1/*
   2 *	Linux INET6 implementation
   3 *	Forwarding Information Database
   4 *
   5 *	Authors:
   6 *	Pedro Roque		<roque@di.fc.ul.pt>
   7 *
   8 *	This program is free software; you can redistribute it and/or
   9 *      modify it under the terms of the GNU General Public License
  10 *      as published by the Free Software Foundation; either version
  11 *      2 of the License, or (at your option) any later version.
  12 *
  13 *	Changes:
  14 *	Yuji SEKIYA @USAGI:	Support default route on router node;
  15 *				remove ip6_null_entry from the top of
  16 *				routing table.
  17 *	Ville Nuorvala:		Fixed routing subtrees.
  18 */
  19
  20#define pr_fmt(fmt) "IPv6: " fmt
  21
 
 
 
 
 
  22#include <linux/errno.h>
  23#include <linux/types.h>
  24#include <linux/net.h>
  25#include <linux/route.h>
  26#include <linux/netdevice.h>
  27#include <linux/in6.h>
  28#include <linux/init.h>
  29#include <linux/list.h>
  30#include <linux/slab.h>
  31
 
 
 
 
  32#include <net/ipv6.h>
  33#include <net/ndisc.h>
  34#include <net/addrconf.h>
  35#include <net/lwtunnel.h>
  36#include <net/fib_notifier.h>
  37
  38#include <net/ip6_fib.h>
  39#include <net/ip6_route.h>
  40
  41static struct kmem_cache *fib6_node_kmem __read_mostly;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  42
  43struct fib6_cleaner {
  44	struct fib6_walker w;
 
  45	struct net *net;
  46	int (*func)(struct rt6_info *, void *arg);
  47	int sernum;
  48	void *arg;
  49};
  50
 
 
  51#ifdef CONFIG_IPV6_SUBTREES
  52#define FWS_INIT FWS_S
  53#else
  54#define FWS_INIT FWS_L
  55#endif
  56
  57static struct rt6_info *fib6_find_prefix(struct net *net,
  58					 struct fib6_table *table,
  59					 struct fib6_node *fn);
  60static struct fib6_node *fib6_repair_tree(struct net *net,
  61					  struct fib6_table *table,
  62					  struct fib6_node *fn);
  63static int fib6_walk(struct net *net, struct fib6_walker *w);
  64static 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
  73static 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
  78static 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
  85static 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
  92static 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
 104enum {
 105	FIB6_NO_SERNUM_CHANGE = 0,
 106};
 107
 108void fib6_update_sernum(struct rt6_info *rt)
 109{
 110	struct net *net = dev_net(rt->dst.dev);
 111	struct fib6_node *fn;
 112
 113	fn = rcu_dereference_protected(rt->rt6i_node,
 114			lockdep_is_held(&rt->rt6i_table->tb6_lock));
 115	if (fn)
 116		fn->fn_sernum = fib6_new_sernum(net);
 117}
 118
 119/*
 120 *	Auxiliary address test functions for the radix tree.
 121 *
 122 *	These assume a 32bit processor (although it will work on
 123 *	64bit processors)
 124 */
 125
 126/*
 127 *	test bit
 128 */
 129#if defined(__LITTLE_ENDIAN)
 130# define BITOP_BE32_SWIZZLE	(0x1F & ~7)
 131#else
 132# define BITOP_BE32_SWIZZLE	0
 133#endif
 134
 135static __be32 addr_bit_set(const void *token, int fn_bit)
 136{
 137	const __be32 *addr = token;
 138	/*
 139	 * Here,
 140	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
 141	 * is optimized version of
 142	 *	htonl(1 << ((~fn_bit)&0x1F))
 143	 * See include/asm-generic/bitops/le.h.
 144	 */
 145	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
 146	       addr[fn_bit >> 5];
 147}
 148
 149static struct fib6_node *node_alloc(struct net *net)
 150{
 151	struct fib6_node *fn;
 152
 153	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 154	if (fn)
 155		net->ipv6.rt6_stats->fib_nodes++;
 156
 157	return fn;
 158}
 159
 160static void node_free_immediate(struct net *net, struct fib6_node *fn)
 161{
 162	kmem_cache_free(fib6_node_kmem, fn);
 163	net->ipv6.rt6_stats->fib_nodes--;
 164}
 165
 166static void node_free_rcu(struct rcu_head *head)
 167{
 168	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
 169
 170	kmem_cache_free(fib6_node_kmem, fn);
 171}
 172
 173static void node_free(struct net *net, struct fib6_node *fn)
 174{
 175	call_rcu(&fn->rcu, node_free_rcu);
 176	net->ipv6.rt6_stats->fib_nodes--;
 177}
 178
 179void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
 180{
 181	int cpu;
 182
 183	if (!non_pcpu_rt->rt6i_pcpu)
 184		return;
 185
 186	for_each_possible_cpu(cpu) {
 187		struct rt6_info **ppcpu_rt;
 188		struct rt6_info *pcpu_rt;
 189
 190		ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
 191		pcpu_rt = *ppcpu_rt;
 192		if (pcpu_rt) {
 193			dst_dev_put(&pcpu_rt->dst);
 194			dst_release(&pcpu_rt->dst);
 195			*ppcpu_rt = NULL;
 196		}
 197	}
 198}
 199EXPORT_SYMBOL_GPL(rt6_free_pcpu);
 200
 201static void fib6_free_table(struct fib6_table *table)
 202{
 203	inetpeer_invalidate_tree(&table->tb6_peers);
 204	kfree(table);
 205}
 206
 207static void fib6_link_table(struct net *net, struct fib6_table *tb)
 208{
 209	unsigned int h;
 210
 211	/*
 212	 * Initialize table lock at a single place to give lockdep a key,
 213	 * tables aren't visible prior to being linked to the list.
 214	 */
 215	spin_lock_init(&tb->tb6_lock);
 
 216	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 217
 218	/*
 219	 * No protection necessary, this is the only list mutatation
 220	 * operation, tables never disappear once they exist.
 221	 */
 222	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 223}
 224
 225#ifdef CONFIG_IPV6_MULTIPLE_TABLES
 226
 227static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 228{
 229	struct fib6_table *table;
 230
 231	table = kzalloc(sizeof(*table), GFP_ATOMIC);
 232	if (table) {
 233		table->tb6_id = id;
 234		rcu_assign_pointer(table->tb6_root.leaf,
 235				   net->ipv6.ip6_null_entry);
 236		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
 237		inet_peer_base_init(&table->tb6_peers);
 238	}
 239
 240	return table;
 241}
 242
 243struct fib6_table *fib6_new_table(struct net *net, u32 id)
 244{
 245	struct fib6_table *tb;
 246
 247	if (id == 0)
 248		id = RT6_TABLE_MAIN;
 249	tb = fib6_get_table(net, id);
 250	if (tb)
 251		return tb;
 252
 253	tb = fib6_alloc_table(net, id);
 254	if (tb)
 255		fib6_link_table(net, tb);
 256
 257	return tb;
 258}
 259EXPORT_SYMBOL_GPL(fib6_new_table);
 260
 261struct fib6_table *fib6_get_table(struct net *net, u32 id)
 262{
 263	struct fib6_table *tb;
 264	struct hlist_head *head;
 
