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 #include <net/fib_notifier.h> 37 38 #include <net/ip6_fib.h> 39 #include <net/ip6_route.h> 40 41 static struct kmem_cache *fib6_node_kmem __read_mostly; 42 43 struct fib6_cleaner { 44 struct fib6_walker w; 45 struct net *net; 46 int (*func)(struct fib6_info *, void *arg); 47 int sernum; 48 void *arg; 49 }; 50 51 #ifdef CONFIG_IPV6_SUBTREES 52 #define FWS_INIT FWS_S 53 #else 54 #define FWS_INIT FWS_L 55 #endif 56 57 static struct fib6_info *fib6_find_prefix(struct net *net, 58 struct fib6_table *table, 59 struct fib6_node *fn); 60 static struct fib6_node *fib6_repair_tree(struct net *net, 61 struct fib6_table *table, 62 struct fib6_node *fn); 63 static int fib6_walk(struct net *net, struct fib6_walker *w); 64 static int fib6_walk_continue(struct fib6_walker *w); 65 66 /* 67 * A routing update causes an increase of the serial number on the 68 * affected subtree. This allows for cached routes to be asynchronously 69 * tested when modifications are made to the destination cache as a 70 * result of redirects, path MTU changes, etc. 71 */ 72 73 static void fib6_gc_timer_cb(struct timer_list *t); 74 75 #define FOR_WALKERS(net, w) \ 76 list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh) 77 78 static void fib6_walker_link(struct net *net, struct fib6_walker *w) 79 { 80 write_lock_bh(&net->ipv6.fib6_walker_lock); 81 list_add(&w->lh, &net->ipv6.fib6_walkers); 82 write_unlock_bh(&net->ipv6.fib6_walker_lock); 83 } 84 85 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w) 86 { 87 write_lock_bh(&net->ipv6.fib6_walker_lock); 88 list_del(&w->lh); 89 write_unlock_bh(&net->ipv6.fib6_walker_lock); 90 } 91 92 static int fib6_new_sernum(struct net *net) 93 { 94 int new, old; 95 96 do { 97 old = atomic_read(&net->ipv6.fib6_sernum); 98 new = old < INT_MAX ? old + 1 : 1; 99 } while (atomic_cmpxchg(&net->ipv6.fib6_sernum, 100 old, new) != old); 101 return new; 102 } 103 104 enum { 105 FIB6_NO_SERNUM_CHANGE = 0, 106 }; 107 108 void fib6_update_sernum(struct net *net, struct fib6_info *f6i) 109 { 110 struct fib6_node *fn; 111 112 fn = rcu_dereference_protected(f6i->fib6_node, 113 lockdep_is_held(&f6i->fib6_table->tb6_lock)); 114 if (fn) 115 fn->fn_sernum = fib6_new_sernum(net); 116 } 117 118 /* 119 * Auxiliary address test functions for the radix tree. 120 * 121 * These assume a 32bit processor (although it will work on 122 * 64bit processors) 123 */ 124 125 /* 126 * test bit 127 */ 128 #if defined(__LITTLE_ENDIAN) 129 # define BITOP_BE32_SWIZZLE (0x1F & ~7) 130 #else 131 # define BITOP_BE32_SWIZZLE 0 132 #endif 133 134 static __be32 addr_bit_set(const void *token, int fn_bit) 135 { 136 const __be32 *addr = token; 137 /* 138 * Here, 139 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f) 140 * is optimized version of 141 * htonl(1 << ((~fn_bit)&0x1F)) 142 * See include/asm-generic/bitops/le.h. 143 */ 144 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) & 145 addr[fn_bit >> 5]; 146 } 147 148 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags) 149 { 150 struct fib6_info *f6i; 151 152 f6i = kzalloc(sizeof(*f6i), gfp_flags); 153 if (!f6i) 154 return NULL; 155 156 f6i->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags); 157 if (!f6i->rt6i_pcpu) { 158 kfree(f6i); 159 return NULL; 160 } 161 162 INIT_LIST_HEAD(&f6i->fib6_siblings); 163 f6i->fib6_metrics = (struct dst_metrics *)&dst_default_metrics; 164 165 atomic_inc(&f6i->fib6_ref); 166 167 return f6i; 168 } 169 170 void fib6_info_destroy_rcu(struct rcu_head *head) 171 { 172 struct fib6_info *f6i = container_of(head, struct fib6_info, rcu); 173 struct rt6_exception_bucket *bucket; 174 struct dst_metrics *m; 175 176 WARN_ON(f6i->fib6_node); 177 178 bucket = rcu_dereference_protected(f6i->rt6i_exception_bucket, 1); 179 if (bucket) { 180 f6i->rt6i_exception_bucket = NULL; 181 kfree(bucket); 182 } 183 184 if (f6i->rt6i_pcpu) { 185 int cpu; 186 187 for_each_possible_cpu(cpu) { 188 struct rt6_info **ppcpu_rt; 189 struct rt6_info *pcpu_rt; 190 191 ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu); 192 pcpu_rt = *ppcpu_rt; 193 if (pcpu_rt) { 194 dst_dev_put(&pcpu_rt->dst); 195 dst_release(&pcpu_rt->dst); 196 *ppcpu_rt = NULL; 197 } 198 } 199 } 200 201 lwtstate_put(f6i->fib6_nh.nh_lwtstate); 202 203 if (f6i->fib6_nh.nh_dev) 204 dev_put(f6i->fib6_nh.nh_dev); 205 206 m = f6i->fib6_metrics; 207 if (m != &dst_default_metrics && refcount_dec_and_test(&m->refcnt)) 208 kfree(m); 209 210 kfree(f6i); 211 } 212 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu); 213 214 static struct fib6_node *node_alloc(struct net *net) 215 { 216 struct fib6_node *fn; 217 218 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC); 219 if (fn) 220 net->ipv6.rt6_stats->fib_nodes++; 221 222 return fn; 223 } 224 225 static void node_free_immediate(struct net *net, struct fib6_node *fn) 226 { 227 kmem_cache_free(fib6_node_kmem, fn); 228 net->ipv6.rt6_stats->fib_nodes--; 229 } 230 231 static void node_free_rcu(struct rcu_head *head) 232 { 233 struct fib6_node *fn = container_of(head, struct fib6_node, rcu); 234 235 kmem_cache_free(fib6_node_kmem, fn); 236 } 237 238 static void node_free(struct net *net, struct fib6_node *fn) 239 { 240 call_rcu(&fn->rcu, node_free_rcu); 241 net->ipv6.rt6_stats->fib_nodes--; 242 } 243 244 static void fib6_free_table(struct fib6_table *table) 245 { 246 inetpeer_invalidate_tree(&table->tb6_peers); 247 kfree(table); 248 } 249 250 static void fib6_link_table(struct net *net, struct fib6_table *tb) 251 { 252 unsigned int h; 253 254 /* 255 * Initialize table lock at a single place to give lockdep a key, 256 * tables aren't visible prior to being linked to the list. 257 */ 258 spin_lock_init(&tb->tb6_lock); 259 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1); 260 261 /* 262 * No protection necessary, this is the only list mutatation 263 * operation, tables never disappear once they exist. 264 */ 265 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]); 266 } 267 268 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 269 270 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id) 271 { 272 struct fib6_table *table; 273 274 table = kzalloc(sizeof(*table), GFP_ATOMIC); 275 if (table) { 276 table->tb6_id = id; 277 rcu_assign_pointer(table->tb6_root.leaf, 278 net->ipv6.fib6_null_entry); 279 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 280 inet_peer_base_init(&table->tb6_peers); 281 } 282 283 return table; 284 } 285 286 struct fib6_table *fib6_new_table(struct net *net, u32 id) 287 { 288 struct fib6_table *tb; 289 290 if (id == 0) 291 id = RT6_TABLE_MAIN; 292 tb = fib6_get_table(net, id); 293 if (tb) 294 return tb; 295 296 tb = fib6_alloc_table(net, id); 297 if (tb) 298 fib6_link_table(net, tb); 299 300 return tb; 301 } 302 EXPORT_SYMBOL_GPL(fib6_new_table); 303 304 struct fib6_table *fib6_get_table(struct net *net, u32 id) 305 { 306 struct fib6_table *tb; 307 struct hlist_head *head; 308 unsigned int h; 309 310 if (id == 0) 311 id = RT6_TABLE_MAIN; 312 h = id & (FIB6_TABLE_HASHSZ - 1); 313 rcu_read_lock(); 314 head = &net->ipv6.fib_table_hash[h]; 315 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 316 if (tb->tb6_id == id) { 317 rcu_read_unlock(); 318 return tb; 319 } 320 } 321 rcu_read_unlock(); 322 323 return NULL; 324 } 325 EXPORT_SYMBOL_GPL(fib6_get_table); 326 327 static void __net_init fib6_tables_init(struct net *net) 328 { 329 fib6_link_table(net, net->ipv6.fib6_main_tbl); 330 fib6_link_table(net, net->ipv6.fib6_local_tbl); 331 } 332 #else 333 334 struct fib6_table *fib6_new_table(struct net *net, u32 id) 335 { 336 return fib6_get_table(net, id); 337 } 338 339 struct fib6_table *fib6_get_table(struct net *net, u32 id) 340 { 341 return net->ipv6.fib6_main_tbl; 342 } 343 344 