xref: /openbmc/linux/net/ipv6/ip6_fib.c (revision 752beb5e)
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 	refcount_set(&f6i->fib6_ref, 1);
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 	kfree(bucket);
179 
180 	if (f6i->rt6i_pcpu) {
181 		int cpu;
182 
183 		for_each_possible_cpu(cpu) {
184 			struct rt6_info **ppcpu_rt;
185 			struct rt6_info *pcpu_rt;
186 
187 			ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
188 			pcpu_rt = *ppcpu_rt;
189 			if (pcpu_rt) {
190 				dst_dev_put(&pcpu_rt->dst);
191 				dst_release(&pcpu_rt->dst);
192 				*ppcpu_rt = NULL;
193 			}
194 		}
195 
196 		free_percpu(f6i->rt6i_pcpu);
197 	}
198 
199 	fib6_nh_release(&f6i->fib6_nh);
200 
201 	ip_fib_metrics_put(f6i->fib6_metrics);
202 
203 	kfree(f6i);
204 }
205 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
206 
207 static struct fib6_node *node_alloc(struct net *net)
208 {
209 	struct fib6_node *fn;
210 
211 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
212 	if (fn)
213 		net->ipv6.rt6_stats->fib_nodes++;
214 
215 	return fn;
216 }
217 
218 static void node_free_immediate(struct net *net, struct fib6_node *fn)
219 {
220 	kmem_cache_free(fib6_node_kmem, fn);
221 	net->ipv6.rt6_stats->fib_nodes--;
222 }
223 
224 static void node_free_rcu(struct rcu_head *head)
225 {
226 	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
227 
228 	kmem_cache_free(fib6_node_kmem, fn);
229 }
230 
231 static void node_free(struct net *net, struct fib6_node *fn)
232 {
233 	call_rcu(&fn->rcu, node_free_rcu);
234 	net->ipv6.rt6_stats->fib_nodes--;
235 }
236 
237 static void fib6_free_table(struct fib6_table *table)
238 {
239 	inetpeer_invalidate_tree(&table->tb6_peers);
240 	kfree(table);
241 }
242 
243 static void fib6_link_table(struct net *net, struct fib6_table *tb)
244 {
245 	unsigned int h;
246 
247 	/*
248 	 * Initialize table lock at a single place to give lockdep a key,
249 	 * tables aren't visible prior to being linked to the list.
250 	 */
251 	spin_lock_init(&tb->tb6_lock);
252 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
253 
254 	/*
255 	 * No protection necessary, this is the only list mutatation
256 	 * operation, tables never disappear once they exist.
257 	 */
258 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
259 }
260 
261 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
262 
263 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
264 {
265 	struct fib6_table *table;
266 
267 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
268 	if (table) {
269 		table->tb6_id = id;
270 		rcu_assign_pointer(table->tb6_root.leaf,
271 				   net->ipv6.fib6_null_entry);
272 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
273 		inet_peer_base_init(&table->tb6_peers);
274 	}
275 
276 	return table;
277 }
278 
279 struct fib6_table *fib6_new_table(struct net *net, u32 id)
280 {
281 	struct fib6_table *tb;
282 
283 	if (id == 0)
284 		id = RT6_TABLE_MAIN;
285 	tb = fib6_get_table(net, id);
286 	if (tb)
287 		return tb;
288 
289 	tb = fib6_alloc_table(net, id);
290 	if (tb)
291 		fib6_link_table(net, tb);
292 
293 	return tb;
294 }
295 EXPORT_SYMBOL_GPL(fib6_new_table);
296 
297 struct fib6_table *fib6_get_table(struct net *net, u32 id)
298 {
299 	struct fib6_table *tb;
300 	struct hlist_head *head;
301 	unsigned int h;
302 
303 	if (id == 0)
304 		id = RT6_TABLE_MAIN;
305 	h = id & (FIB6_TABLE_HASHSZ - 1);
306 	rcu_read_lock();
307 	head = &net->ipv6.fib_table_hash[h];
308 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
309 		if (tb->tb6_id == id) {
310 			rcu_read_unlock();
311 			return tb;
312 		}
313 	}
314 	rcu_read_unlock();
315 
316 	return NULL;
317 }
318 EXPORT_SYMBOL_GPL(fib6_get_table);
319 
320 static void __net_init fib6_tables_init(struct net *net)
321 {
322 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
323 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
324 }
325 #else
326 
327 struct fib6_table *fib6_new_table(struct net *net, u32 id)
328 {
329 	return fib6_get_table(net, id);
330 }
331 
332 struct fib6_table *fib6_get_table(struct net *net, u32 id)
333 {
334 	  return net->ipv6.fib6_main_tbl;
335 }
336 
337 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
338 				   const struct sk_buff *skb,
339 				   int flags, pol_lookup_t lookup)
340 {
341 	struct rt6_info *rt;
342 
343 	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
344 	if (rt->dst.error == -EAGAIN) {
345 		ip6_rt_put(rt);
346 		rt = net->ipv6.ip6_null_entry;
347 		dst_hold(&rt->dst);
348 	}
349 
350 	return &rt->dst;
351 }
352 
353 /* called with rcu lock held; no reference taken on fib6_info */
354 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
355 		struct fib6_result *res, int flags)
356 {
357 	return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
358 				 res, flags);
359 }
360 
361 static void __net_init fib6_tables_init(struct net *net)
362 {
363 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
364 }
365 
366 #endif
367 
368 unsigned int fib6_tables_seq_read(struct net *net)
369 {
370 	unsigned int h, fib_seq = 0;
371 
372 	rcu_read_lock();
373 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
374 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
375 		struct fib6_table *tb;
376 
377 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
378 			fib_seq += tb->fib_seq;
379 	}
380 	rcu_read_unlock();
381 
382 	return fib_seq;
383 }
384 
385 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
386 				    enum fib_event_type event_type,
387 				    struct fib6_info *rt)
388 {
389 	struct fib6_entry_notifier_info info = {
390 		.rt = rt,
391 	};
392 
393 	return call_fib6_notifier(nb, net, event_type, &info.info);
394 }
395 
396 static int call_fib6_entry_notifiers(struct net *net,
397 				     enum fib_event_type event_type,
398 				     struct fib6_info *rt,
399 				     struct netlink_ext_ack *extack)
400 {
401 	struct fib6_entry_notifier_info info = {
402 		.info.extack = extack,
403 		.rt = rt,
404 	};
405 
406 	rt->fib6_table->fib_seq++;
407 	return call_fib6_notifiers(net, event_type, &info.info);
408 }
409 
410 struct fib6_dump_arg {
411 	struct net *net;
412 	struct notifier_block *nb;
413 };
414 
415 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
416 {
417 	if (rt == arg->net->ipv6.fib6_null_entry)
418 		return;
419 	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
420 }
421 
422 static int fib6_node_dump(struct fib6_walker *w)
423 {
424 	struct fib6_info *rt;
425 
426 	for_each_fib6_walker_rt(w)
427 		fib6_rt_dump(rt, w->args);
428 	w->leaf = NULL;
429 	return 0;
430 }
431 
432 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
433 			    struct fib6_walker *w)
434 {
435 	w->root = &tb->tb6_root;
436 	spin_lock_bh(&tb->tb6_lock);
437 	fib6_walk(net, w);
438 	spin_unlock_bh(&tb->tb6_lock);
439 }
440 
441 /* Called with rcu_read_lock() */
442 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
443 {
444 	struct fib6_dump_arg arg;
445 	struct fib6_walker *w;
446 	unsigned int h;
447 
448 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
449 	if (!w)
450 		return -ENOMEM;
451 
452 	w->func = fib6_node_dump;
453 	arg.net = net;
454 	arg.nb = nb;
455 	w->args = &arg;
456 
457 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
458 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
