xref: /openbmc/linux/net/ipv6/ip6_fib.c (revision 4b0aaacee51eb6592a03fdefd5ce97558518e291)
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_rcu(struct rcu_head *head)
171 {
172 	struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
173 	struct rt6_exception_bucket *bucket;
174 	struct dst_metrics *m;
175 
176 	WARN_ON(f6i->fib6_node);
177 
178 	bucket = rcu_dereference_protected(f6i->rt6i_exception_bucket, 1);
179 	if (bucket) {
180 		f6i->rt6i_exception_bucket = NULL;
181 		kfree(bucket);
182 	}
183 
184 	if (f6i->rt6i_pcpu) {
185 		int cpu;
186 
187 		for_each_possible_cpu(cpu) {
188 			struct rt6_info **ppcpu_rt;
189 			struct rt6_info *pcpu_rt;
190 
191 			ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
192 			pcpu_rt = *ppcpu_rt;
193 			if (pcpu_rt) {
194 				dst_dev_put(&pcpu_rt->dst);
195 				dst_release(&pcpu_rt->dst);
196 				*ppcpu_rt = NULL;
197 			}
198 		}
199 	}
200 
201 	lwtstate_put(f6i->fib6_nh.nh_lwtstate);
202 
203 	if (f6i->fib6_nh.nh_dev)
204 		dev_put(f6i->fib6_nh.nh_dev);
205 
206 	m = f6i->fib6_metrics;
207 	if (m != &dst_default_metrics && refcount_dec_and_test(&m->refcnt))
208 		kfree(m);
209 
210 	kfree(f6i);
211 }
212 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
213 
214 static struct fib6_node *node_alloc(struct net *net)
215 {
216 	struct fib6_node *fn;
217 
218 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
219 	if (fn)
220 		net->ipv6.rt6_stats->fib_nodes++;
221 
222 	return fn;
223 }
224 
225 static void node_free_immediate(struct net *net, struct fib6_node *fn)
226 {
227 	kmem_cache_free(fib6_node_kmem, fn);
228 	net->ipv6.rt6_stats->fib_nodes--;
229 }
230 
231 static void node_free_rcu(struct rcu_head *head)
232 {
233 	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
234 
235 	kmem_cache_free(fib6_node_kmem, fn);
236 }
237 
238 static void node_free(struct net *net, struct fib6_node *fn)
239 {
240 	call_rcu(&fn->rcu, node_free_rcu);
241 	net->ipv6.rt6_stats->fib_nodes--;
242 }
243 
244 static void fib6_free_table(struct fib6_table *table)
245 {
246 	inetpeer_invalidate_tree(&table->tb6_peers);
247 	kfree(table);
248 }
249 
250 static void fib6_link_table(struct net *net, struct fib6_table *tb)
251 {
252 	unsigned int h;
253 
254 	/*
255 	 * Initialize table lock at a single place to give lockdep a key,
256 	 * tables aren't visible prior to being linked to the list.
257 	 */
258 	spin_lock_init(&tb->tb6_lock);
259 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
260 
261 	/*
262 	 * No protection necessary, this is the only list mutatation
263 	 * operation, tables never disappear once they exist.
264 	 */
265 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
266 }
267 
268 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
269 
270 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
271 {
272 	struct fib6_table *table;
273 
274 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
275 	if (table) {
276 		table->tb6_id = id;
277 		rcu_assign_pointer(table->tb6_root.leaf,
278 				   net->ipv6.fib6_null_entry);
279 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
280 		inet_peer_base_init(&table->tb6_peers);
281 	}
282 
283 	return table;
284 }
285 
286 struct fib6_table *fib6_new_table(struct net *net, u32 id)
287 {
288 	struct fib6_table *tb;
289 
290 	if (id == 0)
291 		id = RT6_TABLE_MAIN;
292 	tb = fib6_get_table(net, id);
293 	if (tb)
294 		return tb;
295 
296 	tb = fib6_alloc_table(net, id);
297 	if (tb)
298 		fib6_link_table(net, tb);
299 
300 	return tb;
301 }
302 EXPORT_SYMBOL_GPL(fib6_new_table);
303 
304 struct fib6_table *fib6_get_table(struct net *net, u32 id)
305 {
306 	struct fib6_table *tb;
307 	struct hlist_head *head;
308 	unsigned int h;
309 
310 	if (id == 0)
311 		id = RT6_TABLE_MAIN;
312 	h = id & (FIB6_TABLE_HASHSZ - 1);
313 	rcu_read_lock();
314 	head = &net->ipv6.fib_table_hash[h];
315 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
316 		if (tb->tb6_id == id) {
317 			rcu_read_unlock();
318 			return tb;
319 		}
320 	}
321 	rcu_read_unlock();
322 
323 	return NULL;
324 }
325 EXPORT_SYMBOL_GPL(fib6_get_table);
326 
327 static void __net_init fib6_tables_init(struct net *net)
328 {
329 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
330 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
331 }
332 #else
333 
334 struct fib6_table *fib6_new_table(struct net *net, u32 id)
335 {
336 	return fib6_get_table(net, id);
337 }
338 
339 struct fib6_table *fib6_get_table(struct net *net, u32 id)
340 {
341 	  return net->ipv6.fib6_main_tbl;
342 }
343 
344 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
345 				   const struct sk_buff *skb,
346 				   int flags, pol_lookup_t lookup)
347 {
348 	struct rt6_info *rt;
349 
350 	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
351 	if (rt->dst.error == -EAGAIN) {
352 		ip6_rt_put(rt);
353 		rt = net->ipv6.ip6_null_entry;
354 		dst_hold(&rt->dst);
355 	}
356 
357 	return &rt->dst;
358 }
359 
360 /* called with rcu lock held; no reference taken on fib6_info */
361 struct fib6_info *fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
362 			      int flags)
363 {
364 	return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6, flags);
365 }
366 
367 static void __net_init fib6_tables_init(struct net *net)
368 {
369 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
370 }
371 
372 #endif
373 
374 unsigned int fib6_tables_seq_read(struct net *net)
375 {
376 	unsigned int h, fib_seq = 0;
377 
378 	rcu_read_lock();
379 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
380 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
381 		struct fib6_table *tb;
382 
383 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
384 			fib_seq += tb->fib_seq;
385 	}
386 	rcu_read_unlock();
387 
388 	return fib_seq;
389 }
390 
391 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
392 				    enum fib_event_type event_type,
393 				    struct fib6_info *rt)
394 {
395 	struct fib6_entry_notifier_info info = {
396 		.rt = rt,
397 	};
398 
399 	return call_fib6_notifier(nb, net, event_type, &info.info);
400 }
401 
402 static int call_fib6_entry_notifiers(struct net *net,
403 				     enum fib_event_type event_type,
404 				     struct fib6_info *rt,
405 				     struct netlink_ext_ack *extack)
406 {
407 	struct fib6_entry_notifier_info info = {
408 		.info.extack = extack,
409 		.rt = rt,
410 	};
411 
412 	rt->fib6_table->fib_seq++;
413 	return call_fib6_notifiers(net, event_type, &info.info);
414 }
415 
416 struct fib6_dump_arg {
417 	struct net *net;
418 	struct notifier_block *nb;
419 };
420 
421 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
422 {
423 	if (rt == arg->net->ipv6.fib6_null_entry)
424 		return;
425 	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
426 }
427 
428 static int fib6_node_dump(struct fib6_walker *w)
429 {
430 	struct fib6_info *rt;
431 
432 	for_each_fib6_walker_rt(w)
433 		fib6_rt_dump(rt, w->args);
434 	w->leaf = NULL;
435 	return 0;
436 }
437 
438 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
439 			    struct fib6_walker *w)
440 {
441 	w->root = &tb->tb6_root;
442 	spin_lock_bh(&tb->tb6_lock);
443 	fib6_walk(net, w);
444 	spin_unlock_bh(&tb->tb6_lock);
445 }
446 
447 /* Called with rcu_read_lock() */
448 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
449 {
450 	struct fib6_dump_arg arg;
451 	struct fib6_walker *w;
452 	unsigned int h;
453 
454 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
455 	if (!w)
456 		return -ENOMEM;
457 
458 	w->func = fib6_node_dump;
459 	arg.net = net;
460 	arg.nb = nb;
461 	w->args = &arg;
462 
463 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
464 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
465 		struct fib6_table *tb;
466 
