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_free_pcpu(struct rt6_info *non_pcpu_rt) 157 { 158 int cpu; 159 160 if (!non_pcpu_rt->rt6i_pcpu) 161 return; 162 163 for_each_possible_cpu(cpu) { 164 struct rt6_info **ppcpu_rt; 165 struct rt6_info *pcpu_rt; 166 167 ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu); 168 pcpu_rt = *ppcpu_rt; 169 if (pcpu_rt) { 170 dst_dev_put(&pcpu_rt->dst); 171 dst_release(&pcpu_rt->dst); 172 *ppcpu_rt = NULL; 173 } 174 } 175 176 free_percpu(non_pcpu_rt->rt6i_pcpu); 177 non_pcpu_rt->rt6i_pcpu = NULL; 178 } 179 180 static void rt6_release(struct rt6_info *rt) 181 { 182 if (atomic_dec_and_test(&rt->rt6i_ref)) { 183 rt6_free_pcpu(rt); 184 dst_dev_put(&rt->dst); 185 dst_release(&rt->dst); 186 } 187 } 188 189 static void fib6_link_table(struct net *net, struct fib6_table *tb) 190 { 191 unsigned int h; 192 193 /* 194 * Initialize table lock at a single place to give lockdep a key, 195 * tables aren't visible prior to being linked to the list. 196 */ 197 rwlock_init(&tb->tb6_lock); 198 199 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1); 200 201 /* 202 * No protection necessary, this is the only list mutatation 203 * operation, tables never disappear once they exist. 204 */ 205 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]); 206 } 207 208 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 209 210 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id) 211 { 212 struct fib6_table *table; 213 214 table = kzalloc(sizeof(*table), GFP_ATOMIC); 215 if (table) { 216 table->tb6_id = id; 217 table->tb6_root.leaf = net->ipv6.ip6_null_entry; 218 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 219 inet_peer_base_init(&table->tb6_peers); 220 } 221 222 return table; 223 } 224 225 struct fib6_table *fib6_new_table(struct net *net, u32 id) 226 { 227 struct fib6_table *tb; 228 229 if (id == 0) 230 id = RT6_TABLE_MAIN; 231 tb = fib6_get_table(net, id); 232 if (tb) 233 return tb; 234 235 tb = fib6_alloc_table(net, id); 236 if (tb) 237 fib6_link_table(net, tb); 238 239 return tb; 240 } 241 EXPORT_SYMBOL_GPL(fib6_new_table); 242 243 struct fib6_table *fib6_get_table(struct net *net, u32 id) 244 { 245 struct fib6_table *tb; 246 struct hlist_head *head; 247 unsigned int h; 248 249 if (id == 0) 250 id = RT6_TABLE_MAIN; 251 h = id & (FIB6_TABLE_HASHSZ - 1); 252 rcu_read_lock(); 253 head = &net->ipv6.fib_table_hash[h]; 254 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 255 if (tb->tb6_id == id) { 256 rcu_read_unlock(); 257 return tb; 258 } 259 } 260 rcu_read_unlock(); 261 262 return NULL; 263 } 264 EXPORT_SYMBOL_GPL(fib6_get_table); 265 266 static void __net_init fib6_tables_init(struct net *net) 267 { 268 fib6_link_table(net, net->ipv6.fib6_main_tbl); 269 fib6_link_table(net, net->ipv6.fib6_local_tbl); 270 } 271 #else 272 273 struct fib6_table *fib6_new_table(struct net *net, u32 id) 274 { 275 return fib6_get_table(net, id); 276 } 277 278 struct fib6_table *fib6_get_table(struct net *net, u32 id) 279 { 280 return net->ipv6.fib6_main_tbl; 281 } 282 283 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, 284 int flags, pol_lookup_t lookup) 285 { 286 struct rt6_info *rt; 287 288 rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags); 289 if (rt->dst.error == -EAGAIN) { 290 ip6_rt_put(rt); 291 rt = net->ipv6.ip6_null_entry; 292 dst_hold(&rt->dst); 293 } 294 295 return &rt->dst; 296 } 297 298 static void __net_init fib6_tables_init(struct net *net) 299 { 300 fib6_link_table(net, net->ipv6.fib6_main_tbl); 301 } 302 303 #endif 304 305 static int fib6_dump_node(struct fib6_walker *w) 306 { 307 int res; 308 struct rt6_info *rt; 309 310 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 311 res = rt6_dump_route(rt, w->args); 312 if (res < 0) { 313 /* Frame is full, suspend walking */ 314 w->leaf = rt; 315 return 1; 316 } 317 318 /* Multipath routes are dumped in one route with the 319 * RTA_MULTIPATH attribute. Jump 'rt' to point to the 320 * last sibling of this route (no need to dump the 321 * sibling routes again) 322 */ 323 if (rt->rt6i_nsiblings) 324 rt = list_last_entry(&rt->rt6i_siblings, 325 struct rt6_info, 326 rt6i_siblings); 327 } 328 w->leaf = NULL; 329 return 0; 330 } 331 332 static void fib6_dump_end(struct netlink_callback *cb) 333 { 334 struct net *net = sock_net(cb->skb->sk); 335 struct fib6_walker *w = (void *)cb->args[2]; 336 337 if (w) { 338 if (cb->args[4]) { 339 cb->args[4] = 0; 340 fib6_walker_unlink(net, w); 341 } 342 cb->args[2] = 0; 343 kfree(w); 344 } 345 cb->done = (void *)cb->args[3]; 346 cb->args[1] = 3; 347 } 348 349 static int fib6_dump_done(struct netlink_callback *cb) 350 { 351 fib6_dump_end(cb); 352 return cb->done ? cb->done(cb) : 0; 353 } 354 355 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 356 struct netlink_callback *cb) 357 { 358 struct net *net = sock_net(skb->sk); 359 struct fib6_walker *w; 360 int res; 361 362 w = (void *)cb->args[2]; 363 w->root = &table->tb6_root; 364 365 if (cb->args[4] == 0) { 366 w->count = 0; 367 w->skip = 0; 368 369 read_lock_bh(&table->tb6_lock); 370 res = fib6_walk(net, w); 371 read_unlock_bh(&table->tb6_lock); 372 if (res > 0) { 373 cb->args[4] = 1; 374 cb->args[5] = w->root->fn_sernum; 375 } 376 } else { 377 if (cb->args[5] != w->root->fn_sernum) { 378 /* Begin at the root if the tree changed */ 379 cb->args[5] = w->root->fn_sernum; 380 w->state = FWS_INIT; 381 w->node = w->root; 382 w->skip = w->count; 383 } else 384 w->skip = 0; 385 386 read_lock_bh(&table->tb6_lock); 387 res = fib6_walk_continue(w); 388 read_unlock_bh(&table->tb6_lock); 389 if (res <= 0) { 390 fib6_walker_unlink(net, w); 391 cb->args[4] = 0; 392 } 393 } 394 395 return res; 396 } 397 398 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 399 { 400 struct net *net = sock_net(skb->sk); 401 unsigned int h, s_h; 402 unsigned int e = 0, s_e; 403 struct rt6_rtnl_dump_arg arg; 404 struct fib6_walker *w; 405 struct fib6_table *tb; 406 struct hlist_head *head; 407 int res = 0; 408 409 s_h = cb->args[0]; 410 s_e = cb->args[1]; 411 412 w = (void *)cb->args[2]; 413 if (!w) { 414 /* New dump: 415 * 416 * 1. hook callback destructor. 