1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Linux INET6 implementation 4 * Forwarding Information Database 5 * 6 * Authors: 7 * Pedro Roque <roque@di.fc.ul.pt> 8 * 9 * Changes: 10 * Yuji SEKIYA @USAGI: Support default route on router node; 11 * remove ip6_null_entry from the top of 12 * routing table. 13 * Ville Nuorvala: Fixed routing subtrees. 14 */ 15 16 #define pr_fmt(fmt) "IPv6: " fmt 17 18 #include <linux/bpf.h> 19 #include <linux/errno.h> 20 #include <linux/types.h> 21 #include <linux/net.h> 22 #include <linux/route.h> 23 #include <linux/netdevice.h> 24 #include <linux/in6.h> 25 #include <linux/init.h> 26 #include <linux/list.h> 27 #include <linux/slab.h> 28 29 #include <net/ip.h> 30 #include <net/ipv6.h> 31 #include <net/ndisc.h> 32 #include <net/addrconf.h> 33 #include <net/lwtunnel.h> 34 #include <net/fib_notifier.h> 35 36 #include <net/ip_fib.h> 37 #include <net/ip6_fib.h> 38 #include <net/ip6_route.h> 39 40 static struct kmem_cache *fib6_node_kmem __read_mostly; 41 42 struct fib6_cleaner { 43 struct fib6_walker w; 44 struct net *net; 45 int (*func)(struct fib6_info *, void *arg); 46 int sernum; 47 void *arg; 48 bool skip_notify; 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, bool with_fib6_nh) 149 { 150 struct fib6_info *f6i; 151 size_t sz = sizeof(*f6i); 152 153 if (with_fib6_nh) 154 sz += sizeof(struct fib6_nh); 155 156 f6i = kzalloc(sz, gfp_flags); 157 if (!f6i) 158 return NULL; 159 160 /* fib6_siblings is a union with nh_list, so this initializes both */ 161 INIT_LIST_HEAD(&f6i->fib6_siblings); 162 refcount_set(&f6i->fib6_ref, 1); 163 164 return f6i; 165 } 166 167 void fib6_info_destroy_rcu(struct rcu_head *head) 168 { 169 struct fib6_info *f6i = container_of(head, struct fib6_info, rcu); 170 171 WARN_ON(f6i->fib6_node); 172 173 if (f6i->nh) 174 nexthop_put(f6i->nh); 175 else 176 fib6_nh_release(f6i->fib6_nh); 177 178 ip_fib_metrics_put(f6i->fib6_metrics); 179 kfree(f6i); 180 } 181 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu); 182 183 static struct fib6_node *node_alloc(struct net *net) 184 { 185 struct fib6_node *fn; 186 187 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC); 188 if (fn) 189 net->ipv6.rt6_stats->fib_nodes++; 190 191 return fn; 192 } 193 194 static void node_free_immediate(struct net *net, struct fib6_node *fn) 195 { 196 kmem_cache_free(fib6_node_kmem, fn); 197 net->ipv6.rt6_stats->fib_nodes--; 198 } 199 200 static void node_free_rcu(struct rcu_head *head) 201 { 202 struct fib6_node *fn = container_of(head, struct fib6_node, rcu); 203 204 kmem_cache_free(fib6_node_kmem, fn); 205 } 206 207 static void node_free(struct net *net, struct fib6_node *fn) 208 { 209 call_rcu(&fn->rcu, node_free_rcu); 210 net->ipv6.rt6_stats->fib_nodes--; 211 } 212 213 static void fib6_free_table(struct fib6_table *table) 214 { 215 inetpeer_invalidate_tree(&table->tb6_peers); 216 kfree(table); 217 } 218 219 static void fib6_link_table(struct net *net, struct fib6_table *tb) 220 { 221 unsigned int h; 222 223 /* 224 * Initialize table lock at a single place to give lockdep a key, 225 * tables aren't visible prior to being linked to the list. 226 */ 227 spin_lock_init(&tb->tb6_lock); 228 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1); 229 230 /* 231 * No protection necessary, this is the only list mutatation 232 * operation, tables never disappear once they exist. 233 */ 234 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]); 235 } 236 237 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 238 239 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id) 240 { 241 struct fib6_table *table; 242 243 table = kzalloc(sizeof(*table), GFP_ATOMIC); 244 if (table) { 245 table->tb6_id = id; 246 rcu_assign_pointer(table->tb6_root.leaf, 247 net->ipv6.fib6_null_entry); 248 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 249 inet_peer_base_init(&table->tb6_peers); 250 } 251 252 return table; 253 } 254 255 struct fib6_table *fib6_new_table(struct net *net, u32 id) 256 { 257 struct fib6_table *tb; 258 259 if (id == 0) 260 id = RT6_TABLE_MAIN; 261 tb = fib6_get_table(net, id); 262 if (tb) 263 return tb; 264 265 tb = fib6_alloc_table(net, id); 266 if (tb) 267 fib6_link_table(net, tb); 268 269 return tb; 270 } 271 EXPORT_SYMBOL_GPL(fib6_new_table); 272 273 struct fib6_table *fib6_get_table(struct net *net, u32 id) 274 { 275 struct fib6_table *tb; 276 struct hlist_head *head; 277 unsigned int h; 278 279 if (id == 0) 280 id = RT6_TABLE_MAIN; 281 h = id & (FIB6_TABLE_HASHSZ - 1); 282 rcu_read_lock(); 283 head = &net->ipv6.fib_table_hash[h]; 284 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 285 if (tb->tb6_id == id) { 286 rcu_read_unlock(); 287 return tb; 288 } 289 } 290 rcu_read_unlock(); 291 292 return NULL; 293 } 294 EXPORT_SYMBOL_GPL(fib6_get_table); 295 296 static void __net_init fib6_tables_init(struct net *net) 297 { 298 fib6_link_table(net, net->ipv6.fib6_main_tbl); 299 fib6_link_table(net, net->ipv6.fib6_local_tbl); 300 } 301 #else 302 303 struct fib6_table *fib6_new_table(struct net *net, u32 id) 304 { 305 return fib6_get_table(net, id); 306 } 307 308 struct fib6_table *fib6_get_table(struct net *net, u32 id) 309 { 310 return net->ipv6.fib6_main_tbl; 311 } 312 313 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, 314 const struct sk_buff *skb, 315 int flags, pol_lookup_t lookup) 316 { 317 struct rt6_info *rt; 318 319 rt = pol_lookup_func(lookup, 320 net, net->ipv6.fib6_main_tbl, fl6, skb, flags); 321 if (rt->dst.error == -EAGAIN) { 322 ip6_rt_put_flags(rt, flags); 323 rt = net->ipv6.ip6_null_entry; 324 if (!(flags & RT6_LOOKUP_F_DST_NOREF)) 325 dst_hold(&rt->dst); 326 } 327 328 return &rt->dst; 329 } 330 331 /* called with rcu lock held; no reference taken on fib6_info */ 332 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6, 333 struct fib6_result *res, int flags) 334 { 335 return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6, 336 res, flags); 337 } 338 339 static void __net_init fib6_tables_init(struct net *net) 340 { 341 fib6_link_table(net, net->ipv6.fib6_main_tbl); 342 } 343 344 #endif 345 346 unsigned int fib6_tables_seq_read(struct net *net) 347 { 348 unsigned int h, fib_seq = 0; 349 350 rcu_read_lock(); 351 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 352 struct hlist_head *head = &net->ipv6.fib_table_hash[h]; 353 struct fib6_table *tb; 354 355 hlist_for_each_entry_rcu(tb, head, tb6_hlist) 356 fib_seq += tb->fib_seq; 357 } 358 rcu_read_unlock(); 359 360 return fib_seq; 361 } 362 363 static int call_fib6_entry_notifier(struct notifier_block *nb, 364 enum fib_event_type event_type, 365 struct fib6_info *rt, 366 struct netlink_ext_ack *extack) 367 { 368 struct fib6_entry_notifier_info info = { 369 .info.extack = extack, 370 .rt = rt, 371 }; 372 373 return call_fib6_notifier(nb, event_type, &info.info); 374 } 375 376 static int call_fib6_multipath_entry_notifier(struct notifier_block *nb, 377 enum fib_event_type event_type, 378 struct fib6_info *rt, 379 unsigned int nsiblings, 380 struct netlink_ext_ack *extack) 381 { 382 struct fib6_entry_notifier_info info = { 383 .info.extack = extack, 384 .rt = rt, 385 .nsiblings = nsiblings, 386 }; 387 388 return call_fib6_notifier(nb, event_type, &info.info); 389 } 390 391 int call_fib6_entry_notifiers(struct net *net, 392 enum fib_event_type event_type, 393 struct fib6_info *rt, 394 struct netlink_ext_ack *extack) 395 { 396 struct fib6_entry_notifier_info info = { 397 .info.extack = extack, 398 .rt = rt, 399 }; 400 401 rt->fib6_table->fib_seq++; 402 return call_fib6_notifiers(net, event_type, &info.info); 403 } 404 405 int call_fib6_multipath_entry_notifiers(struct net *net, 406 enum fib_event_type event_type, 407 struct fib6_info *rt, 408 unsigned int nsiblings, 409 struct netlink_ext_ack *extack) 410 { 411 struct fib6_entry_notifier_info info = { 412 .info.extack = extack, 413 .rt = rt, 414 .nsiblings = nsiblings, 415 }; 416 417 rt->fib6_table->fib_seq++; 418 return call_fib6_notifiers(net, event_type, &info.info); 419 } 420 421 int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt) 422 { 423 struct fib6_entry_notifier_info info = { 424 .rt = rt, 425 .nsiblings = rt->fib6_nsiblings, 426 }; 427 428 rt->fib6_table->fib_seq++; 429 return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE, &info.info); 430 } 431 432 struct fib6_dump_arg { 433 struct net *net; 434 struct notifier_block *nb; 435 struct netlink_ext_ack *extack; 436 }; 437 438 static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg) 439 { 440 enum fib_event_type fib_event = FIB_EVENT_ENTRY_REPLACE; 441 int err; 442 443 