xref: /openbmc/linux/net/ipv6/ip6_fib.c (revision e3d786a3)
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/ip.h>
33 #include <net/ipv6.h>
34 #include <net/ndisc.h>
35 #include <net/addrconf.h>
36 #include <net/lwtunnel.h>
37 #include <net/fib_notifier.h>
38 
39 #include <net/ip6_fib.h>
40 #include <net/ip6_route.h>
41 
42 static struct kmem_cache *fib6_node_kmem __read_mostly;
43 
44 struct fib6_cleaner {
45 	struct fib6_walker w;
46 	struct net *net;
47 	int (*func)(struct fib6_info *, void *arg);
48 	int sernum;
49 	void *arg;
50 	bool skip_notify;
51 };
52 
53 #ifdef CONFIG_IPV6_SUBTREES
54 #define FWS_INIT FWS_S
55 #else
56 #define FWS_INIT FWS_L
57 #endif
58 
59 static struct fib6_info *fib6_find_prefix(struct net *net,
60 					 struct fib6_table *table,
61 					 struct fib6_node *fn);
62 static struct fib6_node *fib6_repair_tree(struct net *net,
63 					  struct fib6_table *table,
64 					  struct fib6_node *fn);
65 static int fib6_walk(struct net *net, struct fib6_walker *w);
66 static int fib6_walk_continue(struct fib6_walker *w);
67 
68 /*
69  *	A routing update causes an increase of the serial number on the
70  *	affected subtree. This allows for cached routes to be asynchronously
71  *	tested when modifications are made to the destination cache as a
72  *	result of redirects, path MTU changes, etc.
73  */
74 
75 static void fib6_gc_timer_cb(struct timer_list *t);
76 
77 #define FOR_WALKERS(net, w) \
78 	list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
79 
80 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
81 {
82 	write_lock_bh(&net->ipv6.fib6_walker_lock);
83 	list_add(&w->lh, &net->ipv6.fib6_walkers);
84 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
85 }
86 
87 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
88 {
89 	write_lock_bh(&net->ipv6.fib6_walker_lock);
90 	list_del(&w->lh);
91 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
92 }
93 
94 static int fib6_new_sernum(struct net *net)
95 {
96 	int new, old;
97 
98 	do {
99 		old = atomic_read(&net->ipv6.fib6_sernum);
100 		new = old < INT_MAX ? old + 1 : 1;
101 	} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
102 				old, new) != old);
103 	return new;
104 }
105 
106 enum {
107 	FIB6_NO_SERNUM_CHANGE = 0,
108 };
109 
110 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
111 {
112 	struct fib6_node *fn;
113 
114 	fn = rcu_dereference_protected(f6i->fib6_node,
115 			lockdep_is_held(&f6i->fib6_table->tb6_lock));
116 	if (fn)
117 		fn->fn_sernum = fib6_new_sernum(net);
118 }
119 
120 /*
121  *	Auxiliary address test functions for the radix tree.
122  *
123  *	These assume a 32bit processor (although it will work on
124  *	64bit processors)
125  */
126 
127 /*
128  *	test bit
129  */
130 #if defined(__LITTLE_ENDIAN)
131 # define BITOP_BE32_SWIZZLE	(0x1F & ~7)
132 #else
133 # define BITOP_BE32_SWIZZLE	0
134 #endif
135 
136 static __be32 addr_bit_set(const void *token, int fn_bit)
137 {
138 	const __be32 *addr = token;
139 	/*
140 	 * Here,
141 	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
142 	 * is optimized version of
143 	 *	htonl(1 << ((~fn_bit)&0x1F))
144 	 * See include/asm-generic/bitops/le.h.
145 	 */
146 	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
147 	       addr[fn_bit >> 5];
148 }
149 
150 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags)
151 {
152 	struct fib6_info *f6i;
153 
154 	f6i = kzalloc(sizeof(*f6i), gfp_flags);
155 	if (!f6i)
156 		return NULL;
157 
158 	f6i->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
159 	if (!f6i->rt6i_pcpu) {
160 		kfree(f6i);
161 		return NULL;
162 	}
163 
164 	INIT_LIST_HEAD(&f6i->fib6_siblings);
165 	atomic_inc(&f6i->fib6_ref);
166 
167 	return f6i;
168 }
169 
170 void fib6_info_destroy_rcu(struct rcu_head *head)
171 {
172 	struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
173 	struct rt6_exception_bucket *bucket;
174 
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 		free_percpu(f6i->rt6i_pcpu);
200 	}
201 
202 	lwtstate_put(f6i->fib6_nh.nh_lwtstate);
203 
204 	if (f6i->fib6_nh.nh_dev)
205 		dev_put(f6i->fib6_nh.nh_dev);
206 
207 	ip_fib_metrics_put(f6i->fib6_metrics);
208 
209 	kfree(f6i);
210 }
211 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
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 /* called with rcu lock held; no reference taken on fib6_info */
360 struct fib6_info *fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
361 			      int flags)
362 {
363 	return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6, flags);
364 }
365 
366 static void __net_init fib6_tables_init(struct net *net)
367 {
368 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
369 }
370 
371 #endif
372 
373 unsigned int fib6_tables_seq_read(struct net *net)
374 {
375 	unsigned int h, fib_seq = 0;
376 
377 	rcu_read_lock();
378 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
379 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
380 		struct fib6_table *tb;
381 
382 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
383 			fib_seq += tb->fib_seq;
384 	}
385 	rcu_read_unlock();
386 
387 	return fib_seq;
388 }
389 
390 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
391 				    enum fib_event_type event_type,
392 				    struct fib6_info *rt)
393 {
394 	struct fib6_entry_notifier_info info = {
395 		.rt = rt,
396 	};
397 
398 	return call_fib6_notifier(nb, net, event_type, &info.info);
399 }
400 
401 static int call_fib6_entry_notifiers(struct net *net,
402 				     enum fib_event_type event_type,
403 				     struct fib6_info *rt,
404 				     struct netlink_ext_ack *extack)
405 {
406 	struct fib6_entry_notifier_info info = {
407 		.info.extack = extack,
408 		.rt = rt,
409 	};
410 
411 	rt->fib6_table->fib_seq++;
412 	return call_fib6_notifiers(net, event_type, &info.info);
413 }
414 
415 struct fib6_dump_arg {
416 	struct net *net;
417 	struct notifier_block *nb;
418 };
419 
420 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
421 {
422 	if (rt == arg->net->ipv6.fib6_null_entry)
423 		return;
424 	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
425 }
426 
427 static int fib6_node_dump(struct fib6_walker *w)
428 {
429 	struct fib6_info *rt;
430 
431 	for_each_fib6_walker_rt(w)
432 		fib6_rt_dump(rt, w->args);
433 	w->leaf = NULL;
434 	return 0;
435 }
436 
437 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
438 			    struct fib6_walker *w)
439 {
440 	w->root = &tb->tb6_root;
441 	spin_lock_bh(&tb->tb6_lock);
442 	fib6_walk(net, w);
443 	spin_unlock_bh(&tb->tb6_lock);
444 }
445 
446 /* Called with rcu_read_lock() */
447 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
448 {
449 	struct fib6_dump_arg arg;
450 	struct fib6_walker *w;
451 	unsigned int h;
452 
453 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
454 	if (!w)
455 		return -ENOMEM;
456 
457 	w->func = fib6_node_dump;
458 	arg.net = net;
459 	arg.nb = nb;
460 	w->args = &arg;
461 
462 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
463 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
464 		struct fib6_table *tb;
465 
466 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
467 			fib6_table_dump(net, tb, w);
468 	}
469 
470 	kfree(w);
471 
472 	return 0;
473 }
474 
475 static int fib6_dump_node(struct fib6_walker *w)
476 {
477 	int res;
478 	struct fib6_info *rt;
479 
480 	for_each_fib6_walker_rt(w) {
481 		res = rt6_dump_route(rt, w->args);
482 		if (res < 0) {
483 			/* Frame is full, suspend walking */
484 			w->leaf = rt;
485 			return 1;
486 		}
487 
488 		/* Multipath routes are dumped in one route with the
489 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
490 		 * last sibling of this route (no need to dump the
491 		 * sibling routes again)
492 		 */
493 		if (rt->fib6_nsiblings)
494 			rt = list_last_entry(&rt->fib6_siblings,
495 					     struct fib6_info,
