xref: /linux/net/ipv6/ip6_fib.c (revision b04df400c30235fa347313c9e2a0695549bd2c8e)
1 /*
2  *	Linux INET6 implementation
3  *	Forwarding Information Database
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	This program is free software; you can redistribute it and/or
9  *      modify it under the terms of the GNU General Public License
10  *      as published by the Free Software Foundation; either version
11  *      2 of the License, or (at your option) any later version.
12  *
13  *	Changes:
14  *	Yuji SEKIYA @USAGI:	Support default route on router node;
15  *				remove ip6_null_entry from the top of
16  *				routing table.
17  *	Ville Nuorvala:		Fixed routing subtrees.
18  */
19 
20 #define pr_fmt(fmt) "IPv6: " fmt
21 
22 #include <linux/errno.h>
23 #include <linux/types.h>
24 #include <linux/net.h>
25 #include <linux/route.h>
26 #include <linux/netdevice.h>
27 #include <linux/in6.h>
28 #include <linux/init.h>
29 #include <linux/list.h>
30 #include <linux/slab.h>
31 
32 #include <net/ipv6.h>
33 #include <net/ndisc.h>
34 #include <net/addrconf.h>
35 #include <net/lwtunnel.h>
36 #include <net/fib_notifier.h>
37 
38 #include <net/ip6_fib.h>
39 #include <net/ip6_route.h>
40 
41 static struct kmem_cache *fib6_node_kmem __read_mostly;
42 
43 struct fib6_cleaner {
44 	struct fib6_walker w;
45 	struct net *net;
46 	int (*func)(struct fib6_info *, void *arg);
47 	int sernum;
48 	void *arg;
49 };
50 
51 #ifdef CONFIG_IPV6_SUBTREES
52 #define FWS_INIT FWS_S
53 #else
54 #define FWS_INIT FWS_L
55 #endif
56 
57 static struct fib6_info *fib6_find_prefix(struct net *net,
58 					 struct fib6_table *table,
59 					 struct fib6_node *fn);
60 static struct fib6_node *fib6_repair_tree(struct net *net,
61 					  struct fib6_table *table,
62 					  struct fib6_node *fn);
63 static int fib6_walk(struct net *net, struct fib6_walker *w);
64 static int fib6_walk_continue(struct fib6_walker *w);
65 
66 /*
67  *	A routing update causes an increase of the serial number on the
68  *	affected subtree. This allows for cached routes to be asynchronously
69  *	tested when modifications are made to the destination cache as a
70  *	result of redirects, path MTU changes, etc.
71  */
72 
73 static void fib6_gc_timer_cb(struct timer_list *t);
74 
75 #define FOR_WALKERS(net, w) \
76 	list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
77 
78 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
79 {
80 	write_lock_bh(&net->ipv6.fib6_walker_lock);
81 	list_add(&w->lh, &net->ipv6.fib6_walkers);
82 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
83 }
84 
85 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
86 {
87 	write_lock_bh(&net->ipv6.fib6_walker_lock);
88 	list_del(&w->lh);
89 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
90 }
91 
92 static int fib6_new_sernum(struct net *net)
93 {
94 	int new, old;
95 
96 	do {
97 		old = atomic_read(&net->ipv6.fib6_sernum);
98 		new = old < INT_MAX ? old + 1 : 1;
99 	} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
100 				old, new) != old);
101 	return new;
102 }
103 
104 enum {
105 	FIB6_NO_SERNUM_CHANGE = 0,
106 };
107 
108 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
109 {
110 	struct fib6_node *fn;
111 
112 	fn = rcu_dereference_protected(f6i->fib6_node,
113 			lockdep_is_held(&f6i->fib6_table->tb6_lock));
114 	if (fn)
115 		fn->fn_sernum = fib6_new_sernum(net);
116 }
117 
118 /*
119  *	Auxiliary address test functions for the radix tree.
120  *
121  *	These assume a 32bit processor (although it will work on
122  *	64bit processors)
123  */
124 
125 /*
126  *	test bit
127  */
128 #if defined(__LITTLE_ENDIAN)
129 # define BITOP_BE32_SWIZZLE	(0x1F & ~7)
130 #else
131 # define BITOP_BE32_SWIZZLE	0
132 #endif
133 
134 static __be32 addr_bit_set(const void *token, int fn_bit)
135 {
136 	const __be32 *addr = token;
137 	/*
138 	 * Here,
139 	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
140 	 * is optimized version of
141 	 *	htonl(1 << ((~fn_bit)&0x1F))
142 	 * See include/asm-generic/bitops/le.h.
143 	 */
144 	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
145 	       addr[fn_bit >> 5];
146 }
147 
148 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags)
149 {
150 	struct fib6_info *f6i;
151 
152 	f6i = kzalloc(sizeof(*f6i), gfp_flags);
153 	if (!f6i)
154 		return NULL;
155 
156 	f6i->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
157 	if (!f6i->rt6i_pcpu) {
158 		kfree(f6i);
159 		return NULL;
160 	}
161 
162 	INIT_LIST_HEAD(&f6i->fib6_siblings);
163 	f6i->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
164 
165 	atomic_inc(&f6i->fib6_ref);
166 
167 	return f6i;
168 }
169 
170 void fib6_info_destroy(struct fib6_info *f6i)
171 {
172 	struct rt6_exception_bucket *bucket;
173 	struct dst_metrics *m;
174 
175 	WARN_ON(f6i->fib6_node);
176 
177 	bucket = rcu_dereference_protected(f6i->rt6i_exception_bucket, 1);
178 	if (bucket) {
179 		f6i->rt6i_exception_bucket = NULL;
180 		kfree(bucket);
181 	}
182 
183 	if (f6i->rt6i_pcpu) {
184 		int cpu;
185 
186 		for_each_possible_cpu(cpu) {
187 			struct rt6_info **ppcpu_rt;
188 			struct rt6_info *pcpu_rt;
189 
190 			ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
191 			pcpu_rt = *ppcpu_rt;
192 			if (pcpu_rt) {
193 				dst_dev_put(&pcpu_rt->dst);
194 				dst_release(&pcpu_rt->dst);
195 				*ppcpu_rt = NULL;
196 			}
197 		}
198 	}
199 
200 	if (f6i->fib6_nh.nh_dev)
201 		dev_put(f6i->fib6_nh.nh_dev);
202 
203 	m = f6i->fib6_metrics;
204 	if (m != &dst_default_metrics && refcount_dec_and_test(&m->refcnt))
205 		kfree(m);
206 
207 	kfree(f6i);
208 }
209 EXPORT_SYMBOL_GPL(fib6_info_destroy);
210 
211 static struct fib6_node *node_alloc(struct net *net)
212 {
213 	struct fib6_node *fn;
214 
215 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
216 	if (fn)
217 		net->ipv6.rt6_stats->fib_nodes++;
218 
219 	return fn;
220 }
221 
222 static void node_free_immediate(struct net *net, struct fib6_node *fn)
223 {
224 	kmem_cache_free(fib6_node_kmem, fn);
225 	net->ipv6.rt6_stats->fib_nodes--;
226 }
227 
228 static void node_free_rcu(struct rcu_head *head)
229 {
230 	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
231 
232 	kmem_cache_free(fib6_node_kmem, fn);
233 }
234 
235 static void node_free(struct net *net, struct fib6_node *fn)
236 {
237 	call_rcu(&fn->rcu, node_free_rcu);
238 	net->ipv6.rt6_stats->fib_nodes--;
239 }
240 
241 static void fib6_free_table(struct fib6_table *table)
242 {
243 	inetpeer_invalidate_tree(&table->tb6_peers);
244 	kfree(table);
245 }
246 
247 static void fib6_link_table(struct net *net, struct fib6_table *tb)
248 {
249 	unsigned int h;
250 
251 	/*
252 	 * Initialize table lock at a single place to give lockdep a key,
253 	 * tables aren't visible prior to being linked to the list.
254 	 */
255 	spin_lock_init(&tb->tb6_lock);
256 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
257 
258 	/*
259 	 * No protection necessary, this is the only list mutatation
260 	 * operation, tables never disappear once they exist.