if (!rt || rt == arg->net->ipv6.fib6_null_entry) 444 return 0; 445 446 if (rt->fib6_nsiblings) 447 err = call_fib6_multipath_entry_notifier(arg->nb, fib_event, 448 rt, 449 rt->fib6_nsiblings, 450 arg->extack); 451 else 452 err = call_fib6_entry_notifier(arg->nb, fib_event, rt, 453 arg->extack); 454 455 return err; 456 } 457 458 static int fib6_node_dump(struct fib6_walker *w) 459 { 460 int err; 461 462 err = fib6_rt_dump(w->leaf, w->args); 463 w->leaf = NULL; 464 return err; 465 } 466 467 static int fib6_table_dump(struct net *net, struct fib6_table *tb, 468 struct fib6_walker *w) 469 { 470 int err; 471 472 w->root = &tb->tb6_root; 473 spin_lock_bh(&tb->tb6_lock); 474 err = fib6_walk(net, w); 475 spin_unlock_bh(&tb->tb6_lock); 476 return err; 477 } 478 479 /* Called with rcu_read_lock() */ 480 int fib6_tables_dump(struct net *net, struct notifier_block *nb, 481 struct netlink_ext_ack *extack) 482 { 483 struct fib6_dump_arg arg; 484 struct fib6_walker *w; 485 unsigned int h; 486 int err = 0; 487 488 w = kzalloc(sizeof(*w), GFP_ATOMIC); 489 if (!w) 490 return -ENOMEM; 491 492 w->func = fib6_node_dump; 493 arg.net = net; 494 arg.nb = nb; 495 arg.extack = extack; 496 w->args = &arg; 497 498 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 499 struct hlist_head *head = &net->ipv6.fib_table_hash[h]; 500 struct fib6_table *tb; 501 502 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 503 err = fib6_table_dump(net, tb, w); 504 if (err) 505 goto out; 506 } 507 } 508 509 out: 510 kfree(w); 511 512 /* The tree traversal function should never return a positive value. */ 513 return err > 0 ? -EINVAL : err; 514 } 515 516 static int fib6_dump_node(struct fib6_walker *w) 517 { 518 int res; 519 struct fib6_info *rt; 520 521 for_each_fib6_walker_rt(w) { 522 res = rt6_dump_route(rt, w->args, w->skip_in_node); 523 if (res >= 0) { 524 /* Frame is full, suspend walking */ 525 w->leaf = rt; 526 527 /* We'll restart from this node, so if some routes were 528 * already dumped, skip them next time. 529 */ 530 w->skip_in_node += res; 531 532 return 1; 533 } 534 w->skip_in_node = 0; 535 536 /* Multipath routes are dumped in one route with the 537 * RTA_MULTIPATH attribute. Jump 'rt' to point to the 538 * last sibling of this route (no need to dump the 539 * sibling routes again) 540 */ 541 if (rt->fib6_nsiblings) 542 rt = list_last_entry(&rt->fib6_siblings, 543 struct fib6_info, 544 fib6_siblings); 545 } 546 w->leaf = NULL; 547 return 0; 548 } 549 550 static void fib6_dump_end(struct netlink_callback *cb) 551 { 552 struct net *net = sock_net(cb->skb->sk); 553 struct fib6_walker *w = (void *)cb->args[2]; 554 555 if (w) { 556 if (cb->args[4]) { 557 cb->args[4] = 0; 558 fib6_walker_unlink(net, w); 559 } 560 cb->args[2] = 0; 561 kfree(w); 562 } 563 cb->done = (void *)cb->args[3]; 564 cb->args[1] = 3; 565 } 566 567 static int fib6_dump_done(struct netlink_callback *cb) 568 { 569 fib6_dump_end(cb); 570 return cb->done ? cb->done(cb) : 0; 571 } 572 573 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 574 struct netlink_callback *cb) 575 { 576 struct net *net = sock_net(skb->sk); 577 struct fib6_walker *w; 578 int res; 579 580 w = (void *)cb->args[2]; 581 w->root = &table->tb6_root; 582 583 if (cb->args[4] == 0) { 584 w->count = 0; 585 w->skip = 0; 586 w->skip_in_node = 0; 587 588 spin_lock_bh(&table->tb6_lock); 589 res = fib6_walk(net, w); 590 spin_unlock_bh(&table->tb6_lock); 591 if (res > 0) { 592 cb->args[4] = 1; 593 cb->args[5] = w->root->fn_sernum; 594 } 595 } else { 596 if (cb->args[5] != w->root->fn_sernum) { 597 /* Begin at the root if the tree changed */ 598 cb->args[5] = w->root->fn_sernum; 599 w->state = FWS_INIT; 600 w->node = w->root; 601 w->skip = w->count; 602 w->skip_in_node = 0; 603 } else 604 w->skip = 0; 605 606 spin_lock_bh(&table->tb6_lock); 607 res = fib6_walk_continue(w); 608 spin_unlock_bh(&table->tb6_lock); 609 if (res <= 0) { 610 fib6_walker_unlink(net, w); 611 cb->args[4] = 0; 612 } 613 } 614 615 return res; 616 } 617 618 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 619 { 620 struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true, 621 .filter.dump_routes = true }; 622 const struct nlmsghdr *nlh = cb->nlh; 623 struct net *net = sock_net(skb->sk); 624 unsigned int h, s_h; 625 unsigned int e = 0, s_e; 626 struct fib6_walker *w; 627 struct fib6_table *tb; 628 struct hlist_head *head; 629 int res = 0; 630 631 if (cb->strict_check) { 632 int err; 633 634 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb); 635 if (err < 0) 636 return err; 637 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) { 638 struct rtmsg *rtm = nlmsg_data(nlh); 639 640 if (rtm->rtm_flags & RTM_F_PREFIX) 641 arg.filter.flags = RTM_F_PREFIX; 642 } 643 644 w = (void *)cb->args[2]; 645 if (!w) { 646 /* New dump: 647 * 648 * 1. hook callback destructor. 649 */ 650 cb->args[3] = (long)cb->done; 651 cb->done = fib6_dump_done; 652 653 /* 654 * 2. allocate and initialize walker. 655 */ 656 w = kzalloc(sizeof(*w), GFP_ATOMIC); 657 if (!w) 658 return -ENOMEM; 659 w->func = fib6_dump_node; 660 cb->args[2] = (long)w; 661 } 662 663 arg.skb = skb; 664 arg.cb = cb; 665 arg.net = net; 666 w->args = &arg; 667 668 if (arg.filter.table_id) { 669 tb = fib6_get_table(net, arg.filter.table_id); 670 if (!tb) { 671 if (rtnl_msg_family(cb->nlh) != PF_INET6) 672 goto out; 673 674 NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist"); 675 return -ENOENT; 676 } 677 678 if (!cb->args[0]) { 679 res = fib6_dump_table(tb, skb, cb); 680 if (!res) 681 cb->args[0] = 1; 682 } 683 goto out; 684 } 685 686 s_h = cb->args[0]; 687 s_e = cb->args[1]; 688 689 rcu_read_lock(); 690 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) { 691 e = 0; 692 head = &net->ipv6.fib_table_hash[h]; 693 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 694 if (e < s_e) 695 goto next; 696 res = fib6_dump_table(tb, skb, cb); 697 if (res != 0) 698 goto out_unlock; 699 next: 700 e++; 701 } 702 } 703 out_unlock: 704 rcu_read_unlock(); 705 cb->args[1] = e; 706 cb->args[0] = h; 707 out: 708 res = res < 0 ? res : skb->len; 709 if (res <= 0) 710 fib6_dump_end(cb); 711 return res; 712 } 713 714 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val) 715 { 716 if (!f6i) 717 return; 718 719 if (f6i->fib6_metrics == &dst_default_metrics) { 720 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC); 721 722 if (!p) 723 return; 724 725 refcount_set(&p->refcnt, 1); 726 f6i->fib6_metrics = p; 727 } 728 729 f6i->fib6_metrics->metrics[metric - 1] = val; 730 } 731 732 /* 733 * Routing Table 734 * 735 * return the appropriate node for a routing tree "add" operation 736 * by either creating and inserting or by returning an existing 737 * node. 738 */ 739 740 static struct fib6_node *fib6_add_1(struct net *net, 741 struct fib6_table *table, 742 struct fib6_node *root, 743 struct in6_addr *addr, int plen, 744 int offset, int allow_create, 745 int replace_required, 746 struct netlink_ext_ack *extack) 747 { 748 struct fib6_node *fn, *in, *ln; 749 struct fib6_node *pn = NULL; 750 struct rt6key *key; 751 int bit; 752 __be32 dir = 0; 753 754 RT6_TRACE("fib6_add_1\n"); 755 756 /* insert node in tree */ 757 758 fn = root; 759 760 do { 761 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf, 762 lockdep_is_held(&table->tb6_lock)); 763 key = (struct rt6key *)((u8 *)leaf + offset); 764 765 /* 766 * Prefix match 767 */ 768 if (plen < fn->fn_bit || 769 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) { 770 if (!allow_create) { 771 if (replace_required) { 772 NL_SET_ERR_MSG(extack, 773 "Can not replace route - no match found"); 774 pr_warn("Can't replace route, no match found\n"); 775 return ERR_PTR(-ENOENT); 776 } 777 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 778 } 779 goto insert_above; 780 } 781 782 /* 783 * Exact match ? 784 */ 785 786 if (plen == fn->fn_bit) { 787 /* clean up an intermediate node */ 788 if (!(fn->fn_flags & RTN_RTINFO)) { 789 RCU_INIT_POINTER(fn->leaf, NULL); 790 fib6_info_release(leaf); 791 /* remove null_entry in the root node */ 792 } else if (fn->fn_flags & RTN_TL_ROOT && 793 rcu_access_pointer(fn->leaf) == 794 net->ipv6.fib6_null_entry) { 795 RCU_INIT_POINTER(fn->leaf, NULL); 796 } 797 798 return fn; 799 } 800 801 /* 802 * We have more bits to go 803 */ 804 805 /* Try to walk down on tree. */ 806 dir = addr_bit_set(addr, fn->fn_bit); 807 pn = fn; 808 fn = dir ? 