417 */ 418 cb->args[3] = (long)cb->done; 419 cb->done = fib6_dump_done; 420 421 /* 422 * 2. allocate and initialize walker. 423 */ 424 w = kzalloc(sizeof(*w), GFP_ATOMIC); 425 if (!w) 426 return -ENOMEM; 427 w->func = fib6_dump_node; 428 cb->args[2] = (long)w; 429 } 430 431 arg.skb = skb; 432 arg.cb = cb; 433 arg.net = net; 434 w->args = &arg; 435 436 rcu_read_lock(); 437 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) { 438 e = 0; 439 head = &net->ipv6.fib_table_hash[h]; 440 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 441 if (e < s_e) 442 goto next; 443 res = fib6_dump_table(tb, skb, cb); 444 if (res != 0) 445 goto out; 446 next: 447 e++; 448 } 449 } 450 out: 451 rcu_read_unlock(); 452 cb->args[1] = e; 453 cb->args[0] = h; 454 455 res = res < 0 ? res : skb->len; 456 if (res <= 0) 457 fib6_dump_end(cb); 458 return res; 459 } 460 461 /* 462 * Routing Table 463 * 464 * return the appropriate node for a routing tree "add" operation 465 * by either creating and inserting or by returning an existing 466 * node. 467 */ 468 469 static struct fib6_node *fib6_add_1(struct fib6_node *root, 470 struct in6_addr *addr, int plen, 471 int offset, int allow_create, 472 int replace_required, int sernum, 473 struct netlink_ext_ack *extack) 474 { 475 struct fib6_node *fn, *in, *ln; 476 struct fib6_node *pn = NULL; 477 struct rt6key *key; 478 int bit; 479 __be32 dir = 0; 480 481 RT6_TRACE("fib6_add_1\n"); 482 483 /* insert node in tree */ 484 485 fn = root; 486 487 do { 488 key = (struct rt6key *)((u8 *)fn->leaf + offset); 489 490 /* 491 * Prefix match 492 */ 493 if (plen < fn->fn_bit || 494 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) { 495 if (!allow_create) { 496 if (replace_required) { 497 NL_SET_ERR_MSG(extack, 498 "Can not replace route - no match found"); 499 pr_warn("Can't replace route, no match found\n"); 500 return ERR_PTR(-ENOENT); 501 } 502 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 503 } 504 goto insert_above; 505 } 506 507 /* 508 * Exact match ? 509 */ 510 511 if (plen == fn->fn_bit) { 512 /* clean up an intermediate node */ 513 if (!(fn->fn_flags & RTN_RTINFO)) { 514 rt6_release(fn->leaf); 515 fn->leaf = NULL; 516 } 517 518 fn->fn_sernum = sernum; 519 520 return fn; 521 } 522 523 /* 524 * We have more bits to go 525 */ 526 527 /* Try to walk down on tree. */ 528 fn->fn_sernum = sernum; 529 dir = addr_bit_set(addr, fn->fn_bit); 530 pn = fn; 531 fn = dir ? fn->right : fn->left; 532 } while (fn); 533 534 if (!allow_create) { 535 /* We should not create new node because 536 * NLM_F_REPLACE was specified without NLM_F_CREATE 537 * I assume it is safe to require NLM_F_CREATE when 538 * REPLACE flag is used! Later we may want to remove the 539 * check for replace_required, because according 540 * to netlink specification, NLM_F_CREATE 541 * MUST be specified if new route is created. 542 * That would keep IPv6 consistent with IPv4 543 */ 544 if (replace_required) { 545 NL_SET_ERR_MSG(extack, 546 "Can not replace route - no match found"); 547 pr_warn("Can't replace route, no match found\n"); 548 return ERR_PTR(-ENOENT); 549 } 550 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 551 } 552 /* 553 * We walked to the bottom of tree. 554 * Create new leaf node without children. 555 */ 556 557 ln = node_alloc(); 558 559 if (!ln) 560 return ERR_PTR(-ENOMEM); 561 ln->fn_bit = plen; 562 563 ln->parent = pn; 564 ln->fn_sernum = sernum; 565 566 if (dir) 567 pn->right = ln; 568 else 569 pn->left = ln; 570 571 return ln; 572 573 574 insert_above: 575 /* 576 * split since we don't have a common prefix anymore or 577 * we have a less significant route. 578 * we've to insert an intermediate node on the list 579 * this new node will point to the one we need to create 580 * and the current 581 */ 582 583 pn = fn->parent; 584 585 /* find 1st bit in difference between the 2 addrs. 586 587 See comment in __ipv6_addr_diff: bit may be an invalid value, 588 but if it is >= plen, the value is ignored in any case. 589 */ 590 591 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr)); 592 593 /* 594 * (intermediate)[in] 595 * / \ 596 * (new leaf node)[ln] (old node)[fn] 597 */ 598 if (plen > bit) { 599 in = node_alloc(); 600 ln = node_alloc(); 601 602 if (!in || !ln) { 603 if (in) 604 node_free(in); 605 if (ln) 606 node_free(ln); 607 return ERR_PTR(-ENOMEM); 608 } 609 610 /* 611 * new intermediate node. 612 * RTN_RTINFO will 613 * be off since that an address that chooses one of 614 * the branches would not match less specific routes 615 * in the other branch 616 */ 617 618 in->fn_bit = bit; 619 620 in->parent = pn; 621 in->leaf = fn->leaf; 622 atomic_inc(&in->leaf->rt6i_ref); 623 624 in->fn_sernum = sernum; 625 626 /* update parent pointer */ 627 if (dir) 628 pn->right = in; 629 else 630 pn->left = in; 631 632 ln->fn_bit = plen; 633 634 ln->parent = in; 635 fn->parent = in; 636 637 ln->fn_sernum = sernum; 638 639 if (addr_bit_set(addr, bit)) { 640 in->right = ln; 641 in->left = fn; 642 } else { 643 in->left = ln; 644 in->right = fn; 645 } 646 } else { /* plen <= bit */ 647 648 /* 649 * (new leaf node)[ln] 650 * / \ 651 * (old node)[fn] NULL 652 */ 653 654 ln = node_alloc(); 655 656 if (!ln) 657 return ERR_PTR(-ENOMEM); 658 659 ln->fn_bit = plen; 660 661 ln->parent = pn; 662 663 ln->fn_sernum = sernum; 664 665 if (dir) 666 pn->right = ln; 667 else 668 pn->left = ln; 669 670 if (addr_bit_set(&key->addr, plen)) 671 ln->right = fn; 672 else 673 ln->left = fn; 674 675 fn->parent = ln; 676 } 677 return ln; 678 } 679 680 static bool rt6_qualify_for_ecmp(struct rt6_info *rt) 681 { 682 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) == 683 RTF_GATEWAY; 684 } 685 686 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc) 687 { 688 int i; 689 690 for (i = 0; i < RTAX_MAX; i++) { 691 if (test_bit(i, mxc->mx_valid)) 692 mp[i] = mxc->mx[i]; 693 } 694 } 695 696 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc) 