xref: /linux/net/ipv6/ip6_fib.c (revision f412eed9dfdeeb6becd7de2ffe8b5d0a8b3f81ca)
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 fib6_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 struct fib6_info *fib6_find_prefix(struct net *net,
58 					 struct fib6_table *table,
59 					 struct fib6_node *fn);
60 static struct fib6_node *fib6_repair_tree(struct net *net,
61 					  struct fib6_table *table,
62 					  struct fib6_node *fn);
63 static int fib6_walk(struct net *net, struct fib6_walker *w);
64 static int fib6_walk_continue(struct fib6_walker *w);
65 
66 /*
67  *	A routing update causes an increase of the serial number on the
68  *	affected subtree. This allows for cached routes to be asynchronously
69  *	tested when modifications are made to the destination cache as a
70  *	result of redirects, path MTU changes, etc.
71  */
72 
73 static void fib6_gc_timer_cb(struct timer_list *t);
74 
75 #define FOR_WALKERS(net, w) \
76 	list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
77 
78 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
79 {
80 	write_lock_bh(&net->ipv6.fib6_walker_lock);
81 	list_add(&w->lh, &net->ipv6.fib6_walkers);
82 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
83 }
84 
85 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
86 {
87 	write_lock_bh(&net->ipv6.fib6_walker_lock);
88 	list_del(&w->lh);
89 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
90 }
91 
92 static int fib6_new_sernum(struct net *net)
93 {
94 	int new, old;
95 
96 	do {
97 		old = atomic_read(&net->ipv6.fib6_sernum);
98 		new = old < INT_MAX ? old + 1 : 1;
99 	} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
100 				old, new) != old);
101 	return new;
102 }
103 
104 enum {
105 	FIB6_NO_SERNUM_CHANGE = 0,
106 };
107 
108 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
109 {
110 	struct fib6_node *fn;
111 
112 	fn = rcu_dereference_protected(f6i->fib6_node,
113 			lockdep_is_held(&f6i->fib6_table->tb6_lock));
114 	if (fn)
115 		fn->fn_sernum = fib6_new_sernum(net);
116 }
117 
118 /*
119  *	Auxiliary address test functions for the radix tree.
120  *
121  *	These assume a 32bit processor (although it will work on
122  *	64bit processors)
123  */
124 
125 /*
126  *	test bit
127  */
128 #if defined(__LITTLE_ENDIAN)
129 # define BITOP_BE32_SWIZZLE	(0x1F & ~7)
130 #else
131 # define BITOP_BE32_SWIZZLE	0
132 #endif
133 
134 static __be32 addr_bit_set(const void *token, int fn_bit)
135 {
136 	const __be32 *addr = token;
137 	/*
138 	 * Here,
139 	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
140 	 * is optimized version of
141 	 *	htonl(1 << ((~fn_bit)&0x1F))
142 	 * See include/asm-generic/bitops/le.h.
143 	 */
144 	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
145 	       addr[fn_bit >> 5];
146 }
147 
148 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags)
149 {
150 	struct fib6_info *f6i;
151 
152 	f6i = kzalloc(sizeof(*f6i), gfp_flags);
153 	if (!f6i)
154 		return NULL;
155 
156 	f6i->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
157 	if (!f6i->rt6i_pcpu) {
158 		kfree(f6i);
159 		return NULL;
160 	}
161 
162 	INIT_LIST_HEAD(&f6i->fib6_siblings);
163 	f6i->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
164 
165 	atomic_inc(&f6i->fib6_ref);
166 
167 	return f6i;
168 }
169 
170 void fib6_info_destroy(struct fib6_info *f6i)
171 {
172 	struct rt6_exception_bucket *bucket;
173 	struct dst_metrics *m;
174 
175 	WARN_ON(f6i->fib6_node);
176 
177 	bucket = rcu_dereference_protected(f6i->rt6i_exception_bucket, 1);
178 	if (bucket) {
179 		f6i->rt6i_exception_bucket = NULL;
180 		kfree(bucket);
181 	}
182 
183 	if (f6i->rt6i_pcpu) {
184 		int cpu;
185 
186 		for_each_possible_cpu(cpu) {
187 			struct rt6_info **ppcpu_rt;
188 			struct rt6_info *pcpu_rt;
189 
190 			ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
191 			pcpu_rt = *ppcpu_rt;
192 			if (pcpu_rt) {
193 				dst_dev_put(&pcpu_rt->dst);
194 				dst_release(&pcpu_rt->dst);
195 				*ppcpu_rt = NULL;
196 			}
197 		}
198 	}
199 
200 	if (f6i->fib6_nh.nh_dev)
201 		dev_put(f6i->fib6_nh.nh_dev);
202 
203 	m = f6i->fib6_metrics;
204 	if (m != &dst_default_metrics && refcount_dec_and_test(&m->refcnt))
205 		kfree(m);
206 
207 	kfree(f6i);
208 }
209 EXPORT_SYMBOL_GPL(fib6_info_destroy);
210 
211 static struct fib6_node *node_alloc(struct net *net)
212 {
213 	struct fib6_node *fn;
214 
215 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
216 	if (fn)
217 		net->ipv6.rt6_stats->fib_nodes++;
218 
219 	return fn;
220 }
221 
222 static void node_free_immediate(struct net *net, struct fib6_node *fn)
223 {
224 	kmem_cache_free(fib6_node_kmem, fn);
225 	net->ipv6.rt6_stats->fib_nodes--;
226 }
227 
228 static void node_free_rcu(struct rcu_head *head)
229 {
230 	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
231 
232 	kmem_cache_free(fib6_node_kmem, fn);
233 }
234 
235 static void node_free(struct net *net, struct fib6_node *fn)
236 {
237 	call_rcu(&fn->rcu, node_free_rcu);
238 	net->ipv6.rt6_stats->fib_nodes--;
239 }
240 
241 static void fib6_free_table(struct fib6_table *table)
242 {
243 	inetpeer_invalidate_tree(&table->tb6_peers);
244 	kfree(table);
245 }
246 
247 static void fib6_link_table(struct net *net, struct fib6_table *tb)
248 {
249 	unsigned int h;
250 
251 	/*
252 	 * Initialize table lock at a single place to give lockdep a key,
253 	 * tables aren't visible prior to being linked to the list.
254 	 */
255 	spin_lock_init(&tb->tb6_lock);
256 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
257 
258 	/*
259 	 * No protection necessary, this is the only list mutatation
260 	 * operation, tables never disappear once they exist.