809 rcu_dereference_protected(fn->right, 810 lockdep_is_held(&table->tb6_lock)) : 811 rcu_dereference_protected(fn->left, 812 lockdep_is_held(&table->tb6_lock)); 813 } while (fn); 814 815 if (!allow_create) { 816 /* We should not create new node because 817 * NLM_F_REPLACE was specified without NLM_F_CREATE 818 * I assume it is safe to require NLM_F_CREATE when 819 * REPLACE flag is used! Later we may want to remove the 820 * check for replace_required, because according 821 * to netlink specification, NLM_F_CREATE 822 * MUST be specified if new route is created. 823 * That would keep IPv6 consistent with IPv4 824 */ 825 if (replace_required) { 826 NL_SET_ERR_MSG(extack, 827 "Can not replace route - no match found"); 828 pr_warn("Can't replace route, no match found\n"); 829 return ERR_PTR(-ENOENT); 830 } 831 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 832 } 833 /* 834 * We walked to the bottom of tree. 835 * Create new leaf node without children. 836 */ 837 838 ln = node_alloc(net); 839 840 if (!ln) 841 return ERR_PTR(-ENOMEM); 842 ln->fn_bit = plen; 843 RCU_INIT_POINTER(ln->parent, pn); 844 845 if (dir) 846 rcu_assign_pointer(pn->right, ln); 847 else 848 rcu_assign_pointer(pn->left, ln); 849 850 return ln; 851 852 853 insert_above: 854 /* 855 * split since we don't have a common prefix anymore or 856 * we have a less significant route. 857 * we've to insert an intermediate node on the list 858 * this new node will point to the one we need to create 859 * and the current 860 */ 861 862 pn = rcu_dereference_protected(fn->parent, 863 lockdep_is_held(&table->tb6_lock)); 864 865 /* find 1st bit in difference between the 2 addrs. 866 867 See comment in __ipv6_addr_diff: bit may be an invalid value, 868 but if it is >= plen, the value is ignored in any case. 869 */ 870 871 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr)); 872 873 /* 874 * (intermediate)[in] 875 * / \ 876 * (new leaf node)[ln] (old node)[fn] 877 */ 878 if (plen > bit) { 879 in = node_alloc(net); 880 ln = node_alloc(net); 881 882 if (!in || !ln) { 883 if (in) 884 node_free_immediate(net, in); 885 if (ln) 886 node_free_immediate(net, ln); 887 return ERR_PTR(-ENOMEM); 888 } 889 890 /* 891 * new intermediate node. 892 * RTN_RTINFO will 893 * be off since that an address that chooses one of 894 * the branches would not match less specific routes 895 * in the other branch 896 */ 897 898 in->fn_bit = bit; 899 900 RCU_INIT_POINTER(in->parent, pn); 901 in->leaf = fn->leaf; 902 fib6_info_hold(rcu_dereference_protected(in->leaf, 903 lockdep_is_held(&table->tb6_lock))); 904 905 /* update parent pointer */ 906 if (dir) 907 rcu_assign_pointer(pn->right, in); 908 else 909 rcu_assign_pointer(pn->left, in); 910 911 ln->fn_bit = plen; 912 913 RCU_INIT_POINTER(ln->parent, in); 914 rcu_assign_pointer(fn->parent, in); 915 916 if (addr_bit_set(addr, bit)) { 917 rcu_assign_pointer(in->right, ln); 918 rcu_assign_pointer(in->left, fn); 919 } else { 920 rcu_assign_pointer(in->left, ln); 921 rcu_assign_pointer(in->right, fn); 922 } 923 } else { /* plen <= bit */ 924 925 /* 926 * (new leaf node)[ln] 927 * / \ 928 * (old node)[fn] NULL 929 */ 930 931 ln = node_alloc(net); 932 933 if (!ln) 934 return ERR_PTR(-ENOMEM); 935 936 ln->fn_bit = plen; 937 938 RCU_INIT_POINTER(ln->parent, pn); 939 940 if (addr_bit_set(&key->addr, plen)) 941 RCU_INIT_POINTER(ln->right, fn); 942 else 943 RCU_INIT_POINTER(ln->left, fn); 944 945 rcu_assign_pointer(fn->parent, ln); 946 947 if (dir) 948 rcu_assign_pointer(pn->right, ln); 949 else 950 rcu_assign_pointer(pn->left, ln); 951 } 952 return ln; 953 } 954 955 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh, 956 const struct fib6_info *match, 957 const struct fib6_table *table) 958 { 959 int cpu; 960 961 if (!fib6_nh->rt6i_pcpu) 962 return; 963 964 /* release the reference to this fib entry from 965 * all of its cached pcpu routes 966 */ 967 for_each_possible_cpu(cpu) { 968 struct rt6_info **ppcpu_rt; 969 struct rt6_info *pcpu_rt; 970 971 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu); 972 pcpu_rt = *ppcpu_rt; 973 974 /* only dropping the 'from' reference if the cached route 975 * is using 'match'. The cached pcpu_rt->from only changes 976 * from a fib6_info to NULL (ip6_dst_destroy); it can never 977 * change from one fib6_info reference to another 978 */ 979 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) { 980 struct fib6_info *from; 981 982 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL); 983 fib6_info_release(from); 984 } 985 } 986 } 987 988 struct fib6_nh_pcpu_arg { 989 struct fib6_info *from; 990 const struct fib6_table *table; 991 }; 992 993 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg) 994 { 995 struct fib6_nh_pcpu_arg *arg = _arg; 996 997 __fib6_drop_pcpu_from(nh, arg->from, arg->table); 998 return 0; 999 } 1000 1001 static void fib6_drop_pcpu_from(struct fib6_info *f6i, 1002 const struct fib6_table *table) 1003 { 1004 /* Make sure rt6_make_pcpu_route() wont add other percpu routes 1005 * while we are cleaning them here. 1006 */ 1007 f6i->fib6_destroying = 1; 1008 mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */ 1009 1010 if (f6i->nh) { 1011 struct fib6_nh_pcpu_arg arg = { 1012 .from = f6i, 1013 .table = table 1014 }; 1015 1016 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from, 1017 &arg); 1018 } else { 1019 struct fib6_nh *fib6_nh; 1020 1021 fib6_nh = f6i->fib6_nh; 1022 __fib6_drop_pcpu_from(fib6_nh, f6i, table); 1023 } 1024 } 1025 1026 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn, 1027 struct net *net) 1028 { 1029 struct fib6_table *table = rt->fib6_table; 1030 1031 /* Flush all cached dst in exception table */ 1032 rt6_flush_exceptions(rt); 1033 fib6_drop_pcpu_from(rt, table); 1034 1035 if (rt->nh && !list_empty(&rt->nh_list)) 1036 list_del_init(&rt->nh_list); 1037 1038 if (refcount_read(&rt->fib6_ref) != 1) { 1039 /* This route is used as dummy address holder in some split 1040 * nodes. It is not leaked, but it still holds other resources, 1041 * which must be released in time. So, scan ascendant nodes 1042 * and replace dummy references to this route with references 1043 * to still alive ones. 1044 */ 1045 while (fn) { 1046 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf, 1047 lockdep_is_held(&table->tb6_lock)); 1048 struct fib6_info *new_leaf; 1049 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) { 1050 new_leaf = fib6_find_prefix(net, table, fn); 1051 fib6_info_hold(new_leaf); 1052 1053 rcu_assign_pointer(fn->leaf, new_leaf); 1054 fib6_info_release(rt); 1055 } 1056 fn = rcu_dereference_protected(fn->parent, 1057 lockdep_is_held(&table->tb6_lock)); 1058 } 1059 } 1060 } 1061 1062 /* 1063 * Insert routing information in a node. 1064 */ 1065 1066 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt, 1067 struct nl_info *info, 1068 struct netlink_ext_ack *extack) 1069 { 1070 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf, 1071 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1072 struct fib6_info *iter = NULL; 1073 struct fib6_info __rcu **ins; 1074 struct fib6_info __rcu **fallback_ins = NULL; 1075 int replace = (info->nlh && 1076 (info->nlh->nlmsg_flags & NLM_F_REPLACE)); 1077 int add = (!info->nlh || 1078 (info->nlh->nlmsg_flags & NLM_F_CREATE)); 1079 int found = 0; 1080 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt); 1081 bool notify_sibling_rt = false; 1082 u16 nlflags = NLM_F_EXCL; 1083 int err; 1084 1085 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND)) 1086 nlflags |= NLM_F_APPEND; 1087 1088 ins = &fn->leaf; 1089 1090 for (iter = leaf; iter; 1091 iter = rcu_dereference_protected(iter->fib6_next, 1092 lockdep_is_held(&rt->fib6_table->tb6_lock))) { 1093 /* 1094 * Search for duplicates 1095 */ 1096 1097 if (iter->fib6_metric == rt->fib6_metric) { 1098 /* 1099 * Same priority level 1100 */ 1101 if (info->nlh && 1102 (info->nlh->nlmsg_flags & NLM_F_EXCL)) 1103 return -EEXIST; 1104 1105 nlflags &= ~NLM_F_EXCL; 1106 if (replace) { 1107 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) { 1108 found++; 1109 break; 1110 } 1111 fallback_ins = fallback_ins ?: ins; 1112 goto next_iter; 1113 } 1114 1115 if (rt6_duplicate_nexthop(iter, rt)) { 1116 if (rt->fib6_nsiblings) 1117 rt->fib6_nsiblings = 0; 1118 if (!(iter->fib6_flags & RTF_EXPIRES)) 1119 return -EEXIST; 1120 if (!(rt->fib6_flags & RTF_EXPIRES)) 1121 fib6_clean_expires(iter); 1122 else 1123 fib6_set_expires(iter, rt->expires); 1124 1125 if (rt->fib6_pmtu) 1126 fib6_metric_set(iter, RTAX_MTU, 1127 rt->fib6_pmtu); 1128 return -EEXIST; 1129 } 1130 /* If we have the same destination and the same metric, 1131 * but not the same gateway, then the route we try to 1132 * add is sibling to this route, increment our counter 1133 * of siblings, and later we will add our route to the 1134 * list. 1135 * Only static routes (which don't have flag 1136 * RTF_EXPIRES) are used for ECMPv6. 