697 { 698 if (!mxc->mx) 699 return 0; 700 701 if (dst->flags & DST_HOST) { 702 u32 *mp = dst_metrics_write_ptr(dst); 703 704 if (unlikely(!mp)) 705 return -ENOMEM; 706 707 fib6_copy_metrics(mp, mxc); 708 } else { 709 dst_init_metrics(dst, mxc->mx, false); 710 711 /* We've stolen mx now. */ 712 mxc->mx = NULL; 713 } 714 715 return 0; 716 } 717 718 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn, 719 struct net *net) 720 { 721 if (atomic_read(&rt->rt6i_ref) != 1) { 722 /* This route is used as dummy address holder in some split 723 * nodes. It is not leaked, but it still holds other resources, 724 * which must be released in time. So, scan ascendant nodes 725 * and replace dummy references to this route with references 726 * to still alive ones. 727 */ 728 while (fn) { 729 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) { 730 fn->leaf = fib6_find_prefix(net, fn); 731 atomic_inc(&fn->leaf->rt6i_ref); 732 rt6_release(rt); 733 } 734 fn = fn->parent; 735 } 736 /* No more references are possible at this point. */ 737 BUG_ON(atomic_read(&rt->rt6i_ref) != 1); 738 } 739 } 740 741 /* 742 * Insert routing information in a node. 743 */ 744 745 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, 746 struct nl_info *info, struct mx6_config *mxc) 747 { 748 struct rt6_info *iter = NULL; 749 struct rt6_info **ins; 750 struct rt6_info **fallback_ins = NULL; 751 int replace = (info->nlh && 752 (info->nlh->nlmsg_flags & NLM_F_REPLACE)); 753 int add = (!info->nlh || 754 (info->nlh->nlmsg_flags & NLM_F_CREATE)); 755 int found = 0; 756 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt); 757 u16 nlflags = NLM_F_EXCL; 758 int err; 759 760 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND)) 761 nlflags |= NLM_F_APPEND; 762 763 ins = &fn->leaf; 764 765 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) { 766 /* 767 * Search for duplicates 768 */ 769 770 if (iter->rt6i_metric == rt->rt6i_metric) { 771 /* 772 * Same priority level 773 */ 774 if (info->nlh && 775 (info->nlh->nlmsg_flags & NLM_F_EXCL)) 776 return -EEXIST; 777 778 nlflags &= ~NLM_F_EXCL; 779 if (replace) { 780 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) { 781 found++; 782 break; 783 } 784 if (rt_can_ecmp) 785 fallback_ins = fallback_ins ?: ins; 786 goto next_iter; 787 } 788 789 if (rt6_duplicate_nexthop(iter, rt)) { 790 if (rt->rt6i_nsiblings) 791 rt->rt6i_nsiblings = 0; 792 if (!(iter->rt6i_flags & RTF_EXPIRES)) 793 return -EEXIST; 794 if (!(rt->rt6i_flags & RTF_EXPIRES)) 795 rt6_clean_expires(iter); 796 else 797 rt6_set_expires(iter, rt->dst.expires); 798 iter->rt6i_pmtu = rt->rt6i_pmtu; 799 return -EEXIST; 800 } 801 /* If we have the same destination and the same metric, 802 * but not the same gateway, then the route we try to 803 * add is sibling to this route, increment our counter 804 * of siblings, and later we will add our route to the 805 * list. 806 * Only static routes (which don't have flag 807 * RTF_EXPIRES) are used for ECMPv6. 808 * 809 * To avoid long list, we only had siblings if the 810 * route have a gateway. 811 */ 812 if (rt_can_ecmp && 813 rt6_qualify_for_ecmp(iter)) 814 rt->rt6i_nsiblings++; 815 } 816 817 if (iter->rt6i_metric > rt->rt6i_metric) 818 break; 819 820 next_iter: 821 ins = &iter->dst.rt6_next; 822 } 823 824 if (fallback_ins && !found) { 825 /* No ECMP-able route found, replace first non-ECMP one */ 826 ins = fallback_ins; 827 iter = *ins; 828 found++; 829 } 830 831 /* Reset round-robin state, if necessary */ 832 if (ins == &fn->leaf) 833 fn->rr_ptr = NULL; 834 835 /* Link this route to others same route. */ 836 if (rt->rt6i_nsiblings) { 837 unsigned int rt6i_nsiblings; 838 struct rt6_info *sibling, *temp_sibling; 839 840 /* Find the first route that have the same metric */ 841 sibling = fn->leaf; 842 while (sibling) { 843 if (sibling->rt6i_metric == rt->rt6i_metric && 844 rt6_qualify_for_ecmp(sibling)) { 845 list_add_tail(&rt->rt6i_siblings, 846 &sibling->rt6i_siblings); 847 break; 848 } 849 sibling = sibling->dst.rt6_next; 850 } 851 /* For each sibling in the list, increment the counter of 852 * siblings. BUG() if counters does not match, list of siblings 853 * is broken! 854 */ 855 rt6i_nsiblings = 0; 856 list_for_each_entry_safe(sibling, temp_sibling, 857 &rt->rt6i_siblings, rt6i_siblings) { 858 sibling->rt6i_nsiblings++; 859 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings); 860 rt6i_nsiblings++; 861 } 862 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings); 863 } 864 865 /* 866 * insert node 867 */ 868 if (!replace) { 869 if (!add) 870 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 871 872 add: 873 nlflags |= NLM_F_CREATE; 874 err = fib6_commit_metrics(&rt->dst, mxc); 875 if (err) 876 return err; 877 878 rt->dst.rt6_next = iter; 879 *ins = rt; 880 rt->rt6i_node = fn; 881 atomic_inc(&rt->rt6i_ref); 882 if (!info->skip_notify) 883 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags); 884 info->nl_net->ipv6.rt6_stats->fib_rt_entries++; 885 886 if (!(fn->fn_flags & RTN_RTINFO)) { 887 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 888 fn->fn_flags |= RTN_RTINFO; 889 } 890 891 } else { 892 int nsiblings; 893 894 if (!found) { 895 if (add) 896 goto add; 897 pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); 898 return -ENOENT; 899 } 900 901 err = fib6_commit_metrics(&rt->dst, mxc); 902 if (err) 903 return err; 904 905 *ins = rt; 906 rt->rt6i_node = fn; 907 rt->dst.rt6_next = iter->dst.rt6_next; 908 atomic_inc(&rt->rt6i_ref); 909 if (!info->skip_notify) 910 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE); 911 if (!