xref: /openbmc/linux/net/ipv6/ip6_fib.c (revision 38fbeeeeccdb38d0635398e8e344d245f6d8dc52)
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 #include <net/fib_notifier.h>
37 
38 #include <net/ip6_fib.h>
39 #include <net/ip6_route.h>
40 
41 static struct kmem_cache *fib6_node_kmem __read_mostly;
42 
43 struct fib6_cleaner {
44 	struct fib6_walker w;
45 	struct net *net;
46 	int (*func)(struct rt6_info *, void *arg);
47 	int sernum;
48 	void *arg;
49 };
50 
51 #ifdef CONFIG_IPV6_SUBTREES
52 #define FWS_INIT FWS_S
53 #else
54 #define FWS_INIT FWS_L
55 #endif
56 
57 static void fib6_prune_clones(struct net *net, struct fib6_node *fn);
58 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
59 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
60 static int fib6_walk(struct net *net, struct fib6_walker *w);
61 static int fib6_walk_continue(struct fib6_walker *w);
62 
63 /*
64  *	A routing update causes an increase of the serial number on the
65  *	affected subtree. This allows for cached routes to be asynchronously
66  *	tested when modifications are made to the destination cache as a
67  *	result of redirects, path MTU changes, etc.
68  */
69 
70 static void fib6_gc_timer_cb(unsigned long arg);
71 
72 #define FOR_WALKERS(net, w) \
73 	list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
74 
75 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
76 {
77 	write_lock_bh(&net->ipv6.fib6_walker_lock);
78 	list_add(&w->lh, &net->ipv6.fib6_walkers);
79 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
80 }
81 
82 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
83 {
84 	write_lock_bh(&net->ipv6.fib6_walker_lock);
85 	list_del(&w->lh);
86 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
87 }
88 
89 static int fib6_new_sernum(struct net *net)
90 {
91 	int new, old;
92 
93 	do {
94 		old = atomic_read(&net->ipv6.fib6_sernum);
95 		new = old < INT_MAX ? old + 1 : 1;
96 	} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
97 				old, new) != old);
98 	return new;
99 }
100 
101 enum {
102 	FIB6_NO_SERNUM_CHANGE = 0,
103 };
104 
105 void fib6_update_sernum(struct rt6_info *rt)
106 {
107 	struct fib6_table *table = rt->rt6i_table;
108 	struct net *net = dev_net(rt->dst.dev);
109 	struct fib6_node *fn;
110 
111 	write_lock_bh(&table->tb6_lock);
112 	fn = rcu_dereference_protected(rt->rt6i_node,
113 			lockdep_is_held(&table->tb6_lock));
114 	if (fn)
115 		fn->fn_sernum = fib6_new_sernum(net);
116 	write_unlock_bh(&table->tb6_lock);
117 }
118 
119 /*
120  *	Auxiliary address test functions for the radix tree.
121  *
122  *	These assume a 32bit processor (although it will work on
123  *	64bit processors)
124  */
125 
126 /*
127  *	test bit
128  */
129 #if defined(__LITTLE_ENDIAN)
130 # define BITOP_BE32_SWIZZLE	(0x1F & ~7)
131 #else
132 # define BITOP_BE32_SWIZZLE	0
133 #endif
134 
135 static __be32 addr_bit_set(const void *token, int fn_bit)
136 {
137 	const __be32 *addr = token;
138 	/*
139 	 * Here,
140 	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
141 	 * is optimized version of
142 	 *	htonl(1 << ((~fn_bit)&0x1F))
143 	 * See include/asm-generic/bitops/le.h.
144 	 */
145 	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
146 	       addr[fn_bit >> 5];
147 }
148 
149 static struct fib6_node *node_alloc(void)
150 {
151 	struct fib6_node *fn;
152 
153 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
154 
155 	return fn;
156 }
157 
158 static void node_free_immediate(struct fib6_node *fn)
159 {
160 	kmem_cache_free(fib6_node_kmem, fn);
161 }
162 
163 static void node_free_rcu(struct rcu_head *head)
164 {
165 	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
166 
167 	kmem_cache_free(fib6_node_kmem, fn);
168 }
169 
170 static void node_free(struct fib6_node *fn)
171 {
172 	call_rcu(&fn->rcu, node_free_rcu);
173 }
174 
175 void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
176 {
177 	int cpu;
178 
179 	if (!non_pcpu_rt->rt6i_pcpu)
180 		return;
181 
182 	for_each_possible_cpu(cpu) {
183 		struct rt6_info **ppcpu_rt;
184 		struct rt6_info *pcpu_rt;
185 
186 		ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
187 		pcpu_rt = *ppcpu_rt;
188 		if (pcpu_rt) {
189 			dst_dev_put(&pcpu_rt->dst);
190 			dst_release(&pcpu_rt->dst);
191 			*ppcpu_rt = NULL;
192 		}
193 	}
194 
195 	free_percpu(non_pcpu_rt->rt6i_pcpu);
196 	non_pcpu_rt->rt6i_pcpu = NULL;
197 }
198 EXPORT_SYMBOL_GPL(rt6_free_pcpu);
199 
200 static void fib6_free_table(struct fib6_table *table)
201 {
202 	inetpeer_invalidate_tree(&table->tb6_peers);
203 	kfree(table);
204 }
205 
206 static void fib6_link_table(struct net *net, struct fib6_table *tb)
207 {
208 	unsigned int h;
209 
210 	/*
211 	 * Initialize table lock at a single place to give lockdep a key,
212 	 * tables aren't visible prior to being linked to the list.
213 	 */
214 	rwlock_init(&tb->tb6_lock);
215 
216 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
217 
218 	/*
219 	 * No protection necessary, this is the only list mutatation
220 	 * operation, tables never disappear once they exist.
221 	 */
222 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
223 }
224 
225 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
226 
227 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
228 {
229 	struct fib6_table *table;
230 
231 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
232 	if (table) {
233 		table->tb6_id = id;
234 		table->tb6_root.leaf = net->ipv6.ip6_null_entry;
235 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
236 		inet_peer_base_init(&table->tb6_peers);
237 	}
238 
239 	return table;
240 }
241 
242 struct fib6_table *fib6_new_table(struct net *net, u32 id)
243 {
244 	struct fib6_table *tb;
245 
246 	if (id == 0)
247 		id = RT6_TABLE_MAIN;
248 	tb = fib6_get_table(net, id);
249 	if (tb)
250 		return tb;
251 
252 	tb = fib6_alloc_table(net, id);
253 	if (tb)
254 		fib6_link_table(net, tb);
255 
256 	return tb;
257 }
258 EXPORT_SYMBOL_GPL(fib6_new_table);
259 
260 struct fib6_table *fib6_get_table(struct net *net, u32 id)
261 {
262 	struct fib6_table *tb;
263 	struct hlist_head *head;
264 	unsigned int h;
265 
266 	if (id == 0)
267 		id = RT6_TABLE_MAIN;
268 	h = id & (FIB6_TABLE_HASHSZ - 1);
269 	rcu_read_lock();
270 	head = &net->ipv6.fib_table_hash[h];
271 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
272 		if (tb->tb6_id == id) {
273 			rcu_read_unlock();
274 			return tb;
275 		}
276 	}
277 	rcu_read_unlock();
278 
279 	return NULL;
280 }
281 EXPORT_SYMBOL_GPL(fib6_get_table);
282 
283 static void __net_init fib6_tables_init(struct net *net)
284 {
285 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
286 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
287 }
288 #else
289 
290 struct fib6_table *fib6_new_table(struct net *net, u32 id)
291 {
292 	return fib6_get_table(net, id);
293 }
294 
295 struct fib6_table *fib6_get_table(struct net *net, u32 id)
296 {
297 	  return net->ipv6.fib6_main_tbl;
298 }
299 
300 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
301 				   int flags, pol_lookup_t lookup)
302 {
303 	struct rt6_info *rt;
304 
305 	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
306 	if (rt->dst.error == -EAGAIN) {
307 		ip6_rt_put(rt);
308 		rt = net->ipv6.ip6_null_entry;
309 		dst_hold(&rt->dst);
310 	}
311 
312 	return &rt->dst;
313 }
314 
315 static void __net_init fib6_tables_init(struct net *net)
316 {
317 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
318 }
319 
320 #endif
321 
322 unsigned int fib6_tables_seq_read(struct net *net)