467 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
468 			fib6_table_dump(net, tb, w);
469 	}
470 
471 	kfree(w);
472 
473 	return 0;
474 }
475 
476 static int fib6_dump_node(struct fib6_walker *w)
477 {
478 	int res;
479 	struct fib6_info *rt;
480 
481 	for_each_fib6_walker_rt(w) {
482 		res = rt6_dump_route(rt, w->args);
483 		if (res < 0) {
484 			/* Frame is full, suspend walking */
485 			w->leaf = rt;
486 			return 1;
487 		}
488 
489 		/* Multipath routes are dumped in one route with the
490 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
491 		 * last sibling of this route (no need to dump the
492 		 * sibling routes again)
493 		 */
494 		if (rt->fib6_nsiblings)
495 			rt = list_last_entry(&rt->fib6_siblings,
496 					     struct fib6_info,
497 					     fib6_siblings);
498 	}
499 	w->leaf = NULL;
500 	return 0;
501 }
502 
503 static void fib6_dump_end(struct netlink_callback *cb)
504 {
505 	struct net *net = sock_net(cb->skb->sk);
506 	struct fib6_walker *w = (void *)cb->args[2];
507 
508 	if (w) {
509 		if (cb->args[4]) {
510 			cb->args[4] = 0;
511 			fib6_walker_unlink(net, w);
512 		}
513 		cb->args[2] = 0;
514 		kfree(w);
515 	}
516 	cb->done = (void *)cb->args[3];
517 	cb->args[1] = 3;
518 }
519 
520 static int fib6_dump_done(struct netlink_callback *cb)
521 {
522 	fib6_dump_end(cb);
523 	return cb->done ? cb->done(cb) : 0;
524 }
525 
526 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
527 			   struct netlink_callback *cb)
528 {
529 	struct net *net = sock_net(skb->sk);
530 	struct fib6_walker *w;
531 	int res;
532 
533 	w = (void *)cb->args[2];
534 	w->root = &table->tb6_root;
535 
536 	if (cb->args[4] == 0) {
537 		w->count = 0;
538 		w->skip = 0;
539 
540 		spin_lock_bh(&table->tb6_lock);
541 		res = fib6_walk(net, w);
542 		spin_unlock_bh(&table->tb6_lock);
543 		if (res > 0) {
544 			cb->args[4] = 1;
545 			cb->args[5] = w->root->fn_sernum;
546 		}
547 	} else {
548 		if (cb->args[5] != w->root->fn_sernum) {
549 			/* Begin at the root if the tree changed */
550 			cb->args[5] = w->root->fn_sernum;
551 			w->state = FWS_INIT;
552 			w->node = w->root;
553 			w->skip = w->count;
554 		} else
555 			w->skip = 0;
556 
557 		spin_lock_bh(&table->tb6_lock);
558 		res = fib6_walk_continue(w);
559 		spin_unlock_bh(&table->tb6_lock);
560 		if (res <= 0) {
561 			fib6_walker_unlink(net, w);
562 			cb->args[4] = 0;
563 		}
564 	}
565 
566 	return res;
567 }
568 
569 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
570 {
571 	struct net *net = sock_net(skb->sk);
572 	unsigned int h, s_h;
573 	unsigned int e = 0, s_e;
574 	struct rt6_rtnl_dump_arg arg;
575 	struct fib6_walker *w;
576 	struct fib6_table *tb;
577 	struct hlist_head *head;
578 	int res = 0;
579 
580 	s_h = cb->args[0];
581 	s_e = cb->args[1];
582 
583 	w = (void *)cb->args[2];
584 	if (!w) {
585 		/* New dump:
586 		 *
587 		 * 1. hook callback destructor.
588 		 */
589 		cb->args[3] = (long)cb->done;
590 		cb->done = fib6_dump_done;
591 
592 		/*
593 		 * 2. allocate and initialize walker.
594 		 */
595 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
596 		if (!w)
597 			return -ENOMEM;
598 		w->func = fib6_dump_node;
599 		cb->args[2] = (long)w;
600 	}
601 
602 	arg.skb = skb;
603 	arg.cb = cb;
604 	arg.net = net;
605 	w->args = &arg;
606 
607 	rcu_read_lock();
608 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
609 		e = 0;
610 		head = &net->ipv6.fib_table_hash[h];
611 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
612 			if (e < s_e)
613 				goto next;
614 			res = fib6_dump_table(tb, skb, cb);
615 			if (res != 0)
616 				goto out;
617 next:
618 			e++;
619 		}
620 	}
621 out:
622 	rcu_read_unlock();
623 	cb->args[1] = e;
624 	cb->args[0] = h;
625 
626 	res = res < 0 ? res : skb->len;
627 	if (res <= 0)
628 		fib6_dump_end(cb);
629 	return res;
630 }
631 
632 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
633 {
634 	if (!f6i)
635 		return;
636 
637 	if (f6i->fib6_metrics == &dst_default_metrics) {
638 		struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
639 
640 		if (!p)
641 			return;
642 
643 		refcount_set(&p->refcnt, 1);
644 		f6i->fib6_metrics = p;
645 	}
646 
647 	f6i->fib6_metrics->metrics[metric - 1] = val;
648 }
649 
650 /*
651  *	Routing Table
652  *
653  *	return the appropriate node for a routing tree "add" operation
654  *	by either creating and inserting or by returning an existing
655  *	node.
656  */
657 
658 static struct fib6_node *fib6_add_1(struct net *net,
659 				    struct fib6_table *table,
660 				    struct fib6_node *root,
661 				    struct in6_addr *addr, int plen,
662 				    int offset, int allow_create,
663 				    int replace_required,
664 				    struct netlink_ext_ack *extack)
665 {
666 	struct fib6_node *fn, *in, *ln;
667 	struct fib6_node *pn = NULL;
668 	struct rt6key *key;
669 	int	bit;
670 	__be32	dir = 0;
671 
672 	RT6_TRACE("fib6_add_1\n");
673 
674 	/* insert node in tree */
675 
676 	fn = root;
677 
678 	do {
679 		struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
680 					    lockdep_is_held(&table->tb6_lock));
681 		key = (struct rt6key *)((u8 *)leaf + offset);
682 
683 		/*
684 		 *	Prefix match
685 		 */
686 		if (plen < fn->fn_bit ||
687 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
688 			if (!allow_create) {
689 				if (replace_required) {
690 					NL_SET_ERR_MSG(extack,
691 						       "Can not replace route - no match found");
692 					pr_warn("Can't replace route, no match found\n");
693 					return ERR_PTR(-ENOENT);
694 				}
695 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
696 			}
697 			goto insert_above;
698 		}
699 
700 		/*
701 		 *	Exact match ?
702 		 */
703 
704 		if (plen == fn->fn_bit) {
705 			/* clean up an intermediate node */
706 			if (!(fn->fn_flags & RTN_RTINFO)) {
707 				RCU_INIT_POINTER(fn->leaf, NULL);
708 				fib6_info_release(leaf);
709 			/* remove null_entry in the root node */
710 			} else if (fn->fn_flags & RTN_TL_ROOT &&
711 				   rcu_access_pointer(fn->leaf) ==
712 				   net->ipv6.fib6_null_entry) {
713 				RCU_INIT_POINTER(fn->leaf, NULL);
714 			}
715 
716 			return fn;
717 		}
718 
719 		/*
720 		 *	We have more bits to go
721 		 */
722 
723 		/* Try to walk down on tree. */
724 		dir = addr_bit_set(addr, fn->fn_bit);
725 		pn = fn;
726 		fn = dir ?
727 		     rcu_dereference_protected(fn->right,
728 					lockdep_is_held(&table->tb6_lock)) :
729 		     rcu_dereference_protected(fn->left,
730 					lockdep_is_held(&table->tb6_lock));
731 	} while (fn);
732 
733 	if (!allow_create) {
734 		/* We should not create new node because
735 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
736 		 * I assume it is safe to require NLM_F_CREATE when
737 		 * REPLACE flag is used! Later we may want to remove the
738 		 * check for replace_required, because according
739 		 * to netlink specification, NLM_F_CREATE
740 		 * MUST be specified if new route is created.
741 		 * That would keep IPv6 consistent with IPv4
742 		 */
743 		if (replace_required) {
744 			NL_SET_ERR_MSG(extack,
745 				       "Can not replace route - no match found");
746 			pr_warn("Can't replace route, no match found\n");
747 			return ERR_PTR(-ENOENT);
748 		}
749 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
750 	}
751 	/*
752 	 *	We walked to the bottom of tree.
753 	 *	Create new leaf node without children.
754 	 */
755 
756 	ln = node_alloc(net);
757 
758 	if (!ln)
759 		return ERR_PTR(-ENOMEM);
760 	ln->fn_bit = plen;
761 	RCU_INIT_POINTER(ln->parent, pn);
762 
763 	if (dir)
764 		rcu_assign_pointer(pn->right, ln);
765 	else
766 		rcu_assign_pointer(pn->left, ln);
767 
768 	return ln;
769 
770 
771 insert_above:
772 	/*
773 	 * split since we don't have a common prefix anymore or
774 	 * we have a less significant route.
