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 if (fn->rr_ptr == iter) 918 fn->rr_ptr = NULL; 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 if (fn->rr_ptr == iter) 932 fn->rr_ptr = NULL; 933 rt6_release(iter); 934 nsiblings--; 935 } else { 936 ins = &iter->dst.rt6_next; 937 } 938 iter = *ins; 939 } 940 WARN_ON(nsiblings != 0); 941 } 942 } 943 944 return 0; 945 } 946 947 static void fib6_start_gc(struct net *net, struct rt6_info *rt) 948 { 949 if (!timer_pending(&net->ipv6.ip6_fib_timer) && 950 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE))) 951 mod_timer(&net->ipv6.ip6_fib_timer, 952 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 953 } 954 955 void fib6_force_start_gc(struct net *net) 956 { 957 if (!timer_pending(&net->ipv6.ip6_fib_timer)) 958 mod_timer(&net->ipv6.ip6_fib_timer, 959 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 960 } 961 962 /* 963 * Add routing information to the routing tree. 964 * <destination addr>/<source addr> 965 * with source addr info in sub-trees 966 */ 967 968 int fib6_add(struct fib6_node *root, struct rt6_info *rt, 969 struct nl_info *info, struct mx6_config *mxc, 970 struct netlink_ext_ack *extack) 971 { 972 struct fib6_node *fn, *pn = NULL; 973 int err = -ENOMEM; 974 int allow_create = 1; 975 int replace_required = 0; 976 int sernum = fib6_new_sernum(info->nl_net); 977 978 if (WARN_ON_ONCE(!atomic_read(&rt->dst.__refcnt))) 979 return -EINVAL; 980 981 if (info->nlh) { 982 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) 983 allow_create = 0; 984 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) 985 replace_required = 1; 986 } 987 if (!allow_create && !replace_required) 988 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); 989 990 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen, 991 offsetof(struct rt6_info, rt6i_dst), allow_create, 992 replace_required, sernum, extack); 993 if (IS_ERR(fn)) { 994 err = PTR_ERR(fn); 995 fn = NULL; 996 goto out; 997 } 998 999 pn = fn; 1000 1001 #ifdef CONFIG_IPV6_SUBTREES 1002 if (rt->rt6i_src.plen) { 1003 struct fib6_node *sn; 1004 1005 if (!fn->subtree) { 1006 struct fib6_node *sfn; 1007 1008 /* 1009 * Create subtree. 1010 * 1011 * fn[main tree] 1012 * | 1013 * sfn[subtree root] 1014 * \ 1015 * sn[new leaf node] 1016 */ 1017 1018 /* Create subtree root node */ 1019 sfn = node_alloc(); 1020 if (!sfn) 1021 goto failure; 1022 1023 sfn->leaf = info->nl_net->ipv6.ip6_null_entry; 1024 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref); 1025 sfn->fn_flags = RTN_ROOT; 1026 sfn->fn_sernum = sernum; 1027 1028 /* Now add the first leaf node to new subtree */ 1029 1030 sn = fib6_add_1(sfn, &rt->rt6i_src.addr, 1031 rt->rt6i_src.plen, 1032 offsetof(struct rt6_info, rt6i_src), 1033 allow_create, replace_required, sernum, 1034 extack); 1035 1036 if (IS_ERR(sn)) { 1037 /* If it is failed, discard just allocated 1038 root, and then (in failure) stale node 1039 in main tree. 1040 */ 1041 node_free(sfn); 1042 err = PTR_ERR(sn); 1043 goto failure; 1044 } 1045 1046 /* Now link new subtree to main tree */ 1047 sfn->parent = fn; 1048 fn->subtree = sfn; 1049 } else { 1050 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, 1051 rt->rt6i_src.plen, 1052 offsetof(struct rt6_info, rt6i_src), 1053 allow_create, replace_required, sernum, 1054 extack); 1055 1056 if (IS_ERR(sn)) { 1057 err = PTR_ERR(sn); 1058 goto failure; 1059 } 1060 } 1061 1062 if (!fn->leaf) { 1063 fn->leaf = rt; 1064 atomic_inc(&rt->rt6i_ref); 1065 } 1066 fn = sn; 1067 } 1068 #endif 1069 1070 err = fib6_add_rt2node(fn, rt, info, mxc); 1071 if (!err) { 1072 fib6_start_gc(info->nl_net, rt); 1073 if (!