xref: /linux/net/ipv6/ip6_fib.c (revision b9b77222d4ff6b5bb8f5d87fca20de0910618bb9)
1 /*
2  *	Linux INET6 implementation
3  *	Forwarding Information Database
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	This program is free software; you can redistribute it and/or
9  *      modify it under the terms of the GNU General Public License
10  *      as published by the Free Software Foundation; either version
11  *      2 of the License, or (at your option) any later version.
12  *
13  *	Changes:
14  *	Yuji SEKIYA @USAGI:	Support default route on router node;
15  *				remove ip6_null_entry from the top of
16  *				routing table.
17  *	Ville Nuorvala:		Fixed routing subtrees.
18  */
19 
20 #define pr_fmt(fmt) "IPv6: " fmt
21 
22 #include <linux/errno.h>
23 #include <linux/types.h>
24 #include <linux/net.h>
25 #include <linux/route.h>
26 #include <linux/netdevice.h>
27 #include <linux/in6.h>
28 #include <linux/init.h>
29 #include <linux/list.h>
30 #include <linux/slab.h>
31 
32 #include <net/ipv6.h>
33 #include <net/ndisc.h>
34 #include <net/addrconf.h>
35 #include <net/lwtunnel.h>
36 #include <net/fib_notifier.h>
37 
38 #include <net/ip6_fib.h>
39 #include <net/ip6_route.h>
40 
41 static struct kmem_cache *fib6_node_kmem __read_mostly;
42 
43 struct fib6_cleaner {
44 	struct fib6_walker w;
45 	struct net *net;
46 	int (*func)(struct fib6_info *, void *arg);
47 	int sernum;
48 	void *arg;
49 };
50 
51 #ifdef CONFIG_IPV6_SUBTREES
52 #define FWS_INIT FWS_S
53 #else
54 #define FWS_INIT FWS_L
55 #endif
56 
57 static struct fib6_info *fib6_find_prefix(struct net *net,
58 					 struct fib6_table *table,
59 					 struct fib6_node *fn);
60 static struct fib6_node *fib6_repair_tree(struct net *net,
61 					  struct fib6_table *table,
62 					  struct fib6_node *fn);
63 static int fib6_walk(struct net *net, struct fib6_walker *w);
64 static int fib6_walk_continue(struct fib6_walker *w);
65 
66 /*
67  *	A routing update causes an increase of the serial number on the
68  *	affected subtree. This allows for cached routes to be asynchronously
69  *	tested when modifications are made to the destination cache as a
70  *	result of redirects, path MTU changes, etc.
71  */
72 
73 static void fib6_gc_timer_cb(struct timer_list *t);
74 
75 #define FOR_WALKERS(net, w) \
76 	list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
77 
78 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
79 {
80 	write_lock_bh(&net->ipv6.fib6_walker_lock);
81 	list_add(&w->lh, &net->ipv6.fib6_walkers);
82 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
83 }
84 
85 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
86 {
87 	write_lock_bh(&net->ipv6.fib6_walker_lock);
88 	list_del(&w->lh);
89 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
90 }
91 
92 static int fib6_new_sernum(struct net *net)
93 {
94 	int new, old;
95 
96 	do {
97 		old = atomic_read(&net->ipv6.fib6_sernum);
98 		new = old < INT_MAX ? old + 1 : 1;
99 	} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
100 				old, new) != old);
101 	return new;
102 }
103 
104 enum {
105 	FIB6_NO_SERNUM_CHANGE = 0,
106 };
107 
108 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
109 {
110 	struct fib6_node *fn;
111 
112 	fn = rcu_dereference_protected(f6i->fib6_node,
113 			lockdep_is_held(&f6i->fib6_table->tb6_lock));
114 	if (fn)
115 		fn->fn_sernum = fib6_new_sernum(net);
116 }
117 
118 /*
119  *	Auxiliary address test functions for the radix tree.
120  *
121  *	These assume a 32bit processor (although it will work on
122  *	64bit processors)
123  */
124 
125 /*
126  *	test bit
127  */
128 #if defined(__LITTLE_ENDIAN)
129 # define BITOP_BE32_SWIZZLE	(0x1F & ~7)
130 #else
131 # define BITOP_BE32_SWIZZLE	0
132 #endif
133 
134 static __be32 addr_bit_set(const void *token, int fn_bit)
135 {
136 	const __be32 *addr = token;
137 	/*
138 	 * Here,
139 	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
140 	 * is optimized version of
141 	 *	htonl(1 << ((~fn_bit)&0x1F))
142 	 * See include/asm-generic/bitops/le.h.
143 	 */
144 	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
145 	       addr[fn_bit >> 5];
146 }
147 
148 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags)
149 {
150 	struct fib6_info *f6i;
151 
152 	f6i = kzalloc(sizeof(*f6i), gfp_flags);
153 	if (!f6i)
154 		return NULL;
155 
156 	f6i->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
157 	if (!f6i->rt6i_pcpu) {
158 		kfree(f6i);
159 		return NULL;
160 	}
161 
162 	INIT_LIST_HEAD(&f6i->fib6_siblings);
163 	f6i->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
164 
165 	atomic_inc(&f6i->fib6_ref);
166 
167 	return f6i;
168 }
169 
170 void fib6_info_destroy_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 	struct dst_metrics *m;
175 
176 	WARN_ON(f6i->fib6_node);
177 
178 	bucket = rcu_dereference_protected(f6i->rt6i_exception_bucket, 1);
179 	if (bucket) {
180 		f6i->rt6i_exception_bucket = NULL;
181 		kfree(bucket);
182 	}
183 
184 	if (f6i->rt6i_pcpu) {
185 		int cpu;
186 
187 		for_each_possible_cpu(cpu) {
188 			struct rt6_info **ppcpu_rt;
189 			struct rt6_info *pcpu_rt;
190 
191 			ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
192 			pcpu_rt = *ppcpu_rt;
193 			if (pcpu_rt) {
194 				dst_dev_put(&pcpu_rt->dst);
195 				dst_release(&pcpu_rt->dst);
196 				*ppcpu_rt = NULL;
197 			}
198 		}
199 	}
200 
201 	if (f6i->fib6_nh.nh_dev)
202 		dev_put(f6i->fib6_nh.nh_dev);
203 
204 	m = f6i->fib6_metrics;
205 	if (m != &dst_default_metrics && refcount_dec_and_test(&m->refcnt))
206 		kfree(m);
207 
208 	kfree(f6i);
209 }
210 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
211 
212 static struct fib6_node *node_alloc(struct net *net)
213 {
214 	struct fib6_node *fn;
215 
216 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
217 	if (fn)
218 		net->ipv6.rt6_stats->fib_nodes++;
219 
220 	return fn;
221 }
222 
223 static void node_free_immediate(struct net *net, struct fib6_node *fn)
224 {
225 	kmem_cache_free(fib6_node_kmem, fn);
226 	net->ipv6.rt6_stats->fib_nodes--;
227 }
228 
229 static void node_free_rcu(struct rcu_head *head)
230 {
231 	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
232 
233 	kmem_cache_free(fib6_node_kmem, fn);
234 }
235 
236 static void node_free(struct net *net, struct fib6_node *fn)
237 {
238 	call_rcu(&fn->rcu, node_free_rcu);
239 	net->ipv6.rt6_stats->fib_nodes--;
240 }
241 
242 static void fib6_free_table(struct fib6_table *table)
243 {
244 	inetpeer_invalidate_tree(&table->tb6_peers);
245 	kfree(table);
246 }
247 
248 static void fib6_link_table(struct net *net, struct fib6_table *tb)
249 {
250 	unsigned int h;
251 
252 	/*
253 	 * Initialize table lock at a single place to give lockdep a key,
254 	 * tables aren't visible prior to being linked to the list.
255 	 */
256 	spin_lock_init(&tb->tb6_lock);
257 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
258 
259 	/*
260 	 * No protection necessary, this is the only list mutatation
261 	 * operation, tables never disappear once they exist.