459 		struct fib6_table *tb;
460 
461 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
462 			fib6_table_dump(net, tb, w);
463 	}
464 
465 	kfree(w);
466 
467 	return 0;
468 }
469 
470 static int fib6_dump_node(struct fib6_walker *w)
471 {
472 	int res;
473 	struct fib6_info *rt;
474 
475 	for_each_fib6_walker_rt(w) {
476 		res = rt6_dump_route(rt, w->args);
477 		if (res < 0) {
478 			/* Frame is full, suspend walking */
479 			w->leaf = rt;
480 			return 1;
481 		}
482 
483 		/* Multipath routes are dumped in one route with the
484 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
485 		 * last sibling of this route (no need to dump the
486 		 * sibling routes again)
487 		 */
488 		if (rt->fib6_nsiblings)
489 			rt = list_last_entry(&rt->fib6_siblings,
490 					     struct fib6_info,
491 					     fib6_siblings);
492 	}
493 	w->leaf = NULL;
494 	return 0;
495 }
496 
497 static void fib6_dump_end(struct netlink_callback *cb)
498 {
499 	struct net *net = sock_net(cb->skb->sk);
500 	struct fib6_walker *w = (void *)cb->args[2];
501 
502 	if (w) {
503 		if (cb->args[4]) {
504 			cb->args[4] = 0;
505 			fib6_walker_unlink(net, w);
506 		}
507 		cb->args[2] = 0;
508 		kfree(w);
509 	}
510 	cb->done = (void *)cb->args[3];
511 	cb->args[1] = 3;
512 }
513 
514 static int fib6_dump_done(struct netlink_callback *cb)
515 {
516 	fib6_dump_end(cb);
517 	return cb->done ? cb->done(cb) : 0;
518 }
519 
520 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
521 			   struct netlink_callback *cb)
522 {
523 	struct net *net = sock_net(skb->sk);
524 	struct fib6_walker *w;
525 	int res;
526 
527 	w = (void *)cb->args[2];
528 	w->root = &table->tb6_root;
529 
530 	if (cb->args[4] == 0) {
531 		w->count = 0;
532 		w->skip = 0;
533 
534 		spin_lock_bh(&table->tb6_lock);
535 		res = fib6_walk(net, w);
536 		spin_unlock_bh(&table->tb6_lock);
537 		if (res > 0) {
538 			cb->args[4] = 1;
539 			cb->args[5] = w->root->fn_sernum;
540 		}
541 	} else {
542 		if (cb->args[5] != w->root->fn_sernum) {
543 			/* Begin at the root if the tree changed */
544 			cb->args[5] = w->root->fn_sernum;
545 			w->state = FWS_INIT;
546 			w->node = w->root;
547 			w->skip = w->count;
548 		} else
549 			w->skip = 0;
550 
551 		spin_lock_bh(&table->tb6_lock);
552 		res = fib6_walk_continue(w);
553 		spin_unlock_bh(&table->tb6_lock);
554 		if (res <= 0) {
555 			fib6_walker_unlink(net, w);
556 			cb->args[4] = 0;
557 		}
558 	}
559 
560 	return res;
561 }
562 
563 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
564 {
565 	const struct nlmsghdr *nlh = cb->nlh;
566 	struct net *net = sock_net(skb->sk);
567 	struct rt6_rtnl_dump_arg arg = {};
568 	unsigned int h, s_h;
569 	unsigned int e = 0, s_e;
570 	struct fib6_walker *w;
571 	struct fib6_table *tb;
572 	struct hlist_head *head;
573 	int res = 0;
574 
575 	if (cb->strict_check) {
576 		int err;
577 
578 		err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
579 		if (err < 0)
580 			return err;
581 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
582 		struct rtmsg *rtm = nlmsg_data(nlh);
583 
584 		arg.filter.flags = rtm->rtm_flags & (RTM_F_PREFIX|RTM_F_CLONED);
585 	}
586 
587 	/* fib entries are never clones */
588 	if (arg.filter.flags & RTM_F_CLONED)
589 		goto out;
590 
591 	w = (void *)cb->args[2];
592 	if (!w) {
593 		/* New dump:
594 		 *
595 		 * 1. hook callback destructor.
596 		 */
597 		cb->args[3] = (long)cb->done;
598 		cb->done = fib6_dump_done;
599 
600 		/*
601 		 * 2. allocate and initialize walker.
602 		 */
603 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
604 		if (!w)
605 			return -ENOMEM;
606 		w->func = fib6_dump_node;
607 		cb->args[2] = (long)w;
608 	}
609 
610 	arg.skb = skb;
611 	arg.cb = cb;
612 	arg.net = net;
613 	w->args = &arg;
614 
615 	if (arg.filter.table_id) {
616 		tb = fib6_get_table(net, arg.filter.table_id);
617 		if (!tb) {
618 			if (arg.filter.dump_all_families)
619 				goto out;
620 
621 			NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
622 			return -ENOENT;
623 		}
624 
625 		if (!cb->args[0]) {
626 			res = fib6_dump_table(tb, skb, cb);
627 			if (!res)
628 				cb->args[0] = 1;
629 		}
630 		goto out;
631 	}
632 
633 	s_h = cb->args[0];
634 	s_e = cb->args[1];
635 
636 	rcu_read_lock();
637 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
638 		e = 0;
639 		head = &net->ipv6.fib_table_hash[h];
640 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
641 			if (e < s_e)
642 				goto next;
643 			res = fib6_dump_table(tb, skb, cb);
644 			if (res != 0)
645 				goto out_unlock;
646 next:
647 			e++;
648 		}
649 	}
650 out_unlock:
651 	rcu_read_unlock();
652 	cb->args[1] = e;
653 	cb->args[0] = h;
654 out:
655 	res = res < 0 ? res : skb->len;
656 	if (res <= 0)
657 		fib6_dump_end(cb);
658 	return res;
659 }
660 
661 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
662 {
663 	if (!f6i)
664 		return;
665 
666 	if (f6i->fib6_metrics == &dst_default_metrics) {
667 		struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
668 
669 		if (!p)
670 			return;
671 
672 		refcount_set(&p->refcnt, 1);
673 		f6i->fib6_metrics = p;
674 	}
675 
676 	f6i->fib6_metrics->metrics[metric - 1] = val;
677 }
678 
679 /*
680  *	Routing Table
681  *
682  *	return the appropriate node for a routing tree "add" operation
683  *	by either creating and inserting or by returning an existing
684  *	node.
685  */
686 
687 static struct fib6_node *fib6_add_1(struct net *net,
688 				    struct fib6_table *table,
689 				    struct fib6_node *root,
690 				    struct in6_addr *addr, int plen,
691 				    int offset, int allow_create,
692 				    int replace_required,
693 				    struct netlink_ext_ack *extack)
694 {
695 	struct fib6_node *fn, *in, *ln;
696 	struct fib6_node *pn = NULL;
697 	struct rt6key *key;
698 	int	bit;
699 	__be32	dir = 0;
700 
701 	RT6_TRACE("fib6_add_1\n");
702 
703 	/* insert node in tree */
704 
705 	fn = root;
706 
707 	do {
708 		struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
709 					    lockdep_is_held(&table->tb6_lock));
710 		key = (struct rt6key *)((u8 *)leaf + offset);
711 
712 		/*
713 		 *	Prefix match
714 		 */
715 		if (plen < fn->fn_bit ||
716 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
717 			if (!allow_create) {
718 				if (replace_required) {
719 					NL_SET_ERR_MSG(extack,
720 						       "Can not replace route - no match found");
721 					pr_warn("Can't replace route, no match found\n");
722 					return ERR_PTR(-ENOENT);
723 				}
724 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
725 			}
726 			goto insert_above;
727 		}
728 
729 		/*
730 		 *	Exact match ?
731 		 */
732 
733 		if (plen == fn->fn_bit) {
734 			/* clean up an intermediate node */
735 			if (!(fn->fn_flags & RTN_RTINFO)) {
736 				RCU_INIT_POINTER(fn->leaf, NULL);
737 				fib6_info_release(leaf);
738 			/* remove null_entry in the root node */
739 			} else if (fn->fn_flags & RTN_TL_ROOT &&
740 				   rcu_access_pointer(fn->leaf) ==
741 				   net->ipv6.fib6_null_entry) {
742 				RCU_INIT_POINTER(fn->leaf, NULL);
743 			}
744 
745 			return fn;
746 		}
747 
748 		/*
749 		 *	We have more bits to go
750 		 */
751 
752 		/* Try to walk down on tree. */
753 		dir = addr_bit_set(addr, fn->fn_bit);
754 		pn = fn;
755 		fn = dir ?
