xref: /linux/net/ipv6/ip6_fib.c (revision a33f32244d8550da8b4a26e277ce07d5c6d158b5)
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 #include <linux/errno.h>
22 #include <linux/types.h>
23 #include <linux/net.h>
24 #include <linux/route.h>
25 #include <linux/netdevice.h>
26 #include <linux/in6.h>
27 #include <linux/init.h>
28 #include <linux/list.h>
29 #include <linux/slab.h>
30 
31 #ifdef 	CONFIG_PROC_FS
32 #include <linux/proc_fs.h>
33 #endif
34 
35 #include <net/ipv6.h>
36 #include <net/ndisc.h>
37 #include <net/addrconf.h>
38 
39 #include <net/ip6_fib.h>
40 #include <net/ip6_route.h>
41 
42 #define RT6_DEBUG 2
43 
44 #if RT6_DEBUG >= 3
45 #define RT6_TRACE(x...) printk(KERN_DEBUG x)
46 #else
47 #define RT6_TRACE(x...) do { ; } while (0)
48 #endif
49 
50 static struct kmem_cache * fib6_node_kmem __read_mostly;
51 
52 enum fib_walk_state_t
53 {
54 #ifdef CONFIG_IPV6_SUBTREES
55 	FWS_S,
56 #endif
57 	FWS_L,
58 	FWS_R,
59 	FWS_C,
60 	FWS_U
61 };
62 
63 struct fib6_cleaner_t
64 {
65 	struct fib6_walker_t w;
66 	struct net *net;
67 	int (*func)(struct rt6_info *, void *arg);
68 	void *arg;
69 };
70 
71 static DEFINE_RWLOCK(fib6_walker_lock);
72 
73 #ifdef CONFIG_IPV6_SUBTREES
74 #define FWS_INIT FWS_S
75 #else
76 #define FWS_INIT FWS_L
77 #endif
78 
79 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
80 			      struct rt6_info *rt);
81 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
82 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
83 static int fib6_walk(struct fib6_walker_t *w);
84 static int fib6_walk_continue(struct fib6_walker_t *w);
85 
86 /*
87  *	A routing update causes an increase of the serial number on the
88  *	affected subtree. This allows for cached routes to be asynchronously
89  *	tested when modifications are made to the destination cache as a
90  *	result of redirects, path MTU changes, etc.
91  */
92 
93 static __u32 rt_sernum;
94 
95 static void fib6_gc_timer_cb(unsigned long arg);
96 
97 static LIST_HEAD(fib6_walkers);
98 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
99 
100 static inline void fib6_walker_link(struct fib6_walker_t *w)
101 {
102 	write_lock_bh(&fib6_walker_lock);
103 	list_add(&w->lh, &fib6_walkers);
104 	write_unlock_bh(&fib6_walker_lock);
105 }
106 
107 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
108 {
109 	write_lock_bh(&fib6_walker_lock);
110 	list_del(&w->lh);
111 	write_unlock_bh(&fib6_walker_lock);
112 }
113 static __inline__ u32 fib6_new_sernum(void)
114 {
115 	u32 n = ++rt_sernum;
116 	if ((__s32)n <= 0)
117 		rt_sernum = n = 1;
118 	return n;
119 }
120 
121 /*
122  *	Auxiliary address test functions for the radix tree.
123  *
124  *	These assume a 32bit processor (although it will work on
125  *	64bit processors)
126  */
127 
128 /*
129  *	test bit
130  */
131 
132 static __inline__ __be32 addr_bit_set(void *token, int fn_bit)
133 {
134 	__be32 *addr = token;
135 
136 	return htonl(1 << ((~fn_bit)&0x1F)) & addr[fn_bit>>5];
137 }
138 
139 static __inline__ struct fib6_node * node_alloc(void)
140 {
141 	struct fib6_node *fn;
142 
143 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
144 
145 	return fn;
146 }
147 
148 static __inline__ void node_free(struct fib6_node * fn)
149 {
150 	kmem_cache_free(fib6_node_kmem, fn);
151 }
152 
153 static __inline__ void rt6_release(struct rt6_info *rt)
154 {
155 	if (atomic_dec_and_test(&rt->rt6i_ref))
156 		dst_free(&rt->u.dst);
157 }
158 
159 static void fib6_link_table(struct net *net, struct fib6_table *tb)
160 {
161 	unsigned int h;
162 
163 	/*
164 	 * Initialize table lock at a single place to give lockdep a key,
165 	 * tables aren't visible prior to being linked to the list.
166 	 */
167 	rwlock_init(&tb->tb6_lock);
168 
169 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
170 
171 	/*
172 	 * No protection necessary, this is the only list mutatation
173 	 * operation, tables never disappear once they exist.