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, 345 const struct sk_buff *skb, 346 int flags, pol_lookup_t lookup) 347 { 348 struct rt6_info *rt; 349 350 rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags); 351 if (rt->dst.error == -EAGAIN) { 352 ip6_rt_put(rt); 353 rt = net->ipv6.ip6_null_entry; 354 dst_hold(&rt->dst); 355 } 356 357 return &rt->dst; 358 } 359 360 /* called with rcu lock held; no reference taken on fib6_info */ 361 struct fib6_info *fib6_lookup(struct net *net, int oif, struct flowi6 *fl6, 362 int flags) 363 { 364 return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6, flags); 365 } 366 367 static void __net_init fib6_tables_init(struct net *net) 368 { 369 fib6_link_table(net, net->ipv6.fib6_main_tbl); 370 } 371 372 #endif 373 374 unsigned int fib6_tables_seq_read(struct net *net) 375 { 376 unsigned int h, fib_seq = 0; 377 378 rcu_read_lock(); 379 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 380 struct hlist_head *head = &net->ipv6.fib_table_hash[h]; 381 struct fib6_table *tb; 382 383 hlist_for_each_entry_rcu(tb, head, tb6_hlist) 384 fib_seq += tb->fib_seq; 385 } 386 rcu_read_unlock(); 387 388 return fib_seq; 389 } 390 391 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net, 392 enum fib_event_type event_type, 393 struct fib6_info *rt) 394 { 395 struct fib6_entry_notifier_info info = { 396 .rt = rt, 397 }; 398 399 return call_fib6_notifier(nb, net, event_type, &info.info); 400 } 401 402 static int call_fib6_entry_notifiers(struct net *net, 403 enum fib_event_type event_type, 404 struct fib6_info *rt, 405 struct netlink_ext_ack *extack) 406 { 407 struct fib6_entry_notifier_info info = { 408 .info.extack = extack, 409 .rt = rt, 410 }; 411 412 rt->fib6_table->fib_seq++; 413 return call_fib6_notifiers(net, event_type, &info.info); 414 } 415 416 struct fib6_dump_arg { 417 struct net *net; 418 struct notifier_block *nb; 419 }; 420 421 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg) 422 { 423 if (rt == arg->net->ipv6.fib6_null_entry) 424 return; 425 call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt); 426 } 427 428 static int fib6_node_dump(struct fib6_walker *w) 429 { 430 struct fib6_info *rt; 431 432 for_each_fib6_walker_rt(w) 433 fib6_rt_dump(rt, w->args); 434 w->leaf = NULL; 435 return 0; 436 } 437 438 static void fib6_table_dump(struct net *net, struct fib6_table *tb, 439 struct fib6_walker *w) 440 { 441 w->root = &tb->tb6_root; 442 spin_lock_bh(&tb->tb6_lock); 443 fib6_walk(net, w); 444 spin_unlock_bh(&tb->tb6_lock); 445 } 446 447 /* Called with rcu_read_lock() */ 448 int fib6_tables_dump(struct net *net, struct notifier_block *nb) 449 { 450 struct fib6_dump_arg arg; 451 struct fib6_walker *w; 452 unsigned int h; 453 454 w = kzalloc(sizeof(*w), GFP_ATOMIC); 455 if (!w) 456 return -ENOMEM; 457 458 w->func = fib6_node_dump; 459 arg.net = net; 460 arg.nb = nb; 461 w->args = &arg; 462 463 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 464 struct hlist_head *head = &net->ipv6.fib_table_hash[h]; 465 struct fib6_table *tb; 466 467 hlist_for_each_entry_rcu(tb, head, tb6_hlist) 468 fib6_table_dump(net, tb, w); 469 } 470 471 kfree(w); 472 473 return 0; 474 } 475 476 static int fib6_dump_node(struct fib6_walker *w) 477 { 478 int res; 479 struct fib6_info *rt; 480 481 for_each_fib6_walker_rt(w) { 482 res = rt6_dump_route(rt, w->args); 483 if (res < 0) { 484 /* Frame is full, suspend walking */ 485 w->leaf = rt; 486 return 1; 487 } 488 489 /* Multipath routes are dumped in one route with the 490 * RTA_MULTIPATH attribute. Jump 'rt' to point to the 491 * last sibling of this route (no need to dump the 492 * sibling routes again) 493 */ 494 if (rt->fib6_nsiblings) 495 rt = list_last_entry(&rt->fib6_siblings, 496 struct fib6_info, 497 fib6_siblings); 498 } 499 w->leaf = NULL; 500 return 0; 501 } 502 503 static void fib6_dump_end(struct netlink_callback *cb) 504 { 505 struct net *net = sock_net(cb->skb->sk); 506 struct fib6_walker *w = (void *)cb->args[2]; 507 508 if (w) { 509 if (cb->args[4]) { 510 cb->args[4] = 0; 511 fib6_walker_unlink(net, w); 512 } 513 cb->args[2] = 0; 514 kfree(w); 515 } 516 cb->done = (void *)cb->args[3]; 517 cb->args[1] = 3; 518 } 519 520 static int fib6_dump_done(struct netlink_callback *cb) 521 { 522 fib6_dump_end(cb); 523 return cb->done ? cb->done(cb) : 0; 524 } 525 526 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 527 struct netlink_callback *cb) 528 { 529 struct net *net = sock_net(skb->sk); 530 struct fib6_walker *w; 531 int res; 532 533 w = (void *)cb->args[2]; 534 w->root = &table->tb6_root; 535 536 if (cb->args[4] == 0) { 537 w->count = 0; 538 w->skip = 0; 539 540 spin_lock_bh(&table->tb6_lock); 541 res = fib6_walk(net, w); 542 spin_unlock_bh(&table->tb6_lock); 543 if (res > 0) { 544 cb->args[4] = 1; 545 cb->args[5] = w->root->fn_sernum; 546 } 547 } else { 548 if (cb->args[5] != w->root->fn_sernum) { 549 /* Begin at the root if the tree changed */ 550 cb->args[5] = w->root->fn_sernum; 551 w->state = FWS_INIT; 552 w->node = w->root; 553 w->skip = w->count; 554 } else 555 w->skip = 0; 556 557 spin_lock_bh(&table->tb6_lock); 558 res = fib6_walk_continue(w); 559 spin_unlock_bh(&table->tb6_lock); 560 if (res <= 0) { 561 fib6_walker_unlink(net, w); 562 cb->args[4] = 0; 563 } 564 } 565 566 return res; 567 } 568 569 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 570 { 571 struct net *net = sock_net(skb->sk); 572 unsigned int h, s_h; 573 unsigned int e = 0, s_e; 574 struct rt6_rtnl_dump_arg arg; 575 struct fib6_walker *w; 576 struct fib6_table *tb; 577 struct hlist_head *head; 578 int res = 0; 579 580 s_h = cb->args[0]; 581 s_e = cb->args[1]; 582 583 w = (void *)cb->args[2]; 584 if (!w) { 585 /* New dump: 586 * 587 * 1. hook callback destructor. 588 */ 589 cb->args[3] = (long)cb->done; 590 cb->done = fib6_dump_done; 591 592 /* 593 * 2. allocate and initialize walker. 594 */ 595 w = kzalloc(sizeof(*w), GFP_ATOMIC); 596 if (!w) 597 return -ENOMEM; 598 w->func = fib6_dump_node; 599 cb->args[2] = (long)w; 600 } 601 602 arg.skb = skb; 603 arg.cb = cb; 604 arg.net = net; 605 w->args = &arg; 606 607 rcu_read_lock(); 608 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) { 609 e = 0; 610 head = &net->ipv6.fib_table_hash[h]; 611 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 612 if (e < s_e) 613 goto next; 614 res = fib6_dump_table(tb, skb, cb); 615 if (res != 0) 616 goto out; 617 next: 618 e++; 619 } 620 } 621 out: 622 rcu_read_unlock(); 623 cb->args[1] = e; 624 cb->args[0] = h; 625 626 res = res < 0 ? res : skb->len; 627 if (res <= 0) 628 fib6_dump_end(cb); 629 return res; 630 } 631 632 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val) 633 { 634 if (!f6i) 635 return; 636 637 if (f6i->fib6_metrics == &dst_default_metrics) { 638 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC); 639 640 if (!p) 641 return; 642 643 refcount_set(&p->refcnt, 1); 644 f6i->fib6_metrics = p; 645 } 646 647 f6i->fib6_metrics->metrics[metric - 1] = val; 648 } 649 650 /* 651 * Routing Table 652 * 653 * return the appropriate node for a routing tree "add" operation 654 * by either creating and inserting or by returning an existing 655 * node. 656 */ 657 658 static struct fib6_node *fib6_add_1(struct net *net, 659 struct fib6_table *table, 660 struct fib6_node *root, 661 struct in6_addr *addr, int plen, 662 int offset, int allow_create, 663 int replace_required, 664 struct netlink_ext_ack *extack) 665 { 666 struct fib6_node *fn, *in, *ln; 667 struct fib6_node *pn = NULL; 668 struct rt6key *key; 669 int bit; 670 __be32 dir = 0; 671 672 RT6_TRACE("fib6_add_1\n"); 673 674 /* insert node in tree */ 675 676 fn = root; 677 678 do { 679 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf, 680 lockdep_is_held(&table->tb6_lock)); 681 key = (struct rt6key *)((u8 *)leaf + offset); 682 683 /* 684 * Prefix match 685 */ 686 if (plen < fn->fn_bit || 687 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) { 688 if (!allow_create) { 689 if (replace_required) { 690 NL_SET_ERR_MSG(extack, 691 "Can not replace route - no match found"); 692 pr_warn("Can't replace route, no match found\n"); 693 return ERR_PTR(-ENOENT); 694 } 695 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 696 } 697 goto insert_above; 698 } 699 700 /* 701 * Exact match ? 