261 	 */
262 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
263 }
264 
265 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
266 
267 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
268 {
269 	struct fib6_table *table;
270 
271 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
272 	if (table) {
273 		table->tb6_id = id;
274 		rcu_assign_pointer(table->tb6_root.leaf,
275 				   net->ipv6.fib6_null_entry);
276 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
277 		inet_peer_base_init(&table->tb6_peers);
278 	}
279 
280 	return table;
281 }
282 
283 struct fib6_table *fib6_new_table(struct net *net, u32 id)
284 {
285 	struct fib6_table *tb;
286 
287 	if (id == 0)
288 		id = RT6_TABLE_MAIN;
289 	tb = fib6_get_table(net, id);
290 	if (tb)
291 		return tb;
292 
293 	tb = fib6_alloc_table(net, id);
294 	if (tb)
295 		fib6_link_table(net, tb);
296 
297 	return tb;
298 }
299 EXPORT_SYMBOL_GPL(fib6_new_table);
300 
301 struct fib6_table *fib6_get_table(struct net *net, u32 id)
302 {
303 	struct fib6_table *tb;
304 	struct hlist_head *head;
305 	unsigned int h;
306 
307 	if (id == 0)
308 		id = RT6_TABLE_MAIN;
309 	h = id & (FIB6_TABLE_HASHSZ - 1);
310 	rcu_read_lock();
311 	head = &net->ipv6.fib_table_hash[h];
312 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
313 		if (tb->tb6_id == id) {
314 			rcu_read_unlock();
315 			return tb;
316 		}
317 	}
318 	rcu_read_unlock();
319 
320 	return NULL;
321 }
322 EXPORT_SYMBOL_GPL(fib6_get_table);
323 
324 static void __net_init fib6_tables_init(struct net *net)
325 {
326 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
327 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
328 }
329 #else
330 
331 struct fib6_table *fib6_new_table(struct net *net, u32 id)
332 {
333 	return fib6_get_table(net, id);
334 }
335 
336 struct fib6_table *fib6_get_table(struct net *net, u32 id)
337 {
338 	  return net->ipv6.fib6_main_tbl;
339 }
340 
341 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
342 				   const struct sk_buff *skb,
343 				   int flags, pol_lookup_t lookup)
344 {
345 	struct rt6_info *rt;
346 
347 	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
348 	if (rt->dst.error == -EAGAIN) {
349 		ip6_rt_put(rt);
350 		rt = net->ipv6.ip6_null_entry;
351 		dst_hold(&rt->dst);
352 	}
353 
354 	return &rt->dst;
355 }
356 
357 static void __net_init fib6_tables_init(struct net *net)
358 {
359 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
360 }
361 
362 #endif
363 
364 unsigned int fib6_tables_seq_read(struct net *net)
365 {
366 	unsigned int h, fib_seq = 0;
367 
368 	rcu_read_lock();
369 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
370 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
371 		struct fib6_table *tb;
372 
373 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
374 			fib_seq += tb->fib_seq;
375 	}
376 	rcu_read_unlock();
377 
378 	return fib_seq;
379 }
380 
381 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
382 				    enum fib_event_type event_type,
383 				    struct fib6_info *rt)
384 {
385 	struct fib6_entry_notifier_info info = {
386 		.rt = rt,
387 	};
388 
389 	return call_fib6_notifier(nb, net, event_type, &info.info);
390 }
391 
392 static int call_fib6_entry_notifiers(struct net *net,
393 				     enum fib_event_type event_type,
394 				     struct fib6_info *rt,
395 				     struct netlink_ext_ack *extack)
396 {
397 	struct fib6_entry_notifier_info info = {
398 		.info.extack = extack,
399 		.rt = rt,
400 	};
401 
402 	rt->fib6_table->fib_seq++;
403 	return call_fib6_notifiers(net, event_type, &info.info);
404 }
405 
406 struct fib6_dump_arg {
407 	struct net *net;
408 	struct notifier_block *nb;
409 };
410 
411 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
412 {
413 	if (rt == arg->net->ipv6.fib6_null_entry)
414 		return;
415 	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
416 }
417 
418 static int fib6_node_dump(struct fib6_walker *w)
419 {
420 	struct fib6_info *rt;
421 
422 	for_each_fib6_walker_rt(w)
423 		fib6_rt_dump(rt, w->args);
424 	w->leaf = NULL;
425 	return 0;
426 }
427 
428 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
429 			    struct fib6_walker *w)
430 {
431 	w->root = &tb->tb6_root;
432 	spin_lock_bh(&tb->tb6_lock);
433 	fib6_walk(net, w);
434 	spin_unlock_bh(&tb->tb6_lock);
435 }
436 
437 /* Called with rcu_read_lock() */
438 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
439 {
440 	struct fib6_dump_arg arg;
441 	struct fib6_walker *w;
442 	unsigned int h;
443 
444 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
445 	if (!w)
446 		return -ENOMEM;
447 
448 	w->func = fib6_node_dump;
449 	arg.net = net;
450 	arg.nb = nb;
451 	w->args = &arg;
452 
453 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
454 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
455 		struct fib6_table *tb;
456 
457 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
458 			fib6_table_dump(net, tb, w);
459 	}
460 
461 	kfree(w);
462 
463 	return 0;
464 }
465 
466 static int fib6_dump_node(struct fib6_walker *w)
467 {
468 	int res;
469 	struct fib6_info *rt;
470 
471 	for_each_fib6_walker_rt(w) {
472 		res = rt6_dump_route(rt, w->args);
473 		if (res < 0) {
474 			/* Frame is full, suspend walking */
475 			w->leaf = rt;
476 			return 1;
477 		}
478 
479 		/* Multipath routes are dumped in one route with the
480 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
481 		 * last sibling of this route (no need to dump the
482 		 * sibling routes again)
483 		 */
484 		if (rt->fib6_nsiblings)
485 			rt = list_last_entry(&rt->fib6_siblings,
486 					     struct fib6_info,
487 					     fib6_siblings);
488 	}
489 	w->leaf = NULL;
490 	return 0;
491 }
492 
493 static void fib6_dump_end(struct netlink_callback *cb)
494 {
495 	struct net *net = sock_net(cb->skb->sk);
496 	struct fib6_walker *w = (void *)cb->args[2];
497 
498 	if (w) {
499 		if (cb->args[4]) {
500 			cb->args[4] = 0;
501 			fib6_walker_unlink(net, w);
502 		}
503 		cb->args[2] = 0;
504 		kfree(w);
505 	}
506 	cb->done = (void *)cb->args[3];
507 	cb->args[1] = 3;
508 }
509 
510 static int fib6_dump_done(struct netlink_callback *cb)
511 {
512 	fib6_dump_end(cb);
513 	return cb->done ? cb->done(cb) : 0;
514 }
515 
516 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
517 			   struct netlink_callback *cb)
518 {
519 	struct net *net = sock_net(skb->sk);
520 	struct fib6_walker *w;
521 	int res;
522 
523 	w = (void *)cb->args[2];
524 	w->root = &table->tb6_root;
525 
526 	if (cb->args[4] == 0) {
527 		w->count = 0;
528 		w->skip = 0;
529 
530 		spin_lock_bh(&table->tb6_lock);
531 		res = fib6_walk(net, w);
532 		spin_unlock_bh(&table->tb6_lock);
533 		if (res > 0) {
534 			cb->args[4] = 1;
535 			cb->args[5] = w->root->fn_sernum;
536 		}
537 	} else {
538 		if (cb->args[5] != w->root->fn_sernum) {
539 			/* Begin at the root if the tree changed */
540 			cb->args[5] = w->root->fn_sernum;
541 			w->state = FWS_INIT;
542 			w->node = w->root;
543 			w->skip = w->count;
544 		} else
545 			w->skip = 0;
546 
547 		spin_lock_bh(&table->tb6_lock);
548 		res = fib6_walk_continue(w);
549 		spin_unlock_bh(&table->tb6_lock);
550 		if (res <= 0) {
551 			fib6_walker_unlink(net, w);
552 			cb->args[4] = 0;
553 		}
554 	}
555 
556 	return res;
557 }
558 
559 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
560 {
561 	struct net *net = sock_net(skb->sk);
562 	unsigned int h, s_h;
563 	unsigned int e = 0, s_e;
564 	struct rt6_rtnl_dump_arg arg;
565 	struct fib6_walker *w;
566 	struct fib6_table *tb;
567 	struct hlist_head *head;
568 	int res = 0;
569 
570 	s_h = cb->args[0];
571 	s_e = cb->args[1];
572 
573 	w = (void *)cb->args[2];
574 	if (!w) {
575 		/* New dump:
576 		 *
577 		 * 1. hook callback destructor.
578 		 */
579 		cb->args[3] = (long)cb->done;
580 		cb->done = fib6_dump_done;
581 
582 		/*
583 		 * 2. allocate and initialize walker.
584 		 */
585 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
586 		if (!w)
587 			return -ENOMEM;
588 		w->func = fib6_dump_node;
589 		cb->args[2] = (long)w;
590 	}
591 
592 	arg.skb = skb;
593 	arg.cb = cb;
594 	arg.net = net;
595 	w->args = &arg;
596 
597 	rcu_read_lock();
598 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
599 		e = 0;
600 		head = &net->ipv6.fib_table_hash[h];
601 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
602 			if (e < s_e)
603 				goto next;
604 			res = fib6_dump_table(tb, skb, cb);
605 			if (res != 0)
606 				goto out;
607 next:
608 			e++;
609 		}
610 	}
611 out:
612 	rcu_read_unlock();
613 	cb->args[1] = e;
614 	cb->args[0] = h;
615 
616 	res = res < 0 ? res : skb->len;
617 	if (res <= 0)
618 		fib6_dump_end(cb);
619 	return res;
620 }
621 
622 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
623 {
624 	if (!f6i)
625 		return;
626 
627 	if (f6i->fib6_metrics == &dst_default_metrics) {
628 		struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
629 
630 		if (!p)
631 			return;
632 
633 		refcount_set(&p->refcnt, 1);
634 		f6i->fib6_metrics = p;
635 	}
636 
637 	f6i->fib6_metrics->metrics[metric - 1] = val;
638 }
639 
640 /*
641  *	Routing Table
642  *
643  *	return the appropriate node for a routing tree "add" operation
644  *	by either creating and inserting or by returning an existing
645  *	node.
646  */
647 
648 static struct fib6_node *fib6_add_1(struct net *net,
649 				    struct fib6_table *table,
650 				    struct fib6_node *root,
651 				    struct in6_addr *addr, int plen,
652 				    int offset, int allow_create,
653 				    int replace_required,
654 				    struct netlink_ext_ack *extack)
655 {
656 	struct fib6_node *fn, *in, *ln;
657 	struct fib6_node *pn = NULL;
658 	struct rt6key *key;
659 	int	bit;
660 	__be32	dir = 0;
661 
662 	RT6_TRACE("fib6_add_1\n");
663 
664 	/* insert node in tree */
665 
666 	fn = root;
667 
668 	do {
669 		struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
670 					    lockdep_is_held(&table->tb6_lock));
671 		key = (struct rt6key *)((u8 *)leaf + offset);
672 
673 		/*
674 		 *	Prefix match
675 		 */
676 		if (plen < fn->fn_bit ||
677 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
678 			if (!allow_create) {
679 				if (replace_required) {
680 					NL_SET_ERR_MSG(extack,
681 						       "Can not replace route - no match found");
682 					pr_warn("Can't replace route, no match found\n");
683 					return ERR_PTR(-ENOENT);
684 				}
685 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
686 			}
687 			goto insert_above;
688 		}
689 
690 		/*
691 		 *	Exact match ?
