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 = atomic_read(&net->ipv6.fib6_sernum); 95 96 do { 97 new = old < INT_MAX ? old + 1 : 1; 98 } while (!atomic_try_cmpxchg(&net->ipv6.fib6_sernum, &old, new)); 99 100 return new; 101 } 102 103 enum { 104 FIB6_NO_SERNUM_CHANGE = 0, 105 }; 106 107 void fib6_update_sernum(struct net *net, struct fib6_info *f6i) 108 { 109 struct fib6_node *fn; 110 111 fn = rcu_dereference_protected(f6i->fib6_node, 112 lockdep_is_held(&f6i->fib6_table->tb6_lock)); 113 if (fn) 114 WRITE_ONCE(fn->fn_sernum, fib6_new_sernum(net)); 115 } 116 117 /* 118 * Auxiliary address test functions for the radix tree. 119 * 120 * These assume a 32bit processor (although it will work on 121 * 64bit processors) 122 */ 123 124 /* 125 * test bit 126 */ 127 #if defined(__LITTLE_ENDIAN) 128 # define BITOP_BE32_SWIZZLE (0x1F & ~7) 129 #else 130 # define BITOP_BE32_SWIZZLE 0 131 #endif 132 133 static __be32 addr_bit_set(const void *token, int fn_bit) 134 { 135 const __be32 *addr = token; 136 /* 137 * Here, 138 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f) 139 * is optimized version of 140 * htonl(1 << ((~fn_bit)&0x1F)) 141 * See include/asm-generic/bitops/le.h. 142 */ 143 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) & 144 addr[fn_bit >> 5]; 145 } 146 147 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh) 148 { 149 struct fib6_info *f6i; 150 size_t sz = sizeof(*f6i); 151 152 if (with_fib6_nh) 153 sz += sizeof(struct fib6_nh); 154 155 f6i = kzalloc(sz, gfp_flags); 156 if (!f6i) 157 return NULL; 158 159 /* fib6_siblings is a union with nh_list, so this initializes both */ 160 INIT_LIST_HEAD(&f6i->fib6_siblings); 161 refcount_set(&f6i->fib6_ref, 1); 162 163 return f6i; 164 } 165 166 void fib6_info_destroy_rcu(struct rcu_head *head) 167 { 168 struct fib6_info *f6i = container_of(head, struct fib6_info, rcu); 169 170 WARN_ON(f6i->fib6_node); 171 172 if (f6i->nh) 173 nexthop_put(f6i->nh); 174 else 175 fib6_nh_release(f6i->fib6_nh); 176 177 ip_fib_metrics_put(f6i->fib6_metrics); 178 kfree(f6i); 179 } 180 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu); 181 182 static struct fib6_node *node_alloc(struct net *net) 183 { 184 struct fib6_node *fn; 185 186 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC); 187 if (fn) 188 net->ipv6.rt6_stats->fib_nodes++; 189 190 return fn; 191 } 192 193 static void node_free_immediate(struct net *net, struct fib6_node *fn) 194 { 195 kmem_cache_free(fib6_node_kmem, fn); 196 net->ipv6.rt6_stats->fib_nodes--; 197 } 198 199 static void node_free_rcu(struct rcu_head *head) 200 { 201 struct fib6_node *fn = container_of(head, struct fib6_node, rcu); 202 203 kmem_cache_free(fib6_node_kmem, fn); 204 } 205 206 static void node_free(struct net *net, struct fib6_node *fn) 207 { 208 call_rcu(&fn->rcu, node_free_rcu); 209 net->ipv6.rt6_stats->fib_nodes--; 210 } 211 212 static void fib6_free_table(struct fib6_table *table) 213 { 214 inetpeer_invalidate_tree(&table->tb6_peers); 215 kfree(table); 216 } 217 218 static void fib6_link_table(struct net *net, struct fib6_table *tb) 219 { 220 unsigned int h; 221 222 /* 223 * Initialize table lock at a single place to give lockdep a key, 224 * tables aren't visible prior to being linked to the list. 225 */ 226 spin_lock_init(&tb->tb6_lock); 227 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1); 228 229 /* 230 * No protection necessary, this is the only list mutatation 231 * operation, tables never disappear once they exist. 232 */ 233 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]); 234 } 235 236 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 237 238 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id) 239 { 240 struct fib6_table *table; 241 242 table = kzalloc(sizeof(*table), GFP_ATOMIC); 243 if (table) { 244 table->tb6_id = id; 245 rcu_assign_pointer(table->tb6_root.leaf, 246 net->ipv6.fib6_null_entry); 247 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 248 inet_peer_base_init(&table->tb6_peers); 249 } 250 251 return table; 252 } 253 254 struct fib6_table *fib6_new_table(struct net *net, u32 id) 255 { 256 struct fib6_table *tb; 257 258 if (id == 0) 259 id = RT6_TABLE_MAIN; 260 tb = fib6_get_table(net, id); 261 if (tb) 262 return tb; 263 264 tb = fib6_alloc_table(net, id); 265 if (tb) 266 fib6_link_table(net, tb); 267 268 return tb; 269 } 270 EXPORT_SYMBOL_GPL(fib6_new_table); 271 272 struct fib6_table *fib6_get_table(struct net *net, u32 id) 273 { 274 struct fib6_table *tb; 275 struct hlist_head *head; 276 unsigned int h; 277 278 if (id == 0) 279 id = RT6_TABLE_MAIN; 280 h = id & (FIB6_TABLE_HASHSZ - 1); 281 rcu_read_lock(); 282 head = &net->ipv6.fib_table_hash[h]; 283 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 284 if (tb->tb6_id == id) { 285 rcu_read_unlock(); 286 return tb; 287 } 288 } 289 rcu_read_unlock(); 290 291 return NULL; 292 } 293 EXPORT_SYMBOL_GPL(fib6_get_table); 294 295 static void __net_init fib6_tables_init(struct net *net) 296 { 297 fib6_link_table(net, net->ipv6.fib6_main_tbl); 298 fib6_link_table(net, net->ipv6.fib6_local_tbl); 299 } 300 #else 301 302 struct fib6_table *fib6_new_table(struct net *net, u32 id) 303 { 304 return fib6_get_table(net, id); 305 } 306 307 struct fib6_table *fib6_get_table(struct net *net, u32 id) 308 { 309 return net->ipv6.fib6_main_tbl; 310 } 311 312 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, 313 const struct sk_buff *skb, 314 int flags, pol_lookup_t lookup) 315 { 316 struct rt6_info *rt; 317 318 rt = pol_lookup_func(lookup, 319 net, net->ipv6.fib6_main_tbl, fl6, skb, flags); 320 if (rt->dst.error == -EAGAIN) { 321 ip6_rt_put_flags(rt, flags); 322 rt = net->ipv6.ip6_null_entry; 323 if (!(flags & RT6_LOOKUP_F_DST_NOREF)) 324 dst_hold(&rt->dst); 325 } 326 327 return &rt->dst; 328 } 329 330 /* called with rcu lock held; no reference taken on fib6_info */ 331 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6, 332 struct fib6_result *res, int flags) 333 { 334 return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6, 335 res, flags); 336 } 337 338 static void __net_init fib6_tables_init(struct net *net) 339 { 340 fib6_link_table(net, net->ipv6.fib6_main_tbl); 341 } 342 343 #endif 344 345 unsigned int fib6_tables_seq_read(struct net *net) 346 { 347 unsigned int h, fib_seq = 0; 348 349 rcu_read_lock(); 350 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 351 struct hlist_head *head = &net->ipv6.fib_table_hash[h]; 352 struct fib6_table *tb; 353 354 hlist_for_each_entry_rcu(tb, head, tb6_hlist) 355 fib_seq += tb->fib_seq; 356 } 357 rcu_read_unlock(); 358 359 return fib_seq; 360 } 361 362 static int call_fib6_entry_notifier(struct notifier_block *nb, 363 enum fib_event_type event_type, 364 struct fib6_info *rt, 365 struct netlink_ext_ack *extack) 366 { 367 struct fib6_entry_notifier_info info = { 368 .info.extack = extack, 369 .rt = rt, 370 }; 371 372 return call_fib6_notifier(nb, event_type, &info.info); 373 } 374 375 static int call_fib6_multipath_entry_notifier(struct notifier_block *nb, 376 enum fib_event_type event_type, 377 struct fib6_info *rt, 378 unsigned int nsiblings, 379 struct netlink_ext_ack *extack) 380 { 381 struct fib6_entry_notifier_info info = { 382 .info.extack = extack, 383 .rt = rt, 384 .nsiblings = nsiblings, 385 }; 386 387 return call_fib6_notifier(nb, event_type, &info.info); 388 } 389 390 int call_fib6_entry_notifiers(struct net *net, 391 enum fib_event_type event_type, 392 struct fib6_info *rt, 393 struct netlink_ext_ack *extack) 394 { 395 struct fib6_entry_notifier_info info = { 396 .info.extack = extack, 397 .rt = rt, 398 }; 399 400 