xref: /openbmc/linux/net/ipv6/ip6_fib.c (revision 4fc4dca8)
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 	/* Make sure rt6_make_pcpu_route() wont add other percpu routes
908 	 * while we are cleaning them here.
909 	 */
910 	f6i->fib6_destroying = 1;
911 	mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
912 
913 	/* release the reference to this fib entry from
914 	 * all of its cached pcpu routes
915 	 */
916 	for_each_possible_cpu(cpu) {
917 		struct rt6_info **ppcpu_rt;
918 		struct rt6_info *pcpu_rt;
919 
920 		ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
921 		pcpu_rt = *ppcpu_rt;
922 		if (pcpu_rt) {
923 			struct fib6_info *from;
924 
925 			from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
926 			fib6_info_release(from);
927 		}
928 	}
929 }
930 
931 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
932 			  struct net *net)
933 {
934 	struct fib6_table *table = rt->fib6_table;
935 
936 	if (rt->rt6i_pcpu)
937 		fib6_drop_pcpu_from(rt, table);
938 
939 	if (refcount_read(&rt->fib6_ref) != 1) {
940 		/* This route is used as dummy address holder in some split
941 		 * nodes. It is not leaked, but it still holds other resources,
942 		 * which must be released in time. So, scan ascendant nodes
943 		 * and replace dummy references to this route with references
944 		 * to still alive ones.
945 		 */
946 		while (fn) {
947 			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
948 					    lockdep_is_held(&table->tb6_lock));
949 			struct fib6_info *new_leaf;
950 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
951 				new_leaf = fib6_find_prefix(net, table, fn);
952 				fib6_info_hold(new_leaf);
953 
954 				rcu_assign_pointer(fn->leaf, new_leaf);
955 				fib6_info_release(rt);
956 			}
957 			fn = rcu_dereference_protected(fn->parent,
958 				    lockdep_is_held(&table->tb6_lock));
959 		}
960 	}
961 }
962 
963 /*
964  *	Insert routing information in a node.
965  */
966 
967 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
968 			    struct nl_info *info,
969 			    struct netlink_ext_ack *extack)
970 {
971 	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
972 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
973 	struct fib6_info *iter = NULL;
974 	struct fib6_info __rcu **ins;
975 	struct fib6_info __rcu **fallback_ins = NULL;
976 	int replace = (info->nlh &&
977 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
978 	int add = (!info->nlh ||
979 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
980 	int found = 0;
981 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
982 	u16 nlflags = NLM_F_EXCL;
983 	int err;
984 
985 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
986 		nlflags |= NLM_F_APPEND;
987 
988 	ins = &fn->leaf;
989 
990 	for (iter = leaf; iter;
991 	     iter = rcu_dereference_protected(iter->fib6_next,
992 				lockdep_is_held(&rt->fib6_table->tb6_lock))) {
993 		/*
994 		 *	Search for duplicates
995 		 */
996 
997 		if (iter->fib6_metric == rt->fib6_metric) {
998 			/*
999 			 *	Same priority level
1000 			 */
1001 			if (info->nlh &&
1002 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
1003 				return -EEXIST;
1004 
1005 			nlflags &= ~NLM_F_EXCL;
1006 			if (replace) {
1007 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1008 					found++;
1009 					break;
1010 				}
1011 				if (rt_can_ecmp)
1012 					fallback_ins = fallback_ins ?: ins;
1013 				goto next_iter;
1014 			}
1015 
1016 			if (rt6_duplicate_nexthop(iter, rt)) {
1017 				if (rt->fib6_nsiblings)
1018 					rt->fib6_nsiblings = 0;
1019 				if (!(iter->fib6_flags & RTF_EXPIRES))
1020 					return -EEXIST;
1021 				if (!(rt->fib6_flags & RTF_EXPIRES))
1022 					fib6_clean_expires(iter);
1023 				else
1024 					fib6_set_expires(iter, rt->expires);
1025 
1026 				if (rt->fib6_pmtu)
1027 					fib6_metric_set(iter, RTAX_MTU,
1028 							rt->fib6_pmtu);
1029 				return -EEXIST;
1030 			}
1031 			/* If we have the same destination and the same metric,
1032 			 * but not the same gateway, then the route we try to
1033 			 * add is sibling to this route, increment our counter
1034 			 * of siblings, and later we will add our route to the
1035 			 * list.
1036 			 * Only static routes (which don't have flag
1037 			 * RTF_EXPIRES) are used for ECMPv6.
1038 			 *
1039 			 * To avoid long list, we only had siblings if the
1040 			 * route have a gateway.
1041 			 */
1042 			if (rt_can_ecmp &&
1043 			    rt6_qualify_for_ecmp(iter))
1044 				rt->fib6_nsiblings++;
1045 		}
1046 
1047 		if (iter->fib6_metric > rt->fib6_metric)
1048 			break;
1049 
1050 next_iter:
1051 		ins = &iter->fib6_next;
1052 	}
1053 
1054 	if (fallback_ins && !found) {
1055 		/* No ECMP-able route found, replace first non-ECMP one */
1056 		ins = fallback_ins;
1057 		iter = rcu_dereference_protected(*ins,
1058 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1059 		found++;
1060 	}
1061 
1062 	/* Reset round-robin state, if necessary */
1063 	if (ins == &fn->leaf)
1064 		fn->rr_ptr = NULL;
1065 
1066 	/* Link this route to others same route. */
1067 	if (rt->fib6_nsiblings) {
1068 		unsigned int fib6_nsiblings;
1069 		struct fib6_info *sibling, *temp_sibling;
1070 
1071 		/* Find the first route that have the same metric */
1072 		sibling = leaf;
1073 		while (sibling) {
1074 			if (sibling->fib6_metric == rt->fib6_metric &&
1075 			    rt6_qualify_for_ecmp(sibling)) {
1076 				list_add_tail(&rt->fib6_siblings,
1077 					      &sibling->fib6_siblings);
1078 				break;
1079 			}
1080 			sibling = rcu_dereference_protected(sibling->fib6_next,
1081 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1082 		}
1083 		/* For each sibling in the list, increment the counter of
1084 		 * siblings. BUG() if counters does not match, list of siblings
1085 		 * is broken!