702 */ 703 704 if (plen == fn->fn_bit) { 705 /* clean up an intermediate node */ 706 if (!(fn->fn_flags & RTN_RTINFO)) { 707 RCU_INIT_POINTER(fn->leaf, NULL); 708 fib6_info_release(leaf); 709 /* remove null_entry in the root node */ 710 } else if (fn->fn_flags & RTN_TL_ROOT && 711 rcu_access_pointer(fn->leaf) == 712 net->ipv6.fib6_null_entry) { 713 RCU_INIT_POINTER(fn->leaf, NULL); 714 } 715 716 return fn; 717 } 718 719 /* 720 * We have more bits to go 721 */ 722 723 /* Try to walk down on tree. */ 724 dir = addr_bit_set(addr, fn->fn_bit); 725 pn = fn; 726 fn = dir ? 727 rcu_dereference_protected(fn->right, 728 lockdep_is_held(&table->tb6_lock)) : 729 rcu_dereference_protected(fn->left, 730 lockdep_is_held(&table->tb6_lock)); 731 } while (fn); 732 733 if (!allow_create) { 734 /* We should not create new node because 735 * NLM_F_REPLACE was specified without NLM_F_CREATE 736 * I assume it is safe to require NLM_F_CREATE when 737 * REPLACE flag is used! Later we may want to remove the 738 * check for replace_required, because according 739 * to netlink specification, NLM_F_CREATE 740 * MUST be specified if new route is created. 741 * That would keep IPv6 consistent with IPv4 742 */ 743 if (replace_required) { 744 NL_SET_ERR_MSG(extack, 745 "Can not replace route - no match found"); 746 pr_warn("Can't replace route, no match found\n"); 747 return ERR_PTR(-ENOENT); 748 } 749 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 750 } 751 /* 752 * We walked to the bottom of tree. 753 * Create new leaf node without children. 754 */ 755 756 ln = node_alloc(net); 757 758 if (!ln) 759 return ERR_PTR(-ENOMEM); 760 ln->fn_bit = plen; 761 RCU_INIT_POINTER(ln->parent, pn); 762 763 if (dir) 764 rcu_assign_pointer(pn->right, ln); 765 else 766 rcu_assign_pointer(pn->left, ln); 767 768 return ln; 769 770 771 insert_above: 772 /* 773 * split since we don't have a common prefix anymore or 774 * we have a less significant route. 775 * we've to insert an intermediate node on the list 776 * this new node will point to the one we need to create 777 * and the current 778 */ 779 780 pn = rcu_dereference_protected(fn->parent, 781 lockdep_is_held(&table->tb6_lock)); 782 783 /* find 1st bit in difference between the 2 addrs. 784 785 See comment in __ipv6_addr_diff: bit may be an invalid value, 786 but if it is >= plen, the value is ignored in any case. 787 */ 788 789 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr)); 790 791 /* 792 * (intermediate)[in] 793 * / \ 794 * (new leaf node)[ln] (old node)[fn] 795 */ 796 if (plen > bit) { 797 in = node_alloc(net); 798 ln = node_alloc(net); 799 800 if (!in || !ln) { 801 if (in) 802 node_free_immediate(net, in); 803 if (ln) 804 node_free_immediate(net, ln); 805 return ERR_PTR(-ENOMEM); 806 } 807 808 /* 809 * new intermediate node. 810 * RTN_RTINFO will 811 * be off since that an address that chooses one of 812 * the branches would not match less specific routes 813 * in the other branch 814 */ 815 816 in->fn_bit = bit; 817 818 RCU_INIT_POINTER(in->parent, pn); 819 in->leaf = fn->leaf; 820 atomic_inc(&rcu_dereference_protected(in->leaf, 821 lockdep_is_held(&table->tb6_lock))->fib6_ref); 822 823 /* update parent pointer */ 824 if (dir) 825 rcu_assign_pointer(pn->right, in); 826 else 827 rcu_assign_pointer(pn->left, in); 828 829 ln->fn_bit = plen; 830 831 RCU_INIT_POINTER(ln->parent, in); 832 rcu_assign_pointer(fn->parent, in); 833 834 if (addr_bit_set(addr, bit)) { 835 rcu_assign_pointer(in->right, ln); 836 rcu_assign_pointer(in->left, fn); 837 } else { 838 rcu_assign_pointer(in->left, ln); 839 rcu_assign_pointer(in->right, fn); 840 } 841 } else { /* plen <= bit */ 842 843 /* 844 * (new leaf node)[ln] 845 * / \ 846 * (old node)[fn] NULL 847 */ 848 849 ln = node_alloc(net); 850 851 if (!ln) 852 return ERR_PTR(-ENOMEM); 853 854 ln->fn_bit = plen; 855 856 RCU_INIT_POINTER(ln->parent, pn); 857 858 if (addr_bit_set(&key->addr, plen)) 859 RCU_INIT_POINTER(ln->right, fn); 860 else 861 RCU_INIT_POINTER(ln->left, fn); 862 863 rcu_assign_pointer(fn->parent, ln); 864 865 if (dir) 866 rcu_assign_pointer(pn->right, ln); 867 else 868 rcu_assign_pointer(pn->left, ln); 869 } 870 return ln; 871 } 872 873 static void fib6_drop_pcpu_from(struct fib6_info *f6i, 874 const struct fib6_table *table) 875 { 876 int cpu; 877 878 /* release the reference to this fib entry from 879 * all of its cached pcpu routes 880 */ 881 for_each_possible_cpu(cpu) { 882 struct rt6_info **ppcpu_rt; 883 struct rt6_info *pcpu_rt; 884 885 ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu); 886 pcpu_rt = *ppcpu_rt; 887 if (pcpu_rt) { 888 struct fib6_info *from; 889 890 from = rcu_dereference_protected(pcpu_rt->from, 891 lockdep_is_held(&table->tb6_lock)); 892 rcu_assign_pointer(pcpu_rt->from, NULL); 893 fib6_info_release(from); 894 } 895 } 896 } 897 898 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn, 899 struct net *net) 900 { 901 struct fib6_table *table = rt->fib6_table; 902 903 if (atomic_read(&rt->fib6_ref) != 1) { 904 /* This route is used as dummy address holder in some split 905 * nodes. It is not leaked, but it still holds other resources, 906 * which must be released in time. So, scan ascendant nodes 907 * and replace dummy references to this route with references 908 * to still alive ones. 909 */ 910 while (fn) { 911 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf, 912 lockdep_is_held(&table->tb6_lock)); 913 struct fib6_info *new_leaf; 914 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) { 915 new_leaf = fib6_find_prefix(net, table, fn); 916 atomic_inc(&new_leaf->fib6_ref); 917 918 rcu_assign_pointer(fn->leaf, new_leaf); 919 fib6_info_release(rt); 920 } 921 fn = rcu_dereference_protected(fn->parent, 922 lockdep_is_held(&table->tb6_lock)); 923 } 924 925 if (rt->rt6i_pcpu) 926 fib6_drop_pcpu_from(rt, table); 927 } 928 } 929 930 /* 931 * Insert routing information in a node. 932 */ 933 934 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt, 935 struct nl_info *info, 936 struct netlink_ext_ack *extack) 937 { 938 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf, 939 lockdep_is_held(&rt->fib6_table->tb6_lock)); 940 struct fib6_info *iter = NULL; 941 struct fib6_info __rcu **ins; 942 struct fib6_info __rcu **fallback_ins = NULL; 943 int replace = (info->nlh && 944 (info->nlh->nlmsg_flags & NLM_F_REPLACE)); 945 int add = (!info->nlh || 946 (info->nlh->nlmsg_flags & NLM_F_CREATE)); 947 int found = 0; 948 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt); 949 u16 nlflags = NLM_F_EXCL; 950 int err; 951 952 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND)) 953 nlflags |= NLM_F_APPEND; 954 955 ins = &fn->leaf; 956 957 for (iter = leaf; iter; 958 iter = rcu_dereference_protected(iter->fib6_next, 959 lockdep_is_held(&rt->fib6_table->tb6_lock))) { 960 /* 961 * Search for duplicates 962 */ 963 964 if (iter->fib6_metric == rt->fib6_metric) { 965 /* 966 * Same priority level 967 */ 968 if (info->nlh && 969 (info->nlh->nlmsg_flags & NLM_F_EXCL)) 970 return -EEXIST; 971 972 nlflags &= ~NLM_F_EXCL; 973 if (replace) { 974 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) { 975 found++; 976 break; 977 } 978 if (rt_can_ecmp) 979 fallback_ins = fallback_ins ?: ins; 980 goto next_iter; 981 } 982 983 if (rt6_duplicate_nexthop(iter, rt)) { 984 if (rt->fib6_nsiblings) 985 rt->fib6_nsiblings = 0; 986 if (!