xref: /linux/net/ipv6/ip6_fib.c (revision 2eb0f624b709e78ec8e2f4c3412947703db99301)
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 *rt)
109 {
110 	struct fib6_node *fn;
111 
112 	fn = rcu_dereference_protected(rt->rt6i_node,
113 			lockdep_is_held(&rt->rt6i_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->rt6i_siblings);
163 	f6i->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
164 
165 	atomic_inc(&f6i->rt6i_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->rt6i_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->rt6i_idev)
201 		in6_dev_put(f6i->rt6i_idev);
202 	if (f6i->fib6_nh.nh_dev)
203 		dev_put(f6i->fib6_nh.nh_dev);
204 
205 	m = f6i->fib6_metrics;
206 	if (m != &dst_default_metrics && refcount_dec_and_test(&m->refcnt))
207 		kfree(m);
208 
209 	kfree(f6i);
210 }
211 EXPORT_SYMBOL_GPL(fib6_info_destroy);
212 
213 static struct fib6_node *node_alloc(struct net *net)
214 {
215 	struct fib6_node *fn;
216 
217 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
218 	if (fn)
219 		net->ipv6.rt6_stats->fib_nodes++;
220 
221 	return fn;
222 }
223 
224 static void node_free_immediate(struct net *net, struct fib6_node *fn)
225 {
226 	kmem_cache_free(fib6_node_kmem, fn);
227 	net->ipv6.rt6_stats->fib_nodes--;
228 }
229 
230 static void node_free_rcu(struct rcu_head *head)
231 {
232 	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
233 
234 	kmem_cache_free(fib6_node_kmem, fn);
235 }
236 
237 static void node_free(struct net *net, struct fib6_node *fn)
238 {
239 	call_rcu(&fn->rcu, node_free_rcu);
240 	net->ipv6.rt6_stats->fib_nodes--;
241 }
242 
243 static void fib6_free_table(struct fib6_table *table)
244 {
245 	inetpeer_invalidate_tree(&table->tb6_peers);
246 	kfree(table);
247 }
248 
249 static void fib6_link_table(struct net *net, struct fib6_table *tb)
250 {
251 	unsigned int h;
252 
253 	/*
254 	 * Initialize table lock at a single place to give lockdep a key,
255 	 * tables aren't visible prior to being linked to the list.
256 	 */
257 	spin_lock_init(&tb->tb6_lock);
258 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
259 
260 	/*
261 	 * No protection necessary, this is the only list mutatation
262 	 * operation, tables never disappear once they exist.
263 	 */
264 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
265 }
266 
267 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
268 
269 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
270 {
271 	struct fib6_table *table;
272 
273 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
274 	if (table) {
275 		table->tb6_id = id;
276 		rcu_assign_pointer(table->tb6_root.leaf,
277 				   net->ipv6.fib6_null_entry);
278 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
279 		inet_peer_base_init(&table->tb6_peers);
280 	}
281 
282 	return table;
283 }
284 
285 struct fib6_table *fib6_new_table(struct net *net, u32 id)
286 {
287 	struct fib6_table *tb;
288 
289 	if (id == 0)
290 		id = RT6_TABLE_MAIN;
291 	tb = fib6_get_table(net, id);
292 	if (tb)
293 		return tb;
294 
295 	tb = fib6_alloc_table(net, id);
296 	if (tb)
297 		fib6_link_table(net, tb);
298 
299 	return tb;
300 }
301 EXPORT_SYMBOL_GPL(fib6_new_table);
302 
303 struct fib6_table *fib6_get_table(struct net *net, u32 id)
304 {
305 	struct fib6_table *tb;
306 	struct hlist_head *head;
307 	unsigned int h;
308 
309 	if (id == 0)
310 		id = RT6_TABLE_MAIN;
311 	h = id & (FIB6_TABLE_HASHSZ - 1);
312 	rcu_read_lock();
313 	head = &net->ipv6.fib_table_hash[h];
314 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
315 		if (tb->tb6_id == id) {
316 			rcu_read_unlock();
317 			return tb;
318 		}
319 	}
320 	rcu_read_unlock();
321 
322 	return NULL;
323 }
324 EXPORT_SYMBOL_GPL(fib6_get_table);
325 
326 static void __net_init fib6_tables_init(struct net *net)
327 {
328 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
329 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
330 }
331 #else
332 
333 struct fib6_table *fib6_new_table(struct net *net, u32 id)
334 {
335 	return fib6_get_table(net, id);
336 }
337 
338 struct fib6_table *fib6_get_table(struct net *net, u32 id)
339 {
340 	  return net->ipv6.fib6_main_tbl;
341 }
342 
343 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
344 				   const struct sk_buff *skb,
345 				   int flags, pol_lookup_t lookup)
346 {
347 	struct rt6_info *rt;
348 
349 	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
350 	if (rt->dst.error == -EAGAIN) {
351 		ip6_rt_put(rt);
352 		rt = net->ipv6.ip6_null_entry;
353 		dst_hold(&rt->dst);
354 	}
355 
356 	return &rt->dst;
357 }
358 
359 static void __net_init fib6_tables_init(struct net *net)
360 {
361 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
362 }
363 
364 #endif
365 
366 unsigned int fib6_tables_seq_read(struct net *net)
367 {
368 	unsigned int h, fib_seq = 0;
369 
370 	rcu_read_lock();
371 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
372 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
373 		struct fib6_table *tb;
374 
375 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
376 			fib_seq += tb->fib_seq;
377 	}
378 	rcu_read_unlock();
379 
380 	return fib_seq;
381 }
382 
383 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
384 				    enum fib_event_type event_type,
385 				    struct fib6_info *rt)
386 {
387 	struct fib6_entry_notifier_info info = {
388 		.rt = rt,
389 	};
390 
391 	return call_fib6_notifier(nb, net, event_type, &info.info);
392 }
393 
394 static int call_fib6_entry_notifiers(struct net *net,
395 				     enum fib_event_type event_type,
396 				     struct fib6_info *rt,
397 				     struct netlink_ext_ack *extack)
398 {
399 	struct fib6_entry_notifier_info info = {
400 		.info.extack = extack,
401 		.rt = rt,
402 	};
403 
404 	rt->rt6i_table->fib_seq++;
405 	return call_fib6_notifiers(net, event_type, &info.info);
406 }
407 
408 struct fib6_dump_arg {
409 	struct net *net;
410 	struct notifier_block *nb;
411 };
412 
413 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
414 {
415 	if (rt == arg->net->ipv6.fib6_null_entry)
416 		return;
417 	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
418 }
419 
420 static int fib6_node_dump(struct fib6_walker *w)
421 {
422 	struct fib6_info *rt;
423 
424 	for_each_fib6_walker_rt(w)
425 		fib6_rt_dump(rt, w->args);
426 	w->leaf = NULL;
427 	return 0;
428 }
429 
430 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
431 			    struct fib6_walker *w)
432 {
433 	w->root = &tb->tb6_root;
434 	spin_lock_bh(&tb->tb6_lock);
435 	fib6_walk(net, w);
436 	spin_unlock_bh(&tb->tb6_lock);
437 }
438 
439 /* Called with rcu_read_lock() */
440 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
441 {
442 	struct fib6_dump_arg arg;
443 	struct fib6_walker *w;
444 	unsigned int h;
445 
446 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
447 	if (!w)
448 		return -ENOMEM;
449 
450 	w->func = fib6_node_dump;
451 	arg.net = net;
452 	arg.nb = nb;
453 	w->args = &arg;
454 
455 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
456 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
457 		struct fib6_table *tb;
458 
459 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
460 			fib6_table_dump(net, tb, w);
461 	}
462 
463 	kfree(w);
464 
465 	return 0;
466 }
467 
468 static int fib6_dump_node(struct fib6_walker *w)
469 {
470 	int res;
471 	struct fib6_info *rt;
472 
473 	for_each_fib6_walker_rt(w) {
474 		res = rt6_dump_route(rt, w->args);
475 		if (res < 0) {
476 			/* Frame is full, suspend walking */
477 			w->leaf = rt;
478 			return 1;
479 		}
480 
481 		/* Multipath routes are dumped in one route with the
482 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
483 		 * last sibling of this route (no need to dump the
484 		 * sibling routes again)
485 		 */
486 		if (rt->rt6i_nsiblings)
487 			rt = list_last_entry(&rt->rt6i_siblings,
488 					     struct fib6_info,
489 					     rt6i_siblings);
490 	}
491 	w->leaf = NULL;
492 	return 0;
493 }
494 
495 static void fib6_dump_end(struct netlink_callback *cb)
496 {
497 	struct net *net = sock_net(cb->skb->sk);
498 	struct fib6_walker *w = (void *)cb->args[2];
499 
500 	if (w) {
501 		if (cb->args[4]) {
502 			cb->args[4] = 0;
503 			fib6_walker_unlink(net, w);
504 		}
505 		cb->args[2] = 0;
506 		kfree(w);
507 	}
508 	cb->done = (void *)cb->args[3];
509 	cb->args[1] = 3;
510 }
511 
512 static int fib6_dump_done(struct netlink_callback *cb)
513 {
514 	fib6_dump_end(cb);
515 	return cb->done ? cb->done(cb) : 0;
516 }
517 
518 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
519 			   struct netlink_callback *cb)
520 {
521 	struct net *net = sock_net(skb->sk);
522 	struct fib6_walker *w;
523 	int res;
524 
525 	w = (void *)cb->args[2];
526 	w->root = &table->tb6_root;
527 
528 	if (cb->args[4] == 0) {
529 		w->count = 0;
530 		w->skip = 0;
531 
532 		spin_lock_bh(&table->tb6_lock);
533 		res = fib6_walk(net, w);
534 		spin_unlock_bh(&table->tb6_lock);
535 		if (res > 0) {
536 			cb->args[4] = 1;
537 			cb->args[5] = w->root->fn_sernum;
538 		}
539 	} else {
540 		if (cb->args[5] != w->root->fn_sernum) {
541 			/* Begin at the root if the tree changed */
542 			cb->args[5] = w->root->fn_sernum;
543 			w->state = FWS_INIT;
544 			w->node = w->root;
545 			w->skip = w->count;
546 		} else
547 			w->skip = 0;
548 
549 		spin_lock_bh(&table->tb6_lock);
550 		res = fib6_walk_continue(w);
551 		spin_unlock_bh(&table->tb6_lock);
552 		if (res <= 0) {
553 			fib6_walker_unlink(net, w);
554 			cb->args[4] = 0;
555 		}
556 	}
557 
558 	return res;
559 }
560 
561 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
562 {
563 	struct net *net = sock_net(skb->sk);
564 	unsigned int h, s_h;
565 	unsigned int e = 0, s_e;
566 	struct rt6_rtnl_dump_arg arg;
567 	struct fib6_walker *w;
568 	struct fib6_table *tb;
569 	struct hlist_head *head;
570 	int res = 0;
571 
572 	s_h = cb->args[0];
573 	s_e = cb->args[1];
574 
575 	w = (void *)cb->args[2];
576 	if (!w) {
577 		/* New dump:
578 		 *
579 		 * 1. hook callback destructor.