rt->fib6_table->fib_seq++; 401 return call_fib6_notifiers(net, event_type, &info.info); 402 } 403 404 int call_fib6_multipath_entry_notifiers(struct net *net, 405 enum fib_event_type event_type, 406 struct fib6_info *rt, 407 unsigned int nsiblings, 408 struct netlink_ext_ack *extack) 409 { 410 struct fib6_entry_notifier_info info = { 411 .info.extack = extack, 412 .rt = rt, 413 .nsiblings = nsiblings, 414 }; 415 416 rt->fib6_table->fib_seq++; 417 return call_fib6_notifiers(net, event_type, &info.info); 418 } 419 420 int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt) 421 { 422 struct fib6_entry_notifier_info info = { 423 .rt = rt, 424 .nsiblings = rt->fib6_nsiblings, 425 }; 426 427 rt->fib6_table->fib_seq++; 428 return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE, &info.info); 429 } 430 431 struct fib6_dump_arg { 432 struct net *net; 433 struct notifier_block *nb; 434 struct netlink_ext_ack *extack; 435 }; 436 437 static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg) 438 { 439 enum fib_event_type fib_event = FIB_EVENT_ENTRY_REPLACE; 440 int err; 441 442 if (!rt || rt == arg->net->ipv6.fib6_null_entry) 443 return 0; 444 445 if (rt->fib6_nsiblings) 446 err = call_fib6_multipath_entry_notifier(arg->nb, fib_event, 447 rt, 448 rt->fib6_nsiblings, 449 arg->extack); 450 else 451 err = call_fib6_entry_notifier(arg->nb, fib_event, rt, 452 arg->extack); 453 454 return err; 455 } 456 457 static int fib6_node_dump(struct fib6_walker *w) 458 { 459 int err; 460 461 err = fib6_rt_dump(w->leaf, w->args); 462 w->leaf = NULL; 463 return err; 464 } 465 466 static int fib6_table_dump(struct net *net, struct fib6_table *tb, 467 struct fib6_walker *w) 468 { 469 int err; 470 471 w->root = &tb->tb6_root; 472 spin_lock_bh(&tb->tb6_lock); 473 err = fib6_walk(net, w); 474 spin_unlock_bh(&tb->tb6_lock); 475 return err; 476 } 477 478 /* Called with rcu_read_lock() */ 479 int fib6_tables_dump(struct net *net, struct notifier_block *nb, 480 struct netlink_ext_ack *extack) 481 { 482 struct fib6_dump_arg arg; 483 struct fib6_walker *w; 484 unsigned int h; 485 int err = 0; 486 487 w = kzalloc(sizeof(*w), GFP_ATOMIC); 488 if (!w) 489 return -ENOMEM; 490 491 w->func = fib6_node_dump; 492 arg.net = net; 493 arg.nb = nb; 494 arg.extack = extack; 495 w->args = &arg; 496 497 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 498 struct hlist_head *head = &net->ipv6.fib_table_hash[h]; 499 struct fib6_table *tb; 500 501 hlist_for_each_entry_rcu(tb, head, tb6_hlist) { 502 err = fib6_table_dump(net, tb, w); 503 if (err) 504 goto out; 505 } 506 } 507 508 out: 509 kfree(w); 510 511 /* The tree traversal function should never return a positive value. */ 512 return err > 0 ? -EINVAL : err; 513 } 514 515 static int fib6_dump_node(struct fib6_walker *w) 516 { 517 int res; 518 struct fib6_info *rt; 519 520 for_each_fib6_walker_rt(w) { 521 res = rt6_dump_route(rt, w->args, w->skip_in_node); 522 if (res >= 0) { 523 /* Frame is full, suspend walking */ 524 w->leaf = rt; 525 526 /* We'll restart from this node, so if some routes were 527 * already dumped, skip them next time. 528 */ 529 w->skip_in_node += res; 530 531 return 1; 532 } 533 w->skip_in_node = 0; 534 535 /* Multipath routes are dumped in one route with the 536 * RTA_MULTIPATH attribute. Jump 'rt' to point to the 537 * last sibling of this route (no need to dump the 538 * sibling routes again) 539 */ 540 if (rt->fib6_nsiblings) 541 rt = list_last_entry(&rt->fib6_siblings, 542 struct fib6_info, 543 fib6_siblings); 544 } 545 w->leaf = NULL; 546 return 0; 547 } 548 549 static void fib6_dump_end(struct netlink_callback *cb) 550 { 551 struct net *net = sock_net(cb->skb->sk); 552 struct fib6_walker *w = (void *)cb->args[2]; 553 554 if (w) { 555 if (cb->args[4]) { 556 cb->args[4] = 0; 557 fib6_walker_unlink(net, w); 558 } 559 cb->args[2] = 0; 560 kfree(w); 561 } 562 cb->done = (void *)cb->args[3]; 563 cb->args[1] = 3; 564 } 565 566 static int fib6_dump_done(struct netlink_callback *cb) 567 { 568 fib6_dump_end(cb); 569 return cb->done ? cb->done(cb) : 0; 570 } 571 572 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, 573 struct netlink_callback *cb) 574 { 575 struct net *net = sock_net(skb->sk); 576 struct fib6_walker *w; 577 int res; 578 579 w = (void *)cb->args[2]; 580 w->root = &table->tb6_root; 581 582 if (cb->args[4] == 0) { 583 w->count = 0; 584 w->skip = 0; 585 w->skip_in_node = 0; 586 587 spin_lock_bh(&table->tb6_lock); 588 res = fib6_walk(net, w); 589 spin_unlock_bh(&table->tb6_lock); 590 if (res > 0) { 591 cb->args[4] = 1; 592 cb->args[5] = READ_ONCE(w->root->fn_sernum); 593 } 594 } else { 595 int sernum = READ_ONCE(w->root->fn_sernum); 596 if (cb->args[5] != sernum) { 597 /* Begin at the root if the tree changed */ 598 cb->args[5] = 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. allocate and initialize walker. 649 */ 650 w = kzalloc(sizeof(*w), GFP_ATOMIC); 651 if (!w) 652 return -ENOMEM; 653 w->func = fib6_dump_node; 654 cb->args[2] = (long)w; 655 656 /* 2. hook callback destructor. 657 */ 658 cb->args[3] = (long)cb->done; 659 cb->done = fib6_dump_done; 660 661 } 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 rcu_read_lock(); 965 /* release the reference to this fib entry from 966 * all of its cached pcpu routes 967 */ 968 for_each_possible_cpu(cpu) { 969 struct rt6_info **ppcpu_rt; 970 struct rt6_info *pcpu_rt; 971 972 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu); 973 974 /* Paired with xchg() in rt6_get_pcpu_route() */ 975 pcpu_rt = READ_ONCE(*ppcpu_rt); 976 977 /* only dropping the 'from' reference if the cached route 978 * is using 'match'. The cached pcpu_rt->from only changes 979 * from a fib6_info to NULL (ip6_dst_destroy); it can never 980 * change from one fib6_info reference to another 981 */ 982 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) { 983 struct fib6_info *from; 984 985 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL); 986 fib6_info_release(from); 987 } 988 } 989 rcu_read_unlock(); 990 } 991 992 struct fib6_nh_pcpu_arg { 993 struct fib6_info *from; 994 const struct fib6_table *table; 995 }; 996 997 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg) 998 { 999 struct fib6_nh_pcpu_arg *arg = _arg; 1000 1001 __fib6_drop_pcpu_from(nh, arg->from, arg->table); 1002 return 0; 1003 } 1004 1005 static void fib6_drop_pcpu_from(struct fib6_info *f6i, 1006 const struct fib6_table *table) 1007 { 1008 /* Make sure rt6_make_pcpu_route() wont add other percpu routes 1009 * while we are cleaning them here. 1010 */ 1011 f6i->fib6_destroying = 1; 1012 mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */ 1013 1014 if (f6i->nh) { 1015 struct fib6_nh_pcpu_arg arg = { 1016 .from = f6i, 1017 .table = table 1018 }; 1019 1020 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from, 1021 &arg); 1022 } else { 1023 struct fib6_nh *fib6_nh; 1024 1025 fib6_nh = f6i->fib6_nh; 1026 __fib6_drop_pcpu_from(fib6_nh, f6i, table); 1027 } 1028 } 1029 1030 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn, 1031 struct net *net) 1032 { 1033 struct fib6_table *table = rt->fib6_table; 1034 1035 /* Flush all cached dst in exception table */ 1036 rt6_flush_exceptions(rt); 1037 fib6_drop_pcpu_from(rt, table); 1038 1039 if (rt->nh && !list_empty(&rt->nh_list)) 1040 list_del_init(&rt->nh_list); 1041 1042 if (refcount_read(&rt->fib6_ref) != 1) { 1043 /* This route is used as dummy address holder in some split 1044 * nodes. It is not leaked, but it still holds other resources, 1045 * which must be released in time. So, scan ascendant nodes 1046 * and replace dummy references to this route with