xref: /openbmc/linux/net/ipv6/ip6_fib.c (revision 15e47304)
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 
14 /*
15  * 	Changes:
16  * 	Yuji SEKIYA @USAGI:	Support default route on router node;
17  * 				remove ip6_null_entry from the top of
18  * 				routing table.
19  * 	Ville Nuorvala:		Fixed routing subtrees.
20  */
21 
22 #define pr_fmt(fmt) "IPv6: " fmt
23 
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/net.h>
27 #include <linux/route.h>
28 #include <linux/netdevice.h>
29 #include <linux/in6.h>
30 #include <linux/init.h>
31 #include <linux/list.h>
32 #include <linux/slab.h>
33 
34 #include <net/ipv6.h>
35 #include <net/ndisc.h>
36 #include <net/addrconf.h>
37 
38 #include <net/ip6_fib.h>
39 #include <net/ip6_route.h>
40 
41 #define RT6_DEBUG 2
42 
43 #if RT6_DEBUG >= 3
44 #define RT6_TRACE(x...) pr_debug(x)
45 #else
46 #define RT6_TRACE(x...) do { ; } while (0)
47 #endif
48 
49 static struct kmem_cache * fib6_node_kmem __read_mostly;
50 
51 enum fib_walk_state_t
52 {
53 #ifdef CONFIG_IPV6_SUBTREES
54 	FWS_S,
55 #endif
56 	FWS_L,
57 	FWS_R,
58 	FWS_C,
59 	FWS_U
60 };
61 
62 struct fib6_cleaner_t
63 {
64 	struct fib6_walker_t w;
65 	struct net *net;
66 	int (*func)(struct rt6_info *, void *arg);
67 	void *arg;
68 };
69 
70 static DEFINE_RWLOCK(fib6_walker_lock);
71 
72 #ifdef CONFIG_IPV6_SUBTREES
73 #define FWS_INIT FWS_S
74 #else
75 #define FWS_INIT FWS_L
76 #endif
77 
78 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
79 			      struct rt6_info *rt);
80 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
81 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
82 static int fib6_walk(struct fib6_walker_t *w);
83 static int fib6_walk_continue(struct fib6_walker_t *w);
84 
85 /*
86  *	A routing update causes an increase of the serial number on the
87  *	affected subtree. This allows for cached routes to be asynchronously
88  *	tested when modifications are made to the destination cache as a
89  *	result of redirects, path MTU changes, etc.
90  */
91 
92 static __u32 rt_sernum;
93 
94 static void fib6_gc_timer_cb(unsigned long arg);
95 
96 static LIST_HEAD(fib6_walkers);
97 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
98 
99 static inline void fib6_walker_link(struct fib6_walker_t *w)
100 {
101 	write_lock_bh(&fib6_walker_lock);
102 	list_add(&w->lh, &fib6_walkers);
103 	write_unlock_bh(&fib6_walker_lock);
104 }
105 
106 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
107 {
108 	write_lock_bh(&fib6_walker_lock);
109 	list_del(&w->lh);
110 	write_unlock_bh(&fib6_walker_lock);
111 }
112 static __inline__ u32 fib6_new_sernum(void)
113 {
114 	u32 n = ++rt_sernum;
115 	if ((__s32)n <= 0)
116 		rt_sernum = n = 1;
117 	return n;
118 }
119 
120 /*
121  *	Auxiliary address test functions for the radix tree.
122  *
123  *	These assume a 32bit processor (although it will work on
124  *	64bit processors)
125  */
126 
127 /*
128  *	test bit
129  */
130 #if defined(__LITTLE_ENDIAN)
131 # define BITOP_BE32_SWIZZLE	(0x1F & ~7)
132 #else
133 # define BITOP_BE32_SWIZZLE	0
134 #endif
135 
136 static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
137 {
138 	const __be32 *addr = token;
139 	/*
140 	 * Here,
141 	 * 	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
142 	 * is optimized version of
143 	 *	htonl(1 << ((~fn_bit)&0x1F))
144 	 * See include/asm-generic/bitops/le.h.
145 	 */
146 	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
147 	       addr[fn_bit >> 5];
148 }
149 
150 static __inline__ struct fib6_node * node_alloc(void)
151 {
152 	struct fib6_node *fn;
153 
154 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
155 
156 	return fn;
157 }
158 
159 static __inline__ void node_free(struct fib6_node * fn)
160 {
161 	kmem_cache_free(fib6_node_kmem, fn);
162 }
163 
164 static __inline__ void rt6_release(struct rt6_info *rt)
165 {
166 	if (atomic_dec_and_test(&rt->rt6i_ref))
167 		dst_free(&rt->dst);
168 }
169 
170 static void fib6_link_table(struct net *net, struct fib6_table *tb)
171 {
172 	unsigned int h;
173 
174 	/*
175 	 * Initialize table lock at a single place to give lockdep a key,
176 	 * tables aren't visible prior to being linked to the list.
177 	 */
178 	rwlock_init(&tb->tb6_lock);
179 
180 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
181 
182 	/*
183 	 * No protection necessary, this is the only list mutatation
184 	 * operation, tables never disappear once they exist.