756 		     rcu_dereference_protected(fn->right,
757 					lockdep_is_held(&table->tb6_lock)) :
758 		     rcu_dereference_protected(fn->left,
759 					lockdep_is_held(&table->tb6_lock));
760 	} while (fn);
761 
762 	if (!allow_create) {
763 		/* We should not create new node because
764 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
765 		 * I assume it is safe to require NLM_F_CREATE when
766 		 * REPLACE flag is used! Later we may want to remove the
767 		 * check for replace_required, because according
768 		 * to netlink specification, NLM_F_CREATE
769 		 * MUST be specified if new route is created.
770 		 * That would keep IPv6 consistent with IPv4
771 		 */
772 		if (replace_required) {
773 			NL_SET_ERR_MSG(extack,
774 				       "Can not replace route - no match found");
775 			pr_warn("Can't replace route, no match found\n");
776 			return ERR_PTR(-ENOENT);
777 		}
778 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
779 	}
780 	/*
781 	 *	We walked to the bottom of tree.
782 	 *	Create new leaf node without children.
783 	 */
784 
785 	ln = node_alloc(net);
786 
787 	if (!ln)
788 		return ERR_PTR(-ENOMEM);
789 	ln->fn_bit = plen;
790 	RCU_INIT_POINTER(ln->parent, pn);
791 
792 	if (dir)
793 		rcu_assign_pointer(pn->right, ln);
794 	else
795 		rcu_assign_pointer(pn->left, ln);
796 
797 	return ln;
798 
799 
800 insert_above:
801 	/*
802 	 * split since we don't have a common prefix anymore or
803 	 * we have a less significant route.
804 	 * we've to insert an intermediate node on the list
805 	 * this new node will point to the one we need to create
806 	 * and the current
807 	 */
808 
809 	pn = rcu_dereference_protected(fn->parent,
810 				       lockdep_is_held(&table->tb6_lock));
811 
812 	/* find 1st bit in difference between the 2 addrs.
813 
814 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
815 	   but if it is >= plen, the value is ignored in any case.
816 	 */
817 
818 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
819 
820 	/*
821 	 *		(intermediate)[in]
822 	 *	          /	   \
823 	 *	(new leaf node)[ln] (old node)[fn]
824 	 */
825 	if (plen > bit) {
826 		in = node_alloc(net);
827 		ln = node_alloc(net);
828 
829 		if (!in || !ln) {
830 			if (in)
831 				node_free_immediate(net, in);
832 			if (ln)
833 				node_free_immediate(net, ln);
834 			return ERR_PTR(-ENOMEM);
835 		}
836 
837 		/*
838 		 * new intermediate node.
839 		 * RTN_RTINFO will
840 		 * be off since that an address that chooses one of
841 		 * the branches would not match less specific routes
842 		 * in the other branch
843 		 */
844 
845 		in->fn_bit = bit;
846 
847 		RCU_INIT_POINTER(in->parent, pn);
848 		in->leaf = fn->leaf;
849 		fib6_info_hold(rcu_dereference_protected(in->leaf,
850 				lockdep_is_held(&table->tb6_lock)));
851 
852 		/* update parent pointer */
853 		if (dir)
854 			rcu_assign_pointer(pn->right, in);
855 		else
856 			rcu_assign_pointer(pn->left, in);
857 
858 		ln->fn_bit = plen;
859 
860 		RCU_INIT_POINTER(ln->parent, in);
861 		rcu_assign_pointer(fn->parent, in);
862 
863 		if (addr_bit_set(addr, bit)) {
864 			rcu_assign_pointer(in->right, ln);
865 			rcu_assign_pointer(in->left, fn);
866 		} else {
867 			rcu_assign_pointer(in->left, ln);
868 			rcu_assign_pointer(in->right, fn);
869 		}
870 	} else { /* plen <= bit */
871 
872 		/*
873 		 *		(new leaf node)[ln]
874 		 *	          /	   \
875 		 *	     (old node)[fn] NULL
876 		 */
877 
878 		ln = node_alloc(net);
879 
880 		if (!ln)
881 			return ERR_PTR(-ENOMEM);
882 
883 		ln->fn_bit = plen;
884 
885 		RCU_INIT_POINTER(ln->parent, pn);
886 
887 		if (addr_bit_set(&key->addr, plen))
888 			RCU_INIT_POINTER(ln->right, fn);
889 		else
890 			RCU_INIT_POINTER(ln->left, fn);
891 
892 		rcu_assign_pointer(fn->parent, ln);
893 
894 		if (dir)
895 			rcu_assign_pointer(pn->right, ln);
896 		else
897 			rcu_assign_pointer(pn->left, ln);
898 	}
899 	return ln;
900 }
901 
902 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
903 				const struct fib6_table *table)
904 {
905 	int cpu;
906 
907 	/* release the reference to this fib entry from
908 	 * all of its cached pcpu routes
909 	 */
910 	for_each_possible_cpu(cpu) {
911 		struct rt6_info **ppcpu_rt;
912 		struct rt6_info *pcpu_rt;
913 
914 		ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
915 		pcpu_rt = *ppcpu_rt;
916 		if (pcpu_rt) {
917 			struct fib6_info *from;
918 
919 			from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
920 			fib6_info_release(from);
921 		}
922 	}
923 }
924 
925 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
926 			  struct net *net)
927 {
928 	struct fib6_table *table = rt->fib6_table;
929 
930 	if (refcount_read(&rt->fib6_ref) != 1) {
931 		/* This route is used as dummy address holder in some split
932 		 * nodes. It is not leaked, but it still holds other resources,
933 		 * which must be released in time. So, scan ascendant nodes
934 		 * and replace dummy references to this route with references
935 		 * to still alive ones.
936 		 */
937 		while (fn) {
938 			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
939 					    lockdep_is_held(&table->tb6_lock));
940 			struct fib6_info *new_leaf;
941 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
942 				new_leaf = fib6_find_prefix(net, table, fn);
943 				fib6_info_hold(new_leaf);
944 
945 				rcu_assign_pointer(fn->leaf, new_leaf);
946 				fib6_info_release(rt);
947 			}
948 			fn = rcu_dereference_protected(fn->parent,
949 				    lockdep_is_held(&table->tb6_lock));
950 		}
951 
952 		if (rt->rt6i_pcpu)
953 			fib6_drop_pcpu_from(rt, table);
954 	}
955 }
956 
957 /*
958  *	Insert routing information in a node.
959  */
960 
961 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
962 			    struct nl_info *info,
963 			    struct netlink_ext_ack *extack)
964 {
965 	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
966 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
967 	struct fib6_info *iter = NULL;
968 	struct fib6_info __rcu **ins;
969 	struct fib6_info __rcu **fallback_ins = NULL;
970 	int replace = (info->nlh &&
971 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
972 	int add = (!info->nlh ||
973 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
974 	int found = 0;
975 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
976 	u16 nlflags = NLM_F_EXCL;
977 	int err;
978 
979 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
980 		nlflags |= NLM_F_APPEND;
981 
982 	ins = &fn->leaf;
983 
984 	for (iter = leaf; iter;
985 	     iter = rcu_dereference_protected(iter->fib6_next,
986 				lockdep_is_held(&rt->fib6_table->tb6_lock))) {
987 		/*
988 		 *	Search for duplicates
989 		 */
990 
991 		if (iter->fib6_metric == rt->fib6_metric) {
992 			/*
993 			 *	Same priority level
994 			 */
995 			if (info->nlh &&
996 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
997 				return -EEXIST;
998 
999 			nlflags &= ~NLM_F_EXCL;
1000 			if (replace) {
1001 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1002 					found++;
1003 					break;
1004 				}
1005 				if (rt_can_ecmp)
1006 					fallback_ins = fallback_ins ?: ins;
1007 				goto next_iter;
1008 			}
1009 
1010 			if (rt6_duplicate_nexthop(iter, rt)) {
1011 				if (rt->fib6_nsiblings)
1012 					rt->fib6_nsiblings = 0;
1013 				if (!(iter->fib6_flags & RTF_EXPIRES))
1014 					return -EEXIST;
1015 				if (!(rt->fib6_flags & RTF_EXPIRES))
1016 					fib6_clean_expires(iter);
1017 				else
1018 					fib6_set_expires(iter, rt->expires);
1019 
1020 				if (rt->fib6_pmtu)
1021 					fib6_metric_set(iter, RTAX_MTU,
1022 							rt->fib6_pmtu);
1023 				return -EEXIST;
1024 			}
1025 			/* If we have the same destination and the same metric,
1026 			 * but not the same gateway, then the route we try to
1027 			 * add is sibling to this route, increment our counter
1028 			 * of siblings, and later we will add our route to the
1029 			 * list.