261 	 */
262 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
263 }
264 
265 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
266 
267 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
268 {
269 	struct fib6_table *table;
270 
271 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
272 	if (table) {
273 		table->tb6_id = id;
274 		rcu_assign_pointer(table->tb6_root.leaf,
275 				   net->ipv6.fib6_null_entry);
276 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
277 		inet_peer_base_init(&table->tb6_peers);
278 	}
279 
280 	return table;
281 }
282 
283 struct fib6_table *fib6_new_table(struct net *net, u32 id)
284 {
285 	struct fib6_table *tb;
286 
287 	if (id == 0)
288 		id = RT6_TABLE_MAIN;
289 	tb = fib6_get_table(net, id);
290 	if (tb)
291 		return tb;
292 
293 	tb = fib6_alloc_table(net, id);
294 	if (tb)
295 		fib6_link_table(net, tb);
296 
297 	return tb;
298 }
299 EXPORT_SYMBOL_GPL(fib6_new_table);
300 
301 struct fib6_table *fib6_get_table(struct net *net, u32 id)
302 {
303 	struct fib6_table *tb;
304 	struct hlist_head *head;
305 	unsigned int h;
306 
307 	if (id == 0)
308 		id = RT6_TABLE_MAIN;
309 	h = id & (FIB6_TABLE_HASHSZ - 1);
310 	rcu_read_lock();
311 	head = &net->ipv6.fib_table_hash[h];
312 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
313 		if (tb->tb6_id == id) {
314 			rcu_read_unlock();
315 			return tb;
316 		}
317 	}
318 	rcu_read_unlock();
319 
320 	return NULL;
321 }
322 EXPORT_SYMBOL_GPL(fib6_get_table);
323 
324 static void __net_init fib6_tables_init(struct net *net)
325 {
326 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
327 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
328 }
329 #else
330 
331 struct fib6_table *fib6_new_table(struct net *net, u32 id)
332 {
333 	return fib6_get_table(net, id);
334 }
335 
336 struct fib6_table *fib6_get_table(struct net *net, u32 id)
337 {
338 	  return net->ipv6.fib6_main_tbl;
339 }
340 
341 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
342 				   const struct sk_buff *skb,
343 				   int flags, pol_lookup_t lookup)
344 {
345 	struct rt6_info *rt;
346 
347 	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
348 	if (rt->dst.error == -EAGAIN) {
349 		ip6_rt_put(rt);
350 		rt = net->ipv6.ip6_null_entry;
351 		dst_hold(&rt->dst);
352 	}
353 
354 	return &rt->dst;
355 }
356 
357 /* called with rcu lock held; no reference taken on fib6_info */
358 struct fib6_info *fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
359 			      int flags)
360 {
361 	return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6, flags);
362 }
363 
364 static void __net_init fib6_tables_init(struct net *net)
365 {
366 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
367 }
368 
369 #endif
370 
371 unsigned int fib6_tables_seq_read(struct net *net)
372 {
373 	unsigned int h, fib_seq = 0;
374 
375 	rcu_read_lock();
376 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
377 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
378 		struct fib6_table *tb;
379 
380 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
381 			fib_seq += tb->fib_seq;
382 	}
383 	rcu_read_unlock();
384 
385 	return fib_seq;
386 }
387 
388 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
389 				    enum fib_event_type event_type,
390 				    struct fib6_info *rt)
391 {
392 	struct fib6_entry_notifier_info info = {
393 		.rt = rt,
394 	};
395 
396 	return call_fib6_notifier(nb, net, event_type, &info.info);
397 }
398 
399 static int call_fib6_entry_notifiers(struct net *net,
400 				     enum fib_event_type event_type,
401 				     struct fib6_info *rt,
402 				     struct netlink_ext_ack *extack)
403 {
404 	struct fib6_entry_notifier_info info = {
405 		.info.extack = extack,
406 		.rt = rt,
407 	};
408 
409 	rt->fib6_table->fib_seq++;
410 	return call_fib6_notifiers(net, event_type, &info.info);
411 }
412 
413 struct fib6_dump_arg {
414 	struct net *net;
415 	struct notifier_block *nb;
416 };
417 
418 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
419 {
420 	if (rt == arg->net->ipv6.fib6_null_entry)
421 		return;
422 	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
423 }
424 
425 static int fib6_node_dump(struct fib6_walker *w)
426 {
427 	struct fib6_info *rt;
428 
429 	for_each_fib6_walker_rt(w)
430 		fib6_rt_dump(rt, w->args);
431 	w->leaf = NULL;
432 	return 0;
433 }
434 
435 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
436 			    struct fib6_walker *w)
437 {
438 	w->root = &tb->tb6_root;
439 	spin_lock_bh(&tb->tb6_lock);
440 	fib6_walk(net, w);
441 	spin_unlock_bh(&tb->tb6_lock);
442 }
443 
444 /* Called with rcu_read_lock() */
445 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
446 {
447 	struct fib6_dump_arg arg;
448 	struct fib6_walker *w;
449 	unsigned int h;
450 
451 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
452 	if (!w)
453 		return -ENOMEM;
454 
455 	w->func = fib6_node_dump;
456 	arg.net = net;
457 	arg.nb = nb;
458 	w->args = &arg;
459 
460 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
461 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
462 		struct fib6_table *tb;
463 
464 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
465 			fib6_table_dump(net, tb, w);
466 	}
467 
468 	kfree(w);
469 
470 	return 0;
471 }
472 
473 static int fib6_dump_node(struct fib6_walker *w)
474 {
475 	int res;
476 	struct fib6_info *rt;
477 
478 	for_each_fib6_walker_rt(w) {
479 		res = rt6_dump_route(rt, w->args);
480 		if (res < 0) {
481 			/* Frame is full, suspend walking */
482 			w->leaf = rt;
483 			return 1;
484 		}
485 
486 		/* Multipath routes are dumped in one route with the
487 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
488 		 * last sibling of this route (no need to dump the
489 		 * sibling routes again)
490 		 */
491 		if (rt->fib6_nsiblings)
492 			rt = list_last_entry(&rt->fib6_siblings,
493 					     struct fib6_info,
494 					     fib6_siblings);
495 	}
496 	w->leaf = NULL;
497 	return 0;
498 }
499 
500 static void fib6_dump_end(struct netlink_callback *cb)
501 {
502 	struct net *net = sock_net(cb->skb->sk);
503 	struct fib6_walker *w = (void *)cb->args[2];
504 
505 	if (w) {
506 		if (cb->args[4]) {
507 			cb->args[4] = 0;
508 			fib6_walker_unlink(net, w);
509 		}
510 		cb->args[2] = 0;
511 		kfree(w);
512 	}
513 	cb->done = (void *)cb->args[3];
514 	cb->args[1] = 3;
515 }
516 
517 static int fib6_dump_done(struct netlink_callback *cb)
518 {
519 	fib6_dump_end(cb);
520 	return cb->done ? cb->done(cb) : 0;
521 }
522 
523 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
524 			   struct netlink_callback *cb)
525 {
526 	struct net *net = sock_net(skb->sk);
527 	struct fib6_walker *w;
528 	int res;
529 
530 	w = (void *)cb->args[2];
531 	w->root = &table->tb6_root;
532 
533 	if (cb->args[4] == 0) {
534 		w->count = 0;
535 		w->skip = 0;
536 
537 		spin_lock_bh(&table->tb6_lock);
538 		res = fib6_walk(net, w);
539 		spin_unlock_bh(&table->tb6_lock);
540 		if (res > 0) {
541 			cb->args[4] = 1;
542 			cb->args[5] = w->root->fn_sernum;
543 		}
544 	} else {
545 		if (cb->args[5] != w->root->fn_sernum) {
546 			/* Begin at the root if the tree changed */
547 			cb->args[5] = w->root->fn_sernum;
548 			w->state = FWS_INIT;
549 			w->node = w->root;
550 			w->skip = w->count;
551 		} else
552 			w->skip = 0;
553 
554 		spin_lock_bh(&table->tb6_lock);
555 		res = fib6_walk_continue(w);
556 		spin_unlock_bh(&table->tb6_lock);
557 		if (res <= 0) {
558 			fib6_walker_unlink(net, w);
559 			cb->args[4] = 0;
560 		}
561 	}
562 
563 	return res;
564 }
565 
566 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
567 {
568 	struct net *net = sock_net(skb->sk);
569 	unsigned int h, s_h;
570 	unsigned int e = 0, s_e;
571 	struct rt6_rtnl_dump_arg arg;
572 	struct fib6_walker *w;
573 	struct fib6_table *tb;
574 	struct hlist_head *head;
575 	int res = 0;
576 
577 	s_h = cb->args[0];
578 	s_e = cb->args[1];
579 
580 	w = (void *)cb->args[2];
581 	if (!w) {
582 		/* New dump:
583 		 *
584 		 * 1. hook callback destructor.
585 		 */
586 		cb->args[3] = (long)cb->done;
587 		cb->done = fib6_dump_done;
588 
589 		/*
590 		 * 2. allocate and initialize walker.
591 		 */
592 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
593 		if (!w)
594 			return -ENOMEM;
595 		w->func = fib6_dump_node;
596 		cb->args[2] = (long)w;
597 	}
598 
599 	arg.skb = skb;
600 	arg.cb = cb;
601 	arg.net = net;
602 	w->args = &arg;
603 
604 	rcu_read_lock();
605 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
606 		e = 0;
607 		head = &net->ipv6.fib_table_hash[h];
608 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
609 			if (e < s_e)
610 				goto next;
611 			res = fib6_dump_table(tb, skb, cb);
612 			if (res != 0)
613 				goto out;
614 next:
615 			e++;
616 		}
617 	}
618 out:
619 	rcu_read_unlock();
620 	cb->args[1] = e;
621 	cb->args[0] = h;
622 
623 	res = res < 0 ? res : skb->len;
624 	if (res <= 0)
625 		fib6_dump_end(cb);
626 	return res;
627 }
628 
629 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
630 {
631 	if (!f6i)
632 		return;
633 
634 	if (f6i->fib6_metrics == &dst_default_metrics) {
635 		struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
636 
637 		if (!p)
638 			return;
639 
640 		refcount_set(&p->refcnt, 1);
641 		f6i->fib6_metrics = p;
642 	}
643 
644 	f6i->fib6_metrics->metrics[metric - 1] = val;
645 }
646 
647 /*
648  *	Routing Table
649  *
650  *	return the appropriate node for a routing tree "add" operation
651  *	by either creating and inserting or by returning an existing
652  *	node.
