xref: /openbmc/linux/net/ipv6/ip6_fib.c (revision 174cd4b1)
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 #include <net/lwtunnel.h>
36 
37 #include <net/ip6_fib.h>
38 #include <net/ip6_route.h>
39 
40 #define RT6_DEBUG 2
41 
42 #if RT6_DEBUG >= 3
43 #define RT6_TRACE(x...) pr_debug(x)
44 #else
45 #define RT6_TRACE(x...) do { ; } while (0)
46 #endif
47 
48 static struct kmem_cache *fib6_node_kmem __read_mostly;
49 
50 struct fib6_cleaner {
51 	struct fib6_walker w;
52 	struct net *net;
53 	int (*func)(struct rt6_info *, void *arg);
54 	int sernum;
55 	void *arg;
56 };
57 
58 #ifdef CONFIG_IPV6_SUBTREES
59 #define FWS_INIT FWS_S
60 #else
61 #define FWS_INIT FWS_L
62 #endif
63 
64 static void fib6_prune_clones(struct net *net, struct fib6_node *fn);
65 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
66 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
67 static int fib6_walk(struct net *net, struct fib6_walker *w);
68 static int fib6_walk_continue(struct fib6_walker *w);
69 
70 /*
71  *	A routing update causes an increase of the serial number on the
72  *	affected subtree. This allows for cached routes to be asynchronously
73  *	tested when modifications are made to the destination cache as a
74  *	result of redirects, path MTU changes, etc.
75  */
76 
77 static void fib6_gc_timer_cb(unsigned long arg);
78 
79 #define FOR_WALKERS(net, w) \
80 	list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
81 
82 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
83 {
84 	write_lock_bh(&net->ipv6.fib6_walker_lock);
85 	list_add(&w->lh, &net->ipv6.fib6_walkers);
86 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
87 }
88 
89 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
90 {
91 	write_lock_bh(&net->ipv6.fib6_walker_lock);
92 	list_del(&w->lh);
93 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
94 }
95 
96 static int fib6_new_sernum(struct net *net)
97 {
98 	int new, old;
99 
100 	do {
101 		old = atomic_read(&net->ipv6.fib6_sernum);
102 		new = old < INT_MAX ? old + 1 : 1;
103 	} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
104 				old, new) != old);
105 	return new;
106 }
107 
108 enum {
109 	FIB6_NO_SERNUM_CHANGE = 0,
110 };
111 
112 /*
113  *	Auxiliary address test functions for the radix tree.
114  *
115  *	These assume a 32bit processor (although it will work on
116  *	64bit processors)
117  */
118 
119 /*
120  *	test bit
121  */
122 #if defined(__LITTLE_ENDIAN)
123 # define BITOP_BE32_SWIZZLE	(0x1F & ~7)
124 #else
125 # define BITOP_BE32_SWIZZLE	0
126 #endif
127 
128 static __be32 addr_bit_set(const void *token, int fn_bit)
129 {
130 	const __be32 *addr = token;
131 	/*
132 	 * Here,
133 	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
134 	 * is optimized version of
135 	 *	htonl(1 << ((~fn_bit)&0x1F))
136 	 * See include/asm-generic/bitops/le.h.
137 	 */
138 	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
139 	       addr[fn_bit >> 5];
140 }
141 
142 static struct fib6_node *node_alloc(void)
143 {
144 	struct fib6_node *fn;
145 
146 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
147 
148 	return fn;
149 }
150 
151 static void node_free(struct fib6_node *fn)
152 {
153 	kmem_cache_free(fib6_node_kmem, fn);
154 }
155 
156 static void rt6_rcu_free(struct rt6_info *rt)
157 {
158 	call_rcu(&rt->dst.rcu_head, dst_rcu_free);
159 }
160 
161 static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
162 {
163 	int cpu;
164 
165 	if (!non_pcpu_rt->rt6i_pcpu)
166 		return;
167 
168 	for_each_possible_cpu(cpu) {
169 		struct rt6_info **ppcpu_rt;
170 		struct rt6_info *pcpu_rt;
171 
172 		ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
173 		pcpu_rt = *ppcpu_rt;
174 		if (pcpu_rt) {
175 			rt6_rcu_free(pcpu_rt);
176 			*ppcpu_rt = NULL;
177 		}
178 	}
179 
180 	free_percpu(non_pcpu_rt->rt6i_pcpu);
181 	non_pcpu_rt->rt6i_pcpu = NULL;
182 }
183 
184 static void rt6_release(struct rt6_info *rt)
185 {
186 	if (atomic_dec_and_test(&rt->rt6i_ref)) {
187 		rt6_free_pcpu(rt);
188 		rt6_rcu_free(rt);
189 	}
190 }
191 
192 static void fib6_link_table(struct net *net, struct fib6_table *tb)
193 {
194 	unsigned int h;
195 
196 	/*
197 	 * Initialize table lock at a single place to give lockdep a key,
198 	 * tables aren't visible prior to being linked to the list.
199 	 */
200 	rwlock_init(&tb->tb6_lock);
201 
202 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
203 
204 	/*
205 	 * No protection necessary, this is the only list mutatation
206 	 * operation, tables never disappear once they exist.
207 	 */
208 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
209 }
210 
211 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
212 
213 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
214 {
215 	struct fib6_table *table;
216 
217 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
218 	if (table) {
219 		table->tb6_id = id;
220 		table->tb6_root.leaf = net->ipv6.ip6_null_entry;
221 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
222 		inet_peer_base_init(&table->tb6_peers);
223 	}
224 
225 	return table;
226 }
227 
228 struct fib6_table *fib6_new_table(struct net *net, u32 id)
229 {
230 	struct fib6_table *tb;
231 
232 	if (id == 0)
233 		id = RT6_TABLE_MAIN;
234 	tb = fib6_get_table(net, id);
235 	if (tb)
236 		return tb;
237 
238 	tb = fib6_alloc_table(net, id);
239 	if (tb)
240 		fib6_link_table(net, tb);
241 
242 	return tb;
243 }
244 EXPORT_SYMBOL_GPL(fib6_new_table);
245 
246 struct fib6_table *fib6_get_table(struct net *net, u32 id)
247 {
248 	struct fib6_table *tb;
249 	struct hlist_head *head;
250 	unsigned int h;
251 
252 	if (id == 0)
253 		id = RT6_TABLE_MAIN;
254 	h = id & (FIB6_TABLE_HASHSZ - 1);
255 	rcu_read_lock();
256 	head = &net->ipv6.fib_table_hash[h];
257 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
258 		if (tb->tb6_id == id) {
259 			rcu_read_unlock();
260 			return tb;
261 		}
262 	}
263 	rcu_read_unlock();
264 
265 	return NULL;
266 }
267 EXPORT_SYMBOL_GPL(fib6_get_table);
268 
269 static void __net_init fib6_tables_init(struct net *net)
270 {
271 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
272 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
273 }
274 #else
275 
276 struct fib6_table *fib6_new_table(struct net *net, u32 id)
277 {
278 	return fib6_get_table(net, id);
279 }
280 
281 struct fib6_table *fib6_get_table(struct net *net, u32 id)
282 {
283 	  return net->ipv6.fib6_main_tbl;
284 }
285 
286 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
287 				   int flags, pol_lookup_t lookup)
288 {
289 	struct rt6_info *rt;
290 
291 	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
292 	if (rt->rt6i_flags & RTF_REJECT &&
293 	    rt->dst.error == -EAGAIN) {
294 		ip6_rt_put(rt);
295 		rt = net->ipv6.ip6_null_entry;
296 		dst_hold(&rt->dst);
297 	}
298 
299 	return &rt->dst;
300 }
301 
302 static void __net_init fib6_tables_init(struct net *net)