775 	 * we've to insert an intermediate node on the list
776 	 * this new node will point to the one we need to create
777 	 * and the current
778 	 */
779 
780 	pn = rcu_dereference_protected(fn->parent,
781 				       lockdep_is_held(&table->tb6_lock));
782 
783 	/* find 1st bit in difference between the 2 addrs.
784 
785 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
786 	   but if it is >= plen, the value is ignored in any case.
787 	 */
788 
789 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
790 
791 	/*
792 	 *		(intermediate)[in]
793 	 *	          /	   \
794 	 *	(new leaf node)[ln] (old node)[fn]
795 	 */
796 	if (plen > bit) {
797 		in = node_alloc(net);
798 		ln = node_alloc(net);
799 
800 		if (!in || !ln) {
801 			if (in)
802 				node_free_immediate(net, in);
803 			if (ln)
804 				node_free_immediate(net, ln);
805 			return ERR_PTR(-ENOMEM);
806 		}
807 
808 		/*
809 		 * new intermediate node.
810 		 * RTN_RTINFO will
811 		 * be off since that an address that chooses one of
812 		 * the branches would not match less specific routes
813 		 * in the other branch
814 		 */
815 
816 		in->fn_bit = bit;
817 
818 		RCU_INIT_POINTER(in->parent, pn);
819 		in->leaf = fn->leaf;
820 		atomic_inc(&rcu_dereference_protected(in->leaf,
821 				lockdep_is_held(&table->tb6_lock))->fib6_ref);
822 
823 		/* update parent pointer */
824 		if (dir)
825 			rcu_assign_pointer(pn->right, in);
826 		else
827 			rcu_assign_pointer(pn->left, in);
828 
829 		ln->fn_bit = plen;
830 
831 		RCU_INIT_POINTER(ln->parent, in);
832 		rcu_assign_pointer(fn->parent, in);
833 
834 		if (addr_bit_set(addr, bit)) {
835 			rcu_assign_pointer(in->right, ln);
836 			rcu_assign_pointer(in->left, fn);
837 		} else {
838 			rcu_assign_pointer(in->left, ln);
839 			rcu_assign_pointer(in->right, fn);
840 		}
841 	} else { /* plen <= bit */
842 
843 		/*
844 		 *		(new leaf node)[ln]
845 		 *	          /	   \
846 		 *	     (old node)[fn] NULL
847 		 */
848 
849 		ln = node_alloc(net);
850 
851 		if (!ln)
852 			return ERR_PTR(-ENOMEM);
853 
854 		ln->fn_bit = plen;
855 
856 		RCU_INIT_POINTER(ln->parent, pn);
857 
858 		if (addr_bit_set(&key->addr, plen))
859 			RCU_INIT_POINTER(ln->right, fn);
860 		else
861 			RCU_INIT_POINTER(ln->left, fn);
862 
863 		rcu_assign_pointer(fn->parent, ln);
864 
865 		if (dir)
866 			rcu_assign_pointer(pn->right, ln);
867 		else
868 			rcu_assign_pointer(pn->left, ln);
869 	}
870 	return ln;
871 }
872 
873 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
874 				const struct fib6_table *table)
875 {
876 	int cpu;
877 
878 	/* release the reference to this fib entry from
879 	 * all of its cached pcpu routes
880 	 */
881 	for_each_possible_cpu(cpu) {
882 		struct rt6_info **ppcpu_rt;
883 		struct rt6_info *pcpu_rt;
884 
885 		ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
886 		pcpu_rt = *ppcpu_rt;
887 		if (pcpu_rt) {
888 			struct fib6_info *from;
889 
890 			from = rcu_dereference_protected(pcpu_rt->from,
891 					     lockdep_is_held(&table->tb6_lock));
892 			rcu_assign_pointer(pcpu_rt->from, NULL);
893 			fib6_info_release(from);
894 		}
895 	}
896 }
897 
898 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
899 			  struct net *net)
900 {
901 	struct fib6_table *table = rt->fib6_table;
902 
903 	if (atomic_read(&rt->fib6_ref) != 1) {
904 		/* This route is used as dummy address holder in some split
905 		 * nodes. It is not leaked, but it still holds other resources,
906 		 * which must be released in time. So, scan ascendant nodes
907 		 * and replace dummy references to this route with references
908 		 * to still alive ones.
909 		 */
910 		while (fn) {
911 			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
912 					    lockdep_is_held(&table->tb6_lock));
913 			struct fib6_info *new_leaf;
914 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
915 				new_leaf = fib6_find_prefix(net, table, fn);
916 				atomic_inc(&new_leaf->fib6_ref);
917 
918 				rcu_assign_pointer(fn->leaf, new_leaf);
919 				fib6_info_release(rt);
920 			}
921 			fn = rcu_dereference_protected(fn->parent,
922 				    lockdep_is_held(&table->tb6_lock));
923 		}
924 
925 		if (rt->rt6i_pcpu)
926 			fib6_drop_pcpu_from(rt, table);
927 	}
928 }
929 
930 /*
931  *	Insert routing information in a node.
932  */
933 
934 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
935 			    struct nl_info *info,
936 			    struct netlink_ext_ack *extack)
937 {
938 	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
939 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
940 	struct fib6_info *iter = NULL;
941 	struct fib6_info __rcu **ins;
942 	struct fib6_info __rcu **fallback_ins = NULL;
943 	int replace = (info->nlh &&
944 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
945 	int add = (!info->nlh ||
946 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
947 	int found = 0;
948 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
949 	u16 nlflags = NLM_F_EXCL;
950 	int err;
951 
952 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
953 		nlflags |= NLM_F_APPEND;
954 
955 	ins = &fn->leaf;
956 
957 	for (iter = leaf; iter;
958 	     iter = rcu_dereference_protected(iter->fib6_next,
959 				lockdep_is_held(&rt->fib6_table->tb6_lock))) {
960 		/*
961 		 *	Search for duplicates
962 		 */
963 
964 		if (iter->fib6_metric == rt->fib6_metric) {
965 			/*
966 			 *	Same priority level
967 			 */
968 			if (info->nlh &&
969 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
970 				return -EEXIST;
971 
972 			nlflags &= ~NLM_F_EXCL;
973 			if (replace) {
974 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
975 					found++;
976 					break;
977 				}
978 				if (rt_can_ecmp)
979 					fallback_ins = fallback_ins ?: ins;
980 				goto next_iter;
981 			}
982 
983 			if (rt6_duplicate_nexthop(iter, rt)) {
984 				if (rt->fib6_nsiblings)
985 					rt->fib6_nsiblings = 0;
986 				if (!(iter->fib6_flags & RTF_EXPIRES))
987 					return -EEXIST;
988 				if (!(rt->fib6_flags & RTF_EXPIRES))
989 					fib6_clean_expires(iter);
990 				else
991 					fib6_set_expires(iter, rt->expires);
992 
993 				if (rt->fib6_pmtu)
994 					fib6_metric_set(iter, RTAX_MTU,
995 							rt->fib6_pmtu);
996 				return -EEXIST;
997 			}
998 			/* If we have the same destination and the same metric,
999 			 * but not the same gateway, then the route we try to
1000 			 * add is sibling to this route, increment our counter
1001 			 * of siblings, and later we will add our route to the
1002 			 * list.
1003 			 * Only static routes (which don't have flag
1004 			 * RTF_EXPIRES) are used for ECMPv6.
1005 			 *
1006 			 * To avoid long list, we only had siblings if the
1007 			 * route have a gateway.
1008 			 */
1009 			if (rt_can_ecmp &&
1010 			    rt6_qualify_for_ecmp(iter))
1011 				rt->fib6_nsiblings++;
1012 		}
1013 
1014 		if (iter->fib6_metric > rt->fib6_metric)
1015 			break;
1016 
1017 next_iter:
1018 		ins = &iter->fib6_next;
1019 	}
1020 
1021 	if (fallback_ins && !found) {
1022 		/* No ECMP-able route found, replace first non-ECMP one */
1023 		ins = fallback_ins;
1024 		iter = rcu_dereference_protected(*ins,
1025 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1026 		found++;
1027 	}
1028 
1029 	/* Reset round-robin state, if necessary */
1030 	if (ins == &fn->leaf)
1031 		fn->rr_ptr = NULL;
1032 
1033 	/* Link this route to others same route. */
1034 	if (rt->fib6_nsiblings) {
1035 		unsigned int fib6_nsiblings;
1036 		struct fib6_info *sibling, *temp_sibling;
1037 
1038 		/* Find the first route that have the same metric */
1039 		sibling = leaf;
1040 		while (sibling) {
1041 			if (sibling->fib6_metric == rt->fib6_metric &&
1042 			    rt6_qualify_for_ecmp(sibling)) {
1043 				list_add_tail(&rt->fib6_siblings,
1044 					      &sibling->fib6_siblings);
1045 				break;
1046 			}
1047 			sibling = rcu_dereference_protected(sibling->fib6_next,
1048 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1049 		}
1050 		/* For each sibling in the list, increment the counter of
1051 		 * siblings. BUG() if counters does not match, list of siblings
1052 		 * is broken!
