xref: /linux/net/ipv6/ip6_fib.c (revision 005438a8eef063495ac059d128eea71b58de50e5)
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_free_pcpu(struct rt6_info *non_pcpu_rt)
158 {
159 	int cpu;
160 
161 	if (!non_pcpu_rt->rt6i_pcpu)
162 		return;
163 
164 	for_each_possible_cpu(cpu) {
165 		struct rt6_info **ppcpu_rt;
166 		struct rt6_info *pcpu_rt;
167 
168 		ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
169 		pcpu_rt = *ppcpu_rt;
170 		if (pcpu_rt) {
171 			dst_free(&pcpu_rt->dst);
172 			*ppcpu_rt = NULL;
173 		}
174 	}
175 }
176 
177 static void rt6_release(struct rt6_info *rt)
178 {
179 	if (atomic_dec_and_test(&rt->rt6i_ref)) {
180 		rt6_free_pcpu(rt);
181 		dst_free(&rt->dst);
182 	}
183 }
184 
185 static void fib6_link_table(struct net *net, struct fib6_table *tb)
186 {
187 	unsigned int h;
188 
189 	/*
190 	 * Initialize table lock at a single place to give lockdep a key,
191 	 * tables aren't visible prior to being linked to the list.
192 	 */
193 	rwlock_init(&tb->tb6_lock);
194 
195 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
196 
197 	/*
198 	 * No protection necessary, this is the only list mutatation
199 	 * operation, tables never disappear once they exist.
200 	 */
201 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
202 }
203 
204 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
205 
206 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
207 {
208 	struct fib6_table *table;
209 
210 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
211 	if (table) {
212 		table->tb6_id = id;
213 		table->tb6_root.leaf = net->ipv6.ip6_null_entry;
214 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
215 		inet_peer_base_init(&table->tb6_peers);
216 	}
217 
218 	return table;
219 }
220 
221 struct fib6_table *fib6_new_table(struct net *net, u32 id)
222 {
223 	struct fib6_table *tb;
224 
225 	if (id == 0)
226 		id = RT6_TABLE_MAIN;
227 	tb = fib6_get_table(net, id);
228 	if (tb)
229 		return tb;
230 
231 	tb = fib6_alloc_table(net, id);
232 	if (tb)
233 		fib6_link_table(net, tb);
234 
235 	return tb;
236 }
237 
238 struct fib6_table *fib6_get_table(struct net *net, u32 id)
239 {
240 	struct fib6_table *tb;
241 	struct hlist_head *head;
242 	unsigned int h;
243 
244 	if (id == 0)
245 		id = RT6_TABLE_MAIN;
246 	h = id & (FIB6_TABLE_HASHSZ - 1);
247 	rcu_read_lock();
248 	head = &net->ipv6.fib_table_hash[h];
249 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
250 		if (tb->tb6_id == id) {
251 			rcu_read_unlock();
252 			return tb;
253 		}
254 	}
255 	rcu_read_unlock();
256 
257 	return NULL;
258 }
259 
260 static void __net_init fib6_tables_init(struct net *net)
261 {
262 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
263 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
264 }
265 #else
266 
267 struct fib6_table *fib6_new_table(struct net *net, u32 id)
268 {
269 	return fib6_get_table(net, id);
270 }
271 
272 struct fib6_table *fib6_get_table(struct net *net, u32 id)
273 {
274 	  return net->ipv6.fib6_main_tbl;
275 }
276 
277 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
278 				   int flags, pol_lookup_t lookup)
279 {
280 	return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
281 }
282 
283 static void __net_init fib6_tables_init(struct net *net)
284 {
285 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
286 }
287 
288 #endif
289 
290 static int fib6_dump_node(struct fib6_walker *w)
291 {
292 	int res;
293 	struct rt6_info *rt;
294 
295 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
296 		res = rt6_dump_route(rt, w->args);
297 		if (res < 0) {
298 			/* Frame is full, suspend walking */
299 			w->leaf = rt;
300 			return 1;
301 		}
302 	}
303 	w->leaf = NULL;
304 	return 0;
305 }
306 
307 static void fib6_dump_end(struct netlink_callback *cb)
308 {
309 	struct fib6_walker *w = (void *)cb->args[2];
310 
311 	if (w) {
312 		if (cb->args[4]) {
313 			cb->args[4] = 0;
314 			fib6_walker_unlink(w);
315 		}
316 		cb->args[2] = 0;
317 		kfree(w);
318 	}
319 	cb->done = (void *)cb->args[3];
320 	cb->args[1] = 3;
321 }
322 
323 static int fib6_dump_done(struct netlink_callback *cb)
324 {
325 	fib6_dump_end(cb);
326 	return cb->done ? cb->done(cb) : 0;
327 }
328 
329 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
330 			   struct netlink_callback *cb)
331 {
332 	struct fib6_walker *w;
333 	int res;
334 
335 	w = (void *)cb->args[2];
336 	w->root = &table->tb6_root;
337 
338 	if (cb->args[4] == 0) {
339 		w->count = 0;
340 		w->skip = 0;
341 
342 		read_lock_bh(&table->tb6_lock);
343 		res = fib6_walk(w);
344 		read_unlock_bh(&table->tb6_lock);
345 		if (res > 0) {
346 			cb->args[4] = 1;
347 			cb->args[5] = w->root->fn_sernum;
348 		}
349 	} else {
350 		if (cb->args[5] != w->root->fn_sernum) {
351 			/* Begin at the root if the tree changed */
352 			cb->args[5] = w->root->fn_sernum;
353 			w->state = FWS_INIT;
354 			w->node = w->root;
355 			w->skip = w->count;
356 		} else
357 			w->skip = 0;
358 
359 		read_lock_bh(&table->tb6_lock);
360 		res = fib6_walk_continue(w);
361 		read_unlock_bh(&table->tb6_lock);
362 		if (res <= 0) {
363 			fib6_walker_unlink(w);
364 			cb->args[4] = 0;
365 		}
366 	}
367 
368 	return res;
369 }
370 
371 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
372 {
373 	struct net *net = sock_net(skb->sk);
374 	unsigned int h, s_h;
375 	unsigned int e = 0, s_e;
376 	struct rt6_rtnl_dump_arg arg;
377 	struct fib6_walker *w;
378 	struct fib6_table *tb;
379 	struct hlist_head *head;
380 	int res = 0;
381 
382 	s_h = cb->args[0];
383 	s_e = cb->args[1];
384 
385 	w = (void *)cb->args[2];
386 	if (!w) {
387 		/* New dump:
388 		 *
389 		 * 1. hook callback destructor.
390 		 */
391 		cb->args[3] = (long)cb->done;
392 		cb->done = fib6_dump_done;
393 
394 		/*
395 		 * 2. allocate and initialize walker.