185 	 */
186 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
187 }
188 
189 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
190 
191 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
192 {
193 	struct fib6_table *table;
194 
195 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
196 	if (table) {
197 		table->tb6_id = id;
198 		table->tb6_root.leaf = net->ipv6.ip6_null_entry;
199 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
200 		inet_peer_base_init(&table->tb6_peers);
201 	}
202 
203 	return table;
204 }
205 
206 struct fib6_table *fib6_new_table(struct net *net, u32 id)
207 {
208 	struct fib6_table *tb;
209 
210 	if (id == 0)
211 		id = RT6_TABLE_MAIN;
212 	tb = fib6_get_table(net, id);
213 	if (tb)
214 		return tb;
215 
216 	tb = fib6_alloc_table(net, id);
217 	if (tb)
218 		fib6_link_table(net, tb);
219 
220 	return tb;
221 }
222 
223 struct fib6_table *fib6_get_table(struct net *net, u32 id)
224 {
225 	struct fib6_table *tb;
226 	struct hlist_head *head;
227 	struct hlist_node *node;
228 	unsigned int h;
229 
230 	if (id == 0)
231 		id = RT6_TABLE_MAIN;
232 	h = id & (FIB6_TABLE_HASHSZ - 1);
233 	rcu_read_lock();
234 	head = &net->ipv6.fib_table_hash[h];
235 	hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
236 		if (tb->tb6_id == id) {
237 			rcu_read_unlock();
238 			return tb;
239 		}
240 	}
241 	rcu_read_unlock();
242 
243 	return NULL;
244 }
245 
246 static void __net_init fib6_tables_init(struct net *net)
247 {
248 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
249 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
250 }
251 #else
252 
253 struct fib6_table *fib6_new_table(struct net *net, u32 id)
254 {
255 	return fib6_get_table(net, id);
256 }
257 
258 struct fib6_table *fib6_get_table(struct net *net, u32 id)
259 {
260 	  return net->ipv6.fib6_main_tbl;
261 }
262 
263 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
264 				   int flags, pol_lookup_t lookup)
265 {
266 	return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
267 }
268 
269 static void __net_init fib6_tables_init(struct net *net)
270 {
271 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
272 }
273 
274 #endif
275 
276 static int fib6_dump_node(struct fib6_walker_t *w)
277 {
278 	int res;
279 	struct rt6_info *rt;
280 
281 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
282 		res = rt6_dump_route(rt, w->args);
283 		if (res < 0) {
284 			/* Frame is full, suspend walking */
285 			w->leaf = rt;
286 			return 1;
287 		}
288 		WARN_ON(res == 0);
289 	}
290 	w->leaf = NULL;
291 	return 0;
292 }
293 
294 static void fib6_dump_end(struct netlink_callback *cb)
295 {
296 	struct fib6_walker_t *w = (void*)cb->args[2];
297 
298 	if (w) {
299 		if (cb->args[4]) {
300 			cb->args[4] = 0;
301 			fib6_walker_unlink(w);
302 		}
303 		cb->args[2] = 0;
304 		kfree(w);
305 	}
306 	cb->done = (void*)cb->args[3];
307 	cb->args[1] = 3;
308 }
309 
310 static int fib6_dump_done(struct netlink_callback *cb)
311 {
312 	fib6_dump_end(cb);
313 	return cb->done ? cb->done(cb) : 0;
314 }
315 
316 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
317 			   struct netlink_callback *cb)
318 {
319 	struct fib6_walker_t *w;
320 	int res;
321 
322 	w = (void *)cb->args[2];
323 	w->root = &table->tb6_root;
324 
325 	if (cb->args[4] == 0) {
326 		w->count = 0;
327 		w->skip = 0;
328 
329 		read_lock_bh(&table->tb6_lock);
330 		res = fib6_walk(w);
331 		read_unlock_bh(&table->tb6_lock);
332 		if (res > 0) {
333 			cb->args[4] = 1;
334 			cb->args[5] = w->root->fn_sernum;
335 		}
336 	} else {
337 		if (cb->args[5] != w->root->fn_sernum) {
338 			/* Begin at the root if the tree changed */
339 			cb->args[5] = w->root->fn_sernum;
340 			w->state = FWS_INIT;
341 			w->node = w->root;
342 			w->skip = w->count;
343 		} else
344 			w->skip = 0;
345 
346 		read_lock_bh(&table->tb6_lock);
347 		res = fib6_walk_continue(w);
348 		read_unlock_bh(&table->tb6_lock);
349 		if (res <= 0) {
350 			fib6_walker_unlink(w);
351 			cb->args[4] = 0;
352 		}
353 	}
354 
355 	return res;
356 }
357 
358 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
359 {
360 	struct net *net = sock_net(skb->sk);
361 	unsigned int h, s_h;
362 	unsigned int e = 0, s_e;
363 	struct rt6_rtnl_dump_arg arg;
364 	struct fib6_walker_t *w;
365 	struct fib6_table *tb;
366 	struct hlist_node *node;
367 	struct hlist_head *head;
368 	int res = 0;
369 
370 	s_h = cb->args[0];
371 	s_e = cb->args[1];
372 
373 	w = (void *)cb->args[2];
374 	if (!w) {
375 		/* New dump:
376 		 *
377 		 * 1. hook callback destructor.
378 		 */
379 		cb->args[3] = (long)cb->done;
380 		cb->done = fib6_dump_done;
381 
382 		/*
383 		 * 2. allocate and initialize walker.
384 		 */
385 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
386 		if (!w)
387 			return -ENOMEM;
388 		w->func = fib6_dump_node;
389 		cb->args[2] = (long)w;
390 	}
391 
392 	arg.skb = skb;
393 	arg.cb = cb;
394 	arg.net = net;
395 	w->args = &arg;
396 
397 	rcu_read_lock();
398 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
399 		e = 0;
400 		head = &net->ipv6.fib_table_hash[h];
401 		hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
402 			if (e < s_e)
403 				goto next;
404 			res = fib6_dump_table(tb, skb, cb);
405 			if (res != 0)
406 				goto out;
407 next:
408 			e++;
409 		}
410 	}
411 out:
412 	rcu_read_unlock();
413 	cb->args[1] = e;
414 	cb->args[0] = h;
415 
416 	res = res < 0 ? res : skb->len;
417 	if (res <= 0)
418 		fib6_dump_end(cb);
419 	return res;
420 }
421 
422 /*
423  *	Routing Table
424  *
425  *	return the appropriate node for a routing tree "add" operation
426  *	by either creating and inserting or by returning an existing
427  *	node.
428  */
429 
430 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
431 				     int addrlen, int plen,
432 				     int offset, int allow_create,
433 				     int replace_required)
434 {
435 	struct fib6_node *fn, *in, *ln;
436 	struct fib6_node *pn = NULL;
437 	struct rt6key *key;
438 	int	bit;
439 	__be32	dir = 0;
440 	__u32	sernum = fib6_new_sernum();
441 
442 	RT6_TRACE("fib6_add_1\n");
443 
444 	/* insert node in tree */
445 
446 	fn = root;
447 
448 	do {
449 		key = (struct rt6key *)((u8 *)fn->leaf + offset);
450 
451 		/*
452 		 *	Prefix match
453 		 */
454 		if (plen < fn->fn_bit ||
455 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
456 			if (!allow_create) {
457 				if (replace_required) {
458 					pr_warn("Can't replace route, no match found\n");
459 					return ERR_PTR(-ENOENT);
460 				}
461 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
462 			}
463 			goto insert_above;
464 		}
465 
466 		/*
467 		 *	Exact match ?