496 					     fib6_siblings);
497 	}
498 	w->leaf = NULL;
499 	return 0;
500 }
501 
502 static void fib6_dump_end(struct netlink_callback *cb)
503 {
504 	struct net *net = sock_net(cb->skb->sk);
505 	struct fib6_walker *w = (void *)cb->args[2];
506 
507 	if (w) {
508 		if (cb->args[4]) {
509 			cb->args[4] = 0;
510 			fib6_walker_unlink(net, w);
511 		}
512 		cb->args[2] = 0;
513 		kfree(w);
514 	}
515 	cb->done = (void *)cb->args[3];
516 	cb->args[1] = 3;
517 }
518 
519 static int fib6_dump_done(struct netlink_callback *cb)
520 {
521 	fib6_dump_end(cb);
522 	return cb->done ? cb->done(cb) : 0;
523 }
524 
525 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
526 			   struct netlink_callback *cb)
527 {
528 	struct net *net = sock_net(skb->sk);
529 	struct fib6_walker *w;
530 	int res;
531 
532 	w = (void *)cb->args[2];
533 	w->root = &table->tb6_root;
534 
535 	if (cb->args[4] == 0) {
536 		w->count = 0;
537 		w->skip = 0;
538 
539 		spin_lock_bh(&table->tb6_lock);
540 		res = fib6_walk(net, w);
541 		spin_unlock_bh(&table->tb6_lock);
542 		if (res > 0) {
543 			cb->args[4] = 1;
544 			cb->args[5] = w->root->fn_sernum;
545 		}
546 	} else {
547 		if (cb->args[5] != w->root->fn_sernum) {
548 			/* Begin at the root if the tree changed */
549 			cb->args[5] = w->root->fn_sernum;
550 			w->state = FWS_INIT;
551 			w->node = w->root;
552 			w->skip = w->count;
553 		} else
554 			w->skip = 0;
555 
556 		spin_lock_bh(&table->tb6_lock);
557 		res = fib6_walk_continue(w);
558 		spin_unlock_bh(&table->tb6_lock);
559 		if (res <= 0) {
560 			fib6_walker_unlink(net, w);
561 			cb->args[4] = 0;
562 		}
563 	}
564 
565 	return res;
566 }
567 
568 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
569 {
570 	const struct nlmsghdr *nlh = cb->nlh;
571 	struct net *net = sock_net(skb->sk);
572 	struct rt6_rtnl_dump_arg arg = {};
573 	unsigned int h, s_h;
574 	unsigned int e = 0, s_e;
575 	struct fib6_walker *w;
576 	struct fib6_table *tb;
577 	struct hlist_head *head;
578 	int res = 0;
579 
580 	if (cb->strict_check) {
581 		int err;
582 
583 		err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
584 		if (err < 0)
585 			return err;
586 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
587 		struct rtmsg *rtm = nlmsg_data(nlh);
588 
589 		arg.filter.flags = rtm->rtm_flags & (RTM_F_PREFIX|RTM_F_CLONED);
590 	}
591 
592 	/* fib entries are never clones */
593 	if (arg.filter.flags & RTM_F_CLONED)
594 		return skb->len;
595 
596 	w = (void *)cb->args[2];
597 	if (!w) {
598 		/* New dump:
599 		 *
600 		 * 1. hook callback destructor.
601 		 */
602 		cb->args[3] = (long)cb->done;
603 		cb->done = fib6_dump_done;
604 
605 		/*
606 		 * 2. allocate and initialize walker.
607 		 */
608 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
609 		if (!w)
610 			return -ENOMEM;
611 		w->func = fib6_dump_node;
612 		cb->args[2] = (long)w;
613 	}
614 
615 	arg.skb = skb;
616 	arg.cb = cb;
617 	arg.net = net;
618 	w->args = &arg;
619 
620 	if (arg.filter.table_id) {
621 		tb = fib6_get_table(net, arg.filter.table_id);
622 		if (!tb) {
623 			if (arg.filter.dump_all_families)
624 				return skb->len;
625 
626 			NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
627 			return -ENOENT;
628 		}
629 
630 		res = fib6_dump_table(tb, skb, cb);
631 		goto out;
632 	}
633 
634 	s_h = cb->args[0];
635 	s_e = cb->args[1];
636 
637 	rcu_read_lock();
638 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
639 		e = 0;
640 		head = &net->ipv6.fib_table_hash[h];
641 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
642 			if (e < s_e)
643 				goto next;
644 			res = fib6_dump_table(tb, skb, cb);
645 			if (res != 0)
646 				goto out_unlock;
647 next:
648 			e++;
649 		}
650 	}
651 out_unlock:
652 	rcu_read_unlock();
653 	cb->args[1] = e;
654 	cb->args[0] = h;
655 out:
656 	res = res < 0 ? res : skb->len;
657 	if (res <= 0)
658 		fib6_dump_end(cb);
659 	return res;
660 }
661 
662 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
663 {
664 	if (!f6i)
665 		return;
666 
667 	if (f6i->fib6_metrics == &dst_default_metrics) {
668 		struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
669 
670 		if (!p)
671 			return;
672 
673 		refcount_set(&p->refcnt, 1);
674 		f6i->fib6_metrics = p;
675 	}
676 
677 	f6i->fib6_metrics->metrics[metric - 1] = val;
678 }
679 
680 /*
681  *	Routing Table
682  *
683  *	return the appropriate node for a routing tree "add" operation
684  *	by either creating and inserting or by returning an existing
685  *	node.
686  */
687 
688 static struct fib6_node *fib6_add_1(struct net *net,
689 				    struct fib6_table *table,
690 				    struct fib6_node *root,
691 				    struct in6_addr *addr, int plen,
692 				    int offset, int allow_create,
693 				    int replace_required,
694 				    struct netlink_ext_ack *extack)
695 {
696 	struct fib6_node *fn, *in, *ln;
697 	struct fib6_node *pn = NULL;
698 	struct rt6key *key;
699 	int	bit;
700 	__be32	dir = 0;
701 
702 	RT6_TRACE("fib6_add_1\n");
703 
704 	/* insert node in tree */
705 
706 	fn = root;
707 
708 	do {
709 		struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
710 					    lockdep_is_held(&table->tb6_lock));
711 		key = (struct rt6key *)((u8 *)leaf + offset);
712 
713 		/*
714 		 *	Prefix match
715 		 */
716 		if (plen < fn->fn_bit ||
717 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
718 			if (!allow_create) {
719 				if (replace_required) {
720 					NL_SET_ERR_MSG(extack,
721 						       "Can not replace route - no match found");
722 					pr_warn("Can't replace route, no match found\n");
723 					return ERR_PTR(-ENOENT);
724 				}
725 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
726 			}
727 			goto insert_above;
728 		}
729 
730 		/*
731 		 *	Exact match ?
732 		 */
733 
734 		if (plen == fn->fn_bit) {
735 			/* clean up an intermediate node */
736 			if (!(fn->fn_flags & RTN_RTINFO)) {
737 				RCU_INIT_POINTER(fn->leaf, NULL);
738 				fib6_info_release(leaf);
739 			/* remove null_entry in the root node */
740 			} else if (fn->fn_flags & RTN_TL_ROOT &&
741 				   rcu_access_pointer(fn->leaf) ==
742 				   net->ipv6.fib6_null_entry) {
743 				RCU_INIT_POINTER(fn->leaf, NULL);
744 			}
745 
746 			return fn;
747 		}
748 
749 		/*
750 		 *	We have more bits to go
751 		 */
752 
753 		/* Try to walk down on tree. */
754 		dir = addr_bit_set(addr, fn->fn_bit);
755 		pn = fn;
756 		fn = dir ?
757 		     rcu_dereference_protected(fn->right,
758 					lockdep_is_held(&table->tb6_lock)) :
759 		     rcu_dereference_protected(fn->left,
760 					lockdep_is_held(&table->tb6_lock));
761 	} while (fn);
762 
763 	if (!allow_create) {
764 		/* We should not create new node because
765 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
766 		 * I assume it is safe to require NLM_F_CREATE when
767 		 * REPLACE flag is used! Later we may want to remove the
768 		 * check for replace_required, because according
769 		 * to netlink specification, NLM_F_CREATE
770 		 * MUST be specified if new route is created.
771 		 * That would keep IPv6 consistent with IPv4
772 		 */
773 		if (replace_required) {
774 			NL_SET_ERR_MSG(extack,
775 				       "Can not replace route - no match found");
776 			pr_warn("Can't replace route, no match found\n");
777 			return ERR_PTR(-ENOENT);
778 		}
779 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
780 	}
781 	/*
782 	 *	We walked to the bottom of tree.
783 	 *	Create new leaf node without children.
784 	 */
785 
786 	ln = node_alloc(net);
787 
788 	if (!ln)
789 		return ERR_PTR(-ENOMEM);
790 	ln->fn_bit = plen;
791 	RCU_INIT_POINTER(ln->parent, pn);
792 
793 	if (dir)
794 		rcu_assign_pointer(pn->right, ln);
795 	else
796 		rcu_assign_pointer(pn->left, ln);
797 
798 	return ln;
799 
800 
801 insert_above:
802 	/*
803 	 * split since we don't have a common prefix anymore or
804 	 * we have a less significant route.