(rt->rt6i_flags & RTF_CACHE)) 1074 fib6_prune_clones(info->nl_net, pn); 1075 } 1076 1077 out: 1078 if (err) { 1079 #ifdef CONFIG_IPV6_SUBTREES 1080 /* 1081 * If fib6_add_1 has cleared the old leaf pointer in the 1082 * super-tree leaf node we have to find a new one for it. 1083 */ 1084 if (pn != fn && pn->leaf == rt) { 1085 pn->leaf = NULL; 1086 atomic_dec(&rt->rt6i_ref); 1087 } 1088 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) { 1089 pn->leaf = fib6_find_prefix(info->nl_net, pn); 1090 #if RT6_DEBUG >= 2 1091 if (!pn->leaf) { 1092 WARN_ON(pn->leaf == NULL); 1093 pn->leaf = info->nl_net->ipv6.ip6_null_entry; 1094 } 1095 #endif 1096 atomic_inc(&pn->leaf->rt6i_ref); 1097 } 1098 #endif 1099 goto failure; 1100 } 1101 return err; 1102 1103 failure: 1104 /* fn->leaf could be NULL if fn is an intermediate node and we 1105 * failed to add the new route to it in both subtree creation 1106 * failure and fib6_add_rt2node() failure case. 1107 * In both cases, fib6_repair_tree() should be called to fix 1108 * fn->leaf. 1109 */ 1110 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) 1111 fib6_repair_tree(info->nl_net, fn); 1112 /* Always release dst as dst->__refcnt is guaranteed 1113 * to be taken before entering this function 1114 */ 1115 dst_release_immediate(&rt->dst); 1116 return err; 1117 } 1118 1119 /* 1120 * Routing tree lookup 1121 * 1122 */ 1123 1124 struct lookup_args { 1125 int offset; /* key offset on rt6_info */ 1126 const struct in6_addr *addr; /* search key */ 1127 }; 1128 1129 static struct fib6_node *fib6_lookup_1(struct fib6_node *root, 1130 struct lookup_args *args) 1131 { 1132 struct fib6_node *fn; 1133 __be32 dir; 1134 1135 if (unlikely(args->offset == 0)) 1136 return NULL; 1137 1138 /* 1139 * Descend on a tree 1140 */ 1141 1142 fn = root; 1143 1144 for (;;) { 1145 struct fib6_node *next; 1146 1147 dir = addr_bit_set(args->addr, fn->fn_bit); 1148 1149 next = dir ? fn->right : fn->left; 1150 1151 if (next) { 1152 fn = next; 1153 continue; 1154 } 1155 break; 1156 } 1157 1158 while (fn) { 1159 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) { 1160 struct rt6key *key; 1161 1162 key = (struct rt6key *) ((u8 *) fn->leaf + 1163 args->offset); 1164 1165 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 1166 #ifdef CONFIG_IPV6_SUBTREES 1167 if (fn->subtree) { 1168 struct fib6_node *sfn; 1169 sfn = fib6_lookup_1(fn->subtree, 1170 args + 1); 1171 if (!sfn) 1172 goto backtrack; 1173 fn = sfn; 1174 } 1175 #endif 1176 if (fn->fn_flags & RTN_RTINFO) 1177 return fn; 1178 } 1179 } 1180 #ifdef CONFIG_IPV6_SUBTREES 1181 backtrack: 1182 #endif 1183 if (fn->fn_flags & RTN_ROOT) 1184 break; 1185 1186 fn = fn->parent; 1187 } 1188 1189 return NULL; 1190 } 1191 1192 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr, 1193 const struct in6_addr *saddr) 1194 { 1195 struct fib6_node *fn; 1196 struct lookup_args args[] = { 1197 { 1198 .offset = offsetof(struct rt6_info, rt6i_dst), 1199 .addr = daddr, 1200 }, 1201 #ifdef CONFIG_IPV6_SUBTREES 1202 { 1203 .offset = offsetof(struct rt6_info, rt6i_src), 1204 .addr = saddr, 1205 }, 1206 #endif 1207 { 1208 .offset = 0, /* sentinel */ 1209 } 1210 }; 1211 1212 fn = fib6_lookup_1(root, daddr ? args : args + 1); 1213 if (!fn || fn->fn_flags & RTN_TL_ROOT) 1214 fn = root; 1215 1216 return fn; 1217 } 1218 1219 /* 1220 * Get node with specified destination prefix (and source prefix, 1221 * if subtrees are used) 1222 */ 1223 1224 1225 static struct fib6_node *fib6_locate_1(struct fib6_node *root, 1226 const struct in6_addr *addr, 1227 int plen, int offset) 1228 { 1229 struct fib6_node *fn; 1230 1231 for (fn = root; fn ; ) { 1232 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); 1233 1234 /* 1235 * Prefix match 1236 */ 1237 if (plen < fn->fn_bit || 1238 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 1239 return NULL; 1240 1241 if (plen == fn->fn_bit) 1242 return fn; 1243 1244 /* 1245 * We have more bits to go 1246 */ 1247 if (addr_bit_set(addr, fn->fn_bit)) 1248 fn = fn->right; 