1086 		 */
1087 		fib6_nsiblings = 0;
1088 		list_for_each_entry_safe(sibling, temp_sibling,
1089 					 &rt->fib6_siblings, fib6_siblings) {
1090 			sibling->fib6_nsiblings++;
1091 			BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1092 			fib6_nsiblings++;
1093 		}
1094 		BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1095 		rt6_multipath_rebalance(temp_sibling);
1096 	}
1097 
1098 	/*
1099 	 *	insert node
1100 	 */
1101 	if (!replace) {
1102 		if (!add)
1103 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1104 
1105 add:
1106 		nlflags |= NLM_F_CREATE;
1107 
1108 		err = call_fib6_entry_notifiers(info->nl_net,
1109 						FIB_EVENT_ENTRY_ADD,
1110 						rt, extack);
1111 		if (err)
1112 			return err;
1113 
1114 		rcu_assign_pointer(rt->fib6_next, iter);
1115 		fib6_info_hold(rt);
1116 		rcu_assign_pointer(rt->fib6_node, fn);
1117 		rcu_assign_pointer(*ins, rt);
1118 		if (!info->skip_notify)
1119 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1120 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1121 
1122 		if (!(fn->fn_flags & RTN_RTINFO)) {
1123 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1124 			fn->fn_flags |= RTN_RTINFO;
1125 		}
1126 
1127 	} else {
1128 		int nsiblings;
1129 
1130 		if (!found) {
1131 			if (add)
1132 				goto add;
1133 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1134 			return -ENOENT;
1135 		}
1136 
1137 		err = call_fib6_entry_notifiers(info->nl_net,
1138 						FIB_EVENT_ENTRY_REPLACE,
1139 						rt, extack);
1140 		if (err)
1141 			return err;
1142 
1143 		fib6_info_hold(rt);
1144 		rcu_assign_pointer(rt->fib6_node, fn);
1145 		rt->fib6_next = iter->fib6_next;
1146 		rcu_assign_pointer(*ins, rt);
1147 		if (!info->skip_notify)
1148 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1149 		if (!(fn->fn_flags & RTN_RTINFO)) {
1150 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1151 			fn->fn_flags |= RTN_RTINFO;
1152 		}
1153 		nsiblings = iter->fib6_nsiblings;
1154 		iter->fib6_node = NULL;
1155 		fib6_purge_rt(iter, fn, info->nl_net);
1156 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1157 			fn->rr_ptr = NULL;
1158 		fib6_info_release(iter);
1159 
1160 		if (nsiblings) {
1161 			/* Replacing an ECMP route, remove all siblings */
1162 			ins = &rt->fib6_next;
1163 			iter = rcu_dereference_protected(*ins,
1164 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1165 			while (iter) {
1166 				if (iter->fib6_metric > rt->fib6_metric)
1167 					break;
1168 				if (rt6_qualify_for_ecmp(iter)) {
1169 					*ins = iter->fib6_next;
1170 					iter->fib6_node = NULL;
1171 					fib6_purge_rt(iter, fn, info->nl_net);
1172 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1173 						fn->rr_ptr = NULL;
1174 					fib6_info_release(iter);
1175 					nsiblings--;
1176 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1177 				} else {
1178 					ins = &iter->fib6_next;
1179 				}
1180 				iter = rcu_dereference_protected(*ins,
1181 					lockdep_is_held(&rt->fib6_table->tb6_lock));
1182 			}
1183 			WARN_ON(nsiblings != 0);
1184 		}
1185 	}
1186 
1187 	return 0;
1188 }
1189 
1190 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1191 {
1192 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1193 	    (rt->fib6_flags & RTF_EXPIRES))
1194 		mod_timer(&net->ipv6.ip6_fib_timer,
1195 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1196 }
1197 
1198 void fib6_force_start_gc(struct net *net)
1199 {
1200 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1201 		mod_timer(&net->ipv6.ip6_fib_timer,
1202 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1203 }
1204 
1205 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1206 					   int sernum)
1207 {
1208 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1209 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1210 
1211 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1212 	smp_wmb();
1213 	while (fn) {
1214 		fn->fn_sernum = sernum;
1215 		fn = rcu_dereference_protected(fn->parent,
1216 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1217 	}
1218 }
1219 
1220 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1221 {
1222 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1223 }
1224 
1225 /*
1226  *	Add routing information to the routing tree.
1227  *	<destination addr>/<source addr>
1228  *	with source addr info in sub-trees
1229  *	Need to own table->tb6_lock
1230  */
1231 
1232 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1233 	     struct nl_info *info, struct netlink_ext_ack *extack)
1234 {
1235 	struct fib6_table *table = rt->fib6_table;
1236 	struct fib6_node *fn, *pn = NULL;
1237 	int err = -ENOMEM;
1238 	int allow_create = 1;
1239 	int replace_required = 0;
1240 	int sernum = fib6_new_sernum(info->nl_net);
1241 
1242 	if (info->nlh) {
1243 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1244 			allow_create = 0;
1245 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1246 			replace_required = 1;
1247 	}
1248 	if (!allow_create && !replace_required)
1249 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1250 
1251 	fn = fib6_add_1(info->nl_net, table, root,
1252 			&rt->fib6_dst.addr, rt->fib6_dst.plen,
1253 			offsetof(struct fib6_info, fib6_dst), allow_create,
1254 			replace_required, extack);
1255 	if (IS_ERR(fn)) {
1256 		err = PTR_ERR(fn);
1257 		fn = NULL;
1258 		goto out;
1259 	}
1260 
1261 	pn = fn;
1262 
1263 #ifdef CONFIG_IPV6_SUBTREES
1264 	if (rt->fib6_src.plen) {
1265 		struct fib6_node *sn;
1266 
1267 		if (!rcu_access_pointer(fn->subtree)) {
1268 			struct fib6_node *sfn;
1269 
1270 			/*
1271 			 * Create subtree.
