xref: /openbmc/linux/net/ipv6/ip6_fib.c (revision dd5b2498)
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 	fib6_nh_release(&f6i->fib6_nh);
203 
204 	ip_fib_metrics_put(f6i->fib6_metrics);
205 
206 	kfree(f6i);
207 }
208 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
209 
210 static struct fib6_node *node_alloc(struct net *net)
211 {
212 	struct fib6_node *fn;
213 
214 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
215 	if (fn)
216 		net->ipv6.rt6_stats->fib_nodes++;
217 
218 	return fn;
219 }
220 
221 static void node_free_immediate(struct net *net, struct fib6_node *fn)
222 {
223 	kmem_cache_free(fib6_node_kmem, fn);
224 	net->ipv6.rt6_stats->fib_nodes--;
225 }
226 
227 static void node_free_rcu(struct rcu_head *head)
228 {
229 	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
230 
231 	kmem_cache_free(fib6_node_kmem, fn);
232 }
233 
234 static void node_free(struct net *net, struct fib6_node *fn)
235 {
236 	call_rcu(&fn->rcu, node_free_rcu);
237 	net->ipv6.rt6_stats->fib_nodes--;
238 }
239 
240 static void fib6_free_table(struct fib6_table *table)
241 {
242 	inetpeer_invalidate_tree(&table->tb6_peers);
243 	kfree(table);
244 }
245 
246 static void fib6_link_table(struct net *net, struct fib6_table *tb)
247 {
248 	unsigned int h;
249 
250 	/*
251 	 * Initialize table lock at a single place to give lockdep a key,
252 	 * tables aren't visible prior to being linked to the list.
253 	 */
254 	spin_lock_init(&tb->tb6_lock);
255 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
256 
257 	/*
258 	 * No protection necessary, this is the only list mutatation
259 	 * operation, tables never disappear once they exist.
260 	 */
261 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
262 }
263 
264 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
265 
266 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
267 {
268 	struct fib6_table *table;
269 
270 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
271 	if (table) {
272 		table->tb6_id = id;
273 		rcu_assign_pointer(table->tb6_root.leaf,
274 				   net->ipv6.fib6_null_entry);
275 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
276 		inet_peer_base_init(&table->tb6_peers);
277 	}
278 
279 	return table;
280 }
281 
282 struct fib6_table *fib6_new_table(struct net *net, u32 id)
283 {
284 	struct fib6_table *tb;
285 
286 	if (id == 0)
287 		id = RT6_TABLE_MAIN;
288 	tb = fib6_get_table(net, id);
289 	if (tb)
290 		return tb;
291 
292 	tb = fib6_alloc_table(net, id);
293 	if (tb)
294 		fib6_link_table(net, tb);
295 
296 	return tb;
297 }
298 EXPORT_SYMBOL_GPL(fib6_new_table);
299 
300 struct fib6_table *fib6_get_table(struct net *net, u32 id)
301 {
302 	struct fib6_table *tb;
303 	struct hlist_head *head;
304 	unsigned int h;
305 
306 	if (id == 0)
307 		id = RT6_TABLE_MAIN;
308 	h = id & (FIB6_TABLE_HASHSZ - 1);
309 	rcu_read_lock();
310 	head = &net->ipv6.fib_table_hash[h];
311 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
312 		if (tb->tb6_id == id) {
313 			rcu_read_unlock();
314 			return tb;
315 		}
316 	}
317 	rcu_read_unlock();
318 
319 	return NULL;
320 }
321 EXPORT_SYMBOL_GPL(fib6_get_table);
322 
323 static void __net_init fib6_tables_init(struct net *net)
324 {
325 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
326 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
327 }
328 #else
329 
330 struct fib6_table *fib6_new_table(struct net *net, u32 id)
331 {
332 	return fib6_get_table(net, id);
333 }
334 
335 struct fib6_table *fib6_get_table(struct net *net, u32 id)
336 {
337 	  return net->ipv6.fib6_main_tbl;
338 }
339 
340 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
341 				   const struct sk_buff *skb,
342 				   int flags, pol_lookup_t lookup)
343 {
344 	struct rt6_info *rt;
345 
346 	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
347 	if (rt->dst.error == -EAGAIN) {
348 		ip6_rt_put(rt);
349 		rt = net->ipv6.ip6_null_entry;
350 		dst_hold(&rt->dst);
351 	}
352 
353 	return &rt->dst;
354 }
355 
356 /* called with rcu lock held; no reference taken on fib6_info */
357 struct fib6_info *fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
358 			      int flags)
359 {
360 	return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6, flags);
361 }
362 
363 static void __net_init fib6_tables_init(struct net *net)
364 {
365 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
366 }
367 
368 #endif
369 
370 unsigned int fib6_tables_seq_read(struct net *net)
371 {
372 	unsigned int h, fib_seq = 0;
373 
374 	rcu_read_lock();
375 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
376 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
377 		struct fib6_table *tb;
378 
379 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
380 			fib_seq += tb->fib_seq;
381 	}
382 	rcu_read_unlock();
383 
384 	return fib_seq;
385 }
386 
387 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
388 				    enum fib_event_type event_type,
389 				    struct fib6_info *rt)
390 {
391 	struct fib6_entry_notifier_info info = {
392 		.rt = rt,
393 	};
394 
395 	return call_fib6_notifier(nb, net, event_type, &info.info);
396 }
397 
398 static int call_fib6_entry_notifiers(struct net *net,
399 				     enum fib_event_type event_type,
400 				     struct fib6_info *rt,
401 				     struct netlink_ext_ack *extack)
402 {
403 	struct fib6_entry_notifier_info info = {
404 		.info.extack = extack,
405 		.rt = rt,
406 	};
407 
408 	rt->fib6_table->fib_seq++;
409 	return call_fib6_notifiers(net, event_type, &info.info);
410 }
411 
412 struct fib6_dump_arg {
413 	struct net *net;
414 	struct notifier_block *nb;
415 };
416 
417 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
418 {
419 	if (rt == arg->net->ipv6.fib6_null_entry)
420 		return;
421 	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
422 }
423 
424 static int fib6_node_dump(struct fib6_walker *w)
425 {
426 	struct fib6_info *rt;
427 
428 	for_each_fib6_walker_rt(w)
429 		fib6_rt_dump(rt, w->args);
430 	w->leaf = NULL;
431 	return 0;
432 }
433 
434 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
435 			    struct fib6_walker *w)
436 {
437 	w->root = &tb->tb6_root;
438 	spin_lock_bh(&tb->tb6_lock);
439 	fib6_walk(net, w);
440 	spin_unlock_bh(&tb->tb6_lock);
441 }
442 
443 /* Called with rcu_read_lock() */
444 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
445 {
446 	struct fib6_dump_arg arg;
447 	struct fib6_walker *w;
448 	unsigned int h;
449 
450 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
451 	if (!w)
452 		return -ENOMEM;
453 
454 	w->func = fib6_node_dump;
455 	arg.net = net;
456 	arg.nb = nb;
457 	w->args = &arg;
458 
459 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
460 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
461 		struct fib6_table *tb;
462 
463 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
464 			fib6_table_dump(net, tb, w);
465 	}
466 
467 	kfree(w);
468 
469 	return 0;
470 }
471 
472 static int fib6_dump_node(struct fib6_walker *w)
473 {
474 	int res;
475 	struct fib6_info *rt;
476 
477 	for_each_fib6_walker_rt(w) {
478 		res = rt6_dump_route(rt, w->args);
479 		if (res < 0) {
480 			/* Frame is full, suspend walking */
481 			w->leaf = rt;
482 			return 1;
483 		}
484 
485 		/* Multipath routes are dumped in one route with the
486 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
487 		 * last sibling of this route (no need to dump the
488 		 * sibling routes again)
489 		 */
490 		if (rt->fib6_nsiblings)
491 			rt = list_last_entry(&rt->fib6_siblings,
492 					     struct fib6_info,
493 					     fib6_siblings);
494 	}
495 	w->leaf = NULL;
496 	return 0;
497 }
498 
499 static void fib6_dump_end(struct netlink_callback *cb)
500 {
501 	struct net *net = sock_net(cb->skb->sk);
502 	struct fib6_walker *w = (void *)cb->args[2];
503 
504 	if (w) {
505 		if (cb->args[4]) {
506 			cb->args[4] = 0;
507 			fib6_walker_unlink(net, w);
508 		}
509 		cb->args[2] = 0;
510 		kfree(w);
511 	}
512 	cb->done = (void *)cb->args[3];
513 	cb->args[1] = 3;
514 }
515 
516 static int fib6_dump_done(struct netlink_callback *cb)
517 {
518 	fib6_dump_end(cb);
519 	return cb->done ? cb->done(cb) : 0;
520 }
521 
522 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
523 			   struct netlink_callback *cb)
524 {
525 	struct net *net = sock_net(skb->sk);
526 	struct fib6_walker *w;
527 	int res;
528 
529 	w = (void *)cb->args[2];
530 	w->root = &table->tb6_root;
531 
532 	if (cb->args[4] == 0) {
533 		w->count = 0;
534 		w->skip = 0;
535 
536 		spin_lock_bh(&table->tb6_lock);
537 		res = fib6_walk(net, w);
538 		spin_unlock_bh(&table->tb6_lock);
539 		if (res > 0) {
540 			cb->args[4] = 1;
541 			cb->args[5] = w->root->fn_sernum;
542 		}
543 	} else {
544 		if (cb->args[5] != w->root->fn_sernum) {
545 			/* Begin at the root if the tree changed */
546 			cb->args[5] = w->root->fn_sernum;
547 			w->state = FWS_INIT;
548 			w->node = w->root;
549 			w->skip = w->count;
550 		} else
551 			w->skip = 0;
552 
553 		spin_lock_bh(&table->tb6_lock);
554 		res = fib6_walk_continue(w);
555 		spin_unlock_bh(&table->tb6_lock);
556 		if (res <= 0) {
557 			fib6_walker_unlink(net, w);
558 			cb->args[4] = 0;
559 		}
560 	}
561 
562 	return res;
563 }
564 
565 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
566 {
567 	const struct nlmsghdr *nlh = cb->nlh;
568 	struct net *net = sock_net(skb->sk);
569 	struct rt6_rtnl_dump_arg arg = {};
570 	unsigned int h, s_h;
571 	unsigned int e = 0, s_e;
572 	struct fib6_walker *w;
573 	struct fib6_table *tb;
574 	struct hlist_head *head;
575 	int res = 0;
576 
577 	if (cb->strict_check) {
578 		int err;
579 
580 		err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
581 		if (err < 0)
582 			return err;
583 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
584 		struct rtmsg *rtm = nlmsg_data(nlh);
585 
586 		arg.filter.flags = rtm->rtm_flags & (RTM_F_PREFIX|RTM_F_CLONED);
587 	}
588 
589 	/* fib entries are never clones */
590 	if (arg.filter.flags & RTM_F_CLONED)
591 		goto out;
592 
593 	w = (void *)cb->args[2];
594 	if (!w) {
595 		/* New dump:
596 		 *
597 		 * 1. hook callback destructor.