580 		 */
581 		cb->args[3] = (long)cb->done;
582 		cb->done = fib6_dump_done;
583 
584 		/*
585 		 * 2. allocate and initialize walker.
586 		 */
587 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
588 		if (!w)
589 			return -ENOMEM;
590 		w->func = fib6_dump_node;
591 		cb->args[2] = (long)w;
592 	}
593 
594 	arg.skb = skb;
595 	arg.cb = cb;
596 	arg.net = net;
597 	w->args = &arg;
598 
599 	rcu_read_lock();
600 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
601 		e = 0;
602 		head = &net->ipv6.fib_table_hash[h];
603 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
604 			if (e < s_e)
605 				goto next;
606 			res = fib6_dump_table(tb, skb, cb);
607 			if (res != 0)
608 				goto out;
609 next:
610 			e++;
611 		}
612 	}
613 out:
614 	rcu_read_unlock();
615 	cb->args[1] = e;
616 	cb->args[0] = h;
617 
618 	res = res < 0 ? res : skb->len;
619 	if (res <= 0)
620 		fib6_dump_end(cb);
621 	return res;
622 }
623 
624 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
625 {
626 	if (!f6i)
627 		return;
628 
629 	if (f6i->fib6_metrics == &dst_default_metrics) {
630 		struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
631 
632 		if (!p)
633 			return;
634 
635 		refcount_set(&p->refcnt, 1);
636 		f6i->fib6_metrics = p;
637 	}
638 
639 	f6i->fib6_metrics->metrics[metric - 1] = val;
640 }
641 
642 /*
643  *	Routing Table
644  *
645  *	return the appropriate node for a routing tree "add" operation
646  *	by either creating and inserting or by returning an existing
647  *	node.
648  */
649 
650 static struct fib6_node *fib6_add_1(struct net *net,
651 				    struct fib6_table *table,
652 				    struct fib6_node *root,
653 				    struct in6_addr *addr, int plen,
654 				    int offset, int allow_create,
655 				    int replace_required,
656 				    struct netlink_ext_ack *extack)
657 {
658 	struct fib6_node *fn, *in, *ln;
659 	struct fib6_node *pn = NULL;
660 	struct rt6key *key;
661 	int	bit;
662 	__be32	dir = 0;
663 
664 	RT6_TRACE("fib6_add_1\n");
665 
666 	/* insert node in tree */
667 
668 	fn = root;
669 
670 	do {
671 		struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
672 					    lockdep_is_held(&table->tb6_lock));
673 		key = (struct rt6key *)((u8 *)leaf + offset);
674 
675 		/*
676 		 *	Prefix match
677 		 */
678 		if (plen < fn->fn_bit ||
679 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
680 			if (!allow_create) {
681 				if (replace_required) {
682 					NL_SET_ERR_MSG(extack,
683 						       "Can not replace route - no match found");
684 					pr_warn("Can't replace route, no match found\n");
685 					return ERR_PTR(-ENOENT);
686 				}
687 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
688 			}
689 			goto insert_above;
690 		}
691 
692 		/*
693 		 *	Exact match ?
694 		 */
695 
696 		if (plen == fn->fn_bit) {
697 			/* clean up an intermediate node */
698 			if (!(fn->fn_flags & RTN_RTINFO)) {
699 				RCU_INIT_POINTER(fn->leaf, NULL);
700 				fib6_info_release(leaf);
701 			/* remove null_entry in the root node */
702 			} else if (fn->fn_flags & RTN_TL_ROOT &&
703 				   rcu_access_pointer(fn->leaf) ==
704 				   net->ipv6.fib6_null_entry) {
705 				RCU_INIT_POINTER(fn->leaf, NULL);
706 			}
707 
708 			return fn;
709 		}
710 
711 		/*
712 		 *	We have more bits to go
713 		 */
714 
715 		/* Try to walk down on tree. */
716 		dir = addr_bit_set(addr, fn->fn_bit);
717 		pn = fn;
718 		fn = dir ?
719 		     rcu_dereference_protected(fn->right,
720 					lockdep_is_held(&table->tb6_lock)) :
721 		     rcu_dereference_protected(fn->left,
722 					lockdep_is_held(&table->tb6_lock));
723 	} while (fn);
724 
725 	if (!allow_create) {
726 		/* We should not create new node because
727 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
728 		 * I assume it is safe to require NLM_F_CREATE when
729 		 * REPLACE flag is used! Later we may want to remove the
730 		 * check for replace_required, because according
731 		 * to netlink specification, NLM_F_CREATE
732 		 * MUST be specified if new route is created.
733 		 * That would keep IPv6 consistent with IPv4
734 		 */
735 		if (replace_required) {
736 			NL_SET_ERR_MSG(extack,
737 				       "Can not replace route - no match found");
738 			pr_warn("Can't replace route, no match found\n");
739 			return ERR_PTR(-ENOENT);
740 		}
741 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
742 	}
743 	/*
744 	 *	We walked to the bottom of tree.
745 	 *	Create new leaf node without children.
746 	 */
747 
748 	ln = node_alloc(net);
749 
750 	if (!ln)
751 		return ERR_PTR(-ENOMEM);
752 	ln->fn_bit = plen;
753 	RCU_INIT_POINTER(ln->parent, pn);
754 
755 	if (dir)
756 		rcu_assign_pointer(pn->right, ln);
757 	else
758 		rcu_assign_pointer(pn->left, ln);
759 
760 	return ln;
761 
762 
763 insert_above:
764 	/*
765 	 * split since we don't have a common prefix anymore or
766 	 * we have a less significant route.
767 	 * we've to insert an intermediate node on the list
768 	 * this new node will point to the one we need to create
769 	 * and the current
770 	 */
771 
772 	pn = rcu_dereference_protected(fn->parent,
773 				       lockdep_is_held(&table->tb6_lock));
774 
775 	/* find 1st bit in difference between the 2 addrs.
776 
777 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
778 	   but if it is >= plen, the value is ignored in any case.