805 	 * we've to insert an intermediate node on the list
806 	 * this new node will point to the one we need to create
807 	 * and the current
808 	 */
809 
810 	pn = rcu_dereference_protected(fn->parent,
811 				       lockdep_is_held(&table->tb6_lock));
812 
813 	/* find 1st bit in difference between the 2 addrs.
814 
815 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
816 	   but if it is >= plen, the value is ignored in any case.
817 	 */
818 
819 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
820 
821 	/*
822 	 *		(intermediate)[in]
823 	 *	          /	   \
824 	 *	(new leaf node)[ln] (old node)[fn]
825 	 */
826 	if (plen > bit) {
827 		in = node_alloc(net);
828 		ln = node_alloc(net);
829 
830 		if (!in || !ln) {
831 			if (in)
832 				node_free_immediate(net, in);
833 			if (ln)
834 				node_free_immediate(net, ln);
835 			return ERR_PTR(-ENOMEM);
836 		}
837 
838 		/*
839 		 * new intermediate node.
840 		 * RTN_RTINFO will
841 		 * be off since that an address that chooses one of
842 		 * the branches would not match less specific routes
843 		 * in the other branch
844 		 */
845 
846 		in->fn_bit = bit;
847 
848 		RCU_INIT_POINTER(in->parent, pn);
849 		in->leaf = fn->leaf;
850 		atomic_inc(&rcu_dereference_protected(in->leaf,
851 				lockdep_is_held(&table->tb6_lock))->fib6_ref);
852 
853 		/* update parent pointer */
854 		if (dir)
855 			rcu_assign_pointer(pn->right, in);
856 		else
857 			rcu_assign_pointer(pn->left, in);
858 
859 		ln->fn_bit = plen;
860 
861 		RCU_INIT_POINTER(ln->parent, in);
862 		rcu_assign_pointer(fn->parent, in);
863 
864 		if (addr_bit_set(addr, bit)) {
865 			rcu_assign_pointer(in->right, ln);
866 			rcu_assign_pointer(in->left, fn);
867 		} else {
868 			rcu_assign_pointer(in->left, ln);
869 			rcu_assign_pointer(in->right, fn);
870 		}
871 	} else { /* plen <= bit */
872 
873 		/*
874 		 *		(new leaf node)[ln]
875 		 *	          /	   \
876 		 *	     (old node)[fn] NULL
877 		 */
878 
879 		ln = node_alloc(net);
880 
881 		if (!ln)
882 			return ERR_PTR(-ENOMEM);
883 
884 		ln->fn_bit = plen;
885 
886 		RCU_INIT_POINTER(ln->parent, pn);
887 
888 		if (addr_bit_set(&key->addr, plen))
889 			RCU_INIT_POINTER(ln->right, fn);
890 		else
891 			RCU_INIT_POINTER(ln->left, fn);
892 
893 		rcu_assign_pointer(fn->parent, ln);
894 
895 		if (dir)
896 			rcu_assign_pointer(pn->right, ln);
897 		else
898 			rcu_assign_pointer(pn->left, ln);
899 	}
900 	return ln;
901 }
902 
903 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
904 				const struct fib6_table *table)
905 {
906 	int cpu;
907 
908 	/* release the reference to this fib entry from
909 	 * all of its cached pcpu routes
910 	 */
911 	for_each_possible_cpu(cpu) {
912 		struct rt6_info **ppcpu_rt;
913 		struct rt6_info *pcpu_rt;
914 
915 		ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
916 		pcpu_rt = *ppcpu_rt;
917 		if (pcpu_rt) {
918 			struct fib6_info *from;
919 
920 			from = rcu_dereference_protected(pcpu_rt->from,
921 					     lockdep_is_held(&table->tb6_lock));
922 			rcu_assign_pointer(pcpu_rt->from, NULL);
923 			fib6_info_release(from);
924 		}
925 	}
926 }
927 
928 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
929 			  struct net *net)
930 {
931 	struct fib6_table *table = rt->fib6_table;
932 
933 	if (atomic_read(&rt->fib6_ref) != 1) {
934 		/* This route is used as dummy address holder in some split
935 		 * nodes. It is not leaked, but it still holds other resources,
936 		 * which must be released in time. So, scan ascendant nodes
937 		 * and replace dummy references to this route with references
938 		 * to still alive ones.
939 		 */
940 		while (fn) {
941 			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
942 					    lockdep_is_held(&table->tb6_lock));
943 			struct fib6_info *new_leaf;
944 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
945 				new_leaf = fib6_find_prefix(net, table, fn);
946 				atomic_inc(&new_leaf->fib6_ref);
947 
948 				rcu_assign_pointer(fn->leaf, new_leaf);
949 				fib6_info_release(rt);
950 			}
951 			fn = rcu_dereference_protected(fn->parent,
952 				    lockdep_is_held(&table->tb6_lock));
953 		}
954 
955 		if (rt->rt6i_pcpu)
956 			fib6_drop_pcpu_from(rt, table);
957 	}
958 }
959 
960 /*
961  *	Insert routing information in a node.
962  */
963 
964 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
965 			    struct nl_info *info,
966 			    struct netlink_ext_ack *extack)
967 {
968 	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
969 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
970 	struct fib6_info *iter = NULL;
971 	struct fib6_info __rcu **ins;
972 	struct fib6_info __rcu **fallback_ins = NULL;
973 	int replace = (info->nlh &&
974 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
975 	int add = (!info->nlh ||
976 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
977 	int found = 0;
978 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
979 	u16 nlflags = NLM_F_EXCL;
980 	int err;
981 
982 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
983 		nlflags |= NLM_F_APPEND;
984 
985 	ins = &fn->leaf;
986 
987 	for (iter = leaf; iter;
988 	     iter = rcu_dereference_protected(iter->fib6_next,
989 				lockdep_is_held(&rt->fib6_table->tb6_lock))) {
990 		/*
991 		 *	Search for duplicates
992 		 */
993 
994 		if (iter->fib6_metric == rt->fib6_metric) {
995 			/*
996 			 *	Same priority level
997 			 */
998 			if (info->nlh &&
999 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
1000 				return -EEXIST;
1001 
1002 			nlflags &= ~NLM_F_EXCL;
1003 			if (replace) {
1004 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1005 					found++;
1006 					break;
1007 				}
1008 				if (rt_can_ecmp)
1009 					fallback_ins = fallback_ins ?: ins;
1010 				goto next_iter;
1011 			}
1012 
1013 			if (rt6_duplicate_nexthop(iter, rt)) {
1014 				if (rt->fib6_nsiblings)
1015 					rt->fib6_nsiblings = 0;
1016 				if (!(iter->fib6_flags & RTF_EXPIRES))
1017 					return -EEXIST;
1018 				if (!(rt->fib6_flags & RTF_EXPIRES))
1019 					fib6_clean_expires(iter);
1020 				else
1021 					fib6_set_expires(iter, rt->expires);
1022 
1023 				if (rt->fib6_pmtu)
1024 					fib6_metric_set(iter, RTAX_MTU,
1025 							rt->fib6_pmtu);
1026 				return -EEXIST;
1027 			}
1028 			/* If we have the same destination and the same metric,
1029 			 * but not the same gateway, then the route we try to
1030 			 * add is sibling to this route, increment our counter
1031 			 * of siblings, and later we will add our route to the
1032 			 * list.
1033 			 * Only static routes (which don't have flag
1034 			 * RTF_EXPIRES) are used for ECMPv6.
1035 			 *
1036 			 * To avoid long list, we only had siblings if the
1037 			 * route have a gateway.
1038 			 */
1039 			if (rt_can_ecmp &&
1040 			    rt6_qualify_for_ecmp(iter))
1041 				rt->fib6_nsiblings++;
1042 		}
1043 
1044 		if (iter->fib6_metric > rt->fib6_metric)
1045 			break;
1046 
1047 next_iter:
1048 		ins = &iter->fib6_next;
1049 	}
1050 
1051 	if (fallback_ins && !found) {
1052 		/* No ECMP-able route found, replace first non-ECMP one */
1053 		ins = fallback_ins;
1054 		iter = rcu_dereference_protected(*ins,
1055 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1056 		found++;
1057 	}
1058 
1059 	/* Reset round-robin state, if necessary */
1060 	if (ins == &fn->leaf)
1061 		fn->rr_ptr = NULL;
1062 
1063 	/* Link this route to others same route. */
1064 	if (rt->fib6_nsiblings) {
1065 		unsigned int fib6_nsiblings;
1066 		struct fib6_info *sibling, *temp_sibling;
1067 
1068 		/* Find the first route that have the same metric */
1069 		sibling = leaf;
1070 		while (sibling) {
1071 			if (sibling->fib6_metric == rt->fib6_metric &&
1072 			    rt6_qualify_for_ecmp(sibling)) {
1073 				list_add_tail(&rt->fib6_siblings,
1074 					      &sibling->fib6_siblings);
1075 				break;
1076 			}
1077 			sibling = rcu_dereference_protected(sibling->fib6_next,
1078 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1079 		}
1080 		/* For each sibling in the list, increment the counter of
1081 		 * siblings. BUG() if counters does not match, list of siblings
1082 		 * is broken!