598 		 */
599 		cb->args[3] = (long)cb->done;
600 		cb->done = fib6_dump_done;
601 
602 		/*
603 		 * 2. allocate and initialize walker.
604 		 */
605 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
606 		if (!w)
607 			return -ENOMEM;
608 		w->func = fib6_dump_node;
609 		cb->args[2] = (long)w;
610 	}
611 
612 	arg.skb = skb;
613 	arg.cb = cb;
614 	arg.net = net;
615 	w->args = &arg;
616 
617 	if (arg.filter.table_id) {
618 		tb = fib6_get_table(net, arg.filter.table_id);
619 		if (!tb) {
620 			if (arg.filter.dump_all_families)
621 				goto out;
622 
623 			NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
624 			return -ENOENT;
625 		}
626 
627 		if (!cb->args[0]) {
628 			res = fib6_dump_table(tb, skb, cb);
629 			if (!res)
630 				cb->args[0] = 1;
631 		}
632 		goto out;
633 	}
634 
635 	s_h = cb->args[0];
636 	s_e = cb->args[1];
637 
638 	rcu_read_lock();
639 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
640 		e = 0;
641 		head = &net->ipv6.fib_table_hash[h];
642 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
643 			if (e < s_e)
644 				goto next;
645 			res = fib6_dump_table(tb, skb, cb);
646 			if (res != 0)
647 				goto out_unlock;
648 next:
649 			e++;
650 		}
651 	}
652 out_unlock:
653 	rcu_read_unlock();
654 	cb->args[1] = e;
655 	cb->args[0] = h;
656 out:
657 	res = res < 0 ? res : skb->len;
658 	if (res <= 0)
659 		fib6_dump_end(cb);
660 	return res;
661 }
662 
663 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
664 {
665 	if (!f6i)
666 		return;
667 
668 	if (f6i->fib6_metrics == &dst_default_metrics) {
669 		struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
670 
671 		if (!p)
672 			return;
673 
674 		refcount_set(&p->refcnt, 1);
675 		f6i->fib6_metrics = p;
676 	}
677 
678 	f6i->fib6_metrics->metrics[metric - 1] = val;
679 }
680 
681 /*
682  *	Routing Table
683  *
684  *	return the appropriate node for a routing tree "add" operation
685  *	by either creating and inserting or by returning an existing
686  *	node.
687  */
688 
689 static struct fib6_node *fib6_add_1(struct net *net,
690 				    struct fib6_table *table,
691 				    struct fib6_node *root,
692 				    struct in6_addr *addr, int plen,
693 				    int offset, int allow_create,
694 				    int replace_required,
695 				    struct netlink_ext_ack *extack)
696 {
697 	struct fib6_node *fn, *in, *ln;
698 	struct fib6_node *pn = NULL;
699 	struct rt6key *key;
700 	int	bit;
701 	__be32	dir = 0;
702 
703 	RT6_TRACE("fib6_add_1\n");
704 
705 	/* insert node in tree */
706 
707 	fn = root;
708 
709 	do {
710 		struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
711 					    lockdep_is_held(&table->tb6_lock));
712 		key = (struct rt6key *)((u8 *)leaf + offset);
713 
714 		/*
715 		 *	Prefix match
716 		 */
717 		if (plen < fn->fn_bit ||
718 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
719 			if (!allow_create) {
720 				if (replace_required) {
721 					NL_SET_ERR_MSG(extack,
722 						       "Can not replace route - no match found");
723 					pr_warn("Can't replace route, no match found\n");
724 					return ERR_PTR(-ENOENT);
725 				}
726 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
727 			}
728 			goto insert_above;
729 		}
730 
731 		/*
732 		 *	Exact match ?
733 		 */
734 
735 		if (plen == fn->fn_bit) {
736 			/* clean up an intermediate node */
737 			if (!(fn->fn_flags & RTN_RTINFO)) {
738 				RCU_INIT_POINTER(fn->leaf, NULL);
739 				fib6_info_release(leaf);
740 			/* remove null_entry in the root node */
741 			} else if (fn->fn_flags & RTN_TL_ROOT &&
742 				   rcu_access_pointer(fn->leaf) ==
743 				   net->ipv6.fib6_null_entry) {
744 				RCU_INIT_POINTER(fn->leaf, NULL);
745 			}
746 
747 			return fn;
748 		}
749 
750 		/*
751 		 *	We have more bits to go
752 		 */
753 
754 		/* Try to walk down on tree. */
755 		dir = addr_bit_set(addr, fn->fn_bit);
756 		pn = fn;
757 		fn = dir ?
758 		     rcu_dereference_protected(fn->right,
759 					lockdep_is_held(&table->tb6_lock)) :
760 		     rcu_dereference_protected(fn->left,
761 					lockdep_is_held(&table->tb6_lock));
762 	} while (fn);
763 
764 	if (!allow_create) {
765 		/* We should not create new node because
766 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
767 		 * I assume it is safe to require NLM_F_CREATE when
768 		 * REPLACE flag is used! Later we may want to remove the
769 		 * check for replace_required, because according
770 		 * to netlink specification, NLM_F_CREATE
771 		 * MUST be specified if new route is created.
772 		 * That would keep IPv6 consistent with IPv4
773 		 */
774 		if (replace_required) {
775 			NL_SET_ERR_MSG(extack,
776 				       "Can not replace route - no match found");
777 			pr_warn("Can't replace route, no match found\n");
778 			return ERR_PTR(-ENOENT);
779 		}
780 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
781 	}
782 	/*
783 	 *	We walked to the bottom of tree.
784 	 *	Create new leaf node without children.
785 	 */
786 
787 	ln = node_alloc(net);
788 
789 	if (!ln)
790 		return ERR_PTR(-ENOMEM);
791 	ln->fn_bit = plen;
792 	RCU_INIT_POINTER(ln->parent, pn);
793 
794 	if (dir)
795 		rcu_assign_pointer(pn->right, ln);
796 	else
797 		rcu_assign_pointer(pn->left, ln);
798 
799 	return ln;
800 
801 
802 insert_above:
803 	/*
804 	 * split since we don't have a common prefix anymore or
805 	 * we have a less significant route.
806 	 * we've to insert an intermediate node on the list
807 	 * this new node will point to the one we need to create
808 	 * and the current
809 	 */
810 
811 	pn = rcu_dereference_protected(fn->parent,
812 				       lockdep_is_held(&table->tb6_lock));
813 
814 	/* find 1st bit in difference between the 2 addrs.