468 		 */
469 
470 		if (plen == fn->fn_bit) {
471 			/* clean up an intermediate node */
472 			if (!(fn->fn_flags & RTN_RTINFO)) {
473 				rt6_release(fn->leaf);
474 				fn->leaf = NULL;
475 			}
476 
477 			fn->fn_sernum = sernum;
478 
479 			return fn;
480 		}
481 
482 		/*
483 		 *	We have more bits to go
484 		 */
485 
486 		/* Try to walk down on tree. */
487 		fn->fn_sernum = sernum;
488 		dir = addr_bit_set(addr, fn->fn_bit);
489 		pn = fn;
490 		fn = dir ? fn->right: fn->left;
491 	} while (fn);
492 
493 	if (!allow_create) {
494 		/* We should not create new node because
495 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
496 		 * I assume it is safe to require NLM_F_CREATE when
497 		 * REPLACE flag is used! Later we may want to remove the
498 		 * check for replace_required, because according
499 		 * to netlink specification, NLM_F_CREATE
500 		 * MUST be specified if new route is created.
501 		 * That would keep IPv6 consistent with IPv4
502 		 */
503 		if (replace_required) {
504 			pr_warn("Can't replace route, no match found\n");
505 			return ERR_PTR(-ENOENT);
506 		}
507 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
508 	}
509 	/*
510 	 *	We walked to the bottom of tree.
511 	 *	Create new leaf node without children.
512 	 */
513 
514 	ln = node_alloc();
515 
516 	if (!ln)
517 		return NULL;
518 	ln->fn_bit = plen;
519 
520 	ln->parent = pn;
521 	ln->fn_sernum = sernum;
522 
523 	if (dir)
524 		pn->right = ln;
525 	else
526 		pn->left  = ln;
527 
528 	return ln;
529 
530 
531 insert_above:
532 	/*
533 	 * split since we don't have a common prefix anymore or
534 	 * we have a less significant route.
535 	 * we've to insert an intermediate node on the list
536 	 * this new node will point to the one we need to create
537 	 * and the current
538 	 */
539 
540 	pn = fn->parent;
541 
542 	/* find 1st bit in difference between the 2 addrs.
543 
544 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
545 	   but if it is >= plen, the value is ignored in any case.
546 	 */
547 
548 	bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
549 
550 	/*
551 	 *		(intermediate)[in]
552 	 *	          /	   \
553 	 *	(new leaf node)[ln] (old node)[fn]
554 	 */
555 	if (plen > bit) {
556 		in = node_alloc();
557 		ln = node_alloc();
558 
559 		if (!in || !ln) {
560 			if (in)
561 				node_free(in);
562 			if (ln)
563 				node_free(ln);
564 			return NULL;
565 		}
566 
567 		/*
568 		 * new intermediate node.
569 		 * RTN_RTINFO will
570 		 * be off since that an address that chooses one of
571 		 * the branches would not match less specific routes
572 		 * in the other branch
573 		 */
574 
575 		in->fn_bit = bit;
576 
577 		in->parent = pn;
578 		in->leaf = fn->leaf;
579 		atomic_inc(&in->leaf->rt6i_ref);
580 
581 		in->fn_sernum = sernum;
582 
583 		/* update parent pointer */
584 		if (dir)
585 			pn->right = in;
586 		else
587 			pn->left  = in;
588 
589 		ln->fn_bit = plen;
590 
591 		ln->parent = in;
592 		fn->parent = in;
593 
594 		ln->fn_sernum = sernum;
595 
596 		if (addr_bit_set(addr, bit)) {
597 			in->right = ln;
598 			in->left  = fn;
599 		} else {
600 			in->left  = ln;
601 			in->right = fn;
602 		}
603 	} else { /* plen <= bit */
604 
605 		/*
606 		 *		(new leaf node)[ln]
607 		 *	          /	   \
608 		 *	     (old node)[fn] NULL
609 		 */
610 
611 		ln = node_alloc();
612 
613 		if (!ln)
614 			return NULL;
615 
616 		ln->fn_bit = plen;
617 
618 		ln->parent = pn;
619 
620 		ln->fn_sernum = sernum;
621 
622 		if (dir)
623 			pn->right = ln;
624 		else
625 			pn->left  = ln;
626 
627 		if (addr_bit_set(&key->addr, plen))
628 			ln->right = fn;
629 		else
630 			ln->left  = fn;
631 
632 		fn->parent = ln;
633 	}
634 	return ln;
635 }
636 
637 /*
638  *	Insert routing information in a node.