(iter->fib6_flags & RTF_EXPIRES)) 987 return -EEXIST; 988 if (!(rt->fib6_flags & RTF_EXPIRES)) 989 fib6_clean_expires(iter); 990 else 991 fib6_set_expires(iter, rt->expires); 992 fib6_metric_set(iter, RTAX_MTU, rt->fib6_pmtu); 993 return -EEXIST; 994 } 995 /* If we have the same destination and the same metric, 996 * but not the same gateway, then the route we try to 997 * add is sibling to this route, increment our counter 998 * of siblings, and later we will add our route to the 999 * list. 1000 * Only static routes (which don't have flag 1001 * RTF_EXPIRES) are used for ECMPv6. 1002 * 1003 * To avoid long list, we only had siblings if the 1004 * route have a gateway. 1005 */ 1006 if (rt_can_ecmp && 1007 rt6_qualify_for_ecmp(iter)) 1008 rt->fib6_nsiblings++; 1009 } 1010 1011 if (iter->fib6_metric > rt->fib6_metric) 1012 break; 1013 1014 next_iter: 1015 ins = &iter->fib6_next; 1016 } 1017 1018 if (fallback_ins && !found) { 1019 /* No ECMP-able route found, replace first non-ECMP one */ 1020 ins = fallback_ins; 1021 iter = rcu_dereference_protected(*ins, 1022 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1023 found++; 1024 } 1025 1026 /* Reset round-robin state, if necessary */ 1027 if (ins == &fn->leaf) 1028 fn->rr_ptr = NULL; 1029 1030 /* Link this route to others same route. */ 1031 if (rt->fib6_nsiblings) { 1032 unsigned int fib6_nsiblings; 1033 struct fib6_info *sibling, *temp_sibling; 1034 1035 /* Find the first route that have the same metric */ 1036 sibling = leaf; 1037 while (sibling) { 1038 if (sibling->fib6_metric == rt->fib6_metric && 1039 rt6_qualify_for_ecmp(sibling)) { 1040 list_add_tail(&rt->fib6_siblings, 1041 &sibling->fib6_siblings); 1042 break; 1043 } 1044 sibling = rcu_dereference_protected(sibling->fib6_next, 1045 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1046 } 1047 /* For each sibling in the list, increment the counter of 1048 * siblings. BUG() if counters does not match, list of siblings 1049 * is broken! 1050 */ 1051 fib6_nsiblings = 0; 1052 list_for_each_entry_safe(sibling, temp_sibling, 1053 &rt->fib6_siblings, fib6_siblings) { 1054 sibling->fib6_nsiblings++; 1055 BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings); 1056 fib6_nsiblings++; 1057 } 1058 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings); 1059 rt6_multipath_rebalance(temp_sibling); 1060 } 1061 1062 /* 1063 * insert node 1064 */ 1065 if (!replace) { 1066 if (!add) 1067 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 1068 1069 add: 1070 nlflags |= NLM_F_CREATE; 1071 1072 err = call_fib6_entry_notifiers(info->nl_net, 1073 FIB_EVENT_ENTRY_ADD, 1074 rt, extack); 1075 if (err) 1076 return err; 1077 1078 rcu_assign_pointer(rt->fib6_next, iter); 1079 atomic_inc(&rt->fib6_ref); 1080 rcu_assign_pointer(rt->fib6_node, fn); 1081 rcu_assign_pointer(*ins, rt); 1082 if (!info->skip_notify) 1083 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags); 1084 info->nl_net->ipv6.rt6_stats->fib_rt_entries++; 1085 1086 if (!(fn->fn_flags & RTN_RTINFO)) { 1087 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 1088 fn->fn_flags |= RTN_RTINFO; 1089 } 1090 1091 } else { 1092 int nsiblings; 1093 1094 if (!found) { 1095 if (add) 1096 goto add; 1097 pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); 1098 return -ENOENT; 1099 } 1100 1101 err = call_fib6_entry_notifiers(info->nl_net, 1102 FIB_EVENT_ENTRY_REPLACE, 1103 rt, extack); 1104 if (err) 1105 return err; 1106 1107 atomic_inc(&rt->fib6_ref); 1108 rcu_assign_pointer(rt->fib6_node, fn); 1109 rt->fib6_next = iter->fib6_next; 1110 rcu_assign_pointer(*ins, rt); 1111 if (!info->skip_notify) 1112 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE); 1113 if (!(fn->fn_flags & RTN_RTINFO)) { 1114 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 1115 fn->fn_flags |= RTN_RTINFO; 1116 } 1117 nsiblings = iter->fib6_nsiblings; 1118 iter->fib6_node = NULL; 1119 fib6_purge_rt(iter, fn, info->nl_net); 1120 if (rcu_access_pointer(fn->rr_ptr) == iter) 1121 fn->rr_ptr = NULL; 1122 fib6_info_release(iter); 1123 1124 if (nsiblings) { 1125 /* Replacing an ECMP route, remove all siblings */ 1126 ins = &rt->fib6_next; 1127 iter = rcu_dereference_protected(*ins, 1128 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1129 while (iter) { 1130 if (iter->fib6_metric > rt->fib6_metric) 1131 break; 1132 if (rt6_qualify_for_ecmp(iter)) { 1133 *ins = iter->fib6_next; 1134 iter->fib6_node = NULL; 1135 fib6_purge_rt(iter, fn, info->nl_net); 1136 if (rcu_access_pointer(fn->rr_ptr) == iter) 1137 fn->rr_ptr = NULL; 1138 fib6_info_release(iter); 1139 nsiblings--; 1140 info->nl_net->ipv6.rt6_stats->fib_rt_entries--; 1141 } else { 1142 ins = &iter->fib6_next; 1143 } 1144 iter = rcu_dereference_protected(*ins, 1145 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1146 } 1147 WARN_ON(nsiblings != 0); 1148 } 1149 } 1150 1151 return 0; 1152 } 1153 1154 static void fib6_start_gc(struct net *net, struct fib6_info *rt) 1155 { 1156 if (!timer_pending(&net->ipv6.ip6_fib_timer) && 1157 (rt->fib6_flags & RTF_EXPIRES)) 1158 mod_timer(&net->ipv6.ip6_fib_timer, 1159 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 1160 } 1161 1162 void fib6_force_start_gc(struct net *net) 1163 { 1164 if (!timer_pending(&net->ipv6.ip6_fib_timer)) 1165 mod_timer(&net->ipv6.ip6_fib_timer, 1166 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 1167 } 1168 1169 static void __fib6_update_sernum_upto_root(struct fib6_info *rt, 1170 int sernum) 1171 { 1172 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node, 1173 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1174 1175 /* paired with smp_rmb() in rt6_get_cookie_safe() */ 1176 smp_wmb(); 1177 while (fn) { 1178 fn->fn_sernum = sernum; 1179 fn = rcu_dereference_protected(fn->parent, 1180 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1181 } 1182 } 1183 1184 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt) 1185 { 1186 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net)); 1187 } 1188 1189 /* 1190 * Add routing information to the routing tree. 1191 * <destination addr>/<source addr> 1192 * with source addr info in sub-trees 1193 * Need to own table->tb6_lock 1194 */ 1195 1196 int fib6_add(struct fib6_node *root, struct fib6_info *rt, 1197 struct nl_info *info, struct netlink_ext_ack *extack) 1198 { 1199 struct fib6_table *table = rt->fib6_table; 1200 struct fib6_node *fn, *pn = NULL; 1201 int err = -ENOMEM; 1202 int allow_create = 1; 1203 int replace_required = 0; 1204 int sernum = fib6_new_sernum(info->nl_net); 1205 1206 if (info->nlh) { 1207 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) 1208 allow_create = 0; 1209 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) 1210 replace_required = 1; 1211 } 1212 if (!allow_create && !replace_required) 1213 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); 1214 1215 fn = fib6_add_1(info->nl_net, table, root, 1216 &rt->fib6_dst.addr, rt->fib6_dst.plen, 1217 offsetof(struct fib6_info, fib6_dst), allow_create, 1218 replace_required, extack); 1219 if (IS_ERR(fn)) { 1220 err = PTR_ERR(fn); 1221 fn = NULL; 1222 goto out; 1223 } 1224 1225 pn = fn; 1226 1227 #ifdef CONFIG_IPV6_SUBTREES 1228 if (rt->fib6_src.plen) { 1229 struct fib6_node *sn; 1230 1231 if (!rcu_access_pointer(fn->subtree)) { 1232 struct fib6_node *sfn; 1233 1234 /* 1235 * Create subtree. 