xref: /linux/net/ipv6/ip6_fib.c (revision a4f174dee4ae842e07cab7eeec194a3e60925c8d)
1 /*
2  *	Linux INET6 implementation
3  *	Forwarding Information Database
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	This program is free software; you can redistribute it and/or
9  *      modify it under the terms of the GNU General Public License
10  *      as published by the Free Software Foundation; either version
11  *      2 of the License, or (at your option) any later version.
12  *
13  *	Changes:
14  *	Yuji SEKIYA @USAGI:	Support default route on router node;
15  *				remove ip6_null_entry from the top of
16  *				routing table.
17  *	Ville Nuorvala:		Fixed routing subtrees.
18  */
19 
20 #define pr_fmt(fmt) "IPv6: " fmt
21 
22 #include <linux/errno.h>
23 #include <linux/types.h>
24 #include <linux/net.h>
25 #include <linux/route.h>
26 #include <linux/netdevice.h>
27 #include <linux/in6.h>
28 #include <linux/init.h>
29 #include <linux/list.h>
30 #include <linux/slab.h>
31 
32 #include <net/ipv6.h>
33 #include <net/ndisc.h>
34 #include <net/addrconf.h>
35 
36 #include <net/ip6_fib.h>
37 #include <net/ip6_route.h>
38 
39 #define RT6_DEBUG 2
40 
41 #if RT6_DEBUG >= 3
42 #define RT6_TRACE(x...) pr_debug(x)
43 #else
44 #define RT6_TRACE(x...) do { ; } while (0)
45 #endif
46 
47 static struct kmem_cache *fib6_node_kmem __read_mostly;
48 
49 struct fib6_cleaner {
50 	struct fib6_walker w;
51 	struct net *net;
52 	int (*func)(struct rt6_info *, void *arg);
53 	int sernum;
54 	void *arg;
55 };
56 
57 static DEFINE_RWLOCK(fib6_walker_lock);
58 
59 #ifdef CONFIG_IPV6_SUBTREES
60 #define FWS_INIT FWS_S
61 #else
62 #define FWS_INIT FWS_L
63 #endif
64 
65 static void fib6_prune_clones(struct net *net, struct fib6_node *fn);
66 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
67 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
68 static int fib6_walk(struct fib6_walker *w);
69 static int fib6_walk_continue(struct fib6_walker *w);
70 
71 /*
72  *	A routing update causes an increase of the serial number on the
73  *	affected subtree. This allows for cached routes to be asynchronously
74  *	tested when modifications are made to the destination cache as a
75  *	result of redirects, path MTU changes, etc.
76  */
77 
78 static void fib6_gc_timer_cb(unsigned long arg);
79 
80 static LIST_HEAD(fib6_walkers);
81 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
82 
83 static void fib6_walker_link(struct fib6_walker *w)
84 {
85 	write_lock_bh(&fib6_walker_lock);
86 	list_add(&w->lh, &fib6_walkers);
87 	write_unlock_bh(&fib6_walker_lock);
88 }
89 
90 static void fib6_walker_unlink(struct fib6_walker *w)
91 {
92 	write_lock_bh(&fib6_walker_lock);
93 	list_del(&w->lh);
94 	write_unlock_bh(&fib6_walker_lock);
95 }
96 
97 static int fib6_new_sernum(struct net *net)
98 {
99 	int new, old;
100 
101 	do {
102 		old = atomic_read(&net->ipv6.fib6_sernum);
103 		new = old < INT_MAX ? old + 1 : 1;
104 	} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
105 				old, new) != old);
106 	return new;
107 }
108 
109 enum {
110 	FIB6_NO_SERNUM_CHANGE = 0,
111 };
112 
113 /*
114  *	Auxiliary address test functions for the radix tree.
115  *
116  *	These assume a 32bit processor (although it will work on
117  *	64bit processors)
118  */
119 
120 /*
121  *	test bit
122  */
123 #if defined(__LITTLE_ENDIAN)
124 # define BITOP_BE32_SWIZZLE	(0x1F & ~7)
125 #else
126 # define BITOP_BE32_SWIZZLE	0
127 #endif
128 
129 static __be32 addr_bit_set(const void *token, int fn_bit)
130 {
131 	const __be32 *addr = token;
132 	/*
133 	 * Here,
134 	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
135 	 * is optimized version of
136 	 *	htonl(1 << ((~fn_bit)&0x1F))
137 	 * See include/asm-generic/bitops/le.h.
138 	 */
139 	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
140 	       addr[fn_bit >> 5];
141 }
142 
143 static struct fib6_node *node_alloc(void)
144 {
145 	struct fib6_node *fn;
146 
147 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
148 
149 	return fn;
150 }
151 
152 static void node_free(struct fib6_node *fn)
153 {
154 	kmem_cache_free(fib6_node_kmem, fn);
155 }
156 
157 static void rt6_release(struct rt6_info *rt)
158 {
159 	if (atomic_dec_and_test(&rt->rt6i_ref))
160 		dst_free(&rt->dst);
161 }
162 
163 static void fib6_link_table(struct net *net, struct fib6_table *tb)
164 {
165 	unsigned int h;
166 
167 	/*
168 	 * Initialize table lock at a single place to give lockdep a key,
169 	 * tables aren't visible prior to being linked to the list.
170 	 */
171 	rwlock_init(&tb->tb6_lock);
172 
173 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
174 
175 	/*
176 	 * No protection necessary, this is the only list mutatation
177 	 * operation, tables never disappear once they exist.