639  */
640 
641 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
642 			    struct nl_info *info)
643 {
644 	struct rt6_info *iter = NULL;
645 	struct rt6_info **ins;
646 	int replace = (info->nlh &&
647 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
648 	int add = (!info->nlh ||
649 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
650 	int found = 0;
651 
652 	ins = &fn->leaf;
653 
654 	for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
655 		/*
656 		 *	Search for duplicates
657 		 */
658 
659 		if (iter->rt6i_metric == rt->rt6i_metric) {
660 			/*
661 			 *	Same priority level
662 			 */
663 			if (info->nlh &&
664 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
665 				return -EEXIST;
666 			if (replace) {
667 				found++;
668 				break;
669 			}
670 
671 			if (iter->dst.dev == rt->dst.dev &&
672 			    iter->rt6i_idev == rt->rt6i_idev &&
673 			    ipv6_addr_equal(&iter->rt6i_gateway,
674 					    &rt->rt6i_gateway)) {
675 				if (!(iter->rt6i_flags & RTF_EXPIRES))
676 					return -EEXIST;
677 				if (!(rt->rt6i_flags & RTF_EXPIRES))
678 					rt6_clean_expires(iter);
679 				else
680 					rt6_set_expires(iter, rt->dst.expires);
681 				return -EEXIST;
682 			}
683 		}
684 
685 		if (iter->rt6i_metric > rt->rt6i_metric)
686 			break;
687 
688 		ins = &iter->dst.rt6_next;
689 	}
690 
691 	/* Reset round-robin state, if necessary */
692 	if (ins == &fn->leaf)
693 		fn->rr_ptr = NULL;
694 
695 	/*
696 	 *	insert node
697 	 */
698 	if (!replace) {
699 		if (!add)
700 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
701 
702 add:
703 		rt->dst.rt6_next = iter;
704 		*ins = rt;
705 		rt->rt6i_node = fn;
706 		atomic_inc(&rt->rt6i_ref);
707 		inet6_rt_notify(RTM_NEWROUTE, rt, info);
708 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
709 
710 		if (!(fn->fn_flags & RTN_RTINFO)) {
711 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
712 			fn->fn_flags |= RTN_RTINFO;
713 		}
714 
715 	} else {
716 		if (!found) {
717 			if (add)
718 				goto add;
719 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
720 			return -ENOENT;
721 		}
722 		*ins = rt;
723 		rt->rt6i_node = fn;
724 		rt->dst.rt6_next = iter->dst.rt6_next;
725 		atomic_inc(&rt->rt6i_ref);
726 		inet6_rt_notify(RTM_NEWROUTE, rt, info);
727 		rt6_release(iter);
728 		if (!(fn->fn_flags & RTN_RTINFO)) {
729 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
730 			fn->fn_flags |= RTN_RTINFO;
731 		}
732 	}
733 
734 	return 0;
735 }
736 
737 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
738 {
739 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
740 	    (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
741 		mod_timer(&net->ipv6.ip6_fib_timer,
742 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
743 }
744 
745 void fib6_force_start_gc(struct net *net)
746 {
747 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
748 		mod_timer(&net->ipv6.ip6_fib_timer,
749 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
750 }
751 
752 /*
753  *	Add routing information to the routing tree.
754  *	<destination addr>/<source addr>
755  *	with source addr info in sub-trees
756  */
757 
758 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
759 {
760 	struct fib6_node *fn, *pn = NULL;
761 	int err = -ENOMEM;
762 	int allow_create = 1;
763 	int replace_required = 0;
764 
765 	if (info->nlh) {
766 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
767 			allow_create = 0;
768 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
769 			replace_required = 1;
770 	}
771 	if (!allow_create && !replace_required)
772 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
773 
774 	fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
775 			rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst),
776 			allow_create, replace_required);
777 
778 	if (IS_ERR(fn)) {
779 		err = PTR_ERR(fn);
780 		fn = NULL;
781 	}
782 
783 	if (!fn)
784 		goto out;
785 
786 	pn = fn;
787 
788 #ifdef CONFIG_IPV6_SUBTREES
789 	if (rt->rt6i_src.plen) {
790 		struct fib6_node *sn;
791 
792 		if (!fn->subtree) {
793 			struct fib6_node *sfn;
794 
795 			/*
796 			 * Create subtree.
797 			 *
798 			 *		fn[main tree]
799 			 *		|
800 			 *		sfn[subtree root]
801 			 *		   \
802 			 *		    sn[new leaf node]
803 			 */
804 
805 			/* Create subtree root node */
806 			sfn = node_alloc();
807 			if (!sfn)
808 				goto st_failure;
809 
810 			sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
811 			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
812 			sfn->fn_flags = RTN_ROOT;
813 			sfn->fn_sernum = fib6_new_sernum();
814 
815 			/* Now add the first leaf node to new subtree */
816 
817 			sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
818 					sizeof(struct in6_addr), rt->rt6i_src.plen,
819 					offsetof(struct rt6_info, rt6i_src),
820 					allow_create, replace_required);
821 
822 			if (!sn) {
823 				/* If it is failed, discard just allocated
824 				   root, and then (in st_failure) stale node
825 				   in main tree.
826 				 */
827 				node_free(sfn);
828 				goto st_failure;
829 			}
830 
831 			/* Now link new subtree to main tree */
832 			sfn->parent = fn;
833 			fn->subtree = sfn;
834 		} else {
835 			sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
836 					sizeof(struct in6_addr), rt->rt6i_src.plen,
837 					offsetof(struct rt6_info, rt6i_src),
838 					allow_create, replace_required);
839 
840 			if (IS_ERR(sn)) {
841 				err = PTR_ERR(sn);
842 				sn = NULL;
843 			}
844 			if (!sn)
845 				goto st_failure;
846 		}
847 
848 		if (!fn->leaf) {
849 			fn->leaf = rt;
850 			atomic_inc(&rt->rt6i_ref);
851 		}
852 		fn = sn;
853 	}
854 #endif
855 
856 	err = fib6_add_rt2node(fn, rt, info);
857 	if (!err) {
858 		fib6_start_gc(info->nl_net, rt);
859 		if (!(rt->rt6i_flags & RTF_CACHE))
860 			fib6_prune_clones(info->nl_net, pn, rt);
861 	}
862 
863 out:
864 	if (err) {
865 #ifdef CONFIG_IPV6_SUBTREES
866 		/*
867 		 * If fib6_add_1 has cleared the old leaf pointer in the
868 		 * super-tree leaf node we have to find a new one for it.
869 		 */
870 		if (pn != fn && pn->leaf == rt) {
871 			pn->leaf = NULL;
872 			atomic_dec(&rt->rt6i_ref);
873 		}
874 		if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
875 			pn->leaf = fib6_find_prefix(info->nl_net, pn);
876 #if RT6_DEBUG >= 2
877 			if (!pn->leaf) {
878 				WARN_ON(pn->leaf == NULL);
879 				pn->leaf = info->nl_net->ipv6.ip6_null_entry;
880 			}
881 #endif
882 			atomic_inc(&pn->leaf->rt6i_ref);
883 		}
884 #endif
885 		dst_free(&rt->dst);
886 	}
887 	return err;
888 
889 #ifdef CONFIG_IPV6_SUBTREES
890 	/* Subtree creation failed, probably main tree node
891 	   is orphan. If it is, shoot it.