1137 * 1138 * To avoid long list, we only had siblings if the 1139 * route have a gateway. 1140 */ 1141 if (rt_can_ecmp && 1142 rt6_qualify_for_ecmp(iter)) 1143 rt->fib6_nsiblings++; 1144 } 1145 1146 if (iter->fib6_metric > rt->fib6_metric) 1147 break; 1148 1149 next_iter: 1150 ins = &iter->fib6_next; 1151 } 1152 1153 if (fallback_ins && !found) { 1154 /* No matching route with same ecmp-able-ness found, replace 1155 * first matching route 1156 */ 1157 ins = fallback_ins; 1158 iter = rcu_dereference_protected(*ins, 1159 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1160 found++; 1161 } 1162 1163 /* Reset round-robin state, if necessary */ 1164 if (ins == &fn->leaf) 1165 fn->rr_ptr = NULL; 1166 1167 /* Link this route to others same route. */ 1168 if (rt->fib6_nsiblings) { 1169 unsigned int fib6_nsiblings; 1170 struct fib6_info *sibling, *temp_sibling; 1171 1172 /* Find the first route that have the same metric */ 1173 sibling = leaf; 1174 notify_sibling_rt = true; 1175 while (sibling) { 1176 if (sibling->fib6_metric == rt->fib6_metric && 1177 rt6_qualify_for_ecmp(sibling)) { 1178 list_add_tail(&rt->fib6_siblings, 1179 &sibling->fib6_siblings); 1180 break; 1181 } 1182 sibling = rcu_dereference_protected(sibling->fib6_next, 1183 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1184 notify_sibling_rt = false; 1185 } 1186 /* For each sibling in the list, increment the counter of 1187 * siblings. BUG() if counters does not match, list of siblings 1188 * is broken! 1189 */ 1190 fib6_nsiblings = 0; 1191 list_for_each_entry_safe(sibling, temp_sibling, 1192 &rt->fib6_siblings, fib6_siblings) { 1193 sibling->fib6_nsiblings++; 1194 BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings); 1195 fib6_nsiblings++; 1196 } 1197 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings); 1198 rt6_multipath_rebalance(temp_sibling); 1199 } 1200 1201 /* 1202 * insert node 1203 */ 1204 if (!replace) { 1205 if (!add) 1206 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 1207 1208 add: 1209 nlflags |= NLM_F_CREATE; 1210 1211 /* The route should only be notified if it is the first 1212 * route in the node or if it is added as a sibling 1213 * route to the first route in the node. 1214 */ 1215 if (!info->skip_notify_kernel && 1216 (notify_sibling_rt || ins == &fn->leaf)) { 1217 enum fib_event_type fib_event; 1218 1219 if (notify_sibling_rt) 1220 fib_event = FIB_EVENT_ENTRY_APPEND; 1221 else 1222 fib_event = FIB_EVENT_ENTRY_REPLACE; 1223 err = call_fib6_entry_notifiers(info->nl_net, 1224 fib_event, rt, 1225 extack); 1226 if (err) { 1227 struct fib6_info *sibling, *next_sibling; 1228 1229 /* If the route has siblings, then it first 1230 * needs to be unlinked from them. 1231 */ 1232 if (!rt->fib6_nsiblings) 1233 return err; 1234 1235 list_for_each_entry_safe(sibling, next_sibling, 1236 &rt->fib6_siblings, 1237 fib6_siblings) 1238 sibling->fib6_nsiblings--; 1239 rt->fib6_nsiblings = 0; 1240 list_del_init(&rt->fib6_siblings); 1241 rt6_multipath_rebalance(next_sibling); 1242 return err; 1243 } 1244 } 1245 1246 rcu_assign_pointer(rt->fib6_next, iter); 1247 fib6_info_hold(rt); 1248 rcu_assign_pointer(rt->fib6_node, fn); 1249 rcu_assign_pointer(*ins, rt); 1250 if (!info->skip_notify) 1251 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags); 1252 info->nl_net->ipv6.rt6_stats->fib_rt_entries++; 1253 1254 if (!(fn->fn_flags & RTN_RTINFO)) { 1255 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 1256 fn->fn_flags |= RTN_RTINFO; 1257 } 1258 1259 } else { 1260 int nsiblings; 1261 1262 if (!found) { 1263 if (add) 1264 goto add; 1265 pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); 1266 return -ENOENT; 1267 } 1268 1269 if (!info->skip_notify_kernel && ins == &fn->leaf) { 1270 err = call_fib6_entry_notifiers(info->nl_net, 1271 FIB_EVENT_ENTRY_REPLACE, 1272 rt, extack); 1273 if (err) 1274 return err; 1275 } 1276 1277 fib6_info_hold(rt); 1278 rcu_assign_pointer(rt->fib6_node, fn); 1279 rt->fib6_next = iter->fib6_next; 1280 rcu_assign_pointer(*ins, rt); 1281 if (!info->skip_notify) 1282 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE); 1283 if (!(fn->fn_flags & RTN_RTINFO)) { 1284 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 1285 fn->fn_flags |= RTN_RTINFO; 1286 } 1287 nsiblings = iter->fib6_nsiblings; 1288 iter->fib6_node = NULL; 1289 fib6_purge_rt(iter, fn, info->nl_net); 1290 if (rcu_access_pointer(fn->rr_ptr) == iter) 1291 fn->rr_ptr = NULL; 1292 fib6_info_release(iter); 1293 1294 if (nsiblings) { 1295 /* Replacing an ECMP route, remove all siblings */ 1296 ins = &rt->fib6_next; 1297 iter = rcu_dereference_protected(*ins, 1298 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1299 while (iter) { 1300 if (iter->fib6_metric > rt->fib6_metric) 1301 break; 1302 if (rt6_qualify_for_ecmp(iter)) { 1303 *ins = iter->fib6_next; 1304 iter->fib6_node = NULL; 1305 fib6_purge_rt(iter, fn, info->nl_net); 1306 if (rcu_access_pointer(fn->rr_ptr) == iter) 1307 fn->rr_ptr = NULL; 1308 fib6_info_release(iter); 1309 nsiblings--; 1310 info->nl_net->ipv6.rt6_stats->fib_rt_entries--; 1311 } else { 1312 ins = &iter->fib6_next; 1313 } 1314 iter = rcu_dereference_protected(*ins, 1315 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1316 } 1317 WARN_ON(nsiblings != 0); 1318 } 1319 } 1320 1321 return 0; 1322 } 1323 1324 static void fib6_start_gc(struct net *net, struct fib6_info *rt) 1325 { 1326 if (!timer_pending(&net->ipv6.ip6_fib_timer) && 1327 (rt->fib6_flags & RTF_EXPIRES)) 1328 mod_timer(&net->ipv6.ip6_fib_timer, 1329 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 1330 } 1331 1332 void fib6_force_start_gc(struct net *net) 1333 { 1334 if (!timer_pending(&net->ipv6.ip6_fib_timer)) 1335 mod_timer(&net->ipv6.ip6_fib_timer, 1336 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 1337 } 1338 1339 static void __fib6_update_sernum_upto_root(struct fib6_info *rt, 1340 int sernum) 1341 { 1342 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node, 1343 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1344 1345 /* paired with smp_rmb() in fib6_get_cookie_safe() */ 1346 smp_wmb(); 1347 while (fn) { 1348 fn->fn_sernum = sernum; 1349 fn = rcu_dereference_protected(fn->parent, 1350 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1351 } 1352 } 1353 1354 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt) 1355 { 1356 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net)); 1357 } 1358 1359 /* allow ipv4 to update sernum via ipv6_stub */ 1360 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i) 1361 { 1362 spin_lock_bh(&f6i->fib6_table->tb6_lock); 1363 fib6_update_sernum_upto_root(net, f6i); 1364 spin_unlock_bh(&f6i->fib6_table->tb6_lock); 1365 } 1366 1367 /* 1368 * Add routing information to the routing tree. 1369 * <destination addr>/<source addr> 1370 * with source addr info in sub-trees 1371 * Need to own table->tb6_lock 1372 */ 1373 1374 int fib6_add(struct fib6_node *root, struct fib6_info *rt, 1375 struct nl_info *info, struct netlink_ext_ack *extack) 1376 { 1377 struct fib6_table *table = rt->fib6_table; 1378 struct fib6_node *fn, *pn = NULL; 1379 int err = -ENOMEM; 1380 int allow_create = 1; 1381 int replace_required = 0; 1382 1383 if (info->nlh) { 1384 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) 1385 allow_create = 0; 1386 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) 1387 replace_required = 1; 1388 } 1389 if (!allow_create && !replace_required) 1390 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); 1391 1392 fn = fib6_add_1(info->nl_net, table, root, 1393 &rt->fib6_dst.addr, rt->fib6_dst.plen, 1394 offsetof(struct fib6_info, fib6_dst), allow_create, 1395 replace_required, extack); 1396 if (IS_ERR(fn)) { 1397 err = PTR_ERR(fn); 1398 fn = NULL; 1399 goto out; 1400 } 1401 1402 pn = fn; 1403 1404 #ifdef CONFIG_IPV6_SUBTREES 1405 if (rt->fib6_src.plen) { 1406 struct fib6_node *sn; 1407 1408 if (!rcu_access_pointer(fn->subtree)) { 1409 struct fib6_node *sfn; 1410 1411 /* 1412 * Create subtree. 