653  */
654 
655 static struct fib6_node *fib6_add_1(struct net *net,
656 				    struct fib6_table *table,
657 				    struct fib6_node *root,
658 				    struct in6_addr *addr, int plen,
659 				    int offset, int allow_create,
660 				    int replace_required,
661 				    struct netlink_ext_ack *extack)
662 {
663 	struct fib6_node *fn, *in, *ln;
664 	struct fib6_node *pn = NULL;
665 	struct rt6key *key;
666 	int	bit;
667 	__be32	dir = 0;
668 
669 	RT6_TRACE("fib6_add_1\n");
670 
671 	/* insert node in tree */
672 
673 	fn = root;
674 
675 	do {
676 		struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
677 					    lockdep_is_held(&table->tb6_lock));
678 		key = (struct rt6key *)((u8 *)leaf + offset);
679 
680 		/*
681 		 *	Prefix match
682 		 */
683 		if (plen < fn->fn_bit ||
684 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
685 			if (!allow_create) {
686 				if (replace_required) {
687 					NL_SET_ERR_MSG(extack,
688 						       "Can not replace route - no match found");
689 					pr_warn("Can't replace route, no match found\n");
690 					return ERR_PTR(-ENOENT);
691 				}
692 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
693 			}
694 			goto insert_above;
695 		}
696 
697 		/*
698 		 *	Exact match ?
699 		 */
700 
701 		if (plen == fn->fn_bit) {
702 			/* clean up an intermediate node */
703 			if (!(fn->fn_flags & RTN_RTINFO)) {
704 				RCU_INIT_POINTER(fn->leaf, NULL);
705 				fib6_info_release(leaf);
706 			/* remove null_entry in the root node */
707 			} else if (fn->fn_flags & RTN_TL_ROOT &&
708 				   rcu_access_pointer(fn->leaf) ==
709 				   net->ipv6.fib6_null_entry) {
710 				RCU_INIT_POINTER(fn->leaf, NULL);
711 			}
712 
713 			return fn;
714 		}
715 
716 		/*
717 		 *	We have more bits to go
718 		 */
719 
720 		/* Try to walk down on tree. */
721 		dir = addr_bit_set(addr, fn->fn_bit);
722 		pn = fn;
723 		fn = dir ?
724 		     rcu_dereference_protected(fn->right,
725 					lockdep_is_held(&table->tb6_lock)) :
726 		     rcu_dereference_protected(fn->left,
727 					lockdep_is_held(&table->tb6_lock));
728 	} while (fn);
729 
730 	if (!allow_create) {
731 		/* We should not create new node because
732 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
733 		 * I assume it is safe to require NLM_F_CREATE when
734 		 * REPLACE flag is used! Later we may want to remove the
735 		 * check for replace_required, because according
736 		 * to netlink specification, NLM_F_CREATE
737 		 * MUST be specified if new route is created.
738 		 * That would keep IPv6 consistent with IPv4
739 		 */
740 		if (replace_required) {
741 			NL_SET_ERR_MSG(extack,
742 				       "Can not replace route - no match found");
743 			pr_warn("Can't replace route, no match found\n");
744 			return ERR_PTR(-ENOENT);
745 		}
746 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
747 	}
748 	/*
749 	 *	We walked to the bottom of tree.
750 	 *	Create new leaf node without children.
751 	 */
752 
753 	ln = node_alloc(net);
754 
755 	if (!ln)
756 		return ERR_PTR(-ENOMEM);
757 	ln->fn_bit = plen;
758 	RCU_INIT_POINTER(ln->parent, pn);
759 
760 	if (dir)
761 		rcu_assign_pointer(pn->right, ln);
762 	else
763 		rcu_assign_pointer(pn->left, ln);
764 
765 	return ln;
766 
767 
768 insert_above:
769 	/*
770 	 * split since we don't have a common prefix anymore or
771 	 * we have a less significant route.
772 	 * we've to insert an intermediate node on the list
773 	 * this new node will point to the one we need to create
774 	 * and the current
775 	 */
776 
777 	pn = rcu_dereference_protected(fn->parent,
778 				       lockdep_is_held(&table->tb6_lock));
779 
780 	/* find 1st bit in difference between the 2 addrs.
781 
782 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
783 	   but if it is >= plen, the value is ignored in any case.
784 	 */
785 
786 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
787 
788 	/*
789 	 *		(intermediate)[in]
790 	 *	          /	   \
791 	 *	(new leaf node)[ln] (old node)[fn]
792 	 */
793 	if (plen > bit) {
794 		in = node_alloc(net);
795 		ln = node_alloc(net);
796 
797 		if (!in || !ln) {
798 			if (in)
799 				node_free_immediate(net, in);
800 			if (ln)
801 				node_free_immediate(net, ln);
802 			return ERR_PTR(-ENOMEM);
803 		}
804 
805 		/*
806 		 * new intermediate node.
807 		 * RTN_RTINFO will
808 		 * be off since that an address that chooses one of
809 		 * the branches would not match less specific routes
810 		 * in the other branch
811 		 */
812 
813 		in->fn_bit = bit;
814 
815 		RCU_INIT_POINTER(in->parent, pn);
816 		in->leaf = fn->leaf;
817 		atomic_inc(&rcu_dereference_protected(in->leaf,
818 				lockdep_is_held(&table->tb6_lock))->fib6_ref);
819 
820 		/* update parent pointer */
821 		if (dir)
822 			rcu_assign_pointer(pn->right, in);
823 		else
824 			rcu_assign_pointer(pn->left, in);
825 
826 		ln->fn_bit = plen;
827 
828 		RCU_INIT_POINTER(ln->parent, in);
829 		rcu_assign_pointer(fn->parent, in);
830 
831 		if (addr_bit_set(addr, bit)) {
832 			rcu_assign_pointer(in->right, ln);
833 			rcu_assign_pointer(in->left, fn);
834 		} else {
835 			rcu_assign_pointer(in->left, ln);
836 			rcu_assign_pointer(in->right, fn);
837 		}
838 	} else { /* plen <= bit */
839 
840 		/*
841 		 *		(new leaf node)[ln]
842 		 *	          /	   \
843 		 *	     (old node)[fn] NULL
844 		 */
845 
846 		ln = node_alloc(net);
847 
848 		if (!ln)
849 			return ERR_PTR(-ENOMEM);
850 
851 		ln->fn_bit = plen;
852 
853 		RCU_INIT_POINTER(ln->parent, pn);
854 
855 		if (addr_bit_set(&key->addr, plen))
856 			RCU_INIT_POINTER(ln->right, fn);
857 		else
858 			RCU_INIT_POINTER(ln->left, fn);
859 
860 		rcu_assign_pointer(fn->parent, ln);
861 
862 		if (dir)
863 			rcu_assign_pointer(pn->right, ln);
864 		else
865 			rcu_assign_pointer(pn->left, ln);
866 	}
867 	return ln;
868 }
869 
870 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
871 				const struct fib6_table *table)
872 {
873 	int cpu;
874 
875 	/* release the reference to this fib entry from
876 	 * all of its cached pcpu routes
877 	 */
878 	for_each_possible_cpu(cpu) {
879 		struct rt6_info **ppcpu_rt;
880 		struct rt6_info *pcpu_rt;
881 
882 		ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
883 		pcpu_rt = *ppcpu_rt;
884 		if (pcpu_rt) {
885 			struct fib6_info *from;
886 
887 			from = rcu_dereference_protected(pcpu_rt->from,
888 					     lockdep_is_held(&table->tb6_lock));
889 			rcu_assign_pointer(pcpu_rt->from, NULL);
890 			fib6_info_release(from);
891 		}
892 	}
893 }
894 
895 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
896 			  struct net *net)
897 {
898 	struct fib6_table *table = rt->fib6_table;
899 
900 	if (atomic_read(&rt->fib6_ref) != 1) {
901 		/* This route is used as dummy address holder in some split
902 		 * nodes. It is not leaked, but it still holds other resources,
903 		 * which must be released in time. So, scan ascendant nodes
904 		 * and replace dummy references to this route with references
905 		 * to still alive ones.
906 		 */
907 		while (fn) {
908 			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
909 					    lockdep_is_held(&table->tb6_lock));
910 			struct fib6_info *new_leaf;
911 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
912 				new_leaf = fib6_find_prefix(net, table, fn);
913 				atomic_inc(&new_leaf->fib6_ref);
914 
915 				rcu_assign_pointer(fn->leaf, new_leaf);
916 				fib6_info_release(rt);
917 			}
918 			fn = rcu_dereference_protected(fn->parent,
919 				    lockdep_is_held(&table->tb6_lock));
920 		}
921 
922 		if (rt->rt6i_pcpu)
923 			fib6_drop_pcpu_from(rt, table);
924 	}
925 }
926 
927 /*
928  *	Insert routing information in a node.