892 	 */
893 st_failure:
894 	if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
895 		fib6_repair_tree(info->nl_net, fn);
896 	dst_free(&rt->dst);
897 	return err;
898 #endif
899 }
900 
901 /*
902  *	Routing tree lookup
903  *
904  */
905 
906 struct lookup_args {
907 	int			offset;		/* key offset on rt6_info	*/
908 	const struct in6_addr	*addr;		/* search key			*/
909 };
910 
911 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
912 					struct lookup_args *args)
913 {
914 	struct fib6_node *fn;
915 	__be32 dir;
916 
917 	if (unlikely(args->offset == 0))
918 		return NULL;
919 
920 	/*
921 	 *	Descend on a tree
922 	 */
923 
924 	fn = root;
925 
926 	for (;;) {
927 		struct fib6_node *next;
928 
929 		dir = addr_bit_set(args->addr, fn->fn_bit);
930 
931 		next = dir ? fn->right : fn->left;
932 
933 		if (next) {
934 			fn = next;
935 			continue;
936 		}
937 		break;
938 	}
939 
940 	while (fn) {
941 		if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
942 			struct rt6key *key;
943 
944 			key = (struct rt6key *) ((u8 *) fn->leaf +
945 						 args->offset);
946 
947 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
948 #ifdef CONFIG_IPV6_SUBTREES
949 				if (fn->subtree)
950 					fn = fib6_lookup_1(fn->subtree, args + 1);
951 #endif
952 				if (!fn || fn->fn_flags & RTN_RTINFO)
953 					return fn;
954 			}
955 		}
956 
957 		if (fn->fn_flags & RTN_ROOT)
958 			break;
959 
960 		fn = fn->parent;
961 	}
962 
963 	return NULL;
964 }
965 
966 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
967 			       const struct in6_addr *saddr)
968 {
969 	struct fib6_node *fn;
970 	struct lookup_args args[] = {
971 		{
972 			.offset = offsetof(struct rt6_info, rt6i_dst),
973 			.addr = daddr,
974 		},
975 #ifdef CONFIG_IPV6_SUBTREES
976 		{
977 			.offset = offsetof(struct rt6_info, rt6i_src),
978 			.addr = saddr,
979 		},
980 #endif
981 		{
982 			.offset = 0,	/* sentinel */
983 		}
984 	};
985 
986 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
987 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
988 		fn = root;
989 
990 	return fn;
991 }
992 
993 /*
994  *	Get node with specified destination prefix (and source prefix,
995  *	if subtrees are used)
996  */
997 
998 
999 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
1000 					const struct in6_addr *addr,
1001 					int plen, int offset)
1002 {
1003 	struct fib6_node *fn;
1004 
1005 	for (fn = root; fn ; ) {
1006 		struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1007 
1008 		/*
1009 		 *	Prefix match
1010 		 */
1011 		if (plen < fn->fn_bit ||
1012 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1013 			return NULL;
1014 
1015 		if (plen == fn->fn_bit)
1016 			return fn;
1017 
1018 		/*
1019 		 *	We have more bits to go
1020 		 */
1021 		if (addr_bit_set(addr, fn->fn_bit))
1022 			fn = fn->right;
1023 		else
1024 			fn = fn->left;
1025 	}
1026 	return NULL;
1027 }
1028 
1029 struct fib6_node * fib6_locate(struct fib6_node *root,
1030 			       const struct in6_addr *daddr, int dst_len,
1031 			       const struct in6_addr *saddr, int src_len)
1032 {
1033 	struct fib6_node *fn;
1034 
1035 	fn = fib6_locate_1(root, daddr, dst_len,
1036 			   offsetof(struct rt6_info, rt6i_dst));
1037 
1038 #ifdef CONFIG_IPV6_SUBTREES
1039 	if (src_len) {
1040 		WARN_ON(saddr == NULL);
1041 		if (fn && fn->subtree)
1042 			fn = fib6_locate_1(fn->subtree, saddr, src_len,
1043 					   offsetof(struct rt6_info, rt6i_src));
1044 	}
1045 #endif
1046 
1047 	if (fn && fn->fn_flags & RTN_RTINFO)
1048 		return fn;
1049 
1050 	return NULL;
1051 }
1052 
1053 
1054 /*
1055  *	Deletion
1056  *
1057  */
1058 
1059 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1060 {
1061 	if (fn->fn_flags & RTN_ROOT)
1062 		return net->ipv6.ip6_null_entry;
1063 
1064 	while (fn) {
1065 		if (fn->left)
1066 			return fn->left->leaf;
1067 		if (fn->right)
1068 			return fn->right->leaf;
1069 
1070 		fn = FIB6_SUBTREE(fn);
1071 	}
1072 	return NULL;
1073 }
1074 
1075 /*
1076  *	Called to trim the tree of intermediate nodes when possible. "fn"
1077  *	is the node we want to try and remove.