174 	 */
175 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
176 }
177 
178 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
179 
180 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
181 {
182 	struct fib6_table *table;
183 
184 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
185 	if (table != NULL) {
186 		table->tb6_id = id;
187 		table->tb6_root.leaf = net->ipv6.ip6_null_entry;
188 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
189 	}
190 
191 	return table;
192 }
193 
194 struct fib6_table *fib6_new_table(struct net *net, u32 id)
195 {
196 	struct fib6_table *tb;
197 
198 	if (id == 0)
199 		id = RT6_TABLE_MAIN;
200 	tb = fib6_get_table(net, id);
201 	if (tb)
202 		return tb;
203 
204 	tb = fib6_alloc_table(net, id);
205 	if (tb != NULL)
206 		fib6_link_table(net, tb);
207 
208 	return tb;
209 }
210 
211 struct fib6_table *fib6_get_table(struct net *net, u32 id)
212 {
213 	struct fib6_table *tb;
214 	struct hlist_head *head;
215 	struct hlist_node *node;
216 	unsigned int h;
217 
218 	if (id == 0)
219 		id = RT6_TABLE_MAIN;
220 	h = id & (FIB6_TABLE_HASHSZ - 1);
221 	rcu_read_lock();
222 	head = &net->ipv6.fib_table_hash[h];
223 	hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
224 		if (tb->tb6_id == id) {
225 			rcu_read_unlock();
226 			return tb;
227 		}
228 	}
229 	rcu_read_unlock();
230 
231 	return NULL;
232 }
233 
234 static void __net_init fib6_tables_init(struct net *net)
235 {
236 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
237 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
238 }
239 #else
240 
241 struct fib6_table *fib6_new_table(struct net *net, u32 id)
242 {
243 	return fib6_get_table(net, id);
244 }
245 
246 struct fib6_table *fib6_get_table(struct net *net, u32 id)
247 {
248 	  return net->ipv6.fib6_main_tbl;
249 }
250 
251 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi *fl,
252 				   int flags, pol_lookup_t lookup)
253 {
254 	return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl, flags);
255 }
256 
257 static void __net_init fib6_tables_init(struct net *net)
258 {
259 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
260 }
261 
262 #endif
263 
264 static int fib6_dump_node(struct fib6_walker_t *w)
265 {
266 	int res;
267 	struct rt6_info *rt;
268 
269 	for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
270 		res = rt6_dump_route(rt, w->args);
271 		if (res < 0) {
272 			/* Frame is full, suspend walking */
273 			w->leaf = rt;
274 			return 1;
275 		}
276 		WARN_ON(res == 0);
277 	}
278 	w->leaf = NULL;
279 	return 0;
280 }
281 
282 static void fib6_dump_end(struct netlink_callback *cb)
283 {
284 	struct fib6_walker_t *w = (void*)cb->args[2];
285 
286 	if (w) {
287 		if (cb->args[4]) {
288 			cb->args[4] = 0;
289 			fib6_walker_unlink(w);
290 		}
291 		cb->args[2] = 0;
292 		kfree(w);
293 	}
294 	cb->done = (void*)cb->args[3];
295 	cb->args[1] = 3;
296 }
297 
298 static int fib6_dump_done(struct netlink_callback *cb)
299 {
300 	fib6_dump_end(cb);
301 	return cb->done ? cb->done(cb) : 0;
302 }
303 
304 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
305 			   struct netlink_callback *cb)
306 {
307 	struct fib6_walker_t *w;
308 	int res;
309 
310 	w = (void *)cb->args[2];
311 	w->root = &table->tb6_root;
312 
313 	if (cb->args[4] == 0) {
314 		w->count = 0;
315 		w->skip = 0;
316 
317 		read_lock_bh(&table->tb6_lock);
318 		res = fib6_walk(w);
319 		read_unlock_bh(&table->tb6_lock);
320 		if (res > 0) {
321 			cb->args[4] = 1;
322 			cb->args[5] = w->root->fn_sernum;
323 		}
324 	} else {
325 		if (cb->args[5] != w->root->fn_sernum) {
326 			/* Begin at the root if the tree changed */
327 			cb->args[5] = w->root->fn_sernum;
328 			w->state = FWS_INIT;
329 			w->node = w->root;
330 			w->skip = w->count;
331 		} else
332 			w->skip = 0;
333 
334 		read_lock_bh(&table->tb6_lock);
335 		res = fib6_walk_continue(w);
336 		read_unlock_bh(&table->tb6_lock);
337 		if (res <= 0) {
338 			fib6_walker_unlink(w);
339 			cb->args[4] = 0;
340 		}
341 	}
342 
343 	return res;
344 }
345 
346 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
347 {
348 	struct net *net = sock_net(skb->sk);
349 	unsigned int h, s_h;
350 	unsigned int e = 0, s_e;
351 	struct rt6_rtnl_dump_arg arg;
352 	struct fib6_walker_t *w;
353 	struct fib6_table *tb;
354 	struct hlist_node *node;
355 	struct hlist_head *head;
356 	int res = 0;
357 
358 	s_h = cb->args[0];
359 	s_e = cb->args[1];
360 
361 	w = (void *)cb->args[2];
362 	if (w == NULL) {
363 		/* New dump:
364 		 *
365 		 * 1. hook callback destructor.
366 		 */
367 		cb->args[3] = (long)cb->done;
368 		cb->done = fib6_dump_done;
369 
370 		/*
371 		 * 2. allocate and initialize walker.
372 		 */
373 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
374 		if (w == NULL)
375 			return -ENOMEM;
376 		w->func = fib6_dump_node;
377 		cb->args[2] = (long)w;
378 	}
379 
380 	arg.skb = skb;
381 	arg.cb = cb;
382 	arg.net = net;
383 	w->args = &arg;
384 
385 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
386 		e = 0;
387 		head = &net->ipv6.fib_table_hash[h];
388 		hlist_for_each_entry(tb, node, head, tb6_hlist) {
389 			if (e < s_e)
390 				goto next;
391 			res = fib6_dump_table(tb, skb, cb);
392 			if (res != 0)
393 				goto out;
394 next:
395 			e++;
396 		}
397 	}
398 out:
399 	cb->args[1] = e;
400 	cb->args[0] = h;
401 
402 	res = res < 0 ? res : skb->len;
403 	if (res <= 0)
404 		fib6_dump_end(cb);
405 	return res;
406 }
407 
408 /*
409  *	Routing Table
410  *
411  *	return the appropriate node for a routing tree "add" operation
412  *	by either creating and inserting or by returning an existing
413  *	node.