1030 			 * Only static routes (which don't have flag
1031 			 * RTF_EXPIRES) are used for ECMPv6.
1032 			 *
1033 			 * To avoid long list, we only had siblings if the
1034 			 * route have a gateway.
1035 			 */
1036 			if (rt_can_ecmp &&
1037 			    rt6_qualify_for_ecmp(iter))
1038 				rt->fib6_nsiblings++;
1039 		}
1040 
1041 		if (iter->fib6_metric > rt->fib6_metric)
1042 			break;
1043 
1044 next_iter:
1045 		ins = &iter->fib6_next;
1046 	}
1047 
1048 	if (fallback_ins && !found) {
1049 		/* No ECMP-able route found, replace first non-ECMP one */
1050 		ins = fallback_ins;
1051 		iter = rcu_dereference_protected(*ins,
1052 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1053 		found++;
1054 	}
1055 
1056 	/* Reset round-robin state, if necessary */
1057 	if (ins == &fn->leaf)
1058 		fn->rr_ptr = NULL;
1059 
1060 	/* Link this route to others same route. */
1061 	if (rt->fib6_nsiblings) {
1062 		unsigned int fib6_nsiblings;
1063 		struct fib6_info *sibling, *temp_sibling;
1064 
1065 		/* Find the first route that have the same metric */
1066 		sibling = leaf;
1067 		while (sibling) {
1068 			if (sibling->fib6_metric == rt->fib6_metric &&
1069 			    rt6_qualify_for_ecmp(sibling)) {
1070 				list_add_tail(&rt->fib6_siblings,
1071 					      &sibling->fib6_siblings);
1072 				break;
1073 			}
1074 			sibling = rcu_dereference_protected(sibling->fib6_next,
1075 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1076 		}
1077 		/* For each sibling in the list, increment the counter of
1078 		 * siblings. BUG() if counters does not match, list of siblings
1079 		 * is broken!
1080 		 */
1081 		fib6_nsiblings = 0;
1082 		list_for_each_entry_safe(sibling, temp_sibling,
1083 					 &rt->fib6_siblings, fib6_siblings) {
1084 			sibling->fib6_nsiblings++;
1085 			BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1086 			fib6_nsiblings++;
1087 		}
1088 		BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1089 		rt6_multipath_rebalance(temp_sibling);
1090 	}
1091 
1092 	/*
1093 	 *	insert node
1094 	 */
1095 	if (!replace) {
1096 		if (!add)
1097 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1098 
1099 add:
1100 		nlflags |= NLM_F_CREATE;
1101 
1102 		err = call_fib6_entry_notifiers(info->nl_net,
1103 						FIB_EVENT_ENTRY_ADD,
1104 						rt, extack);
1105 		if (err)
1106 			return err;
1107 
1108 		rcu_assign_pointer(rt->fib6_next, iter);
1109 		fib6_info_hold(rt);
1110 		rcu_assign_pointer(rt->fib6_node, fn);
1111 		rcu_assign_pointer(*ins, rt);
1112 		if (!info->skip_notify)
1113 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1114 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1115 
1116 		if (!(fn->fn_flags & RTN_RTINFO)) {
1117 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1118 			fn->fn_flags |= RTN_RTINFO;
1119 		}
1120 
1121 	} else {
1122 		int nsiblings;
1123 
1124 		if (!found) {
1125 			if (add)
1126 				goto add;
1127 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1128 			return -ENOENT;
1129 		}
1130 
1131 		err = call_fib6_entry_notifiers(info->nl_net,
1132 						FIB_EVENT_ENTRY_REPLACE,
1133 						rt, extack);
1134 		if (err)
1135 			return err;
1136 
1137 		fib6_info_hold(rt);
1138 		rcu_assign_pointer(rt->fib6_node, fn);
1139 		rt->fib6_next = iter->fib6_next;
1140 		rcu_assign_pointer(*ins, rt);
1141 		if (!info->skip_notify)
1142 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1143 		if (!(fn->fn_flags & RTN_RTINFO)) {
1144 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1145 			fn->fn_flags |= RTN_RTINFO;
1146 		}
1147 		nsiblings = iter->fib6_nsiblings;
1148 		iter->fib6_node = NULL;
1149 		fib6_purge_rt(iter, fn, info->nl_net);
1150 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1151 			fn->rr_ptr = NULL;
1152 		fib6_info_release(iter);
1153 
1154 		if (nsiblings) {
1155 			/* Replacing an ECMP route, remove all siblings */
1156 			ins = &rt->fib6_next;
1157 			iter = rcu_dereference_protected(*ins,
1158 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1159 			while (iter) {
1160 				if (iter->fib6_metric > rt->fib6_metric)
1161 					break;
1162 				if (rt6_qualify_for_ecmp(iter)) {
1163 					*ins = iter->fib6_next;
1164 					iter->fib6_node = NULL;
1165 					fib6_purge_rt(iter, fn, info->nl_net);
1166 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1167 						fn->rr_ptr = NULL;
1168 					fib6_info_release(iter);
1169 					nsiblings--;
1170 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1171 				} else {
1172 					ins = &iter->fib6_next;
1173 				}
1174 				iter = rcu_dereference_protected(*ins,
1175 					lockdep_is_held(&rt->fib6_table->tb6_lock));
1176 			}
1177 			WARN_ON(nsiblings != 0);
1178 		}
1179 	}
1180 
1181 	return 0;
1182 }
1183 
1184 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1185 {
1186 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1187 	    (rt->fib6_flags & RTF_EXPIRES))
1188 		mod_timer(&net->ipv6.ip6_fib_timer,
1189 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1190 }
1191 
1192 void fib6_force_start_gc(struct net *net)
1193 {
1194 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1195 		mod_timer(&net->ipv6.ip6_fib_timer,
1196 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1197 }
1198 
1199 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1200 					   int sernum)
1201 {
1202 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1203 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1204 
1205 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1206 	smp_wmb();
1207 	while (fn) {
1208 		fn->fn_sernum = sernum;
1209 		fn = rcu_dereference_protected(fn->parent,
1210 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1211 	}
1212 }
1213 
1214 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1215 {
1216 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1217 }
1218 
1219 /*
1220  *	Add routing information to the routing tree.
1221  *	<destination addr>/<source addr>
1222  *	with source addr info in sub-trees
1223  *	Need to own table->tb6_lock
1224  */
1225 
1226 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1227 	     struct nl_info *info, struct netlink_ext_ack *extack)
1228 {
1229 	struct fib6_table *table = rt->fib6_table;
1230 	struct fib6_node *fn, *pn = NULL;
1231 	int err = -ENOMEM;
1232 	int allow_create = 1;
1233 	int replace_required = 0;
1234 	int sernum = fib6_new_sernum(info->nl_net);
1235 
1236 	if (info->nlh) {
1237 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1238 			allow_create = 0;
1239 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1240 			replace_required = 1;
1241 	}
1242 	if (!allow_create && !replace_required)
1243 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1244 
1245 	fn = fib6_add_1(info->nl_net, table, root,
1246 			&rt->fib6_dst.addr, rt->fib6_dst.plen,
1247 			offsetof(struct fib6_info, fib6_dst), allow_create,
1248 			replace_required, extack);
1249 	if (IS_ERR(fn)) {
1250 		err = PTR_ERR(fn);
1251 		fn = NULL;
1252 		goto out;
1253 	}
1254 
1255 	pn = fn;
1256 
1257 #ifdef CONFIG_IPV6_SUBTREES
1258 	if (rt->fib6_src.plen) {
1259 		struct fib6_node *sn;
1260 
1261 		if (!rcu_access_pointer(fn->subtree)) {
1262 			struct fib6_node *sfn;
1263 
1264 			/*
1265 			 * Create subtree.
