1/*
2 *	Linux INET6 implementation
3 *	Forwarding Information Database
4 *
5 *	Authors:
6 *	Pedro Roque		<roque@di.fc.ul.pt>
7 *
8 *	This program is free software; you can redistribute it and/or
9 *      modify it under the terms of the GNU General Public License
10 *      as published by the Free Software Foundation; either version
11 *      2 of the License, or (at your option) any later version.
12 *
13 *	Changes:
14 *	Yuji SEKIYA @USAGI:	Support default route on router node;
15 *				remove ip6_null_entry from the top of
16 *				routing table.
17 *	Ville Nuorvala:		Fixed routing subtrees.
18 */
19
20#define pr_fmt(fmt) "IPv6: " fmt
21
22#include <linux/errno.h>
23#include <linux/types.h>
24#include <linux/net.h>
25#include <linux/route.h>
26#include <linux/netdevice.h>
27#include <linux/in6.h>
28#include <linux/init.h>
29#include <linux/list.h>
30#include <linux/slab.h>
31
32#include <net/ipv6.h>
33#include <net/ndisc.h>
34#include <net/addrconf.h>
35#include <net/lwtunnel.h>
36
37#include <net/ip6_fib.h>
38#include <net/ip6_route.h>
39
40#define RT6_DEBUG 2
41
42#if RT6_DEBUG >= 3
43#define RT6_TRACE(x...) pr_debug(x)
44#else
45#define RT6_TRACE(x...) do { ; } while (0)
46#endif
47
48static struct kmem_cache *fib6_node_kmem __read_mostly;
49
50struct fib6_cleaner {
51	struct fib6_walker w;
52	struct net *net;
53	int (*func)(struct rt6_info *, void *arg);
54	int sernum;
55	void *arg;
56};
57
58static DEFINE_RWLOCK(fib6_walker_lock);
59
60#ifdef CONFIG_IPV6_SUBTREES
61#define FWS_INIT FWS_S
62#else
63#define FWS_INIT FWS_L
64#endif
65
66static void fib6_prune_clones(struct net *net, struct fib6_node *fn);
67static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
68static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
69static int fib6_walk(struct fib6_walker *w);
70static int fib6_walk_continue(struct fib6_walker *w);
71
72/*
73 *	A routing update causes an increase of the serial number on the
74 *	affected subtree. This allows for cached routes to be asynchronously
75 *	tested when modifications are made to the destination cache as a
76 *	result of redirects, path MTU changes, etc.
77 */
78
79static void fib6_gc_timer_cb(unsigned long arg);
80
81static LIST_HEAD(fib6_walkers);
82#define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
83
84static void fib6_walker_link(struct fib6_walker *w)
85{
86	write_lock_bh(&fib6_walker_lock);
87	list_add(&w->lh, &fib6_walkers);
88	write_unlock_bh(&fib6_walker_lock);
89}
90
91static void fib6_walker_unlink(struct fib6_walker *w)
92{
93	write_lock_bh(&fib6_walker_lock);
94	list_del(&w->lh);
95	write_unlock_bh(&fib6_walker_lock);
96}
97
98static int fib6_new_sernum(struct net *net)
99{
100	int new, old;
101
102	do {
103		old = atomic_read(&net->ipv6.fib6_sernum);
104		new = old < INT_MAX ? old + 1 : 1;
105	} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
106				old, new) != old);
107	return new;
108}
109
110enum {
111	FIB6_NO_SERNUM_CHANGE = 0,
112};
113
114/*
115 *	Auxiliary address test functions for the radix tree.
116 *
117 *	These assume a 32bit processor (although it will work on
118 *	64bit processors)
119 */
120
121/*
122 *	test bit
123 */
124#if defined(__LITTLE_ENDIAN)
125# define BITOP_BE32_SWIZZLE	(0x1F & ~7)
126#else
127# define BITOP_BE32_SWIZZLE	0
128#endif
129
130static __be32 addr_bit_set(const void *token, int fn_bit)
131{
132	const __be32 *addr = token;
133	/*
134	 * Here,
135	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
136	 * is optimized version of
137	 *	htonl(1 << ((~fn_bit)&0x1F))
138	 * See include/asm-generic/bitops/le.h.
139	 */
140	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
141	       addr[fn_bit >> 5];
142}
143
144static struct fib6_node *node_alloc(void)
145{
146	struct fib6_node *fn;
147
148	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
149
150	return fn;
151}
152
153static void node_free(struct fib6_node *fn)
154{
155	kmem_cache_free(fib6_node_kmem, fn);
156}
157
158static void rt6_rcu_free(struct rt6_info *rt)
159{
160	call_rcu(&rt->dst.rcu_head, dst_rcu_free);
161}
162
163static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
164{
165	int cpu;
166
167	if (!non_pcpu_rt->rt6i_pcpu)
168		return;
169
170	for_each_possible_cpu(cpu) {
171		struct rt6_info **ppcpu_rt;
172		struct rt6_info *pcpu_rt;
173
174		ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
175		pcpu_rt = *ppcpu_rt;
176		if (pcpu_rt) {
177			rt6_rcu_free(pcpu_rt);
178			*ppcpu_rt = NULL;
179		}
180	}
181
182	non_pcpu_rt->rt6i_pcpu = NULL;
183}
184
185static void rt6_release(struct rt6_info *rt)
186{
187	if (atomic_dec_and_test(&rt->rt6i_ref)) {
188		rt6_free_pcpu(rt);
189		rt6_rcu_free(rt);
190	}
191}
192
193static void fib6_link_table(struct net *net, struct fib6_table *tb)
194{
195	unsigned int h;
196
197	/*
198	 * Initialize table lock at a single place to give lockdep a key,
199	 * tables aren't visible prior to being linked to the list.
200	 */
201	rwlock_init(&tb->tb6_lock);
202
203	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
204
205	/*
206	 * No protection necessary, this is the only list mutatation
207	 * operation, tables never disappear once they exist.