references 1047 * to still alive ones. 1048 */ 1049 while (fn) { 1050 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf, 1051 lockdep_is_held(&table->tb6_lock)); 1052 struct fib6_info *new_leaf; 1053 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) { 1054 new_leaf = fib6_find_prefix(net, table, fn); 1055 fib6_info_hold(new_leaf); 1056 1057 rcu_assign_pointer(fn->leaf, new_leaf); 1058 fib6_info_release(rt); 1059 } 1060 fn = rcu_dereference_protected(fn->parent, 1061 lockdep_is_held(&table->tb6_lock)); 1062 } 1063 } 1064 } 1065 1066 /* 1067 * Insert routing information in a node. 1068 */ 1069 1070 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt, 1071 struct nl_info *info, 1072 struct netlink_ext_ack *extack) 1073 { 1074 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf, 1075 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1076 struct fib6_info *iter = NULL; 1077 struct fib6_info __rcu **ins; 1078 struct fib6_info __rcu **fallback_ins = NULL; 1079 int replace = (info->nlh && 1080 (info->nlh->nlmsg_flags & NLM_F_REPLACE)); 1081 int add = (!info->nlh || 1082 (info->nlh->nlmsg_flags & NLM_F_CREATE)); 1083 int found = 0; 1084 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt); 1085 bool notify_sibling_rt = false; 1086 u16 nlflags = NLM_F_EXCL; 1087 int err; 1088 1089 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND)) 1090 nlflags |= NLM_F_APPEND; 1091 1092 ins = &fn->leaf; 1093 1094 for (iter = leaf; iter; 1095 iter = rcu_dereference_protected(iter->fib6_next, 1096 lockdep_is_held(&rt->fib6_table->tb6_lock))) { 1097 /* 1098 * Search for duplicates 1099 */ 1100 1101 if (iter->fib6_metric == rt->fib6_metric) { 1102 /* 1103 * Same priority level 1104 */ 1105 if (info->nlh && 1106 (info->nlh->nlmsg_flags & NLM_F_EXCL)) 1107 return -EEXIST; 1108 1109 nlflags &= ~NLM_F_EXCL; 1110 if (replace) { 1111 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) { 1112 found++; 1113 break; 1114 } 1115 fallback_ins = fallback_ins ?: ins; 1116 goto next_iter; 1117 } 1118 1119 if (rt6_duplicate_nexthop(iter, rt)) { 1120 if (rt->fib6_nsiblings) 1121 rt->fib6_nsiblings = 0; 1122 if (!(iter->fib6_flags & RTF_EXPIRES)) 1123 return -EEXIST; 1124 if (!(rt->fib6_flags & RTF_EXPIRES)) 1125 fib6_clean_expires(iter); 1126 else 1127 fib6_set_expires(iter, rt->expires); 1128 1129 if (rt->fib6_pmtu) 1130 fib6_metric_set(iter, RTAX_MTU, 1131 rt->fib6_pmtu); 1132 return -EEXIST; 1133 } 1134 /* If we have the same destination and the same metric, 1135 * but not the same gateway, then the route we try to 1136 * add is sibling to this route, increment our counter 1137 * of siblings, and later we will add our route to the 1138 * list. 1139 * Only static routes (which don't have flag 1140 * RTF_EXPIRES) are used for ECMPv6. 1141 * 1142 * To avoid long list, we only had siblings if the 1143 * route have a gateway. 1144 */ 1145 if (rt_can_ecmp && 1146 rt6_qualify_for_ecmp(iter)) 1147 rt->fib6_nsiblings++; 1148 } 1149 1150 if (iter->fib6_metric > rt->fib6_metric) 1151 break; 1152 1153 next_iter: 1154 ins = &iter->fib6_next; 1155 } 1156 1157 if (fallback_ins && !found) { 1158 /* No matching route with same ecmp-able-ness found, replace 1159 * first matching route 1160 */ 1161 ins = fallback_ins; 1162 iter = rcu_dereference_protected(*ins, 1163 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1164 found++; 1165 } 1166 1167 /* Reset round-robin state, if necessary */ 1168 if (ins == &fn->leaf) 1169 fn->rr_ptr = NULL; 1170 1171 /* Link this route to others same route. */ 1172 if (rt->fib6_nsiblings) { 1173 unsigned int fib6_nsiblings; 1174 struct fib6_info *sibling, *temp_sibling; 1175 1176 /* Find the first route that have the same metric */ 1177 sibling = leaf; 1178 notify_sibling_rt = true; 1179 while (sibling) { 1180 if (sibling->fib6_metric == rt->fib6_metric && 1181 rt6_qualify_for_ecmp(sibling)) { 1182 list_add_tail(&rt->fib6_siblings, 1183 &sibling->fib6_siblings); 1184 break; 1185 } 1186 sibling = rcu_dereference_protected(sibling->fib6_next, 1187 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1188 notify_sibling_rt = false; 1189 } 1190 /* For each sibling in the list, increment the counter of 1191 * siblings. BUG() if counters does not match, list of siblings 1192 * is broken! 1193 */ 1194 fib6_nsiblings = 0; 1195 list_for_each_entry_safe(sibling, temp_sibling, 1196 &rt->fib6_siblings, fib6_siblings) { 1197 sibling->fib6_nsiblings++; 1198 BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings); 1199 fib6_nsiblings++; 1200 } 1201 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings); 1202 rt6_multipath_rebalance(temp_sibling); 1203 } 1204 1205 /* 1206 * insert node 1207 */ 1208 if (!replace) { 1209 if (!add) 1210 pr_warn("NLM_F_CREATE should be set when creating new route\n"); 1211 1212 add: 1213 nlflags |= NLM_F_CREATE; 1214 1215 /* The route should only be notified if it is the first 1216 * route in the node or if it is added as a sibling 1217 * route to the first route in the node. 1218 */ 1219 if (!info->skip_notify_kernel && 1220 (notify_sibling_rt || ins == &fn->leaf)) { 1221 enum fib_event_type fib_event; 1222 1223 if (notify_sibling_rt) 1224 fib_event = FIB_EVENT_ENTRY_APPEND; 1225 else 1226 fib_event = FIB_EVENT_ENTRY_REPLACE; 1227 err = call_fib6_entry_notifiers(info->nl_net, 1228 fib_event, rt, 1229 extack); 1230 if (err) { 1231 struct fib6_info *sibling, *next_sibling; 1232 1233 /* If the route has siblings, then it first 1234 * needs to be unlinked from them. 1235 */ 1236 if (!rt->fib6_nsiblings) 1237 return err; 1238 1239 list_for_each_entry_safe(sibling, next_sibling, 1240 &rt->fib6_siblings, 1241 fib6_siblings) 1242 sibling->fib6_nsiblings--; 1243 rt->fib6_nsiblings = 0; 1244 list_del_init(&rt->fib6_siblings); 1245 rt6_multipath_rebalance(next_sibling); 1246 return err; 1247 } 1248 } 1249 1250 rcu_assign_pointer(rt->fib6_next, iter); 1251 fib6_info_hold(rt); 1252 rcu_assign_pointer(rt->fib6_node, fn); 1253 rcu_assign_pointer(*ins, rt); 1254 if (!info->skip_notify) 1255 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags); 1256 info->nl_net->ipv6.rt6_stats->fib_rt_entries++; 1257 1258 if (!(fn->fn_flags & RTN_RTINFO)) { 1259 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 1260 fn->fn_flags |= RTN_RTINFO; 1261 } 1262 1263 } else { 1264 int nsiblings; 1265 1266 if (!found) { 1267 if (add) 1268 goto add; 1269 pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); 1270 return -ENOENT; 1271 } 1272 1273 if (!info->skip_notify_kernel && ins == &fn->leaf) { 1274 err = call_fib6_entry_notifiers(info->nl_net, 1275 FIB_EVENT_ENTRY_REPLACE, 1276 rt, extack); 1277 if (err) 1278 return err; 1279 } 1280 1281 fib6_info_hold(rt); 1282 rcu_assign_pointer(rt->fib6_node, fn); 1283 rt->fib6_next = iter->fib6_next; 1284 rcu_assign_pointer(*ins, rt); 1285 if (!info->skip_notify) 1286 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE); 1287 if (!