779 	 */
780 
781 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
782 
783 	/*
784 	 *		(intermediate)[in]
785 	 *	          /	   \
786 	 *	(new leaf node)[ln] (old node)[fn]
787 	 */
788 	if (plen > bit) {
789 		in = node_alloc(net);
790 		ln = node_alloc(net);
791 
792 		if (!in || !ln) {
793 			if (in)
794 				node_free_immediate(net, in);
795 			if (ln)
796 				node_free_immediate(net, ln);
797 			return ERR_PTR(-ENOMEM);
798 		}
799 
800 		/*
801 		 * new intermediate node.
802 		 * RTN_RTINFO will
803 		 * be off since that an address that chooses one of
804 		 * the branches would not match less specific routes
805 		 * in the other branch
806 		 */
807 
808 		in->fn_bit = bit;
809 
810 		RCU_INIT_POINTER(in->parent, pn);
811 		in->leaf = fn->leaf;
812 		atomic_inc(&rcu_dereference_protected(in->leaf,
813 				lockdep_is_held(&table->tb6_lock))->rt6i_ref);
814 
815 		/* update parent pointer */
816 		if (dir)
817 			rcu_assign_pointer(pn->right, in);
818 		else
819 			rcu_assign_pointer(pn->left, in);
820 
821 		ln->fn_bit = plen;
822 
823 		RCU_INIT_POINTER(ln->parent, in);
824 		rcu_assign_pointer(fn->parent, in);
825 
826 		if (addr_bit_set(addr, bit)) {
827 			rcu_assign_pointer(in->right, ln);
828 			rcu_assign_pointer(in->left, fn);
829 		} else {
830 			rcu_assign_pointer(in->left, ln);
831 			rcu_assign_pointer(in->right, fn);
832 		}
833 	} else { /* plen <= bit */
834 
835 		/*
836 		 *		(new leaf node)[ln]
837 		 *	          /	   \
838 		 *	     (old node)[fn] NULL
839 		 */
840 
841 		ln = node_alloc(net);
842 
843 		if (!ln)
844 			return ERR_PTR(-ENOMEM);
845 
846 		ln->fn_bit = plen;
847 
848 		RCU_INIT_POINTER(ln->parent, pn);
849 
850 		if (addr_bit_set(&key->addr, plen))
851 			RCU_INIT_POINTER(ln->right, fn);
852 		else
853 			RCU_INIT_POINTER(ln->left, fn);
854 
855 		rcu_assign_pointer(fn->parent, ln);
856 
857 		if (dir)
858 			rcu_assign_pointer(pn->right, ln);
859 		else
860 			rcu_assign_pointer(pn->left, ln);
861 	}
862 	return ln;
863 }
864 
865 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
866 			  struct net *net)
867 {
868 	struct fib6_table *table = rt->rt6i_table;
869 
870 	if (atomic_read(&rt->rt6i_ref) != 1) {
871 		/* This route is used as dummy address holder in some split
872 		 * nodes. It is not leaked, but it still holds other resources,
873 		 * which must be released in time. So, scan ascendant nodes
874 		 * and replace dummy references to this route with references
875 		 * to still alive ones.
876 		 */
877 		while (fn) {
878 			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
879 					    lockdep_is_held(&table->tb6_lock));
880 			struct fib6_info *new_leaf;
881 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
882 				new_leaf = fib6_find_prefix(net, table, fn);
883 				atomic_inc(&new_leaf->rt6i_ref);
884 
885 				rcu_assign_pointer(fn->leaf, new_leaf);
886 				fib6_info_release(rt);
887 			}
888 			fn = rcu_dereference_protected(fn->parent,
889 				    lockdep_is_held(&table->tb6_lock));
890 		}
891 
892 		if (rt->rt6i_pcpu) {
893 			int cpu;
894 
895 			/* release the reference to this fib entry from
896 			 * all of its cached pcpu routes
897 			 */
898 			for_each_possible_cpu(cpu) {
899 				struct rt6_info **ppcpu_rt;
900 				struct rt6_info *pcpu_rt;
901 
902 				ppcpu_rt = per_cpu_ptr(rt->rt6i_pcpu, cpu);
903 				pcpu_rt = *ppcpu_rt;
904 				if (pcpu_rt) {
905 					fib6_info_release(pcpu_rt->from);
906 					pcpu_rt->from = NULL;
907 				}
908 			}
909 		}
910 	}
911 }
912 
913 /*
914  *	Insert routing information in a node.
915  */
916 
917 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
918 			    struct nl_info *info,
919 			    struct netlink_ext_ack *extack)
920 {
921 	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
922 				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
923 	struct fib6_info *iter = NULL;
924 	struct fib6_info __rcu **ins;
925 	struct fib6_info __rcu **fallback_ins = NULL;
926 	int replace = (info->nlh &&
927 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
928 	int add = (!info->nlh ||
929 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
930 	int found = 0;
931 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
932 	u16 nlflags = NLM_F_EXCL;
933 	int err;
934 
935 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
936 		nlflags |= NLM_F_APPEND;
937 
938 	ins = &fn->leaf;
939 
940 	for (iter = leaf; iter;
941 	     iter = rcu_dereference_protected(iter->rt6_next,
942 				lockdep_is_held(&rt->rt6i_table->tb6_lock))) {
943 		/*
944 		 *	Search for duplicates
945 		 */
946 
947 		if (iter->rt6i_metric == rt->rt6i_metric) {
948 			/*
949 			 *	Same priority level
950 			 */
951 			if (info->nlh &&
952 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
953 				return -EEXIST;
954 
955 			nlflags &= ~NLM_F_EXCL;
956 			if (replace) {
957 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
958 					found++;
959 					break;
960 				}
961 				if (rt_can_ecmp)
962 					fallback_ins = fallback_ins ?: ins;
963 				goto next_iter;
964 			}
965 
966 			if (rt6_duplicate_nexthop(iter, rt)) {
967 				if (rt->rt6i_nsiblings)
968 					rt->rt6i_nsiblings = 0;
969 				if (!(iter->rt6i_flags & RTF_EXPIRES))
970 					return -EEXIST;
971 				if (!(rt->rt6i_flags & RTF_EXPIRES))
972 					fib6_clean_expires(iter);
973 				else
974 					fib6_set_expires(iter, rt->expires);
975 				fib6_metric_set(iter, RTAX_MTU, rt->fib6_pmtu);
976 				return -EEXIST;
977 			}
978 			/* If we have the same destination and the same metric,
979 			 * but not the same gateway, then the route we try to
980 			 * add is sibling to this route, increment our counter
981 			 * of siblings, and later we will add our route to the
982 			 * list.
983 			 * Only static routes (which don't have flag
984 			 * RTF_EXPIRES) are used for ECMPv6.
985 			 *
986 			 * To avoid long list, we only had siblings if the
987 			 * route have a gateway.
988 			 */
989 			if (rt_can_ecmp &&
990 			    rt6_qualify_for_ecmp(iter))
991 				rt->rt6i_nsiblings++;
992 		}
993 
994 		if (iter->rt6i_metric > rt->rt6i_metric)
995 			break;
996 
997 next_iter:
998 		ins = &iter->rt6_next;
999 	}
1000 
1001 	if (fallback_ins && !found) {
1002 		/* No ECMP-able route found, replace first non-ECMP one */
1003 		ins = fallback_ins;
1004 		iter = rcu_dereference_protected(*ins,
1005 				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
1006 		found++;
1007 	}
1008 
1009 	/* Reset round-robin state, if necessary */
1010 	if (ins == &fn->leaf)
1011 		fn->rr_ptr = NULL;
1012 
1013 	/* Link this route to others same route. */
1014 	if (rt->rt6i_nsiblings) {
1015 		unsigned int rt6i_nsiblings;
1016 		struct fib6_info *sibling, *temp_sibling;
1017 
1018 		/* Find the first route that have the same metric */
1019 		sibling = leaf;
1020 		while (sibling) {
1021 			if (sibling->rt6i_metric == rt->rt6i_metric &&
1022 			    rt6_qualify_for_ecmp(sibling)) {
1023 				list_add_tail(&rt->rt6i_siblings,
1024 					      &sibling->rt6i_siblings);
1025 				break;
1026 			}
1027 			sibling = rcu_dereference_protected(sibling->rt6_next,
1028 				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
1029 		}
1030 		/* For each sibling in the list, increment the counter of
1031 		 * siblings. BUG() if counters does not match, list of siblings
1032 		 * is broken!
