xref: /linux/net/ipv6/ip6_fib.c (revision 148f9bb87745ed45f7a11b2cbd3bc0f017d5d257)
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 	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, 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_head *head;
366 	int res = 0;
367 
368 	s_h = cb->args[0];
369 	s_e = cb->args[1];
370 
371 	w = (void *)cb->args[2];
372 	if (!w) {
373 		/* New dump:
374 		 *
375 		 * 1. hook callback destructor.
376 		 */
377 		cb->args[3] = (long)cb->done;
378 		cb->done = fib6_dump_done;
379 
380 		/*
381 		 * 2. allocate and initialize walker.
382 		 */
383 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
384 		if (!w)
385 			return -ENOMEM;
386 		w->func = fib6_dump_node;
387 		cb->args[2] = (long)w;
388 	}
389 
390 	arg.skb = skb;
391 	arg.cb = cb;
392 	arg.net = net;
393 	w->args = &arg;
394 
395 	rcu_read_lock();
396 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
397 		e = 0;
398 		head = &net->ipv6.fib_table_hash[h];
399 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
400 			if (e < s_e)
401 				goto next;
402 			res = fib6_dump_table(tb, skb, cb);
403 			if (res != 0)
404 				goto out;
405 next:
406 			e++;
407 		}
408 	}
409 out:
410 	rcu_read_unlock();
411 	cb->args[1] = e;
412 	cb->args[0] = h;
413 
414 	res = res < 0 ? res : skb->len;
415 	if (res <= 0)
416 		fib6_dump_end(cb);
417 	return res;
418 }
419 
420 /*
421  *	Routing Table
422  *
423  *	return the appropriate node for a routing tree "add" operation
424  *	by either creating and inserting or by returning an existing
425  *	node.
426  */
427 
428 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
429 				     int addrlen, int plen,
430 				     int offset, int allow_create,
431 				     int replace_required)
432 {
433 	struct fib6_node *fn, *in, *ln;
434 	struct fib6_node *pn = NULL;
435 	struct rt6key *key;
436 	int	bit;
437 	__be32	dir = 0;
438 	__u32	sernum = fib6_new_sernum();
439 
440 	RT6_TRACE("fib6_add_1\n");
441 
442 	/* insert node in tree */
443 
444 	fn = root;
445 
446 	do {
447 		key = (struct rt6key *)((u8 *)fn->leaf + offset);
448 
449 		/*
450 		 *	Prefix match
451 		 */
452 		if (plen < fn->fn_bit ||
453 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
454 			if (!allow_create) {
455 				if (replace_required) {
456 					pr_warn("Can't replace route, no match found\n");
457 					return ERR_PTR(-ENOENT);
458 				}
459 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
460 			}
461 			goto insert_above;
462 		}
463 
464 		/*
465 		 *	Exact match ?
466 		 */
467 
468 		if (plen == fn->fn_bit) {
469 			/* clean up an intermediate node */
470 			if (!(fn->fn_flags & RTN_RTINFO)) {
471 				rt6_release(fn->leaf);
472 				fn->leaf = NULL;
473 			}
474 
475 			fn->fn_sernum = sernum;
476 
477 			return fn;
478 		}
479 
480 		/*
481 		 *	We have more bits to go
482 		 */
483 
484 		/* Try to walk down on tree. */
485 		fn->fn_sernum = sernum;
486 		dir = addr_bit_set(addr, fn->fn_bit);
487 		pn = fn;
488 		fn = dir ? fn->right: fn->left;
489 	} while (fn);
490 
491 	if (!allow_create) {
492 		/* We should not create new node because
493 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
494 		 * I assume it is safe to require NLM_F_CREATE when
495 		 * REPLACE flag is used! Later we may want to remove the
496 		 * check for replace_required, because according
497 		 * to netlink specification, NLM_F_CREATE
498 		 * MUST be specified if new route is created.
499 		 * That would keep IPv6 consistent with IPv4
500 		 */
501 		if (replace_required) {
502 			pr_warn("Can't replace route, no match found\n");
503 			return ERR_PTR(-ENOENT);
504 		}
505 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
506 	}
507 	/*
508 	 *	We walked to the bottom of tree.
509 	 *	Create new leaf node without children.
510 	 */
511 
512 	ln = node_alloc();
513 
514 	if (!ln)
515 		return ERR_PTR(-ENOMEM);
516 	ln->fn_bit = plen;
517 
518 	ln->parent = pn;
519 	ln->fn_sernum = sernum;
520 
521 	if (dir)
522 		pn->right = ln;
523 	else
524 		pn->left  = ln;
525 
526 	return ln;
527 
528 
529 insert_above:
530 	/*
531 	 * split since we don't have a common prefix anymore or
532 	 * we have a less significant route.
533 	 * we've to insert an intermediate node on the list
534 	 * this new node will point to the one we need to create
535 	 * and the current
536 	 */
537 
538 	pn = fn->parent;
539 
540 	/* find 1st bit in difference between the 2 addrs.
541 
542 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
543 	   but if it is >= plen, the value is ignored in any case.
