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_immediate(struct fib6_node *fn) 152 { 153 kmem_cache_free(fib6_node_kmem, fn); 154 } 155 156 static void node_free_rcu(struct rcu_head *head) 157 { 158 struct fib6_node *fn = container_of(head, struct fib6_node, rcu); 159 160 kmem_cache_free(fib6_node_kmem, fn); 161 } 162 163 static void node_free(struct fib6_node *fn) 164 { 165 call_rcu(&fn->rcu, node_free_rcu); 166 } 167 168 static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt) 169 { 170 int cpu; 171 172 if (!non_pcpu_rt->rt6i_pcpu) 173 return; 174 175 for_each_possible_cpu(cpu) { 176 struct rt6_info **ppcpu_rt; 177 struct rt6_info *pcpu_rt; 178 179 ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu); 180 pcpu_rt = *ppcpu_rt; 181 if (pcpu_rt) { 182 dst_dev_put(&pcpu_rt->dst); 183 dst_release(&pcpu_rt->dst); 184 *ppcpu_rt = NULL; 185 } 186 } 187 188 free_percpu(non_pcpu_rt->rt6i_pcpu); 189 non_pcpu_rt->rt6i_pcpu = NULL; 190 } 191 192 static void rt6_release(struct rt6_info *rt) 193 { 194 if (atomic_dec_and_test(&rt->rt6i_ref)) { 195 rt6_free_pcpu(rt); 196 dst_dev_put(&rt->dst); 197 dst_release(&rt->dst); 198 } 199 } 200 201 static void fib6_link_table(struct net *net, struct fib6_table *tb) 202 { 203 unsigned int h; 204 205 /* 206 * Initialize table lock at a single place to give lockdep a key, 207 * tables aren't visible prior to being linked to the list. 208 */ 209 rwlock_init(&tb->tb6_lock); 210 211 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1); 212 213 /* 214 * No protection necessary, this is the only list mutatation 215 * operation, tables never disappear once they exist. 216 */ 217 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]); 218 } 219 220 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 221 222 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id) 223 { 224 struct fib6_table *table; 225 226 table = kzalloc(sizeof(*table), GFP_ATOMIC); 227 if (table) { 228 table->tb6_id = id; 229 table->tb6_root.leaf = net->ipv6.ip6_null_entry; 230 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 231 inet_peer_base_init(&table->tb6_peers); 232 } 233 234 return table; 235 } 236 237 struct fib6_table *fib6_new_table(struct net *net, u32 id) 238 { 239 struct fib6_table *tb; 240 241 if (id == 0) 242 id = RT6_TABLE_MAIN; 243 tb = fib6_get_table(net, id); 244 if (tb) 245 return tb; 246 247 tb = fib6_alloc_table(net, id); 248 if (tb) 249 fib6_link_table(net, tb); 250 251 return tb; 252 } 253 EXPORT_SYMBOL_GPL(fib6_new_table); 254 255 struct fib6_table *fib6_get_table(struct net *net, u32 id) 256 { 257 struct fib6_table *tb; 258 struct hlist_head *head; 259 unsigned int h; 260 261 if (id == 0) 262 id = RT6_TABLE_MAIN; 263 h = id & (FIB6_TABLE_HASHSZ - 1); 264 rcu_read_lock(); 265 head = &net->ipv6.fib_table_hash[h]; 266 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 267 if (tb->tb6_id == id) { 268 rcu_read_unlock(); 269 return tb; 270 } 271 } 272 rcu_read_unlock(); 273 274 return NULL; 275 } 276 EXPORT_SYMBOL_GPL(fib6_get_table); 277 278 static void __net_init fib6_tables_init(struct net *net) 279 { 280 fib6_link_table(net, net->ipv6.fib6_main_tbl); 281 fib6_link_table(net, net->ipv6.fib6_local_tbl); 282 } 283 #else 284 285 struct fib6_table *fib6_new_table(struct net *net, u32 id) 286 { 287 return fib6_get_table(net, id); 288 } 289 290 struct fib6_table *fib6_get_table(struct net *net, u32 id) 291 { 292 return net->ipv6.fib6_main_tbl; 293 } 294 295 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, 296 int flags, pol_lookup_t lookup) 297 { 298 struct rt6_info *rt; 299 300 rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags); 301 if (rt->dst.error == -EAGAIN) { 302 ip6_rt_put(rt); 303 rt = net->ipv6.ip6_null_entry; 304 dst_hold(&rt->dst); 305 } 306 307 return &rt->dst; 308 } 309 310 static void __net_init fib6_tables_init(struct net *net) 311 { 312 fib6_link_table(net, net->ipv6.fib6_main_tbl); 313 } 314 315 #endif 316 317 static int fib6_dump_node(struct fib6_walker *w) 318 { 319 int res; 320 struct rt6_info *rt; 321 322 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 323 res = rt6_dump_route(rt, w->args); 324 if (res < 0) { 325 /* Frame is full, suspend walking */ 326 w->leaf = rt; 327 return 1; 328 } 329 330 /* Multipath routes are dumped in one route with the 331 * RTA_MULTIPATH attribute. Jump 'rt' to point to the 332 * last sibling of this route (no need to dump the 333 * sibling routes again) 334 */ 335 if (rt->rt6i_nsiblings) 336 rt = list_last_entry(&rt->rt6i_siblings, 337 struct rt6_info, 338 rt6i_siblings); 339 } 340 w->leaf = NULL; 341 return 0; 342 } 343 344 static void fib6_dump_end(struct netlink_callback *cb) 345 { 346 struct net *net = sock_net(cb->skb->sk); 347 struct fib6_walker *w = (void *)cb->args[2]; 348 349 if (w) { 350 if (cb->args[4]) { 351 cb->args[4] = 0; 352 fib6_walker_unlink(net, w); 353 } 354 cb->args[2] = 0; 355 kfree(w); 356 } 357 cb->done = (void *)cb->args[3]; 358 cb->args[1] = 3; 359 } 360 361 static int fib6_dump_done(struct netlink_callback *cb) 362 { 363 fib6_dump_end(cb); 364 return cb->done ? cb->done(cb) : 0; 365 } 366 367 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 368 struct netlink_callback *cb) 369 { 370 struct net *net = sock_net(skb->sk); 371 struct fib6_walker *w; 372 int res; 373 374 w = (void *)cb->args[2]; 375 w->root = &table->tb6_root; 376 377 if (cb->args[4] == 0) { 378 w->count = 0; 379 w->skip = 0; 380 381 read_lock_bh(&table->tb6_lock); 382 res = fib6_walk(net, w); 383 read_unlock_bh(&table->tb6_lock); 384 if (res > 0) { 385 cb->args[4] = 1; 386 cb->args[5] = w->root->fn_sernum; 387 } 388 } else { 389 if (cb->args[5] != w->root->fn_sernum) { 390 /* Begin at the root if the tree changed */ 391 cb->args[5] = w->root->fn_sernum; 392 w->state = FWS_INIT; 393 w->node = w->root; 394 w->skip = w->count; 395 } else 396 w->skip = 0; 397 398 read_lock_bh(&table->tb6_lock); 399 res = fib6_walk_continue(w); 400 read_unlock_bh(&table->tb6_lock); 401 if (res <= 0) { 402 fib6_walker_unlink(net, w); 403 cb->args[4] = 0; 404 } 405 } 406 407 return res; 408 } 409 410 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 411 { 412 struct net *net = sock_net(skb->sk); 413 unsigned int h, s_h; 414 unsigned int e = 0, s_e; 415 struct rt6_rtnl_dump_arg arg; 416 struct fib6_walker *w; 417 struct fib6_table *tb; 418 struct hlist_head *head; 419 int res = 0; 420 421 s_h = cb->args[0]; 422 s_e = cb->args[1]; 423 424 w = (void *)cb->args[2]; 425 if (!w) { 426 /* New dump: 427 * 428 * 1. hook callback destructor. 