323 {
324 	unsigned int h, fib_seq = 0;
325 
326 	rcu_read_lock();
327 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
328 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
329 		struct fib6_table *tb;
330 
331 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
332 			read_lock_bh(&tb->tb6_lock);
333 			fib_seq += tb->fib_seq;
334 			read_unlock_bh(&tb->tb6_lock);
335 		}
336 	}
337 	rcu_read_unlock();
338 
339 	return fib_seq;
340 }
341 
342 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
343 				    enum fib_event_type event_type,
344 				    struct rt6_info *rt)
345 {
346 	struct fib6_entry_notifier_info info = {
347 		.rt = rt,
348 	};
349 
350 	return call_fib6_notifier(nb, net, event_type, &info.info);
351 }
352 
353 static int call_fib6_entry_notifiers(struct net *net,
354 				     enum fib_event_type event_type,
355 				     struct rt6_info *rt)
356 {
357 	struct fib6_entry_notifier_info info = {
358 		.rt = rt,
359 	};
360 
361 	rt->rt6i_table->fib_seq++;
362 	return call_fib6_notifiers(net, event_type, &info.info);
363 }
364 
365 struct fib6_dump_arg {
366 	struct net *net;
367 	struct notifier_block *nb;
368 };
369 
370 static void fib6_rt_dump(struct rt6_info *rt, struct fib6_dump_arg *arg)
371 {
372 	if (rt == arg->net->ipv6.ip6_null_entry)
373 		return;
374 	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
375 }
376 
377 static int fib6_node_dump(struct fib6_walker *w)
378 {
379 	struct rt6_info *rt;
380 
381 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next)
382 		fib6_rt_dump(rt, w->args);
383 	w->leaf = NULL;
384 	return 0;
385 }
386 
387 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
388 			    struct fib6_walker *w)
389 {
390 	w->root = &tb->tb6_root;
391 	read_lock_bh(&tb->tb6_lock);
392 	fib6_walk(net, w);
393 	read_unlock_bh(&tb->tb6_lock);
394 }
395 
396 /* Called with rcu_read_lock() */
397 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
398 {
399 	struct fib6_dump_arg arg;
400 	struct fib6_walker *w;
401 	unsigned int h;
402 
403 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
404 	if (!w)
405 		return -ENOMEM;
406 
407 	w->func = fib6_node_dump;
408 	arg.net = net;
409 	arg.nb = nb;
410 	w->args = &arg;
411 
412 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
413 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
414 		struct fib6_table *tb;
415 
416 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
417 			fib6_table_dump(net, tb, w);
418 	}
419 
420 	kfree(w);
421 
422 	return 0;
423 }
424 
425 static int fib6_dump_node(struct fib6_walker *w)
426 {
427 	int res;
428 	struct rt6_info *rt;
429 
430 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
431 		res = rt6_dump_route(rt, w->args);
432 		if (res < 0) {
433 			/* Frame is full, suspend walking */
434 			w->leaf = rt;
435 			return 1;
436 		}
437 
438 		/* Multipath routes are dumped in one route with the
439 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
440 		 * last sibling of this route (no need to dump the
441 		 * sibling routes again)
442 		 */
443 		if (rt->rt6i_nsiblings)
444 			rt = list_last_entry(&rt->rt6i_siblings,
445 					     struct rt6_info,
446 					     rt6i_siblings);
447 	}
448 	w->leaf = NULL;
449 	return 0;
450 }
451 
452 static void fib6_dump_end(struct netlink_callback *cb)
453 {
454 	struct net *net = sock_net(cb->skb->sk);
455 	struct fib6_walker *w = (void *)cb->args[2];
456 
457 	if (w) {
458 		if (cb->args[4]) {
459 			cb->args[4] = 0;
460 			fib6_walker_unlink(net, w);
461 		}
462 		cb->args[2] = 0;
463 		kfree(w);
464 	}
465 	cb->done = (void *)cb->args[3];
466 	cb->args[1] = 3;
467 }
468 
469 static int fib6_dump_done(struct netlink_callback *cb)
470 {
471 	fib6_dump_end(cb);
472 	return cb->done ? cb->done(cb) : 0;
473 }
474 
475 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
476 			   struct netlink_callback *cb)
477 {
478 	struct net *net = sock_net(skb->sk);
479 	struct fib6_walker *w;
480 	int res;
481 
482 	w = (void *)cb->args[2];
483 	w->root = &table->tb6_root;
484 
485 	if (cb->args[4] == 0) {
486 		w->count = 0;
487 		w->skip = 0;
488 
489 		read_lock_bh(&table->tb6_lock);
490 		res = fib6_walk(net, w);
491 		read_unlock_bh(&table->tb6_lock);
492 		if (res > 0) {
493 			cb->args[4] = 1;
494 			cb->args[5] = w->root->fn_sernum;
495 		}
496 	} else {
497 		if (cb->args[5] != w->root->fn_sernum) {
498 			/* Begin at the root if the tree changed */
499 			cb->args[5] = w->root->fn_sernum;
500 			w->state = FWS_INIT;
501 			w->node = w->root;
502 			w->skip = w->count;
503 		} else
504 			w->skip = 0;
505 
506 		read_lock_bh(&table->tb6_lock);
507 		res = fib6_walk_continue(w);
508 		read_unlock_bh(&table->tb6_lock);
509 		if (res <= 0) {
510 			fib6_walker_unlink(net, w);
511 			cb->args[4] = 0;
512 		}
513 	}
514 
515 	return res;
516 }
517 
518 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
519 {
520 	struct net *net = sock_net(skb->sk);
521 	unsigned int h, s_h;
522 	unsigned int e = 0, s_e;
523 	struct rt6_rtnl_dump_arg arg;
524 	struct fib6_walker *w;
525 	struct fib6_table *tb;
526 	struct hlist_head *head;
527 	int res = 0;
528 
529 	s_h = cb->args[0];
530 	s_e = cb->args[1];
531 
532 	w = (void *)cb->args[2];
533 	if (!w) {
534 		/* New dump:
535 		 *
536 		 * 1. hook callback destructor.
537 		 */
538 		cb->args[3] = (long)cb->done;
539 		cb->done = fib6_dump_done;
540 
541 		/*
542 		 * 2. allocate and initialize walker.
543 		 */
544 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
545 		if (!w)
546 			return -ENOMEM;
547 		w->func = fib6_dump_node;
548 		cb->args[2] = (long)w;
549 	}
550 
551 	arg.skb = skb;
552 	arg.cb = cb;
553 	arg.net = net;
554 	w->args = &arg;
555 
556 	rcu_read_lock();
557 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
558 		e = 0;
559 		head = &net->ipv6.fib_table_hash[h];
560 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
561 			if (e < s_e)
562 				goto next;
563 			res = fib6_dump_table(tb, skb, cb);
564 			if (res != 0)
565 				goto out;
566 next:
567 			e++;
568 		}
569 	}
570 out:
571 	rcu_read_unlock();
572 	cb->args[1] = e;
573 	cb->args[0] = h;
574 
575 	res = res < 0 ? res : skb->len;
576 	if (res <= 0)
577 		fib6_dump_end(cb);
578 	return res;
579 }
580 
581 /*
582  *	Routing Table
583  *
584  *	return the appropriate node for a routing tree "add" operation
585  *	by either creating and inserting or by returning an existing
586  *	node.
587  */
588 
589 static struct fib6_node *fib6_add_1(struct fib6_node *root,
590 				     struct in6_addr *addr, int plen,
591 				     int offset, int allow_create,
592 				     int replace_required, int sernum,
593 				     struct netlink_ext_ack *extack)
594 {
595 	struct fib6_node *fn, *in, *ln;
596 	struct fib6_node *pn = NULL;
597 	struct rt6key *key;
598 	int	bit;
599 	__be32	dir = 0;
600 
601 	RT6_TRACE("fib6_add_1\n");
602 
603 	/* insert node in tree */
604 
605 	fn = root;
606 
607 	do {
608 		key = (struct rt6key *)((u8 *)fn->leaf + offset);
609 
610 		/*
611 		 *	Prefix match
612 		 */
613 		if (plen < fn->fn_bit ||
614 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
615 			if (!allow_create) {
616 				if (replace_required) {
617 					NL_SET_ERR_MSG(extack,
618 						       "Can not replace route - no match found");
619 					pr_warn("Can't replace route, no match found\n");
620 					return ERR_PTR(-ENOENT);
621 				}
622 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
623 			}
624 			goto insert_above;
625 		}
626 
627 		/*
628 		 *	Exact match ?