 265	unsigned int h;
 266
 267	if (id == 0)
 268		id = RT6_TABLE_MAIN;
 269	h = id & (FIB6_TABLE_HASHSZ - 1);
 270	rcu_read_lock();
 271	head = &net->ipv6.fib_table_hash[h];
 272	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 273		if (tb->tb6_id == id) {
 274			rcu_read_unlock();
 275			return tb;
 276		}
 277	}
 278	rcu_read_unlock();
 279
 280	return NULL;
 281}
 282EXPORT_SYMBOL_GPL(fib6_get_table);
 283
 284static void __net_init fib6_tables_init(struct net *net)
 285{
 286	fib6_link_table(net, net->ipv6.fib6_main_tbl);
 287	fib6_link_table(net, net->ipv6.fib6_local_tbl);
 288}
 289#else
 290
 291struct fib6_table *fib6_new_table(struct net *net, u32 id)
 292{
 293	return fib6_get_table(net, id);
 294}
 295
 296struct fib6_table *fib6_get_table(struct net *net, u32 id)
 297{
 298	  return net->ipv6.fib6_main_tbl;
 299}
 300
 301struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 302				   const struct sk_buff *skb,
 303				   int flags, pol_lookup_t lookup)
 304{
 305	struct rt6_info *rt;
 306
 307	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
 308	if (rt->dst.error == -EAGAIN) {
 309		ip6_rt_put(rt);
 310		rt = net->ipv6.ip6_null_entry;
 311		dst_hold(&rt->dst);
 312	}
 313
 314	return &rt->dst;
 315}
 316
 317static void __net_init fib6_tables_init(struct net *net)
 318{
 319	fib6_link_table(net, net->ipv6.fib6_main_tbl);
 320}
 321
 322#endif
 323
 324unsigned int fib6_tables_seq_read(struct net *net)
 325{
 326	unsigned int h, fib_seq = 0;
 327
 328	rcu_read_lock();
 329	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 330		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 331		struct fib6_table *tb;
 332
 333		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 334			fib_seq += tb->fib_seq;
 335	}
 336	rcu_read_unlock();
 337
 338	return fib_seq;
 339}
 340
 341static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
 342				    enum fib_event_type event_type,
 343				    struct rt6_info *rt)
 344{
 345	struct fib6_entry_notifier_info info = {
 346		.rt = rt,
 347	};
 348
 349	return call_fib6_notifier(nb, net, event_type, &info.info);
 350}
 351
 352static int call_fib6_entry_notifiers(struct net *net,
 353				     enum fib_event_type event_type,
 354				     struct rt6_info *rt,
 355				     struct netlink_ext_ack *extack)
 356{
 357	struct fib6_entry_notifier_info info = {
 358		.info.extack = extack,
 359		.rt = rt,
 360	};
 361
 362	rt->rt6i_table->fib_seq++;
 363	return call_fib6_notifiers(net, event_type, &info.info);
 364}
 365
 366struct fib6_dump_arg {
 367	struct net *net;
 368	struct notifier_block *nb;
 369};
 370
 371static void fib6_rt_dump(struct rt6_info *rt, struct fib6_dump_arg *arg)
 372{
 373	if (rt == arg->net->ipv6.ip6_null_entry)
 374		return;
 375	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
 376}
 377
 378static int fib6_node_dump(struct fib6_walker *w)
 379{
 380	struct rt6_info *rt;
 381
 382	for_each_fib6_walker_rt(w)
 383		fib6_rt_dump(rt, w->args);
 384	w->leaf = NULL;
 385	return 0;
 386}
 387
 388static void fib6_table_dump(struct net *net, struct fib6_table *tb,
 389			    struct fib6_walker *w)
 390{
 391	w->root = &tb->tb6_root;
 392	spin_lock_bh(&tb->tb6_lock);
 393	fib6_walk(net, w);
 394	spin_unlock_bh(&tb->tb6_lock);
 395}
 396
 397/* Called with rcu_read_lock() */
 398int fib6_tables_dump(struct net *net, struct notifier_block *nb)
 399{
 400	struct fib6_dump_arg arg;
 401	struct fib6_walker *w;
 402	unsigned int h;
 403
 404	w = kzalloc(sizeof(*w), GFP_ATOMIC);
 405	if (!w)
 406		return -ENOMEM;
 407
 408	w->func = fib6_node_dump;
 409	arg.net = net;
 410	arg.nb = nb;
 411	w->args = &arg;
 412
 413	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 414		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 415		struct fib6_table *tb;
 416
 417		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 418			fib6_table_dump(net, tb, w);
 419	}
 420
 421	kfree(w);
 422
 423	return 0;
 424}
 425
 426static int fib6_dump_node(struct fib6_walker *w)
 427{
 428	int res;
 429	struct rt6_info *rt;
 430
 431	for_each_fib6_walker_rt(w) {
 432		res = rt6_dump_route(rt, w->args);
 433		if (res < 0) {
 434			/* Frame is full, suspend walking */
 435			w->leaf = rt;
 436			return 1;
 437		}
 438
 439		/* Multipath routes are dumped in one route with the
 440		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
 441		 * last sibling of this route (no need to dump the
 442		 * sibling routes again)
 443		 */
 444		if (rt->rt6i_nsiblings)
 445			rt = list_last_entry(&rt->rt6i_siblings,
 446					     struct rt6_info,
 447					     rt6i_siblings);
 448	}
 449	w->leaf = NULL;
 450	return 0;
 451}
 452
 453static void fib6_dump_end(struct netlink_callback *cb)
 454{
 455	struct net *net = sock_net(cb->skb->sk);
 456	struct fib6_walker *w = (void *)cb->args[2];
 457
 458	if (w) {
 459		if (cb->args[4]) {
 460			cb->args[4] = 0;
 461			fib6_walker_unlink(net, w);
 462		}
 463		cb->args[2] = 0;
 464		kfree(w);
 465	}
 466	cb->done = (void *)cb->args[3];
 467	cb->args[1] = 3;
 468}
 469
 470static int fib6_dump_done(struct netlink_callback *cb)
 471{
 472	fib6_dump_end(cb);
 473	return cb->done ? cb->done(cb) : 0;
 474}
 475
 476static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 477			   struct netlink_callback *cb)
 478{
 479	struct net *net = sock_net(skb->sk);
 480	struct fib6_walker *w;
 481	int res;
 482
 483	w = (void *)cb->args[2];
 484	w->root = &table->tb6_root;
 485
 486	if (cb->args[4] == 0) {
 487		w->count = 0;
 488		w->skip = 0;
 489
 490		spin_lock_bh(&table->tb6_lock);
 491		res = fib6_walk(net, w);
 492		spin_unlock_bh(&table->tb6_lock);
 493		if (res > 0) {
 494			cb->args[4] = 1;
 495			cb->args[5] = w->root->fn_sernum;
 496		}
 497	} else {
 498		if (cb->args[5] != w->root->fn_sernum) {
 499			/* Begin at the root if the tree changed */
 500			cb->args[5] = w->root->fn_sernum;
 501			w->state = FWS_INIT;
 502			w->node = w->root;
 503			w->skip = w->count;
 504		} else
 505			w->skip = 0;
 506
 507		spin_lock_bh(&table->tb6_lock);
 508		res = fib6_walk_continue(w);
 509		spin_unlock_bh(&table->tb6_lock);
 510		if (res <= 0) {
 511			fib6_walker_unlink(net, w);
 512			cb->args[4] = 0;
 513		}
 514	}
 515
 516	return res;
 517}
 518
 519static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 520{
 521	struct net *net = sock_net(skb->sk);
 522	unsigned int h, s_h;
 523	unsigned int e = 0, s_e;
 524	struct rt6_rtnl_dump_arg arg;
 525	struct fib6_walker *w;
 526	struct fib6_table *tb;
 
 527	struct hlist_head *head;
 528	int res = 0;
 529
 530	s_h = cb->args[0];
 531	s_e = cb->args[1];
 532
 533	w = (void *)cb->args[2];
 534	if (!w) {
 535		/* New dump:
 536		 *
 537		 * 1. hook callback destructor.
 538		 */
 539		cb->args[3] = (long)cb->done;
 540		cb->done = fib6_dump_done;
 541
 542		/*
 543		 * 2. allocate and initialize walker.
 544		 */
 545		w = kzalloc(sizeof(*w), GFP_ATOMIC);
 546		if (!w)
 547			return -ENOMEM;
 548		w->func = fib6_dump_node;
 549		cb->args[2] = (long)w;
 550	}
 551
 552	arg.skb = skb;
 553	arg.cb = cb;
 554	arg.net = net;
 555	w->args = &arg;
 556
 557	rcu_read_lock();
 558	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 559		e = 0;
 560		head = &net->ipv6.fib_table_hash[h];
 561		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 562			if (e < s_e)
 563				goto next;
 564			res = fib6_dump_table(tb, skb, cb);
 565			if (res != 0)
 566				goto out;
 567next:
 568			e++;
 569		}
 570	}
 571out:
 572	rcu_read_unlock();
 573	cb->args[1] = e;
 574	cb->args[0] = h;
 575
 576	res = res < 0 ? res : skb->len;
 577	if (res <= 0)
 578		fib6_dump_end(cb);
 579	return res;
 580}
 581
 582/*
 583 *	Routing Table
 584 *
 585 *	return the appropriate node for a routing tree "add" operation
 586 *	by either creating and inserting or by returning an existing
 587 *	node.
 588 */
 589
 590static struct fib6_node *fib6_add_1(struct net *net,
 591				    struct fib6_table *table,
 592				    struct fib6_node *root,
 593				    struct in6_addr *addr, int plen,
 594				    int offset, int allow_create,
 595				    int replace_required,
 596				    struct netlink_ext_ack *extack)
 597{
 598	struct fib6_node *fn, *in, *ln;
 599	struct fib6_node *pn = NULL;
 600	struct rt6key *key;
 601	int	bit;
 602	__be32	dir = 0;
 
 603
 604	RT6_TRACE("fib6_add_1\n");
 605
 606	/* insert node in tree */
 607
 608	fn = root;
 609
 610	do {
 611		struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
 612					    lockdep_is_held(&table->tb6_lock));
 613		key = (struct rt6key *)((u8 *)leaf + offset);
 614
 615		/*
 616		 *	Prefix match
 617		 */
 618		if (plen < fn->fn_bit ||
 619		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 620			if (!allow_create) {
 621				if (replace_required) {
 622					NL_SET_ERR_MSG(extack,
 623						       "Can not replace route - no match found");
 624					pr_warn("Can't replace route, no match found\n");
 625					return ERR_PTR(-ENOENT);
 626				}
 627				pr_warn("NLM_F_CREATE should be set when creating new route\n");
 628			}
 629			goto insert_above;
 630		}
 631
 632		/*
 633		 *	Exact match ?
 634		 */
 635
 636		if (plen == fn->fn_bit) {
 637			/* clean up an intermediate node */
 638			if (!(fn->fn_flags & RTN_RTINFO)) {
 639				RCU_INIT_POINTER(fn->leaf, NULL);
 640				rt6_release(leaf);
 641			/* remove null_entry in the root node */
 642			} else if (fn->fn_flags & RTN_TL_ROOT &&
 643				   rcu_access_pointer(fn->leaf) ==
 644				   net->ipv6.ip6_null_entry) {
 645				RCU_INIT_POINTER(fn->leaf, NULL);
 646			}
 647
 