178 	 */
179 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
180 }
181 
182 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
183 
184 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
185 {
186 	struct fib6_table *table;
187 
188 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
189 	if (table) {
190 		table->tb6_id = id;
191 		table->tb6_root.leaf = net->ipv6.ip6_null_entry;
192 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
193 		inet_peer_base_init(&table->tb6_peers);
194 	}
195 
196 	return table;
197 }
198 
199 struct fib6_table *fib6_new_table(struct net *net, u32 id)
200 {
201 	struct fib6_table *tb;
202 
203 	if (id == 0)
204 		id = RT6_TABLE_MAIN;
205 	tb = fib6_get_table(net, id);
206 	if (tb)
207 		return tb;
208 
209 	tb = fib6_alloc_table(net, id);
210 	if (tb)
211 		fib6_link_table(net, tb);
212 
213 	return tb;
214 }
215 
216 struct fib6_table *fib6_get_table(struct net *net, u32 id)
217 {
218 	struct fib6_table *tb;
219 	struct hlist_head *head;
220 	unsigned int h;
221 
222 	if (id == 0)
223 		id = RT6_TABLE_MAIN;
224 	h = id & (FIB6_TABLE_HASHSZ - 1);
225 	rcu_read_lock();
226 	head = &net->ipv6.fib_table_hash[h];
227 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
228 		if (tb->tb6_id == id) {
229 			rcu_read_unlock();
230 			return tb;
231 		}
232 	}
233 	rcu_read_unlock();
234 
235 	return NULL;
236 }
237 
238 static void __net_init fib6_tables_init(struct net *net)
239 {
240 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
241 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
242 }
243 #else
244 
245 struct fib6_table *fib6_new_table(struct net *net, u32 id)
246 {
247 	return fib6_get_table(net, id);
248 }
249 
250 struct fib6_table *fib6_get_table(struct net *net, u32 id)
251 {
252 	  return net->ipv6.fib6_main_tbl;
253 }
254 
255 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
256 				   int flags, pol_lookup_t lookup)
257 {
258 	return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
259 }
260 
261 static void __net_init fib6_tables_init(struct net *net)
262 {
263 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
264 }
265 
266 #endif
267 
268 static int fib6_dump_node(struct fib6_walker *w)
269 {
270 	int res;
271 	struct rt6_info *rt;
272 
273 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
274 		res = rt6_dump_route(rt, w->args);
275 		if (res < 0) {
276 			/* Frame is full, suspend walking */
277 			w->leaf = rt;
278 			return 1;
279 		}
280 		WARN_ON(res == 0);
281 	}
282 	w->leaf = NULL;
283 	return 0;
284 }
285 
286 static void fib6_dump_end(struct netlink_callback *cb)
287 {
288 	struct fib6_walker *w = (void *)cb->args[2];
289 
290 	if (w) {
291 		if (cb->args[4]) {
292 			cb->args[4] = 0;
293 			fib6_walker_unlink(w);
294 		}
295 		cb->args[2] = 0;
296 		kfree(w);
297 	}
298 	cb->done = (void *)cb->args[3];
299 	cb->args[1] = 3;
300 }
301 
302 static int fib6_dump_done(struct netlink_callback *cb)
303 {
304 	fib6_dump_end(cb);
305 	return cb->done ? cb->done(cb) : 0;
306 }
307 
308 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
309 			   struct netlink_callback *cb)
310 {
311 	struct fib6_walker *w;
312 	int res;
313 
314 	w = (void *)cb->args[2];
315 	w->root = &table->tb6_root;
316 
317 	if (cb->args[4] == 0) {
318 		w->count = 0;
319 		w->skip = 0;
320 
321 		read_lock_bh(&table->tb6_lock);
322 		res = fib6_walk(w);
323 		read_unlock_bh(&table->tb6_lock);
324 		if (res > 0) {
325 			cb->args[4] = 1;
326 			cb->args[5] = w->root->fn_sernum;
327 		}
328 	} else {
329 		if (cb->args[5] != w->root->fn_sernum) {
330 			/* Begin at the root if the tree changed */
331 			cb->args[5] = w->root->fn_sernum;
332 			w->state = FWS_INIT;
333 			w->node = w->root;
334 			w->skip = w->count;
335 		} else
336 			w->skip = 0;
337 
338 		read_lock_bh(&table->tb6_lock);
339 		res = fib6_walk_continue(w);
340 		read_unlock_bh(&table->tb6_lock);
341 		if (res <= 0) {
342 			fib6_walker_unlink(w);
343 			cb->args[4] = 0;
344 		}
345 	}
346 
347 	return res;
348 }
349 
350 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
351 {
352 	struct net *net = sock_net(skb->sk);
353 	unsigned int h, s_h;
354 	unsigned int e = 0, s_e;
355 	struct rt6_rtnl_dump_arg arg;
356 	struct fib6_walker *w;
357 	struct fib6_table *tb;
358 	struct hlist_head *head;
359 	int res = 0;
360 
361 	s_h = cb->args[0];
362 	s_e = cb->args[1];
363 
364 	w = (void *)cb->args[2];
365 	if (!w) {
366 		/* New dump:
367 		 *
368 		 * 1. hook callback destructor.
369 		 */
370 		cb->args[3] = (long)cb->done;
371 		cb->done = fib6_dump_done;
372 
373 		/*
374 		 * 2. allocate and initialize walker.
375 		 */
376 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
377 		if (!w)
378 			return -ENOMEM;
379 		w->func = fib6_dump_node;
380 		cb->args[2] = (long)w;
381 	}
382 
383 	arg.skb = skb;
384 	arg.cb = cb;
385 	arg.net = net;
386 	w->args = &arg;
387 
388 	rcu_read_lock();
389 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
390 		e = 0;
391 		head = &net->ipv6.fib_table_hash[h];
392 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
393 			if (e < s_e)
394 				goto next;
395 			res = fib6_dump_table(tb, skb, cb);
396 			if (res != 0)
397 				goto out;
398 next:
399 			e++;
400 		}
401 	}
402 out:
403 	rcu_read_unlock();
404 	cb->args[1] = e;
405 	cb->args[0] = h;
406 
407 	res = res < 0 ? res : skb->len;
408 	if (res <= 0)
409 		fib6_dump_end(cb);
410 	return res;
411 }
412 
413 /*
414  *	Routing Table
415  *
416  *	return the appropriate node for a routing tree "add" operation
417  *	by either creating and inserting or by returning an existing
418  *	node.
419  */
420 
421 static struct fib6_node *fib6_add_1(struct fib6_node *root,
422 				     struct in6_addr *addr, int plen,
423 				     int offset, int allow_create,
424 				     int replace_required, int sernum)
425 {
426 	struct fib6_node *fn, *in, *ln;
427 	struct fib6_node *pn = NULL;
428 	struct rt6key *key;
429 	int	bit;
430 	__be32	dir = 0;
431 
432 	RT6_TRACE("fib6_add_1\n");
433 
434 	/* insert node in tree */
435 
436 	fn = root;
437 
438 	do {
439 		key = (struct rt6key *)((u8 *)fn->leaf + offset);
440 
441 		/*
442 		 *	Prefix match
443 		 */
444 		if (plen < fn->fn_bit ||
445 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
446 			if (!allow_create) {
447 				if (replace_required) {
448 					pr_warn("Can't replace route, no match found\n");
449 					return ERR_PTR(-ENOENT);
450 				}
451 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
452 			}
453 			goto insert_above;
454 		}
455 
456 		/*
457 		 *	Exact match ?
458 		 */
459 
460 		if (plen == fn->fn_bit) {
461 			/* clean up an intermediate node */
462 			if (!(fn->fn_flags & RTN_RTINFO)) {
463 				rt6_release(fn->leaf);
464 				fn->leaf = NULL;
465 			}
466 
467 			fn->fn_sernum = sernum;
468 
469 			return fn;
470 		}
471 
472 		/*
473 		 *	We have more bits to go
474 		 */
475 
476 		/* Try to walk down on tree. */
477 		fn->fn_sernum = sernum;
478 		dir = addr_bit_set(addr, fn->fn_bit);
479 		pn = fn;
480 		fn = dir ? fn->right : fn->left;
481 	} while (fn);
482 
483 	if (!allow_create) {
484 		/* We should not create new node because
485 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
486 		 * I assume it is safe to require NLM_F_CREATE when
487 		 * REPLACE flag is used! Later we may want to remove the
488 		 * check for replace_required, because according
489 		 * to netlink specification, NLM_F_CREATE
490 		 * MUST be specified if new route is created.
491 		 * That would keep IPv6 consistent with IPv4
492 		 */
493 		if (replace_required) {
494 			pr_warn("Can't replace route, no match found\n");
495 			return ERR_PTR(-ENOENT);
496 		}
497 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
498 	}
499 	/*
500 	 *	We walked to the bottom of tree.
501 	 *	Create new leaf node without children.
502 	 */
503 
504 	ln = node_alloc();
505 
506 	if (!ln)
507 		return ERR_PTR(-ENOMEM);
508 	ln->fn_bit = plen;
509 
510 	ln->parent = pn;
511 	ln->fn_sernum = sernum;
512 
513 	if (dir)
514 		pn->right = ln;
515 	else
516 		pn->left  = ln;
517 
518 	return ln;
519 
520 
521 insert_above:
522 	/*
523 	 * split since we don't have a common prefix anymore or
524 	 * we have a less significant route.
525 	 * we've to insert an intermediate node on the list
526 	 * this new node will point to the one we need to create
527 	 * and the current
528 	 */
529 
530 	pn = fn->parent;
531 
532 	/* find 1st bit in difference between the 2 addrs.