208	 */
209	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
210}
211
212#ifdef CONFIG_IPV6_MULTIPLE_TABLES
213
214static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
215{
216	struct fib6_table *table;
217
218	table = kzalloc(sizeof(*table), GFP_ATOMIC);
219	if (table) {
220		table->tb6_id = id;
221		table->tb6_root.leaf = net->ipv6.ip6_null_entry;
222		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
223		inet_peer_base_init(&table->tb6_peers);
224	}
225
226	return table;
227}
228
229struct fib6_table *fib6_new_table(struct net *net, u32 id)
230{
231	struct fib6_table *tb;
232
233	if (id == 0)
234		id = RT6_TABLE_MAIN;
235	tb = fib6_get_table(net, id);
236	if (tb)
237		return tb;
238
239	tb = fib6_alloc_table(net, id);
240	if (tb)
241		fib6_link_table(net, tb);
242
243	return tb;
244}
245
246struct fib6_table *fib6_get_table(struct net *net, u32 id)
247{
248	struct fib6_table *tb;
249	struct hlist_head *head;
250	unsigned int h;
251
252	if (id == 0)
253		id = RT6_TABLE_MAIN;
254	h = id & (FIB6_TABLE_HASHSZ - 1);
255	rcu_read_lock();
256	head = &net->ipv6.fib_table_hash[h];
257	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
258		if (tb->tb6_id == id) {
259			rcu_read_unlock();
260			return tb;
261		}
262	}
263	rcu_read_unlock();
264
265	return NULL;
266}
267EXPORT_SYMBOL_GPL(fib6_get_table);
268
269static void __net_init fib6_tables_init(struct net *net)
270{
271	fib6_link_table(net, net->ipv6.fib6_main_tbl);
272	fib6_link_table(net, net->ipv6.fib6_local_tbl);
273}
274#else
275
276struct fib6_table *fib6_new_table(struct net *net, u32 id)
277{
278	return fib6_get_table(net, id);
279}
280
281struct fib6_table *fib6_get_table(struct net *net, u32 id)
282{
283	  return net->ipv6.fib6_main_tbl;
284}
285
286struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
287				   int flags, pol_lookup_t lookup)
288{
289	struct rt6_info *rt;
290
291	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
292	if (rt->rt6i_flags & RTF_REJECT &&
293	    rt->dst.error == -EAGAIN) {
294		ip6_rt_put(rt);
295		rt = net->ipv6.ip6_null_entry;
296		dst_hold(&rt->dst);
297	}
298
299	return &rt->dst;
300}
301
302static void __net_init fib6_tables_init(struct net *net)
303{
304	fib6_link_table(net, net->ipv6.fib6_main_tbl);
305}
306
307#endif
308
309static int fib6_dump_node(struct fib6_walker *w)
310{
311	int res;
312	struct rt6_info *rt;
313
314	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
315		res = rt6_dump_route(rt, w->args);
316		if (res < 0) {
317			/* Frame is full, suspend walking */
318			w->leaf = rt;
319			return 1;
320		}
321	}
322	w->leaf = NULL;
323	return 0;
324}
325
326static void fib6_dump_end(struct netlink_callback *cb)
327{
328	struct fib6_walker *w = (void *)cb->args[2];
329
330	if (w) {
331		if (cb->args[4]) {
332			cb->args[4] = 0;
333			fib6_walker_unlink(w);
334		}
335		cb->args[2] = 0;
336		kfree(w);
337	}
338	cb->done = (void *)cb->args[3];
339	cb->args[1] = 3;
340}
341
342static int fib6_dump_done(struct netlink_callback *cb)
343{
344	fib6_dump_end(cb);
345	return cb->done ? cb->done(cb) : 0;
346}
347
348static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
349			   struct netlink_callback *cb)
350{
351	struct fib6_walker *w;
352	int res;
353
354	w = (void *)cb->args[2];
355	w->root = &table->tb6_root;
356
357	if (cb->args[4] == 0) {
358		w->count = 0;
359		w->skip = 0;
360
361		read_lock_bh(&table->tb6_lock);
362		res = fib6_walk(w);
363		read_unlock_bh(&table->tb6_lock);
364		if (res > 0) {
365			cb->args[4] = 1;
366			cb->args[5] = w->root->fn_sernum;
367		}
368	} else {
369		if (cb->args[5] != w->root->fn_sernum) {
370			/* Begin at the root if the tree changed */
371			cb->args[5] = w->root->fn_sernum;
372			w->state = FWS_INIT;
373			w->node = w->root;
374			w->skip = w->count;
375		} else
376			w->skip = 0;
377
378		read_lock_bh(&table->tb6_lock);
379		res = fib6_walk_continue(w);
380		read_unlock_bh(&table->tb6_lock);
381		if (res <= 0) {
382			fib6_walker_unlink(w);
383			cb->args[4] = 0;
384		}
385	}
386
387	return res;
388}
389
390static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
391{
392	struct net *net = sock_net(skb->sk);
393	unsigned int h, s_h;
394	unsigned int e = 0, s_e;
395	struct rt6_rtnl_dump_arg arg;
396	struct fib6_walker *w;
397	struct fib6_table *tb;
398	struct hlist_head *head;
399	int res = 0;
400
401	s_h = cb->args[0];
402	s_e = cb->args[1];
403
404	w = (void *)cb->args[2];
405	if (!w) {
406		/* New dump:
407		 *
408		 * 1. hook callback destructor.
409		 */
410		cb->args[3] = (long)cb->done;
411		cb->done = fib6_dump_done;
412
413		/*
414		 * 2. allocate and initialize walker.
415		 */
416		w = kzalloc(sizeof(*w), GFP_ATOMIC);
417		if (!w)
418			return -ENOMEM;
419		w->func = fib6_dump_node;
420		cb->args[2] = (long)w;
421	}
422
423	arg.skb = skb;
424	arg.cb = cb;
425	arg.net = net;
426	w->args = &arg;
427
428	rcu_read_lock();
429	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
430		e = 0;
431		head = &net->ipv6.fib_table_hash[h];
432		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
433			if (e < s_e)
434				goto next;
435			res = fib6_dump_table(tb, skb, cb);
436			if (res != 0)
437				goto out;
438next:
439			e++;
440		}
441	}
442out:
443	rcu_read_unlock();
444	cb->args[1] = e;
445	cb->args[0] = h;
446
447	res = res < 0 ? res : skb->len;
448	if (res <= 0)
449		fib6_dump_end(cb);
450	return res;
451}
452
453/*
454 *	Routing Table
455 *
456 *	return the appropriate node for a routing tree "add" operation
457 *	by either creating and inserting or by returning an existing
458 *	node.
459 */
460
461static struct fib6_node *fib6_add_1(struct fib6_node *root,
462				     struct in6_addr *addr, int plen,
463				     int offset, int allow_create,
464				     int replace_required, int sernum)
465{
466	struct fib6_node *fn, *in, *ln;
467	struct fib6_node *pn = NULL;
468	struct rt6key *key;
469	int	bit;
470	__be32	dir = 0;
471
472	RT6_TRACE("fib6_add_1\n");
473
474	/* insert node in tree */
475
476	fn = root;
477
478	do {
479		key = (struct rt6key *)((u8 *)fn->leaf + offset);
480
481		/*
482		 *	Prefix match
483		 */
484		if (plen < fn->fn_bit ||
485		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
486			if (!allow_create) {
487				if (replace_required) {
488					pr_warn("Can't replace route, no match found\n");
489					return ERR_PTR(-ENOENT);
490				}
491				pr_warn("NLM_F_CREATE should be set when creating new route\n");
492			}
493			goto insert_above;
494		}
495
496		/*
497		 *	Exact match ?
498		 */
499
500		if (plen == fn->fn_bit) {
501			/* clean up an intermediate node */
502			if (!(fn->fn_flags & RTN_RTINFO)) {
503				rt6_release(fn->leaf);
504				fn->leaf = NULL;
505			}
506
507			fn->fn_sernum = sernum;
508
509			return fn;
510		}
511
512		/*
513		 *	We have more bits to go
514		 */
515
516		/* Try to walk down on tree. */
517		fn->fn_sernum = sernum;
518		dir = addr_bit_set(addr, fn->fn_bit);
519		pn = fn;
520		fn = dir ? fn->right : fn->left;
521	} while (fn);
522
523	if (!allow_create) {
524		/* We should not create new node because
525		 * NLM_F_REPLACE was specified without NLM_F_CREATE
526		 * I assume it is safe to require NLM_F_CREATE when
527		 * REPLACE flag is used! Later we may want to remove the
528		 * check for replace_required, because according
529		 * to netlink specification, NLM_F_CREATE
530		 * MUST be specified if new route is created.
531		 * That would keep IPv6 consistent with IPv4
532		 */
533		if (replace_required) {
534			pr_warn("Can't replace route, no match found\n");
535			return ERR_PTR(-ENOENT);
536		}
537		pr_warn("NLM_F_CREATE should be set when creating new route\n");
538	}
539	/*
540	 *	We walked to the bottom of tree.