544 	 */
545 
546 	bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
547 
548 	/*
549 	 *		(intermediate)[in]
550 	 *	          /	   \
551 	 *	(new leaf node)[ln] (old node)[fn]
552 	 */
553 	if (plen > bit) {
554 		in = node_alloc();
555 		ln = node_alloc();
556 
557 		if (!in || !ln) {
558 			if (in)
559 				node_free(in);
560 			if (ln)
561 				node_free(ln);
562 			return ERR_PTR(-ENOMEM);
563 		}
564 
565 		/*
566 		 * new intermediate node.
567 		 * RTN_RTINFO will
568 		 * be off since that an address that chooses one of
569 		 * the branches would not match less specific routes
570 		 * in the other branch
571 		 */
572 
573 		in->fn_bit = bit;
574 
575 		in->parent = pn;
576 		in->leaf = fn->leaf;
577 		atomic_inc(&in->leaf->rt6i_ref);
578 
579 		in->fn_sernum = sernum;
580 
581 		/* update parent pointer */
582 		if (dir)
583 			pn->right = in;
584 		else
585 			pn->left  = in;
586 
587 		ln->fn_bit = plen;
588 
589 		ln->parent = in;
590 		fn->parent = in;
591 
592 		ln->fn_sernum = sernum;
593 
594 		if (addr_bit_set(addr, bit)) {
595 			in->right = ln;
596 			in->left  = fn;
597 		} else {
598 			in->left  = ln;
599 			in->right = fn;
600 		}
601 	} else { /* plen <= bit */
602 
603 		/*
604 		 *		(new leaf node)[ln]
605 		 *	          /	   \
606 		 *	     (old node)[fn] NULL
607 		 */
608 
609 		ln = node_alloc();
610 
611 		if (!ln)
612 			return ERR_PTR(-ENOMEM);
613 
614 		ln->fn_bit = plen;
615 
616 		ln->parent = pn;
617 
618 		ln->fn_sernum = sernum;
619 
620 		if (dir)
621 			pn->right = ln;
622 		else
623 			pn->left  = ln;
624 
625 		if (addr_bit_set(&key->addr, plen))
626 			ln->right = fn;
627 		else
628 			ln->left  = fn;
629 
630 		fn->parent = ln;
631 	}
632 	return ln;
633 }
634 
635 static inline bool rt6_qualify_for_ecmp(struct rt6_info *rt)
636 {
637 	return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
638 	       RTF_GATEWAY;
639 }
640 
641 /*
642  *	Insert routing information in a node.
643  */
644 
645 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
646 			    struct nl_info *info)
647 {
648 	struct rt6_info *iter = NULL;
649 	struct rt6_info **ins;
650 	int replace = (info->nlh &&
651 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
652 	int add = (!info->nlh ||
653 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
654 	int found = 0;
655 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
656 
657 	ins = &fn->leaf;
658 
659 	for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
660 		/*
661 		 *	Search for duplicates
662 		 */
663 
664 		if (iter->rt6i_metric == rt->rt6i_metric) {
665 			/*
666 			 *	Same priority level
667 			 */
668 			if (info->nlh &&
669 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
670 				return -EEXIST;
671 			if (replace) {
672 				found++;
673 				break;
674 			}
675 
676 			if (iter->dst.dev == rt->dst.dev &&
677 			    iter->rt6i_idev == rt->rt6i_idev &&
678 			    ipv6_addr_equal(&iter->rt6i_gateway,
679 					    &rt->rt6i_gateway)) {
680 				if (rt->rt6i_nsiblings)
681 					rt->rt6i_nsiblings = 0;
682 				if (!(iter->rt6i_flags & RTF_EXPIRES))
683 					return -EEXIST;
684 				if (!(rt->rt6i_flags & RTF_EXPIRES))
685 					rt6_clean_expires(iter);
686 				else
687 					rt6_set_expires(iter, rt->dst.expires);
688 				return -EEXIST;
689 			}
690 			/* If we have the same destination and the same metric,
691 			 * but not the same gateway, then the route we try to
692 			 * add is sibling to this route, increment our counter
693 			 * of siblings, and later we will add our route to the
694 			 * list.
695 			 * Only static routes (which don't have flag
696 			 * RTF_EXPIRES) are used for ECMPv6.
697 			 *
698 			 * To avoid long list, we only had siblings if the
699 			 * route have a gateway.
700 			 */
701 			if (rt_can_ecmp &&
702 			    rt6_qualify_for_ecmp(iter))
703 				rt->rt6i_nsiblings++;
704 		}
705 
706 		if (iter->rt6i_metric > rt->rt6i_metric)
707 			break;
708 
709 		ins = &iter->dst.rt6_next;
710 	}
711 
712 	/* Reset round-robin state, if necessary */
713 	if (ins == &fn->leaf)
714 		fn->rr_ptr = NULL;
715 
716 	/* Link this route to others same route. */
717 	if (rt->rt6i_nsiblings) {
718 		unsigned int rt6i_nsiblings;
719 		struct rt6_info *sibling, *temp_sibling;
720 
721 		/* Find the first route that have the same metric */
722 		sibling = fn->leaf;
723 		while (sibling) {
724 			if (sibling->rt6i_metric == rt->rt6i_metric &&
725 			    rt6_qualify_for_ecmp(sibling)) {
726 				list_add_tail(&rt->rt6i_siblings,
727 					      &sibling->rt6i_siblings);
728 				break;
729 			}
730 			sibling = sibling->dst.rt6_next;
731 		}
732 		/* For each sibling in the list, increment the counter of
733 		 * siblings. BUG() if counters does not match, list of siblings
734 		 * is broken!