1083 		 */
1084 		fib6_nsiblings = 0;
1085 		list_for_each_entry_safe(sibling, temp_sibling,
1086 					 &rt->fib6_siblings, fib6_siblings) {
1087 			sibling->fib6_nsiblings++;
1088 			BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1089 			fib6_nsiblings++;
1090 		}
1091 		BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1092 		rt6_multipath_rebalance(temp_sibling);
1093 	}
1094 
1095 	/*
1096 	 *	insert node
1097 	 */
1098 	if (!replace) {
1099 		if (!add)
1100 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1101 
1102 add:
1103 		nlflags |= NLM_F_CREATE;
1104 
1105 		err = call_fib6_entry_notifiers(info->nl_net,
1106 						FIB_EVENT_ENTRY_ADD,
1107 						rt, extack);
1108 		if (err)
1109 			return err;
1110 
1111 		rcu_assign_pointer(rt->fib6_next, iter);
1112 		atomic_inc(&rt->fib6_ref);
1113 		rcu_assign_pointer(rt->fib6_node, fn);
1114 		rcu_assign_pointer(*ins, rt);
1115 		if (!info->skip_notify)
1116 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1117 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1118 
1119 		if (!(fn->fn_flags & RTN_RTINFO)) {
1120 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1121 			fn->fn_flags |= RTN_RTINFO;
1122 		}
1123 
1124 	} else {
1125 		int nsiblings;
1126 
1127 		if (!found) {
1128 			if (add)
1129 				goto add;
1130 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1131 			return -ENOENT;
1132 		}
1133 
1134 		err = call_fib6_entry_notifiers(info->nl_net,
1135 						FIB_EVENT_ENTRY_REPLACE,
1136 						rt, extack);
1137 		if (err)
1138 			return err;
1139 
1140 		atomic_inc(&rt->fib6_ref);
1141 		rcu_assign_pointer(rt->fib6_node, fn);
1142 		rt->fib6_next = iter->fib6_next;
1143 		rcu_assign_pointer(*ins, rt);
1144 		if (!info->skip_notify)
1145 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1146 		if (!(fn->fn_flags & RTN_RTINFO)) {
1147 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1148 			fn->fn_flags |= RTN_RTINFO;
1149 		}
1150 		nsiblings = iter->fib6_nsiblings;
1151 		iter->fib6_node = NULL;
1152 		fib6_purge_rt(iter, fn, info->nl_net);
1153 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1154 			fn->rr_ptr = NULL;
1155 		fib6_info_release(iter);
1156 
1157 		if (nsiblings) {
1158 			/* Replacing an ECMP route, remove all siblings */
1159 			ins = &rt->fib6_next;
1160 			iter = rcu_dereference_protected(*ins,
1161 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1162 			while (iter) {
1163 				if (iter->fib6_metric > rt->fib6_metric)
1164 					break;
1165 				if (rt6_qualify_for_ecmp(iter)) {
1166 					*ins = iter->fib6_next;
1167 					iter->fib6_node = NULL;
1168 					fib6_purge_rt(iter, fn, info->nl_net);
1169 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1170 						fn->rr_ptr = NULL;
1171 					fib6_info_release(iter);
1172 					nsiblings--;
1173 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1174 				} else {
1175 					ins = &iter->fib6_next;
1176 				}
1177 				iter = rcu_dereference_protected(*ins,
1178 					lockdep_is_held(&rt->fib6_table->tb6_lock));
1179 			}
1180 			WARN_ON(nsiblings != 0);
1181 		}
1182 	}
1183 
1184 	return 0;
1185 }
1186 
1187 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1188 {
1189 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1190 	    (rt->fib6_flags & RTF_EXPIRES))
1191 		mod_timer(&net->ipv6.ip6_fib_timer,
1192 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1193 }
1194 
1195 void fib6_force_start_gc(struct net *net)
1196 {
1197 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1198 		mod_timer(&net->ipv6.ip6_fib_timer,
1199 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1200 }
1201 
1202 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1203 					   int sernum)
1204 {
1205 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1206 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1207 
1208 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1209 	smp_wmb();
1210 	while (fn) {
1211 		fn->fn_sernum = sernum;
1212 		fn = rcu_dereference_protected(fn->parent,
1213 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1214 	}
1215 }
1216 
1217 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1218 {
1219 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1220 }
1221 
1222 /*
1223  *	Add routing information to the routing tree.
1224  *	<destination addr>/<source addr>
1225  *	with source addr info in sub-trees
1226  *	Need to own table->tb6_lock
1227  */
1228 
1229 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1230 	     struct nl_info *info, struct netlink_ext_ack *extack)
1231 {
1232 	struct fib6_table *table = rt->fib6_table;
1233 	struct fib6_node *fn, *pn = NULL;
1234 	int err = -ENOMEM;
1235 	int allow_create = 1;
1236 	int replace_required = 0;
1237 	int sernum = fib6_new_sernum(info->nl_net);
1238 
1239 	if (info->nlh) {
1240 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1241 			allow_create = 0;
1242 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1243 			replace_required = 1;
1244 	}
1245 	if (!allow_create && !replace_required)
1246 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1247 
1248 	fn = fib6_add_1(info->nl_net, table, root,
1249 			&rt->fib6_dst.addr, rt->fib6_dst.plen,
1250 			offsetof(struct fib6_info, fib6_dst), allow_create,
1251 			replace_required, extack);
1252 	if (IS_ERR(fn)) {
1253 		err = PTR_ERR(fn);
1254 		fn = NULL;
1255 		goto out;
1256 	}
1257 
1258 	pn = fn;
1259 
1260 #ifdef CONFIG_IPV6_SUBTREES
1261 	if (rt->fib6_src.plen) {
1262 		struct fib6_node *sn;
1263 
1264 		if (!rcu_access_pointer(fn->subtree)) {
1265 			struct fib6_node *sfn;
1266 
1267 			/*
1268 			 * Create subtree.
1269 			 *
1270 			 *		fn[main tree]
1271 			 *		|
1272 			 *		sfn[subtree root]
1273 			 *		   \
1274 			 *		    sn[new leaf node]
1275 			 */
1276 
1277 			/* Create subtree root node */
1278 			sfn = node_alloc(info->nl_net);
1279 			if (!sfn)
1280 				goto failure;
1281 
1282 			atomic_inc(&info->nl_net->ipv6.fib6_null_entry->fib6_ref);
1283 			rcu_assign_pointer(sfn->leaf,
1284 					   info->nl_net->ipv6.fib6_null_entry);
1285 			sfn->fn_flags = RTN_ROOT;
1286 
1287 			/* Now add the first leaf node to new subtree */
1288 
1289 			sn = fib6_add_1(info->nl_net, table, sfn,
1290 					&rt->fib6_src.addr, rt->fib6_src.plen,
1291 					offsetof(struct fib6_info, fib6_src),
1292 					allow_create, replace_required, extack);
1293 
1294 			if (IS_ERR(sn)) {
1295 				/* If it is failed, discard just allocated
1296 				   root, and then (in failure) stale node
1297 				   in main tree.
1298 				 */
1299 				node_free_immediate(info->nl_net, sfn);
1300 				err = PTR_ERR(sn);
1301 				goto failure;
1302 			}
1303 
1304 			/* Now link new subtree to main tree */
1305 			rcu_assign_pointer(sfn->parent, fn);
1306 			rcu_assign_pointer(fn->subtree, sfn);
1307 		} else {
1308 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1309 					&rt->fib6_src.addr, rt->fib6_src.plen,
1310 					offsetof(struct fib6_info, fib6_src),
1311 					allow_create, replace_required, extack);
1312 
1313 			if (IS_ERR(sn)) {
1314 				err = PTR_ERR(sn);
1315 				goto failure;
1316 			}
1317 		}
1318 
1319 		if (!rcu_access_pointer(fn->leaf)) {
1320 			if (fn->fn_flags & RTN_TL_ROOT) {
1321 				/* put back null_entry for root node */
1322 				rcu_assign_pointer(fn->leaf,
1323 					    info->nl_net->ipv6.fib6_null_entry);
1324 			} else {
1325 				atomic_inc(&rt->fib6_ref);
1326 				rcu_assign_pointer(fn->leaf, rt);
1327 			}
1328 		}
1329 		fn = sn;
1330 	}
1331 #endif
1332 
1333 	err = fib6_add_rt2node(fn, rt, info, extack);
1334 	if (!err) {
1335 		__fib6_update_sernum_upto_root(rt, sernum);
1336 		fib6_start_gc(info->nl_net, rt);
1337 	}
1338 
1339 out:
1340 	if (err) {
1341 #ifdef CONFIG_IPV6_SUBTREES
1342 		/*
1343 		 * If fib6_add_1 has cleared the old leaf pointer in the
1344 		 * super-tree leaf node we have to find a new one for it.