303 {
304 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
305 }
306 
307 #endif
308 
309 static int fib6_dump_node(struct fib6_walker *w)
310 {
311 	int res;
312 	struct rt6_info *rt;
313 
314 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
315 		res = rt6_dump_route(rt, w->args);
316 		if (res < 0) {
317 			/* Frame is full, suspend walking */
318 			w->leaf = rt;
319 			return 1;
320 		}
321 
322 		/* Multipath routes are dumped in one route with the
323 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
324 		 * last sibling of this route (no need to dump the
325 		 * sibling routes again)
326 		 */
327 		if (rt->rt6i_nsiblings)
328 			rt = list_last_entry(&rt->rt6i_siblings,
329 					     struct rt6_info,
330 					     rt6i_siblings);
331 	}
332 	w->leaf = NULL;
333 	return 0;
334 }
335 
336 static void fib6_dump_end(struct netlink_callback *cb)
337 {
338 	struct net *net = sock_net(cb->skb->sk);
339 	struct fib6_walker *w = (void *)cb->args[2];
340 
341 	if (w) {
342 		if (cb->args[4]) {
343 			cb->args[4] = 0;
344 			fib6_walker_unlink(net, w);
345 		}
346 		cb->args[2] = 0;
347 		kfree(w);
348 	}
349 	cb->done = (void *)cb->args[3];
350 	cb->args[1] = 3;
351 }
352 
353 static int fib6_dump_done(struct netlink_callback *cb)
354 {
355 	fib6_dump_end(cb);
356 	return cb->done ? cb->done(cb) : 0;
357 }
358 
359 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
360 			   struct netlink_callback *cb)
361 {
362 	struct net *net = sock_net(skb->sk);
363 	struct fib6_walker *w;
364 	int res;
365 
366 	w = (void *)cb->args[2];
367 	w->root = &table->tb6_root;
368 
369 	if (cb->args[4] == 0) {
370 		w->count = 0;
371 		w->skip = 0;
372 
373 		read_lock_bh(&table->tb6_lock);
374 		res = fib6_walk(net, w);
375 		read_unlock_bh(&table->tb6_lock);
376 		if (res > 0) {
377 			cb->args[4] = 1;
378 			cb->args[5] = w->root->fn_sernum;
379 		}
380 	} else {
381 		if (cb->args[5] != w->root->fn_sernum) {
382 			/* Begin at the root if the tree changed */
383 			cb->args[5] = w->root->fn_sernum;
384 			w->state = FWS_INIT;
385 			w->node = w->root;
386 			w->skip = w->count;
387 		} else
388 			w->skip = 0;
389 
390 		read_lock_bh(&table->tb6_lock);
391 		res = fib6_walk_continue(w);
392 		read_unlock_bh(&table->tb6_lock);
393 		if (res <= 0) {
394 			fib6_walker_unlink(net, w);
395 			cb->args[4] = 0;
396 		}
397 	}
398 
399 	return res;
400 }
401 
402 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
403 {
404 	struct net *net = sock_net(skb->sk);
405 	unsigned int h, s_h;
406 	unsigned int e = 0, s_e;
407 	struct rt6_rtnl_dump_arg arg;
408 	struct fib6_walker *w;
409 	struct fib6_table *tb;
410 	struct hlist_head *head;
411 	int res = 0;
412 
413 	s_h = cb->args[0];
414 	s_e = cb->args[1];
415 
416 	w = (void *)cb->args[2];
417 	if (!w) {
418 		/* New dump:
419 		 *
420 		 * 1. hook callback destructor.
421 		 */
422 		cb->args[3] = (long)cb->done;
423 		cb->done = fib6_dump_done;
424 
425 		/*
426 		 * 2. allocate and initialize walker.
427 		 */
428 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
429 		if (!w)
430 			return -ENOMEM;
431 		w->func = fib6_dump_node;
432 		cb->args[2] = (long)w;
433 	}
434 
435 	arg.skb = skb;
436 	arg.cb = cb;
437 	arg.net = net;
438 	w->args = &arg;
439 
440 	rcu_read_lock();
441 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
442 		e = 0;
443 		head = &net->ipv6.fib_table_hash[h];
444 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
445 			if (e < s_e)
446 				goto next;
447 			res = fib6_dump_table(tb, skb, cb);
448 			if (res != 0)
449 				goto out;
450 next:
451 			e++;
452 		}
453 	}
454 out:
455 	rcu_read_unlock();
456 	cb->args[1] = e;
457 	cb->args[0] = h;
458 
459 	res = res < 0 ? res : skb->len;
460 	if (res <= 0)
461 		fib6_dump_end(cb);
462 	return res;
463 }
464 
465 /*
466  *	Routing Table
467  *
468  *	return the appropriate node for a routing tree "add" operation
469  *	by either creating and inserting or by returning an existing
470  *	node.
471  */
472 
473 static struct fib6_node *fib6_add_1(struct fib6_node *root,
474 				     struct in6_addr *addr, int plen,
475 				     int offset, int allow_create,
476 				     int replace_required, int sernum)
477 {
478 	struct fib6_node *fn, *in, *ln;
479 	struct fib6_node *pn = NULL;
480 	struct rt6key *key;
481 	int	bit;
482 	__be32	dir = 0;
483 
484 	RT6_TRACE("fib6_add_1\n");
485 
486 	/* insert node in tree */
487 
488 	fn = root;
489 
490 	do {
491 		key = (struct rt6key *)((u8 *)fn->leaf + offset);
492 
493 		/*
494 		 *	Prefix match
495 		 */
496 		if (plen < fn->fn_bit ||
497 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
498 			if (!allow_create) {
499 				if (replace_required) {
500 					pr_warn("Can't replace route, no match found\n");
501 					return ERR_PTR(-ENOENT);
502 				}
503 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
504 			}
505 			goto insert_above;
506 		}
507 
508 		/*
509 		 *	Exact match ?
510 		 */
511 
512 		if (plen == fn->fn_bit) {
513 			/* clean up an intermediate node */
514 			if (!(fn->fn_flags & RTN_RTINFO)) {
515 				rt6_release(fn->leaf);
516 				fn->leaf = NULL;
517 			}
518 
519 			fn->fn_sernum = sernum;
520 
521 			return fn;
522 		}
523 
524 		/*
525 		 *	We have more bits to go
526 		 */
527 
528 		/* Try to walk down on tree. */
529 		fn->fn_sernum = sernum;
530 		dir = addr_bit_set(addr, fn->fn_bit);
531 		pn = fn;
532 		fn = dir ? fn->right : fn->left;
533 	} while (fn);
534 
535 	if (!allow_create) {
536 		/* We should not create new node because
537 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
538 		 * I assume it is safe to require NLM_F_CREATE when
539 		 * REPLACE flag is used! Later we may want to remove the
540 		 * check for replace_required, because according
541 		 * to netlink specification, NLM_F_CREATE
542 		 * MUST be specified if new route is created.
543 		 * That would keep IPv6 consistent with IPv4
544 		 */
545 		if (replace_required) {
546 			pr_warn("Can't replace route, no match found\n");
547 			return ERR_PTR(-ENOENT);
548 		}
549 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
550 	}
551 	/*
552 	 *	We walked to the bottom of tree.
553 	 *	Create new leaf node without children.
554 	 */
555 
556 	ln = node_alloc();
557 
558 	if (!ln)
559 		return ERR_PTR(-ENOMEM);
560 	ln->fn_bit = plen;
561 
562 	ln->parent = pn;
563 	ln->fn_sernum = sernum;
564 
565 	if (dir)
566 		pn->right = ln;
567 	else
568 		pn->left  = ln;
569 
570 	return ln;
571 
572 
573 insert_above:
574 	/*
575 	 * split since we don't have a common prefix anymore or
576 	 * we have a less significant route.
577 	 * we've to insert an intermediate node on the list
578 	 * this new node will point to the one we need to create
579 	 * and the current
580 	 */
581 
582 	pn = fn->parent;
583 
584 	/* find 1st bit in difference between the 2 addrs.