1249 else 1250 fn = fn->left; 1251 } 1252 return NULL; 1253 } 1254 1255 struct fib6_node *fib6_locate(struct fib6_node *root, 1256 const struct in6_addr *daddr, int dst_len, 1257 const struct in6_addr *saddr, int src_len) 1258 { 1259 struct fib6_node *fn; 1260 1261 fn = fib6_locate_1(root, daddr, dst_len, 1262 offsetof(struct rt6_info, rt6i_dst)); 1263 1264 #ifdef CONFIG_IPV6_SUBTREES 1265 if (src_len) { 1266 WARN_ON(saddr == NULL); 1267 if (fn && fn->subtree) 1268 fn = fib6_locate_1(fn->subtree, saddr, src_len, 1269 offsetof(struct rt6_info, rt6i_src)); 1270 } 1271 #endif 1272 1273 if (fn && fn->fn_flags & RTN_RTINFO) 1274 return fn; 1275 1276 return NULL; 1277 } 1278 1279 1280 /* 1281 * Deletion 1282 * 1283 */ 1284 1285 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn) 1286 { 1287 if (fn->fn_flags & RTN_ROOT) 1288 return net->ipv6.ip6_null_entry; 1289 1290 while (fn) { 1291 if (fn->left) 1292 return fn->left->leaf; 1293 if (fn->right) 1294 return fn->right->leaf; 1295 1296 fn = FIB6_SUBTREE(fn); 1297 } 1298 return NULL; 1299 } 1300 1301 /* 1302 * Called to trim the tree of intermediate nodes when possible. "fn" 1303 * is the node we want to try and remove. 1304 */ 1305 1306 static struct fib6_node *fib6_repair_tree(struct net *net, 1307 struct fib6_node *fn) 1308 { 1309 int children; 1310 int nstate; 1311 struct fib6_node *child, *pn; 1312 struct fib6_walker *w; 1313 int iter = 0; 1314 1315 for (;;) { 1316 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1317 iter++; 1318 1319 WARN_ON(fn->fn_flags & RTN_RTINFO); 1320 WARN_ON(fn->fn_flags & RTN_TL_ROOT); 1321 WARN_ON(fn->leaf); 1322 1323 children = 0; 1324 child = NULL; 1325 if (fn->right) 1326 child = fn->right, children |= 1; 1327 if (fn->left) 1328 child = fn->left, children |= 2; 1329 1330 if (children == 3 || FIB6_SUBTREE(fn) 1331 #ifdef CONFIG_IPV6_SUBTREES 1332 /* Subtree root (i.e. fn) may have one child */ 1333 || (children && fn->fn_flags & RTN_ROOT) 1334 #endif 1335 ) { 1336 fn->leaf = fib6_find_prefix(net, fn); 1337 #if RT6_DEBUG >= 2 1338 if (!fn->leaf) { 1339 WARN_ON(!fn->leaf); 1340 fn->leaf = net->ipv6.ip6_null_entry; 1341 } 1342 #endif 1343 atomic_inc(&fn->leaf->rt6i_ref); 1344 return fn->parent; 1345 } 1346 1347 pn = fn->parent; 1348 #ifdef CONFIG_IPV6_SUBTREES 1349 if (FIB6_SUBTREE(pn) == fn) { 1350 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1351 FIB6_SUBTREE(pn) = NULL; 1352 nstate = FWS_L; 1353 } else { 1354 WARN_ON(fn->fn_flags & RTN_ROOT); 1355 #endif 1356 if (pn->right == fn) 1357 pn->right = child; 1358 else if (pn->left == fn) 1359 pn->left = child; 1360 #if RT6_DEBUG >= 2 1361 else 1362 WARN_ON(1); 1363 #endif 1364 if (child) 1365 child->parent = pn; 1366 nstate = FWS_R; 1367 #ifdef CONFIG_IPV6_SUBTREES 1368 } 1369 #endif 1370 1371 read_lock(&net->ipv6.fib6_walker_lock); 1372 FOR_WALKERS(net, w) { 1373 if (!child) { 1374 if (w->root == fn) { 1375 w->root = w->node = NULL; 1376 RT6_TRACE("W %p adjusted by delroot 1\n", w); 1377 } else if (w->node == fn) { 1378 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1379 w->node = pn; 1380 w->state = nstate; 1381 } 1382 } else { 1383 if (w->root == fn) { 1384 w->root = child; 1385 RT6_TRACE("W %p adjusted by delroot 2\n", w); 1386 } 1387 if (w->node == fn) { 1388 w->node = child; 1389 if (children&2) { 1390 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1391 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT; 1392 } else { 1393 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1394 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT; 1395 } 1396 } 1397 } 1398 } 1399 read_unlock(&net->ipv6.fib6_walker_lock); 1400 1401 node_free(fn); 1402 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) 1403 return pn; 1404 1405 rt6_release(pn->leaf); 1406 pn->leaf = NULL; 1407 fn = pn; 1408 } 1409 } 1410 