262 	 */
263 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
264 }
265 
266 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
267 
268 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
269 {
270 	struct fib6_table *table;
271 
272 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
273 	if (table) {
274 		table->tb6_id = id;
275 		rcu_assign_pointer(table->tb6_root.leaf,
276 				   net->ipv6.fib6_null_entry);
277 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
278 		inet_peer_base_init(&table->tb6_peers);
279 	}
280 
281 	return table;
282 }
283 
284 struct fib6_table *fib6_new_table(struct net *net, u32 id)
285 {
286 	struct fib6_table *tb;
287 
288 	if (id == 0)
289 		id = RT6_TABLE_MAIN;
290 	tb = fib6_get_table(net, id);
291 	if (tb)
292 		return tb;
293 
294 	tb = fib6_alloc_table(net, id);
295 	if (tb)
296 		fib6_link_table(net, tb);
297 
298 	return tb;
299 }
300 EXPORT_SYMBOL_GPL(fib6_new_table);
301 
302 struct fib6_table *fib6_get_table(struct net *net, u32 id)
303 {
304 	struct fib6_table *tb;
305 	struct hlist_head *head;
306 	unsigned int h;
307 
308 	if (id == 0)
309 		id = RT6_TABLE_MAIN;
310 	h = id & (FIB6_TABLE_HASHSZ - 1);
311 	rcu_read_lock();
312 	head = &net->ipv6.fib_table_hash[h];
313 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
314 		if (tb->tb6_id == id) {
315 			rcu_read_unlock();
316 			return tb;
317 		}
318 	}
319 	rcu_read_unlock();
320 
321 	return NULL;
322 }
323 EXPORT_SYMBOL_GPL(fib6_get_table);
324 
325 static void __net_init fib6_tables_init(struct net *net)
326 {
327 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
328 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
329 }
330 #else
331 
332 struct fib6_table *fib6_new_table(struct net *net, u32 id)
333 {
334 	return fib6_get_table(net, id);
335 }
336 
337 struct fib6_table *fib6_get_table(struct net *net, u32 id)
338 {
339 	  return net->ipv6.fib6_main_tbl;
340 }
341 
342 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
343 				   const struct sk_buff *skb,
344 				   int flags, pol_lookup_t lookup)
345 {
346 	struct rt6_info *rt;
347 
348 	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
349 	if (rt->dst.error == -EAGAIN) {
350 		ip6_rt_put(rt);
351 		rt = net->ipv6.ip6_null_entry;
352 		dst_hold(&rt->dst);
353 	}
354 
355 	return &rt->dst;
356 }
357 
358 /* called with rcu lock held; no reference taken on fib6_info */
359 struct fib6_info *fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
360 			      int flags)
361 {
362 	return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6, flags);
363 }
364 
365 static void __net_init fib6_tables_init(struct net *net)
366 {
367 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
368 }
369 
370 #endif
371 
372 unsigned int fib6_tables_seq_read(struct net *net)
373 {
374 	unsigned int h, fib_seq = 0;
375 
376 	rcu_read_lock();
377 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
378 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
379 		struct fib6_table *tb;
380 
381 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
382 			fib_seq += tb->fib_seq;
383 	}
384 	rcu_read_unlock();
385 
386 	return fib_seq;
387 }
388 
389 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
390 				    enum fib_event_type event_type,
391 				    struct fib6_info *rt)
392 {
393 	struct fib6_entry_notifier_info info = {
394 		.rt = rt,
395 	};
396 
397 	return call_fib6_notifier(nb, net, event_type, &info.info);
398 }
399 
400 static int call_fib6_entry_notifiers(struct net *net,
401 				     enum fib_event_type event_type,
402 				     struct fib6_info *rt,
403 				     struct netlink_ext_ack *extack)
404 {
405 	struct fib6_entry_notifier_info info = {
406 		.info.extack = extack,
407 		.rt = rt,
408 	};
409 
410 	rt->fib6_table->fib_seq++;
411 	return call_fib6_notifiers(net, event_type, &info.info);
412 }
413 
414 struct fib6_dump_arg {
415 	struct net *net;
416 	struct notifier_block *nb;
417 };
418 
419 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
420 {
421 	if (rt == arg->net->ipv6.fib6_null_entry)
422 		return;
423 	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
424 }
425 
426 static int fib6_node_dump(struct fib6_walker *w)
427 {
428 	struct fib6_info *rt;
429 
430 	for_each_fib6_walker_rt(w)
431 		fib6_rt_dump(rt, w->args);
432 	w->leaf = NULL;
433 	return 0;
434 }
435 
436 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
437 			    struct fib6_walker *w)
438 {
439 	w->root = &tb->tb6_root;
440 	spin_lock_bh(&tb->tb6_lock);
441 	fib6_walk(net, w);
442 	spin_unlock_bh(&tb->tb6_lock);
443 }
444 
445 /* Called with rcu_read_lock() */
446 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
447 {
448 	struct fib6_dump_arg arg;
449 	struct fib6_walker *w;
450 	unsigned int h;
451 
452 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
453 	if (!w)
454 		return -ENOMEM;
455 
456 	w->func = fib6_node_dump;
457 	arg.net = net;
458 	arg.nb = nb;
459 	w->args = &arg;
460 
461 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
462 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
463 		struct fib6_table *tb;
464 
465 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
466 			fib6_table_dump(net, tb, w);
467 	}
468 
469 	kfree(w);
470 
471 	return 0;
472 }
473 
474 static int fib6_dump_node(struct fib6_walker *w)
475 {
476 	int res;
477 	struct fib6_info *rt;
478 
479 	for_each_fib6_walker_rt(w) {
480 		res = rt6_dump_route(rt, w->args);
481 		if (res < 0) {
482 			/* Frame is full, suspend walking */
483 			w->leaf = rt;
484 			return 1;
485 		}
486 
487 		/* Multipath routes are dumped in one route with the
488 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
489 		 * last sibling of this route (no need to dump the
490 		 * sibling routes again)
491 		 */
492 		if (rt->fib6_nsiblings)
493 			rt = list_last_entry(&rt->fib6_siblings,
494 					     struct fib6_info,
495 					     fib6_siblings);
496 	}
497 	w->leaf = NULL;
498 	return 0;
499 }
500 
501 static void fib6_dump_end(struct netlink_callback *cb)
502 {
503 	struct net *net = sock_net(cb->skb->sk);
504 	struct fib6_walker *w = (void *)cb->args[2];
505 
506 	if (w) {
507 		if (cb->args[4]) {
508 			cb->args[4] = 0;
509 			fib6_walker_unlink(net, w);
510 		}
511 		cb->args[2] = 0;
512 		kfree(w);
513 	}
514 	cb->done = (void *)cb->args[3];
515 	cb->args[1] = 3;
516 }
517 
518 static int fib6_dump_done(struct netlink_callback *cb)
519 {
520 	fib6_dump_end(cb);
521 	return cb->done ? cb->done(cb) : 0;
522 }
523 
524 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
525 			   struct netlink_callback *cb)
526 {
527 	struct net *net = sock_net(skb->sk);
528 	struct fib6_walker *w;
529 	int res;
530 
531 	w = (void *)cb->args[2];
532 	w->root = &table->tb6_root;
533 
534 	if (cb->args[4] == 0) {
535 		w->count = 0;
536 		w->skip = 0;
537 
538 		spin_lock_bh(&table->tb6_lock);
539 		res = fib6_walk(net, w);
540 		spin_unlock_bh(&table->tb6_lock);
541 		if (res > 0) {
542 			cb->args[4] = 1;
543 			cb->args[5] = w->root->fn_sernum;
544 		}
545 	} else {
546 		if (cb->args[5] != w->root->fn_sernum) {
547 			/* Begin at the root if the tree changed */
548 			cb->args[5] = w->root->fn_sernum;
549 			w->state = FWS_INIT;
550 			w->node = w->root;
551 			w->skip = w->count;
552 		} else
553 			w->skip = 0;
554 
555 		spin_lock_bh(&table->tb6_lock);
556 		res = fib6_walk_continue(w);
557 		spin_unlock_bh(&table->tb6_lock);
558 		if (res <= 0) {
559 			fib6_walker_unlink(net, w);
560 			cb->args[4] = 0;
561 		}
562 	}
563 
564 	return res;
565 }
566 
567 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
568 {
569 	struct net *net = sock_net(skb->sk);
570 	unsigned int h, s_h;
571 	unsigned int e = 0, s_e;
572 	struct rt6_rtnl_dump_arg arg;
573 	struct fib6_walker *w;
574 	struct fib6_table *tb;
575 	struct hlist_head *head;
576 	int res = 0;
577 
578 	s_h = cb->args[0];
579 	s_e = cb->args[1];
580 
581 	w = (void *)cb->args[2];
582 	if (!w) {
583 		/* New dump:
584 		 *
585 		 * 1. hook callback destructor.
586 		 */
587 		cb->args[3] = (long)cb->done;
588 		cb->done = fib6_dump_done;
589 
590 		/*
591 		 * 2. allocate and initialize walker.
592 		 */
593 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
594 		if (!w)
595 			return -ENOMEM;
596 		w->func = fib6_dump_node;
597 		cb->args[2] = (long)w;
598 	}
599 
600 	arg.skb = skb;
601 	arg.cb = cb;
602 	arg.net = net;
603 	w->args = &arg;
604 
605 	rcu_read_lock();
606 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
607 		e = 0;
608 		head = &net->ipv6.fib_table_hash[h];
609 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
610 			if (e < s_e)
611 				goto next;
612 			res = fib6_dump_table(tb, skb, cb);
613 			if (res != 0)
614 				goto out;
615 next:
616 			e++;
617 		}
618 	}
619 out:
620 	rcu_read_unlock();
621 	cb->args[1] = e;
622 	cb->args[0] = h;
623 
624 	res = res < 0 ? res : skb->len;
625 	if (res <= 0)
626 		fib6_dump_end(cb);
627 	return res;
628 }
629 
630 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
631 {
632 	if (!f6i)
633 		return;
634 
635 	if (f6i->fib6_metrics == &dst_default_metrics) {
636 		struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
637 
638 		if (!p)
639 			return;
640 
641 		refcount_set(&p->refcnt, 1);
642 		f6i->fib6_metrics = p;
643 	}
644 
645 	f6i->fib6_metrics->metrics[metric - 1] = val;
646 }
647 
648 /*
649  *	Routing Table
650  *
651  *	return the appropriate node for a routing tree "add" operation
652  *	by either creating and inserting or by returning an existing
653  *	node.