414  */
415 
416 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
417 				     int addrlen, int plen,
418 				     int offset)
419 {
420 	struct fib6_node *fn, *in, *ln;
421 	struct fib6_node *pn = NULL;
422 	struct rt6key *key;
423 	int	bit;
424 	__be32	dir = 0;
425 	__u32	sernum = fib6_new_sernum();
426 
427 	RT6_TRACE("fib6_add_1\n");
428 
429 	/* insert node in tree */
430 
431 	fn = root;
432 
433 	do {
434 		key = (struct rt6key *)((u8 *)fn->leaf + offset);
435 
436 		/*
437 		 *	Prefix match
438 		 */
439 		if (plen < fn->fn_bit ||
440 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
441 			goto insert_above;
442 
443 		/*
444 		 *	Exact match ?
445 		 */
446 
447 		if (plen == fn->fn_bit) {
448 			/* clean up an intermediate node */
449 			if ((fn->fn_flags & RTN_RTINFO) == 0) {
450 				rt6_release(fn->leaf);
451 				fn->leaf = NULL;
452 			}
453 
454 			fn->fn_sernum = sernum;
455 
456 			return fn;
457 		}
458 
459 		/*
460 		 *	We have more bits to go
461 		 */
462 
463 		/* Try to walk down on tree. */
464 		fn->fn_sernum = sernum;
465 		dir = addr_bit_set(addr, fn->fn_bit);
466 		pn = fn;
467 		fn = dir ? fn->right: fn->left;
468 	} while (fn);
469 
470 	/*
471 	 *	We walked to the bottom of tree.
472 	 *	Create new leaf node without children.
473 	 */
474 
475 	ln = node_alloc();
476 
477 	if (ln == NULL)
478 		return NULL;
479 	ln->fn_bit = plen;
480 
481 	ln->parent = pn;
482 	ln->fn_sernum = sernum;
483 
484 	if (dir)
485 		pn->right = ln;
486 	else
487 		pn->left  = ln;
488 
489 	return ln;
490 
491 
492 insert_above:
493 	/*
494 	 * split since we don't have a common prefix anymore or
495 	 * we have a less significant route.
496 	 * we've to insert an intermediate node on the list
497 	 * this new node will point to the one we need to create
498 	 * and the current
499 	 */
500 
501 	pn = fn->parent;
502 
503 	/* find 1st bit in difference between the 2 addrs.
504 
505 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
506 	   but if it is >= plen, the value is ignored in any case.
507 	 */
508 
509 	bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
510 
511 	/*
512 	 *		(intermediate)[in]
513 	 *	          /	   \
514 	 *	(new leaf node)[ln] (old node)[fn]
515 	 */
516 	if (plen > bit) {
517 		in = node_alloc();
518 		ln = node_alloc();
519 
520 		if (in == NULL || ln == NULL) {
521 			if (in)
522 				node_free(in);
523 			if (ln)
524 				node_free(ln);
525 			return NULL;
526 		}
527 
528 		/*
529 		 * new intermediate node.
530 		 * RTN_RTINFO will
531 		 * be off since that an address that chooses one of
532 		 * the branches would not match less specific routes
533 		 * in the other branch
534 		 */
535 
536 		in->fn_bit = bit;
537 
538 		in->parent = pn;
539 		in->leaf = fn->leaf;
540 		atomic_inc(&in->leaf->rt6i_ref);
541 
542 		in->fn_sernum = sernum;
543 
544 		/* update parent pointer */
545 		if (dir)
546 			pn->right = in;
547 		else
548 			pn->left  = in;
549 
550 		ln->fn_bit = plen;
551 
552 		ln->parent = in;
553 		fn->parent = in;
554 
555 		ln->fn_sernum = sernum;
556 
557 		if (addr_bit_set(addr, bit)) {
558 			in->right = ln;
559 			in->left  = fn;
560 		} else {
561 			in->left  = ln;
562 			in->right = fn;
563 		}
564 	} else { /* plen <= bit */
565 
566 		/*
567 		 *		(new leaf node)[ln]
568 		 *	          /	   \
569 		 *	     (old node)[fn] NULL
570 		 */
571 
572 		ln = node_alloc();
573 
574 		if (ln == NULL)
575 			return NULL;
576 
577 		ln->fn_bit = plen;
578 
579 		ln->parent = pn;
580 
581 		ln->fn_sernum = sernum;
582 
583 		if (dir)
584 			pn->right = ln;
585 		else
586 			pn->left  = ln;
587 
588 		if (addr_bit_set(&key->addr, plen))
589 			ln->right = fn;
590 		else
591 			ln->left  = fn;
592 
593 		fn->parent = ln;
594 	}
595 	return ln;
596 }
597 
598 /*
599  *	Insert routing information in a node.
600  */
601 
602 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
603 			    struct nl_info *info)
604 {
605 	struct rt6_info *iter = NULL;
606 	struct rt6_info **ins;
607 
608 	ins = &fn->leaf;
609 
610 	for (iter = fn->leaf; iter; iter=iter->u.dst.rt6_next) {
611 		/*
612 		 *	Search for duplicates
613 		 */
614 
615 		if (iter->rt6i_metric == rt->rt6i_metric) {
616 			/*
617 			 *	Same priority level
618 			 */
619 
620 			if (iter->rt6i_dev == rt->rt6i_dev &&
621 			    iter->rt6i_idev == rt->rt6i_idev &&
622 			    ipv6_addr_equal(&iter->rt6i_gateway,
623 					    &rt->rt6i_gateway)) {
624 				if (!(iter->rt6i_flags&RTF_EXPIRES))
625 					return -EEXIST;
626 				iter->rt6i_expires = rt->rt6i_expires;
627 				if (!(rt->rt6i_flags&RTF_EXPIRES)) {
628 					iter->rt6i_flags &= ~RTF_EXPIRES;
629 					iter->rt6i_expires = 0;
630 				}
631 				return -EEXIST;
632 			}
633 		}
634 
635 		if (iter->rt6i_metric > rt->rt6i_metric)
636 			break;
637 
638 		ins = &iter->u.dst.rt6_next;
639 	}
640 
641 	/* Reset round-robin state, if necessary */
642 	if (ins == &fn->leaf)
643 		fn->rr_ptr = NULL;
644 
645 	/*
646 	 *	insert node
647 	 */
648 
649 	rt->u.dst.rt6_next = iter;
650 	*ins = rt;
651 	rt->rt6i_node = fn;
652 	atomic_inc(&rt->rt6i_ref);
653 	inet6_rt_notify(RTM_NEWROUTE, rt, info);
654 	info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
655 
656 	if ((fn->fn_flags & RTN_RTINFO) == 0) {
657 		info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
658 		fn->fn_flags |= RTN_RTINFO;
659 	}
660 
661 	return 0;
662 }
663 
664 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
665 {
666 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
667 	    (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
668 		mod_timer(&net->ipv6.ip6_fib_timer,
669 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
670 }
671 
672 void fib6_force_start_gc(struct net *net)
673 {
674 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
675 		mod_timer(&net->ipv6.ip6_fib_timer,
676 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
677 }
678 
679 /*
680  *	Add routing information to the routing tree.