(fn->fn_flags & RTN_RTINFO)) { 912 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 913 fn->fn_flags |= RTN_RTINFO; 914 } 915 nsiblings = iter->rt6i_nsiblings; 916 fib6_purge_rt(iter, fn, info->nl_net); 917 rt6_release(iter); 918 919 if (nsiblings) { 920 /* Replacing an ECMP route, remove all siblings */ 921 ins = &rt->dst.rt6_next; 922 iter = *ins; 923 while (iter) { 924 if (iter->rt6i_metric > rt->rt6i_metric) 925 break; 926 if (rt6_qualify_for_ecmp(iter)) { 927 *ins = iter->dst.rt6_next; 928 fib6_purge_rt(iter, fn, info->nl_net); 929 rt6_release(iter); 930 nsiblings--; 931 } else { 932 ins = &iter->dst.rt6_next; 933 } 934 iter = *ins; 935 } 936 WARN_ON(nsiblings != 0); 937 } 938 } 939 940 return 0; 941 } 942 943 static void fib6_start_gc(struct net *net, struct rt6_info *rt) 944 { 945 if (!timer_pending(&net->ipv6.ip6_fib_timer) && 946 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE))) 947 mod_timer(&net->ipv6.ip6_fib_timer, 948 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 949 } 950 951 void fib6_force_start_gc(struct net *net) 952 { 953 if (!timer_pending(&net->ipv6.ip6_fib_timer)) 954 mod_timer(&net->ipv6.ip6_fib_timer, 955 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 956 } 957 958 /* 959 * Add routing information to the routing tree. 960 * <destination addr>/<source addr> 961 * with source addr info in sub-trees 962 */ 963 964 int fib6_add(struct fib6_node *root, struct rt6_info *rt, 965 struct nl_info *info, struct mx6_config *mxc, 966 struct netlink_ext_ack *extack) 967 { 968 struct fib6_node *fn, *pn = NULL; 969 int err = -ENOMEM; 970 int allow_create = 1; 971 int replace_required = 0; 972 int sernum = fib6_new_sernum(info->nl_net); 973 974 if (WARN_ON_ONCE(!atomic_read(&rt->dst.__refcnt))) 975 return -EINVAL; 976 977 if (info->nlh) { 978 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) 979 allow_create = 0; 980 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) 981 replace_required = 1; 982 } 983 if (!allow_create && !replace_required) 984 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); 985 986 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen, 987 offsetof(struct rt6_info, rt6i_dst), allow_create, 988 replace_required, sernum, extack); 989 if (IS_ERR(fn)) { 990 err = PTR_ERR(fn); 991 fn = NULL; 992 goto out; 993 } 994 995 pn = fn; 996 997 #ifdef CONFIG_IPV6_SUBTREES 998 if (rt->rt6i_src.plen) { 999 struct fib6_node *sn; 1000 1001 if (!fn->subtree) { 1002 struct fib6_node *sfn; 1003 1004 /* 1005 * Create subtree. 1006 * 1007 * fn[main tree] 1008 * | 1009 * sfn[subtree root] 1010 * \ 1011 * sn[new leaf node] 1012 */ 1013 1014 /* Create subtree root node */ 1015 sfn = node_alloc(); 1016 if (!sfn) 1017 goto st_failure; 1018 1019 sfn->leaf = info->nl_net->ipv6.ip6_null_entry; 1020 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref); 1021 sfn->fn_flags = RTN_ROOT; 1022 sfn->fn_sernum = sernum; 1023 1024 /* Now add the first leaf node to new subtree */ 1025 1026 sn = fib6_add_1(sfn, &rt->rt6i_src.addr, 1027 rt->rt6i_src.plen, 1028 offsetof(struct rt6_info, rt6i_src), 1029 allow_create, replace_required, sernum, 1030 extack); 1031 1032 if (IS_ERR(sn)) { 1033 /* If it is failed, discard just allocated 1034 root, and then (in st_failure) stale node 1035 in main tree. 1036 */ 1037 node_free(sfn); 1038 err = PTR_ERR(sn); 1039 goto st_failure; 1040 } 1041 1042 /* Now link new subtree to main tree */ 1043 sfn->parent = fn; 1044 fn->subtree = sfn; 1045 } else { 1046 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, 1047 rt->rt6i_src.plen, 1048 offsetof(struct rt6_info, rt6i_src), 1049 allow_create, replace_required, sernum, 1050 extack); 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 } 1072 1073 out: 1074 if (err) { 1075 #ifdef CONFIG_IPV6_SUBTREES 1076 /* 1077 * If fib6_add_1 has cleared the old leaf pointer in the 1078 * super-tree leaf node we have to find a new one for it. 1079 */ 1080 if (pn != fn && pn->leaf == rt) { 1081 pn->leaf = NULL; 1082 atomic_dec(&rt->rt6i_ref); 1083 } 1084 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) { 1085 pn->leaf = fib6_find_prefix(info->nl_net, pn); 1086 #if RT6_DEBUG >= 2 1087 if (!pn->leaf) { 1088 WARN_ON(pn->leaf == NULL); 1089 pn->leaf = info->nl_net->ipv6.ip6_null_entry; 1090 } 1091 #endif 1092 atomic_inc(&pn->leaf->rt6i_ref); 1093 } 1094 #endif 1095 /* Always release dst as dst->__refcnt is guaranteed 1096 * to be taken before entering this function 1097 */ 1098 dst_release_immediate(&rt->dst); 1099 } 1100 return err; 1101 1102 #ifdef CONFIG_IPV6_SUBTREES 1103 /* Subtree creation failed, probably main tree node 1104 is orphan. If it is, shoot it. 1105 */ 1106 st_failure: 1107 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) 1108 fib6_repair_tree(info->nl_net, fn); 1109 /* Always release dst as dst->__refcnt is guaranteed 1110 * to be taken before entering this function 1111 */ 1112 dst_release_immediate(&rt->dst); 1113 return err; 1114 #endif 1115 } 1116 1117 /* 1118 * Routing tree lookup 1119 * 1120 */ 1121 1122 struct lookup_args { 1123 int offset; /* key offset on rt6_info */ 1124 const struct in6_addr *addr; /* search key */ 1125 }; 1126 1127 static struct fib6_node *fib6_lookup_1(struct fib6_node *root, 1128 struct lookup_args *args) 1129 { 1130 struct fib6_node *fn; 1131 __be32 dir; 1132 1133 if (unlikely(args->offset == 0)) 1134 return NULL; 1135 1136 /* 1137 * Descend on a tree 1138 */ 1139 1140 fn = root; 1141 1142 for (;;) { 1143 struct fib6_node *next; 1144 1145 dir = addr_bit_set(args->addr, fn->fn_bit); 1146 1147 next = dir ? fn->right : fn->left; 1148 1149 if (next) { 1150 fn = next; 1151 continue; 1152 } 1153 break; 1154 } 1155 1156 while (fn) { 1157 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) { 1158 struct rt6key *key; 1159 1160 key = (struct rt6key *) ((u8 *) fn->leaf + 1161 args->offset); 1162 1163 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 1164 #ifdef CONFIG_IPV6_SUBTREES 1165 if (fn->subtree) { 1166 struct fib6_node *sfn; 1167 sfn = fib6_lookup_1(fn->subtree, 1168 args + 1); 1169 if (!sfn) 1170 goto backtrack; 1171 fn = sfn; 1172 } 1173 #endif 1174 if (fn->fn_flags & RTN_RTINFO) 1175 return fn; 1176 } 1177 } 1178 #ifdef CONFIG_IPV6_SUBTREES 1179 backtrack: 1180 #endif 1181 if (fn->fn_flags & RTN_ROOT) 1182 break; 1183 1184 fn = fn->parent; 1185 } 1186 1187 return NULL; 1188 } 1189 1190 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr, 1191 const struct in6_addr *saddr) 1192 { 1193 struct