1033 		 */
1034 		rt6i_nsiblings = 0;
1035 		list_for_each_entry_safe(sibling, temp_sibling,
1036 					 &rt->rt6i_siblings, rt6i_siblings) {
1037 			sibling->rt6i_nsiblings++;
1038 			BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
1039 			rt6i_nsiblings++;
1040 		}
1041 		BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
1042 		rt6_multipath_rebalance(temp_sibling);
1043 	}
1044 
1045 	/*
1046 	 *	insert node
1047 	 */
1048 	if (!replace) {
1049 		if (!add)
1050 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1051 
1052 add:
1053 		nlflags |= NLM_F_CREATE;
1054 
1055 		err = call_fib6_entry_notifiers(info->nl_net,
1056 						FIB_EVENT_ENTRY_ADD,
1057 						rt, extack);
1058 		if (err)
1059 			return err;
1060 
1061 		rcu_assign_pointer(rt->rt6_next, iter);
1062 		atomic_inc(&rt->rt6i_ref);
1063 		rcu_assign_pointer(rt->rt6i_node, fn);
1064 		rcu_assign_pointer(*ins, rt);
1065 		if (!info->skip_notify)
1066 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1067 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1068 
1069 		if (!(fn->fn_flags & RTN_RTINFO)) {
1070 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1071 			fn->fn_flags |= RTN_RTINFO;
1072 		}
1073 
1074 	} else {
1075 		int nsiblings;
1076 
1077 		if (!found) {
1078 			if (add)
1079 				goto add;
1080 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1081 			return -ENOENT;
1082 		}
1083 
1084 		err = call_fib6_entry_notifiers(info->nl_net,
1085 						FIB_EVENT_ENTRY_REPLACE,
1086 						rt, extack);
1087 		if (err)
1088 			return err;
1089 
1090 		atomic_inc(&rt->rt6i_ref);
1091 		rcu_assign_pointer(rt->rt6i_node, fn);
1092 		rt->rt6_next = iter->rt6_next;
1093 		rcu_assign_pointer(*ins, rt);
1094 		if (!info->skip_notify)
1095 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1096 		if (!(fn->fn_flags & RTN_RTINFO)) {
1097 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1098 			fn->fn_flags |= RTN_RTINFO;
1099 		}
1100 		nsiblings = iter->rt6i_nsiblings;
1101 		iter->rt6i_node = NULL;
1102 		fib6_purge_rt(iter, fn, info->nl_net);
1103 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1104 			fn->rr_ptr = NULL;
1105 		fib6_info_release(iter);
1106 
1107 		if (nsiblings) {
1108 			/* Replacing an ECMP route, remove all siblings */
1109 			ins = &rt->rt6_next;
1110 			iter = rcu_dereference_protected(*ins,
1111 				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
1112 			while (iter) {
1113 				if (iter->rt6i_metric > rt->rt6i_metric)
1114 					break;
1115 				if (rt6_qualify_for_ecmp(iter)) {
1116 					*ins = iter->rt6_next;
1117 					iter->rt6i_node = NULL;
1118 					fib6_purge_rt(iter, fn, info->nl_net);
1119 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1120 						fn->rr_ptr = NULL;
1121 					fib6_info_release(iter);
1122 					nsiblings--;
1123 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1124 				} else {
1125 					ins = &iter->rt6_next;
1126 				}
1127 				iter = rcu_dereference_protected(*ins,
1128 					lockdep_is_held(&rt->rt6i_table->tb6_lock));
1129 			}
1130 			WARN_ON(nsiblings != 0);
1131 		}
1132 	}
1133 
1134 	return 0;
1135 }
1136 
1137 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1138 {
1139 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1140 	    (rt->rt6i_flags & RTF_EXPIRES))
1141 		mod_timer(&net->ipv6.ip6_fib_timer,
1142 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1143 }
1144 
1145 void fib6_force_start_gc(struct net *net)
1146 {
1147 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1148 		mod_timer(&net->ipv6.ip6_fib_timer,
1149 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1150 }
1151 
1152 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1153 					   int sernum)
1154 {
1155 	struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1156 				lockdep_is_held(&rt->rt6i_table->tb6_lock));
1157 
1158 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1159 	smp_wmb();
1160 	while (fn) {
1161 		fn->fn_sernum = sernum;
1162 		fn = rcu_dereference_protected(fn->parent,
1163 				lockdep_is_held(&rt->rt6i_table->tb6_lock));
1164 	}
1165 }
1166 
1167 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1168 {
1169 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1170 }
1171 
1172 /*
1173  *	Add routing information to the routing tree.
1174  *	<destination addr>/<source addr>
1175  *	with source addr info in sub-trees
1176  *	Need to own table->tb6_lock
1177  */
1178 
1179 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1180 	     struct nl_info *info, struct netlink_ext_ack *extack)
1181 {
1182 	struct fib6_table *table = rt->rt6i_table;
1183 	struct fib6_node *fn, *pn = NULL;
1184 	int err = -ENOMEM;
1185 	int allow_create = 1;
1186 	int replace_required = 0;
1187 	int sernum = fib6_new_sernum(info->nl_net);
1188 
1189 	if (info->nlh) {
1190 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1191 			allow_create = 0;
1192 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1193 			replace_required = 1;
1194 	}
1195 	if (!allow_create && !replace_required)
1196 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1197 
1198 	fn = fib6_add_1(info->nl_net, table, root,
1199 			&rt->rt6i_dst.addr, rt->rt6i_dst.plen,
1200 			offsetof(struct fib6_info, rt6i_dst), allow_create,
1201 			replace_required, extack);
1202 	if (IS_ERR(fn)) {
1203 		err = PTR_ERR(fn);
1204 		fn = NULL;
1205 		goto out;
1206 	}
1207 
1208 	pn = fn;
1209 
1210 #ifdef CONFIG_IPV6_SUBTREES
1211 	if (rt->rt6i_src.plen) {
1212 		struct fib6_node *sn;
1213 
1214 		if (!rcu_access_pointer(fn->subtree)) {
1215 			struct fib6_node *sfn;
1216 
1217 			/*
1218 			 * Create subtree.
1219 			 *
1220 			 *		fn[main tree]
1221 			 *		|
1222 			 *		sfn[subtree root]
1223 			 *		   \
1224 			 *		    sn[new leaf node]
1225 			 */
1226 
1227 			/* Create subtree root node */
1228 			sfn = node_alloc(info->nl_net);
1229 			if (!sfn)
1230 				goto failure;
1231 
1232 			atomic_inc(&info->nl_net->ipv6.fib6_null_entry->rt6i_ref);
1233 			rcu_assign_pointer(sfn->leaf,
1234 					   info->nl_net->ipv6.fib6_null_entry);
1235 			sfn->fn_flags = RTN_ROOT;
1236 
1237 			/* Now add the first leaf node to new subtree */
1238 
1239 			sn = fib6_add_1(info->nl_net, table, sfn,
1240 					&rt->rt6i_src.addr, rt->rt6i_src.plen,
1241 					offsetof(struct fib6_info, rt6i_src),
1242 					allow_create, replace_required, extack);
1243 
1244 			if (IS_ERR(sn)) {
1245 				/* If it is failed, discard just allocated
1246 				   root, and then (in failure) stale node
1247 				   in main tree.
1248 				 */
1249 				node_free_immediate(info->nl_net, sfn);
1250 				err = PTR_ERR(sn);
1251 				goto failure;
1252 			}
1253 
1254 			/* Now link new subtree to main tree */
1255 			rcu_assign_pointer(sfn->parent, fn);
1256 			rcu_assign_pointer(fn->subtree, sfn);
1257 		} else {
1258 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1259 					&rt->rt6i_src.addr, rt->rt6i_src.plen,
1260 					offsetof(struct fib6_info, rt6i_src),
1261 					allow_create, replace_required, extack);
1262 
1263 			if (IS_ERR(sn)) {
1264 				err = PTR_ERR(sn);
1265 				goto failure;
1266 			}
1267 		}
1268 
1269 		if (!rcu_access_pointer(fn->leaf)) {
1270 			if (fn->fn_flags & RTN_TL_ROOT) {
1271 				/* put back null_entry for root node */
1272 				rcu_assign_pointer(fn->leaf,
1273 					    info->nl_net->ipv6.fib6_null_entry);
1274 			} else {
1275 				atomic_inc(&rt->rt6i_ref);
1276 				rcu_assign_pointer(fn->leaf, rt);
1277 			}
1278 		}
1279 		fn = sn;
1280 	}
1281 #endif
1282 
1283 	err = fib6_add_rt2node(fn, rt, info, extack);
1284 	if (!err) {
1285 		__fib6_update_sernum_upto_root(rt, sernum);
1286 		fib6_start_gc(info->nl_net, rt);
1287 	}
1288 
1289 out:
1290 	if (err) {
1291 #ifdef CONFIG_IPV6_SUBTREES
1292 		/*
1293 		 * If fib6_add_1 has cleared the old leaf pointer in the
1294 		 * super-tree leaf node we have to find a new one for it.