815 
816 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
817 	   but if it is >= plen, the value is ignored in any case.
818 	 */
819 
820 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
821 
822 	/*
823 	 *		(intermediate)[in]
824 	 *	          /	   \
825 	 *	(new leaf node)[ln] (old node)[fn]
826 	 */
827 	if (plen > bit) {
828 		in = node_alloc(net);
829 		ln = node_alloc(net);
830 
831 		if (!in || !ln) {
832 			if (in)
833 				node_free_immediate(net, in);
834 			if (ln)
835 				node_free_immediate(net, ln);
836 			return ERR_PTR(-ENOMEM);
837 		}
838 
839 		/*
840 		 * new intermediate node.
841 		 * RTN_RTINFO will
842 		 * be off since that an address that chooses one of
843 		 * the branches would not match less specific routes
844 		 * in the other branch
845 		 */
846 
847 		in->fn_bit = bit;
848 
849 		RCU_INIT_POINTER(in->parent, pn);
850 		in->leaf = fn->leaf;
851 		atomic_inc(&rcu_dereference_protected(in->leaf,
852 				lockdep_is_held(&table->tb6_lock))->fib6_ref);
853 
854 		/* update parent pointer */
855 		if (dir)
856 			rcu_assign_pointer(pn->right, in);
857 		else
858 			rcu_assign_pointer(pn->left, in);
859 
860 		ln->fn_bit = plen;
861 
862 		RCU_INIT_POINTER(ln->parent, in);
863 		rcu_assign_pointer(fn->parent, in);
864 
865 		if (addr_bit_set(addr, bit)) {
866 			rcu_assign_pointer(in->right, ln);
867 			rcu_assign_pointer(in->left, fn);
868 		} else {
869 			rcu_assign_pointer(in->left, ln);
870 			rcu_assign_pointer(in->right, fn);
871 		}
872 	} else { /* plen <= bit */
873 
874 		/*
875 		 *		(new leaf node)[ln]
876 		 *	          /	   \
877 		 *	     (old node)[fn] NULL
878 		 */
879 
880 		ln = node_alloc(net);
881 
882 		if (!ln)
883 			return ERR_PTR(-ENOMEM);
884 
885 		ln->fn_bit = plen;
886 
887 		RCU_INIT_POINTER(ln->parent, pn);
888 
889 		if (addr_bit_set(&key->addr, plen))
890 			RCU_INIT_POINTER(ln->right, fn);
891 		else
892 			RCU_INIT_POINTER(ln->left, fn);
893 
894 		rcu_assign_pointer(fn->parent, ln);
895 
896 		if (dir)
897 			rcu_assign_pointer(pn->right, ln);
898 		else
899 			rcu_assign_pointer(pn->left, ln);
900 	}
901 	return ln;
902 }
903 
904 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
905 				const struct fib6_table *table)
906 {
907 	int cpu;
908 
909 	/* release the reference to this fib entry from
910 	 * all of its cached pcpu routes
911 	 */
912 	for_each_possible_cpu(cpu) {
913 		struct rt6_info **ppcpu_rt;
914 		struct rt6_info *pcpu_rt;
915 
916 		ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
917 		pcpu_rt = *ppcpu_rt;
918 		if (pcpu_rt) {
919 			struct fib6_info *from;
920 
921 			from = rcu_dereference_protected(pcpu_rt->from,
922 					     lockdep_is_held(&table->tb6_lock));
923 			rcu_assign_pointer(pcpu_rt->from, NULL);
924 			fib6_info_release(from);
925 		}
926 	}
927 }
928 
929 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
930 			  struct net *net)
931 {
932 	struct fib6_table *table = rt->fib6_table;
933 
934 	if (atomic_read(&rt->fib6_ref) != 1) {
935 		/* This route is used as dummy address holder in some split
936 		 * nodes. It is not leaked, but it still holds other resources,
937 		 * which must be released in time. So, scan ascendant nodes
938 		 * and replace dummy references to this route with references
939 		 * to still alive ones.
940 		 */
941 		while (fn) {
942 			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
943 					    lockdep_is_held(&table->tb6_lock));
944 			struct fib6_info *new_leaf;
945 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
946 				new_leaf = fib6_find_prefix(net, table, fn);
947 				atomic_inc(&new_leaf->fib6_ref);
948 
949 				rcu_assign_pointer(fn->leaf, new_leaf);
950 				fib6_info_release(rt);
951 			}
952 			fn = rcu_dereference_protected(fn->parent,
953 				    lockdep_is_held(&table->tb6_lock));
954 		}
955 
956 		if (rt->rt6i_pcpu)
957 			fib6_drop_pcpu_from(rt, table);
958 	}
959 }
960 
961 /*
962  *	Insert routing information in a node.
963  */
964 
965 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
966 			    struct nl_info *info,
967 			    struct netlink_ext_ack *extack)
968 {
969 	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
970 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
971 	struct fib6_info *iter = NULL;
972 	struct fib6_info __rcu **ins;
973 	struct fib6_info __rcu **fallback_ins = NULL;
974 	int replace = (info->nlh &&
975 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
976 	int add = (!info->nlh ||
977 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
978 	int found = 0;
979 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
980 	u16 nlflags = NLM_F_EXCL;
981 	int err;
982 
983 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
984 		nlflags |= NLM_F_APPEND;
985 
986 	ins = &fn->leaf;
987 
988 	for (iter = leaf; iter;
989 	     iter = rcu_dereference_protected(iter->fib6_next,
990 				lockdep_is_held(&rt->fib6_table->tb6_lock))) {
991 		/*
992 		 *	Search for duplicates
993 		 */
994 
995 		if (iter->fib6_metric == rt->fib6_metric) {
996 			/*
997 			 *	Same priority level
998 			 */
999 			if (info->nlh &&
1000 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
1001 				return -EEXIST;
1002 
1003 			nlflags &= ~NLM_F_EXCL;
1004 			if (replace) {
1005 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1006 					found++;
1007 					break;
1008 				}
1009 				if (rt_can_ecmp)
1010 					fallback_ins = fallback_ins ?: ins;
1011 				goto next_iter;
1012 			}
1013 
1014 			if (rt6_duplicate_nexthop(iter, rt)) {
1015 				if (rt->fib6_nsiblings)
1016 					rt->fib6_nsiblings = 0;
1017 				if (!(iter->fib6_flags & RTF_EXPIRES))
1018 					return -EEXIST;
1019 				if (!(rt->fib6_flags & RTF_EXPIRES))
1020 					fib6_clean_expires(iter);
1021 				else
1022 					fib6_set_expires(iter, rt->expires);
1023 
1024 				if (rt->fib6_pmtu)
1025 					fib6_metric_set(iter, RTAX_MTU,
1026 							rt->fib6_pmtu);
1027 				return -EEXIST;
1028 			}
1029 			/* If we have the same destination and the same metric,
1030 			 * but not the same gateway, then the route we try to
1031 			 * add is sibling to this route, increment our counter
1032 			 * of siblings, and later we will add our route to the
1033 			 * list.
1034 			 * Only static routes (which don't have flag
1035 			 * RTF_EXPIRES) are used for ECMPv6.
1036 			 *
1037 			 * To avoid long list, we only had siblings if the
1038 			 * route have a gateway.
1039 			 */
1040 			if (rt_can_ecmp &&
1041 			    rt6_qualify_for_ecmp(iter))
1042 				rt->fib6_nsiblings++;
1043 		}
1044 
1045 		if (iter->fib6_metric > rt->fib6_metric)
1046 			break;
1047 
1048 next_iter:
1049 		ins = &iter->fib6_next;
1050 	}
1051 
1052 	if (fallback_ins && !found) {
1053 		/* No ECMP-able route found, replace first non-ECMP one */
1054 		ins = fallback_ins;
1055 		iter = rcu_dereference_protected(*ins,
1056 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1057 		found++;
1058 	}
1059 
1060 	/* Reset round-robin state, if necessary */
1061 	if (ins == &fn->leaf)
1062 		fn->rr_ptr = NULL;
1063 
1064 	/* Link this route to others same route. */
1065 	if (rt->fib6_nsiblings) {
1066 		unsigned int fib6_nsiblings;
1067 		struct fib6_info *sibling, *temp_sibling;
1068 
1069 		/* Find the first route that have the same metric */
1070 		sibling = leaf;
1071 		while (sibling) {
1072 			if (sibling->fib6_metric == rt->fib6_metric &&
1073 			    rt6_qualify_for_ecmp(sibling)) {
1074 				list_add_tail(&rt->fib6_siblings,
1075 					      &sibling->fib6_siblings);
1076 				break;
1077 			}
1078 			sibling = rcu_dereference_protected(sibling->fib6_next,
1079 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1080 		}
1081 		/* For each sibling in the list, increment the counter of
1082 		 * siblings. BUG() if counters does not match, list of siblings
1083 		 * is broken!