533 
534 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
535 	   but if it is >= plen, the value is ignored in any case.
536 	 */
537 
538 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
539 
540 	/*
541 	 *		(intermediate)[in]
542 	 *	          /	   \
543 	 *	(new leaf node)[ln] (old node)[fn]
544 	 */
545 	if (plen > bit) {
546 		in = node_alloc();
547 		ln = node_alloc();
548 
549 		if (!in || !ln) {
550 			if (in)
551 				node_free(in);
552 			if (ln)
553 				node_free(ln);
554 			return ERR_PTR(-ENOMEM);
555 		}
556 
557 		/*
558 		 * new intermediate node.
559 		 * RTN_RTINFO will
560 		 * be off since that an address that chooses one of
561 		 * the branches would not match less specific routes
562 		 * in the other branch
563 		 */
564 
565 		in->fn_bit = bit;
566 
567 		in->parent = pn;
568 		in->leaf = fn->leaf;
569 		atomic_inc(&in->leaf->rt6i_ref);
570 
571 		in->fn_sernum = sernum;
572 
573 		/* update parent pointer */
574 		if (dir)
575 			pn->right = in;
576 		else
577 			pn->left  = in;
578 
579 		ln->fn_bit = plen;
580 
581 		ln->parent = in;
582 		fn->parent = in;
583 
584 		ln->fn_sernum = sernum;
585 
586 		if (addr_bit_set(addr, bit)) {
587 			in->right = ln;
588 			in->left  = fn;
589 		} else {
590 			in->left  = ln;
591 			in->right = fn;
592 		}
593 	} else { /* plen <= bit */
594 
595 		/*
596 		 *		(new leaf node)[ln]
597 		 *	          /	   \
598 		 *	     (old node)[fn] NULL
599 		 */
600 
601 		ln = node_alloc();
602 
603 		if (!ln)
604 			return ERR_PTR(-ENOMEM);
605 
606 		ln->fn_bit = plen;
607 
608 		ln->parent = pn;
609 
610 		ln->fn_sernum = sernum;
611 
612 		if (dir)
613 			pn->right = ln;
614 		else
615 			pn->left  = ln;
616 
617 		if (addr_bit_set(&key->addr, plen))
618 			ln->right = fn;
619 		else
620 			ln->left  = fn;
621 
622 		fn->parent = ln;
623 	}
624 	return ln;
625 }
626 
627 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
628 {
629 	return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
630 	       RTF_GATEWAY;
631 }
632 
633 static int fib6_commit_metrics(struct dst_entry *dst,
634 			       struct nlattr *mx, int mx_len)
635 {
636 	struct nlattr *nla;
637 	int remaining;
638 	u32 *mp;
639 
640 	if (dst->flags & DST_HOST) {
641 		mp = dst_metrics_write_ptr(dst);
642 	} else {
643 		mp = kzalloc(sizeof(u32) * RTAX_MAX, GFP_ATOMIC);
644 		if (!mp)
645 			return -ENOMEM;
646 		dst_init_metrics(dst, mp, 0);
647 	}
648 
649 	nla_for_each_attr(nla, mx, mx_len, remaining) {
650 		int type = nla_type(nla);
651 
652 		if (type) {
653 			if (type > RTAX_MAX)
654 				return -EINVAL;
655 
656 			mp[type - 1] = nla_get_u32(nla);
657 		}
658 	}
659 	return 0;
660 }
661 
662 /*
663  *	Insert routing information in a node.
664  */
665 
666 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
667 			    struct nl_info *info, struct nlattr *mx, int mx_len)
668 {
669 	struct rt6_info *iter = NULL;
670 	struct rt6_info **ins;
671 	int replace = (info->nlh &&
672 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
673 	int add = (!info->nlh ||
674 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
675 	int found = 0;
676 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
677 	int err;
678 
679 	ins = &fn->leaf;
680 
681 	for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
682 		/*
683 		 *	Search for duplicates
684 		 */
685 
686 		if (iter->rt6i_metric == rt->rt6i_metric) {
687 			/*
688 			 *	Same priority level
689 			 */
690 			if (info->nlh &&
691 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
692 				return -EEXIST;
693 			if (replace) {
694 				found++;
695 				break;
696 			}
697 
698 			if (iter->dst.dev == rt->dst.dev &&
699 			    iter->rt6i_idev == rt->rt6i_idev &&
700 			    ipv6_addr_equal(&iter->rt6i_gateway,
701 					    &rt->rt6i_gateway)) {
702 				if (rt->rt6i_nsiblings)
703 					rt->rt6i_nsiblings = 0;
704 				if (!(iter->rt6i_flags & RTF_EXPIRES))
705 					return -EEXIST;
706 				if (!(rt->rt6i_flags & RTF_EXPIRES))
707 					rt6_clean_expires(iter);
708 				else
709 					rt6_set_expires(iter, rt->dst.expires);
710 				return -EEXIST;
711 			}
712 			/* If we have the same destination and the same metric,
713 			 * but not the same gateway, then the route we try to
714 			 * add is sibling to this route, increment our counter
715 			 * of siblings, and later we will add our route to the
716 			 * list.
717 			 * Only static routes (which don't have flag
718 			 * RTF_EXPIRES) are used for ECMPv6.
719 			 *
720 			 * To avoid long list, we only had siblings if the
721 			 * route have a gateway.
722 			 */
723 			if (rt_can_ecmp &&
724 			    rt6_qualify_for_ecmp(iter))
725 				rt->rt6i_nsiblings++;
726 		}
727 
728 		if (iter->rt6i_metric > rt->rt6i_metric)
729 			break;
730 
731 		ins = &iter->dst.rt6_next;
732 	}
733 
734 	/* Reset round-robin state, if necessary */
735 	if (ins == &fn->leaf)
736 		fn->rr_ptr = NULL;
737 
738 	/* Link this route to others same route. */
739 	if (rt->rt6i_nsiblings) {
740 		unsigned int rt6i_nsiblings;
741 		struct rt6_info *sibling, *temp_sibling;
742 
743 		/* Find the first route that have the same metric */
744 		sibling = fn->leaf;
745 		while (sibling) {
746 			if (sibling->rt6i_metric == rt->rt6i_metric &&
747 			    rt6_qualify_for_ecmp(sibling)) {
748 				list_add_tail(&rt->rt6i_siblings,
749 					      &sibling->rt6i_siblings);
750 				break;
751 			}
752 			sibling = sibling->dst.rt6_next;
753 		}
754 		/* For each sibling in the list, increment the counter of
755 		 * siblings. BUG() if counters does not match, list of siblings
756 		 * is broken!