1295 		 */
1296 		if (pn != fn) {
1297 			struct fib6_info *pn_leaf =
1298 				rcu_dereference_protected(pn->leaf,
1299 				    lockdep_is_held(&table->tb6_lock));
1300 			if (pn_leaf == rt) {
1301 				pn_leaf = NULL;
1302 				RCU_INIT_POINTER(pn->leaf, NULL);
1303 				fib6_info_release(rt);
1304 			}
1305 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1306 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1307 							   pn);
1308 #if RT6_DEBUG >= 2
1309 				if (!pn_leaf) {
1310 					WARN_ON(!pn_leaf);
1311 					pn_leaf =
1312 					    info->nl_net->ipv6.fib6_null_entry;
1313 				}
1314 #endif
1315 				fib6_info_hold(pn_leaf);
1316 				rcu_assign_pointer(pn->leaf, pn_leaf);
1317 			}
1318 		}
1319 #endif
1320 		goto failure;
1321 	}
1322 	return err;
1323 
1324 failure:
1325 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1326 	 * 1. fn is an intermediate node and we failed to add the new
1327 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1328 	 * failure case.
1329 	 * 2. fn is the root node in the table and we fail to add the first
1330 	 * default route to it.
1331 	 */
1332 	if (fn &&
1333 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1334 	     (fn->fn_flags & RTN_TL_ROOT &&
1335 	      !rcu_access_pointer(fn->leaf))))
1336 		fib6_repair_tree(info->nl_net, table, fn);
1337 	return err;
1338 }
1339 
1340 /*
1341  *	Routing tree lookup
1342  *
1343  */
1344 
1345 struct lookup_args {
1346 	int			offset;		/* key offset on fib6_info */
1347 	const struct in6_addr	*addr;		/* search key			*/
1348 };
1349 
1350 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1351 				       struct lookup_args *args)
1352 {
1353 	struct fib6_node *fn;
1354 	__be32 dir;
1355 
1356 	if (unlikely(args->offset == 0))
1357 		return NULL;
1358 
1359 	/*
1360 	 *	Descend on a tree
1361 	 */
1362 
1363 	fn = root;
1364 
1365 	for (;;) {
1366 		struct fib6_node *next;
1367 
1368 		dir = addr_bit_set(args->addr, fn->fn_bit);
1369 
1370 		next = dir ? rcu_dereference(fn->right) :
1371 			     rcu_dereference(fn->left);
1372 
1373 		if (next) {
1374 			fn = next;
1375 			continue;
1376 		}
1377 		break;
1378 	}
1379 
1380 	while (fn) {
1381 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1382 
1383 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1384 			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1385 			struct rt6key *key;
1386 
1387 			if (!leaf)
1388 				goto backtrack;
1389 
1390 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1391 
1392 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1393 #ifdef CONFIG_IPV6_SUBTREES
1394 				if (subtree) {
1395 					struct fib6_node *sfn;
1396 					sfn = fib6_lookup_1(subtree, args + 1);
1397 					if (!sfn)
1398 						goto backtrack;
1399 					fn = sfn;
1400 				}
1401 #endif
1402 				if (fn->fn_flags & RTN_RTINFO)
1403 					return fn;
1404 			}
1405 		}
1406 backtrack:
1407 		if (fn->fn_flags & RTN_ROOT)
1408 			break;
1409 
1410 		fn = rcu_dereference(fn->parent);
1411 	}
1412 
1413 	return NULL;
1414 }
1415 
1416 /* called with rcu_read_lock() held
1417  */
1418 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1419 			      const struct in6_addr *saddr)
1420 {
1421 	struct fib6_node *fn;
1422 	struct lookup_args args[] = {
1423 		{
1424 			.offset = offsetof(struct fib6_info, rt6i_dst),
1425 			.addr = daddr,
1426 		},
1427 #ifdef CONFIG_IPV6_SUBTREES
1428 		{
1429 			.offset = offsetof(struct fib6_info, rt6i_src),
1430 			.addr = saddr,
1431 		},
1432 #endif
1433 		{
1434 			.offset = 0,	/* sentinel */
1435 		}
1436 	};
1437 
1438 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
1439 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1440 		fn = root;
1441 
1442 	return fn;
1443 }
1444 
1445 /*
1446  *	Get node with specified destination prefix (and source prefix,
1447  *	if subtrees are used)
1448  *	exact_match == true means we try to find fn with exact match of
1449  *	the passed in prefix addr
1450  *	exact_match == false means we try to find fn with longest prefix
1451  *	match of the passed in prefix addr. This is useful for finding fn
1452  *	for cached route as it will be stored in the exception table under
1453  *	the node with longest prefix length.
1454  */
1455 
1456 
1457 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1458 				       const struct in6_addr *addr,
1459 				       int plen, int offset,
1460 				       bool exact_match)
1461 {
1462 	struct fib6_node *fn, *prev = NULL;
1463 
1464 	for (fn = root; fn ; ) {
1465 		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1466 		struct rt6key *key;
1467 
1468 		/* This node is being deleted */
1469 		if (!leaf) {
1470 			if (plen <= fn->fn_bit)
1471 				goto out;
1472 			else
1473 				goto next;
1474 		}
1475 
1476 		key = (struct rt6key *)((u8 *)leaf + offset);
1477 
1478 		/*
1479 		 *	Prefix match
1480 		 */
1481 		if (plen < fn->fn_bit ||
1482 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1483 			goto out;
1484 
1485 		if (plen == fn->fn_bit)
1486 			return fn;
1487 
1488 		prev = fn;
1489 
1490 next:
1491 		/*
1492 		 *	We have more bits to go
1493 		 */
1494 		if (addr_bit_set(addr, fn->fn_bit))
1495 			fn = rcu_dereference(fn->right);
1496 		else
1497 			fn = rcu_dereference(fn->left);
1498 	}
1499 out:
1500 	if (exact_match)
1501 		return NULL;
1502 	else
1503 		return prev;
1504 }
1505 
1506 struct fib6_node *fib6_locate(struct fib6_node *root,
1507 			      const struct in6_addr *daddr, int dst_len,
1508 			      const struct in6_addr *saddr, int src_len,
1509 			      bool exact_match)
1510 {
1511 	struct fib6_node *fn;
1512 
1513 	fn = fib6_locate_1(root, daddr, dst_len,
1514 			   offsetof(struct fib6_info, rt6i_dst),
1515 			   exact_match);
1516 
1517 #ifdef CONFIG_IPV6_SUBTREES
1518 	if (src_len) {
1519 		WARN_ON(saddr == NULL);
1520 		if (fn) {
1521 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1522 
1523 			if (subtree) {
1524 				fn = fib6_locate_1(subtree, saddr, src_len,
1525 					   offsetof(struct fib6_info, rt6i_src),
1526 					   exact_match);
1527 			}
1528 		}
1529 	}
1530 #endif
1531 
1532 	if (fn && fn->fn_flags & RTN_RTINFO)
1533 		return fn;
1534 
1535 	return NULL;
1536 }
1537 
1538 
1539 /*
1540  *	Deletion
1541  *
1542  */
1543 
1544 static struct fib6_info *fib6_find_prefix(struct net *net,
1545 					 struct fib6_table *table,
1546 					 struct fib6_node *fn)
1547 {
1548 	struct fib6_node *child_left, *child_right;
1549 
1550 	if (fn->fn_flags & RTN_ROOT)
1551 		return net->ipv6.fib6_null_entry;
1552 
1553 	while (fn) {
1554 		child_left = rcu_dereference_protected(fn->left,
1555 				    lockdep_is_held(&table->tb6_lock));
1556 		child_right = rcu_dereference_protected(fn->right,
1557 				    lockdep_is_held(&table->tb6_lock));
1558 		if (child_left)
1559 			return rcu_dereference_protected(child_left->leaf,
1560 					lockdep_is_held(&table->tb6_lock));
1561 		if (child_right)
1562 			return rcu_dereference_protected(child_right->leaf,
1563 					lockdep_is_held(&table->tb6_lock));
1564 
1565 		fn = FIB6_SUBTREE(fn);
1566 	}
1567 	return NULL;
1568 }
1569 
1570 /*
1571  *	Called to trim the tree of intermediate nodes when possible. "fn"
1572  *	is the node we want to try and remove.