1084 		 */
1085 		fib6_nsiblings = 0;
1086 		list_for_each_entry_safe(sibling, temp_sibling,
1087 					 &rt->fib6_siblings, fib6_siblings) {
1088 			sibling->fib6_nsiblings++;
1089 			BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1090 			fib6_nsiblings++;
1091 		}
1092 		BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1093 		rt6_multipath_rebalance(temp_sibling);
1094 	}
1095 
1096 	/*
1097 	 *	insert node
1098 	 */
1099 	if (!replace) {
1100 		if (!add)
1101 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1102 
1103 add:
1104 		nlflags |= NLM_F_CREATE;
1105 
1106 		err = call_fib6_entry_notifiers(info->nl_net,
1107 						FIB_EVENT_ENTRY_ADD,
1108 						rt, extack);
1109 		if (err)
1110 			return err;
1111 
1112 		rcu_assign_pointer(rt->fib6_next, iter);
1113 		atomic_inc(&rt->fib6_ref);
1114 		rcu_assign_pointer(rt->fib6_node, fn);
1115 		rcu_assign_pointer(*ins, rt);
1116 		if (!info->skip_notify)
1117 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1118 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1119 
1120 		if (!(fn->fn_flags & RTN_RTINFO)) {
1121 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1122 			fn->fn_flags |= RTN_RTINFO;
1123 		}
1124 
1125 	} else {
1126 		int nsiblings;
1127 
1128 		if (!found) {
1129 			if (add)
1130 				goto add;
1131 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1132 			return -ENOENT;
1133 		}
1134 
1135 		err = call_fib6_entry_notifiers(info->nl_net,
1136 						FIB_EVENT_ENTRY_REPLACE,
1137 						rt, extack);
1138 		if (err)
1139 			return err;
1140 
1141 		atomic_inc(&rt->fib6_ref);
1142 		rcu_assign_pointer(rt->fib6_node, fn);
1143 		rt->fib6_next = iter->fib6_next;
1144 		rcu_assign_pointer(*ins, rt);
1145 		if (!info->skip_notify)
1146 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1147 		if (!(fn->fn_flags & RTN_RTINFO)) {
1148 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1149 			fn->fn_flags |= RTN_RTINFO;
1150 		}
1151 		nsiblings = iter->fib6_nsiblings;
1152 		iter->fib6_node = NULL;
1153 		fib6_purge_rt(iter, fn, info->nl_net);
1154 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1155 			fn->rr_ptr = NULL;
1156 		fib6_info_release(iter);
1157 
1158 		if (nsiblings) {
1159 			/* Replacing an ECMP route, remove all siblings */
1160 			ins = &rt->fib6_next;
1161 			iter = rcu_dereference_protected(*ins,
1162 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1163 			while (iter) {
1164 				if (iter->fib6_metric > rt->fib6_metric)
1165 					break;
1166 				if (rt6_qualify_for_ecmp(iter)) {
1167 					*ins = iter->fib6_next;
1168 					iter->fib6_node = NULL;
1169 					fib6_purge_rt(iter, fn, info->nl_net);
1170 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1171 						fn->rr_ptr = NULL;
1172 					fib6_info_release(iter);
1173 					nsiblings--;
1174 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1175 				} else {
1176 					ins = &iter->fib6_next;
1177 				}
1178 				iter = rcu_dereference_protected(*ins,
1179 					lockdep_is_held(&rt->fib6_table->tb6_lock));
1180 			}
1181 			WARN_ON(nsiblings != 0);
1182 		}
1183 	}
1184 
1185 	return 0;
1186 }
1187 
1188 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1189 {
1190 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1191 	    (rt->fib6_flags & RTF_EXPIRES))
1192 		mod_timer(&net->ipv6.ip6_fib_timer,
1193 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1194 }
1195 
1196 void fib6_force_start_gc(struct net *net)
1197 {
1198 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1199 		mod_timer(&net->ipv6.ip6_fib_timer,
1200 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1201 }
1202 
1203 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1204 					   int sernum)
1205 {
1206 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1207 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1208 
1209 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1210 	smp_wmb();
1211 	while (fn) {
1212 		fn->fn_sernum = sernum;
1213 		fn = rcu_dereference_protected(fn->parent,
1214 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1215 	}
1216 }
1217 
1218 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1219 {
1220 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1221 }
1222 
1223 /*
1224  *	Add routing information to the routing tree.
1225  *	<destination addr>/<source addr>
1226  *	with source addr info in sub-trees
1227  *	Need to own table->tb6_lock
1228  */
1229 
1230 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1231 	     struct nl_info *info, struct netlink_ext_ack *extack)
1232 {
1233 	struct fib6_table *table = rt->fib6_table;
1234 	struct fib6_node *fn, *pn = NULL;
1235 	int err = -ENOMEM;
1236 	int allow_create = 1;
1237 	int replace_required = 0;
1238 	int sernum = fib6_new_sernum(info->nl_net);
1239 
1240 	if (info->nlh) {
1241 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1242 			allow_create = 0;
1243 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1244 			replace_required = 1;
1245 	}
1246 	if (!allow_create && !replace_required)
1247 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1248 
1249 	fn = fib6_add_1(info->nl_net, table, root,
1250 			&rt->fib6_dst.addr, rt->fib6_dst.plen,
1251 			offsetof(struct fib6_info, fib6_dst), allow_create,
1252 			replace_required, extack);
1253 	if (IS_ERR(fn)) {
1254 		err = PTR_ERR(fn);
1255 		fn = NULL;
1256 		goto out;
1257 	}
1258 
1259 	pn = fn;
1260 
1261 #ifdef CONFIG_IPV6_SUBTREES
1262 	if (rt->fib6_src.plen) {
1263 		struct fib6_node *sn;
1264 
1265 		if (!rcu_access_pointer(fn->subtree)) {
1266 			struct fib6_node *sfn;
1267 
1268 			/*
1269 			 * Create subtree.
1270 			 *
1271 			 *		fn[main tree]
1272 			 *		|
1273 			 *		sfn[subtree root]
1274 			 *		   \
1275 			 *		    sn[new leaf node]
1276 			 */
1277 
1278 			/* Create subtree root node */
1279 			sfn = node_alloc(info->nl_net);
1280 			if (!sfn)
1281 				goto failure;
1282 
1283 			atomic_inc(&info->nl_net->ipv6.fib6_null_entry->fib6_ref);
1284 			rcu_assign_pointer(sfn->leaf,
1285 					   info->nl_net->ipv6.fib6_null_entry);
1286 			sfn->fn_flags = RTN_ROOT;
1287 
1288 			/* Now add the first leaf node to new subtree */
1289 
1290 			sn = fib6_add_1(info->nl_net, table, sfn,
1291 					&rt->fib6_src.addr, rt->fib6_src.plen,
1292 					offsetof(struct fib6_info, fib6_src),
1293 					allow_create, replace_required, extack);
1294 
1295 			if (IS_ERR(sn)) {
1296 				/* If it is failed, discard just allocated
1297 				   root, and then (in failure) stale node
1298 				   in main tree.
1299 				 */
1300 				node_free_immediate(info->nl_net, sfn);
1301 				err = PTR_ERR(sn);
1302 				goto failure;
1303 			}
1304 
1305 			/* Now link new subtree to main tree */
1306 			rcu_assign_pointer(sfn->parent, fn);
1307 			rcu_assign_pointer(fn->subtree, sfn);
1308 		} else {
1309 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1310 					&rt->fib6_src.addr, rt->fib6_src.plen,
1311 					offsetof(struct fib6_info, fib6_src),
1312 					allow_create, replace_required, extack);
1313 
1314 			if (IS_ERR(sn)) {
1315 				err = PTR_ERR(sn);
1316 				goto failure;
1317 			}
1318 		}
1319 
1320 		if (!rcu_access_pointer(fn->leaf)) {
1321 			if (fn->fn_flags & RTN_TL_ROOT) {
1322 				/* put back null_entry for root node */
1323 				rcu_assign_pointer(fn->leaf,
1324 					    info->nl_net->ipv6.fib6_null_entry);
1325 			} else {
1326 				atomic_inc(&rt->fib6_ref);
1327 				rcu_assign_pointer(fn->leaf, rt);
1328 			}
1329 		}
1330 		fn = sn;
1331 	}
1332 #endif
1333 
1334 	err = fib6_add_rt2node(fn, rt, info, extack);
1335 	if (!err) {
1336 		__fib6_update_sernum_upto_root(rt, sernum);
1337 		fib6_start_gc(info->nl_net, rt);
1338 	}
1339 
1340 out:
1341 	if (err) {
1342 #ifdef CONFIG_IPV6_SUBTREES
1343 		/*
1344 		 * If fib6_add_1 has cleared the old leaf pointer in the
1345 		 * super-tree leaf node we have to find a new one for it.