929  */
930 
931 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
932 			    struct nl_info *info,
933 			    struct netlink_ext_ack *extack)
934 {
935 	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
936 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
937 	struct fib6_info *iter = NULL;
938 	struct fib6_info __rcu **ins;
939 	struct fib6_info __rcu **fallback_ins = NULL;
940 	int replace = (info->nlh &&
941 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
942 	int add = (!info->nlh ||
943 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
944 	int found = 0;
945 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
946 	u16 nlflags = NLM_F_EXCL;
947 	int err;
948 
949 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
950 		nlflags |= NLM_F_APPEND;
951 
952 	ins = &fn->leaf;
953 
954 	for (iter = leaf; iter;
955 	     iter = rcu_dereference_protected(iter->fib6_next,
956 				lockdep_is_held(&rt->fib6_table->tb6_lock))) {
957 		/*
958 		 *	Search for duplicates
959 		 */
960 
961 		if (iter->fib6_metric == rt->fib6_metric) {
962 			/*
963 			 *	Same priority level
964 			 */
965 			if (info->nlh &&
966 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
967 				return -EEXIST;
968 
969 			nlflags &= ~NLM_F_EXCL;
970 			if (replace) {
971 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
972 					found++;
973 					break;
974 				}
975 				if (rt_can_ecmp)
976 					fallback_ins = fallback_ins ?: ins;
977 				goto next_iter;
978 			}
979 
980 			if (rt6_duplicate_nexthop(iter, rt)) {
981 				if (rt->fib6_nsiblings)
982 					rt->fib6_nsiblings = 0;
983 				if (!(iter->fib6_flags & RTF_EXPIRES))
984 					return -EEXIST;
985 				if (!(rt->fib6_flags & RTF_EXPIRES))
986 					fib6_clean_expires(iter);
987 				else
988 					fib6_set_expires(iter, rt->expires);
989 				fib6_metric_set(iter, RTAX_MTU, rt->fib6_pmtu);
990 				return -EEXIST;
991 			}
992 			/* If we have the same destination and the same metric,
993 			 * but not the same gateway, then the route we try to
994 			 * add is sibling to this route, increment our counter
995 			 * of siblings, and later we will add our route to the
996 			 * list.
997 			 * Only static routes (which don't have flag
998 			 * RTF_EXPIRES) are used for ECMPv6.
999 			 *
1000 			 * To avoid long list, we only had siblings if the
1001 			 * route have a gateway.
1002 			 */
1003 			if (rt_can_ecmp &&
1004 			    rt6_qualify_for_ecmp(iter))
1005 				rt->fib6_nsiblings++;
1006 		}
1007 
1008 		if (iter->fib6_metric > rt->fib6_metric)
1009 			break;
1010 
1011 next_iter:
1012 		ins = &iter->fib6_next;
1013 	}
1014 
1015 	if (fallback_ins && !found) {
1016 		/* No ECMP-able route found, replace first non-ECMP one */
1017 		ins = fallback_ins;
1018 		iter = rcu_dereference_protected(*ins,
1019 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1020 		found++;
1021 	}
1022 
1023 	/* Reset round-robin state, if necessary */
1024 	if (ins == &fn->leaf)
1025 		fn->rr_ptr = NULL;
1026 
1027 	/* Link this route to others same route. */
1028 	if (rt->fib6_nsiblings) {
1029 		unsigned int fib6_nsiblings;
1030 		struct fib6_info *sibling, *temp_sibling;
1031 
1032 		/* Find the first route that have the same metric */
1033 		sibling = leaf;
1034 		while (sibling) {
1035 			if (sibling->fib6_metric == rt->fib6_metric &&
1036 			    rt6_qualify_for_ecmp(sibling)) {
1037 				list_add_tail(&rt->fib6_siblings,
1038 					      &sibling->fib6_siblings);
1039 				break;
1040 			}
1041 			sibling = rcu_dereference_protected(sibling->fib6_next,
1042 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1043 		}
1044 		/* For each sibling in the list, increment the counter of
1045 		 * siblings. BUG() if counters does not match, list of siblings
1046 		 * is broken!
1047 		 */
1048 		fib6_nsiblings = 0;
1049 		list_for_each_entry_safe(sibling, temp_sibling,
1050 					 &rt->fib6_siblings, fib6_siblings) {
1051 			sibling->fib6_nsiblings++;
1052 			BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1053 			fib6_nsiblings++;
1054 		}
1055 		BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1056 		rt6_multipath_rebalance(temp_sibling);
1057 	}
1058 
1059 	/*
1060 	 *	insert node
1061 	 */
1062 	if (!replace) {
1063 		if (!add)
1064 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1065 
1066 add:
1067 		nlflags |= NLM_F_CREATE;
1068 
1069 		err = call_fib6_entry_notifiers(info->nl_net,
1070 						FIB_EVENT_ENTRY_ADD,
1071 						rt, extack);
1072 		if (err)
1073 			return err;
1074 
1075 		rcu_assign_pointer(rt->fib6_next, iter);
1076 		atomic_inc(&rt->fib6_ref);
1077 		rcu_assign_pointer(rt->fib6_node, fn);
1078 		rcu_assign_pointer(*ins, rt);
1079 		if (!info->skip_notify)
1080 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1081 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1082 
1083 		if (!(fn->fn_flags & RTN_RTINFO)) {
1084 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1085 			fn->fn_flags |= RTN_RTINFO;
1086 		}
1087 
1088 	} else {
1089 		int nsiblings;
1090 
1091 		if (!found) {
1092 			if (add)
1093 				goto add;
1094 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1095 			return -ENOENT;
1096 		}
1097 
1098 		err = call_fib6_entry_notifiers(info->nl_net,
1099 						FIB_EVENT_ENTRY_REPLACE,
1100 						rt, extack);
1101 		if (err)
1102 			return err;
1103 
1104 		atomic_inc(&rt->fib6_ref);
1105 		rcu_assign_pointer(rt->fib6_node, fn);
1106 		rt->fib6_next = iter->fib6_next;
1107 		rcu_assign_pointer(*ins, rt);
1108 		if (!info->skip_notify)
1109 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1110 		if (!(fn->fn_flags & RTN_RTINFO)) {
1111 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1112 			fn->fn_flags |= RTN_RTINFO;
1113 		}
1114 		nsiblings = iter->fib6_nsiblings;
1115 		iter->fib6_node = NULL;
1116 		fib6_purge_rt(iter, fn, info->nl_net);
1117 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1118 			fn->rr_ptr = NULL;
1119 		fib6_info_release(iter);
1120 
1121 		if (nsiblings) {
1122 			/* Replacing an ECMP route, remove all siblings */
1123 			ins = &rt->fib6_next;
1124 			iter = rcu_dereference_protected(*ins,
1125 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1126 			while (iter) {
1127 				if (iter->fib6_metric > rt->fib6_metric)
1128 					break;
1129 				if (rt6_qualify_for_ecmp(iter)) {
1130 					*ins = iter->fib6_next;
1131 					iter->fib6_node = NULL;
1132 					fib6_purge_rt(iter, fn, info->nl_net);
1133 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1134 						fn->rr_ptr = NULL;
1135 					fib6_info_release(iter);
1136 					nsiblings--;
1137 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1138 				} else {
1139 					ins = &iter->fib6_next;
1140 				}
1141 				iter = rcu_dereference_protected(*ins,
1142 					lockdep_is_held(&rt->fib6_table->tb6_lock));
1143 			}
1144 			WARN_ON(nsiblings != 0);
1145 		}
1146 	}
1147 
1148 	return 0;
1149 }
1150 
1151 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1152 {
1153 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1154 	    (rt->fib6_flags & RTF_EXPIRES))
1155 		mod_timer(&net->ipv6.ip6_fib_timer,
1156 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1157 }
1158 
1159 void fib6_force_start_gc(struct net *net)
1160 {
1161 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1162 		mod_timer(&net->ipv6.ip6_fib_timer,
1163 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1164 }
1165 
1166 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1167 					   int sernum)
1168 {
1169 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1170 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1171 
1172 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1173 	smp_wmb();
1174 	while (fn) {
1175 		fn->fn_sernum = sernum;
1176 		fn = rcu_dereference_protected(fn->parent,
1177 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1178 	}
1179 }
1180 
1181 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1182 {
1183 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1184 }
1185 
1186 /*
1187  *	Add routing information to the routing tree.
1188  *	<destination addr>/<source addr>
1189  *	with source addr info in sub-trees
1190  *	Need to own table->tb6_lock
1191  */
1192 
1193 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1194 	     struct nl_info *info, struct netlink_ext_ack *extack)
1195 {
1196 	struct fib6_table *table = rt->fib6_table;
1197 	struct fib6_node *fn, *pn = NULL;
1198 	int err = -ENOMEM;
1199 	int allow_create = 1;
1200 	int replace_required = 0;
1201 	int sernum = fib6_new_sernum(info->nl_net);
1202 
1203 	if (info->nlh) {
1204 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1205 			allow_create = 0;
1206 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1207 			replace_required = 1;
1208 	}
1209 	if (!allow_create && !replace_required)
1210 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1211 
1212 	fn = fib6_add_1(info->nl_net, table, root,
1213 			&rt->fib6_dst.addr, rt->fib6_dst.plen,
1214 			offsetof(struct fib6_info, fib6_dst), allow_create,
1215 			replace_required, extack);
1216 	if (IS_ERR(fn)) {
1217 		err = PTR_ERR(fn);
1218 		fn = NULL;
1219 		goto out;
1220 	}
1221 
1222 	pn = fn;
1223 
1224 #ifdef CONFIG_IPV6_SUBTREES
1225 	if (rt->fib6_src.plen) {
1226 		struct fib6_node *sn;
1227 
1228 		if (!rcu_access_pointer(fn->subtree)) {
1229 			struct fib6_node *sfn;
1230 
1231 			/*
1232 			 * Create subtree.