1573  *	Need to own table->tb6_lock
1574  */
1575 
1576 static struct fib6_node *fib6_repair_tree(struct net *net,
1577 					  struct fib6_table *table,
1578 					  struct fib6_node *fn)
1579 {
1580 	int children;
1581 	int nstate;
1582 	struct fib6_node *child;
1583 	struct fib6_walker *w;
1584 	int iter = 0;
1585 
1586 	/* Set fn->leaf to null_entry for root node. */
1587 	if (fn->fn_flags & RTN_TL_ROOT) {
1588 		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1589 		return fn;
1590 	}
1591 
1592 	for (;;) {
1593 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1594 					    lockdep_is_held(&table->tb6_lock));
1595 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1596 					    lockdep_is_held(&table->tb6_lock));
1597 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1598 					    lockdep_is_held(&table->tb6_lock));
1599 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1600 					    lockdep_is_held(&table->tb6_lock));
1601 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1602 					    lockdep_is_held(&table->tb6_lock));
1603 		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1604 					    lockdep_is_held(&table->tb6_lock));
1605 		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1606 					    lockdep_is_held(&table->tb6_lock));
1607 		struct fib6_info *new_fn_leaf;
1608 
1609 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1610 		iter++;
1611 
1612 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1613 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1614 		WARN_ON(fn_leaf);
1615 
1616 		children = 0;
1617 		child = NULL;
1618 		if (fn_r)
1619 			child = fn_r, children |= 1;
1620 		if (fn_l)
1621 			child = fn_l, children |= 2;
1622 
1623 		if (children == 3 || FIB6_SUBTREE(fn)
1624 #ifdef CONFIG_IPV6_SUBTREES
1625 		    /* Subtree root (i.e. fn) may have one child */
1626 		    || (children && fn->fn_flags & RTN_ROOT)
1627 #endif
1628 		    ) {
1629 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1630 #if RT6_DEBUG >= 2
1631 			if (!new_fn_leaf) {
1632 				WARN_ON(!new_fn_leaf);
1633 				new_fn_leaf = net->ipv6.fib6_null_entry;
1634 			}
1635 #endif
1636 			fib6_info_hold(new_fn_leaf);
1637 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1638 			return pn;
1639 		}
1640 
1641 #ifdef CONFIG_IPV6_SUBTREES
1642 		if (FIB6_SUBTREE(pn) == fn) {
1643 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1644 			RCU_INIT_POINTER(pn->subtree, NULL);
1645 			nstate = FWS_L;
1646 		} else {
1647 			WARN_ON(fn->fn_flags & RTN_ROOT);
1648 #endif
1649 			if (pn_r == fn)
1650 				rcu_assign_pointer(pn->right, child);
1651 			else if (pn_l == fn)
1652 				rcu_assign_pointer(pn->left, child);
1653 #if RT6_DEBUG >= 2
1654 			else
1655 				WARN_ON(1);
1656 #endif
1657 			if (child)
1658 				rcu_assign_pointer(child->parent, pn);
1659 			nstate = FWS_R;
1660 #ifdef CONFIG_IPV6_SUBTREES
1661 		}
1662 #endif
1663 
1664 		read_lock(&net->ipv6.fib6_walker_lock);
1665 		FOR_WALKERS(net, w) {
1666 			if (!child) {
1667 				if (w->node == fn) {
1668 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1669 					w->node = pn;
1670 					w->state = nstate;
1671 				}
1672 			} else {
1673 				if (w->node == fn) {
1674 					w->node = child;
1675 					if (children&2) {
1676 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1677 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1678 					} else {
1679 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1680 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1681 					}
1682 				}
1683 			}
1684 		}
1685 		read_unlock(&net->ipv6.fib6_walker_lock);
1686 
1687 		node_free(net, fn);
1688 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1689 			return pn;
1690 
1691 		RCU_INIT_POINTER(pn->leaf, NULL);
1692 		fib6_info_release(pn_leaf);
1693 		fn = pn;
1694 	}
1695 }
1696 
1697 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1698 			   struct fib6_info __rcu **rtp, struct nl_info *info)
1699 {
1700 	struct fib6_walker *w;
1701 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1702 				    lockdep_is_held(&table->tb6_lock));
1703 	struct net *net = info->nl_net;
1704 
1705 	RT6_TRACE("fib6_del_route\n");
1706 
1707 	/* Unlink it */
1708 	*rtp = rt->rt6_next;
1709 	rt->rt6i_node = NULL;
1710 	net->ipv6.rt6_stats->fib_rt_entries--;
1711 	net->ipv6.rt6_stats->fib_discarded_routes++;
1712 
1713 	/* Flush all cached dst in exception table */
1714 	rt6_flush_exceptions(rt);
1715 
1716 	/* Reset round-robin state, if necessary */
1717 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1718 		fn->rr_ptr = NULL;
1719 
1720 	/* Remove this entry from other siblings */
1721 	if (rt->rt6i_nsiblings) {
1722 		struct fib6_info *sibling, *next_sibling;
1723 
1724 		list_for_each_entry_safe(sibling, next_sibling,
1725 					 &rt->rt6i_siblings, rt6i_siblings)
1726 			sibling->rt6i_nsiblings--;
1727 		rt->rt6i_nsiblings = 0;
1728 		list_del_init(&rt->rt6i_siblings);
1729 		rt6_multipath_rebalance(next_sibling);
1730 	}
1731 
1732 	/* Adjust walkers */
1733 	read_lock(&net->ipv6.fib6_walker_lock);
1734 	FOR_WALKERS(net, w) {
1735 		if (w->state == FWS_C && w->leaf == rt) {
1736 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1737 			w->leaf = rcu_dereference_protected(rt->rt6_next,
1738 					    lockdep_is_held(&table->tb6_lock));
1739 			if (!w->leaf)
1740 				w->state = FWS_U;
1741 		}
1742 	}
1743 	read_unlock(&net->ipv6.fib6_walker_lock);
1744 
1745 	/* If it was last route, call fib6_repair_tree() to:
1746 	 * 1. For root node, put back null_entry as how the table was created.
1747 	 * 2. For other nodes, expunge its radix tree node.
1748 	 */
1749 	if (!rcu_access_pointer(fn->leaf)) {
1750 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1751 			fn->fn_flags &= ~RTN_RTINFO;
1752 			net->ipv6.rt6_stats->fib_route_nodes--;
1753 		}
1754 		fn = fib6_repair_tree(net, table, fn);
1755 	}
1756 
1757 	fib6_purge_rt(rt, fn, net);
1758 
1759 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1760 	if (!info->skip_notify)
1761 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1762 	fib6_info_release(rt);
1763 }
1764 
1765 /* Need to own table->tb6_lock */
1766 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1767 {
1768 	struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1769 				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
1770 	struct fib6_table *table = rt->rt6i_table;
1771 	struct net *net = info->nl_net;
1772 	struct fib6_info __rcu **rtp;
1773 	struct fib6_info __rcu **rtp_next;
1774 
1775 	if (!fn || rt == net->ipv6.fib6_null_entry)
1776 		return -ENOENT;
1777 
1778 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1779 
1780 	/*
1781 	 *	Walk the leaf entries looking for ourself
1782 	 */
1783 
1784 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1785 		struct fib6_info *cur = rcu_dereference_protected(*rtp,
1786 					lockdep_is_held(&table->tb6_lock));
1787 		if (rt == cur) {
1788 			fib6_del_route(table, fn, rtp, info);
1789 			return 0;
1790 		}
1791 		rtp_next = &cur->rt6_next;
1792 	}
1793 	return -ENOENT;
1794 }
1795 
1796 /*
1797  *	Tree traversal function.
1798  *
1799  *	Certainly, it is not interrupt safe.
1800  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1801  *	It means, that we can modify tree during walking
1802  *	and use this function for garbage collection, clone pruning,
1803  *	cleaning tree when a device goes down etc. etc.
1804  *
1805  *	It guarantees that every node will be traversed,
1806  *	and that it will be traversed only once.
1807  *
1808  *	Callback function w->func may return:
1809  *	0 -> continue walking.
1810  *	positive value -> walking is suspended (used by tree dumps,
1811  *	and probably by gc, if it will be split to several slices)
1812  *	negative value -> terminate walking.
1813  *
1814  *	The function itself returns:
1815  *	0   -> walk is complete.
1816  *	>0  -> walk is incomplete (i.e. suspended)
1817  *	<0  -> walk is terminated by an error.
1818  *
1819  *	This function is called with tb6_lock held.