692 		 */
693 
694 		if (plen == fn->fn_bit) {
695 			/* clean up an intermediate node */
696 			if (!(fn->fn_flags & RTN_RTINFO)) {
697 				RCU_INIT_POINTER(fn->leaf, NULL);
698 				fib6_info_release(leaf);
699 			/* remove null_entry in the root node */
700 			} else if (fn->fn_flags & RTN_TL_ROOT &&
701 				   rcu_access_pointer(fn->leaf) ==
702 				   net->ipv6.fib6_null_entry) {
703 				RCU_INIT_POINTER(fn->leaf, NULL);
704 			}
705 
706 			return fn;
707 		}
708 
709 		/*
710 		 *	We have more bits to go
711 		 */
712 
713 		/* Try to walk down on tree. */
714 		dir = addr_bit_set(addr, fn->fn_bit);
715 		pn = fn;
716 		fn = dir ?
717 		     rcu_dereference_protected(fn->right,
718 					lockdep_is_held(&table->tb6_lock)) :
719 		     rcu_dereference_protected(fn->left,
720 					lockdep_is_held(&table->tb6_lock));
721 	} while (fn);
722 
723 	if (!allow_create) {
724 		/* We should not create new node because
725 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
726 		 * I assume it is safe to require NLM_F_CREATE when
727 		 * REPLACE flag is used! Later we may want to remove the
728 		 * check for replace_required, because according
729 		 * to netlink specification, NLM_F_CREATE
730 		 * MUST be specified if new route is created.
731 		 * That would keep IPv6 consistent with IPv4
732 		 */
733 		if (replace_required) {
734 			NL_SET_ERR_MSG(extack,
735 				       "Can not replace route - no match found");
736 			pr_warn("Can't replace route, no match found\n");
737 			return ERR_PTR(-ENOENT);
738 		}
739 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
740 	}
741 	/*
742 	 *	We walked to the bottom of tree.
743 	 *	Create new leaf node without children.
744 	 */
745 
746 	ln = node_alloc(net);
747 
748 	if (!ln)
749 		return ERR_PTR(-ENOMEM);
750 	ln->fn_bit = plen;
751 	RCU_INIT_POINTER(ln->parent, pn);
752 
753 	if (dir)
754 		rcu_assign_pointer(pn->right, ln);
755 	else
756 		rcu_assign_pointer(pn->left, ln);
757 
758 	return ln;
759 
760 
761 insert_above:
762 	/*
763 	 * split since we don't have a common prefix anymore or
764 	 * we have a less significant route.
765 	 * we've to insert an intermediate node on the list
766 	 * this new node will point to the one we need to create
767 	 * and the current
768 	 */
769 
770 	pn = rcu_dereference_protected(fn->parent,
771 				       lockdep_is_held(&table->tb6_lock));
772 
773 	/* find 1st bit in difference between the 2 addrs.
774 
775 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
776 	   but if it is >= plen, the value is ignored in any case.
777 	 */
778 
779 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
780 
781 	/*
782 	 *		(intermediate)[in]
783 	 *	          /	   \
784 	 *	(new leaf node)[ln] (old node)[fn]
785 	 */
786 	if (plen > bit) {
787 		in = node_alloc(net);
788 		ln = node_alloc(net);
789 
790 		if (!in || !ln) {
791 			if (in)
792 				node_free_immediate(net, in);
793 			if (ln)
794 				node_free_immediate(net, ln);
795 			return ERR_PTR(-ENOMEM);
796 		}
797 
798 		/*
799 		 * new intermediate node.
800 		 * RTN_RTINFO will
801 		 * be off since that an address that chooses one of
802 		 * the branches would not match less specific routes
803 		 * in the other branch
804 		 */
805 
806 		in->fn_bit = bit;
807 
808 		RCU_INIT_POINTER(in->parent, pn);
809 		in->leaf = fn->leaf;
810 		atomic_inc(&rcu_dereference_protected(in->leaf,
811 				lockdep_is_held(&table->tb6_lock))->fib6_ref);
812 
813 		/* update parent pointer */
814 		if (dir)
815 			rcu_assign_pointer(pn->right, in);
816 		else
817 			rcu_assign_pointer(pn->left, in);
818 
819 		ln->fn_bit = plen;
820 
821 		RCU_INIT_POINTER(ln->parent, in);
822 		rcu_assign_pointer(fn->parent, in);
823 
824 		if (addr_bit_set(addr, bit)) {
825 			rcu_assign_pointer(in->right, ln);
826 			rcu_assign_pointer(in->left, fn);
827 		} else {
828 			rcu_assign_pointer(in->left, ln);
829 			rcu_assign_pointer(in->right, fn);
830 		}
831 	} else { /* plen <= bit */
832 
833 		/*
834 		 *		(new leaf node)[ln]
835 		 *	          /	   \
836 		 *	     (old node)[fn] NULL
837 		 */
838 
839 		ln = node_alloc(net);
840 
841 		if (!ln)
842 			return ERR_PTR(-ENOMEM);
843 
844 		ln->fn_bit = plen;
845 
846 		RCU_INIT_POINTER(ln->parent, pn);
847 
848 		if (addr_bit_set(&key->addr, plen))
849 			RCU_INIT_POINTER(ln->right, fn);
850 		else
851 			RCU_INIT_POINTER(ln->left, fn);
852 
853 		rcu_assign_pointer(fn->parent, ln);
854 
855 		if (dir)
856 			rcu_assign_pointer(pn->right, ln);
857 		else
858 			rcu_assign_pointer(pn->left, ln);
859 	}
860 	return ln;
861 }
862 
863 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
864 				const struct fib6_table *table)
865 {
866 	int cpu;
867 
868 	/* release the reference to this fib entry from
869 	 * all of its cached pcpu routes
870 	 */
871 	for_each_possible_cpu(cpu) {
872 		struct rt6_info **ppcpu_rt;
873 		struct rt6_info *pcpu_rt;
874 
875 		ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
876 		pcpu_rt = *ppcpu_rt;
877 		if (pcpu_rt) {
878 			struct fib6_info *from;
879 
880 			from = rcu_dereference_protected(pcpu_rt->from,
881 					     lockdep_is_held(&table->tb6_lock));
882 			rcu_assign_pointer(pcpu_rt->from, NULL);
883 			fib6_info_release(from);
884 		}
885 	}
886 }
887 
888 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
889 			  struct net *net)
890 {
891 	struct fib6_table *table = rt->fib6_table;
892 
893 	if (atomic_read(&rt->fib6_ref) != 1) {
894 		/* This route is used as dummy address holder in some split
895 		 * nodes. It is not leaked, but it still holds other resources,
896 		 * which must be released in time. So, scan ascendant nodes
897 		 * and replace dummy references to this route with references
898 		 * to still alive ones.
899 		 */
900 		while (fn) {
901 			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
902 					    lockdep_is_held(&table->tb6_lock));
903 			struct fib6_info *new_leaf;
904 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
905 				new_leaf = fib6_find_prefix(net, table, fn);
906 				atomic_inc(&new_leaf->fib6_ref);
907 
908 				rcu_assign_pointer(fn->leaf, new_leaf);
909 				fib6_info_release(rt);
910 			}
911 			fn = rcu_dereference_protected(fn->parent,
912 				    lockdep_is_held(&table->tb6_lock));
913 		}
914 
915 		if (rt->rt6i_pcpu)
916 			fib6_drop_pcpu_from(rt, table);
917 	}
918 }
919 
920 /*
921  *	Insert routing information in a node.
922  */
923 
924 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
925 			    struct nl_info *info,
926 			    struct netlink_ext_ack *extack)
927 {
928 	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
929 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
930 	struct fib6_info *iter = NULL;
931 	struct fib6_info __rcu **ins;
932 	struct fib6_info __rcu **fallback_ins = NULL;
933 	int replace = (info->nlh &&
934 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
935 	int add = (!info->nlh ||
936 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
937 	int found = 0;
938 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
939 	u16 nlflags = NLM_F_EXCL;
940 	int err;
941 
942 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
943 		nlflags |= NLM_F_APPEND;
944 
945 	ins = &fn->leaf;
946 
947 	for (iter = leaf; iter;
948 	     iter = rcu_dereference_protected(iter->rt6_next,
949 				lockdep_is_held(&rt->fib6_table->tb6_lock))) {
950 		/*
951 		 *	Search for duplicates
952 		 */
953 
954 		if (iter->fib6_metric == rt->fib6_metric) {
955 			/*
956 			 *	Same priority level
957 			 */
958 			if (info->nlh &&
959 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
960 				return -EEXIST;
961 
962 			nlflags &= ~NLM_F_EXCL;
963 			if (replace) {
964 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
965 					found++;
966 					break;
967 				}
968 				if (rt_can_ecmp)
969 					fallback_ins = fallback_ins ?: ins;
970 				goto next_iter;
971 			}
972 
973 			if (rt6_duplicate_nexthop(iter, rt)) {
974 				if (rt->fib6_nsiblings)
975 					rt->fib6_nsiblings = 0;
976 				if (!(iter->fib6_flags & RTF_EXPIRES))
977 					return -EEXIST;
978 				if (!(rt->fib6_flags & RTF_EXPIRES))
979 					fib6_clean_expires(iter);
980 				else
981 					fib6_set_expires(iter, rt->expires);
982 				fib6_metric_set(iter, RTAX_MTU, rt->fib6_pmtu);
983 				return -EEXIST;
984 			}
985 			/* If we have the same destination and the same metric,
986 			 * but not the same gateway, then the route we try to
987 			 * add is sibling to this route, increment our counter
988 			 * of siblings, and later we will add our route to the
989 			 * list.