fib6_node *fn; 1194 struct lookup_args args[] = { 1195 { 1196 .offset = offsetof(struct rt6_info, rt6i_dst), 1197 .addr = daddr, 1198 }, 1199 #ifdef CONFIG_IPV6_SUBTREES 1200 { 1201 .offset = offsetof(struct rt6_info, rt6i_src), 1202 .addr = saddr, 1203 }, 1204 #endif 1205 { 1206 .offset = 0, /* sentinel */ 1207 } 1208 }; 1209 1210 fn = fib6_lookup_1(root, daddr ? args : args + 1); 1211 if (!fn || fn->fn_flags & RTN_TL_ROOT) 1212 fn = root; 1213 1214 return fn; 1215 } 1216 1217 /* 1218 * Get node with specified destination prefix (and source prefix, 1219 * if subtrees are used) 1220 */ 1221 1222 1223 static struct fib6_node *fib6_locate_1(struct fib6_node *root, 1224 const struct in6_addr *addr, 1225 int plen, int offset) 1226 { 1227 struct fib6_node *fn; 1228 1229 for (fn = root; fn ; ) { 1230 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); 1231 1232 /* 1233 * Prefix match 1234 */ 1235 if (plen < fn->fn_bit || 1236 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 1237 return NULL; 1238 1239 if (plen == fn->fn_bit) 1240 return fn; 1241 1242 /* 1243 * We have more bits to go 1244 */ 1245 if (addr_bit_set(addr, fn->fn_bit)) 1246 fn = fn->right; 1247 else 1248 fn = fn->left; 1249 } 1250 return NULL; 1251 } 1252 1253 struct fib6_node *fib6_locate(struct fib6_node *root, 1254 const struct in6_addr *daddr, int dst_len, 1255 const struct in6_addr *saddr, int src_len) 1256 { 1257 struct fib6_node *fn; 1258 1259 fn = fib6_locate_1(root, daddr, dst_len, 1260 offsetof(struct rt6_info, rt6i_dst)); 1261 1262 #ifdef CONFIG_IPV6_SUBTREES 1263 if (src_len) { 1264 WARN_ON(saddr == NULL); 1265 if (fn && fn->subtree) 1266 fn = fib6_locate_1(fn->subtree, saddr, src_len, 1267 offsetof(struct rt6_info, rt6i_src)); 1268 } 1269 #endif 1270 1271 if (fn && fn->fn_flags & RTN_RTINFO) 1272 return fn; 1273 1274 return NULL; 1275 } 1276 1277 1278 /* 1279 * Deletion 1280 * 1281 */ 1282 1283 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn) 1284 { 1285 if (fn->fn_flags & RTN_ROOT) 1286 return net->ipv6.ip6_null_entry; 1287 1288 while (fn) { 1289 if (fn->left) 1290 return fn->left->leaf; 1291 if (fn->right) 1292 return fn->right->leaf; 1293 1294 fn = FIB6_SUBTREE(fn); 1295 } 1296 return NULL; 1297 } 1298 1299 /* 1300 * Called to trim the tree of intermediate nodes when possible. "fn" 1301 * is the node we want to try and remove. 1302 */ 1303 1304 static struct fib6_node *fib6_repair_tree(struct net *net, 1305 struct fib6_node *fn) 1306 { 1307 int children; 1308 int nstate; 1309 struct fib6_node *child, *pn; 1310 struct fib6_walker *w; 1311 int iter = 0; 1312 1313 for (;;) { 1314 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1315 iter++; 1316 1317 WARN_ON(fn->fn_flags & RTN_RTINFO); 1318 WARN_ON(fn->fn_flags & RTN_TL_ROOT); 1319 WARN_ON(fn->leaf); 1320 1321 children = 0; 1322 child = NULL; 1323 if (fn->right) 1324 child = fn->right, children |= 1; 1325 if (fn->left) 1326 child = fn->left, children |= 2; 1327 1328 if (children == 3 || FIB6_SUBTREE(fn) 1329 #ifdef CONFIG_IPV6_SUBTREES 1330 /* Subtree root (i.e. fn) may have one child */ 1331 || (children && fn->fn_flags & RTN_ROOT) 1332 #endif 1333 ) { 1334 fn->leaf = fib6_find_prefix(net, fn); 1335 #if RT6_DEBUG >= 2 1336 if (!fn->leaf) { 1337 WARN_ON(!fn->leaf); 1338 fn->leaf = net->ipv6.ip6_null_entry; 1339 } 1340 #endif 1341 atomic_inc(&fn->leaf->rt6i_ref); 1342 return fn->parent; 1343 } 1344 1345 pn = fn->parent; 1346 #ifdef CONFIG_IPV6_SUBTREES 1347 if (FIB6_SUBTREE(pn) == fn) { 1348 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1349 FIB6_SUBTREE(pn) = NULL; 1350 nstate = FWS_L; 1351 } else { 1352 WARN_ON(fn->fn_flags & RTN_ROOT); 1353 #endif 1354 if (pn->right == fn) 1355 pn->right = child; 1356 else if (pn->left == fn) 1357 pn->left = child; 1358 #if RT6_DEBUG >= 2 1359 else 1360 WARN_ON(1); 1361 #endif 1362 if (child) 1363 child->parent = pn; 1364 nstate = FWS_R; 1365 #ifdef CONFIG_IPV6_SUBTREES 1366 } 1367 #endif 1368 1369 read_lock(&net->ipv6.fib6_walker_lock); 1370 FOR_WALKERS(net, w) { 1371 if (!child) { 1372 if (w->root == fn) { 1373 w->root = w->node = NULL; 1374 RT6_TRACE("W %p adjusted by delroot 1\n", w); 1375 } else if (w->node == fn) { 1376 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1377 w->node = pn; 1378 w->state = nstate; 1379 } 1380 } else { 1381 if (w->root == fn) { 1382 w->root = child; 1383 RT6_TRACE("W %p adjusted by delroot 2\n", w); 1384 } 1385 if (w->node == fn) { 1386 w->node = child; 1387 if (children&2) { 1388 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1389 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT; 1390 } else { 1391 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1392 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT; 1393 } 1394 } 1395 } 1396 } 1397 read_unlock(&net->ipv6.fib6_walker_lock); 1398 1399 node_free(fn); 1400 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) 1401 return pn; 1402 1403 rt6_release(pn->leaf); 1404 pn->leaf = NULL; 1405 fn = pn; 1406 } 1407 } 1408 1409 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, 1410 struct nl_info *info) 1411 { 1412 struct fib6_walker *w; 1413 struct rt6_info *rt = *rtp; 1414 struct net *net = info->nl_net; 1415 1416 RT6_TRACE("fib6_del_route\n"); 1417 1418 /* Unlink it */ 1419 *rtp = rt->dst.rt6_next; 1420 rt->rt6i_node = NULL; 1421 net->ipv6.rt6_stats->fib_rt_entries--; 1422 net->ipv6.rt6_stats->fib_discarded_routes++; 1423 1424 /* Reset round-robin state, if necessary */ 1425 if (fn->rr_ptr == rt) 1426 fn->rr_ptr = NULL; 1427 1428 /* Remove this entry from other siblings */ 1429 if (rt->rt6i_nsiblings) { 1430 struct rt6_info *sibling, *next_sibling; 1431 1432 list_for_each_entry_safe(sibling, next_sibling, 1433 &rt->rt6i_siblings, rt6i_siblings) 1434 sibling->rt6i_nsiblings--; 1435 rt->rt6i_nsiblings = 0; 1436 list_del_init(&rt->rt6i_siblings); 1437 } 1438 1439 /* Adjust walkers */ 1440 read_lock(&net->ipv6.fib6_walker_lock); 1441 FOR_WALKERS(net, w) { 1442 if (w->state == FWS_C && w->leaf == rt) { 1443 RT6_TRACE("walker %p