1266 			 *
1267 			 *		fn[main tree]
1268 			 *		|
1269 			 *		sfn[subtree root]
1270 			 *		   \
1271 			 *		    sn[new leaf node]
1272 			 */
1273 
1274 			/* Create subtree root node */
1275 			sfn = node_alloc(info->nl_net);
1276 			if (!sfn)
1277 				goto failure;
1278 
1279 			fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1280 			rcu_assign_pointer(sfn->leaf,
1281 					   info->nl_net->ipv6.fib6_null_entry);
1282 			sfn->fn_flags = RTN_ROOT;
1283 
1284 			/* Now add the first leaf node to new subtree */
1285 
1286 			sn = fib6_add_1(info->nl_net, table, sfn,
1287 					&rt->fib6_src.addr, rt->fib6_src.plen,
1288 					offsetof(struct fib6_info, fib6_src),
1289 					allow_create, replace_required, extack);
1290 
1291 			if (IS_ERR(sn)) {
1292 				/* If it is failed, discard just allocated
1293 				   root, and then (in failure) stale node
1294 				   in main tree.
1295 				 */
1296 				node_free_immediate(info->nl_net, sfn);
1297 				err = PTR_ERR(sn);
1298 				goto failure;
1299 			}
1300 
1301 			/* Now link new subtree to main tree */
1302 			rcu_assign_pointer(sfn->parent, fn);
1303 			rcu_assign_pointer(fn->subtree, sfn);
1304 		} else {
1305 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1306 					&rt->fib6_src.addr, rt->fib6_src.plen,
1307 					offsetof(struct fib6_info, fib6_src),
1308 					allow_create, replace_required, extack);
1309 
1310 			if (IS_ERR(sn)) {
1311 				err = PTR_ERR(sn);
1312 				goto failure;
1313 			}
1314 		}
1315 
1316 		if (!rcu_access_pointer(fn->leaf)) {
1317 			if (fn->fn_flags & RTN_TL_ROOT) {
1318 				/* put back null_entry for root node */
1319 				rcu_assign_pointer(fn->leaf,
1320 					    info->nl_net->ipv6.fib6_null_entry);
1321 			} else {
1322 				fib6_info_hold(rt);
1323 				rcu_assign_pointer(fn->leaf, rt);
1324 			}
1325 		}
1326 		fn = sn;
1327 	}
1328 #endif
1329 
1330 	err = fib6_add_rt2node(fn, rt, info, extack);
1331 	if (!err) {
1332 		__fib6_update_sernum_upto_root(rt, sernum);
1333 		fib6_start_gc(info->nl_net, rt);
1334 	}
1335 
1336 out:
1337 	if (err) {
1338 #ifdef CONFIG_IPV6_SUBTREES
1339 		/*
1340 		 * If fib6_add_1 has cleared the old leaf pointer in the
1341 		 * super-tree leaf node we have to find a new one for it.
1342 		 */
1343 		if (pn != fn) {
1344 			struct fib6_info *pn_leaf =
1345 				rcu_dereference_protected(pn->leaf,
1346 				    lockdep_is_held(&table->tb6_lock));
1347 			if (pn_leaf == rt) {
1348 				pn_leaf = NULL;
1349 				RCU_INIT_POINTER(pn->leaf, NULL);
1350 				fib6_info_release(rt);
1351 			}
1352 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1353 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1354 							   pn);
1355 #if RT6_DEBUG >= 2
1356 				if (!pn_leaf) {
1357 					WARN_ON(!pn_leaf);
1358 					pn_leaf =
1359 					    info->nl_net->ipv6.fib6_null_entry;
1360 				}
1361 #endif
1362 				fib6_info_hold(pn_leaf);
1363 				rcu_assign_pointer(pn->leaf, pn_leaf);
1364 			}
1365 		}
1366 #endif
1367 		goto failure;
1368 	}
1369 	return err;
1370 
1371 failure:
1372 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1373 	 * 1. fn is an intermediate node and we failed to add the new
1374 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1375 	 * failure case.
1376 	 * 2. fn is the root node in the table and we fail to add the first
1377 	 * default route to it.
1378 	 */
1379 	if (fn &&
1380 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1381 	     (fn->fn_flags & RTN_TL_ROOT &&
1382 	      !rcu_access_pointer(fn->leaf))))
1383 		fib6_repair_tree(info->nl_net, table, fn);
1384 	return err;
1385 }
1386 
1387 /*
1388  *	Routing tree lookup
1389  *
1390  */
1391 
1392 struct lookup_args {
1393 	int			offset;		/* key offset on fib6_info */
1394 	const struct in6_addr	*addr;		/* search key			*/
1395 };
1396 
1397 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1398 					    struct lookup_args *args)
1399 {
1400 	struct fib6_node *fn;
1401 	__be32 dir;
1402 
1403 	if (unlikely(args->offset == 0))
1404 		return NULL;
1405 
1406 	/*
1407 	 *	Descend on a tree
1408 	 */
1409 
1410 	fn = root;
1411 
1412 	for (;;) {
1413 		struct fib6_node *next;
1414 
1415 		dir = addr_bit_set(args->addr, fn->fn_bit);
1416 
1417 		next = dir ? rcu_dereference(fn->right) :
1418 			     rcu_dereference(fn->left);
1419 
1420 		if (next) {
1421 			fn = next;
1422 			continue;
1423 		}
1424 		break;
1425 	}
1426 
1427 	while (fn) {
1428 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1429 
1430 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1431 			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1432 			struct rt6key *key;
1433 
1434 			if (!leaf)
1435 				goto backtrack;
1436 
1437 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1438 
1439 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1440 #ifdef CONFIG_IPV6_SUBTREES
1441 				if (subtree) {
1442 					struct fib6_node *sfn;
1443 					sfn = fib6_node_lookup_1(subtree,
1444 								 args + 1);
1445 					if (!sfn)
1446 						goto backtrack;
1447 					fn = sfn;
1448 				}
1449 #endif
1450 				if (fn->fn_flags & RTN_RTINFO)
1451 					return fn;
1452 			}
1453 		}
1454 backtrack:
1455 		if (fn->fn_flags & RTN_ROOT)
1456 			break;
1457 
1458 		fn = rcu_dereference(fn->parent);
1459 	}
1460 
1461 	return NULL;
1462 }
1463 
1464 /* called with rcu_read_lock() held
1465  */
1466 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1467 				   const struct in6_addr *daddr,
1468 				   const struct in6_addr *saddr)
1469 {
1470 	struct fib6_node *fn;
1471 	struct lookup_args args[] = {
1472 		{
1473 			.offset = offsetof(struct fib6_info, fib6_dst),
1474 			.addr = daddr,
1475 		},
1476 #ifdef CONFIG_IPV6_SUBTREES
1477 		{
1478 			.offset = offsetof(struct fib6_info, fib6_src),
1479 			.addr = saddr,
1480 		},
1481 #endif
1482 		{
1483 			.offset = 0,	/* sentinel */
1484 		}
1485 	};
1486 
1487 	fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1488 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1489 		fn = root;
1490 
1491 	return fn;
1492 }
1493 
1494 /*
1495  *	Get node with specified destination prefix (and source prefix,
1496  *	if subtrees are used)
1497  *	exact_match == true means we try to find fn with exact match of
1498  *	the passed in prefix addr
1499  *	exact_match == false means we try to find fn with longest prefix
1500  *	match of the passed in prefix addr. This is useful for finding fn
1501  *	for cached route as it will be stored in the exception table under
1502  *	the node with longest prefix length.