629 		 */
630 
631 		if (plen == fn->fn_bit) {
632 			/* clean up an intermediate node */
633 			if (!(fn->fn_flags & RTN_RTINFO)) {
634 				rt6_release(fn->leaf);
635 				fn->leaf = NULL;
636 			}
637 
638 			fn->fn_sernum = sernum;
639 
640 			return fn;
641 		}
642 
643 		/*
644 		 *	We have more bits to go
645 		 */
646 
647 		/* Try to walk down on tree. */
648 		fn->fn_sernum = sernum;
649 		dir = addr_bit_set(addr, fn->fn_bit);
650 		pn = fn;
651 		fn = dir ? fn->right : fn->left;
652 	} while (fn);
653 
654 	if (!allow_create) {
655 		/* We should not create new node because
656 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
657 		 * I assume it is safe to require NLM_F_CREATE when
658 		 * REPLACE flag is used! Later we may want to remove the
659 		 * check for replace_required, because according
660 		 * to netlink specification, NLM_F_CREATE
661 		 * MUST be specified if new route is created.
662 		 * That would keep IPv6 consistent with IPv4
663 		 */
664 		if (replace_required) {
665 			NL_SET_ERR_MSG(extack,
666 				       "Can not replace route - no match found");
667 			pr_warn("Can't replace route, no match found\n");
668 			return ERR_PTR(-ENOENT);
669 		}
670 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
671 	}
672 	/*
673 	 *	We walked to the bottom of tree.
674 	 *	Create new leaf node without children.
675 	 */
676 
677 	ln = node_alloc();
678 
679 	if (!ln)
680 		return ERR_PTR(-ENOMEM);
681 	ln->fn_bit = plen;
682 
683 	ln->parent = pn;
684 	ln->fn_sernum = sernum;
685 
686 	if (dir)
687 		pn->right = ln;
688 	else
689 		pn->left  = ln;
690 
691 	return ln;
692 
693 
694 insert_above:
695 	/*
696 	 * split since we don't have a common prefix anymore or
697 	 * we have a less significant route.
698 	 * we've to insert an intermediate node on the list
699 	 * this new node will point to the one we need to create
700 	 * and the current
701 	 */
702 
703 	pn = fn->parent;
704 
705 	/* find 1st bit in difference between the 2 addrs.
706 
707 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
708 	   but if it is >= plen, the value is ignored in any case.
709 	 */
710 
711 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
712 
713 	/*
714 	 *		(intermediate)[in]
715 	 *	          /	   \
716 	 *	(new leaf node)[ln] (old node)[fn]
717 	 */
718 	if (plen > bit) {
719 		in = node_alloc();
720 		ln = node_alloc();
721 
722 		if (!in || !ln) {
723 			if (in)
724 				node_free_immediate(in);
725 			if (ln)
726 				node_free_immediate(ln);
727 			return ERR_PTR(-ENOMEM);
728 		}
729 
730 		/*
731 		 * new intermediate node.
732 		 * RTN_RTINFO will
733 		 * be off since that an address that chooses one of
734 		 * the branches would not match less specific routes
735 		 * in the other branch
736 		 */
737 
738 		in->fn_bit = bit;
739 
740 		in->parent = pn;
741 		in->leaf = fn->leaf;
742 		atomic_inc(&in->leaf->rt6i_ref);
743 
744 		in->fn_sernum = sernum;
745 
746 		/* update parent pointer */
747 		if (dir)
748 			pn->right = in;
749 		else
750 			pn->left  = in;
751 
752 		ln->fn_bit = plen;
753 
754 		ln->parent = in;
755 		fn->parent = in;
756 
757 		ln->fn_sernum = sernum;
758 
759 		if (addr_bit_set(addr, bit)) {
760 			in->right = ln;
761 			in->left  = fn;
762 		} else {
763 			in->left  = ln;
764 			in->right = fn;
765 		}
766 	} else { /* plen <= bit */
767 
768 		/*
769 		 *		(new leaf node)[ln]
770 		 *	          /	   \
771 		 *	     (old node)[fn] NULL
772 		 */
773 
774 		ln = node_alloc();
775 
776 		if (!ln)
777 			return ERR_PTR(-ENOMEM);
778 
779 		ln->fn_bit = plen;
780 
781 		ln->parent = pn;
782 
783 		ln->fn_sernum = sernum;
784 
785 		if (dir)
786 			pn->right = ln;
787 		else
788 			pn->left  = ln;
789 
790 		if (addr_bit_set(&key->addr, plen))
791 			ln->right = fn;
792 		else
793 			ln->left  = fn;
794 
795 		fn->parent = ln;
796 	}
797 	return ln;
798 }
799 
800 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
801 {
802 	return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
803 	       RTF_GATEWAY;
804 }
805 
806 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
807 {
808 	int i;
809 
810 	for (i = 0; i < RTAX_MAX; i++) {
811 		if (test_bit(i, mxc->mx_valid))
812 			mp[i] = mxc->mx[i];
813 	}
814 }
815 
816 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
817 {
818 	if (!mxc->mx)
819 		return 0;
820 
821 	if (dst->flags & DST_HOST) {
822 		u32 *mp = dst_metrics_write_ptr(dst);
823 
824 		if (unlikely(!mp))
825 			return -ENOMEM;
826 
827 		fib6_copy_metrics(mp, mxc);
828 	} else {
829 		dst_init_metrics(dst, mxc->mx, false);
830 
831 		/* We've stolen mx now. */
832 		mxc->mx = NULL;
833 	}
834 
835 	return 0;
836 }
837 
838 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
839 			  struct net *net)
840 {
841 	if (atomic_read(&rt->rt6i_ref) != 1) {
842 		/* This route is used as dummy address holder in some split
843 		 * nodes. It is not leaked, but it still holds other resources,
844 		 * which must be released in time. So, scan ascendant nodes
845 		 * and replace dummy references to this route with references
846 		 * to still alive ones.
847 		 */
848 		while (fn) {
849 			if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
850 				fn->leaf = fib6_find_prefix(net, fn);
851 				atomic_inc(&fn->leaf->rt6i_ref);
852 				rt6_release(rt);
853 			}
854 			fn = fn->parent;
855 		}
856 	}
857 }
858 
859 /*
860  *	Insert routing information in a node.
861  */
862 
863 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
864 			    struct nl_info *info, struct mx6_config *mxc)
865 {
866 	struct rt6_info *iter = NULL;
867 	struct rt6_info **ins;
868 	struct rt6_info **fallback_ins = NULL;
869 	int replace = (info->nlh &&
870 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
871 	int add = (!info->nlh ||
872 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
873 	int found = 0;
874 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
875 	u16 nlflags = NLM_F_EXCL;
876 	int err;
877 
878 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
879 		nlflags |= NLM_F_APPEND;
880 
881 	ins = &fn->leaf;
882 
883 	for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
884 		/*
885 		 *	Search for duplicates
886 		 */
887 
888 		if (iter->rt6i_metric == rt->rt6i_metric) {
889 			/*
890 			 *	Same priority level
891 			 */
892 			if (info->nlh &&
893 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
894 				return -EEXIST;
895 
896 			nlflags &= ~NLM_F_EXCL;
897 			if (replace) {
898 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
899 					found++;
900 					break;
901 				}
902 				if (rt_can_ecmp)
903 					fallback_ins = fallback_ins ?: ins;
904 				goto next_iter;
905 			}
906 
907 			if (rt6_duplicate_nexthop(iter, rt)) {
908 				if (rt->rt6i_nsiblings)
909 					rt->rt6i_nsiblings = 0;
910 				if (!(iter->rt6i_flags & RTF_EXPIRES))
911 					return -EEXIST;
912 				if (!(rt->rt6i_flags & RTF_EXPIRES))
913 					rt6_clean_expires(iter);
914 				else
915 					rt6_set_expires(iter, rt->dst.expires);
916 				iter->rt6i_pmtu = rt->rt6i_pmtu;
917 				return -EEXIST;
918 			}
919 			/* If we have the same destination and the same metric,
920 			 * but not the same gateway, then the route we try to
921 			 * add is sibling to this route, increment our counter
922 			 * of siblings, and later we will add our route to the
923 			 * list.
924 			 * Only static routes (which don't have flag
925 			 * RTF_EXPIRES) are used for ECMPv6.
926 			 *
927 			 * To avoid long list, we only had siblings if the
928 			 * route have a gateway.