adjusted by delroute\n", w); 1444 w->leaf = rt->dst.rt6_next; 1445 if (!w->leaf) 1446 w->state = FWS_U; 1447 } 1448 } 1449 read_unlock(&net->ipv6.fib6_walker_lock); 1450 1451 rt->dst.rt6_next = NULL; 1452 1453 /* If it was last route, expunge its radix tree node */ 1454 if (!fn->leaf) { 1455 fn->fn_flags &= ~RTN_RTINFO; 1456 net->ipv6.rt6_stats->fib_route_nodes--; 1457 fn = fib6_repair_tree(net, fn); 1458 } 1459 1460 fib6_purge_rt(rt, fn, net); 1461 1462 if (!info->skip_notify) 1463 inet6_rt_notify(RTM_DELROUTE, rt, info, 0); 1464 rt6_release(rt); 1465 } 1466 1467 int fib6_del(struct rt6_info *rt, struct nl_info *info) 1468 { 1469 struct net *net = info->nl_net; 1470 struct fib6_node *fn = rt->rt6i_node; 1471 struct rt6_info **rtp; 1472 1473 #if RT6_DEBUG >= 2 1474 if (rt->dst.obsolete > 0) { 1475 WARN_ON(fn); 1476 return -ENOENT; 1477 } 1478 #endif 1479 if (!fn || rt == net->ipv6.ip6_null_entry) 1480 return -ENOENT; 1481 1482 WARN_ON(!(fn->fn_flags & RTN_RTINFO)); 1483 1484 if (!(rt->rt6i_flags & RTF_CACHE)) { 1485 struct fib6_node *pn = fn; 1486 #ifdef CONFIG_IPV6_SUBTREES 1487 /* clones of this route might be in another subtree */ 1488 if (rt->rt6i_src.plen) { 1489 while (!(pn->fn_flags & RTN_ROOT)) 1490 pn = pn->parent; 1491 pn = pn->parent; 1492 } 1493 #endif 1494 fib6_prune_clones(info->nl_net, pn); 1495 } 1496 1497 /* 1498 * Walk the leaf entries looking for ourself 1499 */ 1500 1501 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) { 1502 if (*rtp == rt) { 1503 fib6_del_route(fn, rtp, info); 1504 return 0; 1505 } 1506 } 1507 return -ENOENT; 1508 } 1509 1510 /* 1511 * Tree traversal function. 1512 * 1513 * Certainly, it is not interrupt safe. 1514 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 1515 * It means, that we can modify tree during walking 1516 * and use this function for garbage collection, clone pruning, 1517 * cleaning tree when a device goes down etc. etc. 1518 * 1519 * It guarantees that every node will be traversed, 1520 * and that it will be traversed only once. 1521 * 1522 * Callback function w->func may return: 1523 * 0 -> continue walking. 1524 * positive value -> walking is suspended (used by tree dumps, 1525 * and probably by gc, if it will be split to several slices) 1526 * negative value -> terminate walking. 1527 * 1528 * The function itself returns: 1529 * 0 -> walk is complete. 1530 * >0 -> walk is incomplete (i.e. suspended) 1531 * <0 -> walk is terminated by an error. 1532 */ 1533 1534 static int fib6_walk_continue(struct fib6_walker *w) 1535 { 1536 struct fib6_node *fn, *pn; 1537 1538 for (;;) { 1539 fn = w->node; 1540 if (!fn) 1541 return 0; 1542 1543 if (w->prune && fn != w->root && 1544 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) { 1545 w->state = FWS_C; 1546 w->leaf = fn->leaf; 1547 } 1548 switch (w->state) { 1549 #ifdef CONFIG_IPV6_SUBTREES 1550 case FWS_S: 1551 if (FIB6_SUBTREE(fn)) { 1552 w->node = FIB6_SUBTREE(fn); 1553 continue; 1554 } 1555 w->state = FWS_L; 1556 #endif 1557 case FWS_L: 1558 if (fn->left) { 1559 w->node = fn->left; 1560 w->state = FWS_INIT; 1561 continue; 1562 } 1563 w->state = FWS_R; 1564 case FWS_R: 1565 if (fn->right) { 1566 w->node = fn->right; 1567 w->state = FWS_INIT; 1568 continue; 1569 } 1570 w->state = FWS_C; 1571 w->leaf = fn->leaf; 1572 case FWS_C: 1573 if (w->leaf && fn->fn_flags & RTN_RTINFO) { 1574 int err; 1575 1576 if (w->skip) { 1577 w->skip--; 1578 goto skip; 1579 } 1580 1581 err = w->func(w); 1582 if (err) 1583 return err; 1584 1585 w->count++; 1586 continue; 1587 } 1588 skip: 1589 w->state = FWS_U; 1590 case FWS_U: 1591 if (fn == w->root) 1592 return 0; 1593 pn = fn->parent; 1594 w->node = pn; 1595 #ifdef CONFIG_IPV6_SUBTREES 1596 if (FIB6_SUBTREE(pn) == fn) { 1597 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1598 w->state = FWS_L; 1599 continue; 1600 } 1601 #endif 1602 if (pn->left == fn) { 1603 w->state = FWS_R; 1604 continue; 1605 } 1606 if (pn->right == fn) { 1607 w->state = FWS_C; 1608 w->leaf = w->node->leaf; 1609 continue; 1610 } 1611 #if RT6_DEBUG >= 2 1612 WARN_ON(1); 1613 #endif 1614 } 1615 } 1616 } 1617 1618 static int fib6_walk(struct net *net, struct fib6_walker *w) 1619 { 1620 int res; 1621 1622 w->state = FWS_INIT; 1623 w->node = w->root; 1624 1625 fib6_walker_link(net, w); 1626 res = fib6_walk_continue(w); 1627 if (res <= 0) 1628 fib6_walker_unlink(net, w); 1629 return res; 1630 } 1631 1632 static int fib6_clean_node(struct fib6_walker *w) 1633 { 1634 int res; 1635 struct rt6_info *rt; 1636 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w); 1637 struct nl_info info = { 1638 .nl_net = c->net, 1639 }; 1640 1641 if (c->sernum != FIB6_NO_SERNUM_CHANGE && 1642 w->node->fn_sernum != c->sernum) 1643 w->node->fn_sernum = c->sernum; 1644 1645 if (!c->func) { 1646 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE); 1647 w->leaf = NULL; 1648 return 0; 1649 } 1650 1651 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 1652 res = c->func(rt, c->arg); 1653 if (res < 0) { 1654 w->leaf = rt; 1655 res = fib6_del(rt, &info); 1656 if (res) { 1657 #if RT6_DEBUG >= 2 1658 pr_debug("%s: del failed: rt=%p@%p err=%d\n", 1659 __func__, rt, rt->rt6i_node, res); 1660 #endif 1661 continue; 1662 } 1663 return 0; 1664 } 1665 WARN_ON(res != 0); 1666 } 1667 w->leaf = rt; 1668 return 0; 1669 } 1670 1671 /* 1672 * Convenient frontend to tree walker. 1673 * 1674 * func is called on each route. 1675 * It may return -1 -> delete this route. 1676 * 0 -> continue walking 1677 * 1678 * prune==1 -> only immediate children of node (certainly, 1679 * ignoring pure split nodes) will be scanned. 