 
 648			return fn;
 649		}
 650
 651		/*
 652		 *	We have more bits to go
 653		 */
 654
 655		/* Try to walk down on tree. */
 
 656		dir = addr_bit_set(addr, fn->fn_bit);
 657		pn = fn;
 658		fn = dir ?
 659		     rcu_dereference_protected(fn->right,
 660					lockdep_is_held(&table->tb6_lock)) :
 661		     rcu_dereference_protected(fn->left,
 662					lockdep_is_held(&table->tb6_lock));
 663	} while (fn);
 664
 665	if (!allow_create) {
 666		/* We should not create new node because
 667		 * NLM_F_REPLACE was specified without NLM_F_CREATE
 668		 * I assume it is safe to require NLM_F_CREATE when
 669		 * REPLACE flag is used! Later we may want to remove the
 670		 * check for replace_required, because according
 671		 * to netlink specification, NLM_F_CREATE
 672		 * MUST be specified if new route is created.
 673		 * That would keep IPv6 consistent with IPv4
 674		 */
 675		if (replace_required) {
 676			NL_SET_ERR_MSG(extack,
 677				       "Can not replace route - no match found");
 678			pr_warn("Can't replace route, no match found\n");
 679			return ERR_PTR(-ENOENT);
 680		}
 681		pr_warn("NLM_F_CREATE should be set when creating new route\n");
 682	}
 683	/*
 684	 *	We walked to the bottom of tree.
 685	 *	Create new leaf node without children.
 686	 */
 687
 688	ln = node_alloc(net);
 689
 690	if (!ln)
 691		return ERR_PTR(-ENOMEM);
 692	ln->fn_bit = plen;
 693	RCU_INIT_POINTER(ln->parent, pn);
 
 
 694
 695	if (dir)
 696		rcu_assign_pointer(pn->right, ln);
 697	else
 698		rcu_assign_pointer(pn->left, ln);
 699
 700	return ln;
 701
 702
 703insert_above:
 704	/*
 705	 * split since we don't have a common prefix anymore or
 706	 * we have a less significant route.
 707	 * we've to insert an intermediate node on the list
 708	 * this new node will point to the one we need to create
 709	 * and the current
 710	 */
 711
 712	pn = rcu_dereference_protected(fn->parent,
 713				       lockdep_is_held(&table->tb6_lock));
 714
 715	/* find 1st bit in difference between the 2 addrs.
 716
 717	   See comment in __ipv6_addr_diff: bit may be an invalid value,
 718	   but if it is >= plen, the value is ignored in any case.
 719	 */
 720
 721	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 722
 723	/*
 724	 *		(intermediate)[in]
 725	 *	          /	   \
 726	 *	(new leaf node)[ln] (old node)[fn]
 727	 */
 728	if (plen > bit) {
 729		in = node_alloc(net);
 730		ln = node_alloc(net);
 731
 732		if (!in || !ln) {
 733			if (in)
 734				node_free_immediate(net, in);
 735			if (ln)
 736				node_free_immediate(net, ln);
 737			return ERR_PTR(-ENOMEM);
 738		}
 739
 740		/*
 741		 * new intermediate node.
 742		 * RTN_RTINFO will
 743		 * be off since that an address that chooses one of
 744		 * the branches would not match less specific routes
 745		 * in the other branch
 746		 */
 747
 748		in->fn_bit = bit;
 749
 750		RCU_INIT_POINTER(in->parent, pn);
 751		in->leaf = fn->leaf;
 752		atomic_inc(&rcu_dereference_protected(in->leaf,
 753				lockdep_is_held(&table->tb6_lock))->rt6i_ref);
 
 754
 755		/* update parent pointer */
 756		if (dir)
 757			rcu_assign_pointer(pn->right, in);
 758		else
 759			rcu_assign_pointer(pn->left, in);
 760
 761		ln->fn_bit = plen;
 762
 763		RCU_INIT_POINTER(ln->parent, in);
 764		rcu_assign_pointer(fn->parent, in);
 
 
 765
 766		if (addr_bit_set(addr, bit)) {
 767			rcu_assign_pointer(in->right, ln);
 768			rcu_assign_pointer(in->left, fn);
 769		} else {
 770			rcu_assign_pointer(in->left, ln);
 771			rcu_assign_pointer(in->right, fn);
 772		}
 773	} else { /* plen <= bit */
 774
 775		/*
 776		 *		(new leaf node)[ln]
 777		 *	          /	   \
 778		 *	     (old node)[fn] NULL
 779		 */
 780
 781		ln = node_alloc(net);
 782
 783		if (!ln)
 784			return ERR_PTR(-ENOMEM);
 785
 786		ln->fn_bit = plen;
 787
 788		RCU_INIT_POINTER(ln->parent, pn);
 789
 790		if (addr_bit_set(&key->addr, plen))
 791			RCU_INIT_POINTER(ln->right, fn);
 792		else
 793			RCU_INIT_POINTER(ln->left, fn);
 794
 795		rcu_assign_pointer(fn->parent, ln);
 796
 797		if (dir)
 798			rcu_assign_pointer(pn->right, ln);
 799		else
 800			rcu_assign_pointer(pn->left, ln);
 801	}
 802	return ln;
 803}
 804
 805static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
 806{
 807	int i;
 808
 809	for (i = 0; i < RTAX_MAX; i++) {
 810		if (test_bit(i, mxc->mx_valid))
 811			mp[i] = mxc->mx[i];
 812	}
 813}
 814
 815static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
 816{
 817	if (!mxc->mx)
 818		return 0;
 819
 820	if (dst->flags & DST_HOST) {
 821		u32 *mp = dst_metrics_write_ptr(dst);
 822
 823		if (unlikely(!mp))
 824			return -ENOMEM;
 825
 826		fib6_copy_metrics(mp, mxc);
 827	} else {
 828		dst_init_metrics(dst, mxc->mx, false);
 829
 830		/* We've stolen mx now. */
 831		mxc->mx = NULL;
 832	}
 833
 834	return 0;
 835}
 836
 837static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
 838			  struct net *net)
 839{
 840	struct fib6_table *table = rt->rt6i_table;
 841
 842	if (atomic_read(&rt->rt6i_ref) != 1) {
 843		/* This route is used as dummy address holder in some split
 844		 * nodes. It is not leaked, but it still holds other resources,
 845		 * which must be released in time. So, scan ascendant nodes
 846		 * and replace dummy references to this route with references
 847		 * to still alive ones.
 848		 */
 849		while (fn) {
 850			struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
 851					    lockdep_is_held(&table->tb6_lock));
 852			struct rt6_info *new_leaf;
 853			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
 854				new_leaf = fib6_find_prefix(net, table, fn);
 855				atomic_inc(&new_leaf->rt6i_ref);
 856				rcu_assign_pointer(fn->leaf, new_leaf);
 857				rt6_release(rt);
 858			}
 859			fn = rcu_dereference_protected(fn->parent,
 860				    lockdep_is_held(&table->tb6_lock));
 861		}
 862	}
 