1413 * 1414 * fn[main tree] 1415 * | 1416 * sfn[subtree root] 1417 * \ 1418 * sn[new leaf node] 1419 */ 1420 1421 /* Create subtree root node */ 1422 sfn = node_alloc(info->nl_net); 1423 if (!sfn) 1424 goto failure; 1425 1426 fib6_info_hold(info->nl_net->ipv6.fib6_null_entry); 1427 rcu_assign_pointer(sfn->leaf, 1428 info->nl_net->ipv6.fib6_null_entry); 1429 sfn->fn_flags = RTN_ROOT; 1430 1431 /* Now add the first leaf node to new subtree */ 1432 1433 sn = fib6_add_1(info->nl_net, table, sfn, 1434 &rt->fib6_src.addr, rt->fib6_src.plen, 1435 offsetof(struct fib6_info, fib6_src), 1436 allow_create, replace_required, extack); 1437 1438 if (IS_ERR(sn)) { 1439 /* If it is failed, discard just allocated 1440 root, and then (in failure) stale node 1441 in main tree. 1442 */ 1443 node_free_immediate(info->nl_net, sfn); 1444 err = PTR_ERR(sn); 1445 goto failure; 1446 } 1447 1448 /* Now link new subtree to main tree */ 1449 rcu_assign_pointer(sfn->parent, fn); 1450 rcu_assign_pointer(fn->subtree, sfn); 1451 } else { 1452 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn), 1453 &rt->fib6_src.addr, rt->fib6_src.plen, 1454 offsetof(struct fib6_info, fib6_src), 1455 allow_create, replace_required, extack); 1456 1457 if (IS_ERR(sn)) { 1458 err = PTR_ERR(sn); 1459 goto failure; 1460 } 1461 } 1462 1463 if (!rcu_access_pointer(fn->leaf)) { 1464 if (fn->fn_flags & RTN_TL_ROOT) { 1465 /* put back null_entry for root node */ 1466 rcu_assign_pointer(fn->leaf, 1467 info->nl_net->ipv6.fib6_null_entry); 1468 } else { 1469 fib6_info_hold(rt); 1470 rcu_assign_pointer(fn->leaf, rt); 1471 } 1472 } 1473 fn = sn; 1474 } 1475 #endif 1476 1477 err = fib6_add_rt2node(fn, rt, info, extack); 1478 if (!err) { 1479 if (rt->nh) 1480 list_add(&rt->nh_list, &rt->nh->f6i_list); 1481 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(info->nl_net)); 1482 fib6_start_gc(info->nl_net, rt); 1483 } 1484 1485 out: 1486 if (err) { 1487 #ifdef CONFIG_IPV6_SUBTREES 1488 /* 1489 * If fib6_add_1 has cleared the old leaf pointer in the 1490 * super-tree leaf node we have to find a new one for it. 1491 */ 1492 if (pn != fn) { 1493 struct fib6_info *pn_leaf = 1494 rcu_dereference_protected(pn->leaf, 1495 lockdep_is_held(&table->tb6_lock)); 1496 if (pn_leaf == rt) { 1497 pn_leaf = NULL; 1498 RCU_INIT_POINTER(pn->leaf, NULL); 1499 fib6_info_release(rt); 1500 } 1501 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) { 1502 pn_leaf = fib6_find_prefix(info->nl_net, table, 1503 pn); 1504 #if RT6_DEBUG >= 2 1505 if (!pn_leaf) { 1506 WARN_ON(!pn_leaf); 1507 pn_leaf = 1508 info->nl_net->ipv6.fib6_null_entry; 1509 } 1510 #endif 1511 fib6_info_hold(pn_leaf); 1512 rcu_assign_pointer(pn->leaf, pn_leaf); 1513 } 1514 } 1515 #endif 1516 goto failure; 1517 } else if (fib6_requires_src(rt)) { 1518 fib6_routes_require_src_inc(info->nl_net); 1519 } 1520 return err; 1521 1522 failure: 1523 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if: 1524 * 1. fn is an intermediate node and we failed to add the new 1525 * route to it in both subtree creation failure and fib6_add_rt2node() 1526 * failure case. 1527 * 2. fn is the root node in the table and we fail to add the first 1528 * default route to it. 1529 */ 1530 if (fn && 1531 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) || 1532 (fn->fn_flags & RTN_TL_ROOT && 1533 !rcu_access_pointer(fn->leaf)))) 1534 fib6_repair_tree(info->nl_net, table, fn); 1535 return err; 1536 } 1537 1538 /* 1539 * Routing tree lookup 1540 * 1541 */ 1542 1543 struct lookup_args { 1544 int offset; /* key offset on fib6_info */ 1545 const struct in6_addr *addr; /* search key */ 1546 }; 1547 1548 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root, 1549 struct lookup_args *args) 1550 { 1551 struct fib6_node *fn; 1552 __be32 dir; 1553 1554 if (unlikely(args->offset == 0)) 1555 return NULL; 1556 1557 /* 1558 * Descend on a tree 1559 */ 1560 1561 fn = root; 1562 1563 for (;;) { 1564 struct fib6_node *next; 1565 1566 dir = addr_bit_set(args->addr, fn->fn_bit); 1567 1568 next = dir ? rcu_dereference(fn->right) : 1569 rcu_dereference(fn->left); 1570 1571 if (next) { 1572 fn = next; 1573 continue; 1574 } 1575 break; 1576 } 1577 1578 while (fn) { 1579 struct fib6_node *subtree = FIB6_SUBTREE(fn); 1580 1581 if (subtree || fn->fn_flags & RTN_RTINFO) { 1582 struct fib6_info *leaf = rcu_dereference(fn->leaf); 1583 struct rt6key *key; 1584 1585 if (!leaf) 1586 goto backtrack; 1587 1588 key = (struct rt6key *) ((u8 *)leaf + args->offset); 1589 1590 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 1591 #ifdef CONFIG_IPV6_SUBTREES 1592 if (subtree) { 1593 struct fib6_node *sfn; 1594 sfn = fib6_node_lookup_1(subtree, 1595 args + 1); 1596 if (!sfn) 1597 goto backtrack; 1598 fn = sfn; 1599 } 1600 #endif 1601 if (fn->fn_flags & RTN_RTINFO) 1602 return fn; 1603 } 1604 } 1605 backtrack: 1606 if (fn->fn_flags & RTN_ROOT) 1607 break; 1608 1609 fn = rcu_dereference(fn->parent); 1610 } 1611 1612 return NULL; 1613 } 1614 1615 /* called with rcu_read_lock() held 1616 */ 1617 struct fib6_node *fib6_node_lookup(struct fib6_node *root, 1618 const struct in6_addr *daddr, 1619 const struct in6_addr *saddr) 1620 { 1621 struct fib6_node *fn; 1622 struct lookup_args args[] = { 1623 { 1624 .offset = offsetof(struct fib6_info, fib6_dst), 1625 .addr = daddr, 1626 }, 1627 #ifdef CONFIG_IPV6_SUBTREES 1628 { 1629 .offset = offsetof(struct fib6_info, fib6_src), 1630 .addr = saddr, 1631 }, 1632 #endif 1633 { 1634 .offset = 0, /* sentinel */ 1635 } 1636 }; 1637 1638 fn = fib6_node_lookup_1(root, daddr ? args : args + 1); 1639 if (!fn || fn->fn_flags & RTN_TL_ROOT) 1640 fn = root; 1641 1642 return fn; 1643 } 1644 1645 /* 1646 * Get node with specified destination prefix (and source prefix, 1647 * if subtrees are used) 1648 * exact_match == true means we try to find fn with exact match of 1649 * the passed in prefix addr 1650 * exact_match == false means we try to find fn with longest prefix 1651 * match of the passed in prefix addr. This is useful for finding fn 1652 * for cached route as it will be stored in the exception table under 1653 * the node with longest prefix length. 1654 */ 1655 1656 1657 static struct fib6_node *fib6_locate_1(struct fib6_node *root, 1658 const struct in6_addr *addr, 1659 int plen, int offset, 1660 bool exact_match) 1661 { 1662 struct fib6_node *fn, *prev = NULL; 1663 1664 for (fn = root; fn ; ) { 1665 struct fib6_info *leaf = rcu_dereference(fn->leaf); 1666 struct rt6key *key; 1667 1668 /* This node is being deleted */ 1669 if (!leaf) { 1670 if (plen <= fn->fn_bit) 1671 goto out; 1672 else 1673 goto next; 1674 } 1675 1676 key = (struct rt6key *)((u8 *)leaf + offset); 1677 1678 /* 1679 * Prefix match 1680 */ 1681 if (plen < fn->fn_bit || 1682 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 1683 goto out; 1684 1685 if (plen == fn->fn_bit) 1686 return fn; 1687 1688 if (fn->fn_flags & RTN_RTINFO) 1689 prev = fn; 1690 1691 next: 1692 /* 1693 * We have more bits to go 1694 */ 1695 if (addr_bit_set(addr, fn->fn_bit)) 1696 fn = rcu_dereference(fn->right); 1697 else 1698 fn = rcu_dereference(fn->left); 1699 } 1700 out: 1701 if (exact_match) 1702 return NULL; 1703 else 1704 return prev; 1705 } 1706 1707 struct fib6_node *fib6_locate(struct fib6_node *root, 1708 const struct in6_addr *daddr, int dst_len, 1709 const struct in6_addr *saddr, int src_len, 1710 bool exact_match) 1711 { 1712 struct fib6_node *fn; 1713 1714 fn = fib6_locate_1(root, daddr, dst_len, 1715 offsetof(struct fib6_info, fib6_dst), 1716 exact_match); 1717 1718 #ifdef CONFIG_IPV6_SUBTREES 1719 if (src_len) { 1720 WARN_ON(saddr == NULL); 1721 if (fn) { 1722 struct fib6_node *subtree = FIB6_SUBTREE(fn); 1723 1724 if (subtree) { 1725 fn = fib6_locate_1(subtree, saddr, src_len, 1726 offsetof(struct fib6_info, fib6_src), 1727 exact_match); 1728 } 1729 } 1730 } 1731 #endif 1732 1733 if (fn && fn->fn_flags & RTN_RTINFO) 1734 return fn; 1735 1736 return NULL; 1737 } 1738 1739 1740 /* 1741 * Deletion 1742 * 1743 */ 1744 1745 static struct fib6_info *fib6_find_prefix(struct net *net, 1746 struct fib6_table *table, 1747 struct fib6_node *fn) 1748 { 1749 struct fib6_node *child_left, *child_right; 1750 1751 if (fn->fn_flags & RTN_ROOT) 1752 return net->ipv6.fib6_null_entry; 1753 1754 while (fn) { 1755 child_left = rcu_dereference_protected(fn->left, 1756 lockdep_is_held(&table->tb6_lock)); 1757 child_right = rcu_dereference_protected(fn->right, 1758 lockdep_is_held(&table->tb6_lock)); 1759 if (child_left) 1760 return rcu_dereference_protected(child_left->leaf, 1761 lockdep_is_held(&table->tb6_lock)); 1762 if (child_right) 1763 return rcu_dereference_protected(child_right->leaf, 1764 lockdep_is_held(&table->tb6_lock)); 1765 1766 fn = FIB6_SUBTREE(fn); 1767 } 1768 return NULL; 1769 } 1770 1771 /* 1772 * Called to trim the tree of intermediate nodes when possible. "fn" 1773 * is the node we want to try and remove. 