735 		 */
736 		rt6i_nsiblings = 0;
737 		list_for_each_entry_safe(sibling, temp_sibling,
738 					 &rt->rt6i_siblings, rt6i_siblings) {
739 			sibling->rt6i_nsiblings++;
740 			BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
741 			rt6i_nsiblings++;
742 		}
743 		BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
744 	}
745 
746 	/*
747 	 *	insert node
748 	 */
749 	if (!replace) {
750 		if (!add)
751 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
752 
753 add:
754 		rt->dst.rt6_next = iter;
755 		*ins = rt;
756 		rt->rt6i_node = fn;
757 		atomic_inc(&rt->rt6i_ref);
758 		inet6_rt_notify(RTM_NEWROUTE, rt, info);
759 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
760 
761 		if (!(fn->fn_flags & RTN_RTINFO)) {
762 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
763 			fn->fn_flags |= RTN_RTINFO;
764 		}
765 
766 	} else {
767 		if (!found) {
768 			if (add)
769 				goto add;
770 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
771 			return -ENOENT;
772 		}
773 		*ins = rt;
774 		rt->rt6i_node = fn;
775 		rt->dst.rt6_next = iter->dst.rt6_next;
776 		atomic_inc(&rt->rt6i_ref);
777 		inet6_rt_notify(RTM_NEWROUTE, rt, info);
778 		rt6_release(iter);
779 		if (!(fn->fn_flags & RTN_RTINFO)) {
780 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
781 			fn->fn_flags |= RTN_RTINFO;
782 		}
783 	}
784 
785 	return 0;
786 }
787 
788 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
789 {
790 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
791 	    (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
792 		mod_timer(&net->ipv6.ip6_fib_timer,
793 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
794 }
795 
796 void fib6_force_start_gc(struct net *net)
797 {
798 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
799 		mod_timer(&net->ipv6.ip6_fib_timer,
800 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
801 }
802 
803 /*
804  *	Add routing information to the routing tree.
805  *	<destination addr>/<source addr>
806  *	with source addr info in sub-trees
807  */
808 
809 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
810 {
811 	struct fib6_node *fn, *pn = NULL;
812 	int err = -ENOMEM;
813 	int allow_create = 1;
814 	int replace_required = 0;
815 
816 	if (info->nlh) {
817 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
818 			allow_create = 0;
819 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
820 			replace_required = 1;
821 	}
822 	if (!allow_create && !replace_required)
823 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
824 
825 	fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
826 			rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst),
827 			allow_create, replace_required);
828 
829 	if (IS_ERR(fn)) {
830 		err = PTR_ERR(fn);
831 		goto out;
832 	}
833 
834 	pn = fn;
835 
836 #ifdef CONFIG_IPV6_SUBTREES
837 	if (rt->rt6i_src.plen) {
838 		struct fib6_node *sn;
839 
840 		if (!fn->subtree) {
841 			struct fib6_node *sfn;
842 
843 			/*
844 			 * Create subtree.
845 			 *
846 			 *		fn[main tree]
847 			 *		|
848 			 *		sfn[subtree root]
849 			 *		   \
850 			 *		    sn[new leaf node]
851 			 */
852 
853 			/* Create subtree root node */
854 			sfn = node_alloc();
855 			if (!sfn)
856 				goto st_failure;
857 
858 			sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
859 			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
860 			sfn->fn_flags = RTN_ROOT;
861 			sfn->fn_sernum = fib6_new_sernum();
862 
863 			/* Now add the first leaf node to new subtree */
864 
865 			sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
866 					sizeof(struct in6_addr), rt->rt6i_src.plen,
867 					offsetof(struct rt6_info, rt6i_src),
868 					allow_create, replace_required);
869 
870 			if (IS_ERR(sn)) {
871 				/* If it is failed, discard just allocated
872 				   root, and then (in st_failure) stale node
873 				   in main tree.
874 				 */
875 				node_free(sfn);
876 				err = PTR_ERR(sn);
877 				goto st_failure;
878 			}
879 
880 			/* Now link new subtree to main tree */
881 			sfn->parent = fn;
882 			fn->subtree = sfn;
883 		} else {
884 			sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
885 					sizeof(struct in6_addr), rt->rt6i_src.plen,
886 					offsetof(struct rt6_info, rt6i_src),
887 					allow_create, replace_required);
888 
889 			if (IS_ERR(sn)) {
890 				err = PTR_ERR(sn);
891 				goto st_failure;
892 			}
893 		}
894 
895 		if (!fn->leaf) {
896 			fn->leaf = rt;
897 			atomic_inc(&rt->rt6i_ref);
898 		}
899 		fn = sn;
900 	}
901 #endif
902 
903 	err = fib6_add_rt2node(fn, rt, info);
904 	if (!err) {
905 		fib6_start_gc(info->nl_net, rt);
906 		if (!(rt->rt6i_flags & RTF_CACHE))
907 			fib6_prune_clones(info->nl_net, pn, rt);
908 	}
909 
910 out:
911 	if (err) {
912 #ifdef CONFIG_IPV6_SUBTREES
913 		/*
914 		 * If fib6_add_1 has cleared the old leaf pointer in the
915 		 * super-tree leaf node we have to find a new one for it.