1272 			 *
1273 			 *		fn[main tree]
1274 			 *		|
1275 			 *		sfn[subtree root]
1276 			 *		   \
1277 			 *		    sn[new leaf node]
1278 			 */
1279 
1280 			/* Create subtree root node */
1281 			sfn = node_alloc(info->nl_net);
1282 			if (!sfn)
1283 				goto failure;
1284 
1285 			fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1286 			rcu_assign_pointer(sfn->leaf,
1287 					   info->nl_net->ipv6.fib6_null_entry);
1288 			sfn->fn_flags = RTN_ROOT;
1289 
1290 			/* Now add the first leaf node to new subtree */
1291 
1292 			sn = fib6_add_1(info->nl_net, table, sfn,
1293 					&rt->fib6_src.addr, rt->fib6_src.plen,
1294 					offsetof(struct fib6_info, fib6_src),
1295 					allow_create, replace_required, extack);
1296 
1297 			if (IS_ERR(sn)) {
1298 				/* If it is failed, discard just allocated
1299 				   root, and then (in failure) stale node
1300 				   in main tree.
1301 				 */
1302 				node_free_immediate(info->nl_net, sfn);
1303 				err = PTR_ERR(sn);
1304 				goto failure;
1305 			}
1306 
1307 			/* Now link new subtree to main tree */
1308 			rcu_assign_pointer(sfn->parent, fn);
1309 			rcu_assign_pointer(fn->subtree, sfn);
1310 		} else {
1311 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1312 					&rt->fib6_src.addr, rt->fib6_src.plen,
1313 					offsetof(struct fib6_info, fib6_src),
1314 					allow_create, replace_required, extack);
1315 
1316 			if (IS_ERR(sn)) {
1317 				err = PTR_ERR(sn);
1318 				goto failure;
1319 			}
1320 		}
1321 
1322 		if (!rcu_access_pointer(fn->leaf)) {
1323 			if (fn->fn_flags & RTN_TL_ROOT) {
1324 				/* put back null_entry for root node */
1325 				rcu_assign_pointer(fn->leaf,
1326 					    info->nl_net->ipv6.fib6_null_entry);
1327 			} else {
1328 				fib6_info_hold(rt);
1329 				rcu_assign_pointer(fn->leaf, rt);
1330 			}
1331 		}
1332 		fn = sn;
1333 	}
1334 #endif
1335 
1336 	err = fib6_add_rt2node(fn, rt, info, extack);
1337 	if (!err) {
1338 		__fib6_update_sernum_upto_root(rt, sernum);
1339 		fib6_start_gc(info->nl_net, rt);
1340 	}
1341 
1342 out:
1343 	if (err) {
1344 #ifdef CONFIG_IPV6_SUBTREES
1345 		/*
1346 		 * If fib6_add_1 has cleared the old leaf pointer in the
1347 		 * super-tree leaf node we have to find a new one for it.
1348 		 */
1349 		if (pn != fn) {
1350 			struct fib6_info *pn_leaf =
1351 				rcu_dereference_protected(pn->leaf,
1352 				    lockdep_is_held(&table->tb6_lock));
1353 			if (pn_leaf == rt) {
1354 				pn_leaf = NULL;
1355 				RCU_INIT_POINTER(pn->leaf, NULL);
1356 				fib6_info_release(rt);
1357 			}
1358 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1359 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1360 							   pn);
1361 #if RT6_DEBUG >= 2
1362 				if (!pn_leaf) {
1363 					WARN_ON(!pn_leaf);
1364 					pn_leaf =
1365 					    info->nl_net->ipv6.fib6_null_entry;
1366 				}
1367 #endif
1368 				fib6_info_hold(pn_leaf);
1369 				rcu_assign_pointer(pn->leaf, pn_leaf);
1370 			}
1371 		}
1372 #endif
1373 		goto failure;
1374 	}
1375 	return err;
1376 
1377 failure:
1378 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1379 	 * 1. fn is an intermediate node and we failed to add the new
1380 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1381 	 * failure case.
1382 	 * 2. fn is the root node in the table and we fail to add the first
1383 	 * default route to it.
1384 	 */
1385 	if (fn &&
1386 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1387 	     (fn->fn_flags & RTN_TL_ROOT &&
1388 	      !rcu_access_pointer(fn->leaf))))
1389 		fib6_repair_tree(info->nl_net, table, fn);
1390 	return err;
1391 }
1392 
1393 /*
1394  *	Routing tree lookup
1395  *
1396  */
1397 
1398 struct lookup_args {
1399 	int			offset;		/* key offset on fib6_info */
1400 	const struct in6_addr	*addr;		/* search key			*/
1401 };
1402 
1403 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1404 					    struct lookup_args *args)
1405 {
1406 	struct fib6_node *fn;
1407 	__be32 dir;
1408 
1409 	if (unlikely(args->offset == 0))
1410 		return NULL;
1411 
1412 	/*
1413 	 *	Descend on a tree
1414 	 */
1415 
1416 	fn = root;
1417 
1418 	for (;;) {
1419 		struct fib6_node *next;
1420 
1421 		dir = addr_bit_set(args->addr, fn->fn_bit);
1422 
1423 		next = dir ? rcu_dereference(fn->right) :
1424 			     rcu_dereference(fn->left);
1425 
1426 		if (next) {
1427 			fn = next;
1428 			continue;
1429 		}
1430 		break;
1431 	}
1432 
1433 	while (fn) {
1434 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1435 
1436 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1437 			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1438 			struct rt6key *key;
1439 
1440 			if (!leaf)
1441 				goto backtrack;
1442 
1443 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1444 
1445 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1446 #ifdef CONFIG_IPV6_SUBTREES
1447 				if (subtree) {
1448 					struct fib6_node *sfn;
1449 					sfn = fib6_node_lookup_1(subtree,
1450 								 args + 1);
1451 					if (!sfn)
1452 						goto backtrack;
1453 					fn = sfn;
1454 				}
1455 #endif
1456 				if (fn->fn_flags & RTN_RTINFO)
1457 					return fn;
1458 			}
1459 		}
1460 backtrack:
1461 		if (fn->fn_flags & RTN_ROOT)
1462 			break;
1463 
1464 		fn = rcu_dereference(fn->parent);
1465 	}
1466 
1467 	return NULL;
1468 }
1469 
1470 /* called with rcu_read_lock() held
1471  */
1472 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1473 				   const struct in6_addr *daddr,
1474 				   const struct in6_addr *saddr)
1475 {
1476 	struct fib6_node *fn;
1477 	struct lookup_args args[] = {
1478 		{
1479 			.offset = offsetof(struct fib6_info, fib6_dst),
1480 			.addr = daddr,
1481 		},
1482 #ifdef CONFIG_IPV6_SUBTREES
1483 		{
1484 			.offset = offsetof(struct fib6_info, fib6_src),
1485 			.addr = saddr,
1486 		},
1487 #endif
1488 		{
1489 			.offset = 0,	/* sentinel */
1490 		}
1491 	};
1492 
1493 	fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1494 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1495 		fn = root;
1496 
1497 	return fn;
1498 }
1499 
1500 /*
1501  *	Get node with specified destination prefix (and source prefix,
1502  *	if subtrees are used)
1503  *	exact_match == true means we try to find fn with exact match of
1504  *	the passed in prefix addr
1505  *	exact_match == false means we try to find fn with longest prefix
1506  *	match of the passed in prefix addr. This is useful for finding fn
1507  *	for cached route as it will be stored in the exception table under
1508  *	the node with longest prefix length.