654  */
655 
656 static struct fib6_node *fib6_add_1(struct net *net,
657 				    struct fib6_table *table,
658 				    struct fib6_node *root,
659 				    struct in6_addr *addr, int plen,
660 				    int offset, int allow_create,
661 				    int replace_required,
662 				    struct netlink_ext_ack *extack)
663 {
664 	struct fib6_node *fn, *in, *ln;
665 	struct fib6_node *pn = NULL;
666 	struct rt6key *key;
667 	int	bit;
668 	__be32	dir = 0;
669 
670 	RT6_TRACE("fib6_add_1\n");
671 
672 	/* insert node in tree */
673 
674 	fn = root;
675 
676 	do {
677 		struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
678 					    lockdep_is_held(&table->tb6_lock));
679 		key = (struct rt6key *)((u8 *)leaf + offset);
680 
681 		/*
682 		 *	Prefix match
683 		 */
684 		if (plen < fn->fn_bit ||
685 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
686 			if (!allow_create) {
687 				if (replace_required) {
688 					NL_SET_ERR_MSG(extack,
689 						       "Can not replace route - no match found");
690 					pr_warn("Can't replace route, no match found\n");
691 					return ERR_PTR(-ENOENT);
692 				}
693 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
694 			}
695 			goto insert_above;
696 		}
697 
698 		/*
699 		 *	Exact match ?
700 		 */
701 
702 		if (plen == fn->fn_bit) {
703 			/* clean up an intermediate node */
704 			if (!(fn->fn_flags & RTN_RTINFO)) {
705 				RCU_INIT_POINTER(fn->leaf, NULL);
706 				fib6_info_release(leaf);
707 			/* remove null_entry in the root node */
708 			} else if (fn->fn_flags & RTN_TL_ROOT &&
709 				   rcu_access_pointer(fn->leaf) ==
710 				   net->ipv6.fib6_null_entry) {
711 				RCU_INIT_POINTER(fn->leaf, NULL);
712 			}
713 
714 			return fn;
715 		}
716 
717 		/*
718 		 *	We have more bits to go
719 		 */
720 
721 		/* Try to walk down on tree. */
722 		dir = addr_bit_set(addr, fn->fn_bit);
723 		pn = fn;
724 		fn = dir ?
725 		     rcu_dereference_protected(fn->right,
726 					lockdep_is_held(&table->tb6_lock)) :
727 		     rcu_dereference_protected(fn->left,
728 					lockdep_is_held(&table->tb6_lock));
729 	} while (fn);
730 
731 	if (!allow_create) {
732 		/* We should not create new node because
733 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
734 		 * I assume it is safe to require NLM_F_CREATE when
735 		 * REPLACE flag is used! Later we may want to remove the
736 		 * check for replace_required, because according
737 		 * to netlink specification, NLM_F_CREATE
738 		 * MUST be specified if new route is created.
739 		 * That would keep IPv6 consistent with IPv4
740 		 */
741 		if (replace_required) {
742 			NL_SET_ERR_MSG(extack,
743 				       "Can not replace route - no match found");
744 			pr_warn("Can't replace route, no match found\n");
745 			return ERR_PTR(-ENOENT);
746 		}
747 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
748 	}
749 	/*
750 	 *	We walked to the bottom of tree.
751 	 *	Create new leaf node without children.
752 	 */
753 
754 	ln = node_alloc(net);
755 
756 	if (!ln)
757 		return ERR_PTR(-ENOMEM);
758 	ln->fn_bit = plen;
759 	RCU_INIT_POINTER(ln->parent, pn);
760 
761 	if (dir)
762 		rcu_assign_pointer(pn->right, ln);
763 	else
764 		rcu_assign_pointer(pn->left, ln);
765 
766 	return ln;
767 
768 
769 insert_above:
770 	/*
771 	 * split since we don't have a common prefix anymore or
772 	 * we have a less significant route.
773 	 * we've to insert an intermediate node on the list
774 	 * this new node will point to the one we need to create
775 	 * and the current
776 	 */
777 
778 	pn = rcu_dereference_protected(fn->parent,
779 				       lockdep_is_held(&table->tb6_lock));
780 
781 	/* find 1st bit in difference between the 2 addrs.
782 
783 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
784 	   but if it is >= plen, the value is ignored in any case.
785 	 */
786 
787 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
788 
789 	/*
790 	 *		(intermediate)[in]
791 	 *	          /	   \
792 	 *	(new leaf node)[ln] (old node)[fn]
793 	 */
794 	if (plen > bit) {
795 		in = node_alloc(net);
796 		ln = node_alloc(net);
797 
798 		if (!in || !ln) {
799 			if (in)
800 				node_free_immediate(net, in);
801 			if (ln)
802 				node_free_immediate(net, ln);
803 			return ERR_PTR(-ENOMEM);
804 		}
805 
806 		/*
807 		 * new intermediate node.
808 		 * RTN_RTINFO will
809 		 * be off since that an address that chooses one of
810 		 * the branches would not match less specific routes
811 		 * in the other branch
812 		 */
813 
814 		in->fn_bit = bit;
815 
816 		RCU_INIT_POINTER(in->parent, pn);
817 		in->leaf = fn->leaf;
818 		atomic_inc(&rcu_dereference_protected(in->leaf,
819 				lockdep_is_held(&table->tb6_lock))->fib6_ref);
820 
821 		/* update parent pointer */
822 		if (dir)
823 			rcu_assign_pointer(pn->right, in);
824 		else
825 			rcu_assign_pointer(pn->left, in);
826 
827 		ln->fn_bit = plen;
828 
829 		RCU_INIT_POINTER(ln->parent, in);
830 		rcu_assign_pointer(fn->parent, in);
831 
832 		if (addr_bit_set(addr, bit)) {
833 			rcu_assign_pointer(in->right, ln);
834 			rcu_assign_pointer(in->left, fn);
835 		} else {
836 			rcu_assign_pointer(in->left, ln);
837 			rcu_assign_pointer(in->right, fn);
838 		}
839 	} else { /* plen <= bit */
840 
841 		/*
842 		 *		(new leaf node)[ln]
843 		 *	          /	   \
844 		 *	     (old node)[fn] NULL
845 		 */
846 
847 		ln = node_alloc(net);
848 
849 		if (!ln)
850 			return ERR_PTR(-ENOMEM);
851 
852 		ln->fn_bit = plen;
853 
854 		RCU_INIT_POINTER(ln->parent, pn);
855 
856 		if (addr_bit_set(&key->addr, plen))
857 			RCU_INIT_POINTER(ln->right, fn);
858 		else
859 			RCU_INIT_POINTER(ln->left, fn);
860 
861 		rcu_assign_pointer(fn->parent, ln);
862 
863 		if (dir)
864 			rcu_assign_pointer(pn->right, ln);
865 		else
866 			rcu_assign_pointer(pn->left, ln);
867 	}
868 	return ln;
869 }
870 
871 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
872 				const struct fib6_table *table)
873 {
874 	int cpu;
875 
876 	/* release the reference to this fib entry from
877 	 * all of its cached pcpu routes
878 	 */
879 	for_each_possible_cpu(cpu) {
880 		struct rt6_info **ppcpu_rt;
881 		struct rt6_info *pcpu_rt;
882 
883 		ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
884 		pcpu_rt = *ppcpu_rt;
885 		if (pcpu_rt) {
886 			struct fib6_info *from;
887 
888 			from = rcu_dereference_protected(pcpu_rt->from,
889 					     lockdep_is_held(&table->tb6_lock));
890 			rcu_assign_pointer(pcpu_rt->from, NULL);
891 			fib6_info_release(from);
892 		}
893 	}
894 }
895 
896 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
897 			  struct net *net)
898 {
899 	struct fib6_table *table = rt->fib6_table;
900 
901 	if (atomic_read(&rt->fib6_ref) != 1) {
902 		/* This route is used as dummy address holder in some split
903 		 * nodes. It is not leaked, but it still holds other resources,
904 		 * which must be released in time. So, scan ascendant nodes
905 		 * and replace dummy references to this route with references
906 		 * to still alive ones.
907 		 */
908 		while (fn) {
909 			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
910 					    lockdep_is_held(&table->tb6_lock));
911 			struct fib6_info *new_leaf;
912 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
913 				new_leaf = fib6_find_prefix(net, table, fn);
914 				atomic_inc(&new_leaf->fib6_ref);
915 
916 				rcu_assign_pointer(fn->leaf, new_leaf);
917 				fib6_info_release(rt);
918 			}
919 			fn = rcu_dereference_protected(fn->parent,
920 				    lockdep_is_held(&table->tb6_lock));
921 		}
922 
923 		if (rt->rt6i_pcpu)
924 			fib6_drop_pcpu_from(rt, table);
925 	}
926 }
927 
928 /*
929  *	Insert routing information in a node.