1233 			 *
1234 			 *		fn[main tree]
1235 			 *		|
1236 			 *		sfn[subtree root]
1237 			 *		   \
1238 			 *		    sn[new leaf node]
1239 			 */
1240 
1241 			/* Create subtree root node */
1242 			sfn = node_alloc(info->nl_net);
1243 			if (!sfn)
1244 				goto failure;
1245 
1246 			atomic_inc(&info->nl_net->ipv6.fib6_null_entry->fib6_ref);
1247 			rcu_assign_pointer(sfn->leaf,
1248 					   info->nl_net->ipv6.fib6_null_entry);
1249 			sfn->fn_flags = RTN_ROOT;
1250 
1251 			/* Now add the first leaf node to new subtree */
1252 
1253 			sn = fib6_add_1(info->nl_net, table, sfn,
1254 					&rt->fib6_src.addr, rt->fib6_src.plen,
1255 					offsetof(struct fib6_info, fib6_src),
1256 					allow_create, replace_required, extack);
1257 
1258 			if (IS_ERR(sn)) {
1259 				/* If it is failed, discard just allocated
1260 				   root, and then (in failure) stale node
1261 				   in main tree.
1262 				 */
1263 				node_free_immediate(info->nl_net, sfn);
1264 				err = PTR_ERR(sn);
1265 				goto failure;
1266 			}
1267 
1268 			/* Now link new subtree to main tree */
1269 			rcu_assign_pointer(sfn->parent, fn);
1270 			rcu_assign_pointer(fn->subtree, sfn);
1271 		} else {
1272 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1273 					&rt->fib6_src.addr, rt->fib6_src.plen,
1274 					offsetof(struct fib6_info, fib6_src),
1275 					allow_create, replace_required, extack);
1276 
1277 			if (IS_ERR(sn)) {
1278 				err = PTR_ERR(sn);
1279 				goto failure;
1280 			}
1281 		}
1282 
1283 		if (!rcu_access_pointer(fn->leaf)) {
1284 			if (fn->fn_flags & RTN_TL_ROOT) {
1285 				/* put back null_entry for root node */
1286 				rcu_assign_pointer(fn->leaf,
1287 					    info->nl_net->ipv6.fib6_null_entry);
1288 			} else {
1289 				atomic_inc(&rt->fib6_ref);
1290 				rcu_assign_pointer(fn->leaf, rt);
1291 			}
1292 		}
1293 		fn = sn;
1294 	}
1295 #endif
1296 
1297 	err = fib6_add_rt2node(fn, rt, info, extack);
1298 	if (!err) {
1299 		__fib6_update_sernum_upto_root(rt, sernum);
1300 		fib6_start_gc(info->nl_net, rt);
1301 	}
1302 
1303 out:
1304 	if (err) {
1305 #ifdef CONFIG_IPV6_SUBTREES
1306 		/*
1307 		 * If fib6_add_1 has cleared the old leaf pointer in the
1308 		 * super-tree leaf node we have to find a new one for it.
1309 		 */
1310 		if (pn != fn) {
1311 			struct fib6_info *pn_leaf =
1312 				rcu_dereference_protected(pn->leaf,
1313 				    lockdep_is_held(&table->tb6_lock));
1314 			if (pn_leaf == rt) {
1315 				pn_leaf = NULL;
1316 				RCU_INIT_POINTER(pn->leaf, NULL);
1317 				fib6_info_release(rt);
1318 			}
1319 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1320 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1321 							   pn);
1322 #if RT6_DEBUG >= 2
1323 				if (!pn_leaf) {
1324 					WARN_ON(!pn_leaf);
1325 					pn_leaf =
1326 					    info->nl_net->ipv6.fib6_null_entry;
1327 				}
1328 #endif
1329 				fib6_info_hold(pn_leaf);
1330 				rcu_assign_pointer(pn->leaf, pn_leaf);
1331 			}
1332 		}
1333 #endif
1334 		goto failure;
1335 	}
1336 	return err;
1337 
1338 failure:
1339 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1340 	 * 1. fn is an intermediate node and we failed to add the new
1341 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1342 	 * failure case.
1343 	 * 2. fn is the root node in the table and we fail to add the first
1344 	 * default route to it.
1345 	 */
1346 	if (fn &&
1347 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1348 	     (fn->fn_flags & RTN_TL_ROOT &&
1349 	      !rcu_access_pointer(fn->leaf))))
1350 		fib6_repair_tree(info->nl_net, table, fn);
1351 	return err;
1352 }
1353 
1354 /*
1355  *	Routing tree lookup
1356  *
1357  */
1358 
1359 struct lookup_args {
1360 	int			offset;		/* key offset on fib6_info */
1361 	const struct in6_addr	*addr;		/* search key			*/
1362 };
1363 
1364 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1365 					    struct lookup_args *args)
1366 {
1367 	struct fib6_node *fn;
1368 	__be32 dir;
1369 
1370 	if (unlikely(args->offset == 0))
1371 		return NULL;
1372 
1373 	/*
1374 	 *	Descend on a tree
1375 	 */
1376 
1377 	fn = root;
1378 
1379 	for (;;) {
1380 		struct fib6_node *next;
1381 
1382 		dir = addr_bit_set(args->addr, fn->fn_bit);
1383 
1384 		next = dir ? rcu_dereference(fn->right) :
1385 			     rcu_dereference(fn->left);
1386 
1387 		if (next) {
1388 			fn = next;
1389 			continue;
1390 		}
1391 		break;
1392 	}
1393 
1394 	while (fn) {
1395 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1396 
1397 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1398 			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1399 			struct rt6key *key;
1400 
1401 			if (!leaf)
1402 				goto backtrack;
1403 
1404 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1405 
1406 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1407 #ifdef CONFIG_IPV6_SUBTREES
1408 				if (subtree) {
1409 					struct fib6_node *sfn;
1410 					sfn = fib6_node_lookup_1(subtree,
1411 								 args + 1);
1412 					if (!sfn)
1413 						goto backtrack;
1414 					fn = sfn;
1415 				}
1416 #endif
1417 				if (fn->fn_flags & RTN_RTINFO)
1418 					return fn;
1419 			}
1420 		}
1421 backtrack:
1422 		if (fn->fn_flags & RTN_ROOT)
1423 			break;
1424 
1425 		fn = rcu_dereference(fn->parent);
1426 	}
1427 
1428 	return NULL;
1429 }
1430 
1431 /* called with rcu_read_lock() held
1432  */
1433 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1434 				   const struct in6_addr *daddr,
1435 				   const struct in6_addr *saddr)
1436 {
1437 	struct fib6_node *fn;
1438 	struct lookup_args args[] = {
1439 		{
1440 			.offset = offsetof(struct fib6_info, fib6_dst),
1441 			.addr = daddr,
1442 		},
1443 #ifdef CONFIG_IPV6_SUBTREES
1444 		{
1445 			.offset = offsetof(struct fib6_info, fib6_src),
1446 			.addr = saddr,
1447 		},
1448 #endif
1449 		{
1450 			.offset = 0,	/* sentinel */
1451 		}
1452 	};
1453 
1454 	fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1455 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1456 		fn = root;
1457 
1458 	return fn;
1459 }
1460 
1461 /*
1462  *	Get node with specified destination prefix (and source prefix,
1463  *	if subtrees are used)
1464  *	exact_match == true means we try to find fn with exact match of
1465  *	the passed in prefix addr
1466  *	exact_match == false means we try to find fn with longest prefix
1467  *	match of the passed in prefix addr. This is useful for finding fn
1468  *	for cached route as it will be stored in the exception table under
1469  *	the node with longest prefix length.
1470  */
1471 
1472 
1473 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1474 				       const struct in6_addr *addr,
1475 				       int plen, int offset,
1476 				       bool exact_match)
1477 {
1478 	struct fib6_node *fn, *prev = NULL;
1479 
1480 	for (fn = root; fn ; ) {
1481 		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1482 		struct rt6key *key;
1483 
1484 		/* This node is being deleted */
1485 		if (!leaf) {
1486 			if (plen <= fn->fn_bit)
1487 				goto out;
1488 			else
1489 				goto next;
1490 		}
1491 
1492 		key = (struct rt6key *)((u8 *)leaf + offset);
1493 
1494 		/*
1495 		 *	Prefix match
1496 		 */
1497 		if (plen < fn->fn_bit ||
1498 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1499 			goto out;
1500 
1501 		if (plen == fn->fn_bit)
1502 			return fn;
1503 
1504 		prev = fn;
1505 
1506 next:
1507 		/*
1508 		 *	We have more bits to go
1509 		 */
1510 		if (addr_bit_set(addr, fn->fn_bit))
1511 			fn = rcu_dereference(fn->right);
1512 		else
1513 			fn = rcu_dereference(fn->left);
1514 	}
1515 out:
1516 	if (exact_match)
1517 		return NULL;
1518 	else
1519 		return prev;
1520 }
1521 
1522 struct fib6_node *fib6_locate(struct fib6_node *root,
1523 			      const struct in6_addr *daddr, int dst_len,
1524 			      const struct in6_addr *saddr, int src_len,
1525 			      bool exact_match)
1526 {
1527 	struct fib6_node *fn;
1528 
1529 	fn = fib6_locate_1(root, daddr, dst_len,
1530 			   offsetof(struct fib6_info, fib6_dst),
1531 			   exact_match);
1532 
1533 #ifdef CONFIG_IPV6_SUBTREES
1534 	if (src_len) {
1535 		WARN_ON(saddr == NULL);
1536 		if (fn) {
1537 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1538 
1539 			if (subtree) {
1540 				fn = fib6_locate_1(subtree, saddr, src_len,
1541 					   offsetof(struct fib6_info, fib6_src),
1542 					   exact_match);
1543 			}
1544 		}
1545 	}
1546 #endif
1547 
1548 	if (fn && fn->fn_flags & RTN_RTINFO)
1549 		return fn;
1550 
1551 	return NULL;
1552 }
1553 
1554 
1555 /*
1556  *	Deletion
1557  *
1558  */
1559 
1560 static struct fib6_info *fib6_find_prefix(struct net *net,
1561 					 struct fib6_table *table,
1562 					 struct fib6_node *fn)
1563 {
1564 	struct fib6_node *child_left, *child_right;
1565 
1566 	if (fn->fn_flags & RTN_ROOT)
1567 		return net->ipv6.fib6_null_entry;
1568 
1569 	while (fn) {
1570 		child_left = rcu_dereference_protected(fn->left,
1571 				    lockdep_is_held(&table->tb6_lock));
1572 		child_right = rcu_dereference_protected(fn->right,
1573 				    lockdep_is_held(&table->tb6_lock));
1574 		if (child_left)
1575 			return rcu_dereference_protected(child_left->leaf,
1576 					lockdep_is_held(&table->tb6_lock));
1577 		if (child_right)
1578 			return rcu_dereference_protected(child_right->leaf,
1579 					lockdep_is_held(&table->tb6_lock));
1580 
1581 		fn = FIB6_SUBTREE(fn);
1582 	}
1583 	return NULL;
1584 }
1585 
1586 /*
1587  *	Called to trim the tree of intermediate nodes when possible. "fn"
1588  *	is the node we want to try and remove.