(fn->fn_flags & RTN_RTINFO)) { 1288 info->nl_net->ipv6.rt6_stats->fib_route_nodes++; 1289 fn->fn_flags |= RTN_RTINFO; 1290 } 1291 nsiblings = iter->fib6_nsiblings; 1292 iter->fib6_node = NULL; 1293 fib6_purge_rt(iter, fn, info->nl_net); 1294 if (rcu_access_pointer(fn->rr_ptr) == iter) 1295 fn->rr_ptr = NULL; 1296 fib6_info_release(iter); 1297 1298 if (nsiblings) { 1299 /* Replacing an ECMP route, remove all siblings */ 1300 ins = &rt->fib6_next; 1301 iter = rcu_dereference_protected(*ins, 1302 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1303 while (iter) { 1304 if (iter->fib6_metric > rt->fib6_metric) 1305 break; 1306 if (rt6_qualify_for_ecmp(iter)) { 1307 *ins = iter->fib6_next; 1308 iter->fib6_node = NULL; 1309 fib6_purge_rt(iter, fn, info->nl_net); 1310 if (rcu_access_pointer(fn->rr_ptr) == iter) 1311 fn->rr_ptr = NULL; 1312 fib6_info_release(iter); 1313 nsiblings--; 1314 info->nl_net->ipv6.rt6_stats->fib_rt_entries--; 1315 } else { 1316 ins = &iter->fib6_next; 1317 } 1318 iter = rcu_dereference_protected(*ins, 1319 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1320 } 1321 WARN_ON(nsiblings != 0); 1322 } 1323 } 1324 1325 return 0; 1326 } 1327 1328 static void fib6_start_gc(struct net *net, struct fib6_info *rt) 1329 { 1330 if (!timer_pending(&net->ipv6.ip6_fib_timer) && 1331 (rt->fib6_flags & RTF_EXPIRES)) 1332 mod_timer(&net->ipv6.ip6_fib_timer, 1333 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 1334 } 1335 1336 void fib6_force_start_gc(struct net *net) 1337 { 1338 if (!timer_pending(&net->ipv6.ip6_fib_timer)) 1339 mod_timer(&net->ipv6.ip6_fib_timer, 1340 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); 1341 } 1342 1343 static void __fib6_update_sernum_upto_root(struct fib6_info *rt, 1344 int sernum) 1345 { 1346 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node, 1347 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1348 1349 /* paired with smp_rmb() in fib6_get_cookie_safe() */ 1350 smp_wmb(); 1351 while (fn) { 1352 WRITE_ONCE(fn->fn_sernum, sernum); 1353 fn = rcu_dereference_protected(fn->parent, 1354 lockdep_is_held(&rt->fib6_table->tb6_lock)); 1355 } 1356 } 1357 1358 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt) 1359 { 1360 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net)); 1361 } 1362 1363 /* allow ipv4 to update sernum via ipv6_stub */ 1364 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i) 1365 { 1366 spin_lock_bh(&f6i->fib6_table->tb6_lock); 1367 fib6_update_sernum_upto_root(net, f6i); 1368 spin_unlock_bh(&f6i->fib6_table->tb6_lock); 1369 } 1370 1371 /* 1372 * Add routing information to the routing tree. 1373 * <destination addr>/<source addr> 1374 * with source addr info in sub-trees 1375 * Need to own table->tb6_lock 1376 */ 1377 1378 int fib6_add(struct fib6_node *root, struct fib6_info *rt, 1379 struct nl_info *info, struct netlink_ext_ack *extack) 1380 { 1381 struct fib6_table *table = rt->fib6_table; 1382 struct fib6_node *fn; 1383 #ifdef CONFIG_IPV6_SUBTREES 1384 struct fib6_node *pn = NULL; 1385 #endif 1386 int err = -ENOMEM; 1387 int allow_create = 1; 1388 int replace_required = 0; 1389 1390 if (info->nlh) { 1391 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) 1392 allow_create = 0; 1393 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) 1394 replace_required = 1; 1395 } 1396 if (!allow_create && !replace_required) 1397 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); 1398 1399 fn = fib6_add_1(info->nl_net, table, root, 1400 &rt->fib6_dst.addr, rt->fib6_dst.plen, 1401 offsetof(struct fib6_info, fib6_dst), allow_create, 1402 replace_required, extack); 1403 if (IS_ERR(fn)) { 1404 err = PTR_ERR(fn); 1405 fn = NULL; 1406 goto out; 1407 } 1408 1409 #ifdef CONFIG_IPV6_SUBTREES 1410 pn = fn; 1411 1412 if (rt->fib6_src.plen) { 1413 struct fib6_node *sn; 1414 1415 if (!rcu_access_pointer(fn->subtree)) { 1416 struct fib6_node *sfn; 1417 1418 /* 1419 * Create subtree. 1420 * 1421 * fn[main tree] 1422 * | 1423 * sfn[subtree root] 1424 * \ 1425 * sn[new leaf node] 1426 */ 1427 1428 /* Create subtree root node */ 1429 sfn = node_alloc(info->nl_net); 1430 if (!sfn) 1431 goto failure; 1432 1433 fib6_info_hold(info->nl_net->ipv6.fib6_null_entry); 1434 rcu_assign_pointer(sfn->leaf, 1435 info->nl_net->ipv6.fib6_null_entry); 1436 sfn->fn_flags = RTN_ROOT; 1437 1438 /* Now add the first leaf node to new subtree */ 1439 1440 sn = fib6_add_1(info->nl_net, table, sfn, 1441 &rt->fib6_src.addr, rt->fib6_src.plen, 1442 offsetof(struct fib6_info, fib6_src), 1443 allow_create, replace_required, extack); 1444 1445 if (IS_ERR(sn)) { 1446 /* If it is failed, discard just allocated 1447 root, and then (in failure) stale node 1448 in main tree. 1449 */ 1450 node_free_immediate(info->nl_net, sfn); 1451 err = PTR_ERR(sn); 1452 goto failure; 1453 } 1454 1455 /* Now link new subtree to main tree */ 1456 rcu_assign_pointer(sfn->parent, fn); 1457 rcu_assign_pointer(fn->subtree, sfn); 1458 } else { 1459 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn), 1460 &rt->fib6_src.addr, rt->fib6_src.plen, 1461 offsetof(struct fib6_info, fib6_src), 1462 allow_create, replace_required, extack); 1463 1464 if (IS_ERR(sn)) { 1465 err = PTR_ERR(sn); 1466 goto failure; 1467 } 1468 } 1469 1470 if (!rcu_access_pointer(fn->leaf)) { 1471 if (fn->fn_flags & RTN_TL_ROOT) { 1472 /* put back null_entry for root node */ 1473 rcu_assign_pointer(fn->leaf, 1474 info->nl_net->ipv6.fib6_null_entry); 1475 } else { 1476 fib6_info_hold(rt); 1477 rcu_assign_pointer(fn->leaf, rt); 1478 } 1479 } 1480 fn = sn; 1481 } 1482 #endif 1483 1484 err = fib6_add_rt2node(fn, rt, info, extack); 1485 if (!err) { 1486 if (rt->nh) 1487 list_add(&rt->nh_list, &rt->nh->f6i_list); 1488 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(info->nl_net)); 1489 fib6_start_gc(info->nl_net, rt); 1490 } 1491 1492 out: 1493 if (err) { 1494 #ifdef CONFIG_IPV6_SUBTREES 1495 /* 1496 * If fib6_add_1 has cleared the old leaf pointer in the 1497 * super-tree leaf node we have to find a new one for it. 1498 */ 1499 if (pn != fn) { 1500 struct fib6_info *pn_leaf = 1501 rcu_dereference_protected(pn->leaf, 1502 lockdep_is_held(&table->tb6_lock)); 1503 if (pn_leaf == rt) { 1504 pn_leaf = NULL; 1505 RCU_INIT_POINTER(pn->leaf, NULL); 1506 fib6_info_release(rt); 1507 } 1508 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) { 1509 pn_leaf = fib6_find_prefix(info->nl_net, table, 1510 pn); 1511 if (!pn_leaf) 1512 pn_leaf = 1513 info->nl_net->ipv6.fib6_null_entry; 1514 fib6_info_hold(pn_leaf); 1515 rcu_assign_pointer(pn->leaf, pn_leaf); 1516 } 1517 } 1518 #endif 1519 goto failure; 1520 } else if (fib6_requires_src(rt)) { 1521 fib6_routes_require_src_inc(info->nl_net); 1522 } 1523 return err; 1524 1525 failure: 1526 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if: 1527 * 1. fn is an intermediate node and we failed to add the new 1528 * route to it in both subtree creation failure and fib6_add_rt2node() 1529 * failure case. 1530 * 2. fn is the root node in the table and we fail to add the first 1531 * default route to it. 1532 */ 1533 if (fn && 1534 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) || 1535 (fn->fn_flags & RTN_TL_ROOT && 1536 !rcu_access_pointer(fn->leaf)))) 1537 fib6_repair_tree(info->nl_net, table, fn); 1538 return err; 1539 } 1540 1541 /* 1542 * Routing tree lookup 1543 * 1544 */ 1545 1546 struct lookup_args { 1547 int offset; /* key offset on fib6_info */ 1548 const struct in6_addr *addr; /* search key */ 1549 }; 1550 1551 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root, 1552 struct lookup_args *args) 1553 { 1554 struct fib6_node *fn; 1555 __be32 dir; 1556 1557 if (unlikely(args->offset == 0)) 1558 return NULL; 1559 1560 /* 1561 * Descend on a tree 1562 */ 1563 1564 fn = root; 1565 1566 for (;;) { 1567 struct fib6_node *next; 1568 1569 dir = addr_bit_set(args->addr, fn->fn_bit); 1570 1571 next = dir ? rcu_dereference(fn->right) : 1572 rcu_dereference(fn->left); 1573 1574 if (next) { 1575 fn = next; 1576 continue; 1577 } 1578 break; 1579 } 1580 1581 while (fn) { 1582 struct fib6_node *subtree = FIB6_SUBTREE(fn); 1583 1584 if (subtree || fn->fn_flags & RTN_RTINFO) { 1585 struct fib6_info *leaf = rcu_dereference(fn->leaf); 1586 struct rt6key *key; 1587 1588 if (!leaf) 1589 goto backtrack; 1590 1591 key = (struct