757 		 */
758 		rt6i_nsiblings = 0;
759 		list_for_each_entry_safe(sibling, temp_sibling,
760 					 &rt->rt6i_siblings, rt6i_siblings) {
761 			sibling->rt6i_nsiblings++;
762 			BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
763 			rt6i_nsiblings++;
764 		}
765 		BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
766 	}
767 
768 	/*
769 	 *	insert node
770 	 */
771 	if (!replace) {
772 		if (!add)
773 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
774 
775 add:
776 		if (mx) {
777 			err = fib6_commit_metrics(&rt->dst, mx, mx_len);
778 			if (err)
779 				return err;
780 		}
781 		rt->dst.rt6_next = iter;
782 		*ins = rt;
783 		rt->rt6i_node = fn;
784 		atomic_inc(&rt->rt6i_ref);
785 		inet6_rt_notify(RTM_NEWROUTE, rt, info);
786 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
787 
788 		if (!(fn->fn_flags & RTN_RTINFO)) {
789 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
790 			fn->fn_flags |= RTN_RTINFO;
791 		}
792 
793 	} else {
794 		if (!found) {
795 			if (add)
796 				goto add;
797 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
798 			return -ENOENT;
799 		}
800 		if (mx) {
801 			err = fib6_commit_metrics(&rt->dst, mx, mx_len);
802 			if (err)
803 				return err;
804 		}
805 		*ins = rt;
806 		rt->rt6i_node = fn;
807 		rt->dst.rt6_next = iter->dst.rt6_next;
808 		atomic_inc(&rt->rt6i_ref);
809 		inet6_rt_notify(RTM_NEWROUTE, rt, info);
810 		rt6_release(iter);
811 		if (!(fn->fn_flags & RTN_RTINFO)) {
812 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
813 			fn->fn_flags |= RTN_RTINFO;
814 		}
815 	}
816 
817 	return 0;
818 }
819 
820 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
821 {
822 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
823 	    (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
824 		mod_timer(&net->ipv6.ip6_fib_timer,
825 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
826 }
827 
828 void fib6_force_start_gc(struct net *net)
829 {
830 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
831 		mod_timer(&net->ipv6.ip6_fib_timer,
832 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
833 }
834 
835 /*
836  *	Add routing information to the routing tree.
837  *	<destination addr>/<source addr>
838  *	with source addr info in sub-trees
839  */
840 
841 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info,
842 	     struct nlattr *mx, int mx_len)
843 {
844 	struct fib6_node *fn, *pn = NULL;
845 	int err = -ENOMEM;
846 	int allow_create = 1;
847 	int replace_required = 0;
848 	int sernum = fib6_new_sernum(info->nl_net);
849 
850 	if (info->nlh) {
851 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
852 			allow_create = 0;
853 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
854 			replace_required = 1;
855 	}
856 	if (!allow_create && !replace_required)
857 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
858 
859 	fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
860 			offsetof(struct rt6_info, rt6i_dst), allow_create,
861 			replace_required, sernum);
862 	if (IS_ERR(fn)) {
863 		err = PTR_ERR(fn);
864 		fn = NULL;
865 		goto out;
866 	}
867 
868 	pn = fn;
869 
870 #ifdef CONFIG_IPV6_SUBTREES
871 	if (rt->rt6i_src.plen) {
872 		struct fib6_node *sn;
873 
874 		if (!fn->subtree) {
875 			struct fib6_node *sfn;
876 
877 			/*
878 			 * Create subtree.
879 			 *
880 			 *		fn[main tree]
881 			 *		|
882 			 *		sfn[subtree root]
883 			 *		   \
884 			 *		    sn[new leaf node]
885 			 */
886 
887 			/* Create subtree root node */
888 			sfn = node_alloc();
889 			if (!sfn)
890 				goto st_failure;
891 
892 			sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
893 			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
894 			sfn->fn_flags = RTN_ROOT;
895 			sfn->fn_sernum = sernum;
896 
897 			/* Now add the first leaf node to new subtree */
898 
899 			sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
900 					rt->rt6i_src.plen,
901 					offsetof(struct rt6_info, rt6i_src),
902 					allow_create, replace_required, sernum);
903 
904 			if (IS_ERR(sn)) {
905 				/* If it is failed, discard just allocated
906 				   root, and then (in st_failure) stale node
907 				   in main tree.
908 				 */
909 				node_free(sfn);
910 				err = PTR_ERR(sn);
911 				goto st_failure;
912 			}
913 
914 			/* Now link new subtree to main tree */
915 			sfn->parent = fn;
916 			fn->subtree = sfn;
917 		} else {
918 			sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
919 					rt->rt6i_src.plen,
920 					offsetof(struct rt6_info, rt6i_src),
921 					allow_create, replace_required, sernum);
922 
923 			if (IS_ERR(sn)) {
924 				err = PTR_ERR(sn);
925 				goto st_failure;
926 			}
927 		}
928 
929 		if (!fn->leaf) {
930 			fn->leaf = rt;
931 			atomic_inc(&rt->rt6i_ref);
932 		}
933 		fn = sn;
934 	}
935 #endif
936 
937 	err = fib6_add_rt2node(fn, rt, info, mx, mx_len);
938 	if (!err) {
939 		fib6_start_gc(info->nl_net, rt);
940 		if (!(rt->rt6i_flags & RTF_CACHE))
941 			fib6_prune_clones(info->nl_net, pn);
942 	}
943 
944 out:
945 	if (err) {
946 #ifdef CONFIG_IPV6_SUBTREES
947 		/*
948 		 * If fib6_add_1 has cleared the old leaf pointer in the
949 		 * super-tree leaf node we have to find a new one for it.
950 		 */
951 		if (pn != fn && pn->leaf == rt) {
952 			pn->leaf = NULL;
953 			atomic_dec(&rt->rt6i_ref);
954 		}
955 		if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
956 			pn->leaf = fib6_find_prefix(info->nl_net, pn);
957 #if RT6_DEBUG >= 2
958 			if (!pn->leaf) {
959 				WARN_ON(pn->leaf == NULL);
960 				pn->leaf = info->nl_net->ipv6.ip6_null_entry;
961 			}
962 #endif
963 			atomic_inc(&pn->leaf->rt6i_ref);
964 		}
965 #endif
966 		dst_free(&rt->dst);
967 	}
968 	return err;
969 
970 #ifdef CONFIG_IPV6_SUBTREES
971 	/* Subtree creation failed, probably main tree node
972 	   is orphan. If it is, shoot it.