990 			 * Only static routes (which don't have flag
991 			 * RTF_EXPIRES) are used for ECMPv6.
992 			 *
993 			 * To avoid long list, we only had siblings if the
994 			 * route have a gateway.
995 			 */
996 			if (rt_can_ecmp &&
997 			    rt6_qualify_for_ecmp(iter))
998 				rt->fib6_nsiblings++;
999 		}
1000 
1001 		if (iter->fib6_metric > rt->fib6_metric)
1002 			break;
1003 
1004 next_iter:
1005 		ins = &iter->rt6_next;
1006 	}
1007 
1008 	if (fallback_ins && !found) {
1009 		/* No ECMP-able route found, replace first non-ECMP one */
1010 		ins = fallback_ins;
1011 		iter = rcu_dereference_protected(*ins,
1012 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1013 		found++;
1014 	}
1015 
1016 	/* Reset round-robin state, if necessary */
1017 	if (ins == &fn->leaf)
1018 		fn->rr_ptr = NULL;
1019 
1020 	/* Link this route to others same route. */
1021 	if (rt->fib6_nsiblings) {
1022 		unsigned int fib6_nsiblings;
1023 		struct fib6_info *sibling, *temp_sibling;
1024 
1025 		/* Find the first route that have the same metric */
1026 		sibling = leaf;
1027 		while (sibling) {
1028 			if (sibling->fib6_metric == rt->fib6_metric &&
1029 			    rt6_qualify_for_ecmp(sibling)) {
1030 				list_add_tail(&rt->fib6_siblings,
1031 					      &sibling->fib6_siblings);
1032 				break;
1033 			}
1034 			sibling = rcu_dereference_protected(sibling->rt6_next,
1035 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1036 		}
1037 		/* For each sibling in the list, increment the counter of
1038 		 * siblings. BUG() if counters does not match, list of siblings
1039 		 * is broken!
1040 		 */
1041 		fib6_nsiblings = 0;
1042 		list_for_each_entry_safe(sibling, temp_sibling,
1043 					 &rt->fib6_siblings, fib6_siblings) {
1044 			sibling->fib6_nsiblings++;
1045 			BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1046 			fib6_nsiblings++;
1047 		}
1048 		BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1049 		rt6_multipath_rebalance(temp_sibling);
1050 	}
1051 
1052 	/*
1053 	 *	insert node
1054 	 */
1055 	if (!replace) {
1056 		if (!add)
1057 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1058 
1059 add:
1060 		nlflags |= NLM_F_CREATE;
1061 
1062 		err = call_fib6_entry_notifiers(info->nl_net,
1063 						FIB_EVENT_ENTRY_ADD,
1064 						rt, extack);
1065 		if (err)
1066 			return err;
1067 
1068 		rcu_assign_pointer(rt->rt6_next, iter);
1069 		atomic_inc(&rt->fib6_ref);
1070 		rcu_assign_pointer(rt->fib6_node, fn);
1071 		rcu_assign_pointer(*ins, rt);
1072 		if (!info->skip_notify)
1073 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1074 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1075 
1076 		if (!(fn->fn_flags & RTN_RTINFO)) {
1077 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1078 			fn->fn_flags |= RTN_RTINFO;
1079 		}
1080 
1081 	} else {
1082 		int nsiblings;
1083 
1084 		if (!found) {
1085 			if (add)
1086 				goto add;
1087 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1088 			return -ENOENT;
1089 		}
1090 
1091 		err = call_fib6_entry_notifiers(info->nl_net,
1092 						FIB_EVENT_ENTRY_REPLACE,
1093 						rt, extack);
1094 		if (err)
1095 			return err;
1096 
1097 		atomic_inc(&rt->fib6_ref);
1098 		rcu_assign_pointer(rt->fib6_node, fn);
1099 		rt->rt6_next = iter->rt6_next;
1100 		rcu_assign_pointer(*ins, rt);
1101 		if (!info->skip_notify)
1102 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1103 		if (!(fn->fn_flags & RTN_RTINFO)) {
1104 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1105 			fn->fn_flags |= RTN_RTINFO;
1106 		}
1107 		nsiblings = iter->fib6_nsiblings;
1108 		iter->fib6_node = NULL;
1109 		fib6_purge_rt(iter, fn, info->nl_net);
1110 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1111 			fn->rr_ptr = NULL;
1112 		fib6_info_release(iter);
1113 
1114 		if (nsiblings) {
1115 			/* Replacing an ECMP route, remove all siblings */
1116 			ins = &rt->rt6_next;
1117 			iter = rcu_dereference_protected(*ins,
1118 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1119 			while (iter) {
1120 				if (iter->fib6_metric > rt->fib6_metric)
1121 					break;
1122 				if (rt6_qualify_for_ecmp(iter)) {
1123 					*ins = iter->rt6_next;
1124 					iter->fib6_node = NULL;
1125 					fib6_purge_rt(iter, fn, info->nl_net);
1126 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1127 						fn->rr_ptr = NULL;
1128 					fib6_info_release(iter);
1129 					nsiblings--;
1130 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1131 				} else {
1132 					ins = &iter->rt6_next;
1133 				}
1134 				iter = rcu_dereference_protected(*ins,
1135 					lockdep_is_held(&rt->fib6_table->tb6_lock));
1136 			}
1137 			WARN_ON(nsiblings != 0);
1138 		}
1139 	}
1140 
1141 	return 0;
1142 }
1143 
1144 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1145 {
1146 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1147 	    (rt->fib6_flags & RTF_EXPIRES))
1148 		mod_timer(&net->ipv6.ip6_fib_timer,
1149 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1150 }
1151 
1152 void fib6_force_start_gc(struct net *net)
1153 {
1154 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1155 		mod_timer(&net->ipv6.ip6_fib_timer,
1156 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1157 }
1158 
1159 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1160 					   int sernum)
1161 {
1162 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1163 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1164 
1165 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1166 	smp_wmb();
1167 	while (fn) {
1168 		fn->fn_sernum = sernum;
1169 		fn = rcu_dereference_protected(fn->parent,
1170 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1171 	}
1172 }
1173 
1174 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1175 {
1176 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1177 }
1178 
1179 /*
1180  *	Add routing information to the routing tree.
1181  *	<destination addr>/<source addr>
1182  *	with source addr info in sub-trees
1183  *	Need to own table->tb6_lock
1184  */
1185 
1186 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1187 	     struct nl_info *info, struct netlink_ext_ack *extack)
1188 {
1189 	struct fib6_table *table = rt->fib6_table;
1190 	struct fib6_node *fn, *pn = NULL;
1191 	int err = -ENOMEM;
1192 	int allow_create = 1;
1193 	int replace_required = 0;
1194 	int sernum = fib6_new_sernum(info->nl_net);
1195 
1196 	if (info->nlh) {
1197 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1198 			allow_create = 0;
1199 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1200 			replace_required = 1;
1201 	}
1202 	if (!allow_create && !replace_required)
1203 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1204 
1205 	fn = fib6_add_1(info->nl_net, table, root,
1206 			&rt->fib6_dst.addr, rt->fib6_dst.plen,
1207 			offsetof(struct fib6_info, fib6_dst), allow_create,
1208 			replace_required, extack);
1209 	if (IS_ERR(fn)) {
1210 		err = PTR_ERR(fn);
1211 		fn = NULL;
1212 		goto out;
1213 	}
1214 
1215 	pn = fn;
1216 
1217 #ifdef CONFIG_IPV6_SUBTREES
1218 	if (rt->fib6_src.plen) {
1219 		struct fib6_node *sn;
1220 
1221 		if (!rcu_access_pointer(fn->subtree)) {
1222 			struct fib6_node *sfn;
1223 
1224 			/*
1225 			 * Create subtree.
1226 			 *
1227 			 *		fn[main tree]
1228 			 *		|
1229 			 *		sfn[subtree root]
1230 			 *		   \
1231 			 *		    sn[new leaf node]
1232 			 */
1233 
1234 			/* Create subtree root node */
1235 			sfn = node_alloc(info->nl_net);
1236 			if (!sfn)
1237 				goto failure;
1238 
1239 			atomic_inc(&info->nl_net->ipv6.fib6_null_entry->fib6_ref);
1240 			rcu_assign_pointer(sfn->leaf,
1241 					   info->nl_net->ipv6.fib6_null_entry);
1242 			sfn->fn_flags = RTN_ROOT;
1243 
1244 			/* Now add the first leaf node to new subtree */
1245 
1246 			sn = fib6_add_1(info->nl_net, table, sfn,
1247 					&rt->fib6_src.addr, rt->fib6_src.plen,
1248 					offsetof(struct fib6_info, fib6_src),
1249 					allow_create, replace_required, extack);
1250 
1251 			if (IS_ERR(sn)) {
1252 				/* If it is failed, discard just allocated
1253 				   root, and then (in failure) stale node
1254 				   in main tree.