396 		 */
397 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
398 		if (!w)
399 			return -ENOMEM;
400 		w->func = fib6_dump_node;
401 		cb->args[2] = (long)w;
402 	}
403 
404 	arg.skb = skb;
405 	arg.cb = cb;
406 	arg.net = net;
407 	w->args = &arg;
408 
409 	rcu_read_lock();
410 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
411 		e = 0;
412 		head = &net->ipv6.fib_table_hash[h];
413 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
414 			if (e < s_e)
415 				goto next;
416 			res = fib6_dump_table(tb, skb, cb);
417 			if (res != 0)
418 				goto out;
419 next:
420 			e++;
421 		}
422 	}
423 out:
424 	rcu_read_unlock();
425 	cb->args[1] = e;
426 	cb->args[0] = h;
427 
428 	res = res < 0 ? res : skb->len;
429 	if (res <= 0)
430 		fib6_dump_end(cb);
431 	return res;
432 }
433 
434 /*
435  *	Routing Table
436  *
437  *	return the appropriate node for a routing tree "add" operation
438  *	by either creating and inserting or by returning an existing
439  *	node.
440  */
441 
442 static struct fib6_node *fib6_add_1(struct fib6_node *root,
443 				     struct in6_addr *addr, int plen,
444 				     int offset, int allow_create,
445 				     int replace_required, int sernum)
446 {
447 	struct fib6_node *fn, *in, *ln;
448 	struct fib6_node *pn = NULL;
449 	struct rt6key *key;
450 	int	bit;
451 	__be32	dir = 0;
452 
453 	RT6_TRACE("fib6_add_1\n");
454 
455 	/* insert node in tree */
456 
457 	fn = root;
458 
459 	do {
460 		key = (struct rt6key *)((u8 *)fn->leaf + offset);
461 
462 		/*
463 		 *	Prefix match
464 		 */
465 		if (plen < fn->fn_bit ||
466 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
467 			if (!allow_create) {
468 				if (replace_required) {
469 					pr_warn("Can't replace route, no match found\n");
470 					return ERR_PTR(-ENOENT);
471 				}
472 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
473 			}
474 			goto insert_above;
475 		}
476 
477 		/*
478 		 *	Exact match ?
479 		 */
480 
481 		if (plen == fn->fn_bit) {
482 			/* clean up an intermediate node */
483 			if (!(fn->fn_flags & RTN_RTINFO)) {
484 				rt6_release(fn->leaf);
485 				fn->leaf = NULL;
486 			}
487 
488 			fn->fn_sernum = sernum;
489 
490 			return fn;
491 		}
492 
493 		/*
494 		 *	We have more bits to go
495 		 */
496 
497 		/* Try to walk down on tree. */
498 		fn->fn_sernum = sernum;
499 		dir = addr_bit_set(addr, fn->fn_bit);
500 		pn = fn;
501 		fn = dir ? fn->right : fn->left;
502 	} while (fn);
503 
504 	if (!allow_create) {
505 		/* We should not create new node because
506 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
507 		 * I assume it is safe to require NLM_F_CREATE when
508 		 * REPLACE flag is used! Later we may want to remove the
509 		 * check for replace_required, because according
510 		 * to netlink specification, NLM_F_CREATE
511 		 * MUST be specified if new route is created.
512 		 * That would keep IPv6 consistent with IPv4
513 		 */
514 		if (replace_required) {
515 			pr_warn("Can't replace route, no match found\n");
516 			return ERR_PTR(-ENOENT);
517 		}
518 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
519 	}
520 	/*
521 	 *	We walked to the bottom of tree.
522 	 *	Create new leaf node without children.
523 	 */
524 
525 	ln = node_alloc();
526 
527 	if (!ln)
528 		return ERR_PTR(-ENOMEM);
529 	ln->fn_bit = plen;
530 
531 	ln->parent = pn;
532 	ln->fn_sernum = sernum;
533 
534 	if (dir)
535 		pn->right = ln;
536 	else
537 		pn->left  = ln;
538 
539 	return ln;
540 
541 
542 insert_above:
543 	/*
544 	 * split since we don't have a common prefix anymore or
545 	 * we have a less significant route.
546 	 * we've to insert an intermediate node on the list
547 	 * this new node will point to the one we need to create
548 	 * and the current
549 	 */
550 
551 	pn = fn->parent;
552 
553 	/* find 1st bit in difference between the 2 addrs.
554 
555 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
556 	   but if it is >= plen, the value is ignored in any case.
557 	 */
558 
559 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
560 
561 	/*
562 	 *		(intermediate)[in]
563 	 *	          /	   \
564 	 *	(new leaf node)[ln] (old node)[fn]
565 	 */
566 	if (plen > bit) {
567 		in = node_alloc();
568 		ln = node_alloc();
569 
570 		if (!in || !ln) {
571 			if (in)
572 				node_free(in);
573 			if (ln)
574 				node_free(ln);
575 			return ERR_PTR(-ENOMEM);
576 		}
577 
578 		/*
579 		 * new intermediate node.
580 		 * RTN_RTINFO will
581 		 * be off since that an address that chooses one of
582 		 * the branches would not match less specific routes
583 		 * in the other branch
584 		 */
585 
586 		in->fn_bit = bit;
587 
588 		in->parent = pn;
589 		in->leaf = fn->leaf;
590 		atomic_inc(&in->leaf->rt6i_ref);
591 
592 		in->fn_sernum = sernum;
593 
594 		/* update parent pointer */
595 		if (dir)
596 			pn->right = in;
597 		else
598 			pn->left  = in;
599 
600 		ln->fn_bit = plen;
601 
602 		ln->parent = in;
603 		fn->parent = in;
604 
605 		ln->fn_sernum = sernum;
606 
607 		if (addr_bit_set(addr, bit)) {
608 			in->right = ln;
609 			in->left  = fn;
610 		} else {
611 			in->left  = ln;
612 			in->right = fn;
613 		}
614 	} else { /* plen <= bit */
615 
616 		/*
617 		 *		(new leaf node)[ln]
618 		 *	          /	   \
619 		 *	     (old node)[fn] NULL
620 		 */
621 
622 		ln = node_alloc();
623 
624 		if (!ln)
625 			return ERR_PTR(-ENOMEM);
626 
627 		ln->fn_bit = plen;
628 
629 		ln->parent = pn;
630 
631 		ln->fn_sernum = sernum;
632 
633 		if (dir)
634 			pn->right = ln;
635 		else
636 			pn->left  = ln;
637 
638 		if (addr_bit_set(&key->addr, plen))
639 			ln->right = fn;
640 		else
641 			ln->left  = fn;
642 
643 		fn->parent = ln;
644 	}
645 	return ln;
646 }
647 
648 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
649 {
650 	return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
651 	       RTF_GATEWAY;
652 }
653 
654 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
655 {
656 	int i;
657 
658 	for (i = 0; i < RTAX_MAX; i++) {
659 		if (test_bit(i, mxc->mx_valid))
660 			mp[i] = mxc->mx[i];
661 	}
662 }
663 
664 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
665 {
666 	if (!mxc->mx)
667 		return 0;
668 
669 	if (dst->flags & DST_HOST) {
670 		u32 *mp = dst_metrics_write_ptr(dst);
671 
672 		if (unlikely(!mp))
673 			return -ENOMEM;
674 
675 		fib6_copy_metrics(mp, mxc);
676 	} else {
677 		dst_init_metrics(dst, mxc->mx, false);
678 
679 		/* We've stolen mx now. */
680 		mxc->mx = NULL;
681 	}
682 
683 	return 0;
684 }
685 
686 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
687 			  struct net *net)
688 {
689 	if (atomic_read(&rt->rt6i_ref) != 1) {
690 		/* This route is used as dummy address holder in some split
691 		 * nodes. It is not leaked, but it still holds other resources,
692 		 * which must be released in time. So, scan ascendant nodes
693 		 * and replace dummy references to this route with references
694 		 * to still alive ones.