541	 *	Create new leaf node without children.
542	 */
543
544	ln = node_alloc();
545
546	if (!ln)
547		return ERR_PTR(-ENOMEM);
548	ln->fn_bit = plen;
549
550	ln->parent = pn;
551	ln->fn_sernum = sernum;
552
553	if (dir)
554		pn->right = ln;
555	else
556		pn->left  = ln;
557
558	return ln;
559
560
561insert_above:
562	/*
563	 * split since we don't have a common prefix anymore or
564	 * we have a less significant route.
565	 * we've to insert an intermediate node on the list
566	 * this new node will point to the one we need to create
567	 * and the current
568	 */
569
570	pn = fn->parent;
571
572	/* find 1st bit in difference between the 2 addrs.
573
574	   See comment in __ipv6_addr_diff: bit may be an invalid value,
575	   but if it is >= plen, the value is ignored in any case.
576	 */
577
578	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
579
580	/*
581	 *		(intermediate)[in]
582	 *	          /	   \
583	 *	(new leaf node)[ln] (old node)[fn]
584	 */
585	if (plen > bit) {
586		in = node_alloc();
587		ln = node_alloc();
588
589		if (!in || !ln) {
590			if (in)
591				node_free(in);
592			if (ln)
593				node_free(ln);
594			return ERR_PTR(-ENOMEM);
595		}
596
597		/*
598		 * new intermediate node.
599		 * RTN_RTINFO will
600		 * be off since that an address that chooses one of
601		 * the branches would not match less specific routes
602		 * in the other branch
603		 */
604
605		in->fn_bit = bit;
606
607		in->parent = pn;
608		in->leaf = fn->leaf;
609		atomic_inc(&in->leaf->rt6i_ref);
610
611		in->fn_sernum = sernum;
612
613		/* update parent pointer */
614		if (dir)
615			pn->right = in;
616		else
617			pn->left  = in;
618
619		ln->fn_bit = plen;
620
621		ln->parent = in;
622		fn->parent = in;
623
624		ln->fn_sernum = sernum;
625
626		if (addr_bit_set(addr, bit)) {
627			in->right = ln;
628			in->left  = fn;
629		} else {
630			in->left  = ln;
631			in->right = fn;
632		}
633	} else { /* plen <= bit */
634
635		/*
636		 *		(new leaf node)[ln]
637		 *	          /	   \
638		 *	     (old node)[fn] NULL
639		 */
640
641		ln = node_alloc();
642
643		if (!ln)
644			return ERR_PTR(-ENOMEM);
645
646		ln->fn_bit = plen;
647
648		ln->parent = pn;
649
650		ln->fn_sernum = sernum;
651
652		if (dir)
653			pn->right = ln;
654		else
655			pn->left  = ln;
656
657		if (addr_bit_set(&key->addr, plen))
658			ln->right = fn;
659		else
660			ln->left  = fn;
661
662		fn->parent = ln;
663	}
664	return ln;
665}
666
667static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
668{
669	return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
670	       RTF_GATEWAY;
671}
672
673static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
674{
675	int i;
676
677	for (i = 0; i < RTAX_MAX; i++) {
678		if (test_bit(i, mxc->mx_valid))
679			mp[i] = mxc->mx[i];
680	}
681}
682
683static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
684{
685	if (!mxc->mx)
686		return 0;
687
688	if (dst->flags & DST_HOST) {
689		u32 *mp = dst_metrics_write_ptr(dst);
690
691		if (unlikely(!mp))
692			return -ENOMEM;
693
694		fib6_copy_metrics(mp, mxc);
695	} else {
696		dst_init_metrics(dst, mxc->mx, false);
697
698		/* We've stolen mx now. */
699		mxc->mx = NULL;
700	}
701
702	return 0;
703}
704
705static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
706			  struct net *net)
707{
708	if (atomic_read(&rt->rt6i_ref) != 1) {
709		/* This route is used as dummy address holder in some split
710		 * nodes. It is not leaked, but it still holds other resources,
711		 * which must be released in time. So, scan ascendant nodes
712		 * and replace dummy references to this route with references
713		 * to still alive ones.
714		 */
715		while (fn) {
716			if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
717				fn->leaf = fib6_find_prefix(net, fn);
718				atomic_inc(&fn->leaf->rt6i_ref);
719				rt6_release(rt);
720			}
721			fn = fn->parent;
722		}
723		/* No more references are possible at this point. */
724		BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
725	}
726}
727
728/*
729 *	Insert routing information in a node.
730 */
731
732static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
733			    struct nl_info *info, struct mx6_config *mxc)
734{
735	struct rt6_info *iter = NULL;
736	struct rt6_info **ins;
737	struct rt6_info **fallback_ins = NULL;
738	int replace = (info->nlh &&
739		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
740	int add = (!info->nlh ||
741		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
742	int found = 0;
743	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
744	int err;
745
746	ins = &fn->leaf;
747
748	for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
749		/*
750		 *	Search for duplicates
751		 */
752
753		if (iter->rt6i_metric == rt->rt6i_metric) {
754			/*
755			 *	Same priority level
756			 */
757			if (info->nlh &&
758			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
759				return -EEXIST;
760			if (replace) {
761				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
762					found++;
763					break;
764				}
765				if (rt_can_ecmp)
766					fallback_ins = fallback_ins ?: ins;
767				goto next_iter;
768			}
769
770			if (iter->dst.dev == rt->dst.dev &&
771			    iter->rt6i_idev == rt->rt6i_idev &&
772			    ipv6_addr_equal(&iter->rt6i_gateway,
773					    &rt->rt6i_gateway)) {
774				if (rt->rt6i_nsiblings)
775					rt->rt6i_nsiblings = 0;
776				if (!(iter->rt6i_flags & RTF_EXPIRES))
777					return -EEXIST;
778				if (!(rt->rt6i_flags & RTF_EXPIRES))
779					rt6_clean_expires(iter);
780				else
781					rt6_set_expires(iter, rt->dst.expires);
782				iter->rt6i_pmtu = rt->rt6i_pmtu;
783				return -EEXIST;
784			}
785			/* If we have the same destination and the same metric,
786			 * but not the same gateway, then the route we try to
787			 * add is sibling to this route, increment our counter
788			 * of siblings, and later we will add our route to the
789			 * list.
790			 * Only static routes (which don't have flag
791			 * RTF_EXPIRES) are used for ECMPv6.
792			 *
793			 * To avoid long list, we only had siblings if the
794			 * route have a gateway.
795			 */
796			if (rt_can_ecmp &&
797			    rt6_qualify_for_ecmp(iter))
798				rt->rt6i_nsiblings++;
799		}
800
801		if (iter->rt6i_metric > rt->rt6i_metric)
802			break;
803
804next_iter:
805		ins = &iter->dst.rt6_next;
806	}
807
808	if (fallback_ins && !found) {
809		/* No ECMP-able route found, replace first non-ECMP one */
810		ins = fallback_ins;
811		iter = *ins;
812		found++;
813	}
814
815	/* Reset round-robin state, if necessary */
816	if (ins == &fn->leaf)
817		fn->rr_ptr = NULL;
818
819	/* Link this route to others same route. */
820	if (rt->rt6i_nsiblings) {
821		unsigned int rt6i_nsiblings;
822		struct rt6_info *sibling, *temp_sibling;
823
824		/* Find the first route that have the same metric */
825		sibling = fn->leaf;
826		while (sibling) {
827			if (sibling->rt6i_metric == rt->rt6i_metric &&
828			    rt6_qualify_for_ecmp(sibling)) {
829				list_add_tail(&rt->rt6i_siblings,
830					      &sibling->rt6i_siblings);
831				break;
832			}
833			sibling = sibling->dst.rt6_next;
834		}
835		/* For each sibling in the list, increment the counter of
836		 * siblings. BUG() if counters does not match, list of siblings
837		 * is broken!