1255 				 */
1256 				node_free_immediate(info->nl_net, sfn);
1257 				err = PTR_ERR(sn);
1258 				goto failure;
1259 			}
1260 
1261 			/* Now link new subtree to main tree */
1262 			rcu_assign_pointer(sfn->parent, fn);
1263 			rcu_assign_pointer(fn->subtree, sfn);
1264 		} else {
1265 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1266 					&rt->fib6_src.addr, rt->fib6_src.plen,
1267 					offsetof(struct fib6_info, fib6_src),
1268 					allow_create, replace_required, extack);
1269 
1270 			if (IS_ERR(sn)) {
1271 				err = PTR_ERR(sn);
1272 				goto failure;
1273 			}
1274 		}
1275 
1276 		if (!rcu_access_pointer(fn->leaf)) {
1277 			if (fn->fn_flags & RTN_TL_ROOT) {
1278 				/* put back null_entry for root node */
1279 				rcu_assign_pointer(fn->leaf,
1280 					    info->nl_net->ipv6.fib6_null_entry);
1281 			} else {
1282 				atomic_inc(&rt->fib6_ref);
1283 				rcu_assign_pointer(fn->leaf, rt);
1284 			}
1285 		}
1286 		fn = sn;
1287 	}
1288 #endif
1289 
1290 	err = fib6_add_rt2node(fn, rt, info, extack);
1291 	if (!err) {
1292 		__fib6_update_sernum_upto_root(rt, sernum);
1293 		fib6_start_gc(info->nl_net, rt);
1294 	}
1295 
1296 out:
1297 	if (err) {
1298 #ifdef CONFIG_IPV6_SUBTREES
1299 		/*
1300 		 * If fib6_add_1 has cleared the old leaf pointer in the
1301 		 * super-tree leaf node we have to find a new one for it.
1302 		 */
1303 		if (pn != fn) {
1304 			struct fib6_info *pn_leaf =
1305 				rcu_dereference_protected(pn->leaf,
1306 				    lockdep_is_held(&table->tb6_lock));
1307 			if (pn_leaf == rt) {
1308 				pn_leaf = NULL;
1309 				RCU_INIT_POINTER(pn->leaf, NULL);
1310 				fib6_info_release(rt);
1311 			}
1312 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1313 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1314 							   pn);
1315 #if RT6_DEBUG >= 2
1316 				if (!pn_leaf) {
1317 					WARN_ON(!pn_leaf);
1318 					pn_leaf =
1319 					    info->nl_net->ipv6.fib6_null_entry;
1320 				}
1321 #endif
1322 				fib6_info_hold(pn_leaf);
1323 				rcu_assign_pointer(pn->leaf, pn_leaf);
1324 			}
1325 		}
1326 #endif
1327 		goto failure;
1328 	}
1329 	return err;
1330 
1331 failure:
1332 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1333 	 * 1. fn is an intermediate node and we failed to add the new
1334 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1335 	 * failure case.
1336 	 * 2. fn is the root node in the table and we fail to add the first
1337 	 * default route to it.
1338 	 */
1339 	if (fn &&
1340 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1341 	     (fn->fn_flags & RTN_TL_ROOT &&
1342 	      !rcu_access_pointer(fn->leaf))))
1343 		fib6_repair_tree(info->nl_net, table, fn);
1344 	return err;
1345 }
1346 
1347 /*
1348  *	Routing tree lookup
1349  *
1350  */
1351 
1352 struct lookup_args {
1353 	int			offset;		/* key offset on fib6_info */
1354 	const struct in6_addr	*addr;		/* search key			*/
1355 };
1356 
1357 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1358 				       struct lookup_args *args)
1359 {
1360 	struct fib6_node *fn;
1361 	__be32 dir;
1362 
1363 	if (unlikely(args->offset == 0))
1364 		return NULL;
1365 
1366 	/*
1367 	 *	Descend on a tree
1368 	 */
1369 
1370 	fn = root;
1371 
1372 	for (;;) {
1373 		struct fib6_node *next;
1374 
1375 		dir = addr_bit_set(args->addr, fn->fn_bit);
1376 
1377 		next = dir ? rcu_dereference(fn->right) :
1378 			     rcu_dereference(fn->left);
1379 
1380 		if (next) {
1381 			fn = next;
1382 			continue;
1383 		}
1384 		break;
1385 	}
1386 
1387 	while (fn) {
1388 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1389 
1390 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1391 			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1392 			struct rt6key *key;
1393 
1394 			if (!leaf)
1395 				goto backtrack;
1396 
1397 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1398 
1399 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1400 #ifdef CONFIG_IPV6_SUBTREES
1401 				if (subtree) {
1402 					struct fib6_node *sfn;
1403 					sfn = fib6_lookup_1(subtree, args + 1);
1404 					if (!sfn)
1405 						goto backtrack;
1406 					fn = sfn;
1407 				}
1408 #endif
1409 				if (fn->fn_flags & RTN_RTINFO)
1410 					return fn;
1411 			}
1412 		}
1413 backtrack:
1414 		if (fn->fn_flags & RTN_ROOT)
1415 			break;
1416 
1417 		fn = rcu_dereference(fn->parent);
1418 	}
1419 
1420 	return NULL;
1421 }
1422 
1423 /* called with rcu_read_lock() held
1424  */
1425 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1426 			      const struct in6_addr *saddr)
1427 {
1428 	struct fib6_node *fn;
1429 	struct lookup_args args[] = {
1430 		{
1431 			.offset = offsetof(struct fib6_info, fib6_dst),
1432 			.addr = daddr,
1433 		},
1434 #ifdef CONFIG_IPV6_SUBTREES
1435 		{
1436 			.offset = offsetof(struct fib6_info, fib6_src),
1437 			.addr = saddr,
1438 		},
1439 #endif
1440 		{
1441 			.offset = 0,	/* sentinel */
1442 		}
1443 	};
1444 
1445 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
1446 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1447 		fn = root;
1448 
1449 	return fn;
1450 }
1451 
1452 /*
1453  *	Get node with specified destination prefix (and source prefix,
1454  *	if subtrees are used)
1455  *	exact_match == true means we try to find fn with exact match of
1456  *	the passed in prefix addr
1457  *	exact_match == false means we try to find fn with longest prefix
1458  *	match of the passed in prefix addr. This is useful for finding fn
1459  *	for cached route as it will be stored in the exception table under
1460  *	the node with longest prefix length.
1461  */
1462 
1463 
1464 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1465 				       const struct in6_addr *addr,
1466 				       int plen, int offset,
1467 				       bool exact_match)
1468 {
1469 	struct fib6_node *fn, *prev = NULL;
1470 
1471 	for (fn = root; fn ; ) {
1472 		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1473 		struct rt6key *key;
1474 
1475 		/* This node is being deleted */
1476 		if (!leaf) {
1477 			if (plen <= fn->fn_bit)
1478 				goto out;
1479 			else
1480 				goto next;
1481 		}
1482 
1483 		key = (struct rt6key *)((u8 *)leaf + offset);
1484 
1485 		/*
1486 		 *	Prefix match
1487 		 */
1488 		if (plen < fn->fn_bit ||
1489 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1490 			goto out;
1491 
1492 		if (plen == fn->fn_bit)
1493 			return fn;
1494 
1495 		prev = fn;
1496 
1497 next:
1498 		/*
1499 		 *	We have more bits to go
1500 		 */
1501 		if (addr_bit_set(addr, fn->fn_bit))
1502 			fn = rcu_dereference(fn->right);
1503 		else
1504 			fn = rcu_dereference(fn->left);
1505 	}
1506 out:
1507 	if (exact_match)
1508 		return NULL;
1509 	else
1510 		return prev;
1511 }
1512 
1513 struct fib6_node *fib6_locate(struct fib6_node *root,
1514 			      const struct in6_addr *daddr, int dst_len,
1515 			      const struct in6_addr *saddr, int src_len,
1516 			      bool exact_match)
1517 {
1518 	struct fib6_node *fn;
1519 
1520 	fn = fib6_locate_1(root, daddr, dst_len,
1521 			   offsetof(struct fib6_info, fib6_dst),
1522 			   exact_match);
1523 
1524 #ifdef CONFIG_IPV6_SUBTREES
1525 	if (src_len) {
1526 		WARN_ON(saddr == NULL);
1527 		if (fn) {
1528 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1529 
1530 			if (subtree) {
1531 				fn = fib6_locate_1(subtree, saddr, src_len,
1532 					   offsetof(struct fib6_info, fib6_src),
1533 					   exact_match);
1534 			}
1535 		}
1536 	}
1537 #endif
1538 
1539 	if (fn && fn->fn_flags & RTN_RTINFO)
1540 		return fn;
1541 
1542 	return NULL;
1543 }
1544 
1545 
1546 /*
1547  *	Deletion
1548  *
1549  */
1550 
1551 static struct fib6_info *fib6_find_prefix(struct net *net,
1552 					 struct fib6_table *table,
1553 					 struct fib6_node *fn)
1554 {
1555 	struct fib6_node *child_left, *child_right;
1556 
1557 	if (fn->fn_flags & RTN_ROOT)
1558 		return net->ipv6.fib6_null_entry;
1559 
1560 	while (fn) {
1561 		child_left = rcu_dereference_protected(fn->left,
1562 				    lockdep_is_held(&table->tb6_lock));
1563 		child_right = rcu_dereference_protected(fn->right,
1564 				    lockdep_is_held(&table->tb6_lock));
1565 		if (child_left)
1566 			return rcu_dereference_protected(child_left->leaf,
1567 					lockdep_is_held(&table->tb6_lock));
1568 		if (child_right)
1569 			return rcu_dereference_protected(child_right->leaf,
1570 					lockdep_is_held(&table->tb6_lock));
1571 
1572 		fn = FIB6_SUBTREE(fn);
1573 	}
1574 	return NULL;
1575 }
1576 
1577 /*
1578  *	Called to trim the tree of intermediate nodes when possible. "fn"
1579  *	is the node we want to try and remove.