1053 		 */
1054 		fib6_nsiblings = 0;
1055 		list_for_each_entry_safe(sibling, temp_sibling,
1056 					 &rt->fib6_siblings, fib6_siblings) {
1057 			sibling->fib6_nsiblings++;
1058 			BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1059 			fib6_nsiblings++;
1060 		}
1061 		BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1062 		rt6_multipath_rebalance(temp_sibling);
1063 	}
1064 
1065 	/*
1066 	 *	insert node
1067 	 */
1068 	if (!replace) {
1069 		if (!add)
1070 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1071 
1072 add:
1073 		nlflags |= NLM_F_CREATE;
1074 
1075 		err = call_fib6_entry_notifiers(info->nl_net,
1076 						FIB_EVENT_ENTRY_ADD,
1077 						rt, extack);
1078 		if (err)
1079 			return err;
1080 
1081 		rcu_assign_pointer(rt->fib6_next, iter);
1082 		atomic_inc(&rt->fib6_ref);
1083 		rcu_assign_pointer(rt->fib6_node, fn);
1084 		rcu_assign_pointer(*ins, rt);
1085 		if (!info->skip_notify)
1086 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1087 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1088 
1089 		if (!(fn->fn_flags & RTN_RTINFO)) {
1090 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1091 			fn->fn_flags |= RTN_RTINFO;
1092 		}
1093 
1094 	} else {
1095 		int nsiblings;
1096 
1097 		if (!found) {
1098 			if (add)
1099 				goto add;
1100 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1101 			return -ENOENT;
1102 		}
1103 
1104 		err = call_fib6_entry_notifiers(info->nl_net,
1105 						FIB_EVENT_ENTRY_REPLACE,
1106 						rt, extack);
1107 		if (err)
1108 			return err;
1109 
1110 		atomic_inc(&rt->fib6_ref);
1111 		rcu_assign_pointer(rt->fib6_node, fn);
1112 		rt->fib6_next = iter->fib6_next;
1113 		rcu_assign_pointer(*ins, rt);
1114 		if (!info->skip_notify)
1115 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1116 		if (!(fn->fn_flags & RTN_RTINFO)) {
1117 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1118 			fn->fn_flags |= RTN_RTINFO;
1119 		}
1120 		nsiblings = iter->fib6_nsiblings;
1121 		iter->fib6_node = NULL;
1122 		fib6_purge_rt(iter, fn, info->nl_net);
1123 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1124 			fn->rr_ptr = NULL;
1125 		fib6_info_release(iter);
1126 
1127 		if (nsiblings) {
1128 			/* Replacing an ECMP route, remove all siblings */
1129 			ins = &rt->fib6_next;
1130 			iter = rcu_dereference_protected(*ins,
1131 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1132 			while (iter) {
1133 				if (iter->fib6_metric > rt->fib6_metric)
1134 					break;
1135 				if (rt6_qualify_for_ecmp(iter)) {
1136 					*ins = iter->fib6_next;
1137 					iter->fib6_node = NULL;
1138 					fib6_purge_rt(iter, fn, info->nl_net);
1139 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1140 						fn->rr_ptr = NULL;
1141 					fib6_info_release(iter);
1142 					nsiblings--;
1143 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1144 				} else {
1145 					ins = &iter->fib6_next;
1146 				}
1147 				iter = rcu_dereference_protected(*ins,
1148 					lockdep_is_held(&rt->fib6_table->tb6_lock));
1149 			}
1150 			WARN_ON(nsiblings != 0);
1151 		}
1152 	}
1153 
1154 	return 0;
1155 }
1156 
1157 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1158 {
1159 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1160 	    (rt->fib6_flags & RTF_EXPIRES))
1161 		mod_timer(&net->ipv6.ip6_fib_timer,
1162 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1163 }
1164 
1165 void fib6_force_start_gc(struct net *net)
1166 {
1167 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1168 		mod_timer(&net->ipv6.ip6_fib_timer,
1169 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1170 }
1171 
1172 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1173 					   int sernum)
1174 {
1175 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1176 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1177 
1178 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1179 	smp_wmb();
1180 	while (fn) {
1181 		fn->fn_sernum = sernum;
1182 		fn = rcu_dereference_protected(fn->parent,
1183 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1184 	}
1185 }
1186 
1187 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1188 {
1189 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1190 }
1191 
1192 /*
1193  *	Add routing information to the routing tree.
1194  *	<destination addr>/<source addr>
1195  *	with source addr info in sub-trees
1196  *	Need to own table->tb6_lock
1197  */
1198 
1199 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1200 	     struct nl_info *info, struct netlink_ext_ack *extack)
1201 {
1202 	struct fib6_table *table = rt->fib6_table;
1203 	struct fib6_node *fn, *pn = NULL;
1204 	int err = -ENOMEM;
1205 	int allow_create = 1;
1206 	int replace_required = 0;
1207 	int sernum = fib6_new_sernum(info->nl_net);
1208 
1209 	if (info->nlh) {
1210 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1211 			allow_create = 0;
1212 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1213 			replace_required = 1;
1214 	}
1215 	if (!allow_create && !replace_required)
1216 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1217 
1218 	fn = fib6_add_1(info->nl_net, table, root,
1219 			&rt->fib6_dst.addr, rt->fib6_dst.plen,
1220 			offsetof(struct fib6_info, fib6_dst), allow_create,
1221 			replace_required, extack);
1222 	if (IS_ERR(fn)) {
1223 		err = PTR_ERR(fn);
1224 		fn = NULL;
1225 		goto out;
1226 	}
1227 
1228 	pn = fn;
1229 
1230 #ifdef CONFIG_IPV6_SUBTREES
1231 	if (rt->fib6_src.plen) {
1232 		struct fib6_node *sn;
1233 
1234 		if (!rcu_access_pointer(fn->subtree)) {
1235 			struct fib6_node *sfn;
1236 
1237 			/*
1238 			 * Create subtree.
1239 			 *
1240 			 *		fn[main tree]
1241 			 *		|
1242 			 *		sfn[subtree root]
1243 			 *		   \
1244 			 *		    sn[new leaf node]
1245 			 */
1246 
1247 			/* Create subtree root node */
1248 			sfn = node_alloc(info->nl_net);
1249 			if (!sfn)
1250 				goto failure;
1251 
1252 			atomic_inc(&info->nl_net->ipv6.fib6_null_entry->fib6_ref);
1253 			rcu_assign_pointer(sfn->leaf,
1254 					   info->nl_net->ipv6.fib6_null_entry);
1255 			sfn->fn_flags = RTN_ROOT;
1256 
1257 			/* Now add the first leaf node to new subtree */
1258 
1259 			sn = fib6_add_1(info->nl_net, table, sfn,
1260 					&rt->fib6_src.addr, rt->fib6_src.plen,
1261 					offsetof(struct fib6_info, fib6_src),
1262 					allow_create, replace_required, extack);
1263 
1264 			if (IS_ERR(sn)) {
1265 				/* If it is failed, discard just allocated
1266 				   root, and then (in failure) stale node
1267 				   in main tree.
1268 				 */
1269 				node_free_immediate(info->nl_net, sfn);
1270 				err = PTR_ERR(sn);
1271 				goto failure;
1272 			}
1273 
1274 			/* Now link new subtree to main tree */
1275 			rcu_assign_pointer(sfn->parent, fn);
1276 			rcu_assign_pointer(fn->subtree, sfn);
1277 		} else {
1278 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1279 					&rt->fib6_src.addr, rt->fib6_src.plen,
1280 					offsetof(struct fib6_info, fib6_src),
1281 					allow_create, replace_required, extack);
1282 
1283 			if (IS_ERR(sn)) {
1284 				err = PTR_ERR(sn);
1285 				goto failure;
1286 			}
1287 		}
1288 
1289 		if (!rcu_access_pointer(fn->leaf)) {
1290 			if (fn->fn_flags & RTN_TL_ROOT) {
1291 				/* put back null_entry for root node */
1292 				rcu_assign_pointer(fn->leaf,
1293 					    info->nl_net->ipv6.fib6_null_entry);
1294 			} else {
1295 				atomic_inc(&rt->fib6_ref);
1296 				rcu_assign_pointer(fn->leaf, rt);
1297 			}
1298 		}
1299 		fn = sn;
1300 	}
1301 #endif
1302 
1303 	err = fib6_add_rt2node(fn, rt, info, extack);
1304 	if (!err) {
1305 		__fib6_update_sernum_upto_root(rt, sernum);
1306 		fib6_start_gc(info->nl_net, rt);
1307 	}
1308 
1309 out:
1310 	if (err) {
1311 #ifdef CONFIG_IPV6_SUBTREES
1312 		/*
1313 		 * If fib6_add_1 has cleared the old leaf pointer in the
1314 		 * super-tree leaf node we have to find a new one for it.