695 		 */
696 		while (fn) {
697 			if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
698 				fn->leaf = fib6_find_prefix(net, fn);
699 				atomic_inc(&fn->leaf->rt6i_ref);
700 				rt6_release(rt);
701 			}
702 			fn = fn->parent;
703 		}
704 		/* No more references are possible at this point. */
705 		BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
706 	}
707 }
708 
709 /*
710  *	Insert routing information in a node.
711  */
712 
713 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
714 			    struct nl_info *info, struct mx6_config *mxc)
715 {
716 	struct rt6_info *iter = NULL;
717 	struct rt6_info **ins;
718 	struct rt6_info **fallback_ins = NULL;
719 	int replace = (info->nlh &&
720 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
721 	int add = (!info->nlh ||
722 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
723 	int found = 0;
724 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
725 	int err;
726 
727 	ins = &fn->leaf;
728 
729 	for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
730 		/*
731 		 *	Search for duplicates
732 		 */
733 
734 		if (iter->rt6i_metric == rt->rt6i_metric) {
735 			/*
736 			 *	Same priority level
737 			 */
738 			if (info->nlh &&
739 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
740 				return -EEXIST;
741 			if (replace) {
742 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
743 					found++;
744 					break;
745 				}
746 				if (rt_can_ecmp)
747 					fallback_ins = fallback_ins ?: ins;
748 				goto next_iter;
749 			}
750 
751 			if (iter->dst.dev == rt->dst.dev &&
752 			    iter->rt6i_idev == rt->rt6i_idev &&
753 			    ipv6_addr_equal(&iter->rt6i_gateway,
754 					    &rt->rt6i_gateway)) {
755 				if (rt->rt6i_nsiblings)
756 					rt->rt6i_nsiblings = 0;
757 				if (!(iter->rt6i_flags & RTF_EXPIRES))
758 					return -EEXIST;
759 				if (!(rt->rt6i_flags & RTF_EXPIRES))
760 					rt6_clean_expires(iter);
761 				else
762 					rt6_set_expires(iter, rt->dst.expires);
763 				iter->rt6i_pmtu = rt->rt6i_pmtu;
764 				return -EEXIST;
765 			}
766 			/* If we have the same destination and the same metric,
767 			 * but not the same gateway, then the route we try to
768 			 * add is sibling to this route, increment our counter
769 			 * of siblings, and later we will add our route to the
770 			 * list.
771 			 * Only static routes (which don't have flag
772 			 * RTF_EXPIRES) are used for ECMPv6.
773 			 *
774 			 * To avoid long list, we only had siblings if the
775 			 * route have a gateway.
776 			 */
777 			if (rt_can_ecmp &&
778 			    rt6_qualify_for_ecmp(iter))
779 				rt->rt6i_nsiblings++;
780 		}
781 
782 		if (iter->rt6i_metric > rt->rt6i_metric)
783 			break;
784 
785 next_iter:
786 		ins = &iter->dst.rt6_next;
787 	}
788 
789 	if (fallback_ins && !found) {
790 		/* No ECMP-able route found, replace first non-ECMP one */
791 		ins = fallback_ins;
792 		iter = *ins;
793 		found++;
794 	}
795 
796 	/* Reset round-robin state, if necessary */
797 	if (ins == &fn->leaf)
798 		fn->rr_ptr = NULL;
799 
800 	/* Link this route to others same route. */
801 	if (rt->rt6i_nsiblings) {
802 		unsigned int rt6i_nsiblings;
803 		struct rt6_info *sibling, *temp_sibling;
804 
805 		/* Find the first route that have the same metric */
806 		sibling = fn->leaf;
807 		while (sibling) {
808 			if (sibling->rt6i_metric == rt->rt6i_metric &&
809 			    rt6_qualify_for_ecmp(sibling)) {
810 				list_add_tail(&rt->rt6i_siblings,
811 					      &sibling->rt6i_siblings);
812 				break;
813 			}
814 			sibling = sibling->dst.rt6_next;
815 		}
816 		/* For each sibling in the list, increment the counter of
817 		 * siblings. BUG() if counters does not match, list of siblings
818 		 * is broken!
819 		 */
820 		rt6i_nsiblings = 0;
821 		list_for_each_entry_safe(sibling, temp_sibling,
822 					 &rt->rt6i_siblings, rt6i_siblings) {
823 			sibling->rt6i_nsiblings++;
824 			BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
825 			rt6i_nsiblings++;
826 		}
827 		BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
828 	}
829 
830 	/*
831 	 *	insert node
832 	 */
833 	if (!replace) {
834 		if (!add)
835 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
836 
837 add:
838 		err = fib6_commit_metrics(&rt->dst, mxc);
839 		if (err)
840 			return err;
841 
842 		rt->dst.rt6_next = iter;
843 		*ins = rt;
844 		rt->rt6i_node = fn;
845 		atomic_inc(&rt->rt6i_ref);
846 		inet6_rt_notify(RTM_NEWROUTE, rt, info);
847 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
848 
849 		if (!(fn->fn_flags & RTN_RTINFO)) {
850 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
851 			fn->fn_flags |= RTN_RTINFO;
852 		}
853 
854 	} else {
855 		int nsiblings;
856 
857 		if (!found) {
858 			if (add)
859 				goto add;
860 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
861 			return -ENOENT;
862 		}
863 
864 		err = fib6_commit_metrics(&rt->dst, mxc);
865 		if (err)
866 			return err;
867 
868 		*ins = rt;
869 		rt->rt6i_node = fn;
870 		rt->dst.rt6_next = iter->dst.rt6_next;
871 		atomic_inc(&rt->rt6i_ref);
872 		inet6_rt_notify(RTM_NEWROUTE, rt, info);
873 		if (!(fn->fn_flags & RTN_RTINFO)) {
874 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
875 			fn->fn_flags |= RTN_RTINFO;
876 		}
877 		nsiblings = iter->rt6i_nsiblings;
878 		fib6_purge_rt(iter, fn, info->nl_net);
879 		rt6_release(iter);
880 
881 		if (nsiblings) {
882 			/* Replacing an ECMP route, remove all siblings */
883 			ins = &rt->dst.rt6_next;
884 			iter = *ins;
885 			while (iter) {
886 				if (rt6_qualify_for_ecmp(iter)) {
887 					*ins = iter->dst.rt6_next;
888 					fib6_purge_rt(iter, fn, info->nl_net);
889 					rt6_release(iter);
890 					nsiblings--;
891 				} else {
892 					ins = &iter->dst.rt6_next;
893 				}
894 				iter = *ins;
895 			}
896 			WARN_ON(nsiblings != 0);
897 		}
898 	}
899 
900 	return 0;
901 }
902 
903 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
904 {
905 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
906 	    (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
907 		mod_timer(&net->ipv6.ip6_fib_timer,
908 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
909 }
910 
911 void fib6_force_start_gc(struct net *net)
912 {
913 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
914 		mod_timer(&net->ipv6.ip6_fib_timer,
915 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
916 }
917 
918 /*
919  *	Add routing information to the routing tree.