681  *	<destination addr>/<source addr>
682  *	with source addr info in sub-trees
683  */
684 
685 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
686 {
687 	struct fib6_node *fn, *pn = NULL;
688 	int err = -ENOMEM;
689 
690 	fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
691 			rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
692 
693 	if (fn == NULL)
694 		goto out;
695 
696 	pn = fn;
697 
698 #ifdef CONFIG_IPV6_SUBTREES
699 	if (rt->rt6i_src.plen) {
700 		struct fib6_node *sn;
701 
702 		if (fn->subtree == NULL) {
703 			struct fib6_node *sfn;
704 
705 			/*
706 			 * Create subtree.
707 			 *
708 			 *		fn[main tree]
709 			 *		|
710 			 *		sfn[subtree root]
711 			 *		   \
712 			 *		    sn[new leaf node]
713 			 */
714 
715 			/* Create subtree root node */
716 			sfn = node_alloc();
717 			if (sfn == NULL)
718 				goto st_failure;
719 
720 			sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
721 			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
722 			sfn->fn_flags = RTN_ROOT;
723 			sfn->fn_sernum = fib6_new_sernum();
724 
725 			/* Now add the first leaf node to new subtree */
726 
727 			sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
728 					sizeof(struct in6_addr), rt->rt6i_src.plen,
729 					offsetof(struct rt6_info, rt6i_src));
730 
731 			if (sn == NULL) {
732 				/* If it is failed, discard just allocated
733 				   root, and then (in st_failure) stale node
734 				   in main tree.
735 				 */
736 				node_free(sfn);
737 				goto st_failure;
738 			}
739 
740 			/* Now link new subtree to main tree */
741 			sfn->parent = fn;
742 			fn->subtree = sfn;
743 		} else {
744 			sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
745 					sizeof(struct in6_addr), rt->rt6i_src.plen,
746 					offsetof(struct rt6_info, rt6i_src));
747 
748 			if (sn == NULL)
749 				goto st_failure;
750 		}
751 
752 		if (fn->leaf == NULL) {
753 			fn->leaf = rt;
754 			atomic_inc(&rt->rt6i_ref);
755 		}
756 		fn = sn;
757 	}
758 #endif
759 
760 	err = fib6_add_rt2node(fn, rt, info);
761 
762 	if (err == 0) {
763 		fib6_start_gc(info->nl_net, rt);
764 		if (!(rt->rt6i_flags&RTF_CACHE))
765 			fib6_prune_clones(info->nl_net, pn, rt);
766 	}
767 
768 out:
769 	if (err) {
770 #ifdef CONFIG_IPV6_SUBTREES
771 		/*
772 		 * If fib6_add_1 has cleared the old leaf pointer in the
773 		 * super-tree leaf node we have to find a new one for it.
774 		 */
775 		if (pn != fn && pn->leaf == rt) {
776 			pn->leaf = NULL;
777 			atomic_dec(&rt->rt6i_ref);
778 		}
779 		if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
780 			pn->leaf = fib6_find_prefix(info->nl_net, pn);
781 #if RT6_DEBUG >= 2
782 			if (!pn->leaf) {
783 				WARN_ON(pn->leaf == NULL);
784 				pn->leaf = info->nl_net->ipv6.ip6_null_entry;
785 			}
786 #endif
787 			atomic_inc(&pn->leaf->rt6i_ref);
788 		}
789 #endif
790 		dst_free(&rt->u.dst);
791 	}
792 	return err;
793 
794 #ifdef CONFIG_IPV6_SUBTREES
795 	/* Subtree creation failed, probably main tree node
796 	   is orphan. If it is, shoot it.