1315 		 */
1316 		if (pn != fn) {
1317 			struct fib6_info *pn_leaf =
1318 				rcu_dereference_protected(pn->leaf,
1319 				    lockdep_is_held(&table->tb6_lock));
1320 			if (pn_leaf == rt) {
1321 				pn_leaf = NULL;
1322 				RCU_INIT_POINTER(pn->leaf, NULL);
1323 				fib6_info_release(rt);
1324 			}
1325 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1326 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1327 							   pn);
1328 #if RT6_DEBUG >= 2
1329 				if (!pn_leaf) {
1330 					WARN_ON(!pn_leaf);
1331 					pn_leaf =
1332 					    info->nl_net->ipv6.fib6_null_entry;
1333 				}
1334 #endif
1335 				fib6_info_hold(pn_leaf);
1336 				rcu_assign_pointer(pn->leaf, pn_leaf);
1337 			}
1338 		}
1339 #endif
1340 		goto failure;
1341 	}
1342 	return err;
1343 
1344 failure:
1345 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1346 	 * 1. fn is an intermediate node and we failed to add the new
1347 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1348 	 * failure case.
1349 	 * 2. fn is the root node in the table and we fail to add the first
1350 	 * default route to it.
1351 	 */
1352 	if (fn &&
1353 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1354 	     (fn->fn_flags & RTN_TL_ROOT &&
1355 	      !rcu_access_pointer(fn->leaf))))
1356 		fib6_repair_tree(info->nl_net, table, fn);
1357 	return err;
1358 }
1359 
1360 /*
1361  *	Routing tree lookup
1362  *
1363  */
1364 
1365 struct lookup_args {
1366 	int			offset;		/* key offset on fib6_info */
1367 	const struct in6_addr	*addr;		/* search key			*/
1368 };
1369 
1370 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1371 					    struct lookup_args *args)
1372 {
1373 	struct fib6_node *fn;
1374 	__be32 dir;
1375 
1376 	if (unlikely(args->offset == 0))
1377 		return NULL;
1378 
1379 	/*
1380 	 *	Descend on a tree
1381 	 */
1382 
1383 	fn = root;
1384 
1385 	for (;;) {
1386 		struct fib6_node *next;
1387 
1388 		dir = addr_bit_set(args->addr, fn->fn_bit);
1389 
1390 		next = dir ? rcu_dereference(fn->right) :
1391 			     rcu_dereference(fn->left);
1392 
1393 		if (next) {
1394 			fn = next;
1395 			continue;
1396 		}
1397 		break;
1398 	}
1399 
1400 	while (fn) {
1401 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1402 
1403 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1404 			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1405 			struct rt6key *key;
1406 
1407 			if (!leaf)
1408 				goto backtrack;
1409 
1410 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1411 
1412 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1413 #ifdef CONFIG_IPV6_SUBTREES
1414 				if (subtree) {
1415 					struct fib6_node *sfn;
1416 					sfn = fib6_node_lookup_1(subtree,
1417 								 args + 1);
1418 					if (!sfn)
1419 						goto backtrack;
1420 					fn = sfn;
1421 				}
1422 #endif
1423 				if (fn->fn_flags & RTN_RTINFO)
1424 					return fn;
1425 			}
1426 		}
1427 backtrack:
1428 		if (fn->fn_flags & RTN_ROOT)
1429 			break;
1430 
1431 		fn = rcu_dereference(fn->parent);
1432 	}
1433 
1434 	return NULL;
1435 }
1436 
1437 /* called with rcu_read_lock() held
1438  */
1439 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1440 				   const struct in6_addr *daddr,
1441 				   const struct in6_addr *saddr)
1442 {
1443 	struct fib6_node *fn;
1444 	struct lookup_args args[] = {
1445 		{
1446 			.offset = offsetof(struct fib6_info, fib6_dst),
1447 			.addr = daddr,
1448 		},
1449 #ifdef CONFIG_IPV6_SUBTREES
1450 		{
1451 			.offset = offsetof(struct fib6_info, fib6_src),
1452 			.addr = saddr,
1453 		},
1454 #endif
1455 		{
1456 			.offset = 0,	/* sentinel */
1457 		}
1458 	};
1459 
1460 	fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1461 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1462 		fn = root;
1463 
1464 	return fn;
1465 }
1466 
1467 /*
1468  *	Get node with specified destination prefix (and source prefix,
1469  *	if subtrees are used)
1470  *	exact_match == true means we try to find fn with exact match of
1471  *	the passed in prefix addr
1472  *	exact_match == false means we try to find fn with longest prefix
1473  *	match of the passed in prefix addr. This is useful for finding fn
1474  *	for cached route as it will be stored in the exception table under
1475  *	the node with longest prefix length.
1476  */
1477 
1478 
1479 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1480 				       const struct in6_addr *addr,
1481 				       int plen, int offset,
1482 				       bool exact_match)
1483 {
1484 	struct fib6_node *fn, *prev = NULL;
1485 
1486 	for (fn = root; fn ; ) {
1487 		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1488 		struct rt6key *key;
1489 
1490 		/* This node is being deleted */
1491 		if (!leaf) {
1492 			if (plen <= fn->fn_bit)
1493 				goto out;
1494 			else
1495 				goto next;
1496 		}
1497 
1498 		key = (struct rt6key *)((u8 *)leaf + offset);
1499 
1500 		/*
1501 		 *	Prefix match
1502 		 */
1503 		if (plen < fn->fn_bit ||
1504 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1505 			goto out;
1506 
1507 		if (plen == fn->fn_bit)
1508 			return fn;
1509 
1510 		prev = fn;
1511 
1512 next:
1513 		/*
1514 		 *	We have more bits to go
1515 		 */
1516 		if (addr_bit_set(addr, fn->fn_bit))
1517 			fn = rcu_dereference(fn->right);
1518 		else
1519 			fn = rcu_dereference(fn->left);
1520 	}
1521 out:
1522 	if (exact_match)
1523 		return NULL;
1524 	else
1525 		return prev;
1526 }
1527 
1528 struct fib6_node *fib6_locate(struct fib6_node *root,
1529 			      const struct in6_addr *daddr, int dst_len,
1530 			      const struct in6_addr *saddr, int src_len,
1531 			      bool exact_match)
1532 {
1533 	struct fib6_node *fn;
1534 
1535 	fn = fib6_locate_1(root, daddr, dst_len,
1536 			   offsetof(struct fib6_info, fib6_dst),
1537 			   exact_match);
1538 
1539 #ifdef CONFIG_IPV6_SUBTREES
1540 	if (src_len) {
1541 		WARN_ON(saddr == NULL);
1542 		if (fn) {
1543 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1544 
1545 			if (subtree) {
1546 				fn = fib6_locate_1(subtree, saddr, src_len,
1547 					   offsetof(struct fib6_info, fib6_src),
1548 					   exact_match);
1549 			}
1550 		}
1551 	}
1552 #endif
1553 
1554 	if (fn && fn->fn_flags & RTN_RTINFO)
1555 		return fn;
1556 
1557 	return NULL;
1558 }
1559 
1560 
1561 /*
1562  *	Deletion
1563  *
1564  */
1565 
1566 static struct fib6_info *fib6_find_prefix(struct net *net,
1567 					 struct fib6_table *table,
1568 					 struct fib6_node *fn)
1569 {
1570 	struct fib6_node *child_left, *child_right;
1571 
1572 	if (fn->fn_flags & RTN_ROOT)
1573 		return net->ipv6.fib6_null_entry;
1574 
1575 	while (fn) {
1576 		child_left = rcu_dereference_protected(fn->left,
1577 				    lockdep_is_held(&table->tb6_lock));
1578 		child_right = rcu_dereference_protected(fn->right,
1579 				    lockdep_is_held(&table->tb6_lock));
1580 		if (child_left)
1581 			return rcu_dereference_protected(child_left->leaf,
1582 					lockdep_is_held(&table->tb6_lock));
1583 		if (child_right)
1584 			return rcu_dereference_protected(child_right->leaf,
1585 					lockdep_is_held(&table->tb6_lock));
1586 
1587 		fn = FIB6_SUBTREE(fn);
1588 	}
1589 	return NULL;
1590 }
1591 
1592 /*
1593  *	Called to trim the tree of intermediate nodes when possible. "fn"
1594  *	is the node we want to try and remove.