585 
586 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
587 	   but if it is >= plen, the value is ignored in any case.
588 	 */
589 
590 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
591 
592 	/*
593 	 *		(intermediate)[in]
594 	 *	          /	   \
595 	 *	(new leaf node)[ln] (old node)[fn]
596 	 */
597 	if (plen > bit) {
598 		in = node_alloc();
599 		ln = node_alloc();
600 
601 		if (!in || !ln) {
602 			if (in)
603 				node_free(in);
604 			if (ln)
605 				node_free(ln);
606 			return ERR_PTR(-ENOMEM);
607 		}
608 
609 		/*
610 		 * new intermediate node.
611 		 * RTN_RTINFO will
612 		 * be off since that an address that chooses one of
613 		 * the branches would not match less specific routes
614 		 * in the other branch
615 		 */
616 
617 		in->fn_bit = bit;
618 
619 		in->parent = pn;
620 		in->leaf = fn->leaf;
621 		atomic_inc(&in->leaf->rt6i_ref);
622 
623 		in->fn_sernum = sernum;
624 
625 		/* update parent pointer */
626 		if (dir)
627 			pn->right = in;
628 		else
629 			pn->left  = in;
630 
631 		ln->fn_bit = plen;
632 
633 		ln->parent = in;
634 		fn->parent = in;
635 
636 		ln->fn_sernum = sernum;
637 
638 		if (addr_bit_set(addr, bit)) {
639 			in->right = ln;
640 			in->left  = fn;
641 		} else {
642 			in->left  = ln;
643 			in->right = fn;
644 		}
645 	} else { /* plen <= bit */
646 
647 		/*
648 		 *		(new leaf node)[ln]
649 		 *	          /	   \
650 		 *	     (old node)[fn] NULL
651 		 */
652 
653 		ln = node_alloc();
654 
655 		if (!ln)
656 			return ERR_PTR(-ENOMEM);
657 
658 		ln->fn_bit = plen;
659 
660 		ln->parent = pn;
661 
662 		ln->fn_sernum = sernum;
663 
664 		if (dir)
665 			pn->right = ln;
666 		else
667 			pn->left  = ln;
668 
669 		if (addr_bit_set(&key->addr, plen))
670 			ln->right = fn;
671 		else
672 			ln->left  = fn;
673 
674 		fn->parent = ln;
675 	}
676 	return ln;
677 }
678 
679 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
680 {
681 	return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
682 	       RTF_GATEWAY;
683 }
684 
685 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
686 {
687 	int i;
688 
689 	for (i = 0; i < RTAX_MAX; i++) {
690 		if (test_bit(i, mxc->mx_valid))
691 			mp[i] = mxc->mx[i];
692 	}
693 }
694 
695 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
696 {
697 	if (!mxc->mx)
698 		return 0;
699 
700 	if (dst->flags & DST_HOST) {
701 		u32 *mp = dst_metrics_write_ptr(dst);
702 
703 		if (unlikely(!mp))
704 			return -ENOMEM;
705 
706 		fib6_copy_metrics(mp, mxc);
707 	} else {
708 		dst_init_metrics(dst, mxc->mx, false);
709 
710 		/* We've stolen mx now. */
711 		mxc->mx = NULL;
712 	}
713 
714 	return 0;
715 }
716 
717 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
718 			  struct net *net)
719 {
720 	if (atomic_read(&rt->rt6i_ref) != 1) {
721 		/* This route is used as dummy address holder in some split
722 		 * nodes. It is not leaked, but it still holds other resources,
723 		 * which must be released in time. So, scan ascendant nodes
724 		 * and replace dummy references to this route with references
725 		 * to still alive ones.
726 		 */
727 		while (fn) {
728 			if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
729 				fn->leaf = fib6_find_prefix(net, fn);
730 				atomic_inc(&fn->leaf->rt6i_ref);
731 				rt6_release(rt);
732 			}
733 			fn = fn->parent;
734 		}
735 		/* No more references are possible at this point. */
736 		BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
737 	}
738 }
739 
740 /*
741  *	Insert routing information in a node.
742  */
743 
744 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
745 			    struct nl_info *info, struct mx6_config *mxc)
746 {
747 	struct rt6_info *iter = NULL;
748 	struct rt6_info **ins;
749 	struct rt6_info **fallback_ins = NULL;
750 	int replace = (info->nlh &&
751 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
752 	int add = (!info->nlh ||
753 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
754 	int found = 0;
755 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
756 	u16 nlflags = NLM_F_EXCL;
757 	int err;
758 
759 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
760 		nlflags |= NLM_F_APPEND;
761 
762 	ins = &fn->leaf;
763 
764 	for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
765 		/*
766 		 *	Search for duplicates
767 		 */
768 
769 		if (iter->rt6i_metric == rt->rt6i_metric) {
770 			/*
771 			 *	Same priority level
772 			 */
773 			if (info->nlh &&
774 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
775 				return -EEXIST;
776 
777 			nlflags &= ~NLM_F_EXCL;
778 			if (replace) {
779 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
780 					found++;
781 					break;
782 				}
783 				if (rt_can_ecmp)
784 					fallback_ins = fallback_ins ?: ins;
785 				goto next_iter;
786 			}
787 
788 			if (iter->dst.dev == rt->dst.dev &&
789 			    iter->rt6i_idev == rt->rt6i_idev &&
790 			    ipv6_addr_equal(&iter->rt6i_gateway,
791 					    &rt->rt6i_gateway)) {
792 				if (rt->rt6i_nsiblings)
793 					rt->rt6i_nsiblings = 0;
794 				if (!(iter->rt6i_flags & RTF_EXPIRES))
795 					return -EEXIST;
796 				if (!(rt->rt6i_flags & RTF_EXPIRES))
797 					rt6_clean_expires(iter);
798 				else
799 					rt6_set_expires(iter, rt->dst.expires);
800 				iter->rt6i_pmtu = rt->rt6i_pmtu;
801 				return -EEXIST;
802 			}
803 			/* If we have the same destination and the same metric,
804 			 * but not the same gateway, then the route we try to
805 			 * add is sibling to this route, increment our counter
806 			 * of siblings, and later we will add our route to the
807 			 * list.
808 			 * Only static routes (which don't have flag
809 			 * RTF_EXPIRES) are used for ECMPv6.
810 			 *
811 			 * To avoid long list, we only had siblings if the
812 			 * route have a gateway.
813 			 */
814 			if (rt_can_ecmp &&
815 			    rt6_qualify_for_ecmp(iter))
816 				rt->rt6i_nsiblings++;
817 		}
818 
819 		if (iter->rt6i_metric > rt->rt6i_metric)
820 			break;
821 
822 next_iter:
823 		ins = &iter->dst.rt6_next;
824 	}
825 
826 	if (fallback_ins && !found) {
827 		/* No ECMP-able route found, replace first non-ECMP one */
828 		ins = fallback_ins;
829 		iter = *ins;
830 		found++;
831 	}
832 
833 	/* Reset round-robin state, if necessary */
834 	if (ins == &fn->leaf)
835 		fn->rr_ptr = NULL;
836 
837 	/* Link this route to others same route. */
838 	if (rt->rt6i_nsiblings) {
839 		unsigned int rt6i_nsiblings;
840 		struct rt6_info *sibling, *temp_sibling;
841 
842 		/* Find the first route that have the same metric */
843 		sibling = fn->leaf;
844 		while (sibling) {
845 			if (sibling->rt6i_metric == rt->rt6i_metric &&
846 			    rt6_qualify_for_ecmp(sibling)) {
847 				list_add_tail(&rt->rt6i_siblings,
848 					      &sibling->rt6i_siblings);
849 				break;
850 			}
851 			sibling = sibling->dst.rt6_next;
852 		}
853 		/* For each sibling in the list, increment the counter of
854 		 * siblings. BUG() if counters does not match, list of siblings
855 		 * is broken!