1236 * 1237 * fn[main tree] 1238 * | 1239 * sfn[subtree root] 1240 * \ 1241 * sn[new leaf node] 1242 */ 1243 1244 /* Create subtree root node */ 1245 sfn = node_alloc(info->nl_net); 1246 if (!sfn) 1247 goto failure; 1248 1249 atomic_inc(&info->nl_net->ipv6.fib6_null_entry->fib6_ref); 1250 rcu_assign_pointer(sfn->leaf, 1251 info->nl_net->ipv6.fib6_null_entry); 1252 sfn->fn_flags = RTN_ROOT; 1253 1254 /* Now add the first leaf node to new subtree */ 1255 1256 sn = fib6_add_1(info->nl_net, table, sfn, 1257 &rt->fib6_src.addr, rt->fib6_src.plen, 1258 offsetof(struct fib6_info, fib6_src), 1259 allow_create, replace_required, extack); 1260 1261 if (IS_ERR(sn)) { 1262 /* If it is failed, discard just allocated 1263 root, and then (in failure) stale node 1264 in main tree. 1265 */ 1266 node_free_immediate(info->nl_net, sfn); 1267 err = PTR_ERR(sn); 1268 goto failure; 1269 } 1270 1271 /* Now link new subtree to main tree */ 1272 rcu_assign_pointer(sfn->parent, fn); 1273 rcu_assign_pointer(fn->subtree, sfn); 1274 } else { 1275 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn), 1276 &rt->fib6_src.addr, rt->fib6_src.plen, 1277 offsetof(struct fib6_info, fib6_src), 1278 allow_create, replace_required, extack); 1279 1280 if (IS_ERR(sn)) { 1281 err = PTR_ERR(sn); 1282 goto failure; 1283 } 1284 } 1285 1286 if (!rcu_access_pointer(fn->leaf)) { 1287 if (fn->fn_flags & RTN_TL_ROOT) { 1288 /* put back null_entry for root node */ 1289 rcu_assign_pointer(fn->leaf, 1290 info->nl_net->ipv6.fib6_null_entry); 1291 } else { 1292 atomic_inc(&rt->fib6_ref); 1293 rcu_assign_pointer(fn->leaf, rt); 1294 } 1295 } 1296 fn = sn; 1297 } 1298 #endif 1299 1300 err = fib6_add_rt2node(fn, rt, info, extack); 1301 if (!err) { 1302 __fib6_update_sernum_upto_root(rt, sernum); 1303 fib6_start_gc(info->nl_net, rt); 1304 } 1305 1306 out: 1307 if (err) { 1308 #ifdef CONFIG_IPV6_SUBTREES 1309 /* 1310 * If fib6_add_1 has cleared the old leaf pointer in the 1311 * super-tree leaf node we have to find a new one for it. 1312 */ 1313 if (pn != fn) { 1314 struct fib6_info *pn_leaf = 1315 rcu_dereference_protected(pn->leaf, 1316 lockdep_is_held(&table->tb6_lock)); 1317 if (pn_leaf == rt) { 1318 pn_leaf = NULL; 1319 RCU_INIT_POINTER(pn->leaf, NULL); 1320 fib6_info_release(rt); 1321 } 1322 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) { 1323 pn_leaf = fib6_find_prefix(info->nl_net, table, 1324 pn); 1325 #if RT6_DEBUG >= 2 1326 if (!pn_leaf) { 1327 WARN_ON(!pn_leaf); 1328 pn_leaf = 1329 info->nl_net->ipv6.fib6_null_entry; 1330 } 1331 #endif 1332 fib6_info_hold(pn_leaf); 1333 rcu_assign_pointer(pn->leaf, pn_leaf); 1334 } 1335 } 1336 #endif 1337 goto failure; 1338 } 1339 return err; 1340 1341 failure: 1342 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if: 1343 * 1. fn is an intermediate node and we failed to add the new 1344 * route to it in both subtree creation failure and fib6_add_rt2node() 1345 * failure case. 1346 * 2. fn is the root node in the table and we fail to add the first 1347 * default route to it. 1348 */ 1349 if (fn && 1350 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) || 1351 (fn->fn_flags & RTN_TL_ROOT && 1352 !rcu_access_pointer(fn->leaf)))) 1353 fib6_repair_tree(info->nl_net, table, fn); 1354 return err; 1355 } 1356 1357 /* 1358 * Routing tree lookup 1359 * 1360 */ 1361 1362 struct lookup_args { 1363 int offset; /* key offset on fib6_info */ 1364 const struct in6_addr *addr; /* search key */ 1365 }; 1366 1367 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root, 1368 struct lookup_args *args) 1369 { 1370 struct fib6_node *fn; 1371 __be32 dir; 1372 1373 if (unlikely(args->offset == 0)) 1374 return NULL; 1375 1376 /* 1377 * Descend on a tree 1378 */ 1379 1380 fn = root; 1381 1382 for (;;) { 1383 struct fib6_node *next; 1384 1385 dir = addr_bit_set(args->addr, fn->fn_bit); 1386 1387 next = dir ? rcu_dereference(fn->right) : 1388 rcu_dereference(fn->left); 1389 1390 if (next) { 1391 fn = next; 1392 continue; 1393 } 1394 break; 1395 } 1396 1397 while (fn) { 1398 struct fib6_node *subtree = FIB6_SUBTREE(fn); 1399 1400 if (subtree || fn->fn_flags & RTN_RTINFO) { 1401 struct fib6_info *leaf = rcu_dereference(fn->leaf); 1402 struct rt6key *key; 1403 1404 if (!leaf) 1405 goto backtrack; 1406 1407 key = (struct rt6key *) ((u8 *)leaf + args->offset); 1408 1409 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 1410 #ifdef CONFIG_IPV6_SUBTREES 1411 if (subtree) { 1412 struct fib6_node *sfn; 1413 sfn = fib6_node_lookup_1(subtree, 1414 args + 1); 1415 if (!sfn) 1416 goto backtrack; 1417 fn = sfn; 1418 } 1419 #endif 1420 if (fn->fn_flags & RTN_RTINFO) 1421 return fn; 1422 } 1423 } 1424 backtrack: 1425 if (fn->fn_flags & RTN_ROOT) 1426 break; 1427 1428 fn = rcu_dereference(fn->parent); 1429 } 1430 1431 return NULL; 1432 } 1433 1434 /* called with rcu_read_lock() held 1435 */ 1436 struct fib6_node *fib6_node_lookup(struct fib6_node *root, 1437 const struct in6_addr *daddr, 1438 const struct in6_addr *saddr) 1439 { 1440 struct fib6_node *fn; 1441 struct lookup_args args[] = { 1442 { 1443 .offset = offsetof(struct fib6_info, fib6_dst), 1444 .addr = daddr, 1445 }, 1446 #ifdef CONFIG_IPV6_SUBTREES 1447 { 1448 .offset = offsetof(struct fib6_info, fib6_src), 1449 .addr = saddr, 1450 }, 1451 #endif 1452 { 1453 .offset = 0, /* sentinel */ 1454 } 1455 }; 1456 1457 fn = fib6_node_lookup_1(root, daddr ? args : args + 1); 1458 if (!fn || fn->fn_flags & RTN_TL_ROOT) 1459 fn = root; 1460 1461 return fn; 1462 } 1463 1464 /* 1465 * Get node with specified destination prefix (and source prefix, 1466 * if subtrees are used) 1467 * exact_match == true means we try to find fn with exact match of 1468 * the passed in prefix addr 1469 * exact_match == false means we try to find fn with longest prefix 1470 * match of the passed in prefix addr. This is useful for finding fn 1471 * for cached route as it will be stored in the exception table under 1472 * the node with longest prefix length. 1473 */ 1474 1475 1476 static struct fib6_node *fib6_locate_1(struct fib6_node *root, 1477 const struct in6_addr *addr, 1478 int plen, int offset, 1479 bool exact_match) 1480 { 1481 struct fib6_node *fn, *prev = NULL; 1482 1483 for (fn = root; fn ; ) { 1484 struct fib6_info *leaf = rcu_dereference(fn->leaf); 1485 struct rt6key *key; 1486 1487 /* This node is being deleted */ 1488 if (!leaf) { 1489 if (plen <= fn->fn_bit) 1490 goto out; 1491 else 1492 goto next; 1493 } 1494 1495 key = (struct rt6key *)((u8 *)leaf + offset); 1496 1497 /* 1498 * Prefix match 1499 */ 1500 if (plen < fn->fn_bit || 1501 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 1502 goto out; 1503 1504 if (plen == fn->fn_bit) 1505 return fn; 1506 1507 prev = fn; 1508 1509 next: 1510 /* 1511 * We have more bits to go 1512 */ 1513 if (addr_bit_set(addr, fn->fn_bit)) 1514 fn = rcu_dereference(fn->right); 1515 else 1516 fn = rcu_dereference(fn->left); 1517 } 1518 out: 1519 if (exact_match) 1520 return NULL; 1521 else 1522 return prev; 1523 } 1524 1525 struct fib6_node *fib6_locate(struct fib6_node *root, 1526 const struct in6_addr *daddr, int dst_len, 1527 const struct in6_addr *saddr, int src_len, 1528 bool exact_match) 1529 { 1530 struct fib6_node *fn; 1531 1532 fn = fib6_locate_1(root, daddr, dst_len, 1533 offsetof(struct fib6_info, fib6_dst), 1534 exact_match); 1535 1536 #ifdef CONFIG_IPV6_SUBTREES 1537 if (src_len) { 1538 WARN_ON(saddr == NULL); 1539 if (fn) { 1540 struct fib6_node *subtree = FIB6_SUBTREE(fn); 1541 1542 if (subtree) { 1543 fn = fib6_locate_1(subtree, saddr, src_len, 1544 offsetof(struct fib6_info, fib6_src), 1545 exact_match); 1546 } 1547 } 1548 } 1549 #endif 1550 1551 if (fn && fn->fn_flags & RTN_RTINFO) 1552 return fn; 1553 1554 return NULL; 1555 } 1556 1557 1558 /* 1559 * Deletion 1560 * 1561 */ 1562 1563 static struct fib6_info *fib6_find_prefix(struct net *net, 1564 struct fib6_table *table, 1565 struct fib6_node *fn) 1566 { 1567 struct fib6_node *child_left, *child_right; 1568 1569 if (fn->fn_flags & RTN_ROOT) 1570 return net->ipv6.fib6_null_entry; 1571 1572 while (fn) { 1573 child_left = rcu_dereference_protected(fn->left, 1574 lockdep_is_held(&table->tb6_lock)); 1575 child_right = rcu_dereference_protected(fn->right, 1576 lockdep_is_held(&table->tb6_lock)); 1577 if (child_left) 1578 return rcu_dereference_protected(child_left->leaf, 1579 lockdep_is_held(&table->tb6_lock)); 1580 if (child_right) 1581 return rcu_dereference_protected(child_right->leaf, 1582 lockdep_is_held(&table->tb6_lock)); 1583 1584 fn = FIB6_SUBTREE(fn); 1585 } 1586 return NULL; 1587 } 1588 1589 /* 1590 * Called to trim the tree of intermediate nodes when possible. "fn" 1591 * is the node we want to try and remove. 