1595  *	Need to own table->tb6_lock
1596  */
1597 
1598 static struct fib6_node *fib6_repair_tree(struct net *net,
1599 					  struct fib6_table *table,
1600 					  struct fib6_node *fn)
1601 {
1602 	int children;
1603 	int nstate;
1604 	struct fib6_node *child;
1605 	struct fib6_walker *w;
1606 	int iter = 0;
1607 
1608 	/* Set fn->leaf to null_entry for root node. */
1609 	if (fn->fn_flags & RTN_TL_ROOT) {
1610 		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1611 		return fn;
1612 	}
1613 
1614 	for (;;) {
1615 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1616 					    lockdep_is_held(&table->tb6_lock));
1617 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1618 					    lockdep_is_held(&table->tb6_lock));
1619 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1620 					    lockdep_is_held(&table->tb6_lock));
1621 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1622 					    lockdep_is_held(&table->tb6_lock));
1623 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1624 					    lockdep_is_held(&table->tb6_lock));
1625 		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1626 					    lockdep_is_held(&table->tb6_lock));
1627 		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1628 					    lockdep_is_held(&table->tb6_lock));
1629 		struct fib6_info *new_fn_leaf;
1630 
1631 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1632 		iter++;
1633 
1634 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1635 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1636 		WARN_ON(fn_leaf);
1637 
1638 		children = 0;
1639 		child = NULL;
1640 		if (fn_r)
1641 			child = fn_r, children |= 1;
1642 		if (fn_l)
1643 			child = fn_l, children |= 2;
1644 
1645 		if (children == 3 || FIB6_SUBTREE(fn)
1646 #ifdef CONFIG_IPV6_SUBTREES
1647 		    /* Subtree root (i.e. fn) may have one child */
1648 		    || (children && fn->fn_flags & RTN_ROOT)
1649 #endif
1650 		    ) {
1651 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1652 #if RT6_DEBUG >= 2
1653 			if (!new_fn_leaf) {
1654 				WARN_ON(!new_fn_leaf);
1655 				new_fn_leaf = net->ipv6.fib6_null_entry;
1656 			}
1657 #endif
1658 			fib6_info_hold(new_fn_leaf);
1659 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1660 			return pn;
1661 		}
1662 
1663 #ifdef CONFIG_IPV6_SUBTREES
1664 		if (FIB6_SUBTREE(pn) == fn) {
1665 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1666 			RCU_INIT_POINTER(pn->subtree, NULL);
1667 			nstate = FWS_L;
1668 		} else {
1669 			WARN_ON(fn->fn_flags & RTN_ROOT);
1670 #endif
1671 			if (pn_r == fn)
1672 				rcu_assign_pointer(pn->right, child);
1673 			else if (pn_l == fn)
1674 				rcu_assign_pointer(pn->left, child);
1675 #if RT6_DEBUG >= 2
1676 			else
1677 				WARN_ON(1);
1678 #endif
1679 			if (child)
1680 				rcu_assign_pointer(child->parent, pn);
1681 			nstate = FWS_R;
1682 #ifdef CONFIG_IPV6_SUBTREES
1683 		}
1684 #endif
1685 
1686 		read_lock(&net->ipv6.fib6_walker_lock);
1687 		FOR_WALKERS(net, w) {
1688 			if (!child) {
1689 				if (w->node == fn) {
1690 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1691 					w->node = pn;
1692 					w->state = nstate;
1693 				}
1694 			} else {
1695 				if (w->node == fn) {
1696 					w->node = child;
1697 					if (children&2) {
1698 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1699 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1700 					} else {
1701 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1702 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1703 					}
1704 				}
1705 			}
1706 		}
1707 		read_unlock(&net->ipv6.fib6_walker_lock);
1708 
1709 		node_free(net, fn);
1710 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1711 			return pn;
1712 
1713 		RCU_INIT_POINTER(pn->leaf, NULL);
1714 		fib6_info_release(pn_leaf);
1715 		fn = pn;
1716 	}
1717 }
1718 
1719 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1720 			   struct fib6_info __rcu **rtp, struct nl_info *info)
1721 {
1722 	struct fib6_walker *w;
1723 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1724 				    lockdep_is_held(&table->tb6_lock));
1725 	struct net *net = info->nl_net;
1726 
1727 	RT6_TRACE("fib6_del_route\n");
1728 
1729 	/* Unlink it */
1730 	*rtp = rt->fib6_next;
1731 	rt->fib6_node = NULL;
1732 	net->ipv6.rt6_stats->fib_rt_entries--;
1733 	net->ipv6.rt6_stats->fib_discarded_routes++;
1734 
1735 	/* Flush all cached dst in exception table */
1736 	rt6_flush_exceptions(rt);
1737 
1738 	/* Reset round-robin state, if necessary */
1739 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1740 		fn->rr_ptr = NULL;
1741 
1742 	/* Remove this entry from other siblings */
1743 	if (rt->fib6_nsiblings) {
1744 		struct fib6_info *sibling, *next_sibling;
1745 
1746 		list_for_each_entry_safe(sibling, next_sibling,
1747 					 &rt->fib6_siblings, fib6_siblings)
1748 			sibling->fib6_nsiblings--;
1749 		rt->fib6_nsiblings = 0;
1750 		list_del_init(&rt->fib6_siblings);
1751 		rt6_multipath_rebalance(next_sibling);
1752 	}
1753 
1754 	/* Adjust walkers */
1755 	read_lock(&net->ipv6.fib6_walker_lock);
1756 	FOR_WALKERS(net, w) {
1757 		if (w->state == FWS_C && w->leaf == rt) {
1758 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1759 			w->leaf = rcu_dereference_protected(rt->fib6_next,
1760 					    lockdep_is_held(&table->tb6_lock));
1761 			if (!w->leaf)
1762 				w->state = FWS_U;
1763 		}
1764 	}
1765 	read_unlock(&net->ipv6.fib6_walker_lock);
1766 
1767 	/* If it was last route, call fib6_repair_tree() to:
1768 	 * 1. For root node, put back null_entry as how the table was created.
1769 	 * 2. For other nodes, expunge its radix tree node.
1770 	 */
1771 	if (!rcu_access_pointer(fn->leaf)) {
1772 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1773 			fn->fn_flags &= ~RTN_RTINFO;
1774 			net->ipv6.rt6_stats->fib_route_nodes--;
1775 		}
1776 		fn = fib6_repair_tree(net, table, fn);
1777 	}
1778 
1779 	fib6_purge_rt(rt, fn, net);
1780 
1781 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1782 	if (!info->skip_notify)
1783 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1784 	fib6_info_release(rt);
1785 }
1786 
1787 /* Need to own table->tb6_lock */
1788 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1789 {
1790 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1791 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1792 	struct fib6_table *table = rt->fib6_table;
1793 	struct net *net = info->nl_net;
1794 	struct fib6_info __rcu **rtp;
1795 	struct fib6_info __rcu **rtp_next;
1796 
1797 	if (!fn || rt == net->ipv6.fib6_null_entry)
1798 		return -ENOENT;
1799 
1800 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1801 
1802 	/*
1803 	 *	Walk the leaf entries looking for ourself
1804 	 */
1805 
1806 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1807 		struct fib6_info *cur = rcu_dereference_protected(*rtp,
1808 					lockdep_is_held(&table->tb6_lock));
1809 		if (rt == cur) {
1810 			fib6_del_route(table, fn, rtp, info);
1811 			return 0;
1812 		}
1813 		rtp_next = &cur->fib6_next;
1814 	}
1815 	return -ENOENT;
1816 }
1817 
1818 /*
1819  *	Tree traversal function.
1820  *
1821  *	Certainly, it is not interrupt safe.
1822  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1823  *	It means, that we can modify tree during walking
1824  *	and use this function for garbage collection, clone pruning,
1825  *	cleaning tree when a device goes down etc. etc.
1826  *
1827  *	It guarantees that every node will be traversed,
1828  *	and that it will be traversed only once.
1829  *
1830  *	Callback function w->func may return:
1831  *	0 -> continue walking.
1832  *	positive value -> walking is suspended (used by tree dumps,
1833  *	and probably by gc, if it will be split to several slices)
1834  *	negative value -> terminate walking.
1835  *
1836  *	The function itself returns:
1837  *	0   -> walk is complete.
1838  *	>0  -> walk is incomplete (i.e. suspended)
1839  *	<0  -> walk is terminated by an error.
1840  *
1841  *	This function is called with tb6_lock held.
