xref: /linux/net/ipv6/ip6_fib.c (revision a508da6cc0093171833efb8376b00473f24221b9)
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 	}
201 
202 	return table;
203 }
204 
205 struct fib6_table *fib6_new_table(struct net *net, u32 id)
206 {
207 	struct fib6_table *tb;
208 
209 	if (id == 0)
210 		id = RT6_TABLE_MAIN;
211 	tb = fib6_get_table(net, id);
212 	if (tb)
213 		return tb;
214 
215 	tb = fib6_alloc_table(net, id);
216 	if (tb)
217 		fib6_link_table(net, tb);
218 
219 	return tb;
220 }
221 
222 struct fib6_table *fib6_get_table(struct net *net, u32 id)
223 {
224 	struct fib6_table *tb;
225 	struct hlist_head *head;
226 	struct hlist_node *node;
227 	unsigned int h;
228 
229 	if (id == 0)
230 		id = RT6_TABLE_MAIN;
231 	h = id & (FIB6_TABLE_HASHSZ - 1);
232 	rcu_read_lock();
233 	head = &net->ipv6.fib_table_hash[h];
234 	hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
235 		if (tb->tb6_id == id) {
236 			rcu_read_unlock();
237 			return tb;
238 		}
239 	}
240 	rcu_read_unlock();
241 
242 	return NULL;
243 }
244 
245 static void __net_init fib6_tables_init(struct net *net)
246 {
247 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
248 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
249 }
250 #else
251 
252 struct fib6_table *fib6_new_table(struct net *net, u32 id)
253 {
254 	return fib6_get_table(net, id);
255 }
256 
257 struct fib6_table *fib6_get_table(struct net *net, u32 id)
258 {
259 	  return net->ipv6.fib6_main_tbl;
260 }
261 
262 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
263 				   int flags, pol_lookup_t lookup)
264 {
265 	return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
266 }
267 
268 static void __net_init fib6_tables_init(struct net *net)
269 {
270 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
271 }
272 
273 #endif
274 
275 static int fib6_dump_node(struct fib6_walker_t *w)
276 {
277 	int res;
278 	struct rt6_info *rt;
279 
280 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
281 		res = rt6_dump_route(rt, w->args);
282 		if (res < 0) {
283 			/* Frame is full, suspend walking */
284 			w->leaf = rt;
285 			return 1;
286 		}
287 		WARN_ON(res == 0);
288 	}
289 	w->leaf = NULL;
290 	return 0;
291 }
292 
293 static void fib6_dump_end(struct netlink_callback *cb)
294 {
295 	struct fib6_walker_t *w = (void*)cb->args[2];
296 
297 	if (w) {
298 		if (cb->args[4]) {
299 			cb->args[4] = 0;
300 			fib6_walker_unlink(w);
301 		}
302 		cb->args[2] = 0;
303 		kfree(w);
304 	}
305 	cb->done = (void*)cb->args[3];
306 	cb->args[1] = 3;
307 }
308 
309 static int fib6_dump_done(struct netlink_callback *cb)
310 {
311 	fib6_dump_end(cb);
312 	return cb->done ? cb->done(cb) : 0;
313 }
314 
315 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
316 			   struct netlink_callback *cb)
317 {
318 	struct fib6_walker_t *w;
319 	int res;
320 
321 	w = (void *)cb->args[2];
322 	w->root = &table->tb6_root;
323 
324 	if (cb->args[4] == 0) {
325 		w->count = 0;
326 		w->skip = 0;
327 
328 		read_lock_bh(&table->tb6_lock);
329 		res = fib6_walk(w);
330 		read_unlock_bh(&table->tb6_lock);
331 		if (res > 0) {
332 			cb->args[4] = 1;
333 			cb->args[5] = w->root->fn_sernum;
334 		}
335 	} else {
336 		if (cb->args[5] != w->root->fn_sernum) {
337 			/* Begin at the root if the tree changed */
338 			cb->args[5] = w->root->fn_sernum;
339 			w->state = FWS_INIT;
340 			w->node = w->root;
341 			w->skip = w->count;
342 		} else
343 			w->skip = 0;
344 
345 		read_lock_bh(&table->tb6_lock);
346 		res = fib6_walk_continue(w);
347 		read_unlock_bh(&table->tb6_lock);
348 		if (res <= 0) {
349 			fib6_walker_unlink(w);
350 			cb->args[4] = 0;
351 		}
352 	}
353 
354 	return res;
355 }
356 
357 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
358 {
359 	struct net *net = sock_net(skb->sk);
360 	unsigned int h, s_h;
361 	unsigned int e = 0, s_e;
362 	struct rt6_rtnl_dump_arg arg;
363 	struct fib6_walker_t *w;
364 	struct fib6_table *tb;
365 	struct hlist_node *node;
366 	struct hlist_head *head;
367 	int res = 0;
368 
369 	s_h = cb->args[0];
370 	s_e = cb->args[1];
371 
372 	w = (void *)cb->args[2];
373 	if (!w) {
374 		/* New dump:
375 		 *
376 		 * 1. hook callback destructor.
377 		 */
378 		cb->args[3] = (long)cb->done;
379 		cb->done = fib6_dump_done;
380 
381 		/*
382 		 * 2. allocate and initialize walker.