 863}
 864
 865/*
 866 *	Insert routing information in a node.
 867 */
 868
 869static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
 870			    struct nl_info *info, struct mx6_config *mxc,
 871			    struct netlink_ext_ack *extack)
 872{
 873	struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
 874				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
 875	struct rt6_info *iter = NULL;
 876	struct rt6_info __rcu **ins;
 877	struct rt6_info __rcu **fallback_ins = NULL;
 878	int replace = (info->nlh &&
 879		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
 880	int add = (!info->nlh ||
 881		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
 882	int found = 0;
 883	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
 884	u16 nlflags = NLM_F_EXCL;
 885	int err;
 886
 887	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
 888		nlflags |= NLM_F_APPEND;
 889
 890	ins = &fn->leaf;
 891
 892	for (iter = leaf; iter;
 893	     iter = rcu_dereference_protected(iter->rt6_next,
 894				lockdep_is_held(&rt->rt6i_table->tb6_lock))) {
 895		/*
 896		 *	Search for duplicates
 897		 */
 898
 899		if (iter->rt6i_metric == rt->rt6i_metric) {
 900			/*
 901			 *	Same priority level
 902			 */
 903			if (info->nlh &&
 904			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
 905				return -EEXIST;
 906
 907			nlflags &= ~NLM_F_EXCL;
 908			if (replace) {
 909				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
 910					found++;
 911					break;
 912				}
 913				if (rt_can_ecmp)
 914					fallback_ins = fallback_ins ?: ins;
 915				goto next_iter;
 916			}
 917
 918			if (rt6_duplicate_nexthop(iter, rt)) {
 919				if (rt->rt6i_nsiblings)
 920					rt->rt6i_nsiblings = 0;
 921				if (!(iter->rt6i_flags & RTF_EXPIRES))
 922					return -EEXIST;
 923				if (!(rt->rt6i_flags & RTF_EXPIRES))
 924					rt6_clean_expires(iter);
 925				else
 926					rt6_set_expires(iter, rt->dst.expires);
 927				iter->rt6i_pmtu = rt->rt6i_pmtu;
 928				return -EEXIST;
 929			}
 930			/* If we have the same destination and the same metric,
 931			 * but not the same gateway, then the route we try to
 932			 * add is sibling to this route, increment our counter
 933			 * of siblings, and later we will add our route to the
 934			 * list.
 935			 * Only static routes (which don't have flag
 936			 * RTF_EXPIRES) are used for ECMPv6.
 937			 *
 938			 * To avoid long list, we only had siblings if the
 939			 * route have a gateway.
 940			 */
 941			if (rt_can_ecmp &&
 942			    rt6_qualify_for_ecmp(iter))
 943				rt->rt6i_nsiblings++;
 944		}
 945
 946		if (iter->rt6i_metric > rt->rt6i_metric)
 947			break;
 948
 949next_iter:
 950		ins = &iter->rt6_next;
 951	}
 952
 953	if (fallback_ins && !found) {
 954		/* No ECMP-able route found, replace first non-ECMP one */
 955		ins = fallback_ins;
 956		iter = rcu_dereference_protected(*ins,
 957				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
 958		found++;
 959	}
 960
 961	/* Reset round-robin state, if necessary */
 962	if (ins == &fn->leaf)
 963		fn->rr_ptr = NULL;
 964
 965	/* Link this route to others same route. */
 966	if (rt->rt6i_nsiblings) {
 967		unsigned int rt6i_nsiblings;
 968		struct rt6_info *sibling, *temp_sibling;
 969
 970		/* Find the first route that have the same metric */
 971		sibling = leaf;
 972		while (sibling) {
 973			if (sibling->rt6i_metric == rt->rt6i_metric &&
 974			    rt6_qualify_for_ecmp(sibling)) {
 975				list_add_tail(&rt->rt6i_siblings,
 976					      &sibling->rt6i_siblings);
 977				break;
 978			}
 979			sibling = rcu_dereference_protected(sibling->rt6_next,
 980				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
 981		}
 982		/* For each sibling in the list, increment the counter of
 983		 * siblings. BUG() if counters does not match, list of siblings
 984		 * is broken!
 985		 */
 986		rt6i_nsiblings = 0;
 987		list_for_each_entry_safe(sibling, temp_sibling,
 988					 &rt->rt6i_siblings, rt6i_siblings) {
 989			sibling->rt6i_nsiblings++;
 990			BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
 991			rt6i_nsiblings++;
 992		}
 993		BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
 994		rt6_multipath_rebalance(temp_sibling);
 995	}
 996
 997	/*
 998	 *	insert node
 999	 */
1000	if (!replace) {
1001		if (!add)
1002			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1003
1004add:
1005		nlflags |= NLM_F_CREATE;
1006		err = fib6_commit_metrics(&rt->dst, mxc);
1007		if (err)
1008			return err;
1009
1010		err = call_fib6_entry_notifiers(info->nl_net,
1011						FIB_EVENT_ENTRY_ADD,
1012						rt, extack);
1013		if (err)
1014			return err;
1015
1016		rcu_assign_pointer(rt->rt6_next, iter);
1017		atomic_inc(&rt->rt6i_ref);
1018		rcu_assign_pointer(rt->rt6i_node, fn);
1019		rcu_assign_pointer(*ins, rt);
1020		if (!info->skip_notify)
1021			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1022		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1023
1024		if (!(fn->fn_flags & RTN_RTINFO)) {
1025			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1026			fn->fn_flags |= RTN_RTINFO;
1027		}
1028
1029	} else {
1030		int nsiblings;
1031
1032		if (!found) {
1033			if (add)
1034				goto add;
1035			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1036			return -ENOENT;
1037		}
1038
1039		err = fib6_commit_metrics(&rt->dst, mxc);
1040		if (err)
1041			return err;
1042
1043		err = call_fib6_entry_notifiers(info->nl_net,
1044						FIB_EVENT_ENTRY_REPLACE,
1045						rt, extack);
1046		if (err)
1047			return err;
1048
1049		atomic_inc(&rt->rt6i_ref);
1050		rcu_assign_pointer(rt->rt6i_node, fn);
1051		rt->rt6_next = iter->rt6_next;
1052		rcu_assign_pointer(*ins, rt);
1053		if (!info->skip_notify)
1054			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1055		if (!(fn->fn_flags & RTN_RTINFO)) {
1056			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1057			fn->fn_flags |= RTN_RTINFO;
1058		}
1059		nsiblings = iter->rt6i_nsiblings;
1060		iter->rt6i_node = NULL;
1061		fib6_purge_rt(iter, fn, info->nl_net);
1062		if (rcu_access_pointer(fn->rr_ptr) == iter)
1063			fn->rr_ptr = NULL;
1064		rt6_release(iter);
1065
1066		if (nsiblings) {
1067			/* Replacing an ECMP route, remove all siblings */
1068			ins = &rt->rt6_next;
1069			iter = rcu_dereference_protected(*ins,
1070				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
1071			while (iter) {
1072				if (iter->rt6i_metric > rt->rt6i_metric)
1073					break;
1074				if (rt6_qualify_for_ecmp(iter)) {
1075					*ins = iter->rt6_next;
1076					iter->rt6i_node = NULL;
1077					fib6_purge_rt(iter, fn, info->nl_net);
1078					if (rcu_access_pointer(fn->rr_ptr) == iter)
1079						fn->rr_ptr = NULL;
1080					rt6_release(iter);
1081					nsiblings--;
1082					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1083				} else {
1084					ins = &iter->rt6_next;
1085				}
1086				iter = rcu_dereference_protected(*ins,
1087					lockdep_is_held(&rt->rt6i_table->tb6_lock));
1088			}
1089			WARN_ON(nsiblings != 0);
1090		}
1091	}
1092
1093	return 0;
1094}
1095
1096static void fib6_start_gc(struct net *net, struct rt6_info *rt)
1097{
1098	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1099	    (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
1100		mod_timer(&net->ipv6.ip6_fib_timer,
1101			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1102}
1103
1104void fib6_force_start_gc(struct net *net)
1105{
1106	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1107		mod_timer(&net->ipv6.ip6_fib_timer,
1108			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1109}
1110
1111static void __fib6_update_sernum_upto_root(struct rt6_info *rt,
1112					   int sernum)
1113{
1114	struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1115				lockdep_is_held(&rt->rt6i_table->tb6_lock));
1116
1117	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1118	smp_wmb();
1119	while (fn) {
1120		fn->fn_sernum = sernum;
1121		fn = rcu_dereference_protected(fn->parent,
1122				lockdep_is_held(&rt->rt6i_table->tb6_lock));
1123	}
1124}
1125
1126void fib6_update_sernum_upto_root(struct net *net, struct rt6_info *rt)
1127{
1128	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1129}
1130
1131/*
1132 *	Add routing information to the routing tree.
1133 *	<destination addr>/<source addr>
1134 *	with source addr info in sub-trees
1135 *	Need to own table->tb6_lock
1136 */
1137
1138int fib6_add(struct fib6_node *root, struct rt6_info *rt,
1139	     struct nl_info *info, struct mx6_config *mxc,
1140	     struct netlink_ext_ack *extack)
1141{
1142	struct fib6_table *table = rt->rt6i_table;
1143	struct fib6_node *fn, *pn = NULL;
1144	int err = -ENOMEM;
1145	int allow_create = 1;
1146	int replace_required = 0;
1147	int sernum = fib6_new_sernum(info->nl_net);
1148
1149	if (WARN_ON_ONCE(!atomic_read(&rt->dst.__refcnt)))
1150		return -EINVAL;
1151	if (WARN_ON_ONCE(rt->rt6i_flags & RTF_CACHE))
1152		return -EINVAL;
1153
1154	if (info->nlh) {
1155		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1156			allow_create = 0;
1157		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1158			replace_required = 1;
1159	}
1160	if (!allow_create && !replace_required)
1161		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1162
1163	fn = fib6_add_1(info->nl_net, table, root,
1164			&rt->rt6i_dst.addr, rt->rt6i_dst.plen,
1165			offsetof(struct rt6_info, rt6i_dst), allow_create,
1166			replace_required, extack);
1167	if (IS_ERR(fn)) {
1168		err = PTR_ERR(fn);
1169		fn = NULL;
1170		goto out;
1171	}
1172
1173	pn = fn;
1174
1175#ifdef CONFIG_IPV6_SUBTREES
1176	if (rt->rt6i_src.plen) {
1177		struct fib6_node *sn;
1178
1179		if (!rcu_access_pointer(fn->subtree)) {
1180			struct fib6_node *sfn;
1181
1182			/*
1183			 * Create subtree.
1184			 *
1185			 *		fn[main tree]
1186			 *		|
1187			 *		sfn[subtree root]
1188			 *		   \
1189			 *		    sn[new leaf node]
1190			 */
1191
1192			/* Create subtree root node */
1193			sfn = node_alloc(info->nl_net);
1194			if (!sfn)
1195				goto failure;
1196
 
1197			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1198			rcu_assign_pointer(sfn->leaf,
1199					   info->nl_net->ipv6.ip6_null_entry);
1200			sfn->fn_flags = RTN_ROOT;
 