973 	 */
974 st_failure:
975 	if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
976 		fib6_repair_tree(info->nl_net, fn);
977 	dst_free(&rt->dst);
978 	return err;
979 #endif
980 }
981 
982 /*
983  *	Routing tree lookup
984  *
985  */
986 
987 struct lookup_args {
988 	int			offset;		/* key offset on rt6_info	*/
989 	const struct in6_addr	*addr;		/* search key			*/
990 };
991 
992 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
993 				       struct lookup_args *args)
994 {
995 	struct fib6_node *fn;
996 	__be32 dir;
997 
998 	if (unlikely(args->offset == 0))
999 		return NULL;
1000 
1001 	/*
1002 	 *	Descend on a tree
1003 	 */
1004 
1005 	fn = root;
1006 
1007 	for (;;) {
1008 		struct fib6_node *next;
1009 
1010 		dir = addr_bit_set(args->addr, fn->fn_bit);
1011 
1012 		next = dir ? fn->right : fn->left;
1013 
1014 		if (next) {
1015 			fn = next;
1016 			continue;
1017 		}
1018 		break;
1019 	}
1020 
1021 	while (fn) {
1022 		if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1023 			struct rt6key *key;
1024 
1025 			key = (struct rt6key *) ((u8 *) fn->leaf +
1026 						 args->offset);
1027 
1028 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1029 #ifdef CONFIG_IPV6_SUBTREES
1030 				if (fn->subtree) {
1031 					struct fib6_node *sfn;
1032 					sfn = fib6_lookup_1(fn->subtree,
1033 							    args + 1);
1034 					if (!sfn)
1035 						goto backtrack;
1036 					fn = sfn;
1037 				}
1038 #endif
1039 				if (fn->fn_flags & RTN_RTINFO)
1040 					return fn;
1041 			}
1042 		}
1043 #ifdef CONFIG_IPV6_SUBTREES
1044 backtrack:
1045 #endif
1046 		if (fn->fn_flags & RTN_ROOT)
1047 			break;
1048 
1049 		fn = fn->parent;
1050 	}
1051 
1052 	return NULL;
1053 }
1054 
1055 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1056 			      const struct in6_addr *saddr)
1057 {
1058 	struct fib6_node *fn;
1059 	struct lookup_args args[] = {
1060 		{
1061 			.offset = offsetof(struct rt6_info, rt6i_dst),
1062 			.addr = daddr,
1063 		},
1064 #ifdef CONFIG_IPV6_SUBTREES
1065 		{
1066 			.offset = offsetof(struct rt6_info, rt6i_src),
1067 			.addr = saddr,
1068 		},
1069 #endif
1070 		{
1071 			.offset = 0,	/* sentinel */
1072 		}
1073 	};
1074 
1075 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
1076 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1077 		fn = root;
1078 
1079 	return fn;
1080 }
1081 
1082 /*
1083  *	Get node with specified destination prefix (and source prefix,
1084  *	if subtrees are used)
1085  */
1086 
1087 
1088 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1089 				       const struct in6_addr *addr,
1090 				       int plen, int offset)
1091 {
1092 	struct fib6_node *fn;
1093 
1094 	for (fn = root; fn ; ) {
1095 		struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1096 
1097 		/*
1098 		 *	Prefix match
1099 		 */
1100 		if (plen < fn->fn_bit ||
1101 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1102 			return NULL;
1103 
1104 		if (plen == fn->fn_bit)
1105 			return fn;
1106 
1107 		/*
1108 		 *	We have more bits to go
1109 		 */
1110 		if (addr_bit_set(addr, fn->fn_bit))
1111 			fn = fn->right;
1112 		else
1113 			fn = fn->left;
1114 	}
1115 	return NULL;
1116 }
1117 
1118 struct fib6_node *fib6_locate(struct fib6_node *root,
1119 			      const struct in6_addr *daddr, int dst_len,
1120 			      const struct in6_addr *saddr, int src_len)
1121 {
1122 	struct fib6_node *fn;
1123 
1124 	fn = fib6_locate_1(root, daddr, dst_len,
1125 			   offsetof(struct rt6_info, rt6i_dst));
1126 
1127 #ifdef CONFIG_IPV6_SUBTREES
1128 	if (src_len) {
1129 		WARN_ON(saddr == NULL);
1130 		if (fn && fn->subtree)
1131 			fn = fib6_locate_1(fn->subtree, saddr, src_len,
1132 					   offsetof(struct rt6_info, rt6i_src));
1133 	}
1134 #endif
1135 
1136 	if (fn && fn->fn_flags & RTN_RTINFO)
1137 		return fn;
1138 
1139 	return NULL;
1140 }
1141 
1142 
1143 /*
1144  *	Deletion
1145  *
1146  */
1147 
1148 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1149 {
1150 	if (fn->fn_flags & RTN_ROOT)
1151 		return net->ipv6.ip6_null_entry;
1152 
1153 	while (fn) {
1154 		if (fn->left)
1155 			return fn->left->leaf;
1156 		if (fn->right)
1157 			return fn->right->leaf;
1158 
1159 		fn = FIB6_SUBTREE(fn);
1160 	}
1161 	return NULL;
1162 }
1163 
1164 /*
1165  *	Called to trim the tree of intermediate nodes when possible. "fn"
1166  *	is the node we want to try and remove.
1167  */
1168 
1169 static struct fib6_node *fib6_repair_tree(struct net *net,
1170 					   struct fib6_node *fn)
1171 {
1172 	int children;
1173 	int nstate;
1174 	struct fib6_node *child, *pn;
1175 	struct fib6_walker *w;
1176 	int iter = 0;
1177 
1178 	for (;;) {
1179 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1180 		iter++;
1181 
1182 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1183 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1184 		WARN_ON(fn->leaf != NULL);
1185 
1186 		children = 0;
1187 		child = NULL;
1188 		if (fn->right)
1189 			child = fn->right, children |= 1;
1190 		if (fn->left)
1191 			child = fn->left, children |= 2;
1192 
1193 		if (children == 3 || FIB6_SUBTREE(fn)
1194 #ifdef CONFIG_IPV6_SUBTREES
1195 		    /* Subtree root (i.e. fn) may have one child */
1196 		    || (children && fn->fn_flags & RTN_ROOT)
1197 #endif
1198 		    ) {
1199 			fn->leaf = fib6_find_prefix(net, fn);
1200 #if RT6_DEBUG >= 2
1201 			if (!fn->leaf) {
1202 				WARN_ON(!fn->leaf);
1203 				fn->leaf = net->ipv6.ip6_null_entry;
1204 			}
1205 #endif
1206 			atomic_inc(&fn->leaf->rt6i_ref);
1207 			return fn->parent;
1208 		}
1209 
1210 		pn = fn->parent;
1211 #ifdef CONFIG_IPV6_SUBTREES
1212 		if (FIB6_SUBTREE(pn) == fn) {
1213 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1214 			FIB6_SUBTREE(pn) = NULL;
1215 			nstate = FWS_L;
1216 		} else {
1217 			WARN_ON(fn->fn_flags & RTN_ROOT);
1218 #endif
1219 			if (pn->right == fn)
1220 				pn->right = child;
1221 			else if (pn->left == fn)
1222 				pn->left = child;
1223 #if RT6_DEBUG >= 2
1224 			else
1225 				WARN_ON(1);
1226 #endif
1227 			if (child)
1228 				child->parent = pn;
1229 			nstate = FWS_R;
1230 #ifdef CONFIG_IPV6_SUBTREES
1231 		}
1232 #endif
1233 
1234 		read_lock(&fib6_walker_lock);
1235 		FOR_WALKERS(w) {
1236 			if (!child) {
1237 				if (w->root == fn) {
1238 					w->root = w->node = NULL;
1239 					RT6_TRACE("W %p adjusted by delroot 1\n", w);
1240 				} else if (w->node == fn) {
1241 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1242 					w->node = pn;
1243 					w->state = nstate;
1244 				}
1245 			} else {
1246 				if (w->root == fn) {
1247 					w->root = child;
1248 					RT6_TRACE("W %p adjusted by delroot 2\n", w);
1249 				}
1250 				if (w->node == fn) {
1251 					w->node = child;
1252 					if (children&2) {
1253 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1254 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1255 					} else {
1256 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1257 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1258 					}
1259 				}
1260 			}
1261 		}
1262 		read_unlock(&fib6_walker_lock);
1263 
1264 		node_free(fn);
1265 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1266 			return pn;
1267 
1268 		rt6_release(pn->leaf);
1269 		pn->leaf = NULL;
1270 		fn = pn;
1271 	}
1272 }
1273 
1274 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1275 			   struct nl_info *info)
1276 {
1277 	struct fib6_walker *w;
1278 	struct rt6_info *rt = *rtp;
1279 	struct net *net = info->nl_net;
1280 
1281 	RT6_TRACE("fib6_del_route\n");
1282 
1283 	/* Unlink it */
1284 	*rtp = rt->dst.rt6_next;
1285 	rt->rt6i_node = NULL;
1286 	net->ipv6.rt6_stats->fib_rt_entries--;
1287 	net->ipv6.rt6_stats->fib_discarded_routes++;
1288 
1289 	/* Reset round-robin state, if necessary */
1290 	if (fn->rr_ptr == rt)
1291 		fn->rr_ptr = NULL;
1292 
1293 	/* Remove this entry from other siblings */
1294 	if (rt->rt6i_nsiblings) {
1295 		struct rt6_info *sibling, *next_sibling;
1296 
1297 		list_for_each_entry_safe(sibling, next_sibling,
1298 					 &rt->rt6i_siblings, rt6i_siblings)
1299 			sibling->rt6i_nsiblings--;
1300 		rt->rt6i_nsiblings = 0;
1301 		list_del_init(&rt->rt6i_siblings);
1302 	}
1303 
1304 	/* Adjust walkers */
1305 	read_lock(&fib6_walker_lock);
1306 	FOR_WALKERS(w) {
1307 		if (w->state == FWS_C && w->leaf == rt) {
1308 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1309 			w->leaf = rt->dst.rt6_next;
1310 			if (!w->leaf)
1311 				w->state = FWS_U;
1312 		}
1313 	}
1314 	read_unlock(&fib6_walker_lock);
1315 
1316 	rt->dst.rt6_next = NULL;
1317 
1318 	/* If it was last route, expunge its radix tree node */
1319 	if (!fn->leaf) {
1320 		fn->fn_flags &= ~RTN_RTINFO;
1321 		net->ipv6.rt6_stats->fib_route_nodes--;
1322 		fn = fib6_repair_tree(net, fn);
1323 	}
1324 
1325 	if (atomic_read(&rt->rt6i_ref) != 1) {
1326 		/* This route is used as dummy address holder in some split
1327 		 * nodes. It is not leaked, but it still holds other resources,
1328 		 * which must be released in time. So, scan ascendant nodes
1329 		 * and replace dummy references to this route with references
1330 		 * to still alive ones.