1346 		 */
1347 		if (pn != fn) {
1348 			struct fib6_info *pn_leaf =
1349 				rcu_dereference_protected(pn->leaf,
1350 				    lockdep_is_held(&table->tb6_lock));
1351 			if (pn_leaf == rt) {
1352 				pn_leaf = NULL;
1353 				RCU_INIT_POINTER(pn->leaf, NULL);
1354 				fib6_info_release(rt);
1355 			}
1356 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1357 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1358 							   pn);
1359 #if RT6_DEBUG >= 2
1360 				if (!pn_leaf) {
1361 					WARN_ON(!pn_leaf);
1362 					pn_leaf =
1363 					    info->nl_net->ipv6.fib6_null_entry;
1364 				}
1365 #endif
1366 				fib6_info_hold(pn_leaf);
1367 				rcu_assign_pointer(pn->leaf, pn_leaf);
1368 			}
1369 		}
1370 #endif
1371 		goto failure;
1372 	}
1373 	return err;
1374 
1375 failure:
1376 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1377 	 * 1. fn is an intermediate node and we failed to add the new
1378 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1379 	 * failure case.
1380 	 * 2. fn is the root node in the table and we fail to add the first
1381 	 * default route to it.
1382 	 */
1383 	if (fn &&
1384 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1385 	     (fn->fn_flags & RTN_TL_ROOT &&
1386 	      !rcu_access_pointer(fn->leaf))))
1387 		fib6_repair_tree(info->nl_net, table, fn);
1388 	return err;
1389 }
1390 
1391 /*
1392  *	Routing tree lookup
1393  *
1394  */
1395 
1396 struct lookup_args {
1397 	int			offset;		/* key offset on fib6_info */
1398 	const struct in6_addr	*addr;		/* search key			*/
1399 };
1400 
1401 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1402 					    struct lookup_args *args)
1403 {
1404 	struct fib6_node *fn;
1405 	__be32 dir;
1406 
1407 	if (unlikely(args->offset == 0))
1408 		return NULL;
1409 
1410 	/*
1411 	 *	Descend on a tree
1412 	 */
1413 
1414 	fn = root;
1415 
1416 	for (;;) {
1417 		struct fib6_node *next;
1418 
1419 		dir = addr_bit_set(args->addr, fn->fn_bit);
1420 
1421 		next = dir ? rcu_dereference(fn->right) :
1422 			     rcu_dereference(fn->left);
1423 
1424 		if (next) {
1425 			fn = next;
1426 			continue;
1427 		}
1428 		break;
1429 	}
1430 
1431 	while (fn) {
1432 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1433 
1434 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1435 			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1436 			struct rt6key *key;
1437 
1438 			if (!leaf)
1439 				goto backtrack;
1440 
1441 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1442 
1443 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1444 #ifdef CONFIG_IPV6_SUBTREES
1445 				if (subtree) {
1446 					struct fib6_node *sfn;
1447 					sfn = fib6_node_lookup_1(subtree,
1448 								 args + 1);
1449 					if (!sfn)
1450 						goto backtrack;
1451 					fn = sfn;
1452 				}
1453 #endif
1454 				if (fn->fn_flags & RTN_RTINFO)
1455 					return fn;
1456 			}
1457 		}
1458 backtrack:
1459 		if (fn->fn_flags & RTN_ROOT)
1460 			break;
1461 
1462 		fn = rcu_dereference(fn->parent);
1463 	}
1464 
1465 	return NULL;
1466 }
1467 
1468 /* called with rcu_read_lock() held
1469  */
1470 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1471 				   const struct in6_addr *daddr,
1472 				   const struct in6_addr *saddr)
1473 {
1474 	struct fib6_node *fn;
1475 	struct lookup_args args[] = {
1476 		{
1477 			.offset = offsetof(struct fib6_info, fib6_dst),
1478 			.addr = daddr,
1479 		},
1480 #ifdef CONFIG_IPV6_SUBTREES
1481 		{
1482 			.offset = offsetof(struct fib6_info, fib6_src),
1483 			.addr = saddr,
1484 		},
1485 #endif
1486 		{
1487 			.offset = 0,	/* sentinel */
1488 		}
1489 	};
1490 
1491 	fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1492 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1493 		fn = root;
1494 
1495 	return fn;
1496 }
1497 
1498 /*
1499  *	Get node with specified destination prefix (and source prefix,
1500  *	if subtrees are used)
1501  *	exact_match == true means we try to find fn with exact match of
1502  *	the passed in prefix addr
1503  *	exact_match == false means we try to find fn with longest prefix
1504  *	match of the passed in prefix addr. This is useful for finding fn
1505  *	for cached route as it will be stored in the exception table under
1506  *	the node with longest prefix length.
1507  */
1508 
1509 
1510 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1511 				       const struct in6_addr *addr,
1512 				       int plen, int offset,
1513 				       bool exact_match)
1514 {
1515 	struct fib6_node *fn, *prev = NULL;
1516 
1517 	for (fn = root; fn ; ) {
1518 		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1519 		struct rt6key *key;
1520 
1521 		/* This node is being deleted */
1522 		if (!leaf) {
1523 			if (plen <= fn->fn_bit)
1524 				goto out;
1525 			else
1526 				goto next;
1527 		}
1528 
1529 		key = (struct rt6key *)((u8 *)leaf + offset);
1530 
1531 		/*
1532 		 *	Prefix match
1533 		 */
1534 		if (plen < fn->fn_bit ||
1535 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1536 			goto out;
1537 
1538 		if (plen == fn->fn_bit)
1539 			return fn;
1540 
1541 		prev = fn;
1542 
1543 next:
1544 		/*
1545 		 *	We have more bits to go
1546 		 */
1547 		if (addr_bit_set(addr, fn->fn_bit))
1548 			fn = rcu_dereference(fn->right);
1549 		else
1550 			fn = rcu_dereference(fn->left);
1551 	}
1552 out:
1553 	if (exact_match)
1554 		return NULL;
1555 	else
1556 		return prev;
1557 }
1558 
1559 struct fib6_node *fib6_locate(struct fib6_node *root,
1560 			      const struct in6_addr *daddr, int dst_len,
1561 			      const struct in6_addr *saddr, int src_len,
1562 			      bool exact_match)
1563 {
1564 	struct fib6_node *fn;
1565 
1566 	fn = fib6_locate_1(root, daddr, dst_len,
1567 			   offsetof(struct fib6_info, fib6_dst),
1568 			   exact_match);
1569 
1570 #ifdef CONFIG_IPV6_SUBTREES
1571 	if (src_len) {
1572 		WARN_ON(saddr == NULL);
1573 		if (fn) {
1574 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1575 
1576 			if (subtree) {
1577 				fn = fib6_locate_1(subtree, saddr, src_len,
1578 					   offsetof(struct fib6_info, fib6_src),
1579 					   exact_match);
1580 			}
1581 		}
1582 	}
1583 #endif
1584 
1585 	if (fn && fn->fn_flags & RTN_RTINFO)
1586 		return fn;
1587 
1588 	return NULL;
1589 }
1590 
1591 
1592 /*
1593  *	Deletion
1594  *
1595  */
1596 
1597 static struct fib6_info *fib6_find_prefix(struct net *net,
1598 					 struct fib6_table *table,
1599 					 struct fib6_node *fn)
1600 {
1601 	struct fib6_node *child_left, *child_right;
1602 
1603 	if (fn->fn_flags & RTN_ROOT)
1604 		return net->ipv6.fib6_null_entry;
1605 
1606 	while (fn) {
1607 		child_left = rcu_dereference_protected(fn->left,
1608 				    lockdep_is_held(&table->tb6_lock));
1609 		child_right = rcu_dereference_protected(fn->right,
1610 				    lockdep_is_held(&table->tb6_lock));
1611 		if (child_left)
1612 			return rcu_dereference_protected(child_left->leaf,
1613 					lockdep_is_held(&table->tb6_lock));
1614 		if (child_right)
1615 			return rcu_dereference_protected(child_right->leaf,
1616 					lockdep_is_held(&table->tb6_lock));
1617 
1618 		fn = FIB6_SUBTREE(fn);
1619 	}
1620 	return NULL;
1621 }
1622 
1623 /*
1624  *	Called to trim the tree of intermediate nodes when possible. "fn"
1625  *	is the node we want to try and remove.