429 */ 430 cb->args[3] = (long)cb->done; 431 cb->done = fib6_dump_done; 432 433 /* 434 * 2. allocate and initialize walker. 435 */ 436 w = kzalloc(sizeof(*w), GFP_ATOMIC); 437 if (!w) 438 return -ENOMEM; 439 w->func = fib6_dump_node; 440 cb->args[2] = (long)w; 441 } 442 443 arg.skb = skb; 444 arg.cb = cb; 445 arg.net = net; 446 w->args = &arg; 447 448 rcu_read_lock(); 449 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) { 450 e = 0; 451 head = &net->ipv6.fib_table_hash[h]; 452 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 453 if (e < s_e) 454 goto next; 455 res = fib6_dump_table(tb, skb, cb); 456 if (res != 0) 457 goto out; 458 next: 459 e++; 460 } 461 } 462 out: 463 rcu_read_unlock(); 464 cb->args[1] = e; 465 cb->args[0] = h; 466 467 res = res < 0 ? res : skb->len; 468 if (res <= 0) 469 fib6_dump_end(cb); 470 return res; 471 } 472 473 /* 474 * Routing Table 475 * 476 * return the appropriate node for a routing tree "add" operation 477 * by either creating and inserting or by returning an existing 478 * node. 479 */ 480 481 static struct fib6_node *fib6_add_1(struct fib6_node *root, 482 struct in6_addr *addr, int plen, 483 int offset, int allow_create, 484 int replace_required, int sernum, 485 struct netlink_ext_ack *extack) 486 { 487 struct fib6_node *fn, *in, *ln; 488 struct fib6_node *pn = NULL; 489 struct rt6key *key; 490 int bit; 491 __be32 dir = 0; 492 493 RT6_TRACE("fib6_add_1\n"); 494 495 /* insert node in tree */ 496 497 fn = root; 498 499 do { 500 key = (struct rt6key *)((u8 *)fn->leaf + offset); 501 502 /* 503 * Prefix match 504 */ 505 if (plen < fn->fn_bit || 506 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) { 507 if (!allow_create) { 508 if (replace_required) { 509 NL_SET_ERR_MSG(extack, 510 "Can not replace route - no match found"); 511 pr_warn("Can't replace route, no match found\n"); 512 return ERR_PTR(-ENOENT); 513 } 514 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 515 } 516 goto insert_above; 517 } 518 519 /* 520 * Exact match ? 521 */ 522 523 if (plen == fn->fn_bit) { 524 /* clean up an intermediate node */ 525 if (!(fn->fn_flags & RTN_RTINFO)) { 526 rt6_release(fn->leaf); 527 fn->leaf = NULL; 528 } 529 530 fn->fn_sernum = sernum; 531 532 return fn; 533 } 534 535 /* 536 * We have more bits to go 537 */ 538 539 /* Try to walk down on tree. */ 540 fn->fn_sernum = sernum; 541 dir = addr_bit_set(addr, fn->fn_bit); 542 pn = fn; 543 fn = dir ? fn->right : fn->left; 544 } while (fn); 545 546 if (!allow_create) { 547 /* We should not create new node because 548 * NLM_F_REPLACE was specified without NLM_F_CREATE 549 * I assume it is safe to require NLM_F_CREATE when 550 * REPLACE flag is used! Later we may want to remove the 551 * check for replace_required, because according 552 * to netlink specification, NLM_F_CREATE 553 * MUST be specified if new route is created. 554 * That would keep IPv6 consistent with IPv4 555 */ 556 if (replace_required) { 557 NL_SET_ERR_MSG(extack, 558 "Can not replace route - no match found"); 559 pr_warn("Can't replace route, no match found\n"); 560 return ERR_PTR(-ENOENT); 561 } 562 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 563 } 564 /* 565 * We walked to the bottom of tree. 566 * Create new leaf node without children. 567 */ 568 569 ln = node_alloc(); 570 571 if (!ln) 572 return ERR_PTR(-ENOMEM); 573 ln->fn_bit = plen; 574 575 ln->parent = pn; 576 ln->fn_sernum = sernum; 577 578 if (dir) 579 pn->right = ln; 580 else 581 pn->left = ln; 582 583 return ln; 584 585 586 insert_above: 587 /* 588 * split since we don't have a common prefix anymore or 589 * we have a less significant route. 590 * we've to insert an intermediate node on the list 591 * this new node will point to the one we need to create 592 * and the current 593 */ 594 595 pn = fn->parent; 596 597 /* find 1st bit in difference between the 2 addrs. 598 599 See comment in __ipv6_addr_diff: bit may be an invalid value, 600 but if it is >= plen, the value is ignored in any case. 601 */ 602 603 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr)); 604 605 /* 606 * (intermediate)[in] 607 * / \ 608 * (new leaf node)[ln] (old node)[fn] 609 */ 610 if (plen > bit) { 611 in = node_alloc(); 612 ln = node_alloc(); 613 614 if (!in || !ln) { 615 if (in) 616 node_free_immediate(in); 617 if (ln) 618 node_free_immediate(ln); 619 return ERR_PTR(-ENOMEM); 620 } 621 622 /* 623 * new intermediate node. 624 * RTN_RTINFO will 625 * be off since that an address that chooses one of 626 * the branches would not match less specific routes 627 * in the other branch 628 */ 629 630 in->fn_bit = bit; 631 632 in->parent = pn; 633 in->leaf = fn->leaf; 634 atomic_inc(&in->leaf->rt6i_ref); 635 636 in->fn_sernum = sernum; 637 638 /* update parent pointer */ 639 if (dir) 640 pn->right = in; 641 else 642 pn->left = in; 643 644 ln->fn_bit = plen; 645 646 ln->parent = in; 647 fn->parent = in; 648 649 ln->fn_sernum = sernum; 650 651 if (addr_bit_set(addr, bit)) { 652 in->right = ln; 653 in->left = fn; 654 } else { 655 in->left = ln; 656 in->right = fn; 657 } 658 } else { /* plen <= bit */ 659 660 /* 661 * (new leaf node)[ln] 662 * / \ 663 * (old node)[fn] NULL 664 */ 665 666 ln = node_alloc(); 667 668 if (!ln) 669 return ERR_PTR(-ENOMEM); 670 671 ln->fn_bit = plen; 672 673 ln->parent = pn; 674 675 ln->fn_sernum = sernum; 676 677 if (dir) 678 pn->right = ln; 679 else 680 pn->left = ln; 681 682 if (addr_bit_set(&key->addr, plen)) 683 ln->right = fn; 684 else 685 ln->left = fn; 686 687 fn->parent = ln; 688 } 689 return ln; 690 } 691 692 static bool rt6_qualify_for_ecmp(struct rt6_info *rt) 693 { 694 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) == 695 RTF_GATEWAY; 696 } 697 698 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc) 699 { 700 int i; 701 702 for (i = 0; i < RTAX_MAX; i++) { 