1592 * Need to own table->tb6_lock 1593 */ 1594 1595 static struct fib6_node *fib6_repair_tree(struct net *net, 1596 struct fib6_table *table, 1597 struct fib6_node *fn) 1598 { 1599 int children; 1600 int nstate; 1601 struct fib6_node *child; 1602 struct fib6_walker *w; 1603 int iter = 0; 1604 1605 /* Set fn->leaf to null_entry for root node. */ 1606 if (fn->fn_flags & RTN_TL_ROOT) { 1607 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry); 1608 return fn; 1609 } 1610 1611 for (;;) { 1612 struct fib6_node *fn_r = rcu_dereference_protected(fn->right, 1613 lockdep_is_held(&table->tb6_lock)); 1614 struct fib6_node *fn_l = rcu_dereference_protected(fn->left, 1615 lockdep_is_held(&table->tb6_lock)); 1616 struct fib6_node *pn = rcu_dereference_protected(fn->parent, 1617 lockdep_is_held(&table->tb6_lock)); 1618 struct fib6_node *pn_r = rcu_dereference_protected(pn->right, 1619 lockdep_is_held(&table->tb6_lock)); 1620 struct fib6_node *pn_l = rcu_dereference_protected(pn->left, 1621 lockdep_is_held(&table->tb6_lock)); 1622 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf, 1623 lockdep_is_held(&table->tb6_lock)); 1624 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf, 1625 lockdep_is_held(&table->tb6_lock)); 1626 struct fib6_info *new_fn_leaf; 1627 1628 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1629 iter++; 1630 1631 WARN_ON(fn->fn_flags & RTN_RTINFO); 1632 WARN_ON(fn->fn_flags & RTN_TL_ROOT); 1633 WARN_ON(fn_leaf); 1634 1635 children = 0; 1636 child = NULL; 1637 if (fn_r) 1638 child = fn_r, children |= 1; 1639 if (fn_l) 1640 child = fn_l, children |= 2; 1641 1642 if (children == 3 || FIB6_SUBTREE(fn) 1643 #ifdef CONFIG_IPV6_SUBTREES 1644 /* Subtree root (i.e. fn) may have one child */ 1645 || (children && fn->fn_flags & RTN_ROOT) 1646 #endif 1647 ) { 1648 new_fn_leaf = fib6_find_prefix(net, table, fn); 1649 #if RT6_DEBUG >= 2 1650 if (!new_fn_leaf) { 1651 WARN_ON(!new_fn_leaf); 1652 new_fn_leaf = net->ipv6.fib6_null_entry; 1653 } 1654 #endif 1655 fib6_info_hold(new_fn_leaf); 1656 rcu_assign_pointer(fn->leaf, new_fn_leaf); 1657 return pn; 1658 } 1659 1660 #ifdef CONFIG_IPV6_SUBTREES 1661 if (FIB6_SUBTREE(pn) == fn) { 1662 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1663 RCU_INIT_POINTER(pn->subtree, NULL); 1664 nstate = FWS_L; 1665 } else { 1666 WARN_ON(fn->fn_flags & RTN_ROOT); 1667 #endif 1668 if (pn_r == fn) 1669 rcu_assign_pointer(pn->right, child); 1670 else if (pn_l == fn) 1671 rcu_assign_pointer(pn->left, child); 1672 #if RT6_DEBUG >= 2 1673 else 1674 WARN_ON(1); 1675 #endif 1676 if (child) 1677 rcu_assign_pointer(child->parent, pn); 1678 nstate = FWS_R; 1679 #ifdef CONFIG_IPV6_SUBTREES 1680 } 1681 #endif 1682 1683 read_lock(&net->ipv6.fib6_walker_lock); 1684 FOR_WALKERS(net, w) { 1685 if (!child) { 1686 if (w->node == fn) { 1687 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1688 w->node = pn; 1689 w->state = nstate; 1690 } 1691 } else { 1692 if (w->node == fn) { 1693 w->node = child; 1694 if (children&2) { 1695 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1696 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT; 1697 } else { 1698 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1699 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT; 1700 } 1701 } 1702 } 1703 } 1704 read_unlock(&net->ipv6.fib6_walker_lock); 1705 1706 node_free(net, fn); 1707 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) 1708 return pn; 1709 1710 RCU_INIT_POINTER(pn->leaf, NULL); 1711 fib6_info_release(pn_leaf); 1712 fn = pn; 1713 } 1714 } 1715 1716 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn, 1717 struct fib6_info __rcu **rtp, struct nl_info *info) 1718 { 1719 struct fib6_walker *w; 1720 struct fib6_info *rt = rcu_dereference_protected(*rtp, 1721 lockdep_is_held(&table->tb6_lock)); 1722 struct net *net = info->nl_net; 1723 1724 RT6_TRACE("fib6_del_route\n"); 1725 1726 /* Unlink it */ 1727 *rtp = rt->fib6_next; 1728 rt->fib6_node = NULL; 1729 net->ipv6.rt6_stats->fib_rt_entries--; 1730 net->ipv6.rt6_stats->fib_discarded_routes++; 1731 1732 /* Flush all cached dst in exception table */ 1733 rt6_flush_exceptions(rt); 1734 1735 /* Reset round-robin state, if necessary */ 1736 if (rcu_access_pointer(fn->rr_ptr) == rt) 1737 fn->rr_ptr = NULL; 1738 1739 /* Remove this entry from other siblings */ 1740 if (rt->fib6_nsiblings) { 1741 struct fib6_info *sibling, *next_sibling; 1742 1743 list_for_each_entry_safe(sibling, next_sibling, 1744 &rt->fib6_siblings, fib6_siblings) 1745 sibling->fib6_nsiblings--; 1746 rt->fib6_nsiblings = 0; 1747 list_del_init(&rt->fib6_siblings); 1748 rt6_multipath_rebalance(next_sibling); 1749 } 1750 1751 /* Adjust walkers */ 1752 read_lock(&net->ipv6.fib6_walker_lock); 1753 FOR_WALKERS(net, w) { 1754 if (w->state == FWS_C && w->leaf == rt) { 1755 RT6_TRACE("walker %p adjusted by delroute\n", w); 1756 w->leaf = rcu_dereference_protected(rt->fib6_next, 1757 lockdep_is_held(&table->tb6_lock)); 1758 if (!w->leaf) 1759 w->state = FWS_U; 1760 } 1761 } 1762 read_unlock(&net->ipv6.fib6_walker_lock); 1763 1764 /* If it was last route, call fib6_repair_tree() to: 1765 * 1. For root node, put back null_entry as how the table was created. 1766 * 2. For other nodes, expunge its radix tree node. 1767 */ 1768 if (!rcu_access_pointer(fn->leaf)) { 1769 if (!(fn->fn_flags & RTN_TL_ROOT)) { 1770 fn->fn_flags &= ~RTN_RTINFO; 1771 net->ipv6.rt6_stats->fib_route_nodes--; 1772 } 1773 fn = fib6_repair_tree(net, table, fn); 1774 } 1775 1776 fib6_purge_rt(rt, fn, net); 1777 1778 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL); 1779 if (!info->skip_notify) 1780 inet6_rt_notify(RTM_DELROUTE, rt, info, 0); 1781 fib6_info_release(rt); 1782 } 1783 1784 /* Need to own table->tb6_lock */ 1785 int fib6_del(struct fib6_info *rt, struct nl_info *info) 1786 { 1787 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node, 1788 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1789 struct fib6_table *table = rt->fib6_table; 1790 struct net *net = info->nl_net; 1791 struct fib6_info __rcu **rtp; 1792 struct fib6_info __rcu **rtp_next; 1793 1794 if (!fn || rt == net->ipv6.fib6_null_entry) 1795 return -ENOENT; 1796 1797 WARN_ON(!(fn->fn_flags & RTN_RTINFO)); 1798 1799 /* 1800 * Walk the leaf entries looking for ourself 1801 */ 1802 1803 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) { 1804 struct fib6_info *cur = rcu_dereference_protected(*rtp, 1805 lockdep_is_held(&table->tb6_lock)); 1806 if (rt == cur) { 1807 fib6_del_route(table, fn, rtp, info); 1808 return 0; 1809 } 1810 rtp_next = &cur->fib6_next; 1811 } 1812 return -ENOENT; 1813 } 1814 1815 /* 1816 * Tree traversal function. 1817 * 1818 * Certainly, it is not interrupt safe. 1819 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 1820 * It means, that we can modify tree during walking 1821 * and use this function for garbage collection, clone pruning, 1822 * cleaning tree when a device goes down etc. etc. 1823 * 1824 * It guarantees that every node will be traversed, 1825 * and that it will be traversed only once. 1826 * 1827 * Callback function w->func may return: 1828 * 0 -> continue walking. 