838		 */
839		rt6i_nsiblings = 0;
840		list_for_each_entry_safe(sibling, temp_sibling,
841					 &rt->rt6i_siblings, rt6i_siblings) {
842			sibling->rt6i_nsiblings++;
843			BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
844			rt6i_nsiblings++;
845		}
846		BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
847	}
848
849	/*
850	 *	insert node
851	 */
852	if (!replace) {
853		if (!add)
854			pr_warn("NLM_F_CREATE should be set when creating new route\n");
855
856add:
857		err = fib6_commit_metrics(&rt->dst, mxc);
858		if (err)
859			return err;
860
861		rt->dst.rt6_next = iter;
862		*ins = rt;
863		rt->rt6i_node = fn;
864		atomic_inc(&rt->rt6i_ref);
865		inet6_rt_notify(RTM_NEWROUTE, rt, info, 0);
866		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
867
868		if (!(fn->fn_flags & RTN_RTINFO)) {
869			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
870			fn->fn_flags |= RTN_RTINFO;
871		}
872
873	} else {
874		int nsiblings;
875
876		if (!found) {
877			if (add)
878				goto add;
879			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
880			return -ENOENT;
881		}
882
883		err = fib6_commit_metrics(&rt->dst, mxc);
884		if (err)
885			return err;
886
887		*ins = rt;
888		rt->rt6i_node = fn;
889		rt->dst.rt6_next = iter->dst.rt6_next;
890		atomic_inc(&rt->rt6i_ref);
891		inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
892		if (!(fn->fn_flags & RTN_RTINFO)) {
893			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
894			fn->fn_flags |= RTN_RTINFO;
895		}
896		nsiblings = iter->rt6i_nsiblings;
897		fib6_purge_rt(iter, fn, info->nl_net);
898		rt6_release(iter);
899
900		if (nsiblings) {
901			/* Replacing an ECMP route, remove all siblings */
902			ins = &rt->dst.rt6_next;
903			iter = *ins;
904			while (iter) {
905				if (rt6_qualify_for_ecmp(iter)) {
906					*ins = iter->dst.rt6_next;
907					fib6_purge_rt(iter, fn, info->nl_net);
908					rt6_release(iter);
909					nsiblings--;
910				} else {
911					ins = &iter->dst.rt6_next;
912				}
913				iter = *ins;
914			}
915			WARN_ON(nsiblings != 0);
916		}
917	}
918
919	return 0;
920}
921
922static void fib6_start_gc(struct net *net, struct rt6_info *rt)
923{
924	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
925	    (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
926		mod_timer(&net->ipv6.ip6_fib_timer,
927			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
928}
929
930void fib6_force_start_gc(struct net *net)
931{
932	if (!timer_pending(&net->ipv6.ip6_fib_timer))
933		mod_timer(&net->ipv6.ip6_fib_timer,
934			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
935}
936
937/*
938 *	Add routing information to the routing tree.
939 *	<destination addr>/<source addr>
940 *	with source addr info in sub-trees
941 */
942
943int fib6_add(struct fib6_node *root, struct rt6_info *rt,
944	     struct nl_info *info, struct mx6_config *mxc)
945{
946	struct fib6_node *fn, *pn = NULL;
947	int err = -ENOMEM;
948	int allow_create = 1;
949	int replace_required = 0;
950	int sernum = fib6_new_sernum(info->nl_net);
951
952	if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) &&
953			 !atomic_read(&rt->dst.__refcnt)))
954		return -EINVAL;
955
956	if (info->nlh) {
957		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
958			allow_create = 0;
959		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
960			replace_required = 1;
961	}
962	if (!allow_create && !replace_required)
963		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
964
965	fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
966			offsetof(struct rt6_info, rt6i_dst), allow_create,
967			replace_required, sernum);
968	if (IS_ERR(fn)) {
969		err = PTR_ERR(fn);
970		fn = NULL;
971		goto out;
972	}
973
974	pn = fn;
975
976#ifdef CONFIG_IPV6_SUBTREES
977	if (rt->rt6i_src.plen) {
978		struct fib6_node *sn;
979
980		if (!fn->subtree) {
981			struct fib6_node *sfn;
982
983			/*
984			 * Create subtree.
985			 *
986			 *		fn[main tree]
987			 *		|
988			 *		sfn[subtree root]
989			 *		   \
990			 *		    sn[new leaf node]
991			 */
992
993			/* Create subtree root node */
994			sfn = node_alloc();
995			if (!sfn)
996				goto st_failure;
997
998			sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
999			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1000			sfn->fn_flags = RTN_ROOT;
1001			sfn->fn_sernum = sernum;
1002
1003			/* Now add the first leaf node to new subtree */
1004
1005			sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
1006					rt->rt6i_src.plen,
1007					offsetof(struct rt6_info, rt6i_src),
1008					allow_create, replace_required, sernum);
1009
1010			if (IS_ERR(sn)) {
1011				/* If it is failed, discard just allocated
1012				   root, and then (in st_failure) stale node
1013				   in main tree.
1014				 */
1015				node_free(sfn);
1016				err = PTR_ERR(sn);
1017				goto st_failure;
1018			}
1019
1020			/* Now link new subtree to main tree */
1021			sfn->parent = fn;
1022			fn->subtree = sfn;
1023		} else {
1024			sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1025					rt->rt6i_src.plen,
1026					offsetof(struct rt6_info, rt6i_src),
1027					allow_create, replace_required, sernum);
1028
1029			if (IS_ERR(sn)) {
1030				err = PTR_ERR(sn);
1031				goto st_failure;
1032			}
1033		}
1034
1035		if (!fn->leaf) {
1036			fn->leaf = rt;
1037			atomic_inc(&rt->rt6i_ref);
1038		}
1039		fn = sn;
1040	}
1041#endif
1042
1043	err = fib6_add_rt2node(fn, rt, info, mxc);
1044	if (!err) {
1045		fib6_start_gc(info->nl_net, rt);
1046		if (!(rt->rt6i_flags & RTF_CACHE))
1047			fib6_prune_clones(info->nl_net, pn);
1048		rt->dst.flags &= ~DST_NOCACHE;
1049	}
1050
1051out:
1052	if (err) {
1053#ifdef CONFIG_IPV6_SUBTREES
1054		/*
1055		 * If fib6_add_1 has cleared the old leaf pointer in the
1056		 * super-tree leaf node we have to find a new one for it.
1057		 */
1058		if (pn != fn && pn->leaf == rt) {
1059			pn->leaf = NULL;
1060			atomic_dec(&rt->rt6i_ref);
1061		}
1062		if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1063			pn->leaf = fib6_find_prefix(info->nl_net, pn);
1064#if RT6_DEBUG >= 2
1065			if (!pn->leaf) {
1066				WARN_ON(pn->leaf == NULL);
1067				pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1068			}
1069#endif
1070			atomic_inc(&pn->leaf->rt6i_ref);
1071		}
1072#endif
1073		if (!(rt->dst.flags & DST_NOCACHE))
1074			dst_free(&rt->dst);
1075	}
1076	return err;
1077
1078#ifdef CONFIG_IPV6_SUBTREES
1079	/* Subtree creation failed, probably main tree node
1080	   is orphan. If it is, shoot it.