383 		 */
384 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
385 		if (!w)
386 			return -ENOMEM;
387 		w->func = fib6_dump_node;
388 		cb->args[2] = (long)w;
389 	}
390 
391 	arg.skb = skb;
392 	arg.cb = cb;
393 	arg.net = net;
394 	w->args = &arg;
395 
396 	rcu_read_lock();
397 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
398 		e = 0;
399 		head = &net->ipv6.fib_table_hash[h];
400 		hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
401 			if (e < s_e)
402 				goto next;
403 			res = fib6_dump_table(tb, skb, cb);
404 			if (res != 0)
405 				goto out;
406 next:
407 			e++;
408 		}
409 	}
410 out:
411 	rcu_read_unlock();
412 	cb->args[1] = e;
413 	cb->args[0] = h;
414 
415 	res = res < 0 ? res : skb->len;
416 	if (res <= 0)
417 		fib6_dump_end(cb);
418 	return res;
419 }
420 
421 /*
422  *	Routing Table
423  *
424  *	return the appropriate node for a routing tree "add" operation
425  *	by either creating and inserting or by returning an existing
426  *	node.
427  */
428 
429 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
430 				     int addrlen, int plen,
431 				     int offset, int allow_create,
432 				     int replace_required)
433 {
434 	struct fib6_node *fn, *in, *ln;
435 	struct fib6_node *pn = NULL;
436 	struct rt6key *key;
437 	int	bit;
438 	__be32	dir = 0;
439 	__u32	sernum = fib6_new_sernum();
440 
441 	RT6_TRACE("fib6_add_1\n");
442 
443 	/* insert node in tree */
444 
445 	fn = root;
446 
447 	do {
448 		key = (struct rt6key *)((u8 *)fn->leaf + offset);
449 
450 		/*
451 		 *	Prefix match
452 		 */
453 		if (plen < fn->fn_bit ||
454 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
455 			if (!allow_create) {
456 				if (replace_required) {
457 					pr_warn("Can't replace route, no match found\n");
458 					return ERR_PTR(-ENOENT);
459 				}
460 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
461 			}
462 			goto insert_above;
463 		}
464 
465 		/*
466 		 *	Exact match ?
467 		 */
468 
469 		if (plen == fn->fn_bit) {
470 			/* clean up an intermediate node */
471 			if (!(fn->fn_flags & RTN_RTINFO)) {
472 				rt6_release(fn->leaf);
473 				fn->leaf = NULL;
474 			}
475 
476 			fn->fn_sernum = sernum;
477 
478 			return fn;
479 		}
480 
481 		/*
482 		 *	We have more bits to go
483 		 */
484 
485 		/* Try to walk down on tree. */
486 		fn->fn_sernum = sernum;
487 		dir = addr_bit_set(addr, fn->fn_bit);
488 		pn = fn;
489 		fn = dir ? fn->right: fn->left;
490 	} while (fn);
491 
492 	if (!allow_create) {
493 		/* We should not create new node because
494 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
495 		 * I assume it is safe to require NLM_F_CREATE when
496 		 * REPLACE flag is used! Later we may want to remove the
497 		 * check for replace_required, because according
498 		 * to netlink specification, NLM_F_CREATE
499 		 * MUST be specified if new route is created.
500 		 * That would keep IPv6 consistent with IPv4
501 		 */
502 		if (replace_required) {
503 			pr_warn("Can't replace route, no match found\n");
504 			return ERR_PTR(-ENOENT);
505 		}
506 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
507 	}
508 	/*
509 	 *	We walked to the bottom of tree.
510 	 *	Create new leaf node without children.
511 	 */
512 
513 	ln = node_alloc();
514 
515 	if (!ln)
516 		return NULL;
517 	ln->fn_bit = plen;
518 
519 	ln->parent = pn;
520 	ln->fn_sernum = sernum;
521 
522 	if (dir)
523 		pn->right = ln;
524 	else
525 		pn->left  = ln;
526 
527 	return ln;
528 
529 
530 insert_above:
531 	/*
532 	 * split since we don't have a common prefix anymore or
533 	 * we have a less significant route.
534 	 * we've to insert an intermediate node on the list
535 	 * this new node will point to the one we need to create
536 	 * and the current
537 	 */
538 
539 	pn = fn->parent;
540 
541 	/* find 1st bit in difference between the 2 addrs.
542 
543 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
544 	   but if it is >= plen, the value is ignored in any case.
545 	 */
546 
547 	bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
548 
549 	/*
550 	 *		(intermediate)[in]
551 	 *	          /	   \
552 	 *	(new leaf node)[ln] (old node)[fn]
553 	 */
554 	if (plen > bit) {
555 		in = node_alloc();
556 		ln = node_alloc();
557 
558 		if (!in || !ln) {
559 			if (in)
560 				node_free(in);
561 			if (ln)
562 				node_free(ln);
563 			return NULL;
564 		}
565 
566 		/*
567 		 * new intermediate node.
568 		 * RTN_RTINFO will
569 		 * be off since that an address that chooses one of
570 		 * the branches would not match less specific routes
571 		 * in the other branch
572 		 */
573 
574 		in->fn_bit = bit;
575 
576 		in->parent = pn;
577 		in->leaf = fn->leaf;
578 		atomic_inc(&in->leaf->rt6i_ref);
579 
580 		in->fn_sernum = sernum;
581 
582 		/* update parent pointer */
583 		if (dir)
584 			pn->right = in;
585 		else
586 			pn->left  = in;
587 
588 		ln->fn_bit = plen;
589 
590 		ln->parent = in;
591 		fn->parent = in;
592 
593 		ln->fn_sernum = sernum;
594 
595 		if (addr_bit_set(addr, bit)) {
596 			in->right = ln;
597 			in->left  = fn;
598 		} else {
599 			in->left  = ln;
600 			in->right = fn;
601 		}
602 	} else { /* plen <= bit */
603 
604 		/*
605 		 *		(new leaf node)[ln]
606 		 *	          /	   \
607 		 *	     (old node)[fn] NULL
608 		 */
609 
610 		ln = node_alloc();
611 
612 		if (!ln)
613 			return NULL;
614 
615 		ln->fn_bit = plen;
616 
617 		ln->parent = pn;
618 
619 		ln->fn_sernum = sernum;
620 
621 		if (dir)
622 			pn->right = ln;
623 		else
624 			pn->left  = ln;
625 
626 		if (addr_bit_set(&key->addr, plen))
627 			ln->right = fn;
628 		else
629 			ln->left  = fn;
630 
631 		fn->parent = ln;
632 	}
633 	return ln;
634 }
635 
636 /*
637  *	Insert routing information in a node.