1580  *	Need to own table->tb6_lock
1581  */
1582 
1583 static struct fib6_node *fib6_repair_tree(struct net *net,
1584 					  struct fib6_table *table,
1585 					  struct fib6_node *fn)
1586 {
1587 	int children;
1588 	int nstate;
1589 	struct fib6_node *child;
1590 	struct fib6_walker *w;
1591 	int iter = 0;
1592 
1593 	/* Set fn->leaf to null_entry for root node. */
1594 	if (fn->fn_flags & RTN_TL_ROOT) {
1595 		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1596 		return fn;
1597 	}
1598 
1599 	for (;;) {
1600 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1601 					    lockdep_is_held(&table->tb6_lock));
1602 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1603 					    lockdep_is_held(&table->tb6_lock));
1604 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1605 					    lockdep_is_held(&table->tb6_lock));
1606 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1607 					    lockdep_is_held(&table->tb6_lock));
1608 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1609 					    lockdep_is_held(&table->tb6_lock));
1610 		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1611 					    lockdep_is_held(&table->tb6_lock));
1612 		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1613 					    lockdep_is_held(&table->tb6_lock));
1614 		struct fib6_info *new_fn_leaf;
1615 
1616 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1617 		iter++;
1618 
1619 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1620 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1621 		WARN_ON(fn_leaf);
1622 
1623 		children = 0;
1624 		child = NULL;
1625 		if (fn_r)
1626 			child = fn_r, children |= 1;
1627 		if (fn_l)
1628 			child = fn_l, children |= 2;
1629 
1630 		if (children == 3 || FIB6_SUBTREE(fn)
1631 #ifdef CONFIG_IPV6_SUBTREES
1632 		    /* Subtree root (i.e. fn) may have one child */
1633 		    || (children && fn->fn_flags & RTN_ROOT)
1634 #endif
1635 		    ) {
1636 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1637 #if RT6_DEBUG >= 2
1638 			if (!new_fn_leaf) {
1639 				WARN_ON(!new_fn_leaf);
1640 				new_fn_leaf = net->ipv6.fib6_null_entry;
1641 			}
1642 #endif
1643 			fib6_info_hold(new_fn_leaf);
1644 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1645 			return pn;
1646 		}
1647 
1648 #ifdef CONFIG_IPV6_SUBTREES
1649 		if (FIB6_SUBTREE(pn) == fn) {
1650 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1651 			RCU_INIT_POINTER(pn->subtree, NULL);
1652 			nstate = FWS_L;
1653 		} else {
1654 			WARN_ON(fn->fn_flags & RTN_ROOT);
1655 #endif
1656 			if (pn_r == fn)
1657 				rcu_assign_pointer(pn->right, child);
1658 			else if (pn_l == fn)
1659 				rcu_assign_pointer(pn->left, child);
1660 #if RT6_DEBUG >= 2
1661 			else
1662 				WARN_ON(1);
1663 #endif
1664 			if (child)
1665 				rcu_assign_pointer(child->parent, pn);
1666 			nstate = FWS_R;
1667 #ifdef CONFIG_IPV6_SUBTREES
1668 		}
1669 #endif
1670 
1671 		read_lock(&net->ipv6.fib6_walker_lock);
1672 		FOR_WALKERS(net, w) {
1673 			if (!child) {
1674 				if (w->node == fn) {
1675 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1676 					w->node = pn;
1677 					w->state = nstate;
1678 				}
1679 			} else {
1680 				if (w->node == fn) {
1681 					w->node = child;
1682 					if (children&2) {
1683 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1684 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1685 					} else {
1686 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1687 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1688 					}
1689 				}
1690 			}
1691 		}
1692 		read_unlock(&net->ipv6.fib6_walker_lock);
1693 
1694 		node_free(net, fn);
1695 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1696 			return pn;
1697 
1698 		RCU_INIT_POINTER(pn->leaf, NULL);
1699 		fib6_info_release(pn_leaf);
1700 		fn = pn;
1701 	}
1702 }
1703 
1704 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1705 			   struct fib6_info __rcu **rtp, struct nl_info *info)
1706 {
1707 	struct fib6_walker *w;
1708 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1709 				    lockdep_is_held(&table->tb6_lock));
1710 	struct net *net = info->nl_net;
1711 
1712 	RT6_TRACE("fib6_del_route\n");
1713 
1714 	/* Unlink it */
1715 	*rtp = rt->rt6_next;
1716 	rt->fib6_node = NULL;
1717 	net->ipv6.rt6_stats->fib_rt_entries--;
1718 	net->ipv6.rt6_stats->fib_discarded_routes++;
1719 
1720 	/* Flush all cached dst in exception table */
1721 	rt6_flush_exceptions(rt);
1722 
1723 	/* Reset round-robin state, if necessary */
1724 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1725 		fn->rr_ptr = NULL;
1726 
1727 	/* Remove this entry from other siblings */
1728 	if (rt->fib6_nsiblings) {
1729 		struct fib6_info *sibling, *next_sibling;
1730 
1731 		list_for_each_entry_safe(sibling, next_sibling,
1732 					 &rt->fib6_siblings, fib6_siblings)
1733 			sibling->fib6_nsiblings--;
1734 		rt->fib6_nsiblings = 0;
1735 		list_del_init(&rt->fib6_siblings);
1736 		rt6_multipath_rebalance(next_sibling);
1737 	}
1738 
1739 	/* Adjust walkers */
1740 	read_lock(&net->ipv6.fib6_walker_lock);
1741 	FOR_WALKERS(net, w) {
1742 		if (w->state == FWS_C && w->leaf == rt) {
1743 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1744 			w->leaf = rcu_dereference_protected(rt->rt6_next,
1745 					    lockdep_is_held(&table->tb6_lock));
1746 			if (!w->leaf)
1747 				w->state = FWS_U;
1748 		}
1749 	}
1750 	read_unlock(&net->ipv6.fib6_walker_lock);
1751 
1752 	/* If it was last route, call fib6_repair_tree() to:
1753 	 * 1. For root node, put back null_entry as how the table was created.
1754 	 * 2. For other nodes, expunge its radix tree node.
1755 	 */
1756 	if (!rcu_access_pointer(fn->leaf)) {
1757 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1758 			fn->fn_flags &= ~RTN_RTINFO;
1759 			net->ipv6.rt6_stats->fib_route_nodes--;
1760 		}
1761 		fn = fib6_repair_tree(net, table, fn);
1762 	}
1763 
1764 	fib6_purge_rt(rt, fn, net);
1765 
1766 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1767 	if (!info->skip_notify)
1768 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1769 	fib6_info_release(rt);
1770 }
1771 
1772 /* Need to own table->tb6_lock */
1773 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1774 {
1775 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1776 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1777 	struct fib6_table *table = rt->fib6_table;
1778 	struct net *net = info->nl_net;
1779 	struct fib6_info __rcu **rtp;
1780 	struct fib6_info __rcu **rtp_next;
1781 
1782 	if (!fn || rt == net->ipv6.fib6_null_entry)
1783 		return -ENOENT;
1784 
1785 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1786 
1787 	/*
1788 	 *	Walk the leaf entries looking for ourself
1789 	 */
1790 
1791 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1792 		struct fib6_info *cur = rcu_dereference_protected(*rtp,
1793 					lockdep_is_held(&table->tb6_lock));
1794 		if (rt == cur) {
1795 			fib6_del_route(table, fn, rtp, info);
1796 			return 0;
1797 		}
1798 		rtp_next = &cur->rt6_next;
1799 	}
1800 	return -ENOENT;
1801 }
1802 
1803 /*
1804  *	Tree traversal function.
1805  *
1806  *	Certainly, it is not interrupt safe.
1807  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1808  *	It means, that we can modify tree during walking
1809  *	and use this function for garbage collection, clone pruning,
1810  *	cleaning tree when a device goes down etc. etc.
1811  *
1812  *	It guarantees that every node will be traversed,
1813  *	and that it will be traversed only once.
1814  *
1815  *	Callback function w->func may return:
1816  *	0 -> continue walking.
1817  *	positive value -> walking is suspended (used by tree dumps,
1818  *	and probably by gc, if it will be split to several slices)
1819  *	negative value -> terminate walking.