703 if (test_bit(i, mxc->mx_valid)) 704 mp[i] = mxc->mx[i]; 705 } 706 } 707 708 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc) 709 { 710 if (!mxc->mx) 711 return 0; 712 713 if (dst->flags & DST_HOST) { 714 u32 *mp = dst_metrics_write_ptr(dst); 715 716 if (unlikely(!mp)) 717 return -ENOMEM; 718 719 fib6_copy_metrics(mp, mxc); 720 } else { 721 dst_init_metrics(dst, mxc->mx, false); 722 723 /* We've stolen mx now. */ 724 mxc->mx = NULL; 725 } 726 727 return 0; 728 } 729 730 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn, 731 struct net *net) 732 { 733 if (atomic_read(&rt->rt6i_ref) != 1) { 734 /* This route is used as dummy address holder in some split 735 * nodes. It is not leaked, but it still holds other resources, 736 * which must be released in time. So, scan ascendant nodes 737 * and replace dummy references to this route with references 738 * to still alive ones. 739 */ 740 while (fn) { 741 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) { 742 fn->leaf = fib6_find_prefix(net, fn); 743 atomic_inc(&fn->leaf->rt6i_ref); 744 rt6_release(rt); 745 } 746 fn = fn->parent; 747 } 748 /* No more references are possible at this point. */ 749 BUG_ON(atomic_read(&rt->rt6i_ref) != 1); 750 } 751 } 752 753 /* 754 * Insert routing information in a node. 755 */ 756 757 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, 758 struct nl_info *info, struct mx6_config *mxc) 759 { 760 struct rt6_info *iter = NULL; 761 struct rt6_info **ins; 762 struct rt6_info **fallback_ins = NULL; 763 int replace = (info->nlh && 764 (info->nlh->nlmsg_flags & NLM_F_REPLACE)); 765 int add = (!info->nlh || 766 (info->nlh->nlmsg_flags & NLM_F_CREATE)); 767 int found = 0; 768 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt); 769 u16 nlflags = NLM_F_EXCL; 770 int err; 771 772 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND)) 773 nlflags |= NLM_F_APPEND; 774 775 ins = &fn->leaf; 776 777 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) { 778 /* 779 * Search for duplicates 780 */ 781 782 if (iter->rt6i_metric == rt->rt6i_metric) { 783 /* 784 * Same priority level 785 */ 786 if (info->nlh && 787 (info->nlh->nlmsg_flags & NLM_F_EXCL)) 788 return -EEXIST; 789 790 nlflags &= ~NLM_F_EXCL; 791 if (replace) { 792 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) { 793 found++; 794 break; 795 } 796 if (rt_can_ecmp) 797 fallback_ins = fallback_ins ?: ins; 798 goto next_iter; 799 } 800 801 if (rt6_duplicate_nexthop(iter, rt)) { 802 if (rt->rt6i_nsiblings) 803 rt->rt6i_nsiblings = 0; 804 if (!(iter->rt6i_flags & RTF_EXPIRES)) 805 return -EEXIST; 806 if (!(rt->rt6i_flags & RTF_EXPIRES)) 807 rt6_clean_expires(iter); 808 else 809 rt6_set_expires(iter, rt->dst.expires); 810 iter->rt6i_pmtu = rt->rt6i_pmtu; 811 return -EEXIST; 812 } 813 /* If we have the same destination and the same metric, 814 * but not the same gateway, then the route we try to 815 * add is sibling to this route, increment our counter 816 * of siblings, and later we will add our route to the 817 * list. 818 * Only static routes (which don't have flag 819 * RTF_EXPIRES) are used for ECMPv6. 820 * 821 * To avoid long list, we only had siblings if the 822 * route have a gateway. 823 */ 824 if (rt_can_ecmp && 825 rt6_qualify_for_ecmp(iter)) 826 rt->rt6i_nsiblings++; 827 } 828 829 if (iter->rt6i_metric > rt->rt6i_metric) 830 break; 831 832 next_iter: 833 ins = &iter->dst.rt6_next; 834 } 835 836 if (fallback_ins && !found) { 837 /* No ECMP-able route found, replace first non-ECMP one */ 838 ins = fallback_ins; 839 iter = *ins; 840 found++; 841 } 842 843 /* Reset round-robin state, if necessary */ 844 if (ins == &fn->leaf) 845 fn->rr_ptr = NULL; 846 847 /* Link this route to others same route. */ 848 if (rt->rt6i_nsiblings) { 849 unsigned int rt6i_nsiblings; 850 struct rt6_info *sibling, *temp_sibling; 851 852 /* Find the first route that have the same metric */ 853 sibling = fn->leaf; 854 while (sibling) { 855 if (sibling->rt6i_metric == rt->rt6i_metric && 856 rt6_qualify_for_ecmp(sibling)) { 857 list_add_tail(&rt->rt6i_siblings, 858 &sibling->rt6i_siblings); 859 break; 860 } 861 sibling = sibling->dst.rt6_next; 862 } 863 /* For each sibling in the list, increment the counter of 864 * siblings. BUG() if counters does not match, list of siblings 865 * is broken! 866 */ 867 rt6i_nsiblings = 0; 868 list_for_each_entry_safe(sibling, temp_sibling, 869 &rt->rt6i_siblings, rt6i_siblings) { 870 sibling->rt6i_nsiblings++; 871 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings); 872 rt6i_nsiblings++; 873 } 874 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings); 875 } 876 877 /* 878 * insert node 879 */ 880 if (!replace) { 881 if (!add) 882 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 883 884 add: 885 nlflags |= NLM_F_CREATE; 886 err = fib6_commit_metrics(&rt->dst, mxc); 887 if (err) 888 return err; 889 890 rt->dst.rt6_next = iter; 891 *ins = rt; 892 rcu_assign_pointer(rt->rt6i_node, fn); 893 atomic_inc(&rt->rt6i_ref); 894 if (!info->skip_notify) 895 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags); 896 info->nl_net->ipv6.rt6_stats->fib_rt_entries++; 897 898 if (!(fn->fn_flags & RTN_RTINFO)) { 899 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 900 fn->fn_flags |= RTN_RTINFO; 901 } 902 903 } else { 904 int nsiblings; 905 906 if (!found) { 907 if (add) 908 goto add; 909 pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); 910 return -ENOENT; 911 } 912 913 err = fib6_commit_metrics(&rt->dst, mxc); 914 if (err) 915 return err; 916 917 *ins = rt; 918 rcu_assign_pointer(rt->rt6i_node, fn); 919 rt->dst.rt6_next = iter->dst.rt6_next; 920 atomic_inc(&rt->rt6i_ref); 921 if (!info->skip_notify) 922 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE); 923 if (!