1078  */
1079 
1080 static struct fib6_node *fib6_repair_tree(struct net *net,
1081 					   struct fib6_node *fn)
1082 {
1083 	int children;
1084 	int nstate;
1085 	struct fib6_node *child, *pn;
1086 	struct fib6_walker_t *w;
1087 	int iter = 0;
1088 
1089 	for (;;) {
1090 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1091 		iter++;
1092 
1093 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1094 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1095 		WARN_ON(fn->leaf != NULL);
1096 
1097 		children = 0;
1098 		child = NULL;
1099 		if (fn->right) child = fn->right, children |= 1;
1100 		if (fn->left) child = fn->left, children |= 2;
1101 
1102 		if (children == 3 || FIB6_SUBTREE(fn)
1103 #ifdef CONFIG_IPV6_SUBTREES
1104 		    /* Subtree root (i.e. fn) may have one child */
1105 		    || (children && fn->fn_flags & RTN_ROOT)
1106 #endif
1107 		    ) {
1108 			fn->leaf = fib6_find_prefix(net, fn);
1109 #if RT6_DEBUG >= 2
1110 			if (!fn->leaf) {
1111 				WARN_ON(!fn->leaf);
1112 				fn->leaf = net->ipv6.ip6_null_entry;
1113 			}
1114 #endif
1115 			atomic_inc(&fn->leaf->rt6i_ref);
1116 			return fn->parent;
1117 		}
1118 
1119 		pn = fn->parent;
1120 #ifdef CONFIG_IPV6_SUBTREES
1121 		if (FIB6_SUBTREE(pn) == fn) {
1122 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1123 			FIB6_SUBTREE(pn) = NULL;
1124 			nstate = FWS_L;
1125 		} else {
1126 			WARN_ON(fn->fn_flags & RTN_ROOT);
1127 #endif
1128 			if (pn->right == fn) pn->right = child;
1129 			else if (pn->left == fn) pn->left = child;
1130 #if RT6_DEBUG >= 2
1131 			else
1132 				WARN_ON(1);
1133 #endif
1134 			if (child)
1135 				child->parent = pn;
1136 			nstate = FWS_R;
1137 #ifdef CONFIG_IPV6_SUBTREES
1138 		}
1139 #endif
1140 
1141 		read_lock(&fib6_walker_lock);
1142 		FOR_WALKERS(w) {
1143 			if (!child) {
1144 				if (w->root == fn) {
1145 					w->root = w->node = NULL;
1146 					RT6_TRACE("W %p adjusted by delroot 1\n", w);
1147 				} else if (w->node == fn) {
1148 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1149 					w->node = pn;
1150 					w->state = nstate;
1151 				}
1152 			} else {
1153 				if (w->root == fn) {
1154 					w->root = child;
1155 					RT6_TRACE("W %p adjusted by delroot 2\n", w);
1156 				}
1157 				if (w->node == fn) {
1158 					w->node = child;
1159 					if (children&2) {
1160 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1161 						w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1162 					} else {
1163 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1164 						w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1165 					}
1166 				}
1167 			}
1168 		}
1169 		read_unlock(&fib6_walker_lock);
1170 
1171 		node_free(fn);
1172 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1173 			return pn;
1174 
1175 		rt6_release(pn->leaf);
1176 		pn->leaf = NULL;
1177 		fn = pn;
1178 	}
1179 }
1180 
1181 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1182 			   struct nl_info *info)
1183 {
1184 	struct fib6_walker_t *w;
1185 	struct rt6_info *rt = *rtp;
1186 	struct net *net = info->nl_net;
1187 
1188 	RT6_TRACE("fib6_del_route\n");
1189 
1190 	/* Unlink it */
1191 	*rtp = rt->dst.rt6_next;
1192 	rt->rt6i_node = NULL;
1193 	net->ipv6.rt6_stats->fib_rt_entries--;
1194 	net->ipv6.rt6_stats->fib_discarded_routes++;
1195 
1196 	/* Reset round-robin state, if necessary */
1197 	if (fn->rr_ptr == rt)
1198 		fn->rr_ptr = NULL;
1199 
1200 	/* Adjust walkers */
1201 	read_lock(&fib6_walker_lock);
1202 	FOR_WALKERS(w) {
1203 		if (w->state == FWS_C && w->leaf == rt) {
1204 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1205 			w->leaf = rt->dst.rt6_next;
1206 			if (!w->leaf)
1207 				w->state = FWS_U;
1208 		}
1209 	}
1210 	read_unlock(&fib6_walker_lock);
1211 
1212 	rt->dst.rt6_next = NULL;
1213 
1214 	/* If it was last route, expunge its radix tree node */
1215 	if (!fn->leaf) {
1216 		fn->fn_flags &= ~RTN_RTINFO;
1217 		net->ipv6.rt6_stats->fib_route_nodes--;
1218 		fn = fib6_repair_tree(net, fn);
1219 	}
1220 
1221 	if (atomic_read(&rt->rt6i_ref) != 1) {
1222 		/* This route is used as dummy address holder in some split
1223 		 * nodes. It is not leaked, but it still holds other resources,
1224 		 * which must be released in time. So, scan ascendant nodes
1225 		 * and replace dummy references to this route with references
1226 		 * to still alive ones.
1227 		 */
1228 		while (fn) {
1229 			if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
1230 				fn->leaf = fib6_find_prefix(net, fn);
1231 				atomic_inc(&fn->leaf->rt6i_ref);
1232 				rt6_release(rt);
1233 			}
1234 			fn = fn->parent;
1235 		}
1236 		/* No more references are possible at this point. */
1237 		BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1238 	}
1239 
1240 	inet6_rt_notify(RTM_DELROUTE, rt, info);
1241 	rt6_release(rt);
1242 }
1243 
1244 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1245 {
1246 	struct net *net = info->nl_net;
1247 	struct fib6_node *fn = rt->rt6i_node;
1248 	struct rt6_info **rtp;
1249 
1250 #if RT6_DEBUG >= 2
1251 	if (rt->dst.obsolete>0) {
1252 		WARN_ON(fn != NULL);
1253 		return -ENOENT;
1254 	}
1255 #endif
1256 	if (!fn || rt == net->ipv6.ip6_null_entry)
1257 		return -ENOENT;
1258 
1259 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1260 
1261 	if (!(rt->rt6i_flags & RTF_CACHE)) {
1262 		struct fib6_node *pn = fn;
1263 #ifdef CONFIG_IPV6_SUBTREES
1264 		/* clones of this route might be in another subtree */
1265 		if (rt->rt6i_src.plen) {
1266 			while (!(pn->fn_flags & RTN_ROOT))
1267 				pn = pn->parent;
1268 			pn = pn->parent;
1269 		}
1270 #endif
1271 		fib6_prune_clones(info->nl_net, pn, rt);
1272 	}
1273 
1274 	/*
1275 	 *	Walk the leaf entries looking for ourself
1276 	 */
1277 
1278 	for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1279 		if (*rtp == rt) {
1280 			fib6_del_route(fn, rtp, info);
1281 			return 0;
1282 		}
1283 	}
1284 	return -ENOENT;
1285 }
1286 
1287 /*
1288  *	Tree traversal function.
1289  *
1290  *	Certainly, it is not interrupt safe.
1291  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1292  *	It means, that we can modify tree during walking
1293  *	and use this function for garbage collection, clone pruning,
1294  *	cleaning tree when a device goes down etc. etc.
1295  *
1296  *	It guarantees that every node will be traversed,
1297  *	and that it will be traversed only once.
1298  *
1299  *	Callback function w->func may return:
1300  *	0 -> continue walking.
1301  *	positive value -> walking is suspended (used by tree dumps,
1302  *	and probably by gc, if it will be split to several slices)
1303  *	negative value -> terminate walking.
1304  *
1305  *	The function itself returns:
1306  *	0   -> walk is complete.