797 	 */
798 st_failure:
799 	if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
800 		fib6_repair_tree(info->nl_net, fn);
801 	dst_free(&rt->u.dst);
802 	return err;
803 #endif
804 }
805 
806 /*
807  *	Routing tree lookup
808  *
809  */
810 
811 struct lookup_args {
812 	int		offset;		/* key offset on rt6_info	*/
813 	struct in6_addr	*addr;		/* search key			*/
814 };
815 
816 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
817 					struct lookup_args *args)
818 {
819 	struct fib6_node *fn;
820 	__be32 dir;
821 
822 	if (unlikely(args->offset == 0))
823 		return NULL;
824 
825 	/*
826 	 *	Descend on a tree
827 	 */
828 
829 	fn = root;
830 
831 	for (;;) {
832 		struct fib6_node *next;
833 
834 		dir = addr_bit_set(args->addr, fn->fn_bit);
835 
836 		next = dir ? fn->right : fn->left;
837 
838 		if (next) {
839 			fn = next;
840 			continue;
841 		}
842 
843 		break;
844 	}
845 
846 	while(fn) {
847 		if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
848 			struct rt6key *key;
849 
850 			key = (struct rt6key *) ((u8 *) fn->leaf +
851 						 args->offset);
852 
853 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
854 #ifdef CONFIG_IPV6_SUBTREES
855 				if (fn->subtree)
856 					fn = fib6_lookup_1(fn->subtree, args + 1);
857 #endif
858 				if (!fn || fn->fn_flags & RTN_RTINFO)
859 					return fn;
860 			}
861 		}
862 
863 		if (fn->fn_flags & RTN_ROOT)
864 			break;
865 
866 		fn = fn->parent;
867 	}
868 
869 	return NULL;
870 }
871 
872 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr,
873 			       struct in6_addr *saddr)
874 {
875 	struct fib6_node *fn;
876 	struct lookup_args args[] = {
877 		{
878 			.offset = offsetof(struct rt6_info, rt6i_dst),
879 			.addr = daddr,
880 		},
881 #ifdef CONFIG_IPV6_SUBTREES
882 		{
883 			.offset = offsetof(struct rt6_info, rt6i_src),
884 			.addr = saddr,
885 		},
886 #endif
887 		{
888 			.offset = 0,	/* sentinel */
889 		}
890 	};
891 
892 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
893 
894 	if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
895 		fn = root;
896 
897 	return fn;
898 }
899 
900 /*
901  *	Get node with specified destination prefix (and source prefix,
902  *	if subtrees are used)
903  */
904 
905 
906 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
907 					struct in6_addr *addr,
908 					int plen, int offset)
909 {
910 	struct fib6_node *fn;
911 
912 	for (fn = root; fn ; ) {
913 		struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
914 
915 		/*
916 		 *	Prefix match
917 		 */
918 		if (plen < fn->fn_bit ||
919 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
920 			return NULL;
921 
922 		if (plen == fn->fn_bit)
923 			return fn;
924 
925 		/*
926 		 *	We have more bits to go
927 		 */
928 		if (addr_bit_set(addr, fn->fn_bit))
929 			fn = fn->right;
930 		else
931 			fn = fn->left;
932 	}
933 	return NULL;
934 }
935 
936 struct fib6_node * fib6_locate(struct fib6_node *root,
937 			       struct in6_addr *daddr, int dst_len,
938 			       struct in6_addr *saddr, int src_len)
939 {
940 	struct fib6_node *fn;
941 
942 	fn = fib6_locate_1(root, daddr, dst_len,
943 			   offsetof(struct rt6_info, rt6i_dst));
944 
945 #ifdef CONFIG_IPV6_SUBTREES
946 	if (src_len) {
947 		WARN_ON(saddr == NULL);
948 		if (fn && fn->subtree)
949 			fn = fib6_locate_1(fn->subtree, saddr, src_len,
950 					   offsetof(struct rt6_info, rt6i_src));
951 	}
952 #endif
953 
954 	if (fn && fn->fn_flags&RTN_RTINFO)
955 		return fn;
956 
957 	return NULL;
958 }
959 
960 
961 /*
962  *	Deletion
963  *
964  */
965 
966 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
967 {
968 	if (fn->fn_flags&RTN_ROOT)
969 		return net->ipv6.ip6_null_entry;
970 
971 	while(fn) {
972 		if(fn->left)
973 			return fn->left->leaf;
974 
975 		if(fn->right)
976 			return fn->right->leaf;
977 
978 		fn = FIB6_SUBTREE(fn);
979 	}
980 	return NULL;
981 }
982 
983 /*
984  *	Called to trim the tree of intermediate nodes when possible. "fn"
985  *	is the node we want to try and remove.
986  */
987 
988 static struct fib6_node *fib6_repair_tree(struct net *net,
989 					   struct fib6_node *fn)
990 {
991 	int children;
992 	int nstate;
993 	struct fib6_node *child, *pn;
994 	struct fib6_walker_t *w;
995 	int iter = 0;
996 
997 	for (;;) {
998 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
999 		iter++;
1000 
1001 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1002 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1003 		WARN_ON(fn->leaf != NULL);
1004 
1005 		children = 0;
1006 		child = NULL;
1007 		if (fn->right) child = fn->right, children |= 1;
1008 		if (fn->left) child = fn->left, children |= 2;
1009 
1010 		if (children == 3 || FIB6_SUBTREE(fn)
1011 #ifdef CONFIG_IPV6_SUBTREES
1012 		    /* Subtree root (i.e. fn) may have one child */
1013 		    || (children && fn->fn_flags&RTN_ROOT)
1014 #endif
1015 		    ) {
1016 			fn->leaf = fib6_find_prefix(net, fn);
1017 #if RT6_DEBUG >= 2
1018 			if (fn->leaf==NULL) {
1019 				WARN_ON(!fn->leaf);
1020 				fn->leaf = net->ipv6.ip6_null_entry;
1021 			}
1022 #endif
1023 			atomic_inc(&fn->leaf->rt6i_ref);
1024 			return fn->parent;
1025 		}
1026 
1027 		pn = fn->parent;
1028 #ifdef CONFIG_IPV6_SUBTREES
1029 		if (FIB6_SUBTREE(pn) == fn) {
1030 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1031 			FIB6_SUBTREE(pn) = NULL;