916 		 */
917 		if (pn != fn && pn->leaf == rt) {
918 			pn->leaf = NULL;
919 			atomic_dec(&rt->rt6i_ref);
920 		}
921 		if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
922 			pn->leaf = fib6_find_prefix(info->nl_net, pn);
923 #if RT6_DEBUG >= 2
924 			if (!pn->leaf) {
925 				WARN_ON(pn->leaf == NULL);
926 				pn->leaf = info->nl_net->ipv6.ip6_null_entry;
927 			}
928 #endif
929 			atomic_inc(&pn->leaf->rt6i_ref);
930 		}
931 #endif
932 		dst_free(&rt->dst);
933 	}
934 	return err;
935 
936 #ifdef CONFIG_IPV6_SUBTREES
937 	/* Subtree creation failed, probably main tree node
938 	   is orphan. If it is, shoot it.
939 	 */
940 st_failure:
941 	if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
942 		fib6_repair_tree(info->nl_net, fn);
943 	dst_free(&rt->dst);
944 	return err;
945 #endif
946 }
947 
948 /*
949  *	Routing tree lookup
950  *
951  */
952 
953 struct lookup_args {
954 	int			offset;		/* key offset on rt6_info	*/
955 	const struct in6_addr	*addr;		/* search key			*/
956 };
957 
958 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
959 					struct lookup_args *args)
960 {
961 	struct fib6_node *fn;
962 	__be32 dir;
963 
964 	if (unlikely(args->offset == 0))
965 		return NULL;
966 
967 	/*
968 	 *	Descend on a tree
969 	 */
970 
971 	fn = root;
972 
973 	for (;;) {
974 		struct fib6_node *next;
975 
976 		dir = addr_bit_set(args->addr, fn->fn_bit);
977 
978 		next = dir ? fn->right : fn->left;
979 
980 		if (next) {
981 			fn = next;
982 			continue;
983 		}
984 		break;
985 	}
986 
987 	while (fn) {
988 		if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
989 			struct rt6key *key;
990 
991 			key = (struct rt6key *) ((u8 *) fn->leaf +
992 						 args->offset);
993 
994 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
995 #ifdef CONFIG_IPV6_SUBTREES
996 				if (fn->subtree)
997 					fn = fib6_lookup_1(fn->subtree, args + 1);
998 #endif
999 				if (!fn || fn->fn_flags & RTN_RTINFO)
1000 					return fn;
1001 			}
1002 		}
1003 
1004 		if (fn->fn_flags & RTN_ROOT)
1005 			break;
1006 
1007 		fn = fn->parent;
1008 	}
1009 
1010 	return NULL;
1011 }
1012 
1013 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1014 			       const struct in6_addr *saddr)
1015 {
1016 	struct fib6_node *fn;
1017 	struct lookup_args args[] = {
1018 		{
1019 			.offset = offsetof(struct rt6_info, rt6i_dst),
1020 			.addr = daddr,
1021 		},
1022 #ifdef CONFIG_IPV6_SUBTREES
1023 		{
1024 			.offset = offsetof(struct rt6_info, rt6i_src),
1025 			.addr = saddr,
1026 		},
1027 #endif
1028 		{
1029 			.offset = 0,	/* sentinel */
1030 		}
1031 	};
1032 
1033 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
1034 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1035 		fn = root;
1036 
1037 	return fn;
1038 }
1039 
1040 /*
1041  *	Get node with specified destination prefix (and source prefix,
1042  *	if subtrees are used)
1043  */
1044 
1045 
1046 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
1047 					const struct in6_addr *addr,
1048 					int plen, int offset)
1049 {
1050 	struct fib6_node *fn;
1051 
1052 	for (fn = root; fn ; ) {
1053 		struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1054 
1055 		/*
1056 		 *	Prefix match
1057 		 */
1058 		if (plen < fn->fn_bit ||
1059 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1060 			return NULL;
1061 
1062 		if (plen == fn->fn_bit)
1063 			return fn;
1064 
1065 		/*
1066 		 *	We have more bits to go
1067 		 */
1068 		if (addr_bit_set(addr, fn->fn_bit))
1069 			fn = fn->right;
1070 		else
1071 			fn = fn->left;
1072 	}
1073 	return NULL;
1074 }
1075 
1076 struct fib6_node * fib6_locate(struct fib6_node *root,
1077 			       const struct in6_addr *daddr, int dst_len,
1078 			       const struct in6_addr *saddr, int src_len)
1079 {
1080 	struct fib6_node *fn;
1081 
1082 	fn = fib6_locate_1(root, daddr, dst_len,
1083 			   offsetof(struct rt6_info, rt6i_dst));
1084 
1085 #ifdef CONFIG_IPV6_SUBTREES
1086 	if (src_len) {
1087 		WARN_ON(saddr == NULL);
1088 		if (fn && fn->subtree)
1089 			fn = fib6_locate_1(fn->subtree, saddr, src_len,
1090 					   offsetof(struct rt6_info, rt6i_src));
1091 	}
1092 #endif
1093 
1094 	if (fn && fn->fn_flags & RTN_RTINFO)
1095 		return fn;
1096 
1097 	return NULL;
1098 }
1099 
1100 
1101 /*
1102  *	Deletion
1103  *
1104  */
1105 
1106 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1107 {
1108 	if (fn->fn_flags & RTN_ROOT)
1109 		return net->ipv6.ip6_null_entry;
1110 
1111 	while (fn) {
1112 		if (fn->left)
1113 			return fn->left->leaf;
1114 		if (fn->right)
1115 			return fn->right->leaf;
1116 
1117 		fn = FIB6_SUBTREE(fn);
1118 	}
1119 	return NULL;
1120 }
1121 
1122 /*
1123  *	Called to trim the tree of intermediate nodes when possible. "fn"
1124  *	is the node we want to try and remove.