1345 		 */
1346 		if (pn != fn) {
1347 			struct fib6_info *pn_leaf =
1348 				rcu_dereference_protected(pn->leaf,
1349 				    lockdep_is_held(&table->tb6_lock));
1350 			if (pn_leaf == rt) {
1351 				pn_leaf = NULL;
1352 				RCU_INIT_POINTER(pn->leaf, NULL);
1353 				fib6_info_release(rt);
1354 			}
1355 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1356 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1357 							   pn);
1358 #if RT6_DEBUG >= 2
1359 				if (!pn_leaf) {
1360 					WARN_ON(!pn_leaf);
1361 					pn_leaf =
1362 					    info->nl_net->ipv6.fib6_null_entry;
1363 				}
1364 #endif
1365 				fib6_info_hold(pn_leaf);
1366 				rcu_assign_pointer(pn->leaf, pn_leaf);
1367 			}
1368 		}
1369 #endif
1370 		goto failure;
1371 	}
1372 	return err;
1373 
1374 failure:
1375 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1376 	 * 1. fn is an intermediate node and we failed to add the new
1377 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1378 	 * failure case.
1379 	 * 2. fn is the root node in the table and we fail to add the first
1380 	 * default route to it.
1381 	 */
1382 	if (fn &&
1383 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1384 	     (fn->fn_flags & RTN_TL_ROOT &&
1385 	      !rcu_access_pointer(fn->leaf))))
1386 		fib6_repair_tree(info->nl_net, table, fn);
1387 	return err;
1388 }
1389 
1390 /*
1391  *	Routing tree lookup
1392  *
1393  */
1394 
1395 struct lookup_args {
1396 	int			offset;		/* key offset on fib6_info */
1397 	const struct in6_addr	*addr;		/* search key			*/
1398 };
1399 
1400 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1401 					    struct lookup_args *args)
1402 {
1403 	struct fib6_node *fn;
1404 	__be32 dir;
1405 
1406 	if (unlikely(args->offset == 0))
1407 		return NULL;
1408 
1409 	/*
1410 	 *	Descend on a tree
1411 	 */
1412 
1413 	fn = root;
1414 
1415 	for (;;) {
1416 		struct fib6_node *next;
1417 
1418 		dir = addr_bit_set(args->addr, fn->fn_bit);
1419 
1420 		next = dir ? rcu_dereference(fn->right) :
1421 			     rcu_dereference(fn->left);
1422 
1423 		if (next) {
1424 			fn = next;
1425 			continue;
1426 		}
1427 		break;
1428 	}
1429 
1430 	while (fn) {
1431 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1432 
1433 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1434 			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1435 			struct rt6key *key;
1436 
1437 			if (!leaf)
1438 				goto backtrack;
1439 
1440 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1441 
1442 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1443 #ifdef CONFIG_IPV6_SUBTREES
1444 				if (subtree) {
1445 					struct fib6_node *sfn;
1446 					sfn = fib6_node_lookup_1(subtree,
1447 								 args + 1);
1448 					if (!sfn)
1449 						goto backtrack;
1450 					fn = sfn;
1451 				}
1452 #endif
1453 				if (fn->fn_flags & RTN_RTINFO)
1454 					return fn;
1455 			}
1456 		}
1457 backtrack:
1458 		if (fn->fn_flags & RTN_ROOT)
1459 			break;
1460 
1461 		fn = rcu_dereference(fn->parent);
1462 	}
1463 
1464 	return NULL;
1465 }
1466 
1467 /* called with rcu_read_lock() held
1468  */
1469 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1470 				   const struct in6_addr *daddr,
1471 				   const struct in6_addr *saddr)
1472 {
1473 	struct fib6_node *fn;
1474 	struct lookup_args args[] = {
1475 		{
1476 			.offset = offsetof(struct fib6_info, fib6_dst),
1477 			.addr = daddr,
1478 		},
1479 #ifdef CONFIG_IPV6_SUBTREES
1480 		{
1481 			.offset = offsetof(struct fib6_info, fib6_src),
1482 			.addr = saddr,
1483 		},
1484 #endif
1485 		{
1486 			.offset = 0,	/* sentinel */
1487 		}
1488 	};
1489 
1490 	fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1491 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1492 		fn = root;
1493 
1494 	return fn;
1495 }
1496 
1497 /*
1498  *	Get node with specified destination prefix (and source prefix,
1499  *	if subtrees are used)
1500  *	exact_match == true means we try to find fn with exact match of
1501  *	the passed in prefix addr
1502  *	exact_match == false means we try to find fn with longest prefix
1503  *	match of the passed in prefix addr. This is useful for finding fn
1504  *	for cached route as it will be stored in the exception table under
1505  *	the node with longest prefix length.
1506  */
1507 
1508 
1509 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1510 				       const struct in6_addr *addr,
1511 				       int plen, int offset,
1512 				       bool exact_match)
1513 {
1514 	struct fib6_node *fn, *prev = NULL;
1515 
1516 	for (fn = root; fn ; ) {
1517 		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1518 		struct rt6key *key;
1519 
1520 		/* This node is being deleted */
1521 		if (!leaf) {
1522 			if (plen <= fn->fn_bit)
1523 				goto out;
1524 			else
1525 				goto next;
1526 		}
1527 
1528 		key = (struct rt6key *)((u8 *)leaf + offset);
1529 
1530 		/*
1531 		 *	Prefix match
1532 		 */
1533 		if (plen < fn->fn_bit ||
1534 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1535 			goto out;
1536 
1537 		if (plen == fn->fn_bit)
1538 			return fn;
1539 
1540 		prev = fn;
1541 
1542 next:
1543 		/*
1544 		 *	We have more bits to go
1545 		 */
1546 		if (addr_bit_set(addr, fn->fn_bit))
1547 			fn = rcu_dereference(fn->right);
1548 		else
1549 			fn = rcu_dereference(fn->left);
1550 	}
1551 out:
1552 	if (exact_match)
1553 		return NULL;
1554 	else
1555 		return prev;
1556 }
1557 
1558 struct fib6_node *fib6_locate(struct fib6_node *root,
1559 			      const struct in6_addr *daddr, int dst_len,
1560 			      const struct in6_addr *saddr, int src_len,
1561 			      bool exact_match)
1562 {
1563 	struct fib6_node *fn;
1564 
1565 	fn = fib6_locate_1(root, daddr, dst_len,
1566 			   offsetof(struct fib6_info, fib6_dst),
1567 			   exact_match);
1568 
1569 #ifdef CONFIG_IPV6_SUBTREES
1570 	if (src_len) {
1571 		WARN_ON(saddr == NULL);
1572 		if (fn) {
1573 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1574 
1575 			if (subtree) {
1576 				fn = fib6_locate_1(subtree, saddr, src_len,
1577 					   offsetof(struct fib6_info, fib6_src),
1578 					   exact_match);
1579 			}
1580 		}
1581 	}
1582 #endif
1583 
1584 	if (fn && fn->fn_flags & RTN_RTINFO)
1585 		return fn;
1586 
1587 	return NULL;
1588 }
1589 
1590 
1591 /*
1592  *	Deletion
1593  *
1594  */
1595 
1596 static struct fib6_info *fib6_find_prefix(struct net *net,
1597 					 struct fib6_table *table,
1598 					 struct fib6_node *fn)
1599 {
1600 	struct fib6_node *child_left, *child_right;
1601 
1602 	if (fn->fn_flags & RTN_ROOT)
1603 		return net->ipv6.fib6_null_entry;
1604 
1605 	while (fn) {
1606 		child_left = rcu_dereference_protected(fn->left,
1607 				    lockdep_is_held(&table->tb6_lock));
1608 		child_right = rcu_dereference_protected(fn->right,
1609 				    lockdep_is_held(&table->tb6_lock));
1610 		if (child_left)
1611 			return rcu_dereference_protected(child_left->leaf,
1612 					lockdep_is_held(&table->tb6_lock));
1613 		if (child_right)
1614 			return rcu_dereference_protected(child_right->leaf,
1615 					lockdep_is_held(&table->tb6_lock));
1616 
1617 		fn = FIB6_SUBTREE(fn);
1618 	}
1619 	return NULL;
1620 }
1621 
1622 /*
1623  *	Called to trim the tree of intermediate nodes when possible. "fn"
1624  *	is the node we want to try and remove.