1081	 */
1082st_failure:
1083	if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1084		fib6_repair_tree(info->nl_net, fn);
1085	if (!(rt->dst.flags & DST_NOCACHE))
1086		dst_free(&rt->dst);
1087	return err;
1088#endif
1089}
1090
1091/*
1092 *	Routing tree lookup
1093 *
1094 */
1095
1096struct lookup_args {
1097	int			offset;		/* key offset on rt6_info	*/
1098	const struct in6_addr	*addr;		/* search key			*/
1099};
1100
1101static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1102				       struct lookup_args *args)
1103{
1104	struct fib6_node *fn;
1105	__be32 dir;
1106
1107	if (unlikely(args->offset == 0))
1108		return NULL;
1109
1110	/*
1111	 *	Descend on a tree
1112	 */
1113
1114	fn = root;
1115
1116	for (;;) {
1117		struct fib6_node *next;
1118
1119		dir = addr_bit_set(args->addr, fn->fn_bit);
1120
1121		next = dir ? fn->right : fn->left;
1122
1123		if (next) {
1124			fn = next;
1125			continue;
1126		}
1127		break;
1128	}
1129
1130	while (fn) {
1131		if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1132			struct rt6key *key;
1133
1134			key = (struct rt6key *) ((u8 *) fn->leaf +
1135						 args->offset);
1136
1137			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1138#ifdef CONFIG_IPV6_SUBTREES
1139				if (fn->subtree) {
1140					struct fib6_node *sfn;
1141					sfn = fib6_lookup_1(fn->subtree,
1142							    args + 1);
1143					if (!sfn)
1144						goto backtrack;
1145					fn = sfn;
1146				}
1147#endif
1148				if (fn->fn_flags & RTN_RTINFO)
1149					return fn;
1150			}
1151		}
1152#ifdef CONFIG_IPV6_SUBTREES
1153backtrack:
1154#endif
1155		if (fn->fn_flags & RTN_ROOT)
1156			break;
1157
1158		fn = fn->parent;
1159	}
1160
1161	return NULL;
1162}
1163
1164struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1165			      const struct in6_addr *saddr)
1166{
1167	struct fib6_node *fn;
1168	struct lookup_args args[] = {
1169		{
1170			.offset = offsetof(struct rt6_info, rt6i_dst),
1171			.addr = daddr,
1172		},
1173#ifdef CONFIG_IPV6_SUBTREES
1174		{
1175			.offset = offsetof(struct rt6_info, rt6i_src),
1176			.addr = saddr,
1177		},
1178#endif
1179		{
1180			.offset = 0,	/* sentinel */
1181		}
1182	};
1183
1184	fn = fib6_lookup_1(root, daddr ? args : args + 1);
1185	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1186		fn = root;
1187
1188	return fn;
1189}
1190
1191/*
1192 *	Get node with specified destination prefix (and source prefix,
1193 *	if subtrees are used)
1194 */
1195
1196
1197static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1198				       const struct in6_addr *addr,
1199				       int plen, int offset)
1200{
1201	struct fib6_node *fn;
1202
1203	for (fn = root; fn ; ) {
1204		struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1205
1206		/*
1207		 *	Prefix match
1208		 */
1209		if (plen < fn->fn_bit ||
1210		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1211			return NULL;
1212
1213		if (plen == fn->fn_bit)
1214			return fn;
1215
1216		/*
1217		 *	We have more bits to go
1218		 */
1219		if (addr_bit_set(addr, fn->fn_bit))
1220			fn = fn->right;
1221		else
1222			fn = fn->left;
1223	}
1224	return NULL;
1225}
1226
1227struct fib6_node *fib6_locate(struct fib6_node *root,
1228			      const struct in6_addr *daddr, int dst_len,
1229			      const struct in6_addr *saddr, int src_len)
1230{
1231	struct fib6_node *fn;
1232
1233	fn = fib6_locate_1(root, daddr, dst_len,
1234			   offsetof(struct rt6_info, rt6i_dst));
1235
1236#ifdef CONFIG_IPV6_SUBTREES
1237	if (src_len) {
1238		WARN_ON(saddr == NULL);
1239		if (fn && fn->subtree)
1240			fn = fib6_locate_1(fn->subtree, saddr, src_len,
1241					   offsetof(struct rt6_info, rt6i_src));
1242	}
1243#endif
1244
1245	if (fn && fn->fn_flags & RTN_RTINFO)
1246		return fn;
1247
1248	return NULL;
1249}
1250
1251
1252/*
1253 *	Deletion
1254 *
1255 */
1256
1257static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1258{
1259	if (fn->fn_flags & RTN_ROOT)
1260		return net->ipv6.ip6_null_entry;
1261
1262	while (fn) {
1263		if (fn->left)
1264			return fn->left->leaf;
1265		if (fn->right)
1266			return fn->right->leaf;
1267
1268		fn = FIB6_SUBTREE(fn);
1269	}
1270	return NULL;
1271}
1272
1273/*
1274 *	Called to trim the tree of intermediate nodes when possible. "fn"
1275 *	is the node we want to try and remove.
1276 */
1277
1278static struct fib6_node *fib6_repair_tree(struct net *net,
1279					   struct fib6_node *fn)
1280{
1281	int children;
1282	int nstate;
1283	struct fib6_node *child, *pn;
1284	struct fib6_walker *w;
1285	int iter = 0;
1286
1287	for (;;) {
1288		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1289		iter++;
1290
1291		WARN_ON(fn->fn_flags & RTN_RTINFO);
1292		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1293		WARN_ON(fn->leaf);
1294
1295		children = 0;
1296		child = NULL;
1297		if (fn->right)
1298			child = fn->right, children |= 1;
1299		if (fn->left)
1300			child = fn->left, children |= 2;
1301
1302		if (children == 3 || FIB6_SUBTREE(fn)
1303#ifdef CONFIG_IPV6_SUBTREES
1304		    /* Subtree root (i.e. fn) may have one child */
1305		    || (children && fn->fn_flags & RTN_ROOT)
1306#endif
1307		    ) {
1308			fn->leaf = fib6_find_prefix(net, fn);
1309#if RT6_DEBUG >= 2
1310			if (!fn->leaf) {
1311				WARN_ON(!fn->leaf);
1312				fn->leaf = net->ipv6.ip6_null_entry;
1313			}
1314#endif
1315			atomic_inc(&fn->leaf->rt6i_ref);
1316			return fn->parent;
1317		}
1318
1319		pn = fn->parent;
1320#ifdef CONFIG_IPV6_SUBTREES
1321		if (FIB6_SUBTREE(pn) == fn) {
1322			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1323			FIB6_SUBTREE(pn) = NULL;
1324			nstate = FWS_L;
1325		} else {
1326			WARN_ON(fn->fn_flags & RTN_ROOT);
1327#endif
1328			if (pn->right == fn)
1329				pn->right = child;
1330			else if (pn->left == fn)
1331				pn->left = child;
1332#if RT6_DEBUG >= 2
1333			else
1334				WARN_ON(1);
1335#endif
1336			if (child)
1337				child->parent = pn;
1338			nstate = FWS_R;
1339#ifdef CONFIG_IPV6_SUBTREES
1340		}
1341#endif
1342
1343		read_lock(&fib6_walker_lock);
1344		FOR_WALKERS(w) {
1345			if (!child) {
1346				if (w->root == fn) {
1347					w->root = w->node = NULL;
1348					RT6_TRACE("W %p adjusted by delroot 1\n", w);
1349				} else if (w->node == fn) {
1350					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1351					w->node = pn;
1352					w->state = nstate;
1353				}
1354			} else {
1355				if (w->root == fn) {
1356					w->root = child;
1357					RT6_TRACE("W %p adjusted by delroot 2\n", w);
1358				}
1359				if (w->node == fn) {
1360					w->node = child;
1361					if (children&2) {
1362						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1363						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1364					} else {
1365						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1366						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1367					}