920  *	<destination addr>/<source addr>
921  *	with source addr info in sub-trees
922  */
923 
924 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
925 	     struct nl_info *info, struct mx6_config *mxc)
926 {
927 	struct fib6_node *fn, *pn = NULL;
928 	int err = -ENOMEM;
929 	int allow_create = 1;
930 	int replace_required = 0;
931 	int sernum = fib6_new_sernum(info->nl_net);
932 
933 	if (info->nlh) {
934 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
935 			allow_create = 0;
936 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
937 			replace_required = 1;
938 	}
939 	if (!allow_create && !replace_required)
940 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
941 
942 	fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
943 			offsetof(struct rt6_info, rt6i_dst), allow_create,
944 			replace_required, sernum);
945 	if (IS_ERR(fn)) {
946 		err = PTR_ERR(fn);
947 		fn = NULL;
948 		goto out;
949 	}
950 
951 	pn = fn;
952 
953 #ifdef CONFIG_IPV6_SUBTREES
954 	if (rt->rt6i_src.plen) {
955 		struct fib6_node *sn;
956 
957 		if (!fn->subtree) {
958 			struct fib6_node *sfn;
959 
960 			/*
961 			 * Create subtree.
962 			 *
963 			 *		fn[main tree]
964 			 *		|
965 			 *		sfn[subtree root]
966 			 *		   \
967 			 *		    sn[new leaf node]
968 			 */
969 
970 			/* Create subtree root node */
971 			sfn = node_alloc();
972 			if (!sfn)
973 				goto st_failure;
974 
975 			sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
976 			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
977 			sfn->fn_flags = RTN_ROOT;
978 			sfn->fn_sernum = sernum;
979 
980 			/* Now add the first leaf node to new subtree */
981 
982 			sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
983 					rt->rt6i_src.plen,
984 					offsetof(struct rt6_info, rt6i_src),
985 					allow_create, replace_required, sernum);
986 
987 			if (IS_ERR(sn)) {
988 				/* If it is failed, discard just allocated
989 				   root, and then (in st_failure) stale node
990 				   in main tree.
991 				 */
992 				node_free(sfn);
993 				err = PTR_ERR(sn);
994 				goto st_failure;
995 			}
996 
997 			/* Now link new subtree to main tree */
998 			sfn->parent = fn;
999 			fn->subtree = sfn;
1000 		} else {
1001 			sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1002 					rt->rt6i_src.plen,
1003 					offsetof(struct rt6_info, rt6i_src),
1004 					allow_create, replace_required, sernum);
1005 
1006 			if (IS_ERR(sn)) {
1007 				err = PTR_ERR(sn);
1008 				goto st_failure;
1009 			}
1010 		}
1011 
1012 		if (!fn->leaf) {
1013 			fn->leaf = rt;
1014 			atomic_inc(&rt->rt6i_ref);
1015 		}
1016 		fn = sn;
1017 	}
1018 #endif
1019 
1020 	err = fib6_add_rt2node(fn, rt, info, mxc);
1021 	if (!err) {
1022 		fib6_start_gc(info->nl_net, rt);
1023 		if (!(rt->rt6i_flags & RTF_CACHE))
1024 			fib6_prune_clones(info->nl_net, pn);
1025 	}
1026 
1027 out:
1028 	if (err) {
1029 #ifdef CONFIG_IPV6_SUBTREES
1030 		/*
1031 		 * If fib6_add_1 has cleared the old leaf pointer in the
1032 		 * super-tree leaf node we have to find a new one for it.
1033 		 */
1034 		if (pn != fn && pn->leaf == rt) {
1035 			pn->leaf = NULL;
1036 			atomic_dec(&rt->rt6i_ref);
1037 		}
1038 		if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1039 			pn->leaf = fib6_find_prefix(info->nl_net, pn);
1040 #if RT6_DEBUG >= 2
1041 			if (!pn->leaf) {
1042 				WARN_ON(pn->leaf == NULL);
1043 				pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1044 			}
1045 #endif
1046 			atomic_inc(&pn->leaf->rt6i_ref);
1047 		}
1048 #endif
1049 		dst_free(&rt->dst);
1050 	}
1051 	return err;
1052 
1053 #ifdef CONFIG_IPV6_SUBTREES
1054 	/* Subtree creation failed, probably main tree node
1055 	   is orphan. If it is, shoot it.
1056 	 */
1057 st_failure:
1058 	if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1059 		fib6_repair_tree(info->nl_net, fn);
1060 	dst_free(&rt->dst);
1061 	return err;
1062 #endif
1063 }
1064 
1065 /*
1066  *	Routing tree lookup
1067  *
1068  */
1069 
1070 struct lookup_args {
1071 	int			offset;		/* key offset on rt6_info	*/
1072 	const struct in6_addr	*addr;		/* search key			*/
1073 };
1074 
1075 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1076 				       struct lookup_args *args)
1077 {
1078 	struct fib6_node *fn;
1079 	__be32 dir;
1080 
1081 	if (unlikely(args->offset == 0))
1082 		return NULL;
1083 
1084 	/*
1085 	 *	Descend on a tree
1086 	 */
1087 
1088 	fn = root;
1089 
1090 	for (;;) {
1091 		struct fib6_node *next;
1092 
1093 		dir = addr_bit_set(args->addr, fn->fn_bit);
1094 
1095 		next = dir ? fn->right : fn->left;
1096 
1097 		if (next) {
1098 			fn = next;
1099 			continue;
1100 		}
1101 		break;
1102 	}
1103 
1104 	while (fn) {
1105 		if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1106 			struct rt6key *key;
1107 
1108 			key = (struct rt6key *) ((u8 *) fn->leaf +
1109 						 args->offset);
1110 
1111 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1112 #ifdef CONFIG_IPV6_SUBTREES
1113 				if (fn->subtree) {
1114 					struct fib6_node *sfn;
1115 					sfn = fib6_lookup_1(fn->subtree,
1116 							    args + 1);
1117 					if (!sfn)
1118 						goto backtrack;
1119 					fn = sfn;
1120 				}
1121 #endif
1122 				if (fn->fn_flags & RTN_RTINFO)
1123 					return fn;
1124 			}
1125 		}
1126 #ifdef CONFIG_IPV6_SUBTREES
1127 backtrack:
1128 #endif
1129 		if (fn->fn_flags & RTN_ROOT)
1130 			break;
1131 
1132 		fn = fn->parent;
1133 	}
1134 
1135 	return NULL;
1136 }
1137 
1138 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1139 			      const struct in6_addr *saddr)
1140 {
1141 	struct fib6_node *fn;
1142 	struct lookup_args args[] = {
1143 		{
1144 			.offset = offsetof(struct rt6_info, rt6i_dst),
1145 			.addr = daddr,
1146 		},
1147 #ifdef CONFIG_IPV6_SUBTREES
1148 		{
1149 			.offset = offsetof(struct rt6_info, rt6i_src),
1150 			.addr = saddr,
1151 		},
1152 #endif
1153 		{
1154 			.offset = 0,	/* sentinel */
1155 		}
1156 	};