1820  */
1821 
1822 static int fib6_walk_continue(struct fib6_walker *w)
1823 {
1824 	struct fib6_node *fn, *pn, *left, *right;
1825 
1826 	/* w->root should always be table->tb6_root */
1827 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1828 
1829 	for (;;) {
1830 		fn = w->node;
1831 		if (!fn)
1832 			return 0;
1833 
1834 		switch (w->state) {
1835 #ifdef CONFIG_IPV6_SUBTREES
1836 		case FWS_S:
1837 			if (FIB6_SUBTREE(fn)) {
1838 				w->node = FIB6_SUBTREE(fn);
1839 				continue;
1840 			}
1841 			w->state = FWS_L;
1842 #endif
1843 			/* fall through */
1844 		case FWS_L:
1845 			left = rcu_dereference_protected(fn->left, 1);
1846 			if (left) {
1847 				w->node = left;
1848 				w->state = FWS_INIT;
1849 				continue;
1850 			}
1851 			w->state = FWS_R;
1852 			/* fall through */
1853 		case FWS_R:
1854 			right = rcu_dereference_protected(fn->right, 1);
1855 			if (right) {
1856 				w->node = right;
1857 				w->state = FWS_INIT;
1858 				continue;
1859 			}
1860 			w->state = FWS_C;
1861 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1862 			/* fall through */
1863 		case FWS_C:
1864 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1865 				int err;
1866 
1867 				if (w->skip) {
1868 					w->skip--;
1869 					goto skip;
1870 				}
1871 
1872 				err = w->func(w);
1873 				if (err)
1874 					return err;
1875 
1876 				w->count++;
1877 				continue;
1878 			}
1879 skip:
1880 			w->state = FWS_U;
1881 			/* fall through */
1882 		case FWS_U:
1883 			if (fn == w->root)
1884 				return 0;
1885 			pn = rcu_dereference_protected(fn->parent, 1);
1886 			left = rcu_dereference_protected(pn->left, 1);
1887 			right = rcu_dereference_protected(pn->right, 1);
1888 			w->node = pn;
1889 #ifdef CONFIG_IPV6_SUBTREES
1890 			if (FIB6_SUBTREE(pn) == fn) {
1891 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1892 				w->state = FWS_L;
1893 				continue;
1894 			}
1895 #endif
1896 			if (left == fn) {
1897 				w->state = FWS_R;
1898 				continue;
1899 			}
1900 			if (right == fn) {
1901 				w->state = FWS_C;
1902 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1903 				continue;
1904 			}
1905 #if RT6_DEBUG >= 2
1906 			WARN_ON(1);
1907 #endif
1908 		}
1909 	}
1910 }
1911 
1912 static int fib6_walk(struct net *net, struct fib6_walker *w)
1913 {
1914 	int res;
1915 
1916 	w->state = FWS_INIT;
1917 	w->node = w->root;
1918 
1919 	fib6_walker_link(net, w);
1920 	res = fib6_walk_continue(w);
1921 	if (res <= 0)
1922 		fib6_walker_unlink(net, w);
1923 	return res;
1924 }
1925 
1926 static int fib6_clean_node(struct fib6_walker *w)
1927 {
1928 	int res;
1929 	struct fib6_info *rt;
1930 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1931 	struct nl_info info = {
1932 		.nl_net = c->net,
1933 	};
1934 
1935 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1936 	    w->node->fn_sernum != c->sernum)
1937 		w->node->fn_sernum = c->sernum;
1938 
1939 	if (!c->func) {
1940 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1941 		w->leaf = NULL;
1942 		return 0;
1943 	}
1944 
1945 	for_each_fib6_walker_rt(w) {
1946 		res = c->func(rt, c->arg);
1947 		if (res == -1) {
1948 			w->leaf = rt;
1949 			res = fib6_del(rt, &info);
1950 			if (res) {
1951 #if RT6_DEBUG >= 2
1952 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1953 					 __func__, rt,
1954 					 rcu_access_pointer(rt->rt6i_node),
1955 					 res);
1956 #endif
1957 				continue;
1958 			}
1959 			return 0;
1960 		} else if (res == -2) {
1961 			if (WARN_ON(!rt->rt6i_nsiblings))
1962 				continue;
1963 			rt = list_last_entry(&rt->rt6i_siblings,
1964 					     struct fib6_info, rt6i_siblings);
1965 			continue;
1966 		}
1967 		WARN_ON(res != 0);
1968 	}
1969 	w->leaf = rt;
1970 	return 0;
1971 }
1972 
1973 /*
1974  *	Convenient frontend to tree walker.
1975  *
1976  *	func is called on each route.
1977  *		It may return -2 -> skip multipath route.
1978  *			      -1 -> delete this route.
1979  *		              0  -> continue walking
1980  */
1981 
1982 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1983 			    int (*func)(struct fib6_info *, void *arg),
1984 			    int sernum, void *arg)
1985 {
1986 	struct fib6_cleaner c;
1987 
1988 	c.w.root = root;
1989 	c.w.func = fib6_clean_node;
1990 	c.w.count = 0;
1991 	c.w.skip = 0;
1992 	c.func = func;
1993 	c.sernum = sernum;
1994 	c.arg = arg;
1995 	c.net = net;
1996 
1997 	fib6_walk(net, &c.w);
1998 }
1999 
2000 static void __fib6_clean_all(struct net *net,
2001 			     int (*func)(struct fib6_info *, void *),
2002 			     int sernum, void *arg)
2003 {
2004 	struct fib6_table *table;
2005 	struct hlist_head *head;
2006 	unsigned int h;
2007 
2008 	rcu_read_lock();
2009 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2010 		head = &net->ipv6.fib_table_hash[h];
2011 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2012 			spin_lock_bh(&table->tb6_lock);
2013 			fib6_clean_tree(net, &table->tb6_root,
2014 					func, sernum, arg);
2015 			spin_unlock_bh(&table->tb6_lock);
2016 		}
2017 	}
2018 	rcu_read_unlock();
2019 }
2020 
2021 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2022 		    void *arg)
2023 {
2024 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
2025 }
2026 
2027 static void fib6_flush_trees(struct net *net)
2028 {
2029 	int new_sernum = fib6_new_sernum(net);
2030 
2031 	__fib6_clean_all(net, NULL, new_sernum, NULL);
2032 }
2033 
2034 /*
2035  *	Garbage collection
2036  */
2037 
2038 static int fib6_age(struct fib6_info *rt, void *arg)
2039 {
2040 	struct fib6_gc_args *gc_args = arg;
2041 	unsigned long now = jiffies;
2042 
2043 	/*
2044 	 *	check addrconf expiration here.
2045 	 *	Routes are expired even if they are in use.
2046 	 */
2047 
2048 	if (rt->rt6i_flags & RTF_EXPIRES && rt->expires) {
2049 		if (time_after(now, rt->expires)) {
2050 			RT6_TRACE("expiring %p\n", rt);
2051 			return -1;
2052 		}
2053 		gc_args->more++;
2054 	}
2055 
2056 	/*	Also age clones in the exception table.
2057 	 *	Note, that clones are aged out
2058 	 *	only if they are not in use now.