930  */
931 
932 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
933 			    struct nl_info *info,
934 			    struct netlink_ext_ack *extack)
935 {
936 	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
937 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
938 	struct fib6_info *iter = NULL;
939 	struct fib6_info __rcu **ins;
940 	struct fib6_info __rcu **fallback_ins = NULL;
941 	int replace = (info->nlh &&
942 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
943 	int add = (!info->nlh ||
944 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
945 	int found = 0;
946 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
947 	u16 nlflags = NLM_F_EXCL;
948 	int err;
949 
950 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
951 		nlflags |= NLM_F_APPEND;
952 
953 	ins = &fn->leaf;
954 
955 	for (iter = leaf; iter;
956 	     iter = rcu_dereference_protected(iter->fib6_next,
957 				lockdep_is_held(&rt->fib6_table->tb6_lock))) {
958 		/*
959 		 *	Search for duplicates
960 		 */
961 
962 		if (iter->fib6_metric == rt->fib6_metric) {
963 			/*
964 			 *	Same priority level
965 			 */
966 			if (info->nlh &&
967 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
968 				return -EEXIST;
969 
970 			nlflags &= ~NLM_F_EXCL;
971 			if (replace) {
972 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
973 					found++;
974 					break;
975 				}
976 				if (rt_can_ecmp)
977 					fallback_ins = fallback_ins ?: ins;
978 				goto next_iter;
979 			}
980 
981 			if (rt6_duplicate_nexthop(iter, rt)) {
982 				if (rt->fib6_nsiblings)
983 					rt->fib6_nsiblings = 0;
984 				if (!(iter->fib6_flags & RTF_EXPIRES))
985 					return -EEXIST;
986 				if (!(rt->fib6_flags & RTF_EXPIRES))
987 					fib6_clean_expires(iter);
988 				else
989 					fib6_set_expires(iter, rt->expires);
990 				fib6_metric_set(iter, RTAX_MTU, rt->fib6_pmtu);
991 				return -EEXIST;
992 			}
993 			/* If we have the same destination and the same metric,
994 			 * but not the same gateway, then the route we try to
995 			 * add is sibling to this route, increment our counter
996 			 * of siblings, and later we will add our route to the
997 			 * list.
998 			 * Only static routes (which don't have flag
999 			 * RTF_EXPIRES) are used for ECMPv6.
1000 			 *
1001 			 * To avoid long list, we only had siblings if the
1002 			 * route have a gateway.
1003 			 */
1004 			if (rt_can_ecmp &&
1005 			    rt6_qualify_for_ecmp(iter))
1006 				rt->fib6_nsiblings++;
1007 		}
1008 
1009 		if (iter->fib6_metric > rt->fib6_metric)
1010 			break;
1011 
1012 next_iter:
1013 		ins = &iter->fib6_next;
1014 	}
1015 
1016 	if (fallback_ins && !found) {
1017 		/* No ECMP-able route found, replace first non-ECMP one */
1018 		ins = fallback_ins;
1019 		iter = rcu_dereference_protected(*ins,
1020 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1021 		found++;
1022 	}
1023 
1024 	/* Reset round-robin state, if necessary */
1025 	if (ins == &fn->leaf)
1026 		fn->rr_ptr = NULL;
1027 
1028 	/* Link this route to others same route. */
1029 	if (rt->fib6_nsiblings) {
1030 		unsigned int fib6_nsiblings;
1031 		struct fib6_info *sibling, *temp_sibling;
1032 
1033 		/* Find the first route that have the same metric */
1034 		sibling = leaf;
1035 		while (sibling) {
1036 			if (sibling->fib6_metric == rt->fib6_metric &&
1037 			    rt6_qualify_for_ecmp(sibling)) {
1038 				list_add_tail(&rt->fib6_siblings,
1039 					      &sibling->fib6_siblings);
1040 				break;
1041 			}
1042 			sibling = rcu_dereference_protected(sibling->fib6_next,
1043 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1044 		}
1045 		/* For each sibling in the list, increment the counter of
1046 		 * siblings. BUG() if counters does not match, list of siblings
1047 		 * is broken!
1048 		 */
1049 		fib6_nsiblings = 0;
1050 		list_for_each_entry_safe(sibling, temp_sibling,
1051 					 &rt->fib6_siblings, fib6_siblings) {
1052 			sibling->fib6_nsiblings++;
1053 			BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1054 			fib6_nsiblings++;
1055 		}
1056 		BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1057 		rt6_multipath_rebalance(temp_sibling);
1058 	}
1059 
1060 	/*
1061 	 *	insert node
1062 	 */
1063 	if (!replace) {
1064 		if (!add)
1065 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1066 
1067 add:
1068 		nlflags |= NLM_F_CREATE;
1069 
1070 		err = call_fib6_entry_notifiers(info->nl_net,
1071 						FIB_EVENT_ENTRY_ADD,
1072 						rt, extack);
1073 		if (err)
1074 			return err;
1075 
1076 		rcu_assign_pointer(rt->fib6_next, iter);
1077 		atomic_inc(&rt->fib6_ref);
1078 		rcu_assign_pointer(rt->fib6_node, fn);
1079 		rcu_assign_pointer(*ins, rt);
1080 		if (!info->skip_notify)
1081 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1082 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1083 
1084 		if (!(fn->fn_flags & RTN_RTINFO)) {
1085 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1086 			fn->fn_flags |= RTN_RTINFO;
1087 		}
1088 
1089 	} else {
1090 		int nsiblings;
1091 
1092 		if (!found) {
1093 			if (add)
1094 				goto add;
1095 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1096 			return -ENOENT;
1097 		}
1098 
1099 		err = call_fib6_entry_notifiers(info->nl_net,
1100 						FIB_EVENT_ENTRY_REPLACE,
1101 						rt, extack);
1102 		if (err)
1103 			return err;
1104 
1105 		atomic_inc(&rt->fib6_ref);
1106 		rcu_assign_pointer(rt->fib6_node, fn);
1107 		rt->fib6_next = iter->fib6_next;
1108 		rcu_assign_pointer(*ins, rt);
1109 		if (!info->skip_notify)
1110 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1111 		if (!(fn->fn_flags & RTN_RTINFO)) {
1112 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1113 			fn->fn_flags |= RTN_RTINFO;
1114 		}
1115 		nsiblings = iter->fib6_nsiblings;
1116 		iter->fib6_node = NULL;
1117 		fib6_purge_rt(iter, fn, info->nl_net);
1118 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1119 			fn->rr_ptr = NULL;
1120 		fib6_info_release(iter);
1121 
1122 		if (nsiblings) {
1123 			/* Replacing an ECMP route, remove all siblings */
1124 			ins = &rt->fib6_next;
1125 			iter = rcu_dereference_protected(*ins,
1126 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1127 			while (iter) {
1128 				if (iter->fib6_metric > rt->fib6_metric)
1129 					break;
1130 				if (rt6_qualify_for_ecmp(iter)) {
1131 					*ins = iter->fib6_next;
1132 					iter->fib6_node = NULL;
1133 					fib6_purge_rt(iter, fn, info->nl_net);
1134 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1135 						fn->rr_ptr = NULL;
1136 					fib6_info_release(iter);
1137 					nsiblings--;
1138 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1139 				} else {
1140 					ins = &iter->fib6_next;
1141 				}
1142 				iter = rcu_dereference_protected(*ins,
1143 					lockdep_is_held(&rt->fib6_table->tb6_lock));
1144 			}
1145 			WARN_ON(nsiblings != 0);
1146 		}
1147 	}
1148 
1149 	return 0;
1150 }
1151 
1152 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1153 {
1154 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1155 	    (rt->fib6_flags & RTF_EXPIRES))
1156 		mod_timer(&net->ipv6.ip6_fib_timer,
1157 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1158 }
1159 
1160 void fib6_force_start_gc(struct net *net)
1161 {
1162 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1163 		mod_timer(&net->ipv6.ip6_fib_timer,
1164 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1165 }
1166 
1167 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1168 					   int sernum)
1169 {
1170 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1171 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1172 
1173 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1174 	smp_wmb();
1175 	while (fn) {
1176 		fn->fn_sernum = sernum;
1177 		fn = rcu_dereference_protected(fn->parent,
1178 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1179 	}
1180 }
1181 
1182 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1183 {
1184 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1185 }
1186 
1187 /*
1188  *	Add routing information to the routing tree.
1189  *	<destination addr>/<source addr>
1190  *	with source addr info in sub-trees
1191  *	Need to own table->tb6_lock
1192  */
1193 
1194 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1195 	     struct nl_info *info, struct netlink_ext_ack *extack)
1196 {
1197 	struct fib6_table *table = rt->fib6_table;
1198 	struct fib6_node *fn, *pn = NULL;
1199 	int err = -ENOMEM;
1200 	int allow_create = 1;
1201 	int replace_required = 0;
1202 	int sernum = fib6_new_sernum(info->nl_net);
1203 
1204 	if (info->nlh) {
1205 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1206 			allow_create = 0;
1207 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1208 			replace_required = 1;
1209 	}
1210 	if (!allow_create && !replace_required)
1211 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1212 
1213 	fn = fib6_add_1(info->nl_net, table, root,
1214 			&rt->fib6_dst.addr, rt->fib6_dst.plen,
1215 			offsetof(struct fib6_info, fib6_dst), allow_create,
1216 			replace_required, extack);
1217 	if (IS_ERR(fn)) {
1218 		err = PTR_ERR(fn);
1219 		fn = NULL;
1220 		goto out;
1221 	}
1222 
1223 	pn = fn;
1224 
1225 #ifdef CONFIG_IPV6_SUBTREES
1226 	if (rt->fib6_src.plen) {
1227 		struct fib6_node *sn;
1228 
1229 		if (!rcu_access_pointer(fn->subtree)) {
1230 			struct fib6_node *sfn;
1231 
1232 			/*
1233 			 * Create subtree.