1157 
1158 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
1159 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1160 		fn = root;
1161 
1162 	return fn;
1163 }
1164 
1165 /*
1166  *	Get node with specified destination prefix (and source prefix,
1167  *	if subtrees are used)
1168  */
1169 
1170 
1171 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1172 				       const struct in6_addr *addr,
1173 				       int plen, int offset)
1174 {
1175 	struct fib6_node *fn;
1176 
1177 	for (fn = root; fn ; ) {
1178 		struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1179 
1180 		/*
1181 		 *	Prefix match
1182 		 */
1183 		if (plen < fn->fn_bit ||
1184 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1185 			return NULL;
1186 
1187 		if (plen == fn->fn_bit)
1188 			return fn;
1189 
1190 		/*
1191 		 *	We have more bits to go
1192 		 */
1193 		if (addr_bit_set(addr, fn->fn_bit))
1194 			fn = fn->right;
1195 		else
1196 			fn = fn->left;
1197 	}
1198 	return NULL;
1199 }
1200 
1201 struct fib6_node *fib6_locate(struct fib6_node *root,
1202 			      const struct in6_addr *daddr, int dst_len,
1203 			      const struct in6_addr *saddr, int src_len)
1204 {
1205 	struct fib6_node *fn;
1206 
1207 	fn = fib6_locate_1(root, daddr, dst_len,
1208 			   offsetof(struct rt6_info, rt6i_dst));
1209 
1210 #ifdef CONFIG_IPV6_SUBTREES
1211 	if (src_len) {
1212 		WARN_ON(saddr == NULL);
1213 		if (fn && fn->subtree)
1214 			fn = fib6_locate_1(fn->subtree, saddr, src_len,
1215 					   offsetof(struct rt6_info, rt6i_src));
1216 	}
1217 #endif
1218 
1219 	if (fn && fn->fn_flags & RTN_RTINFO)
1220 		return fn;
1221 
1222 	return NULL;
1223 }
1224 
1225 
1226 /*
1227  *	Deletion
1228  *
1229  */
1230 
1231 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1232 {
1233 	if (fn->fn_flags & RTN_ROOT)
1234 		return net->ipv6.ip6_null_entry;
1235 
1236 	while (fn) {
1237 		if (fn->left)
1238 			return fn->left->leaf;
1239 		if (fn->right)
1240 			return fn->right->leaf;
1241 
1242 		fn = FIB6_SUBTREE(fn);
1243 	}
1244 	return NULL;
1245 }
1246 
1247 /*
1248  *	Called to trim the tree of intermediate nodes when possible. "fn"
1249  *	is the node we want to try and remove.
1250  */
1251 
1252 static struct fib6_node *fib6_repair_tree(struct net *net,
1253 					   struct fib6_node *fn)
1254 {
1255 	int children;
1256 	int nstate;
1257 	struct fib6_node *child, *pn;
1258 	struct fib6_walker *w;
1259 	int iter = 0;
1260 
1261 	for (;;) {
1262 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1263 		iter++;
1264 
1265 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1266 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1267 		WARN_ON(fn->leaf);
1268 
1269 		children = 0;
1270 		child = NULL;
1271 		if (fn->right)
1272 			child = fn->right, children |= 1;
1273 		if (fn->left)
1274 			child = fn->left, children |= 2;
1275 
1276 		if (children == 3 || FIB6_SUBTREE(fn)
1277 #ifdef CONFIG_IPV6_SUBTREES
1278 		    /* Subtree root (i.e. fn) may have one child */
1279 		    || (children && fn->fn_flags & RTN_ROOT)
1280 #endif
1281 		    ) {
1282 			fn->leaf = fib6_find_prefix(net, fn);
1283 #if RT6_DEBUG >= 2
1284 			if (!fn->leaf) {
1285 				WARN_ON(!fn->leaf);
1286 				fn->leaf = net->ipv6.ip6_null_entry;
1287 			}
1288 #endif
1289 			atomic_inc(&fn->leaf->rt6i_ref);
1290 			return fn->parent;
1291 		}
1292 
1293 		pn = fn->parent;
1294 #ifdef CONFIG_IPV6_SUBTREES
1295 		if (FIB6_SUBTREE(pn) == fn) {
1296 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1297 			FIB6_SUBTREE(pn) = NULL;
1298 			nstate = FWS_L;
1299 		} else {
1300 			WARN_ON(fn->fn_flags & RTN_ROOT);
1301 #endif
1302 			if (pn->right == fn)
1303 				pn->right = child;
1304 			else if (pn->left == fn)
1305 				pn->left = child;
1306 #if RT6_DEBUG >= 2
1307 			else
1308 				WARN_ON(1);
1309 #endif
1310 			if (child)
1311 				child->parent = pn;
1312 			nstate = FWS_R;
1313 #ifdef CONFIG_IPV6_SUBTREES
1314 		}
1315 #endif
1316 
1317 		read_lock(&fib6_walker_lock);
1318 		FOR_WALKERS(w) {
1319 			if (!child) {
1320 				if (w->root == fn) {
1321 					w->root = w->node = NULL;
1322 					RT6_TRACE("W %p adjusted by delroot 1\n", w);
1323 				} else if (w->node == fn) {
1324 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1325 					w->node = pn;
1326 					w->state = nstate;
1327 				}
1328 			} else {
1329 				if (w->root == fn) {
1330 					w->root = child;
1331 					RT6_TRACE("W %p adjusted by delroot 2\n", w);
1332 				}
1333 				if (w->node == fn) {
1334 					w->node = child;
1335 					if (children&2) {
1336 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1337 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1338 					} else {
1339 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1340 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1341 					}
1342 				}
1343 			}
1344 		}
1345 		read_unlock(&fib6_walker_lock);
1346 
1347 		node_free(fn);
1348 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1349 			return pn;
1350 
1351 		rt6_release(pn->leaf);
1352 		pn->leaf = NULL;
1353 		fn = pn;
1354 	}
1355 }
1356 
1357 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1358 			   struct nl_info *info)
1359 {
1360 	struct fib6_walker *w;
1361 	struct rt6_info *rt = *rtp;
1362 	struct net *net = info->nl_net;
1363 
1364 	RT6_TRACE("fib6_del_route\n");
1365 
1366 	/* Unlink it */
1367 	*rtp = rt->dst.rt6_next;
1368 	rt->rt6i_node = NULL;
1369 	net->ipv6.rt6_stats->fib_rt_entries--;
1370 	net->ipv6.rt6_stats->fib_discarded_routes++;
1371 
1372 	/* Reset round-robin state, if necessary */
1373 	if (fn->rr_ptr == rt)
1374 		fn->rr_ptr = NULL;
1375 
1376 	/* Remove this entry from other siblings */
1377 	if (rt->rt6i_nsiblings) {
1378 		struct rt6_info *sibling, *next_sibling;
1379 
1380 		list_for_each_entry_safe(sibling, next_sibling,
1381 					 &rt->rt6i_siblings, rt6i_siblings)
1382 			sibling->rt6i_nsiblings--;
1383 		rt->rt6i_nsiblings = 0;