1820  *
1821  *	The function itself returns:
1822  *	0   -> walk is complete.
1823  *	>0  -> walk is incomplete (i.e. suspended)
1824  *	<0  -> walk is terminated by an error.
1825  *
1826  *	This function is called with tb6_lock held.
1827  */
1828 
1829 static int fib6_walk_continue(struct fib6_walker *w)
1830 {
1831 	struct fib6_node *fn, *pn, *left, *right;
1832 
1833 	/* w->root should always be table->tb6_root */
1834 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1835 
1836 	for (;;) {
1837 		fn = w->node;
1838 		if (!fn)
1839 			return 0;
1840 
1841 		switch (w->state) {
1842 #ifdef CONFIG_IPV6_SUBTREES
1843 		case FWS_S:
1844 			if (FIB6_SUBTREE(fn)) {
1845 				w->node = FIB6_SUBTREE(fn);
1846 				continue;
1847 			}
1848 			w->state = FWS_L;
1849 #endif
1850 			/* fall through */
1851 		case FWS_L:
1852 			left = rcu_dereference_protected(fn->left, 1);
1853 			if (left) {
1854 				w->node = left;
1855 				w->state = FWS_INIT;
1856 				continue;
1857 			}
1858 			w->state = FWS_R;
1859 			/* fall through */
1860 		case FWS_R:
1861 			right = rcu_dereference_protected(fn->right, 1);
1862 			if (right) {
1863 				w->node = right;
1864 				w->state = FWS_INIT;
1865 				continue;
1866 			}
1867 			w->state = FWS_C;
1868 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1869 			/* fall through */
1870 		case FWS_C:
1871 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1872 				int err;
1873 
1874 				if (w->skip) {
1875 					w->skip--;
1876 					goto skip;
1877 				}
1878 
1879 				err = w->func(w);
1880 				if (err)
1881 					return err;
1882 
1883 				w->count++;
1884 				continue;
1885 			}
1886 skip:
1887 			w->state = FWS_U;
1888 			/* fall through */
1889 		case FWS_U:
1890 			if (fn == w->root)
1891 				return 0;
1892 			pn = rcu_dereference_protected(fn->parent, 1);
1893 			left = rcu_dereference_protected(pn->left, 1);
1894 			right = rcu_dereference_protected(pn->right, 1);
1895 			w->node = pn;
1896 #ifdef CONFIG_IPV6_SUBTREES
1897 			if (FIB6_SUBTREE(pn) == fn) {
1898 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1899 				w->state = FWS_L;
1900 				continue;
1901 			}
1902 #endif
1903 			if (left == fn) {
1904 				w->state = FWS_R;
1905 				continue;
1906 			}
1907 			if (right == fn) {
1908 				w->state = FWS_C;
1909 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1910 				continue;
1911 			}
1912 #if RT6_DEBUG >= 2
1913 			WARN_ON(1);
1914 #endif
1915 		}
1916 	}
1917 }
1918 
1919 static int fib6_walk(struct net *net, struct fib6_walker *w)
1920 {
1921 	int res;
1922 
1923 	w->state = FWS_INIT;
1924 	w->node = w->root;
1925 
1926 	fib6_walker_link(net, w);
1927 	res = fib6_walk_continue(w);
1928 	if (res <= 0)
1929 		fib6_walker_unlink(net, w);
1930 	return res;
1931 }
1932 
1933 static int fib6_clean_node(struct fib6_walker *w)
1934 {
1935 	int res;
1936 	struct fib6_info *rt;
1937 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1938 	struct nl_info info = {
1939 		.nl_net = c->net,
1940 	};
1941 
1942 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1943 	    w->node->fn_sernum != c->sernum)
1944 		w->node->fn_sernum = c->sernum;
1945 
1946 	if (!c->func) {
1947 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1948 		w->leaf = NULL;
1949 		return 0;
1950 	}
1951 
1952 	for_each_fib6_walker_rt(w) {
1953 		res = c->func(rt, c->arg);
1954 		if (res == -1) {
1955 			w->leaf = rt;
1956 			res = fib6_del(rt, &info);
1957 			if (res) {
1958 #if RT6_DEBUG >= 2
1959 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1960 					 __func__, rt,
1961 					 rcu_access_pointer(rt->fib6_node),
1962 					 res);
1963 #endif
1964 				continue;
1965 			}
1966 			return 0;
1967 		} else if (res == -2) {
1968 			if (WARN_ON(!rt->fib6_nsiblings))
1969 				continue;
1970 			rt = list_last_entry(&rt->fib6_siblings,
1971 					     struct fib6_info, fib6_siblings);
1972 			continue;
1973 		}
1974 		WARN_ON(res != 0);
1975 	}
1976 	w->leaf = rt;
1977 	return 0;
1978 }
1979 
1980 /*
1981  *	Convenient frontend to tree walker.
1982  *
1983  *	func is called on each route.
1984  *		It may return -2 -> skip multipath route.
1985  *			      -1 -> delete this route.
1986  *		              0  -> continue walking
1987  */
1988 
1989 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1990 			    int (*func)(struct fib6_info *, void *arg),
1991 			    int sernum, void *arg)
1992 {
1993 	struct fib6_cleaner c;
1994 
1995 	c.w.root = root;
1996 	c.w.func = fib6_clean_node;
1997 	c.w.count = 0;
1998 	c.w.skip = 0;
1999 	c.func = func;
2000 	c.sernum = sernum;
2001 	c.arg = arg;
2002 	c.net = net;
2003 
2004 	fib6_walk(net, &c.w);
2005 }
2006 
2007 static void __fib6_clean_all(struct net *net,
2008 			     int (*func)(struct fib6_info *, void *),
2009 			     int sernum, void *arg)
2010 {
2011 	struct fib6_table *table;
2012 	struct hlist_head *head;
2013 	unsigned int h;
2014 
2015 	rcu_read_lock();
2016 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2017 		head = &net->ipv6.fib_table_hash[h];
2018 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2019 			spin_lock_bh(&table->tb6_lock);
2020 			fib6_clean_tree(net, &table->tb6_root,
2021 					func, sernum, arg);
2022 			spin_unlock_bh(&table->tb6_lock);
2023 		}
2024 	}
2025 	rcu_read_unlock();
2026 }
2027 
2028 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2029 		    void *arg)
2030 {
2031 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
2032 }
2033 
2034 static void fib6_flush_trees(struct net *net)
2035 {
2036 	int new_sernum = fib6_new_sernum(net);
2037 
2038 	__fib6_clean_all(net, NULL, new_sernum, NULL);
2039 }
2040 
2041 /*
2042  *	Garbage collection
2043  */
2044 
2045 static int fib6_age(struct fib6_info *rt, void *arg)
2046 {
2047 	struct fib6_gc_args *gc_args = arg;
2048 	unsigned long now = jiffies;
2049 
2050 	/*
2051 	 *	check addrconf expiration here.
2052 	 *	Routes are expired even if they are in use.
2053 	 */
2054 
2055 	if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2056 		if (time_after(now, rt->expires)) {
2057 			RT6_TRACE("expiring %p\n", rt);
2058 			return -1;
2059 		}
2060 		gc_args->more++;
2061 	}
2062 
2063 	/*	Also age clones in the exception table.
2064 	 *	Note, that clones are aged out
2065 	 *	only if they are not in use now.