1234 			 *
1235 			 *		fn[main tree]
1236 			 *		|
1237 			 *		sfn[subtree root]
1238 			 *		   \
1239 			 *		    sn[new leaf node]
1240 			 */
1241 
1242 			/* Create subtree root node */
1243 			sfn = node_alloc(info->nl_net);
1244 			if (!sfn)
1245 				goto failure;
1246 
1247 			atomic_inc(&info->nl_net->ipv6.fib6_null_entry->fib6_ref);
1248 			rcu_assign_pointer(sfn->leaf,
1249 					   info->nl_net->ipv6.fib6_null_entry);
1250 			sfn->fn_flags = RTN_ROOT;
1251 
1252 			/* Now add the first leaf node to new subtree */
1253 
1254 			sn = fib6_add_1(info->nl_net, table, sfn,
1255 					&rt->fib6_src.addr, rt->fib6_src.plen,
1256 					offsetof(struct fib6_info, fib6_src),
1257 					allow_create, replace_required, extack);
1258 
1259 			if (IS_ERR(sn)) {
1260 				/* If it is failed, discard just allocated
1261 				   root, and then (in failure) stale node
1262 				   in main tree.
1263 				 */
1264 				node_free_immediate(info->nl_net, sfn);
1265 				err = PTR_ERR(sn);
1266 				goto failure;
1267 			}
1268 
1269 			/* Now link new subtree to main tree */
1270 			rcu_assign_pointer(sfn->parent, fn);
1271 			rcu_assign_pointer(fn->subtree, sfn);
1272 		} else {
1273 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1274 					&rt->fib6_src.addr, rt->fib6_src.plen,
1275 					offsetof(struct fib6_info, fib6_src),
1276 					allow_create, replace_required, extack);
1277 
1278 			if (IS_ERR(sn)) {
1279 				err = PTR_ERR(sn);
1280 				goto failure;
1281 			}
1282 		}
1283 
1284 		if (!rcu_access_pointer(fn->leaf)) {
1285 			if (fn->fn_flags & RTN_TL_ROOT) {
1286 				/* put back null_entry for root node */
1287 				rcu_assign_pointer(fn->leaf,
1288 					    info->nl_net->ipv6.fib6_null_entry);
1289 			} else {
1290 				atomic_inc(&rt->fib6_ref);
1291 				rcu_assign_pointer(fn->leaf, rt);
1292 			}
1293 		}
1294 		fn = sn;
1295 	}
1296 #endif
1297 
1298 	err = fib6_add_rt2node(fn, rt, info, extack);
1299 	if (!err) {
1300 		__fib6_update_sernum_upto_root(rt, sernum);
1301 		fib6_start_gc(info->nl_net, rt);
1302 	}
1303 
1304 out:
1305 	if (err) {
1306 #ifdef CONFIG_IPV6_SUBTREES
1307 		/*
1308 		 * If fib6_add_1 has cleared the old leaf pointer in the
1309 		 * super-tree leaf node we have to find a new one for it.
1310 		 */
1311 		if (pn != fn) {
1312 			struct fib6_info *pn_leaf =
1313 				rcu_dereference_protected(pn->leaf,
1314 				    lockdep_is_held(&table->tb6_lock));
1315 			if (pn_leaf == rt) {
1316 				pn_leaf = NULL;
1317 				RCU_INIT_POINTER(pn->leaf, NULL);
1318 				fib6_info_release(rt);
1319 			}
1320 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1321 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1322 							   pn);
1323 #if RT6_DEBUG >= 2
1324 				if (!pn_leaf) {
1325 					WARN_ON(!pn_leaf);
1326 					pn_leaf =
1327 					    info->nl_net->ipv6.fib6_null_entry;
1328 				}
1329 #endif
1330 				fib6_info_hold(pn_leaf);
1331 				rcu_assign_pointer(pn->leaf, pn_leaf);
1332 			}
1333 		}
1334 #endif
1335 		goto failure;
1336 	}
1337 	return err;
1338 
1339 failure:
1340 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1341 	 * 1. fn is an intermediate node and we failed to add the new
1342 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1343 	 * failure case.
1344 	 * 2. fn is the root node in the table and we fail to add the first
1345 	 * default route to it.
1346 	 */
1347 	if (fn &&
1348 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1349 	     (fn->fn_flags & RTN_TL_ROOT &&
1350 	      !rcu_access_pointer(fn->leaf))))
1351 		fib6_repair_tree(info->nl_net, table, fn);
1352 	return err;
1353 }
1354 
1355 /*
1356  *	Routing tree lookup
1357  *
1358  */
1359 
1360 struct lookup_args {
1361 	int			offset;		/* key offset on fib6_info */
1362 	const struct in6_addr	*addr;		/* search key			*/
1363 };
1364 
1365 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1366 					    struct lookup_args *args)
1367 {
1368 	struct fib6_node *fn;
1369 	__be32 dir;
1370 
1371 	if (unlikely(args->offset == 0))
1372 		return NULL;
1373 
1374 	/*
1375 	 *	Descend on a tree
1376 	 */
1377 
1378 	fn = root;
1379 
1380 	for (;;) {
1381 		struct fib6_node *next;
1382 
1383 		dir = addr_bit_set(args->addr, fn->fn_bit);
1384 
1385 		next = dir ? rcu_dereference(fn->right) :
1386 			     rcu_dereference(fn->left);
1387 
1388 		if (next) {
1389 			fn = next;
1390 			continue;
1391 		}
1392 		break;
1393 	}
1394 
1395 	while (fn) {
1396 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1397 
1398 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1399 			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1400 			struct rt6key *key;
1401 
1402 			if (!leaf)
1403 				goto backtrack;
1404 
1405 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1406 
1407 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1408 #ifdef CONFIG_IPV6_SUBTREES
1409 				if (subtree) {
1410 					struct fib6_node *sfn;
1411 					sfn = fib6_node_lookup_1(subtree,
1412 								 args + 1);
1413 					if (!sfn)
1414 						goto backtrack;
1415 					fn = sfn;
1416 				}
1417 #endif
1418 				if (fn->fn_flags & RTN_RTINFO)
1419 					return fn;
1420 			}
1421 		}
1422 backtrack:
1423 		if (fn->fn_flags & RTN_ROOT)
1424 			break;
1425 
1426 		fn = rcu_dereference(fn->parent);
1427 	}
1428 
1429 	return NULL;
1430 }
1431 
1432 /* called with rcu_read_lock() held
1433  */
1434 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1435 				   const struct in6_addr *daddr,
1436 				   const struct in6_addr *saddr)
1437 {
1438 	struct fib6_node *fn;
1439 	struct lookup_args args[] = {
1440 		{
1441 			.offset = offsetof(struct fib6_info, fib6_dst),
1442 			.addr = daddr,
1443 		},
1444 #ifdef CONFIG_IPV6_SUBTREES
1445 		{
1446 			.offset = offsetof(struct fib6_info, fib6_src),
1447 			.addr = saddr,
1448 		},
1449 #endif
1450 		{
1451 			.offset = 0,	/* sentinel */
1452 		}
1453 	};
1454 
1455 	fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1456 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1457 		fn = root;
1458 
1459 	return fn;
1460 }
1461 
1462 /*
1463  *	Get node with specified destination prefix (and source prefix,
1464  *	if subtrees are used)
1465  *	exact_match == true means we try to find fn with exact match of
1466  *	the passed in prefix addr
1467  *	exact_match == false means we try to find fn with longest prefix
1468  *	match of the passed in prefix addr. This is useful for finding fn
1469  *	for cached route as it will be stored in the exception table under
1470  *	the node with longest prefix length.