1589  *	Need to own table->tb6_lock
1590  */
1591 
1592 static struct fib6_node *fib6_repair_tree(struct net *net,
1593 					  struct fib6_table *table,
1594 					  struct fib6_node *fn)
1595 {
1596 	int children;
1597 	int nstate;
1598 	struct fib6_node *child;
1599 	struct fib6_walker *w;
1600 	int iter = 0;
1601 
1602 	/* Set fn->leaf to null_entry for root node. */
1603 	if (fn->fn_flags & RTN_TL_ROOT) {
1604 		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1605 		return fn;
1606 	}
1607 
1608 	for (;;) {
1609 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1610 					    lockdep_is_held(&table->tb6_lock));
1611 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1612 					    lockdep_is_held(&table->tb6_lock));
1613 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1614 					    lockdep_is_held(&table->tb6_lock));
1615 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1616 					    lockdep_is_held(&table->tb6_lock));
1617 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1618 					    lockdep_is_held(&table->tb6_lock));
1619 		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1620 					    lockdep_is_held(&table->tb6_lock));
1621 		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1622 					    lockdep_is_held(&table->tb6_lock));
1623 		struct fib6_info *new_fn_leaf;
1624 
1625 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1626 		iter++;
1627 
1628 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1629 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1630 		WARN_ON(fn_leaf);
1631 
1632 		children = 0;
1633 		child = NULL;
1634 		if (fn_r)
1635 			child = fn_r, children |= 1;
1636 		if (fn_l)
1637 			child = fn_l, children |= 2;
1638 
1639 		if (children == 3 || FIB6_SUBTREE(fn)
1640 #ifdef CONFIG_IPV6_SUBTREES
1641 		    /* Subtree root (i.e. fn) may have one child */
1642 		    || (children && fn->fn_flags & RTN_ROOT)
1643 #endif
1644 		    ) {
1645 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1646 #if RT6_DEBUG >= 2
1647 			if (!new_fn_leaf) {
1648 				WARN_ON(!new_fn_leaf);
1649 				new_fn_leaf = net->ipv6.fib6_null_entry;
1650 			}
1651 #endif
1652 			fib6_info_hold(new_fn_leaf);
1653 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1654 			return pn;
1655 		}
1656 
1657 #ifdef CONFIG_IPV6_SUBTREES
1658 		if (FIB6_SUBTREE(pn) == fn) {
1659 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1660 			RCU_INIT_POINTER(pn->subtree, NULL);
1661 			nstate = FWS_L;
1662 		} else {
1663 			WARN_ON(fn->fn_flags & RTN_ROOT);
1664 #endif
1665 			if (pn_r == fn)
1666 				rcu_assign_pointer(pn->right, child);
1667 			else if (pn_l == fn)
1668 				rcu_assign_pointer(pn->left, child);
1669 #if RT6_DEBUG >= 2
1670 			else
1671 				WARN_ON(1);
1672 #endif
1673 			if (child)
1674 				rcu_assign_pointer(child->parent, pn);
1675 			nstate = FWS_R;
1676 #ifdef CONFIG_IPV6_SUBTREES
1677 		}
1678 #endif
1679 
1680 		read_lock(&net->ipv6.fib6_walker_lock);
1681 		FOR_WALKERS(net, w) {
1682 			if (!child) {
1683 				if (w->node == fn) {
1684 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1685 					w->node = pn;
1686 					w->state = nstate;
1687 				}
1688 			} else {
1689 				if (w->node == fn) {
1690 					w->node = child;
1691 					if (children&2) {
1692 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1693 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1694 					} else {
1695 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1696 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1697 					}
1698 				}
1699 			}
1700 		}
1701 		read_unlock(&net->ipv6.fib6_walker_lock);
1702 
1703 		node_free(net, fn);
1704 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1705 			return pn;
1706 
1707 		RCU_INIT_POINTER(pn->leaf, NULL);
1708 		fib6_info_release(pn_leaf);
1709 		fn = pn;
1710 	}
1711 }
1712 
1713 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1714 			   struct fib6_info __rcu **rtp, struct nl_info *info)
1715 {
1716 	struct fib6_walker *w;
1717 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1718 				    lockdep_is_held(&table->tb6_lock));
1719 	struct net *net = info->nl_net;
1720 
1721 	RT6_TRACE("fib6_del_route\n");
1722 
1723 	/* Unlink it */
1724 	*rtp = rt->fib6_next;
1725 	rt->fib6_node = NULL;
1726 	net->ipv6.rt6_stats->fib_rt_entries--;
1727 	net->ipv6.rt6_stats->fib_discarded_routes++;
1728 
1729 	/* Flush all cached dst in exception table */
1730 	rt6_flush_exceptions(rt);
1731 
1732 	/* Reset round-robin state, if necessary */
1733 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1734 		fn->rr_ptr = NULL;
1735 
1736 	/* Remove this entry from other siblings */
1737 	if (rt->fib6_nsiblings) {
1738 		struct fib6_info *sibling, *next_sibling;
1739 
1740 		list_for_each_entry_safe(sibling, next_sibling,
1741 					 &rt->fib6_siblings, fib6_siblings)
1742 			sibling->fib6_nsiblings--;
1743 		rt->fib6_nsiblings = 0;
1744 		list_del_init(&rt->fib6_siblings);
1745 		rt6_multipath_rebalance(next_sibling);
1746 	}
1747 
1748 	/* Adjust walkers */
1749 	read_lock(&net->ipv6.fib6_walker_lock);
1750 	FOR_WALKERS(net, w) {
1751 		if (w->state == FWS_C && w->leaf == rt) {
1752 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1753 			w->leaf = rcu_dereference_protected(rt->fib6_next,
1754 					    lockdep_is_held(&table->tb6_lock));
1755 			if (!w->leaf)
1756 				w->state = FWS_U;
1757 		}
1758 	}
1759 	read_unlock(&net->ipv6.fib6_walker_lock);
1760 
1761 	/* If it was last route, call fib6_repair_tree() to:
1762 	 * 1. For root node, put back null_entry as how the table was created.
1763 	 * 2. For other nodes, expunge its radix tree node.
1764 	 */
1765 	if (!rcu_access_pointer(fn->leaf)) {
1766 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1767 			fn->fn_flags &= ~RTN_RTINFO;
1768 			net->ipv6.rt6_stats->fib_route_nodes--;
1769 		}
1770 		fn = fib6_repair_tree(net, table, fn);
1771 	}
1772 
1773 	fib6_purge_rt(rt, fn, net);
1774 
1775 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1776 	if (!info->skip_notify)
1777 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1778 	fib6_info_release(rt);
1779 }
1780 
1781 /* Need to own table->tb6_lock */
1782 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1783 {
1784 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1785 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1786 	struct fib6_table *table = rt->fib6_table;
1787 	struct net *net = info->nl_net;
1788 	struct fib6_info __rcu **rtp;
1789 	struct fib6_info __rcu **rtp_next;
1790 
1791 	if (!fn || rt == net->ipv6.fib6_null_entry)
1792 		return -ENOENT;
1793 
1794 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1795 
1796 	/*
1797 	 *	Walk the leaf entries looking for ourself
1798 	 */
1799 
1800 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1801 		struct fib6_info *cur = rcu_dereference_protected(*rtp,
1802 					lockdep_is_held(&table->tb6_lock));
1803 		if (rt == cur) {
1804 			fib6_del_route(table, fn, rtp, info);
1805 			return 0;
1806 		}
1807 		rtp_next = &cur->fib6_next;
1808 	}
1809 	return -ENOENT;
1810 }
1811 
1812 /*
1813  *	Tree traversal function.