1201
1202			/* Now add the first leaf node to new subtree */
1203
1204			sn = fib6_add_1(info->nl_net, table, sfn,
1205					&rt->rt6i_src.addr, rt->rt6i_src.plen,
1206					offsetof(struct rt6_info, rt6i_src),
1207					allow_create, replace_required, extack);
1208
1209			if (IS_ERR(sn)) {
1210				/* If it is failed, discard just allocated
1211				   root, and then (in failure) stale node
1212				   in main tree.
1213				 */
1214				node_free_immediate(info->nl_net, sfn);
1215				err = PTR_ERR(sn);
1216				goto failure;
1217			}
1218
1219			/* Now link new subtree to main tree */
1220			rcu_assign_pointer(sfn->parent, fn);
1221			rcu_assign_pointer(fn->subtree, sfn);
1222		} else {
1223			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1224					&rt->rt6i_src.addr, rt->rt6i_src.plen,
1225					offsetof(struct rt6_info, rt6i_src),
1226					allow_create, replace_required, extack);
1227
1228			if (IS_ERR(sn)) {
1229				err = PTR_ERR(sn);
1230				goto failure;
1231			}
1232		}
1233
1234		if (!rcu_access_pointer(fn->leaf)) {
1235			if (fn->fn_flags & RTN_TL_ROOT) {
1236				/* put back null_entry for root node */
1237				rcu_assign_pointer(fn->leaf,
1238					    info->nl_net->ipv6.ip6_null_entry);
1239			} else {
1240				atomic_inc(&rt->rt6i_ref);
1241				rcu_assign_pointer(fn->leaf, rt);
1242			}
1243		}
1244		fn = sn;
1245	}
1246#endif
1247
1248	err = fib6_add_rt2node(fn, rt, info, mxc, extack);
1249	if (!err) {
1250		__fib6_update_sernum_upto_root(rt, sernum);
1251		fib6_start_gc(info->nl_net, rt);
 
 
1252	}
1253
1254out:
1255	if (err) {
1256#ifdef CONFIG_IPV6_SUBTREES
1257		/*
1258		 * If fib6_add_1 has cleared the old leaf pointer in the
1259		 * super-tree leaf node we have to find a new one for it.
1260		 */
1261		if (pn != fn) {
1262			struct rt6_info *pn_leaf =
1263				rcu_dereference_protected(pn->leaf,
1264				    lockdep_is_held(&table->tb6_lock));
1265			if (pn_leaf == rt) {
1266				pn_leaf = NULL;
1267				RCU_INIT_POINTER(pn->leaf, NULL);
1268				atomic_dec(&rt->rt6i_ref);
1269			}
1270			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1271				pn_leaf = fib6_find_prefix(info->nl_net, table,
1272							   pn);
1273#if RT6_DEBUG >= 2
1274				if (!pn_leaf) {
1275					WARN_ON(!pn_leaf);
1276					pn_leaf =
1277					    info->nl_net->ipv6.ip6_null_entry;
1278				}
1279#endif
1280				atomic_inc(&pn_leaf->rt6i_ref);
1281				rcu_assign_pointer(pn->leaf, pn_leaf);
1282			}
 
 
1283		}
1284#endif
1285		goto failure;
1286	}
1287	return err;
1288
1289failure:
1290	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1291	 * 1. fn is an intermediate node and we failed to add the new
1292	 * route to it in both subtree creation failure and fib6_add_rt2node()
1293	 * failure case.
1294	 * 2. fn is the root node in the table and we fail to add the first
1295	 * default route to it.
1296	 */
1297	if (fn &&
1298	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1299	     (fn->fn_flags & RTN_TL_ROOT &&
1300	      !rcu_access_pointer(fn->leaf))))
1301		fib6_repair_tree(info->nl_net, table, fn);
1302	/* Always release dst as dst->__refcnt is guaranteed
1303	 * to be taken before entering this function
1304	 */
1305	dst_release_immediate(&rt->dst);
 
 
 
1306	return err;
 
1307}
1308
1309/*
1310 *	Routing tree lookup
1311 *
1312 */
1313
1314struct lookup_args {
1315	int			offset;		/* key offset on rt6_info	*/
1316	const struct in6_addr	*addr;		/* search key			*/
1317};
1318
1319static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1320				       struct lookup_args *args)
1321{
1322	struct fib6_node *fn;
1323	__be32 dir;
1324
1325	if (unlikely(args->offset == 0))
1326		return NULL;
1327
1328	/*
1329	 *	Descend on a tree
1330	 */
1331
1332	fn = root;
1333
1334	for (;;) {
1335		struct fib6_node *next;
1336
1337		dir = addr_bit_set(args->addr, fn->fn_bit);
1338
1339		next = dir ? rcu_dereference(fn->right) :
1340			     rcu_dereference(fn->left);
1341
1342		if (next) {
1343			fn = next;
1344			continue;
1345		}
 
1346		break;
1347	}
1348
1349	while (fn) {
1350		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1351
1352		if (subtree || fn->fn_flags & RTN_RTINFO) {
1353			struct rt6_info *leaf = rcu_dereference(fn->leaf);
1354			struct rt6key *key;
1355
1356			if (!leaf)
1357				goto backtrack;
1358
1359			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1360
1361			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1362#ifdef CONFIG_IPV6_SUBTREES
1363				if (subtree) {
1364					struct fib6_node *sfn;
1365					sfn = fib6_lookup_1(subtree, args + 1);
1366					if (!sfn)
1367						goto backtrack;
1368					fn = sfn;
1369				}
1370#endif
1371				if (fn->fn_flags & RTN_RTINFO)
1372					return fn;
1373			}
1374		}
1375backtrack:
1376		if (fn->fn_flags & RTN_ROOT)
1377			break;
1378
1379		fn = rcu_dereference(fn->parent);
1380	}
1381
1382	return NULL;
1383}
1384
1385/* called with rcu_read_lock() held
1386 */
1387struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1388			      const struct in6_addr *saddr)
1389{
1390	struct fib6_node *fn;
1391	struct lookup_args args[] = {
1392		{
1393			.offset = offsetof(struct rt6_info, rt6i_dst),
1394			.addr = daddr,
1395		},
1396#ifdef CONFIG_IPV6_SUBTREES
1397		{
1398			.offset = offsetof(struct rt6_info, rt6i_src),
1399			.addr = saddr,
1400		},
1401#endif
1402		{
1403			.offset = 0,	/* sentinel */
1404		}
1405	};
1406
1407	fn = fib6_lookup_1(root, daddr ? args : args + 1);
1408	if (!fn || fn->fn_flags & RTN_TL_ROOT)
 