1384 		list_del_init(&rt->rt6i_siblings);
1385 	}
1386 
1387 	/* Adjust walkers */
1388 	read_lock(&fib6_walker_lock);
1389 	FOR_WALKERS(w) {
1390 		if (w->state == FWS_C && w->leaf == rt) {
1391 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1392 			w->leaf = rt->dst.rt6_next;
1393 			if (!w->leaf)
1394 				w->state = FWS_U;
1395 		}
1396 	}
1397 	read_unlock(&fib6_walker_lock);
1398 
1399 	rt->dst.rt6_next = NULL;
1400 
1401 	/* If it was last route, expunge its radix tree node */
1402 	if (!fn->leaf) {
1403 		fn->fn_flags &= ~RTN_RTINFO;
1404 		net->ipv6.rt6_stats->fib_route_nodes--;
1405 		fn = fib6_repair_tree(net, fn);
1406 	}
1407 
1408 	fib6_purge_rt(rt, fn, net);
1409 
1410 	inet6_rt_notify(RTM_DELROUTE, rt, info);
1411 	rt6_release(rt);
1412 }
1413 
1414 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1415 {
1416 	struct net *net = info->nl_net;
1417 	struct fib6_node *fn = rt->rt6i_node;
1418 	struct rt6_info **rtp;
1419 
1420 #if RT6_DEBUG >= 2
1421 	if (rt->dst.obsolete > 0) {
1422 		WARN_ON(fn);
1423 		return -ENOENT;
1424 	}
1425 #endif
1426 	if (!fn || rt == net->ipv6.ip6_null_entry)
1427 		return -ENOENT;
1428 
1429 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1430 
1431 	if (!(rt->rt6i_flags & RTF_CACHE)) {
1432 		struct fib6_node *pn = fn;
1433 #ifdef CONFIG_IPV6_SUBTREES
1434 		/* clones of this route might be in another subtree */
1435 		if (rt->rt6i_src.plen) {
1436 			while (!(pn->fn_flags & RTN_ROOT))
1437 				pn = pn->parent;
1438 			pn = pn->parent;
1439 		}
1440 #endif
1441 		fib6_prune_clones(info->nl_net, pn);
1442 	}
1443 
1444 	/*
1445 	 *	Walk the leaf entries looking for ourself
1446 	 */
1447 
1448 	for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1449 		if (*rtp == rt) {
1450 			fib6_del_route(fn, rtp, info);
1451 			return 0;
1452 		}
1453 	}
1454 	return -ENOENT;
1455 }
1456 
1457 /*
1458  *	Tree traversal function.
1459  *
1460  *	Certainly, it is not interrupt safe.
1461  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1462  *	It means, that we can modify tree during walking
1463  *	and use this function for garbage collection, clone pruning,
1464  *	cleaning tree when a device goes down etc. etc.
1465  *
1466  *	It guarantees that every node will be traversed,
1467  *	and that it will be traversed only once.
1468  *
1469  *	Callback function w->func may return:
1470  *	0 -> continue walking.
1471  *	positive value -> walking is suspended (used by tree dumps,
1472  *	and probably by gc, if it will be split to several slices)
1473  *	negative value -> terminate walking.
1474  *
1475  *	The function itself returns:
1476  *	0   -> walk is complete.
1477  *	>0  -> walk is incomplete (i.e. suspended)
1478  *	<0  -> walk is terminated by an error.
1479  */
1480 
1481 static int fib6_walk_continue(struct fib6_walker *w)
1482 {
1483 	struct fib6_node *fn, *pn;
1484 
1485 	for (;;) {
1486 		fn = w->node;
1487 		if (!fn)
1488 			return 0;
1489 
1490 		if (w->prune && fn != w->root &&
1491 		    fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1492 			w->state = FWS_C;
1493 			w->leaf = fn->leaf;
1494 		}
1495 		switch (w->state) {
1496 #ifdef CONFIG_IPV6_SUBTREES
1497 		case FWS_S:
1498 			if (FIB6_SUBTREE(fn)) {
1499 				w->node = FIB6_SUBTREE(fn);
1500 				continue;
1501 			}
1502 			w->state = FWS_L;
1503 #endif
1504 		case FWS_L:
1505 			if (fn->left) {
1506 				w->node = fn->left;
1507 				w->state = FWS_INIT;
1508 				continue;
1509 			}
1510 			w->state = FWS_R;
1511 		case FWS_R:
1512 			if (fn->right) {
1513 				w->node = fn->right;
1514 				w->state = FWS_INIT;
1515 				continue;
1516 			}
1517 			w->state = FWS_C;
1518 			w->leaf = fn->leaf;
1519 		case FWS_C:
1520 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1521 				int err;
1522 
1523 				if (w->skip) {
1524 					w->skip--;
1525 					goto skip;
1526 				}
1527 
1528 				err = w->func(w);
1529 				if (err)
1530 					return err;
1531 
1532 				w->count++;
1533 				continue;
1534 			}
1535 skip:
1536 			w->state = FWS_U;
1537 		case FWS_U:
1538 			if (fn == w->root)
1539 				return 0;
1540 			pn = fn->parent;
1541 			w->node = pn;
1542 #ifdef CONFIG_IPV6_SUBTREES
1543 			if (FIB6_SUBTREE(pn) == fn) {
1544 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1545 				w->state = FWS_L;
1546 				continue;
1547 			}
1548 #endif
1549 			if (pn->left == fn) {
1550 				w->state = FWS_R;
1551 				continue;
1552 			}
1553 			if (pn->right == fn) {
1554 				w->state = FWS_C;
1555 				w->leaf = w->node->leaf;
1556 				continue;
1557 			}
1558 #if RT6_DEBUG >= 2
1559 			WARN_ON(1);
1560 #endif
1561 		}
1562 	}
1563 }
1564 
1565 static int fib6_walk(struct fib6_walker *w)
1566 {
1567 	int res;
1568 
1569 	w->state = FWS_INIT;
1570 	w->node = w->root;
1571 
1572 	fib6_walker_link(w);
1573 	res = fib6_walk_continue(w);
1574 	if (res <= 0)
1575 		fib6_walker_unlink(w);
1576 	return res;
1577 }
1578 
1579 static int fib6_clean_node(struct fib6_walker *w)
1580 {
1581 	int res;
1582 	struct rt6_info *rt;
1583 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1584 	struct nl_info info = {
1585 		.nl_net = c->net,
1586 	};
1587 
1588 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1589 	    w->node->fn_sernum != c->sernum)
1590 		w->node->fn_sernum = c->sernum;
1591 
1592 	if (!c->func) {
1593 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1594 		w->leaf = NULL;
1595 		return 0;
1596 	}
1597 
1598 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1599 		res = c->func(rt, c->arg);
1600 		if (res < 0) {
1601 			w->leaf = rt;
1602 			res = fib6_del(rt, &info);
1603 			if (res) {
1604 #if RT6_DEBUG >= 2
1605 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1606 					 __func__, rt, rt->rt6i_node, res);
1607 #endif
1608 				continue;
1609 			}
1610 			return 0;
1611 		}
1612 		WARN_ON(res != 0);
1613 	}
1614 	w->leaf = rt;
1615 	return 0;
1616 }
1617 
1618 /*
1619  *	Convenient frontend to tree walker.