(fn->fn_flags & RTN_RTINFO)) { 924 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 925 fn->fn_flags |= RTN_RTINFO; 926 } 927 nsiblings = iter->rt6i_nsiblings; 928 fib6_purge_rt(iter, fn, info->nl_net); 929 if (fn->rr_ptr == iter) 930 fn->rr_ptr = NULL; 931 rt6_release(iter); 932 933 if (nsiblings) { 934 /* Replacing an ECMP route, remove all siblings */ 935 ins = &rt->dst.rt6_next; 936 iter = *ins; 937 while (iter) { 938 if (iter->rt6i_metric > rt->rt6i_metric) 939 break; 940 if (rt6_qualify_for_ecmp(iter)) { 941 *ins = iter->dst.rt6_next; 942 fib6_purge_rt(iter, fn, info->nl_net); 943 if (fn->rr_ptr == iter) 944 fn->rr_ptr = NULL; 945 rt6_release(iter); 946 nsiblings--; 947 } else { 948 ins = &iter->dst.rt6_next; 949 } 950 iter = *ins; 951 } 952 WARN_ON(nsiblings != 0); 953 } 954 } 955 956 return 0; 957 } 958 959 static void fib6_start_gc(struct net *net, struct rt6_info *rt) 960 { 961 if (!timer_pending(&net->ipv6.ip6_fib_timer) && 962 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE))) 963 mod_timer(&net->ipv6.ip6_fib_timer, 964 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 965 } 966 967 void fib6_force_start_gc(struct net *net) 968 { 969 if (!timer_pending(&net->ipv6.ip6_fib_timer)) 970 mod_timer(&net->ipv6.ip6_fib_timer, 971 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 972 } 973 974 /* 975 * Add routing information to the routing tree. 976 * <destination addr>/<source addr> 977 * with source addr info in sub-trees 978 */ 979 980 int fib6_add(struct fib6_node *root, struct rt6_info *rt, 981 struct nl_info *info, struct mx6_config *mxc, 982 struct netlink_ext_ack *extack) 983 { 984 struct fib6_node *fn, *pn = NULL; 985 int err = -ENOMEM; 986 int allow_create = 1; 987 int replace_required = 0; 988 int sernum = fib6_new_sernum(info->nl_net); 989 990 if (WARN_ON_ONCE(!atomic_read(&rt->dst.__refcnt))) 991 return -EINVAL; 992 993 if (info->nlh) { 994 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) 995 allow_create = 0; 996 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) 997 replace_required = 1; 998 } 999 if (!allow_create && !replace_required) 1000 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); 1001 1002 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen, 1003 offsetof(struct rt6_info, rt6i_dst), allow_create, 1004 replace_required, sernum, extack); 1005 if (IS_ERR(fn)) { 1006 err = PTR_ERR(fn); 1007 fn = NULL; 1008 goto out; 1009 } 1010 1011 pn = fn; 1012 1013 #ifdef CONFIG_IPV6_SUBTREES 1014 if (rt->rt6i_src.plen) { 1015 struct fib6_node *sn; 1016 1017 if (!fn->subtree) { 1018 struct fib6_node *sfn; 1019 1020 /* 1021 * Create subtree. 1022 * 1023 * fn[main tree] 1024 * | 1025 * sfn[subtree root] 1026 * \ 1027 * sn[new leaf node] 1028 */ 1029 1030 /* Create subtree root node */ 1031 sfn = node_alloc(); 1032 if (!sfn) 1033 goto failure; 1034 1035 sfn->leaf = info->nl_net->ipv6.ip6_null_entry; 1036 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref); 1037 sfn->fn_flags = RTN_ROOT; 1038 sfn->fn_sernum = sernum; 1039 1040 /* Now add the first leaf node to new subtree */ 1041 1042 sn = fib6_add_1(sfn, &rt->rt6i_src.addr, 1043 rt->rt6i_src.plen, 1044 offsetof(struct rt6_info, rt6i_src), 1045 allow_create, replace_required, sernum, 1046 extack); 1047 1048 if (IS_ERR(sn)) { 1049 /* If it is failed, discard just allocated 1050 root, and then (in failure) stale node 1051 in main tree. 1052 */ 1053 node_free_immediate(sfn); 1054 err = PTR_ERR(sn); 1055 goto failure; 1056 } 1057 1058 /* Now link new subtree to main tree */ 1059 sfn->parent = fn; 1060 fn->subtree = sfn; 1061 } else { 1062 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, 1063 rt->rt6i_src.plen, 1064 offsetof(struct rt6_info, rt6i_src), 1065 allow_create, replace_required, sernum, 1066 extack); 1067 1068 if (IS_ERR(sn)) { 1069 err = PTR_ERR(sn); 1070 goto failure; 1071 } 1072 } 1073 1074 if (!fn->leaf) { 1075 fn->leaf = rt; 1076 atomic_inc(&rt->rt6i_ref); 1077 } 1078 fn = sn; 1079 } 1080 #endif 1081 1082 err = fib6_add_rt2node(fn, rt, info, mxc); 1083 if (!err) { 1084 fib6_start_gc(info->nl_net, rt); 1085 if (!(rt->rt6i_flags & RTF_CACHE)) 1086 fib6_prune_clones(info->nl_net, pn); 1087 } 1088 1089 out: 1090 if (err) { 1091 #ifdef CONFIG_IPV6_SUBTREES 1092 /* 1093 * If fib6_add_1 has cleared the old leaf pointer in the 1094 * super-tree leaf node we have to find a new one for it. 1095 */ 1096 if (pn != fn && pn->leaf == rt) { 1097 pn->leaf = NULL; 1098 atomic_dec(&rt->rt6i_ref); 1099 } 1100 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) { 1101 pn->leaf = fib6_find_prefix(info->nl_net, pn); 1102 #if RT6_DEBUG >= 2 1103 if (!pn->leaf) { 1104 WARN_ON(pn->leaf == NULL); 1105 pn->leaf = info->nl_net->ipv6.ip6_null_entry; 1106 } 1107 #endif 1108 atomic_inc(&pn->leaf->rt6i_ref); 1109 } 1110 #endif 1111 goto failure; 1112 } 1113 return err; 1114 1115 failure: 1116 /* fn->leaf could be NULL if fn is an intermediate node and we 1117 * failed to add the new route to it in both subtree creation 1118 * failure and fib6_add_rt2node() failure case. 1119 * In both cases, fib6_repair_tree() should be called to fix 1120 * fn->leaf. 1121 */ 1122 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) 1123 fib6_repair_tree(info->nl_net, fn); 1124 /* Always release dst as dst->__refcnt is guaranteed 1125 * to be taken before entering this function 1126 */ 1127 dst_release_immediate(&rt->dst); 1128 return err; 1129 } 1130 1131 /* 1132 * Routing tree lookup 1133 * 1134 */ 1135 1136 struct lookup_args { 1137 int offset; /* key offset on rt6_info */ 1138 const struct in6_addr *addr; /* search key */ 1139 }; 1140 1141 static struct fib6_node *fib6_lookup_1(struct fib6_node *root, 1142 struct lookup_args *args) 1143 { 1144 struct fib6_node *fn; 1145 __be32 dir; 1146 1147 if (unlikely(args->offset == 0)) 1148 return NULL; 1149 1150 /* 1151 * Descend on a tree 1152 */ 1153 1154 fn = root; 1155 1156 for (;;) { 1157 struct fib6_node *next; 1158 1159 dir = addr_bit_set(args->addr, fn->fn_bit); 1160 1161 next = dir ? fn->right : fn->left; 1162 1163 if (next) { 1164 fn = next; 1165 continue; 1166 } 1167 break; 1168 } 1169 1170 while (fn) { 1171 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) { 1172 struct rt6key *key; 1173 1174 key = (struct rt6key *) ((u8 *) fn->leaf + 1175 args->offset); 1176 1177 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 1178 #ifdef CONFIG_IPV6_SUBTREES 1179 if (fn->subtree) { 1180 struct fib6_node *sfn; 1181 sfn = fib6_lookup_1(fn->subtree, 1182 args + 1); 1183 if (!sfn) 1184 goto backtrack; 1185 fn = sfn; 1186 } 1187 #endif 1188 if (fn->fn_flags & RTN_RTINFO) 1189 return fn; 1190 } 1191 } 1192 #ifdef CONFIG_IPV6_SUBTREES 1193 backtrack: 1194 #endif 1195 if (fn->fn_flags & RTN_ROOT) 1196 break; 1197 1198 fn = fn->parent; 1199 } 1200 1201 return NULL; 1202 } 1203 1204 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr, 1205 const struct in6_addr *saddr) 1206 { 1207 struct fib6_node *fn; 1208 struct lookup_args args[] = { 1209 { 1210 .offset = offsetof(struct rt6_info, rt6i_dst), 1211 .addr = daddr, 1212 }, 1213 #ifdef CONFIG_IPV6_SUBTREES 1214 { 1215 .offset = offsetof(struct rt6_info, rt6i_src), 1216 .addr = saddr, 1217 }, 1218 #endif 1219 { 1220 .offset = 0, /* sentinel */ 1221 } 1222 }; 1223 1224 fn = fib6_lookup_1(root, daddr ? args : args + 1); 1225 if (!fn || fn->fn_flags & RTN_TL_ROOT) 1226 fn = root; 1227 1228 return fn; 1229 } 1230 1231 /* 1232 * Get node with specified destination prefix (and source prefix, 1233 * if subtrees are used) 1234 */ 1235 1236 1237 static struct fib6_node *fib6_locate_1(struct fib6_node *root, 1238 const struct in6_addr *addr, 1239 int plen, int offset) 1240 { 1241 struct fib6_node *fn; 1242 1243 for (fn = root; fn ; ) { 1244 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); 1245 1246 /* 1247 * Prefix match 1248 */ 1249 if (plen < fn->fn_bit || 1250 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 1251 return NULL; 1252 1253 if (plen == fn->fn_bit) 1254 return fn; 1255 1256 /* 1257 * We have more bits to go 1258 */ 1259 if (addr_bit_set(addr, fn->fn_bit)) 1260 fn = fn->right; 1261 else 1262 fn = fn->left; 1263 } 1264 return NULL; 1265 } 1266 1267 struct fib6_node *fib6_locate(struct fib6_node *root, 1268 const struct in6_addr *daddr, int dst_len, 1269 const struct in6_addr *saddr, int src_len) 1270 { 1271 struct fib6_node *fn; 1272 1273 fn = fib6_locate_1(root, daddr, dst_len, 1274 offsetof(struct rt6_info, rt6i_dst)); 1275 1276 #ifdef CONFIG_IPV6_SUBTREES 1277 if (src_len) { 1278 WARN_ON(saddr == NULL); 1279 if (fn && fn->subtree) 1280 fn = fib6_locate_1(fn->subtree, saddr, src_len, 1281 offsetof(struct rt6_info, rt6i_src)); 1282 } 1283 #endif 1284 1285 if (fn && fn->fn_flags & RTN_RTINFO) 1286 return fn; 1287 1288 return NULL; 1289 } 1290 1291 1292 /* 1293 * Deletion 1294 * 1295 */ 1296 1297 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn) 1298 { 1299 if (fn->fn_flags & RTN_ROOT) 1300 return net->ipv6.ip6_null_entry; 1301 1302 while (fn) { 1303 if (fn->left) 1304 return fn->left->leaf; 1305 if (fn->right) 1306 return fn->right->leaf; 1307 1308 fn = FIB6_SUBTREE(fn); 1309 } 1310 return NULL; 1311 } 1312 1313 /* 1314 * Called to trim the tree of intermediate nodes when possible. "fn" 1315 * is the node we want to try and remove. 1316 */ 1317 1318 static struct fib6_node *fib6_repair_tree(struct net *net, 1319 struct fib6_node *fn) 1320 { 1321 int children; 1322 int nstate; 1323 struct fib6_node *child, *pn; 1324 struct fib6_walker *w; 1325 int iter = 0; 1326 1327 for (;;) { 1328 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1329 iter++; 1330 1331 WARN_ON(fn->fn_flags & RTN_RTINFO); 1332 WARN_ON(fn->fn_flags & RTN_TL_ROOT); 1333 WARN_ON(fn->leaf); 1334 1335 children = 0; 1336 child = NULL; 1337 if (fn->right) 1338 child = fn->right, children |= 1; 1339 if (fn->left) 1340 child = fn->left, children |= 2; 1341 1342 if (children == 3 || FIB6_SUBTREE(fn) 1343 #ifdef CONFIG_IPV6_SUBTREES 1344 /* Subtree root (i.e. fn) may have one child */ 1345 || (children && fn->fn_flags & RTN_ROOT) 1346 #endif 1347 ) { 1348 fn->leaf = fib6_find_prefix(net, fn); 1349 #if RT6_DEBUG >= 2 1350 if (!fn->leaf) { 1351 WARN_ON(!fn->leaf); 1352 fn->leaf = net->ipv6.ip6_null_entry; 1353 } 1354 #endif 1355 atomic_inc(&fn->leaf->rt6i_ref); 1356 return fn->parent; 1357 } 1358 1359 pn = fn->parent; 1360 #ifdef CONFIG_IPV6_SUBTREES 1361 if (FIB6_SUBTREE(pn) == fn) { 1362 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1363 FIB6_SUBTREE(pn) = NULL; 1364 nstate = FWS_L; 1365 } else { 1366 WARN_ON(fn->fn_flags & RTN_ROOT); 1367 #endif 1368 if (pn->right == fn) 1369 pn->right = child; 1370 else if (pn->left == fn) 1371 pn->left = child; 1372 #if RT6_DEBUG >= 2 1373 else 1374 WARN_ON(1); 1375 #endif 1376 if (child) 1377 child->parent = pn; 1378 nstate = FWS_R; 1379 #ifdef CONFIG_IPV6_SUBTREES 1380 } 1381 #endif 1382 1383 read_lock(&net->ipv6.fib6_walker_lock); 1384 FOR_WALKERS(net, w) { 1385 if (!child) { 1386 if (w->root == fn) { 1387 w->root = w->node = NULL; 1388 RT6_TRACE("W %p adjusted by delroot 1\n", w); 1389 } else if (w->node == fn) { 1390 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1391 w->node = pn; 1392 w->state = nstate; 1393 } 1394 } else { 1395 if (w->root == fn) { 1396 w->root = child; 1397 RT6_TRACE("W %p adjusted by delroot 2\n", w); 1398 } 1399 if (w->node == fn) { 1400 w->node = child; 1401 if (children&2) { 1402 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1403 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT; 1404 } else { 1405 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1406 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT; 1407 } 1408 } 1409 } 1410 } 1411 read_unlock(&net->ipv6.fib6_walker_lock); 1412 1413 node_free(fn); 1414 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) 1415 return pn; 1416 1417 rt6_release(pn->leaf); 1418 pn->leaf = NULL; 1419 fn = pn; 1420 } 1421 } 1422 1423 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, 1424 struct nl_info *info) 1425 { 1426 struct fib6_walker *w; 1427 struct rt6_info *rt = *rtp; 1428 struct net *net = info->nl_net; 1429 1430 RT6_TRACE("fib6_del_route\n"); 1431 1432 /* Unlink it */ 1433 *rtp = rt->dst.rt6_next; 1434 rt->rt6i_node = NULL; 1435 net->ipv6.rt6_stats->fib_rt_entries--; 1436 net->ipv6.rt6_stats->fib_discarded_routes++; 1437 1438 /* Reset round-robin state, if necessary */ 1439 if (fn->rr_ptr == rt) 1440 fn->rr_ptr = NULL; 1441 1442 /* Remove this entry from other siblings */ 1443 if (rt->rt6i_nsiblings) { 1444 struct rt6_info *sibling, *next_sibling; 1445 1446 list_for_each_entry_safe(sibling, next_sibling, 1447 &rt->rt6i_siblings, rt6i_siblings) 1448 sibling->rt6i_nsiblings--; 1449 rt->rt6i_nsiblings = 0; 1450 