1626  *	Need to own table->tb6_lock
1627  */
1628 
1629 static struct fib6_node *fib6_repair_tree(struct net *net,
1630 					  struct fib6_table *table,
1631 					  struct fib6_node *fn)
1632 {
1633 	int children;
1634 	int nstate;
1635 	struct fib6_node *child;
1636 	struct fib6_walker *w;
1637 	int iter = 0;
1638 
1639 	/* Set fn->leaf to null_entry for root node. */
1640 	if (fn->fn_flags & RTN_TL_ROOT) {
1641 		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1642 		return fn;
1643 	}
1644 
1645 	for (;;) {
1646 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1647 					    lockdep_is_held(&table->tb6_lock));
1648 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1649 					    lockdep_is_held(&table->tb6_lock));
1650 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1651 					    lockdep_is_held(&table->tb6_lock));
1652 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1653 					    lockdep_is_held(&table->tb6_lock));
1654 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1655 					    lockdep_is_held(&table->tb6_lock));
1656 		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1657 					    lockdep_is_held(&table->tb6_lock));
1658 		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1659 					    lockdep_is_held(&table->tb6_lock));
1660 		struct fib6_info *new_fn_leaf;
1661 
1662 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1663 		iter++;
1664 
1665 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1666 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1667 		WARN_ON(fn_leaf);
1668 
1669 		children = 0;
1670 		child = NULL;
1671 		if (fn_r)
1672 			child = fn_r, children |= 1;
1673 		if (fn_l)
1674 			child = fn_l, children |= 2;
1675 
1676 		if (children == 3 || FIB6_SUBTREE(fn)
1677 #ifdef CONFIG_IPV6_SUBTREES
1678 		    /* Subtree root (i.e. fn) may have one child */
1679 		    || (children && fn->fn_flags & RTN_ROOT)
1680 #endif
1681 		    ) {
1682 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1683 #if RT6_DEBUG >= 2
1684 			if (!new_fn_leaf) {
1685 				WARN_ON(!new_fn_leaf);
1686 				new_fn_leaf = net->ipv6.fib6_null_entry;
1687 			}
1688 #endif
1689 			fib6_info_hold(new_fn_leaf);
1690 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1691 			return pn;
1692 		}
1693 
1694 #ifdef CONFIG_IPV6_SUBTREES
1695 		if (FIB6_SUBTREE(pn) == fn) {
1696 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1697 			RCU_INIT_POINTER(pn->subtree, NULL);
1698 			nstate = FWS_L;
1699 		} else {
1700 			WARN_ON(fn->fn_flags & RTN_ROOT);
1701 #endif
1702 			if (pn_r == fn)
1703 				rcu_assign_pointer(pn->right, child);
1704 			else if (pn_l == fn)
1705 				rcu_assign_pointer(pn->left, child);
1706 #if RT6_DEBUG >= 2
1707 			else
1708 				WARN_ON(1);
1709 #endif
1710 			if (child)
1711 				rcu_assign_pointer(child->parent, pn);
1712 			nstate = FWS_R;
1713 #ifdef CONFIG_IPV6_SUBTREES
1714 		}
1715 #endif
1716 
1717 		read_lock(&net->ipv6.fib6_walker_lock);
1718 		FOR_WALKERS(net, w) {
1719 			if (!child) {
1720 				if (w->node == fn) {
1721 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1722 					w->node = pn;
1723 					w->state = nstate;
1724 				}
1725 			} else {
1726 				if (w->node == fn) {
1727 					w->node = child;
1728 					if (children&2) {
1729 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1730 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1731 					} else {
1732 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1733 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1734 					}
1735 				}
1736 			}
1737 		}
1738 		read_unlock(&net->ipv6.fib6_walker_lock);
1739 
1740 		node_free(net, fn);
1741 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1742 			return pn;
1743 
1744 		RCU_INIT_POINTER(pn->leaf, NULL);
1745 		fib6_info_release(pn_leaf);
1746 		fn = pn;
1747 	}
1748 }
1749 
1750 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1751 			   struct fib6_info __rcu **rtp, struct nl_info *info)
1752 {
1753 	struct fib6_walker *w;
1754 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1755 				    lockdep_is_held(&table->tb6_lock));
1756 	struct net *net = info->nl_net;
1757 
1758 	RT6_TRACE("fib6_del_route\n");
1759 
1760 	/* Unlink it */
1761 	*rtp = rt->fib6_next;
1762 	rt->fib6_node = NULL;
1763 	net->ipv6.rt6_stats->fib_rt_entries--;
1764 	net->ipv6.rt6_stats->fib_discarded_routes++;
1765 
1766 	/* Flush all cached dst in exception table */
1767 	rt6_flush_exceptions(rt);
1768 
1769 	/* Reset round-robin state, if necessary */
1770 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1771 		fn->rr_ptr = NULL;
1772 
1773 	/* Remove this entry from other siblings */
1774 	if (rt->fib6_nsiblings) {
1775 		struct fib6_info *sibling, *next_sibling;
1776 
1777 		list_for_each_entry_safe(sibling, next_sibling,
1778 					 &rt->fib6_siblings, fib6_siblings)
1779 			sibling->fib6_nsiblings--;
1780 		rt->fib6_nsiblings = 0;
1781 		list_del_init(&rt->fib6_siblings);
1782 		rt6_multipath_rebalance(next_sibling);
1783 	}
1784 
1785 	/* Adjust walkers */
1786 	read_lock(&net->ipv6.fib6_walker_lock);
1787 	FOR_WALKERS(net, w) {
1788 		if (w->state == FWS_C && w->leaf == rt) {
1789 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1790 			w->leaf = rcu_dereference_protected(rt->fib6_next,
1791 					    lockdep_is_held(&table->tb6_lock));
1792 			if (!w->leaf)
1793 				w->state = FWS_U;
1794 		}
1795 	}
1796 	read_unlock(&net->ipv6.fib6_walker_lock);
1797 
1798 	/* If it was last route, call fib6_repair_tree() to:
1799 	 * 1. For root node, put back null_entry as how the table was created.
1800 	 * 2. For other nodes, expunge its radix tree node.
1801 	 */
1802 	if (!rcu_access_pointer(fn->leaf)) {
1803 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1804 			fn->fn_flags &= ~RTN_RTINFO;
1805 			net->ipv6.rt6_stats->fib_route_nodes--;
1806 		}
1807 		fn = fib6_repair_tree(net, table, fn);
1808 	}
1809 
1810 	fib6_purge_rt(rt, fn, net);
1811 
1812 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1813 	if (!info->skip_notify)
1814 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1815 	fib6_info_release(rt);
1816 }
1817 
1818 /* Need to own table->tb6_lock */
1819 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1820 {
1821 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1822 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1823 	struct fib6_table *table = rt->fib6_table;
1824 	struct net *net = info->nl_net;
1825 	struct fib6_info __rcu **rtp;
1826 	struct fib6_info __rcu **rtp_next;
1827 
1828 	if (!fn || rt == net->ipv6.fib6_null_entry)
1829 		return -ENOENT;
1830 
1831 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1832 
1833 	/*
1834 	 *	Walk the leaf entries looking for ourself
1835 	 */
1836 
1837 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1838 		struct fib6_info *cur = rcu_dereference_protected(*rtp,
1839 					lockdep_is_held(&table->tb6_lock));
1840 		if (rt == cur) {
1841 			fib6_del_route(table, fn, rtp, info);
1842 			return 0;
1843 		}
1844 		rtp_next = &cur->fib6_next;
1845 	}
1846 	return -ENOENT;
1847 }
1848 
1849 /*
1850  *	Tree traversal function.
1851  *
1852  *	Certainly, it is not interrupt safe.
1853  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1854  *	It means, that we can modify tree during walking
1855  *	and use this function for garbage collection, clone pruning,
1856  *	cleaning tree when a device goes down etc. etc.
1857  *
1858  *	It guarantees that every node will be traversed,
1859  *	and that it will be traversed only once.
1860  *
1861  *	Callback function w->func may return:
1862  *	0 -> continue walking.
1863  *	positive value -> walking is suspended (used by tree dumps,
1864  *	and probably by gc, if it will be split to several slices)
1865  *	negative value -> terminate walking.
1866  *
1867  *	The function itself returns:
1868  *	0   -> walk is complete.
1869  *	>0  -> walk is incomplete (i.e. suspended)
1870  *	<0  -> walk is terminated by an error.
1871  *
1872  *	This function is called with tb6_lock held.