1625  *	Need to own table->tb6_lock
1626  */
1627 
1628 static struct fib6_node *fib6_repair_tree(struct net *net,
1629 					  struct fib6_table *table,
1630 					  struct fib6_node *fn)
1631 {
1632 	int children;
1633 	int nstate;
1634 	struct fib6_node *child;
1635 	struct fib6_walker *w;
1636 	int iter = 0;
1637 
1638 	/* Set fn->leaf to null_entry for root node. */
1639 	if (fn->fn_flags & RTN_TL_ROOT) {
1640 		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1641 		return fn;
1642 	}
1643 
1644 	for (;;) {
1645 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1646 					    lockdep_is_held(&table->tb6_lock));
1647 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1648 					    lockdep_is_held(&table->tb6_lock));
1649 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1650 					    lockdep_is_held(&table->tb6_lock));
1651 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1652 					    lockdep_is_held(&table->tb6_lock));
1653 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1654 					    lockdep_is_held(&table->tb6_lock));
1655 		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1656 					    lockdep_is_held(&table->tb6_lock));
1657 		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1658 					    lockdep_is_held(&table->tb6_lock));
1659 		struct fib6_info *new_fn_leaf;
1660 
1661 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1662 		iter++;
1663 
1664 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1665 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1666 		WARN_ON(fn_leaf);
1667 
1668 		children = 0;
1669 		child = NULL;
1670 		if (fn_r)
1671 			child = fn_r, children |= 1;
1672 		if (fn_l)
1673 			child = fn_l, children |= 2;
1674 
1675 		if (children == 3 || FIB6_SUBTREE(fn)
1676 #ifdef CONFIG_IPV6_SUBTREES
1677 		    /* Subtree root (i.e. fn) may have one child */
1678 		    || (children && fn->fn_flags & RTN_ROOT)
1679 #endif
1680 		    ) {
1681 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1682 #if RT6_DEBUG >= 2
1683 			if (!new_fn_leaf) {
1684 				WARN_ON(!new_fn_leaf);
1685 				new_fn_leaf = net->ipv6.fib6_null_entry;
1686 			}
1687 #endif
1688 			fib6_info_hold(new_fn_leaf);
1689 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1690 			return pn;
1691 		}
1692 
1693 #ifdef CONFIG_IPV6_SUBTREES
1694 		if (FIB6_SUBTREE(pn) == fn) {
1695 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1696 			RCU_INIT_POINTER(pn->subtree, NULL);
1697 			nstate = FWS_L;
1698 		} else {
1699 			WARN_ON(fn->fn_flags & RTN_ROOT);
1700 #endif
1701 			if (pn_r == fn)
1702 				rcu_assign_pointer(pn->right, child);
1703 			else if (pn_l == fn)
1704 				rcu_assign_pointer(pn->left, child);
1705 #if RT6_DEBUG >= 2
1706 			else
1707 				WARN_ON(1);
1708 #endif
1709 			if (child)
1710 				rcu_assign_pointer(child->parent, pn);
1711 			nstate = FWS_R;
1712 #ifdef CONFIG_IPV6_SUBTREES
1713 		}
1714 #endif
1715 
1716 		read_lock(&net->ipv6.fib6_walker_lock);
1717 		FOR_WALKERS(net, w) {
1718 			if (!child) {
1719 				if (w->node == fn) {
1720 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1721 					w->node = pn;
1722 					w->state = nstate;
1723 				}
1724 			} else {
1725 				if (w->node == fn) {
1726 					w->node = child;
1727 					if (children&2) {
1728 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1729 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1730 					} else {
1731 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1732 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1733 					}
1734 				}
1735 			}
1736 		}
1737 		read_unlock(&net->ipv6.fib6_walker_lock);
1738 
1739 		node_free(net, fn);
1740 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1741 			return pn;
1742 
1743 		RCU_INIT_POINTER(pn->leaf, NULL);
1744 		fib6_info_release(pn_leaf);
1745 		fn = pn;
1746 	}
1747 }
1748 
1749 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1750 			   struct fib6_info __rcu **rtp, struct nl_info *info)
1751 {
1752 	struct fib6_walker *w;
1753 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1754 				    lockdep_is_held(&table->tb6_lock));
1755 	struct net *net = info->nl_net;
1756 
1757 	RT6_TRACE("fib6_del_route\n");
1758 
1759 	/* Unlink it */
1760 	*rtp = rt->fib6_next;
1761 	rt->fib6_node = NULL;
1762 	net->ipv6.rt6_stats->fib_rt_entries--;
1763 	net->ipv6.rt6_stats->fib_discarded_routes++;
1764 
1765 	/* Flush all cached dst in exception table */
1766 	rt6_flush_exceptions(rt);
1767 
1768 	/* Reset round-robin state, if necessary */
1769 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1770 		fn->rr_ptr = NULL;
1771 
1772 	/* Remove this entry from other siblings */
1773 	if (rt->fib6_nsiblings) {
1774 		struct fib6_info *sibling, *next_sibling;
1775 
1776 		list_for_each_entry_safe(sibling, next_sibling,
1777 					 &rt->fib6_siblings, fib6_siblings)
1778 			sibling->fib6_nsiblings--;
1779 		rt->fib6_nsiblings = 0;
1780 		list_del_init(&rt->fib6_siblings);
1781 		rt6_multipath_rebalance(next_sibling);
1782 	}
1783 
1784 	/* Adjust walkers */
1785 	read_lock(&net->ipv6.fib6_walker_lock);
1786 	FOR_WALKERS(net, w) {
1787 		if (w->state == FWS_C && w->leaf == rt) {
1788 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1789 			w->leaf = rcu_dereference_protected(rt->fib6_next,
1790 					    lockdep_is_held(&table->tb6_lock));
1791 			if (!w->leaf)
1792 				w->state = FWS_U;
1793 		}
1794 	}
1795 	read_unlock(&net->ipv6.fib6_walker_lock);
1796 
1797 	/* If it was last route, call fib6_repair_tree() to:
1798 	 * 1. For root node, put back null_entry as how the table was created.
1799 	 * 2. For other nodes, expunge its radix tree node.
1800 	 */
1801 	if (!rcu_access_pointer(fn->leaf)) {
1802 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1803 			fn->fn_flags &= ~RTN_RTINFO;
1804 			net->ipv6.rt6_stats->fib_route_nodes--;
1805 		}
1806 		fn = fib6_repair_tree(net, table, fn);
1807 	}
1808 
1809 	fib6_purge_rt(rt, fn, net);
1810 
1811 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1812 	if (!info->skip_notify)
1813 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1814 	fib6_info_release(rt);
1815 }
1816 
1817 /* Need to own table->tb6_lock */
1818 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1819 {
1820 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1821 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1822 	struct fib6_table *table = rt->fib6_table;
1823 	struct net *net = info->nl_net;
1824 	struct fib6_info __rcu **rtp;
1825 	struct fib6_info __rcu **rtp_next;
1826 
1827 	if (!fn || rt == net->ipv6.fib6_null_entry)
1828 		return -ENOENT;
1829 
1830 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1831 
1832 	/*
1833 	 *	Walk the leaf entries looking for ourself
1834 	 */
1835 
1836 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1837 		struct fib6_info *cur = rcu_dereference_protected(*rtp,
1838 					lockdep_is_held(&table->tb6_lock));
1839 		if (rt == cur) {
1840 			fib6_del_route(table, fn, rtp, info);
1841 			return 0;
1842 		}
1843 		rtp_next = &cur->fib6_next;
1844 	}
1845 	return -ENOENT;
1846 }
1847 
1848 /*
1849  *	Tree traversal function.
1850  *
1851  *	Certainly, it is not interrupt safe.
1852  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1853  *	It means, that we can modify tree during walking
1854  *	and use this function for garbage collection, clone pruning,
1855  *	cleaning tree when a device goes down etc. etc.
1856  *
1857  *	It guarantees that every node will be traversed,
1858  *	and that it will be traversed only once.
1859  *
1860  *	Callback function w->func may return:
1861  *	0 -> continue walking.
1862  *	positive value -> walking is suspended (used by tree dumps,
1863  *	and probably by gc, if it will be split to several slices)
1864  *	negative value -> terminate walking.
1865  *
1866  *	The function itself returns:
1867  *	0   -> walk is complete.
1868  *	>0  -> walk is incomplete (i.e. suspended)
1869  *	<0  -> walk is terminated by an error.
1870  *
1871  *	This function is called with tb6_lock held.