1411 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, 1412 struct nl_info *info) 1413 { 1414 struct fib6_walker *w; 1415 struct rt6_info *rt = *rtp; 1416 struct net *net = info->nl_net; 1417 1418 RT6_TRACE("fib6_del_route\n"); 1419 1420 /* Unlink it */ 1421 *rtp = rt->dst.rt6_next; 1422 rt->rt6i_node = NULL; 1423 net->ipv6.rt6_stats->fib_rt_entries--; 1424 net->ipv6.rt6_stats->fib_discarded_routes++; 1425 1426 /* Reset round-robin state, if necessary */ 1427 if (fn->rr_ptr == rt) 1428 fn->rr_ptr = NULL; 1429 1430 /* Remove this entry from other siblings */ 1431 if (rt->rt6i_nsiblings) { 1432 struct rt6_info *sibling, *next_sibling; 1433 1434 list_for_each_entry_safe(sibling, next_sibling, 1435 &rt->rt6i_siblings, rt6i_siblings) 1436 sibling->rt6i_nsiblings--; 1437 rt->rt6i_nsiblings = 0; 1438 list_del_init(&rt->rt6i_siblings); 1439 } 1440 1441 /* Adjust walkers */ 1442 read_lock(&net->ipv6.fib6_walker_lock); 1443 FOR_WALKERS(net, w) { 1444 if (w->state == FWS_C && w->leaf == rt) { 1445 RT6_TRACE("walker %p adjusted by delroute\n", w); 1446 w->leaf = rt->dst.rt6_next; 1447 if (!w->leaf) 1448 w->state = FWS_U; 1449 } 1450 } 1451 read_unlock(&net->ipv6.fib6_walker_lock); 1452 1453 rt->dst.rt6_next = NULL; 1454 1455 /* If it was last route, expunge its radix tree node */ 1456 if (!fn->leaf) { 1457 fn->fn_flags &= ~RTN_RTINFO; 1458 net->ipv6.rt6_stats->fib_route_nodes--; 1459 fn = fib6_repair_tree(net, fn); 1460 } 1461 1462 fib6_purge_rt(rt, fn, net); 1463 1464 if (!info->skip_notify) 1465 inet6_rt_notify(RTM_DELROUTE, rt, info, 0); 1466 rt6_release(rt); 1467 } 1468 1469 int fib6_del(struct rt6_info *rt, struct nl_info *info) 1470 { 1471 struct net *net = info->nl_net; 1472 struct fib6_node *fn = rt->rt6i_node; 1473 struct rt6_info **rtp; 1474 1475 #if RT6_DEBUG >= 2 1476 if (rt->dst.obsolete > 0) { 1477 WARN_ON(fn); 1478 return -ENOENT; 1479 } 1480 #endif 1481 if (!fn || rt == net->ipv6.ip6_null_entry) 1482 return -ENOENT; 1483 1484 WARN_ON(!(fn->fn_flags & RTN_RTINFO)); 1485 1486 if (!(rt->rt6i_flags & RTF_CACHE)) { 1487 struct fib6_node *pn = fn; 1488 #ifdef CONFIG_IPV6_SUBTREES 1489 /* clones of this route might be in another subtree */ 1490 if (rt->rt6i_src.plen) { 1491 while (!(pn->fn_flags & RTN_ROOT)) 1492 pn = pn->parent; 1493 pn = pn->parent; 1494 } 1495 #endif 1496 fib6_prune_clones(info->nl_net, pn); 1497 } 1498 1499 /* 1500 * Walk the leaf entries looking for ourself 1501 */ 1502 1503 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) { 1504 if (*rtp == rt) { 1505 fib6_del_route(fn, rtp, info); 1506 return 0; 1507 } 1508 } 1509 return -ENOENT; 1510 } 1511 1512 /* 1513 * Tree traversal function. 1514 * 1515 * Certainly, it is not interrupt safe. 1516 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 1517 * It means, that we can modify tree during walking 1518 * and use this function for garbage collection, clone pruning, 1519 * cleaning tree when a device goes down etc. etc. 1520 * 1521 * It guarantees that every node will be traversed, 1522 * and that it will be traversed only once. 1523 * 1524 * Callback function w->func may return: 1525 * 0 -> continue walking. 1526 * positive value -> walking is suspended (used by tree dumps, 1527 * and probably by gc, if it will be split to several slices) 1528 * negative value -> terminate walking. 1529 * 1530 * The function itself returns: 1531 * 0 -> walk is complete. 1532 * >0 -> walk is incomplete (i.e. suspended) 1533 * <0 -> walk is terminated by an error. 