1331 		 */
1332 		while (fn) {
1333 			if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
1334 				fn->leaf = fib6_find_prefix(net, fn);
1335 				atomic_inc(&fn->leaf->rt6i_ref);
1336 				rt6_release(rt);
1337 			}
1338 			fn = fn->parent;
1339 		}
1340 		/* No more references are possible at this point. */
1341 		BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1342 	}
1343 
1344 	inet6_rt_notify(RTM_DELROUTE, rt, info);
1345 	rt6_release(rt);
1346 }
1347 
1348 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1349 {
1350 	struct net *net = info->nl_net;
1351 	struct fib6_node *fn = rt->rt6i_node;
1352 	struct rt6_info **rtp;
1353 
1354 #if RT6_DEBUG >= 2
1355 	if (rt->dst.obsolete > 0) {
1356 		WARN_ON(fn != NULL);
1357 		return -ENOENT;
1358 	}
1359 #endif
1360 	if (!fn || rt == net->ipv6.ip6_null_entry)
1361 		return -ENOENT;
1362 
1363 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1364 
1365 	if (!(rt->rt6i_flags & RTF_CACHE)) {
1366 		struct fib6_node *pn = fn;
1367 #ifdef CONFIG_IPV6_SUBTREES
1368 		/* clones of this route might be in another subtree */
1369 		if (rt->rt6i_src.plen) {
1370 			while (!(pn->fn_flags & RTN_ROOT))
1371 				pn = pn->parent;
1372 			pn = pn->parent;
1373 		}
1374 #endif
1375 		fib6_prune_clones(info->nl_net, pn);
1376 	}
1377 
1378 	/*
1379 	 *	Walk the leaf entries looking for ourself
1380 	 */
1381 
1382 	for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1383 		if (*rtp == rt) {
1384 			fib6_del_route(fn, rtp, info);
1385 			return 0;
1386 		}
1387 	}
1388 	return -ENOENT;
1389 }
1390 
1391 /*
1392  *	Tree traversal function.
1393  *
1394  *	Certainly, it is not interrupt safe.
1395  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1396  *	It means, that we can modify tree during walking
1397  *	and use this function for garbage collection, clone pruning,
1398  *	cleaning tree when a device goes down etc. etc.
1399  *
1400  *	It guarantees that every node will be traversed,
1401  *	and that it will be traversed only once.
1402  *
1403  *	Callback function w->func may return:
1404  *	0 -> continue walking.
1405  *	positive value -> walking is suspended (used by tree dumps,
1406  *	and probably by gc, if it will be split to several slices)
1407  *	negative value -> terminate walking.
1408  *
1409  *	The function itself returns:
1410  *	0   -> walk is complete.
1411  *	>0  -> walk is incomplete (i.e. suspended)
1412  *	<0  -> walk is terminated by an error.
1413  */
1414 
1415 static int fib6_walk_continue(struct fib6_walker *w)
1416 {
1417 	struct fib6_node *fn, *pn;
1418 
1419 	for (;;) {
1420 		fn = w->node;
1421 		if (!fn)
1422 			return 0;
1423 
1424 		if (w->prune && fn != w->root &&
1425 		    fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1426 			w->state = FWS_C;
1427 			w->leaf = fn->leaf;
1428 		}
1429 		switch (w->state) {
1430 #ifdef CONFIG_IPV6_SUBTREES
1431 		case FWS_S:
1432 			if (FIB6_SUBTREE(fn)) {
1433 				w->node = FIB6_SUBTREE(fn);
1434 				continue;
1435 			}
1436 			w->state = FWS_L;
1437 #endif
1438 		case FWS_L:
1439 			if (fn->left) {
1440 				w->node = fn->left;
1441 				w->state = FWS_INIT;
1442 				continue;
1443 			}
1444 			w->state = FWS_R;
1445 		case FWS_R:
1446 			if (fn->right) {
1447 				w->node = fn->right;
1448 				w->state = FWS_INIT;
1449 				continue;
1450 			}
1451 			w->state = FWS_C;
1452 			w->leaf = fn->leaf;
1453 		case FWS_C:
1454 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1455 				int err;
1456 
1457 				if (w->skip) {
1458 					w->skip--;
1459 					goto skip;
1460 				}
1461 
1462 				err = w->func(w);
1463 				if (err)
1464 					return err;
1465 
1466 				w->count++;
1467 				continue;
1468 			}
1469 skip:
1470 			w->state = FWS_U;
1471 		case FWS_U:
1472 			if (fn == w->root)
1473 				return 0;
1474 			pn = fn->parent;
1475 			w->node = pn;
1476 #ifdef CONFIG_IPV6_SUBTREES
1477 			if (FIB6_SUBTREE(pn) == fn) {
1478 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1479 				w->state = FWS_L;
1480 				continue;
1481 			}
1482 #endif
1483 			if (pn->left == fn) {
1484 				w->state = FWS_R;
1485 				continue;
1486 			}
1487 			if (pn->right == fn) {
1488 				w->state = FWS_C;
1489 				w->leaf = w->node->leaf;
1490 				continue;
1491 			}
1492 #if RT6_DEBUG >= 2
1493 			WARN_ON(1);
1494 #endif
1495 		}
1496 	}
1497 }
1498 
1499 static int fib6_walk(struct fib6_walker *w)
1500 {
1501 	int res;
1502 
1503 	w->state = FWS_INIT;
1504 	w->node = w->root;
1505 
1506 	fib6_walker_link(w);
1507 	res = fib6_walk_continue(w);
1508 	if (res <= 0)
1509 		fib6_walker_unlink(w);
1510 	return res;
1511 }
1512 
1513 static int fib6_clean_node(struct fib6_walker *w)
1514 {
1515 	int res;
1516 	struct rt6_info *rt;
1517 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1518 	struct nl_info info = {
1519 		.nl_net = c->net,
1520 	};
1521 
1522 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1523 	    w->node->fn_sernum != c->sernum)
1524 		w->node->fn_sernum = c->sernum;
1525 
1526 	if (!c->func) {
1527 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1528 		w->leaf = NULL;
1529 		return 0;
1530 	}
1531 
1532 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1533 		res = c->func(rt, c->arg);
1534 		if (res < 0) {
1535 			w->leaf = rt;
1536 			res = fib6_del(rt, &info);
1537 			if (res) {
1538 #if RT6_DEBUG >= 2
1539 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1540 					 __func__, rt, rt->rt6i_node, res);
1541 #endif
1542 				continue;
1543 			}
1544 			return 0;
1545 		}
1546 		WARN_ON(res != 0);
1547 	}
1548 	w->leaf = rt;
1549 	return 0;
1550 }
1551 
1552 /*
1553  *	Convenient frontend to tree walker.