1032 			nstate = FWS_L;
1033 		} else {
1034 			WARN_ON(fn->fn_flags & RTN_ROOT);
1035 #endif
1036 			if (pn->right == fn) pn->right = child;
1037 			else if (pn->left == fn) pn->left = child;
1038 #if RT6_DEBUG >= 2
1039 			else
1040 				WARN_ON(1);
1041 #endif
1042 			if (child)
1043 				child->parent = pn;
1044 			nstate = FWS_R;
1045 #ifdef CONFIG_IPV6_SUBTREES
1046 		}
1047 #endif
1048 
1049 		read_lock(&fib6_walker_lock);
1050 		FOR_WALKERS(w) {
1051 			if (child == NULL) {
1052 				if (w->root == fn) {
1053 					w->root = w->node = NULL;
1054 					RT6_TRACE("W %p adjusted by delroot 1\n", w);
1055 				} else if (w->node == fn) {
1056 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1057 					w->node = pn;
1058 					w->state = nstate;
1059 				}
1060 			} else {
1061 				if (w->root == fn) {
1062 					w->root = child;
1063 					RT6_TRACE("W %p adjusted by delroot 2\n", w);
1064 				}
1065 				if (w->node == fn) {
1066 					w->node = child;
1067 					if (children&2) {
1068 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1069 						w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1070 					} else {
1071 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1072 						w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1073 					}
1074 				}
1075 			}
1076 		}
1077 		read_unlock(&fib6_walker_lock);
1078 
1079 		node_free(fn);
1080 		if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1081 			return pn;
1082 
1083 		rt6_release(pn->leaf);
1084 		pn->leaf = NULL;
1085 		fn = pn;
1086 	}
1087 }
1088 
1089 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1090 			   struct nl_info *info)
1091 {
1092 	struct fib6_walker_t *w;
1093 	struct rt6_info *rt = *rtp;
1094 	struct net *net = info->nl_net;
1095 
1096 	RT6_TRACE("fib6_del_route\n");
1097 
1098 	/* Unlink it */
1099 	*rtp = rt->u.dst.rt6_next;
1100 	rt->rt6i_node = NULL;
1101 	net->ipv6.rt6_stats->fib_rt_entries--;
1102 	net->ipv6.rt6_stats->fib_discarded_routes++;
1103 
1104 	/* Reset round-robin state, if necessary */
1105 	if (fn->rr_ptr == rt)
1106 		fn->rr_ptr = NULL;
1107 
1108 	/* Adjust walkers */
1109 	read_lock(&fib6_walker_lock);
1110 	FOR_WALKERS(w) {
1111 		if (w->state == FWS_C && w->leaf == rt) {
1112 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1113 			w->leaf = rt->u.dst.rt6_next;
1114 			if (w->leaf == NULL)
1115 				w->state = FWS_U;
1116 		}
1117 	}
1118 	read_unlock(&fib6_walker_lock);
1119 
1120 	rt->u.dst.rt6_next = NULL;
1121 
1122 	/* If it was last route, expunge its radix tree node */
1123 	if (fn->leaf == NULL) {
1124 		fn->fn_flags &= ~RTN_RTINFO;
1125 		net->ipv6.rt6_stats->fib_route_nodes--;
1126 		fn = fib6_repair_tree(net, fn);
1127 	}
1128 
1129 	if (atomic_read(&rt->rt6i_ref) != 1) {
1130 		/* This route is used as dummy address holder in some split
1131 		 * nodes. It is not leaked, but it still holds other resources,
1132 		 * which must be released in time. So, scan ascendant nodes
1133 		 * and replace dummy references to this route with references
1134 		 * to still alive ones.
1135 		 */
1136 		while (fn) {
1137 			if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1138 				fn->leaf = fib6_find_prefix(net, fn);
1139 				atomic_inc(&fn->leaf->rt6i_ref);
1140 				rt6_release(rt);
1141 			}
1142 			fn = fn->parent;
1143 		}
1144 		/* No more references are possible at this point. */
1145 		BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1146 	}
1147 
1148 	inet6_rt_notify(RTM_DELROUTE, rt, info);
1149 	rt6_release(rt);
1150 }
1151 
1152 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1153 {
1154 	struct net *net = info->nl_net;
1155 	struct fib6_node *fn = rt->rt6i_node;
1156 	struct rt6_info **rtp;
1157 
1158 #if RT6_DEBUG >= 2
1159 	if (rt->u.dst.obsolete>0) {
1160 		WARN_ON(fn != NULL);
1161 		return -ENOENT;
1162 	}
1163 #endif
1164 	if (fn == NULL || rt == net->ipv6.ip6_null_entry)
1165 		return -ENOENT;
1166 
1167 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1168 
1169 	if (!(rt->rt6i_flags&RTF_CACHE)) {
1170 		struct fib6_node *pn = fn;
1171 #ifdef CONFIG_IPV6_SUBTREES
1172 		/* clones of this route might be in another subtree */
1173 		if (rt->rt6i_src.plen) {
1174 			while (!(pn->fn_flags&RTN_ROOT))
1175 				pn = pn->parent;
1176 			pn = pn->parent;
1177 		}
1178 #endif
1179 		fib6_prune_clones(info->nl_net, pn, rt);
1180 	}
1181 
1182 	/*
1183 	 *	Walk the leaf entries looking for ourself
1184 	 */
1185 
1186 	for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.dst.rt6_next) {
1187 		if (*rtp == rt) {
1188 			fib6_del_route(fn, rtp, info);
1189 			return 0;
1190 		}
1191 	}
1192 	return -ENOENT;
1193 }
1194 
1195 /*
1196  *	Tree traversal function.
1197  *
1198  *	Certainly, it is not interrupt safe.
1199  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1200  *	It means, that we can modify tree during walking
1201  *	and use this function for garbage collection, clone pruning,
1202  *	cleaning tree when a device goes down etc. etc.
1203  *
1204  *	It guarantees that every node will be traversed,
1205  *	and that it will be traversed only once.
1206  *
1207  *	Callback function w->func may return:
1208  *	0 -> continue walking.
1209  *	positive value -> walking is suspended (used by tree dumps,
1210  *	and probably by gc, if it will be split to several slices)
1211  *	negative value -> terminate walking.