1125  */
1126 
1127 static struct fib6_node *fib6_repair_tree(struct net *net,
1128 					   struct fib6_node *fn)
1129 {
1130 	int children;
1131 	int nstate;
1132 	struct fib6_node *child, *pn;
1133 	struct fib6_walker_t *w;
1134 	int iter = 0;
1135 
1136 	for (;;) {
1137 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1138 		iter++;
1139 
1140 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1141 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1142 		WARN_ON(fn->leaf != NULL);
1143 
1144 		children = 0;
1145 		child = NULL;
1146 		if (fn->right) child = fn->right, children |= 1;
1147 		if (fn->left) child = fn->left, children |= 2;
1148 
1149 		if (children == 3 || FIB6_SUBTREE(fn)
1150 #ifdef CONFIG_IPV6_SUBTREES
1151 		    /* Subtree root (i.e. fn) may have one child */
1152 		    || (children && fn->fn_flags & RTN_ROOT)
1153 #endif
1154 		    ) {
1155 			fn->leaf = fib6_find_prefix(net, fn);
1156 #if RT6_DEBUG >= 2
1157 			if (!fn->leaf) {
1158 				WARN_ON(!fn->leaf);
1159 				fn->leaf = net->ipv6.ip6_null_entry;
1160 			}
1161 #endif
1162 			atomic_inc(&fn->leaf->rt6i_ref);
1163 			return fn->parent;
1164 		}
1165 
1166 		pn = fn->parent;
1167 #ifdef CONFIG_IPV6_SUBTREES
1168 		if (FIB6_SUBTREE(pn) == fn) {
1169 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1170 			FIB6_SUBTREE(pn) = NULL;
1171 			nstate = FWS_L;
1172 		} else {
1173 			WARN_ON(fn->fn_flags & RTN_ROOT);
1174 #endif
1175 			if (pn->right == fn) pn->right = child;
1176 			else if (pn->left == fn) pn->left = child;
1177 #if RT6_DEBUG >= 2
1178 			else
1179 				WARN_ON(1);
1180 #endif
1181 			if (child)
1182 				child->parent = pn;
1183 			nstate = FWS_R;
1184 #ifdef CONFIG_IPV6_SUBTREES
1185 		}
1186 #endif
1187 
1188 		read_lock(&fib6_walker_lock);
1189 		FOR_WALKERS(w) {
1190 			if (!child) {
1191 				if (w->root == fn) {
1192 					w->root = w->node = NULL;
1193 					RT6_TRACE("W %p adjusted by delroot 1\n", w);
1194 				} else if (w->node == fn) {
1195 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1196 					w->node = pn;
1197 					w->state = nstate;
1198 				}
1199 			} else {
1200 				if (w->root == fn) {
1201 					w->root = child;
1202 					RT6_TRACE("W %p adjusted by delroot 2\n", w);
1203 				}
1204 				if (w->node == fn) {
1205 					w->node = child;
1206 					if (children&2) {
1207 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1208 						w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1209 					} else {
1210 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1211 						w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1212 					}
1213 				}
1214 			}
1215 		}
1216 		read_unlock(&fib6_walker_lock);
1217 
1218 		node_free(fn);
1219 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1220 			return pn;
1221 
1222 		rt6_release(pn->leaf);
1223 		pn->leaf = NULL;
1224 		fn = pn;
1225 	}
1226 }
1227 
1228 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1229 			   struct nl_info *info)
1230 {
1231 	struct fib6_walker_t *w;
1232 	struct rt6_info *rt = *rtp;
1233 	struct net *net = info->nl_net;
1234 
1235 	RT6_TRACE("fib6_del_route\n");
1236 
1237 	/* Unlink it */
1238 	*rtp = rt->dst.rt6_next;
1239 	rt->rt6i_node = NULL;
1240 	net->ipv6.rt6_stats->fib_rt_entries--;
1241 	net->ipv6.rt6_stats->fib_discarded_routes++;
1242 
1243 	/* Reset round-robin state, if necessary */
1244 	if (fn->rr_ptr == rt)
1245 		fn->rr_ptr = NULL;
1246 
1247 	/* Remove this entry from other siblings */
1248 	if (rt->rt6i_nsiblings) {
1249 		struct rt6_info *sibling, *next_sibling;
1250 
1251 		list_for_each_entry_safe(sibling, next_sibling,
1252 					 &rt->rt6i_siblings, rt6i_siblings)
1253 			sibling->rt6i_nsiblings--;
1254 		rt->rt6i_nsiblings = 0;
1255 		list_del_init(&rt->rt6i_siblings);
1256 	}
1257 
1258 	/* Adjust walkers */
1259 	read_lock(&fib6_walker_lock);
1260 	FOR_WALKERS(w) {
1261 		if (w->state == FWS_C && w->leaf == rt) {
1262 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1263 			w->leaf = rt->dst.rt6_next;
1264 			if (!w->leaf)
1265 				w->state = FWS_U;
1266 		}
1267 	}
1268 	read_unlock(&fib6_walker_lock);
1269 
1270 	rt->dst.rt6_next = NULL;
1271 
1272 	/* If it was last route, expunge its radix tree node */
1273 	if (!fn->leaf) {
1274 		fn->fn_flags &= ~RTN_RTINFO;
1275 		net->ipv6.rt6_stats->fib_route_nodes--;
1276 		fn = fib6_repair_tree(net, fn);
1277 	}
1278 
1279 	if (atomic_read(&rt->rt6i_ref) != 1) {
1280 		/* This route is used as dummy address holder in some split
1281 		 * nodes. It is not leaked, but it still holds other resources,
1282 		 * which must be released in time. So, scan ascendant nodes
1283 		 * and replace dummy references to this route with references
1284 		 * to still alive ones.