1509  */
1510 
1511 
1512 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1513 				       const struct in6_addr *addr,
1514 				       int plen, int offset,
1515 				       bool exact_match)
1516 {
1517 	struct fib6_node *fn, *prev = NULL;
1518 
1519 	for (fn = root; fn ; ) {
1520 		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1521 		struct rt6key *key;
1522 
1523 		/* This node is being deleted */
1524 		if (!leaf) {
1525 			if (plen <= fn->fn_bit)
1526 				goto out;
1527 			else
1528 				goto next;
1529 		}
1530 
1531 		key = (struct rt6key *)((u8 *)leaf + offset);
1532 
1533 		/*
1534 		 *	Prefix match
1535 		 */
1536 		if (plen < fn->fn_bit ||
1537 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1538 			goto out;
1539 
1540 		if (plen == fn->fn_bit)
1541 			return fn;
1542 
1543 		prev = fn;
1544 
1545 next:
1546 		/*
1547 		 *	We have more bits to go
1548 		 */
1549 		if (addr_bit_set(addr, fn->fn_bit))
1550 			fn = rcu_dereference(fn->right);
1551 		else
1552 			fn = rcu_dereference(fn->left);
1553 	}
1554 out:
1555 	if (exact_match)
1556 		return NULL;
1557 	else
1558 		return prev;
1559 }
1560 
1561 struct fib6_node *fib6_locate(struct fib6_node *root,
1562 			      const struct in6_addr *daddr, int dst_len,
1563 			      const struct in6_addr *saddr, int src_len,
1564 			      bool exact_match)
1565 {
1566 	struct fib6_node *fn;
1567 
1568 	fn = fib6_locate_1(root, daddr, dst_len,
1569 			   offsetof(struct fib6_info, fib6_dst),
1570 			   exact_match);
1571 
1572 #ifdef CONFIG_IPV6_SUBTREES
1573 	if (src_len) {
1574 		WARN_ON(saddr == NULL);
1575 		if (fn) {
1576 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1577 
1578 			if (subtree) {
1579 				fn = fib6_locate_1(subtree, saddr, src_len,
1580 					   offsetof(struct fib6_info, fib6_src),
1581 					   exact_match);
1582 			}
1583 		}
1584 	}
1585 #endif
1586 
1587 	if (fn && fn->fn_flags & RTN_RTINFO)
1588 		return fn;
1589 
1590 	return NULL;
1591 }
1592 
1593 
1594 /*
1595  *	Deletion
1596  *
1597  */
1598 
1599 static struct fib6_info *fib6_find_prefix(struct net *net,
1600 					 struct fib6_table *table,
1601 					 struct fib6_node *fn)
1602 {
1603 	struct fib6_node *child_left, *child_right;
1604 
1605 	if (fn->fn_flags & RTN_ROOT)
1606 		return net->ipv6.fib6_null_entry;
1607 
1608 	while (fn) {
1609 		child_left = rcu_dereference_protected(fn->left,
1610 				    lockdep_is_held(&table->tb6_lock));
1611 		child_right = rcu_dereference_protected(fn->right,
1612 				    lockdep_is_held(&table->tb6_lock));
1613 		if (child_left)
1614 			return rcu_dereference_protected(child_left->leaf,
1615 					lockdep_is_held(&table->tb6_lock));
1616 		if (child_right)
1617 			return rcu_dereference_protected(child_right->leaf,
1618 					lockdep_is_held(&table->tb6_lock));
1619 
1620 		fn = FIB6_SUBTREE(fn);
1621 	}
1622 	return NULL;
1623 }
1624 
1625 /*
1626  *	Called to trim the tree of intermediate nodes when possible. "fn"
1627  *	is the node we want to try and remove.