1680 */ 1681 1682 static void fib6_clean_tree(struct net *net, struct fib6_node *root, 1683 int (*func)(struct rt6_info *, void *arg), 1684 bool prune, int sernum, void *arg) 1685 { 1686 struct fib6_cleaner c; 1687 1688 c.w.root = root; 1689 c.w.func = fib6_clean_node; 1690 c.w.prune = prune; 1691 c.w.count = 0; 1692 c.w.skip = 0; 1693 c.func = func; 1694 c.sernum = sernum; 1695 c.arg = arg; 1696 c.net = net; 1697 1698 fib6_walk(net, &c.w); 1699 } 1700 1701 static void __fib6_clean_all(struct net *net, 1702 int (*func)(struct rt6_info *, void *), 1703 int sernum, void *arg) 1704 { 1705 struct fib6_table *table; 1706 struct hlist_head *head; 1707 unsigned int h; 1708 1709 rcu_read_lock(); 1710 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 1711 head = &net->ipv6.fib_table_hash[h]; 1712 hlist_for_each_entry_rcu(table, head, tb6_hlist) { 1713 write_lock_bh(&table->tb6_lock); 1714 fib6_clean_tree(net, &table->tb6_root, 1715 func, false, sernum, arg); 1716 write_unlock_bh(&table->tb6_lock); 1717 } 1718 } 1719 rcu_read_unlock(); 1720 } 1721 1722 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *), 1723 void *arg) 1724 { 1725 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg); 1726 } 1727 1728 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1729 { 1730 if (rt->rt6i_flags & RTF_CACHE) { 1731 RT6_TRACE("pruning clone %p\n", rt); 1732 return -1; 1733 } 1734 1735 return 0; 1736 } 1737 1738 static void fib6_prune_clones(struct net *net, struct fib6_node *fn) 1739 { 1740 fib6_clean_tree(net, fn, fib6_prune_clone, true, 1741 FIB6_NO_SERNUM_CHANGE, NULL); 1742 } 1743 1744 static void fib6_flush_trees(struct net *net) 1745 { 1746 int new_sernum = fib6_new_sernum(net); 1747 1748 __fib6_clean_all(net, NULL, new_sernum, NULL); 1749 } 1750 1751 /* 1752 * Garbage collection 1753 */ 1754 1755 struct fib6_gc_args 1756 { 1757 int timeout; 1758 int more; 1759 }; 1760 1761 static int fib6_age(struct rt6_info *rt, void *arg) 1762 { 1763 struct fib6_gc_args *gc_args = arg; 1764 unsigned long now = jiffies; 1765 1766 /* 1767 * check addrconf expiration here. 1768 * Routes are expired even if they are in use. 1769 * 1770 * Also age clones. Note, that clones are aged out 1771 * only if they are not in use now. 1772 */ 1773 1774 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) { 1775 if (time_after(now, rt->dst.expires)) { 1776 RT6_TRACE("expiring %p\n", rt); 1777 return -1; 1778 } 1779 gc_args->more++; 1780 } else if (rt->rt6i_flags & RTF_CACHE) { 1781 if (atomic_read(&rt->dst.__refcnt) == 1 && 1782 time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) { 1783 RT6_TRACE("aging clone %p\n", rt); 1784 return -1; 1785 } else if (rt->rt6i_flags & RTF_GATEWAY) { 1786 struct neighbour *neigh; 1787 __u8 neigh_flags = 0; 1788 1789 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway); 1790 if (neigh) { 1791 neigh_flags = neigh->flags; 1792 neigh_release(neigh); 1793 } 1794 if (!(neigh_flags & NTF_ROUTER)) { 1795 RT6_TRACE("purging route %p via non-router but gateway\n", 1796 rt); 1797 return -1; 1798 } 1799 } 1800 gc_args->more++; 1801 } 1802 1803 return 0; 1804 } 1805 1806 void fib6_run_gc(unsigned long expires, struct net *net, bool force) 1807 { 1808 struct fib6_gc_args gc_args; 1809 unsigned long now; 1810 1811 if (force) { 1812 spin_lock_bh(&net->ipv6.fib6_gc_lock); 1813 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) { 1814 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); 1815 return; 1816 } 1817 gc_args.timeout = expires ? (int)expires : 1818 net->ipv6.sysctl.ip6_rt_gc_interval; 1819 gc_args.more = 0; 1820 1821 fib6_clean_all(net, fib6_age, &gc_args); 1822 now = jiffies; 1823 net->ipv6.ip6_rt_last_gc = now; 1824 1825 if (gc_args.more) 1826 mod_timer(&net->ipv6.ip6_fib_timer, 1827 round_jiffies(now 1828 + net->ipv6.sysctl.ip6_rt_gc_interval)); 1829 else 1830 del_timer(&net->ipv6.ip6_fib_timer); 1831 spin_unlock_bh(&net->ipv6.fib6_gc_lock); 1832 } 1833 1834 static void fib6_gc_timer_cb(unsigned long arg) 1835 { 1836 fib6_run_gc(0, (struct net *)arg, true); 1837 } 1838 1839 static int __net_init fib6_net_init(struct net *net) 1840 { 1841 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; 1842 1843 spin_lock_init(&net->ipv6.fib6_gc_lock); 1844 rwlock_init(&net->ipv6.fib6_walker_lock); 1845 INIT_LIST_HEAD(&net->ipv6.fib6_walkers); 1846 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net); 1847 1848 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); 1849 if (!net->ipv6.rt6_stats) 1850 goto out_timer; 1851 1852 /* Avoid false sharing : Use at least a full cache line */ 1853 size = max_t(size_t, size, L1_CACHE_BYTES); 1854 1855 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); 1856 if (!net->ipv6.fib_table_hash) 1857 goto out_rt6_stats; 1858 1859 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), 1860 GFP_KERNEL); 1861 if (!net->ipv6.fib6_main_tbl) 1862 goto out_fib_table_hash; 1863 1864 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; 1865 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1866 net->ipv6.fib6_main_tbl->tb6_root.fn_flags = 1867 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1868 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); 1869 1870 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1871 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), 1872 GFP_KERNEL); 1873 if (!net->ipv6.fib6_local_tbl) 1874 goto out_fib6_main_tbl; 1875 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; 1876 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1877 net->ipv6.fib6_local_tbl->tb6_root.fn_flags = 1878 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1879 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); 1880 #endif 1881 fib6_tables_init(net); 1882 1883 return 0; 1884 1885 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1886 out_fib6_main_tbl: 1887 kfree(net->ipv6.fib6_main_tbl); 1888 #endif 1889 out_fib_table_hash: 1890 kfree(net->ipv6.fib_table_hash); 1891 out_rt6_stats: 1892 kfree(net->ipv6.rt6_stats); 1893 out_timer: 1894 return -ENOMEM; 1895 } 1896 1897 static void fib6_net_exit(struct net *net) 1898 { 1899 rt6_ifdown(net, NULL); 1900 del_timer_sync(&net->ipv6.ip6_fib_timer); 1901 1902 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1903 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers); 1904 kfree(net->ipv6.fib6_local_tbl); 1905 #endif 1906 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers); 1907 kfree(net->ipv6.fib6_main_tbl); 