rt6key *) ((u8 *)leaf + args->offset); 1592 1593 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { 1594 #ifdef CONFIG_IPV6_SUBTREES 1595 if (subtree) { 1596 struct fib6_node *sfn; 1597 sfn = fib6_node_lookup_1(subtree, 1598 args + 1); 1599 if (!sfn) 1600 goto backtrack; 1601 fn = sfn; 1602 } 1603 #endif 1604 if (fn->fn_flags & RTN_RTINFO) 1605 return fn; 1606 } 1607 } 1608 backtrack: 1609 if (fn->fn_flags & RTN_ROOT) 1610 break; 1611 1612 fn = rcu_dereference(fn->parent); 1613 } 1614 1615 return NULL; 1616 } 1617 1618 /* called with rcu_read_lock() held 1619 */ 1620 struct fib6_node *fib6_node_lookup(struct fib6_node *root, 1621 const struct in6_addr *daddr, 1622 const struct in6_addr *saddr) 1623 { 1624 struct fib6_node *fn; 1625 struct lookup_args args[] = { 1626 { 1627 .offset = offsetof(struct fib6_info, fib6_dst), 1628 .addr = daddr, 1629 }, 1630 #ifdef CONFIG_IPV6_SUBTREES 1631 { 1632 .offset = offsetof(struct fib6_info, fib6_src), 1633 .addr = saddr, 1634 }, 1635 #endif 1636 { 1637 .offset = 0, /* sentinel */ 1638 } 1639 }; 1640 1641 fn = fib6_node_lookup_1(root, daddr ? args : args + 1); 1642 if (!fn || fn->fn_flags & RTN_TL_ROOT) 1643 fn = root; 1644 1645 return fn; 1646 } 1647 1648 /* 1649 * Get node with specified destination prefix (and source prefix, 1650 * if subtrees are used) 1651 * exact_match == true means we try to find fn with exact match of 1652 * the passed in prefix addr 1653 * exact_match == false means we try to find fn with longest prefix 1654 * match of the passed in prefix addr. This is useful for finding fn 1655 * for cached route as it will be stored in the exception table under 1656 * the node with longest prefix length. 1657 */ 1658 1659 1660 static struct fib6_node *fib6_locate_1(struct fib6_node *root, 1661 const struct in6_addr *addr, 1662 int plen, int offset, 1663 bool exact_match) 1664 { 1665 struct fib6_node *fn, *prev = NULL; 1666 1667 for (fn = root; fn ; ) { 1668 struct fib6_info *leaf = rcu_dereference(fn->leaf); 1669 struct rt6key *key; 1670 1671 /* This node is being deleted */ 1672 if (!leaf) { 1673 if (plen <= fn->fn_bit) 1674 goto out; 1675 else 1676 goto next; 1677 } 1678 1679 key = (struct rt6key *)((u8 *)leaf + offset); 1680 1681 /* 1682 * Prefix match 1683 */ 1684 if (plen < fn->fn_bit || 1685 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) 1686 goto out; 1687 1688 if (plen == fn->fn_bit) 1689 return fn; 1690 1691 if (fn->fn_flags & RTN_RTINFO) 1692 prev = fn; 1693 1694 next: 1695 /* 1696 * We have more bits to go 1697 */ 1698 if (addr_bit_set(addr, fn->fn_bit)) 1699 fn = rcu_dereference(fn->right); 1700 else 1701 fn = rcu_dereference(fn->left); 1702 } 1703 out: 1704 if (exact_match) 1705 return NULL; 1706 else 1707 return prev; 1708 } 1709 1710 struct fib6_node *fib6_locate(struct fib6_node *root, 1711 const struct in6_addr *daddr, int dst_len, 1712 const struct in6_addr *saddr, int src_len, 1713 bool exact_match) 1714 { 1715 struct fib6_node *fn; 1716 1717 fn = fib6_locate_1(root, daddr, dst_len, 1718 offsetof(struct fib6_info, fib6_dst), 1719 exact_match); 1720 1721 #ifdef CONFIG_IPV6_SUBTREES 1722 if (src_len) { 1723 WARN_ON(saddr == NULL); 1724 if (fn) { 1725 struct fib6_node *subtree = FIB6_SUBTREE(fn); 1726 1727 if (subtree) { 1728 fn = fib6_locate_1(subtree, saddr, src_len, 1729 offsetof(struct fib6_info, fib6_src), 1730 exact_match); 1731 } 1732 } 1733 } 1734 #endif 1735 1736 if (fn && fn->fn_flags & RTN_RTINFO) 1737 return fn; 1738 1739 return NULL; 1740 } 1741 1742 1743 /* 1744 * Deletion 1745 * 1746 */ 1747 1748 static struct fib6_info *fib6_find_prefix(struct net *net, 1749 struct fib6_table *table, 1750 struct fib6_node *fn) 1751 { 1752 struct fib6_node *child_left, *child_right; 1753 1754 if (fn->fn_flags & RTN_ROOT) 1755 return net->ipv6.fib6_null_entry; 1756 1757 while (fn) { 1758 child_left = rcu_dereference_protected(fn->left, 1759 lockdep_is_held(&table->tb6_lock)); 1760 child_right = rcu_dereference_protected(fn->right, 1761 lockdep_is_held(&table->tb6_lock)); 1762 if (child_left) 1763 return rcu_dereference_protected(child_left->leaf, 1764 lockdep_is_held(&table->tb6_lock)); 1765 if (child_right) 1766 return rcu_dereference_protected(child_right->leaf, 1767 lockdep_is_held(&table->tb6_lock)); 1768 1769 fn = FIB6_SUBTREE(fn); 1770 } 1771 return NULL; 1772 } 1773 1774 /* 1775 * Called to trim the tree of intermediate nodes when possible. "fn" 1776 * is the node we want to try and remove. 1777 * Need to own table->tb6_lock 1778 */ 1779 1780 static struct fib6_node *fib6_repair_tree(struct net *net, 1781 struct fib6_table *table, 1782 struct fib6_node *fn) 1783 { 1784 int children; 1785 int nstate; 1786 struct fib6_node *child; 1787 struct fib6_walker *w; 1788 int iter = 0; 1789 1790 /* Set fn->leaf to null_entry for root node. */ 1791 if (fn->fn_flags & RTN_TL_ROOT) { 1792 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry); 1793 return fn; 1794 } 1795 1796 for (;;) { 1797 struct fib6_node *fn_r = rcu_dereference_protected(fn->right, 1798 lockdep_is_held(&table->tb6_lock)); 1799 struct fib6_node *fn_l = rcu_dereference_protected(fn->left, 1800 lockdep_is_held(&table->tb6_lock)); 1801 struct fib6_node *pn = rcu_dereference_protected(fn->parent, 1802 lockdep_is_held(&table->tb6_lock)); 1803 struct fib6_node *pn_r = rcu_dereference_protected(pn->right, 1804 lockdep_is_held(&table->tb6_lock)); 1805 struct fib6_node *pn_l = rcu_dereference_protected(pn->left, 1806 lockdep_is_held(&table->tb6_lock)); 1807 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf, 1808 lockdep_is_held(&table->tb6_lock)); 1809 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf, 1810 lockdep_is_held(&table->tb6_lock)); 1811 struct fib6_info *new_fn_leaf; 1812 1813 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); 1814 iter++; 1815 1816 WARN_ON(fn->fn_flags & RTN_RTINFO); 1817 WARN_ON(fn->fn_flags & RTN_TL_ROOT); 1818 WARN_ON(fn_leaf); 1819 1820 children = 0; 1821 child = NULL; 1822 if (fn_r) { 1823 child = fn_r; 1824 children |= 1; 1825 } 1826 if (fn_l) { 1827 child = fn_l; 1828 children |= 2; 1829 } 1830 1831 if (children == 3 || FIB6_SUBTREE(fn) 1832 #ifdef CONFIG_IPV6_SUBTREES 1833 /* Subtree root (i.e. fn) may have one child */ 1834 || (children && fn->fn_flags & RTN_ROOT) 1835 #endif 1836 ) { 1837 new_fn_leaf = fib6_find_prefix(net, table, fn); 1838 #if RT6_DEBUG >= 2 1839 if (!new_fn_leaf) { 1840 WARN_ON(!new_fn_leaf); 1841 new_fn_leaf = net->ipv6.fib6_null_entry; 1842 } 1843 #endif 1844 fib6_info_hold(new_fn_leaf); 1845 rcu_assign_pointer(fn->leaf, new_fn_leaf); 1846 return pn; 1847 } 1848 1849 #ifdef CONFIG_IPV6_SUBTREES 1850 if (FIB6_SUBTREE(pn) == fn) { 1851 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 1852 RCU_INIT_POINTER(pn->subtree, NULL); 1853 nstate = FWS_L; 1854 } else { 1855 WARN_ON(fn->fn_flags & RTN_ROOT); 1856 #endif 1857 if (pn_r == fn) 1858 rcu_assign_pointer(pn->right, child); 1859 else if (pn_l == fn) 1860 rcu_assign_pointer(pn->left, child); 1861 #if RT6_DEBUG >= 2 1862 else 1863 WARN_ON(1); 1864 #endif 1865 if (child) 1866 rcu_assign_pointer(child->parent, pn); 1867 nstate = FWS_R; 1868 #ifdef CONFIG_IPV6_SUBTREES 1869 } 1870 #endif 1871 1872 read_lock(&net->ipv6.fib6_walker_lock); 1873 FOR_WALKERS(net, w) { 1874 if (!child) { 1875 if (w->node == fn) { 1876 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); 1877 w->node = pn; 1878 w->state = nstate; 1879 } 1880 } else { 1881 if (w->node == fn) { 1882 w->node = child; 1883 if (children&2) { 1884 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1885 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT; 1886 } else { 1887 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); 1888 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT; 1889 } 1890 } 1891 } 1892 } 1893 read_unlock(&net->ipv6.fib6_walker_lock); 1894 1895 node_free(net, fn); 1896 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) 1897 return pn; 1898 1899 RCU_INIT_POINTER(pn->leaf, NULL); 1900 fib6_info_release(pn_leaf); 1901 fn = pn; 1902 } 1903 } 1904 1905 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn, 1906 struct fib6_info __rcu **rtp, struct nl_info *info) 1907 { 1908 struct fib6_info *leaf, *replace_rt = NULL; 1909 struct fib6_walker *w; 1910 struct fib6_info *rt = rcu_dereference_protected(*rtp, 1911 lockdep_is_held(&table->tb6_lock)); 1912 struct net *net = info->nl_net; 1913 bool notify_del = false; 1914 1915 RT6_TRACE("fib6_del_route\n"); 1916 1917 /* If the deleted route is the first in the node and it is not part of 1918 * a multipath route, then we need to replace it with the next route 1919 * in the node, if exists. 