list_del_init(&rt->rt6i_siblings); 1451 } 1452 1453 /* Adjust walkers */ 1454 read_lock(&net->ipv6.fib6_walker_lock); 1455 FOR_WALKERS(net, w) { 1456 if (w->state == FWS_C && w->leaf == rt) { 1457 RT6_TRACE("walker %p adjusted by delroute\n", w); 1458 w->leaf = rt->dst.rt6_next; 1459 if (!w->leaf) 1460 w->state = FWS_U; 1461 } 1462 } 1463 read_unlock(&net->ipv6.fib6_walker_lock); 1464 1465 rt->dst.rt6_next = NULL; 1466 1467 /* If it was last route, expunge its radix tree node */ 1468 if (!fn->leaf) { 1469 fn->fn_flags &= ~RTN_RTINFO; 1470 net->ipv6.rt6_stats->fib_route_nodes--; 1471 fn = fib6_repair_tree(net, fn); 1472 } 1473 1474 fib6_purge_rt(rt, fn, net); 1475 1476 if (!info->skip_notify) 1477 inet6_rt_notify(RTM_DELROUTE, rt, info, 0); 1478 rt6_release(rt); 1479 } 1480 1481 int fib6_del(struct rt6_info *rt, struct nl_info *info) 1482 { 1483 struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node, 1484 lockdep_is_held(&rt->rt6i_table->tb6_lock)); 1485 struct net *net = info->nl_net; 1486 struct rt6_info **rtp; 1487 1488 #if RT6_DEBUG >= 2 1489 if (rt->dst.obsolete > 0) { 1490 WARN_ON(fn); 1491 return -ENOENT; 1492 } 1493 #endif 1494 if (!fn || rt == net->ipv6.ip6_null_entry) 1495 return -ENOENT; 1496 1497 WARN_ON(!(fn->fn_flags & RTN_RTINFO)); 1498 1499 if (!(rt->rt6i_flags & RTF_CACHE)) { 1500 struct fib6_node *pn = fn; 1501 #ifdef CONFIG_IPV6_SUBTREES 1502 /* clones of this route might be in another subtree */ 1503 if (rt->rt6i_src.plen) { 1504 while (!(pn->fn_flags & RTN_ROOT)) 1505 pn = pn->parent; 1506 pn = pn->parent; 1507 } 1508 #endif 1509 fib6_prune_clones(info->nl_net, pn); 1510 } 1511 1512 /* 1513 * Walk the leaf entries looking for ourself 1514 */ 1515 1516 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) { 1517 if (*rtp == rt) { 1518 fib6_del_route(fn, rtp, info); 1519 return 0; 1520 } 1521 } 1522 return -ENOENT; 1523 } 1524 1525 /* 1526 * Tree traversal function. 1527 * 1528 * Certainly, it is not interrupt safe. 1529 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 1530 * It means, that we can modify tree during walking 1531 * and use this function for garbage collection, clone pruning, 1532 * cleaning tree when a device goes down etc. etc. 1533 * 1534 * It guarantees that every node will be traversed, 1535 * and that it will be traversed only once. 1536 * 1537 * Callback function w->func may return: 1538 * 0 -> continue walking. 1539 * positive value -> walking is suspended (used by tree dumps, 1540 * and probably by gc, if it will be split to several slices) 1541 * negative value -> terminate walking. 1542 * 1543 * The function itself returns: 1544 * 0 -> walk is complete. 1545 * >0 -> walk is incomplete (i.e. suspended) 1546 * <0 -> walk is terminated by an error. 1547 */ 1548 1549 static int fib6_walk_continue(struct fib6_walker *w) 1550 { 1551 struct fib6_node *fn, *pn; 1552 1553 for (;;) { 1554 fn = w->node; 1555 if (!fn) 1556 return 0; 1557 1558 if (w->prune && fn != w->root && 1559 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) { 1560 w->state = FWS_C; 1561 w->leaf = fn->leaf; 1562 } 1563 switch (w->state) { 1564 #ifdef CONFIG_IPV6_SUBTREES 1565 case FWS_S: 1566 if (FIB6_SUBTREE(fn)) { 1567 w->node = FIB6_SUBTREE(fn); 1568 continue; 1569 } 1570 w->state = FWS_L; 1571 #endif 1572 case FWS_L: 1573 if (fn->left) { 1574 w->node = fn->left; 1575 w->state = FWS_INIT; 1576 continue; 1577 } 1578 w->state = FWS_R; 1579 case FWS_R: 1580 if (fn->right) { 1581 w->node = fn->right; 1582 w->state = FWS_INIT; 1583 continue; 1584 } 1585 w->state = FWS_C; 1586 w->leaf = fn->leaf; 1587 case FWS_C: 1588 if (w->leaf && fn->fn_flags & RTN_RTINFO) { 1589 int err; 1590 1591 if (w->skip) { 1592 w->skip--; 1593 goto skip; 1594 } 1595 1596 err = w->func(w); 1597 if (err) 1598 return err; 1599 1600 w->count++; 1601 continue; 1602 } 1603 skip: 1604 w->state = FWS_U; 1605 case FWS_U: 1606 if (fn == w->root) 1607 return 0; 1608 pn = fn->parent; 1609 w->node = pn; 1610 #ifdef CONFIG_IPV6_SUBTREES 1611 if (FIB6_SUBTREE(pn) == fn) { 1612 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1613 w->state = FWS_L; 1614 continue; 1615 } 1616 #endif 1617 if (pn->left == fn) { 1618 w->state = FWS_R; 1619 continue; 1620 } 1621 if (pn->right == fn) { 1622 w->state = FWS_C; 1623 w->leaf = w->node->leaf; 1624 continue; 1625 } 1626 #if RT6_DEBUG >= 2 1627 WARN_ON(1); 1628 #endif 1629 } 1630 } 1631 } 1632 1633 static int fib6_walk(struct net *net, struct fib6_walker *w) 1634 { 1635 int res; 1636 1637 w->state = FWS_INIT; 1638 w->node = w->root; 1639 1640 fib6_walker_link(net, w); 1641 res = fib6_walk_continue(w); 1642 if (res <= 0) 1643 fib6_walker_unlink(net, w); 1644 return res; 1645 } 1646 1647 static int fib6_clean_node(struct fib6_walker *w) 1648 { 1649 int res; 1650 struct rt6_info *rt; 1651 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w); 1652 struct nl_info info = { 1653 .nl_net = c->net, 1654 }; 1655 1656 if (c->sernum != FIB6_NO_SERNUM_CHANGE && 1657 w->node->fn_sernum != c->sernum) 1658 w->node->fn_sernum = c->sernum; 1659 1660 if (!c->func) { 1661 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE); 1662 w->leaf = NULL; 1663 return 0; 1664 } 1665 1666 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { 1667 res = c->func(rt, c->arg); 1668 if (res < 0) { 1669 w->leaf = rt; 1670 res = fib6_del(rt, &info); 1671 if (res) { 1672 #if RT6_DEBUG >= 2 1673 pr_debug("%s: del failed: rt=%p@%p err=%d\n", 1674 __func__, rt, 1675 rcu_access_pointer(rt->rt6i_node), 1676 res); 1677 #endif 1678 continue; 1679 } 1680 return 0; 1681 } 1682 WARN_ON(res != 0); 1683 } 1684 w->leaf = rt; 1685 return 0; 1686 } 1687 1688 /* 1689 * Convenient frontend to tree walker. 1690 * 1691 * func is called on each route. 1692 * It may return -1 -> delete this route. 1693 * 0 -> continue walking 1694 * 1695 * prune==1 -> only immediate children of node (certainly, 1696 * ignoring pure split nodes) will be scanned. 