1554  *
1555  *	func is called on each route.
1556  *		It may return -1 -> delete this route.
1557  *		              0  -> continue walking
1558  *
1559  *	prune==1 -> only immediate children of node (certainly,
1560  *	ignoring pure split nodes) will be scanned.
1561  */
1562 
1563 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1564 			    int (*func)(struct rt6_info *, void *arg),
1565 			    bool prune, int sernum, void *arg)
1566 {
1567 	struct fib6_cleaner c;
1568 
1569 	c.w.root = root;
1570 	c.w.func = fib6_clean_node;
1571 	c.w.prune = prune;
1572 	c.w.count = 0;
1573 	c.w.skip = 0;
1574 	c.func = func;
1575 	c.sernum = sernum;
1576 	c.arg = arg;
1577 	c.net = net;
1578 
1579 	fib6_walk(&c.w);
1580 }
1581 
1582 static void __fib6_clean_all(struct net *net,
1583 			     int (*func)(struct rt6_info *, void *),
1584 			     int sernum, void *arg)
1585 {
1586 	struct fib6_table *table;
1587 	struct hlist_head *head;
1588 	unsigned int h;
1589 
1590 	rcu_read_lock();
1591 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1592 		head = &net->ipv6.fib_table_hash[h];
1593 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1594 			write_lock_bh(&table->tb6_lock);
1595 			fib6_clean_tree(net, &table->tb6_root,
1596 					func, false, sernum, arg);
1597 			write_unlock_bh(&table->tb6_lock);
1598 		}
1599 	}
1600 	rcu_read_unlock();
1601 }
1602 
1603 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1604 		    void *arg)
1605 {
1606 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1607 }
1608 
1609 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1610 {
1611 	if (rt->rt6i_flags & RTF_CACHE) {
1612 		RT6_TRACE("pruning clone %p\n", rt);
1613 		return -1;
1614 	}
1615 
1616 	return 0;
1617 }
1618 
1619 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1620 {
1621 	fib6_clean_tree(net, fn, fib6_prune_clone, true,
1622 			FIB6_NO_SERNUM_CHANGE, NULL);
1623 }
1624 
1625 static void fib6_flush_trees(struct net *net)
1626 {
1627 	int new_sernum = fib6_new_sernum(net);
1628 
1629 	__fib6_clean_all(net, NULL, new_sernum, NULL);
1630 }
1631 
1632 /*
1633  *	Garbage collection
1634  */
1635 
1636 static struct fib6_gc_args
1637 {
1638 	int			timeout;
1639 	int			more;
1640 } gc_args;
1641 
1642 static int fib6_age(struct rt6_info *rt, void *arg)
1643 {
1644 	unsigned long now = jiffies;
1645 
1646 	/*
1647 	 *	check addrconf expiration here.
1648 	 *	Routes are expired even if they are in use.
1649 	 *
1650 	 *	Also age clones. Note, that clones are aged out
1651 	 *	only if they are not in use now.
1652 	 */
1653 
1654 	if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1655 		if (time_after(now, rt->dst.expires)) {
1656 			RT6_TRACE("expiring %p\n", rt);
1657 			return -1;
1658 		}
1659 		gc_args.more++;
1660 	} else if (rt->rt6i_flags & RTF_CACHE) {
1661 		if (atomic_read(&rt->dst.__refcnt) == 0 &&
1662 		    time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1663 			RT6_TRACE("aging clone %p\n", rt);
1664 			return -1;
1665 		} else if (rt->rt6i_flags & RTF_GATEWAY) {
1666 			struct neighbour *neigh;
1667 			__u8 neigh_flags = 0;
1668 
1669 			neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1670 			if (neigh) {
1671 				neigh_flags = neigh->flags;
1672 				neigh_release(neigh);
1673 			}
1674 			if (!(neigh_flags & NTF_ROUTER)) {
1675 				RT6_TRACE("purging route %p via non-router but gateway\n",
1676 					  rt);
1677 				return -1;
1678 			}
1679 		}
1680 		gc_args.more++;
1681 	}
1682 
1683 	return 0;
1684 }
1685 
1686 static DEFINE_SPINLOCK(fib6_gc_lock);
1687 
1688 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1689 {
1690 	unsigned long now;
1691 
1692 	if (force) {
1693 		spin_lock_bh(&fib6_gc_lock);
1694 	} else if (!spin_trylock_bh(&fib6_gc_lock)) {
1695 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1696 		return;
1697 	}
1698 	gc_args.timeout = expires ? (int)expires :
1699 			  net->ipv6.sysctl.ip6_rt_gc_interval;
1700 
1701 	gc_args.more = icmp6_dst_gc();
1702 
1703 	fib6_clean_all(net, fib6_age, NULL);
1704 	now = jiffies;
1705 	net->ipv6.ip6_rt_last_gc = now;
1706 
1707 	if (gc_args.more)
1708 		mod_timer(&net->ipv6.ip6_fib_timer,
1709 			  round_jiffies(now
1710 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
1711 	else
1712 		del_timer(&net->ipv6.ip6_fib_timer);
1713 	spin_unlock_bh(&fib6_gc_lock);
1714 }
1715 
1716 static void fib6_gc_timer_cb(unsigned long arg)
1717 {
1718 	fib6_run_gc(0, (struct net *)arg, true);
1719 }
1720 
1721 static int __net_init fib6_net_init(struct net *net)
1722 {
1723 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1724 
1725 	setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1726 
1727 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1728 	if (!net->ipv6.rt6_stats)
1729 		goto out_timer;
1730 
1731 	/* Avoid false sharing : Use at least a full cache line */
1732 	size = max_t(size_t, size, L1_CACHE_BYTES);
1733 
1734 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1735 	if (!net->ipv6.fib_table_hash)
1736 		goto out_rt6_stats;
1737 
1738 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1739 					  GFP_KERNEL);
1740 	if (!net->ipv6.fib6_main_tbl)
1741 		goto out_fib_table_hash;
1742 
1743 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1744 	net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1745 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1746 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1747 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1748 
1749 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1750 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1751 					   GFP_KERNEL);
1752 	if (!net->ipv6.fib6_local_tbl)
1753 		goto out_fib6_main_tbl;
1754 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1755 	net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1756 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1757 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1758 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1759 #endif
1760 	fib6_tables_init(net);
1761 
1762 	return 0;
1763 
1764 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1765 out_fib6_main_tbl:
1766 	kfree(net->ipv6.fib6_main_tbl);
1767 #endif
1768 out_fib_table_hash:
1769 	kfree(net->ipv6.fib_table_hash);
1770 out_rt6_stats:
1771 	kfree(net->ipv6.rt6_stats);
1772 out_timer:
1773 	return -ENOMEM;
1774 }
1775 
1776 static void fib6_net_exit(struct net *net)
1777 {
1778 	rt6_ifdown(net, NULL);