1829 * positive value -> walking is suspended (used by tree dumps, 1830 * and probably by gc, if it will be split to several slices) 1831 * negative value -> terminate walking. 1832 * 1833 * The function itself returns: 1834 * 0 -> walk is complete. 1835 * >0 -> walk is incomplete (i.e. suspended) 1836 * <0 -> walk is terminated by an error. 1837 * 1838 * This function is called with tb6_lock held. 1839 */ 1840 1841 static int fib6_walk_continue(struct fib6_walker *w) 1842 { 1843 struct fib6_node *fn, *pn, *left, *right; 1844 1845 /* w->root should always be table->tb6_root */ 1846 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT)); 1847 1848 for (;;) { 1849 fn = w->node; 1850 if (!fn) 1851 return 0; 1852 1853 switch (w->state) { 1854 #ifdef CONFIG_IPV6_SUBTREES 1855 case FWS_S: 1856 if (FIB6_SUBTREE(fn)) { 1857 w->node = FIB6_SUBTREE(fn); 1858 continue; 1859 } 1860 w->state = FWS_L; 1861 #endif 1862 /* fall through */ 1863 case FWS_L: 1864 left = rcu_dereference_protected(fn->left, 1); 1865 if (left) { 1866 w->node = left; 1867 w->state = FWS_INIT; 1868 continue; 1869 } 1870 w->state = FWS_R; 1871 /* fall through */ 1872 case FWS_R: 1873 right = rcu_dereference_protected(fn->right, 1); 1874 if (right) { 1875 w->node = right; 1876 w->state = FWS_INIT; 1877 continue; 1878 } 1879 w->state = FWS_C; 1880 w->leaf = rcu_dereference_protected(fn->leaf, 1); 1881 /* fall through */ 1882 case FWS_C: 1883 if (w->leaf && fn->fn_flags & RTN_RTINFO) { 1884 int err; 1885 1886 if (w->skip) { 1887 w->skip--; 1888 goto skip; 1889 } 1890 1891 err = w->func(w); 1892 if (err) 1893 return err; 1894 1895 w->count++; 1896 continue; 1897 } 1898 skip: 1899 w->state = FWS_U; 1900 /* fall through */ 1901 case FWS_U: 1902 if (fn == w->root) 1903 return 0; 1904 pn = rcu_dereference_protected(fn->parent, 1); 1905 left = rcu_dereference_protected(pn->left, 1); 1906 right = rcu_dereference_protected(pn->right, 1); 1907 w->node = pn; 1908 #ifdef CONFIG_IPV6_SUBTREES 1909 if (FIB6_SUBTREE(pn) == fn) { 1910 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1911 w->state = FWS_L; 1912 continue; 1913 } 1914 #endif 1915 if (left == fn) { 1916 w->state = FWS_R; 1917 continue; 1918 } 1919 if (right == fn) { 1920 w->state = FWS_C; 1921 w->leaf = rcu_dereference_protected(w->node->leaf, 1); 1922 continue; 1923 } 1924 #if RT6_DEBUG >= 2 1925 WARN_ON(1); 1926 #endif 1927 } 1928 } 1929 } 1930 1931 static int fib6_walk(struct net *net, struct fib6_walker *w) 1932 { 1933 int res; 1934 1935 w->state = FWS_INIT; 1936 w->node = w->root; 1937 1938 fib6_walker_link(net, w); 1939 res = fib6_walk_continue(w); 1940 if (res <= 0) 1941 fib6_walker_unlink(net, w); 1942 return res; 1943 } 1944 1945 static int fib6_clean_node(struct fib6_walker *w) 1946 { 1947 int res; 1948 struct fib6_info *rt; 1949 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w); 1950 struct nl_info info = { 1951 .nl_net = c->net, 1952 }; 1953 1954 if (c->sernum != FIB6_NO_SERNUM_CHANGE && 1955 w->node->fn_sernum != c->sernum) 1956 w->node->fn_sernum = c->sernum; 1957 1958 if (!c->func) { 1959 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE); 1960 w->leaf = NULL; 1961 return 0; 1962 } 1963 1964 for_each_fib6_walker_rt(w) { 1965 res = c->func(rt, c->arg); 1966 if (res == -1) { 1967 w->leaf = rt; 1968 res = fib6_del(rt, &info); 1969 if (res) { 1970 #if RT6_DEBUG >= 2 1971 pr_debug("%s: del failed: rt=%p@%p err=%d\n", 1972 __func__, rt, 1973 rcu_access_pointer(rt->fib6_node), 1974 res); 1975 #endif 1976 continue; 1977 } 1978 return 0; 1979 } else if (res == -2) { 1980 if (WARN_ON(!rt->fib6_nsiblings)) 1981 continue; 1982 rt = list_last_entry(&rt->fib6_siblings, 1983 struct fib6_info, fib6_siblings); 1984 continue; 1985 } 1986 WARN_ON(res != 0); 1987 } 1988 w->leaf = rt; 1989 return 0; 1990 } 1991 1992 /* 1993 * Convenient frontend to tree walker. 1994 * 1995 * func is called on each route. 1996 * It may return -2 -> skip multipath route. 1997 * -1 -> delete this route. 1998 * 0 -> continue walking 1999 */ 2000 2001 static void fib6_clean_tree(struct net *net, struct fib6_node *root, 2002 int (*func)(struct fib6_info *, void *arg), 2003 int sernum, void *arg) 2004 { 2005 struct fib6_cleaner c; 2006 2007 c.w.root = root; 2008 c.w.func = fib6_clean_node; 2009 c.w.count = 0; 2010 c.w.skip = 0; 2011 c.func = func; 2012 c.sernum = sernum; 2013 c.arg = arg; 2014 c.net = net; 2015 2016 fib6_walk(net, &c.w); 2017 } 2018 2019 static void __fib6_clean_all(struct net *net, 2020 int (*func)(struct fib6_info *, void *), 2021 int sernum, void *arg) 2022 { 2023 struct fib6_table *table; 2024 struct hlist_head *head; 2025 unsigned int h; 2026 2027 rcu_read_lock(); 2028 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 2029 head = &net->ipv6.fib_table_hash[h]; 2030 hlist_for_each_entry_rcu(table, head, tb6_hlist) { 2031 spin_lock_bh(&table->tb6_lock); 2032 fib6_clean_tree(net, &table->tb6_root, 2033 func, sernum, arg); 2034 spin_unlock_bh(&table->tb6_lock); 2035 } 2036 } 2037 rcu_read_unlock(); 2038 } 2039 2040 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *), 2041 void *arg) 2042 { 2043 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg); 2044 } 2045 2046 static void fib6_flush_trees(struct net *net) 2047 { 2048 int new_sernum = fib6_new_sernum(net); 2049 2050 __fib6_clean_all(net, NULL, new_sernum, NULL); 2051 } 2052 2053 /* 2054 * Garbage collection 2055 */ 2056 2057 static int fib6_age(struct fib6_info *rt, void *arg) 2058 { 2059 struct fib6_gc_args *gc_args = arg; 2060 unsigned long now = jiffies; 2061 2062 /* 2063 * check addrconf expiration here. 2064 * Routes are expired even if they are in use. 2065 */ 2066 2067 if (rt->fib6_flags & RTF_EXPIRES && rt->expires) { 2068 if (time_after(now, rt->expires)) { 2069 RT6_TRACE("expiring %p\n", rt); 2070 return -1; 2071 } 2072 gc_args->more++; 2073 } 2074 2075 /* Also age clones in the exception table. 2076 * Note, that clones are aged out 2077 * only if they are not in use now. 2078 */ 2079 rt6_age_exceptions(rt, gc_args, now); 2080 2081 return 0; 2082 } 2083 2084 void fib6_run_gc(unsigned long expires, struct net *net, bool force) 2085 { 2086 struct fib6_gc_args gc_args; 2087 unsigned long now; 2088 2089 if (force) { 2090 spin_lock_bh(&net->ipv6.fib6_gc_lock); 2091 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) { 2092 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); 2093 return; 2094 } 2095 gc_args.timeout = expires ? (int)expires : 2096 net->ipv6.sysctl.ip6_rt_gc_interval; 2097 gc_args.more = 0; 2098 2099 fib6_clean_all(net, fib6_age, &gc_args); 2100 now = jiffies; 2101 net->ipv6.ip6_rt_last_gc = now; 2102 2103 if (gc_args.more) 2104 mod_timer(&net->ipv6.ip6_fib_timer, 2105 round_jiffies(now 2106 + net->ipv6.sysctl.ip6_rt_gc_interval)); 2107 else 2108 del_timer(&net->ipv6.ip6_fib_timer); 2109 spin_unlock_bh(&net->ipv6.fib6_gc_lock); 2110 } 2111 2112 static void fib6_gc_timer_cb(struct timer_list *t) 2113 { 2114 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer); 2115 2116 fib6_run_gc(0, arg, true); 2117 } 2118 2119 static int __net_init fib6_net_init(struct net *net) 2120 { 2121 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; 2122 int err; 2123 2124 err = fib6_notifier_init(net); 2125 if (err) 2126 return err; 2127 2128 spin_lock_init(&net->ipv6.fib6_gc_lock); 2129 rwlock_init(&net->ipv6.fib6_walker_lock); 2130 INIT_LIST_HEAD(&net->ipv6.fib6_walkers); 2131 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0); 2132 2133 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); 2134 if (!net->ipv6.rt6_stats) 2135 goto out_timer; 2136 2137 /* Avoid false sharing : Use at least a full cache line */ 2138 size = max_t(size_t, size, L1_CACHE_BYTES); 2139 2140 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); 