1471  */
1472 
1473 
1474 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1475 				       const struct in6_addr *addr,
1476 				       int plen, int offset,
1477 				       bool exact_match)
1478 {
1479 	struct fib6_node *fn, *prev = NULL;
1480 
1481 	for (fn = root; fn ; ) {
1482 		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1483 		struct rt6key *key;
1484 
1485 		/* This node is being deleted */
1486 		if (!leaf) {
1487 			if (plen <= fn->fn_bit)
1488 				goto out;
1489 			else
1490 				goto next;
1491 		}
1492 
1493 		key = (struct rt6key *)((u8 *)leaf + offset);
1494 
1495 		/*
1496 		 *	Prefix match
1497 		 */
1498 		if (plen < fn->fn_bit ||
1499 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1500 			goto out;
1501 
1502 		if (plen == fn->fn_bit)
1503 			return fn;
1504 
1505 		prev = fn;
1506 
1507 next:
1508 		/*
1509 		 *	We have more bits to go
1510 		 */
1511 		if (addr_bit_set(addr, fn->fn_bit))
1512 			fn = rcu_dereference(fn->right);
1513 		else
1514 			fn = rcu_dereference(fn->left);
1515 	}
1516 out:
1517 	if (exact_match)
1518 		return NULL;
1519 	else
1520 		return prev;
1521 }
1522 
1523 struct fib6_node *fib6_locate(struct fib6_node *root,
1524 			      const struct in6_addr *daddr, int dst_len,
1525 			      const struct in6_addr *saddr, int src_len,
1526 			      bool exact_match)
1527 {
1528 	struct fib6_node *fn;
1529 
1530 	fn = fib6_locate_1(root, daddr, dst_len,
1531 			   offsetof(struct fib6_info, fib6_dst),
1532 			   exact_match);
1533 
1534 #ifdef CONFIG_IPV6_SUBTREES
1535 	if (src_len) {
1536 		WARN_ON(saddr == NULL);
1537 		if (fn) {
1538 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1539 
1540 			if (subtree) {
1541 				fn = fib6_locate_1(subtree, saddr, src_len,
1542 					   offsetof(struct fib6_info, fib6_src),
1543 					   exact_match);
1544 			}
1545 		}
1546 	}
1547 #endif
1548 
1549 	if (fn && fn->fn_flags & RTN_RTINFO)
1550 		return fn;
1551 
1552 	return NULL;
1553 }
1554 
1555 
1556 /*
1557  *	Deletion
1558  *
1559  */
1560 
1561 static struct fib6_info *fib6_find_prefix(struct net *net,
1562 					 struct fib6_table *table,
1563 					 struct fib6_node *fn)
1564 {
1565 	struct fib6_node *child_left, *child_right;
1566 
1567 	if (fn->fn_flags & RTN_ROOT)
1568 		return net->ipv6.fib6_null_entry;
1569 
1570 	while (fn) {
1571 		child_left = rcu_dereference_protected(fn->left,
1572 				    lockdep_is_held(&table->tb6_lock));
1573 		child_right = rcu_dereference_protected(fn->right,
1574 				    lockdep_is_held(&table->tb6_lock));
1575 		if (child_left)
1576 			return rcu_dereference_protected(child_left->leaf,
1577 					lockdep_is_held(&table->tb6_lock));
1578 		if (child_right)
1579 			return rcu_dereference_protected(child_right->leaf,
1580 					lockdep_is_held(&table->tb6_lock));
1581 
1582 		fn = FIB6_SUBTREE(fn);
1583 	}
1584 	return NULL;
1585 }
1586 
1587 /*
1588  *	Called to trim the tree of intermediate nodes when possible. "fn"
1589  *	is the node we want to try and remove.
1590  *	Need to own table->tb6_lock
1591  */
1592 
1593 static struct fib6_node *fib6_repair_tree(struct net *net,
1594 					  struct fib6_table *table,
1595 					  struct fib6_node *fn)
1596 {
1597 	int children;
1598 	int nstate;
1599 	struct fib6_node *child;
1600 	struct fib6_walker *w;
1601 	int iter = 0;
1602 
1603 	/* Set fn->leaf to null_entry for root node. */
1604 	if (fn->fn_flags & RTN_TL_ROOT) {
1605 		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1606 		return fn;
1607 	}
1608 
1609 	for (;;) {
1610 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1611 					    lockdep_is_held(&table->tb6_lock));
1612 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1613 					    lockdep_is_held(&table->tb6_lock));
1614 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1615 					    lockdep_is_held(&table->tb6_lock));
1616 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1617 					    lockdep_is_held(&table->tb6_lock));
1618 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1619 					    lockdep_is_held(&table->tb6_lock));
1620 		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1621 					    lockdep_is_held(&table->tb6_lock));
1622 		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1623 					    lockdep_is_held(&table->tb6_lock));
1624 		struct fib6_info *new_fn_leaf;
1625 
1626 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1627 		iter++;
1628 
1629 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1630 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1631 		WARN_ON(fn_leaf);
1632 
1633 		children = 0;
1634 		child = NULL;
1635 		if (fn_r)
1636 			child = fn_r, children |= 1;
1637 		if (fn_l)
1638 			child = fn_l, children |= 2;
1639 
1640 		if (children == 3 || FIB6_SUBTREE(fn)
1641 #ifdef CONFIG_IPV6_SUBTREES
1642 		    /* Subtree root (i.e. fn) may have one child */
1643 		    || (children && fn->fn_flags & RTN_ROOT)
1644 #endif
1645 		    ) {
1646 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1647 #if RT6_DEBUG >= 2
1648 			if (!new_fn_leaf) {
1649 				WARN_ON(!new_fn_leaf);
1650 				new_fn_leaf = net->ipv6.fib6_null_entry;
1651 			}
1652 #endif
1653 			fib6_info_hold(new_fn_leaf);
1654 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1655 			return pn;
1656 		}
1657 
1658 #ifdef CONFIG_IPV6_SUBTREES
1659 		if (FIB6_SUBTREE(pn) == fn) {
1660 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1661 			RCU_INIT_POINTER(pn->subtree, NULL);
1662 			nstate = FWS_L;
1663 		} else {
1664 			WARN_ON(fn->fn_flags & RTN_ROOT);
1665 #endif
1666 			if (pn_r == fn)
1667 				rcu_assign_pointer(pn->right, child);
1668 			else if (pn_l == fn)
1669 				rcu_assign_pointer(pn->left, child);
1670 #if RT6_DEBUG >= 2
1671 			else
1672 				WARN_ON(1);
1673 #endif
1674 			if (child)
1675 				rcu_assign_pointer(child->parent, pn);
1676 			nstate = FWS_R;
1677 #ifdef CONFIG_IPV6_SUBTREES
1678 		}
1679 #endif
1680 
1681 		read_lock(&net->ipv6.fib6_walker_lock);
1682 		FOR_WALKERS(net, w) {
1683 			if (!child) {
1684 				if (w->node == fn) {
1685 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1686 					w->node = pn;
1687 					w->state = nstate;
1688 				}
1689 			} else {
1690 				if (w->node == fn) {
1691 					w->node = child;
1692 					if (children&2) {
1693 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1694 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1695 					} else {
1696 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1697 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1698 					}
1699 				}
1700 			}
1701 		}
1702 		read_unlock(&net->ipv6.fib6_walker_lock);
1703 
1704 		node_free(net, fn);
1705 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1706 			return pn;
1707 
1708 		RCU_INIT_POINTER(pn->leaf, NULL);
1709 		fib6_info_release(pn_leaf);
1710 		fn = pn;
1711 	}
1712 }
1713 
1714 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1715 			   struct fib6_info __rcu **rtp, struct nl_info *info)
1716 {
1717 	struct fib6_walker *w;
1718 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1719 				    lockdep_is_held(&table->tb6_lock));
1720 	struct net *net = info->nl_net;
1721 
1722 	RT6_TRACE("fib6_del_route\n");
1723 
1724 	/* Unlink it */
1725 	*rtp = rt->fib6_next;
1726 	rt->fib6_node = NULL;
1727 	net->ipv6.rt6_stats->fib_rt_entries--;
1728 	net->ipv6.rt6_stats->fib_discarded_routes++;
1729 
1730 	/* Flush all cached dst in exception table */
1731 	rt6_flush_exceptions(rt);
1732 
1733 	/* Reset round-robin state, if necessary */
1734 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1735 		fn->rr_ptr = NULL;
1736 
1737 	/* Remove this entry from other siblings */
1738 	if (rt->fib6_nsiblings) {
1739 		struct fib6_info *sibling, *next_sibling;
1740 
1741 		list_for_each_entry_safe(sibling, next_sibling,
1742 					 &rt->fib6_siblings, fib6_siblings)
1743 			sibling->fib6_nsiblings--;
1744 		rt->fib6_nsiblings = 0;
1745 		list_del_init(&rt->fib6_siblings);
1746 		rt6_multipath_rebalance(next_sibling);
1747 	}
1748 
1749 	/* Adjust walkers */
1750 	read_lock(&net->ipv6.fib6_walker_lock);
1751 	FOR_WALKERS(net, w) {
1752 		if (w->state == FWS_C && w->leaf == rt) {
1753 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1754 			w->leaf = rcu_dereference_protected(rt->fib6_next,
1755 					    lockdep_is_held(&table->tb6_lock));
1756 			if (!w->leaf)
1757 				w->state = FWS_U;
1758 		}
1759 	}
1760 	read_unlock(&net->ipv6.fib6_walker_lock);
1761 
1762 	/* If it was last route, call fib6_repair_tree() to:
1763 	 * 1. For root node, put back null_entry as how the table was created.
1764 	 * 2. For other nodes, expunge its radix tree node.
1765 	 */
1766 	if (!rcu_access_pointer(fn->leaf)) {
1767 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1768 			fn->fn_flags &= ~RTN_RTINFO;
1769 			net->ipv6.rt6_stats->fib_route_nodes--;
1770 		}
1771 		fn = fib6_repair_tree(net, table, fn);
1772 	}
1773 
1774 	fib6_purge_rt(rt, fn, net);
1775 
1776 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1777 	if (!info->skip_notify)
1778 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1779 	fib6_info_release(rt);
1780 }
1781 
1782 /* Need to own table->tb6_lock */
1783 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1784 {
1785 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1786 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1787 	struct fib6_table *table = rt->fib6_table;
1788 	struct net *net = info->nl_net;
1789 	struct fib6_info __rcu **rtp;
1790 	struct fib6_info __rcu **rtp_next;
1791 
1792 	if (!fn || rt == net->ipv6.fib6_null_entry)
1793 		return -ENOENT;
1794 
1795 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1796 
1797 	/*
1798 	 *	Walk the leaf entries looking for ourself
1799 	 */
1800 
1801 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1802 		struct fib6_info *cur = rcu_dereference_protected(*rtp,
1803 					lockdep_is_held(&table->tb6_lock));
1804 		if (rt == cur) {
1805 			fib6_del_route(table, fn, rtp, info);
1806 			return 0;
1807 		}
1808 		rtp_next = &cur->fib6_next;
1809 	}
1810 	return -ENOENT;
1811 }
1812 
1813 /*
1814  *	Tree traversal function.