1920 */ 1921 leaf = rcu_dereference_protected(fn->leaf, 1922 lockdep_is_held(&table->tb6_lock)); 1923 if (leaf == rt && !rt->fib6_nsiblings) { 1924 if (rcu_access_pointer(rt->fib6_next)) 1925 replace_rt = rcu_dereference_protected(rt->fib6_next, 1926 lockdep_is_held(&table->tb6_lock)); 1927 else 1928 notify_del = true; 1929 } 1930 1931 /* Unlink it */ 1932 *rtp = rt->fib6_next; 1933 rt->fib6_node = NULL; 1934 net->ipv6.rt6_stats->fib_rt_entries--; 1935 net->ipv6.rt6_stats->fib_discarded_routes++; 1936 1937 /* Reset round-robin state, if necessary */ 1938 if (rcu_access_pointer(fn->rr_ptr) == rt) 1939 fn->rr_ptr = NULL; 1940 1941 /* Remove this entry from other siblings */ 1942 if (rt->fib6_nsiblings) { 1943 struct fib6_info *sibling, *next_sibling; 1944 1945 /* The route is deleted from a multipath route. If this 1946 * multipath route is the first route in the node, then we need 1947 * to emit a delete notification. Otherwise, we need to skip 1948 * the notification. 1949 */ 1950 if (rt->fib6_metric == leaf->fib6_metric && 1951 rt6_qualify_for_ecmp(leaf)) 1952 notify_del = true; 1953 list_for_each_entry_safe(sibling, next_sibling, 1954 &rt->fib6_siblings, fib6_siblings) 1955 sibling->fib6_nsiblings--; 1956 rt->fib6_nsiblings = 0; 1957 list_del_init(&rt->fib6_siblings); 1958 rt6_multipath_rebalance(next_sibling); 1959 } 1960 1961 /* Adjust walkers */ 1962 read_lock(&net->ipv6.fib6_walker_lock); 1963 FOR_WALKERS(net, w) { 1964 if (w->state == FWS_C && w->leaf == rt) { 1965 RT6_TRACE("walker %p adjusted by delroute\n", w); 1966 w->leaf = rcu_dereference_protected(rt->fib6_next, 1967 lockdep_is_held(&table->tb6_lock)); 1968 if (!w->leaf) 1969 w->state = FWS_U; 1970 } 1971 } 1972 read_unlock(&net->ipv6.fib6_walker_lock); 1973 1974 /* If it was last route, call fib6_repair_tree() to: 1975 * 1. For root node, put back null_entry as how the table was created. 1976 * 2. For other nodes, expunge its radix tree node. 1977 */ 1978 if (!rcu_access_pointer(fn->leaf)) { 1979 if (!(fn->fn_flags & RTN_TL_ROOT)) { 1980 fn->fn_flags &= ~RTN_RTINFO; 1981 net->ipv6.rt6_stats->fib_route_nodes--; 1982 } 1983 fn = fib6_repair_tree(net, table, fn); 1984 } 1985 1986 fib6_purge_rt(rt, fn, net); 1987 1988 if (!info->skip_notify_kernel) { 1989 if (notify_del) 1990 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, 1991 rt, NULL); 1992 else if (replace_rt) 1993 call_fib6_entry_notifiers_replace(net, replace_rt); 1994 } 1995 if (!info->skip_notify) 1996 inet6_rt_notify(RTM_DELROUTE, rt, info, 0); 1997 1998 fib6_info_release(rt); 1999 } 2000 2001 /* Need to own table->tb6_lock */ 2002 int fib6_del(struct fib6_info *rt, struct nl_info *info) 2003 { 2004 struct net *net = info->nl_net; 2005 struct fib6_info __rcu **rtp; 2006 struct fib6_info __rcu **rtp_next; 2007 struct fib6_table *table; 2008 struct fib6_node *fn; 2009 2010 if (rt == net->ipv6.fib6_null_entry) 2011 return -ENOENT; 2012 2013 table = rt->fib6_table; 2014 fn = rcu_dereference_protected(rt->fib6_node, 2015 lockdep_is_held(&table->tb6_lock)); 2016 if (!fn) 2017 return -ENOENT; 2018 2019 WARN_ON(!(fn->fn_flags & RTN_RTINFO)); 2020 2021 /* 2022 * Walk the leaf entries looking for ourself 2023 */ 2024 2025 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) { 2026 struct fib6_info *cur = rcu_dereference_protected(*rtp, 2027 lockdep_is_held(&table->tb6_lock)); 2028 if (rt == cur) { 2029 if (fib6_requires_src(cur)) 2030 fib6_routes_require_src_dec(info->nl_net); 2031 fib6_del_route(table, fn, rtp, info); 2032 return 0; 2033 } 2034 rtp_next = &cur->fib6_next; 2035 } 2036 return -ENOENT; 2037 } 2038 2039 /* 2040 * Tree traversal function. 2041 * 2042 * Certainly, it is not interrupt safe. 2043 * However, it is internally reenterable wrt itself and fib6_add/fib6_del. 2044 * It means, that we can modify tree during walking 2045 * and use this function for garbage collection, clone pruning, 2046 * cleaning tree when a device goes down etc. etc. 2047 * 2048 * It guarantees that every node will be traversed, 2049 * and that it will be traversed only once. 2050 * 2051 * Callback function w->func may return: 2052 * 0 -> continue walking. 2053 * positive value -> walking is suspended (used by tree dumps, 2054 * and probably by gc, if it will be split to several slices) 2055 * negative value -> terminate walking. 2056 * 2057 * The function itself returns: 2058 * 0 -> walk is complete. 2059 * >0 -> walk is incomplete (i.e. suspended) 2060 * <0 -> walk is terminated by an error. 2061 * 2062 * This function is called with tb6_lock held. 2063 */ 2064 2065 static int fib6_walk_continue(struct fib6_walker *w) 2066 { 2067 struct fib6_node *fn, *pn, *left, *right; 2068 2069 /* w->root should always be table->tb6_root */ 2070 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT)); 2071 2072 for (;;) { 2073 fn = w->node; 2074 if (!fn) 2075 return 0; 2076 2077 switch (w->state) { 2078 #ifdef CONFIG_IPV6_SUBTREES 2079 case FWS_S: 2080 if (FIB6_SUBTREE(fn)) { 2081 w->node = FIB6_SUBTREE(fn); 2082 continue; 2083 } 2084 w->state = FWS_L; 2085 fallthrough; 2086 #endif 2087 case FWS_L: 2088 left = rcu_dereference_protected(fn->left, 1); 2089 if (left) { 2090 w->node = left; 2091 w->state = FWS_INIT; 2092 continue; 2093 } 2094 w->state = FWS_R; 2095 fallthrough; 2096 case FWS_R: 2097 right = rcu_dereference_protected(fn->right, 1); 2098 if (right) { 2099 w->node = right; 2100 w->state = FWS_INIT; 2101 continue; 2102 } 2103 w->state = FWS_C; 2104 w->leaf = rcu_dereference_protected(fn->leaf, 1); 2105 fallthrough; 2106 case FWS_C: 2107 if (w->leaf && fn->fn_flags & RTN_RTINFO) { 2108 int err; 2109 2110 if (w->skip) { 2111 w->skip--; 2112 goto skip; 2113 } 2114 2115 err = w->func(w); 2116 if (err) 2117 return err; 2118 2119 w->count++; 2120 continue; 2121 } 2122 skip: 2123 w->state = FWS_U; 2124 fallthrough; 2125 case FWS_U: 2126 if (fn == w->root) 2127 return 0; 2128 pn = rcu_dereference_protected(fn->parent, 1); 2129 left = rcu_dereference_protected(pn->left, 1); 2130 right = rcu_dereference_protected(pn->right, 1); 2131 w->node = pn; 2132 #ifdef CONFIG_IPV6_SUBTREES 2133 if (FIB6_SUBTREE(pn) == fn) { 2134 WARN_ON(!