1534 */ 1535 1536 static int fib6_walk_continue(struct fib6_walker *w) 1537 { 1538 struct fib6_node *fn, *pn; 1539 1540 for (;;) { 1541 fn = w->node; 1542 if (!fn) 1543 return 0; 1544 1545 if (w->prune && fn != w->root && 1546 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) { 1547 w->state = FWS_C; 1548 w->leaf = fn->leaf; 1549 } 1550 switch (w->state) { 1551 #ifdef CONFIG_IPV6_SUBTREES 1552 case FWS_S: 1553 if (FIB6_SUBTREE(fn)) { 1554 w->node = FIB6_SUBTREE(fn); 1555 continue; 1556 } 1557 w->state = FWS_L; 1558 #endif 1559 case FWS_L: 1560 if (fn->left) { 1561 w->node = fn->left; 1562 w->state = FWS_INIT; 1563 continue; 1564 } 1565 w->state = FWS_R; 1566 case FWS_R: 1567 if (fn->right) { 1568 w->node = fn->right; 1569 w->state = FWS_INIT; 1570 continue; 1571 } 1572 w->state = FWS_C; 1573 w->leaf = fn->leaf; 1574 case FWS_C: 1575 if (w->leaf && fn->fn_flags & RTN_RTINFO) { 1576 int err; 1577 1578 if (w->skip) { 1579 w->skip--; 1580 goto skip; 1581 } 1582 1583 err = w->func(w); 1584 if (err) 1585 return err; 1586 1587 w->count++; 1588 continue; 1589 } 1590 skip: 1591 w->state = FWS_U; 1592 case FWS_U: 1593 if (fn == w->root) 1594 return 0; 1595 pn = fn->parent; 1596 w->node = pn; 1597 #ifdef CONFIG_IPV6_SUBTREES 1598 if (FIB6_SUBTREE(pn) == fn) { 1599 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1600 w->state = FWS_L; 1601 continue; 1602 } 1603 #endif 1604 if (pn->left == fn) { 1605 w->state = FWS_R; 1606 continue; 1607 } 1608 if (pn->right == fn) { 1609 w->state = FWS_C; 1610 w->leaf = w->node->leaf; 1611 continue; 1612 } 1613 #if RT6_DEBUG >= 2 1614 WARN_ON(1); 1615 #endif 1616 } 1617 } 1618 } 1619 1620 static int fib6_walk(struct net *net, struct fib6_walker *w) 1621 { 1622 int res; 1623 1624 w->state = FWS_INIT; 1625 w->node = w->root; 1626 1627 fib6_walker_link(net, w); 1628 res = fib6_walk_continue(w); 1629 if (res <= 0) 1630 fib6_walker_unlink(net, w); 1631 return res; 1632 } 1633 1634 static int fib6_clean_node(struct fib6_walker *w) 1635 { 1636 int res; 1637 struct rt6_info *rt; 1638 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w); 1639 struct nl_info info = { 1640 .nl_net = c->net, 1641 }; 1642 1643 if (c->sernum != FIB6_NO_SERNUM_CHANGE && 1644 w->node->fn_sernum != c->sernum) 1645 w->node->fn_sernum = c->sernum; 1646 1647 if (!c->func) { 1648 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE); 1649 w->leaf = NULL; 1650 return 0; 1651 } 1652 1653 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 1654 res = c->func(rt, c->arg); 1655 if (res < 0) { 1656 w->leaf = rt; 1657 res = fib6_del(rt, &info); 1658 if (res) { 1659 #if RT6_DEBUG >= 2 1660 pr_debug("%s: del failed: rt=%p@%p err=%d\n", 1661 __func__, rt, rt->rt6i_node, res); 1662 #endif 1663 continue; 1664 } 1665 return 0; 1666 } 1667 WARN_ON(res != 0); 1668 } 1669 w->leaf = rt; 1670 return 0; 1671 } 1672 1673 /* 1674 * Convenient frontend to tree walker. 1675 * 1676 * func is called on each route. 1677 * It may return -1 -> delete this route. 1678 * 0 -> continue walking 1679 * 1680 * prune==1 -> only immediate children of node (certainly, 1681 * ignoring pure split nodes) will be scanned. 1682 */ 1683 1684 static void fib6_clean_tree(struct net *net, struct fib6_node *root, 1685 int (*func)(struct rt6_info *, void *arg), 1686 bool prune, int sernum, void *arg) 1687 { 1688 struct fib6_cleaner c; 1689 1690 c.w.root = root; 1691 c.w.func = fib6_clean_node; 1692 c.w.prune = prune; 1693 c.w.count = 0; 1694 c.w.skip = 0; 1695 c.func = func; 1696 c.sernum = sernum; 1697 c.arg = arg; 1698 c.net = net; 1699 1700 fib6_walk(net, &c.w); 1701 } 1702 1703 static void __fib6_clean_all(struct net *net, 1704 int (*func)(struct rt6_info *, void *), 1705 int sernum, void *arg) 1706 { 1707 struct fib6_table *table; 1708 struct hlist_head *head; 1709 unsigned int h; 1710 1711 rcu_read_lock(); 1712 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 1713 head = &net->ipv6.fib_table_hash[h]; 1714 hlist_for_each_entry_rcu(table, head, tb6_hlist) { 1715 write_lock_bh(&table->tb6_lock); 1716 fib6_clean_tree(net, &table->tb6_root, 1717 func, false, sernum, arg); 1718 write_unlock_bh(&table->tb6_lock); 1719 } 1720 } 1721 rcu_read_unlock(); 1722 } 1723 1724 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *), 1725 void *arg) 1726 { 1727 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg); 1728 } 1729 1730 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1731 { 1732 if (rt->rt6i_flags & RTF_CACHE) { 1733 RT6_TRACE("pruning clone %p\n", rt); 1734 return -1; 1735 } 1736 1737 return 0; 1738 } 1739 1740 static void fib6_prune_clones(struct net *net, struct fib6_node *fn) 1741 { 1742 fib6_clean_tree(net, fn, fib6_prune_clone, true, 1743 FIB6_NO_SERNUM_CHANGE, NULL); 1744 } 1745 1746 static void fib6_flush_trees(struct net *net) 1747 { 1748 int new_sernum = fib6_new_sernum(net); 1749 1750 __fib6_clean_all(net, NULL, new_sernum, NULL); 1751 } 1752 1753 /* 1754 * Garbage collection 1755 */ 1756 1757 struct fib6_gc_args 1758 { 1759 int timeout; 1760 int more; 1761 }; 1762 1763 static int fib6_age(struct rt6_info *rt, void *arg) 1764 { 1765 struct fib6_gc_args *gc_args = arg; 1766 unsigned long now = jiffies; 1767 1768 /* 1769 * check addrconf expiration here. 1770 * Routes are expired even if they are in use. 1771 * 1772 * Also age clones. Note, that clones are aged out 1773 * only if they are not in use now. 1774 */ 1775 1776 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) { 1777 if (time_after(now, rt->dst.expires)) { 1778 RT6_TRACE("expiring %p\n", rt); 1779 return -1; 1780 } 1781 gc_args->more++; 1782 } else if (rt->rt6i_flags & RTF_CACHE) { 1783 if (atomic_read(&rt->dst.__refcnt) == 1 && 1784 time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) { 1785 RT6_TRACE("aging clone %p\n", rt); 1786 return -1; 1787 } else if (rt->rt6i_flags & RTF_GATEWAY) { 1788 struct neighbour *neigh; 1789 __u8 neigh_flags = 0; 1790 1791 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway); 1792 if (neigh) { 1793 neigh_flags = neigh->flags; 1794 neigh_release(neigh); 1795 } 1796 if (!(neigh_flags & NTF_ROUTER)) { 1797 RT6_TRACE("purging route %p via non-router but gateway\n", 1798 rt); 1799 return -1; 1800 } 1801 } 1802 gc_args->more++; 1803 } 1804 1805 return 0; 1806 } 1807 1808 void fib6_run_gc(unsigned long expires, struct net *net, bool force) 1809 { 1810 struct fib6_gc_args gc_args; 1811 unsigned long now; 1812 1813 if (force) { 1814 spin_lock_bh(&net->ipv6.fib6_gc_lock); 1815 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) { 1816 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); 1817 return; 1818 } 1819 gc_args.timeout = expires ? (int)expires : 1820 net->ipv6.sysctl.ip6_rt_gc_interval; 1821 gc_args.more = 0; 1822 1823 fib6_clean_all(net, fib6_age, &gc_args); 1824 now = jiffies; 1825 net->ipv6.ip6_rt_last_gc = now; 1826 1827 if (gc_args.more) 1828 mod_timer(&net->ipv6.ip6_fib_timer, 1829 round_jiffies(now 1830 + net->ipv6.sysctl.ip6_rt_gc_interval)); 1831 else 1832 del_timer(&net->ipv6.ip6_fib_timer); 1833 spin_unlock_bh(&net->ipv6.fib6_gc_lock); 1834 } 1835 1836 static void fib6_gc_timer_cb(unsigned long arg) 1837 { 1838 fib6_run_gc(0, (struct net *)arg, true); 1839 } 1840 1841 static int __net_init fib6_net_init(struct net *net) 1842 { 1843 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; 1844 1845 spin_lock_init(&net->ipv6.fib6_gc_lock); 1846 rwlock_init(&net->ipv6.fib6_walker_lock); 1847 INIT_LIST_HEAD(&net->ipv6.fib6_walkers); 1848 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net); 1849 1850 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); 1851 if (!net->ipv6.rt6_stats) 1852 goto out_timer; 1853 1854 /* Avoid false sharing : Use at least a full cache line */ 1855 size = max_t(size_t, size, L1_CACHE_BYTES); 1856 1857 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); 1858 if (!net->ipv6.fib_table_hash) 1859 goto out_rt6_stats; 1860 1861 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), 1862 GFP_KERNEL); 1863 