1815  *
1816  *	Certainly, it is not interrupt safe.
1817  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1818  *	It means, that we can modify tree during walking
1819  *	and use this function for garbage collection, clone pruning,
1820  *	cleaning tree when a device goes down etc. etc.
1821  *
1822  *	It guarantees that every node will be traversed,
1823  *	and that it will be traversed only once.
1824  *
1825  *	Callback function w->func may return:
1826  *	0 -> continue walking.
1827  *	positive value -> walking is suspended (used by tree dumps,
1828  *	and probably by gc, if it will be split to several slices)
1829  *	negative value -> terminate walking.
1830  *
1831  *	The function itself returns:
1832  *	0   -> walk is complete.
1833  *	>0  -> walk is incomplete (i.e. suspended)
1834  *	<0  -> walk is terminated by an error.
1835  *
1836  *	This function is called with tb6_lock held.
1837  */
1838 
1839 static int fib6_walk_continue(struct fib6_walker *w)
1840 {
1841 	struct fib6_node *fn, *pn, *left, *right;
1842 
1843 	/* w->root should always be table->tb6_root */
1844 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1845 
1846 	for (;;) {
1847 		fn = w->node;
1848 		if (!fn)
1849 			return 0;
1850 
1851 		switch (w->state) {
1852 #ifdef CONFIG_IPV6_SUBTREES
1853 		case FWS_S:
1854 			if (FIB6_SUBTREE(fn)) {
1855 				w->node = FIB6_SUBTREE(fn);
1856 				continue;
1857 			}
1858 			w->state = FWS_L;
1859 #endif
1860 			/* fall through */
1861 		case FWS_L:
1862 			left = rcu_dereference_protected(fn->left, 1);
1863 			if (left) {
1864 				w->node = left;
1865 				w->state = FWS_INIT;
1866 				continue;
1867 			}
1868 			w->state = FWS_R;
1869 			/* fall through */
1870 		case FWS_R:
1871 			right = rcu_dereference_protected(fn->right, 1);
1872 			if (right) {
1873 				w->node = right;
1874 				w->state = FWS_INIT;
1875 				continue;
1876 			}
1877 			w->state = FWS_C;
1878 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1879 			/* fall through */
1880 		case FWS_C:
1881 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1882 				int err;
1883 
1884 				if (w->skip) {
1885 					w->skip--;
1886 					goto skip;
1887 				}
1888 
1889 				err = w->func(w);
1890 				if (err)
1891 					return err;
1892 
1893 				w->count++;
1894 				continue;
1895 			}
1896 skip:
1897 			w->state = FWS_U;
1898 			/* fall through */
1899 		case FWS_U:
1900 			if (fn == w->root)
1901 				return 0;
1902 			pn = rcu_dereference_protected(fn->parent, 1);
1903 			left = rcu_dereference_protected(pn->left, 1);
1904 			right = rcu_dereference_protected(pn->right, 1);
1905 			w->node = pn;
1906 #ifdef CONFIG_IPV6_SUBTREES
1907 			if (FIB6_SUBTREE(pn) == fn) {
1908 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1909 				w->state = FWS_L;
1910 				continue;
1911 			}
1912 #endif
1913 			if (left == fn) {
1914 				w->state = FWS_R;
1915 				continue;
1916 			}
1917 			if (right == fn) {
1918 				w->state = FWS_C;
1919 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1920 				continue;
1921 			}
1922 #if RT6_DEBUG >= 2
1923 			WARN_ON(1);
1924 #endif
1925 		}
1926 	}
1927 }
1928 
1929 static int fib6_walk(struct net *net, struct fib6_walker *w)
1930 {
1931 	int res;
1932 
1933 	w->state = FWS_INIT;
1934 	w->node = w->root;
1935 
1936 	fib6_walker_link(net, w);
1937 	res = fib6_walk_continue(w);
1938 	if (res <= 0)
1939 		fib6_walker_unlink(net, w);
1940 	return res;
1941 }
1942 
1943 static int fib6_clean_node(struct fib6_walker *w)
1944 {
1945 	int res;
1946 	struct fib6_info *rt;
1947 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1948 	struct nl_info info = {
1949 		.nl_net = c->net,
1950 	};
1951 
1952 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1953 	    w->node->fn_sernum != c->sernum)
1954 		w->node->fn_sernum = c->sernum;
1955 
1956 	if (!c->func) {
1957 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1958 		w->leaf = NULL;
1959 		return 0;
1960 	}
1961 
1962 	for_each_fib6_walker_rt(w) {
1963 		res = c->func(rt, c->arg);
1964 		if (res == -1) {
1965 			w->leaf = rt;
1966 			res = fib6_del(rt, &info);
1967 			if (res) {
1968 #if RT6_DEBUG >= 2
1969 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1970 					 __func__, rt,
1971 					 rcu_access_pointer(rt->fib6_node),
1972 					 res);
1973 #endif
1974 				continue;
1975 			}
1976 			return 0;
1977 		} else if (res == -2) {
1978 			if (WARN_ON(!rt->fib6_nsiblings))
1979 				continue;
1980 			rt = list_last_entry(&rt->fib6_siblings,
1981 					     struct fib6_info, fib6_siblings);
1982 			continue;
1983 		}
1984 		WARN_ON(res != 0);
1985 	}
1986 	w->leaf = rt;
1987 	return 0;
1988 }
1989 
1990 /*
1991  *	Convenient frontend to tree walker.
1992  *
1993  *	func is called on each route.
1994  *		It may return -2 -> skip multipath route.
1995  *			      -1 -> delete this route.
1996  *		              0  -> continue walking
1997  */
1998 
1999 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2000 			    int (*func)(struct fib6_info *, void *arg),
2001 			    int sernum, void *arg)
2002 {
2003 	struct fib6_cleaner c;
2004 
2005 	c.w.root = root;
2006 	c.w.func = fib6_clean_node;
2007 	c.w.count = 0;
2008 	c.w.skip = 0;
2009 	c.func = func;
2010 	c.sernum = sernum;
2011 	c.arg = arg;
2012 	c.net = net;
2013 
2014 	fib6_walk(net, &c.w);
2015 }
2016 
2017 static void __fib6_clean_all(struct net *net,
2018 			     int (*func)(struct fib6_info *, void *),
2019 			     int sernum, void *arg)
2020 {
2021 	struct fib6_table *table;
2022 	struct hlist_head *head;
2023 	unsigned int h;
2024 
2025 	rcu_read_lock();
2026 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2027 		head = &net->ipv6.fib_table_hash[h];
2028 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2029 			spin_lock_bh(&table->tb6_lock);
2030 			fib6_clean_tree(net, &table->tb6_root,
2031 					func, sernum, arg);
2032 			spin_unlock_bh(&table->tb6_lock);
2033 		}
2034 	}
2035 	rcu_read_unlock();
2036 }
2037 
2038 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2039 		    void *arg)
2040 {
2041 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
2042 }
2043 
2044 static void fib6_flush_trees(struct net *net)
2045 {
2046 	int new_sernum = fib6_new_sernum(net);
2047 
2048 	__fib6_clean_all(net, NULL, new_sernum, NULL);
2049 }
2050 
2051 /*
2052  *	Garbage collection
2053  */
2054 
2055 static int fib6_age(struct fib6_info *rt, void *arg)
2056 {
2057 	struct fib6_gc_args *gc_args = arg;
2058 	unsigned long now = jiffies;
2059 
2060 	/*
2061 	 *	check addrconf expiration here.
2062 	 *	Routes are expired even if they are in use.
2063 	 */
2064 
2065 	if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2066 		if (time_after(now, rt->expires)) {
2067 			RT6_TRACE("expiring %p\n", rt);
2068 			return -1;
2069 		}
2070 		gc_args->more++;
2071 	}
2072 
2073 	/*	Also age clones in the exception table.
2074 	 *	Note, that clones are aged out
2075 	 *	only if they are not in use now.