929 			 */
930 			if (rt_can_ecmp &&
931 			    rt6_qualify_for_ecmp(iter))
932 				rt->rt6i_nsiblings++;
933 		}
934 
935 		if (iter->rt6i_metric > rt->rt6i_metric)
936 			break;
937 
938 next_iter:
939 		ins = &iter->dst.rt6_next;
940 	}
941 
942 	if (fallback_ins && !found) {
943 		/* No ECMP-able route found, replace first non-ECMP one */
944 		ins = fallback_ins;
945 		iter = *ins;
946 		found++;
947 	}
948 
949 	/* Reset round-robin state, if necessary */
950 	if (ins == &fn->leaf)
951 		fn->rr_ptr = NULL;
952 
953 	/* Link this route to others same route. */
954 	if (rt->rt6i_nsiblings) {
955 		unsigned int rt6i_nsiblings;
956 		struct rt6_info *sibling, *temp_sibling;
957 
958 		/* Find the first route that have the same metric */
959 		sibling = fn->leaf;
960 		while (sibling) {
961 			if (sibling->rt6i_metric == rt->rt6i_metric &&
962 			    rt6_qualify_for_ecmp(sibling)) {
963 				list_add_tail(&rt->rt6i_siblings,
964 					      &sibling->rt6i_siblings);
965 				break;
966 			}
967 			sibling = sibling->dst.rt6_next;
968 		}
969 		/* For each sibling in the list, increment the counter of
970 		 * siblings. BUG() if counters does not match, list of siblings
971 		 * is broken!
972 		 */
973 		rt6i_nsiblings = 0;
974 		list_for_each_entry_safe(sibling, temp_sibling,
975 					 &rt->rt6i_siblings, rt6i_siblings) {
976 			sibling->rt6i_nsiblings++;
977 			BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
978 			rt6i_nsiblings++;
979 		}
980 		BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
981 	}
982 
983 	/*
984 	 *	insert node
985 	 */
986 	if (!replace) {
987 		if (!add)
988 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
989 
990 add:
991 		nlflags |= NLM_F_CREATE;
992 		err = fib6_commit_metrics(&rt->dst, mxc);
993 		if (err)
994 			return err;
995 
996 		rt->dst.rt6_next = iter;
997 		*ins = rt;
998 		rcu_assign_pointer(rt->rt6i_node, fn);
999 		atomic_inc(&rt->rt6i_ref);
1000 		call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_ADD,
1001 					  rt);
1002 		if (!info->skip_notify)
1003 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1004 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1005 
1006 		if (!(fn->fn_flags & RTN_RTINFO)) {
1007 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1008 			fn->fn_flags |= RTN_RTINFO;
1009 		}
1010 
1011 	} else {
1012 		int nsiblings;
1013 
1014 		if (!found) {
1015 			if (add)
1016 				goto add;
1017 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1018 			return -ENOENT;
1019 		}
1020 
1021 		err = fib6_commit_metrics(&rt->dst, mxc);
1022 		if (err)
1023 			return err;
1024 
1025 		*ins = rt;
1026 		rcu_assign_pointer(rt->rt6i_node, fn);
1027 		rt->dst.rt6_next = iter->dst.rt6_next;
1028 		atomic_inc(&rt->rt6i_ref);
1029 		call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_REPLACE,
1030 					  rt);
1031 		if (!info->skip_notify)
1032 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1033 		if (!(fn->fn_flags & RTN_RTINFO)) {
1034 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1035 			fn->fn_flags |= RTN_RTINFO;
1036 		}
1037 		nsiblings = iter->rt6i_nsiblings;
1038 		iter->rt6i_node = NULL;
1039 		fib6_purge_rt(iter, fn, info->nl_net);
1040 		if (fn->rr_ptr == iter)
1041 			fn->rr_ptr = NULL;
1042 		rt6_release(iter);
1043 
1044 		if (nsiblings) {
1045 			/* Replacing an ECMP route, remove all siblings */
1046 			ins = &rt->dst.rt6_next;
1047 			iter = *ins;
1048 			while (iter) {
1049 				if (iter->rt6i_metric > rt->rt6i_metric)
1050 					break;
1051 				if (rt6_qualify_for_ecmp(iter)) {
1052 					*ins = iter->dst.rt6_next;
1053 					iter->rt6i_node = NULL;
1054 					fib6_purge_rt(iter, fn, info->nl_net);
1055 					if (fn->rr_ptr == iter)
1056 						fn->rr_ptr = NULL;
1057 					rt6_release(iter);
1058 					nsiblings--;
1059 				} else {
1060 					ins = &iter->dst.rt6_next;
1061 				}
1062 				iter = *ins;
1063 			}
1064 			WARN_ON(nsiblings != 0);
1065 		}
1066 	}
1067 
1068 	return 0;
1069 }
1070 
1071 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
1072 {
1073 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1074 	    (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
1075 		mod_timer(&net->ipv6.ip6_fib_timer,
1076 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1077 }
1078 
1079 void fib6_force_start_gc(struct net *net)
1080 {
1081 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1082 		mod_timer(&net->ipv6.ip6_fib_timer,
1083 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1084 }
1085 
1086 /*
1087  *	Add routing information to the routing tree.
1088  *	<destination addr>/<source addr>
1089  *	with source addr info in sub-trees
1090  */
1091 
1092 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
1093 	     struct nl_info *info, struct mx6_config *mxc,
1094 	     struct netlink_ext_ack *extack)
1095 {
1096 	struct fib6_node *fn, *pn = NULL;
1097 	int err = -ENOMEM;
1098 	int allow_create = 1;
1099 	int replace_required = 0;
1100 	int sernum = fib6_new_sernum(info->nl_net);
1101 
1102 	if (WARN_ON_ONCE(!atomic_read(&rt->dst.__refcnt)))
1103 		return -EINVAL;
1104 
1105 	if (info->nlh) {
1106 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1107 			allow_create = 0;
1108 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1109 			replace_required = 1;
1110 	}
1111 	if (!allow_create && !replace_required)
1112 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1113 
1114 	fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
1115 			offsetof(struct rt6_info, rt6i_dst), allow_create,
1116 			replace_required, sernum, extack);
1117 	if (IS_ERR(fn)) {
1118 		err = PTR_ERR(fn);
1119 		fn = NULL;
1120 		goto out;
1121 	}
1122 
1123 	pn = fn;
1124 
1125 #ifdef CONFIG_IPV6_SUBTREES
1126 	if (rt->rt6i_src.plen) {
1127 		struct fib6_node *sn;
1128 
1129 		if (!fn->subtree) {
1130 			struct fib6_node *sfn;
1131 
1132 			/*
1133 			 * Create subtree.
1134 			 *
1135 			 *		fn[main tree]
1136 			 *		|
1137 			 *		sfn[subtree root]
1138 			 *		   \
1139 			 *		    sn[new leaf node]
1140 			 */
1141 
1142 			/* Create subtree root node */
1143 			sfn = node_alloc();
1144 			if (!sfn)
1145 				goto failure;
1146 
1147 			sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
1148 			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1149 			sfn->fn_flags = RTN_ROOT;
1150 			sfn->fn_sernum = sernum;
1151 
1152 			/* Now add the first leaf node to new subtree */
1153 
1154 			sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
1155 					rt->rt6i_src.plen,
1156 					offsetof(struct rt6_info, rt6i_src),
1157 					allow_create, replace_required, sernum,
1158 					extack);
1159 
1160 			if (IS_ERR(sn)) {
1161 				/* If it is failed, discard just allocated
1162 				   root, and then (in failure) stale node
1163 				   in main tree.
1164 				 */
1165 				node_free_immediate(sfn);
1166 				err = PTR_ERR(sn);
1167 				goto failure;
1168 			}
1169 
1170 			/* Now link new subtree to main tree */
1171 			sfn->parent = fn;
1172 			fn->subtree = sfn;
1173 		} else {
1174 			sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1175 					rt->rt6i_src.plen,
1176 					offsetof(struct rt6_info, rt6i_src),
1177 					allow_create, replace_required, sernum,
1178 					extack);
1179 
1180 			if (IS_ERR(sn)) {
1181 				err = PTR_ERR(sn);
1182 				goto failure;
1183 			}
1184 		}
1185 
1186 		if (!fn->leaf) {
1187 			fn->leaf = rt;
1188 			atomic_inc(&rt->rt6i_ref);
1189 		}
1190 		fn = sn;
1191 	}
1192 #endif
1193 
1194 	err = fib6_add_rt2node(fn, rt, info, mxc);
1195 	if (!err) {
1196 		fib6_start_gc(info->nl_net, rt);
1197 		if (!(rt->rt6i_flags & RTF_CACHE))
1198 			fib6_prune_clones(info->nl_net, pn);
1199 	}
1200 
1201 out:
1202 	if (err) {
1203 #ifdef CONFIG_IPV6_SUBTREES
1204 		/*
1205 		 * If fib6_add_1 has cleared the old leaf pointer in the
1206 		 * super-tree leaf node we have to find a new one for it.