1368				}
1369			}
1370		}
1371		read_unlock(&fib6_walker_lock);
1372
1373		node_free(fn);
1374		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1375			return pn;
1376
1377		rt6_release(pn->leaf);
1378		pn->leaf = NULL;
1379		fn = pn;
1380	}
1381}
1382
1383static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1384			   struct nl_info *info)
1385{
1386	struct fib6_walker *w;
1387	struct rt6_info *rt = *rtp;
1388	struct net *net = info->nl_net;
1389
1390	RT6_TRACE("fib6_del_route\n");
1391
1392	/* Unlink it */
1393	*rtp = rt->dst.rt6_next;
1394	rt->rt6i_node = NULL;
1395	net->ipv6.rt6_stats->fib_rt_entries--;
1396	net->ipv6.rt6_stats->fib_discarded_routes++;
1397
1398	/* Reset round-robin state, if necessary */
1399	if (fn->rr_ptr == rt)
1400		fn->rr_ptr = NULL;
1401
1402	/* Remove this entry from other siblings */
1403	if (rt->rt6i_nsiblings) {
1404		struct rt6_info *sibling, *next_sibling;
1405
1406		list_for_each_entry_safe(sibling, next_sibling,
1407					 &rt->rt6i_siblings, rt6i_siblings)
1408			sibling->rt6i_nsiblings--;
1409		rt->rt6i_nsiblings = 0;
1410		list_del_init(&rt->rt6i_siblings);
1411	}
1412
1413	/* Adjust walkers */
1414	read_lock(&fib6_walker_lock);
1415	FOR_WALKERS(w) {
1416		if (w->state == FWS_C && w->leaf == rt) {
1417			RT6_TRACE("walker %p adjusted by delroute\n", w);
1418			w->leaf = rt->dst.rt6_next;
1419			if (!w->leaf)
1420				w->state = FWS_U;
1421		}
1422	}
1423	read_unlock(&fib6_walker_lock);
1424
1425	rt->dst.rt6_next = NULL;
1426
1427	/* If it was last route, expunge its radix tree node */
1428	if (!fn->leaf) {
1429		fn->fn_flags &= ~RTN_RTINFO;
1430		net->ipv6.rt6_stats->fib_route_nodes--;
1431		fn = fib6_repair_tree(net, fn);
1432	}
1433
1434	fib6_purge_rt(rt, fn, net);
1435
1436	inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1437	rt6_release(rt);
1438}
1439
1440int fib6_del(struct rt6_info *rt, struct nl_info *info)
1441{
1442	struct net *net = info->nl_net;
1443	struct fib6_node *fn = rt->rt6i_node;
1444	struct rt6_info **rtp;
1445
1446#if RT6_DEBUG >= 2
1447	if (rt->dst.obsolete > 0) {
1448		WARN_ON(fn);
1449		return -ENOENT;
1450	}
1451#endif
1452	if (!fn || rt == net->ipv6.ip6_null_entry)
1453		return -ENOENT;
1454
1455	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1456
1457	if (!(rt->rt6i_flags & RTF_CACHE)) {
1458		struct fib6_node *pn = fn;
1459#ifdef CONFIG_IPV6_SUBTREES
1460		/* clones of this route might be in another subtree */
1461		if (rt->rt6i_src.plen) {
1462			while (!(pn->fn_flags & RTN_ROOT))
1463				pn = pn->parent;
1464			pn = pn->parent;
1465		}
1466#endif
1467		fib6_prune_clones(info->nl_net, pn);
1468	}
1469
1470	/*
1471	 *	Walk the leaf entries looking for ourself
1472	 */
1473
1474	for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1475		if (*rtp == rt) {
1476			fib6_del_route(fn, rtp, info);
1477			return 0;
1478		}
1479	}
1480	return -ENOENT;
1481}
1482
1483/*
1484 *	Tree traversal function.
1485 *
1486 *	Certainly, it is not interrupt safe.
1487 *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1488 *	It means, that we can modify tree during walking
1489 *	and use this function for garbage collection, clone pruning,
1490 *	cleaning tree when a device goes down etc. etc.
1491 *
1492 *	It guarantees that every node will be traversed,
1493 *	and that it will be traversed only once.
1494 *
1495 *	Callback function w->func may return:
1496 *	0 -> continue walking.
1497 *	positive value -> walking is suspended (used by tree dumps,
1498 *	and probably by gc, if it will be split to several slices)
1499 *	negative value -> terminate walking.
1500 *
1501 *	The function itself returns:
1502 *	0   -> walk is complete.
1503 *	>0  -> walk is incomplete (i.e. suspended)
1504 *	<0  -> walk is terminated by an error.
1505 */
1506
1507static int fib6_walk_continue(struct fib6_walker *w)
1508{
1509	struct fib6_node *fn, *pn;
1510
1511	for (;;) {
1512		fn = w->node;
1513		if (!fn)
1514			return 0;
1515
1516		if (w->prune && fn != w->root &&
1517		    fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1518			w->state = FWS_C;
1519			w->leaf = fn->leaf;
1520		}
1521		switch (w->state) {
1522#ifdef CONFIG_IPV6_SUBTREES
1523		case FWS_S:
1524			if (FIB6_SUBTREE(fn)) {
1525				w->node = FIB6_SUBTREE(fn);
1526				continue;
1527			}
1528			w->state = FWS_L;
1529#endif
1530		case FWS_L:
1531			if (fn->left) {
1532				w->node = fn->left;
1533				w->state = FWS_INIT;
1534				continue;
1535			}
1536			w->state = FWS_R;
1537		case FWS_R:
1538			if (fn->right) {
1539				w->node = fn->right;
1540				w->state = FWS_INIT;
1541				continue;
1542			}
1543			w->state = FWS_C;
1544			w->leaf = fn->leaf;
1545		case FWS_C:
1546			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1547				int err;
1548
1549				if (w->skip) {
1550					w->skip--;
1551					goto skip;
1552				}
1553
1554				err = w->func(w);
1555				if (err)
1556					return err;
1557
1558				w->count++;
1559				continue;
1560			}
1561skip:
1562			w->state = FWS_U;
1563		case FWS_U:
1564			if (fn == w->root)
1565				return 0;
1566			pn = fn->parent;
1567			w->node = pn;
1568#ifdef CONFIG_IPV6_SUBTREES
1569			if (FIB6_SUBTREE(pn) == fn) {
1570				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1571				w->state = FWS_L;
1572				continue;
1573			}
1574#endif
1575			if (pn->left == fn) {
1576				w->state = FWS_R;
1577				continue;
1578			}
1579			if (pn->right == fn) {
1580				w->state = FWS_C;
1581				w->leaf = w->node->leaf;
1582				continue;
1583			}
1584#if RT6_DEBUG >= 2
1585			WARN_ON(1);
1586#endif
1587		}
1588	}
1589}
1590
1591static int fib6_walk(struct fib6_walker *w)
1592{
1593	int res;
1594
1595	w->state = FWS_INIT;
1596	w->node = w->root;
1597
1598	fib6_walker_link(w);
1599	res = fib6_walk_continue(w);
1600	if (res <= 0)
1601		fib6_walker_unlink(w);
1602	return res;
1603}
1604
1605static int fib6_clean_node(struct fib6_walker *w)
1606{
1607	int res;
1608	struct rt6_info *rt;
1609	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1610	struct nl_info info = {
1611		.nl_net = c->net,
1612	};
1613
1614	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1615	    w->node->fn_sernum != c->sernum)
1616		w->node->fn_sernum = c->sernum;
1617
1618	if (!c->func) {
1619		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1620		w->leaf = NULL;
1621		return 0;
1622	}
1623
1624	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1625		res = c->func(rt, c->arg);
1626		if (res < 0) {
1627			w->leaf = rt;
1628			res = fib6_del(rt, &info);
1629			if (res) {
1630#if RT6_DEBUG >= 2
1631				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1632					 __func__, rt, rt->rt6i_node, res);
1633#endif
1634				continue;
1635			}
1636			return 0;
1637		}
1638		WARN_ON(res != 0);
1639	}
1640	w->leaf = rt;
1641	return 0;
1642}
1643
1644/*
1645 *	Convenient frontend to tree walker.