856 		 */
857 		rt6i_nsiblings = 0;
858 		list_for_each_entry_safe(sibling, temp_sibling,
859 					 &rt->rt6i_siblings, rt6i_siblings) {
860 			sibling->rt6i_nsiblings++;
861 			BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
862 			rt6i_nsiblings++;
863 		}
864 		BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
865 	}
866 
867 	/*
868 	 *	insert node
869 	 */
870 	if (!replace) {
871 		if (!add)
872 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
873 
874 add:
875 		nlflags |= NLM_F_CREATE;
876 		err = fib6_commit_metrics(&rt->dst, mxc);
877 		if (err)
878 			return err;
879 
880 		rt->dst.rt6_next = iter;
881 		*ins = rt;
882 		rt->rt6i_node = fn;
883 		atomic_inc(&rt->rt6i_ref);
884 		if (!info->skip_notify)
885 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
886 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
887 
888 		if (!(fn->fn_flags & RTN_RTINFO)) {
889 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
890 			fn->fn_flags |= RTN_RTINFO;
891 		}
892 
893 	} else {
894 		int nsiblings;
895 
896 		if (!found) {
897 			if (add)
898 				goto add;
899 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
900 			return -ENOENT;
901 		}
902 
903 		err = fib6_commit_metrics(&rt->dst, mxc);
904 		if (err)
905 			return err;
906 
907 		*ins = rt;
908 		rt->rt6i_node = fn;
909 		rt->dst.rt6_next = iter->dst.rt6_next;
910 		atomic_inc(&rt->rt6i_ref);
911 		if (!info->skip_notify)
912 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
913 		if (!(fn->fn_flags & RTN_RTINFO)) {
914 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
915 			fn->fn_flags |= RTN_RTINFO;
916 		}
917 		nsiblings = iter->rt6i_nsiblings;
918 		fib6_purge_rt(iter, fn, info->nl_net);
919 		rt6_release(iter);
920 
921 		if (nsiblings) {
922 			/* Replacing an ECMP route, remove all siblings */
923 			ins = &rt->dst.rt6_next;
924 			iter = *ins;
925 			while (iter) {
926 				if (rt6_qualify_for_ecmp(iter)) {
927 					*ins = iter->dst.rt6_next;
928 					fib6_purge_rt(iter, fn, info->nl_net);
929 					rt6_release(iter);
930 					nsiblings--;
931 				} else {
932 					ins = &iter->dst.rt6_next;
933 				}
934 				iter = *ins;
935 			}
936 			WARN_ON(nsiblings != 0);
937 		}
938 	}
939 
940 	return 0;
941 }
942 
943 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
944 {
945 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
946 	    (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
947 		mod_timer(&net->ipv6.ip6_fib_timer,
948 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
949 }
950 
951 void fib6_force_start_gc(struct net *net)
952 {
953 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
954 		mod_timer(&net->ipv6.ip6_fib_timer,
955 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
956 }
957 
958 /*
959  *	Add routing information to the routing tree.
960  *	<destination addr>/<source addr>
961  *	with source addr info in sub-trees
962  */
963 
964 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
965 	     struct nl_info *info, struct mx6_config *mxc)
966 {
967 	struct fib6_node *fn, *pn = NULL;
968 	int err = -ENOMEM;
969 	int allow_create = 1;
970 	int replace_required = 0;
971 	int sernum = fib6_new_sernum(info->nl_net);
972 
973 	if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) &&
974 			 !atomic_read(&rt->dst.__refcnt)))
975 		return -EINVAL;
976 
977 	if (info->nlh) {
978 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
979 			allow_create = 0;
980 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
981 			replace_required = 1;
982 	}
983 	if (!allow_create && !replace_required)
984 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
985 
986 	fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
987 			offsetof(struct rt6_info, rt6i_dst), allow_create,
988 			replace_required, sernum);
989 	if (IS_ERR(fn)) {
990 		err = PTR_ERR(fn);
991 		fn = NULL;
992 		goto out;
993 	}
994 
995 	pn = fn;
996 
997 #ifdef CONFIG_IPV6_SUBTREES
998 	if (rt->rt6i_src.plen) {
999 		struct fib6_node *sn;
1000 
1001 		if (!fn->subtree) {
1002 			struct fib6_node *sfn;
1003 
1004 			/*
1005 			 * Create subtree.
1006 			 *
1007 			 *		fn[main tree]
1008 			 *		|
1009 			 *		sfn[subtree root]
1010 			 *		   \
1011 			 *		    sn[new leaf node]
1012 			 */
1013 
1014 			/* Create subtree root node */
1015 			sfn = node_alloc();
1016 			if (!sfn)
1017 				goto st_failure;
1018 
1019 			sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
1020 			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1021 			sfn->fn_flags = RTN_ROOT;
1022 			sfn->fn_sernum = sernum;
1023 
1024 			/* Now add the first leaf node to new subtree */
1025 
1026 			sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
1027 					rt->rt6i_src.plen,
1028 					offsetof(struct rt6_info, rt6i_src),
1029 					allow_create, replace_required, sernum);
1030 
1031 			if (IS_ERR(sn)) {
1032 				/* If it is failed, discard just allocated
1033 				   root, and then (in st_failure) stale node
1034 				   in main tree.
1035 				 */
1036 				node_free(sfn);
1037 				err = PTR_ERR(sn);
1038 				goto st_failure;
1039 			}
1040 
1041 			/* Now link new subtree to main tree */
1042 			sfn->parent = fn;
1043 			fn->subtree = sfn;
1044 		} else {
1045 			sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1046 					rt->rt6i_src.plen,
1047 					offsetof(struct rt6_info, rt6i_src),
1048 					allow_create, replace_required, sernum);
1049 
1050 			if (IS_ERR(sn)) {
1051 				err = PTR_ERR(sn);
1052 				goto st_failure;
1053 			}
1054 		}
1055 
1056 		if (!fn->leaf) {
1057 			fn->leaf = rt;
1058 			atomic_inc(&rt->rt6i_ref);
1059 		}
1060 		fn = sn;
1061 	}
1062 #endif
1063 
1064 	err = fib6_add_rt2node(fn, rt, info, mxc);
1065 	if (!err) {
1066 		fib6_start_gc(info->nl_net, rt);
1067 		if (!(rt->rt6i_flags & RTF_CACHE))
1068 			fib6_prune_clones(info->nl_net, pn);
1069 		rt->dst.flags &= ~DST_NOCACHE;
1070 	}
1071 
1072 out:
1073 	if (err) {
1074 #ifdef CONFIG_IPV6_SUBTREES
1075 		/*
1076 		 * If fib6_add_1 has cleared the old leaf pointer in the
1077 		 * super-tree leaf node we have to find a new one for it.
1078 		 */
1079 		if (pn != fn && pn->leaf == rt) {
1080 			pn->leaf = NULL;
1081 			atomic_dec(&rt->rt6i_ref);
1082 		}
1083 		if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1084 			pn->leaf = fib6_find_prefix(info->nl_net, pn);
1085 #if RT6_DEBUG >= 2
1086 			if (!pn->leaf) {
1087 				WARN_ON(pn->leaf == NULL);
1088 				pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1089 			}
1090 #endif
1091 			atomic_inc(&pn->leaf->rt6i_ref);
1092 		}
1093 #endif
1094 		if (!(rt->dst.flags & DST_NOCACHE))
1095 			dst_free(&rt->dst);
1096 	}
1097 	return err;
1098 
1099 #ifdef CONFIG_IPV6_SUBTREES
1100 	/* Subtree creation failed, probably main tree node
1101 	   is orphan. If it is, shoot it.