2059 	 */
2060 	rt6_age_exceptions(rt, gc_args, now);
2061 
2062 	return 0;
2063 }
2064 
2065 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2066 {
2067 	struct fib6_gc_args gc_args;
2068 	unsigned long now;
2069 
2070 	if (force) {
2071 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2072 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2073 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2074 		return;
2075 	}
2076 	gc_args.timeout = expires ? (int)expires :
2077 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2078 	gc_args.more = 0;
2079 
2080 	fib6_clean_all(net, fib6_age, &gc_args);
2081 	now = jiffies;
2082 	net->ipv6.ip6_rt_last_gc = now;
2083 
2084 	if (gc_args.more)
2085 		mod_timer(&net->ipv6.ip6_fib_timer,
2086 			  round_jiffies(now
2087 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2088 	else
2089 		del_timer(&net->ipv6.ip6_fib_timer);
2090 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2091 }
2092 
2093 static void fib6_gc_timer_cb(struct timer_list *t)
2094 {
2095 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2096 
2097 	fib6_run_gc(0, arg, true);
2098 }
2099 
2100 static int __net_init fib6_net_init(struct net *net)
2101 {
2102 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2103 	int err;
2104 
2105 	err = fib6_notifier_init(net);
2106 	if (err)
2107 		return err;
2108 
2109 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2110 	rwlock_init(&net->ipv6.fib6_walker_lock);
2111 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2112 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2113 
2114 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2115 	if (!net->ipv6.rt6_stats)
2116 		goto out_timer;
2117 
2118 	/* Avoid false sharing : Use at least a full cache line */
2119 	size = max_t(size_t, size, L1_CACHE_BYTES);
2120 
2121 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2122 	if (!net->ipv6.fib_table_hash)
2123 		goto out_rt6_stats;
2124 
2125 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2126 					  GFP_KERNEL);
2127 	if (!net->ipv6.fib6_main_tbl)
2128 		goto out_fib_table_hash;
2129 
2130 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2131 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2132 			   net->ipv6.fib6_null_entry);
2133 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2134 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2135 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2136 
2137 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2138 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2139 					   GFP_KERNEL);
2140 	if (!net->ipv6.fib6_local_tbl)
2141 		goto out_fib6_main_tbl;
2142 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2143 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2144 			   net->ipv6.fib6_null_entry);
2145 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2146 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2147 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2148 #endif
2149 	fib6_tables_init(net);
2150 
2151 	return 0;
2152 
2153 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2154 out_fib6_main_tbl:
2155 	kfree(net->ipv6.fib6_main_tbl);
2156 #endif
2157 out_fib_table_hash:
2158 	kfree(net->ipv6.fib_table_hash);
2159 out_rt6_stats:
2160 	kfree(net->ipv6.rt6_stats);
2161 out_timer:
2162 	fib6_notifier_exit(net);
2163 	return -ENOMEM;
2164 }
2165 
2166 static void fib6_net_exit(struct net *net)
2167 {
2168 	unsigned int i;
2169 
2170 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2171 
2172 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2173 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2174 		struct hlist_node *tmp;
2175 		struct fib6_table *tb;
2176 
2177 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2178 			hlist_del(&tb->tb6_hlist);
2179 			fib6_free_table(tb);
2180 		}
2181 	}
2182 
2183 	kfree(net->ipv6.fib_table_hash);
2184 	kfree(net->ipv6.rt6_stats);
2185 	fib6_notifier_exit(net);
2186 }
2187 
2188 static struct pernet_operations fib6_net_ops = {
2189 	.init = fib6_net_init,
2190 	.exit = fib6_net_exit,
2191 };
2192 
2193 int __init fib6_init(void)
2194 {
2195 	int ret = -ENOMEM;
2196 
2197 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2198 					   sizeof(struct fib6_node),
2199 					   0, SLAB_HWCACHE_ALIGN,
2200 					   NULL);
2201 	if (!fib6_node_kmem)
2202 		goto out;
2203 
2204 	ret = register_pernet_subsys(&fib6_net_ops);
2205 	if (ret)
2206 		goto out_kmem_cache_create;
2207 
2208 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2209 				   inet6_dump_fib, 0);
2210 	if (ret)
2211 		goto out_unregister_subsys;
2212 
2213 	__fib6_flush_trees = fib6_flush_trees;
2214 out:
2215 	return ret;
2216 
2217 out_unregister_subsys:
2218 	unregister_pernet_subsys(&fib6_net_ops);
2219 out_kmem_cache_create:
2220 	kmem_cache_destroy(fib6_node_kmem);
2221 	goto out;
2222 }
2223 
2224 void fib6_gc_cleanup(void)
2225 {
2226 	unregister_pernet_subsys(&fib6_net_ops);
2227 	kmem_cache_destroy(fib6_node_kmem);
2228 }
2229 
2230 #ifdef CONFIG_PROC_FS
2231 
2232 struct ipv6_route_iter {
2233 	struct seq_net_private p;
2234 	struct fib6_walker w;
2235 	loff_t skip;
2236 	struct fib6_table *tbl;
2237 	int sernum;
2238 };
2239 
2240 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2241 {
2242 	struct fib6_info *rt = v;
2243 	struct ipv6_route_iter *iter = seq->private;
2244 	const struct net_device *dev;
2245 
2246 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
2247 
2248 #ifdef CONFIG_IPV6_SUBTREES
2249 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
2250 #else
2251 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2252 #endif
2253 	if (rt->rt6i_flags & RTF_GATEWAY)
2254 		seq_printf(seq, "%pi6", &rt->fib6_nh.nh_gw);
2255 	else
2256 		seq_puts(seq, "00000000000000000000000000000000");
2257 
2258 	dev = rt->fib6_nh.nh_dev;
2259 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2260 		   rt->rt6i_metric, atomic_read(&rt->rt6i_ref), 0,
2261 		   rt->rt6i_flags, dev ? dev->name : "");
2262 	iter->w.leaf = NULL;
2263 	return 0;
2264 }
2265 
2266 static int ipv6_route_yield(struct fib6_walker *w)
2267 {
2268 	struct ipv6_route_iter *iter = w->args;
2269 
2270 	if (!iter->skip)
2271 		return 1;
2272 
2273 	do {
2274 		iter->w.leaf = rcu_dereference_protected(
2275 				iter->w.leaf->rt6_next,
2276 				lockdep_is_held(&iter->tbl->tb6_lock));
2277 		iter->skip--;
2278 		if (!iter->skip && iter->w.leaf)
2279 			return 1;
2280 	} while (iter->w.leaf);
2281 
2282 	return 0;
2283 }
2284 
2285 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2286 				      struct net *net)
2287 {
2288 	memset(&iter->w, 0, sizeof(iter->w));
2289 	iter->w.func = ipv6_route_yield;
2290 	iter->w.root = &iter->tbl->tb6_root;
2291 	iter->w.state = FWS_INIT;
2292 	iter->w.node = iter->w.root;
2293 	iter->w.args = iter;
2294 	iter->sernum = iter->w.root->fn_sernum;
2295 	INIT_LIST_HEAD(&iter->w.lh);
2296 	fib6_walker_link(net, &iter->w);
2297 }
2298 
2299 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2300 						    struct net *net)
2301 {
2302 	unsigned int h;
2303 	struct hlist_node *node;
2304 
2305 	if (tbl) {
2306 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2307 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2308 	} else {
2309 		h = 0;
2310 		node = NULL;
2311 	}
2312 
2313 	while (!node && h < FIB6_TABLE_HASHSZ) {
2314 		node = rcu_dereference_bh(
2315 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2316 	}
2317 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2318 }
2319 
2320 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2321 {
2322 	if (iter->sernum != iter->w.root->fn_sernum) {
2323 		iter->sernum = iter->w.root->fn_sernum;
2324 		iter->w.state = FWS_INIT;
2325 		iter->w.node = iter->w.root;
2326 		WARN_ON(iter->w.skip);
2327 		iter->w.skip = iter->w.count;
2328 	}
2329 }
2330 
2331 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2332 {
2333 	int r;
2334 	struct fib6_info *n;
2335 	struct net *net = seq_file_net(seq);
2336 	struct ipv6_route_iter *iter = seq->private;
2337 
2338 	if (!v)
2339 		goto iter_table;
2340 
2341 	n = rcu_dereference_bh(((struct fib6_info *)v)->rt6_next);
2342 	if (n) {
2343 		++*pos;
2344 		return n;
2345 	}
2346 
2347 iter_table:
2348 	ipv6_route_check_sernum(iter);
2349 	spin_lock_bh(&iter->tbl->tb6_lock);
2350 	r = fib6_walk_continue(&iter->w);
2351 	spin_unlock_bh(&iter->tbl->tb6_lock);
2352 	if (r > 0) {
2353 		if (v)
2354 			++*pos;
2355 		return iter->w.leaf;
2356 	} else if (r < 0) {
2357 		fib6_walker_unlink(net, &iter->w);
2358 		return NULL;
2359 	}
2360 	fib6_walker_unlink(net, &iter->w);
2361 
2362 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2363 	if (!iter->tbl)
2364 		return NULL;
2365 
2366 	ipv6_route_seq_setup_walk(iter, net);
2367 	goto iter_table;
2368 }
2369 
2370 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2371 	__acquires(RCU_BH)
2372 {
2373 	struct net *net = seq_file_net(seq);
2374 	struct ipv6_route_iter *iter = seq->private;
2375 
2376 	rcu_read_lock_bh();
2377 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2378 	iter->skip = *pos;
2379 
2380 	if (iter->tbl) {
2381 		ipv6_route_seq_setup_walk(iter, net);
2382 		return ipv6_route_seq_next(seq, NULL, pos);
2383 	} else {
2384 		return NULL;
2385 	}
2386 }
2387 
2388 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2389 {
2390 	struct fib6_walker *w = &iter->w;
2391 	return w->node && !(w->state == FWS_U && w->node == w->root);
2392 }
2393 
2394 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2395 	__releases(RCU_BH)
2396 {
2397 	struct net *net = seq_file_net(seq);
2398 	struct ipv6_route_iter *iter = seq->private;
2399 
2400 	if (ipv6_route_iter_active(iter))
2401 		fib6_walker_unlink(net, &iter->w);
2402 
2403 	rcu_read_unlock_bh();
2404 }
2405 
2406 static const struct seq_operations ipv6_route_seq_ops = {
2407 	.start	= ipv6_route_seq_start,
2408 	.next	= ipv6_route_seq_next,
2409 	.stop	= ipv6_route_seq_stop,
2410 	.show	= ipv6_route_seq_show
2411 };
2412 
2413 int ipv6_route_open(struct inode *inode, struct file *file)
2414 {
2415 	return seq_open_net(inode, file, &ipv6_route_seq_ops,
2416 			    sizeof(struct ipv6_route_iter));
2417 }
2418 
2419 #endif /* CONFIG_PROC_FS */
2420