1628  *	Need to own table->tb6_lock
1629  */
1630 
1631 static struct fib6_node *fib6_repair_tree(struct net *net,
1632 					  struct fib6_table *table,
1633 					  struct fib6_node *fn)
1634 {
1635 	int children;
1636 	int nstate;
1637 	struct fib6_node *child;
1638 	struct fib6_walker *w;
1639 	int iter = 0;
1640 
1641 	/* Set fn->leaf to null_entry for root node. */
1642 	if (fn->fn_flags & RTN_TL_ROOT) {
1643 		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1644 		return fn;
1645 	}
1646 
1647 	for (;;) {
1648 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1649 					    lockdep_is_held(&table->tb6_lock));
1650 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1651 					    lockdep_is_held(&table->tb6_lock));
1652 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1653 					    lockdep_is_held(&table->tb6_lock));
1654 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1655 					    lockdep_is_held(&table->tb6_lock));
1656 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1657 					    lockdep_is_held(&table->tb6_lock));
1658 		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1659 					    lockdep_is_held(&table->tb6_lock));
1660 		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1661 					    lockdep_is_held(&table->tb6_lock));
1662 		struct fib6_info *new_fn_leaf;
1663 
1664 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1665 		iter++;
1666 
1667 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1668 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1669 		WARN_ON(fn_leaf);
1670 
1671 		children = 0;
1672 		child = NULL;
1673 		if (fn_r)
1674 			child = fn_r, children |= 1;
1675 		if (fn_l)
1676 			child = fn_l, children |= 2;
1677 
1678 		if (children == 3 || FIB6_SUBTREE(fn)
1679 #ifdef CONFIG_IPV6_SUBTREES
1680 		    /* Subtree root (i.e. fn) may have one child */
1681 		    || (children && fn->fn_flags & RTN_ROOT)
1682 #endif
1683 		    ) {
1684 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1685 #if RT6_DEBUG >= 2
1686 			if (!new_fn_leaf) {
1687 				WARN_ON(!new_fn_leaf);
1688 				new_fn_leaf = net->ipv6.fib6_null_entry;
1689 			}
1690 #endif
1691 			fib6_info_hold(new_fn_leaf);
1692 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1693 			return pn;
1694 		}
1695 
1696 #ifdef CONFIG_IPV6_SUBTREES
1697 		if (FIB6_SUBTREE(pn) == fn) {
1698 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1699 			RCU_INIT_POINTER(pn->subtree, NULL);
1700 			nstate = FWS_L;
1701 		} else {
1702 			WARN_ON(fn->fn_flags & RTN_ROOT);
1703 #endif
1704 			if (pn_r == fn)
1705 				rcu_assign_pointer(pn->right, child);
1706 			else if (pn_l == fn)
1707 				rcu_assign_pointer(pn->left, child);
1708 #if RT6_DEBUG >= 2
1709 			else
1710 				WARN_ON(1);
1711 #endif
1712 			if (child)
1713 				rcu_assign_pointer(child->parent, pn);
1714 			nstate = FWS_R;
1715 #ifdef CONFIG_IPV6_SUBTREES
1716 		}
1717 #endif
1718 
1719 		read_lock(&net->ipv6.fib6_walker_lock);
1720 		FOR_WALKERS(net, w) {
1721 			if (!child) {
1722 				if (w->node == fn) {
1723 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1724 					w->node = pn;
1725 					w->state = nstate;
1726 				}
1727 			} else {
1728 				if (w->node == fn) {
1729 					w->node = child;
1730 					if (children&2) {
1731 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1732 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1733 					} else {
1734 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1735 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1736 					}
1737 				}
1738 			}
1739 		}
1740 		read_unlock(&net->ipv6.fib6_walker_lock);
1741 
1742 		node_free(net, fn);
1743 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1744 			return pn;
1745 
1746 		RCU_INIT_POINTER(pn->leaf, NULL);
1747 		fib6_info_release(pn_leaf);
1748 		fn = pn;
1749 	}
1750 }
1751 
1752 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1753 			   struct fib6_info __rcu **rtp, struct nl_info *info)
1754 {
1755 	struct fib6_walker *w;
1756 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1757 				    lockdep_is_held(&table->tb6_lock));
1758 	struct net *net = info->nl_net;
1759 
1760 	RT6_TRACE("fib6_del_route\n");
1761 
1762 	/* Unlink it */
1763 	*rtp = rt->fib6_next;
1764 	rt->fib6_node = NULL;
1765 	net->ipv6.rt6_stats->fib_rt_entries--;
1766 	net->ipv6.rt6_stats->fib_discarded_routes++;
1767 
1768 	/* Flush all cached dst in exception table */
1769 	rt6_flush_exceptions(rt);
1770 
1771 	/* Reset round-robin state, if necessary */
1772 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1773 		fn->rr_ptr = NULL;
1774 
1775 	/* Remove this entry from other siblings */
1776 	if (rt->fib6_nsiblings) {
1777 		struct fib6_info *sibling, *next_sibling;
1778 
1779 		list_for_each_entry_safe(sibling, next_sibling,
1780 					 &rt->fib6_siblings, fib6_siblings)
1781 			sibling->fib6_nsiblings--;
1782 		rt->fib6_nsiblings = 0;
1783 		list_del_init(&rt->fib6_siblings);
1784 		rt6_multipath_rebalance(next_sibling);
1785 	}
1786 
1787 	/* Adjust walkers */
1788 	read_lock(&net->ipv6.fib6_walker_lock);
1789 	FOR_WALKERS(net, w) {
1790 		if (w->state == FWS_C && w->leaf == rt) {
1791 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1792 			w->leaf = rcu_dereference_protected(rt->fib6_next,
1793 					    lockdep_is_held(&table->tb6_lock));
1794 			if (!w->leaf)
1795 				w->state = FWS_U;
1796 		}
1797 	}
1798 	read_unlock(&net->ipv6.fib6_walker_lock);
1799 
1800 	/* If it was last route, call fib6_repair_tree() to:
1801 	 * 1. For root node, put back null_entry as how the table was created.
1802 	 * 2. For other nodes, expunge its radix tree node.
1803 	 */
1804 	if (!rcu_access_pointer(fn->leaf)) {
1805 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1806 			fn->fn_flags &= ~RTN_RTINFO;
1807 			net->ipv6.rt6_stats->fib_route_nodes--;
1808 		}
1809 		fn = fib6_repair_tree(net, table, fn);
1810 	}
1811 
1812 	fib6_purge_rt(rt, fn, net);
1813 
1814 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1815 	if (!info->skip_notify)
1816 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1817 	fib6_info_release(rt);
1818 }
1819 
1820 /* Need to own table->tb6_lock */
1821 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1822 {
1823 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1824 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1825 	struct fib6_table *table = rt->fib6_table;
1826 	struct net *net = info->nl_net;
1827 	struct fib6_info __rcu **rtp;
1828 	struct fib6_info __rcu **rtp_next;
1829 
1830 	if (!fn || rt == net->ipv6.fib6_null_entry)
1831 		return -ENOENT;
1832 
1833 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1834 
1835 	/*
1836 	 *	Walk the leaf entries looking for ourself
1837 	 */
1838 
1839 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1840 		struct fib6_info *cur = rcu_dereference_protected(*rtp,
1841 					lockdep_is_held(&table->tb6_lock));
1842 		if (rt == cur) {
1843 			fib6_del_route(table, fn, rtp, info);
1844 			return 0;
1845 		}
1846 		rtp_next = &cur->fib6_next;
1847 	}
1848 	return -ENOENT;
1849 }
1850 
1851 /*
1852  *	Tree traversal function.
1853  *
1854  *	Certainly, it is not interrupt safe.