1779 	del_timer_sync(&net->ipv6.ip6_fib_timer);
1780 
1781 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1782 	inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1783 	kfree(net->ipv6.fib6_local_tbl);
1784 #endif
1785 	inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1786 	kfree(net->ipv6.fib6_main_tbl);
1787 	kfree(net->ipv6.fib_table_hash);
1788 	kfree(net->ipv6.rt6_stats);
1789 }
1790 
1791 static struct pernet_operations fib6_net_ops = {
1792 	.init = fib6_net_init,
1793 	.exit = fib6_net_exit,
1794 };
1795 
1796 int __init fib6_init(void)
1797 {
1798 	int ret = -ENOMEM;
1799 
1800 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
1801 					   sizeof(struct fib6_node),
1802 					   0, SLAB_HWCACHE_ALIGN,
1803 					   NULL);
1804 	if (!fib6_node_kmem)
1805 		goto out;
1806 
1807 	ret = register_pernet_subsys(&fib6_net_ops);
1808 	if (ret)
1809 		goto out_kmem_cache_create;
1810 
1811 	ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1812 			      NULL);
1813 	if (ret)
1814 		goto out_unregister_subsys;
1815 
1816 	__fib6_flush_trees = fib6_flush_trees;
1817 out:
1818 	return ret;
1819 
1820 out_unregister_subsys:
1821 	unregister_pernet_subsys(&fib6_net_ops);
1822 out_kmem_cache_create:
1823 	kmem_cache_destroy(fib6_node_kmem);
1824 	goto out;
1825 }
1826 
1827 void fib6_gc_cleanup(void)
1828 {
1829 	unregister_pernet_subsys(&fib6_net_ops);
1830 	kmem_cache_destroy(fib6_node_kmem);
1831 }
1832 
1833 #ifdef CONFIG_PROC_FS
1834 
1835 struct ipv6_route_iter {
1836 	struct seq_net_private p;
1837 	struct fib6_walker w;
1838 	loff_t skip;
1839 	struct fib6_table *tbl;
1840 	int sernum;
1841 };
1842 
1843 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1844 {
1845 	struct rt6_info *rt = v;
1846 	struct ipv6_route_iter *iter = seq->private;
1847 
1848 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1849 
1850 #ifdef CONFIG_IPV6_SUBTREES
1851 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1852 #else
1853 	seq_puts(seq, "00000000000000000000000000000000 00 ");
1854 #endif
1855 	if (rt->rt6i_flags & RTF_GATEWAY)
1856 		seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1857 	else
1858 		seq_puts(seq, "00000000000000000000000000000000");
1859 
1860 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1861 		   rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1862 		   rt->dst.__use, rt->rt6i_flags,
1863 		   rt->dst.dev ? rt->dst.dev->name : "");
1864 	iter->w.leaf = NULL;
1865 	return 0;
1866 }
1867 
1868 static int ipv6_route_yield(struct fib6_walker *w)
1869 {
1870 	struct ipv6_route_iter *iter = w->args;
1871 
1872 	if (!iter->skip)
1873 		return 1;
1874 
1875 	do {
1876 		iter->w.leaf = iter->w.leaf->dst.rt6_next;
1877 		iter->skip--;
1878 		if (!iter->skip && iter->w.leaf)
1879 			return 1;
1880 	} while (iter->w.leaf);
1881 
1882 	return 0;
1883 }
1884 
1885 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
1886 {
1887 	memset(&iter->w, 0, sizeof(iter->w));
1888 	iter->w.func = ipv6_route_yield;
1889 	iter->w.root = &iter->tbl->tb6_root;
1890 	iter->w.state = FWS_INIT;
1891 	iter->w.node = iter->w.root;
1892 	iter->w.args = iter;
1893 	iter->sernum = iter->w.root->fn_sernum;
1894 	INIT_LIST_HEAD(&iter->w.lh);
1895 	fib6_walker_link(&iter->w);
1896 }
1897 
1898 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1899 						    struct net *net)
1900 {
1901 	unsigned int h;
1902 	struct hlist_node *node;
1903 
1904 	if (tbl) {
1905 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
1906 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
1907 	} else {
1908 		h = 0;
1909 		node = NULL;
1910 	}
1911 
1912 	while (!node && h < FIB6_TABLE_HASHSZ) {
1913 		node = rcu_dereference_bh(
1914 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
1915 	}
1916 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
1917 }
1918 
1919 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
1920 {
1921 	if (iter->sernum != iter->w.root->fn_sernum) {
1922 		iter->sernum = iter->w.root->fn_sernum;
1923 		iter->w.state = FWS_INIT;
1924 		iter->w.node = iter->w.root;
1925 		WARN_ON(iter->w.skip);
1926 		iter->w.skip = iter->w.count;
1927 	}
1928 }
1929 
1930 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1931 {
1932 	int r;
1933 	struct rt6_info *n;
1934 	struct net *net = seq_file_net(seq);
1935 	struct ipv6_route_iter *iter = seq->private;
1936 
1937 	if (!v)
1938 		goto iter_table;
1939 
1940 	n = ((struct rt6_info *)v)->dst.rt6_next;
1941 	if (n) {
1942 		++*pos;
1943 		return n;
1944 	}
1945 
1946 iter_table:
1947 	ipv6_route_check_sernum(iter);
1948 	read_lock(&iter->tbl->tb6_lock);
1949 	r = fib6_walk_continue(&iter->w);
1950 	read_unlock(&iter->tbl->tb6_lock);
1951 	if (r > 0) {
1952 		if (v)
1953 			++*pos;
1954 		return iter->w.leaf;
1955 	} else if (r < 0) {
1956 		fib6_walker_unlink(&iter->w);
1957 		return NULL;
1958 	}
1959 	fib6_walker_unlink(&iter->w);
1960 
1961 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
1962 	if (!iter->tbl)
1963 		return NULL;
1964 
1965 	ipv6_route_seq_setup_walk(iter);
1966 	goto iter_table;
1967 }
1968 
1969 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
1970 	__acquires(RCU_BH)
1971 {
1972 	struct net *net = seq_file_net(seq);
1973 	struct ipv6_route_iter *iter = seq->private;
1974 
1975 	rcu_read_lock_bh();
1976 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
1977 	iter->skip = *pos;
1978 
1979 	if (iter->tbl) {
1980 		ipv6_route_seq_setup_walk(iter);
1981 		return ipv6_route_seq_next(seq, NULL, pos);
1982 	} else {
1983 		return NULL;
1984 	}
1985 }
1986 
1987 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
1988 {
1989 	struct fib6_walker *w = &iter->w;
1990 	return w->node && !(w->state == FWS_U && w->node == w->root);
1991 }
1992 
1993 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
1994 	__releases(RCU_BH)
1995 {
1996 	struct ipv6_route_iter *iter = seq->private;
1997 
1998 	if (ipv6_route_iter_active(iter))
1999 		fib6_walker_unlink(&iter->w);
2000 
2001 	rcu_read_unlock_bh();
2002 }
2003 
2004 static const struct seq_operations ipv6_route_seq_ops = {
2005 	.start	= ipv6_route_seq_start,
2006 	.next	= ipv6_route_seq_next,
2007 	.stop	= ipv6_route_seq_stop,
2008 	.show	= ipv6_route_seq_show
2009 };
2010 
2011 int ipv6_route_open(struct inode *inode, struct file *file)
2012 {
2013 	return seq_open_net(inode, file, &ipv6_route_seq_ops,
2014 			    sizeof(struct ipv6_route_iter));
2015 }
2016 
2017 #endif /* CONFIG_PROC_FS */
2018