1620  *
1621  *	func is called on each route.
1622  *		It may return -1 -> delete this route.
1623  *		              0  -> continue walking
1624  *
1625  *	prune==1 -> only immediate children of node (certainly,
1626  *	ignoring pure split nodes) will be scanned.
1627  */
1628 
1629 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1630 			    int (*func)(struct rt6_info *, void *arg),
1631 			    bool prune, int sernum, void *arg)
1632 {
1633 	struct fib6_cleaner c;
1634 
1635 	c.w.root = root;
1636 	c.w.func = fib6_clean_node;
1637 	c.w.prune = prune;
1638 	c.w.count = 0;
1639 	c.w.skip = 0;
1640 	c.func = func;
1641 	c.sernum = sernum;
1642 	c.arg = arg;
1643 	c.net = net;
1644 
1645 	fib6_walk(&c.w);
1646 }
1647 
1648 static void __fib6_clean_all(struct net *net,
1649 			     int (*func)(struct rt6_info *, void *),
1650 			     int sernum, void *arg)
1651 {
1652 	struct fib6_table *table;
1653 	struct hlist_head *head;
1654 	unsigned int h;
1655 
1656 	rcu_read_lock();
1657 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1658 		head = &net->ipv6.fib_table_hash[h];
1659 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1660 			write_lock_bh(&table->tb6_lock);
1661 			fib6_clean_tree(net, &table->tb6_root,
1662 					func, false, sernum, arg);
1663 			write_unlock_bh(&table->tb6_lock);
1664 		}
1665 	}
1666 	rcu_read_unlock();
1667 }
1668 
1669 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1670 		    void *arg)
1671 {
1672 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1673 }
1674 
1675 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1676 {
1677 	if (rt->rt6i_flags & RTF_CACHE) {
1678 		RT6_TRACE("pruning clone %p\n", rt);
1679 		return -1;
1680 	}
1681 
1682 	return 0;
1683 }
1684 
1685 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1686 {
1687 	fib6_clean_tree(net, fn, fib6_prune_clone, true,
1688 			FIB6_NO_SERNUM_CHANGE, NULL);
1689 }
1690 
1691 static void fib6_flush_trees(struct net *net)
1692 {
1693 	int new_sernum = fib6_new_sernum(net);
1694 
1695 	__fib6_clean_all(net, NULL, new_sernum, NULL);
1696 }
1697 
1698 /*
1699  *	Garbage collection
1700  */
1701 
1702 static struct fib6_gc_args
1703 {
1704 	int			timeout;
1705 	int			more;
1706 } gc_args;
1707 
1708 static int fib6_age(struct rt6_info *rt, void *arg)
1709 {
1710 	unsigned long now = jiffies;
1711 
1712 	/*
1713 	 *	check addrconf expiration here.
1714 	 *	Routes are expired even if they are in use.
1715 	 *
1716 	 *	Also age clones. Note, that clones are aged out
1717 	 *	only if they are not in use now.
1718 	 */
1719 
1720 	if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1721 		if (time_after(now, rt->dst.expires)) {
1722 			RT6_TRACE("expiring %p\n", rt);
1723 			return -1;
1724 		}
1725 		gc_args.more++;
1726 	} else if (rt->rt6i_flags & RTF_CACHE) {
1727 		if (atomic_read(&rt->dst.__refcnt) == 0 &&
1728 		    time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1729 			RT6_TRACE("aging clone %p\n", rt);
1730 			return -1;
1731 		} else if (rt->rt6i_flags & RTF_GATEWAY) {
1732 			struct neighbour *neigh;
1733 			__u8 neigh_flags = 0;
1734 
1735 			neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1736 			if (neigh) {
1737 				neigh_flags = neigh->flags;
1738 				neigh_release(neigh);
1739 			}
1740 			if (!(neigh_flags & NTF_ROUTER)) {
1741 				RT6_TRACE("purging route %p via non-router but gateway\n",
1742 					  rt);
1743 				return -1;
1744 			}
1745 		}
1746 		gc_args.more++;
1747 	}
1748 
1749 	return 0;
1750 }
1751 
1752 static DEFINE_SPINLOCK(fib6_gc_lock);
1753 
1754 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1755 {
1756 	unsigned long now;
1757 
1758 	if (force) {
1759 		spin_lock_bh(&fib6_gc_lock);
1760 	} else if (!spin_trylock_bh(&fib6_gc_lock)) {
1761 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1762 		return;
1763 	}
1764 	gc_args.timeout = expires ? (int)expires :
1765 			  net->ipv6.sysctl.ip6_rt_gc_interval;
1766 
1767 	gc_args.more = icmp6_dst_gc();
1768 
1769 	fib6_clean_all(net, fib6_age, NULL);
1770 	now = jiffies;
1771 	net->ipv6.ip6_rt_last_gc = now;
1772 
1773 	if (gc_args.more)
1774 		mod_timer(&net->ipv6.ip6_fib_timer,
1775 			  round_jiffies(now
1776 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
1777 	else
1778 		del_timer(&net->ipv6.ip6_fib_timer);
1779 	spin_unlock_bh(&fib6_gc_lock);
1780 }
1781 
1782 static void fib6_gc_timer_cb(unsigned long arg)
1783 {
1784 	fib6_run_gc(0, (struct net *)arg, true);
1785 }
1786 
1787 static int __net_init fib6_net_init(struct net *net)
1788 {
1789 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1790 
1791 	setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1792 
1793 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1794 	if (!net->ipv6.rt6_stats)
1795 		goto out_timer;
1796 
1797 	/* Avoid false sharing : Use at least a full cache line */
1798 	size = max_t(size_t, size, L1_CACHE_BYTES);
1799 
1800 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1801 	if (!net->ipv6.fib_table_hash)
1802 		goto out_rt6_stats;
1803 
1804 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1805 					  GFP_KERNEL);
1806 	if (!net->ipv6.fib6_main_tbl)
1807 		goto out_fib_table_hash;
1808 
1809 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1810 	net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1811 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1812 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1813 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1814 
1815 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1816 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1817 					   GFP_KERNEL);
1818 	if (!net->ipv6.fib6_local_tbl)
1819 		goto out_fib6_main_tbl;
1820 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1821 	net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1822 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1823 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1824 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1825 #endif
1826 	fib6_tables_init(net);
1827 
1828 	return 0;
1829 
1830 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1831 out_fib6_main_tbl:
1832 	kfree(net->ipv6.fib6_main_tbl);
1833 #endif
1834 out_fib_table_hash:
1835 	kfree(net->ipv6.fib_table_hash);
1836 out_rt6_stats:
1837 	kfree(net->ipv6.rt6_stats);
1838 out_timer:
1839 	return -ENOMEM;
1840 }
1841 
1842 static void fib6_net_exit(struct net *net)
1843 {
1844 	rt6_ifdown(net, NULL);