1102 	 */
1103 st_failure:
1104 	if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1105 		fib6_repair_tree(info->nl_net, fn);
1106 	if (!(rt->dst.flags & DST_NOCACHE))
1107 		dst_free(&rt->dst);
1108 	return err;
1109 #endif
1110 }
1111 
1112 /*
1113  *	Routing tree lookup
1114  *
1115  */
1116 
1117 struct lookup_args {
1118 	int			offset;		/* key offset on rt6_info	*/
1119 	const struct in6_addr	*addr;		/* search key			*/
1120 };
1121 
1122 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1123 				       struct lookup_args *args)
1124 {
1125 	struct fib6_node *fn;
1126 	__be32 dir;
1127 
1128 	if (unlikely(args->offset == 0))
1129 		return NULL;
1130 
1131 	/*
1132 	 *	Descend on a tree
1133 	 */
1134 
1135 	fn = root;
1136 
1137 	for (;;) {
1138 		struct fib6_node *next;
1139 
1140 		dir = addr_bit_set(args->addr, fn->fn_bit);
1141 
1142 		next = dir ? fn->right : fn->left;
1143 
1144 		if (next) {
1145 			fn = next;
1146 			continue;
1147 		}
1148 		break;
1149 	}
1150 
1151 	while (fn) {
1152 		if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1153 			struct rt6key *key;
1154 
1155 			key = (struct rt6key *) ((u8 *) fn->leaf +
1156 						 args->offset);
1157 
1158 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1159 #ifdef CONFIG_IPV6_SUBTREES
1160 				if (fn->subtree) {
1161 					struct fib6_node *sfn;
1162 					sfn = fib6_lookup_1(fn->subtree,
1163 							    args + 1);
1164 					if (!sfn)
1165 						goto backtrack;
1166 					fn = sfn;
1167 				}
1168 #endif
1169 				if (fn->fn_flags & RTN_RTINFO)
1170 					return fn;
1171 			}
1172 		}
1173 #ifdef CONFIG_IPV6_SUBTREES
1174 backtrack:
1175 #endif
1176 		if (fn->fn_flags & RTN_ROOT)
1177 			break;
1178 
1179 		fn = fn->parent;
1180 	}
1181 
1182 	return NULL;
1183 }
1184 
1185 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1186 			      const struct in6_addr *saddr)
1187 {
1188 	struct fib6_node *fn;
1189 	struct lookup_args args[] = {
1190 		{
1191 			.offset = offsetof(struct rt6_info, rt6i_dst),
1192 			.addr = daddr,
1193 		},
1194 #ifdef CONFIG_IPV6_SUBTREES
1195 		{
1196 			.offset = offsetof(struct rt6_info, rt6i_src),
1197 			.addr = saddr,
1198 		},
1199 #endif
1200 		{
1201 			.offset = 0,	/* sentinel */
1202 		}
1203 	};
1204 
1205 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
1206 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1207 		fn = root;
1208 
1209 	return fn;
1210 }
1211 
1212 /*
1213  *	Get node with specified destination prefix (and source prefix,
1214  *	if subtrees are used)
1215  */
1216 
1217 
1218 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1219 				       const struct in6_addr *addr,
1220 				       int plen, int offset)
1221 {
1222 	struct fib6_node *fn;
1223 
1224 	for (fn = root; fn ; ) {
1225 		struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1226 
1227 		/*
1228 		 *	Prefix match
1229 		 */
1230 		if (plen < fn->fn_bit ||
1231 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1232 			return NULL;
1233 
1234 		if (plen == fn->fn_bit)
1235 			return fn;
1236 
1237 		/*
1238 		 *	We have more bits to go
1239 		 */
1240 		if (addr_bit_set(addr, fn->fn_bit))
1241 			fn = fn->right;
1242 		else
1243 			fn = fn->left;
1244 	}
1245 	return NULL;
1246 }
1247 
1248 struct fib6_node *fib6_locate(struct fib6_node *root,
1249 			      const struct in6_addr *daddr, int dst_len,
1250 			      const struct in6_addr *saddr, int src_len)
1251 {
1252 	struct fib6_node *fn;
1253 
1254 	fn = fib6_locate_1(root, daddr, dst_len,
1255 			   offsetof(struct rt6_info, rt6i_dst));
1256 
1257 #ifdef CONFIG_IPV6_SUBTREES
1258 	if (src_len) {
1259 		WARN_ON(saddr == NULL);
1260 		if (fn && fn->subtree)
1261 			fn = fib6_locate_1(fn->subtree, saddr, src_len,
1262 					   offsetof(struct rt6_info, rt6i_src));
1263 	}
1264 #endif
1265 
1266 	if (fn && fn->fn_flags & RTN_RTINFO)
1267 		return fn;
1268 
1269 	return NULL;
1270 }
1271 
1272 
1273 /*
1274  *	Deletion
1275  *
1276  */
1277 
1278 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1279 {
1280 	if (fn->fn_flags & RTN_ROOT)
1281 		return net->ipv6.ip6_null_entry;
1282 
1283 	while (fn) {
1284 		if (fn->left)
1285 			return fn->left->leaf;
1286 		if (fn->right)
1287 			return fn->right->leaf;
1288 
1289 		fn = FIB6_SUBTREE(fn);
1290 	}
1291 	return NULL;
1292 }
1293 
1294 /*
1295  *	Called to trim the tree of intermediate nodes when possible. "fn"
1296  *	is the node we want to try and remove.
1297  */
1298 
1299 static struct fib6_node *fib6_repair_tree(struct net *net,
1300 					   struct fib6_node *fn)
1301 {
1302 	int children;
1303 	int nstate;
1304 	struct fib6_node *child, *pn;
1305 	struct fib6_walker *w;
1306 	int iter = 0;
1307 
1308 	for (;;) {
1309 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1310 		iter++;
1311 
1312 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1313 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1314 		WARN_ON(fn->leaf);
1315 
1316 		children = 0;
1317 		child = NULL;
1318 		if (fn->right)
1319 			child = fn->right, children |= 1;
1320 		if (fn->left)
1321 			child = fn->left, children |= 2;
1322 
1323 		if (children == 3 || FIB6_SUBTREE(fn)
1324 #ifdef CONFIG_IPV6_SUBTREES
1325 		    /* Subtree root (i.e. fn) may have one child */
1326 		    || (children && fn->fn_flags & RTN_ROOT)
1327 #endif
1328 		    ) {
1329 			fn->leaf = fib6_find_prefix(net, fn);
1330 #if RT6_DEBUG >= 2
1331 			if (!fn->leaf) {
1332 				WARN_ON(!fn->leaf);
1333 				fn->leaf = net->ipv6.ip6_null_entry;
1334 			}
1335 #endif
1336 			atomic_inc(&fn->leaf->rt6i_ref);
1337 			return fn->parent;
1338 		}
1339 
1340 		pn = fn->parent;
1341 #ifdef CONFIG_IPV6_SUBTREES
1342 		if (FIB6_SUBTREE(pn) == fn) {
1343 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1344 			FIB6_SUBTREE(pn) = NULL;
1345 			nstate = FWS_L;
1346 		} else {
1347 			WARN_ON(fn->fn_flags & RTN_ROOT);
1348 #endif
1349 			if (pn->right == fn)
1350 				pn->right = child;
1351 			else if (pn->left == fn)
1352 				pn->left = child;
1353 #if RT6_DEBUG >= 2
1354 			else
1355 				WARN_ON(1);
1356 #endif
1357 			if (child)
1358 				child->parent = pn;
1359 			nstate = FWS_R;
1360 #ifdef CONFIG_IPV6_SUBTREES
1361 		}
1362 #endif
1363 
1364 		read_lock(&net->ipv6.fib6_walker_lock);
1365 		FOR_WALKERS(net, w) {
1366 			if (!child) {
1367 				if (w->root == fn) {
1368 					w->root = w->node = NULL;
1369 					RT6_TRACE("W %p adjusted by delroot 1\n", w);
1370 				} else if (w->node == fn) {
1371 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1372 					w->node = pn;
1373 					w->state = nstate;
1374 				}
1375 			} else {
1376 				if (w->root == fn) {
1377 					w->root = child;
1378 					RT6_TRACE("W %p adjusted by delroot 2\n", w);
1379 				}
1380 				if (w->node == fn) {
1381 					w->node = child;
1382 					if (children&2) {
1383 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1384 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1385 					} else {
1386 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1387 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1388 					}
1389 				}
1390 			}
1391 		}