(fn->fn_flags & RTN_ROOT)); 2135 w->state = FWS_L; 2136 continue; 2137 } 2138 #endif 2139 if (left == fn) { 2140 w->state = FWS_R; 2141 continue; 2142 } 2143 if (right == fn) { 2144 w->state = FWS_C; 2145 w->leaf = rcu_dereference_protected(w->node->leaf, 1); 2146 continue; 2147 } 2148 #if RT6_DEBUG >= 2 2149 WARN_ON(1); 2150 #endif 2151 } 2152 } 2153 } 2154 2155 static int fib6_walk(struct net *net, struct fib6_walker *w) 2156 { 2157 int res; 2158 2159 w->state = FWS_INIT; 2160 w->node = w->root; 2161 2162 fib6_walker_link(net, w); 2163 res = fib6_walk_continue(w); 2164 if (res <= 0) 2165 fib6_walker_unlink(net, w); 2166 return res; 2167 } 2168 2169 static int fib6_clean_node(struct fib6_walker *w) 2170 { 2171 int res; 2172 struct fib6_info *rt; 2173 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w); 2174 struct nl_info info = { 2175 .nl_net = c->net, 2176 .skip_notify = c->skip_notify, 2177 }; 2178 2179 if (c->sernum != FIB6_NO_SERNUM_CHANGE && 2180 READ_ONCE(w->node->fn_sernum) != c->sernum) 2181 WRITE_ONCE(w->node->fn_sernum, c->sernum); 2182 2183 if (!c->func) { 2184 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE); 2185 w->leaf = NULL; 2186 return 0; 2187 } 2188 2189 for_each_fib6_walker_rt(w) { 2190 res = c->func(rt, c->arg); 2191 if (res == -1) { 2192 w->leaf = rt; 2193 res = fib6_del(rt, &info); 2194 if (res) { 2195 #if RT6_DEBUG >= 2 2196 pr_debug("%s: del failed: rt=%p@%p err=%d\n", 2197 __func__, rt, 2198 rcu_access_pointer(rt->fib6_node), 2199 res); 2200 #endif 2201 continue; 2202 } 2203 return 0; 2204 } else if (res == -2) { 2205 if (WARN_ON(!rt->fib6_nsiblings)) 2206 continue; 2207 rt = list_last_entry(&rt->fib6_siblings, 2208 struct fib6_info, fib6_siblings); 2209 continue; 2210 } 2211 WARN_ON(res != 0); 2212 } 2213 w->leaf = rt; 2214 return 0; 2215 } 2216 2217 /* 2218 * Convenient frontend to tree walker. 2219 * 2220 * func is called on each route. 2221 * It may return -2 -> skip multipath route. 2222 * -1 -> delete this route. 2223 * 0 -> continue walking 2224 */ 2225 2226 static void fib6_clean_tree(struct net *net, struct fib6_node *root, 2227 int (*func)(struct fib6_info *, void *arg), 2228 int sernum, void *arg, bool skip_notify) 2229 { 2230 struct fib6_cleaner c; 2231 2232 c.w.root = root; 2233 c.w.func = fib6_clean_node; 2234 c.w.count = 0; 2235 c.w.skip = 0; 2236 c.w.skip_in_node = 0; 2237 c.func = func; 2238 c.sernum = sernum; 2239 c.arg = arg; 2240 c.net = net; 2241 c.skip_notify = skip_notify; 2242 2243 fib6_walk(net, &c.w); 2244 } 2245 2246 static void __fib6_clean_all(struct net *net, 2247 int (*func)(struct fib6_info *, void *), 2248 int sernum, void *arg, bool skip_notify) 2249 { 2250 struct fib6_table *table; 2251 struct hlist_head *head; 2252 unsigned int h; 2253 2254 rcu_read_lock(); 2255 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { 2256 head = &net->ipv6.fib_table_hash[h]; 2257 hlist_for_each_entry_rcu(table, head, tb6_hlist) { 2258 spin_lock_bh(&table->tb6_lock); 2259 fib6_clean_tree(net, &table->tb6_root, 2260 func, sernum, arg, skip_notify); 2261 spin_unlock_bh(&table->tb6_lock); 2262 } 2263 } 2264 rcu_read_unlock(); 2265 } 2266 2267 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *), 2268 void *arg) 2269 { 2270 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false); 2271 } 2272 2273 void fib6_clean_all_skip_notify(struct net *net, 2274 int (*func)(struct fib6_info *, void *), 2275 void *arg) 2276 { 2277 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true); 2278 } 2279 2280 static void fib6_flush_trees(struct net *net) 2281 { 2282 int new_sernum = fib6_new_sernum(net); 2283 2284 __fib6_clean_all(net, NULL, new_sernum, NULL, false); 2285 } 2286 2287 /* 2288 * Garbage collection 2289 */ 2290 2291 static int fib6_age(struct fib6_info *rt, void *arg) 2292 { 2293 struct fib6_gc_args *gc_args = arg; 2294 unsigned long now = jiffies; 2295 2296 /* 2297 * check addrconf expiration here. 2298 * Routes are expired even if they are in use. 2299 */ 2300 2301 if (rt->fib6_flags & RTF_EXPIRES && rt->expires) { 2302 if (time_after(now, rt->expires)) { 2303 RT6_TRACE("expiring %p\n", rt); 2304 return -1; 2305 } 2306 gc_args->more++; 2307 } 2308 2309 /* Also age clones in the exception table. 2310 * Note, that clones are aged out 2311 * only if they are not in use now. 2312 */ 2313 rt6_age_exceptions(rt, gc_args, now); 2314 2315 return 0; 2316 } 2317 2318 void fib6_run_gc(unsigned long expires, struct net *net, bool force) 2319 { 2320 struct fib6_gc_args gc_args; 2321 unsigned long now; 2322 2323 if (force) { 2324 spin_lock_bh(&net->ipv6.fib6_gc_lock); 2325 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) { 2326 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); 2327 return; 2328 } 2329 gc_args.timeout = expires ? (int)expires : 2330 net->ipv6.sysctl.ip6_rt_gc_interval; 2331 gc_args.more = 0; 2332 2333 fib6_clean_all(net, fib6_age, &gc_args); 2334 now = jiffies; 2335 net->ipv6.ip6_rt_last_gc = now; 2336 2337 if (gc_args.more) 2338 mod_timer(&net->ipv6.ip6_fib_timer, 2339 round_jiffies(now 2340 + net->ipv6.sysctl.ip6_rt_gc_interval)); 2341 else 2342 del_timer(&net->ipv6.ip6_fib_timer); 2343 spin_unlock_bh(&net->ipv6.fib6_gc_lock); 2344 } 2345 2346 static void fib6_gc_timer_cb(struct timer_list *t) 2347 { 2348 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer); 2349 2350 fib6_run_gc(0, arg, true); 2351 } 2352 2353 static int __net_init fib6_net_init(struct net *net) 2354 { 2355 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; 2356 int err; 2357 2358 err = fib6_notifier_init(net); 2359 if (err) 2360 return err; 2361 2362 /* Default to 3-tuple */ 2363 net->ipv6.sysctl.multipath_hash_fields = 2364 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK; 2365 2366 spin_lock_init(&net->ipv6.fib6_gc_lock); 2367 rwlock_init(&net->ipv6.fib6_walker_lock); 2368 INIT_LIST_HEAD(&net->ipv6.fib6_walkers); 2369 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0); 2370 2371 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); 2372 if (!net->ipv6.rt6_stats) 2373 goto out_notifier; 2374 2375 /* Avoid false sharing : Use at least a full cache line */ 2376 size = max_t(size_t, size, L1_CACHE_BYTES); 2377 2378 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); 2379 if (!net->ipv6.fib_table_hash) 2380 goto out_rt6_stats; 2381 2382 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), 2383 GFP_KERNEL); 2384 if (!net->ipv6.fib6_main_tbl) 2385 goto out_fib_table_hash; 2386 2387 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; 2388 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf, 2389 net->ipv6.fib6_null_entry); 2390 net->ipv6.fib6_main_tbl->tb6_root.fn_flags = 2391 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 2392 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); 2393 2394 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 2395 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), 2396 GFP_KERNEL); 2397 if (!net->ipv6.fib6_local_tbl) 2398 goto out_fib6_main_tbl; 2399 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; 2400 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf, 2401 net->ipv6.fib6_null_entry); 2402 net->ipv6.fib6_local_tbl->tb6_root.fn_flags = 2403 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; 2404 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); 2405 #endif 2406 fib6_tables_init(net); 2407 2408 return 0; 2409 2410 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 2411 out_fib6_main_tbl: 2412 kfree(net->ipv6.fib6_main_tbl); 2413 #endif 2414 out_fib_table_hash: 2415 kfree(net->ipv6.fib_table_hash); 2416 out_rt6_stats: 2417 kfree(net->ipv6.rt6_stats); 2418 out_notifier: 2419 fib6_notifier_exit(net); 2420 return -ENOMEM; 2421 } 2422 2423 static void fib6_net_exit(struct net *net) 2424 { 2425 unsigned int i; 2426 2427 del_timer_sync(&net->ipv6.ip6_fib_timer); 2428 2429 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) { 2430 struct hlist_head *head = &net->ipv6.fib_table_hash[i]; 2431 struct hlist_node *tmp; 2432 struct fib6_table *tb; 2433 2434 