1697 */ 1698 1699 static void fib6_clean_tree(struct net *net, struct fib6_node *root, 1700 int (*func)(struct rt6_info *, void *arg), 1701 bool prune, int sernum, void *arg) 1702 { 1703 struct fib6_cleaner c; 1704 1705 c.w.root = root; 1706 c.w.func = fib6_clean_node; 1707 c.w.prune = prune; 1708 c.w.count = 0; 1709 c.w.skip = 0; 1710 c.func = func; 1711 c.sernum = sernum; 1712 c.arg = arg; 1713 c.net = net; 1714 1715 fib6_walk(net, &c.w); 1716 } 1717 1718 static void __fib6_clean_all(struct net *net, 1719 int (*func)(struct rt6_info *, void *), 1720 int sernum, void *arg) 1721 { 1722 struct fib6_table *table; 1723 struct hlist_head *head; 1724 unsigned int h; 1725 1726 rcu_read_lock(); 1727 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 1728 head = &net->ipv6.fib_table_hash[h]; 1729 hlist_for_each_entry_rcu(table, head, tb6_hlist) { 1730 write_lock_bh(&table->tb6_lock); 1731 fib6_clean_tree(net, &table->tb6_root, 1732 func, false, sernum, arg); 1733 write_unlock_bh(&table->tb6_lock); 1734 } 1735 } 1736 rcu_read_unlock(); 1737 } 1738 1739 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *), 1740 void *arg) 1741 { 1742 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg); 1743 } 1744 1745 static int fib6_prune_clone(struct rt6_info *rt, void *arg) 1746 { 1747 if (rt->rt6i_flags & RTF_CACHE) { 1748 RT6_TRACE("pruning clone %p\n", rt); 1749 return -1; 1750 } 1751 1752 return 0; 1753 } 1754 1755 static void fib6_prune_clones(struct net *net, struct fib6_node *fn) 1756 { 1757 fib6_clean_tree(net, fn, fib6_prune_clone, true, 1758 FIB6_NO_SERNUM_CHANGE, NULL); 1759 } 1760 1761 static void fib6_flush_trees(struct net *net) 1762 { 1763 int new_sernum = fib6_new_sernum(net); 1764 1765 __fib6_clean_all(net, NULL, new_sernum, NULL); 1766 } 1767 1768 /* 1769 * Garbage collection 1770 */ 1771 1772 struct fib6_gc_args 1773 { 1774 int timeout; 1775 int more; 1776 }; 1777 1778 static int fib6_age(struct rt6_info *rt, void *arg) 1779 { 1780 struct fib6_gc_args *gc_args = arg; 1781 unsigned long now = jiffies; 1782 1783 /* 1784 * check addrconf expiration here. 1785 * Routes are expired even if they are in use. 1786 * 1787 * Also age clones. Note, that clones are aged out 1788 * only if they are not in use now. 1789 */ 1790 1791 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) { 1792 if (time_after(now, rt->dst.expires)) { 1793 RT6_TRACE("expiring %p\n", rt); 1794 return -1; 1795 } 1796 gc_args->more++; 1797 } else if (rt->rt6i_flags & RTF_CACHE) { 1798 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) 1799 rt->dst.obsolete = DST_OBSOLETE_KILL; 1800 if (atomic_read(&rt->dst.__refcnt) == 1 && 1801 rt->dst.obsolete == DST_OBSOLETE_KILL) { 1802 RT6_TRACE("aging clone %p\n", rt); 1803 return -1; 1804 } else if (rt->rt6i_flags & RTF_GATEWAY) { 1805 struct neighbour *neigh; 1806 __u8 neigh_flags = 0; 1807 1808 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway); 1809 if (neigh) { 1810 neigh_flags = neigh->flags; 1811 neigh_release(neigh); 1812 } 1813 if (!(neigh_flags & NTF_ROUTER)) { 1814 RT6_TRACE("purging route %p via non-router but gateway\n", 1815 rt); 1816 return -1; 1817 } 1818 } 1819 gc_args->more++; 1820 } 1821 1822 return 0; 1823 } 1824 1825 void fib6_run_gc(unsigned long expires, struct net *net, bool force) 1826 { 1827 struct fib6_gc_args gc_args; 1828 unsigned long now; 1829 1830 if (force) { 1831 spin_lock_bh(&net->ipv6.fib6_gc_lock); 1832 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) { 1833 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); 1834 return; 1835 } 1836 gc_args.timeout = expires ? (int)expires : 1837 net->ipv6.sysctl.ip6_rt_gc_interval; 1838 gc_args.more = 0; 1839 1840 fib6_clean_all(net, fib6_age, &gc_args); 1841 now = jiffies; 1842 net->ipv6.ip6_rt_last_gc = now; 1843 1844 if (gc_args.more) 1845 mod_timer(&net->ipv6.ip6_fib_timer, 1846 round_jiffies(now 1847 + net->ipv6.sysctl.ip6_rt_gc_interval)); 1848 else 1849 del_timer(&net->ipv6.ip6_fib_timer); 1850 spin_unlock_bh(&net->ipv6.fib6_gc_lock); 1851 } 1852 1853 static void fib6_gc_timer_cb(unsigned long arg) 1854 { 1855 fib6_run_gc(0, (struct net *)arg, true); 1856 } 1857 1858 static int __net_init fib6_net_init(struct net *net) 1859 { 1860 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; 1861 1862 spin_lock_init(&net->ipv6.fib6_gc_lock); 1863 rwlock_init(&net->ipv6.fib6_walker_lock); 1864 INIT_LIST_HEAD(&net->ipv6.fib6_walkers); 1865 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net); 1866 1867 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); 1868 if (!net->ipv6.rt6_stats) 1869 goto out_timer; 1870 1871 /* Avoid false sharing : Use at least a full cache line */ 1872 size = max_t(size_t, size, L1_CACHE_BYTES); 1873 1874 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); 1875 if (!net->ipv6.fib_table_hash) 1876 goto out_rt6_stats; 1877 1878 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), 1879 GFP_KERNEL); 1880 if (!net->ipv6.fib6_main_tbl) 1881 goto out_fib_table_hash; 1882 1883 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; 1884 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1885 net->ipv6.fib6_main_tbl->tb6_root.fn_flags = 1886 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1887 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); 1888 1889 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1890 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), 1891 GFP_KERNEL); 1892 if (!net->ipv6.fib6_local_tbl) 1893 goto out_fib6_main_tbl; 1894 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; 1895 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; 1896 net->ipv6.fib6_local_tbl->tb6_root.fn_flags = 1897 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 1898 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); 1899 #endif 1900 fib6_tables_init(net); 1901 1902 return 0; 1903 1904 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1905 out_fib6_main_tbl: 1906 kfree(net->ipv6.fib6_main_tbl); 1907 #endif 1908 out_fib_table_hash: 1909 kfree(net->ipv6.fib_table_hash); 1910 out_rt6_stats: 1911 kfree(net->ipv6.rt6_stats); 1912 out_timer: 1913 return -ENOMEM; 1914 } 1915 1916 static void fib6_net_exit(struct net *net) 1917 { 1918 rt6_ifdown(net, NULL); 1919 del_timer_sync(&net->ipv6.ip6_fib_timer); 1920 1921 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 1922 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers); 1923 