1855  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1856  *	It means, that we can modify tree during walking
1857  *	and use this function for garbage collection, clone pruning,
1858  *	cleaning tree when a device goes down etc. etc.
1859  *
1860  *	It guarantees that every node will be traversed,
1861  *	and that it will be traversed only once.
1862  *
1863  *	Callback function w->func may return:
1864  *	0 -> continue walking.
1865  *	positive value -> walking is suspended (used by tree dumps,
1866  *	and probably by gc, if it will be split to several slices)
1867  *	negative value -> terminate walking.
1868  *
1869  *	The function itself returns:
1870  *	0   -> walk is complete.
1871  *	>0  -> walk is incomplete (i.e. suspended)
1872  *	<0  -> walk is terminated by an error.
1873  *
1874  *	This function is called with tb6_lock held.
1875  */
1876 
1877 static int fib6_walk_continue(struct fib6_walker *w)
1878 {
1879 	struct fib6_node *fn, *pn, *left, *right;
1880 
1881 	/* w->root should always be table->tb6_root */
1882 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1883 
1884 	for (;;) {
1885 		fn = w->node;
1886 		if (!fn)
1887 			return 0;
1888 
1889 		switch (w->state) {
1890 #ifdef CONFIG_IPV6_SUBTREES
1891 		case FWS_S:
1892 			if (FIB6_SUBTREE(fn)) {
1893 				w->node = FIB6_SUBTREE(fn);
1894 				continue;
1895 			}
1896 			w->state = FWS_L;
1897 #endif
1898 			/* fall through */
1899 		case FWS_L:
1900 			left = rcu_dereference_protected(fn->left, 1);
1901 			if (left) {
1902 				w->node = left;
1903 				w->state = FWS_INIT;
1904 				continue;
1905 			}
1906 			w->state = FWS_R;
1907 			/* fall through */
1908 		case FWS_R:
1909 			right = rcu_dereference_protected(fn->right, 1);
1910 			if (right) {
1911 				w->node = right;
1912 				w->state = FWS_INIT;
1913 				continue;
1914 			}
1915 			w->state = FWS_C;
1916 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1917 			/* fall through */
1918 		case FWS_C:
1919 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1920 				int err;
1921 
1922 				if (w->skip) {
1923 					w->skip--;
1924 					goto skip;
1925 				}
1926 
1927 				err = w->func(w);
1928 				if (err)
1929 					return err;
1930 
1931 				w->count++;
1932 				continue;
1933 			}
1934 skip:
1935 			w->state = FWS_U;
1936 			/* fall through */
1937 		case FWS_U:
1938 			if (fn == w->root)
1939 				return 0;
1940 			pn = rcu_dereference_protected(fn->parent, 1);
1941 			left = rcu_dereference_protected(pn->left, 1);
1942 			right = rcu_dereference_protected(pn->right, 1);
1943 			w->node = pn;
1944 #ifdef CONFIG_IPV6_SUBTREES
1945 			if (FIB6_SUBTREE(pn) == fn) {
1946 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1947 				w->state = FWS_L;
1948 				continue;
1949 			}
1950 #endif
1951 			if (left == fn) {
1952 				w->state = FWS_R;
1953 				continue;
1954 			}
1955 			if (right == fn) {
1956 				w->state = FWS_C;
1957 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1958 				continue;
1959 			}
1960 #if RT6_DEBUG >= 2
1961 			WARN_ON(1);
1962 #endif
1963 		}
1964 	}
1965 }
1966 
1967 static int fib6_walk(struct net *net, struct fib6_walker *w)
1968 {
1969 	int res;
1970 
1971 	w->state = FWS_INIT;
1972 	w->node = w->root;
1973 
1974 	fib6_walker_link(net, w);
1975 	res = fib6_walk_continue(w);
1976 	if (res <= 0)
1977 		fib6_walker_unlink(net, w);
1978 	return res;
1979 }
1980 
1981 static int fib6_clean_node(struct fib6_walker *w)
1982 {
1983 	int res;
1984 	struct fib6_info *rt;
1985 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1986 	struct nl_info info = {
1987 		.nl_net = c->net,
1988 		.skip_notify = c->skip_notify,
1989 	};
1990 
1991 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1992 	    w->node->fn_sernum != c->sernum)
1993 		w->node->fn_sernum = c->sernum;
1994 
1995 	if (!c->func) {
1996 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1997 		w->leaf = NULL;
1998 		return 0;
1999 	}
2000 
2001 	for_each_fib6_walker_rt(w) {
2002 		res = c->func(rt, c->arg);
2003 		if (res == -1) {
2004 			w->leaf = rt;
2005 			res = fib6_del(rt, &info);
2006 			if (res) {
2007 #if RT6_DEBUG >= 2
2008 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2009 					 __func__, rt,
2010 					 rcu_access_pointer(rt->fib6_node),
2011 					 res);
2012 #endif
2013 				continue;
2014 			}
2015 			return 0;
2016 		} else if (res == -2) {
2017 			if (WARN_ON(!rt->fib6_nsiblings))
2018 				continue;
2019 			rt = list_last_entry(&rt->fib6_siblings,
2020 					     struct fib6_info, fib6_siblings);
2021 			continue;
2022 		}
2023 		WARN_ON(res != 0);
2024 	}
2025 	w->leaf = rt;
2026 	return 0;
2027 }
2028 
2029 /*
2030  *	Convenient frontend to tree walker.
2031  *
2032  *	func is called on each route.
2033  *		It may return -2 -> skip multipath route.
2034  *			      -1 -> delete this route.