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) { 2435 hlist_del(&tb->tb6_hlist); 2436 fib6_free_table(tb); 2437 } 2438 } 2439 2440 kfree(net->ipv6.fib_table_hash); 2441 kfree(net->ipv6.rt6_stats); 2442 fib6_notifier_exit(net); 2443 } 2444 2445 static struct pernet_operations fib6_net_ops = { 2446 .init = fib6_net_init, 2447 .exit = fib6_net_exit, 2448 }; 2449 2450 int __init fib6_init(void) 2451 { 2452 int ret = -ENOMEM; 2453 2454 fib6_node_kmem = kmem_cache_create("fib6_nodes", 2455 sizeof(struct fib6_node), 0, 2456 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, 2457 NULL); 2458 if (!fib6_node_kmem) 2459 goto out; 2460 2461 ret = register_pernet_subsys(&fib6_net_ops); 2462 if (ret) 2463 goto out_kmem_cache_create; 2464 2465 ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL, 2466 inet6_dump_fib, 0); 2467 if (ret) 2468 goto out_unregister_subsys; 2469 2470 __fib6_flush_trees = fib6_flush_trees; 2471 out: 2472 return ret; 2473 2474 out_unregister_subsys: 2475 unregister_pernet_subsys(&fib6_net_ops); 2476 out_kmem_cache_create: 2477 kmem_cache_destroy(fib6_node_kmem); 2478 goto out; 2479 } 2480 2481 void fib6_gc_cleanup(void) 2482 { 2483 unregister_pernet_subsys(&fib6_net_ops); 2484 kmem_cache_destroy(fib6_node_kmem); 2485 } 2486 2487 #ifdef CONFIG_PROC_FS 2488 static int ipv6_route_native_seq_show(struct seq_file *seq, void *v) 2489 { 2490 struct fib6_info *rt = v; 2491 struct ipv6_route_iter *iter = seq->private; 2492 struct fib6_nh *fib6_nh = rt->fib6_nh; 2493 unsigned int flags = rt->fib6_flags; 2494 const struct net_device *dev; 2495 2496 if (rt->nh) 2497 fib6_nh = nexthop_fib6_nh(rt->nh); 2498 2499 seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen); 2500 2501 #ifdef CONFIG_IPV6_SUBTREES 2502 seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen); 2503 #else 2504 seq_puts(seq, "00000000000000000000000000000000 00 "); 2505 #endif 2506 if (fib6_nh->fib_nh_gw_family) { 2507 flags |= RTF_GATEWAY; 2508 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6); 2509 } else { 2510 seq_puts(seq, "00000000000000000000000000000000"); 2511 } 2512 2513 dev = fib6_nh->fib_nh_dev; 2514 seq_printf(seq, " %08x %08x %08x %08x %8s\n", 2515 rt->fib6_metric, refcount_read(&rt->fib6_ref), 0, 2516 flags, dev ? dev->name : ""); 2517 iter->w.leaf = NULL; 2518 return 0; 2519 } 2520 2521 static int ipv6_route_yield(struct fib6_walker *w) 2522 { 2523 struct ipv6_route_iter *iter = w->args; 2524 2525 if (!iter->skip) 2526 return 1; 2527 2528 do { 2529 iter->w.leaf = rcu_dereference_protected( 2530 iter->w.leaf->fib6_next, 2531 lockdep_is_held(&iter->tbl->tb6_lock)); 2532 iter->skip--; 2533 if (!iter->skip && iter->w.leaf) 2534 return 1; 2535 } while (iter->w.leaf); 2536 2537 return 0; 2538 } 2539 2540 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter, 2541 struct net *net) 2542 { 2543 memset(&iter->w, 0, sizeof(iter->w)); 2544 iter->w.func = ipv6_route_yield; 2545 iter->w.root = &iter->tbl->tb6_root; 2546 iter->w.state = FWS_INIT; 2547 iter->w.node = iter->w.root; 2548 iter->w.args = iter; 2549 iter->sernum = READ_ONCE(iter->w.root->fn_sernum); 2550 INIT_LIST_HEAD(&iter->w.lh); 2551 fib6_walker_link(net, &iter->w); 2552 } 2553 2554 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl, 2555 struct net *net) 2556 { 2557 unsigned int h; 2558 struct hlist_node *node; 2559 2560 if (tbl) { 2561 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1; 2562 node = rcu_dereference(hlist_next_rcu(&tbl->tb6_hlist)); 2563 } else { 2564 h = 0; 2565 node = NULL; 2566 } 2567 2568 while (!node && h < FIB6_TABLE_HASHSZ) { 2569 node = rcu_dereference( 2570 hlist_first_rcu(&net->ipv6.fib_table_hash[h++])); 2571 } 2572 return hlist_entry_safe(node, struct fib6_table, tb6_hlist); 2573 } 2574 2575 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter) 2576 { 2577 int sernum = READ_ONCE(iter->w.root->fn_sernum); 2578 2579 if (iter->sernum != sernum) { 2580 iter->sernum = sernum; 2581 iter->w.state = FWS_INIT; 2582 iter->w.node = iter->w.root; 2583 WARN_ON(iter->w.skip); 2584 iter->w.skip = iter->w.count; 2585 } 2586 } 2587 2588 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2589 { 2590 int r; 2591 struct fib6_info *n; 2592 struct net *net = seq_file_net(seq); 2593 struct ipv6_route_iter *iter = seq->private; 2594 2595 ++(*pos); 2596 if (!v) 2597 goto iter_table; 2598 2599 n = rcu_dereference(((struct fib6_info *)v)->fib6_next); 2600 if (n) 2601 return n; 2602 2603 iter_table: 2604 ipv6_route_check_sernum(iter); 2605 spin_lock_bh(&iter->tbl->tb6_lock); 2606 r = fib6_walk_continue(&iter->w); 2607 spin_unlock_bh(&iter->tbl->tb6_lock); 2608 if (r > 0) { 2609 return iter->w.leaf; 2610 } else if (r < 0) { 2611 fib6_walker_unlink(net, &iter->w); 2612 return NULL; 2613 } 2614 fib6_walker_unlink(net, &iter->w); 2615 2616 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net); 2617 if (!iter->tbl) 2618 return NULL; 2619 2620 ipv6_route_seq_setup_walk(iter, net); 2621 goto iter_table; 2622 } 2623 2624 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos) 2625 __acquires(RCU) 2626 { 2627 struct net *net = seq_file_net(seq); 2628 struct ipv6_route_iter *iter = seq->private; 2629 2630 rcu_read_lock(); 2631 iter->tbl = ipv6_route_seq_next_table(NULL, net); 2632 iter->skip = *pos; 2633 2634 if (iter->tbl) { 2635 loff_t p = 0; 2636 2637 ipv6_route_seq_setup_walk(iter, net); 2638 return ipv6_route_seq_next(seq, NULL, &p); 2639 } else { 2640 return NULL; 2641 } 2642 } 2643 2644 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter) 2645 { 2646 struct fib6_walker *w = &iter->w; 2647 return w->node && !(w->state == FWS_U && w->node == w->root); 2648 } 2649 2650 static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v) 2651 __releases(RCU) 2652 { 2653 struct net *net = seq_file_net(seq); 2654 struct ipv6_route_iter *iter = seq->private; 2655 2656 if (ipv6_route_iter_active(iter)) 2657 fib6_walker_unlink(net, &iter->w); 2658 2659 rcu_read_unlock(); 2660 } 2661 2662 #if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL) 2663 static int ipv6_route_prog_seq_show(struct bpf_prog *prog, 2664 struct bpf_iter_meta *meta, 2665 void *v) 2666 { 2667 struct bpf_iter__ipv6_route ctx; 2668 2669 ctx.meta = meta; 2670 ctx.rt = v; 2671 return bpf_iter_run_prog(prog, &ctx); 2672 } 2673 2674 static int ipv6_route_seq_show(struct seq_file *seq, void *v) 2675 { 2676 struct ipv6_route_iter *iter = seq->private; 2677 struct bpf_iter_meta meta; 2678 struct bpf_prog *prog; 2679 int ret; 2680 2681 meta.seq = seq; 2682 prog = bpf_iter_get_info(&meta, false); 2683 if (!prog) 2684 return ipv6_route_native_seq_show(seq, v); 2685 2686 ret = ipv6_route_prog_seq_show(prog, &meta, v); 2687 iter->w.leaf = NULL; 2688 2689 return ret; 2690 } 2691 2692 static void ipv6_route_seq_stop(struct seq_file *seq, void *v) 2693 { 2694 struct bpf_iter_meta meta; 2695 struct bpf_prog *prog; 2696 2697 if (!v) { 2698 meta.seq = seq; 2699 prog = bpf_iter_get_info(&meta, true); 2700 if (prog) 2701 (void)ipv6_route_prog_seq_show(prog, &meta, v); 2702 } 2703 2704 ipv6_route_native_seq_stop(seq, v); 2705 } 2706 #else 2707 static int ipv6_route_seq_show(struct seq_file *seq, void *v) 2708 { 2709 return ipv6_route_native_seq_show(seq, v); 2710 } 2711 2712 static void ipv6_route_seq_stop(struct seq_file *seq, void *v) 2713 { 2714 ipv6_route_native_seq_stop(seq, v); 2715 } 2716 #endif 2717 2718 const struct seq_operations ipv6_route_seq_ops = { 2719 .start = ipv6_route_seq_start, 2720 .next = ipv6_route_seq_next, 2721 .stop = ipv6_route_seq_stop, 2722 .show = ipv6_route_seq_show 2723 }; 2724 #endif /* CONFIG_PROC_FS */ 2725