638  */
639 
640 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
641 			    struct nl_info *info)
642 {
643 	struct rt6_info *iter = NULL;
644 	struct rt6_info **ins;
645 	int replace = (info->nlh &&
646 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
647 	int add = (!info->nlh ||
648 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
649 	int found = 0;
650 
651 	ins = &fn->leaf;
652 
653 	for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
654 		/*
655 		 *	Search for duplicates
656 		 */
657 
658 		if (iter->rt6i_metric == rt->rt6i_metric) {
659 			/*
660 			 *	Same priority level
661 			 */
662 			if (info->nlh &&
663 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
664 				return -EEXIST;
665 			if (replace) {
666 				found++;
667 				break;
668 			}
669 
670 			if (iter->dst.dev == rt->dst.dev &&
671 			    iter->rt6i_idev == rt->rt6i_idev &&
672 			    ipv6_addr_equal(&iter->rt6i_gateway,
673 					    &rt->rt6i_gateway)) {
674 				if (!(iter->rt6i_flags & RTF_EXPIRES))
675 					return -EEXIST;
676 				if (!(rt->rt6i_flags & RTF_EXPIRES))
677 					rt6_clean_expires(iter);
678 				else
679 					rt6_set_expires(iter, rt->dst.expires);
680 				return -EEXIST;
681 			}
682 		}
683 
684 		if (iter->rt6i_metric > rt->rt6i_metric)
685 			break;
686 
687 		ins = &iter->dst.rt6_next;
688 	}
689 
690 	/* Reset round-robin state, if necessary */
691 	if (ins == &fn->leaf)
692 		fn->rr_ptr = NULL;
693 
694 	/*
695 	 *	insert node
696 	 */
697 	if (!replace) {
698 		if (!add)
699 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
700 
701 add:
702 		rt->dst.rt6_next = iter;
703 		*ins = rt;
704 		rt->rt6i_node = fn;
705 		atomic_inc(&rt->rt6i_ref);
706 		inet6_rt_notify(RTM_NEWROUTE, rt, info);
707 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
708 
709 		if (!(fn->fn_flags & RTN_RTINFO)) {
710 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
711 			fn->fn_flags |= RTN_RTINFO;
712 		}
713 
714 	} else {
715 		if (!found) {
716 			if (add)
717 				goto add;
718 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
719 			return -ENOENT;
720 		}
721 		*ins = rt;
722 		rt->rt6i_node = fn;
723 		rt->dst.rt6_next = iter->dst.rt6_next;
724 		atomic_inc(&rt->rt6i_ref);
725 		inet6_rt_notify(RTM_NEWROUTE, rt, info);
726 		rt6_release(iter);
727 		if (!(fn->fn_flags & RTN_RTINFO)) {
728 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
729 			fn->fn_flags |= RTN_RTINFO;
730 		}
731 	}
732 
733 	return 0;
734 }
735 
736 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
737 {
738 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
739 	    (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
740 		mod_timer(&net->ipv6.ip6_fib_timer,
741 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
742 }
743 
744 void fib6_force_start_gc(struct net *net)
745 {
746 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
747 		mod_timer(&net->ipv6.ip6_fib_timer,
748 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
749 }
750 
751 /*
752  *	Add routing information to the routing tree.
753  *	<destination addr>/<source addr>
754  *	with source addr info in sub-trees
755  */
756 
757 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
758 {
759 	struct fib6_node *fn, *pn = NULL;
760 	int err = -ENOMEM;
761 	int allow_create = 1;
762 	int replace_required = 0;
763 
764 	if (info->nlh) {
765 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
766 			allow_create = 0;
767 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
768 			replace_required = 1;
769 	}
770 	if (!allow_create && !replace_required)
771 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
772 
773 	fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
774 			rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst),
775 			allow_create, replace_required);
776 
777 	if (IS_ERR(fn)) {
778 		err = PTR_ERR(fn);
779 		fn = NULL;
780 	}
781 
782 	if (!fn)
783 		goto out;
784 
785 	pn = fn;
786 
787 #ifdef CONFIG_IPV6_SUBTREES
788 	if (rt->rt6i_src.plen) {
789 		struct fib6_node *sn;
790 
791 		if (!fn->subtree) {
792 			struct fib6_node *sfn;
793 
794 			/*
795 			 * Create subtree.
796 			 *
797 			 *		fn[main tree]
798 			 *		|
799 			 *		sfn[subtree root]
800 			 *		   \
801 			 *		    sn[new leaf node]
802 			 */
803 
804 			/* Create subtree root node */
805 			sfn = node_alloc();
806 			if (!sfn)
807 				goto st_failure;
808 
809 			sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
810 			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
811 			sfn->fn_flags = RTN_ROOT;
812 			sfn->fn_sernum = fib6_new_sernum();
813 
814 			/* Now add the first leaf node to new subtree */
815 
816 			sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
817 					sizeof(struct in6_addr), rt->rt6i_src.plen,
818 					offsetof(struct rt6_info, rt6i_src),
819 					allow_create, replace_required);
820 
821 			if (!sn) {
822 				/* If it is failed, discard just allocated
823 				   root, and then (in st_failure) stale node
824 				   in main tree.