1503  */
1504 
1505 
1506 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1507 				       const struct in6_addr *addr,
1508 				       int plen, int offset,
1509 				       bool exact_match)
1510 {
1511 	struct fib6_node *fn, *prev = NULL;
1512 
1513 	for (fn = root; fn ; ) {
1514 		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1515 		struct rt6key *key;
1516 
1517 		/* This node is being deleted */
1518 		if (!leaf) {
1519 			if (plen <= fn->fn_bit)
1520 				goto out;
1521 			else
1522 				goto next;
1523 		}
1524 
1525 		key = (struct rt6key *)((u8 *)leaf + offset);
1526 
1527 		/*
1528 		 *	Prefix match
1529 		 */
1530 		if (plen < fn->fn_bit ||
1531 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1532 			goto out;
1533 
1534 		if (plen == fn->fn_bit)
1535 			return fn;
1536 
1537 		prev = fn;
1538 
1539 next:
1540 		/*
1541 		 *	We have more bits to go
1542 		 */
1543 		if (addr_bit_set(addr, fn->fn_bit))
1544 			fn = rcu_dereference(fn->right);
1545 		else
1546 			fn = rcu_dereference(fn->left);
1547 	}
1548 out:
1549 	if (exact_match)
1550 		return NULL;
1551 	else
1552 		return prev;
1553 }
1554 
1555 struct fib6_node *fib6_locate(struct fib6_node *root,
1556 			      const struct in6_addr *daddr, int dst_len,
1557 			      const struct in6_addr *saddr, int src_len,
1558 			      bool exact_match)
1559 {
1560 	struct fib6_node *fn;
1561 
1562 	fn = fib6_locate_1(root, daddr, dst_len,
1563 			   offsetof(struct fib6_info, fib6_dst),
1564 			   exact_match);
1565 
1566 #ifdef CONFIG_IPV6_SUBTREES
1567 	if (src_len) {
1568 		WARN_ON(saddr == NULL);
1569 		if (fn) {
1570 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1571 
1572 			if (subtree) {
1573 				fn = fib6_locate_1(subtree, saddr, src_len,
1574 					   offsetof(struct fib6_info, fib6_src),
1575 					   exact_match);
1576 			}
1577 		}
1578 	}
1579 #endif
1580 
1581 	if (fn && fn->fn_flags & RTN_RTINFO)
1582 		return fn;
1583 
1584 	return NULL;
1585 }
1586 
1587 
1588 /*
1589  *	Deletion
1590  *
1591  */
1592 
1593 static struct fib6_info *fib6_find_prefix(struct net *net,
1594 					 struct fib6_table *table,
1595 					 struct fib6_node *fn)
1596 {
1597 	struct fib6_node *child_left, *child_right;
1598 
1599 	if (fn->fn_flags & RTN_ROOT)
1600 		return net->ipv6.fib6_null_entry;
1601 
1602 	while (fn) {
1603 		child_left = rcu_dereference_protected(fn->left,
1604 				    lockdep_is_held(&table->tb6_lock));
1605 		child_right = rcu_dereference_protected(fn->right,
1606 				    lockdep_is_held(&table->tb6_lock));
1607 		if (child_left)
1608 			return rcu_dereference_protected(child_left->leaf,
1609 					lockdep_is_held(&table->tb6_lock));
1610 		if (child_right)
1611 			return rcu_dereference_protected(child_right->leaf,
1612 					lockdep_is_held(&table->tb6_lock));
1613 
1614 		fn = FIB6_SUBTREE(fn);
1615 	}
1616 	return NULL;
1617 }
1618 
1619 /*
1620  *	Called to trim the tree of intermediate nodes when possible. "fn"
1621  *	is the node we want to try and remove.
1622  *	Need to own table->tb6_lock
1623  */
1624 
1625 static struct fib6_node *fib6_repair_tree(struct net *net,
1626 					  struct fib6_table *table,
1627 					  struct fib6_node *fn)
1628 {
1629 	int children;
1630 	int nstate;
1631 	struct fib6_node *child;
1632 	struct fib6_walker *w;
1633 	int iter = 0;
1634 
1635 	/* Set fn->leaf to null_entry for root node. */
1636 	if (fn->fn_flags & RTN_TL_ROOT) {
1637 		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1638 		return fn;
1639 	}
1640 
1641 	for (;;) {
1642 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1643 					    lockdep_is_held(&table->tb6_lock));
1644 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1645 					    lockdep_is_held(&table->tb6_lock));
1646 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1647 					    lockdep_is_held(&table->tb6_lock));
1648 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1649 					    lockdep_is_held(&table->tb6_lock));
1650 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1651 					    lockdep_is_held(&table->tb6_lock));
1652 		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1653 					    lockdep_is_held(&table->tb6_lock));
1654 		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1655 					    lockdep_is_held(&table->tb6_lock));
1656 		struct fib6_info *new_fn_leaf;
1657 
1658 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1659 		iter++;
1660 
1661 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1662 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1663 		WARN_ON(fn_leaf);
1664 
1665 		children = 0;
1666 		child = NULL;
1667 		if (fn_r)
1668 			child = fn_r, children |= 1;
1669 		if (fn_l)
1670 			child = fn_l, children |= 2;
1671 
1672 		if (children == 3 || FIB6_SUBTREE(fn)
1673 #ifdef CONFIG_IPV6_SUBTREES
1674 		    /* Subtree root (i.e. fn) may have one child */
1675 		    || (children && fn->fn_flags & RTN_ROOT)
1676 #endif
1677 		    ) {
1678 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1679 #if RT6_DEBUG >= 2
1680 			if (!new_fn_leaf) {
1681 				WARN_ON(!new_fn_leaf);
1682 				new_fn_leaf = net->ipv6.fib6_null_entry;
1683 			}
1684 #endif
1685 			fib6_info_hold(new_fn_leaf);
1686 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1687 			return pn;
1688 		}
1689 
1690 #ifdef CONFIG_IPV6_SUBTREES
1691 		if (FIB6_SUBTREE(pn) == fn) {
1692 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1693 			RCU_INIT_POINTER(pn->subtree, NULL);
1694 			nstate = FWS_L;
1695 		} else {
1696 			WARN_ON(fn->fn_flags & RTN_ROOT);
1697 #endif
1698 			if (pn_r == fn)
1699 				rcu_assign_pointer(pn->right, child);
1700 			else if (pn_l == fn)
1701 				rcu_assign_pointer(pn->left, child);
1702 #if RT6_DEBUG >= 2
1703 			else
1704 				WARN_ON(1);
1705 #endif
1706 			if (child)
1707 				rcu_assign_pointer(child->parent, pn);
1708 			nstate = FWS_R;
1709 #ifdef CONFIG_IPV6_SUBTREES
1710 		}
1711 #endif
1712 
1713 		read_lock(&net->ipv6.fib6_walker_lock);
1714 		FOR_WALKERS(net, w) {
1715 			if (!child) {
1716 				if (w->node == fn) {
1717 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1718 					w->node = pn;
1719 					w->state = nstate;
1720 				}
1721 			} else {
1722 				if (w->node == fn) {
1723 					w->node = child;
1724 					if (children&2) {
1725 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1726 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1727 					} else {
1728 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1729 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1730 					}
1731 				}
1732 			}
1733 		}
1734 		read_unlock(&net->ipv6.fib6_walker_lock);
1735 
1736 		node_free(net, fn);
1737 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1738 			return pn;
1739 
1740 		RCU_INIT_POINTER(pn->leaf, NULL);
1741 		fib6_info_release(pn_leaf);
1742 		fn = pn;
1743 	}
1744 }
1745 
1746 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1747 			   struct fib6_info __rcu **rtp, struct nl_info *info)
1748 {
1749 	struct fib6_walker *w;
1750 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1751 				    lockdep_is_held(&table->tb6_lock));
1752 	struct net *net = info->nl_net;
1753 
1754 	RT6_TRACE("fib6_del_route\n");
1755 
1756 	/* Unlink it */
1757 	*rtp = rt->fib6_next;
1758 	rt->fib6_node = NULL;
1759 	net->ipv6.rt6_stats->fib_rt_entries--;
1760 	net->ipv6.rt6_stats->fib_discarded_routes++;
1761 
1762 	/* Flush all cached dst in exception table */
1763 	rt6_flush_exceptions(rt);
1764 
1765 	/* Reset round-robin state, if necessary */
1766 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1767 		fn->rr_ptr = NULL;
1768 
1769 	/* Remove this entry from other siblings */
1770 	if (rt->fib6_nsiblings) {
1771 		struct fib6_info *sibling, *next_sibling;
1772 
1773 		list_for_each_entry_safe(sibling, next_sibling,
1774 					 &rt->fib6_siblings, fib6_siblings)
1775 			sibling->fib6_nsiblings--;
1776 		rt->fib6_nsiblings = 0;
1777 		list_del_init(&rt->fib6_siblings);
1778 		rt6_multipath_rebalance(next_sibling);
1779 	}
1780 
1781 	/* Adjust walkers */
1782 	read_lock(&net->ipv6.fib6_walker_lock);
1783 	FOR_WALKERS(net, w) {
1784 		if (w->state == FWS_C && w->leaf == rt) {
1785 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1786 			w->leaf = rcu_dereference_protected(rt->fib6_next,
1787 					    lockdep_is_held(&table->tb6_lock));
1788 			if (!w->leaf)
1789 				w->state = FWS_U;
1790 		}
1791 	}
1792 	read_unlock(&net->ipv6.fib6_walker_lock);
1793 
1794 	/* If it was last route, call fib6_repair_tree() to:
1795 	 * 1. For root node, put back null_entry as how the table was created.