1207 		 */
1208 		if (pn != fn && pn->leaf == rt) {
1209 			pn->leaf = NULL;
1210 			atomic_dec(&rt->rt6i_ref);
1211 		}
1212 		if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1213 			pn->leaf = fib6_find_prefix(info->nl_net, pn);
1214 #if RT6_DEBUG >= 2
1215 			if (!pn->leaf) {
1216 				WARN_ON(pn->leaf == NULL);
1217 				pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1218 			}
1219 #endif
1220 			atomic_inc(&pn->leaf->rt6i_ref);
1221 		}
1222 #endif
1223 		goto failure;
1224 	}
1225 	return err;
1226 
1227 failure:
1228 	/* fn->leaf could be NULL if fn is an intermediate node and we
1229 	 * failed to add the new route to it in both subtree creation
1230 	 * failure and fib6_add_rt2node() failure case.
1231 	 * In both cases, fib6_repair_tree() should be called to fix
1232 	 * fn->leaf.
1233 	 */
1234 	if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1235 		fib6_repair_tree(info->nl_net, fn);
1236 	/* Always release dst as dst->__refcnt is guaranteed
1237 	 * to be taken before entering this function
1238 	 */
1239 	dst_release_immediate(&rt->dst);
1240 	return err;
1241 }
1242 
1243 /*
1244  *	Routing tree lookup
1245  *
1246  */
1247 
1248 struct lookup_args {
1249 	int			offset;		/* key offset on rt6_info	*/
1250 	const struct in6_addr	*addr;		/* search key			*/
1251 };
1252 
1253 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1254 				       struct lookup_args *args)
1255 {
1256 	struct fib6_node *fn;
1257 	__be32 dir;
1258 
1259 	if (unlikely(args->offset == 0))
1260 		return NULL;
1261 
1262 	/*
1263 	 *	Descend on a tree
1264 	 */
1265 
1266 	fn = root;
1267 
1268 	for (;;) {
1269 		struct fib6_node *next;
1270 
1271 		dir = addr_bit_set(args->addr, fn->fn_bit);
1272 
1273 		next = dir ? fn->right : fn->left;
1274 
1275 		if (next) {
1276 			fn = next;
1277 			continue;
1278 		}
1279 		break;
1280 	}
1281 
1282 	while (fn) {
1283 		if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1284 			struct rt6key *key;
1285 
1286 			key = (struct rt6key *) ((u8 *) fn->leaf +
1287 						 args->offset);
1288 
1289 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1290 #ifdef CONFIG_IPV6_SUBTREES
1291 				if (fn->subtree) {
1292 					struct fib6_node *sfn;
1293 					sfn = fib6_lookup_1(fn->subtree,
1294 							    args + 1);
1295 					if (!sfn)
1296 						goto backtrack;
1297 					fn = sfn;
1298 				}
1299 #endif
1300 				if (fn->fn_flags & RTN_RTINFO)
1301 					return fn;
1302 			}
1303 		}
1304 #ifdef CONFIG_IPV6_SUBTREES
1305 backtrack:
1306 #endif
1307 		if (fn->fn_flags & RTN_ROOT)
1308 			break;
1309 
1310 		fn = fn->parent;
1311 	}
1312 
1313 	return NULL;
1314 }
1315 
1316 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1317 			      const struct in6_addr *saddr)
1318 {
1319 	struct fib6_node *fn;
1320 	struct lookup_args args[] = {
1321 		{
1322 			.offset = offsetof(struct rt6_info, rt6i_dst),
1323 			.addr = daddr,
1324 		},
1325 #ifdef CONFIG_IPV6_SUBTREES
1326 		{
1327 			.offset = offsetof(struct rt6_info, rt6i_src),
1328 			.addr = saddr,
1329 		},
1330 #endif
1331 		{
1332 			.offset = 0,	/* sentinel */
1333 		}
1334 	};
1335 
1336 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
1337 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1338 		fn = root;
1339 
1340 	return fn;
1341 }
1342 
1343 /*
1344  *	Get node with specified destination prefix (and source prefix,
1345  *	if subtrees are used)
1346  *	exact_match == true means we try to find fn with exact match of
1347  *	the passed in prefix addr
1348  *	exact_match == false means we try to find fn with longest prefix
1349  *	match of the passed in prefix addr. This is useful for finding fn
1350  *	for cached route as it will be stored in the exception table under
1351  *	the node with longest prefix length.
1352  */
1353 
1354 
1355 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1356 				       const struct in6_addr *addr,
1357 				       int plen, int offset,
1358 				       bool exact_match)
1359 {
1360 	struct fib6_node *fn, *prev = NULL;
1361 
1362 	for (fn = root; fn ; ) {
1363 		struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1364 
1365 		/*
1366 		 *	Prefix match
1367 		 */
1368 		if (plen < fn->fn_bit ||
1369 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1370 			goto out;
1371 
1372 		if (plen == fn->fn_bit)
1373 			return fn;
1374 
1375 		prev = fn;
1376 
1377 		/*
1378 		 *	We have more bits to go
1379 		 */
1380 		if (addr_bit_set(addr, fn->fn_bit))
1381 			fn = fn->right;
1382 		else
1383 			fn = fn->left;
1384 	}
1385 out:
1386 	if (exact_match)
1387 		return NULL;
1388 	else
1389 		return prev;
1390 }
1391 
1392 struct fib6_node *fib6_locate(struct fib6_node *root,
1393 			      const struct in6_addr *daddr, int dst_len,
1394 			      const struct in6_addr *saddr, int src_len,
1395 			      bool exact_match)
1396 {
1397 	struct fib6_node *fn;
1398 
1399 	fn = fib6_locate_1(root, daddr, dst_len,
1400 			   offsetof(struct rt6_info, rt6i_dst),
1401 			   exact_match);
1402 
1403 #ifdef CONFIG_IPV6_SUBTREES
1404 	if (src_len) {
1405 		WARN_ON(saddr == NULL);
1406 		if (fn && fn->subtree)
1407 			fn = fib6_locate_1(fn->subtree, saddr, src_len,
1408 					   offsetof(struct rt6_info, rt6i_src),
1409 					   exact_match);
1410 	}
1411 #endif
1412 
1413 	if (fn && fn->fn_flags & RTN_RTINFO)
1414 		return fn;
1415 
1416 	return NULL;
1417 }
1418 
1419 
1420 /*
1421  *	Deletion
1422  *
1423  */
1424 
1425 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1426 {
1427 	if (fn->fn_flags & RTN_ROOT)
1428 		return net->ipv6.ip6_null_entry;
1429 
1430 	while (fn) {
1431 		if (fn->left)
1432 			return fn->left->leaf;
1433 		if (fn->right)
1434 			return fn->right->leaf;
1435 
1436 		fn = FIB6_SUBTREE(fn);
1437 	}
1438 	return NULL;
1439 }
1440 
1441 /*
1442  *	Called to trim the tree of intermediate nodes when possible. "fn"
1443  *	is the node we want to try and remove.