1872  */
1873 
1874 static int fib6_walk_continue(struct fib6_walker *w)
1875 {
1876 	struct fib6_node *fn, *pn, *left, *right;
1877 
1878 	/* w->root should always be table->tb6_root */
1879 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1880 
1881 	for (;;) {
1882 		fn = w->node;
1883 		if (!fn)
1884 			return 0;
1885 
1886 		switch (w->state) {
1887 #ifdef CONFIG_IPV6_SUBTREES
1888 		case FWS_S:
1889 			if (FIB6_SUBTREE(fn)) {
1890 				w->node = FIB6_SUBTREE(fn);
1891 				continue;
1892 			}
1893 			w->state = FWS_L;
1894 #endif
1895 			/* fall through */
1896 		case FWS_L:
1897 			left = rcu_dereference_protected(fn->left, 1);
1898 			if (left) {
1899 				w->node = left;
1900 				w->state = FWS_INIT;
1901 				continue;
1902 			}
1903 			w->state = FWS_R;
1904 			/* fall through */
1905 		case FWS_R:
1906 			right = rcu_dereference_protected(fn->right, 1);
1907 			if (right) {
1908 				w->node = right;
1909 				w->state = FWS_INIT;
1910 				continue;
1911 			}
1912 			w->state = FWS_C;
1913 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1914 			/* fall through */
1915 		case FWS_C:
1916 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1917 				int err;
1918 
1919 				if (w->skip) {
1920 					w->skip--;
1921 					goto skip;
1922 				}
1923 
1924 				err = w->func(w);
1925 				if (err)
1926 					return err;
1927 
1928 				w->count++;
1929 				continue;
1930 			}
1931 skip:
1932 			w->state = FWS_U;
1933 			/* fall through */
1934 		case FWS_U:
1935 			if (fn == w->root)
1936 				return 0;
1937 			pn = rcu_dereference_protected(fn->parent, 1);
1938 			left = rcu_dereference_protected(pn->left, 1);
1939 			right = rcu_dereference_protected(pn->right, 1);
1940 			w->node = pn;
1941 #ifdef CONFIG_IPV6_SUBTREES
1942 			if (FIB6_SUBTREE(pn) == fn) {
1943 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1944 				w->state = FWS_L;
1945 				continue;
1946 			}
1947 #endif
1948 			if (left == fn) {
1949 				w->state = FWS_R;
1950 				continue;
1951 			}
1952 			if (right == fn) {
1953 				w->state = FWS_C;
1954 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1955 				continue;
1956 			}
1957 #if RT6_DEBUG >= 2
1958 			WARN_ON(1);
1959 #endif
1960 		}
1961 	}
1962 }
1963 
1964 static int fib6_walk(struct net *net, struct fib6_walker *w)
1965 {
1966 	int res;
1967 
1968 	w->state = FWS_INIT;
1969 	w->node = w->root;
1970 
1971 	fib6_walker_link(net, w);
1972 	res = fib6_walk_continue(w);
1973 	if (res <= 0)
1974 		fib6_walker_unlink(net, w);
1975 	return res;
1976 }
1977 
1978 static int fib6_clean_node(struct fib6_walker *w)
1979 {
1980 	int res;
1981 	struct fib6_info *rt;
1982 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1983 	struct nl_info info = {
1984 		.nl_net = c->net,
1985 		.skip_notify = c->skip_notify,
1986 	};
1987 
1988 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1989 	    w->node->fn_sernum != c->sernum)
1990 		w->node->fn_sernum = c->sernum;
1991 
1992 	if (!c->func) {
1993 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1994 		w->leaf = NULL;
1995 		return 0;
1996 	}
1997 
1998 	for_each_fib6_walker_rt(w) {
1999 		res = c->func(rt, c->arg);
2000 		if (res == -1) {
2001 			w->leaf = rt;
2002 			res = fib6_del(rt, &info);
2003 			if (res) {
2004 #if RT6_DEBUG >= 2
2005 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2006 					 __func__, rt,
2007 					 rcu_access_pointer(rt->fib6_node),
2008 					 res);
2009 #endif
2010 				continue;
2011 			}
2012 			return 0;
2013 		} else if (res == -2) {
2014 			if (WARN_ON(!rt->fib6_nsiblings))
2015 				continue;
2016 			rt = list_last_entry(&rt->fib6_siblings,
2017 					     struct fib6_info, fib6_siblings);
2018 			continue;
2019 		}
2020 		WARN_ON(res != 0);
2021 	}
2022 	w->leaf = rt;
2023 	return 0;
2024 }
2025 
2026 /*
2027  *	Convenient frontend to tree walker.
2028  *
2029  *	func is called on each route.
2030  *		It may return -2 -> skip multipath route.
2031  *			      -1 -> delete this route.
2032  *		              0  -> continue walking
2033  */
2034 
2035 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2036 			    int (*func)(struct fib6_info *, void *arg),
2037 			    int sernum, void *arg, bool skip_notify)
2038 {
2039 	struct fib6_cleaner c;
2040 
2041 	c.w.root = root;
2042 	c.w.func = fib6_clean_node;
2043 	c.w.count = 0;
2044 	c.w.skip = 0;
2045 	c.func = func;
2046 	c.sernum = sernum;
2047 	c.arg = arg;
2048 	c.net = net;
2049 	c.skip_notify = skip_notify;
2050 
2051 	fib6_walk(net, &c.w);
2052 }
2053 
2054 static void __fib6_clean_all(struct net *net,
2055 			     int (*func)(struct fib6_info *, void *),
2056 			     int sernum, void *arg, bool skip_notify)
2057 {
2058 	struct fib6_table *table;
2059 	struct hlist_head *head;
2060 	unsigned int h;
2061 
2062 	rcu_read_lock();
2063 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2064 		head = &net->ipv6.fib_table_hash[h];
2065 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2066 			spin_lock_bh(&table->tb6_lock);
2067 			fib6_clean_tree(net, &table->tb6_root,
2068 					func, sernum, arg, skip_notify);
2069 			spin_unlock_bh(&table->tb6_lock);
2070 		}
2071 	}
2072 	rcu_read_unlock();
2073 }
2074 
2075 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2076 		    void *arg)
2077 {
2078 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2079 }
2080 
2081 void fib6_clean_all_skip_notify(struct net *net,
2082 				int (*func)(struct fib6_info *, void *),
2083 				void *arg)
2084 {
2085 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2086 }
2087 
2088 static void fib6_flush_trees(struct net *net)
2089 {
2090 	int new_sernum = fib6_new_sernum(net);
2091 
2092 	__fib6_clean_all(net, NULL, new_sernum, NULL, false);
2093 }
2094 
2095 /*
2096  *	Garbage collection
2097  */
2098 
2099 static int fib6_age(struct fib6_info *rt, void *arg)
2100 {
2101 	struct fib6_gc_args *gc_args = arg;
2102 	unsigned long now = jiffies;
2103 
2104 	/*
2105 	 *	check addrconf expiration here.
2106 	 *	Routes are expired even if they are in use.
2107 	 */
2108 
2109 	if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2110 		if (time_after(now, rt->expires)) {
2111 			RT6_TRACE("expiring %p\n", rt);
2112 			return -1;
2113 		}
2114 		gc_args->more++;
2115 	}
2116 
2117 	/*	Also age clones in the exception table.
2118 	 *	Note, that clones are aged out
2119 	 *	only if they are not in use now.