1409		fn = root;
1410
1411	return fn;
1412}
1413
1414/*
1415 *	Get node with specified destination prefix (and source prefix,
1416 *	if subtrees are used)
1417 *	exact_match == true means we try to find fn with exact match of
1418 *	the passed in prefix addr
1419 *	exact_match == false means we try to find fn with longest prefix
1420 *	match of the passed in prefix addr. This is useful for finding fn
1421 *	for cached route as it will be stored in the exception table under
1422 *	the node with longest prefix length.
1423 */
1424
1425
1426static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1427				       const struct in6_addr *addr,
1428				       int plen, int offset,
1429				       bool exact_match)
1430{
1431	struct fib6_node *fn, *prev = NULL;
1432
1433	for (fn = root; fn ; ) {
1434		struct rt6_info *leaf = rcu_dereference(fn->leaf);
1435		struct rt6key *key;
1436
1437		/* This node is being deleted */
1438		if (!leaf) {
1439			if (plen <= fn->fn_bit)
1440				goto out;
1441			else
1442				goto next;
1443		}
1444
1445		key = (struct rt6key *)((u8 *)leaf + offset);
1446
1447		/*
1448		 *	Prefix match
1449		 */
1450		if (plen < fn->fn_bit ||
1451		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1452			goto out;
1453
1454		if (plen == fn->fn_bit)
1455			return fn;
1456
1457		prev = fn;
1458
1459next:
1460		/*
1461		 *	We have more bits to go
1462		 */
1463		if (addr_bit_set(addr, fn->fn_bit))
1464			fn = rcu_dereference(fn->right);
1465		else
1466			fn = rcu_dereference(fn->left);
1467	}
1468out:
1469	if (exact_match)
1470		return NULL;
1471	else
1472		return prev;
1473}
1474
1475struct fib6_node *fib6_locate(struct fib6_node *root,
1476			      const struct in6_addr *daddr, int dst_len,
1477			      const struct in6_addr *saddr, int src_len,
1478			      bool exact_match)
1479{
1480	struct fib6_node *fn;
1481
1482	fn = fib6_locate_1(root, daddr, dst_len,
1483			   offsetof(struct rt6_info, rt6i_dst),
1484			   exact_match);
1485
1486#ifdef CONFIG_IPV6_SUBTREES
1487	if (src_len) {
1488		WARN_ON(saddr == NULL);
1489		if (fn) {
1490			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1491
1492			if (subtree) {
1493				fn = fib6_locate_1(subtree, saddr, src_len,
1494					   offsetof(struct rt6_info, rt6i_src),
1495					   exact_match);
1496			}
1497		}
1498	}
1499#endif
1500
1501	if (fn && fn->fn_flags & RTN_RTINFO)
1502		return fn;
1503
1504	return NULL;
1505}
1506
1507
1508/*
1509 *	Deletion
1510 *
1511 */
1512
1513static struct rt6_info *fib6_find_prefix(struct net *net,
1514					 struct fib6_table *table,
1515					 struct fib6_node *fn)
1516{
1517	struct fib6_node *child_left, *child_right;
1518
1519	if (fn->fn_flags & RTN_ROOT)
1520		return net->ipv6.ip6_null_entry;
1521
1522	while (fn) {
1523		child_left = rcu_dereference_protected(fn->left,
1524				    lockdep_is_held(&table->tb6_lock));
1525		child_right = rcu_dereference_protected(fn->right,
1526				    lockdep_is_held(&table->tb6_lock));
1527		if (child_left)
1528			return rcu_dereference_protected(child_left->leaf,
1529					lockdep_is_held(&table->tb6_lock));
1530		if (child_right)
1531			return rcu_dereference_protected(child_right->leaf,
1532					lockdep_is_held(&table->tb6_lock));
1533
1534		fn = FIB6_SUBTREE(fn);
1535	}
1536	return NULL;
1537}
1538
1539/*
1540 *	Called to trim the tree of intermediate nodes when possible. "fn"
1541 *	is the node we want to try and remove.
1542 *	Need to own table->tb6_lock
1543 */
1544
1545static struct fib6_node *fib6_repair_tree(struct net *net,
1546					  struct fib6_table *table,
1547					  struct fib6_node *fn)
1548{
1549	int children;
1550	int nstate;
1551	struct fib6_node *child;
1552	struct fib6_walker *w;
1553	int iter = 0;
1554
1555	/* Set fn->leaf to null_entry for root node. */
1556	if (fn->fn_flags & RTN_TL_ROOT) {
1557		rcu_assign_pointer(fn->leaf, net->ipv6.ip6_null_entry);
1558		return fn;
1559	}
1560
1561	for (;;) {
1562		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1563					    lockdep_is_held(&table->tb6_lock));
1564		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1565					    lockdep_is_held(&table->tb6_lock));
1566		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1567					    lockdep_is_held(&table->tb6_lock));
1568		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1569					    lockdep_is_held(&table->tb6_lock));
1570		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1571					    lockdep_is_held(&table->tb6_lock));
1572		struct rt6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1573					    lockdep_is_held(&table->tb6_lock));
1574		struct rt6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1575					    lockdep_is_held(&table->tb6_lock));
1576		struct rt6_info *new_fn_leaf;
1577
1578		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1579		iter++;
1580
1581		WARN_ON(fn->fn_flags & RTN_RTINFO);
1582		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1583		WARN_ON(fn_leaf);
1584
1585		children = 0;
1586		child = NULL;
1587		if (fn_r)
1588			child = fn_r, children |= 1;
1589		if (fn_l)
1590			child = fn_l, children |= 2;
1591
1592		if (children == 3 || FIB6_SUBTREE(fn)
1593#ifdef CONFIG_IPV6_SUBTREES
1594		    /* Subtree root (i.e. fn) may have one child */
1595		    || (children && fn->fn_flags & RTN_ROOT)
1596#endif
1597		    ) {
1598			new_fn_leaf = fib6_find_prefix(net, table, fn);
1599#if RT6_DEBUG >= 2
1600			if (!new_fn_leaf) {
1601				WARN_ON(!new_fn_leaf);
1602				new_fn_leaf = net->ipv6.ip6_null_entry;
1603			}
1604#endif
1605			atomic_inc(&new_fn_leaf->rt6i_ref);
1606			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1607			return pn;
1608		}
1609
 
1610#ifdef CONFIG_IPV6_SUBTREES
1611		if (FIB6_SUBTREE(pn) == fn) {
1612			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1613			RCU_INIT_POINTER(pn->subtree, NULL);
1614			nstate = FWS_L;
1615		} else {
1616			WARN_ON(fn->fn_flags & RTN_ROOT);
1617#endif
1618			if (pn_r == fn)
1619				rcu_assign_pointer(pn->right, child);
1620			else if (pn_l == fn)
1621				rcu_assign_pointer(pn->left, child);
1622#if RT6_DEBUG >= 2
1623			else
1624				WARN_ON(1);
1625#endif
1626			if (child)
1627				rcu_assign_pointer(child->parent, pn);
1628			nstate = FWS_R;
1629#ifdef CONFIG_IPV6_SUBTREES
1630		}
1631#endif
1632
1633		read_lock(&net->ipv6.fib6_walker_lock);
1634		FOR_WALKERS(net, w) {
1635			if (!child) {
1636				if (w->node == fn) {
 
 
 
1637					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1638					w->node = pn;
1639					w->state = nstate;
1640				}
1641			} else {
 
 
 
 
1642				if (w->node == fn) {
1643					w->node = child;
1644					if (children&2) {
1645						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1646						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1647					} else {
1648						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1649						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1650					}
1651				}
1652			}
1653		}
1654		read_unlock(&net->ipv6.fib6_walker_lock);
1655
1656		node_free(net, fn);
1657		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1658			return pn;
1659
1660		RCU_INIT_POINTER(pn->leaf, NULL);
1661		rt6_release(pn_leaf);
1662		fn = pn;
1663	}
1664}
1665
1666static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1667			   struct rt6_info __rcu **rtp, struct nl_info *info)
1668{
1669	struct fib6_walker *w;
1670	struct rt6_info *rt = rcu_dereference_protected(*rtp,
1671				    lockdep_is_held(&table->tb6_lock));
1672	struct net *net = info->nl_net;
1673
1674	RT6_TRACE("fib6_del_route\n");
1675
1676	WARN_ON_ONCE(rt->rt6i_flags & RTF_CACHE);
1677
1678	/* Unlink it */
1679	*rtp = rt->rt6_next;
1680	rt->rt6i_node = NULL;
1681	net->ipv6.rt6_stats->fib_rt_entries--;
1682	net->ipv6.rt6_stats->fib_discarded_routes++;
1683
1684	/* Flush all cached dst in exception table */
1685	rt6_flush_exceptions(rt);
1686
1687	/* Reset round-robin state, if necessary */
1688	if (rcu_access_pointer(fn->rr_ptr) == rt)
1689		fn->rr_ptr = NULL;
1690
1691	/* Remove this entry from other siblings */
1692	if (rt->rt6i_nsiblings) {
1693		struct rt6_info *sibling, *next_sibling;
1694
1695		list_for_each_entry_safe(sibling, next_sibling,
1696					 &rt->rt6i_siblings, rt6i_siblings)
1697			sibling->rt6i_nsiblings--;
1698		rt->rt6i_nsiblings = 0;
1699		list_del_init(&rt->rt6i_siblings);
1700		rt6_multipath_rebalance(next_sibling);
1701	}
1702
1703	/* Adjust walkers */
1704	read_lock(&net->ipv6.fib6_walker_lock);
1705	FOR_WALKERS(net, w) {
1706		if (w->state == FWS_C && w->leaf == rt) {
1707			RT6_TRACE("walker %p adjusted by delroute\n", w);
1708			w->leaf = rcu_dereference_protected(rt->rt6_next,
1709					    lockdep_is_held(&table->tb6_lock));
1710			if (!w->leaf)
1711				w->state = FWS_U;
1712		}
1713	}
1714	read_unlock(&net->ipv6.fib6_walker_lock);
1715
1716	/* If it was last route, call fib6_repair_tree() to:
1717	 * 1. For root node, put back null_entry as how the table was created.
1718	 * 2. For other nodes, expunge its radix tree node.
1719	 */
1720	if (!rcu_access_pointer(fn->leaf)) {
1721		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1722			fn->fn_flags &= ~RTN_RTINFO;
1723			net->ipv6.rt6_stats->fib_route_nodes--;
1724		}
1725		fn = fib6_repair_tree(net, table, fn);
1726	}
1727
1728	fib6_purge_rt(rt, fn, net);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1729
1730	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1731	if (!info->skip_notify)
1732		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1733	rt6_release(rt);
1734}
1735
1736/* Need to own table->tb6_lock */
1737int fib6_del(struct rt6_info *rt, struct nl_info *info)
1738{
1739	struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1740				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
1741	struct fib6_table *table = rt->rt6i_table;
1742	struct net *net = info->nl_net;
1743	struct rt6_info __rcu **rtp;
1744	struct rt6_info __rcu **rtp_next;
1745
1746#if RT6_DEBUG >= 2
1747	if (rt->dst.obsolete > 0) {
1748		WARN_ON(fn);
1749		return -ENOENT;
1750	}
1751#endif
1752	if (!fn || rt == net->ipv6.ip6_null_entry)
1753		return -ENOENT;
1754
1755	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1756
1757	/* remove cached dst from exception table */
1758	if (rt->rt6i_flags & RTF_CACHE)
1759		return rt6_remove_exception_rt(rt);
 
 
 
 
 
 
 
 
 