825 				 */
826 				node_free(sfn);
827 				goto st_failure;
828 			}
829 
830 			/* Now link new subtree to main tree */
831 			sfn->parent = fn;
832 			fn->subtree = sfn;
833 		} else {
834 			sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
835 					sizeof(struct in6_addr), rt->rt6i_src.plen,
836 					offsetof(struct rt6_info, rt6i_src),
837 					allow_create, replace_required);
838 
839 			if (IS_ERR(sn)) {
840 				err = PTR_ERR(sn);
841 				sn = NULL;
842 			}
843 			if (!sn)
844 				goto st_failure;
845 		}
846 
847 		if (!fn->leaf) {
848 			fn->leaf = rt;
849 			atomic_inc(&rt->rt6i_ref);
850 		}
851 		fn = sn;
852 	}
853 #endif
854 
855 	err = fib6_add_rt2node(fn, rt, info);
856 	if (!err) {
857 		fib6_start_gc(info->nl_net, rt);
858 		if (!(rt->rt6i_flags & RTF_CACHE))
859 			fib6_prune_clones(info->nl_net, pn, rt);
860 	}
861 
862 out:
863 	if (err) {
864 #ifdef CONFIG_IPV6_SUBTREES
865 		/*
866 		 * If fib6_add_1 has cleared the old leaf pointer in the
867 		 * super-tree leaf node we have to find a new one for it.
868 		 */
869 		if (pn != fn && pn->leaf == rt) {
870 			pn->leaf = NULL;
871 			atomic_dec(&rt->rt6i_ref);
872 		}
873 		if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
874 			pn->leaf = fib6_find_prefix(info->nl_net, pn);
875 #if RT6_DEBUG >= 2
876 			if (!pn->leaf) {
877 				WARN_ON(pn->leaf == NULL);
878 				pn->leaf = info->nl_net->ipv6.ip6_null_entry;
879 			}
880 #endif
881 			atomic_inc(&pn->leaf->rt6i_ref);
882 		}
883 #endif
884 		dst_free(&rt->dst);
885 	}
886 	return err;
887 
888 #ifdef CONFIG_IPV6_SUBTREES
889 	/* Subtree creation failed, probably main tree node
890 	   is orphan. If it is, shoot it.
891 	 */
892 st_failure:
893 	if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
894 		fib6_repair_tree(info->nl_net, fn);
895 	dst_free(&rt->dst);
896 	return err;
897 #endif
898 }
899 
900 /*
901  *	Routing tree lookup
902  *
903  */
904 
905 struct lookup_args {
906 	int			offset;		/* key offset on rt6_info	*/
907 	const struct in6_addr	*addr;		/* search key			*/
908 };
909 
910 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
911 					struct lookup_args *args)
912 {
913 	struct fib6_node *fn;
914 	__be32 dir;
915 
916 	if (unlikely(args->offset == 0))
917 		return NULL;
918 
919 	/*
920 	 *	Descend on a tree
921 	 */
922 
923 	fn = root;
924 
925 	for (;;) {
926 		struct fib6_node *next;
927 
928 		dir = addr_bit_set(args->addr, fn->fn_bit);
929 
930 		next = dir ? fn->right : fn->left;
931 
932 		if (next) {
933 			fn = next;
934 			continue;
935 		}
936 		break;
937 	}
938 
939 	while (fn) {
940 		if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
941 			struct rt6key *key;
942 
943 			key = (struct rt6key *) ((u8 *) fn->leaf +
944 						 args->offset);
945 
946 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
947 #ifdef CONFIG_IPV6_SUBTREES
948 				if (fn->subtree)
949 					fn = fib6_lookup_1(fn->subtree, args + 1);
950 #endif
951 				if (!fn || fn->fn_flags & RTN_RTINFO)
952 					return fn;
953 			}
954 		}
955 
956 		if (fn->fn_flags & RTN_ROOT)
957 			break;
958 
959 		fn = fn->parent;
960 	}
961 
962 	return NULL;
963 }
964 
965 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
966 			       const struct in6_addr *saddr)
967 {
968 	struct fib6_node *fn;
969 	struct lookup_args args[] = {
970 		{
971 			.offset = offsetof(struct rt6_info, rt6i_dst),
972 			.addr = daddr,
973 		},
974 #ifdef CONFIG_IPV6_SUBTREES
975 		{
976 			.offset = offsetof(struct rt6_info, rt6i_src),
977 			.addr = saddr,
978 		},
979 #endif
980 		{
981 			.offset = 0,	/* sentinel */
982 		}
983 	};
984 
985 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
986 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
987 		fn = root;
988 
989 	return fn;
990 }
991 
992 /*
993  *	Get node with specified destination prefix (and source prefix,
994  *	if subtrees are used)
995  */
996 
997 
998 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
999 					const struct in6_addr *addr,
1000 					int plen, int offset)
1001 {
1002 	struct fib6_node *fn;
1003 
1004 	for (fn = root; fn ; ) {
1005 		struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1006 
1007 		/*
1008 		 *	Prefix match
1009 		 */
1010 		if (plen < fn->fn_bit ||
1011 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1012 			return NULL;
1013 
1014 		if (plen == fn->fn_bit)
1015 			return fn;
1016 
1017 		/*
1018 		 *	We have more bits to go
1019 		 */
1020 		if (addr_bit_set(addr, fn->fn_bit))
1021 			fn = fn->right;
1022 		else
1023 			fn = fn->left;
1024 	}
1025 	return NULL;
1026 }
1027 
1028 struct fib6_node * fib6_locate(struct fib6_node *root,
1029 			       const struct in6_addr *daddr, int dst_len,
1030 			       const struct in6_addr *saddr, int src_len)
1031 {
1032 	struct fib6_node *fn;
1033 
1034 	fn = fib6_locate_1(root, daddr, dst_len,
1035 			   offsetof(struct rt6_info, rt6i_dst));
1036 
1037 #ifdef CONFIG_IPV6_SUBTREES
1038 	if (src_len) {
1039 		WARN_ON(saddr == NULL);
1040 		if (fn && fn->subtree)
1041 			fn = fib6_locate_1(fn->subtree, saddr, src_len,