2066 	 */
2067 	rt6_age_exceptions(rt, gc_args, now);
2068 
2069 	return 0;
2070 }
2071 
2072 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2073 {
2074 	struct fib6_gc_args gc_args;
2075 	unsigned long now;
2076 
2077 	if (force) {
2078 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2079 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2080 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2081 		return;
2082 	}
2083 	gc_args.timeout = expires ? (int)expires :
2084 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2085 	gc_args.more = 0;
2086 
2087 	fib6_clean_all(net, fib6_age, &gc_args);
2088 	now = jiffies;
2089 	net->ipv6.ip6_rt_last_gc = now;
2090 
2091 	if (gc_args.more)
2092 		mod_timer(&net->ipv6.ip6_fib_timer,
2093 			  round_jiffies(now
2094 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2095 	else
2096 		del_timer(&net->ipv6.ip6_fib_timer);
2097 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2098 }
2099 
2100 static void fib6_gc_timer_cb(struct timer_list *t)
2101 {
2102 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2103 
2104 	fib6_run_gc(0, arg, true);
2105 }
2106 
2107 static int __net_init fib6_net_init(struct net *net)
2108 {
2109 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2110 	int err;
2111 
2112 	err = fib6_notifier_init(net);
2113 	if (err)
2114 		return err;
2115 
2116 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2117 	rwlock_init(&net->ipv6.fib6_walker_lock);
2118 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2119 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2120 
2121 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2122 	if (!net->ipv6.rt6_stats)
2123 		goto out_timer;
2124 
2125 	/* Avoid false sharing : Use at least a full cache line */
2126 	size = max_t(size_t, size, L1_CACHE_BYTES);
2127 
2128 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2129 	if (!net->ipv6.fib_table_hash)
2130 		goto out_rt6_stats;
2131 
2132 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2133 					  GFP_KERNEL);
2134 	if (!net->ipv6.fib6_main_tbl)
2135 		goto out_fib_table_hash;
2136 
2137 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2138 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2139 			   net->ipv6.fib6_null_entry);
2140 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2141 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2142 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2143 
2144 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2145 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2146 					   GFP_KERNEL);
2147 	if (!net->ipv6.fib6_local_tbl)
2148 		goto out_fib6_main_tbl;
2149 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2150 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2151 			   net->ipv6.fib6_null_entry);
2152 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2153 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2154 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2155 #endif
2156 	fib6_tables_init(net);
2157 
2158 	return 0;
2159 
2160 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2161 out_fib6_main_tbl:
2162 	kfree(net->ipv6.fib6_main_tbl);
2163 #endif
2164 out_fib_table_hash:
2165 	kfree(net->ipv6.fib_table_hash);
2166 out_rt6_stats:
2167 	kfree(net->ipv6.rt6_stats);
2168 out_timer:
2169 	fib6_notifier_exit(net);
2170 	return -ENOMEM;
2171 }
2172 
2173 static void fib6_net_exit(struct net *net)
2174 {
2175 	unsigned int i;
2176 
2177 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2178 
2179 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2180 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2181 		struct hlist_node *tmp;
2182 		struct fib6_table *tb;
2183 
2184 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2185 			hlist_del(&tb->tb6_hlist);
2186 			fib6_free_table(tb);
2187 		}
2188 	}
2189 
2190 	kfree(net->ipv6.fib_table_hash);
2191 	kfree(net->ipv6.rt6_stats);
2192 	fib6_notifier_exit(net);
2193 }
2194 
2195 static struct pernet_operations fib6_net_ops = {
2196 	.init = fib6_net_init,
2197 	.exit = fib6_net_exit,
2198 };
2199 
2200 int __init fib6_init(void)
2201 {
2202 	int ret = -ENOMEM;
2203 
2204 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2205 					   sizeof(struct fib6_node),
2206 					   0, SLAB_HWCACHE_ALIGN,
2207 					   NULL);
2208 	if (!fib6_node_kmem)
2209 		goto out;
2210 
2211 	ret = register_pernet_subsys(&fib6_net_ops);
2212 	if (ret)
2213 		goto out_kmem_cache_create;
2214 
2215 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2216 				   inet6_dump_fib, 0);
2217 	if (ret)
2218 		goto out_unregister_subsys;
2219 
2220 	__fib6_flush_trees = fib6_flush_trees;
2221 out:
2222 	return ret;
2223 
2224 out_unregister_subsys:
2225 	unregister_pernet_subsys(&fib6_net_ops);
2226 out_kmem_cache_create:
2227 	kmem_cache_destroy(fib6_node_kmem);
2228 	goto out;
2229 }
2230 
2231 void fib6_gc_cleanup(void)
2232 {
2233 	unregister_pernet_subsys(&fib6_net_ops);
2234 	kmem_cache_destroy(fib6_node_kmem);
2235 }
2236 
2237 #ifdef CONFIG_PROC_FS
2238 
2239 struct ipv6_route_iter {
2240 	struct seq_net_private p;
2241 	struct fib6_walker w;
2242 	loff_t skip;
2243 	struct fib6_table *tbl;
2244 	int sernum;
2245 };
2246 
2247 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2248 {
2249 	struct fib6_info *rt = v;
2250 	struct ipv6_route_iter *iter = seq->private;
2251 	const struct net_device *dev;
2252 
2253 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2254 
2255 #ifdef CONFIG_IPV6_SUBTREES
2256 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2257 #else
2258 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2259 #endif
2260 	if (rt->fib6_flags & RTF_GATEWAY)
2261 		seq_printf(seq, "%pi6", &rt->fib6_nh.nh_gw);
2262 	else
2263 		seq_puts(seq, "00000000000000000000000000000000");
2264 
2265 	dev = rt->fib6_nh.nh_dev;
2266 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2267 		   rt->fib6_metric, atomic_read(&rt->fib6_ref), 0,
2268 		   rt->fib6_flags, dev ? dev->name : "");
2269 	iter->w.leaf = NULL;
2270 	return 0;
2271 }
2272 
2273 static int ipv6_route_yield(struct fib6_walker *w)
2274 {
2275 	struct ipv6_route_iter *iter = w->args;
2276 
2277 	if (!iter->skip)
2278 		return 1;
2279 
2280 	do {
2281 		iter->w.leaf = rcu_dereference_protected(
2282 				iter->w.leaf->rt6_next,
2283 				lockdep_is_held(&iter->tbl->tb6_lock));
2284 		iter->skip--;
2285 		if (!iter->skip && iter->w.leaf)
2286 			return 1;
2287 	} while (iter->w.leaf);
2288 
2289 	return 0;
2290 }
2291 
2292 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2293 				      struct net *net)
2294 {
2295 	memset(&iter->w, 0, sizeof(iter->w));
2296 	iter->w.func = ipv6_route_yield;
2297 	iter->w.root = &iter->tbl->tb6_root;
2298 	iter->w.state = FWS_INIT;
2299 	iter->w.node = iter->w.root;
2300 	iter->w.args = iter;
2301 	iter->sernum = iter->w.root->fn_sernum;
2302 	INIT_LIST_HEAD(&iter->w.lh);
2303 	fib6_walker_link(net, &iter->w);
2304 }
2305 
2306 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2307 						    struct net *net)
2308 {
2309 	unsigned int h;
2310 	struct hlist_node *node;
2311 
2312 	if (tbl) {
2313 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2314 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2315 	} else {
2316 		h = 0;
2317 		node = NULL;
2318 	}
2319 
2320 	while (!node && h < FIB6_TABLE_HASHSZ) {
2321 		node = rcu_dereference_bh(
2322 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2323 	}
2324 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2325 }
2326 
2327 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2328 {
2329 	if (iter->sernum != iter->w.root->fn_sernum) {
2330 		iter->sernum = iter->w.root->fn_sernum;
2331 		iter->w.state = FWS_INIT;
2332 		iter->w.node = iter->w.root;
2333 		WARN_ON(iter->w.skip);
2334 		iter->w.skip = iter->w.count;
2335 	}
2336 }
2337 
2338 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2339 {
2340 	int r;
2341 	struct fib6_info *n;
2342 	struct net *net = seq_file_net(seq);
2343 	struct ipv6_route_iter *iter = seq->private;
2344 
2345 	if (!v)
2346 		goto iter_table;
2347 
2348 	n = rcu_dereference_bh(((struct fib6_info *)v)->rt6_next);
2349 	if (n) {
2350 		++*pos;
2351 		return n;
2352 	}
2353 
2354 iter_table:
2355 	ipv6_route_check_sernum(iter);
2356 	spin_lock_bh(&iter->tbl->tb6_lock);
2357 	r = fib6_walk_continue(&iter->w);
2358 	spin_unlock_bh(&iter->tbl->tb6_lock);
2359 	if (r > 0) {
2360 		if (v)
2361 			++*pos;
2362 		return iter->w.leaf;
2363 	} else if (r < 0) {
2364 		fib6_walker_unlink(net, &iter->w);
2365 		return NULL;
2366 	}
2367 	fib6_walker_unlink(net, &iter->w);
2368 
2369 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2370 	if (!iter->tbl)
2371 		return NULL;
2372 
2373 	ipv6_route_seq_setup_walk(iter, net);
2374 	goto iter_table;
2375 }
2376 
2377 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2378 	__acquires(RCU_BH)
2379 {
2380 	struct net *net = seq_file_net(seq);
2381 	struct ipv6_route_iter *iter = seq->private;
2382 
2383 	rcu_read_lock_bh();
2384 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2385 	iter->skip = *pos;
2386 
2387 	if (iter->tbl) {
2388 		ipv6_route_seq_setup_walk(iter, net);
2389 		return ipv6_route_seq_next(seq, NULL, pos);
2390 	} else {
2391 		return NULL;
2392 	}
2393 }
2394 
2395 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2396 {
2397 	struct fib6_walker *w = &iter->w;
2398 	return w->node && !(w->state == FWS_U && w->node == w->root);
2399 }
2400 
2401 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2402 	__releases(RCU_BH)
2403 {
2404 	struct net *net = seq_file_net(seq);
2405 	struct ipv6_route_iter *iter = seq->private;
2406 
2407 	if (ipv6_route_iter_active(iter))
2408 		fib6_walker_unlink(net, &iter->w);
2409 
2410 	rcu_read_unlock_bh();
2411 }
2412 
2413 static const struct seq_operations ipv6_route_seq_ops = {
2414 	.start	= ipv6_route_seq_start,
2415 	.next	= ipv6_route_seq_next,
2416 	.stop	= ipv6_route_seq_stop,
2417 	.show	= ipv6_route_seq_show
2418 };
2419 
2420 int ipv6_route_open(struct inode *inode, struct file *file)
2421 {
2422 	return seq_open_net(inode, file, &ipv6_route_seq_ops,
2423 			    sizeof(struct ipv6_route_iter));
2424 }
2425 
2426 #endif /* CONFIG_PROC_FS */
2427