kfree(net->ipv6.fib6_local_tbl); 1924 #endif 1925 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers); 1926 kfree(net->ipv6.fib6_main_tbl); 1927 kfree(net->ipv6.fib_table_hash); 1928 kfree(net->ipv6.rt6_stats); 1929 } 1930 1931 static struct pernet_operations fib6_net_ops = { 1932 .init = fib6_net_init, 1933 .exit = fib6_net_exit, 1934 }; 1935 1936 int __init fib6_init(void) 1937 { 1938 int ret = -ENOMEM; 1939 1940 fib6_node_kmem = kmem_cache_create("fib6_nodes", 1941 sizeof(struct fib6_node), 1942 0, SLAB_HWCACHE_ALIGN, 1943 NULL); 1944 if (!fib6_node_kmem) 1945 goto out; 1946 1947 ret = register_pernet_subsys(&fib6_net_ops); 1948 if (ret) 1949 goto out_kmem_cache_create; 1950 1951 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib, 1952 NULL); 1953 if (ret) 1954 goto out_unregister_subsys; 1955 1956 __fib6_flush_trees = fib6_flush_trees; 1957 out: 1958 return ret; 1959 1960 out_unregister_subsys: 1961 unregister_pernet_subsys(&fib6_net_ops); 1962 out_kmem_cache_create: 1963 kmem_cache_destroy(fib6_node_kmem); 1964 goto out; 1965 } 1966 1967 void fib6_gc_cleanup(void) 1968 { 1969 unregister_pernet_subsys(&fib6_net_ops); 1970 kmem_cache_destroy(fib6_node_kmem); 1971 } 1972 1973 #ifdef CONFIG_PROC_FS 1974 1975 struct ipv6_route_iter { 1976 struct seq_net_private p; 1977 struct fib6_walker w; 1978 loff_t skip; 1979 struct fib6_table *tbl; 1980 int sernum; 1981 }; 1982 1983 static int ipv6_route_seq_show(struct seq_file *seq, void *v) 1984 { 1985 struct rt6_info *rt = v; 1986 struct ipv6_route_iter *iter = seq->private; 1987 1988 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen); 1989 1990 #ifdef CONFIG_IPV6_SUBTREES 1991 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen); 1992 #else 1993 seq_puts(seq, "00000000000000000000000000000000 00 "); 1994 #endif 1995 if (rt->rt6i_flags & RTF_GATEWAY) 1996 seq_printf(seq, "%pi6", &rt->rt6i_gateway); 1997 else 1998 seq_puts(seq, "00000000000000000000000000000000"); 1999 2000 seq_printf(seq, " %08x %08x %08x %08x %8s\n", 2001 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt), 2002 rt->dst.__use, rt->rt6i_flags, 2003 rt->dst.dev ? rt->dst.dev->name : ""); 2004 iter->w.leaf = NULL; 2005 return 0; 2006 } 2007 2008 static int ipv6_route_yield(struct fib6_walker *w) 2009 { 2010 struct ipv6_route_iter *iter = w->args; 2011 2012 if (!iter->skip) 2013 return 1; 2014 2015 do { 2016 iter->w.leaf = iter->w.leaf->dst.rt6_next; 2017 iter->skip--; 2018 if (!iter->skip && iter->w.leaf) 2019 return 1; 2020 } while (iter->w.leaf); 2021 2022 return 0; 2023 } 2024 2025 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter, 2026 struct net *net) 2027 { 2028 memset(&iter->w, 0, sizeof(iter->w)); 2029 iter->w.func = ipv6_route_yield; 2030 iter->w.root = &iter->tbl->tb6_root; 2031 iter->w.state = FWS_INIT; 2032 iter->w.node = iter->w.root; 2033 iter->w.args = iter; 2034 iter->sernum = iter->w.root->fn_sernum; 2035 INIT_LIST_HEAD(&iter->w.lh); 2036 fib6_walker_link(net, &iter->w); 2037 } 2038 2039 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl, 2040 struct net *net) 2041 { 2042 unsigned int h; 2043 struct hlist_node *node; 2044 2045 if (tbl) { 2046 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1; 2047 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist)); 2048 } else { 2049 h = 0; 2050 node = NULL; 2051 } 2052 2053 while (!node && h < FIB6_TABLE_HASHSZ) { 2054 node = rcu_dereference_bh( 2055 hlist_first_rcu(&net->ipv6.fib_table_hash[h++])); 2056 } 2057 return hlist_entry_safe(node, struct fib6_table, tb6_hlist); 2058 } 2059 2060 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter) 2061 { 2062 if (iter->sernum != iter->w.root->fn_sernum) { 2063 iter->sernum = iter->w.root->fn_sernum; 2064 iter->w.state = FWS_INIT; 2065 iter->w.node = iter->w.root; 2066 WARN_ON(iter->w.skip); 2067 iter->w.skip = iter->w.count; 2068 } 2069 } 2070 2071 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2072 { 2073 int r; 2074 struct rt6_info *n; 2075 struct net *net = seq_file_net(seq); 2076 struct ipv6_route_iter *iter = seq->private; 2077 2078 if (!v) 2079 goto iter_table; 2080 2081 n = ((struct rt6_info *)v)->dst.rt6_next; 2082 if (n) { 2083 ++*pos; 2084 return n; 2085 } 2086 2087 iter_table: 2088 ipv6_route_check_sernum(iter); 2089 read_lock(&iter->tbl->tb6_lock); 2090 r = fib6_walk_continue(&iter->w); 2091 read_unlock(&iter->tbl->tb6_lock); 2092 if (r > 0) { 2093 if (v) 2094 ++*pos; 2095 return iter->w.leaf; 2096 } else if (r < 0) { 2097 fib6_walker_unlink(net, &iter->w); 2098 return NULL; 2099 } 2100 fib6_walker_unlink(net, &iter->w); 2101 2102 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net); 2103 if (!iter->tbl) 2104 return NULL; 2105 2106 ipv6_route_seq_setup_walk(iter, net); 2107 goto iter_table; 2108 } 2109 2110 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos) 2111 __acquires(RCU_BH) 2112 { 2113 struct net *net = seq_file_net(seq); 2114 struct ipv6_route_iter *iter = seq->private; 2115 2116 rcu_read_lock_bh(); 2117 iter->tbl = ipv6_route_seq_next_table(NULL, net); 2118 iter->skip = *pos; 2119 2120 if (iter->tbl) { 2121 ipv6_route_seq_setup_walk(iter, net); 2122 return ipv6_route_seq_next(seq, NULL, pos); 2123 } else { 2124 return NULL; 2125 } 2126 } 2127 2128 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter) 2129 { 2130 struct fib6_walker *w = &iter->w; 2131 return w->node && !(w->state == FWS_U && w->node == w->root); 2132 } 2133 2134 static void ipv6_route_seq_stop(struct seq_file *seq, void *v) 2135 __releases(RCU_BH) 2136 { 2137 struct net *net = seq_file_net(seq); 2138 struct ipv6_route_iter *iter = seq->private; 2139 2140 if (ipv6_route_iter_active(iter)) 2141 fib6_walker_unlink(net, &iter->w); 2142 2143 rcu_read_unlock_bh(); 2144 } 2145 2146 static const struct seq_operations ipv6_route_seq_ops = { 2147 .start = ipv6_route_seq_start, 2148 .next = ipv6_route_seq_next, 2149 .stop = ipv6_route_seq_stop, 2150 .show = ipv6_route_seq_show 2151 }; 2152 2153 int ipv6_route_open(struct inode *inode, struct file *file) 2154 { 2155 return seq_open_net(inode, file, &ipv6_route_seq_ops, 2156 sizeof(struct ipv6_route_iter)); 2157 } 2158 2159 #endif /* CONFIG_PROC_FS */ 2160