2035  *		              0  -> continue walking
2036  */
2037 
2038 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2039 			    int (*func)(struct fib6_info *, void *arg),
2040 			    int sernum, void *arg, bool skip_notify)
2041 {
2042 	struct fib6_cleaner c;
2043 
2044 	c.w.root = root;
2045 	c.w.func = fib6_clean_node;
2046 	c.w.count = 0;
2047 	c.w.skip = 0;
2048 	c.func = func;
2049 	c.sernum = sernum;
2050 	c.arg = arg;
2051 	c.net = net;
2052 	c.skip_notify = skip_notify;
2053 
2054 	fib6_walk(net, &c.w);
2055 }
2056 
2057 static void __fib6_clean_all(struct net *net,
2058 			     int (*func)(struct fib6_info *, void *),
2059 			     int sernum, void *arg, bool skip_notify)
2060 {
2061 	struct fib6_table *table;
2062 	struct hlist_head *head;
2063 	unsigned int h;
2064 
2065 	rcu_read_lock();
2066 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2067 		head = &net->ipv6.fib_table_hash[h];
2068 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2069 			spin_lock_bh(&table->tb6_lock);
2070 			fib6_clean_tree(net, &table->tb6_root,
2071 					func, sernum, arg, skip_notify);
2072 			spin_unlock_bh(&table->tb6_lock);
2073 		}
2074 	}
2075 	rcu_read_unlock();
2076 }
2077 
2078 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2079 		    void *arg)
2080 {
2081 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2082 }
2083 
2084 void fib6_clean_all_skip_notify(struct net *net,
2085 				int (*func)(struct fib6_info *, void *),
2086 				void *arg)
2087 {
2088 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2089 }
2090 
2091 static void fib6_flush_trees(struct net *net)
2092 {
2093 	int new_sernum = fib6_new_sernum(net);
2094 
2095 	__fib6_clean_all(net, NULL, new_sernum, NULL, false);
2096 }
2097 
2098 /*
2099  *	Garbage collection
2100  */
2101 
2102 static int fib6_age(struct fib6_info *rt, void *arg)
2103 {
2104 	struct fib6_gc_args *gc_args = arg;
2105 	unsigned long now = jiffies;
2106 
2107 	/*
2108 	 *	check addrconf expiration here.
2109 	 *	Routes are expired even if they are in use.
2110 	 */
2111 
2112 	if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2113 		if (time_after(now, rt->expires)) {
2114 			RT6_TRACE("expiring %p\n", rt);
2115 			return -1;
2116 		}
2117 		gc_args->more++;
2118 	}
2119 
2120 	/*	Also age clones in the exception table.
2121 	 *	Note, that clones are aged out
2122 	 *	only if they are not in use now.
2123 	 */
2124 	rt6_age_exceptions(rt, gc_args, now);
2125 
2126 	return 0;
2127 }
2128 
2129 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2130 {
2131 	struct fib6_gc_args gc_args;
2132 	unsigned long now;
2133 
2134 	if (force) {
2135 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2136 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2137 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2138 		return;
2139 	}
2140 	gc_args.timeout = expires ? (int)expires :
2141 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2142 	gc_args.more = 0;
2143 
2144 	fib6_clean_all(net, fib6_age, &gc_args);
2145 	now = jiffies;
2146 	net->ipv6.ip6_rt_last_gc = now;
2147 
2148 	if (gc_args.more)
2149 		mod_timer(&net->ipv6.ip6_fib_timer,
2150 			  round_jiffies(now
2151 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2152 	else
2153 		del_timer(&net->ipv6.ip6_fib_timer);
2154 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2155 }
2156 
2157 static void fib6_gc_timer_cb(struct timer_list *t)
2158 {
2159 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2160 
2161 	fib6_run_gc(0, arg, true);
2162 }
2163 
2164 static int __net_init fib6_net_init(struct net *net)
2165 {
2166 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2167 	int err;
2168 
2169 	err = fib6_notifier_init(net);
2170 	if (err)
2171 		return err;
2172 
2173 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2174 	rwlock_init(&net->ipv6.fib6_walker_lock);
2175 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2176 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2177 
2178 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2179 	if (!net->ipv6.rt6_stats)
2180 		goto out_timer;
2181 
2182 	/* Avoid false sharing : Use at least a full cache line */
2183 	size = max_t(size_t, size, L1_CACHE_BYTES);
2184 
2185 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2186 	if (!net->ipv6.fib_table_hash)
2187 		goto out_rt6_stats;
2188 
2189 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2190 					  GFP_KERNEL);
2191 	if (!net->ipv6.fib6_main_tbl)
2192 		goto out_fib_table_hash;
2193 
2194 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2195 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2196 			   net->ipv6.fib6_null_entry);
2197 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2198 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2199 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2200 
2201 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2202 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2203 					   GFP_KERNEL);
2204 	if (!net->ipv6.fib6_local_tbl)
2205 		goto out_fib6_main_tbl;
2206 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2207 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2208 			   net->ipv6.fib6_null_entry);
2209 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2210 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2211 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2212 #endif
2213 	fib6_tables_init(net);
2214 
2215 	return 0;
2216 
2217 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2218 out_fib6_main_tbl:
2219 	kfree(net->ipv6.fib6_main_tbl);
2220 #endif
2221 out_fib_table_hash:
2222 	kfree(net->ipv6.fib_table_hash);
2223 out_rt6_stats:
2224 	kfree(net->ipv6.rt6_stats);
2225 out_timer:
2226 	fib6_notifier_exit(net);
2227 	return -ENOMEM;
2228 }
2229 
2230 static void fib6_net_exit(struct net *net)
2231 {
2232 	unsigned int i;
2233 
2234 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2235 
2236 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2237 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2238 		struct hlist_node *tmp;
2239 		struct fib6_table *tb;
2240 
2241 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2242 			hlist_del(&tb->tb6_hlist);