1796 	 * 2. For other nodes, expunge its radix tree node.
1797 	 */
1798 	if (!rcu_access_pointer(fn->leaf)) {
1799 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1800 			fn->fn_flags &= ~RTN_RTINFO;
1801 			net->ipv6.rt6_stats->fib_route_nodes--;
1802 		}
1803 		fn = fib6_repair_tree(net, table, fn);
1804 	}
1805 
1806 	fib6_purge_rt(rt, fn, net);
1807 
1808 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1809 	if (!info->skip_notify)
1810 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1811 	fib6_info_release(rt);
1812 }
1813 
1814 /* Need to own table->tb6_lock */
1815 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1816 {
1817 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1818 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1819 	struct fib6_table *table = rt->fib6_table;
1820 	struct net *net = info->nl_net;
1821 	struct fib6_info __rcu **rtp;
1822 	struct fib6_info __rcu **rtp_next;
1823 
1824 	if (!fn || rt == net->ipv6.fib6_null_entry)
1825 		return -ENOENT;
1826 
1827 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1828 
1829 	/*
1830 	 *	Walk the leaf entries looking for ourself
1831 	 */
1832 
1833 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1834 		struct fib6_info *cur = rcu_dereference_protected(*rtp,
1835 					lockdep_is_held(&table->tb6_lock));
1836 		if (rt == cur) {
1837 			fib6_del_route(table, fn, rtp, info);
1838 			return 0;
1839 		}
1840 		rtp_next = &cur->fib6_next;
1841 	}
1842 	return -ENOENT;
1843 }
1844 
1845 /*
1846  *	Tree traversal function.
1847  *
1848  *	Certainly, it is not interrupt safe.
1849  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1850  *	It means, that we can modify tree during walking
1851  *	and use this function for garbage collection, clone pruning,
1852  *	cleaning tree when a device goes down etc. etc.
1853  *
1854  *	It guarantees that every node will be traversed,
1855  *	and that it will be traversed only once.
1856  *
1857  *	Callback function w->func may return:
1858  *	0 -> continue walking.
1859  *	positive value -> walking is suspended (used by tree dumps,
1860  *	and probably by gc, if it will be split to several slices)
1861  *	negative value -> terminate walking.
1862  *
1863  *	The function itself returns:
1864  *	0   -> walk is complete.
1865  *	>0  -> walk is incomplete (i.e. suspended)
1866  *	<0  -> walk is terminated by an error.
1867  *
1868  *	This function is called with tb6_lock held.
1869  */
1870 
1871 static int fib6_walk_continue(struct fib6_walker *w)
1872 {
1873 	struct fib6_node *fn, *pn, *left, *right;
1874 
1875 	/* w->root should always be table->tb6_root */
1876 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1877 
1878 	for (;;) {
1879 		fn = w->node;
1880 		if (!fn)
1881 			return 0;
1882 
1883 		switch (w->state) {
1884 #ifdef CONFIG_IPV6_SUBTREES
1885 		case FWS_S:
1886 			if (FIB6_SUBTREE(fn)) {
1887 				w->node = FIB6_SUBTREE(fn);
1888 				continue;
1889 			}
1890 			w->state = FWS_L;
1891 #endif
1892 			/* fall through */
1893 		case FWS_L:
1894 			left = rcu_dereference_protected(fn->left, 1);
1895 			if (left) {
1896 				w->node = left;
1897 				w->state = FWS_INIT;
1898 				continue;
1899 			}
1900 			w->state = FWS_R;
1901 			/* fall through */
1902 		case FWS_R:
1903 			right = rcu_dereference_protected(fn->right, 1);
1904 			if (right) {
1905 				w->node = right;
1906 				w->state = FWS_INIT;
1907 				continue;
1908 			}
1909 			w->state = FWS_C;
1910 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1911 			/* fall through */
1912 		case FWS_C:
1913 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1914 				int err;
1915 
1916 				if (w->skip) {
1917 					w->skip--;
1918 					goto skip;
1919 				}
1920 
1921 				err = w->func(w);
1922 				if (err)
1923 					return err;
1924 
1925 				w->count++;
1926 				continue;
1927 			}
1928 skip:
1929 			w->state = FWS_U;
1930 			/* fall through */
1931 		case FWS_U:
1932 			if (fn == w->root)
1933 				return 0;
1934 			pn = rcu_dereference_protected(fn->parent, 1);
1935 			left = rcu_dereference_protected(pn->left, 1);
1936 			right = rcu_dereference_protected(pn->right, 1);
1937 			w->node = pn;
1938 #ifdef CONFIG_IPV6_SUBTREES
1939 			if (FIB6_SUBTREE(pn) == fn) {
1940 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1941 				w->state = FWS_L;
1942 				continue;
1943 			}
1944 #endif
1945 			if (left == fn) {
1946 				w->state = FWS_R;
1947 				continue;
1948 			}
1949 			if (right == fn) {
1950 				w->state = FWS_C;
1951 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1952 				continue;
1953 			}
1954 #if RT6_DEBUG >= 2
1955 			WARN_ON(1);
1956 #endif
1957 		}
1958 	}
1959 }
1960 
1961 static int fib6_walk(struct net *net, struct fib6_walker *w)
1962 {
1963 	int res;
1964 
1965 	w->state = FWS_INIT;
1966 	w->node = w->root;
1967 
1968 	fib6_walker_link(net, w);
1969 	res = fib6_walk_continue(w);
1970 	if (res <= 0)
1971 		fib6_walker_unlink(net, w);
1972 	return res;
1973 }
1974 
1975 static int fib6_clean_node(struct fib6_walker *w)
1976 {
1977 	int res;
1978 	struct fib6_info *rt;
1979 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1980 	struct nl_info info = {
1981 		.nl_net = c->net,
1982 		.skip_notify = c->skip_notify,
1983 	};
1984 
1985 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1986 	    w->node->fn_sernum != c->sernum)
1987 		w->node->fn_sernum = c->sernum;
1988 
1989 	if (!c->func) {
1990 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1991 		w->leaf = NULL;
1992 		return 0;
1993 	}
1994 
1995 	for_each_fib6_walker_rt(w) {
1996 		res = c->func(rt, c->arg);
1997 		if (res == -1) {
1998 			w->leaf = rt;
1999 			res = fib6_del(rt, &info);
2000 			if (res) {
2001 #if RT6_DEBUG >= 2
2002 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2003 					 __func__, rt,
2004 					 rcu_access_pointer(rt->fib6_node),
2005 					 res);
2006 #endif
2007 				continue;
2008 			}
2009 			return 0;
2010 		} else if (res == -2) {
2011 			if (WARN_ON(!rt->fib6_nsiblings))
2012 				continue;
2013 			rt = list_last_entry(&rt->fib6_siblings,
2014 					     struct fib6_info, fib6_siblings);
2015 			continue;
2016 		}
2017 		WARN_ON(res != 0);
2018 	}
2019 	w->leaf = rt;
2020 	return 0;
2021 }
2022 
2023 /*
2024  *	Convenient frontend to tree walker.
2025  *
2026  *	func is called on each route.
2027  *		It may return -2 -> skip multipath route.
2028  *			      -1 -> delete this route.