1392 		read_unlock(&net->ipv6.fib6_walker_lock);
1393 
1394 		node_free(fn);
1395 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1396 			return pn;
1397 
1398 		rt6_release(pn->leaf);
1399 		pn->leaf = NULL;
1400 		fn = pn;
1401 	}
1402 }
1403 
1404 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1405 			   struct nl_info *info)
1406 {
1407 	struct fib6_walker *w;
1408 	struct rt6_info *rt = *rtp;
1409 	struct net *net = info->nl_net;
1410 
1411 	RT6_TRACE("fib6_del_route\n");
1412 
1413 	/* Unlink it */
1414 	*rtp = rt->dst.rt6_next;
1415 	rt->rt6i_node = NULL;
1416 	net->ipv6.rt6_stats->fib_rt_entries--;
1417 	net->ipv6.rt6_stats->fib_discarded_routes++;
1418 
1419 	/* Reset round-robin state, if necessary */
1420 	if (fn->rr_ptr == rt)
1421 		fn->rr_ptr = NULL;
1422 
1423 	/* Remove this entry from other siblings */
1424 	if (rt->rt6i_nsiblings) {
1425 		struct rt6_info *sibling, *next_sibling;
1426 
1427 		list_for_each_entry_safe(sibling, next_sibling,
1428 					 &rt->rt6i_siblings, rt6i_siblings)
1429 			sibling->rt6i_nsiblings--;
1430 		rt->rt6i_nsiblings = 0;
1431 		list_del_init(&rt->rt6i_siblings);
1432 	}
1433 
1434 	/* Adjust walkers */
1435 	read_lock(&net->ipv6.fib6_walker_lock);
1436 	FOR_WALKERS(net, w) {
1437 		if (w->state == FWS_C && w->leaf == rt) {
1438 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1439 			w->leaf = rt->dst.rt6_next;
1440 			if (!w->leaf)
1441 				w->state = FWS_U;
1442 		}
1443 	}
1444 	read_unlock(&net->ipv6.fib6_walker_lock);
1445 
1446 	rt->dst.rt6_next = NULL;
1447 
1448 	/* If it was last route, expunge its radix tree node */
1449 	if (!fn->leaf) {
1450 		fn->fn_flags &= ~RTN_RTINFO;
1451 		net->ipv6.rt6_stats->fib_route_nodes--;
1452 		fn = fib6_repair_tree(net, fn);
1453 	}
1454 
1455 	fib6_purge_rt(rt, fn, net);
1456 
1457 	if (!info->skip_notify)
1458 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1459 	rt6_release(rt);
1460 }
1461 
1462 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1463 {
1464 	struct net *net = info->nl_net;
1465 	struct fib6_node *fn = rt->rt6i_node;
1466 	struct rt6_info **rtp;
1467 
1468 #if RT6_DEBUG >= 2
1469 	if (rt->dst.obsolete > 0) {
1470 		WARN_ON(fn);
1471 		return -ENOENT;
1472 	}
1473 #endif
1474 	if (!fn || rt == net->ipv6.ip6_null_entry)
1475 		return -ENOENT;
1476 
1477 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1478 
1479 	if (!(rt->rt6i_flags & RTF_CACHE)) {
1480 		struct fib6_node *pn = fn;
1481 #ifdef CONFIG_IPV6_SUBTREES
1482 		/* clones of this route might be in another subtree */
1483 		if (rt->rt6i_src.plen) {
1484 			while (!(pn->fn_flags & RTN_ROOT))
1485 				pn = pn->parent;
1486 			pn = pn->parent;
1487 		}
1488 #endif
1489 		fib6_prune_clones(info->nl_net, pn);
1490 	}
1491 
1492 	/*
1493 	 *	Walk the leaf entries looking for ourself
1494 	 */
1495 
1496 	for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1497 		if (*rtp == rt) {
1498 			fib6_del_route(fn, rtp, info);
1499 			return 0;
1500 		}
1501 	}
1502 	return -ENOENT;
1503 }
1504 
1505 /*
1506  *	Tree traversal function.
1507  *
1508  *	Certainly, it is not interrupt safe.
1509  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1510  *	It means, that we can modify tree during walking
1511  *	and use this function for garbage collection, clone pruning,
1512  *	cleaning tree when a device goes down etc. etc.
1513  *
1514  *	It guarantees that every node will be traversed,
1515  *	and that it will be traversed only once.
1516  *
1517  *	Callback function w->func may return:
1518  *	0 -> continue walking.
1519  *	positive value -> walking is suspended (used by tree dumps,
1520  *	and probably by gc, if it will be split to several slices)
1521  *	negative value -> terminate walking.
1522  *
1523  *	The function itself returns:
1524  *	0   -> walk is complete.
1525  *	>0  -> walk is incomplete (i.e. suspended)
1526  *	<0  -> walk is terminated by an error.
1527  */
1528 
1529 static int fib6_walk_continue(struct fib6_walker *w)
1530 {
1531 	struct fib6_node *fn, *pn;
1532 
1533 	for (;;) {
1534 		fn = w->node;
1535 		if (!fn)
1536 			return 0;
1537 
1538 		if (w->prune && fn != w->root &&
1539 		    fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1540 			w->state = FWS_C;
1541 			w->leaf = fn->leaf;
1542 		}
1543 		switch (w->state) {
1544 #ifdef CONFIG_IPV6_SUBTREES
1545 		case FWS_S:
1546 			if (FIB6_SUBTREE(fn)) {
1547 				w->node = FIB6_SUBTREE(fn);
1548 				continue;
1549 			}
1550 			w->state = FWS_L;
1551 #endif
1552 		case FWS_L:
1553 			if (fn->left) {
1554 				w->node = fn->left;
1555 				w->state = FWS_INIT;
1556 				continue;
1557 			}
1558 			w->state = FWS_R;
1559 		case FWS_R:
1560 			if (fn->right) {
1561 				w->node = fn->right;
1562 				w->state = FWS_INIT;
1563 				continue;
1564 			}
1565 			w->state = FWS_C;
1566 			w->leaf = fn->leaf;
1567 		case FWS_C:
1568 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1569 				int err;
1570 
1571 				if (w->skip) {
1572 					w->skip--;
1573 					goto skip;
1574 				}
1575 
1576 				err = w->func(w);
1577 				if (err)
1578 					return err;
1579 
1580 				w->count++;
1581 				continue;
1582 			}
1583 skip:
1584 			w->state = FWS_U;
1585 		case FWS_U:
1586 			if (fn == w->root)
1587 				return 0;
1588 			pn = fn->parent;
1589 			w->node = pn;
1590 #ifdef CONFIG_IPV6_SUBTREES
1591 			if (FIB6_SUBTREE(pn) == fn) {
1592 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1593 				w->state = FWS_L;
1594 				continue;
1595 			}
1596 #endif
1597 			if (pn->left == fn) {
1598 				w->state = FWS_R;
1599 				continue;
1600 			}
1601 			if (pn->right == fn) {
1602 				w->state = FWS_C;
1603 				w->leaf = w->node->leaf;
1604 				continue;
1605 			}
1606 #if RT6_DEBUG >= 2
1607 			WARN_ON(1);
1608 #endif
1609 		}
1610 	}
1611 }
1612 
1613 static int fib6_walk(struct net *net, struct fib6_walker *w)
1614 {
1615 	int res;
1616 
1617 	w->state = FWS_INIT;
1618 	w->node = w->root;
1619 
1620 	fib6_walker_link(net, w);
1621 	res = fib6_walk_continue(w);
1622 	if (res <= 0)
1623 		fib6_walker_unlink(net, w);
1624 	return res;
1625 }
1626 
1627 static int fib6_clean_node(struct fib6_walker *w)
1628 {
1629 	int res;
1630 	struct rt6_info *rt;
1631 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1632 	struct nl_info info = {
1633 		.nl_net = c->net,
1634 	};
1635 
1636 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1637 	    w->node->fn_sernum != c->sernum)
1638 		w->node->fn_sernum = c->sernum;
1639 
1640 	if (!c->func) {
1641 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1642 		w->leaf = NULL;
1643 		return 0;
1644 	}
1645 
1646 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1647 		res = c->func(rt, c->arg);
1648 		if (res < 0) {
1649 			w->leaf = rt;
1650 			res = fib6_del(rt, &info);
1651 			if (res) {
1652 #if RT6_DEBUG >= 2
1653 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1654 					 __func__, rt, rt->rt6i_node, res);
1655 #endif
1656 				continue;
1657 			}
1658 			return 0;
1659 		}
1660 		WARN_ON(res != 0);
1661 	}
1662 	w->leaf = rt;
1663 	return 0;
1664 }
1665 
1666 /*
1667  *	Convenient frontend to tree walker.