if (!net->ipv6.fib6_main_tbl) 1864 goto out_fib_table_hash; 1865 1866 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; 1867 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1868 net->ipv6.fib6_main_tbl->tb6_root.fn_flags = 1869 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1870 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); 1871 1872 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1873 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), 1874 GFP_KERNEL); 1875 if (!net->ipv6.fib6_local_tbl) 1876 goto out_fib6_main_tbl; 1877 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; 1878 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1879 net->ipv6.fib6_local_tbl->tb6_root.fn_flags = 1880 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1881 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); 1882 #endif 1883 fib6_tables_init(net); 1884 1885 return 0; 1886 1887 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1888 out_fib6_main_tbl: 1889 kfree(net->ipv6.fib6_main_tbl); 1890 #endif 1891 out_fib_table_hash: 1892 kfree(net->ipv6.fib_table_hash); 1893 out_rt6_stats: 1894 kfree(net->ipv6.rt6_stats); 1895 out_timer: 1896 return -ENOMEM; 1897 } 1898 1899 static void fib6_net_exit(struct net *net) 1900 { 1901 rt6_ifdown(net, NULL); 1902 del_timer_sync(&net->ipv6.ip6_fib_timer); 1903 1904 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1905 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers); 1906 kfree(net->ipv6.fib6_local_tbl); 1907 #endif 1908 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers); 1909 kfree(net->ipv6.fib6_main_tbl); 1910 kfree(net->ipv6.fib_table_hash); 1911 kfree(net->ipv6.rt6_stats); 1912 } 1913 1914 static struct pernet_operations fib6_net_ops = { 1915 .init = fib6_net_init, 1916 .exit = fib6_net_exit, 1917 }; 1918 1919 int __init fib6_init(void) 1920 { 1921 int ret = -ENOMEM; 1922 1923 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1924 sizeof(struct fib6_node), 1925 0, SLAB_HWCACHE_ALIGN, 1926 NULL); 1927 if (!fib6_node_kmem) 1928 goto out; 1929 1930 ret = register_pernet_subsys(&fib6_net_ops); 1931 if (ret) 1932 goto out_kmem_cache_create; 1933 1934 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib, 1935 NULL); 1936 if (ret) 1937 goto out_unregister_subsys; 1938 1939 __fib6_flush_trees = fib6_flush_trees; 1940 out: 1941 return ret; 1942 1943 out_unregister_subsys: 1944 unregister_pernet_subsys(&fib6_net_ops); 1945 out_kmem_cache_create: 1946 kmem_cache_destroy(fib6_node_kmem); 1947 goto out; 1948 } 1949 1950 void fib6_gc_cleanup(void) 1951 { 1952 unregister_pernet_subsys(&fib6_net_ops); 1953 kmem_cache_destroy(fib6_node_kmem); 1954 } 1955 1956 #ifdef CONFIG_PROC_FS 1957 1958 struct ipv6_route_iter { 1959 struct seq_net_private p; 1960 struct fib6_walker w; 1961 loff_t skip; 1962 struct fib6_table *tbl; 1963 int sernum; 1964 }; 1965 1966 static int ipv6_route_seq_show(struct seq_file *seq, void *v) 1967 { 1968 struct rt6_info *rt = v; 1969 struct ipv6_route_iter *iter = seq->private; 1970 1971 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen); 1972 1973 #ifdef CONFIG_IPV6_SUBTREES 1974 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen); 1975 #else 1976 seq_puts(seq, "00000000000000000000000000000000 00 "); 1977 #endif 1978 if (rt->rt6i_flags & RTF_GATEWAY) 1979 seq_printf(seq, "%pi6", &rt->rt6i_gateway); 1980 else 1981 seq_puts(seq, "00000000000000000000000000000000"); 1982 1983 seq_printf(seq, " %08x %08x %08x %08x %8s\n", 1984 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt), 1985 rt->dst.