1042 					   offsetof(struct rt6_info, rt6i_src));
1043 	}
1044 #endif
1045 
1046 	if (fn && fn->fn_flags & RTN_RTINFO)
1047 		return fn;
1048 
1049 	return NULL;
1050 }
1051 
1052 
1053 /*
1054  *	Deletion
1055  *
1056  */
1057 
1058 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1059 {
1060 	if (fn->fn_flags & RTN_ROOT)
1061 		return net->ipv6.ip6_null_entry;
1062 
1063 	while (fn) {
1064 		if (fn->left)
1065 			return fn->left->leaf;
1066 		if (fn->right)
1067 			return fn->right->leaf;
1068 
1069 		fn = FIB6_SUBTREE(fn);
1070 	}
1071 	return NULL;
1072 }
1073 
1074 /*
1075  *	Called to trim the tree of intermediate nodes when possible. "fn"
1076  *	is the node we want to try and remove.
1077  */
1078 
1079 static struct fib6_node *fib6_repair_tree(struct net *net,
1080 					   struct fib6_node *fn)
1081 {
1082 	int children;
1083 	int nstate;
1084 	struct fib6_node *child, *pn;
1085 	struct fib6_walker_t *w;
1086 	int iter = 0;
1087 
1088 	for (;;) {
1089 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1090 		iter++;
1091 
1092 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1093 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1094 		WARN_ON(fn->leaf != NULL);
1095 
1096 		children = 0;
1097 		child = NULL;
1098 		if (fn->right) child = fn->right, children |= 1;
1099 		if (fn->left) child = fn->left, children |= 2;
1100 
1101 		if (children == 3 || FIB6_SUBTREE(fn)
1102 #ifdef CONFIG_IPV6_SUBTREES
1103 		    /* Subtree root (i.e. fn) may have one child */
1104 		    || (children && fn->fn_flags & RTN_ROOT)
1105 #endif
1106 		    ) {
1107 			fn->leaf = fib6_find_prefix(net, fn);
1108 #if RT6_DEBUG >= 2
1109 			if (!fn->leaf) {
1110 				WARN_ON(!fn->leaf);
1111 				fn->leaf = net->ipv6.ip6_null_entry;
1112 			}
1113 #endif
1114 			atomic_inc(&fn->leaf->rt6i_ref);
1115 			return fn->parent;
1116 		}
1117 
1118 		pn = fn->parent;
1119 #ifdef CONFIG_IPV6_SUBTREES
1120 		if (FIB6_SUBTREE(pn) == fn) {
1121 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1122 			FIB6_SUBTREE(pn) = NULL;
1123 			nstate = FWS_L;
1124 		} else {
1125 			WARN_ON(fn->fn_flags & RTN_ROOT);
1126 #endif
1127 			if (pn->right == fn) pn->right = child;
1128 			else if (pn->left == fn) pn->left = child;
1129 #if RT6_DEBUG >= 2
1130 			else
1131 				WARN_ON(1);
1132 #endif
1133 			if (child)
1134 				child->parent = pn;
1135 			nstate = FWS_R;
1136 #ifdef CONFIG_IPV6_SUBTREES
1137 		}
1138 #endif
1139 
1140 		read_lock(&fib6_walker_lock);
1141 		FOR_WALKERS(w) {
1142 			if (!child) {
1143 				if (w->root == fn) {
1144 					w->root = w->node = NULL;
1145 					RT6_TRACE("W %p adjusted by delroot 1\n", w);
1146 				} else if (w->node == fn) {
1147 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1148 					w->node = pn;
1149 					w->state = nstate;
1150 				}
1151 			} else {
1152 				if (w->root == fn) {
1153 					w->root = child;
1154 					RT6_TRACE("W %p adjusted by delroot 2\n", w);
1155 				}
1156 				if (w->node == fn) {
1157 					w->node = child;
1158 					if (children&2) {
1159 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1160 						w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1161 					} else {
1162 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1163 						w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1164 					}
1165 				}
1166 			}
1167 		}
1168 		read_unlock(&fib6_walker_lock);
1169 
1170 		node_free(fn);
1171 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1172 			return pn;
1173 
1174 		rt6_release(pn->leaf);
1175 		pn->leaf = NULL;
1176 		fn = pn;
1177 	}
1178 }
1179 
1180 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1181 			   struct nl_info *info)
1182 {
1183 	struct fib6_walker_t *w;
1184 	struct rt6_info *rt = *rtp;
1185 	struct net *net = info->nl_net;
1186 
1187 	RT6_TRACE("fib6_del_route\n");
1188 
1189 	/* Unlink it */
1190 	*rtp = rt->dst.rt6_next;
1191 	rt->rt6i_node = NULL;
1192 	net->ipv6.rt6_stats->fib_rt_entries--;
1193 	net->ipv6.rt6_stats->fib_discarded_routes++;
1194 
1195 	/* Reset round-robin state, if necessary */
1196 	if (fn->rr_ptr == rt)
1197 		fn->rr_ptr = NULL;
1198 
1199 	/* Adjust walkers */
1200 	read_lock(&fib6_walker_lock);
1201 	FOR_WALKERS(w) {
1202 		if (w->state == FWS_C && w->leaf == rt) {
1203 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1204 			w->leaf = rt->dst.rt6_next;
1205 			if (!w->leaf)
1206 				w->state = FWS_U;
1207 		}
1208 	}
1209 	read_unlock(&fib6_walker_lock);
1210 
1211 	rt->dst.rt6_next = NULL;
1212 
1213 	/* If it was last route, expunge its radix tree node */
1214 	if (!fn->leaf) {
1215 		fn->fn_flags &= ~RTN_RTINFO;
1216 		net->ipv6.rt6_stats->fib_route_nodes--;
1217 		fn = fib6_repair_tree(net, fn);
1218 	}
1219 
1220 	if (atomic_read(&rt->rt6i_ref) != 1) {
1221 		/* This route is used as dummy address holder in some split
1222 		 * nodes. It is not leaked, but it still holds other resources,
1223 		 * which must be released in time. So, scan ascendant nodes
1224 		 * and replace dummy references to this route with references
1225 		 * to still alive ones.