1307  *	>0  -> walk is incomplete (i.e. suspended)
1308  *	<0  -> walk is terminated by an error.
1309  */
1310 
1311 static int fib6_walk_continue(struct fib6_walker_t *w)
1312 {
1313 	struct fib6_node *fn, *pn;
1314 
1315 	for (;;) {
1316 		fn = w->node;
1317 		if (!fn)
1318 			return 0;
1319 
1320 		if (w->prune && fn != w->root &&
1321 		    fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1322 			w->state = FWS_C;
1323 			w->leaf = fn->leaf;
1324 		}
1325 		switch (w->state) {
1326 #ifdef CONFIG_IPV6_SUBTREES
1327 		case FWS_S:
1328 			if (FIB6_SUBTREE(fn)) {
1329 				w->node = FIB6_SUBTREE(fn);
1330 				continue;
1331 			}
1332 			w->state = FWS_L;
1333 #endif
1334 		case FWS_L:
1335 			if (fn->left) {
1336 				w->node = fn->left;
1337 				w->state = FWS_INIT;
1338 				continue;
1339 			}
1340 			w->state = FWS_R;
1341 		case FWS_R:
1342 			if (fn->right) {
1343 				w->node = fn->right;
1344 				w->state = FWS_INIT;
1345 				continue;
1346 			}
1347 			w->state = FWS_C;
1348 			w->leaf = fn->leaf;
1349 		case FWS_C:
1350 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1351 				int err;
1352 
1353 				if (w->skip) {
1354 					w->skip--;
1355 					continue;
1356 				}
1357 
1358 				err = w->func(w);
1359 				if (err)
1360 					return err;
1361 
1362 				w->count++;
1363 				continue;
1364 			}
1365 			w->state = FWS_U;
1366 		case FWS_U:
1367 			if (fn == w->root)
1368 				return 0;
1369 			pn = fn->parent;
1370 			w->node = pn;
1371 #ifdef CONFIG_IPV6_SUBTREES
1372 			if (FIB6_SUBTREE(pn) == fn) {
1373 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1374 				w->state = FWS_L;
1375 				continue;
1376 			}
1377 #endif
1378 			if (pn->left == fn) {
1379 				w->state = FWS_R;
1380 				continue;
1381 			}
1382 			if (pn->right == fn) {
1383 				w->state = FWS_C;
1384 				w->leaf = w->node->leaf;
1385 				continue;
1386 			}
1387 #if RT6_DEBUG >= 2
1388 			WARN_ON(1);
1389 #endif
1390 		}
1391 	}
1392 }
1393 
1394 static int fib6_walk(struct fib6_walker_t *w)
1395 {
1396 	int res;
1397 
1398 	w->state = FWS_INIT;
1399 	w->node = w->root;
1400 
1401 	fib6_walker_link(w);
1402 	res = fib6_walk_continue(w);
1403 	if (res <= 0)
1404 		fib6_walker_unlink(w);
1405 	return res;
1406 }
1407 
1408 static int fib6_clean_node(struct fib6_walker_t *w)
1409 {
1410 	int res;
1411 	struct rt6_info *rt;
1412 	struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1413 	struct nl_info info = {
1414 		.nl_net = c->net,
1415 	};
1416 
1417 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1418 		res = c->func(rt, c->arg);
1419 		if (res < 0) {
1420 			w->leaf = rt;
1421 			res = fib6_del(rt, &info);
1422 			if (res) {
1423 #if RT6_DEBUG >= 2
1424 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1425 					 __func__, rt, rt->rt6i_node, res);
1426 #endif
1427 				continue;
1428 			}
1429 			return 0;
1430 		}
1431 		WARN_ON(res != 0);
1432 	}
1433 	w->leaf = rt;
1434 	return 0;
1435 }
1436 
1437 /*
1438  *	Convenient frontend to tree walker.
1439  *
1440  *	func is called on each route.
1441  *		It may return -1 -> delete this route.
1442  *		              0  -> continue walking
1443  *
1444  *	prune==1 -> only immediate children of node (certainly,
1445  *	ignoring pure split nodes) will be scanned.
1446  */
1447 
1448 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1449 			    int (*func)(struct rt6_info *, void *arg),
1450 			    int prune, void *arg)
1451 {
1452 	struct fib6_cleaner_t c;
1453 
1454 	c.w.root = root;
1455 	c.w.func = fib6_clean_node;
1456 	c.w.prune = prune;
1457 	c.w.count = 0;
1458 	c.w.skip = 0;
1459 	c.func = func;
1460 	c.arg = arg;
1461 	c.net = net;
1462 
1463 	fib6_walk(&c.w);
1464 }
1465 
1466 void fib6_clean_all_ro(struct net *net, int (*func)(struct rt6_info *, void *arg),
1467 		    int prune, void *arg)
1468 {
1469 	struct fib6_table *table;
1470 	struct hlist_node *node;
1471 	struct hlist_head *head;
1472 	unsigned int h;
1473 
1474 	rcu_read_lock();
1475 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1476 		head = &net->ipv6.fib_table_hash[h];
1477 		hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1478 			read_lock_bh(&table->tb6_lock);
1479 			fib6_clean_tree(net, &table->tb6_root,
1480 					func, prune, arg);
1481 			read_unlock_bh(&table->tb6_lock);
1482 		}
1483 	}
1484 	rcu_read_unlock();
1485 }
1486 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1487 		    int prune, void *arg)
1488 {
1489 	struct fib6_table *table;
1490 	struct hlist_node *node;
1491 	struct hlist_head *head;
1492 	unsigned int h;
1493 
1494 	rcu_read_lock();
1495 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1496 		head = &net->ipv6.fib_table_hash[h];
1497 		hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1498 			write_lock_bh(&table->tb6_lock);
1499 			fib6_clean_tree(net, &table->tb6_root,
1500 					func, prune, arg);
1501 			write_unlock_bh(&table->tb6_lock);
1502 		}
1503 	}
1504 	rcu_read_unlock();
1505 }
1506 
1507 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1508 {
1509 	if (rt->rt6i_flags & RTF_CACHE) {
1510 		RT6_TRACE("pruning clone %p\n", rt);
1511 		return -1;
1512 	}
1513 
1514 	return 0;
1515 }
1516 
1517 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1518 			      struct rt6_info *rt)
1519 {
1520 	fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1521 }
1522 
1523 /*
1524  *	Garbage collection
1525  */
1526 
1527 static struct fib6_gc_args
1528 {
1529 	int			timeout;
1530 	int			more;
1531 } gc_args;
1532 
1533 static int fib6_age(struct rt6_info *rt, void *arg)
1534 {
1535 	unsigned long now = jiffies;
1536 
1537 	/*
1538 	 *	check addrconf expiration here.