1285 		 */
1286 		while (fn) {
1287 			if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
1288 				fn->leaf = fib6_find_prefix(net, fn);
1289 				atomic_inc(&fn->leaf->rt6i_ref);
1290 				rt6_release(rt);
1291 			}
1292 			fn = fn->parent;
1293 		}
1294 		/* No more references are possible at this point. */
1295 		BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1296 	}
1297 
1298 	inet6_rt_notify(RTM_DELROUTE, rt, info);
1299 	rt6_release(rt);
1300 }
1301 
1302 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1303 {
1304 	struct net *net = info->nl_net;
1305 	struct fib6_node *fn = rt->rt6i_node;
1306 	struct rt6_info **rtp;
1307 
1308 #if RT6_DEBUG >= 2
1309 	if (rt->dst.obsolete>0) {
1310 		WARN_ON(fn != NULL);
1311 		return -ENOENT;
1312 	}
1313 #endif
1314 	if (!fn || rt == net->ipv6.ip6_null_entry)
1315 		return -ENOENT;
1316 
1317 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1318 
1319 	if (!(rt->rt6i_flags & RTF_CACHE)) {
1320 		struct fib6_node *pn = fn;
1321 #ifdef CONFIG_IPV6_SUBTREES
1322 		/* clones of this route might be in another subtree */
1323 		if (rt->rt6i_src.plen) {
1324 			while (!(pn->fn_flags & RTN_ROOT))
1325 				pn = pn->parent;
1326 			pn = pn->parent;
1327 		}
1328 #endif
1329 		fib6_prune_clones(info->nl_net, pn, rt);
1330 	}
1331 
1332 	/*
1333 	 *	Walk the leaf entries looking for ourself
1334 	 */
1335 
1336 	for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1337 		if (*rtp == rt) {
1338 			fib6_del_route(fn, rtp, info);
1339 			return 0;
1340 		}
1341 	}
1342 	return -ENOENT;
1343 }
1344 
1345 /*
1346  *	Tree traversal function.
1347  *
1348  *	Certainly, it is not interrupt safe.
1349  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1350  *	It means, that we can modify tree during walking
1351  *	and use this function for garbage collection, clone pruning,
1352  *	cleaning tree when a device goes down etc. etc.
1353  *
1354  *	It guarantees that every node will be traversed,
1355  *	and that it will be traversed only once.
1356  *
1357  *	Callback function w->func may return:
1358  *	0 -> continue walking.
1359  *	positive value -> walking is suspended (used by tree dumps,
1360  *	and probably by gc, if it will be split to several slices)
1361  *	negative value -> terminate walking.
1362  *
1363  *	The function itself returns:
1364  *	0   -> walk is complete.
1365  *	>0  -> walk is incomplete (i.e. suspended)
1366  *	<0  -> walk is terminated by an error.
1367  */
1368 
1369 static int fib6_walk_continue(struct fib6_walker_t *w)
1370 {
1371 	struct fib6_node *fn, *pn;
1372 
1373 	for (;;) {
1374 		fn = w->node;
1375 		if (!fn)
1376 			return 0;
1377 
1378 		if (w->prune && fn != w->root &&
1379 		    fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1380 			w->state = FWS_C;
1381 			w->leaf = fn->leaf;
1382 		}
1383 		switch (w->state) {
1384 #ifdef CONFIG_IPV6_SUBTREES
1385 		case FWS_S:
1386 			if (FIB6_SUBTREE(fn)) {
1387 				w->node = FIB6_SUBTREE(fn);
1388 				continue;
1389 			}
1390 			w->state = FWS_L;
1391 #endif
1392 		case FWS_L:
1393 			if (fn->left) {
1394 				w->node = fn->left;
1395 				w->state = FWS_INIT;
1396 				continue;
1397 			}
1398 			w->state = FWS_R;
1399 		case FWS_R:
1400 			if (fn->right) {
1401 				w->node = fn->right;
1402 				w->state = FWS_INIT;
1403 				continue;
1404 			}
1405 			w->state = FWS_C;
1406 			w->leaf = fn->leaf;
1407 		case FWS_C:
1408 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1409 				int err;
1410 
1411 				if (w->skip) {
1412 					w->skip--;
1413 					continue;
1414 				}
1415 
1416 				err = w->func(w);
1417 				if (err)
1418 					return err;
1419 
1420 				w->count++;
1421 				continue;
1422 			}
1423 			w->state = FWS_U;
1424 		case FWS_U:
1425 			if (fn == w->root)
1426 				return 0;
1427 			pn = fn->parent;
1428 			w->node = pn;
1429 #ifdef CONFIG_IPV6_SUBTREES
1430 			if (FIB6_SUBTREE(pn) == fn) {
1431 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1432 				w->state = FWS_L;
1433 				continue;
1434 			}
1435 #endif
1436 			if (pn->left == fn) {