1226 		 */
1227 		while (fn) {
1228 			if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
1229 				fn->leaf = fib6_find_prefix(net, fn);
1230 				atomic_inc(&fn->leaf->rt6i_ref);
1231 				rt6_release(rt);
1232 			}
1233 			fn = fn->parent;
1234 		}
1235 		/* No more references are possible at this point. */
1236 		BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1237 	}
1238 
1239 	inet6_rt_notify(RTM_DELROUTE, rt, info);
1240 	rt6_release(rt);
1241 }
1242 
1243 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1244 {
1245 	struct net *net = info->nl_net;
1246 	struct fib6_node *fn = rt->rt6i_node;
1247 	struct rt6_info **rtp;
1248 
1249 #if RT6_DEBUG >= 2
1250 	if (rt->dst.obsolete>0) {
1251 		WARN_ON(fn != NULL);
1252 		return -ENOENT;
1253 	}
1254 #endif
1255 	if (!fn || rt == net->ipv6.ip6_null_entry)
1256 		return -ENOENT;
1257 
1258 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1259 
1260 	if (!(rt->rt6i_flags & RTF_CACHE)) {
1261 		struct fib6_node *pn = fn;
1262 #ifdef CONFIG_IPV6_SUBTREES
1263 		/* clones of this route might be in another subtree */
1264 		if (rt->rt6i_src.plen) {
1265 			while (!(pn->fn_flags & RTN_ROOT))
1266 				pn = pn->parent;
1267 			pn = pn->parent;
1268 		}
1269 #endif
1270 		fib6_prune_clones(info->nl_net, pn, rt);
1271 	}
1272 
1273 	/*
1274 	 *	Walk the leaf entries looking for ourself
1275 	 */
1276 
1277 	for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1278 		if (*rtp == rt) {
1279 			fib6_del_route(fn, rtp, info);
1280 			return 0;
1281 		}
1282 	}
1283 	return -ENOENT;
1284 }
1285 
1286 /*
1287  *	Tree traversal function.
1288  *
1289  *	Certainly, it is not interrupt safe.
1290  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1291  *	It means, that we can modify tree during walking
1292  *	and use this function for garbage collection, clone pruning,
1293  *	cleaning tree when a device goes down etc. etc.
1294  *
1295  *	It guarantees that every node will be traversed,
1296  *	and that it will be traversed only once.
1297  *
1298  *	Callback function w->func may return:
1299  *	0 -> continue walking.
1300  *	positive value -> walking is suspended (used by tree dumps,
1301  *	and probably by gc, if it will be split to several slices)
1302  *	negative value -> terminate walking.
1303  *
1304  *	The function itself returns:
1305  *	0   -> walk is complete.
1306  *	>0  -> walk is incomplete (i.e. suspended)
1307  *	<0  -> walk is terminated by an error.