2120 	 */
2121 	rt6_age_exceptions(rt, gc_args, now);
2122 
2123 	return 0;
2124 }
2125 
2126 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2127 {
2128 	struct fib6_gc_args gc_args;
2129 	unsigned long now;
2130 
2131 	if (force) {
2132 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2133 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2134 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2135 		return;
2136 	}
2137 	gc_args.timeout = expires ? (int)expires :
2138 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2139 	gc_args.more = 0;
2140 
2141 	fib6_clean_all(net, fib6_age, &gc_args);
2142 	now = jiffies;
2143 	net->ipv6.ip6_rt_last_gc = now;
2144 
2145 	if (gc_args.more)
2146 		mod_timer(&net->ipv6.ip6_fib_timer,
2147 			  round_jiffies(now
2148 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2149 	else
2150 		del_timer(&net->ipv6.ip6_fib_timer);
2151 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2152 }
2153 
2154 static void fib6_gc_timer_cb(struct timer_list *t)
2155 {
2156 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2157 
2158 	fib6_run_gc(0, arg, true);
2159 }
2160 
2161 static int __net_init fib6_net_init(struct net *net)
2162 {
2163 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2164 	int err;
2165 
2166 	err = fib6_notifier_init(net);
2167 	if (err)
2168 		return err;
2169 
2170 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2171 	rwlock_init(&net->ipv6.fib6_walker_lock);
2172 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2173 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2174 
2175 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2176 	if (!net->ipv6.rt6_stats)
2177 		goto out_timer;
2178 
2179 	/* Avoid false sharing : Use at least a full cache line */
2180 	size = max_t(size_t, size, L1_CACHE_BYTES);
2181 
2182 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2183 	if (!net->ipv6.fib_table_hash)
2184 		goto out_rt6_stats;
2185 
2186 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2187 					  GFP_KERNEL);
2188 	if (!net->ipv6.fib6_main_tbl)
2189 		goto out_fib_table_hash;
2190 
2191 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2192 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2193 			   net->ipv6.fib6_null_entry);
2194 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2195 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2196 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2197 
2198 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2199 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2200 					   GFP_KERNEL);
2201 	if (!net->ipv6.fib6_local_tbl)
2202 		goto out_fib6_main_tbl;
2203 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2204 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2205 			   net->ipv6.fib6_null_entry);
2206 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2207 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2208 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2209 #endif
2210 	fib6_tables_init(net);
2211 
2212 	return 0;
2213 
2214 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2215 out_fib6_main_tbl:
2216 	kfree(net->ipv6.fib6_main_tbl);
2217 #endif
2218 out_fib_table_hash:
2219 	kfree(net->ipv6.fib_table_hash);
2220 out_rt6_stats:
2221 	kfree(net->ipv6.rt6_stats);
2222 out_timer:
2223 	fib6_notifier_exit(net);
2224 	return -ENOMEM;
2225 }
2226 
2227 static void fib6_net_exit(struct net *net)
2228 {
2229 	unsigned int i;
2230 
2231 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2232 
2233 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2234 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2235 		struct hlist_node *tmp;
2236 		struct fib6_table *tb;
2237 
2238 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2239 			hlist_del(&tb->tb6_hlist);
2240 			fib6_free_table(tb);
2241 		}
2242 	}
2243 
2244 	kfree(net->ipv6.fib_table_hash);
2245 	kfree(net->ipv6.rt6_stats);
2246 	fib6_notifier_exit(net);
2247 }
2248 
2249 static struct pernet_operations fib6_net_ops = {
2250 	.init = fib6_net_init,
2251 	.exit = fib6_net_exit,
2252 };
2253 
2254 int __init fib6_init(void)
2255 {
2256 	int ret = -ENOMEM;
2257 
2258 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2259 					   sizeof(struct fib6_node),
2260 					   0, SLAB_HWCACHE_ALIGN,
2261 					   NULL);
2262 	if (!fib6_node_kmem)
2263 		goto out;
2264 
2265 	ret = register_pernet_subsys(&fib6_net_ops);
2266 	if (ret)
2267 		goto out_kmem_cache_create;
2268 
2269 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2270 				   inet6_dump_fib, 0);
2271 	if (ret)
2272 		goto out_unregister_subsys;
2273 
2274 	__fib6_flush_trees = fib6_flush_trees;
2275 out:
2276 	return ret;
2277 
2278 out_unregister_subsys:
2279 	unregister_pernet_subsys(&fib6_net_ops);
2280 out_kmem_cache_create:
2281 	kmem_cache_destroy(fib6_node_kmem);
2282 	goto out;
2283 }
2284 
2285 void fib6_gc_cleanup(void)
2286 {
2287 	unregister_pernet_subsys(&fib6_net_ops);
2288 	kmem_cache_destroy(fib6_node_kmem);
2289 }
2290 
2291 #ifdef CONFIG_PROC_FS
2292 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2293 {
2294 	struct fib6_info *rt = v;
2295 	struct ipv6_route_iter *iter = seq->private;
2296 	const struct net_device *dev;
2297 
2298 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2299 
2300 #ifdef CONFIG_IPV6_SUBTREES
2301 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2302 #else
2303 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2304 #endif
2305 	if (rt->fib6_flags & RTF_GATEWAY)
2306 		seq_printf(seq, "%pi6", &rt->fib6_nh.nh_gw);
2307 	else
2308 		seq_puts(seq, "00000000000000000000000000000000");
2309 
2310 	dev = rt->fib6_nh.nh_dev;
2311 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2312 		   rt->fib6_metric, atomic_read(&rt->fib6_ref), 0,
2313 		   rt->fib6_flags, dev ? dev->name : "");
2314 	iter->w.leaf = NULL;
2315 	return 0;
2316 }
2317 
2318 static int ipv6_route_yield(struct fib6_walker *w)
2319 {
2320 	struct ipv6_route_iter *iter = w->args;
2321 
2322 	if (!iter->skip)
2323 		return 1;
2324 
2325 	do {
2326 		iter->w.leaf = rcu_dereference_protected(
2327 				iter->w.leaf->fib6_next,
2328 				lockdep_is_held(&iter->tbl->tb6_lock));
2329 		iter->skip--;
2330 		if (!iter->skip && iter->w.leaf)
2331 			return 1;
2332 	} while (iter->w.leaf);
2333 
2334 	return 0;
2335 }
2336 
2337 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2338 				      struct net *net)
2339 {
2340 	memset(&iter->w, 0, sizeof(iter->w));
2341 	iter->w.func = ipv6_route_yield;
2342 	iter->w.root = &iter->tbl->tb6_root;
2343 	iter->w.state = FWS_INIT;
2344 	iter->w.node = iter->w.root;
2345 	iter->w.args = iter;
2346 	iter->sernum = iter->w.root->fn_sernum;
2347 	INIT_LIST_HEAD(&iter->w.lh);
2348 	fib6_walker_link(net, &iter->w);
2349 }
2350 
2351 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2352 						    struct net *net)
2353 {
2354 	unsigned int h;
2355 	struct hlist_node *node;
2356 
2357 	if (tbl) {
2358 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2359 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2360 	} else {
2361 		h = 0;
2362 		node = NULL;
2363 	}
2364 
2365 	while (!node && h < FIB6_TABLE_HASHSZ) {
2366 		node = rcu_dereference_bh(
2367 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2368 	}
2369 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2370 }
2371 
2372 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2373 {
2374 	if (iter->sernum != iter->w.root->fn_sernum) {
2375 		iter->sernum = iter->w.root->fn_sernum;
2376 		iter->w.state = FWS_INIT;
2377 		iter->w.node = iter->w.root;
2378 		WARN_ON(iter->w.skip);
2379 		iter->w.skip = iter->w.count;
2380 	}
2381 }
2382 
2383 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2384 {
2385 	int r;
2386 	struct fib6_info *n;
2387 	struct net *net = seq_file_net(seq);
2388 	struct ipv6_route_iter *iter = seq->private;
2389 
2390 	if (!v)
2391 		goto iter_table;
2392 
2393 	n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2394 	if (n) {
2395 		++*pos;
2396 		return n;
2397 	}
2398 
2399 iter_table:
2400 	ipv6_route_check_sernum(iter);
2401 	spin_lock_bh(&iter->tbl->tb6_lock);
2402 	r = fib6_walk_continue(&iter->w);
2403 	spin_unlock_bh(&iter->tbl->tb6_lock);
2404 	if (r > 0) {
2405 		if (v)
2406 			++*pos;
2407 		return iter->w.leaf;
2408 	} else if (r < 0) {
2409 		fib6_walker_unlink(net, &iter->w);
2410 		return NULL;
2411 	}
2412 	fib6_walker_unlink(net, &iter->w);
2413 
2414 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2415 	if (!iter->tbl)
2416 		return NULL;
2417 
2418 	ipv6_route_seq_setup_walk(iter, net);
2419 	goto iter_table;
2420 }
2421 
2422 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2423 	__acquires(RCU_BH)
2424 {
2425 	struct net *net = seq_file_net(seq);
2426 	struct ipv6_route_iter *iter = seq->private;
2427 
2428 	rcu_read_lock_bh();
2429 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2430 	iter->skip = *pos;
2431 
2432 	if (iter->tbl) {
2433 		ipv6_route_seq_setup_walk(iter, net);
2434 		return ipv6_route_seq_next(seq, NULL, pos);
2435 	} else {
2436 		return NULL;
2437 	}
2438 }
2439 
2440 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2441 {
2442 	struct fib6_walker *w = &iter->w;
2443 	return w->node && !(w->state == FWS_U && w->node == w->root);
2444 }
2445 
2446 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2447 	__releases(RCU_BH)
2448 {
2449 	struct net *net = seq_file_net(seq);
2450 	struct ipv6_route_iter *iter = seq->private;
2451 
2452 	if (ipv6_route_iter_active(iter))
2453 		fib6_walker_unlink(net, &iter->w);
2454 
2455 	rcu_read_unlock_bh();
2456 }
2457 
2458 const struct seq_operations ipv6_route_seq_ops = {
2459 	.start	= ipv6_route_seq_start,
2460 	.next	= ipv6_route_seq_next,
2461 	.stop	= ipv6_route_seq_stop,
2462 	.show	= ipv6_route_seq_show
2463 };
2464 #endif /* CONFIG_PROC_FS */
2465