1539 	 *	Routes are expired even if they are in use.
1540 	 *
1541 	 *	Also age clones. Note, that clones are aged out
1542 	 *	only if they are not in use now.
1543 	 */
1544 
1545 	if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1546 		if (time_after(now, rt->dst.expires)) {
1547 			RT6_TRACE("expiring %p\n", rt);
1548 			return -1;
1549 		}
1550 		gc_args.more++;
1551 	} else if (rt->rt6i_flags & RTF_CACHE) {
1552 		if (atomic_read(&rt->dst.__refcnt) == 0 &&
1553 		    time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1554 			RT6_TRACE("aging clone %p\n", rt);
1555 			return -1;
1556 		} else if (rt->rt6i_flags & RTF_GATEWAY) {
1557 			struct neighbour *neigh;
1558 			__u8 neigh_flags = 0;
1559 
1560 			neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1561 			if (neigh) {
1562 				neigh_flags = neigh->flags;
1563 				neigh_release(neigh);
1564 			}
1565 			if (!(neigh_flags & NTF_ROUTER)) {
1566 				RT6_TRACE("purging route %p via non-router but gateway\n",
1567 					  rt);
1568 				return -1;
1569 			}
1570 		}
1571 		gc_args.more++;
1572 	}
1573 
1574 	return 0;
1575 }
1576 
1577 static DEFINE_SPINLOCK(fib6_gc_lock);
1578 
1579 void fib6_run_gc(unsigned long expires, struct net *net)
1580 {
1581 	if (expires != ~0UL) {
1582 		spin_lock_bh(&fib6_gc_lock);
1583 		gc_args.timeout = expires ? (int)expires :
1584 			net->ipv6.sysctl.ip6_rt_gc_interval;
1585 	} else {
1586 		if (!spin_trylock_bh(&fib6_gc_lock)) {
1587 			mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1588 			return;
1589 		}
1590 		gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1591 	}
1592 
1593 	gc_args.more = icmp6_dst_gc();
1594 
1595 	fib6_clean_all(net, fib6_age, 0, NULL);
1596 
1597 	if (gc_args.more)
1598 		mod_timer(&net->ipv6.ip6_fib_timer,
1599 			  round_jiffies(jiffies
1600 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
1601 	else
1602 		del_timer(&net->ipv6.ip6_fib_timer);
1603 	spin_unlock_bh(&fib6_gc_lock);
1604 }
1605 
1606 static void fib6_gc_timer_cb(unsigned long arg)
1607 {
1608 	fib6_run_gc(0, (struct net *)arg);
1609 }
1610 
1611 static int __net_init fib6_net_init(struct net *net)
1612 {
1613 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1614 
1615 	setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1616 
1617 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1618 	if (!net->ipv6.rt6_stats)
1619 		goto out_timer;
1620 
1621 	/* Avoid false sharing : Use at least a full cache line */
1622 	size = max_t(size_t, size, L1_CACHE_BYTES);
1623 
1624 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1625 	if (!net->ipv6.fib_table_hash)
1626 		goto out_rt6_stats;
1627 
1628 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1629 					  GFP_KERNEL);
1630 	if (!net->ipv6.fib6_main_tbl)
1631 		goto out_fib_table_hash;
1632 
1633 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1634 	net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1635 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1636 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1637 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1638 
1639 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1640 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1641 					   GFP_KERNEL);
1642 	if (!net->ipv6.fib6_local_tbl)
1643 		goto out_fib6_main_tbl;
1644 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1645 	net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1646 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1647 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1648 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1649 #endif
1650 	fib6_tables_init(net);
1651 
1652 	return 0;
1653 
1654 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1655 out_fib6_main_tbl:
1656 	kfree(net->ipv6.fib6_main_tbl);
1657 #endif
1658 out_fib_table_hash:
1659 	kfree(net->ipv6.fib_table_hash);
1660 out_rt6_stats:
1661 	kfree(net->ipv6.rt6_stats);
1662 out_timer:
1663 	return -ENOMEM;
1664  }
1665 
1666 static void fib6_net_exit(struct net *net)
1667 {
1668 	rt6_ifdown(net, NULL);
1669 	del_timer_sync(&net->ipv6.ip6_fib_timer);
1670 
1671 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1672 	inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1673 	kfree(net->ipv6.fib6_local_tbl);
1674 #endif
1675 	inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1676 	kfree(net->ipv6.fib6_main_tbl);
1677 	kfree(net->ipv6.fib_table_hash);
1678 	kfree(net->ipv6.rt6_stats);
1679 }
1680 
1681 static struct pernet_operations fib6_net_ops = {
1682 	.init = fib6_net_init,
1683 	.exit = fib6_net_exit,
1684 };
1685 
1686 int __init fib6_init(void)
1687 {
1688 	int ret = -ENOMEM;
1689 
1690 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
1691 					   sizeof(struct fib6_node),
1692 					   0, SLAB_HWCACHE_ALIGN,
1693 					   NULL);
1694 	if (!fib6_node_kmem)
1695 		goto out;
1696 
1697 	ret = register_pernet_subsys(&fib6_net_ops);
1698 	if (ret)
1699 		goto out_kmem_cache_create;
1700 
1701 	ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1702 			      NULL);
1703 	if (ret)
1704 		goto out_unregister_subsys;
1705 out:
1706 	return ret;
1707 
1708 out_unregister_subsys:
1709 	unregister_pernet_subsys(&fib6_net_ops);
1710 out_kmem_cache_create:
1711 	kmem_cache_destroy(fib6_node_kmem);
1712 	goto out;
1713 }
1714 
1715 void fib6_gc_cleanup(void)
1716 {
1717 	unregister_pernet_subsys(&fib6_net_ops);
1718 	kmem_cache_destroy(fib6_node_kmem);
1719 }
1720