1873  */
1874 
1875 static int fib6_walk_continue(struct fib6_walker *w)
1876 {
1877 	struct fib6_node *fn, *pn, *left, *right;
1878 
1879 	/* w->root should always be table->tb6_root */
1880 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1881 
1882 	for (;;) {
1883 		fn = w->node;
1884 		if (!fn)
1885 			return 0;
1886 
1887 		switch (w->state) {
1888 #ifdef CONFIG_IPV6_SUBTREES
1889 		case FWS_S:
1890 			if (FIB6_SUBTREE(fn)) {
1891 				w->node = FIB6_SUBTREE(fn);
1892 				continue;
1893 			}
1894 			w->state = FWS_L;
1895 #endif
1896 			/* fall through */
1897 		case FWS_L:
1898 			left = rcu_dereference_protected(fn->left, 1);
1899 			if (left) {
1900 				w->node = left;
1901 				w->state = FWS_INIT;
1902 				continue;
1903 			}
1904 			w->state = FWS_R;
1905 			/* fall through */
1906 		case FWS_R:
1907 			right = rcu_dereference_protected(fn->right, 1);
1908 			if (right) {
1909 				w->node = right;
1910 				w->state = FWS_INIT;
1911 				continue;
1912 			}
1913 			w->state = FWS_C;
1914 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1915 			/* fall through */
1916 		case FWS_C:
1917 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1918 				int err;
1919 
1920 				if (w->skip) {
1921 					w->skip--;
1922 					goto skip;
1923 				}
1924 
1925 				err = w->func(w);
1926 				if (err)
1927 					return err;
1928 
1929 				w->count++;
1930 				continue;
1931 			}
1932 skip:
1933 			w->state = FWS_U;
1934 			/* fall through */
1935 		case FWS_U:
1936 			if (fn == w->root)
1937 				return 0;
1938 			pn = rcu_dereference_protected(fn->parent, 1);
1939 			left = rcu_dereference_protected(pn->left, 1);
1940 			right = rcu_dereference_protected(pn->right, 1);
1941 			w->node = pn;
1942 #ifdef CONFIG_IPV6_SUBTREES
1943 			if (FIB6_SUBTREE(pn) == fn) {
1944 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1945 				w->state = FWS_L;
1946 				continue;
1947 			}
1948 #endif
1949 			if (left == fn) {
1950 				w->state = FWS_R;
1951 				continue;
1952 			}
1953 			if (right == fn) {
1954 				w->state = FWS_C;
1955 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1956 				continue;
1957 			}
1958 #if RT6_DEBUG >= 2
1959 			WARN_ON(1);
1960 #endif
1961 		}
1962 	}
1963 }
1964 
1965 static int fib6_walk(struct net *net, struct fib6_walker *w)
1966 {
1967 	int res;
1968 
1969 	w->state = FWS_INIT;
1970 	w->node = w->root;
1971 
1972 	fib6_walker_link(net, w);
1973 	res = fib6_walk_continue(w);
1974 	if (res <= 0)
1975 		fib6_walker_unlink(net, w);
1976 	return res;
1977 }
1978 
1979 static int fib6_clean_node(struct fib6_walker *w)
1980 {
1981 	int res;
1982 	struct fib6_info *rt;
1983 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1984 	struct nl_info info = {
1985 		.nl_net = c->net,
1986 		.skip_notify = c->skip_notify,
1987 	};
1988 
1989 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1990 	    w->node->fn_sernum != c->sernum)
1991 		w->node->fn_sernum = c->sernum;
1992 
1993 	if (!c->func) {
1994 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1995 		w->leaf = NULL;
1996 		return 0;
1997 	}
1998 
1999 	for_each_fib6_walker_rt(w) {
2000 		res = c->func(rt, c->arg);
2001 		if (res == -1) {
2002 			w->leaf = rt;
2003 			res = fib6_del(rt, &info);
2004 			if (res) {
2005 #if RT6_DEBUG >= 2
2006 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2007 					 __func__, rt,
2008 					 rcu_access_pointer(rt->fib6_node),
2009 					 res);
2010 #endif
2011 				continue;
2012 			}
2013 			return 0;
2014 		} else if (res == -2) {
2015 			if (WARN_ON(!rt->fib6_nsiblings))
2016 				continue;
2017 			rt = list_last_entry(&rt->fib6_siblings,
2018 					     struct fib6_info, fib6_siblings);
2019 			continue;
2020 		}
2021 		WARN_ON(res != 0);
2022 	}
2023 	w->leaf = rt;
2024 	return 0;
2025 }
2026 
2027 /*
2028  *	Convenient frontend to tree walker.
2029  *
2030  *	func is called on each route.
2031  *		It may return -2 -> skip multipath route.
2032  *			      -1 -> delete this route.
2033  *		              0  -> continue walking
2034  */
2035 
2036 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2037 			    int (*func)(struct fib6_info *, void *arg),
2038 			    int sernum, void *arg, bool skip_notify)
2039 {
2040 	struct fib6_cleaner c;
2041 
2042 	c.w.root = root;
2043 	c.w.func = fib6_clean_node;
2044 	c.w.count = 0;
2045 	c.w.skip = 0;
2046 	c.func = func;
2047 	c.sernum = sernum;
2048 	c.arg = arg;
2049 	c.net = net;
2050 	c.skip_notify = skip_notify;
2051 
2052 	fib6_walk(net, &c.w);
2053 }
2054 
2055 static void __fib6_clean_all(struct net *net,
2056 			     int (*func)(struct fib6_info *, void *),
2057 			     int sernum, void *arg, bool skip_notify)
2058 {
2059 	struct fib6_table *table;
2060 	struct hlist_head *head;
2061 	unsigned int h;
2062 
2063 	rcu_read_lock();
2064 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2065 		head = &net->ipv6.fib_table_hash[h];
2066 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2067 			spin_lock_bh(&table->tb6_lock);
2068 			fib6_clean_tree(net, &table->tb6_root,
2069 					func, sernum, arg, skip_notify);
2070 			spin_unlock_bh(&table->tb6_lock);
2071 		}
2072 	}
2073 	rcu_read_unlock();
2074 }
2075 
2076 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2077 		    void *arg)
2078 {
2079 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2080 }
2081 
2082 void fib6_clean_all_skip_notify(struct net *net,
2083 				int (*func)(struct fib6_info *, void *),
2084 				void *arg)
2085 {
2086 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2087 }
2088 
2089 static void fib6_flush_trees(struct net *net)
2090 {
2091 	int new_sernum = fib6_new_sernum(net);
2092 
2093 	__fib6_clean_all(net, NULL, new_sernum, NULL, false);
2094 }
2095 
2096 /*
2097  *	Garbage collection
2098  */
2099 
2100 static int fib6_age(struct fib6_info *rt, void *arg)
2101 {
2102 	struct fib6_gc_args *gc_args = arg;
2103 	unsigned long now = jiffies;
2104 
2105 	/*
2106 	 *	check addrconf expiration here.
2107 	 *	Routes are expired even if they are in use.
2108 	 */
2109 
2110 	if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2111 		if (time_after(now, rt->expires)) {
2112 			RT6_TRACE("expiring %p\n", rt);
2113 			return -1;
2114 		}
2115 		gc_args->more++;
2116 	}
2117 
2118 	/*	Also age clones in the exception table.
2119 	 *	Note, that clones are aged out
2120 	 *	only if they are not in use now.