1437 				w->state = FWS_R;
1438 				continue;
1439 			}
1440 			if (pn->right == fn) {
1441 				w->state = FWS_C;
1442 				w->leaf = w->node->leaf;
1443 				continue;
1444 			}
1445 #if RT6_DEBUG >= 2
1446 			WARN_ON(1);
1447 #endif
1448 		}
1449 	}
1450 }
1451 
1452 static int fib6_walk(struct fib6_walker_t *w)
1453 {
1454 	int res;
1455 
1456 	w->state = FWS_INIT;
1457 	w->node = w->root;
1458 
1459 	fib6_walker_link(w);
1460 	res = fib6_walk_continue(w);
1461 	if (res <= 0)
1462 		fib6_walker_unlink(w);
1463 	return res;
1464 }
1465 
1466 static int fib6_clean_node(struct fib6_walker_t *w)
1467 {
1468 	int res;
1469 	struct rt6_info *rt;
1470 	struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1471 	struct nl_info info = {
1472 		.nl_net = c->net,
1473 	};
1474 
1475 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1476 		res = c->func(rt, c->arg);
1477 		if (res < 0) {
1478 			w->leaf = rt;
1479 			res = fib6_del(rt, &info);
1480 			if (res) {
1481 #if RT6_DEBUG >= 2
1482 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1483 					 __func__, rt, rt->rt6i_node, res);
1484 #endif
1485 				continue;
1486 			}
1487 			return 0;
1488 		}
1489 		WARN_ON(res != 0);
1490 	}
1491 	w->leaf = rt;
1492 	return 0;
1493 }
1494 
1495 /*
1496  *	Convenient frontend to tree walker.
1497  *
1498  *	func is called on each route.
1499  *		It may return -1 -> delete this route.
1500  *		              0  -> continue walking
1501  *
1502  *	prune==1 -> only immediate children of node (certainly,
1503  *	ignoring pure split nodes) will be scanned.
1504  */
1505 
1506 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1507 			    int (*func)(struct rt6_info *, void *arg),
1508 			    int prune, void *arg)
1509 {
1510 	struct fib6_cleaner_t c;
1511 
1512 	c.w.root = root;
1513 	c.w.func = fib6_clean_node;
1514 	c.w.prune = prune;
1515 	c.w.count = 0;
1516 	c.w.skip = 0;
1517 	c.func = func;
1518 	c.arg = arg;
1519 	c.net = net;
1520 
1521 	fib6_walk(&c.w);
1522 }
1523 
1524 void fib6_clean_all_ro(struct net *net, int (*func)(struct rt6_info *, void *arg),
1525 		    int prune, void *arg)
1526 {
1527 	struct fib6_table *table;
1528 	struct hlist_head *head;
1529 	unsigned int h;
1530 
1531 	rcu_read_lock();
1532 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1533 		head = &net->ipv6.fib_table_hash[h];
1534 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1535 			read_lock_bh(&table->tb6_lock);
1536 			fib6_clean_tree(net, &table->tb6_root,
1537 					func, prune, arg);
1538 			read_unlock_bh(&table->tb6_lock);
1539 		}
1540 	}
1541 	rcu_read_unlock();
1542 }
1543 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1544 		    int prune, void *arg)
1545 {
1546 	struct fib6_table *table;
1547 	struct hlist_head *head;
1548 	unsigned int h;
1549 
1550 	rcu_read_lock();
1551 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1552 		head = &net->ipv6.fib_table_hash[h];
1553 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1554 			write_lock_bh(&table->tb6_lock);
1555 			fib6_clean_tree(net, &table->tb6_root,
1556 					func, prune, arg);
1557 			write_unlock_bh(&table->tb6_lock);
1558 		}
1559 	}
1560 	rcu_read_unlock();
1561 }
1562 
1563 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1564 {
1565 	if (rt->rt6i_flags & RTF_CACHE) {
1566 		RT6_TRACE("pruning clone %p\n", rt);
1567 		return -1;
1568 	}
1569 
1570 	return 0;
1571 }
1572 
1573 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1574 			      struct rt6_info *rt)
1575 {
1576 	fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1577 }
1578 
1579 /*
1580  *	Garbage collection
1581  */
1582 
1583 static struct fib6_gc_args
1584 {
1585 	int			timeout;
1586 	int			more;
1587 } gc_args;
1588 
1589 static int fib6_age(struct rt6_info *rt, void *arg)
1590 {
1591 	unsigned long now = jiffies;
1592 
1593 	/*
1594 	 *	check addrconf expiration here.
1595 	 *	Routes are expired even if they are in use.
1596 	 *
1597 	 *	Also age clones. Note, that clones are aged out
1598 	 *	only if they are not in use now.