2076 	 */
2077 	rt6_age_exceptions(rt, gc_args, now);
2078 
2079 	return 0;
2080 }
2081 
2082 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2083 {
2084 	struct fib6_gc_args gc_args;
2085 	unsigned long now;
2086 
2087 	if (force) {
2088 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2089 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2090 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2091 		return;
2092 	}
2093 	gc_args.timeout = expires ? (int)expires :
2094 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2095 	gc_args.more = 0;
2096 
2097 	fib6_clean_all(net, fib6_age, &gc_args);
2098 	now = jiffies;
2099 	net->ipv6.ip6_rt_last_gc = now;
2100 
2101 	if (gc_args.more)
2102 		mod_timer(&net->ipv6.ip6_fib_timer,
2103 			  round_jiffies(now
2104 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2105 	else
2106 		del_timer(&net->ipv6.ip6_fib_timer);
2107 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2108 }
2109 
2110 static void fib6_gc_timer_cb(struct timer_list *t)
2111 {
2112 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2113 
2114 	fib6_run_gc(0, arg, true);
2115 }
2116 
2117 static int __net_init fib6_net_init(struct net *net)
2118 {
2119 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2120 	int err;
2121 
2122 	err = fib6_notifier_init(net);
2123 	if (err)
2124 		return err;
2125 
2126 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2127 	rwlock_init(&net->ipv6.fib6_walker_lock);
2128 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2129 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2130 
2131 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2132 	if (!net->ipv6.rt6_stats)
2133 		goto out_timer;
2134 
2135 	/* Avoid false sharing : Use at least a full cache line */
2136 	size = max_t(size_t, size, L1_CACHE_BYTES);
2137 
2138 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2139 	if (!net->ipv6.fib_table_hash)
2140 		goto out_rt6_stats;
2141 
2142 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2143 					  GFP_KERNEL);
2144 	if (!net->ipv6.fib6_main_tbl)
2145 		goto out_fib_table_hash;
2146 
2147 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2148 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2149 			   net->ipv6.fib6_null_entry);
2150 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2151 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2152 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2153 
2154 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2155 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2156 					   GFP_KERNEL);
2157 	if (!net->ipv6.fib6_local_tbl)
2158 		goto out_fib6_main_tbl;
2159 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2160 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2161 			   net->ipv6.fib6_null_entry);
2162 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2163 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2164 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2165 #endif
2166 	fib6_tables_init(net);
2167 
2168 	return 0;
2169 
2170 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2171 out_fib6_main_tbl:
2172 	kfree(net->ipv6.fib6_main_tbl);
2173 #endif
2174 out_fib_table_hash:
2175 	kfree(net->ipv6.fib_table_hash);
2176 out_rt6_stats:
2177 	kfree(net->ipv6.rt6_stats);
2178 out_timer:
2179 	fib6_notifier_exit(net);
2180 	return -ENOMEM;
2181 }
2182 
2183 static void fib6_net_exit(struct net *net)
2184 {
2185 	unsigned int i;
2186 
2187 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2188 
2189 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2190 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2191 		struct hlist_node *tmp;
2192 		struct fib6_table *tb;
2193 
2194 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2195 			hlist_del(&tb->tb6_hlist);
2196 			fib6_free_table(tb);
2197 		}
2198 	}
2199 
2200 	kfree(net->ipv6.fib_table_hash);
2201 	kfree(net->ipv6.rt6_stats);
2202 	fib6_notifier_exit(net);
2203 }
2204 
2205 static struct pernet_operations fib6_net_ops = {
2206 	.init = fib6_net_init,
2207 	.exit = fib6_net_exit,
2208 };
2209 
2210 int __init fib6_init(void)
2211 {
2212 	int ret = -ENOMEM;
2213 
2214 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2215 					   sizeof(struct fib6_node),
2216 					   0, SLAB_HWCACHE_ALIGN,
2217 					   NULL);
2218 	if (!fib6_node_kmem)
2219 		goto out;
2220 
2221 	ret = register_pernet_subsys(&fib6_net_ops);
2222 	if (ret)
2223 		goto out_kmem_cache_create;
2224 
2225 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2226 				   inet6_dump_fib, 0);
2227 	if (ret)
2228 		goto out_unregister_subsys;
2229 
2230 	__fib6_flush_trees = fib6_flush_trees;
2231 out:
2232 	return ret;
2233 
2234 out_unregister_subsys:
2235 	unregister_pernet_subsys(&fib6_net_ops);
2236 out_kmem_cache_create:
2237 	kmem_cache_destroy(fib6_node_kmem);
2238 	goto out;
2239 }
2240 
2241 void fib6_gc_cleanup(void)
2242 {
2243 	unregister_pernet_subsys(&fib6_net_ops);
2244 	kmem_cache_destroy(fib6_node_kmem);
2245 }
2246 
2247 #ifdef CONFIG_PROC_FS
2248 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2249 {
2250 	struct fib6_info *rt = v;
2251 	struct ipv6_route_iter *iter = seq->private;
2252 	const struct net_device *dev;
2253 
2254 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2255 
2256 #ifdef CONFIG_IPV6_SUBTREES
2257 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2258 #else
2259 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2260 #endif
2261 	if (rt->fib6_flags & RTF_GATEWAY)
2262 		seq_printf(seq, "%pi6", &rt->fib6_nh.nh_gw);
2263 	else
2264 		seq_puts(seq, "00000000000000000000000000000000");
2265 
2266 	dev = rt->fib6_nh.nh_dev;
2267 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2268 		   rt->fib6_metric, atomic_read(&rt->fib6_ref), 0,
2269 		   rt->fib6_flags, dev ? dev->name : "");
2270 	iter->w.leaf = NULL;
2271 	return 0;
2272 }
2273 
2274 static int ipv6_route_yield(struct fib6_walker *w)
2275 {
2276 	struct ipv6_route_iter *iter = w->args;
2277 
2278 	if (!iter->skip)
2279 		return 1;
2280 
2281 	do {
2282 		iter->w.leaf = rcu_dereference_protected(
2283 				iter->w.leaf->fib6_next,
2284 				lockdep_is_held(&iter->tbl->tb6_lock));
2285 		iter->skip--;
2286 		if (!iter->skip && iter->w.leaf)
2287 			return 1;
2288 	} while (iter->w.leaf);
2289 
2290 	return 0;
2291 }
2292 
2293 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2294 				      struct net *net)
2295 {
2296 	memset(&iter->w, 0, sizeof(iter->w));
2297 	iter->w.func = ipv6_route_yield;
2298 	iter->w.root = &iter->tbl->tb6_root;
2299 	iter->w.state = FWS_INIT;
2300 	iter->w.node = iter->w.root;
2301 	iter->w.args = iter;
2302 	iter->sernum = iter->w.root->fn_sernum;
2303 	INIT_LIST_HEAD(&iter->w.lh);
2304 	fib6_walker_link(net, &iter->w);
2305 }
2306 
2307 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2308 						    struct net *net)
2309 {
2310 	unsigned int h;
2311 	struct hlist_node *node;
2312 
2313 	if (tbl) {
2314 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2315 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2316 	} else {
2317 		h = 0;
2318 		node = NULL;
2319 	}
2320 
2321 	while (!node && h < FIB6_TABLE_HASHSZ) {
2322 		node = rcu_dereference_bh(
2323 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2324 	}
2325 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2326 }
2327 
2328 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2329 {
2330 	if (iter->sernum != iter->w.root->fn_sernum) {
2331 		iter->sernum = iter->w.root->fn_sernum;
2332 		iter->w.state = FWS_INIT;
2333 		iter->w.node = iter->w.root;
2334 		WARN_ON(iter->w.skip);
2335 		iter->w.skip = iter->w.count;
2336 	}
2337 }
2338 
2339 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2340 {
2341 	int r;
2342 	struct fib6_info *n;
2343 	struct net *net = seq_file_net(seq);
2344 	struct ipv6_route_iter *iter = seq->private;
2345 
2346 	if (!v)
2347 		goto iter_table;
2348 
2349 	n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2350 	if (n) {
2351 		++*pos;
2352 		return n;
2353 	}
2354 
2355 iter_table:
2356 	ipv6_route_check_sernum(iter);
2357 	spin_lock_bh(&iter->tbl->tb6_lock);
2358 	r = fib6_walk_continue(&iter->w);
2359 	spin_unlock_bh(&iter->tbl->tb6_lock);
2360 	if (r > 0) {
2361 		if (v)
2362 			++*pos;
2363 		return iter->w.leaf;
2364 	} else if (r < 0) {
2365 		fib6_walker_unlink(net, &iter->w);
2366 		return NULL;
2367 	}
2368 	fib6_walker_unlink(net, &iter->w);
2369 
2370 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2371 	if (!iter->tbl)
2372 		return NULL;
2373 
2374 	ipv6_route_seq_setup_walk(iter, net);
2375 	goto iter_table;
2376 }
2377 
2378 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2379 	__acquires(RCU_BH)
2380 {
2381 	struct net *net = seq_file_net(seq);
2382 	struct ipv6_route_iter *iter = seq->private;
2383 
2384 	rcu_read_lock_bh();
2385 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2386 	iter->skip = *pos;
2387 
2388 	if (iter->tbl) {
2389 		ipv6_route_seq_setup_walk(iter, net);
2390 		return ipv6_route_seq_next(seq, NULL, pos);
2391 	} else {
2392 		return NULL;
2393 	}
2394 }
2395 
2396 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2397 {
2398 	struct fib6_walker *w = &iter->w;
2399 	return w->node && !(w->state == FWS_U && w->node == w->root);
2400 }
2401 
2402 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2403 	__releases(RCU_BH)
2404 {
2405 	struct net *net = seq_file_net(seq);
2406 	struct ipv6_route_iter *iter = seq->private;
2407 
2408 	if (ipv6_route_iter_active(iter))
2409 		fib6_walker_unlink(net, &iter->w);
2410 
2411 	rcu_read_unlock_bh();
2412 }
2413 
2414 const struct seq_operations ipv6_route_seq_ops = {
2415 	.start	= ipv6_route_seq_start,
2416 	.next	= ipv6_route_seq_next,
2417 	.stop	= ipv6_route_seq_stop,
2418 	.show	= ipv6_route_seq_show
2419 };
2420 #endif /* CONFIG_PROC_FS */
2421