1308  */
1309 
1310 static int fib6_walk_continue(struct fib6_walker_t *w)
1311 {
1312 	struct fib6_node *fn, *pn;
1313 
1314 	for (;;) {
1315 		fn = w->node;
1316 		if (!fn)
1317 			return 0;
1318 
1319 		if (w->prune && fn != w->root &&
1320 		    fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1321 			w->state = FWS_C;
1322 			w->leaf = fn->leaf;
1323 		}
1324 		switch (w->state) {
1325 #ifdef CONFIG_IPV6_SUBTREES
1326 		case FWS_S:
1327 			if (FIB6_SUBTREE(fn)) {
1328 				w->node = FIB6_SUBTREE(fn);
1329 				continue;
1330 			}
1331 			w->state = FWS_L;
1332 #endif
1333 		case FWS_L:
1334 			if (fn->left) {
1335 				w->node = fn->left;
1336 				w->state = FWS_INIT;
1337 				continue;
1338 			}
1339 			w->state = FWS_R;
1340 		case FWS_R:
1341 			if (fn->right) {
1342 				w->node = fn->right;
1343 				w->state = FWS_INIT;
1344 				continue;
1345 			}
1346 			w->state = FWS_C;
1347 			w->leaf = fn->leaf;
1348 		case FWS_C:
1349 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1350 				int err;
1351 
1352 				if (w->count < w->skip) {
1353 					w->count++;
1354 					continue;
1355 				}
1356 
1357 				err = w->func(w);
1358 				if (err)
1359 					return err;
1360 
1361 				w->count++;
1362 				continue;
1363 			}
1364 			w->state = FWS_U;
1365 		case FWS_U:
1366 			if (fn == w->root)
1367 				return 0;
1368 			pn = fn->parent;
1369 			w->node = pn;
1370 #ifdef CONFIG_IPV6_SUBTREES
1371 			if (FIB6_SUBTREE(pn) == fn) {
1372 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1373 				w->state = FWS_L;
1374 				continue;
1375 			}
1376 #endif
1377 			if (pn->left == fn) {
1378 				w->state = FWS_R;
1379 				continue;
1380 			}
1381 			if (pn->right == fn) {
1382 				w->state = FWS_C;
1383 				w->leaf = w->node->leaf;
1384 				continue;
1385 			}
1386 #if RT6_DEBUG >= 2
1387 			WARN_ON(1);
1388 #endif
1389 		}
1390 	}
1391 }
1392 
1393 static int fib6_walk(struct fib6_walker_t *w)
1394 {
1395 	int res;
1396 
1397 	w->state = FWS_INIT;
1398 	w->node = w->root;
1399 
1400 	fib6_walker_link(w);
1401 	res = fib6_walk_continue(w);
1402 	if (res <= 0)
1403 		fib6_walker_unlink(w);
1404 	return res;
1405 }
1406 
1407 static int fib6_clean_node(struct fib6_walker_t *w)
1408 {
1409 	int res;
1410 	struct rt6_info *rt;
1411 	struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1412 	struct nl_info info = {
1413 		.nl_net = c->net,
1414 	};
1415 
1416 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1417 		res = c->func(rt, c->arg);
1418 		if (res < 0) {
1419 			w->leaf = rt;
1420 			res = fib6_del(rt, &info);
1421 			if (res) {
1422 #if RT6_DEBUG >= 2
1423 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1424 					 __func__, rt, rt->rt6i_node, res);
1425 #endif
1426 				continue;
1427 			}
1428 			return 0;
1429 		}
1430 		WARN_ON(res != 0);
1431 	}
1432 	w->leaf = rt;
1433 	return 0;
1434 }
1435 
1436 /*
1437  *	Convenient frontend to tree walker.
1438  *
1439  *	func is called on each route.
1440  *		It may return -1 -> delete this route.
1441  *		              0  -> continue walking
1442  *
1443  *	prune==1 -> only immediate children of node (certainly,
1444  *	ignoring pure split nodes) will be scanned.
1445  */
1446 
1447 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1448 			    int (*func)(struct rt6_info *, void *arg),
1449 			    int prune, void *arg)
1450 {
1451 	struct fib6_cleaner_t c;
1452 
1453 	c.w.root = root;
1454 	c.w.func = fib6_clean_node;
1455 	c.w.prune = prune;
1456 	c.w.count = 0;
1457 	c.w.skip = 0;
1458 	c.func = func;
1459 	c.arg = arg;
1460 	c.net = net;
1461 
1462 	fib6_walk(&c.w);
1463 }
1464 
1465 void fib6_clean_all_ro(struct net *net, int (*func)(struct rt6_info *, void *arg),
1466 		    int prune, void *arg)
1467 {
1468 	struct fib6_table *table;
1469 	struct hlist_node *node;
1470 	struct hlist_head *head;
1471 	unsigned int h;
1472 
1473 	rcu_read_lock();
1474 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1475 		head = &net->ipv6.fib_table_hash[h];
1476 		hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1477 			read_lock_bh(&table->tb6_lock);
1478 			fib6_clean_tree(net, &table->tb6_root,
1479 					func, prune, arg);
1480 			read_unlock_bh(&table->tb6_lock);
1481 		}
1482 	}
1483 	rcu_read_unlock();
1484 }
1485 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1486 		    int prune, void *arg)
1487 {
1488 	struct fib6_table *table;
1489 	struct hlist_node *node;
1490 	struct hlist_head *head;
1491 	unsigned int h;
1492 
1493 	rcu_read_lock();
1494 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1495 		head = &net->ipv6.fib_table_hash[h];
1496 		hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1497 			write_lock_bh(&table->tb6_lock);
1498 			fib6_clean_tree(net, &table->tb6_root,
1499 					func, prune, arg);
1500 			write_unlock_bh(&table->tb6_lock);
1501 		}
1502 	}
1503 	rcu_read_unlock();
1504 }
1505 
1506 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1507 {
1508 	if (rt->rt6i_flags & RTF_CACHE) {
1509 		RT6_TRACE("pruning clone %p\n", rt);
1510 		return -1;
1511 	}
1512 
1513 	return 0;
1514 }
1515 
1516 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1517 			      struct rt6_info *rt)
1518 {
1519 	fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1520 }
1521 
1522 /*
1523  *	Garbage collection
1524  */
1525 
1526 static struct fib6_gc_args
1527 {
1528 	int			timeout;
1529 	int			more;
1530 } gc_args;
1531 
1532 static int fib6_age(struct rt6_info *rt, void *arg)
1533 {
1534 	unsigned long now = jiffies;
1535 
1536 	/*
1537 	 *	check addrconf expiration here.
1538 	 *	Routes are expired even if they are in use.
1539 	 *
1540 	 *	Also age clones. Note, that clones are aged out
1541 	 *	only if they are not in use now.