1760
1761	/*
1762	 *	Walk the leaf entries looking for ourself
1763	 */
1764
1765	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1766		struct rt6_info *cur = rcu_dereference_protected(*rtp,
1767					lockdep_is_held(&table->tb6_lock));
1768		if (rt == cur) {
1769			fib6_del_route(table, fn, rtp, info);
1770			return 0;
1771		}
1772		rtp_next = &cur->rt6_next;
1773	}
1774	return -ENOENT;
1775}
1776
1777/*
1778 *	Tree traversal function.
1779 *
1780 *	Certainly, it is not interrupt safe.
1781 *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1782 *	It means, that we can modify tree during walking
1783 *	and use this function for garbage collection, clone pruning,
1784 *	cleaning tree when a device goes down etc. etc.
1785 *
1786 *	It guarantees that every node will be traversed,
1787 *	and that it will be traversed only once.
1788 *
1789 *	Callback function w->func may return:
1790 *	0 -> continue walking.
1791 *	positive value -> walking is suspended (used by tree dumps,
1792 *	and probably by gc, if it will be split to several slices)
1793 *	negative value -> terminate walking.
1794 *
1795 *	The function itself returns:
1796 *	0   -> walk is complete.
1797 *	>0  -> walk is incomplete (i.e. suspended)
1798 *	<0  -> walk is terminated by an error.
1799 *
1800 *	This function is called with tb6_lock held.
1801 */
1802
1803static int fib6_walk_continue(struct fib6_walker *w)
1804{
1805	struct fib6_node *fn, *pn, *left, *right;
1806
1807	/* w->root should always be table->tb6_root */
1808	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1809
1810	for (;;) {
1811		fn = w->node;
1812		if (!fn)
1813			return 0;
1814
 
 
 
 
 
1815		switch (w->state) {
1816#ifdef CONFIG_IPV6_SUBTREES
1817		case FWS_S:
1818			if (FIB6_SUBTREE(fn)) {
1819				w->node = FIB6_SUBTREE(fn);
1820				continue;
1821			}
1822			w->state = FWS_L;
1823#endif
1824			/* fall through */
1825		case FWS_L:
1826			left = rcu_dereference_protected(fn->left, 1);
1827			if (left) {
1828				w->node = left;
1829				w->state = FWS_INIT;
1830				continue;
1831			}
1832			w->state = FWS_R;
1833			/* fall through */
1834		case FWS_R:
1835			right = rcu_dereference_protected(fn->right, 1);
1836			if (right) {
1837				w->node = right;
1838				w->state = FWS_INIT;
1839				continue;
1840			}
1841			w->state = FWS_C;
1842			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1843			/* fall through */
1844		case FWS_C:
1845			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1846				int err;
1847
1848				if (w->skip) {
1849					w->skip--;
1850					goto skip;
1851				}
1852
1853				err = w->func(w);
1854				if (err)
1855					return err;
1856
1857				w->count++;
1858				continue;
1859			}
1860skip:
1861			w->state = FWS_U;
1862			/* fall through */
1863		case FWS_U:
1864			if (fn == w->root)
1865				return 0;
1866			pn = rcu_dereference_protected(fn->parent, 1);
1867			left = rcu_dereference_protected(pn->left, 1);
1868			right = rcu_dereference_protected(pn->right, 1);
1869			w->node = pn;
1870#ifdef CONFIG_IPV6_SUBTREES
1871			if (FIB6_SUBTREE(pn) == fn) {
1872				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1873				w->state = FWS_L;
1874				continue;
1875			}
1876#endif
1877			if (left == fn) {
1878				w->state = FWS_R;
1879				continue;
1880			}
1881			if (right == fn) {
1882				w->state = FWS_C;
1883				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1884				continue;
1885			}
1886#if RT6_DEBUG >= 2
1887			WARN_ON(1);
1888#endif
1889		}
1890	}
1891}
1892
1893static int fib6_walk(struct net *net, struct fib6_walker *w)
1894{
1895	int res;
1896
1897	w->state = FWS_INIT;
1898	w->node = w->root;
1899
1900	fib6_walker_link(net, w);
1901	res = fib6_walk_continue(w);
1902	if (res <= 0)
1903		fib6_walker_unlink(net, w);
1904	return res;
1905}
1906
1907static int fib6_clean_node(struct fib6_walker *w)
1908{
1909	int res;
1910	struct rt6_info *rt;
1911	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1912	struct nl_info info = {
1913		.nl_net = c->net,
1914	};
1915
1916	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1917	    w->node->fn_sernum != c->sernum)
1918		w->node->fn_sernum = c->sernum;
1919
1920	if (!c->func) {
1921		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1922		w->leaf = NULL;
1923		return 0;
1924	}
1925
1926	for_each_fib6_walker_rt(w) {
1927		res = c->func(rt, c->arg);
1928		if (res == -1) {
1929			w->leaf = rt;
1930			res = fib6_del(rt, &info);
1931			if (res) {
1932#if RT6_DEBUG >= 2
1933				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1934					 __func__, rt,
1935					 rcu_access_pointer(rt->rt6i_node),
1936					 res);
1937#endif
1938				continue;
1939			}
1940			return 0;
1941		} else if (res == -2) {
1942			if (WARN_ON(!rt->rt6i_nsiblings))
1943				continue;
1944			rt = list_last_entry(&rt->rt6i_siblings,
1945					     struct rt6_info, rt6i_siblings);
1946			continue;
1947		}
1948		WARN_ON(res != 0);
1949	}
1950	w->leaf = rt;
1951	return 0;
1952}
1953
1954/*
1955 *	Convenient frontend to tree walker.
1956 *
1957 *	func is called on each route.
1958 *		It may return -2 -> skip multipath route.
1959 *			      -1 -> delete this route.
1960 *		              0  -> continue walking
 
 
 
1961 */
1962
1963static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1964			    int (*func)(struct rt6_info *, void *arg),
1965			    int sernum, void *arg)
1966{
1967	struct fib6_cleaner c;
1968
1969	c.w.root = root;
1970	c.w.func = fib6_clean_node;
 
1971	c.w.count = 0;
1972	c.w.skip = 0;
1973	c.func = func;
1974	c.sernum = sernum;
1975	c.arg = arg;
1976	c.net = net;
1977
1978	fib6_walk(net, &c.w);
1979}
1980
1981static void __fib6_clean_all(struct net *net,
1982			     int (*func)(struct rt6_info *, void *),
1983			     int sernum, void *arg)
1984{
1985	struct fib6_table *table;
 
1986	struct hlist_head *head;
1987	unsigned int h;
1988
1989	rcu_read_lock();
1990	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1991		head = &net->ipv6.fib_table_hash[h];
1992		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1993			spin_lock_bh(&table->tb6_lock);
1994			fib6_clean_tree(net, &table->tb6_root,
1995					func, sernum, arg);
1996			spin_unlock_bh(&table->tb6_lock);
1997		}
1998	}
1999	rcu_read_unlock();
2000}
2001
2002void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
2003		    void *arg)
2004{
2005	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
 
 
 
 
 
2006}
2007
2008static void fib6_flush_trees(struct net *net)
 
2009{
2010	int new_sernum = fib6_new_sernum(net);
2011
2012	__fib6_clean_all(net, NULL, new_sernum, NULL);
2013}
2014
2015/*
2016 *	Garbage collection
2017 */
2018
 
 
 
 
 
 
2019static int fib6_age(struct rt6_info *rt, void *arg)
2020{
2021	struct fib6_gc_args *gc_args = arg;
2022	unsigned long now = jiffies;
2023
2024	/*
2025	 *	check addrconf expiration here.
2026	 *	Routes are expired even if they are in use.
 
 
 
2027	 */
2028
2029	if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
2030		if (time_after(now, rt->dst.expires)) {
2031			RT6_TRACE("expiring %p\n", rt);
2032			return -1;
2033		}
2034		gc_args->more++;
 
 
 
 
 
 
 
 
 
 
 
 
2035	}
2036
2037	/*	Also age clones in the exception table.
2038	 *	Note, that clones are aged out
2039	 *	only if they are not in use now.
2040	 */
2041	rt6_age_exceptions(rt, gc_args, now);
2042
2043	return 0;
2044}
2045
2046void fib6_run_gc(unsigned long expires, struct net *net, bool force)
 
 
2047{
2048	struct fib6_gc_args gc_args;
2049	unsigned long now;
 
 
 
 
 
 
 
 
 
 
 