1845 	del_timer_sync(&net->ipv6.ip6_fib_timer);
1846 
1847 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1848 	inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1849 	kfree(net->ipv6.fib6_local_tbl);
1850 #endif
1851 	inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1852 	kfree(net->ipv6.fib6_main_tbl);
1853 	kfree(net->ipv6.fib_table_hash);
1854 	kfree(net->ipv6.rt6_stats);
1855 }
1856 
1857 static struct pernet_operations fib6_net_ops = {
1858 	.init = fib6_net_init,
1859 	.exit = fib6_net_exit,
1860 };
1861 
1862 int __init fib6_init(void)
1863 {
1864 	int ret = -ENOMEM;
1865 
1866 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
1867 					   sizeof(struct fib6_node),
1868 					   0, SLAB_HWCACHE_ALIGN,
1869 					   NULL);
1870 	if (!fib6_node_kmem)
1871 		goto out;
1872 
1873 	ret = register_pernet_subsys(&fib6_net_ops);
1874 	if (ret)
1875 		goto out_kmem_cache_create;
1876 
1877 	ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1878 			      NULL);
1879 	if (ret)
1880 		goto out_unregister_subsys;
1881 
1882 	__fib6_flush_trees = fib6_flush_trees;
1883 out:
1884 	return ret;
1885 
1886 out_unregister_subsys:
1887 	unregister_pernet_subsys(&fib6_net_ops);
1888 out_kmem_cache_create:
1889 	kmem_cache_destroy(fib6_node_kmem);
1890 	goto out;
1891 }
1892 
1893 void fib6_gc_cleanup(void)
1894 {
1895 	unregister_pernet_subsys(&fib6_net_ops);
1896 	kmem_cache_destroy(fib6_node_kmem);
1897 }
1898 
1899 #ifdef CONFIG_PROC_FS
1900 
1901 struct ipv6_route_iter {
1902 	struct seq_net_private p;
1903 	struct fib6_walker w;
1904 	loff_t skip;
1905 	struct fib6_table *tbl;
1906 	int sernum;
1907 };
1908 
1909 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1910 {
1911 	struct rt6_info *rt = v;
1912 	struct ipv6_route_iter *iter = seq->private;
1913 
1914 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1915 
1916 #ifdef CONFIG_IPV6_SUBTREES
1917 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1918 #else
1919 	seq_puts(seq, "00000000000000000000000000000000 00 ");
1920 #endif
1921 	if (rt->rt6i_flags & RTF_GATEWAY)
1922 		seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1923 	else
1924 		seq_puts(seq, "00000000000000000000000000000000");
1925 
1926 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1927 		   rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1928 		   rt->dst.__use, rt->rt6i_flags,
1929 		   rt->dst.dev ? rt->dst.dev->name : "");
1930 	iter->w.leaf = NULL;
1931 	return 0;
1932 }
1933 
1934 static int ipv6_route_yield(struct fib6_walker *w)
1935 {
1936 	struct ipv6_route_iter *iter = w->args;
1937 
1938 	if (!iter->skip)
1939 		return 1;
1940 
1941 	do {
1942 		iter->w.leaf = iter->w.leaf->dst.rt6_next;
1943 		iter->skip--;
1944 		if (!iter->skip && iter->w.leaf)
1945 			return 1;
1946 	} while (iter->w.leaf);
1947 
1948 	return 0;
1949 }
1950 
1951 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
1952 {
1953 	memset(&iter->w, 0, sizeof(iter->w));
1954 	iter->w.func = ipv6_route_yield;
1955 	iter->w.root = &iter->tbl->tb6_root;
1956 	iter->w.state = FWS_INIT;
1957 	iter->w.node = iter->w.root;
1958 	iter->w.args = iter;
1959 	iter->sernum = iter->w.root->fn_sernum;
1960 	INIT_LIST_HEAD(&iter->w.lh);
1961 	fib6_walker_link(&iter->w);
1962 }
1963 
1964 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1965 						    struct net *net)
1966 {
1967 	unsigned int h;
1968 	struct hlist_node *node;
1969 
1970 	if (tbl) {
1971 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
1972 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
1973 	} else {
1974 		h = 0;
1975 		node = NULL;
1976 	}
1977 
1978 	while (!node && h < FIB6_TABLE_HASHSZ) {
1979 		node = rcu_dereference_bh(
1980 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
1981 	}
1982 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
1983 }
1984 
1985 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
1986 {
1987 	if (iter->sernum != iter->w.root->fn_sernum) {
1988 		iter->sernum = iter->w.root->fn_sernum;
1989 		iter->w.state = FWS_INIT;
1990 		iter->w.node = iter->w.root;
1991 		WARN_ON(iter->w.skip);
1992 		iter->w.skip = iter->w.count;
1993 	}
1994 }
1995 
1996 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1997 {
1998 	int r;
1999 	struct rt6_info *n;
2000 	struct net *net = seq_file_net(seq);
2001 	struct ipv6_route_iter *iter = seq->private;
2002 
2003 	if (!v)
2004 		goto iter_table;
2005 
2006 	n = ((struct rt6_info *)v)->dst.rt6_next;
2007 	if (n) {
2008 		++*pos;
2009 		return n;
2010 	}
2011 
2012 iter_table:
2013 	ipv6_route_check_sernum(iter);
2014 	read_lock(&iter->tbl->tb6_lock);
2015 	r = fib6_walk_continue(&iter->w);
2016 	read_unlock(&iter->tbl->tb6_lock);
2017 	if (r > 0) {
2018 		if (v)
2019 			++*pos;
2020 		return iter->w.leaf;
2021 	} else if (r < 0) {
2022 		fib6_walker_unlink(&iter->w);
2023 		return NULL;
2024 	}
2025 	fib6_walker_unlink(&iter->w);
2026 
2027 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2028 	if (!iter->tbl)
2029 		return NULL;
2030 
2031 	ipv6_route_seq_setup_walk(iter);
2032 	goto iter_table;
2033 }
2034 
2035 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2036 	__acquires(RCU_BH)
2037 {
2038 	struct net *net = seq_file_net(seq);
2039 	struct ipv6_route_iter *iter = seq->private;
2040 
2041 	rcu_read_lock_bh();
2042 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2043 	iter->skip = *pos;
2044 
2045 	if (iter->tbl) {
2046 		ipv6_route_seq_setup_walk(iter);
2047 		return ipv6_route_seq_next(seq, NULL, pos);
2048 	} else {
2049 		return NULL;
2050 	}
2051 }
2052 
2053 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2054 {
2055 	struct fib6_walker *w = &iter->w;
2056 	return w->node && !(w->state == FWS_U && w->node == w->root);
2057 }
2058 
2059 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2060 	__releases(RCU_BH)
2061 {
2062 	struct ipv6_route_iter *iter = seq->private;
2063 
2064 	if (ipv6_route_iter_active(iter))
2065 		fib6_walker_unlink(&iter->w);
2066 
2067 	rcu_read_unlock_bh();
2068 }
2069 
2070 static const struct seq_operations ipv6_route_seq_ops = {
2071 	.start	= ipv6_route_seq_start,
2072 	.next	= ipv6_route_seq_next,
2073 	.stop	= ipv6_route_seq_stop,
2074 	.show	= ipv6_route_seq_show
2075 };
2076 
2077 int ipv6_route_open(struct inode *inode, struct file *file)
2078 {
2079 	return seq_open_net(inode, file, &ipv6_route_seq_ops,
2080 			    sizeof(struct ipv6_route_iter));
2081 }
2082 
2083 #endif /* CONFIG_PROC_FS */
2084