1444  */
1445 
1446 static struct fib6_node *fib6_repair_tree(struct net *net,
1447 					   struct fib6_node *fn)
1448 {
1449 	int children;
1450 	int nstate;
1451 	struct fib6_node *child, *pn;
1452 	struct fib6_walker *w;
1453 	int iter = 0;
1454 
1455 	for (;;) {
1456 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1457 		iter++;
1458 
1459 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1460 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1461 		WARN_ON(fn->leaf);
1462 
1463 		children = 0;
1464 		child = NULL;
1465 		if (fn->right)
1466 			child = fn->right, children |= 1;
1467 		if (fn->left)
1468 			child = fn->left, children |= 2;
1469 
1470 		if (children == 3 || FIB6_SUBTREE(fn)
1471 #ifdef CONFIG_IPV6_SUBTREES
1472 		    /* Subtree root (i.e. fn) may have one child */
1473 		    || (children && fn->fn_flags & RTN_ROOT)
1474 #endif
1475 		    ) {
1476 			fn->leaf = fib6_find_prefix(net, fn);
1477 #if RT6_DEBUG >= 2
1478 			if (!fn->leaf) {
1479 				WARN_ON(!fn->leaf);
1480 				fn->leaf = net->ipv6.ip6_null_entry;
1481 			}
1482 #endif
1483 			atomic_inc(&fn->leaf->rt6i_ref);
1484 			return fn->parent;
1485 		}
1486 
1487 		pn = fn->parent;
1488 #ifdef CONFIG_IPV6_SUBTREES
1489 		if (FIB6_SUBTREE(pn) == fn) {
1490 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1491 			FIB6_SUBTREE(pn) = NULL;
1492 			nstate = FWS_L;
1493 		} else {
1494 			WARN_ON(fn->fn_flags & RTN_ROOT);
1495 #endif
1496 			if (pn->right == fn)
1497 				pn->right = child;
1498 			else if (pn->left == fn)
1499 				pn->left = child;
1500 #if RT6_DEBUG >= 2
1501 			else
1502 				WARN_ON(1);
1503 #endif
1504 			if (child)
1505 				child->parent = pn;
1506 			nstate = FWS_R;
1507 #ifdef CONFIG_IPV6_SUBTREES
1508 		}
1509 #endif
1510 
1511 		read_lock(&net->ipv6.fib6_walker_lock);
1512 		FOR_WALKERS(net, w) {
1513 			if (!child) {
1514 				if (w->root == fn) {
1515 					w->root = w->node = NULL;
1516 					RT6_TRACE("W %p adjusted by delroot 1\n", w);
1517 				} else if (w->node == fn) {
1518 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1519 					w->node = pn;
1520 					w->state = nstate;
1521 				}
1522 			} else {
1523 				if (w->root == fn) {
1524 					w->root = child;
1525 					RT6_TRACE("W %p adjusted by delroot 2\n", w);
1526 				}
1527 				if (w->node == fn) {
1528 					w->node = child;
1529 					if (children&2) {
1530 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1531 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1532 					} else {
1533 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1534 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1535 					}
1536 				}
1537 			}
1538 		}
1539 		read_unlock(&net->ipv6.fib6_walker_lock);
1540 
1541 		node_free(fn);
1542 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1543 			return pn;
1544 
1545 		rt6_release(pn->leaf);
1546 		pn->leaf = NULL;
1547 		fn = pn;
1548 	}
1549 }
1550 
1551 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1552 			   struct nl_info *info)
1553 {
1554 	struct fib6_walker *w;
1555 	struct rt6_info *rt = *rtp;
1556 	struct net *net = info->nl_net;
1557 
1558 	RT6_TRACE("fib6_del_route\n");
1559 
1560 	/* Unlink it */
1561 	*rtp = rt->dst.rt6_next;
1562 	rt->rt6i_node = NULL;
1563 	net->ipv6.rt6_stats->fib_rt_entries--;
1564 	net->ipv6.rt6_stats->fib_discarded_routes++;
1565 
1566 	/* Reset round-robin state, if necessary */
1567 	if (fn->rr_ptr == rt)
1568 		fn->rr_ptr = NULL;
1569 
1570 	/* Remove this entry from other siblings */
1571 	if (rt->rt6i_nsiblings) {
1572 		struct rt6_info *sibling, *next_sibling;
1573 
1574 		list_for_each_entry_safe(sibling, next_sibling,
1575 					 &rt->rt6i_siblings, rt6i_siblings)
1576 			sibling->rt6i_nsiblings--;
1577 		rt->rt6i_nsiblings = 0;
1578 		list_del_init(&rt->rt6i_siblings);
1579 	}
1580 
1581 	/* Adjust walkers */
1582 	read_lock(&net->ipv6.fib6_walker_lock);
1583 	FOR_WALKERS(net, w) {
1584 		if (w->state == FWS_C && w->leaf == rt) {
1585 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1586 			w->leaf = rt->dst.rt6_next;
1587 			if (!w->leaf)
1588 				w->state = FWS_U;
1589 		}
1590 	}
1591 	read_unlock(&net->ipv6.fib6_walker_lock);
1592 
1593 	rt->dst.rt6_next = NULL;
1594 
1595 	/* If it was last route, expunge its radix tree node */
1596 	if (!fn->leaf) {
1597 		fn->fn_flags &= ~RTN_RTINFO;
1598 		net->ipv6.rt6_stats->fib_route_nodes--;
1599 		fn = fib6_repair_tree(net, fn);
1600 	}
1601 
1602 	fib6_purge_rt(rt, fn, net);
1603 
1604 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt);
1605 	if (!info->skip_notify)
1606 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1607 	rt6_release(rt);
1608 }
1609 
1610 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1611 {
1612 	struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1613 				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
1614 	struct net *net = info->nl_net;
1615 	struct rt6_info **rtp;
1616 
1617 #if RT6_DEBUG >= 2
1618 	if (rt->dst.obsolete > 0) {
1619 		WARN_ON(fn);
1620 		return -ENOENT;
1621 	}
1622 #endif
1623 	if (!fn || rt == net->ipv6.ip6_null_entry)
1624 		return -ENOENT;
1625 
1626 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1627 
1628 	if (!(rt->rt6i_flags & RTF_CACHE)) {
1629 		struct fib6_node *pn = fn;
1630 #ifdef CONFIG_IPV6_SUBTREES
1631 		/* clones of this route might be in another subtree */
1632 		if (rt->rt6i_src.plen) {
1633 			while (!(pn->fn_flags & RTN_ROOT))
1634 				pn = pn->parent;
1635 			pn = pn->parent;
1636 		}
1637 #endif
1638 		fib6_prune_clones(info->nl_net, pn);
1639 	}
1640 
1641 	/*
1642 	 *	Walk the leaf entries looking for ourself
1643 	 */
1644 
1645 	for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1646 		if (*rtp == rt) {
1647 			fib6_del_route(fn, rtp, info);
1648 			return 0;
1649 		}
1650 	}
1651 	return -ENOENT;
1652 }
1653 
1654 /*
1655  *	Tree traversal function.
1656  *
1657  *	Certainly, it is not interrupt safe.
1658  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1659  *	It means, that we can modify tree during walking
1660  *	and use this function for garbage collection, clone pruning,
1661  *	cleaning tree when a device goes down etc. etc.
1662  *
1663  *	It guarantees that every node will be traversed,
1664  *	and that it will be traversed only once.
1665  *
1666  *	Callback function w->func may return:
1667  *	0 -> continue walking.
1668  *	positive value -> walking is suspended (used by tree dumps,
1669  *	and probably by gc, if it will be split to several slices)
1670  *	negative value -> terminate walking.
1671  *
1672  *	The function itself returns:
1673  *	0   -> walk is complete.
1674  *	>0  -> walk is incomplete (i.e. suspended)
1675  *	<0  -> walk is terminated by an error.