1646 *
1647 *	func is called on each route.
1648 *		It may return -1 -> delete this route.
1649 *		              0  -> continue walking
1650 *
1651 *	prune==1 -> only immediate children of node (certainly,
1652 *	ignoring pure split nodes) will be scanned.
1653 */
1654
1655static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1656			    int (*func)(struct rt6_info *, void *arg),
1657			    bool prune, int sernum, void *arg)
1658{
1659	struct fib6_cleaner c;
1660
1661	c.w.root = root;
1662	c.w.func = fib6_clean_node;
1663	c.w.prune = prune;
1664	c.w.count = 0;
1665	c.w.skip = 0;
1666	c.func = func;
1667	c.sernum = sernum;
1668	c.arg = arg;
1669	c.net = net;
1670
1671	fib6_walk(&c.w);
1672}
1673
1674static void __fib6_clean_all(struct net *net,
1675			     int (*func)(struct rt6_info *, void *),
1676			     int sernum, void *arg)
1677{
1678	struct fib6_table *table;
1679	struct hlist_head *head;
1680	unsigned int h;
1681
1682	rcu_read_lock();
1683	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1684		head = &net->ipv6.fib_table_hash[h];
1685		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1686			write_lock_bh(&table->tb6_lock);
1687			fib6_clean_tree(net, &table->tb6_root,
1688					func, false, sernum, arg);
1689			write_unlock_bh(&table->tb6_lock);
1690		}
1691	}
1692	rcu_read_unlock();
1693}
1694
1695void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1696		    void *arg)
1697{
1698	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1699}
1700
1701static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1702{
1703	if (rt->rt6i_flags & RTF_CACHE) {
1704		RT6_TRACE("pruning clone %p\n", rt);
1705		return -1;
1706	}
1707
1708	return 0;
1709}
1710
1711static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1712{
1713	fib6_clean_tree(net, fn, fib6_prune_clone, true,
1714			FIB6_NO_SERNUM_CHANGE, NULL);
1715}
1716
1717static void fib6_flush_trees(struct net *net)
1718{
1719	int new_sernum = fib6_new_sernum(net);
1720
1721	__fib6_clean_all(net, NULL, new_sernum, NULL);
1722}
1723
1724/*
1725 *	Garbage collection
1726 */
1727
1728static struct fib6_gc_args
1729{
1730	int			timeout;
1731	int			more;
1732} gc_args;
1733
1734static int fib6_age(struct rt6_info *rt, void *arg)
1735{
1736	unsigned long now = jiffies;
1737
1738	/*
1739	 *	check addrconf expiration here.
1740	 *	Routes are expired even if they are in use.
1741	 *
1742	 *	Also age clones. Note, that clones are aged out
1743	 *	only if they are not in use now.
1744	 */
1745
1746	if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1747		if (time_after(now, rt->dst.expires)) {
1748			RT6_TRACE("expiring %p\n", rt);
1749			return -1;
1750		}
1751		gc_args.more++;
1752	} else if (rt->rt6i_flags & RTF_CACHE) {
1753		if (atomic_read(&rt->dst.__refcnt) == 0 &&
1754		    time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1755			RT6_TRACE("aging clone %p\n", rt);
1756			return -1;
1757		} else if (rt->rt6i_flags & RTF_GATEWAY) {
1758			struct neighbour *neigh;
1759			__u8 neigh_flags = 0;
1760
1761			neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1762			if (neigh) {
1763				neigh_flags = neigh->flags;
1764				neigh_release(neigh);
1765			}
1766			if (!(neigh_flags & NTF_ROUTER)) {
1767				RT6_TRACE("purging route %p via non-router but gateway\n",
1768					  rt);
1769				return -1;
1770			}
1771		}
1772		gc_args.more++;
1773	}
1774
1775	return 0;
1776}
1777
1778static DEFINE_SPINLOCK(fib6_gc_lock);
1779
1780void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1781{
1782	unsigned long now;
1783
1784	if (force) {
1785		spin_lock_bh(&fib6_gc_lock);
1786	} else if (!spin_trylock_bh(&fib6_gc_lock)) {
1787		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1788		return;
1789	}
1790	gc_args.timeout = expires ? (int)expires :
1791			  net->ipv6.sysctl.ip6_rt_gc_interval;
1792
1793	gc_args.more = icmp6_dst_gc();
1794
1795	fib6_clean_all(net, fib6_age, NULL);
1796	now = jiffies;
1797	net->ipv6.ip6_rt_last_gc = now;
1798
1799	if (gc_args.more)
1800		mod_timer(&net->ipv6.ip6_fib_timer,
1801			  round_jiffies(now
1802					+ net->ipv6.sysctl.ip6_rt_gc_interval));
1803	else
1804		del_timer(&net->ipv6.ip6_fib_timer);
1805	spin_unlock_bh(&fib6_gc_lock);
1806}
1807
1808static void fib6_gc_timer_cb(unsigned long arg)
1809{
1810	fib6_run_gc(0, (struct net *)arg, true);
1811}
1812
1813static int __net_init fib6_net_init(struct net *net)
1814{
1815	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1816
1817	setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1818
1819	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1820	if (!net->ipv6.rt6_stats)
1821		goto out_timer;
1822
1823	/* Avoid false sharing : Use at least a full cache line */
1824	size = max_t(size_t, size, L1_CACHE_BYTES);
1825
1826	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1827	if (!net->ipv6.fib_table_hash)
1828		goto out_rt6_stats;
1829
1830	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1831					  GFP_KERNEL);
1832	if (!net->ipv6.fib6_main_tbl)
1833		goto out_fib_table_hash;
1834
1835	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1836	net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1837	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1838		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1839	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1840
1841#ifdef CONFIG_IPV6_MULTIPLE_TABLES
1842	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1843					   GFP_KERNEL);
1844	if (!net->ipv6.fib6_local_tbl)
1845		goto out_fib6_main_tbl;
1846	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1847	net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1848	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1849		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1850	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1851#endif
1852	fib6_tables_init(net);
1853
1854	return 0;
1855
1856#ifdef CONFIG_IPV6_MULTIPLE_TABLES
1857out_fib6_main_tbl:
1858	kfree(net->ipv6.fib6_main_tbl);
1859#endif
1860out_fib_table_hash:
1861	kfree(net->ipv6.fib_table_hash);
1862out_rt6_stats:
1863	kfree(net->ipv6.rt6_stats);
1864out_timer:
1865	return -ENOMEM;
1866}
1867
1868static void fib6_net_exit(struct net *net)
1869{
1870	rt6_ifdown(net, NULL);