1212  *
1213  *	The function itself returns:
1214  *	0   -> walk is complete.
1215  *	>0  -> walk is incomplete (i.e. suspended)
1216  *	<0  -> walk is terminated by an error.
1217  */
1218 
1219 static int fib6_walk_continue(struct fib6_walker_t *w)
1220 {
1221 	struct fib6_node *fn, *pn;
1222 
1223 	for (;;) {
1224 		fn = w->node;
1225 		if (fn == NULL)
1226 			return 0;
1227 
1228 		if (w->prune && fn != w->root &&
1229 		    fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1230 			w->state = FWS_C;
1231 			w->leaf = fn->leaf;
1232 		}
1233 		switch (w->state) {
1234 #ifdef CONFIG_IPV6_SUBTREES
1235 		case FWS_S:
1236 			if (FIB6_SUBTREE(fn)) {
1237 				w->node = FIB6_SUBTREE(fn);
1238 				continue;
1239 			}
1240 			w->state = FWS_L;
1241 #endif
1242 		case FWS_L:
1243 			if (fn->left) {
1244 				w->node = fn->left;
1245 				w->state = FWS_INIT;
1246 				continue;
1247 			}
1248 			w->state = FWS_R;
1249 		case FWS_R:
1250 			if (fn->right) {
1251 				w->node = fn->right;
1252 				w->state = FWS_INIT;
1253 				continue;
1254 			}
1255 			w->state = FWS_C;
1256 			w->leaf = fn->leaf;
1257 		case FWS_C:
1258 			if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1259 				int err;
1260 
1261 				if (w->count < w->skip) {
1262 					w->count++;
1263 					continue;
1264 				}
1265 
1266 				err = w->func(w);
1267 				if (err)
1268 					return err;
1269 
1270 				w->count++;
1271 				continue;
1272 			}
1273 			w->state = FWS_U;
1274 		case FWS_U:
1275 			if (fn == w->root)
1276 				return 0;
1277 			pn = fn->parent;
1278 			w->node = pn;
1279 #ifdef CONFIG_IPV6_SUBTREES
1280 			if (FIB6_SUBTREE(pn) == fn) {
1281 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1282 				w->state = FWS_L;
1283 				continue;
1284 			}
1285 #endif
1286 			if (pn->left == fn) {
1287 				w->state = FWS_R;
1288 				continue;
1289 			}
1290 			if (pn->right == fn) {
1291 				w->state = FWS_C;
1292 				w->leaf = w->node->leaf;
1293 				continue;
1294 			}
1295 #if RT6_DEBUG >= 2
1296 			WARN_ON(1);
1297 #endif
1298 		}
1299 	}
1300 }
1301 
1302 static int fib6_walk(struct fib6_walker_t *w)
1303 {
1304 	int res;
1305 
1306 	w->state = FWS_INIT;
1307 	w->node = w->root;
1308 
1309 	fib6_walker_link(w);
1310 	res = fib6_walk_continue(w);
1311 	if (res <= 0)
1312 		fib6_walker_unlink(w);
1313 	return res;
1314 }
1315 
1316 static int fib6_clean_node(struct fib6_walker_t *w)
1317 {
1318 	int res;
1319 	struct rt6_info *rt;
1320 	struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1321 	struct nl_info info = {
1322 		.nl_net = c->net,
1323 	};
1324 
1325 	for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
1326 		res = c->func(rt, c->arg);
1327 		if (res < 0) {
1328 			w->leaf = rt;
1329 			res = fib6_del(rt, &info);
1330 			if (res) {
1331 #if RT6_DEBUG >= 2
1332 				printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1333 #endif
1334 				continue;
1335 			}
1336 			return 0;
1337 		}
1338 		WARN_ON(res != 0);
1339 	}
1340 	w->leaf = rt;
1341 	return 0;
1342 }
1343 
1344 /*
1345  *	Convenient frontend to tree walker.
1346  *
1347  *	func is called on each route.
1348  *		It may return -1 -> delete this route.
1349  *		              0  -> continue walking
1350  *
1351  *	prune==1 -> only immediate children of node (certainly,
1352  *	ignoring pure split nodes) will be scanned.
1353  */
1354 
1355 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1356 			    int (*func)(struct rt6_info *, void *arg),
1357 			    int prune, void *arg)
1358 {
1359 	struct fib6_cleaner_t c;
1360 
1361 	c.w.root = root;
1362 	c.w.func = fib6_clean_node;
1363 	c.w.prune = prune;
1364 	c.w.count = 0;
1365 	c.w.skip = 0;
1366 	c.func = func;
1367 	c.arg = arg;
1368 	c.net = net;
1369 
1370 	fib6_walk(&c.w);
1371 }
1372 
1373 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1374 		    int prune, void *arg)
1375 {
1376 	struct fib6_table *table;
1377 	struct hlist_node *node;
1378 	struct hlist_head *head;
1379 	unsigned int h;
1380 
1381 	rcu_read_lock();
1382 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1383 		head = &net->ipv6.fib_table_hash[h];
1384 		hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1385 			write_lock_bh(&table->tb6_lock);
1386 			fib6_clean_tree(net, &table->tb6_root,
1387 					func, prune, arg);
1388 			write_unlock_bh(&table->tb6_lock);
1389 		}
1390 	}
1391 	rcu_read_unlock();
1392 }
1393 
1394 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1395 {
1396 	if (rt->rt6i_flags & RTF_CACHE) {
1397 		RT6_TRACE("pruning clone %p\n", rt);
1398 		return -1;
1399 	}
1400 
1401 	return 0;
1402 }
1403 
1404 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1405 			      struct rt6_info *rt)
1406 {
1407 	fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1408 }
1409 
1410 /*
1411  *	Garbage collection
1412  */
1413 
1414 static struct fib6_gc_args
1415 {
1416 	int			timeout;
1417 	int			more;
1418 } gc_args;
1419 
1420 static int fib6_age(struct rt6_info *rt, void *arg)
1421 {
1422 	unsigned long now = jiffies;
1423 
1424 	/*
1425 	 *	check addrconf expiration here.