__use, rt->rt6i_flags, 1986 rt->dst.dev ? rt->dst.dev->name : ""); 1987 iter->w.leaf = NULL; 1988 return 0; 1989 } 1990 1991 static int ipv6_route_yield(struct fib6_walker *w) 1992 { 1993 struct ipv6_route_iter *iter = w->args; 1994 1995 if (!iter->skip) 1996 return 1; 1997 1998 do { 1999 iter->w.leaf = iter->w.leaf->dst.rt6_next; 2000 iter->skip--; 2001 if (!iter->skip && iter->w.leaf) 2002 return 1; 2003 } while (iter->w.leaf); 2004 2005 return 0; 2006 } 2007 2008 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter, 2009 struct net *net) 2010 { 2011 memset(&iter->w, 0, sizeof(iter->w)); 2012 iter->w.func = ipv6_route_yield; 2013 iter->w.root = &iter->tbl->tb6_root; 2014 iter->w.state = FWS_INIT; 2015 iter->w.node = iter->w.root; 2016 iter->w.args = iter; 2017 iter->sernum = iter->w.root->fn_sernum; 2018 INIT_LIST_HEAD(&iter->w.lh); 2019 fib6_walker_link(net, &iter->w); 2020 } 2021 2022 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl, 2023 struct net *net) 2024 { 2025 unsigned int h; 2026 struct hlist_node *node; 2027 2028 if (tbl) { 2029 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1; 2030 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist)); 2031 } else { 2032 h = 0; 2033 node = NULL; 2034 } 2035 2036 while (!node && h < FIB6_TABLE_HASHSZ) { 2037 node = rcu_dereference_bh( 2038 hlist_first_rcu(&net->ipv6.fib_table_hash[h++])); 2039 } 2040 return hlist_entry_safe(node, struct fib6_table, tb6_hlist); 2041 } 2042 2043 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter) 2044 { 2045 if (iter->sernum != iter->w.root->fn_sernum) { 2046 iter->sernum = iter->w.root->fn_sernum; 2047 iter->w.state = FWS_INIT; 2048 iter->w.node = iter->w.root; 2049 WARN_ON(iter->w.skip); 2050 iter->w.skip = iter->w.count; 2051 } 2052 } 2053 2054 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2055 { 2056 int r; 2057 struct rt6_info *n; 2058 struct net *net = seq_file_net(seq); 2059 struct ipv6_route_iter *iter = seq->private; 2060 2061 if (!v) 2062 goto iter_table; 2063 2064 n = ((struct rt6_info *)v)->dst.rt6_next; 2065 if (n) { 2066 ++*pos; 2067 return n; 2068 } 2069 2070 iter_table: 2071 ipv6_route_check_sernum(iter); 2072 read_lock(&iter->tbl->tb6_lock); 2073 r = fib6_walk_continue(&iter->w); 2074 read_unlock(&iter->tbl->tb6_lock); 2075 if (r > 0) { 2076 if (v) 2077 ++*pos; 2078 return iter->w.leaf; 2079 } else if (r < 0) { 2080 fib6_walker_unlink(net, &iter->w); 2081 return NULL; 2082 } 2083 fib6_walker_unlink(net, &iter->w); 2084 2085 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net); 2086 if (!iter->tbl) 2087 return NULL; 2088 2089 ipv6_route_seq_setup_walk(iter, net); 2090 goto iter_table; 2091 } 2092 2093 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos) 2094 __acquires(RCU_BH) 2095 { 2096 struct net *net = seq_file_net(seq); 2097 struct ipv6_route_iter *iter = seq->private; 2098 2099 rcu_read_lock_bh(); 2100 iter->tbl = ipv6_route_seq_next_table(NULL, net); 2101 iter->skip = *pos; 2102 2103 if (iter->tbl) { 2104 ipv6_route_seq_setup_walk(iter, net); 2105 return ipv6_route_seq_next(seq, NULL, pos); 2106 } else { 2107 return NULL; 2108 } 2109 } 2110 2111 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter) 2112 { 2113 struct fib6_walker *w = &iter->w; 2114 return w->node && !(w->state == FWS_U && w->node == w->root); 2115 } 2116 2117 static void ipv6_route_seq_stop(struct seq_file *seq, void *v) 2118 __releases(RCU_BH) 2119 { 2120 struct net *net = seq_file_net(seq); 2121 struct ipv6_route_iter *iter = seq->private; 2122 2123 if (ipv6_route_iter_active(iter)) 2124 fib6_walker_unlink(net, &iter->w); 2125 2126 rcu_read_unlock_bh(); 2127 } 2128 2129 static const struct seq_operations ipv6_route_seq_ops = { 2130 .start = ipv6_route_seq_start, 2131 .next = ipv6_route_seq_next, 2132 .stop = ipv6_route_seq_stop, 2133 .show = ipv6_route_seq_show 2134 }; 2135 2136 int ipv6_route_open(struct inode *inode, struct file *file) 2137 { 2138 return seq_open_net(inode, file, &ipv6_route_seq_ops, 2139 sizeof(struct ipv6_route_iter)); 2140 } 2141 2142 #endif /* CONFIG_PROC_FS */ 2143