(&iter->w, 0, sizeof(iter->w)); 2008 iter->w.func = ipv6_route_yield; 2009 iter->w.root = &iter->tbl->tb6_root; 2010 iter->w.state = FWS_INIT; 2011 iter->w.node = iter->w.root; 2012 iter->w.args = iter; 2013 iter->sernum = iter->w.root->fn_sernum; 2014 INIT_LIST_HEAD(&iter->w.lh); 2015 fib6_walker_link(net, &iter->w); 2016 } 2017 2018 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl, 2019 struct net *net) 2020 { 2021 unsigned int h; 2022 struct hlist_node *node; 2023 2024 if (tbl) { 2025 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1; 2026 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist)); 2027 } else { 2028 h = 0; 2029 node = NULL; 2030 } 2031 2032 while (!node && h < FIB6_TABLE_HASHSZ) { 2033 node = rcu_dereference_bh( 2034 hlist_first_rcu(&net->ipv6.fib_table_hash[h++])); 2035 } 2036 return hlist_entry_safe(node, struct fib6_table, tb6_hlist); 2037 } 2038 2039 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter) 2040 { 2041 if (iter->sernum != iter->w.root->fn_sernum) { 2042 iter->sernum = iter->w.root->fn_sernum; 2043 iter->w.state = FWS_INIT; 2044 iter->w.node = iter->w.root; 2045 WARN_ON(iter->w.skip); 2046 iter->w.skip = iter->w.count; 2047 } 2048 } 2049 2050 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2051 { 2052 int r; 2053 struct rt6_info *n; 2054 struct net *net = seq_file_net(seq); 2055 struct ipv6_route_iter *iter = seq->private; 2056 2057 if (!v) 2058 goto iter_table; 2059 2060 n = ((struct rt6_info *)v)->dst.rt6_next; 2061 if (n) { 2062 ++*pos; 2063 return n; 2064 } 2065 2066 iter_table: 2067 ipv6_route_check_sernum(iter); 2068 read_lock(&iter->tbl->tb6_lock); 2069 r = fib6_walk_continue(&iter->w); 2070 read_unlock(&iter->tbl->tb6_lock); 2071 if (r > 0) { 2072 if (v) 2073 ++*pos; 2074 return iter->w.leaf; 2075 } else if (r < 0) { 2076 fib6_walker_unlink(net, &iter->w); 2077 return NULL; 2078 } 2079 fib6_walker_unlink(net, &iter->w); 2080 2081 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net); 2082 if (!iter->tbl) 2083 return NULL; 2084 2085 ipv6_route_seq_setup_walk(iter, net); 2086 goto iter_table; 2087 } 2088 2089 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos) 2090 __acquires(RCU_BH) 2091 { 2092 struct net *net = seq_file_net(seq); 2093 struct ipv6_route_iter *iter = seq->private; 2094 2095 rcu_read_lock_bh(); 2096 iter->tbl = ipv6_route_seq_next_table(NULL, net); 2097 iter->skip = *pos; 2098 2099 if (iter->tbl) { 2100 ipv6_route_seq_setup_walk(iter, net); 2101 return ipv6_route_seq_next(seq, NULL, pos); 2102 } else { 2103 return NULL; 2104 } 2105 } 2106 2107 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter) 2108 { 2109 struct fib6_walker *w = &iter->w; 2110 return w->node && !(w->state == FWS_U && w->node == w->root); 2111 } 2112 2113 static void ipv6_route_seq_stop(struct seq_file *seq, void *v) 2114 __releases(RCU_BH) 2115 { 2116 struct net *net = seq_file_net(seq); 2117 struct ipv6_route_iter *iter = seq->private; 2118 2119 if (ipv6_route_iter_active(iter)) 2120 fib6_walker_unlink(net, &iter->w); 2121 2122 rcu_read_unlock_bh(); 2123 } 2124 2125 static const struct seq_operations ipv6_route_seq_ops = { 2126 .start = ipv6_route_seq_start, 2127 .next = ipv6_route_seq_next, 2128 .stop = ipv6_route_seq_stop, 2129 .show = ipv6_route_seq_show 2130 }; 2131 2132 int ipv6_route_open(struct inode *inode, struct file *file) 2133 { 2134 return seq_open_net(inode, file, &ipv6_route_seq_ops, 2135 sizeof(struct ipv6_route_iter)); 2136 } 2137 2138 #endif /* CONFIG_PROC_FS */ 2139