1668  *
1669  *	func is called on each route.
1670  *		It may return -1 -> delete this route.
1671  *		              0  -> continue walking
1672  *
1673  *	prune==1 -> only immediate children of node (certainly,
1674  *	ignoring pure split nodes) will be scanned.
1675  */
1676 
1677 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1678 			    int (*func)(struct rt6_info *, void *arg),
1679 			    bool prune, int sernum, void *arg)
1680 {
1681 	struct fib6_cleaner c;
1682 
1683 	c.w.root = root;
1684 	c.w.func = fib6_clean_node;
1685 	c.w.prune = prune;
1686 	c.w.count = 0;
1687 	c.w.skip = 0;
1688 	c.func = func;
1689 	c.sernum = sernum;
1690 	c.arg = arg;
1691 	c.net = net;
1692 
1693 	fib6_walk(net, &c.w);
1694 }
1695 
1696 static void __fib6_clean_all(struct net *net,
1697 			     int (*func)(struct rt6_info *, void *),
1698 			     int sernum, void *arg)
1699 {
1700 	struct fib6_table *table;
1701 	struct hlist_head *head;
1702 	unsigned int h;
1703 
1704 	rcu_read_lock();
1705 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1706 		head = &net->ipv6.fib_table_hash[h];
1707 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1708 			write_lock_bh(&table->tb6_lock);
1709 			fib6_clean_tree(net, &table->tb6_root,
1710 					func, false, sernum, arg);
1711 			write_unlock_bh(&table->tb6_lock);
1712 		}
1713 	}
1714 	rcu_read_unlock();
1715 }
1716 
1717 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1718 		    void *arg)
1719 {
1720 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1721 }
1722 
1723 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1724 {
1725 	if (rt->rt6i_flags & RTF_CACHE) {
1726 		RT6_TRACE("pruning clone %p\n", rt);
1727 		return -1;
1728 	}
1729 
1730 	return 0;
1731 }
1732 
1733 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1734 {
1735 	fib6_clean_tree(net, fn, fib6_prune_clone, true,
1736 			FIB6_NO_SERNUM_CHANGE, NULL);
1737 }
1738 
1739 static void fib6_flush_trees(struct net *net)
1740 {
1741 	int new_sernum = fib6_new_sernum(net);
1742 
1743 	__fib6_clean_all(net, NULL, new_sernum, NULL);
1744 }
1745 
1746 /*
1747  *	Garbage collection
1748  */
1749 
1750 struct fib6_gc_args
1751 {
1752 	int			timeout;
1753 	int			more;
1754 };
1755 
1756 static int fib6_age(struct rt6_info *rt, void *arg)
1757 {
1758 	struct fib6_gc_args *gc_args = arg;
1759 	unsigned long now = jiffies;
1760 
1761 	/*
1762 	 *	check addrconf expiration here.
1763 	 *	Routes are expired even if they are in use.
1764 	 *
1765 	 *	Also age clones. Note, that clones are aged out
1766 	 *	only if they are not in use now.
1767 	 */
1768 
1769 	if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1770 		if (time_after(now, rt->dst.expires)) {
1771 			RT6_TRACE("expiring %p\n", rt);
1772 			return -1;
1773 		}
1774 		gc_args->more++;
1775 	} else if (rt->rt6i_flags & RTF_CACHE) {
1776 		if (atomic_read(&rt->dst.__refcnt) == 0 &&
1777 		    time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
1778 			RT6_TRACE("aging clone %p\n", rt);
1779 			return -1;
1780 		} else if (rt->rt6i_flags & RTF_GATEWAY) {
1781 			struct neighbour *neigh;
1782 			__u8 neigh_flags = 0;
1783 
1784 			neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1785 			if (neigh) {
1786 				neigh_flags = neigh->flags;
1787 				neigh_release(neigh);
1788 			}
1789 			if (!(neigh_flags & NTF_ROUTER)) {
1790 				RT6_TRACE("purging route %p via non-router but gateway\n",
1791 					  rt);
1792 				return -1;
1793 			}
1794 		}
1795 		gc_args->more++;
1796 	}
1797 
1798 	return 0;
1799 }
1800 
1801 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1802 {
1803 	struct fib6_gc_args gc_args;
1804 	unsigned long now;
1805 
1806 	if (force) {
1807 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
1808 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
1809 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1810 		return;
1811 	}
1812 	gc_args.timeout = expires ? (int)expires :
1813 			  net->ipv6.sysctl.ip6_rt_gc_interval;
1814 
1815 	gc_args.more = icmp6_dst_gc();
1816 
1817 	fib6_clean_all(net, fib6_age, &gc_args);
1818 	now = jiffies;
1819 	net->ipv6.ip6_rt_last_gc = now;
1820 
1821 	if (gc_args.more)
1822 		mod_timer(&net->ipv6.ip6_fib_timer,
1823 			  round_jiffies(now
1824 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
1825 	else
1826 		del_timer(&net->ipv6.ip6_fib_timer);
1827 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
1828 }
1829 
1830 static void fib6_gc_timer_cb(unsigned long arg)
1831 {
1832 	fib6_run_gc(0, (struct net *)arg, true);
1833 }
1834 
1835 static int __net_init fib6_net_init(struct net *net)
1836 {
1837 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1838 
1839 	spin_lock_init(&net->ipv6.fib6_gc_lock);
1840 	rwlock_init(&net->ipv6.fib6_walker_lock);
1841 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
1842 	setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1843 
1844 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1845 	if (!net->ipv6.rt6_stats)
1846 		goto out_timer;
1847 
1848 	/* Avoid false sharing : Use at least a full cache line */
1849 	size = max_t(size_t, size, L1_CACHE_BYTES);
1850 
1851 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1852 	if (!net->ipv6.fib_table_hash)
1853 		goto out_rt6_stats;
1854 
1855 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1856 					  GFP_KERNEL);
1857 	if (!net->ipv6.fib6_main_tbl)
1858 		goto out_fib_table_hash;
1859 
1860 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1861 	net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1862 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1863 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1864 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1865 
1866 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1867 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1868 					   GFP_KERNEL);
1869 	if (!net->ipv6.fib6_local_tbl)
1870 		goto out_fib6_main_tbl;
1871 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1872 	net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1873 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1874 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1875 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1876 #endif
1877 	fib6_tables_init(net);
1878 
1879 	return 0;
1880 
1881 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1882 out_fib6_main_tbl:
1883 	kfree(net->ipv6.fib6_main_tbl);
1884 #endif
1885 out_fib_table_hash:
1886 	kfree(net->ipv6.fib_table_hash);
1887 out_rt6_stats:
1888 	kfree(net->ipv6.rt6_stats);
1889 out_timer:
1890 	return -ENOMEM;
1891 }
1892 