1599 	 */
1600 
1601 	if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1602 		if (time_after(now, rt->dst.expires)) {
1603 			RT6_TRACE("expiring %p\n", rt);
1604 			return -1;
1605 		}
1606 		gc_args.more++;
1607 	} else if (rt->rt6i_flags & RTF_CACHE) {
1608 		if (atomic_read(&rt->dst.__refcnt) == 0 &&
1609 		    time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1610 			RT6_TRACE("aging clone %p\n", rt);
1611 			return -1;
1612 		} else if (rt->rt6i_flags & RTF_GATEWAY) {
1613 			struct neighbour *neigh;
1614 			__u8 neigh_flags = 0;
1615 
1616 			neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1617 			if (neigh) {
1618 				neigh_flags = neigh->flags;
1619 				neigh_release(neigh);
1620 			}
1621 			if (!(neigh_flags & NTF_ROUTER)) {
1622 				RT6_TRACE("purging route %p via non-router but gateway\n",
1623 					  rt);
1624 				return -1;
1625 			}
1626 		}
1627 		gc_args.more++;
1628 	}
1629 
1630 	return 0;
1631 }
1632 
1633 static DEFINE_SPINLOCK(fib6_gc_lock);
1634 
1635 void fib6_run_gc(unsigned long expires, struct net *net)
1636 {
1637 	if (expires != ~0UL) {
1638 		spin_lock_bh(&fib6_gc_lock);
1639 		gc_args.timeout = expires ? (int)expires :
1640 			net->ipv6.sysctl.ip6_rt_gc_interval;
1641 	} else {
1642 		if (!spin_trylock_bh(&fib6_gc_lock)) {
1643 			mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1644 			return;
1645 		}
1646 		gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1647 	}
1648 
1649 	gc_args.more = icmp6_dst_gc();
1650 
1651 	fib6_clean_all(net, fib6_age, 0, NULL);
1652 
1653 	if (gc_args.more)
1654 		mod_timer(&net->ipv6.ip6_fib_timer,
1655 			  round_jiffies(jiffies
1656 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
1657 	else
1658 		del_timer(&net->ipv6.ip6_fib_timer);
1659 	spin_unlock_bh(&fib6_gc_lock);
1660 }
1661 
1662 static void fib6_gc_timer_cb(unsigned long arg)
1663 {
1664 	fib6_run_gc(0, (struct net *)arg);
1665 }
1666 
1667 static int __net_init fib6_net_init(struct net *net)
1668 {
1669 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1670 
1671 	setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1672 
1673 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1674 	if (!net->ipv6.rt6_stats)
1675 		goto out_timer;
1676 
1677 	/* Avoid false sharing : Use at least a full cache line */
1678 	size = max_t(size_t, size, L1_CACHE_BYTES);
1679 
1680 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1681 	if (!net->ipv6.fib_table_hash)
1682 		goto out_rt6_stats;
1683 
1684 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1685 					  GFP_KERNEL);
1686 	if (!net->ipv6.fib6_main_tbl)
1687 		goto out_fib_table_hash;
1688 
1689 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1690 	net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1691 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1692 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1693 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1694 
1695 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1696 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1697 					   GFP_KERNEL);
1698 	if (!net->ipv6.fib6_local_tbl)
1699 		goto out_fib6_main_tbl;
1700 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1701 	net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1702 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1703 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1704 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1705 #endif
1706 	fib6_tables_init(net);
1707 
1708 	return 0;
1709 
1710 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1711 out_fib6_main_tbl:
1712 	kfree(net->ipv6.fib6_main_tbl);
1713 #endif
1714 out_fib_table_hash:
1715 	kfree(net->ipv6.fib_table_hash);
1716 out_rt6_stats:
1717 	kfree(net->ipv6.rt6_stats);
1718 out_timer:
1719 	return -ENOMEM;
1720  }
1721 
1722 static void fib6_net_exit(struct net *net)
1723 {
1724 	rt6_ifdown(net, NULL);
1725 	del_timer_sync(&net->ipv6.ip6_fib_timer);
1726 
1727 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1728 	inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1729 	kfree(net->ipv6.fib6_local_tbl);
1730 #endif
1731 	inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1732 	kfree(net->ipv6.fib6_main_tbl);
1733 	kfree(net->ipv6.fib_table_hash);
1734 	kfree(net->ipv6.rt6_stats);
1735 }
1736 
1737 static struct pernet_operations fib6_net_ops = {
1738 	.init = fib6_net_init,
1739 	.exit = fib6_net_exit,
1740 };
1741 
1742 int __init fib6_init(void)
1743 {
1744 	int ret = -ENOMEM;
1745 
1746 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
1747 					   sizeof(struct fib6_node),
1748 					   0, SLAB_HWCACHE_ALIGN,
1749 					   NULL);
1750 	if (!fib6_node_kmem)
1751 		goto out;
1752 
1753 	ret = register_pernet_subsys(&fib6_net_ops);
1754 	if (ret)
1755 		goto out_kmem_cache_create;
1756 
1757 	ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1758 			      NULL);
1759 	if (ret)
1760 		goto out_unregister_subsys;
1761 out:
1762 	return ret;
1763 
1764 out_unregister_subsys:
1765 	unregister_pernet_subsys(&fib6_net_ops);
1766 out_kmem_cache_create:
1767 	kmem_cache_destroy(fib6_node_kmem);
1768 	goto out;
1769 }
1770 
1771 void fib6_gc_cleanup(void)
1772 {
1773 	unregister_pernet_subsys(&fib6_net_ops);
1774 	kmem_cache_destroy(fib6_node_kmem);
1775 }
1776