1871	del_timer_sync(&net->ipv6.ip6_fib_timer);
1872
1873#ifdef CONFIG_IPV6_MULTIPLE_TABLES
1874	inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1875	kfree(net->ipv6.fib6_local_tbl);
1876#endif
1877	inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1878	kfree(net->ipv6.fib6_main_tbl);
1879	kfree(net->ipv6.fib_table_hash);
1880	kfree(net->ipv6.rt6_stats);
1881}
1882
1883static struct pernet_operations fib6_net_ops = {
1884	.init = fib6_net_init,
1885	.exit = fib6_net_exit,
1886};
1887
1888int __init fib6_init(void)
1889{
1890	int ret = -ENOMEM;
1891
1892	fib6_node_kmem = kmem_cache_create("fib6_nodes",
1893					   sizeof(struct fib6_node),
1894					   0, SLAB_HWCACHE_ALIGN,
1895					   NULL);
1896	if (!fib6_node_kmem)
1897		goto out;
1898
1899	ret = register_pernet_subsys(&fib6_net_ops);
1900	if (ret)
1901		goto out_kmem_cache_create;
1902
1903	ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1904			      NULL);
1905	if (ret)
1906		goto out_unregister_subsys;
1907
1908	__fib6_flush_trees = fib6_flush_trees;
1909out:
1910	return ret;
1911
1912out_unregister_subsys:
1913	unregister_pernet_subsys(&fib6_net_ops);
1914out_kmem_cache_create:
1915	kmem_cache_destroy(fib6_node_kmem);
1916	goto out;
1917}
1918
1919void fib6_gc_cleanup(void)
1920{
1921	unregister_pernet_subsys(&fib6_net_ops);
1922	kmem_cache_destroy(fib6_node_kmem);
1923}
1924
1925#ifdef CONFIG_PROC_FS
1926
1927struct ipv6_route_iter {
1928	struct seq_net_private p;
1929	struct fib6_walker w;
1930	loff_t skip;
1931	struct fib6_table *tbl;
1932	int sernum;
1933};
1934
1935static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1936{
1937	struct rt6_info *rt = v;
1938	struct ipv6_route_iter *iter = seq->private;
1939
1940	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1941
1942#ifdef CONFIG_IPV6_SUBTREES
1943	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1944#else
1945	seq_puts(seq, "00000000000000000000000000000000 00 ");
1946#endif
1947	if (rt->rt6i_flags & RTF_GATEWAY)
1948		seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1949	else
1950		seq_puts(seq, "00000000000000000000000000000000");
1951
1952	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1953		   rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1954		   rt->dst.__use, rt->rt6i_flags,
1955		   rt->dst.dev ? rt->dst.dev->name : "");
1956	iter->w.leaf = NULL;
1957	return 0;
1958}
1959
1960static int ipv6_route_yield(struct fib6_walker *w)
1961{
1962	struct ipv6_route_iter *iter = w->args;
1963
1964	if (!iter->skip)
1965		return 1;
1966
1967	do {
1968		iter->w.leaf = iter->w.leaf->dst.rt6_next;
1969		iter->skip--;
1970		if (!iter->skip && iter->w.leaf)
1971			return 1;
1972	} while (iter->w.leaf);
1973
1974	return 0;
1975}
1976
1977static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
1978{
1979	memset(&iter->w, 0, sizeof(iter->w));
1980	iter->w.func = ipv6_route_yield;
1981	iter->w.root = &iter->tbl->tb6_root;
1982	iter->w.state = FWS_INIT;
1983	iter->w.node = iter->w.root;
1984	iter->w.args = iter;
1985	iter->sernum = iter->w.root->fn_sernum;
1986	INIT_LIST_HEAD(&iter->w.lh);
1987	fib6_walker_link(&iter->w);
1988}
1989
1990static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1991						    struct net *net)
1992{
1993	unsigned int h;
1994	struct hlist_node *node;
1995
1996	if (tbl) {
1997		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
1998		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
1999	} else {
2000		h = 0;
2001		node = NULL;
2002	}
2003
2004	while (!node && h < FIB6_TABLE_HASHSZ) {
2005		node = rcu_dereference_bh(
2006			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2007	}
2008	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2009}
2010
2011static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2012{
2013	if (iter->sernum != iter->w.root->fn_sernum) {
2014		iter->sernum = iter->w.root->fn_sernum;
2015		iter->w.state = FWS_INIT;
2016		iter->w.node = iter->w.root;
2017		WARN_ON(iter->w.skip);
2018		iter->w.skip = iter->w.count;
2019	}
2020}
2021
2022static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2023{
2024	int r;
2025	struct rt6_info *n;
2026	struct net *net = seq_file_net(seq);
2027	struct ipv6_route_iter *iter = seq->private;
2028
2029	if (!v)
2030		goto iter_table;
2031
2032	n = ((struct rt6_info *)v)->dst.rt6_next;
2033	if (n) {
2034		++*pos;
2035		return n;
2036	}
2037
2038iter_table:
2039	ipv6_route_check_sernum(iter);
2040	read_lock(&iter->tbl->tb6_lock);
2041	r = fib6_walk_continue(&iter->w);
2042	read_unlock(&iter->tbl->tb6_lock);
2043	if (r > 0) {
2044		if (v)
2045			++*pos;
2046		return iter->w.leaf;
2047	} else if (r < 0) {
2048		fib6_walker_unlink(&iter->w);
2049		return NULL;
2050	}
2051	fib6_walker_unlink(&iter->w);
2052
2053	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2054	if (!iter->tbl)
2055		return NULL;
2056
2057	ipv6_route_seq_setup_walk(iter);
2058	goto iter_table;
2059}
2060
2061static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2062	__acquires(RCU_BH)
2063{
2064	struct net *net = seq_file_net(seq);
2065	struct ipv6_route_iter *iter = seq->private;
2066
2067	rcu_read_lock_bh();
2068	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2069	iter->skip = *pos;
2070
2071	if (iter->tbl) {
2072		ipv6_route_seq_setup_walk(iter);
2073		return ipv6_route_seq_next(seq, NULL, pos);
2074	} else {
2075		return NULL;
2076	}
2077}
2078
2079static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2080{
2081	struct fib6_walker *w = &iter->w;
2082	return w->node && !(w->state == FWS_U && w->node == w->root);
2083}
2084
2085static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2086	__releases(RCU_BH)
2087{
2088	struct ipv6_route_iter *iter = seq->private;
2089
2090	if (ipv6_route_iter_active(iter))
2091		fib6_walker_unlink(&iter->w);
2092
2093	rcu_read_unlock_bh();
2094}
2095
2096static const struct seq_operations ipv6_route_seq_ops = {
2097	.start	= ipv6_route_seq_start,
2098	.next	= ipv6_route_seq_next,
2099	.stop	= ipv6_route_seq_stop,
2100	.show	= ipv6_route_seq_show
2101};
2102
2103int ipv6_route_open(struct inode *inode, struct file *file)
2104{
2105	return seq_open_net(inode, file, &ipv6_route_seq_ops,
2106			    sizeof(struct ipv6_route_iter));
2107}
2108
2109#endif /* CONFIG_PROC_FS */
2110