1842  */
1843 
1844 static int fib6_walk_continue(struct fib6_walker *w)
1845 {
1846 	struct fib6_node *fn, *pn, *left, *right;
1847 
1848 	/* w->root should always be table->tb6_root */
1849 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1850 
1851 	for (;;) {
1852 		fn = w->node;
1853 		if (!fn)
1854 			return 0;
1855 
1856 		switch (w->state) {
1857 #ifdef CONFIG_IPV6_SUBTREES
1858 		case FWS_S:
1859 			if (FIB6_SUBTREE(fn)) {
1860 				w->node = FIB6_SUBTREE(fn);
1861 				continue;
1862 			}
1863 			w->state = FWS_L;
1864 #endif
1865 			/* fall through */
1866 		case FWS_L:
1867 			left = rcu_dereference_protected(fn->left, 1);
1868 			if (left) {
1869 				w->node = left;
1870 				w->state = FWS_INIT;
1871 				continue;
1872 			}
1873 			w->state = FWS_R;
1874 			/* fall through */
1875 		case FWS_R:
1876 			right = rcu_dereference_protected(fn->right, 1);
1877 			if (right) {
1878 				w->node = right;
1879 				w->state = FWS_INIT;
1880 				continue;
1881 			}
1882 			w->state = FWS_C;
1883 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1884 			/* fall through */
1885 		case FWS_C:
1886 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1887 				int err;
1888 
1889 				if (w->skip) {
1890 					w->skip--;
1891 					goto skip;
1892 				}
1893 
1894 				err = w->func(w);
1895 				if (err)
1896 					return err;
1897 
1898 				w->count++;
1899 				continue;
1900 			}
1901 skip:
1902 			w->state = FWS_U;
1903 			/* fall through */
1904 		case FWS_U:
1905 			if (fn == w->root)
1906 				return 0;
1907 			pn = rcu_dereference_protected(fn->parent, 1);
1908 			left = rcu_dereference_protected(pn->left, 1);
1909 			right = rcu_dereference_protected(pn->right, 1);
1910 			w->node = pn;
1911 #ifdef CONFIG_IPV6_SUBTREES
1912 			if (FIB6_SUBTREE(pn) == fn) {
1913 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1914 				w->state = FWS_L;
1915 				continue;
1916 			}
1917 #endif
1918 			if (left == fn) {
1919 				w->state = FWS_R;
1920 				continue;
1921 			}
1922 			if (right == fn) {
1923 				w->state = FWS_C;
1924 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1925 				continue;
1926 			}
1927 #if RT6_DEBUG >= 2
1928 			WARN_ON(1);
1929 #endif
1930 		}
1931 	}
1932 }
1933 
1934 static int fib6_walk(struct net *net, struct fib6_walker *w)
1935 {
1936 	int res;
1937 
1938 	w->state = FWS_INIT;
1939 	w->node = w->root;
1940 
1941 	fib6_walker_link(net, w);
1942 	res = fib6_walk_continue(w);
1943 	if (res <= 0)
1944 		fib6_walker_unlink(net, w);
1945 	return res;
1946 }
1947 
1948 static int fib6_clean_node(struct fib6_walker *w)
1949 {
1950 	int res;
1951 	struct fib6_info *rt;
1952 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1953 	struct nl_info info = {
1954 		.nl_net = c->net,
1955 	};
1956 
1957 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1958 	    w->node->fn_sernum != c->sernum)
1959 		w->node->fn_sernum = c->sernum;
1960 
1961 	if (!c->func) {
1962 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1963 		w->leaf = NULL;
1964 		return 0;
1965 	}
1966 
1967 	for_each_fib6_walker_rt(w) {
1968 		res = c->func(rt, c->arg);
1969 		if (res == -1) {
1970 			w->leaf = rt;
1971 			res = fib6_del(rt, &info);
1972 			if (res) {
1973 #if RT6_DEBUG >= 2
1974 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1975 					 __func__, rt,
1976 					 rcu_access_pointer(rt->fib6_node),
1977 					 res);
1978 #endif
1979 				continue;
1980 			}
1981 			return 0;
1982 		} else if (res == -2) {
1983 			if (WARN_ON(!rt->fib6_nsiblings))
1984 				continue;
1985 			rt = list_last_entry(&rt->fib6_siblings,
1986 					     struct fib6_info, fib6_siblings);
1987 			continue;
1988 		}
1989 		WARN_ON(res != 0);
1990 	}
1991 	w->leaf = rt;
1992 	return 0;
1993 }
1994 
1995 /*
1996  *	Convenient frontend to tree walker.
1997  *
1998  *	func is called on each route.
1999  *		It may return -2 -> skip multipath route.
2000  *			      -1 -> delete this route.
2001  *		              0  -> continue walking
2002  */
2003 
2004 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2005 			    int (*func)(struct fib6_info *, void *arg),
2006 			    int sernum, void *arg)
2007 {
2008 	struct fib6_cleaner c;
2009 
2010 	c.w.root = root;
2011 	c.w.func = fib6_clean_node;
2012 	c.w.count = 0;
2013 	c.w.skip = 0;
2014 	c.func = func;
2015 	c.sernum = sernum;
2016 	c.arg = arg;
2017 	c.net = net;
2018 
2019 	fib6_walk(net, &c.w);
2020 }
2021 
2022 static void __fib6_clean_all(struct net *net,
2023 			     int (*func)(struct fib6_info *, void *),
2024 			     int sernum, void *arg)
2025 {
2026 	struct fib6_table *table;
2027 	struct hlist_head *head;
2028 	unsigned int h;
2029 
2030 	rcu_read_lock();
2031 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2032 		head = &net->ipv6.fib_table_hash[h];
2033 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2034 			spin_lock_bh(&table->tb6_lock);
2035 			fib6_clean_tree(net, &table->tb6_root,
2036 					func, sernum, arg);
2037 			spin_unlock_bh(&table->tb6_lock);
2038 		}
2039 	}
2040 	rcu_read_unlock();
2041 }
2042 
2043 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2044 		    void *arg)
2045 {
2046 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
2047 }
2048 
2049 static void fib6_flush_trees(struct net *net)
2050 {
2051 	int new_sernum = fib6_new_sernum(net);
2052 
2053 	__fib6_clean_all(net, NULL, new_sernum, NULL);
2054 }
2055 
2056 /*
2057  *	Garbage collection
2058  */
2059 
2060 static int fib6_age(struct fib6_info *rt, void *arg)
2061 {
2062 	struct fib6_gc_args *gc_args = arg;
2063 	unsigned long now = jiffies;
2064 
2065 	/*
2066 	 *	check addrconf expiration here.
2067 	 *	Routes are expired even if they are in use.
2068 	 */
2069 
2070 	if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2071 		if (time_after(now, rt->expires)) {
2072 			RT6_TRACE("expiring %p\n", rt);
2073 			return -1;
2074 		}
2075 		gc_args->more++;
2076 	}
2077 
2078 	/*	Also age clones in the exception table.
2079 	 *	Note, that clones are aged out
2080 	 *	only if they are not in use now.