1908 kfree(net->ipv6.fib_table_hash); 1909 kfree(net->ipv6.rt6_stats); 1910 } 1911 1912 static struct pernet_operations fib6_net_ops = { 1913 .init = fib6_net_init, 1914 .exit = fib6_net_exit, 1915 }; 1916 1917 int __init fib6_init(void) 1918 { 1919 int ret = -ENOMEM; 1920 1921 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1922 sizeof(struct fib6_node), 1923 0, SLAB_HWCACHE_ALIGN, 1924 NULL); 1925 if (!fib6_node_kmem) 1926 goto out; 1927 1928 ret = register_pernet_subsys(&fib6_net_ops); 1929 if (ret) 1930 goto out_kmem_cache_create; 1931 1932 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib, 1933 NULL); 1934 if (ret) 1935 goto out_unregister_subsys; 1936 1937 __fib6_flush_trees = fib6_flush_trees; 1938 out: 1939 return ret; 1940 1941 out_unregister_subsys: 1942 unregister_pernet_subsys(&fib6_net_ops); 1943 out_kmem_cache_create: 1944 kmem_cache_destroy(fib6_node_kmem); 1945 goto out; 1946 } 1947 1948 void fib6_gc_cleanup(void) 1949 { 1950 unregister_pernet_subsys(&fib6_net_ops); 1951 kmem_cache_destroy(fib6_node_kmem); 1952 } 1953 1954 #ifdef CONFIG_PROC_FS 1955 1956 struct ipv6_route_iter { 1957 struct seq_net_private p; 1958 struct fib6_walker w; 1959 loff_t skip; 1960 struct fib6_table *tbl; 1961 int sernum; 1962 }; 1963 1964 static int ipv6_route_seq_show(struct seq_file *seq, void *v) 1965 { 1966 struct rt6_info *rt = v; 1967 struct ipv6_route_iter *iter = seq->private; 1968 1969 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen); 1970 1971 #ifdef CONFIG_IPV6_SUBTREES 1972 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen); 1973 #else 1974 seq_puts(seq, "00000000000000000000000000000000 00 "); 1975 #endif 1976 if (rt->rt6i_flags & RTF_GATEWAY) 1977 seq_printf(seq, "%pi6", &rt->rt6i_gateway); 1978 else 1979 seq_puts(seq, "00000000000000000000000000000000"); 1980 1981 seq_printf(seq, " %08x %08x %08x %08x %8s\n", 1982 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt), 1983 rt->dst.__use, rt->rt6i_flags, 1984 rt->dst.dev ? rt->dst.dev->name : ""); 1985 iter->w.leaf = NULL; 1986 return 0; 1987 } 1988 1989 static int ipv6_route_yield(struct fib6_walker *w) 1990 { 1991 struct ipv6_route_iter *iter = w->args; 1992 1993 if (!iter->skip) 1994 return 1; 1995 1996 do { 1997 iter->w.leaf = iter->w.leaf->dst.rt6_next; 1998 iter->skip--; 1999 if (!iter->skip && iter->w.leaf) 2000 return 1; 2001 } while (iter->w.leaf); 2002 2003 return 0; 2004 } 2005 2006 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter, 2007 struct net *net) 2008 { 2009 memset(&iter->w, 0, sizeof(iter->w)); 2010 iter->w.func = ipv6_route_yield; 2011 iter->w.root = &iter->tbl->tb6_root; 2012 iter->w.state = FWS_INIT; 2013 iter->w.node = iter->w.root; 2014 iter->w.args = iter; 2015 iter->sernum = iter->w.root->fn_sernum; 2016 INIT_LIST_HEAD(&iter->w.lh); 2017 fib6_walker_link(net, &iter->w); 2018 } 2019 2020 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl, 2021 struct net *net) 2022 { 2023 unsigned int h; 2024 struct hlist_node *node; 2025 2026 if (tbl) { 2027 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1; 2028 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist)); 2029 } else { 2030 h = 0; 2031 node = NULL; 2032 } 2033 2034 while (!node && h < FIB6_TABLE_HASHSZ) { 2035 node = rcu_dereference_bh( 2036 hlist_first_rcu(&net->ipv6.fib_table_hash[h++])); 2037 } 2038 return hlist_entry_safe(node, struct fib6_table, tb6_hlist); 2039 } 2040 2041 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter) 2042 { 2043 if (iter->sernum != iter->w.root->fn_sernum) { 2044 iter->sernum = iter->w.root->fn_sernum; 2045 iter->w.state = FWS_INIT; 2046 iter->w.node = iter->w.root; 2047 WARN_ON(iter->w.skip); 2048 iter->w.skip = iter->w.count; 2049 } 2050 } 2051 2052 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2053 { 2054 int r; 2055 struct rt6_info *n; 2056 struct net *net = seq_file_net(seq); 2057 struct ipv6_route_iter *iter = seq->private; 2058 2059 if (!v) 2060 goto iter_table; 2061 2062 n = ((struct rt6_info *)v)->dst.rt6_next; 2063 if (n) { 2064 ++*pos; 2065 return n; 2066 } 2067 2068 iter_table: 2069 ipv6_route_check_sernum(iter); 2070 read_lock(&iter->tbl->tb6_lock); 2071 r = fib6_walk_continue(&iter->w); 2072 read_unlock(&iter->tbl->tb6_lock); 2073 if (r > 0) { 2074 if (v) 2075 ++*pos; 2076 return iter->w.leaf; 2077 } else if (r < 0) { 2078 fib6_walker_unlink(net, &iter->w); 2079 return NULL; 2080 } 2081 fib6_walker_unlink(net, &iter->w); 2082 2083 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net); 2084 if (!iter->tbl) 2085 return NULL; 2086 2087 ipv6_route_seq_setup_walk(iter, net); 2088 goto iter_table; 2089 } 2090 2091 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos) 2092 __acquires(RCU_BH) 2093 { 2094 struct net *net = seq_file_net(seq); 2095 struct ipv6_route_iter *iter = seq->private; 2096 2097 rcu_read_lock_bh(); 2098 iter->tbl = ipv6_route_seq_next_table(NULL, net); 2099 iter->skip = *pos; 2100 2101 if (iter->tbl) { 2102 ipv6_route_seq_setup_walk(iter, net); 2103 return ipv6_route_seq_next(seq, NULL, pos); 2104 } else { 2105 return NULL; 2106 } 2107 } 2108 2109 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter) 2110 { 2111 struct fib6_walker *w = &iter->w; 2112 return w->node && !(w->state == FWS_U && w->node == w->root); 2113 } 2114 2115 static void ipv6_route_seq_stop(struct seq_file *seq, void *v) 2116 __releases(RCU_BH) 2117 { 2118 struct net *net = seq_file_net(seq); 2119 struct ipv6_route_iter *iter = seq->private; 2120 2121 if (ipv6_route_iter_active(iter)) 2122 fib6_walker_unlink(net, &iter->w); 2123 2124 rcu_read_unlock_bh(); 2125 } 2126 2127 static const struct seq_operations ipv6_route_seq_ops = { 2128 .start = ipv6_route_seq_start, 2129 .next = ipv6_route_seq_next, 2130 .stop = ipv6_route_seq_stop, 2131 .show = ipv6_route_seq_show 2132 }; 2133 2134 int ipv6_route_open(struct inode *inode, struct file *file) 2135 { 2136 return seq_open_net(inode, file, &ipv6_route_seq_ops, 2137 sizeof(struct ipv6_route_iter)); 2138 } 2139 2140 #endif /* CONFIG_PROC_FS */ 2141