1676  */
1677 
1678 static int fib6_walk_continue(struct fib6_walker *w)
1679 {
1680 	struct fib6_node *fn, *pn;
1681 
1682 	for (;;) {
1683 		fn = w->node;
1684 		if (!fn)
1685 			return 0;
1686 
1687 		if (w->prune && fn != w->root &&
1688 		    fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1689 			w->state = FWS_C;
1690 			w->leaf = fn->leaf;
1691 		}
1692 		switch (w->state) {
1693 #ifdef CONFIG_IPV6_SUBTREES
1694 		case FWS_S:
1695 			if (FIB6_SUBTREE(fn)) {
1696 				w->node = FIB6_SUBTREE(fn);
1697 				continue;
1698 			}
1699 			w->state = FWS_L;
1700 #endif
1701 		case FWS_L:
1702 			if (fn->left) {
1703 				w->node = fn->left;
1704 				w->state = FWS_INIT;
1705 				continue;
1706 			}
1707 			w->state = FWS_R;
1708 		case FWS_R:
1709 			if (fn->right) {
1710 				w->node = fn->right;
1711 				w->state = FWS_INIT;
1712 				continue;
1713 			}
1714 			w->state = FWS_C;
1715 			w->leaf = fn->leaf;
1716 		case FWS_C:
1717 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1718 				int err;
1719 
1720 				if (w->skip) {
1721 					w->skip--;
1722 					goto skip;
1723 				}
1724 
1725 				err = w->func(w);
1726 				if (err)
1727 					return err;
1728 
1729 				w->count++;
1730 				continue;
1731 			}
1732 skip:
1733 			w->state = FWS_U;
1734 		case FWS_U:
1735 			if (fn == w->root)
1736 				return 0;
1737 			pn = fn->parent;
1738 			w->node = pn;
1739 #ifdef CONFIG_IPV6_SUBTREES
1740 			if (FIB6_SUBTREE(pn) == fn) {
1741 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1742 				w->state = FWS_L;
1743 				continue;
1744 			}
1745 #endif
1746 			if (pn->left == fn) {
1747 				w->state = FWS_R;
1748 				continue;
1749 			}
1750 			if (pn->right == fn) {
1751 				w->state = FWS_C;
1752 				w->leaf = w->node->leaf;
1753 				continue;
1754 			}
1755 #if RT6_DEBUG >= 2
1756 			WARN_ON(1);
1757 #endif
1758 		}
1759 	}
1760 }
1761 
1762 static int fib6_walk(struct net *net, struct fib6_walker *w)
1763 {
1764 	int res;
1765 
1766 	w->state = FWS_INIT;
1767 	w->node = w->root;
1768 
1769 	fib6_walker_link(net, w);
1770 	res = fib6_walk_continue(w);
1771 	if (res <= 0)
1772 		fib6_walker_unlink(net, w);
1773 	return res;
1774 }
1775 
1776 static int fib6_clean_node(struct fib6_walker *w)
1777 {
1778 	int res;
1779 	struct rt6_info *rt;
1780 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1781 	struct nl_info info = {
1782 		.nl_net = c->net,
1783 	};
1784 
1785 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1786 	    w->node->fn_sernum != c->sernum)
1787 		w->node->fn_sernum = c->sernum;
1788 
1789 	if (!c->func) {
1790 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1791 		w->leaf = NULL;
1792 		return 0;
1793 	}
1794 
1795 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1796 		res = c->func(rt, c->arg);
1797 		if (res < 0) {
1798 			w->leaf = rt;
1799 			res = fib6_del(rt, &info);
1800 			if (res) {
1801 #if RT6_DEBUG >= 2
1802 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1803 					 __func__, rt,
1804 					 rcu_access_pointer(rt->rt6i_node),
1805 					 res);
1806 #endif
1807 				continue;
1808 			}
1809 			return 0;
1810 		}
1811 		WARN_ON(res != 0);
1812 	}
1813 	w->leaf = rt;
1814 	return 0;
1815 }
1816 
1817 /*
1818  *	Convenient frontend to tree walker.
1819  *
1820  *	func is called on each route.
1821  *		It may return -1 -> delete this route.
1822  *		              0  -> continue walking
1823  *
1824  *	prune==1 -> only immediate children of node (certainly,
1825  *	ignoring pure split nodes) will be scanned.
1826  */
1827 
1828 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1829 			    int (*func)(struct rt6_info *, void *arg),
1830 			    bool prune, int sernum, void *arg)
1831 {
1832 	struct fib6_cleaner c;
1833 
1834 	c.w.root = root;
1835 	c.w.func = fib6_clean_node;
1836 	c.w.prune = prune;
1837 	c.w.count = 0;
1838 	c.w.skip = 0;
1839 	c.func = func;
1840 	c.sernum = sernum;
1841 	c.arg = arg;
1842 	c.net = net;
1843 
1844 	fib6_walk(net, &c.w);
1845 }
1846 
1847 static void __fib6_clean_all(struct net *net,
1848 			     int (*func)(struct rt6_info *, void *),
1849 			     int sernum, void *arg)
1850 {
1851 	struct fib6_table *table;
1852 	struct hlist_head *head;
1853 	unsigned int h;
1854 
1855 	rcu_read_lock();
1856 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1857 		head = &net->ipv6.fib_table_hash[h];
1858 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1859 			write_lock_bh(&table->tb6_lock);
1860 			fib6_clean_tree(net, &table->tb6_root,
1861 					func, false, sernum, arg);
1862 			write_unlock_bh(&table->tb6_lock);
1863 		}
1864 	}
1865 	rcu_read_unlock();
1866 }
1867 
1868 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1869 		    void *arg)
1870 {
1871 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1872 }
1873 
1874 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1875 {
1876 	if (rt->rt6i_flags & RTF_CACHE) {
1877 		RT6_TRACE("pruning clone %p\n", rt);
1878 		return -1;
1879 	}
1880 
1881 	return 0;
1882 }
1883 
1884 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1885 {
1886 	fib6_clean_tree(net, fn, fib6_prune_clone, true,
1887 			FIB6_NO_SERNUM_CHANGE, NULL);
1888 }
1889 
1890 static void fib6_flush_trees(struct net *net)
1891 {
1892 	int new_sernum = fib6_new_sernum(net);
1893 
1894 	__fib6_clean_all(net, NULL, new_sernum, NULL);
1895 }
1896 
1897 /*
1898  *	Garbage collection
1899  */
1900 
1901 static int fib6_age(struct rt6_info *rt, void *arg)
1902 {
1903 	struct fib6_gc_args *gc_args = arg;
1904 	unsigned long now = jiffies;
1905 
1906 	/*
1907 	 *	check addrconf expiration here.
1908 	 *	Routes are expired even if they are in use.
1909 	 */
1910 
1911 	if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1912 		if (time_after(now, rt->dst.expires)) {
1913 			RT6_TRACE("expiring %p\n", rt);
1914 			return -1;
1915 		}
1916 		gc_args->more++;
1917 	/* The following part will soon be removed when the exception
1918 	 * table is hooked up to store all cached routes.
1919 	 */
1920 	} else if (rt->rt6i_flags & RTF_CACHE) {
1921 		if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout))
1922 			rt->dst.obsolete = DST_OBSOLETE_KILL;
1923 		if (atomic_read(&rt->dst.__refcnt) == 1 &&
1924 		    rt->dst.obsolete == DST_OBSOLETE_KILL) {
1925 			RT6_TRACE("aging clone %p\n", rt);
1926 			return -1;
1927 		} else if (rt->rt6i_flags & RTF_GATEWAY) {
1928 			struct neighbour *neigh;
1929 			__u8 neigh_flags = 0;
1930 
1931 			neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1932 			if (neigh) {
1933 				neigh_flags = neigh->flags;
1934 				neigh_release(neigh);
1935 			}
1936 			if (!(neigh_flags & NTF_ROUTER)) {
1937 				RT6_TRACE("purging route %p via non-router but gateway\n",
1938 					  rt);
1939 				return -1;
1940 			}
1941 		}
1942 		gc_args->more++;
1943 	}
1944 
1945 	/*	Also age clones in the exception table.
1946 	 *	Note, that clones are aged out
1947 	 *	only if they are not in use now.