1774 * Need to own table->tb6_lock 1775 */ 1776 1777 static struct fib6_node *fib6_repair_tree(struct net *net, 1778 struct fib6_table *table, 1779 struct fib6_node *fn) 1780 { 1781 int children; 1782 int nstate; 1783 struct fib6_node *child; 1784 struct fib6_walker *w; 1785 int iter = 0; 1786 1787 /* Set fn->leaf to null_entry for root node. */ 1788 if (fn->fn_flags & RTN_TL_ROOT) { 1789 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry); 1790 return fn; 1791 } 1792 1793 for (;;) { 1794 struct fib6_node *fn_r = rcu_dereference_protected(fn->right, 1795 lockdep_is_held(&table->tb6_lock)); 1796 struct fib6_node *fn_l = rcu_dereference_protected(fn->left, 1797 lockdep_is_held(&table->tb6_lock)); 1798 struct fib6_node *pn = rcu_dereference_protected(fn->parent, 1799 lockdep_is_held(&table->tb6_lock)); 1800 struct fib6_node *pn_r = rcu_dereference_protected(pn->right, 1801 lockdep_is_held(&table->tb6_lock)); 1802 struct fib6_node *pn_l = rcu_dereference_protected(pn->left, 1803 lockdep_is_held(&table->tb6_lock)); 1804 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf, 1805 lockdep_is_held(&table->tb6_lock)); 1806 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf, 1807 lockdep_is_held(&table->tb6_lock)); 1808 struct fib6_info *new_fn_leaf; 1809 1810 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1811 iter++; 1812 1813 WARN_ON(fn->fn_flags & RTN_RTINFO); 1814 WARN_ON(fn->fn_flags & RTN_TL_ROOT); 1815 WARN_ON(fn_leaf); 1816 1817 children = 0; 1818 child = NULL; 1819 if (fn_r) { 1820 child = fn_r; 1821 children |= 1; 1822 } 1823 if (fn_l) { 1824 child = fn_l; 1825 children |= 2; 1826 } 1827 1828 if (children == 3 || FIB6_SUBTREE(fn) 1829 #ifdef CONFIG_IPV6_SUBTREES 1830 /* Subtree root (i.e. fn) may have one child */ 1831 || (children && fn->fn_flags & RTN_ROOT) 1832 #endif 1833 ) { 1834 new_fn_leaf = fib6_find_prefix(net, table, fn); 1835 #if RT6_DEBUG >= 2 1836 if (!new_fn_leaf) { 1837 WARN_ON(!new_fn_leaf); 1838 new_fn_leaf = net->ipv6.fib6_null_entry; 1839 } 1840 #endif 1841 fib6_info_hold(new_fn_leaf); 1842 rcu_assign_pointer(fn->leaf, new_fn_leaf); 1843 return pn; 1844 } 1845 1846 #ifdef CONFIG_IPV6_SUBTREES 1847 if (FIB6_SUBTREE(pn) == fn) { 1848 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1849 RCU_INIT_POINTER(pn->subtree, NULL); 1850 nstate = FWS_L; 1851 } else { 1852 WARN_ON(fn->fn_flags & RTN_ROOT); 1853 #endif 1854 if (pn_r == fn) 1855 rcu_assign_pointer(pn->right, child); 1856 else if (pn_l == fn) 1857 rcu_assign_pointer(pn->left, child); 1858 #if RT6_DEBUG >= 2 1859 else 1860 WARN_ON(1); 1861 #endif 1862 if (child) 1863 rcu_assign_pointer(child->parent, pn); 1864 nstate = FWS_R; 1865 #ifdef CONFIG_IPV6_SUBTREES 1866 } 1867 #endif 1868 1869 read_lock(&net->ipv6.fib6_walker_lock); 1870 FOR_WALKERS(net, w) { 1871 if (!child) { 1872 if (w->node == fn) { 1873 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1874 w->node = pn; 1875 w->state = nstate; 1876 } 1877 } else { 1878 if (w->node == fn) { 1879 w->node = child; 1880 if (children&2) { 1881 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1882 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT; 1883 } else { 1884 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1885 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT; 1886 } 1887 } 1888 } 1889 } 1890 read_unlock(&net->ipv6.fib6_walker_lock); 1891 1892 node_free(net, fn); 1893 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) 1894 return pn; 1895 1896 RCU_INIT_POINTER(pn->leaf, NULL); 1897 fib6_info_release(pn_leaf); 1898 fn = pn; 1899 } 1900 } 1901 1902 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn, 1903 struct fib6_info __rcu **rtp, struct nl_info *info) 1904 { 1905 struct fib6_info *leaf, *replace_rt = NULL; 1906 struct fib6_walker *w; 1907 struct fib6_info *rt = rcu_dereference_protected(*rtp, 1908 lockdep_is_held(&table->tb6_lock)); 1909 struct net *net = info->nl_net; 1910 bool notify_del = false; 1911 1912 RT6_TRACE("fib6_del_route\n"); 1913 1914 /* If the deleted route is the first in the node and it is not part of 1915 * a multipath route, then we need to replace it with the next route 1916 * in the node, if exists. 1917 */ 1918 leaf = rcu_dereference_protected(fn->leaf, 1919 lockdep_is_held(&table->tb6_lock)); 1920 if (leaf == rt && !rt->fib6_nsiblings) { 1921 if (rcu_access_pointer(rt->fib6_next)) 1922 replace_rt = rcu_dereference_protected(rt->fib6_next, 1923 lockdep_is_held(&table->tb6_lock)); 1924 else 1925 notify_del = true; 1926 } 1927 1928 /* Unlink it */ 1929 *rtp = rt->fib6_next; 1930 rt->fib6_node = NULL; 1931 net->ipv6.rt6_stats->fib_rt_entries--; 1932 net->ipv6.rt6_stats->fib_discarded_routes++; 1933 1934 /* Reset round-robin state, if necessary */ 1935 if (rcu_access_pointer(fn->rr_ptr) == rt) 1936 fn->rr_ptr = NULL; 1937 1938 /* Remove this entry from other siblings */ 1939 if (rt->fib6_nsiblings) { 1940 struct fib6_info *sibling, *next_sibling; 1941 1942 /* The route is deleted from a multipath route. If this 1943 * multipath route is the first route in the node, then we need 1944 * to emit a delete notification. Otherwise, we need to skip 1945 * the notification. 1946 */ 1947 if (rt->fib6_metric == leaf->fib6_metric && 1948 rt6_qualify_for_ecmp(leaf)) 1949 notify_del = true; 1950 list_for_each_entry_safe(sibling, next_sibling, 1951 &rt->fib6_siblings, fib6_siblings) 1952 sibling->fib6_nsiblings--; 1953 rt->fib6_nsiblings = 0; 1954 list_del_init(&rt->fib6_siblings); 1955 rt6_multipath_rebalance(next_sibling); 1956 } 1957 1958 /* Adjust walkers */ 1959 read_lock(&net->ipv6.fib6_walker_lock); 1960 FOR_WALKERS(net, w) { 1961 if (w->state == FWS_C && w->leaf == rt) { 1962 RT6_TRACE("walker %p adjusted by delroute\n", w); 1963 w->leaf = rcu_dereference_protected(rt->fib6_next, 1964 lockdep_is_held(&table->tb6_lock)); 1965 if (!w->leaf) 1966 w->state = FWS_U; 1967 } 1968 } 1969 read_unlock(&net->ipv6.fib6_walker_lock); 1970 1971 /* If it was last route, call fib6_repair_tree() to: 1972 * 1. For root node, put back null_entry as how the table was created. 1973 * 2. For other nodes, expunge its radix tree node. 1974 */ 1975 if (!rcu_access_pointer(fn->leaf)) { 1976 if (!(fn->fn_flags & RTN_TL_ROOT)) { 1977 fn->fn_flags &= ~RTN_RTINFO; 1978 net->ipv6.rt6_stats->fib_route_nodes--; 1979 } 1980 fn = fib6_repair_tree(net, table, fn); 1981 } 1982 1983 fib6_purge_rt(rt, fn, net); 1984 1985 if (!info->skip_notify_kernel) { 1986 if (notify_del) 1987 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, 1988 rt, NULL); 1989 else if (replace_rt) 1990 call_fib6_entry_notifiers_replace(net, replace_rt); 1991 } 1992 if (!info->skip_notify) 1993 inet6_rt_notify(RTM_DELROUTE, rt, info, 0); 1994 1995 fib6_info_release(rt); 1996 } 1997 1998 /* Need to own table->tb6_lock */ 1999 int fib6_del(struct fib6_info *rt, struct nl_info *info) 2000 { 2001 struct net *net = info->nl_net; 2002 struct fib6_info __rcu **rtp; 2003 struct fib6_info __rcu **rtp_next; 2004 struct fib6_table *table; 2005 struct fib6_node *fn; 2006 2007 if (rt == net->ipv6.fib6_null_entry) 2008 return -ENOENT; 2009 2010 table = rt->fib6_table; 2011 fn = rcu_dereference_protected(rt->fib6_node, 2012 lockdep_is_held(&table->tb6_lock)); 2013 if (!fn) 2014 return -ENOENT; 2015 2016 WARN_ON(!(fn->fn_flags & RTN_RTINFO)); 2017 2018 /* 2019 * Walk the leaf entries looking for ourself 2020 */ 2021 2022 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) { 2023 struct fib6_info *cur = rcu_dereference_protected(*rtp, 2024 lockdep_is_held(&table->tb6_lock)); 2025 if (rt == cur) { 2026 if (fib6_requires_src(cur)) 2027 fib6_routes_require_src_dec(info->nl_net); 2028 fib6_del_route(table, fn, rtp, info); 2029 return 0; 2030 } 2031 rtp_next = &cur->fib6_next; 2032 } 2033 return -ENOENT; 2034 } 2035 2036 /* 2037 * Tree traversal function. 2038 * 2039 * Certainly, it is not interrupt safe. 2040 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 2041 * It means, that we can modify tree during walking 2042 * and use this function for garbage collection, clone pruning, 2043 * cleaning tree when a device goes down etc. etc. 2044 * 2045 * It guarantees that every node will be traversed, 2046 * and that it will be traversed only once. 2047 * 2048 * Callback function w->func may return: 2049 * 0 -> continue walking. 2050 * positive value -> walking is suspended (used by tree dumps, 2051 * and probably by gc, if it will be split to several slices) 2052 * negative value -> terminate walking. 2053 * 2054 * The function itself returns: 2055 * 0 -> walk is complete. 2056 * >0 -> walk is incomplete (i.e. suspended) 2057 * <0 -> walk is terminated by an error. 