2081 	 */
2082 	rt6_age_exceptions(rt, gc_args, now);
2083 
2084 	return 0;
2085 }
2086 
2087 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2088 {
2089 	struct fib6_gc_args gc_args;
2090 	unsigned long now;
2091 
2092 	if (force) {
2093 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2094 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2095 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2096 		return;
2097 	}
2098 	gc_args.timeout = expires ? (int)expires :
2099 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2100 	gc_args.more = 0;
2101 
2102 	fib6_clean_all(net, fib6_age, &gc_args);
2103 	now = jiffies;
2104 	net->ipv6.ip6_rt_last_gc = now;
2105 
2106 	if (gc_args.more)
2107 		mod_timer(&net->ipv6.ip6_fib_timer,
2108 			  round_jiffies(now
2109 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2110 	else
2111 		del_timer(&net->ipv6.ip6_fib_timer);
2112 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2113 }
2114 
2115 static void fib6_gc_timer_cb(struct timer_list *t)
2116 {
2117 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2118 
2119 	fib6_run_gc(0, arg, true);
2120 }
2121 
2122 static int __net_init fib6_net_init(struct net *net)
2123 {
2124 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2125 	int err;
2126 
2127 	err = fib6_notifier_init(net);
2128 	if (err)
2129 		return err;
2130 
2131 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2132 	rwlock_init(&net->ipv6.fib6_walker_lock);
2133 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2134 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2135 
2136 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2137 	if (!net->ipv6.rt6_stats)
2138 		goto out_timer;
2139 
2140 	/* Avoid false sharing : Use at least a full cache line */
2141 	size = max_t(size_t, size, L1_CACHE_BYTES);
2142 
2143 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2144 	if (!net->ipv6.fib_table_hash)
2145 		goto out_rt6_stats;
2146 
2147 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2148 					  GFP_KERNEL);
2149 	if (!net->ipv6.fib6_main_tbl)
2150 		goto out_fib_table_hash;
2151 
2152 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2153 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2154 			   net->ipv6.fib6_null_entry);
2155 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2156 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2157 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2158 
2159 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2160 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2161 					   GFP_KERNEL);
2162 	if (!net->ipv6.fib6_local_tbl)
2163 		goto out_fib6_main_tbl;
2164 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2165 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2166 			   net->ipv6.fib6_null_entry);
2167 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2168 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2169 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2170 #endif
2171 	fib6_tables_init(net);
2172 
2173 	return 0;
2174 
2175 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2176 out_fib6_main_tbl:
2177 	kfree(net->ipv6.fib6_main_tbl);
2178 #endif
2179 out_fib_table_hash:
2180 	kfree(net->ipv6.fib_table_hash);
2181 out_rt6_stats:
2182 	kfree(net->ipv6.rt6_stats);
2183 out_timer:
2184 	fib6_notifier_exit(net);
2185 	return -ENOMEM;
2186 }
2187 
2188 static void fib6_net_exit(struct net *net)
2189 {
2190 	unsigned int i;
2191 
2192 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2193 
2194 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2195 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2196 		struct hlist_node *tmp;
2197 		struct fib6_table *tb;
2198 
2199 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2200 			hlist_del(&tb->tb6_hlist);
2201 			fib6_free_table(tb);
2202 		}
2203 	}
2204 
2205 	kfree(net->ipv6.fib_table_hash);
2206 	kfree(net->ipv6.rt6_stats);
2207 	fib6_notifier_exit(net);
2208 }
2209 
2210 static struct pernet_operations fib6_net_ops = {
2211 	.init = fib6_net_init,
2212 	.exit = fib6_net_exit,
2213 };
2214 
2215 int __init fib6_init(void)
2216 {
2217 	int ret = -ENOMEM;
2218 
2219 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2220 					   sizeof(struct fib6_node),
2221 					   0, SLAB_HWCACHE_ALIGN,
2222 					   NULL);
2223 	if (!fib6_node_kmem)
2224 		goto out;
2225 
2226 	ret = register_pernet_subsys(&fib6_net_ops);
2227 	if (ret)
2228 		goto out_kmem_cache_create;
2229 
2230 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2231 				   inet6_dump_fib, 0);
2232 	if (ret)
2233 		goto out_unregister_subsys;
2234 
2235 	__fib6_flush_trees = fib6_flush_trees;
2236 out:
2237 	return ret;
2238 
2239 out_unregister_subsys:
2240 	unregister_pernet_subsys(&fib6_net_ops);
2241 out_kmem_cache_create:
2242 	kmem_cache_destroy(fib6_node_kmem);
2243 	goto out;
2244 }
2245 
2246 void fib6_gc_cleanup(void)
2247 {
2248 	unregister_pernet_subsys(&fib6_net_ops);
2249 	kmem_cache_destroy(fib6_node_kmem);
2250 }
2251 
2252 #ifdef CONFIG_PROC_FS
2253 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2254 {
2255 	struct fib6_info *rt = v;
2256 	struct ipv6_route_iter *iter = seq->private;
2257 	const struct net_device *dev;
2258 
2259 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2260 
2261 #ifdef CONFIG_IPV6_SUBTREES
2262 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2263 #else
2264 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2265 #endif
2266 	if (rt->fib6_flags & RTF_GATEWAY)
2267 		seq_printf(seq, "%pi6", &rt->fib6_nh.nh_gw);
2268 	else
2269 		seq_puts(seq, "00000000000000000000000000000000");
2270 
2271 	dev = rt->fib6_nh.nh_dev;
2272 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2273 		   rt->fib6_metric, atomic_read(&rt->fib6_ref), 0,
2274 		   rt->fib6_flags, dev ? dev->name : "");
2275 	iter->w.leaf = NULL;
2276 	return 0;
2277 }
2278 
2279 static int ipv6_route_yield(struct fib6_walker *w)
2280 {
2281 	struct ipv6_route_iter *iter = w->args;
2282 
2283 	if (!iter->skip)
2284 		return 1;
2285 
2286 	do {
2287 		iter->w.leaf = rcu_dereference_protected(
2288 				iter->w.leaf->fib6_next,
2289 				lockdep_is_held(&iter->tbl->tb6_lock));
2290 		iter->skip--;
2291 		if (!iter->skip && iter->w.leaf)
2292 			return 1;
2293 	} while (iter->w.leaf);
2294 
2295 	return 0;
2296 }
2297 
2298 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2299 				      struct net *net)
2300 {
2301 	memset(&iter->w, 0, sizeof(iter->w));
2302 	iter->w.func = ipv6_route_yield;
2303 	iter->w.root = &iter->tbl->tb6_root;
2304 	iter->w.state = FWS_INIT;
2305 	iter->w.node = iter->w.root;
2306 	iter->w.args = iter;
2307 	iter->sernum = iter->w.root->fn_sernum;
2308 	INIT_LIST_HEAD(&iter->w.lh);
2309 	fib6_walker_link(net, &iter->w);
2310 }
2311 
2312 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2313 						    struct net *net)
2314 {
2315 	unsigned int h;
2316 	struct hlist_node *node;
2317 
2318 	if (tbl) {
2319 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2320 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2321 	} else {
2322 		h = 0;
2323 		node = NULL;
2324 	}
2325 
2326 	while (!node && h < FIB6_TABLE_HASHSZ) {
2327 		node = rcu_dereference_bh(
2328 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2329 	}
2330 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2331 }
2332 
2333 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2334 {
2335 	if (iter->sernum != iter->w.root->fn_sernum) {
2336 		iter->sernum = iter->w.root->fn_sernum;
2337 		iter->w.state = FWS_INIT;
2338 		iter->w.node = iter->w.root;
2339 		WARN_ON(iter->w.skip);
2340 		iter->w.skip = iter->w.count;
2341 	}
2342 }
2343 
2344 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2345 {
2346 	int r;
2347 	struct fib6_info *n;
2348 	struct net *net = seq_file_net(seq);
2349 	struct ipv6_route_iter *iter = seq->private;
2350 
2351 	if (!v)
2352 		goto iter_table;
2353 
2354 	n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2355 	if (n) {
2356 		++*pos;
2357 		return n;
2358 	}
2359 
2360 iter_table:
2361 	ipv6_route_check_sernum(iter);
2362 	spin_lock_bh(&iter->tbl->tb6_lock);
2363 	r = fib6_walk_continue(&iter->w);
2364 	spin_unlock_bh(&iter->tbl->tb6_lock);
2365 	if (r > 0) {
2366 		if (v)
2367 			++*pos;
2368 		return iter->w.leaf;
2369 	} else if (r < 0) {
2370 		fib6_walker_unlink(net, &iter->w);
2371 		return NULL;
2372 	}
2373 	fib6_walker_unlink(net, &iter->w);
2374 
2375 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2376 	if (!iter->tbl)
2377 		return NULL;
2378 
2379 	ipv6_route_seq_setup_walk(iter, net);
2380 	goto iter_table;
2381 }
2382 
2383 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2384 	__acquires(RCU_BH)
2385 {
2386 	struct net *net = seq_file_net(seq);
2387 	struct ipv6_route_iter *iter = seq->private;
2388 
2389 	rcu_read_lock_bh();
2390 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2391 	iter->skip = *pos;
2392 
2393 	if (iter->tbl) {
2394 		ipv6_route_seq_setup_walk(iter, net);
2395 		return ipv6_route_seq_next(seq, NULL, pos);
2396 	} else {
2397 		return NULL;
2398 	}
2399 }
2400 
2401 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2402 {
2403 	struct fib6_walker *w = &iter->w;
2404 	return w->node && !(w->state == FWS_U && w->node == w->root);
2405 }
2406 
2407 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2408 	__releases(RCU_BH)
2409 {
2410 	struct net *net = seq_file_net(seq);
2411 	struct ipv6_route_iter *iter = seq->private;
2412 
2413 	if (ipv6_route_iter_active(iter))
2414 		fib6_walker_unlink(net, &iter->w);
2415 
2416 	rcu_read_unlock_bh();
2417 }
2418 
2419 const struct seq_operations ipv6_route_seq_ops = {
2420 	.start	= ipv6_route_seq_start,
2421 	.next	= ipv6_route_seq_next,
2422 	.stop	= ipv6_route_seq_stop,
2423 	.show	= ipv6_route_seq_show
2424 };
2425 #endif /* CONFIG_PROC_FS */
2426