1814  *
1815  *	Certainly, it is not interrupt safe.
1816  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1817  *	It means, that we can modify tree during walking
1818  *	and use this function for garbage collection, clone pruning,
1819  *	cleaning tree when a device goes down etc. etc.
1820  *
1821  *	It guarantees that every node will be traversed,
1822  *	and that it will be traversed only once.
1823  *
1824  *	Callback function w->func may return:
1825  *	0 -> continue walking.
1826  *	positive value -> walking is suspended (used by tree dumps,
1827  *	and probably by gc, if it will be split to several slices)
1828  *	negative value -> terminate walking.
1829  *
1830  *	The function itself returns:
1831  *	0   -> walk is complete.
1832  *	>0  -> walk is incomplete (i.e. suspended)
1833  *	<0  -> walk is terminated by an error.
1834  *
1835  *	This function is called with tb6_lock held.
1836  */
1837 
1838 static int fib6_walk_continue(struct fib6_walker *w)
1839 {
1840 	struct fib6_node *fn, *pn, *left, *right;
1841 
1842 	/* w->root should always be table->tb6_root */
1843 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1844 
1845 	for (;;) {
1846 		fn = w->node;
1847 		if (!fn)
1848 			return 0;
1849 
1850 		switch (w->state) {
1851 #ifdef CONFIG_IPV6_SUBTREES
1852 		case FWS_S:
1853 			if (FIB6_SUBTREE(fn)) {
1854 				w->node = FIB6_SUBTREE(fn);
1855 				continue;
1856 			}
1857 			w->state = FWS_L;
1858 #endif
1859 			/* fall through */
1860 		case FWS_L:
1861 			left = rcu_dereference_protected(fn->left, 1);
1862 			if (left) {
1863 				w->node = left;
1864 				w->state = FWS_INIT;
1865 				continue;
1866 			}
1867 			w->state = FWS_R;
1868 			/* fall through */
1869 		case FWS_R:
1870 			right = rcu_dereference_protected(fn->right, 1);
1871 			if (right) {
1872 				w->node = right;
1873 				w->state = FWS_INIT;
1874 				continue;
1875 			}
1876 			w->state = FWS_C;
1877 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1878 			/* fall through */
1879 		case FWS_C:
1880 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1881 				int err;
1882 
1883 				if (w->skip) {
1884 					w->skip--;
1885 					goto skip;
1886 				}
1887 
1888 				err = w->func(w);
1889 				if (err)
1890 					return err;
1891 
1892 				w->count++;
1893 				continue;
1894 			}
1895 skip:
1896 			w->state = FWS_U;
1897 			/* fall through */
1898 		case FWS_U:
1899 			if (fn == w->root)
1900 				return 0;
1901 			pn = rcu_dereference_protected(fn->parent, 1);
1902 			left = rcu_dereference_protected(pn->left, 1);
1903 			right = rcu_dereference_protected(pn->right, 1);
1904 			w->node = pn;
1905 #ifdef CONFIG_IPV6_SUBTREES
1906 			if (FIB6_SUBTREE(pn) == fn) {
1907 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1908 				w->state = FWS_L;
1909 				continue;
1910 			}
1911 #endif
1912 			if (left == fn) {
1913 				w->state = FWS_R;
1914 				continue;
1915 			}
1916 			if (right == fn) {
1917 				w->state = FWS_C;
1918 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1919 				continue;
1920 			}
1921 #if RT6_DEBUG >= 2
1922 			WARN_ON(1);
1923 #endif
1924 		}
1925 	}
1926 }
1927 
1928 static int fib6_walk(struct net *net, struct fib6_walker *w)
1929 {
1930 	int res;
1931 
1932 	w->state = FWS_INIT;
1933 	w->node = w->root;
1934 
1935 	fib6_walker_link(net, w);
1936 	res = fib6_walk_continue(w);
1937 	if (res <= 0)
1938 		fib6_walker_unlink(net, w);
1939 	return res;
1940 }
1941 
1942 static int fib6_clean_node(struct fib6_walker *w)
1943 {
1944 	int res;
1945 	struct fib6_info *rt;
1946 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1947 	struct nl_info info = {
1948 		.nl_net = c->net,
1949 	};
1950 
1951 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1952 	    w->node->fn_sernum != c->sernum)
1953 		w->node->fn_sernum = c->sernum;
1954 
1955 	if (!c->func) {
1956 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1957 		w->leaf = NULL;
1958 		return 0;
1959 	}
1960 
1961 	for_each_fib6_walker_rt(w) {
1962 		res = c->func(rt, c->arg);
1963 		if (res == -1) {
1964 			w->leaf = rt;
1965 			res = fib6_del(rt, &info);
1966 			if (res) {
1967 #if RT6_DEBUG >= 2
1968 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1969 					 __func__, rt,
1970 					 rcu_access_pointer(rt->fib6_node),
1971 					 res);
1972 #endif
1973 				continue;
1974 			}
1975 			return 0;
1976 		} else if (res == -2) {
1977 			if (WARN_ON(!rt->fib6_nsiblings))
1978 				continue;
1979 			rt = list_last_entry(&rt->fib6_siblings,
1980 					     struct fib6_info, fib6_siblings);
1981 			continue;
1982 		}
1983 		WARN_ON(res != 0);
1984 	}
1985 	w->leaf = rt;
1986 	return 0;
1987 }
1988 
1989 /*
1990  *	Convenient frontend to tree walker.
1991  *
1992  *	func is called on each route.
1993  *		It may return -2 -> skip multipath route.
1994  *			      -1 -> delete this route.
1995  *		              0  -> continue walking
1996  */
1997 
1998 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1999 			    int (*func)(struct fib6_info *, void *arg),
2000 			    int sernum, void *arg)
2001 {
2002 	struct fib6_cleaner c;
2003 
2004 	c.w.root = root;
2005 	c.w.func = fib6_clean_node;
2006 	c.w.count = 0;
2007 	c.w.skip = 0;
2008 	c.func = func;
2009 	c.sernum = sernum;
2010 	c.arg = arg;
2011 	c.net = net;
2012 
2013 	fib6_walk(net, &c.w);
2014 }
2015 
2016 static void __fib6_clean_all(struct net *net,
2017 			     int (*func)(struct fib6_info *, void *),
2018 			     int sernum, void *arg)
2019 {
2020 	struct fib6_table *table;
2021 	struct hlist_head *head;
2022 	unsigned int h;
2023 
2024 	rcu_read_lock();
2025 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2026 		head = &net->ipv6.fib_table_hash[h];
2027 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2028 			spin_lock_bh(&table->tb6_lock);
2029 			fib6_clean_tree(net, &table->tb6_root,
2030 					func, sernum, arg);
2031 			spin_unlock_bh(&table->tb6_lock);
2032 		}
2033 	}
2034 	rcu_read_unlock();
2035 }
2036 
2037 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2038 		    void *arg)
2039 {
2040 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
2041 }
2042 
2043 static void fib6_flush_trees(struct net *net)
2044 {
2045 	int new_sernum = fib6_new_sernum(net);
2046 
2047 	__fib6_clean_all(net, NULL, new_sernum, NULL);
2048 }
2049 
2050 /*
2051  *	Garbage collection
2052  */
2053 
2054 static int fib6_age(struct fib6_info *rt, void *arg)
2055 {
2056 	struct fib6_gc_args *gc_args = arg;
2057 	unsigned long now = jiffies;
2058 
2059 	/*
2060 	 *	check addrconf expiration here.
2061 	 *	Routes are expired even if they are in use.
2062 	 */
2063 
2064 	if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2065 		if (time_after(now, rt->expires)) {
2066 			RT6_TRACE("expiring %p\n", rt);
2067 			return -1;
2068 		}
2069 		gc_args->more++;
2070 	}
2071 
2072 	/*	Also age clones in the exception table.
2073 	 *	Note, that clones are aged out
2074 	 *	only if they are not in use now.