2058 * 2059 * This function is called with tb6_lock held. 2060 */ 2061 2062 static int fib6_walk_continue(struct fib6_walker *w) 2063 { 2064 struct fib6_node *fn, *pn, *left, *right; 2065 2066 /* w->root should always be table->tb6_root */ 2067 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT)); 2068 2069 for (;;) { 2070 fn = w->node; 2071 if (!fn) 2072 return 0; 2073 2074 switch (w->state) { 2075 #ifdef CONFIG_IPV6_SUBTREES 2076 case FWS_S: 2077 if (FIB6_SUBTREE(fn)) { 2078 w->node = FIB6_SUBTREE(fn); 2079 continue; 2080 } 2081 w->state = FWS_L; 2082 fallthrough; 2083 #endif 2084 case FWS_L: 2085 left = rcu_dereference_protected(fn->left, 1); 2086 if (left) { 2087 w->node = left; 2088 w->state = FWS_INIT; 2089 continue; 2090 } 2091 w->state = FWS_R; 2092 fallthrough; 2093 case FWS_R: 2094 right = rcu_dereference_protected(fn->right, 1); 2095 if (right) { 2096 w->node = right; 2097 w->state = FWS_INIT; 2098 continue; 2099 } 2100 w->state = FWS_C; 2101 w->leaf = rcu_dereference_protected(fn->leaf, 1); 2102 fallthrough; 2103 case FWS_C: 2104 if (w->leaf && fn->fn_flags & RTN_RTINFO) { 2105 int err; 2106 2107 if (w->skip) { 2108 w->skip--; 2109 goto skip; 2110 } 2111 2112 err = w->func(w); 2113 if (err) 2114 return err; 2115 2116 w->count++; 2117 continue; 2118 } 2119 skip: 2120 w->state = FWS_U; 2121 fallthrough; 2122 case FWS_U: 2123 if (fn == w->root) 2124 return 0; 2125 pn = rcu_dereference_protected(fn->parent, 1); 2126 left = rcu_dereference_protected(pn->left, 1); 2127 right = rcu_dereference_protected(pn->right, 1); 2128 w->node = pn; 2129 #ifdef CONFIG_IPV6_SUBTREES 2130 if (FIB6_SUBTREE(pn) == fn) { 2131 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 2132 w->state = FWS_L; 2133 continue; 2134 } 2135 #endif 2136 if (left == fn) { 2137 w->state = FWS_R; 2138 continue; 2139 } 2140 if (right == fn) { 2141 w->state = FWS_C; 2142 w->leaf = rcu_dereference_protected(w->node->leaf, 1); 2143 continue; 2144 } 2145 #if RT6_DEBUG >= 2 2146 WARN_ON(1); 2147 #endif 2148 } 2149 } 2150 } 2151 2152 static int fib6_walk(struct net *net, struct fib6_walker *w) 2153 { 2154 int res; 2155 2156 w->state = FWS_INIT; 2157 w->node = w->root; 2158 2159 fib6_walker_link(net, w); 2160 res = fib6_walk_continue(w); 2161 if (res <= 0) 2162 fib6_walker_unlink(net, w); 2163 return res; 2164 } 2165 2166 static int fib6_clean_node(struct fib6_walker *w) 2167 { 2168 int res; 2169 struct fib6_info *rt; 2170 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w); 2171 struct nl_info info = { 2172 .nl_net = c->net, 2173 .skip_notify = c->skip_notify, 2174 }; 2175 2176 if (c->sernum != FIB6_NO_SERNUM_CHANGE && 2177 w->node->fn_sernum != c->sernum) 2178 w->node->fn_sernum = c->sernum; 2179 2180 if (!c->func) { 2181 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE); 2182 w->leaf = NULL; 2183 return 0; 2184 } 2185 2186 for_each_fib6_walker_rt(w) { 2187 res = c->func(rt, c->arg); 2188 if (res == -1) { 2189 w->leaf = rt; 2190 res = fib6_del(rt, &info); 2191 if (res) { 2192 #if RT6_DEBUG >= 2 2193 pr_debug("%s: del failed: rt=%p@%p err=%d\n", 2194 __func__, rt, 2195 rcu_access_pointer(rt->fib6_node), 2196 res); 2197 #endif 2198 continue; 2199 } 2200 return 0; 2201 } else if (res == -2) { 2202 if (WARN_ON(!rt->fib6_nsiblings)) 2203 continue; 2204 rt = list_last_entry(&rt->fib6_siblings, 2205 struct fib6_info, fib6_siblings); 2206 continue; 2207 } 2208 WARN_ON(res != 0); 2209 } 2210 w->leaf = rt; 2211 return 0; 2212 } 2213 2214 /* 2215 * Convenient frontend to tree walker. 2216 * 2217 * func is called on each route. 2218 * It may return -2 -> skip multipath route. 2219 * -1 -> delete this route. 2220 * 0 -> continue walking 2221 */ 2222 2223 static void fib6_clean_tree(struct net *net, struct fib6_node *root, 2224 int (*func)(struct fib6_info *, void *arg), 2225 int sernum, void *arg, bool skip_notify) 2226 { 2227 struct fib6_cleaner c; 2228 2229 c.w.root = root; 2230 c.w.func = fib6_clean_node; 2231 c.w.count = 0; 2232 c.w.skip = 0; 2233 c.w.skip_in_node = 0; 2234 c.func = func; 2235 c.sernum = sernum; 2236 c.arg = arg; 2237 c.net = net; 2238 c.skip_notify = skip_notify; 2239 2240 fib6_walk(net, &c.w); 2241 } 2242 2243 static void __fib6_clean_all(struct net *net, 2244 int (*func)(struct fib6_info *, void *), 2245 int sernum, void *arg, bool skip_notify) 2246 { 2247 struct fib6_table *table; 2248 struct hlist_head *head; 2249 unsigned int h; 2250 2251 rcu_read_lock(); 2252 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 2253 head = &net->ipv6.fib_table_hash[h]; 2254 hlist_for_each_entry_rcu(table, head, tb6_hlist) { 2255 spin_lock_bh(&table->tb6_lock); 2256 fib6_clean_tree(net, &table->tb6_root, 2257 func, sernum, arg, skip_notify); 2258 spin_unlock_bh(&table->tb6_lock); 2259 } 2260 } 2261 rcu_read_unlock(); 2262 } 2263 2264 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *), 2265 void *arg) 2266 { 2267 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false); 2268 } 2269 2270 void fib6_clean_all_skip_notify(struct net *net, 2271 int (*func)(struct fib6_info *, void *), 2272 void *arg) 2273 { 2274 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true); 2275 } 2276 2277 static void fib6_flush_trees(struct net *net) 2278 { 2279 int new_sernum = fib6_new_sernum(net); 2280 2281 __fib6_clean_all(net, NULL, new_sernum, NULL, false); 2282 } 2283 2284 /* 2285 * Garbage collection 2286 */ 2287 2288 static int fib6_age(struct fib6_info *rt, void *arg) 2289 { 2290 struct fib6_gc_args *gc_args = arg; 2291 unsigned long now = jiffies; 2292 2293 /* 2294 * check addrconf expiration here. 2295 * Routes are expired even if they are in use. 2296 */ 2297 2298 if (rt->fib6_flags & RTF_EXPIRES && rt->expires) { 2299 if (time_after(now, rt->expires)) { 2300 RT6_TRACE("expiring %p\n", rt); 2301 return -1; 2302 } 2303 gc_args->more++; 2304 } 2305 2306 /* Also age clones in the exception table. 2307 * Note, that clones are aged out 2308 * only if they are not in use now. 2309 */ 2310 rt6_age_exceptions(rt, gc_args, now); 2311 2312 return 0; 2313 } 2314 2315 void fib6_run_gc(unsigned long expires, struct net *net, bool force) 2316 { 2317 struct fib6_gc_args gc_args; 2318 unsigned long now; 2319 2320 if (force) { 2321 spin_lock_bh(&net->ipv6.fib6_gc_lock); 2322 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) { 2323 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); 2324 return; 2325 } 2326 gc_args.timeout = expires ? (int)expires : 2327 net->ipv6.sysctl.ip6_rt_gc_interval; 2328 gc_args.more = 0; 2329 2330 fib6_clean_all(net, fib6_age, &gc_args); 2331 now = jiffies; 2332 net->ipv6.ip6_rt_last_gc = now; 2333 2334 if (gc_args.more) 2335 mod_timer(&net->ipv6.ip6_fib_timer, 2336 round_jiffies(now 2337 + net->ipv6.sysctl.ip6_rt_gc_interval)); 2338 else 2339 del_timer(&net->ipv6.ip6_fib_timer); 2340 spin_unlock_bh(&net->ipv6.fib6_gc_lock); 2341 } 2342 2343 static void fib6_gc_timer_cb(struct timer_list *t) 2344 { 2345 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer); 2346 2347 fib6_run_gc(0, arg, true); 2348 } 2349 2350 static int __net_init fib6_net_init(struct net *net) 2351 { 2352 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; 2353 int err; 2354 2355 err = fib6_notifier_init(net); 2356 if (err) 2357 return err; 2358 2359 /* Default to 3-tuple */ 2360 net->ipv6.sysctl.multipath_hash_fields = 2361 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK; 2362 2363 spin_lock_init(&net->ipv6.fib6_gc_lock); 2364 rwlock_init(&net->ipv6.fib6_walker_lock); 2365 INIT_LIST_HEAD(&net->ipv6.fib6_walkers); 2366 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0); 2367 2368 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); 2369 if (!net->ipv6.rt6_stats) 2370 goto out_notifier; 2371 2372 /* Avoid false sharing : Use at least a full cache line */ 2373 size = max_t(size_t, size, L1_CACHE_BYTES); 2374 2375 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); 2376 if (!net->ipv6.fib_table_hash) 2377 goto out_rt6_stats; 2378 2379 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), 2380 GFP_KERNEL); 2381 if (!net->ipv6.fib6_main_tbl) 2382 goto out_fib_table_hash; 2383 2384 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; 2385 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf, 2386 net->ipv6.fib6_null_entry); 2387 net->ipv6.fib6_main_tbl->tb6_root.fn_flags = 2388 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 2389 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); 2390 2391 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 2392 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), 2393 GFP_KERNEL); 2394 if (!net->ipv6.fib6_local_tbl) 2395 goto out_fib6_main_tbl; 2396 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; 