1542 	 */
1543 
1544 	if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1545 		if (time_after(now, rt->dst.expires)) {
1546 			RT6_TRACE("expiring %p\n", rt);
1547 			return -1;
1548 		}
1549 		gc_args.more++;
1550 	} else if (rt->rt6i_flags & RTF_CACHE) {
1551 		if (atomic_read(&rt->dst.__refcnt) == 0 &&
1552 		    time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1553 			RT6_TRACE("aging clone %p\n", rt);
1554 			return -1;
1555 		} else if (rt->rt6i_flags & RTF_GATEWAY) {
1556 			struct neighbour *neigh;
1557 			__u8 neigh_flags = 0;
1558 
1559 			neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1560 			if (neigh) {
1561 				neigh_flags = neigh->flags;
1562 				neigh_release(neigh);
1563 			}
1564 			if (neigh_flags & NTF_ROUTER) {
1565 				RT6_TRACE("purging route %p via non-router but gateway\n",
1566 					  rt);
1567 				return -1;
1568 			}
1569 		}
1570 		gc_args.more++;
1571 	}
1572 
1573 	return 0;
1574 }
1575 
1576 static DEFINE_SPINLOCK(fib6_gc_lock);
1577 
1578 void fib6_run_gc(unsigned long expires, struct net *net)
1579 {
1580 	if (expires != ~0UL) {
1581 		spin_lock_bh(&fib6_gc_lock);
1582 		gc_args.timeout = expires ? (int)expires :
1583 			net->ipv6.sysctl.ip6_rt_gc_interval;
1584 	} else {
1585 		if (!spin_trylock_bh(&fib6_gc_lock)) {
1586 			mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1587 			return;
1588 		}
1589 		gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1590 	}
1591 
1592 	gc_args.more = icmp6_dst_gc();
1593 
1594 	fib6_clean_all(net, fib6_age, 0, NULL);
1595 
1596 	if (gc_args.more)
1597 		mod_timer(&net->ipv6.ip6_fib_timer,
1598 			  round_jiffies(jiffies
1599 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
1600 	else
1601 		del_timer(&net->ipv6.ip6_fib_timer);
1602 	spin_unlock_bh(&fib6_gc_lock);
1603 }
1604 
1605 static void fib6_gc_timer_cb(unsigned long arg)
1606 {
1607 	fib6_run_gc(0, (struct net *)arg);
1608 }
1609 
1610 static int __net_init fib6_net_init(struct net *net)
1611 {
1612 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1613 
1614 	setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1615 
1616 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1617 	if (!net->ipv6.rt6_stats)
1618 		goto out_timer;
1619 
1620 	/* Avoid false sharing : Use at least a full cache line */
1621 	size = max_t(size_t, size, L1_CACHE_BYTES);
1622 
1623 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1624 	if (!net->ipv6.fib_table_hash)
1625 		goto out_rt6_stats;
1626 
1627 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1628 					  GFP_KERNEL);
1629 	if (!net->ipv6.fib6_main_tbl)
1630 		goto out_fib_table_hash;
1631 
1632 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1633 	net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1634 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1635 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1636 
1637 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1638 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1639 					   GFP_KERNEL);
1640 	if (!net->ipv6.fib6_local_tbl)
1641 		goto out_fib6_main_tbl;
1642 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1643 	net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1644 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1645 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1646 #endif
1647 	fib6_tables_init(net);
1648 
1649 	return 0;
1650 
1651 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1652 out_fib6_main_tbl:
1653 	kfree(net->ipv6.fib6_main_tbl);
1654 #endif
1655 out_fib_table_hash:
1656 	kfree(net->ipv6.fib_table_hash);
1657 out_rt6_stats:
1658 	kfree(net->ipv6.rt6_stats);
1659 out_timer:
1660 	return -ENOMEM;
1661  }
1662 
1663 static void fib6_net_exit(struct net *net)
1664 {
1665 	rt6_ifdown(net, NULL);
1666 	del_timer_sync(&net->ipv6.ip6_fib_timer);
1667 
1668 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1669 	kfree(net->ipv6.fib6_local_tbl);
1670 #endif
1671 	kfree(net->ipv6.fib6_main_tbl);
1672 	kfree(net->ipv6.fib_table_hash);
1673 	kfree(net->ipv6.rt6_stats);
1674 }
1675 
1676 static struct pernet_operations fib6_net_ops = {
1677 	.init = fib6_net_init,
1678 	.exit = fib6_net_exit,
1679 };
1680 
1681 int __init fib6_init(void)
1682 {
1683 	int ret = -ENOMEM;
1684 
1685 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
1686 					   sizeof(struct fib6_node),
1687 					   0, SLAB_HWCACHE_ALIGN,
1688 					   NULL);
1689 	if (!fib6_node_kmem)
1690 		goto out;
1691 
1692 	ret = register_pernet_subsys(&fib6_net_ops);
1693 	if (ret)
1694 		goto out_kmem_cache_create;
1695 
1696 	ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1697 			      NULL);
1698 	if (ret)
1699 		goto out_unregister_subsys;
1700 out:
1701 	return ret;
1702 
1703 out_unregister_subsys:
1704 	unregister_pernet_subsys(&fib6_net_ops);
1705 out_kmem_cache_create:
1706 	kmem_cache_destroy(fib6_node_kmem);
1707 	goto out;
1708 }
1709 
1710 void fib6_gc_cleanup(void)
1711 {
1712 	unregister_pernet_subsys(&fib6_net_ops);
1713 	kmem_cache_destroy(fib6_node_kmem);
1714 }
1715