2075 	 */
2076 	rt6_age_exceptions(rt, gc_args, now);
2077 
2078 	return 0;
2079 }
2080 
2081 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2082 {
2083 	struct fib6_gc_args gc_args;
2084 	unsigned long now;
2085 
2086 	if (force) {
2087 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2088 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2089 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2090 		return;
2091 	}
2092 	gc_args.timeout = expires ? (int)expires :
2093 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2094 	gc_args.more = 0;
2095 
2096 	fib6_clean_all(net, fib6_age, &gc_args);
2097 	now = jiffies;
2098 	net->ipv6.ip6_rt_last_gc = now;
2099 
2100 	if (gc_args.more)
2101 		mod_timer(&net->ipv6.ip6_fib_timer,
2102 			  round_jiffies(now
2103 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2104 	else
2105 		del_timer(&net->ipv6.ip6_fib_timer);
2106 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2107 }
2108 
2109 static void fib6_gc_timer_cb(struct timer_list *t)
2110 {
2111 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2112 
2113 	fib6_run_gc(0, arg, true);
2114 }
2115 
2116 static int __net_init fib6_net_init(struct net *net)
2117 {
2118 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2119 	int err;
2120 
2121 	err = fib6_notifier_init(net);
2122 	if (err)
2123 		return err;
2124 
2125 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2126 	rwlock_init(&net->ipv6.fib6_walker_lock);
2127 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2128 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2129 
2130 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2131 	if (!net->ipv6.rt6_stats)
2132 		goto out_timer;
2133 
2134 	/* Avoid false sharing : Use at least a full cache line */
2135 	size = max_t(size_t, size, L1_CACHE_BYTES);
2136 
2137 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2138 	if (!net->ipv6.fib_table_hash)
2139 		goto out_rt6_stats;
2140 
2141 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2142 					  GFP_KERNEL);
2143 	if (!net->ipv6.fib6_main_tbl)
2144 		goto out_fib_table_hash;
2145 
2146 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2147 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2148 			   net->ipv6.fib6_null_entry);
2149 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2150 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2151 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2152 
2153 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2154 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2155 					   GFP_KERNEL);
2156 	if (!net->ipv6.fib6_local_tbl)
2157 		goto out_fib6_main_tbl;
2158 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2159 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2160 			   net->ipv6.fib6_null_entry);
2161 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2162 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2163 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2164 #endif
2165 	fib6_tables_init(net);
2166 
2167 	return 0;
2168 
2169 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2170 out_fib6_main_tbl:
2171 	kfree(net->ipv6.fib6_main_tbl);
2172 #endif
2173 out_fib_table_hash:
2174 	kfree(net->ipv6.fib_table_hash);
2175 out_rt6_stats:
2176 	kfree(net->ipv6.rt6_stats);
2177 out_timer:
2178 	fib6_notifier_exit(net);
2179 	return -ENOMEM;
2180 }
2181 
2182 static void fib6_net_exit(struct net *net)
2183 {
2184 	unsigned int i;
2185 
2186 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2187 
2188 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2189 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2190 		struct hlist_node *tmp;
2191 		struct fib6_table *tb;
2192 
2193 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2194 			hlist_del(&tb->tb6_hlist);
2195 			fib6_free_table(tb);
2196 		}
2197 	}
2198 
2199 	kfree(net->ipv6.fib_table_hash);
2200 	kfree(net->ipv6.rt6_stats);
2201 	fib6_notifier_exit(net);
2202 }
2203 
2204 static struct pernet_operations fib6_net_ops = {
2205 	.init = fib6_net_init,
2206 	.exit = fib6_net_exit,
2207 };
2208 
2209 int __init fib6_init(void)
2210 {
2211 	int ret = -ENOMEM;
2212 
2213 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2214 					   sizeof(struct fib6_node),
2215 					   0, SLAB_HWCACHE_ALIGN,
2216 					   NULL);
2217 	if (!fib6_node_kmem)
2218 		goto out;
2219 
2220 	ret = register_pernet_subsys(&fib6_net_ops);
2221 	if (ret)
2222 		goto out_kmem_cache_create;
2223 
2224 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2225 				   inet6_dump_fib, 0);
2226 	if (ret)
2227 		goto out_unregister_subsys;
2228 
2229 	__fib6_flush_trees = fib6_flush_trees;
2230 out:
2231 	return ret;
2232 
2233 out_unregister_subsys:
2234 	unregister_pernet_subsys(&fib6_net_ops);
2235 out_kmem_cache_create:
2236 	kmem_cache_destroy(fib6_node_kmem);
2237 	goto out;
2238 }
2239 
2240 void fib6_gc_cleanup(void)
2241 {
2242 	unregister_pernet_subsys(&fib6_net_ops);
2243 	kmem_cache_destroy(fib6_node_kmem);
2244 }
2245 
2246 #ifdef CONFIG_PROC_FS
2247 
2248 struct ipv6_route_iter {
2249 	struct seq_net_private p;
2250 	struct fib6_walker w;
2251 	loff_t skip;
2252 	struct fib6_table *tbl;
2253 	int sernum;
2254 };
2255 
2256 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2257 {
2258 	struct fib6_info *rt = v;
2259 	struct ipv6_route_iter *iter = seq->private;
2260 	const struct net_device *dev;
2261 
2262 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2263 
2264 #ifdef CONFIG_IPV6_SUBTREES
2265 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2266 #else
2267 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2268 #endif
2269 	if (rt->fib6_flags & RTF_GATEWAY)
2270 		seq_printf(seq, "%pi6", &rt->fib6_nh.nh_gw);
2271 	else
2272 		seq_puts(seq, "00000000000000000000000000000000");
2273 
2274 	dev = rt->fib6_nh.nh_dev;
2275 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2276 		   rt->fib6_metric, atomic_read(&rt->fib6_ref), 0,
2277 		   rt->fib6_flags, dev ? dev->name : "");
2278 	iter->w.leaf = NULL;
2279 	return 0;
2280 }
2281 
2282 static int ipv6_route_yield(struct fib6_walker *w)
2283 {
2284 	struct ipv6_route_iter *iter = w->args;
2285 
2286 	if (!iter->skip)
2287 		return 1;
2288 
2289 	do {
2290 		iter->w.leaf = rcu_dereference_protected(
2291 				iter->w.leaf->fib6_next,
2292 				lockdep_is_held(&iter->tbl->tb6_lock));
2293 		iter->skip--;
2294 		if (!iter->skip && iter->w.leaf)
2295 			return 1;
2296 	} while (iter->w.leaf);
2297 
2298 	return 0;
2299 }
2300 
2301 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2302 				      struct net *net)
2303 {
2304 	memset(&iter->w, 0, sizeof(iter->w));
2305 	iter->w.func = ipv6_route_yield;
2306 	iter->w.root = &iter->tbl->tb6_root;
2307 	iter->w.state = FWS_INIT;
2308 	iter->w.node = iter->w.root;
2309 	iter->w.args = iter;
2310 	iter->sernum = iter->w.root->fn_sernum;
2311 	INIT_LIST_HEAD(&iter->w.lh);
2312 	fib6_walker_link(net, &iter->w);
2313 }
2314 
2315 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2316 						    struct net *net)
2317 {
2318 	unsigned int h;
2319 	struct hlist_node *node;
2320 
2321 	if (tbl) {
2322 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2323 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2324 	} else {
2325 		h = 0;
2326 		node = NULL;
2327 	}
2328 
2329 	while (!node && h < FIB6_TABLE_HASHSZ) {
2330 		node = rcu_dereference_bh(
2331 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2332 	}
2333 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2334 }
2335 
2336 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2337 {
2338 	if (iter->sernum != iter->w.root->fn_sernum) {
2339 		iter->sernum = iter->w.root->fn_sernum;
2340 		iter->w.state = FWS_INIT;
2341 		iter->w.node = iter->w.root;
2342 		WARN_ON(iter->w.skip);
2343 		iter->w.skip = iter->w.count;
2344 	}
2345 }
2346 
2347 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2348 {
2349 	int r;
2350 	struct fib6_info *n;
2351 	struct net *net = seq_file_net(seq);
2352 	struct ipv6_route_iter *iter = seq->private;
2353 
2354 	if (!v)
2355 		goto iter_table;
2356 
2357 	n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2358 	if (n) {
2359 		++*pos;
2360 		return n;
2361 	}
2362 
2363 iter_table:
2364 	ipv6_route_check_sernum(iter);
2365 	spin_lock_bh(&iter->tbl->tb6_lock);
2366 	r = fib6_walk_continue(&iter->w);
2367 	spin_unlock_bh(&iter->tbl->tb6_lock);
2368 	if (r > 0) {
2369 		if (v)
2370 			++*pos;
2371 		return iter->w.leaf;
2372 	} else if (r < 0) {
2373 		fib6_walker_unlink(net, &iter->w);
2374 		return NULL;
2375 	}
2376 	fib6_walker_unlink(net, &iter->w);
2377 
2378 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2379 	if (!iter->tbl)
2380 		return NULL;
2381 
2382 	ipv6_route_seq_setup_walk(iter, net);
2383 	goto iter_table;
2384 }
2385 
2386 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2387 	__acquires(RCU_BH)
2388 {
2389 	struct net *net = seq_file_net(seq);
2390 	struct ipv6_route_iter *iter = seq->private;
2391 
2392 	rcu_read_lock_bh();
2393 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2394 	iter->skip = *pos;
2395 
2396 	if (iter->tbl) {
2397 		ipv6_route_seq_setup_walk(iter, net);
2398 		return ipv6_route_seq_next(seq, NULL, pos);
2399 	} else {
2400 		return NULL;
2401 	}
2402 }
2403 
2404 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2405 {
2406 	struct fib6_walker *w = &iter->w;
2407 	return w->node && !(w->state == FWS_U && w->node == w->root);
2408 }
2409 
2410 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2411 	__releases(RCU_BH)
2412 {
2413 	struct net *net = seq_file_net(seq);
2414 	struct ipv6_route_iter *iter = seq->private;
2415 
2416 	if (ipv6_route_iter_active(iter))
2417 		fib6_walker_unlink(net, &iter->w);
2418 
2419 	rcu_read_unlock_bh();
2420 }
2421 
2422 static const struct seq_operations ipv6_route_seq_ops = {
2423 	.start	= ipv6_route_seq_start,
2424 	.next	= ipv6_route_seq_next,
2425 	.stop	= ipv6_route_seq_stop,
2426 	.show	= ipv6_route_seq_show
2427 };
2428 
2429 int ipv6_route_open(struct inode *inode, struct file *file)
2430 {
2431 	return seq_open_net(inode, file, &ipv6_route_seq_ops,
2432 			    sizeof(struct ipv6_route_iter));
2433 }
2434 
2435 #endif /* CONFIG_PROC_FS */
2436