2050
2051	if (force) {
2052		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2053	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2054		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2055		return;
2056	}
2057	gc_args.timeout = expires ? (int)expires :
2058			  net->ipv6.sysctl.ip6_rt_gc_interval;
2059	gc_args.more = 0;
2060
2061	fib6_clean_all(net, fib6_age, &gc_args);
2062	now = jiffies;
2063	net->ipv6.ip6_rt_last_gc = now;
2064
2065	if (gc_args.more)
2066		mod_timer(&net->ipv6.ip6_fib_timer,
2067			  round_jiffies(now
2068					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2069	else
2070		del_timer(&net->ipv6.ip6_fib_timer);
2071	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2072}
2073
2074static void fib6_gc_timer_cb(struct timer_list *t)
2075{
2076	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2077
2078	fib6_run_gc(0, arg, true);
2079}
2080
2081static int __net_init fib6_net_init(struct net *net)
2082{
2083	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2084	int err;
2085
2086	err = fib6_notifier_init(net);
2087	if (err)
2088		return err;
2089
2090	spin_lock_init(&net->ipv6.fib6_gc_lock);
2091	rwlock_init(&net->ipv6.fib6_walker_lock);
2092	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2093	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2094
2095	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2096	if (!net->ipv6.rt6_stats)
2097		goto out_timer;
2098
2099	/* Avoid false sharing : Use at least a full cache line */
2100	size = max_t(size_t, size, L1_CACHE_BYTES);
2101
2102	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2103	if (!net->ipv6.fib_table_hash)
2104		goto out_rt6_stats;
2105
2106	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2107					  GFP_KERNEL);
2108	if (!net->ipv6.fib6_main_tbl)
2109		goto out_fib_table_hash;
2110
2111	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2112	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2113			   net->ipv6.ip6_null_entry);
2114	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2115		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2116	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2117
2118#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2119	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2120					   GFP_KERNEL);
2121	if (!net->ipv6.fib6_local_tbl)
2122		goto out_fib6_main_tbl;
2123	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2124	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2125			   net->ipv6.ip6_null_entry);
2126	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2127		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2128	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2129#endif
2130	fib6_tables_init(net);
2131
2132	return 0;
2133
2134#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2135out_fib6_main_tbl:
2136	kfree(net->ipv6.fib6_main_tbl);
2137#endif
2138out_fib_table_hash:
2139	kfree(net->ipv6.fib_table_hash);
2140out_rt6_stats:
2141	kfree(net->ipv6.rt6_stats);
2142out_timer:
2143	fib6_notifier_exit(net);
2144	return -ENOMEM;
2145}
2146
2147static void fib6_net_exit(struct net *net)
2148{
2149	unsigned int i;
2150
2151	del_timer_sync(&net->ipv6.ip6_fib_timer);
2152
2153	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2154		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2155		struct hlist_node *tmp;
2156		struct fib6_table *tb;
2157
2158		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2159			hlist_del(&tb->tb6_hlist);
2160			fib6_free_table(tb);
2161		}
2162	}
2163
2164	kfree(net->ipv6.fib_table_hash);
2165	kfree(net->ipv6.rt6_stats);
2166	fib6_notifier_exit(net);
2167}
2168
2169static struct pernet_operations fib6_net_ops = {
2170	.init = fib6_net_init,
2171	.exit = fib6_net_exit,
2172};
2173
2174int __init fib6_init(void)
2175{
2176	int ret = -ENOMEM;
2177
2178	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2179					   sizeof(struct fib6_node),
2180					   0, SLAB_HWCACHE_ALIGN,
2181					   NULL);
2182	if (!fib6_node_kmem)
2183		goto out;
2184
2185	ret = register_pernet_subsys(&fib6_net_ops);
2186	if (ret)
2187		goto out_kmem_cache_create;
2188
2189	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2190				   inet6_dump_fib, 0);
2191	if (ret)
2192		goto out_unregister_subsys;
2193
2194	__fib6_flush_trees = fib6_flush_trees;
2195out:
2196	return ret;
2197
2198out_unregister_subsys:
2199	unregister_pernet_subsys(&fib6_net_ops);
2200out_kmem_cache_create:
2201	kmem_cache_destroy(fib6_node_kmem);
2202	goto out;
2203}
2204
2205void fib6_gc_cleanup(void)
2206{
2207	unregister_pernet_subsys(&fib6_net_ops);
2208	kmem_cache_destroy(fib6_node_kmem);
2209}
2210
2211#ifdef CONFIG_PROC_FS
2212
2213struct ipv6_route_iter {
2214	struct seq_net_private p;
2215	struct fib6_walker w;
2216	loff_t skip;
2217	struct fib6_table *tbl;
2218	int sernum;
2219};
2220
2221static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2222{
2223	struct rt6_info *rt = v;
2224	struct ipv6_route_iter *iter = seq->private;
2225
2226	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
2227
2228#ifdef CONFIG_IPV6_SUBTREES
2229	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
2230#else
2231	seq_puts(seq, "00000000000000000000000000000000 00 ");
2232#endif
2233	if (rt->rt6i_flags & RTF_GATEWAY)
2234		seq_printf(seq, "%pi6", &rt->rt6i_gateway);
2235	else
2236		seq_puts(seq, "00000000000000000000000000000000");
2237
2238	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2239		   rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
2240		   rt->dst.__use, rt->rt6i_flags,
2241		   rt->dst.dev ? rt->dst.dev->name : "");
2242	iter->w.leaf = NULL;
2243	return 0;
2244}
2245
2246static int ipv6_route_yield(struct fib6_walker *w)
2247{
2248	struct ipv6_route_iter *iter = w->args;
2249
2250	if (!iter->skip)
2251		return 1;
2252
2253	do {
2254		iter->w.leaf = rcu_dereference_protected(
2255				iter->w.leaf->rt6_next,
2256				lockdep_is_held(&iter->tbl->tb6_lock));
2257		iter->skip--;
2258		if (!iter->skip && iter->w.leaf)
2259			return 1;
2260	} while (iter->w.leaf);
2261
2262	return 0;
2263}
2264
2265static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2266				      struct net *net)
2267{
2268	memset(&iter->w, 0, sizeof(iter->w));
2269	iter->w.func = ipv6_route_yield;
2270	iter->w.root = &iter->tbl->tb6_root;
2271	iter->w.state = FWS_INIT;
2272	iter->w.node = iter->w.root;
2273	iter->w.args = iter;
2274	iter->sernum = iter->w.root->fn_sernum;
2275	INIT_LIST_HEAD(&iter->w.lh);
2276	fib6_walker_link(net, &iter->w);
2277}
2278
2279static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2280						    struct net *net)
2281{
2282	unsigned int h;
2283	struct hlist_node *node;
2284
2285	if (tbl) {
2286		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2287		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2288	} else {
2289		h = 0;
2290		node = NULL;
2291	}
2292
2293	while (!node && h < FIB6_TABLE_HASHSZ) {
2294		node = rcu_dereference_bh(
2295			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2296	}
2297	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2298}
2299
2300static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2301{
2302	if (iter->sernum != iter->w.root->fn_sernum) {
2303		iter->sernum = iter->w.root->fn_sernum;
2304		iter->w.state = FWS_INIT;
2305		iter->w.node = iter->w.root;
2306		WARN_ON(iter->w.skip);
2307		iter->w.skip = iter->w.count;
2308	}
2309}
2310
2311static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2312{
2313	int r;
2314	struct rt6_info *n;
2315	struct net *net = seq_file_net(seq);
2316	struct ipv6_route_iter *iter = seq->private;
2317
2318	if (!v)
2319		goto iter_table;
2320
2321	n = rcu_dereference_bh(((struct rt6_info *)v)->rt6_next);
2322	if (n) {
2323		++*pos;
2324		return n;
2325	}
2326
2327iter_table:
2328	ipv6_route_check_sernum(iter);
2329	spin_lock_bh(&iter->tbl->tb6_lock);
2330	r = fib6_walk_continue(&iter->w);
2331	spin_unlock_bh(&iter->tbl->tb6_lock);
2332	if (r > 0) {
2333		if (v)
2334			++*pos;
2335		return iter->w.leaf;
2336	} else if (r < 0) {
2337		fib6_walker_unlink(net, &iter->w);
2338		return NULL;
2339	}
2340	fib6_walker_unlink(net, &iter->w);
2341
2342	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2343	if (!iter->tbl)
2344		return NULL;
2345
2346	ipv6_route_seq_setup_walk(iter, net);
2347	goto iter_table;
2348}
2349
2350static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2351	__acquires(RCU_BH)
2352{
2353	struct net *net = seq_file_net(seq);
2354	struct ipv6_route_iter *iter = seq->private;
2355
2356	rcu_read_lock_bh();
2357	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2358	iter->skip = *pos;
2359
2360	if (iter->tbl) {
2361		ipv6_route_seq_setup_walk(iter, net);
2362		return ipv6_route_seq_next(seq, NULL, pos);
2363	} else {
2364		return NULL;
2365	}
2366}
2367
2368static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2369{
2370	struct fib6_walker *w = &iter->w;
2371	return w->node && !(w->state == FWS_U && w->node == w->root);
2372}
2373
2374static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2375	__releases(RCU_BH)
2376{
2377	struct net *net = seq_file_net(seq);
2378	struct ipv6_route_iter *iter = seq->private;
2379
2380	if (ipv6_route_iter_active(iter))
2381		fib6_walker_unlink(net, &iter->w);
2382
2383	rcu_read_unlock_bh();
2384}
2385
2386static const struct seq_operations ipv6_route_seq_ops = {
2387	.start	= ipv6_route_seq_start,
2388	.next	= ipv6_route_seq_next,
2389	.stop	= ipv6_route_seq_stop,
2390	.show	= ipv6_route_seq_show
2391};
2392
2393int ipv6_route_open(struct inode *inode, struct file *file)
2394{
2395	return seq_open_net(inode, file, &ipv6_route_seq_ops,
2396			    sizeof(struct ipv6_route_iter));
2397}
2398
2399#endif /* CONFIG_PROC_FS */