1426 	 *	Routes are expired even if they are in use.
1427 	 *
1428 	 *	Also age clones. Note, that clones are aged out
1429 	 *	only if they are not in use now.
1430 	 */
1431 
1432 	if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1433 		if (time_after(now, rt->rt6i_expires)) {
1434 			RT6_TRACE("expiring %p\n", rt);
1435 			return -1;
1436 		}
1437 		gc_args.more++;
1438 	} else if (rt->rt6i_flags & RTF_CACHE) {
1439 		if (atomic_read(&rt->u.dst.__refcnt) == 0 &&
1440 		    time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) {
1441 			RT6_TRACE("aging clone %p\n", rt);
1442 			return -1;
1443 		} else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1444 			   (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) {
1445 			RT6_TRACE("purging route %p via non-router but gateway\n",
1446 				  rt);
1447 			return -1;
1448 		}
1449 		gc_args.more++;
1450 	}
1451 
1452 	return 0;
1453 }
1454 
1455 static DEFINE_SPINLOCK(fib6_gc_lock);
1456 
1457 void fib6_run_gc(unsigned long expires, struct net *net)
1458 {
1459 	if (expires != ~0UL) {
1460 		spin_lock_bh(&fib6_gc_lock);
1461 		gc_args.timeout = expires ? (int)expires :
1462 			net->ipv6.sysctl.ip6_rt_gc_interval;
1463 	} else {
1464 		if (!spin_trylock_bh(&fib6_gc_lock)) {
1465 			mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1466 			return;
1467 		}
1468 		gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1469 	}
1470 
1471 	gc_args.more = icmp6_dst_gc();
1472 
1473 	fib6_clean_all(net, fib6_age, 0, NULL);
1474 
1475 	if (gc_args.more)
1476 		mod_timer(&net->ipv6.ip6_fib_timer,
1477 			  round_jiffies(jiffies
1478 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
1479 	else
1480 		del_timer(&net->ipv6.ip6_fib_timer);
1481 	spin_unlock_bh(&fib6_gc_lock);
1482 }
1483 
1484 static void fib6_gc_timer_cb(unsigned long arg)
1485 {
1486 	fib6_run_gc(0, (struct net *)arg);
1487 }
1488 
1489 static int __net_init fib6_net_init(struct net *net)
1490 {
1491 	setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1492 
1493 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1494 	if (!net->ipv6.rt6_stats)
1495 		goto out_timer;
1496 
1497 	net->ipv6.fib_table_hash = kcalloc(FIB6_TABLE_HASHSZ,
1498 					   sizeof(*net->ipv6.fib_table_hash),
1499 					   GFP_KERNEL);
1500 	if (!net->ipv6.fib_table_hash)
1501 		goto out_rt6_stats;
1502 
1503 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1504 					  GFP_KERNEL);
1505 	if (!net->ipv6.fib6_main_tbl)
1506 		goto out_fib_table_hash;
1507 
1508 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1509 	net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1510 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1511 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1512 
1513 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1514 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1515 					   GFP_KERNEL);
1516 	if (!net->ipv6.fib6_local_tbl)
1517 		goto out_fib6_main_tbl;
1518 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1519 	net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1520 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1521 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1522 #endif
1523 	fib6_tables_init(net);
1524 
1525 	return 0;
1526 
1527 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1528 out_fib6_main_tbl:
1529 	kfree(net->ipv6.fib6_main_tbl);
1530 #endif
1531 out_fib_table_hash:
1532 	kfree(net->ipv6.fib_table_hash);
1533 out_rt6_stats:
1534 	kfree(net->ipv6.rt6_stats);
1535 out_timer:
1536 	return -ENOMEM;
1537  }
1538 
1539 static void fib6_net_exit(struct net *net)
1540 {
1541 	rt6_ifdown(net, NULL);
1542 	del_timer_sync(&net->ipv6.ip6_fib_timer);
1543 
1544 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1545 	kfree(net->ipv6.fib6_local_tbl);
1546 #endif
1547 	kfree(net->ipv6.fib6_main_tbl);
1548 	kfree(net->ipv6.fib_table_hash);
1549 	kfree(net->ipv6.rt6_stats);
1550 }
1551 
1552 static struct pernet_operations fib6_net_ops = {
1553 	.init = fib6_net_init,
1554 	.exit = fib6_net_exit,
1555 };
1556 
1557 int __init fib6_init(void)
1558 {
1559 	int ret = -ENOMEM;
1560 
1561 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
1562 					   sizeof(struct fib6_node),
1563 					   0, SLAB_HWCACHE_ALIGN,
1564 					   NULL);
1565 	if (!fib6_node_kmem)
1566 		goto out;
1567 
1568 	ret = register_pernet_subsys(&fib6_net_ops);
1569 	if (ret)
1570 		goto out_kmem_cache_create;
1571 
1572 	ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib);
1573 	if (ret)
1574 		goto out_unregister_subsys;
1575 out:
1576 	return ret;
1577 
1578 out_unregister_subsys:
1579 	unregister_pernet_subsys(&fib6_net_ops);
1580 out_kmem_cache_create:
1581 	kmem_cache_destroy(fib6_node_kmem);
1582 	goto out;
1583 }
1584 
1585 void fib6_gc_cleanup(void)
1586 {
1587 	unregister_pernet_subsys(&fib6_net_ops);
1588 	kmem_cache_destroy(fib6_node_kmem);
1589 }
1590