2397 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf, 2398 net->ipv6.fib6_null_entry); 2399 net->ipv6.fib6_local_tbl->tb6_root.fn_flags = 2400 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 2401 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); 2402 #endif 2403 fib6_tables_init(net); 2404 2405 return 0; 2406 2407 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 2408 out_fib6_main_tbl: 2409 kfree(net->ipv6.fib6_main_tbl); 2410 #endif 2411 out_fib_table_hash: 2412 kfree(net->ipv6.fib_table_hash); 2413 out_rt6_stats: 2414 kfree(net->ipv6.rt6_stats); 2415 out_notifier: 2416 fib6_notifier_exit(net); 2417 return -ENOMEM; 2418 } 2419 2420 static void fib6_net_exit(struct net *net) 2421 { 2422 unsigned int i; 2423 2424 del_timer_sync(&net->ipv6.ip6_fib_timer); 2425 2426 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) { 2427 struct hlist_head *head = &net->ipv6.fib_table_hash[i]; 2428 struct hlist_node *tmp; 2429 struct fib6_table *tb; 2430 2431 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) { 2432 hlist_del(&tb->tb6_hlist); 2433 fib6_free_table(tb); 2434 } 2435 } 2436 2437 kfree(net->ipv6.fib_table_hash); 2438 kfree(net->ipv6.rt6_stats); 2439 fib6_notifier_exit(net); 2440 } 2441 2442 static struct pernet_operations fib6_net_ops = { 2443 .init = fib6_net_init, 2444 .exit = fib6_net_exit, 2445 }; 2446 2447 int __init fib6_init(void) 2448 { 2449 int ret = -ENOMEM; 2450 2451 fib6_node_kmem = kmem_cache_create("fib6_nodes", 2452 sizeof(struct fib6_node), 0, 2453 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, 2454 NULL); 2455 if (!fib6_node_kmem) 2456 goto out; 2457 2458 ret = register_pernet_subsys(&fib6_net_ops); 2459 if (ret) 2460 goto out_kmem_cache_create; 2461 2462 ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL, 2463 inet6_dump_fib, 0); 2464 if (ret) 2465 goto out_unregister_subsys; 2466 2467 __fib6_flush_trees = fib6_flush_trees; 2468 out: 2469 return ret; 2470 2471 out_unregister_subsys: 2472 unregister_pernet_subsys(&fib6_net_ops); 2473 out_kmem_cache_create: 2474 kmem_cache_destroy(fib6_node_kmem); 2475 goto out; 2476 } 2477 2478 void fib6_gc_cleanup(void) 2479 { 2480 unregister_pernet_subsys(&fib6_net_ops); 2481 kmem_cache_destroy(fib6_node_kmem); 2482 } 2483 2484 #ifdef CONFIG_PROC_FS 2485 static int ipv6_route_native_seq_show(struct seq_file *seq, void *v) 2486 { 2487 struct fib6_info *rt = v; 2488 struct ipv6_route_iter *iter = seq->private; 2489 struct fib6_nh *fib6_nh = rt->fib6_nh; 2490 unsigned int flags = rt->fib6_flags; 2491 const struct net_device *dev; 2492 2493 if (rt->nh) 2494 fib6_nh = nexthop_fib6_nh_bh(rt->nh); 2495 2496 seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen); 2497 2498 #ifdef CONFIG_IPV6_SUBTREES 2499 seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen); 2500 #else 2501 seq_puts(seq, "00000000000000000000000000000000 00 "); 2502 #endif 2503 if (fib6_nh->fib_nh_gw_family) { 2504 flags |= RTF_GATEWAY; 2505 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6); 2506 } else { 2507 seq_puts(seq, "00000000000000000000000000000000"); 2508 } 2509 2510 dev = fib6_nh->fib_nh_dev; 2511 seq_printf(seq, " %08x %08x %08x %08x %8s\n", 2512 rt->fib6_metric, refcount_read(&rt->fib6_ref), 0, 2513 flags, dev ? dev->name : ""); 2514 iter->w.leaf = NULL; 2515 return 0; 2516 } 2517 2518 static int ipv6_route_yield(struct fib6_walker *w) 2519 { 2520 struct ipv6_route_iter *iter = w->args; 2521 2522 if (!iter->skip) 2523 return 1; 2524 2525 do { 2526 iter->w.leaf = rcu_dereference_protected( 2527 iter->w.leaf->fib6_next, 2528 lockdep_is_held(&iter->tbl->tb6_lock)); 2529 iter->skip--; 2530 if (!iter->skip && iter->w.leaf) 2531 return 1; 2532 } while (iter->w.leaf); 2533 2534 return 0; 2535 } 2536 2537 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter, 2538 struct net *net) 2539 { 2540 memset(&iter->w, 0, sizeof(iter->w)); 2541 iter->w.func = ipv6_route_yield; 2542 iter->w.root = &iter->tbl->tb6_root; 2543 iter->w.state = FWS_INIT; 2544 iter->w.node = iter->w.root; 2545 iter->w.args = iter; 2546 iter->sernum = iter->w.root->fn_sernum; 2547 INIT_LIST_HEAD(&iter->w.lh); 2548 fib6_walker_link(net, &iter->w); 2549 } 2550 2551 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl, 2552 struct net *net) 2553 { 2554 unsigned int h; 2555 struct hlist_node *node; 2556 2557 if (tbl) { 2558 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1; 2559 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist)); 2560 } else { 2561 h = 0; 2562 node = NULL; 2563 } 2564 2565 while (!node && h < FIB6_TABLE_HASHSZ) { 2566 node = rcu_dereference_bh( 2567 hlist_first_rcu(&net->ipv6.fib_table_hash[h++])); 2568 } 2569 return hlist_entry_safe(node, struct fib6_table, tb6_hlist); 2570 } 2571 2572 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter) 2573 { 2574 if (iter->sernum != iter->w.root->fn_sernum) { 2575 iter->sernum = iter->w.root->fn_sernum; 2576 iter->w.state = FWS_INIT; 2577 iter->w.node = iter->w.root; 2578 WARN_ON(iter->w.skip); 2579 iter->w.skip = iter->w.count; 2580 } 2581 } 2582 2583 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2584 { 2585 int r; 2586 struct fib6_info *n; 2587 struct net *net = seq_file_net(seq); 2588 struct ipv6_route_iter *iter = seq->private; 2589 2590 ++(*pos); 2591 if (!v) 2592 goto iter_table; 2593 2594 n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next); 2595 if (n) 2596 return n; 2597 2598 iter_table: 2599 ipv6_route_check_sernum(iter); 2600 spin_lock_bh(&iter->tbl->tb6_lock); 2601 r = fib6_walk_continue(&iter->w); 2602 spin_unlock_bh(&iter->tbl->tb6_lock); 2603 if (r > 0) { 2604 return iter->w.leaf; 2605 } else if (r < 0) { 2606 fib6_walker_unlink(net, &iter->w); 2607 return NULL; 2608 } 2609 fib6_walker_unlink(net, &iter->w); 2610 2611 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net); 2612 if (!iter->tbl) 2613 return NULL; 2614 2615 ipv6_route_seq_setup_walk(iter, net); 2616 goto iter_table; 2617 } 2618 2619 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos) 2620 __acquires(RCU_BH) 2621 { 2622 struct net *net = seq_file_net(seq); 2623 struct ipv6_route_iter *iter = seq->private; 2624 2625 rcu_read_lock_bh(); 2626 iter->tbl = ipv6_route_seq_next_table(NULL, net); 2627 iter->skip = *pos; 2628 2629 if (iter->tbl) { 2630 loff_t p = 0; 2631 2632 ipv6_route_seq_setup_walk(iter, net); 2633 return ipv6_route_seq_next(seq, NULL, &p); 2634 } else { 2635 return NULL; 2636 } 2637 } 2638 2639 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter) 2640 { 2641 struct fib6_walker *w = &iter->w; 2642 return w->node && !(w->state == FWS_U && w->node == w->root); 2643 } 2644 2645 static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v) 2646 __releases(RCU_BH) 2647 { 2648 struct net *net = seq_file_net(seq); 2649 struct ipv6_route_iter *iter = seq->private; 2650 2651 if (ipv6_route_iter_active(iter)) 2652 fib6_walker_unlink(net, &iter->w); 2653 2654 rcu_read_unlock_bh(); 2655 } 2656 2657 #if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL) 2658 static int ipv6_route_prog_seq_show(struct bpf_prog *prog, 2659 struct bpf_iter_meta *meta, 2660 void *v) 2661 { 2662 struct bpf_iter__ipv6_route ctx; 2663 2664 ctx.meta = meta; 2665 ctx.rt = v; 2666 return bpf_iter_run_prog(prog, &ctx); 2667 } 2668 2669 static int ipv6_route_seq_show(struct seq_file *seq, void *v) 2670 { 2671 struct ipv6_route_iter *iter = seq->private; 2672 struct bpf_iter_meta meta; 2673 struct bpf_prog *prog; 2674 int ret; 2675 2676 meta.seq = seq; 2677 prog = bpf_iter_get_info(&meta, false); 2678 if (!prog) 2679 return ipv6_route_native_seq_show(seq, v); 2680 2681 ret = ipv6_route_prog_seq_show(prog, &meta, v); 2682 iter->w.leaf = NULL; 2683 2684 return ret; 2685 } 2686 2687 static void ipv6_route_seq_stop(struct seq_file *seq, void *v) 2688 { 2689 struct bpf_iter_meta meta; 2690 struct bpf_prog *prog; 2691 2692 if (!v) { 2693 meta.seq = seq; 2694 prog = bpf_iter_get_info(&meta, true); 2695 if (prog) 2696 (void)ipv6_route_prog_seq_show(prog, &meta, v); 2697 } 2698 2699 ipv6_route_native_seq_stop(seq, v); 2700 } 2701 #else 2702 static int ipv6_route_seq_show(struct seq_file *seq, void *v) 2703 { 2704 return ipv6_route_native_seq_show(seq, v); 2705 } 2706 2707 static void ipv6_route_seq_stop(struct seq_file *seq, void *v) 2708 { 2709 ipv6_route_native_seq_stop(seq, v); 2710 } 2711 #endif 2712 2713 const struct seq_operations ipv6_route_seq_ops = { 2714 .start = ipv6_route_seq_start, 2715 .next = ipv6_route_seq_next, 2716 .stop = ipv6_route_seq_stop, 2717 .show = ipv6_route_seq_show 2718 }; 2719 #endif /* CONFIG_PROC_FS */ 2720