2029  *		              0  -> continue walking
2030  */
2031 
2032 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2033 			    int (*func)(struct fib6_info *, void *arg),
2034 			    int sernum, void *arg, bool skip_notify)
2035 {
2036 	struct fib6_cleaner c;
2037 
2038 	c.w.root = root;
2039 	c.w.func = fib6_clean_node;
2040 	c.w.count = 0;
2041 	c.w.skip = 0;
2042 	c.func = func;
2043 	c.sernum = sernum;
2044 	c.arg = arg;
2045 	c.net = net;
2046 	c.skip_notify = skip_notify;
2047 
2048 	fib6_walk(net, &c.w);
2049 }
2050 
2051 static void __fib6_clean_all(struct net *net,
2052 			     int (*func)(struct fib6_info *, void *),
2053 			     int sernum, void *arg, bool skip_notify)
2054 {
2055 	struct fib6_table *table;
2056 	struct hlist_head *head;
2057 	unsigned int h;
2058 
2059 	rcu_read_lock();
2060 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2061 		head = &net->ipv6.fib_table_hash[h];
2062 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2063 			spin_lock_bh(&table->tb6_lock);
2064 			fib6_clean_tree(net, &table->tb6_root,
2065 					func, sernum, arg, skip_notify);
2066 			spin_unlock_bh(&table->tb6_lock);
2067 		}
2068 	}
2069 	rcu_read_unlock();
2070 }
2071 
2072 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2073 		    void *arg)
2074 {
2075 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2076 }
2077 
2078 void fib6_clean_all_skip_notify(struct net *net,
2079 				int (*func)(struct fib6_info *, void *),
2080 				void *arg)
2081 {
2082 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2083 }
2084 
2085 static void fib6_flush_trees(struct net *net)
2086 {
2087 	int new_sernum = fib6_new_sernum(net);
2088 
2089 	__fib6_clean_all(net, NULL, new_sernum, NULL, false);
2090 }
2091 
2092 /*
2093  *	Garbage collection
2094  */
2095 
2096 static int fib6_age(struct fib6_info *rt, void *arg)
2097 {
2098 	struct fib6_gc_args *gc_args = arg;
2099 	unsigned long now = jiffies;
2100 
2101 	/*
2102 	 *	check addrconf expiration here.
2103 	 *	Routes are expired even if they are in use.
2104 	 */
2105 
2106 	if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2107 		if (time_after(now, rt->expires)) {
2108 			RT6_TRACE("expiring %p\n", rt);
2109 			return -1;
2110 		}
2111 		gc_args->more++;
2112 	}
2113 
2114 	/*	Also age clones in the exception table.
2115 	 *	Note, that clones are aged out
2116 	 *	only if they are not in use now.
2117 	 */
2118 	rt6_age_exceptions(rt, gc_args, now);
2119 
2120 	return 0;
2121 }
2122 
2123 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2124 {
2125 	struct fib6_gc_args gc_args;
2126 	unsigned long now;
2127 
2128 	if (force) {
2129 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2130 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2131 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2132 		return;
2133 	}
2134 	gc_args.timeout = expires ? (int)expires :
2135 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2136 	gc_args.more = 0;
2137 
2138 	fib6_clean_all(net, fib6_age, &gc_args);
2139 	now = jiffies;
2140 	net->ipv6.ip6_rt_last_gc = now;
2141 
2142 	if (gc_args.more)
2143 		mod_timer(&net->ipv6.ip6_fib_timer,
2144 			  round_jiffies(now
2145 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2146 	else
2147 		del_timer(&net->ipv6.ip6_fib_timer);
2148 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2149 }
2150 
2151 static void fib6_gc_timer_cb(struct timer_list *t)
2152 {
2153 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2154 
2155 	fib6_run_gc(0, arg, true);
2156 }
2157 
2158 static int __net_init fib6_net_init(struct net *net)
2159 {
2160 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2161 	int err;
2162 
2163 	err = fib6_notifier_init(net);
2164 	if (err)
2165 		return err;
2166 
2167 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2168 	rwlock_init(&net->ipv6.fib6_walker_lock);
2169 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2170 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2171 
2172 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2173 	if (!net->ipv6.rt6_stats)
2174 		goto out_timer;
2175 
2176 	/* Avoid false sharing : Use at least a full cache line */
2177 	size = max_t(size_t, size, L1_CACHE_BYTES);
2178 
2179 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2180 	if (!net->ipv6.fib_table_hash)
2181 		goto out_rt6_stats;
2182 
2183 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2184 					  GFP_KERNEL);
2185 	if (!net->ipv6.fib6_main_tbl)
2186 		goto out_fib_table_hash;
2187 
2188 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2189 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2190 			   net->ipv6.fib6_null_entry);
2191 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2192 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2193 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2194 
2195 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2196 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2197 					   GFP_KERNEL);
2198 	if (!net->ipv6.fib6_local_tbl)
2199 		goto out_fib6_main_tbl;
2200 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2201 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2202 			   net->ipv6.fib6_null_entry);
2203 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2204 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2205 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2206 #endif
2207 	fib6_tables_init(net);
2208 
2209 	return 0;
2210 
2211 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2212 out_fib6_main_tbl:
2213 	kfree(net->ipv6.fib6_main_tbl);
2214 #endif
2215 out_fib_table_hash:
2216 	kfree(net->ipv6.fib_table_hash);
2217 out_rt6_stats:
2218 	kfree(net->ipv6.rt6_stats);
2219 out_timer:
2220 	fib6_notifier_exit(net);
2221 	return -ENOMEM;
2222 }
2223 
2224 static void fib6_net_exit(struct net *net)
2225 {
2226 	unsigned int i;
2227 
2228 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2229 
2230 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2231 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2232 		struct hlist_node *tmp;
2233 		struct fib6_table *tb;
2234 
2235 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2236 			hlist_del(&tb->tb6_hlist);
2237 			fib6_free_table(tb);
2238 		}
2239 	}
2240 
2241 	kfree(net->ipv6.fib_table_hash);
2242 	kfree(net->ipv6.rt6_stats);
2243 	fib6_notifier_exit(net);
2244 }
2245 
2246 static struct pernet_operations fib6_net_ops = {
2247 	.init = fib6_net_init,
2248 	.exit = fib6_net_exit,
2249 };
2250 
2251 int __init fib6_init(void)
2252 {
2253 	int ret = -ENOMEM;
2254 
2255 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2256 					   sizeof(struct fib6_node),
2257 					   0, SLAB_HWCACHE_ALIGN,
2258 					   NULL);
2259 	if (!fib6_node_kmem)
2260 		goto out;
2261 
2262 	ret = register_pernet_subsys(&fib6_net_ops);
2263 	if (ret)
2264 		goto out_kmem_cache_create;
2265 
2266 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2267 				   inet6_dump_fib, 0);
2268 	if (ret)
2269 		goto out_unregister_subsys;
2270 
2271 	__fib6_flush_trees = fib6_flush_trees;
2272 out:
2273 	return ret;
2274 
2275 out_unregister_subsys:
2276 	unregister_pernet_subsys(&fib6_net_ops);
2277 out_kmem_cache_create:
2278 	kmem_cache_destroy(fib6_node_kmem);
2279 	goto out;
2280 }
2281 
2282 void fib6_gc_cleanup(void)
2283 {
2284 	unregister_pernet_subsys(&fib6_net_ops);
2285 	kmem_cache_destroy(fib6_node_kmem);
2286 }
2287 
2288 #ifdef CONFIG_PROC_FS
2289 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2290 {
2291 	struct fib6_info *rt = v;
2292 	struct ipv6_route_iter *iter = seq->private;
2293 	unsigned int flags = rt->fib6_flags;
2294 	const struct net_device *dev;
2295 
2296 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2297 
2298 #ifdef CONFIG_IPV6_SUBTREES
2299 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2300 #else
2301 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2302 #endif
2303 	if (rt->fib6_nh.fib_nh_gw_family) {
2304 		flags |= RTF_GATEWAY;
2305 		seq_printf(seq, "%pi6", &rt->fib6_nh.fib_nh_gw6);
2306 	} else {
2307 		seq_puts(seq, "00000000000000000000000000000000");
2308 	}
2309 
2310 	dev = rt->fib6_nh.fib_nh_dev;
2311 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2312 		   rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2313 		   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