2243 			fib6_free_table(tb);
2244 		}
2245 	}
2246 
2247 	kfree(net->ipv6.fib_table_hash);
2248 	kfree(net->ipv6.rt6_stats);
2249 	fib6_notifier_exit(net);
2250 }
2251 
2252 static struct pernet_operations fib6_net_ops = {
2253 	.init = fib6_net_init,
2254 	.exit = fib6_net_exit,
2255 };
2256 
2257 int __init fib6_init(void)
2258 {
2259 	int ret = -ENOMEM;
2260 
2261 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2262 					   sizeof(struct fib6_node),
2263 					   0, SLAB_HWCACHE_ALIGN,
2264 					   NULL);
2265 	if (!fib6_node_kmem)
2266 		goto out;
2267 
2268 	ret = register_pernet_subsys(&fib6_net_ops);
2269 	if (ret)
2270 		goto out_kmem_cache_create;
2271 
2272 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2273 				   inet6_dump_fib, 0);
2274 	if (ret)
2275 		goto out_unregister_subsys;
2276 
2277 	__fib6_flush_trees = fib6_flush_trees;
2278 out:
2279 	return ret;
2280 
2281 out_unregister_subsys:
2282 	unregister_pernet_subsys(&fib6_net_ops);
2283 out_kmem_cache_create:
2284 	kmem_cache_destroy(fib6_node_kmem);
2285 	goto out;
2286 }
2287 
2288 void fib6_gc_cleanup(void)
2289 {
2290 	unregister_pernet_subsys(&fib6_net_ops);
2291 	kmem_cache_destroy(fib6_node_kmem);
2292 }
2293 
2294 #ifdef CONFIG_PROC_FS
2295 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2296 {
2297 	struct fib6_info *rt = v;
2298 	struct ipv6_route_iter *iter = seq->private;
2299 	unsigned int flags = rt->fib6_flags;
2300 	const struct net_device *dev;
2301 
2302 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2303 
2304 #ifdef CONFIG_IPV6_SUBTREES
2305 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2306 #else
2307 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2308 #endif
2309 	if (rt->fib6_nh.fib_nh_gw_family) {
2310 		flags |= RTF_GATEWAY;
2311 		seq_printf(seq, "%pi6", &rt->fib6_nh.fib_nh_gw6);
2312 	} else {
2313 		seq_puts(seq, "00000000000000000000000000000000");
2314 	}
2315 
2316 	dev = rt->fib6_nh.fib_nh_dev;
2317 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2318 		   rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2319 		   flags, dev ? dev->name : "");
2320 	iter->w.leaf = NULL;
2321 	return 0;
2322 }
2323 
2324 static int ipv6_route_yield(struct fib6_walker *w)
2325 {
2326 	struct ipv6_route_iter *iter = w->args;
2327 
2328 	if (!iter->skip)
2329 		return 1;
2330 
2331 	do {
2332 		iter->w.leaf = rcu_dereference_protected(
2333 				iter->w.leaf->fib6_next,
2334 				lockdep_is_held(&iter->tbl->tb6_lock));
2335 		iter->skip--;
2336 		if (!iter->skip && iter->w.leaf)
2337 			return 1;
2338 	} while (iter->w.leaf);
2339 
2340 	return 0;
2341 }
2342 
2343 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2344 				      struct net *net)
2345 {
2346 	memset(&iter->w, 0, sizeof(iter->w));
2347 	iter->w.func = ipv6_route_yield;
2348 	iter->w.root = &iter->tbl->tb6_root;
2349 	iter->w.state = FWS_INIT;
2350 	iter->w.node = iter->w.root;
2351 	iter->w.args = iter;
2352 	iter->sernum = iter->w.root->fn_sernum;
2353 	INIT_LIST_HEAD(&iter->w.lh);
2354 	fib6_walker_link(net, &iter->w);
2355 }
2356 
2357 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2358 						    struct net *net)
2359 {
2360 	unsigned int h;
2361 	struct hlist_node *node;
2362 
2363 	if (tbl) {
2364 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2365 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2366 	} else {
2367 		h = 0;
2368 		node = NULL;
2369 	}
2370 
2371 	while (!node && h < FIB6_TABLE_HASHSZ) {
2372 		node = rcu_dereference_bh(
2373 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2374 	}
2375 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2376 }
2377 
2378 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2379 {
2380 	if (iter->sernum != iter->w.root->fn_sernum) {
2381 		iter->sernum = iter->w.root->fn_sernum;
2382 		iter->w.state = FWS_INIT;
2383 		iter->w.node = iter->w.root;
2384 		WARN_ON(iter->w.skip);
2385 		iter->w.skip = iter->w.count;
2386 	}
2387 }
2388 
2389 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2390 {
2391 	int r;
2392 	struct fib6_info *n;
2393 	struct net *net = seq_file_net(seq);
2394 	struct ipv6_route_iter *iter = seq->private;
2395 
2396 	if (!v)
2397 		goto iter_table;
2398 
2399 	n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2400 	if (n) {
2401 		++*pos;
2402 		return n;
2403 	}
2404 
2405 iter_table:
2406 	ipv6_route_check_sernum(iter);
2407 	spin_lock_bh(&iter->tbl->tb6_lock);
2408 	r = fib6_walk_continue(&iter->w);
2409 	spin_unlock_bh(&iter->tbl->tb6_lock);
2410 	if (r > 0) {
2411 		if (v)
2412 			++*pos;
2413 		return iter->w.leaf;
2414 	} else if (r < 0) {
2415 		fib6_walker_unlink(net, &iter->w);
2416 		return NULL;
2417 	}
2418 	fib6_walker_unlink(net, &iter->w);
2419 
2420 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2421 	if (!iter->tbl)
2422 		return NULL;
2423 
2424 	ipv6_route_seq_setup_walk(iter, net);
2425 	goto iter_table;
2426 }
2427 
2428 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2429 	__acquires(RCU_BH)
2430 {
2431 	struct net *net = seq_file_net(seq);
2432 	struct ipv6_route_iter *iter = seq->private;
2433 
2434 	rcu_read_lock_bh();
2435 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2436 	iter->skip = *pos;
2437 
2438 	if (iter->tbl) {
2439 		ipv6_route_seq_setup_walk(iter, net);
2440 		return ipv6_route_seq_next(seq, NULL, pos);
2441 	} else {
2442 		return NULL;
2443 	}
2444 }
2445 
2446 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2447 {
2448 	struct fib6_walker *w = &iter->w;
2449 	return w->node && !(w->state == FWS_U && w->node == w->root);
2450 }
2451 
2452 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2453 	__releases(RCU_BH)
2454 {
2455 	struct net *net = seq_file_net(seq);
2456 	struct ipv6_route_iter *iter = seq->private;
2457 
2458 	if (ipv6_route_iter_active(iter))
2459 		fib6_walker_unlink(net, &iter->w);
2460 
2461 	rcu_read_unlock_bh();
2462 }
2463 
2464 const struct seq_operations ipv6_route_seq_ops = {
2465 	.start	= ipv6_route_seq_start,
2466 	.next	= ipv6_route_seq_next,
2467 	.stop	= ipv6_route_seq_stop,
2468 	.show	= ipv6_route_seq_show
2469 };
2470 #endif /* CONFIG_PROC_FS */
2471