1948 	 */
1949 	rt6_age_exceptions(rt, gc_args, now);
1950 
1951 	return 0;
1952 }
1953 
1954 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1955 {
1956 	struct fib6_gc_args gc_args;
1957 	unsigned long now;
1958 
1959 	if (force) {
1960 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
1961 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
1962 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1963 		return;
1964 	}
1965 	gc_args.timeout = expires ? (int)expires :
1966 			  net->ipv6.sysctl.ip6_rt_gc_interval;
1967 	gc_args.more = 0;
1968 
1969 	fib6_clean_all(net, fib6_age, &gc_args);
1970 	now = jiffies;
1971 	net->ipv6.ip6_rt_last_gc = now;
1972 
1973 	if (gc_args.more)
1974 		mod_timer(&net->ipv6.ip6_fib_timer,
1975 			  round_jiffies(now
1976 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
1977 	else
1978 		del_timer(&net->ipv6.ip6_fib_timer);
1979 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
1980 }
1981 
1982 static void fib6_gc_timer_cb(unsigned long arg)
1983 {
1984 	fib6_run_gc(0, (struct net *)arg, true);
1985 }
1986 
1987 static int __net_init fib6_net_init(struct net *net)
1988 {
1989 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1990 	int err;
1991 
1992 	err = fib6_notifier_init(net);
1993 	if (err)
1994 		return err;
1995 
1996 	spin_lock_init(&net->ipv6.fib6_gc_lock);
1997 	rwlock_init(&net->ipv6.fib6_walker_lock);
1998 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
1999 	setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
2000 
2001 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2002 	if (!net->ipv6.rt6_stats)
2003 		goto out_timer;
2004 
2005 	/* Avoid false sharing : Use at least a full cache line */
2006 	size = max_t(size_t, size, L1_CACHE_BYTES);
2007 
2008 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2009 	if (!net->ipv6.fib_table_hash)
2010 		goto out_rt6_stats;
2011 
2012 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2013 					  GFP_KERNEL);
2014 	if (!net->ipv6.fib6_main_tbl)
2015 		goto out_fib_table_hash;
2016 
2017 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2018 	net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
2019 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2020 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2021 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2022 
2023 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2024 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2025 					   GFP_KERNEL);
2026 	if (!net->ipv6.fib6_local_tbl)
2027 		goto out_fib6_main_tbl;
2028 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2029 	net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
2030 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2031 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2032 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2033 #endif
2034 	fib6_tables_init(net);
2035 
2036 	return 0;
2037 
2038 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2039 out_fib6_main_tbl:
2040 	kfree(net->ipv6.fib6_main_tbl);
2041 #endif
2042 out_fib_table_hash:
2043 	kfree(net->ipv6.fib_table_hash);
2044 out_rt6_stats:
2045 	kfree(net->ipv6.rt6_stats);
2046 out_timer:
2047 	fib6_notifier_exit(net);
2048 	return -ENOMEM;
2049 }
2050 
2051 static void fib6_net_exit(struct net *net)
2052 {
2053 	unsigned int i;
2054 
2055 	rt6_ifdown(net, NULL);
2056 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2057 
2058 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2059 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2060 		struct hlist_node *tmp;
2061 		struct fib6_table *tb;
2062 
2063 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2064 			hlist_del(&tb->tb6_hlist);
2065 			fib6_free_table(tb);
2066 		}
2067 	}
2068 
2069 	kfree(net->ipv6.fib_table_hash);
2070 	kfree(net->ipv6.rt6_stats);
2071 	fib6_notifier_exit(net);
2072 }
2073 
2074 static struct pernet_operations fib6_net_ops = {
2075 	.init = fib6_net_init,
2076 	.exit = fib6_net_exit,
2077 };
2078 
2079 int __init fib6_init(void)
2080 {
2081 	int ret = -ENOMEM;
2082 
2083 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2084 					   sizeof(struct fib6_node),
2085 					   0, SLAB_HWCACHE_ALIGN,
2086 					   NULL);
2087 	if (!fib6_node_kmem)
2088 		goto out;
2089 
2090 	ret = register_pernet_subsys(&fib6_net_ops);
2091 	if (ret)
2092 		goto out_kmem_cache_create;
2093 
2094 	ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
2095 			      0);
2096 	if (ret)
2097 		goto out_unregister_subsys;
2098 
2099 	__fib6_flush_trees = fib6_flush_trees;
2100 out:
2101 	return ret;
2102 
2103 out_unregister_subsys:
2104 	unregister_pernet_subsys(&fib6_net_ops);
2105 out_kmem_cache_create:
2106 	kmem_cache_destroy(fib6_node_kmem);
2107 	goto out;
2108 }
2109 
2110 void fib6_gc_cleanup(void)
2111 {
2112 	unregister_pernet_subsys(&fib6_net_ops);
2113 	kmem_cache_destroy(fib6_node_kmem);
2114 }
2115 
2116 #ifdef CONFIG_PROC_FS
2117 
2118 struct ipv6_route_iter {
2119 	struct seq_net_private p;
2120 	struct fib6_walker w;
2121 	loff_t skip;
2122 	struct fib6_table *tbl;
2123 	int sernum;
2124 };
2125 
2126 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2127 {
2128 	struct rt6_info *rt = v;
2129 	struct ipv6_route_iter *iter = seq->private;
2130 
2131 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
2132 
2133 #ifdef CONFIG_IPV6_SUBTREES
2134 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
2135 #else
2136 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2137 #endif
2138 	if (rt->rt6i_flags & RTF_GATEWAY)
2139 		seq_printf(seq, "%pi6", &rt->rt6i_gateway);
2140 	else
2141 		seq_puts(seq, "00000000000000000000000000000000");
2142 
2143 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2144 		   rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
2145 		   rt->dst.__use, rt->rt6i_flags,
2146 		   rt->dst.dev ? rt->dst.dev->name : "");
2147 	iter->w.leaf = NULL;
2148 	return 0;
2149 }
2150 
2151 static int ipv6_route_yield(struct fib6_walker *w)
2152 {
2153 	struct ipv6_route_iter *iter = w->args;
2154 
2155 	if (!iter->skip)
2156 		return 1;
2157 
2158 	do {
2159 		iter->w.leaf = iter->w.leaf->dst.rt6_next;
2160 		iter->skip--;
2161 		if (!iter->skip && iter->w.leaf)
2162 			return 1;
2163 	} while (iter->w.leaf);
2164 
2165 	return 0;
2166 }
2167 
2168 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2169 				      struct net *net)
2170 {
2171 	memset(&iter->w, 0, sizeof(iter->w));
2172 	iter->w.func = ipv6_route_yield;
2173 	iter->w.root = &iter->tbl->tb6_root;
2174 	iter->w.state = FWS_INIT;
2175 	iter->w.node = iter->w.root;
2176 	iter->w.args = iter;
2177 	iter->sernum = iter->w.root->fn_sernum;
2178 	INIT_LIST_HEAD(&iter->w.lh);
2179 	fib6_walker_link(net, &iter->w);
2180 }
2181 
2182 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2183 						    struct net *net)
2184 {
2185 	unsigned int h;
2186 	struct hlist_node *node;
2187 
2188 	if (tbl) {
2189 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2190 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2191 	} else {
2192 		h = 0;
2193 		node = NULL;
2194 	}
2195 
2196 	while (!node && h < FIB6_TABLE_HASHSZ) {
2197 		node = rcu_dereference_bh(
2198 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2199 	}
2200 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2201 }
2202 
2203 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2204 {
2205 	if (iter->sernum != iter->w.root->fn_sernum) {
2206 		iter->sernum = iter->w.root->fn_sernum;
2207 		iter->w.state = FWS_INIT;
2208 		iter->w.node = iter->w.root;
2209 		WARN_ON(iter->w.skip);
2210 		iter->w.skip = iter->w.count;
2211 	}
2212 }
2213 
2214 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2215 {
2216 	int r;
2217 	struct rt6_info *n;
2218 	struct net *net = seq_file_net(seq);
2219 	struct ipv6_route_iter *iter = seq->private;
2220 
2221 	if (!v)
2222 		goto iter_table;
2223 
2224 	n = ((struct rt6_info *)v)->dst.rt6_next;
2225 	if (n) {
2226 		++*pos;
2227 		return n;
2228 	}
2229 
2230 iter_table:
2231 	ipv6_route_check_sernum(iter);
2232 	read_lock(&iter->tbl->tb6_lock);
2233 	r = fib6_walk_continue(&iter->w);
2234 	read_unlock(&iter->tbl->tb6_lock);
2235 	if (r > 0) {
2236 		if (v)
2237 			++*pos;
2238 		return iter->w.leaf;
2239 	} else if (r < 0) {
2240 		fib6_walker_unlink(net, &iter->w);
2241 		return NULL;
2242 	}
2243 	fib6_walker_unlink(net, &iter->w);
2244 
2245 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2246 	if (!iter->tbl)
2247 		return NULL;
2248 
2249 	ipv6_route_seq_setup_walk(iter, net);
2250 	goto iter_table;
2251 }
2252 
2253 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2254 	__acquires(RCU_BH)
2255 {
2256 	struct net *net = seq_file_net(seq);
2257 	struct ipv6_route_iter *iter = seq->private;
2258 
2259 	rcu_read_lock_bh();
2260 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2261 	iter->skip = *pos;
2262 
2263 	if (iter->tbl) {
2264 		ipv6_route_seq_setup_walk(iter, net);
2265 		return ipv6_route_seq_next(seq, NULL, pos);
2266 	} else {
2267 		return NULL;
2268 	}
2269 }
2270 
2271 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2272 {
2273 	struct fib6_walker *w = &iter->w;
2274 	return w->node && !(w->state == FWS_U && w->node == w->root);
2275 }
2276 
2277 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2278 	__releases(RCU_BH)
2279 {
2280 	struct net *net = seq_file_net(seq);
2281 	struct ipv6_route_iter *iter = seq->private;
2282 
2283 	if (ipv6_route_iter_active(iter))
2284 		fib6_walker_unlink(net, &iter->w);
2285 
2286 	rcu_read_unlock_bh();
2287 }
2288 
2289 static const struct seq_operations ipv6_route_seq_ops = {
2290 	.start	= ipv6_route_seq_start,
2291 	.next	= ipv6_route_seq_next,
2292 	.stop	= ipv6_route_seq_stop,
2293 	.show	= ipv6_route_seq_show
2294 };
2295 
2296 int ipv6_route_open(struct inode *inode, struct file *file)
2297 {
2298 	return seq_open_net(inode, file, &ipv6_route_seq_ops,
2299 			    sizeof(struct ipv6_route_iter));
2300 }
2301 
2302 #endif /* CONFIG_PROC_FS */
2303