1893 static void fib6_net_exit(struct net *net)
1894 {
1895 	rt6_ifdown(net, NULL);
1896 	del_timer_sync(&net->ipv6.ip6_fib_timer);
1897 
1898 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1899 	inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1900 	kfree(net->ipv6.fib6_local_tbl);
1901 #endif
1902 	inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1903 	kfree(net->ipv6.fib6_main_tbl);
1904 	kfree(net->ipv6.fib_table_hash);
1905 	kfree(net->ipv6.rt6_stats);
1906 }
1907 
1908 static struct pernet_operations fib6_net_ops = {
1909 	.init = fib6_net_init,
1910 	.exit = fib6_net_exit,
1911 };
1912 
1913 int __init fib6_init(void)
1914 {
1915 	int ret = -ENOMEM;
1916 
1917 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
1918 					   sizeof(struct fib6_node),
1919 					   0, SLAB_HWCACHE_ALIGN,
1920 					   NULL);
1921 	if (!fib6_node_kmem)
1922 		goto out;
1923 
1924 	ret = register_pernet_subsys(&fib6_net_ops);
1925 	if (ret)
1926 		goto out_kmem_cache_create;
1927 
1928 	ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1929 			      NULL);
1930 	if (ret)
1931 		goto out_unregister_subsys;
1932 
1933 	__fib6_flush_trees = fib6_flush_trees;
1934 out:
1935 	return ret;
1936 
1937 out_unregister_subsys:
1938 	unregister_pernet_subsys(&fib6_net_ops);
1939 out_kmem_cache_create:
1940 	kmem_cache_destroy(fib6_node_kmem);
1941 	goto out;
1942 }
1943 
1944 void fib6_gc_cleanup(void)
1945 {
1946 	unregister_pernet_subsys(&fib6_net_ops);
1947 	kmem_cache_destroy(fib6_node_kmem);
1948 }
1949 
1950 #ifdef CONFIG_PROC_FS
1951 
1952 struct ipv6_route_iter {
1953 	struct seq_net_private p;
1954 	struct fib6_walker w;
1955 	loff_t skip;
1956 	struct fib6_table *tbl;
1957 	int sernum;
1958 };
1959 
1960 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1961 {
1962 	struct rt6_info *rt = v;
1963 	struct ipv6_route_iter *iter = seq->private;
1964 
1965 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1966 
1967 #ifdef CONFIG_IPV6_SUBTREES
1968 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1969 #else
1970 	seq_puts(seq, "00000000000000000000000000000000 00 ");
1971 #endif
1972 	if (rt->rt6i_flags & RTF_GATEWAY)
1973 		seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1974 	else
1975 		seq_puts(seq, "00000000000000000000000000000000");
1976 
1977 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1978 		   rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1979 		   rt->dst.__use, rt->rt6i_flags,
1980 		   rt->dst.dev ? rt->dst.dev->name : "");
1981 	iter->w.leaf = NULL;
1982 	return 0;
1983 }
1984 
1985 static int ipv6_route_yield(struct fib6_walker *w)
1986 {
1987 	struct ipv6_route_iter *iter = w->args;
1988 
1989 	if (!iter->skip)
1990 		return 1;
1991 
1992 	do {
1993 		iter->w.leaf = iter->w.leaf->dst.rt6_next;
1994 		iter->skip--;
1995 		if (!iter->skip && iter->w.leaf)
1996 			return 1;
1997 	} while (iter->w.leaf);
1998 
1999 	return 0;
2000 }
2001 
2002 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2003 				      struct net *net)
2004 {
2005 	memset(&iter->w, 0, sizeof(iter->w));
2006 	iter->w.func = ipv6_route_yield;
2007 	iter->w.root = &iter->tbl->tb6_root;
2008 	iter->w.state = FWS_INIT;
2009 	iter->w.node = iter->w.root;
2010 	iter->w.args = iter;
2011 	iter->sernum = iter->w.root->fn_sernum;
2012 	INIT_LIST_HEAD(&iter->w.lh);
2013 	fib6_walker_link(net, &iter->w);
2014 }
2015 
2016 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2017 						    struct net *net)
2018 {
2019 	unsigned int h;
2020 	struct hlist_node *node;
2021 
2022 	if (tbl) {
2023 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2024 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2025 	} else {
2026 		h = 0;
2027 		node = NULL;
2028 	}
2029 
2030 	while (!node && h < FIB6_TABLE_HASHSZ) {
2031 		node = rcu_dereference_bh(
2032 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2033 	}
2034 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2035 }
2036 
2037 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2038 {
2039 	if (iter->sernum != iter->w.root->fn_sernum) {
2040 		iter->sernum = iter->w.root->fn_sernum;
2041 		iter->w.state = FWS_INIT;
2042 		iter->w.node = iter->w.root;
2043 		WARN_ON(iter->w.skip);
2044 		iter->w.skip = iter->w.count;
2045 	}
2046 }
2047 
2048 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2049 {
2050 	int r;
2051 	struct rt6_info *n;
2052 	struct net *net = seq_file_net(seq);
2053 	struct ipv6_route_iter *iter = seq->private;
2054 
2055 	if (!v)
2056 		goto iter_table;
2057 
2058 	n = ((struct rt6_info *)v)->dst.rt6_next;
2059 	if (n) {
2060 		++*pos;
2061 		return n;
2062 	}
2063 
2064 iter_table:
2065 	ipv6_route_check_sernum(iter);
2066 	read_lock(&iter->tbl->tb6_lock);
2067 	r = fib6_walk_continue(&iter->w);
2068 	read_unlock(&iter->tbl->tb6_lock);
2069 	if (r > 0) {
2070 		if (v)
2071 			++*pos;
2072 		return iter->w.leaf;
2073 	} else if (r < 0) {
2074 		fib6_walker_unlink(net, &iter->w);
2075 		return NULL;
2076 	}
2077 	fib6_walker_unlink(net, &iter->w);
2078 
2079 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2080 	if (!iter->tbl)
2081 		return NULL;
2082 
2083 	ipv6_route_seq_setup_walk(iter, net);
2084 	goto iter_table;
2085 }
2086 
2087 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2088 	__acquires(RCU_BH)
2089 {
2090 	struct net *net = seq_file_net(seq);
2091 	struct ipv6_route_iter *iter = seq->private;
2092 
2093 	rcu_read_lock_bh();
2094 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2095 	iter->skip = *pos;
2096 
2097 	if (iter->tbl) {
2098 		ipv6_route_seq_setup_walk(iter, net);
2099 		return ipv6_route_seq_next(seq, NULL, pos);
2100 	} else {
2101 		return NULL;
2102 	}
2103 }
2104 
2105 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2106 {
2107 	struct fib6_walker *w = &iter->w;
2108 	return w->node && !(w->state == FWS_U && w->node == w->root);
2109 }
2110 
2111 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2112 	__releases(RCU_BH)
2113 {
2114 	struct net *net = seq_file_net(seq);
2115 	struct ipv6_route_iter *iter = seq->private;
2116 
2117 	if (ipv6_route_iter_active(iter))
2118 		fib6_walker_unlink(net, &iter->w);
2119 
2120 	rcu_read_unlock_bh();
2121 }
2122 
2123 static const struct seq_operations ipv6_route_seq_ops = {
2124 	.start	= ipv6_route_seq_start,
2125 	.next	= ipv6_route_seq_next,
2126 	.stop	= ipv6_route_seq_stop,
2127 	.show	= ipv6_route_seq_show
2128 };
2129 
2130 int ipv6_route_open(struct inode *inode, struct file *file)
2131 {
2132 	return seq_open_net(inode, file, &ipv6_route_seq_ops,
2133 			    sizeof(struct ipv6_route_iter));
2134 }
2135 
2136 #endif /* CONFIG_PROC_FS */
2137