2141 if (!net->ipv6.fib_table_hash) 2142 goto out_rt6_stats; 2143 2144 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), 2145 GFP_KERNEL); 2146 if (!net->ipv6.fib6_main_tbl) 2147 goto out_fib_table_hash; 2148 2149 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; 2150 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf, 2151 net->ipv6.fib6_null_entry); 2152 net->ipv6.fib6_main_tbl->tb6_root.fn_flags = 2153 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 2154 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); 2155 2156 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 2157 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), 2158 GFP_KERNEL); 2159 if (!net->ipv6.fib6_local_tbl) 2160 goto out_fib6_main_tbl; 2161 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; 2162 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf, 2163 net->ipv6.fib6_null_entry); 2164 net->ipv6.fib6_local_tbl->tb6_root.fn_flags = 2165 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 2166 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); 2167 #endif 2168 fib6_tables_init(net); 2169 2170 return 0; 2171 2172 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 2173 out_fib6_main_tbl: 2174 kfree(net->ipv6.fib6_main_tbl); 2175 #endif 2176 out_fib_table_hash: 2177 kfree(net->ipv6.fib_table_hash); 2178 out_rt6_stats: 2179 kfree(net->ipv6.rt6_stats); 2180 out_timer: 2181 fib6_notifier_exit(net); 2182 return -ENOMEM; 2183 } 2184 2185 static void fib6_net_exit(struct net *net) 2186 { 2187 unsigned int i; 2188 2189 del_timer_sync(&net->ipv6.ip6_fib_timer); 2190 2191 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) { 2192 struct hlist_head *head = &net->ipv6.fib_table_hash[i]; 2193 struct hlist_node *tmp; 2194 struct fib6_table *tb; 2195 2196 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) { 2197 hlist_del(&tb->tb6_hlist); 2198 fib6_free_table(tb); 2199 } 2200 } 2201 2202 kfree(net->ipv6.fib_table_hash); 2203 kfree(net->ipv6.rt6_stats); 2204 fib6_notifier_exit(net); 2205 } 2206 2207 static struct pernet_operations fib6_net_ops = { 2208 .init = fib6_net_init, 2209 .exit = fib6_net_exit, 2210 }; 2211 2212 int __init fib6_init(void) 2213 { 2214 int ret = -ENOMEM; 2215 2216 fib6_node_kmem = kmem_cache_create("fib6_nodes", 2217 sizeof(struct fib6_node), 2218 0, SLAB_HWCACHE_ALIGN, 2219 NULL); 2220 if (!fib6_node_kmem) 2221 goto out; 2222 2223 ret = register_pernet_subsys(&fib6_net_ops); 2224 if (ret) 2225 goto out_kmem_cache_create; 2226 2227 ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL, 2228 inet6_dump_fib, 0); 2229 if (ret) 2230 goto out_unregister_subsys; 2231 2232 __fib6_flush_trees = fib6_flush_trees; 2233 out: 2234 return ret; 2235 2236 out_unregister_subsys: 2237 unregister_pernet_subsys(&fib6_net_ops); 2238 out_kmem_cache_create: 2239 kmem_cache_destroy(fib6_node_kmem); 2240 goto out; 2241 } 2242 2243 void fib6_gc_cleanup(void) 2244 { 2245 unregister_pernet_subsys(&fib6_net_ops); 2246 kmem_cache_destroy(fib6_node_kmem); 2247 } 2248 2249 #ifdef CONFIG_PROC_FS 2250 static int ipv6_route_seq_show(struct seq_file *seq, void *v) 2251 { 2252 struct fib6_info *rt = v; 2253 struct ipv6_route_iter *iter = seq->private; 2254 const struct net_device *dev; 2255 2256 seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen); 2257 2258 #ifdef CONFIG_IPV6_SUBTREES 2259 seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen); 2260 #else 2261 seq_puts(seq, "00000000000000000000000000000000 00 "); 2262 #endif 2263 if (rt->fib6_flags & RTF_GATEWAY) 2264 seq_printf(seq, "%pi6", &rt->fib6_nh.nh_gw); 2265 else 2266 seq_puts(seq, "00000000000000000000000000000000"); 2267 2268 dev = rt->fib6_nh.nh_dev; 2269 seq_printf(seq, " %08x %08x %08x %08x %8s\n", 2270 rt->fib6_metric, atomic_read(&rt->fib6_ref), 0, 2271 rt->fib6_flags, dev ? dev->name : ""); 2272 iter->w.leaf = NULL; 2273 return 0; 2274 } 2275 2276 static int ipv6_route_yield(struct fib6_walker *w) 2277 { 2278 struct ipv6_route_iter *iter = w->args; 2279 2280 if (!iter->skip) 2281 return 1; 2282 2283 do { 2284 iter->w.leaf = rcu_dereference_protected( 2285 iter->w.leaf->fib6_next, 2286 lockdep_is_held(&iter->tbl->tb6_lock)); 2287 iter->skip--; 2288 if (!iter->skip && iter->w.leaf) 2289 return 1; 2290 } while (iter->w.leaf); 2291 2292 return 0; 2293 } 2294 2295 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter, 2296 struct net *net) 2297 { 2298 memset(&iter->w, 0, sizeof(iter->w)); 2299 iter->w.func = ipv6_route_yield; 2300 iter->w.root = &iter->tbl->tb6_root; 2301 iter->w.state = FWS_INIT; 2302 iter->w.node = iter->w.root; 2303 iter->w.args = iter; 2304 iter->sernum = iter->w.root->fn_sernum; 2305 INIT_LIST_HEAD(&iter->w.lh); 2306 fib6_walker_link(net, &iter->w); 2307 } 2308 2309 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl, 2310 struct net *net) 2311 { 2312 unsigned int h; 2313 struct hlist_node *node; 2314 2315 if (tbl) { 2316 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1; 2317 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist)); 2318 } else { 2319 h = 0; 2320 node = NULL; 2321 } 2322 2323 while (!node && h < FIB6_TABLE_HASHSZ) { 2324 node = rcu_dereference_bh( 2325 hlist_first_rcu(&net->ipv6.fib_table_hash[h++])); 2326 } 2327 return hlist_entry_safe(node, struct fib6_table, tb6_hlist); 2328 } 2329 2330 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter) 2331 { 2332 if (iter->sernum != iter->w.root->fn_sernum) { 2333 iter->sernum = iter->w.root->fn_sernum; 2334 iter->w.state = FWS_INIT; 2335 iter->w.node = iter->w.root; 2336 WARN_ON(iter->w.skip); 2337 iter->w.skip = iter->w.count; 2338 } 2339 } 2340 2341 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2342 { 2343 int r; 2344 struct fib6_info *n; 2345 struct net *net = seq_file_net(seq); 2346 struct ipv6_route_iter *iter = seq->private; 2347 2348 if (!v) 2349 goto iter_table; 2350 2351 n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next); 2352 if (n) { 2353 ++*pos; 2354 return n; 2355 } 2356 2357 iter_table: 2358 ipv6_route_check_sernum(iter); 2359 spin_lock_bh(&iter->tbl->tb6_lock); 2360 r = fib6_walk_continue(&iter->w); 2361 spin_unlock_bh(&iter->tbl->tb6_lock); 2362 if (r > 0) { 2363 if (v) 2364 ++*pos; 2365 return iter->w.leaf; 2366 } else if (r < 0) { 2367 fib6_walker_unlink(net, &iter->w); 2368 return NULL; 2369 } 2370 fib6_walker_unlink(net, &iter->w); 2371 2372 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net); 2373 if (!iter->tbl) 2374 return NULL; 2375 2376 ipv6_route_seq_setup_walk(iter, net); 2377 goto iter_table; 2378 } 2379 2380 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos) 2381 __acquires(RCU_BH) 2382 { 2383 struct net *net = seq_file_net(seq); 2384 struct ipv6_route_iter *iter = seq->private; 2385 2386 rcu_read_lock_bh(); 2387 iter->tbl = ipv6_route_seq_next_table(NULL, net); 2388 iter->skip = *pos; 2389 2390 if (iter->tbl) { 2391 ipv6_route_seq_setup_walk(iter, net); 2392 return ipv6_route_seq_next(seq, NULL, pos); 2393 } else { 2394 return NULL; 2395 } 2396 } 2397 2398 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter) 2399 { 2400 struct fib6_walker *w = &iter->w; 2401 return w->node && !(w->state == FWS_U && w->node == w->root); 2402 } 2403 2404 static void ipv6_route_seq_stop(struct seq_file *seq, void *v) 2405 __releases(RCU_BH) 2406 { 2407 struct net *net = seq_file_net(seq); 2408 struct ipv6_route_iter *iter = seq->private; 2409 2410 if (ipv6_route_iter_active(iter)) 2411 fib6_walker_unlink(net, &iter->w); 2412 2413 rcu_read_unlock_bh(); 2414 } 2415 2416 const struct seq_operations ipv6_route_seq_ops = { 2417 .start = ipv6_route_seq_start, 2418 .next = ipv6_route_seq_next, 2419 .stop = ipv6_route_seq_stop, 2420 .show = ipv6_route_seq_show 2421 }; 2422 #endif /* CONFIG_PROC_FS */ 2423