2121 	 */
2122 	rt6_age_exceptions(rt, gc_args, now);
2123 
2124 	return 0;
2125 }
2126 
2127 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2128 {
2129 	struct fib6_gc_args gc_args;
2130 	unsigned long now;
2131 
2132 	if (force) {
2133 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2134 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2135 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2136 		return;
2137 	}
2138 	gc_args.timeout = expires ? (int)expires :
2139 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2140 	gc_args.more = 0;
2141 
2142 	fib6_clean_all(net, fib6_age, &gc_args);
2143 	now = jiffies;
2144 	net->ipv6.ip6_rt_last_gc = now;
2145 
2146 	if (gc_args.more)
2147 		mod_timer(&net->ipv6.ip6_fib_timer,
2148 			  round_jiffies(now
2149 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2150 	else
2151 		del_timer(&net->ipv6.ip6_fib_timer);
2152 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2153 }
2154 
2155 static void fib6_gc_timer_cb(struct timer_list *t)
2156 {
2157 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2158 
2159 	fib6_run_gc(0, arg, true);
2160 }
2161 
2162 static int __net_init fib6_net_init(struct net *net)
2163 {
2164 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2165 	int err;
2166 
2167 	err = fib6_notifier_init(net);
2168 	if (err)
2169 		return err;
2170 
2171 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2172 	rwlock_init(&net->ipv6.fib6_walker_lock);
2173 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2174 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2175 
2176 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2177 	if (!net->ipv6.rt6_stats)
2178 		goto out_timer;
2179 
2180 	/* Avoid false sharing : Use at least a full cache line */
2181 	size = max_t(size_t, size, L1_CACHE_BYTES);
2182 
2183 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2184 	if (!net->ipv6.fib_table_hash)
2185 		goto out_rt6_stats;
2186 
2187 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2188 					  GFP_KERNEL);
2189 	if (!net->ipv6.fib6_main_tbl)
2190 		goto out_fib_table_hash;
2191 
2192 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2193 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2194 			   net->ipv6.fib6_null_entry);
2195 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2196 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2197 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2198 
2199 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2200 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2201 					   GFP_KERNEL);
2202 	if (!net->ipv6.fib6_local_tbl)
2203 		goto out_fib6_main_tbl;
2204 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2205 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2206 			   net->ipv6.fib6_null_entry);
2207 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2208 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2209 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2210 #endif
2211 	fib6_tables_init(net);
2212 
2213 	return 0;
2214 
2215 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2216 out_fib6_main_tbl:
2217 	kfree(net->ipv6.fib6_main_tbl);
2218 #endif
2219 out_fib_table_hash:
2220 	kfree(net->ipv6.fib_table_hash);
2221 out_rt6_stats:
2222 	kfree(net->ipv6.rt6_stats);
2223 out_timer:
2224 	fib6_notifier_exit(net);
2225 	return -ENOMEM;
2226 }
2227 
2228 static void fib6_net_exit(struct net *net)
2229 {
2230 	unsigned int i;
2231 
2232 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2233 
2234 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2235 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2236 		struct hlist_node *tmp;
2237 		struct fib6_table *tb;
2238 
2239 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2240 			hlist_del(&tb->tb6_hlist);
2241 			fib6_free_table(tb);
2242 		}
2243 	}
2244 
2245 	kfree(net->ipv6.fib_table_hash);
2246 	kfree(net->ipv6.rt6_stats);
2247 	fib6_notifier_exit(net);
2248 }
2249 
2250 static struct pernet_operations fib6_net_ops = {
2251 	.init = fib6_net_init,
2252 	.exit = fib6_net_exit,
2253 };
2254 
2255 int __init fib6_init(void)
2256 {
2257 	int ret = -ENOMEM;
2258 
2259 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2260 					   sizeof(struct fib6_node),
2261 					   0, SLAB_HWCACHE_ALIGN,
2262 					   NULL);
2263 	if (!fib6_node_kmem)
2264 		goto out;
2265 
2266 	ret = register_pernet_subsys(&fib6_net_ops);
2267 	if (ret)
2268 		goto out_kmem_cache_create;
2269 
2270 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2271 				   inet6_dump_fib, 0);
2272 	if (ret)
2273 		goto out_unregister_subsys;
2274 
2275 	__fib6_flush_trees = fib6_flush_trees;
2276 out:
2277 	return ret;
2278 
2279 out_unregister_subsys:
2280 	unregister_pernet_subsys(&fib6_net_ops);
2281 out_kmem_cache_create:
2282 	kmem_cache_destroy(fib6_node_kmem);
2283 	goto out;
2284 }
2285 
2286 void fib6_gc_cleanup(void)
2287 {
2288 	unregister_pernet_subsys(&fib6_net_ops);
2289 	kmem_cache_destroy(fib6_node_kmem);
2290 }
2291 
2292 #ifdef CONFIG_PROC_FS
2293 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2294 {
2295 	struct fib6_info *rt = v;
2296 	struct ipv6_route_iter *iter = seq->private;
2297 	unsigned int flags = rt->fib6_flags;
2298 	const struct net_device *dev;
2299 
2300 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2301 
2302 #ifdef CONFIG_IPV6_SUBTREES
2303 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2304 #else
2305 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2306 #endif
2307 	if (rt->fib6_nh.fib_nh_gw_family) {
2308 		flags |= RTF_GATEWAY;
2309 		seq_printf(seq, "%pi6", &rt->fib6_nh.fib_nh_gw6);
2310 	} else {
2311 		seq_puts(seq, "00000000000000000000000000000000");
2312 	}
2313 
2314 	dev = rt->fib6_nh.fib_nh_dev;
2315 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2316 		   rt->fib6_metric, atomic_read(&rt->fib6_ref), 0,
2317 		   flags, dev ? dev->name : "");
2318 	iter->w.leaf = NULL;
2319 	return 0;
2320 }
2321 
2322 static int ipv6_route_yield(struct fib6_walker *w)
2323 {
2324 	struct ipv6_route_iter *iter = w->args;
2325 
2326 	if (!iter->skip)
2327 		return 1;
2328 
2329 	do {
2330 		iter->w.leaf = rcu_dereference_protected(
2331 				iter->w.leaf->fib6_next,
2332 				lockdep_is_held(&iter->tbl->tb6_lock));
2333 		iter->skip--;
2334 		if (!iter->skip && iter->w.leaf)
2335 			return 1;
2336 	} while (iter->w.leaf);
2337 
2338 	return 0;
2339 }
2340 
2341 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2342 				      struct net *net)
2343 {
2344 	memset(&iter->w, 0, sizeof(iter->w));
2345 	iter->w.func = ipv6_route_yield;
2346 	iter->w.root = &iter->tbl->tb6_root;
2347 	iter->w.state = FWS_INIT;
2348 	iter->w.node = iter->w.root;
2349 	iter->w.args = iter;
2350 	iter->sernum = iter->w.root->fn_sernum;
2351 	INIT_LIST_HEAD(&iter->w.lh);
2352 	fib6_walker_link(net, &iter->w);
2353 }
2354 
2355 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2356 						    struct net *net)
2357 {
2358 	unsigned int h;
2359 	struct hlist_node *node;
2360 
2361 	if (tbl) {
2362 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2363 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2364 	} else {
2365 		h = 0;
2366 		node = NULL;
2367 	}
2368 
2369 	while (!node && h < FIB6_TABLE_HASHSZ) {
2370 		node = rcu_dereference_bh(
2371 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2372 	}
2373 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2374 }
2375 
2376 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2377 {
2378 	if (iter->sernum != iter->w.root->fn_sernum) {
2379 		iter->sernum = iter->w.root->fn_sernum;
2380 		iter->w.state = FWS_INIT;
2381 		iter->w.node = iter->w.root;
2382 		WARN_ON(iter->w.skip);
2383 		iter->w.skip = iter->w.count;
2384 	}
2385 }
2386 
2387 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2388 {
2389 	int r;
2390 	struct fib6_info *n;
2391 	struct net *net = seq_file_net(seq);
2392 	struct ipv6_route_iter *iter = seq->private;
2393 
2394 	if (!v)
2395 		goto iter_table;
2396 
2397 	n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2398 	if (n) {
2399 		++*pos;
2400 		return n;
2401 	}
2402 
2403 iter_table:
2404 	ipv6_route_check_sernum(iter);
2405 	spin_lock_bh(&iter->tbl->tb6_lock);
2406 	r = fib6_walk_continue(&iter->w);
2407 	spin_unlock_bh(&iter->tbl->tb6_lock);
2408 	if (r > 0) {
2409 		if (v)
2410 			++*pos;
2411 		return iter->w.leaf;
2412 	} else if (r < 0) {
2413 		fib6_walker_unlink(net, &iter->w);
2414 		return NULL;
2415 	}
2416 	fib6_walker_unlink(net, &iter->w);
2417 
2418 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2419 	if (!iter->tbl)
2420 		return NULL;
2421 
2422 	ipv6_route_seq_setup_walk(iter, net);
2423 	goto iter_table;
2424 }
2425 
2426 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2427 	__acquires(RCU_BH)
2428 {
2429 	struct net *net = seq_file_net(seq);
2430 	struct ipv6_route_iter *iter = seq->private;
2431 
2432 	rcu_read_lock_bh();
2433 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2434 	iter->skip = *pos;
2435 
2436 	if (iter->tbl) {
2437 		ipv6_route_seq_setup_walk(iter, net);
2438 		return ipv6_route_seq_next(seq, NULL, pos);
2439 	} else {
2440 		return NULL;
2441 	}
2442 }
2443 
2444 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2445 {
2446 	struct fib6_walker *w = &iter->w;
2447 	return w->node && !(w->state == FWS_U && w->node == w->root);
2448 }
2449 
2450 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2451 	__releases(RCU_BH)
2452 {
2453 	struct net *net = seq_file_net(seq);
2454 	struct ipv6_route_iter *iter = seq->private;
2455 
2456 	if (ipv6_route_iter_active(iter))
2457 		fib6_walker_unlink(net, &iter->w);
2458 
2459 	rcu_read_unlock_bh();
2460 }
2461 
2462 const struct seq_operations ipv6_route_seq_ops = {
2463 	.start	= ipv6_route_seq_start,
2464 	.next	= ipv6_route_seq_next,
2465 	.stop	= ipv6_route_seq_stop,
2466 	.show	= ipv6_route_seq_show
2467 };
2468 #endif /* CONFIG_PROC_FS */
2469